Skip to contentSkip to conversation

drm/scheduler: Add job dependency trace.

PATCH In review Andrey Grodzovsky 4 participants 4 messages Jul 31, 2018 – Aug 1, 2018 1 message not in the archive

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 drm/scheduler: Add job dependency trace.

During debug sessions I encountered a need to trace back a job dependecy a few steps back to the first failing job. This trace helpped me a lot.

---
 drivers/gpu/drm/scheduler/gpu_scheduler.c       |  8 ++++++--
 drivers/gpu/drm/scheduler/gpu_scheduler_trace.h | 24 ++++++++++++++++++++++++
 2 files changed, 30 insertions(+), 2 deletions(-)
drivers/gpu/drm/scheduler/gpu_scheduler.c +6 -2
index 3f2fc5e..45703b9 100644
@@ -506,9 +506,13 @@ drm_sched_entity_pop_job(struct drm_sched_entity *entity)
 	if (!sched_job)
 		return NULL;
 
-	while ((entity->dependency = sched->ops->dependency(sched_job, entity)))
-		if (drm_sched_entity_add_dependency_cb(entity))
+	while ((entity->dependency = sched->ops->dependency(sched_job, entity))) {
+		if (drm_sched_entity_add_dependency_cb(entity)) {
+
+			trace_drm_sched_job_wait_dep(sched_job, entity->dependency);
 			return NULL;
+		}
+	}
 
 	/* skip jobs from entity that marked guilty */
 	if (entity->guilty && atomic_read(entity->guilty))
drivers/gpu/drm/scheduler/gpu_scheduler_trace.h +24 -0
index 4998ad9..1626f39 100644
@@ -74,6 +74,30 @@ TRACE_EVENT(drm_sched_process_job,
 	    TP_printk("fence=%p signaled", __entry->fence)
 );
 
+TRACE_EVENT(drm_sched_job_wait_dep,
+	    TP_PROTO(struct drm_sched_job *sched_job, struct dma_fence *fence),
+	    TP_ARGS(sched_job, fence),
+	    TP_STRUCT__entry(
+			     __field(const char *,name)
+			     __field(uint64_t, id)
+			     __field(struct dma_fence *, fence)
+			     __field(uint64_t, ctx)
+			     __field(unsigned, seqno)
+			     ),
+
+	    TP_fast_assign(
+			   __entry->name = sched_job->sched->name;
+			   __entry->id = sched_job->id;
+			   __entry->fence = fence;
+			   __entry->ctx = fence->context;
+			   __entry->seqno = fence->seqno;
+			   ),
+	    TP_printk("job ring=%s, id=%llu, depends fence=%p, context=%llu, seq=%u",
+		      __entry->name, __entry->id,
+		      __entry->fence, __entry->ctx,
+		      __entry->seqno)
+);
+
 #endif
 
 /* This part must be outside protection */
Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#1 lore raw

PATCH 2/2 drm/amdgpu: Add job pipe sync dependecy trace

It's useful to trace any dependency a job has on prevoius jobs.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c    |  5 +++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h | 24 ++++++++++++++++++++++++
 2 files changed, 29 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +5 -0
index 5518e62..5c22cfd 100644
@@ -32,6 +32,7 @@
 #include <drm/amdgpu_drm.h>
 #include "amdgpu.h"
 #include "atom.h"
+#include "amdgpu_trace.h"
 
 #define AMDGPU_IB_TEST_TIMEOUT	msecs_to_jiffies(1000)
 
@@ -168,6 +169,10 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 	    ((tmp = amdgpu_sync_get_fence(&job->sched_sync, NULL)) ||
 	     amdgpu_vm_need_pipeline_sync(ring, job))) {
 		need_pipe_sync = true;
+
+		if (tmp)
+			trace_amdgpu_ib_pipe_sync(job, tmp);
+
 		dma_fence_put(tmp);
 	}
 
drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +24 -0
index 11f262f..3cffe53 100644
@@ -457,6 +457,30 @@ TRACE_EVENT(amdgpu_bo_move,
 			__entry->new_placement, __entry->bo_size)
 );
 
+TRACE_EVENT(amdgpu_ib_pipe_sync,
+	    TP_PROTO(struct amdgpu_job *sched_job, struct dma_fence *fence),
+	    TP_ARGS(sched_job, fence),
+	    TP_STRUCT__entry(
+			     __field(const char *,name)
+			     __field(uint64_t, id)
+			     __field(struct dma_fence *, fence)
+			     __field(uint64_t, ctx)
+			     __field(unsigned, seqno)
+			     ),
+
+	    TP_fast_assign(
+			   __entry->name = sched_job->base.sched->name;
+			   __entry->id = sched_job->base.id;
+			   __entry->fence = fence;
+			   __entry->ctx = fence->context;
+			   __entry->seqno = fence->seqno;
+			   ),
+	    TP_printk("job ring=%s, id=%llu, need pipe sync to fence=%p, context=%llu, seq=%u",
+		      __entry->name, __entry->id,
+		      __entry->fence, __entry->ctx,
+		      __entry->seqno)
+);
+
 #undef AMDGPU_JOB_GET_TIMELINE_NAME
 #endif
 
Signature · 6 lines
2.7.4

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

A message that is not in the archive (sent off-list or not yet received): 1533049014-6934-1-git-send-email-andrey.grodzovsky-5C7GfCeVMHo@public.gmane.org

Reviewed-by
#3 lore raw

Am 31.07.2018 um 16:56 schrieb Andrey Grodzovsky:

Quoted 5 lines

During debug sessions I encountered a need to trace back a job dependecy a few steps back to the first failing job. This trace helpped me a lot.

Quoted 50 lines
---
  drivers/gpu/drm/scheduler/gpu_scheduler.c       |  8 ++++++--
  drivers/gpu/drm/scheduler/gpu_scheduler_trace.h | 24 ++++++++++++++++++++++++
  2 files changed, 30 insertions(+), 2 deletions(-)
drivers/gpu/drm/scheduler/gpu_scheduler.c +6 -2
index 3f2fc5e..45703b9 100644
@@ -506,9 +506,13 @@ drm_sched_entity_pop_job(struct drm_sched_entity *entity)
  	if (!sched_job)
  		return NULL;
 
-	while ((entity->dependency = sched->ops->dependency(sched_job, entity)))
-		if (drm_sched_entity_add_dependency_cb(entity))
+	while ((entity->dependency = sched->ops->dependency(sched_job, entity))) {
+		if (drm_sched_entity_add_dependency_cb(entity)) {
+
+			trace_drm_sched_job_wait_dep(sched_job, entity->dependency);
  			return NULL;
+		}
+	}
 
  	/* skip jobs from entity that marked guilty */
  	if (entity->guilty && atomic_read(entity->guilty))
drivers/gpu/drm/scheduler/gpu_scheduler_trace.h +17 -0
index 4998ad9..1626f39 100644
@@ -74,6 +74,30 @@ TRACE_EVENT(drm_sched_process_job,
  	    TP_printk("fence=%p signaled", __entry->fence)
  );
 
+TRACE_EVENT(drm_sched_job_wait_dep,
+	    TP_PROTO(struct drm_sched_job *sched_job, struct dma_fence *fence),
+	    TP_ARGS(sched_job, fence),
+	    TP_STRUCT__entry(
+			     __field(const char *,name)
+			     __field(uint64_t, id)
+			     __field(struct dma_fence *, fence)
+			     __field(uint64_t, ctx)
+			     __field(unsigned, seqno)
+			     ),
+
+	    TP_fast_assign(
+			   __entry->name = sched_job->sched->name;
+			   __entry->id = sched_job->id;
+			   __entry->fence = fence;
+			   __entry->ctx = fence->context;
+			   __entry->seqno = fence->seqno;

drivers/gpu/drm/scheduler/gpu_scheduler_trace.h TRACE_EVENT(drm_sched_process_job, · line 103

Quoted 10 lines
drivers/gpu/drm/scheduler/gpu_scheduler_trace.h +7 -0
+			   ),
+	    TP_printk("job ring=%s, id=%llu, depends fence=%p, context=%llu, seq=%u",
+		      __entry->name, __entry->id,
+		      __entry->fence, __entry->ctx,
+		      __entry->seqno)
+);
+
  #endif
 
  /* This part must be outside protection */

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

#4 lore raw
In reply to Andrey Grodzovsky
Quoted 5 lines · Andrey Grodzovsky

During debug sessions I encountered a need to trace back a job dependecy a few steps back to the first failing job. This trace helpped me a lot.

Series are Reviewed-by: Huang Rui <[email protected]>

Quoted 67 lines
---
 drivers/gpu/drm/scheduler/gpu_scheduler.c       |  8 ++++++--
 drivers/gpu/drm/scheduler/gpu_scheduler_trace.h | 24 ++++++++++++++++++++++++
 2 files changed, 30 insertions(+), 2 deletions(-)
drivers/gpu/drm/scheduler/gpu_scheduler.c +6 -2
index 3f2fc5e..45703b9 100644
@@ -506,9 +506,13 @@ drm_sched_entity_pop_job(struct drm_sched_entity *entity)
 	if (!sched_job)
 		return NULL;
 
-	while ((entity->dependency = sched->ops->dependency(sched_job, entity)))
-		if (drm_sched_entity_add_dependency_cb(entity))
+	while ((entity->dependency = sched->ops->dependency(sched_job, entity))) {
+		if (drm_sched_entity_add_dependency_cb(entity)) {
+
+			trace_drm_sched_job_wait_dep(sched_job, entity->dependency);
 			return NULL;
+		}
+	}
 
 	/* skip jobs from entity that marked guilty */
 	if (entity->guilty && atomic_read(entity->guilty))
drivers/gpu/drm/scheduler/gpu_scheduler_trace.h +24 -0
index 4998ad9..1626f39 100644
@@ -74,6 +74,30 @@ TRACE_EVENT(drm_sched_process_job,
 	    TP_printk("fence=%p signaled", __entry->fence)
 );
 
+TRACE_EVENT(drm_sched_job_wait_dep,
+	    TP_PROTO(struct drm_sched_job *sched_job, struct dma_fence *fence),
+	    TP_ARGS(sched_job, fence),
+	    TP_STRUCT__entry(
+			     __field(const char *,name)
+			     __field(uint64_t, id)
+			     __field(struct dma_fence *, fence)
+			     __field(uint64_t, ctx)
+			     __field(unsigned, seqno)
+			     ),
+
+	    TP_fast_assign(
+			   __entry->name = sched_job->sched->name;
+			   __entry->id = sched_job->id;
+			   __entry->fence = fence;
+			   __entry->ctx = fence->context;
+			   __entry->seqno = fence->seqno;
+			   ),
+	    TP_printk("job ring=%s, id=%llu, depends fence=%p, context=%llu, seq=%u",
+		      __entry->name, __entry->id,
+		      __entry->fence, __entry->ctx,
+		      __entry->seqno)
+);
+
 #endif
 
 /* This part must be outside protection */
Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel