backends/native: Add basic KMS abstraction building blocks
The intention with KMS abstraction is to hide away accessing the drm
functions behind an API that allows us to have different kind of KMS
implementations, including legacy non-atomic and atomic. The intention
is also that the code interacting with the drm device should be able to
be run in a different thread than the main thread. This means that we
need to make sure that all drm*() API usage must only occur from within
tasks that eventually can be run in the dedicated thread.
The idea here is that MetaKms provides a outward facing API other places
of mutter can use (e.g. MetaGpuKms and friends), while MetaKmsImpl is
an internal implementation that only gets interacted with via "tasks"
posted via the MetaKms object. These tasks will in the future
potentially be run on the dedicated KMS thread. Initially, we don't
create any new threads.
Likewise, MetaKmsDevice is a outward facing representation of a KMS
device, while MetaKmsImplDevice is the corresponding implementation,
which only runs from within the MetaKmsImpl tasks.
This commit only moves opening and closing the device to this new API,
while leaking the fd outside of the impl enclosure, effectively making
the isolation for drm*() calls pointless. This, however, is necessary to
allow gradual porting of drm interaction, and eventually the file
descriptor in MetaGpuKms will be removed. For now, it's harmless, since
everything still run in the main thread.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-01-29 09:24:44 +00:00
|
|
|
/*
|
2020-07-17 07:38:11 +00:00
|
|
|
* Copyright (C) 2019-2020 Red Hat
|
backends/native: Add basic KMS abstraction building blocks
The intention with KMS abstraction is to hide away accessing the drm
functions behind an API that allows us to have different kind of KMS
implementations, including legacy non-atomic and atomic. The intention
is also that the code interacting with the drm device should be able to
be run in a different thread than the main thread. This means that we
need to make sure that all drm*() API usage must only occur from within
tasks that eventually can be run in the dedicated thread.
The idea here is that MetaKms provides a outward facing API other places
of mutter can use (e.g. MetaGpuKms and friends), while MetaKmsImpl is
an internal implementation that only gets interacted with via "tasks"
posted via the MetaKms object. These tasks will in the future
potentially be run on the dedicated KMS thread. Initially, we don't
create any new threads.
Likewise, MetaKmsDevice is a outward facing representation of a KMS
device, while MetaKmsImplDevice is the corresponding implementation,
which only runs from within the MetaKmsImpl tasks.
This commit only moves opening and closing the device to this new API,
while leaking the fd outside of the impl enclosure, effectively making
the isolation for drm*() calls pointless. This, however, is necessary to
allow gradual porting of drm interaction, and eventually the file
descriptor in MetaGpuKms will be removed. For now, it's harmless, since
everything still run in the main thread.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-01-29 09:24:44 +00:00
|
|
|
*
|
|
|
|
|
* This program is free software; you can redistribute it and/or
|
|
|
|
|
* modify it under the terms of the GNU General Public License as
|
|
|
|
|
* published by the Free Software Foundation; either version 2 of the
|
|
|
|
|
* License, or (at your option) any later version.
|
|
|
|
|
*
|
|
|
|
|
* This program is distributed in the hope that it will be useful, but
|
|
|
|
|
* WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
|
* General Public License for more details.
|
|
|
|
|
*
|
|
|
|
|
* You should have received a copy of the GNU General Public License
|
|
|
|
|
* along with this program; if not, write to the Free Software
|
|
|
|
|
* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
|
|
|
|
|
* 02111-1307, USA.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#include "config.h"
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
#include "backends/native/meta-kms-impl-device-simple.h"
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
2021-04-06 10:14:24 +00:00
|
|
|
#include "backends/native/meta-backend-native-private.h"
|
2020-09-21 17:32:46 +00:00
|
|
|
#include "backends/native/meta-drm-buffer-gbm.h"
|
2020-07-14 14:48:47 +00:00
|
|
|
#include "backends/native/meta-kms-connector-private.h"
|
2020-07-17 07:38:11 +00:00
|
|
|
#include "backends/native/meta-kms-crtc-private.h"
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
#include "backends/native/meta-kms-device-private.h"
|
2020-07-17 07:38:11 +00:00
|
|
|
#include "backends/native/meta-kms-mode-private.h"
|
2020-07-14 13:10:50 +00:00
|
|
|
#include "backends/native/meta-kms-plane-private.h"
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
#include "backends/native/meta-kms-private.h"
|
|
|
|
|
#include "backends/native/meta-kms-update-private.h"
|
|
|
|
|
#include "backends/native/meta-kms-utils.h"
|
|
|
|
|
|
2021-01-19 18:26:14 +00:00
|
|
|
typedef gboolean (* MetaKmsSimpleProcessFunc) (MetaKmsImplDevice *impl_device,
|
|
|
|
|
MetaKmsUpdate *update,
|
|
|
|
|
gpointer entry_data,
|
|
|
|
|
GError **error);
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
typedef struct _CachedModeSet
|
|
|
|
|
{
|
|
|
|
|
GList *connectors;
|
|
|
|
|
drmModeModeInfo *drm_mode;
|
2021-12-08 20:20:52 +00:00
|
|
|
|
|
|
|
|
int width;
|
|
|
|
|
int height;
|
|
|
|
|
int stride;
|
|
|
|
|
uint32_t format;
|
|
|
|
|
uint64_t modifier;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
} CachedModeSet;
|
backends/native: Add basic KMS abstraction building blocks
The intention with KMS abstraction is to hide away accessing the drm
functions behind an API that allows us to have different kind of KMS
implementations, including legacy non-atomic and atomic. The intention
is also that the code interacting with the drm device should be able to
be run in a different thread than the main thread. This means that we
need to make sure that all drm*() API usage must only occur from within
tasks that eventually can be run in the dedicated thread.
The idea here is that MetaKms provides a outward facing API other places
of mutter can use (e.g. MetaGpuKms and friends), while MetaKmsImpl is
an internal implementation that only gets interacted with via "tasks"
posted via the MetaKms object. These tasks will in the future
potentially be run on the dedicated KMS thread. Initially, we don't
create any new threads.
Likewise, MetaKmsDevice is a outward facing representation of a KMS
device, while MetaKmsImplDevice is the corresponding implementation,
which only runs from within the MetaKmsImpl tasks.
This commit only moves opening and closing the device to this new API,
while leaking the fd outside of the impl enclosure, effectively making
the isolation for drm*() calls pointless. This, however, is necessary to
allow gradual porting of drm interaction, and eventually the file
descriptor in MetaGpuKms will be removed. For now, it's harmless, since
everything still run in the main thread.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-01-29 09:24:44 +00:00
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
struct _MetaKmsImplDeviceSimple
|
backends/native: Add basic KMS abstraction building blocks
The intention with KMS abstraction is to hide away accessing the drm
functions behind an API that allows us to have different kind of KMS
implementations, including legacy non-atomic and atomic. The intention
is also that the code interacting with the drm device should be able to
be run in a different thread than the main thread. This means that we
need to make sure that all drm*() API usage must only occur from within
tasks that eventually can be run in the dedicated thread.
The idea here is that MetaKms provides a outward facing API other places
of mutter can use (e.g. MetaGpuKms and friends), while MetaKmsImpl is
an internal implementation that only gets interacted with via "tasks"
posted via the MetaKms object. These tasks will in the future
potentially be run on the dedicated KMS thread. Initially, we don't
create any new threads.
Likewise, MetaKmsDevice is a outward facing representation of a KMS
device, while MetaKmsImplDevice is the corresponding implementation,
which only runs from within the MetaKmsImpl tasks.
This commit only moves opening and closing the device to this new API,
while leaking the fd outside of the impl enclosure, effectively making
the isolation for drm*() calls pointless. This, however, is necessary to
allow gradual porting of drm interaction, and eventually the file
descriptor in MetaGpuKms will be removed. For now, it's harmless, since
everything still run in the main thread.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-01-29 09:24:44 +00:00
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
MetaKmsImplDevice parent;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
|
|
|
|
GSource *mode_set_fallback_feedback_source;
|
|
|
|
|
GList *mode_set_fallback_page_flip_datas;
|
|
|
|
|
|
|
|
|
|
GList *pending_page_flip_retries;
|
|
|
|
|
GSource *retry_page_flips_source;
|
|
|
|
|
|
|
|
|
|
GList *postponed_page_flip_datas;
|
|
|
|
|
GList *postponed_mode_set_fallback_datas;
|
|
|
|
|
|
2021-04-10 23:36:42 +00:00
|
|
|
GList *posted_page_flip_datas;
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
GHashTable *cached_mode_sets;
|
backends/native: Add basic KMS abstraction building blocks
The intention with KMS abstraction is to hide away accessing the drm
functions behind an API that allows us to have different kind of KMS
implementations, including legacy non-atomic and atomic. The intention
is also that the code interacting with the drm device should be able to
be run in a different thread than the main thread. This means that we
need to make sure that all drm*() API usage must only occur from within
tasks that eventually can be run in the dedicated thread.
The idea here is that MetaKms provides a outward facing API other places
of mutter can use (e.g. MetaGpuKms and friends), while MetaKmsImpl is
an internal implementation that only gets interacted with via "tasks"
posted via the MetaKms object. These tasks will in the future
potentially be run on the dedicated KMS thread. Initially, we don't
create any new threads.
Likewise, MetaKmsDevice is a outward facing representation of a KMS
device, while MetaKmsImplDevice is the corresponding implementation,
which only runs from within the MetaKmsImpl tasks.
This commit only moves opening and closing the device to this new API,
while leaking the fd outside of the impl enclosure, effectively making
the isolation for drm*() calls pointless. This, however, is necessary to
allow gradual porting of drm interaction, and eventually the file
descriptor in MetaGpuKms will be removed. For now, it's harmless, since
everything still run in the main thread.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-01-29 09:24:44 +00:00
|
|
|
};
|
|
|
|
|
|
2021-04-06 10:14:24 +00:00
|
|
|
static GInitableIface *initable_parent_iface;
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
static void
|
2020-07-17 07:38:11 +00:00
|
|
|
initable_iface_init (GInitableIface *iface);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
G_DEFINE_TYPE_WITH_CODE (MetaKmsImplDeviceSimple, meta_kms_impl_device_simple,
|
|
|
|
|
META_TYPE_KMS_IMPL_DEVICE,
|
|
|
|
|
G_IMPLEMENT_INTERFACE (G_TYPE_INITABLE,
|
|
|
|
|
initable_iface_init))
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
flush_postponed_page_flip_datas (MetaKmsImplDeviceSimple *impl_device_simple);
|
backends/native: Add basic KMS abstraction building blocks
The intention with KMS abstraction is to hide away accessing the drm
functions behind an API that allows us to have different kind of KMS
implementations, including legacy non-atomic and atomic. The intention
is also that the code interacting with the drm device should be able to
be run in a different thread than the main thread. This means that we
need to make sure that all drm*() API usage must only occur from within
tasks that eventually can be run in the dedicated thread.
The idea here is that MetaKms provides a outward facing API other places
of mutter can use (e.g. MetaGpuKms and friends), while MetaKmsImpl is
an internal implementation that only gets interacted with via "tasks"
posted via the MetaKms object. These tasks will in the future
potentially be run on the dedicated KMS thread. Initially, we don't
create any new threads.
Likewise, MetaKmsDevice is a outward facing representation of a KMS
device, while MetaKmsImplDevice is the corresponding implementation,
which only runs from within the MetaKmsImpl tasks.
