The i915_request holds a reference to intel_context, which in turn holds a reference on the VM. But the dma-resv update for VM_BIND feature would require VM hold a reference to the i915_request through dma-resv fences of VM_PRIVATE objects (which share a per VM dma-resv object).
Thus, we have a circular reference pattern causing the VM reference to never reach 0, hence VM is not destroyed.
Break this by reverting the below patch which is making the i915_request to hold a reference on intel_context. "drm/i915: Hold reference to intel_context over life of i915_request"
This means we can't access rq->engine in i915_fence_get_driver_name() as user do not hold a reference on rq->engine here. So, instead store required device private pointer in 'rq->i915' and use it.
Niranjana Vishwanathapura (2): drm/i915: Do not access rq->engine without a reference Revert "drm/i915: Hold reference to intel_context over life of i915_request"
Ramalingam C (1): drm/i915: Do not use reserved requests for virtual engines
drivers/gpu/drm/i915/i915_request.c | 55 ++++++++++++++++++----------- drivers/gpu/drm/i915/i915_request.h | 2 ++ 2 files changed, 36 insertions(+), 21 deletions(-)
From: Niranjana Vishwanathapura niranjana.vishwanathapura@intel.com
In i915_fence_get_driver_name(), user may not hold a reference to rq->engine. Hence do not access it. Instead, store required device private pointer in 'rq->i915' and use it.
Signed-off-by: Niranjana Vishwanathapura niranjana.vishwanathapura@intel.com Suggested-by: Matthew Brost matthew.brost@intel.com --- drivers/gpu/drm/i915/i915_request.c | 3 ++- drivers/gpu/drm/i915/i915_request.h | 2 ++ 2 files changed, 4 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c index 73d5195146b0..7f6998bf390c 100644 --- a/drivers/gpu/drm/i915/i915_request.c +++ b/drivers/gpu/drm/i915/i915_request.c @@ -60,7 +60,7 @@ static struct kmem_cache *slab_execute_cbs;
static const char *i915_fence_get_driver_name(struct dma_fence *fence) { - return dev_name(to_request(fence)->engine->i915->drm.dev); + return dev_name(to_request(fence)->i915->drm.dev); }
static const char *i915_fence_get_timeline_name(struct dma_fence *fence) @@ -937,6 +937,7 @@ __i915_request_create(struct intel_context *ce, gfp_t gfp) rq->engine = ce->engine; rq->ring = ce->ring; rq->execution_mask = ce->engine->mask; + rq->i915 = ce->engine->i915;
ret = intel_timeline_get_seqno(tl, rq, &seqno); if (ret) diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h index 28b1f9db5487..47041ec68df8 100644 --- a/drivers/gpu/drm/i915/i915_request.h +++ b/drivers/gpu/drm/i915/i915_request.h @@ -196,6 +196,8 @@ struct i915_request { struct dma_fence fence; spinlock_t lock;
+ struct drm_i915_private *i915; + /** * Context and ring buffer related to this request * Contexts are refcounted, so when this request is associated with a
From: Niranjana Vishwanathapura niranjana.vishwanathapura@intel.com
This reverts commit 1e98d8c52ed5dfbaf273c4423c636525c2ce59e7.
The problem with this patch is that it makes i915_request to hold a reference to intel_context, which in turn holds a reference on the VM. This strong back referencing can lead to reference loops which leads to resource leak.
An example is the upcoming VM_BIND work which requires VM to hold a reference to some shared VM specific BO. But this BO's dma-resv fences holds reference to the i915_request thus leading to reference loop.
Signed-off-by: Niranjana Vishwanathapura niranjana.vishwanathapura@intel.com Signed-off-by: Ramalingam C ramalingam.c@intel.com Suggested-by: Matthew Brost matthew.brost@intel.com --- drivers/gpu/drm/i915/i915_request.c | 55 +++++++++++++++++------------ 1 file changed, 32 insertions(+), 23 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c index 7f6998bf390c..c71905d8e154 100644 --- a/drivers/gpu/drm/i915/i915_request.c +++ b/drivers/gpu/drm/i915/i915_request.c @@ -134,17 +134,39 @@ 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, - * do not use with virtual engines as this really is only needed for - * kernel contexts. + * 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. */ - if (!intel_engine_is_virtual(rq->engine) && - !cmpxchg(&rq->engine->request_pool, NULL, rq)) { - intel_context_put(rq->context); + if (is_power_of_2(rq->execution_mask) && + !cmpxchg(&rq->engine->request_pool, NULL, rq)) return; - } - - intel_context_put(rq->context);
kmem_cache_free(slab_requests, rq); } @@ -921,19 +943,7 @@ __i915_request_create(struct intel_context *ce, gfp_t gfp) } }
- /* - * 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. In execlist mode the request always - * eventually points to a physical engine so this isn't an issue. - */ - rq->context = intel_context_get(ce); + rq->context = ce; rq->engine = ce->engine; rq->ring = ce->ring; rq->execution_mask = ce->engine->mask; @@ -1009,7 +1019,6 @@ __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(slab_requests, rq); err_unreserve: intel_context_unpin(ce);
Do not use reserved requests for virtual engines as this is only needed for kernel contexts.
Signed-off-by: Ramalingam C ramalingam.c@intel.com Suggested-by: Matthew Brost matthew.brost@intel.com --- drivers/gpu/drm/i915/i915_request.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c index c71905d8e154..f0392b053bca 100644 --- a/drivers/gpu/drm/i915/i915_request.c +++ b/drivers/gpu/drm/i915/i915_request.c @@ -135,6 +135,8 @@ static void i915_fence_release(struct dma_fence *fence)
/* * 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. * * 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 @@ -164,7 +166,8 @@ static void i915_fence_release(struct dma_fence *fence) * know that if the rq->execution_mask is a single bit, rq->engine * can be a physical engine with the exact corresponding mask. */ - if (is_power_of_2(rq->execution_mask) && + if (!intel_engine_is_virtual(rq->engine) && + is_power_of_2(rq->execution_mask) && !cmpxchg(&rq->engine->request_pool, NULL, rq)) return;
dri-devel@lists.freedesktop.org
-
Ramalingam C