diff mbox series

[RFC,61/97] drm/i915: Hold reference to intel_context over life of i915_request

Message ID 20210506191451.77768-62-matthew.brost@intel.com (mailing list archive)
State New, archived
Headers show
Series Basic GuC submission support in the i915 | expand

Commit Message

Matthew Brost May 6, 2021, 7:14 p.m. UTC
Hold a reference to the intel_context over life of an i915_request.
Without this an i915_request can exist after the context has been
destroyed (e.g. request retired, context closed, but user space holds a
reference to the request from an out fence). In the case of GuC
submission + virtual engine, the engine that the request references is
also destroyed which can trigger bad pointer dref in fence ops (e.g.
i915_fence_get_driver_name). We could likely change
i915_fence_get_driver_name to avoid touching the engine but let's just
be safe and hold the intel_context reference.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
---
 drivers/gpu/drm/i915/i915_request.c | 54 ++++++++++++-----------------
 1 file changed, 22 insertions(+), 32 deletions(-)

Comments

Tvrtko Ursulin June 2, 2021, 12:18 p.m. UTC | #1
On 06/05/2021 20:14, Matthew Brost wrote:
> Hold a reference to the intel_context over life of an i915_request.
> Without this an i915_request can exist after the context has been
> destroyed (e.g. request retired, context closed, but user space holds a
> reference to the request from an out fence). In the case of GuC
> submission + virtual engine, the engine that the request references is
> also destroyed which can trigger bad pointer dref in fence ops (e.g.
> i915_fence_get_driver_name). We could likely change
> i915_fence_get_driver_name to avoid touching the engine but let's just
> be safe and hold the intel_context reference.

Isn't this a bug in present upstream as well? Like calling sync fence 
info on retired requests or something else?

If it is a bug in upstream then I think a single patch to deal with the 
issue should be posted independently. It may be as simple as checking 
for the signaled bit in i915_fence_get_driver_name and dereferencing 
with rcu protection.

Regards,

Tvrtko

> Signed-off-by: Matthew Brost <matthew.brost@intel.com>
> ---
>   drivers/gpu/drm/i915/i915_request.c | 54 ++++++++++++-----------------
>   1 file changed, 22 insertions(+), 32 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
> index 127d60b36422..0b96b824ea06 100644
> --- a/drivers/gpu/drm/i915/i915_request.c
> +++ b/drivers/gpu/drm/i915/i915_request.c
> @@ -126,39 +126,17 @@ static void i915_fence_release(struct dma_fence *fence)
>   	i915_sw_fence_fini(&rq->semaphore);
>   
>   	/*
> -	 * Keep one request on each engine for reserved use under mempressure
> -	 *
> -	 * We do not hold a reference to the engine here and so have to be
> -	 * very careful in what rq->engine we poke. The virtual engine is
> -	 * referenced via the rq->context and we released that ref during
> -	 * i915_request_retire(), ergo we must not dereference a virtual
> -	 * engine here. Not that we would want to, as the only consumer of
> -	 * the reserved engine->request_pool is the power management parking,
> -	 * which must-not-fail, and that is only run on the physical engines.
> -	 *
> -	 * Since the request must have been executed to be have completed,
> -	 * we know that it will have been processed by the HW and will
> -	 * not be unsubmitted again, so rq->engine and rq->execution_mask
> -	 * at this point is stable. rq->execution_mask will be a single
> -	 * bit if the last and _only_ engine it could execution on was a
> -	 * physical engine, if it's multiple bits then it started on and
> -	 * could still be on a virtual engine. Thus if the mask is not a
> -	 * power-of-two we assume that rq->engine may still be a virtual
> -	 * engine and so a dangling invalid pointer that we cannot dereference
> -	 *
> -	 * For example, consider the flow of a bonded request through a virtual
> -	 * engine. The request is created with a wide engine mask (all engines
> -	 * that we might execute on). On processing the bond, the request mask
> -	 * is reduced to one or more engines. If the request is subsequently
> -	 * bound to a single engine, it will then be constrained to only
> -	 * execute on that engine and never returned to the virtual engine
> -	 * after timeslicing away, see __unwind_incomplete_requests(). Thus we
> -	 * know that if the rq->execution_mask is a single bit, rq->engine
> -	 * can be a physical engine with the exact corresponding mask.
> +	 * Keep one request on each engine for reserved use under mempressure,
> +	 * do not use with virtual engines as this really is only needed for
> +	 * kernel contexts.
>   	 */
> -	if (is_power_of_2(rq->execution_mask) &&
> -	    !cmpxchg(&rq->engine->request_pool, NULL, rq))
> +	if (!intel_engine_is_virtual(rq->engine) &&
> +	    !cmpxchg(&rq->engine->request_pool, NULL, rq)) {
> +		intel_context_put(rq->context);
>   		return;
> +	}
> +
> +	intel_context_put(rq->context);
>   
>   	kmem_cache_free(global.slab_requests, rq);
>   }
> @@ -977,7 +955,18 @@ __i915_request_create(struct intel_context *ce, gfp_t gfp)
>   		}
>   	}
>   
> -	rq->context = ce;
> +	/*
> +	 * Hold a reference to the intel_context over life of an i915_request.
> +	 * Without this an i915_request can exist after the context has been
> +	 * destroyed (e.g. request retired, context closed, but user space holds
> +	 * a reference to the request from an out fence). In the case of GuC
> +	 * submission + virtual engine, the engine that the request references
> +	 * is also destroyed which can trigger bad pointer dref in fence ops
> +	 * (e.g. i915_fence_get_driver_name). We could likely change these
> +	 * functions to avoid touching the engine but let's just be safe and
> +	 * hold the intel_context reference.
> +	 */
> +	rq->context = intel_context_get(ce);
>   	rq->engine = ce->engine;
>   	rq->ring = ce->ring;
>   	rq->execution_mask = ce->engine->mask;
> @@ -1054,6 +1043,7 @@ __i915_request_create(struct intel_context *ce, gfp_t gfp)
>   	GEM_BUG_ON(!list_empty(&rq->sched.waiters_list));
>   
>   err_free:
> +	intel_context_put(ce);
>   	kmem_cache_free(global.slab_requests, rq);
>   err_unreserve:
>   	intel_context_unpin(ce);
>
diff mbox series

Patch

diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index 127d60b36422..0b96b824ea06 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -126,39 +126,17 @@  static void i915_fence_release(struct dma_fence *fence)
 	i915_sw_fence_fini(&rq->semaphore);
 
 	/*
-	 * Keep one request on each engine for reserved use under mempressure
-	 *
-	 * We do not hold a reference to the engine here and so have to be
-	 * very careful in what rq->engine we poke. The virtual engine is
-	 * referenced via the rq->context and we released that ref during
-	 * i915_request_retire(), ergo we must not dereference a virtual
-	 * engine here. Not that we would want to, as the only consumer of
-	 * the reserved engine->request_pool is the power management parking,
-	 * which must-not-fail, and that is only run on the physical engines.
-	 *
-	 * Since the request must have been executed to be have completed,
-	 * we know that it will have been processed by the HW and will
-	 * not be unsubmitted again, so rq->engine and rq->execution_mask
-	 * at this point is stable. rq->execution_mask will be a single
-	 * bit if the last and _only_ engine it could execution on was a
-	 * physical engine, if it's multiple bits then it started on and
-	 * could still be on a virtual engine. Thus if the mask is not a
-	 * power-of-two we assume that rq->engine may still be a virtual
-	 * engine and so a dangling invalid pointer that we cannot dereference
-	 *
-	 * For example, consider the flow of a bonded request through a virtual
-	 * engine. The request is created with a wide engine mask (all engines
-	 * that we might execute on). On processing the bond, the request mask
-	 * is reduced to one or more engines. If the request is subsequently
-	 * bound to a single engine, it will then be constrained to only
-	 * execute on that engine and never returned to the virtual engine
-	 * after timeslicing away, see __unwind_incomplete_requests(). Thus we
-	 * know that if the rq->execution_mask is a single bit, rq->engine
-	 * can be a physical engine with the exact corresponding mask.
+	 * Keep one request on each engine for reserved use under mempressure,
+	 * do not use with virtual engines as this really is only needed for
+	 * kernel contexts.
 	 */
-	if (is_power_of_2(rq->execution_mask) &&
-	    !cmpxchg(&rq->engine->request_pool, NULL, rq))
+	if (!intel_engine_is_virtual(rq->engine) &&
+	    !cmpxchg(&rq->engine->request_pool, NULL, rq)) {
+		intel_context_put(rq->context);
 		return;
+	}
+
+	intel_context_put(rq->context);
 
 	kmem_cache_free(global.slab_requests, rq);
 }
@@ -977,7 +955,18 @@  __i915_request_create(struct intel_context *ce, gfp_t gfp)
 		}
 	}
 
-	rq->context = ce;
+	/*
+	 * Hold a reference to the intel_context over life of an i915_request.
+	 * Without this an i915_request can exist after the context has been
+	 * destroyed (e.g. request retired, context closed, but user space holds
+	 * a reference to the request from an out fence). In the case of GuC
+	 * submission + virtual engine, the engine that the request references
+	 * is also destroyed which can trigger bad pointer dref in fence ops
+	 * (e.g. i915_fence_get_driver_name). We could likely change these
+	 * functions to avoid touching the engine but let's just be safe and
+	 * hold the intel_context reference.
+	 */
+	rq->context = intel_context_get(ce);
 	rq->engine = ce->engine;
 	rq->ring = ce->ring;
 	rq->execution_mask = ce->engine->mask;
@@ -1054,6 +1043,7 @@  __i915_request_create(struct intel_context *ce, gfp_t gfp)
 	GEM_BUG_ON(!list_empty(&rq->sched.waiters_list));
 
 err_free:
+	intel_context_put(ce);
 	kmem_cache_free(global.slab_requests, rq);
 err_unreserve:
 	intel_context_unpin(ce);