This commit only moves opening and closing the device to this new API,
while leaking the fd outside of the impl enclosure, effectively making
the isolation for drm*() calls pointless. This, however, is necessary to
allow gradual porting of drm interaction, and eventually the file
descriptor in MetaGpuKms will be removed. For now, it's harmless, since
everything still run in the main thread.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-01-29 09:24:44 +00:00
|
|
|
|
2020-10-21 15:38:44 +00:00
|
|
|
static gboolean
|
|
|
|
|
get_connector_property (MetaKmsImplDevice *impl_device,
|
|
|
|
|
MetaKmsConnector *connector,
|
|
|
|
|
MetaKmsConnectorProp prop,
|
|
|
|
|
uint64_t *value,
|
|
|
|
|
GError **error)
|
|
|
|
|
{
|
|
|
|
|
uint32_t prop_id;
|
|
|
|
|
int fd;
|
|
|
|
|
drmModeConnector *drm_connector;
|
|
|
|
|
int i;
|
|
|
|
|
gboolean found;
|
|
|
|
|
|
|
|
|
|
prop_id = meta_kms_connector_get_prop_id (connector, prop);
|
|
|
|
|
if (!prop_id)
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, G_IO_ERROR, G_IO_ERROR_NOT_SUPPORTED,
|
|
|
|
|
"Property (%s) not found on connector %u",
|
|
|
|
|
meta_kms_connector_get_prop_name (connector, prop),
|
|
|
|
|
meta_kms_connector_get_id (connector));
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
fd = meta_kms_impl_device_get_fd (impl_device);
|
|
|
|
|
|
|
|
|
|
drm_connector = drmModeGetConnector (fd,
|
|
|
|
|
meta_kms_connector_get_id (connector));
|
|
|
|
|
if (!drm_connector)
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, G_IO_ERROR, g_io_error_from_errno (errno),
|
|
|
|
|
"Failed to get connector %u resources: %s",
|
|
|
|
|
meta_kms_connector_get_id (connector),
|
|
|
|
|
g_strerror (errno));
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
found = FALSE;
|
|
|
|
|
for (i = 0; i < drm_connector->count_props; i++)
|
|
|
|
|
{
|
|
|
|
|
if (drm_connector->props[i] == prop_id)
|
|
|
|
|
{
|
|
|
|
|
*value = drm_connector->prop_values[i];
|
|
|
|
|
found = TRUE;
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
drmModeFreeConnector (drm_connector);
|
|
|
|
|
|
|
|
|
|
if (!found)
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, G_IO_ERROR, G_IO_ERROR_NOT_FOUND,
|
|
|
|
|
"Connector property %u not found", prop_id);
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
static gboolean
|
2020-07-17 07:38:11 +00:00
|
|
|
set_connector_property (MetaKmsImplDevice *impl_device,
|
|
|
|
|
MetaKmsConnector *connector,
|
2020-07-14 14:48:47 +00:00
|
|
|
MetaKmsConnectorProp prop,
|
|
|
|
|
uint64_t value,
|
|
|
|
|
GError **error)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2020-07-14 14:48:47 +00:00
|
|
|
uint32_t prop_id;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
int fd;
|
|
|
|
|
int ret;
|
|
|
|
|
|
2020-07-14 14:48:47 +00:00
|
|
|
prop_id = meta_kms_connector_get_prop_id (connector, prop);
|
|
|
|
|
if (!prop_id)
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, G_IO_ERROR, G_IO_ERROR_NOT_SUPPORTED,
|
|
|
|
|
"Property (%s) not found on connector %u",
|
|
|
|
|
meta_kms_connector_get_prop_name (connector, prop),
|
|
|
|
|
meta_kms_connector_get_id (connector));
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
fd = meta_kms_impl_device_get_fd (impl_device);
|
|
|
|
|
|
|
|
|
|
ret = drmModeObjectSetProperty (fd,
|
|
|
|
|
meta_kms_connector_get_id (connector),
|
|
|
|
|
DRM_MODE_OBJECT_CONNECTOR,
|
2020-07-14 14:48:47 +00:00
|
|
|
prop_id,
|
|
|
|
|
value);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
if (ret != 0)
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, G_IO_ERROR, g_io_error_from_errno (-ret),
|
|
|
|
|
"Failed to set connector %u property %u: %s",
|
|
|
|
|
meta_kms_connector_get_id (connector),
|
2020-07-14 14:48:47 +00:00
|
|
|
prop_id,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
g_strerror (-ret));
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
2020-07-14 14:48:47 +00:00
|
|
|
static gboolean
|
2020-10-21 15:38:44 +00:00
|
|
|
process_power_save (MetaKmsImplDevice *impl_device,
|
|
|
|
|
GError **error)
|
2020-07-14 14:48:47 +00:00
|
|
|
{
|
2020-10-21 15:38:44 +00:00
|
|
|
GList *l;
|
2020-07-14 14:48:47 +00:00
|
|
|
|
2020-10-21 15:38:44 +00:00
|
|
|
for (l = meta_kms_impl_device_peek_connectors (impl_device); l; l = l->next)
|
2020-07-14 14:48:47 +00:00
|
|
|
{
|
2020-10-21 15:38:44 +00:00
|
|
|
MetaKmsConnector *connector = l->data;
|
|
|
|
|
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
2020-10-21 15:38:44 +00:00
|
|
|
"[simple] Setting DPMS of connector %u (%s) to OFF",
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_kms_connector_get_id (connector),
|
2020-10-21 15:38:44 +00:00
|
|
|
meta_kms_impl_device_get_path (impl_device));
|
2020-10-02 16:37:18 +00:00
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!set_connector_property (impl_device,
|
|
|
|
|
connector,
|
2020-07-14 14:48:47 +00:00
|
|
|
META_KMS_CONNECTOR_PROP_DPMS,
|
2020-10-21 15:38:44 +00:00
|
|
|
DRM_MODE_DPMS_OFF,
|
2020-07-14 14:48:47 +00:00
|
|
|
error))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
2020-10-21 15:38:44 +00:00
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static gboolean
|
|
|
|
|
process_connector_update (MetaKmsImplDevice *impl_device,
|
|
|
|
|
MetaKmsUpdate *update,
|
|
|
|
|
gpointer update_entry,
|
|
|
|
|
GError **error)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsConnectorUpdate *connector_update = update_entry;
|
|
|
|
|
MetaKmsConnector *connector = connector_update->connector;
|
|
|
|
|
|
2020-07-14 14:48:47 +00:00
|
|
|
if (connector_update->underscanning.has_update &&
|
|
|
|
|
connector_update->underscanning.is_active)
|
|
|
|
|
{
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Setting underscanning on connector %u (%s) to "
|
|
|
|
|
"%" G_GUINT64_FORMAT "x%" G_GUINT64_FORMAT,
|
|
|
|
|
meta_kms_connector_get_id (connector),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device),
|
|
|
|
|
connector_update->underscanning.hborder,
|
|
|
|
|
connector_update->underscanning.vborder);
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!set_connector_property (impl_device,
|
|
|
|
|
connector,
|
2020-07-14 14:48:47 +00:00
|
|
|
META_KMS_CONNECTOR_PROP_UNDERSCAN,
|
|
|
|
|
1,
|
|
|
|
|
error))
|
|
|
|
|
return FALSE;
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!set_connector_property (impl_device,
|
|
|
|
|
connector,
|
2020-07-14 14:48:47 +00:00
|
|
|
META_KMS_CONNECTOR_PROP_UNDERSCAN_HBORDER,
|
|
|
|
|
connector_update->underscanning.hborder,
|
|
|
|
|
error))
|
|
|
|
|
return FALSE;
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!set_connector_property (impl_device,
|
|
|
|
|
connector,
|
2020-07-14 14:48:47 +00:00
|
|
|
META_KMS_CONNECTOR_PROP_UNDERSCAN_VBORDER,
|
|
|
|
|
connector_update->underscanning.vborder,
|
|
|
|
|
error))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
else if (connector_update->underscanning.has_update)
|
|
|
|
|
{
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Unsetting underscanning on connector %u (%s)",
|
|
|
|
|
meta_kms_connector_get_id (connector),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device));
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!set_connector_property (impl_device,
|
|
|
|
|
connector,
|
2020-07-14 14:48:47 +00:00
|
|
|
META_KMS_CONNECTOR_PROP_UNDERSCAN,
|
|
|
|
|
0,
|
|
|
|
|
error))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
static CachedModeSet *
|
|
|
|
|
cached_mode_set_new (GList *connectors,
|
2021-12-08 20:20:52 +00:00
|
|
|
const drmModeModeInfo *drm_mode,
|
|
|
|
|
MetaDrmBuffer *buffer)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
|
|
|
|
CachedModeSet *cached_mode_set;
|
|
|
|
|
|
2021-12-08 20:20:52 +00:00
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
cached_mode_set = g_new0 (CachedModeSet, 1);
|
|
|
|
|
*cached_mode_set = (CachedModeSet) {
|
|
|
|
|
.connectors = g_list_copy (connectors),
|
2021-02-04 17:45:59 +00:00
|
|
|
.drm_mode = g_memdup2 (drm_mode, sizeof *drm_mode),
|
2021-12-08 20:20:52 +00:00
|
|
|
.width = meta_drm_buffer_get_width (buffer),
|
|
|
|
|
.height = meta_drm_buffer_get_height (buffer),
|
|
|
|
|
.stride = meta_drm_buffer_get_stride (buffer),
|
|
|
|
|
.format = meta_drm_buffer_get_format (buffer),
|
|
|
|
|
.modifier = meta_drm_buffer_get_modifier (buffer),
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
return cached_mode_set;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
cached_mode_set_free (CachedModeSet *cached_mode_set)
|
|
|
|
|
{
|
|
|
|
|
g_list_free (cached_mode_set->connectors);
|
|
|
|
|
g_free (cached_mode_set->drm_mode);
|
|
|
|
|
g_free (cached_mode_set);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
fill_connector_ids_array (GList *connectors,
|
|
|
|
|
uint32_t **out_connectors,
|
|
|
|
|
int *out_n_connectors)
|
|
|
|
|
{
|
|
|
|
|
GList *l;
|
|
|
|
|
int i;
|
|
|
|
|
|
|
|
|
|
*out_n_connectors = g_list_length (connectors);
|
|
|
|
|
*out_connectors = g_new0 (uint32_t, *out_n_connectors);
|
|
|
|
|
i = 0;
|
|
|
|
|
for (l = connectors; l; l = l->next)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsConnector *connector = l->data;
|
|
|
|
|
|
|
|
|
|
(*out_connectors)[i++] = meta_kms_connector_get_id (connector);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2020-07-14 13:10:50 +00:00
|
|
|
static gboolean
|
2020-07-17 07:38:11 +00:00
|
|
|
set_plane_rotation (MetaKmsImplDevice *impl_device,
|
|
|
|
|
MetaKmsPlane *plane,
|
|
|
|
|
uint64_t rotation,
|
|
|
|
|
GError **error)
|
2020-07-14 13:10:50 +00:00
|
|
|
{
|
|
|
|
|
int fd;
|
|
|
|
|
uint32_t rotation_prop_id;
|
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
|
|
fd = meta_kms_impl_device_get_fd (impl_device);
|
|
|
|
|
|
|
|
|
|
rotation_prop_id = meta_kms_plane_get_prop_id (plane,
|
|
|
|
|
META_KMS_PLANE_PROP_ROTATION);
|
2020-10-02 16:37:18 +00:00
|
|
|
|
|
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Setting plane %u (%s) rotation to %" G_GUINT64_FORMAT,
|
|
|
|
|
meta_kms_plane_get_id (plane),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device),
|
|
|
|
|
rotation);
|
|
|
|
|
|
2020-07-14 13:10:50 +00:00
|
|
|
ret = drmModeObjectSetProperty (fd,
|
|
|
|
|
meta_kms_plane_get_id (plane),
|
|
|
|
|
DRM_MODE_OBJECT_PLANE,
|
|
|
|
|
rotation_prop_id,
|
|
|
|
|
rotation);
|
|
|
|
|
if (ret != 0)
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, G_IO_ERROR, g_io_error_from_errno (-ret),
|
|
|
|
|
"Failed to rotation property (%u) to %" G_GUINT64_FORMAT
|
|
|
|
|
" on plane %u: %s",
|
|
|
|
|
rotation_prop_id,
|
|
|
|
|
rotation,
|
|
|
|
|
meta_kms_plane_get_id (plane),
|
|
|
|
|
g_strerror (-ret));
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
static gboolean
|
2020-07-17 07:38:11 +00:00
|
|
|
process_mode_set (MetaKmsImplDevice *impl_device,
|
|
|
|
|
MetaKmsUpdate *update,
|
|
|
|
|
gpointer update_entry,
|
|
|
|
|
GError **error)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
MetaKmsImplDeviceSimple *impl_device_simple =
|
|
|
|
|
META_KMS_IMPL_DEVICE_SIMPLE (impl_device);
|
2019-11-08 10:30:18 +00:00
|
|
|
MetaKmsModeSet *mode_set = update_entry;
|
2019-06-24 08:35:47 +00:00
|
|
|
MetaKmsCrtc *crtc = mode_set->crtc;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
g_autofree uint32_t *connectors = NULL;
|
|
|
|
|
int n_connectors;
|
|
|
|
|
MetaKmsPlaneAssignment *plane_assignment;
|
2021-12-08 20:20:52 +00:00
|
|
|
MetaDrmBuffer *buffer;
|
2020-07-02 13:58:59 +00:00
|
|
|
drmModeModeInfo *drm_mode;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
uint32_t x, y;
|
|
|
|
|
uint32_t fb_id;
|
|
|
|
|
int fd;
|
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
|
|
crtc = mode_set->crtc;
|
|
|
|
|
|
2020-07-02 13:58:59 +00:00
|
|
|
if (mode_set->mode)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2020-10-21 15:38:44 +00:00
|
|
|
GList *l;
|
2020-09-21 17:32:46 +00:00
|
|
|
|
2020-07-02 13:58:59 +00:00
|
|
|
drm_mode = g_alloca (sizeof *drm_mode);
|
|
|
|
|
*drm_mode = *meta_kms_mode_get_drm_mode (mode_set->mode);
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
fill_connector_ids_array (mode_set->connectors,
|
|
|
|
|
&connectors,
|
|
|
|
|
&n_connectors);
|
|
|
|
|
|
2019-10-04 09:48:46 +00:00
|
|
|
plane_assignment = meta_kms_update_get_primary_plane_assignment (update,
|
|
|
|
|
crtc);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
if (!plane_assignment)
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, G_IO_ERROR, G_IO_ERROR_FAILED,
|
|
|
|
|
"Missing primary plane assignment for legacy mode set on CRTC %u",
|
|
|
|
|
meta_kms_crtc_get_id (crtc));
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
x = meta_fixed_16_to_int (plane_assignment->src_rect.x);
|
|
|
|
|
y = meta_fixed_16_to_int (plane_assignment->src_rect.y);
|
|
|
|
|
|
2020-07-14 13:10:50 +00:00
|
|
|
if (plane_assignment->rotation)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!set_plane_rotation (impl_device,
|
2020-07-14 13:10:50 +00:00
|
|
|
plane_assignment->plane,
|
|
|
|
|
plane_assignment->rotation,
|
|
|
|
|
error))
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
2020-09-21 17:32:46 +00:00
|
|
|
buffer = plane_assignment->buffer;
|
2021-05-05 12:35:48 +00:00
|
|
|
if (!meta_drm_buffer_ensure_fb_id (buffer, error))
|
|
|
|
|
return FALSE;
|
|
|
|
|
|
2020-09-21 17:32:46 +00:00
|
|
|
fb_id = meta_drm_buffer_get_fb_id (buffer);
|
2020-10-02 16:37:18 +00:00
|
|
|
|
2020-10-21 15:38:44 +00:00
|
|
|
for (l = mode_set->connectors; l; l = l->next)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsConnector *connector = l->data;
|
|
|
|
|
uint64_t dpms_value;
|
|
|
|
|
|
|
|
|
|
if (!get_connector_property (impl_device,
|
|
|
|
|
connector,
|
|
|
|
|
META_KMS_CONNECTOR_PROP_DPMS,
|
|
|
|
|
&dpms_value,
|
|
|
|
|
error))
|
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
|
|
if (dpms_value != DRM_MODE_DPMS_ON)
|
|
|
|
|
{
|
|
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Setting DPMS of connector %u (%s) to ON",
|
|
|
|
|
meta_kms_connector_get_id (connector),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device));
|
|
|
|
|
|
|
|
|
|
if (!set_connector_property (impl_device,
|
|
|
|
|
connector,
|
|
|
|
|
META_KMS_CONNECTOR_PROP_DPMS,
|
|
|
|
|
DRM_MODE_DPMS_ON,
|
|
|
|
|
error))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Setting mode of CRTC %u (%s) to %s",
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device),
|
|
|
|
|
drm_mode->name);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
2021-12-08 20:20:52 +00:00
|
|
|
buffer = NULL;
|
2020-07-02 13:58:59 +00:00
|
|
|
drm_mode = NULL;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
x = y = 0;
|
|
|
|
|
n_connectors = 0;
|
|
|
|
|
connectors = NULL;
|
|
|
|
|
fb_id = 0;
|
2020-10-02 16:37:18 +00:00
|
|
|
|
|
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Unsetting mode of CRTC %u (%s)",
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device));
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
fd = meta_kms_impl_device_get_fd (impl_device);
|
|
|
|
|
ret = drmModeSetCrtc (fd,
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
fb_id,
|
|
|
|
|
x, y,
|
|
|
|
|
connectors, n_connectors,
|
2020-07-02 13:58:59 +00:00
|
|
|
drm_mode);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
if (ret != 0)
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, G_IO_ERROR, g_io_error_from_errno (-ret),
|
2020-01-14 10:23:04 +00:00
|
|
|
"Failed to set mode %s on CRTC %u: %s",
|
2020-07-02 13:58:59 +00:00
|
|
|
drm_mode ? drm_mode->name : "off",
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
g_strerror (-ret));
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
2020-07-02 13:58:59 +00:00
|
|
|
if (drm_mode)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
g_hash_table_replace (impl_device_simple->cached_mode_sets,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
crtc,
|
|
|
|
|
cached_mode_set_new (mode_set->connectors,
|
2021-12-08 20:20:52 +00:00
|
|
|
drm_mode,
|
|
|
|
|
buffer));
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
g_hash_table_remove (impl_device_simple->cached_mode_sets, crtc);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
2019-05-03 17:20:21 +00:00
|
|
|
static gboolean
|
2020-07-17 07:38:11 +00:00
|
|
|
process_crtc_gamma (MetaKmsImplDevice *impl_device,
|
|
|
|
|
MetaKmsUpdate *update,
|
|
|
|
|
gpointer update_entry,
|
|
|
|
|
GError **error)
|
2019-05-03 17:20:21 +00:00
|
|
|
{
|
2019-11-08 10:30:18 +00:00
|
|
|
MetaKmsCrtcGamma *gamma = update_entry;
|
2019-05-03 17:20:21 +00:00
|
|
|
MetaKmsCrtc *crtc = gamma->crtc;
|
|
|
|
|
int fd;
|
|
|
|
|
int ret;
|
|
|
|
|
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Setting CRTC %u (%s) gamma, size: %d",
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device),
|
|
|
|
|
gamma->size);
|
|
|
|
|
|
2019-05-03 17:20:21 +00:00
|
|
|
fd = meta_kms_impl_device_get_fd (impl_device);
|
|
|
|
|
ret = drmModeCrtcSetGamma (fd, meta_kms_crtc_get_id (crtc),
|
|
|
|
|
gamma->size,
|
|
|
|
|
gamma->red,
|
|
|
|
|
gamma->green,
|
|
|
|
|
gamma->blue);
|
|
|
|
|
if (ret != 0)
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, G_IO_ERROR, g_io_error_from_errno (-ret),
|
|
|
|
|
"drmModeCrtcSetGamma on CRTC %u failed: %s",
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
g_strerror (-ret));
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
static gboolean
|
|
|
|
|
is_timestamp_earlier_than (uint64_t ts1,
|
|
|
|
|
uint64_t ts2)
|
|
|
|
|
{
|
|
|
|
|
if (ts1 == ts2)
|
|
|
|
|
return FALSE;
|
|
|
|
|
else
|
|
|
|
|
return ts2 - ts1 < UINT64_MAX / 2;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef struct _RetryPageFlipData
|
|
|
|
|
{
|
|
|
|
|
MetaKmsCrtc *crtc;
|
|
|
|
|
uint32_t fb_id;
|
|
|
|
|
MetaKmsPageFlipData *page_flip_data;
|
|
|
|
|
float refresh_rate;
|
|
|
|
|
uint64_t retry_time_us;
|
2021-01-22 10:31:18 +00:00
|
|
|
MetaKmsCustomPageFlip *custom_page_flip;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
} RetryPageFlipData;
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
retry_page_flip_data_free (RetryPageFlipData *retry_page_flip_data)
|
|
|
|
|
{
|
|
|
|
|
g_assert (!retry_page_flip_data->page_flip_data);
|
2021-01-22 10:31:18 +00:00
|
|
|
g_clear_pointer (&retry_page_flip_data->custom_page_flip,
|
|
|
|
|
meta_kms_custom_page_flip_free);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
g_free (retry_page_flip_data);
|
|
|
|
|
}
|
|
|
|
|
|
2020-01-14 10:16:09 +00:00
|
|
|
static CachedModeSet *
|
2020-07-17 07:38:11 +00:00
|
|
|
get_cached_mode_set (MetaKmsImplDeviceSimple *impl_device_simple,
|
|
|
|
|
MetaKmsCrtc *crtc)
|
2020-01-14 10:16:09 +00:00
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
return g_hash_table_lookup (impl_device_simple->cached_mode_sets, crtc);
|
2020-01-14 10:16:09 +00:00
|
|
|
}
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
static float
|
2020-07-17 07:38:11 +00:00
|
|
|
get_cached_crtc_refresh_rate (MetaKmsImplDeviceSimple *impl_device_simple,
|
|
|
|
|
MetaKmsCrtc *crtc)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
|
|
|
|
CachedModeSet *cached_mode_set;
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
cached_mode_set = g_hash_table_lookup (impl_device_simple->cached_mode_sets,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
crtc);
|
|
|
|
|
g_assert (cached_mode_set);
|
|
|
|
|
|
|
|
|
|
return meta_calculate_drm_mode_refresh_rate (cached_mode_set->drm_mode);
|
|
|
|
|
}
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
#define meta_assert_in_kms_impl(kms) \
|
|
|
|
|
g_assert (meta_kms_in_impl_task (kms))
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
static gboolean
|
|
|
|
|
retry_page_flips (gpointer user_data)
|
|
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
MetaKmsImplDeviceSimple *impl_device_simple =
|
|
|
|
|
META_KMS_IMPL_DEVICE_SIMPLE (user_data);
|
|
|
|
|
MetaKmsImplDevice *impl_device = META_KMS_IMPL_DEVICE (impl_device_simple);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
uint64_t now_us;
|
|
|
|
|
GList *l;
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
now_us = g_source_get_time (impl_device_simple->retry_page_flips_source);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
l = impl_device_simple->pending_page_flip_retries;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
while (l)
|
|
|
|
|
{
|
|
|
|
|
RetryPageFlipData *retry_page_flip_data = l->data;
|
|
|
|
|
MetaKmsCrtc *crtc = retry_page_flip_data->crtc;
|
|
|
|
|
GList *l_next = l->next;
|
|
|
|
|
int fd;
|
|
|
|
|
int ret;
|
|
|
|
|
MetaKmsPageFlipData *page_flip_data;
|
2021-01-22 10:31:18 +00:00
|
|
|
MetaKmsCustomPageFlip *custom_page_flip;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
|
|
|
|
if (is_timestamp_earlier_than (now_us,
|
|
|
|
|
retry_page_flip_data->retry_time_us))
|
|
|
|
|
{
|
|
|
|
|
l = l_next;
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
2021-01-22 10:31:18 +00:00
|
|
|
custom_page_flip = retry_page_flip_data->custom_page_flip;
|
|
|
|
|
if (custom_page_flip)
|
2020-07-17 20:38:55 +00:00
|
|
|
{
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Retrying custom page flip on CRTC %u (%s)",
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device));
|
2021-01-22 10:31:18 +00:00
|
|
|
ret = custom_page_flip->func (custom_page_flip->user_data,
|
|
|
|
|
retry_page_flip_data->page_flip_data);
|
2020-07-17 20:38:55 +00:00
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Retrying page flip on CRTC %u (%s) with %u",
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device),
|
|
|
|
|
retry_page_flip_data->fb_id);
|
|
|
|
|
|
2020-07-17 20:38:55 +00:00
|
|
|
fd = meta_kms_impl_device_get_fd (impl_device);
|
|
|
|
|
ret = drmModePageFlip (fd,
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
retry_page_flip_data->fb_id,
|
|
|
|
|
DRM_MODE_PAGE_FLIP_EVENT,
|
|
|
|
|
retry_page_flip_data->page_flip_data);
|
|
|
|
|
}
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
if (ret == -EBUSY)
|
|
|
|
|
{
|
|
|
|
|
float refresh_rate;
|
|
|
|
|
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Rescheduling page flip retry on CRTC %u (%s)",
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device));
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
refresh_rate =
|
|
|
|
|
get_cached_crtc_refresh_rate (impl_device_simple, crtc);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
retry_page_flip_data->retry_time_us +=
|
|
|
|
|
(uint64_t) (G_USEC_PER_SEC / refresh_rate);
|
|
|
|
|
l = l_next;
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
impl_device_simple->pending_page_flip_retries =
|
|
|
|
|
g_list_remove_link (impl_device_simple->pending_page_flip_retries, l);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
|
|
|
|
page_flip_data = g_steal_pointer (&retry_page_flip_data->page_flip_data);
|
|
|
|
|
if (ret != 0)
|
|
|
|
|
{
|
|
|
|
|
g_autoptr (GError) error = NULL;
|
|
|
|
|
|
|
|
|
|
g_set_error (&error, G_IO_ERROR, g_io_error_from_errno (-ret),
|
|
|
|
|
"drmModePageFlip on CRTC %u failed: %s",
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
g_strerror (-ret));
|
|
|
|
|
if (!g_error_matches (error,
|
|
|
|
|
G_IO_ERROR,
|
|
|
|
|
G_IO_ERROR_PERMISSION_DENIED))
|
|
|
|
|
g_critical ("Failed to page flip: %s", error->message);
|
|
|
|
|
|
|
|
|
|
meta_kms_page_flip_data_discard_in_impl (page_flip_data, error);
|
2021-04-12 14:25:53 +00:00
|
|
|
meta_kms_impl_device_unhold_fd (impl_device);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
2021-04-10 23:36:42 +00:00
|
|
|
else
|
|
|
|
|
{
|
|
|
|
|
impl_device_simple->posted_page_flip_datas =
|
|
|
|
|
g_list_prepend (impl_device_simple->posted_page_flip_datas,
|
|
|
|
|
page_flip_data);
|
|
|
|
|
}
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
|
|
|
|
retry_page_flip_data_free (retry_page_flip_data);
|
|
|
|
|
|
|
|
|
|
l = l_next;
|
|
|
|
|
}
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
if (impl_device_simple->pending_page_flip_retries)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
|
|
|
|
GList *l;
|
|
|
|
|
uint64_t earliest_retry_time_us = 0;
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
for (l = impl_device_simple->pending_page_flip_retries; l; l = l->next)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
|
|
|
|
RetryPageFlipData *retry_page_flip_data = l->data;
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
if (l == impl_device_simple->pending_page_flip_retries ||
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
is_timestamp_earlier_than (retry_page_flip_data->retry_time_us,
|
|
|
|
|
earliest_retry_time_us))
|
|
|
|
|
earliest_retry_time_us = retry_page_flip_data->retry_time_us;
|
|
|
|
|
}
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
g_source_set_ready_time (impl_device_simple->retry_page_flips_source,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
earliest_retry_time_us);
|
|
|
|
|
return G_SOURCE_CONTINUE;
|
|
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
g_clear_pointer (&impl_device_simple->retry_page_flips_source,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
g_source_unref);
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
flush_postponed_page_flip_datas (impl_device_simple);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
|
|
|
|
return G_SOURCE_REMOVE;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2021-01-22 10:31:18 +00:00
|
|
|
schedule_retry_page_flip (MetaKmsImplDeviceSimple *impl_device_simple,
|
|
|
|
|
MetaKmsCrtc *crtc,
|
|
|
|
|
uint32_t fb_id,
|
|
|
|
|
float refresh_rate,
|
|
|
|
|
MetaKmsPageFlipData *page_flip_data,
|
|
|
|
|
MetaKmsCustomPageFlip *custom_page_flip)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
|
|
|
|
RetryPageFlipData *retry_page_flip_data;
|
|
|
|
|
uint64_t now_us;
|
|
|
|
|
uint64_t retry_time_us;
|
|
|
|
|
|
|
|
|
|
now_us = g_get_monotonic_time ();
|
|
|
|
|
retry_time_us = now_us + (uint64_t) (G_USEC_PER_SEC / refresh_rate);
|
|
|
|
|
|
|
|
|
|
retry_page_flip_data = g_new0 (RetryPageFlipData, 1);
|
|
|
|
|
*retry_page_flip_data = (RetryPageFlipData) {
|
|
|
|
|
.crtc = crtc,
|
|
|
|
|
.fb_id = fb_id,
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
.page_flip_data = page_flip_data,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
.refresh_rate = refresh_rate,
|
|
|
|
|
.retry_time_us = retry_time_us,
|
2021-01-22 10:31:18 +00:00
|
|
|
.custom_page_flip = custom_page_flip,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
};
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!impl_device_simple->retry_page_flips_source)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
MetaKmsImplDevice *impl_device =
|
|
|
|
|
META_KMS_IMPL_DEVICE (impl_device_simple);
|
|
|
|
|
MetaKmsDevice *device = meta_kms_impl_device_get_device (impl_device);
|
|
|
|
|
MetaKms *kms = meta_kms_device_get_kms (device);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
GSource *source;
|
|
|
|
|
|
|
|
|
|
source = meta_kms_add_source_in_impl (kms, retry_page_flips,
|
2020-07-17 07:38:11 +00:00
|
|
|
impl_device_simple, NULL);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
g_source_set_ready_time (source, retry_time_us);
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
impl_device_simple->retry_page_flips_source = source;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
|
|
|
|
GList *l;
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
for (l = impl_device_simple->pending_page_flip_retries; l; l = l->next)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
|
|
|
|
RetryPageFlipData *pending_retry_page_flip_data = l->data;
|
|
|
|
|
uint64_t pending_retry_time_us =
|
|
|
|
|
pending_retry_page_flip_data->retry_time_us;
|
|
|
|
|
|
|
|
|
|
if (is_timestamp_earlier_than (retry_time_us, pending_retry_time_us))
|
|
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
g_source_set_ready_time (impl_device_simple->retry_page_flips_source,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
retry_time_us);
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
impl_device_simple->pending_page_flip_retries =
|
|
|
|
|
g_list_append (impl_device_simple->pending_page_flip_retries,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
retry_page_flip_data);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2021-04-10 23:32:13 +00:00
|
|
|
dispatch_page_flip_datas (GList **page_flip_datas,
|
|
|
|
|
GFunc func,
|
|
|
|
|
gpointer user_data)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2021-04-10 23:32:13 +00:00
|
|
|
g_list_foreach (*page_flip_datas, func, user_data);
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
g_clear_pointer (page_flip_datas, g_list_free);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static gboolean
|
|
|
|
|
mode_set_fallback_feedback_idle (gpointer user_data)
|
|
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
MetaKmsImplDeviceSimple *impl_device_simple = user_data;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
g_clear_pointer (&impl_device_simple->mode_set_fallback_feedback_source,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
g_source_unref);
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
if (impl_device_simple->pending_page_flip_retries)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
impl_device_simple->postponed_mode_set_fallback_datas =
|
|
|
|
|
g_steal_pointer (&impl_device_simple->mode_set_fallback_page_flip_datas);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
2021-04-10 23:32:13 +00:00
|
|
|
dispatch_page_flip_datas (&impl_device_simple->mode_set_fallback_page_flip_datas,
|
|
|
|
|
(GFunc) meta_kms_page_flip_data_mode_set_fallback_in_impl,
|
|
|
|
|
NULL);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return G_SOURCE_REMOVE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static gboolean
|
2020-07-17 07:38:11 +00:00
|
|
|
mode_set_fallback (MetaKmsImplDeviceSimple *impl_device_simple,
|
|
|
|
|
MetaKmsUpdate *update,
|
|
|
|
|
MetaKmsPlaneAssignment *plane_assignment,
|
|
|
|
|
MetaKmsPageFlipData *page_flip_data,
|
|
|
|
|
GError **error)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
MetaKmsImplDevice *impl_device = META_KMS_IMPL_DEVICE (impl_device_simple);
|
|
|
|
|
MetaKmsDevice *device = meta_kms_impl_device_get_device (impl_device);
|
|
|
|
|
MetaKms *kms = meta_kms_device_get_kms (device);
|
2020-10-02 14:06:35 +00:00
|
|
|
MetaKmsCrtc *crtc = meta_kms_page_flip_data_get_crtc (page_flip_data);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
CachedModeSet *cached_mode_set;
|
|
|
|
|
g_autofree uint32_t *connectors = NULL;
|
|
|
|
|
int n_connectors;
|
2020-09-21 17:32:46 +00:00
|
|
|
uint32_t fb_id;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
uint32_t x, y;
|
|
|
|
|
int fd;
|
|
|
|
|
int ret;
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
cached_mode_set = g_hash_table_lookup (impl_device_simple->cached_mode_sets,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
crtc);
|
2020-01-17 07:28:17 +00:00
|
|
|
if (!cached_mode_set)
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, G_IO_ERROR, G_IO_ERROR_FAILED,
|
|
|
|
|
"Missing mode set for page flip fallback");
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
2021-05-05 12:35:48 +00:00
|
|
|
if (!meta_drm_buffer_ensure_fb_id (plane_assignment->buffer, error))
|
|
|
|
|
return FALSE;
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
fill_connector_ids_array (cached_mode_set->connectors,
|
|
|
|
|
&connectors,
|
|
|
|
|
&n_connectors);
|
|
|
|
|
|
2020-09-21 17:32:46 +00:00
|
|
|
fb_id = meta_drm_buffer_get_fb_id (plane_assignment->buffer);
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
x = meta_fixed_16_to_int (plane_assignment->src_rect.x);
|
|
|
|
|
y = meta_fixed_16_to_int (plane_assignment->src_rect.y);
|
|
|
|
|
|
|
|
|
|
fd = meta_kms_impl_device_get_fd (impl_device);
|
|
|
|
|
ret = drmModeSetCrtc (fd,
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
2020-09-21 17:32:46 +00:00
|
|
|
fb_id,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
x, y,
|
|
|
|
|
connectors, n_connectors,
|
|
|
|
|
cached_mode_set->drm_mode);
|
|
|
|
|
if (ret != 0)
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, G_IO_ERROR, g_io_error_from_errno (-ret),
|
|
|
|
|
"drmModeSetCrtc mode '%s' on CRTC %u failed: %s",
|
|
|
|
|
cached_mode_set->drm_mode->name,
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
g_strerror (-ret));
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!impl_device_simple->mode_set_fallback_feedback_source)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
|
|
|
|
GSource *source;
|
|
|
|
|
|
|
|
|
|
source = meta_kms_add_source_in_impl (kms,
|
|
|
|
|
mode_set_fallback_feedback_idle,
|
2020-07-17 07:38:11 +00:00
|
|
|
impl_device_simple,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
NULL);
|
2020-07-17 07:38:11 +00:00
|
|
|
impl_device_simple->mode_set_fallback_feedback_source = source;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
impl_device_simple->mode_set_fallback_page_flip_datas =
|
|
|
|
|
g_list_prepend (impl_device_simple->mode_set_fallback_page_flip_datas,
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
page_flip_data);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
2020-10-09 22:52:11 +00:00
|
|
|
static gboolean
|
|
|
|
|
symbolic_page_flip_idle (gpointer user_data)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsPageFlipData *page_flip_data = user_data;
|
|
|
|
|
MetaKmsImplDevice *impl_device;
|
|
|
|
|
MetaKmsCrtc *crtc;
|
|
|
|
|
|
|
|
|
|
impl_device = meta_kms_page_flip_data_get_impl_device (page_flip_data);
|
|
|
|
|
crtc = meta_kms_page_flip_data_get_crtc (page_flip_data);
|
|
|
|
|
|
|
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Handling symbolic page flip callback from %s, data: %p, CRTC: %u",
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device),
|
|
|
|
|
page_flip_data,
|
|
|
|
|
meta_kms_crtc_get_id (crtc));
|
|
|
|
|
|
|
|
|
|
meta_kms_impl_device_handle_page_flip_callback (impl_device, page_flip_data);
|
|
|
|
|
|
|
|
|
|
return G_SOURCE_REMOVE;
|
|
|
|
|
}
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
static gboolean
|
2020-10-02 14:06:35 +00:00
|
|
|
dispatch_page_flip (MetaKmsImplDevice *impl_device,
|
|
|
|
|
MetaKmsUpdate *update,
|
|
|
|
|
MetaKmsPageFlipData *page_flip_data,
|
|
|
|
|
GError **error)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
MetaKmsImplDeviceSimple *impl_device_simple =
|
|
|
|
|
META_KMS_IMPL_DEVICE_SIMPLE (impl_device);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
MetaKmsCrtc *crtc;
|
|
|
|
|
MetaKmsPlaneAssignment *plane_assignment;
|
2021-01-22 10:31:18 +00:00
|
|
|
g_autoptr (MetaKmsCustomPageFlip) custom_page_flip = NULL;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
int fd;
|
|
|
|
|
int ret;
|
|
|
|
|
|
2020-10-02 14:06:35 +00:00
|
|
|
crtc = meta_kms_page_flip_data_get_crtc (page_flip_data);
|
2019-10-04 09:48:46 +00:00
|
|
|
plane_assignment = meta_kms_update_get_primary_plane_assignment (update,
|
|
|
|
|
crtc);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
2021-01-22 10:31:18 +00:00
|
|
|
custom_page_flip = meta_kms_update_take_custom_page_flip_func (update);
|
2020-10-09 22:52:11 +00:00
|
|
|
|
2021-01-22 10:31:18 +00:00
|
|
|
if (!plane_assignment && !custom_page_flip)
|
2020-10-09 22:52:11 +00:00
|
|
|
{
|
|
|
|
|
MetaKmsDevice *device = meta_kms_impl_device_get_device (impl_device);
|
|
|
|
|
MetaKms *kms = meta_kms_device_get_kms (device);
|
|
|
|
|
GSource *source;
|
|
|
|
|
|
|
|
|
|
meta_kms_page_flip_data_make_symbolic (page_flip_data);
|
|
|
|
|
|
|
|
|
|
source = meta_kms_add_source_in_impl (kms,
|
|
|
|
|
symbolic_page_flip_idle,
|
|
|
|
|
page_flip_data,
|
|
|
|
|
NULL);
|
|
|
|
|
|
|
|
|
|
g_source_set_ready_time (source, 0);
|
|
|
|
|
g_source_unref (source);
|
|
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
2021-05-05 12:35:48 +00:00
|
|
|
if (plane_assignment && plane_assignment->buffer &&
|
|
|
|
|
!meta_drm_buffer_ensure_fb_id (plane_assignment->buffer, error))
|
|
|
|
|
return FALSE;
|
|
|
|
|
|
2020-10-09 22:52:11 +00:00
|
|
|
fd = meta_kms_impl_device_get_fd (impl_device);
|
2021-01-22 10:31:18 +00:00
|
|
|
if (custom_page_flip)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Invoking custom page flip on CRTC %u (%s)",
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device));
|
2021-01-22 10:31:18 +00:00
|
|
|
ret = custom_page_flip->func (custom_page_flip->user_data,
|
|
|
|
|
page_flip_data);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
2020-09-21 17:32:46 +00:00
|
|
|
uint32_t fb_id;
|
|
|
|
|
|
|
|
|
|
fb_id = meta_drm_buffer_get_fb_id (plane_assignment->buffer);
|
2020-10-02 16:37:18 +00:00
|
|
|
|
|
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Page flipping CRTC %u (%s) with %u, data: %p",
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device),
|
|
|
|
|
fb_id,
|
|
|
|
|
page_flip_data);
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
ret = drmModePageFlip (fd,
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
2020-09-21 17:32:46 +00:00
|
|
|
fb_id,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
DRM_MODE_PAGE_FLIP_EVENT,
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
page_flip_data);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (ret == -EBUSY)
|
|
|
|
|
{
|
2020-01-14 10:16:09 +00:00
|
|
|
CachedModeSet *cached_mode_set;
|
|
|
|
|
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Scheduling page flip retry on CRTC %u (%s)",
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device));
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
cached_mode_set = get_cached_mode_set (impl_device_simple, crtc);
|
2020-01-14 10:16:09 +00:00
|
|
|
if (cached_mode_set)
|
|
|
|
|
{
|
2020-09-21 17:32:46 +00:00
|
|
|
uint32_t fb_id;
|
2020-01-14 10:16:09 +00:00
|
|
|
drmModeModeInfo *drm_mode;
|
|
|
|
|
float refresh_rate;
|
|
|
|
|
|
2020-09-21 17:32:46 +00:00
|
|
|
if (plane_assignment)
|
|
|
|
|
fb_id = meta_drm_buffer_get_fb_id (plane_assignment->buffer);
|
|
|
|
|
else
|
|
|
|
|
fb_id = 0;
|
2020-01-14 10:16:09 +00:00
|
|
|
drm_mode = cached_mode_set->drm_mode;
|
|
|
|
|
refresh_rate = meta_calculate_drm_mode_refresh_rate (drm_mode);
|
2021-04-12 14:25:53 +00:00
|
|
|
meta_kms_impl_device_hold_fd (impl_device);
|
2020-07-17 07:38:11 +00:00
|
|
|
schedule_retry_page_flip (impl_device_simple,
|
2020-01-14 10:16:09 +00:00
|
|
|
crtc,
|
2020-09-21 17:32:46 +00:00
|
|
|
fb_id,
|
2020-01-14 10:16:09 +00:00
|
|
|
refresh_rate,
|
2020-07-17 20:38:55 +00:00
|
|
|
page_flip_data,
|
2021-01-22 10:31:18 +00:00
|
|
|
g_steal_pointer (&custom_page_flip));
|
2021-04-10 23:36:42 +00:00
|
|
|
return TRUE;
|
2020-01-14 10:16:09 +00:00
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, G_IO_ERROR, G_IO_ERROR_FAILED,
|
|
|
|
|
"Page flip of %u failed, and no mode set available",
|
|
|
|
|
meta_kms_crtc_get_id (crtc));
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
else if (ret == -EINVAL)
|
|
|
|
|
{
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Falling back to mode set on CRTC %u (%s)",
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device));
|
|
|
|
|
|
2021-04-10 23:36:42 +00:00
|
|
|
return mode_set_fallback (impl_device_simple,
|
|
|
|
|
update,
|
|
|
|
|
plane_assignment,
|
|
|
|
|
page_flip_data,
|
|
|
|
|
error);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
else if (ret != 0)
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, G_IO_ERROR, g_io_error_from_errno (-ret),
|
|
|
|
|
"drmModePageFlip on CRTC %u failed: %s",
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
g_strerror (-ret));
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
2021-04-10 23:36:42 +00:00
|
|
|
else
|
|
|
|
|
{
|
2021-04-12 14:25:53 +00:00
|
|
|
meta_kms_impl_device_hold_fd (impl_device);
|
|
|
|
|
|
2021-04-10 23:36:42 +00:00
|
|
|
impl_device_simple->posted_page_flip_datas =
|
|
|
|
|
g_list_prepend (impl_device_simple->posted_page_flip_datas,
|
|
|
|
|
page_flip_data);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
2021-04-10 23:36:42 +00:00
|
|
|
return TRUE;
|
|
|
|
|
}
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
|
2020-10-02 14:06:35 +00:00
|
|
|
static GList *
|
|
|
|
|
generate_page_flip_datas (MetaKmsImplDevice *impl_device,
|
|
|
|
|
MetaKmsUpdate *update)
|
|
|
|
|
{
|
|
|
|
|
GList *listeners;
|
|
|
|
|
GList *page_flip_datas = NULL;
|
|
|
|
|
|
|
|
|
|
listeners = g_list_copy (meta_kms_update_get_page_flip_listeners (update));
|
|
|
|
|
|
|
|
|
|
while (listeners)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsPageFlipListener *listener = listeners->data;
|
|
|
|
|
MetaKmsCrtc *crtc = listener->crtc;
|
|
|
|
|
MetaKmsPageFlipData *page_flip_data;
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
gpointer user_data;
|
|
|
|
|
GDestroyNotify destroy_notify;
|
2020-10-02 14:06:35 +00:00
|
|
|
GList *l;
|
|
|
|
|
|
|
|
|
|
page_flip_data = meta_kms_page_flip_data_new (impl_device, crtc);
|
|
|
|
|
page_flip_datas = g_list_append (page_flip_datas, page_flip_data);
|
|
|
|
|
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
user_data = g_steal_pointer (&listener->user_data);
|
|
|
|
|
destroy_notify = g_steal_pointer (&listener->destroy_notify);
|
2020-10-02 14:06:35 +00:00
|
|
|
meta_kms_page_flip_data_add_listener (page_flip_data,
|
|
|
|
|
listener->vtable,
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
listener->flags,
|
|
|
|
|
user_data,
|
|
|
|
|
destroy_notify);
|
2020-10-02 14:06:35 +00:00
|
|
|
|
|
|
|
|
listeners = g_list_delete_link (listeners, listeners);
|
|
|
|
|
|
|
|
|
|
l = listeners;
|
|
|
|
|
while (l)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsPageFlipListener *other_listener = l->data;
|
|
|
|
|
GList *l_next = l->next;
|
|
|
|
|
|
|
|
|
|
if (other_listener->crtc == crtc)
|
|
|
|
|
{
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
gpointer other_user_data;
|
|
|
|
|
GDestroyNotify other_destroy_notify;
|
|
|
|
|
|
|
|
|
|
other_user_data = g_steal_pointer (&other_listener->user_data);
|
|
|
|
|
other_destroy_notify =
|
|
|
|
|
g_steal_pointer (&other_listener->destroy_notify);
|
2020-10-02 14:06:35 +00:00
|
|
|
meta_kms_page_flip_data_add_listener (page_flip_data,
|
|
|
|
|
other_listener->vtable,
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
other_listener->flags,
|
|
|
|
|
other_user_data,
|
|
|
|
|
other_destroy_notify);
|
2020-10-02 14:06:35 +00:00
|
|
|
listeners = g_list_delete_link (listeners, l);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
l = l_next;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return page_flip_datas;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static gboolean
|
|
|
|
|
maybe_dispatch_page_flips (MetaKmsImplDevice *impl_device,
|
|
|
|
|
MetaKmsUpdate *update,
|
|
|
|
|
GList **failed_planes,
|
kms: Gracefully handle page flipping direct scanouts failing
When drmModePageFlip() or drmModeAtomicCommit() unexpectedly failed (e.g.
ENOSPC, which has been seen in the wild), this failure was not handled
very gracefully. The page flip listener for the scanout was left in the
MetaKmsUpdate, meaning when the primary plane composition was later page
flipped, two page flip listeners were added, one for the primary plane,
and one for the scanout. This caused the 'page-flipped' event to be
handled twice, the second time being fatal.
Handle this by making 'no-discard' listener flag be somewhat reversed,
and say 'drop-on-error', and then drop all 'drop-on-error' listeners
when a MetaKmsUpdate failed to be processed.
Also for a "preserve" flagged update, don't ever trigger "discard"
callbacks just yet, as preserved updates are used again for the primary
plane composition, in order to not miss e.g. CRTC gamma updates, or
cursor plane updates, which were added separately.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/1809
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1910>
2021-07-01 17:45:03 +00:00
|
|
|
MetaKmsUpdateFlag flags,
|
2020-10-02 14:06:35 +00:00
|
|
|
GError **error)
|
|
|
|
|
{
|
|
|
|
|
g_autoptr (GList) page_flip_datas = NULL;
|
|
|
|
|
GList *l;
|
|
|
|
|
|
|
|
|
|
page_flip_datas = generate_page_flip_datas (impl_device, update);
|
|
|
|
|
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
while (page_flip_datas)
|
2020-10-02 14:06:35 +00:00
|
|
|
{
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
g_autoptr (GList) l = NULL;
|
|
|
|
|
MetaKmsPageFlipData *page_flip_data;
|
|
|
|
|
|
|
|
|
|
l = page_flip_datas;
|
|
|
|
|
page_flip_datas = g_list_remove_link (page_flip_datas, l);
|
|
|
|
|
page_flip_data = g_steal_pointer (&l->data);
|
2020-10-02 14:06:35 +00:00
|
|
|
|
|
|
|
|
if (!dispatch_page_flip (impl_device, update, page_flip_data, error))
|
|
|
|
|
{
|
|
|
|
|
if (!g_error_matches (*error,
|
|
|
|
|
G_IO_ERROR,
|
|
|
|
|
G_IO_ERROR_PERMISSION_DENIED))
|
|
|
|
|
{
|
|
|
|
|
MetaKmsCrtc *crtc =
|
|
|
|
|
meta_kms_page_flip_data_get_crtc (page_flip_data);
|
|
|
|
|
MetaKmsPlaneAssignment *plane_assignment;
|
|
|
|
|
MetaKmsPlaneFeedback *plane_feedback;
|
|
|
|
|
|
|
|
|
|
plane_assignment =
|
|
|
|
|
meta_kms_update_get_primary_plane_assignment (update, crtc);
|
|
|
|
|
|
|
|
|
|
plane_feedback =
|
|
|
|
|
meta_kms_plane_feedback_new_take_error (plane_assignment->plane,
|
|
|
|
|
plane_assignment->crtc,
|
|
|
|
|
g_error_copy (*error));
|
|
|
|
|
*failed_planes = g_list_prepend (*failed_planes, plane_feedback);
|
|
|
|
|
}
|
|
|
|
|
|
kms: Gracefully handle page flipping direct scanouts failing
When drmModePageFlip() or drmModeAtomicCommit() unexpectedly failed (e.g.
ENOSPC, which has been seen in the wild), this failure was not handled
very gracefully. The page flip listener for the scanout was left in the
MetaKmsUpdate, meaning when the primary plane composition was later page
flipped, two page flip listeners were added, one for the primary plane,
and one for the scanout. This caused the 'page-flipped' event to be
handled twice, the second time being fatal.
Handle this by making 'no-discard' listener flag be somewhat reversed,
and say 'drop-on-error', and then drop all 'drop-on-error' listeners
when a MetaKmsUpdate failed to be processed.
Also for a "preserve" flagged update, don't ever trigger "discard"
callbacks just yet, as preserved updates are used again for the primary
plane composition, in order to not miss e.g. CRTC gamma updates, or
cursor plane updates, which were added separately.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/1809
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1910>
2021-07-01 17:45:03 +00:00
|
|
|
if (!(flags & META_KMS_UPDATE_FLAG_PRESERVE_ON_ERROR))
|
|
|
|
|
meta_kms_page_flip_data_discard_in_impl (page_flip_data, *error);
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
|
|
|
|
|
goto err;
|
2020-10-02 14:06:35 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return TRUE;
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
|
|
|
|
|
err:
|
kms: Gracefully handle page flipping direct scanouts failing
When drmModePageFlip() or drmModeAtomicCommit() unexpectedly failed (e.g.
ENOSPC, which has been seen in the wild), this failure was not handled
very gracefully. The page flip listener for the scanout was left in the
MetaKmsUpdate, meaning when the primary plane composition was later page
flipped, two page flip listeners were added, one for the primary plane,
and one for the scanout. This caused the 'page-flipped' event to be
handled twice, the second time being fatal.
Handle this by making 'no-discard' listener flag be somewhat reversed,
and say 'drop-on-error', and then drop all 'drop-on-error' listeners
when a MetaKmsUpdate failed to be processed.
Also for a "preserve" flagged update, don't ever trigger "discard"
callbacks just yet, as preserved updates are used again for the primary
plane composition, in order to not miss e.g. CRTC gamma updates, or
cursor plane updates, which were added separately.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/1809
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1910>
2021-07-01 17:45:03 +00:00
|
|
|
if (!(flags & META_KMS_UPDATE_FLAG_PRESERVE_ON_ERROR))
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
{
|
kms: Gracefully handle page flipping direct scanouts failing
When drmModePageFlip() or drmModeAtomicCommit() unexpectedly failed (e.g.
ENOSPC, which has been seen in the wild), this failure was not handled
very gracefully. The page flip listener for the scanout was left in the
MetaKmsUpdate, meaning when the primary plane composition was later page
flipped, two page flip listeners were added, one for the primary plane,
and one for the scanout. This caused the 'page-flipped' event to be
handled twice, the second time being fatal.
Handle this by making 'no-discard' listener flag be somewhat reversed,
and say 'drop-on-error', and then drop all 'drop-on-error' listeners
when a MetaKmsUpdate failed to be processed.
Also for a "preserve" flagged update, don't ever trigger "discard"
callbacks just yet, as preserved updates are used again for the primary
plane composition, in order to not miss e.g. CRTC gamma updates, or
cursor plane updates, which were added separately.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/1809
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1910>
2021-07-01 17:45:03 +00:00
|
|
|
for (l = page_flip_datas; l; l = l->next)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsPageFlipData *page_flip_data = l->data;
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
|
kms: Gracefully handle page flipping direct scanouts failing
When drmModePageFlip() or drmModeAtomicCommit() unexpectedly failed (e.g.
ENOSPC, which has been seen in the wild), this failure was not handled
very gracefully. The page flip listener for the scanout was left in the
MetaKmsUpdate, meaning when the primary plane composition was later page
flipped, two page flip listeners were added, one for the primary plane,
and one for the scanout. This caused the 'page-flipped' event to be
handled twice, the second time being fatal.
Handle this by making 'no-discard' listener flag be somewhat reversed,
and say 'drop-on-error', and then drop all 'drop-on-error' listeners
when a MetaKmsUpdate failed to be processed.
Also for a "preserve" flagged update, don't ever trigger "discard"
callbacks just yet, as preserved updates are used again for the primary
plane composition, in order to not miss e.g. CRTC gamma updates, or
cursor plane updates, which were added separately.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/1809
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1910>
2021-07-01 17:45:03 +00:00
|
|
|
meta_kms_page_flip_data_discard_in_impl (page_flip_data, *error);
|
|
|
|
|
}
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
}
|
|
|
|
|
g_list_free (page_flip_datas);
|
|
|
|
|
|
|
|
|
|
return FALSE;
|
2020-10-02 14:06:35 +00:00
|
|
|
}
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
static gboolean
|
2021-01-19 18:26:14 +00:00
|
|
|
process_entries (MetaKmsImplDevice *impl_device,
|
|
|
|
|
MetaKmsUpdate *update,
|
|
|
|
|
GList *entries,
|
|
|
|
|
MetaKmsSimpleProcessFunc func,
|
|
|
|
|
GError **error)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
|
|
|
|
GList *l;
|
|
|
|
|
|
2019-11-08 10:30:18 +00:00
|
|
|
for (l = entries; l; l = l->next)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!func (impl_device, update, l->data, error))
|
2019-11-08 10:30:18 +00:00
|
|
|
return FALSE;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
|
2019-11-08 10:30:18 +00:00
|
|
|
return TRUE;
|
|
|
|
|
}
|
2019-05-03 17:20:21 +00:00
|
|
|
|
2019-11-08 23:17:33 +00:00
|
|
|
static gboolean
|
2020-07-17 07:38:11 +00:00
|
|
|
process_cursor_plane_assignment (MetaKmsImplDevice *impl_device,
|
2019-11-08 23:17:33 +00:00
|
|
|
MetaKmsUpdate *update,
|
|
|
|
|
MetaKmsPlaneAssignment *plane_assignment,
|
|
|
|
|
GError **error)
|
|
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
uint32_t crtc_id;
|
2019-11-08 23:17:33 +00:00
|
|
|
int fd;
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
crtc_id = meta_kms_crtc_get_id (plane_assignment->crtc),
|
2019-11-08 23:17:33 +00:00
|
|
|
fd = meta_kms_impl_device_get_fd (impl_device);
|
|
|
|
|
|
|
|
|
|
if (!(plane_assignment->flags & META_KMS_ASSIGN_PLANE_FLAG_FB_UNCHANGED))
|
|
|
|
|
{
|
|
|
|
|
int width, height;
|
kms/plane-assignment: Add API to set cursor hotspot metadata
The transactional KMS API has been modelled after atomic KMS. Atomic KMS
currently doesn't support forwarding cursor hotspot metadata, thus it
was left out of the transactional KMS API having the user set the simply
create a plane assigment with the cursor sprite assigned to a cursor
plane using regular coordinates.
This, however, proved to be inadequate for virtual machines using
"seamless mouse mode" where they rely on the cursor position to
correspond to the actual cursor position of the virtual machine, not the
cursor plane. In effect, this caused cursor positions to look "shifted".
Fix this by adding back the hotspot metadata, right now as a optional
field to the plane assignment. In the legacy KMS implementation, this is
translated into drmModeSetCursor2() just as before, while still falling
back to drmModeSetCursor() with the plane coordinates, if either there
was no hotspot set, or if drmModeSetCursor2() failed.
Eventually, the atomic KMS API will learn about hotspots, but when
adding our own atomic KMS backend to the transacitonal KMS API, we must
until then still fall back to legacy KMS for virtual machines.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1136
2020-03-19 08:01:03 +00:00
|
|
|
int ret = -1;
|
2020-09-21 17:32:46 +00:00
|
|
|
uint32_t handle_u32;
|
2019-11-08 23:17:33 +00:00
|
|
|
|
2020-09-21 16:00:36 +00:00
|
|
|
width = plane_assignment->dst_rect.width;
|
|
|
|
|
height = plane_assignment->dst_rect.height;
|
2019-11-08 23:17:33 +00:00
|
|
|
|
2020-09-21 17:32:46 +00:00
|
|
|
if (plane_assignment->buffer)
|
|
|
|
|
{
|
|
|
|
|
MetaDrmBufferGbm *buffer_gbm =
|
|
|
|
|
META_DRM_BUFFER_GBM (plane_assignment->buffer);
|
|
|
|
|
struct gbm_bo *bo;
|
|
|
|
|
union gbm_bo_handle handle;
|
|
|
|
|
|
|
|
|
|
bo = meta_drm_buffer_gbm_get_bo (buffer_gbm);
|
|
|
|
|
handle = gbm_bo_get_handle (bo);
|
|
|
|
|
handle_u32 = handle.u32;
|
|
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
|
|
|
|
handle_u32 = 0;
|
|
|
|
|
}
|
|
|
|
|
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Setting HW cursor of CRTC %u (%s) to %u "
|
|
|
|
|
"(size: %dx%d, hot: (%d, %d))",
|
|
|
|
|
crtc_id,
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device),
|
|
|
|
|
handle_u32,
|
|
|
|
|
width, height,
|
|
|
|
|
plane_assignment->cursor_hotspot.x,
|
|
|
|
|
plane_assignment->cursor_hotspot.y);
|
|
|
|
|
|
kms/plane-assignment: Add API to set cursor hotspot metadata
The transactional KMS API has been modelled after atomic KMS. Atomic KMS
currently doesn't support forwarding cursor hotspot metadata, thus it
was left out of the transactional KMS API having the user set the simply
create a plane assigment with the cursor sprite assigned to a cursor
plane using regular coordinates.
This, however, proved to be inadequate for virtual machines using
"seamless mouse mode" where they rely on the cursor position to
correspond to the actual cursor position of the virtual machine, not the
cursor plane. In effect, this caused cursor positions to look "shifted".
Fix this by adding back the hotspot metadata, right now as a optional
field to the plane assignment. In the legacy KMS implementation, this is
translated into drmModeSetCursor2() just as before, while still falling
back to drmModeSetCursor() with the plane coordinates, if either there
was no hotspot set, or if drmModeSetCursor2() failed.
Eventually, the atomic KMS API will learn about hotspots, but when
adding our own atomic KMS backend to the transacitonal KMS API, we must
until then still fall back to legacy KMS for virtual machines.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1136
2020-03-19 08:01:03 +00:00
|
|
|
if (plane_assignment->cursor_hotspot.is_valid)
|
|
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
ret = drmModeSetCursor2 (fd,
|
|
|
|
|
crtc_id,
|
2020-09-21 17:32:46 +00:00
|
|
|
handle_u32,
|
kms/plane-assignment: Add API to set cursor hotspot metadata
The transactional KMS API has been modelled after atomic KMS. Atomic KMS
currently doesn't support forwarding cursor hotspot metadata, thus it
was left out of the transactional KMS API having the user set the simply
create a plane assigment with the cursor sprite assigned to a cursor
plane using regular coordinates.
This, however, proved to be inadequate for virtual machines using
"seamless mouse mode" where they rely on the cursor position to
correspond to the actual cursor position of the virtual machine, not the
cursor plane. In effect, this caused cursor positions to look "shifted".
Fix this by adding back the hotspot metadata, right now as a optional
field to the plane assignment. In the legacy KMS implementation, this is
translated into drmModeSetCursor2() just as before, while still falling
back to drmModeSetCursor() with the plane coordinates, if either there
was no hotspot set, or if drmModeSetCursor2() failed.
Eventually, the atomic KMS API will learn about hotspots, but when
adding our own atomic KMS backend to the transacitonal KMS API, we must
until then still fall back to legacy KMS for virtual machines.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1136
2020-03-19 08:01:03 +00:00
|
|
|
width, height,
|
|
|
|
|
plane_assignment->cursor_hotspot.x,
|
|
|
|
|
plane_assignment->cursor_hotspot.y);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (ret != 0)
|
|
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
ret = drmModeSetCursor (fd, crtc_id,
|
2020-09-21 17:32:46 +00:00
|
|
|
handle_u32,
|
kms/plane-assignment: Add API to set cursor hotspot metadata
The transactional KMS API has been modelled after atomic KMS. Atomic KMS
currently doesn't support forwarding cursor hotspot metadata, thus it
was left out of the transactional KMS API having the user set the simply
create a plane assigment with the cursor sprite assigned to a cursor
plane using regular coordinates.
This, however, proved to be inadequate for virtual machines using
"seamless mouse mode" where they rely on the cursor position to
correspond to the actual cursor position of the virtual machine, not the
cursor plane. In effect, this caused cursor positions to look "shifted".
Fix this by adding back the hotspot metadata, right now as a optional
field to the plane assignment. In the legacy KMS implementation, this is
translated into drmModeSetCursor2() just as before, while still falling
back to drmModeSetCursor() with the plane coordinates, if either there
was no hotspot set, or if drmModeSetCursor2() failed.
Eventually, the atomic KMS API will learn about hotspots, but when
adding our own atomic KMS backend to the transacitonal KMS API, we must
until then still fall back to legacy KMS for virtual machines.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1136
2020-03-19 08:01:03 +00:00
|
|
|
width, height);
|
|
|
|
|
}
|
|
|
|
|
|
2019-11-08 23:17:33 +00:00
|
|
|
if (ret != 0)
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, G_IO_ERROR, g_io_error_from_errno (-ret),
|
|
|
|
|
"drmModeSetCursor failed: %s", g_strerror (-ret));
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Moving HW cursor of CRTC %u (%s) to (%d, %d)",
|
|
|
|
|
crtc_id,
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device),
|
|
|
|
|
plane_assignment->dst_rect.x,
|
|
|
|
|
plane_assignment->dst_rect.y);
|
|
|
|
|
|
2019-11-08 23:17:33 +00:00
|
|
|
drmModeMoveCursor (fd,
|
2020-07-17 07:38:11 +00:00
|
|
|
crtc_id,
|
2020-09-21 16:00:36 +00:00
|
|
|
plane_assignment->dst_rect.x,
|
|
|
|
|
plane_assignment->dst_rect.y);
|
2019-11-08 23:17:33 +00:00
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static gboolean
|
2020-07-17 07:38:11 +00:00
|
|
|
process_plane_assignment (MetaKmsImplDevice *impl_device,
|
2019-11-08 23:17:33 +00:00
|
|
|
MetaKmsUpdate *update,
|
|
|
|
|
MetaKmsPlaneAssignment *plane_assignment,
|
|
|
|
|
MetaKmsPlaneFeedback **plane_feedback)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsPlane *plane;
|
|
|
|
|
MetaKmsPlaneType plane_type;
|
|
|
|
|
GError *error = NULL;
|
|
|
|
|
|
|
|
|
|
plane = plane_assignment->plane;
|
|
|
|
|
plane_type = meta_kms_plane_get_plane_type (plane);
|
|
|
|
|
switch (plane_type)
|
|
|
|
|
{
|
|
|
|
|
case META_KMS_PLANE_TYPE_PRIMARY:
|
|
|
|
|
/* Handled as part of the mode-set and page flip. */
|
|
|
|
|
return TRUE;
|
|
|
|
|
case META_KMS_PLANE_TYPE_CURSOR:
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!process_cursor_plane_assignment (impl_device, update,
|
2019-11-08 23:17:33 +00:00
|
|
|
plane_assignment,
|
|
|
|
|
&error))
|
|
|
|
|
{
|
|
|
|
|
*plane_feedback =
|
|
|
|
|
meta_kms_plane_feedback_new_take_error (plane,
|
|
|
|
|
plane_assignment->crtc,
|
|
|
|
|
g_steal_pointer (&error));
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
case META_KMS_PLANE_TYPE_OVERLAY:
|
|
|
|
|
error = g_error_new_literal (G_IO_ERROR, G_IO_ERROR_FAILED,
|
|
|
|
|
"Overlay planes cannot be assigned");
|
|
|
|
|
*plane_feedback =
|
|
|
|
|
meta_kms_plane_feedback_new_take_error (plane,
|
|
|
|
|
plane_assignment->crtc,
|
|
|
|
|
g_steal_pointer (&error));
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
g_assert_not_reached ();
|
|
|
|
|
}
|
|
|
|
|
|
2020-10-02 13:45:28 +00:00
|
|
|
static gboolean
|
|
|
|
|
process_plane_assignments (MetaKmsImplDevice *impl_device,
|
|
|
|
|
MetaKmsUpdate *update,
|
|
|
|
|
GList **failed_planes,
|
|
|
|
|
GError **error)
|
2019-11-08 23:17:33 +00:00
|
|
|
{
|
|
|
|
|
GList *l;
|
|
|
|
|
|
|
|
|
|
for (l = meta_kms_update_get_plane_assignments (update); l; l = l->next)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsPlaneAssignment *plane_assignment = l->data;
|
|
|
|
|
MetaKmsPlaneFeedback *plane_feedback;
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!process_plane_assignment (impl_device, update, plane_assignment,
|
2019-11-08 23:17:33 +00:00
|
|
|
&plane_feedback))
|
|
|
|
|
{
|
2020-10-02 13:45:28 +00:00
|
|
|
if (g_error_matches (plane_feedback->error,
|
|
|
|
|
G_IO_ERROR,
|
|
|
|
|
G_IO_ERROR_PERMISSION_DENIED))
|
|
|
|
|
{
|
|
|
|
|
g_propagate_error (error,
|
|
|
|
|
g_steal_pointer (&plane_feedback->error));
|
|
|
|
|
meta_kms_plane_feedback_free (plane_feedback);
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
2019-11-08 23:17:33 +00:00
|
|
|
|
2020-10-02 13:45:28 +00:00
|
|
|
*failed_planes = g_list_prepend (*failed_planes, plane_feedback);
|
|
|
|
|
if (plane_assignment->flags & META_KMS_ASSIGN_PLANE_FLAG_ALLOW_FAIL)
|
|
|
|
|
{
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
|
|
|
|
g_propagate_error (error, g_error_copy (plane_feedback->error));
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
}
|
2019-11-08 23:17:33 +00:00
|
|
|
}
|
|
|
|
|
|
2020-10-02 13:45:28 +00:00
|
|
|
return TRUE;
|
2019-11-08 23:17:33 +00:00
|
|
|
}
|
|
|
|
|
|
2020-10-02 14:56:10 +00:00
|
|
|
static void
|
|
|
|
|
page_flip_handler (int fd,
|
|
|
|
|
unsigned int sequence,
|
|
|
|
|
unsigned int tv_sec,
|
|
|
|
|
unsigned int tv_usec,
|
|
|
|
|
void *user_data)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsPageFlipData *page_flip_data = user_data;
|
|
|
|
|
MetaKmsImplDevice *impl_device;
|
2021-04-10 23:36:42 +00:00
|
|
|
MetaKmsImplDeviceSimple *impl_device_simple;
|
2020-10-02 16:37:18 +00:00
|
|
|
MetaKmsCrtc *crtc;
|
2020-10-02 14:56:10 +00:00
|
|
|
|
|
|
|
|
meta_kms_page_flip_data_set_timings_in_impl (page_flip_data,
|
|
|
|
|
sequence, tv_sec, tv_usec);
|
|
|
|
|
|
|
|
|
|
impl_device = meta_kms_page_flip_data_get_impl_device (page_flip_data);
|
2021-04-10 23:36:42 +00:00
|
|
|
impl_device_simple = META_KMS_IMPL_DEVICE_SIMPLE (impl_device);
|
2020-10-02 16:37:18 +00:00
|
|
|
crtc = meta_kms_page_flip_data_get_crtc (page_flip_data);
|
|
|
|
|
|
|
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Handling page flip callback from %s, data: %p, CRTC: %u",
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device),
|
|
|
|
|
page_flip_data,
|
|
|
|
|
meta_kms_crtc_get_id (crtc));
|
|
|
|
|
|
2021-04-12 14:25:53 +00:00
|
|
|
meta_kms_impl_device_unhold_fd (impl_device);
|
|
|
|
|
|
2020-10-02 14:56:10 +00:00
|
|
|
meta_kms_impl_device_handle_page_flip_callback (impl_device, page_flip_data);
|
2021-04-10 23:36:42 +00:00
|
|
|
impl_device_simple->posted_page_flip_datas =
|
|
|
|
|
g_list_remove (impl_device_simple->posted_page_flip_datas,
|
|
|
|
|
page_flip_data);
|
2020-10-02 14:56:10 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2021-04-01 09:56:22 +00:00
|
|
|
meta_kms_impl_device_simple_setup_drm_event_context (MetaKmsImplDevice *impl_device,
|
2020-10-02 14:56:10 +00:00
|
|
|
drmEventContext *drm_event_context)
|
|
|
|
|
{
|
|
|
|
|
drm_event_context->version = 2;
|
|
|
|
|
drm_event_context->page_flip_handler = page_flip_handler;
|
|
|
|
|
}
|
|
|
|
|
|
2021-12-08 20:20:52 +00:00
|
|
|
static MetaKmsFeedback *
|
|
|
|
|
perform_update_test (MetaKmsImplDevice *impl_device,
|
|
|
|
|
MetaKmsUpdate *update)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsImplDeviceSimple *impl_device_simple =
|
|
|
|
|
META_KMS_IMPL_DEVICE_SIMPLE (impl_device);
|
|
|
|
|
GList *failed_planes = NULL;
|
|
|
|
|
GList *l;
|
|
|
|
|
|
|
|
|
|
for (l = meta_kms_update_get_plane_assignments (update); l; l = l->next)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsPlaneAssignment *plane_assignment = l->data;
|
|
|
|
|
MetaKmsPlane *plane = plane_assignment->plane;
|
|
|
|
|
MetaKmsCrtc *crtc = plane_assignment->crtc;
|
|
|
|
|
MetaDrmBuffer *buffer = plane_assignment->buffer;
|
|
|
|
|
CachedModeSet *cached_mode_set;
|
|
|
|
|
|
|
|
|
|
if (!plane_assignment->crtc ||
|
|
|
|
|
!plane_assignment->buffer)
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
cached_mode_set = get_cached_mode_set (impl_device_simple,
|
|
|
|
|
plane_assignment->crtc);
|
|
|
|
|
if (!cached_mode_set)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsPlaneFeedback *plane_feedback;
|
|
|
|
|
|
|
|
|
|
plane_feedback =
|
|
|
|
|
meta_kms_plane_feedback_new_failed (plane, crtc,
|
|
|
|
|
"No existing mode set");
|
|
|
|
|
failed_planes = g_list_append (failed_planes, plane_feedback);
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (meta_drm_buffer_get_width (buffer) != cached_mode_set->width ||
|
|
|
|
|
meta_drm_buffer_get_height (buffer) != cached_mode_set->height ||
|
|
|
|
|
meta_drm_buffer_get_stride (buffer) != cached_mode_set->stride ||
|
|
|
|
|
meta_drm_buffer_get_format (buffer) != cached_mode_set->format ||
|
|
|
|
|
meta_drm_buffer_get_modifier (buffer) != cached_mode_set->modifier)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsPlaneFeedback *plane_feedback;
|
|
|
|
|
|
|
|
|
|
plane_feedback =
|
|
|
|
|
meta_kms_plane_feedback_new_failed (plane, crtc,
|
|
|
|
|
"Incompatible buffer");
|
|
|
|
|
failed_planes = g_list_append (failed_planes, plane_feedback);
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (failed_planes)
|
|
|
|
|
{
|
|
|
|
|
GError *error;
|
|
|
|
|
|
|
|
|
|
error = g_error_new_literal (G_IO_ERROR,
|
|
|
|
|
G_IO_ERROR_FAILED,
|
|
|
|
|
"One or more buffers incompatible");
|
|
|
|
|
return meta_kms_feedback_new_failed (failed_planes, error);
|
|
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
|
|
|
|
return meta_kms_feedback_new_passed (NULL);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2019-11-08 23:14:36 +00:00
|
|
|
static MetaKmsFeedback *
|
2020-07-17 07:38:11 +00:00
|
|
|
meta_kms_impl_device_simple_process_update (MetaKmsImplDevice *impl_device,
|
kms: Gracefully handle page flipping direct scanouts failing
When drmModePageFlip() or drmModeAtomicCommit() unexpectedly failed (e.g.
ENOSPC, which has been seen in the wild), this failure was not handled
very gracefully. The page flip listener for the scanout was left in the
MetaKmsUpdate, meaning when the primary plane composition was later page
flipped, two page flip listeners were added, one for the primary plane,
and one for the scanout. This caused the 'page-flipped' event to be
handled twice, the second time being fatal.
Handle this by making 'no-discard' listener flag be somewhat reversed,
and say 'drop-on-error', and then drop all 'drop-on-error' listeners
when a MetaKmsUpdate failed to be processed.
