While the labels may mislead the casual reader, the tail of the function etnaviv_ioctl_gem_submit is always executed, as a lot of the structures set up in this function need to be cleaned up regardless of whether the submit succeeded or failed.
An exception is the newly added drm_sched_job_cleanup, which must only be called when the submit failed before handing the job to the scheduler.
Fixes: b827c84f5e84 ("drm/etnaviv: Use scheduler dependency handling") Reported-by: Michael Walle michael@walle.cc Signed-off-by: Lucas Stach l.stach@pengutronix.de --- Daniel, please apply to the appropriate drm-misc branch to make sure this ends up in the same pullrequest as the offending commit. --- drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c b/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c index 98bb5c9239de..bf2a400f59c9 100644 --- a/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c +++ b/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c @@ -607,7 +607,8 @@ int etnaviv_ioctl_gem_submit(struct drm_device *dev, void *data, args->fence = submit->out_fence_id;
err_submit_job: - drm_sched_job_cleanup(&submit->sched_job); + if (ret) + drm_sched_job_cleanup(&submit->sched_job); err_submit_put: etnaviv_submit_put(submit);
On Fri, Apr 29, 2022 at 09:17:31PM +0200, Lucas Stach wrote:
While the labels may mislead the casual reader, the tail of the function etnaviv_ioctl_gem_submit is always executed, as a lot of the structures set up in this function need to be cleaned up regardless of whether the submit succeeded or failed.
An exception is the newly added drm_sched_job_cleanup, which must only be called when the submit failed before handing the job to the scheduler.
Fixes: b827c84f5e84 ("drm/etnaviv: Use scheduler dependency handling") Reported-by: Michael Walle michael@walle.cc Signed-off-by: Lucas Stach l.stach@pengutronix.de
Daniel, please apply to the appropriate drm-misc branch to make sure this ends up in the same pullrequest as the offending commit.
Apologies on being behind on emails, I just came to the same conclusion.
Except there's one issue, the error path for sync_file_create is after a successful call to drm_sched_entity_push_job(), so in that case we must again _not_ call drm_sched_job_cleanup. So we need to change that goto err_submit_job to goto err_submit_put and maybe add a comment about why.
Can you pls respin?
Thanks, Daniel
drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c b/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c index 98bb5c9239de..bf2a400f59c9 100644 --- a/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c +++ b/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c @@ -607,7 +607,8 @@ int etnaviv_ioctl_gem_submit(struct drm_device *dev, void *data, args->fence = submit->out_fence_id;
err_submit_job:
- drm_sched_job_cleanup(&submit->sched_job);
- if (ret)
drm_sched_job_cleanup(&submit->sched_job);
err_submit_put: etnaviv_submit_put(submit);
-- 2.30.2
Am Mittwoch, dem 04.05.2022 um 10:14 +0200 schrieb Daniel Vetter:
On Fri, Apr 29, 2022 at 09:17:31PM +0200, Lucas Stach wrote:
While the labels may mislead the casual reader, the tail of the function etnaviv_ioctl_gem_submit is always executed, as a lot of the structures set up in this function need to be cleaned up regardless of whether the submit succeeded or failed.
An exception is the newly added drm_sched_job_cleanup, which must only be called when the submit failed before handing the job to the scheduler.
Fixes: b827c84f5e84 ("drm/etnaviv: Use scheduler dependency handling") Reported-by: Michael Walle michael@walle.cc Signed-off-by: Lucas Stach l.stach@pengutronix.de
Daniel, please apply to the appropriate drm-misc branch to make sure this ends up in the same pullrequest as the offending commit.
Apologies on being behind on emails, I just came to the same conclusion.
Except there's one issue, the error path for sync_file_create is after a successful call to drm_sched_entity_push_job(), so in that case we must again _not_ call drm_sched_job_cleanup. So we need to change that goto err_submit_job to goto err_submit_put and maybe add a comment about why.
Can you pls respin?
Just did that.
In general I don't like this freestanding and quite non-obvious error handling for the sched_job and would like to move this into the etnaviv submit_cleanup routine, but to do so I would need a reliable way to detect if drm_sched_job_init was successfully called on the job. This seems like a bit more change than I would like to add on top of the current patches at this point. I'll take a note to clean this up later, once the regression is fixed.
Regards, Lucas
Thanks, Daniel
drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c b/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c index 98bb5c9239de..bf2a400f59c9 100644 --- a/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c +++ b/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c @@ -607,7 +607,8 @@ int etnaviv_ioctl_gem_submit(struct drm_device *dev, void *data, args->fence = submit->out_fence_id;
err_submit_job:
- drm_sched_job_cleanup(&submit->sched_job);
- if (ret)
drm_sched_job_cleanup(&submit->sched_job);
err_submit_put: etnaviv_submit_put(submit);
-- 2.30.2
dri-devel@lists.freedesktop.org