drm/amd/display: put commit when -ERESTARTSYS received
Subsystems (2)
PATCH 1/1 drm/amd/display: put commit when -ERESTARTSYS received
During page flip atomic_check and atomic_commit can return -ERESTARTSYS to restart the ioctl. When this happens we fail to put the commit object leading to a memory leak.
--- drivers/gpu/drm/drm_atomic.c | 25 +++++++++++++++++++++++-- 1 file changed, 23 insertions(+), 2 deletions(-)
drivers/gpu/drm/drm_atomic.c +23 -2
@@ -1642,14 +1642,35 @@ int drm_atomic_nonblocking_commit(struct drm_atomic_state *state) {struct drm_mode_config *config = &state->dev->mode_config;int ret; +int i; +struct drm_crtc *crtc; +struct drm_crtc_state *crtc_state; ret =drm_atomic_check_only (state); -if (ret) +if (ret) { +if (ret == -ERESTARTSYS ) + goto fail; +return ret; + }DRM_DEBUG_ATOMIC ("commiting %p nonblocking\n" , state); -return config->funcs->atomic_commit (state->dev, state, true); + ret = config->funcs->atomic_commit (state->dev, state, true); +if (ret == -ERESTARTSYS ) + goto fail; + +return ret; + +/* cleanup commit object if commit fails with ERESTARTSYS */ +fail: +for_each_crtc_in_state (state, crtc, crtc_state, i) { +if (state->crtcs[i].commit) { +drm_crtc_commit_put (state->crtcs[i].commit); + } + } + +return ret; }EXPORT_SYMBOL (drm_atomic_nonblocking_commit);
Signature · 6 lines
2.11.0 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Quoted 5 lines · Jerry Zuo
During page flip
atomic_checkandatomic_commitcan return-ERESTARTSYSto restart the ioctl. When this happens we fail to put the commit object leading to a memory leak.Signed-off-by Jerry Zuo <[email protected]>
The subject should be:
drm/atomic: put commit when -ERESTARTSYS received
Since the change is in the core atomic code.
Alex
Quoted 49 lines
--- drivers/gpu/drm/drm_atomic.c | 25 +++++++++++++++++++++++-- 1 file changed, 23 insertions(+), 2 deletions(-)drivers/gpu/drm/drm_atomic.c +23
-3@@ -1642,14 +1642,35 @@ int drm_atomic_nonblocking_commit(struct drm_atomic_state *state) {struct drm_mode_config *config = &state->dev->mode_config;int ret; +int i; +struct drm_crtc *crtc; +struct drm_crtc_state *crtc_state; ret =drm_atomic_check_only (state); -if (ret) +if (ret) { +if (ret == -ERESTARTSYS ) + goto fail; +return ret; + }DRM_DEBUG_ATOMIC ("commiting %p nonblocking\n" , state); -return config->funcs->atomic_commit (state->dev, state, true); + ret = config->funcs->atomic_commit (state->dev, state, true); +if (ret == -ERESTARTSYS ) + goto fail; + +return ret; + +/* cleanup commit object if commit fails with ERESTARTSYS */ +fail: +for_each_crtc_in_state (state, crtc, crtc_state, i) { +if (state->crtcs[i].commit) { +drm_crtc_commit_put (state->crtcs[i].commit); + } + } + +return ret; }EXPORT_SYMBOL (drm_atomic_nonblocking_commit); --2.11.0
Quoted 4 lines
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Quoted 10 lines · Alex Deucher
Quoted 5 lines · Jerry Zuo
During page flip
atomic_checkandatomic_commitcan return-ERESTARTSYSto restart the ioctl. When this happens we fail to put the commit object leading to a memory leak.Signed-off-by Jerry Zuo <[email protected]>The subject should be:
drm/atomic: put commit when-ERESTARTSYSreceived
Since the change is in the core atomic code.
Do we have an igt testcase to exercise this? This is the kind of error case handling igt really is made for, and igt has ready-made helpers to interrupt ioctls. I think Maarten was even working on such a testcase, adding him.