Also for a "preserve" flagged update, don't ever trigger "discard"
callbacks just yet, as preserved updates are used again for the primary
plane composition, in order to not miss e.g. CRTC gamma updates, or
cursor plane updates, which were added separately.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/1809
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1910>
2021-07-01 17:45:03 +00:00
|
|
|
MetaKmsUpdate *update,
|
|
|
|
|
MetaKmsUpdateFlag flags)
|
2019-11-08 10:30:18 +00:00
|
|
|
{
|
2019-11-08 23:14:36 +00:00
|
|
|
GError *error = NULL;
|
2020-10-02 13:45:28 +00:00
|
|
|
GList *failed_planes = NULL;
|
2019-05-03 17:20:21 +00:00
|
|
|
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Processing update %" G_GUINT64_FORMAT,
|
|
|
|
|
meta_kms_update_get_sequence_number (update));
|
|
|
|
|
|
2021-12-08 20:20:52 +00:00
|
|
|
if (flags & META_KMS_UPDATE_FLAG_TEST_ONLY)
|
|
|
|
|
return perform_update_test (impl_device, update);
|
|
|
|
|
|
2020-10-21 15:38:44 +00:00
|
|
|
if (meta_kms_update_is_power_save (update))
|
|
|
|
|
{
|
|
|
|
|
if (!process_power_save (impl_device, &error))
|
|
|
|
|
goto err;
|
|
|
|
|
goto out;
|
|
|
|
|
}
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!process_entries (impl_device,
|
2019-11-08 10:30:18 +00:00
|
|
|
update,
|
2020-10-21 15:38:44 +00:00
|
|
|
meta_kms_update_get_mode_sets (update),
|
|
|
|
|
process_mode_set,
|
2019-11-08 23:14:36 +00:00
|
|
|
&error))
|
2020-10-02 13:45:28 +00:00
|
|
|
goto err;
|
2019-11-08 10:30:18 +00:00
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!process_entries (impl_device,
|
2019-11-08 10:30:18 +00:00
|
|
|
update,
|
2020-10-21 15:38:44 +00:00
|
|
|
meta_kms_update_get_connector_updates (update),
|
|
|
|
|
process_connector_update,
|
2019-11-08 23:14:36 +00:00
|
|
|
&error))
|
2020-10-02 13:45:28 +00:00
|
|
|
goto err;
|
2019-11-08 10:30:18 +00:00
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!process_entries (impl_device,
|
2019-11-08 10:30:18 +00:00
|
|
|
update,
|
|
|
|
|
meta_kms_update_get_crtc_gammas (update),
|
|
|
|
|
process_crtc_gamma,
|
2019-11-08 23:14:36 +00:00
|
|
|
&error))
|
2020-10-02 13:45:28 +00:00
|
|
|
goto err;
|
2019-11-08 23:17:33 +00:00
|
|
|
|
2020-10-02 13:45:28 +00:00
|
|
|
if (!process_plane_assignments (impl_device, update, &failed_planes, &error))
|
|
|
|
|
goto err;
|
2019-11-08 10:30:18 +00:00
|
|
|
|
kms: Gracefully handle page flipping direct scanouts failing
When drmModePageFlip() or drmModeAtomicCommit() unexpectedly failed (e.g.
ENOSPC, which has been seen in the wild), this failure was not handled
very gracefully. The page flip listener for the scanout was left in the
MetaKmsUpdate, meaning when the primary plane composition was later page
flipped, two page flip listeners were added, one for the primary plane,
and one for the scanout. This caused the 'page-flipped' event to be
handled twice, the second time being fatal.
Handle this by making 'no-discard' listener flag be somewhat reversed,
and say 'drop-on-error', and then drop all 'drop-on-error' listeners
when a MetaKmsUpdate failed to be processed.
Also for a "preserve" flagged update, don't ever trigger "discard"
callbacks just yet, as preserved updates are used again for the primary
plane composition, in order to not miss e.g. CRTC gamma updates, or
cursor plane updates, which were added separately.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/1809
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1910>
2021-07-01 17:45:03 +00:00
|
|
|
if (!maybe_dispatch_page_flips (impl_device, update, &failed_planes, flags,
|
|
|
|
|
&error))
|
2020-10-02 13:45:28 +00:00
|
|
|
goto err;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
2020-10-21 15:38:44 +00:00
|
|
|
out:
|
2020-10-02 13:45:28 +00:00
|
|
|
return meta_kms_feedback_new_passed (failed_planes);
|
2019-11-08 23:17:33 +00:00
|
|
|
|
2020-10-02 13:45:28 +00:00
|
|
|
err:
|
2019-11-08 23:17:33 +00:00
|
|
|
return meta_kms_feedback_new_failed (failed_planes, error);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2020-07-17 07:38:11 +00:00
|
|
|
flush_postponed_page_flip_datas (MetaKmsImplDeviceSimple *impl_device_simple)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2021-04-10 23:32:13 +00:00
|
|
|
dispatch_page_flip_datas (&impl_device_simple->postponed_page_flip_datas,
|
|
|
|
|
(GFunc) meta_kms_page_flip_data_flipped_in_impl,
|
|
|
|
|
NULL);
|
|
|
|
|
dispatch_page_flip_datas (&impl_device_simple->postponed_mode_set_fallback_datas,
|
|
|
|
|
(GFunc) meta_kms_page_flip_data_mode_set_fallback_in_impl,
|
|
|
|
|
NULL);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2020-07-17 07:38:11 +00:00
|
|
|
meta_kms_impl_device_simple_handle_page_flip_callback (MetaKmsImplDevice *impl_device,
|
|
|
|
|
MetaKmsPageFlipData *page_flip_data)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
MetaKmsImplDeviceSimple *impl_device_simple =
|
|
|
|
|
META_KMS_IMPL_DEVICE_SIMPLE (impl_device);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
if (impl_device_simple->pending_page_flip_retries)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
impl_device_simple->postponed_page_flip_datas =
|
|
|
|
|
g_list_append (impl_device_simple->postponed_page_flip_datas,
|
kms/page-flip: Pass ownership of listener user data along with closure
In order to reliably manage the reference count of the user data passed
to page flip listeners - being the stage view - make the ownership of
this data travel through the different objects that take responsibility
of the next step.
Initially this is the MetaKmsPageFlipListener that belongs to a
MetaKmsUpdate.
When a page flip is successfully queued, the ownership is transferred to
a MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. In the
simple impl device, the MetaKmsPageFlipData is passed to
drmModePageFlip(), then returned back via the DRM event. In the future
atomic impl device, the MetaKmsPageFlipData is stored in a table, then
retrieved when DRM event are handled.
When the DRM events are handled, the page flip listener's interface
callbacks are invoked, and after that, the user data is freed using the
passed GDestroyNotify function, in the main context, the same as where
the interface callbacks were called.
When a page flip fails, the ownership is also transferred to a
MetaKmsPageFlipClosure that is part of a MetaKmsPageFlipData. This page
flip data will be passed to the main context via a callback, where it
will discard the page flip, and free the user data using the provided
GDestroyNotify.
Note that this adds back a page flip listener type flag for telling the
KMS implementation whether to actively discard a page flip via the
interface, or just free the user data. Avoiding discarding via the
interface is needed for the direct scanout case, where we immediately
need to know the result in order to fall back to the composite pipeline
if the direct scanout failed. We do in fact also need active discard via
the interface paths, e.g. in the simple impl device when we're
asynchronously retrying a page flip, so replace the ad-hoc discard paths
in meta-renderer-native.c and replace them by not asking for no-discard
page flip error handling.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1488>
2020-12-16 07:41:14 +00:00
|
|
|
page_flip_data);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
|
|
|
|
meta_kms_page_flip_data_flipped_in_impl (page_flip_data);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2021-04-12 14:25:53 +00:00
|
|
|
static void
|
|
|
|
|
dispose_page_flip_data (MetaKmsPageFlipData *page_flip_data,
|
|
|
|
|
MetaKmsImplDevice *impl_device)
|
|
|
|
|
{
|
|
|
|
|
meta_kms_page_flip_data_discard_in_impl (page_flip_data, NULL);
|
|
|
|
|
meta_kms_impl_device_unhold_fd (impl_device);
|
|
|
|
|
}
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
static void
|
2020-07-17 07:38:11 +00:00
|
|
|
meta_kms_impl_device_simple_discard_pending_page_flips (MetaKmsImplDevice *impl_device)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
MetaKmsImplDeviceSimple *impl_device_simple =
|
|
|
|
|
META_KMS_IMPL_DEVICE_SIMPLE (impl_device);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
GList *l;
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
if (!impl_device_simple->pending_page_flip_retries)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
return;
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
for (l = impl_device_simple->pending_page_flip_retries; l; l = l->next)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
|
|
|
|
RetryPageFlipData *retry_page_flip_data = l->data;
|
|
|
|
|
MetaKmsPageFlipData *page_flip_data;
|
|
|
|
|
|
|
|
|
|
page_flip_data = g_steal_pointer (&retry_page_flip_data->page_flip_data);
|
2020-10-02 16:37:18 +00:00
|
|
|
|
|
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Discarding page flip retry for CRTC %u (%s)",
|
|
|
|
|
meta_kms_crtc_get_id (
|
|
|
|
|
meta_kms_page_flip_data_get_crtc (page_flip_data)),
|
|
|
|
|
meta_kms_impl_device_get_path (
|
|
|
|
|
meta_kms_page_flip_data_get_impl_device (page_flip_data)));
|
|
|
|
|
|
2021-04-12 14:25:53 +00:00
|
|
|
dispose_page_flip_data (page_flip_data, impl_device);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
retry_page_flip_data_free (retry_page_flip_data);
|
|
|
|
|
}
|
2020-07-17 07:38:11 +00:00
|
|
|
g_clear_pointer (&impl_device_simple->pending_page_flip_retries, g_list_free);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
g_clear_pointer (&impl_device_simple->retry_page_flips_source,
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
g_source_destroy);
|
|
|
|
|
}
|
|
|
|
|
|
2021-04-13 16:18:46 +00:00
|
|
|
static void
|
|
|
|
|
meta_kms_impl_device_simple_prepare_shutdown (MetaKmsImplDevice *impl_device)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsImplDeviceSimple *impl_device_simple =
|
|
|
|
|
META_KMS_IMPL_DEVICE_SIMPLE (impl_device);
|
|
|
|
|
|
|
|
|
|
g_list_foreach (impl_device_simple->posted_page_flip_datas,
|
2021-04-12 14:25:53 +00:00
|
|
|
(GFunc) dispose_page_flip_data,
|
|
|
|
|
impl_device);
|
2021-04-13 16:18:46 +00:00
|
|
|
g_clear_list (&impl_device_simple->posted_page_flip_datas, NULL);
|
|
|
|
|
}
|
|
|
|
|
|
2020-02-21 11:06:28 +00:00
|
|
|
static void
|
2020-07-17 07:38:11 +00:00
|
|
|
meta_kms_impl_device_simple_finalize (GObject *object)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsImplDeviceSimple *impl_device_simple =
|
|
|
|
|
META_KMS_IMPL_DEVICE_SIMPLE (object);
|
2021-04-12 14:25:53 +00:00
|
|
|
MetaKmsImplDevice *impl_device = META_KMS_IMPL_DEVICE (impl_device_simple);
|
2020-07-17 07:38:11 +00:00
|
|
|
|
|
|
|
|
g_list_free_full (impl_device_simple->pending_page_flip_retries,
|
|
|
|
|
(GDestroyNotify) retry_page_flip_data_free);
|
2021-04-15 14:32:53 +00:00
|
|
|
dispatch_page_flip_datas (&impl_device_simple->postponed_page_flip_datas,
|
2021-04-12 14:25:53 +00:00
|
|
|
(GFunc) dispose_page_flip_data,
|
|
|
|
|
impl_device);
|
2021-04-15 14:32:53 +00:00
|
|
|
dispatch_page_flip_datas (&impl_device_simple->postponed_mode_set_fallback_datas,
|
2021-04-12 14:25:53 +00:00
|
|
|
(GFunc) dispose_page_flip_data,
|
|
|
|
|
impl_device);
|
2020-07-17 07:38:11 +00:00
|
|
|
|
2021-04-13 16:18:46 +00:00
|
|
|
g_assert (!impl_device_simple->posted_page_flip_datas);
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
g_clear_pointer (&impl_device_simple->mode_set_fallback_feedback_source,
|
|
|
|
|
g_source_destroy);
|
|
|
|
|
g_hash_table_destroy (impl_device_simple->cached_mode_sets);
|
|
|
|
|
|
|
|
|
|
G_OBJECT_CLASS (meta_kms_impl_device_simple_parent_class)->finalize (object);
|
|
|
|
|
}
|
|
|
|
|
|
2021-04-06 10:14:24 +00:00
|
|
|
static MetaDeviceFile *
|
|
|
|
|
meta_kms_impl_device_simple_open_device_file (MetaKmsImplDevice *impl_device,
|
|
|
|
|
const char *path,
|
|
|
|
|
GError **error)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsDevice *device = meta_kms_impl_device_get_device (impl_device);
|
|
|
|
|
MetaKms *kms = meta_kms_device_get_kms (device);
|
|
|
|
|
MetaBackend *backend = meta_kms_get_backend (kms);
|
|
|
|
|
MetaDevicePool *device_pool =
|
|
|
|
|
meta_backend_native_get_device_pool (META_BACKEND_NATIVE (backend));
|
2021-04-11 10:15:23 +00:00
|
|
|
g_autoptr (MetaDeviceFile) device_file = NULL;
|
2021-04-06 10:14:24 +00:00
|
|
|
|
2021-04-11 10:15:23 +00:00
|
|
|
device_file = meta_device_pool_open (device_pool, path,
|
|
|
|
|
META_DEVICE_FILE_FLAG_TAKE_CONTROL,
|
|
|
|
|
error);
|
|
|
|
|
if (!device_file)
|
|
|
|
|
return NULL;
|
|
|
|
|
|
2021-04-12 14:25:53 +00:00
|
|
|
if (!meta_device_file_has_tag (device_file,
|
|
|
|
|
META_DEVICE_FILE_TAG_KMS,
|
|
|
|
|
META_KMS_DEVICE_FILE_TAG_SIMPLE))
|
2021-04-11 10:15:23 +00:00
|
|
|
{
|
2021-04-12 14:25:53 +00:00
|
|
|
int fd = meta_device_file_get_fd (device_file);
|
|
|
|
|
|
|
|
|
|
g_warn_if_fail (!meta_device_file_has_tag (device_file,
|
|
|
|
|
META_DEVICE_FILE_TAG_KMS,
|
|
|
|
|
META_KMS_DEVICE_FILE_TAG_ATOMIC));
|
|
|
|
|
|
|
|
|
|
if (drmSetClientCap (fd, DRM_CLIENT_CAP_UNIVERSAL_PLANES, 1) != 0)
|
|
|
|
|
{
|
|
|
|
|
g_set_error (error, META_KMS_ERROR, META_KMS_ERROR_NOT_SUPPORTED,
|
|
|
|
|
"DRM_CLIENT_CAP_UNIVERSAL_PLANES not supported");
|
|
|
|
|
return NULL;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
meta_device_file_tag (device_file,
|
|
|
|
|
META_DEVICE_FILE_TAG_KMS,
|
|
|
|
|
META_KMS_DEVICE_FILE_TAG_SIMPLE);
|
2021-04-11 10:15:23 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return g_steal_pointer (&device_file);
|
2021-04-06 10:14:24 +00:00
|
|
|
}
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
static gboolean
|
|
|
|
|
meta_kms_impl_device_simple_initable_init (GInitable *initable,
|
|
|
|
|
GCancellable *cancellable,
|
|
|
|
|
GError **error)
|
2020-02-21 11:06:28 +00:00
|
|
|
{
|
2020-07-17 07:38:11 +00:00
|
|
|
MetaKmsImplDeviceSimple *impl_device_simple =
|
|
|
|
|
META_KMS_IMPL_DEVICE_SIMPLE (initable);
|
|
|
|
|
MetaKmsImplDevice *impl_device = META_KMS_IMPL_DEVICE (impl_device_simple);
|
|
|
|
|
MetaKmsDevice *device = meta_kms_impl_device_get_device (impl_device);
|
2020-02-21 11:06:28 +00:00
|
|
|
GList *l;
|
|
|
|
|
|
2021-04-06 10:14:24 +00:00
|
|
|
if (!initable_parent_iface->init (initable, cancellable, error))
|
|
|
|
|
return FALSE;
|
|
|
|
|
|
2021-01-18 17:27:07 +00:00
|
|
|
if (!meta_kms_impl_device_init_mode_setting (impl_device, error))
|
2020-07-17 07:38:11 +00:00
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
|
|
impl_device_simple->cached_mode_sets =
|
|
|
|
|
g_hash_table_new_full (NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
(GDestroyNotify) cached_mode_set_free);
|
|
|
|
|
|
2020-02-21 11:06:28 +00:00
|
|
|
for (l = meta_kms_device_get_crtcs (device); l; l = l->next)
|
|
|
|
|
{
|
|
|
|
|
MetaKmsCrtc *crtc = l->data;
|
|
|
|
|
MetaKmsPlane *plane;
|
|
|
|
|
|
|
|
|
|
plane = meta_kms_device_get_cursor_plane_for (device, crtc);
|
|
|
|
|
if (plane)
|
|
|
|
|
continue;
|
|
|
|
|
|
2020-10-02 16:37:18 +00:00
|
|
|
meta_topic (META_DEBUG_KMS,
|
|
|
|
|
"[simple] Adding fake cursor plane for CRTC %u (%s)",
|
|
|
|
|
meta_kms_crtc_get_id (crtc),
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device));
|
|
|
|
|
|
2020-02-21 11:06:28 +00:00
|
|
|
meta_kms_device_add_fake_plane_in_impl (device,
|
|
|
|
|
META_KMS_PLANE_TYPE_CURSOR,
|
|
|
|
|
crtc);
|
|
|
|
|
}
|
2020-07-17 07:38:11 +00:00
|
|
|
|
2021-04-06 10:14:24 +00:00
|
|
|
g_message ("Added device '%s' (%s) using non-atomic mode setting.",
|
|
|
|
|
meta_kms_impl_device_get_path (impl_device),
|
|
|
|
|
meta_kms_impl_device_get_driver_name (impl_device));
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
return TRUE;
|
2020-02-21 11:06:28 +00:00
|
|
|
}
|
|
|
|
|
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
static void
|
2020-07-17 07:38:11 +00:00
|
|
|
meta_kms_impl_device_simple_init (MetaKmsImplDeviceSimple *impl_device_simple)
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
|
backends/native: Add basic KMS abstraction building blocks
The intention with KMS abstraction is to hide away accessing the drm
functions behind an API that allows us to have different kind of KMS
implementations, including legacy non-atomic and atomic. The intention
is also that the code interacting with the drm device should be able to
be run in a different thread than the main thread. This means that we
need to make sure that all drm*() API usage must only occur from within
tasks that eventually can be run in the dedicated thread.
The idea here is that MetaKms provides a outward facing API other places
of mutter can use (e.g. MetaGpuKms and friends), while MetaKmsImpl is
an internal implementation that only gets interacted with via "tasks"
posted via the MetaKms object. These tasks will in the future
potentially be run on the dedicated KMS thread. Initially, we don't
create any new threads.
Likewise, MetaKmsDevice is a outward facing representation of a KMS
device, while MetaKmsImplDevice is the corresponding implementation,
which only runs from within the MetaKmsImpl tasks.
This commit only moves opening and closing the device to this new API,
while leaking the fd outside of the impl enclosure, effectively making
the isolation for drm*() calls pointless. This, however, is necessary to
allow gradual porting of drm interaction, and eventually the file
descriptor in MetaGpuKms will be removed. For now, it's harmless, since
everything still run in the main thread.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-01-29 09:24:44 +00:00
|
|
|
static void
|
2020-07-17 07:38:11 +00:00
|
|
|
initable_iface_init (GInitableIface *iface)
|
backends/native: Add basic KMS abstraction building blocks
The intention with KMS abstraction is to hide away accessing the drm
functions behind an API that allows us to have different kind of KMS
implementations, including legacy non-atomic and atomic. The intention
is also that the code interacting with the drm device should be able to
be run in a different thread than the main thread. This means that we
need to make sure that all drm*() API usage must only occur from within
tasks that eventually can be run in the dedicated thread.
The idea here is that MetaKms provides a outward facing API other places
of mutter can use (e.g. MetaGpuKms and friends), while MetaKmsImpl is
an internal implementation that only gets interacted with via "tasks"
posted via the MetaKms object. These tasks will in the future
potentially be run on the dedicated KMS thread. Initially, we don't
create any new threads.
Likewise, MetaKmsDevice is a outward facing representation of a KMS
device, while MetaKmsImplDevice is the corresponding implementation,
which only runs from within the MetaKmsImpl tasks.
This commit only moves opening and closing the device to this new API,
while leaking the fd outside of the impl enclosure, effectively making
the isolation for drm*() calls pointless. This, however, is necessary to
allow gradual porting of drm interaction, and eventually the file
descriptor in MetaGpuKms will be removed. For now, it's harmless, since
everything still run in the main thread.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-01-29 09:24:44 +00:00
|
|
|
{
|
2021-04-06 10:14:24 +00:00
|
|
|
initable_parent_iface = g_type_interface_peek_parent (iface);
|
|
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
iface->init = meta_kms_impl_device_simple_initable_init;
|
backends/native: Add basic KMS abstraction building blocks
The intention with KMS abstraction is to hide away accessing the drm
functions behind an API that allows us to have different kind of KMS
implementations, including legacy non-atomic and atomic. The intention
is also that the code interacting with the drm device should be able to
be run in a different thread than the main thread. This means that we
need to make sure that all drm*() API usage must only occur from within
tasks that eventually can be run in the dedicated thread.
The idea here is that MetaKms provides a outward facing API other places
of mutter can use (e.g. MetaGpuKms and friends), while MetaKmsImpl is
an internal implementation that only gets interacted with via "tasks"
posted via the MetaKms object. These tasks will in the future
potentially be run on the dedicated KMS thread. Initially, we don't
create any new threads.
Likewise, MetaKmsDevice is a outward facing representation of a KMS
device, while MetaKmsImplDevice is the corresponding implementation,
which only runs from within the MetaKmsImpl tasks.
This commit only moves opening and closing the device to this new API,
while leaking the fd outside of the impl enclosure, effectively making
the isolation for drm*() calls pointless. This, however, is necessary to
allow gradual porting of drm interaction, and eventually the file
descriptor in MetaGpuKms will be removed. For now, it's harmless, since
everything still run in the main thread.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-01-29 09:24:44 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2020-07-17 07:38:11 +00:00
|
|
|
meta_kms_impl_device_simple_class_init (MetaKmsImplDeviceSimpleClass *klass)
|
backends/native: Add basic KMS abstraction building blocks
The intention with KMS abstraction is to hide away accessing the drm
functions behind an API that allows us to have different kind of KMS
implementations, including legacy non-atomic and atomic. The intention
is also that the code interacting with the drm device should be able to
be run in a different thread than the main thread. This means that we
need to make sure that all drm*() API usage must only occur from within
tasks that eventually can be run in the dedicated thread.
The idea here is that MetaKms provides a outward facing API other places
of mutter can use (e.g. MetaGpuKms and friends), while MetaKmsImpl is
an internal implementation that only gets interacted with via "tasks"
posted via the MetaKms object. These tasks will in the future
potentially be run on the dedicated KMS thread. Initially, we don't
create any new threads.
Likewise, MetaKmsDevice is a outward facing representation of a KMS
device, while MetaKmsImplDevice is the corresponding implementation,
which only runs from within the MetaKmsImpl tasks.
This commit only moves opening and closing the device to this new API,
while leaking the fd outside of the impl enclosure, effectively making
the isolation for drm*() calls pointless. This, however, is necessary to
allow gradual porting of drm interaction, and eventually the file
descriptor in MetaGpuKms will be removed. For now, it's harmless, since
everything still run in the main thread.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-01-29 09:24:44 +00:00
|
|
|
{
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
GObjectClass *object_class = G_OBJECT_CLASS (klass);
|
2020-07-17 07:38:11 +00:00
|
|
|
MetaKmsImplDeviceClass *impl_device_class =
|
|
|
|
|
META_KMS_IMPL_DEVICE_CLASS (klass);
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
2020-07-17 07:38:11 +00:00
|
|
|
object_class->finalize = meta_kms_impl_device_simple_finalize;
|
backend/native: Add and use transactional KMS API
This commit introduces, and makes use of, a transactional API used for
setting up KMS state, later to be applied, potentially atomically. From
an API point of view, so is always the case, but in the current
implementation, it still uses legacy drmMode* API to apply the state
non-atomically.
The API consists of various buliding blocks:
* MetaKmsUpdate - a set of configuration changes, the higher level
handle for handing over configuration to the impl backend. It's used to
set mode, assign framebuffers to planes, queue page flips and set
connector properties.
* MetaKmsPlaneAssignment - the assignment of a framebuffer to a plane.
Currently used to map a framebuffer to the primary plane of a CRTC. In
the legacy KMS implementation, the plane assignment is used to derive
the framebuffer used for mode setting and page flipping.
This also means various high level changes:
State, excluding configuring the cursor plane and creating/destroying
DRM framebuffer handles, are applied in the end of a clutter frame, in
one go. From an API point of view, this is done atomically, but as
mentioned, only the non-atomic implementation exists so far.
From MetaRendererNative's point of view, a page flip now initially
always succeeds; the handling of EBUSY errors are done asynchronously in
the MetaKmsImpl backend (still by retrying at refresh rate, but
postponing flip callbacks instead of manipulating the frame clock).
Handling of falling back to mode setting instead of page flipping is
notified after the fact by a more precise page flip feedback API.
EGLStream based page flipping relies on the impl backend not being
atomic, as the page flipping is done in the EGLStream backend (e.g.
nvidia driver). It uses a 'custom' page flip queueing method, keeping
the EGLStream logic inside meta-renderer-native.c.
Page flip handling is moved to meta-kms-impl-device.c from
meta-gpu-kms.c. It goes via an extra idle callback before reaching
meta-renderer-native.c to make sure callbacks are invoked outside of the
impl context.
While dummy power save page flipping is kept in meta-renderer-native.c, the
EBUSY handling is moved to meta-kms-impl-simple.c. Instead of freezing the
frame clock, actual page flip callbacks are postponed until all EBUSY retries
have either succeeded or failed due to some other error than EBUSY. This
effectively inhibits new frames to be drawn, meaning we won't stall waiting on
the file descriptor for pending page flips.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-04-04 20:36:41 +00:00
|
|
|
|
2021-04-06 10:14:24 +00:00
|
|
|
impl_device_class->open_device_file =
|
|
|
|
|
meta_kms_impl_device_simple_open_device_file;
|
2020-10-02 14:56:10 +00:00
|
|
|
impl_device_class->setup_drm_event_context =
|
|
|
|
|
meta_kms_impl_device_simple_setup_drm_event_context;
|
2020-07-17 07:38:11 +00:00
|
|
|
impl_device_class->process_update =
|
|
|
|
|
meta_kms_impl_device_simple_process_update;
|
|
|
|
|
impl_device_class->handle_page_flip_callback =
|
|
|
|
|
meta_kms_impl_device_simple_handle_page_flip_callback;
|
|
|
|
|
impl_device_class->discard_pending_page_flips =
|
|
|
|
|
meta_kms_impl_device_simple_discard_pending_page_flips;
|
2021-04-13 16:18:46 +00:00
|
|
|
impl_device_class->prepare_shutdown =
|
|
|
|
|
meta_kms_impl_device_simple_prepare_shutdown;
|
backends/native: Add basic KMS abstraction building blocks
The intention with KMS abstraction is to hide away accessing the drm
functions behind an API that allows us to have different kind of KMS
implementations, including legacy non-atomic and atomic. The intention
is also that the code interacting with the drm device should be able to
be run in a different thread than the main thread. This means that we
need to make sure that all drm*() API usage must only occur from within
tasks that eventually can be run in the dedicated thread.
The idea here is that MetaKms provides a outward facing API other places
of mutter can use (e.g. MetaGpuKms and friends), while MetaKmsImpl is
an internal implementation that only gets interacted with via "tasks"
posted via the MetaKms object. These tasks will in the future
potentially be run on the dedicated KMS thread. Initially, we don't
create any new threads.
Likewise, MetaKmsDevice is a outward facing representation of a KMS
device, while MetaKmsImplDevice is the corresponding implementation,
which only runs from within the MetaKmsImpl tasks.
This commit only moves opening and closing the device to this new API,
while leaking the fd outside of the impl enclosure, effectively making
the isolation for drm*() calls pointless. This, however, is necessary to
allow gradual porting of drm interaction, and eventually the file
descriptor in MetaGpuKms will be removed. For now, it's harmless, since
everything still run in the main thread.
https://gitlab.gnome.org/GNOME/mutter/issues/548
https://gitlab.gnome.org/GNOME/mutter/merge_requests/525
2019-01-29 09:24:44 +00:00
|
|
|
}
|
