drm/amdgpu: fix incorrect mutex usage v3
Before this patch the scheduler fence was created when we push the job
into the queue, so we could only get the fence after pushing it.
The mutex now was necessary to prevent the thread pushing the jobs to
the hardware from running faster than the thread pushing the jobs into
the queue.
Otherwise the thread pushing jobs into the queue would have accessed
possible freed up memory when it tries to get a reference to the fence.
So what you get in the end is thread A:
mutex_lock(&job->lock);
...
Kick of thread B.
...
mutex_unlock(&job->lock);
And thread B:
mutex_lock(&job->lock);
....
mutex_unlock(&job->lock);
kfree(job);
I'm actually not sure if I'm still up to date on this, but this usage
pattern used to be not allowed with mutexes. See here as well
https://lwn.net/Articles/575460/.
v2: remove unrelated changes, fix missing owner
v3: rebased, add more commit message
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 7b02e34..0f18702 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1225,7 +1225,7 @@
struct amdgpu_device *adev;
struct amdgpu_ib *ibs;
uint32_t num_ibs;
- struct mutex job_lock;
+ void *owner;
struct amdgpu_user_fence uf;
int (*free_job)(struct amdgpu_job *job);
};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 2ae73d5..44cf977 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -845,8 +845,9 @@
goto out;
if (amdgpu_enable_scheduler && parser.num_ibs) {
- struct amdgpu_job *job;
struct amdgpu_ring * ring = parser.ibs->ring;
+ struct amd_sched_fence *fence;
+ struct amdgpu_job *job;
job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
if (!job) {
@@ -859,37 +860,41 @@
job->adev = parser.adev;
job->ibs = parser.ibs;
job->num_ibs = parser.num_ibs;
- job->base.owner = parser.filp;
- mutex_init(&job->job_lock);
+ job->owner = parser.filp;
+ job->free_job = amdgpu_cs_free_job;
+
if (job->ibs[job->num_ibs - 1].user) {
job->uf = parser.uf;
job->ibs[job->num_ibs - 1].user = &job->uf;
parser.uf.bo = NULL;
}
- parser.ibs = NULL;
- parser.num_ibs = 0;
-
- job->free_job = amdgpu_cs_free_job;
- mutex_lock(&job->job_lock);
- r = amd_sched_entity_push_job(&job->base);
- if (r) {
- mutex_unlock(&job->job_lock);
+ fence = amd_sched_fence_create(job->base.s_entity,
+ parser.filp);
+ if (!fence) {
+ r = -ENOMEM;
amdgpu_cs_free_job(job);
kfree(job);
goto out;
}
- cs->out.handle =
- amdgpu_ctx_add_fence(parser.ctx, ring,
- &job->base.s_fence->base);
+ job->base.s_fence = fence;
+ fence_get(&fence->base);
+
+ cs->out.handle = amdgpu_ctx_add_fence(parser.ctx, ring,
+ &fence->base);
job->ibs[job->num_ibs - 1].sequence = cs->out.handle;
- list_sort(NULL, &parser.validated, cmp_size_smaller_first);
- ttm_eu_fence_buffer_objects(&parser.ticket,
- &parser.validated,
- &job->base.s_fence->base);
+ parser.ibs = NULL;
+ parser.num_ibs = 0;
+
trace_amdgpu_cs_ioctl(job);
- mutex_unlock(&job->job_lock);
+ amd_sched_entity_push_job(&job->base);
+
+ list_sort(NULL, &parser.validated, cmp_size_smaller_first);
+ ttm_eu_fence_buffer_objects(&parser.ticket, &parser.validated,
+ &fence->base);
+ fence_put(&fence->base);
+
amdgpu_cs_parser_fini_late(&parser);
mutex_unlock(&vm->mutex);
return 0;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
index 8ef9e44..438c052 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
@@ -45,12 +45,8 @@
return NULL;
}
job = to_amdgpu_job(sched_job);
- mutex_lock(&job->job_lock);
trace_amdgpu_sched_run_job(job);
- r = amdgpu_ib_schedule(job->adev,
- job->num_ibs,
- job->ibs,
- job->base.owner);
+ r = amdgpu_ib_schedule(job->adev, job->num_ibs, job->ibs, job->owner);
if (r) {
DRM_ERROR("Error scheduling IBs (%d)\n", r);
goto err;
@@ -63,7 +59,6 @@
if (job->free_job)
job->free_job(job);
- mutex_unlock(&job->job_lock);
kfree(job);
return fence ? &fence->base : NULL;
}
@@ -89,21 +84,19 @@
return -ENOMEM;
job->base.sched = &ring->sched;
job->base.s_entity = &adev->kernel_ctx.rings[ring->idx].entity;
+ job->base.s_fence = amd_sched_fence_create(job->base.s_entity, owner);
+ if (!job->base.s_fence) {
+ kfree(job);
+ return -ENOMEM;
+ }
+ *f = fence_get(&job->base.s_fence->base);
+
job->adev = adev;
job->ibs = ibs;
job->num_ibs = num_ibs;
- job->base.owner = owner;
- mutex_init(&job->job_lock);
+ job->owner = owner;
job->free_job = free_job;
- mutex_lock(&job->job_lock);
- r = amd_sched_entity_push_job(&job->base);
- if (r) {
- mutex_unlock(&job->job_lock);
- kfree(job);
- return r;
- }
- *f = fence_get(&job->base.s_fence->base);
- mutex_unlock(&job->job_lock);
+ amd_sched_entity_push_job(&job->base);
} else {
r = amdgpu_ib_schedule(adev, num_ibs, ibs, owner);
if (r)