-Daniel
Quoted 56 lines
Alex
Quoted 53 lines
--- drivers/gpu/drm/drm_atomic.c | 25 +++++++++++++++++++++++-- 1 file changed, 23 insertions(+), 2 deletions(-)drivers/gpu/drm/drm_atomic.c +23
-3@@ -1642,14 +1642,35 @@ int drm_atomic_nonblocking_commit(struct drm_atomic_state *state) {struct drm_mode_config *config = &state->dev->mode_config;int ret; +int i; +struct drm_crtc *crtc; +struct drm_crtc_state *crtc_state; ret =drm_atomic_check_only (state); -if (ret) +if (ret) { +if (ret == -ERESTARTSYS ) + goto fail; +return ret; + }DRM_DEBUG_ATOMIC ("commiting %p nonblocking\n" , state); -return config->funcs->atomic_commit (state->dev, state, true); + ret = config->funcs->atomic_commit (state->dev, state, true); +if (ret == -ERESTARTSYS ) + goto fail; + +return ret; + +/* cleanup commit object if commit fails with ERESTARTSYS */ +fail: +for_each_crtc_in_state (state, crtc, crtc_state, i) { +if (state->crtcs[i].commit) { +drm_crtc_commit_put (state->crtcs[i].commit); + } + } + +return ret; }EXPORT_SYMBOL (drm_atomic_nonblocking_commit); --2.11.0
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
Quoted 8 lines · Daniel Vetter
Quoted 7 lines · Alex Deucher
Quoted 5 lines · Jerry Zuo
During page flip
atomic_checkandatomic_commitcan return-ERESTARTSYSto restart the ioctl. When this happens we fail to put the commit object leading to a memory leak.Signed-off-by Jerry Zuo <[email protected]>
Quoted 8 lines
Quoted 3 lines
The subject should be:
drm/atomic: put commit when-ERESTARTSYSreceived
Since the change is in the core atomic code.Do we have an igt testcase to exercise this? This is the kind of error case handling igt really is made for, and igt has ready-made helpers to interrupt ioctls. I think Maarten was even working on such a testcase, adding him.
I'm not aware of an IGT test for this. We triggered this scenario when running mode changes while glxgears is running. We observed no user-visible issue, only a memory leak. Is IGT able to capture those (semi-)reliably?
I agree that it would be great to have an IGT test for these corner cases.
Harry
Quoted 50 lines
-Daniel
Quoted 49 lines
Quoted 1 line
Alex
Quoted 46 lines
--- drivers/gpu/drm/drm_atomic.c | 25 +++++++++++++++++++++++-- 1 file changed, 23 insertions(+), 2 deletions(-)drivers/gpu/drm/drm_atomic.c +23
-2@@ -1642,14 +1642,35 @@ int drm_atomic_nonblocking_commit(struct drm_atomic_state *state) {struct drm_mode_config *config = &state->dev->mode_config;int ret; +int i; +struct drm_crtc *crtc; +struct drm_crtc_state *crtc_state; ret =drm_atomic_check_only (state); -if (ret) +if (ret) { +if (ret == -ERESTARTSYS ) + goto fail; +return ret; + }DRM_DEBUG_ATOMIC ("commiting %p nonblocking\n" , state); -return config->funcs->atomic_commit (state->dev, state, true); + ret = config->funcs->atomic_commit (state->dev, state, true); +if (ret == -ERESTARTSYS ) + goto fail; + +return ret; + +/* cleanup commit object if commit fails with ERESTARTSYS */ +fail: +for_each_crtc_in_state (state, crtc, crtc_state, i) { +if (state->crtcs[i].commit) { +drm_crtc_commit_put (state->crtcs[i].commit); + } + } + +return ret; }EXPORT_SYMBOL (drm_atomic_nonblocking_commit);
Quoted 10 lines
Quoted 7 lines
Quoted 7 lines
--
2.11.0_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel