Re: [PATCH 1/3] drm/v3d: Take a lock across GPU scheduler job creation and queuing.

From: Christian KÃnig
Date: Wed Jun 06 2018 - 04:52:16 EST


Am 06.06.2018 um 10:46 schrieb Lucas Stach:
Am Dienstag, den 05.06.2018, 12:03 -0700 schrieb Eric Anholt:
Between creation and queueing of a job, you need to prevent any other
job from being created and queued.ÂÂOtherwise the scheduler's fences
may be signaled out of seqno order.

Signed-off-by: Eric Anholt <eric@xxxxxxxxxx>
Fixes: 57692c94dcbe ("drm/v3d: Introduce a new DRM driver for Broadcom V3D V3.x+")
---

ccing amd-gfx due to interaction of this series with the scheduler.

Âdrivers/gpu/drm/v3d/v3d_drv.h |ÂÂ5 +++++
Âdrivers/gpu/drm/v3d/v3d_gem.c | 11 +++++++++--
Â2 files changed, 14 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/v3d/v3d_drv.h b/drivers/gpu/drm/v3d/v3d_drv.h
index a043ac3aae98..26005abd9c5d 100644
--- a/drivers/gpu/drm/v3d/v3d_drv.h
+++ b/drivers/gpu/drm/v3d/v3d_drv.h
@@ -85,6 +85,11 @@ struct v3d_dev {
 Â*/
 struct mutex reset_lock;
+ /* Lock taken when creating and pushing the GPU scheduler
+ Â* jobs, to keep the sched-fence seqnos in order.
+ Â*/
+ struct mutex sched_lock;
+
 struct {
 u32 num_allocated;
 u32 pages_allocated;
diff --git a/drivers/gpu/drm/v3d/v3d_gem.c b/drivers/gpu/drm/v3d/v3d_gem.c
index b513f9189caf..9ea83bdb9a30 100644
--- a/drivers/gpu/drm/v3d/v3d_gem.c
+++ b/drivers/gpu/drm/v3d/v3d_gem.c
@@ -550,13 +550,16 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 if (ret)
 goto fail;
+ mutex_lock(&v3d->sched_lock);
 if (exec->bin.start != exec->bin.end) {
 ret = drm_sched_job_init(&exec->bin.base,
 Â&v3d->queue[V3D_BIN].sched,
 Â&v3d_priv->sched_entity[V3D_BIN],
 Âv3d_priv);
- if (ret)
+ if (ret) {
+ mutex_unlock(&v3d->sched_lock);
 goto fail_unreserve;
I don't see any path where you would go to fail_unreserve with the
mutex not yet locked, so you could just fold the mutex_unlock into this
error path for a bit less code duplication.

Otherwise this looks fine.

Yeah, agree that could be cleaned up.

I can't judge the correctness of the driver, but at least the scheduler handling looks good to me.

Regards,
Christian.


Regards,
Lucas

+ }
 exec->bin_done_fence =
 dma_fence_get(&exec->bin.base.s_fence->finished);
@@ -570,12 +573,15 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 Â&v3d->queue[V3D_RENDER].sched,
 Â&v3d_priv->sched_entity[V3D_RENDER],
 Âv3d_priv);
- if (ret)
+ if (ret) {
+ mutex_unlock(&v3d->sched_lock);
 goto fail_unreserve;
+ }
 kref_get(&exec->refcount); /* put by scheduler job completion */
 drm_sched_entity_push_job(&exec->render.base,
 ÂÂ&v3d_priv->sched_entity[V3D_RENDER]);
+ mutex_unlock(&v3d->sched_lock);
 v3d_attach_object_fences(exec);
@@ -615,6 +621,7 @@ v3d_gem_init(struct drm_device *dev)
 spin_lock_init(&v3d->job_lock);
 mutex_init(&v3d->bo_lock);
 mutex_init(&v3d->reset_lock);
+ mutex_init(&v3d->sched_lock);
 /* Note: We don't allocate address 0.ÂÂVarious bits of HW
 Â* treat 0 as special, such as the occlusion query counters
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel