[Intel-gfx] [PATCH] drm/i915: Record the current requests queue for execlists upon hang

Mika Kuoppala mika.kuoppala at linux.intel.com
Thu Oct 13 09:51:26 UTC 2016


Chris Wilson <chris at chris-wilson.co.uk> writes:

> Mika wanted to know what requests were pending at the time of a hang as
> we now track which requests we have submitted to the hardware.
>
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Mika Kuoppala <mika.kuoppala at intel.com>
> ---
>  drivers/gpu/drm/i915/i915_drv.h       |  3 +-
>  drivers/gpu/drm/i915/i915_gpu_error.c | 64 ++++++++++++++++++++++++-----------
>  2 files changed, 47 insertions(+), 20 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index bf397b643cc0..6360e807c6ba 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -832,10 +832,11 @@ struct drm_i915_error_state {
>  		struct drm_i915_error_request {
>  			long jiffies;
>  			pid_t pid;
> +			u32 context;
>  			u32 seqno;
>  			u32 head;
>  			u32 tail;
> -		} *requests;
> +		} *requests, execlist[2];
>  
>  		struct drm_i915_error_waiter {
>  			char comm[TASK_COMM_LEN];
> diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
> index 78cc13b9b2a5..026b78c66219 100644
> --- a/drivers/gpu/drm/i915/i915_gpu_error.c
> +++ b/drivers/gpu/drm/i915/i915_gpu_error.c
> @@ -363,6 +363,20 @@ static void error_print_instdone(struct drm_i915_error_state_buf *m,
>  			   ee->instdone.row[slice][subslice]);
>  }
>  
> +static void error_print_request(struct drm_i915_error_state_buf *m,
> +				const char *prefix,
> +				struct drm_i915_error_request *erq)
> +{
> +	if (!erq->seqno)
> +		return;
> +
> +	err_printf(m, "%s pid %d, seqno %8x:%08x, emitted %dms ago, head %08x, tail %08x\n",
> +		   prefix, erq->pid,
> +		   erq->context, erq->seqno,
> +		   jiffies_to_msecs(jiffies - erq->jiffies),
> +		   erq->head, erq->tail);
> +}
> +
>  static void error_print_engine(struct drm_i915_error_state_buf *m,
>  			       struct drm_i915_error_engine *ee)
>  {
> @@ -434,6 +448,8 @@ static void error_print_engine(struct drm_i915_error_state_buf *m,
>  	err_printf(m, "  hangcheck: %s [%d]\n",
>  		   hangcheck_action_to_str(ee->hangcheck_action),
>  		   ee->hangcheck_score);
> +	error_print_request(m, "  ELSP[0]: ", &ee->execlist[0]);
> +	error_print_request(m, "  ELSP[1]: ", &ee->execlist[1]);
>  }
>  
>  void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
> @@ -649,14 +665,8 @@ int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
>  			err_printf(m, "%s --- %d requests\n",
>  				   dev_priv->engine[i].name,
>  				   ee->num_requests);
> -			for (j = 0; j < ee->num_requests; j++) {
> -				err_printf(m, "  pid %d, seqno 0x%08x, emitted %ld, head 0x%08x, tail 0x%08x\n",
> -					   ee->requests[j].pid,
> -					   ee->requests[j].seqno,
> -					   ee->requests[j].jiffies,
> -					   ee->requests[j].head,
> -					   ee->requests[j].tail);
> -			}
> +			for (j = 0; j < ee->num_requests; j++)
> +				error_print_request(m, " ", &ee->requests[j]);
>  		}
>  
>  		if (IS_ERR(ee->waiters)) {
> @@ -1155,6 +1165,20 @@ static void error_record_engine_registers(struct drm_i915_error_state *error,
>  	}
>  }
>  
> +static void record_request(struct drm_i915_gem_request *request,
> +			   struct drm_i915_error_request *erq)
> +{
> +	erq->context = request->ctx->hw_id;
> +	erq->seqno = request->fence.seqno;
> +	erq->jiffies = request->emitted_jiffies;
> +	erq->head = request->head;
> +	erq->tail = request->tail;
> +
> +	rcu_read_lock();
> +	erq->pid = request->ctx->pid ? pid_nr(request->ctx->pid) : 0;

This lock is only for the pid_nr and nothing to do with ctx dereference?
Not that it was added by this patch...

> +	rcu_read_unlock();
> +}
> +
>  static void engine_record_requests(struct intel_engine_cs *engine,
>  				   struct drm_i915_gem_request *first,
>  				   struct drm_i915_error_engine *ee)
> @@ -1178,8 +1202,6 @@ static void engine_record_requests(struct intel_engine_cs *engine,
>  	count = 0;
>  	request = first;
>  	list_for_each_entry_from(request, &engine->request_list, link) {
> -		struct drm_i915_error_request *erq;
> -
>  		if (count >= ee->num_requests) {
>  			/*
>  			 * If the ring request list was changed in
> @@ -1199,19 +1221,22 @@ static void engine_record_requests(struct intel_engine_cs *engine,
>  			break;
>  		}
>  
> -		erq = &ee->requests[count++];
> -		erq->seqno = request->fence.seqno;
> -		erq->jiffies = request->emitted_jiffies;
> -		erq->head = request->head;
> -		erq->tail = request->tail;
> -
> -		rcu_read_lock();
> -		erq->pid = request->ctx->pid ? pid_nr(request->ctx->pid) : 0;
> -		rcu_read_unlock();
> +		record_request(request, &ee->requests[count++]);
>  	}
>  	ee->num_requests = count;
>  }
>  
> +static void error_record_engine_execlists(struct intel_engine_cs *engine,
> +					  struct drm_i915_error_engine *ee)
> +{
> +	unsigned int n;
> +
> +	for (n = 0; n < ARRAY_SIZE(engine->execlist_port); n++)
> +		if (engine->execlist_port[n].request)
> +			record_request(engine->execlist_port[n].request,
> +				       &ee->execlist[n]);

Ok even if we get interrupt at around here and reset the ports,
the pointer should stay in request_list and at that part we should be
safe.

And with retirement, we are in no more unsafer waters as with the other
requests.

Reviewed-by: Mika Kuoppala <mika.kuoppala at intel.com>

> +}
> +
>  static void i915_gem_record_rings(struct drm_i915_private *dev_priv,
>  				  struct drm_i915_error_state *error)
>  {
> @@ -1236,6 +1261,7 @@ static void i915_gem_record_rings(struct drm_i915_private *dev_priv,
>  
>  		error_record_engine_registers(error, engine, ee);
>  		error_record_engine_waiters(engine, ee);
> +		error_record_engine_execlists(engine, ee);
>  
>  		request = i915_gem_find_active_request(engine);
>  		if (request) {
> -- 
> 2.9.3


More information about the Intel-gfx mailing list