From b5773a3616d10bdef60662b8780ded912ee5d7b6 Mon Sep 17 00:00:00 2001 From: Chris Wilson Date: Thu, 28 Feb 2019 22:06:39 +0000 Subject: [PATCH] drm/i915/execlists: Suppress mere WAIT preemption WAIT is occasionally suppressed by virtue of preempted requests being promoted to NEWCLIENT if they have not all ready received that boost. Make this consistent for all WAIT boosts that they are not allowed to preempt executing contexts and are merely granted the right to be at the front of the queue for the next execution slot. This is in keeping with the desire that the WAIT boost be a minor tweak that does not give excessive promotion to its user and open ourselves to trivial abuse. The problem with the inconsistent WAIT preemption becomes more apparent as the preemption is propagated across the engines, where one engine may preempt and the other not, and we be relying on the exact execution order being consistent across engines (e.g. using HW semaphores to coordinate parallel execution). v2: Also protect GuC submission from false preemption loops. v3: Build bug safeguards and better debug messages for st. v4: Do the priority bumping in unsubmit (i.e. on preemption/reset unwind), applying it earlier during submit causes out-of-order execution combined with execute fences. v5: Call sw_fence_fini for our dummy request (Matthew) Signed-off-by: Chris Wilson Cc: Tvrtko Ursulin Reviewed-by: Tvrtko Ursulin Cc: Matthew Auld Link: https://patchwork.freedesktop.org/patch/msgid/20190228220639.3173-1-chris@chris-wilson.co.uk --- drivers/gpu/drm/i915/i915_request.c | 15 ++ drivers/gpu/drm/i915/i915_scheduler.c | 1 - drivers/gpu/drm/i915/i915_scheduler.h | 2 + drivers/gpu/drm/i915/intel_guc_submission.c | 2 +- drivers/gpu/drm/i915/intel_lrc.c | 9 +- drivers/gpu/drm/i915/selftests/intel_lrc.c | 163 ++++++++++++++++++++ 6 files changed, 189 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c index a011bf4be48e..c65f6c990fdd 100644 --- a/drivers/gpu/drm/i915/i915_request.c +++ b/drivers/gpu/drm/i915/i915_request.c @@ -358,11 +358,14 @@ void __i915_request_submit(struct i915_request *request) /* We may be recursing from the signal callback of another i915 fence */ spin_lock_nested(&request->lock, SINGLE_DEPTH_NESTING); + GEM_BUG_ON(test_bit(I915_FENCE_FLAG_ACTIVE, &request->fence.flags)); set_bit(I915_FENCE_FLAG_ACTIVE, &request->fence.flags); + if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &request->fence.flags) && !i915_request_enable_breadcrumb(request)) intel_engine_queue_breadcrumbs(engine); + spin_unlock(&request->lock); engine->emit_fini_breadcrumb(request, @@ -406,10 +409,22 @@ void __i915_request_unsubmit(struct i915_request *request) /* We may be recursing from the signal callback of another i915 fence */ spin_lock_nested(&request->lock, SINGLE_DEPTH_NESTING); + + /* + * As we do not allow WAIT to preempt inflight requests, + * once we have executed a request, along with triggering + * any execution callbacks, we must preserve its ordering + * within the non-preemptible FIFO. + */ + BUILD_BUG_ON(__NO_PREEMPTION & ~I915_PRIORITY_MASK); /* only internal */ + request->sched.attr.priority |= __NO_PREEMPTION; + if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &request->fence.flags)) i915_request_cancel_breadcrumb(request); + GEM_BUG_ON(!test_bit(I915_FENCE_FLAG_ACTIVE, &request->fence.flags)); clear_bit(I915_FENCE_FLAG_ACTIVE, &request->fence.flags); + spin_unlock(&request->lock); /* Transfer back from the global per-engine timeline to per-context */ diff --git a/drivers/gpu/drm/i915/i915_scheduler.c b/drivers/gpu/drm/i915/i915_scheduler.c index 0dd720593f9c..50018ad30233 100644 --- a/drivers/gpu/drm/i915/i915_scheduler.c +++ b/drivers/gpu/drm/i915/i915_scheduler.c @@ -324,7 +324,6 @@ static void __i915_schedule(struct i915_request *rq, if (node_signaled(p->signaler)) continue; - GEM_BUG_ON(p->signaler->attr.priority < node->attr.priority); if (prio > READ_ONCE(p->signaler->attr.priority)) list_move_tail(&p->dfs_link, &dfs); } diff --git a/drivers/gpu/drm/i915/i915_scheduler.h b/drivers/gpu/drm/i915/i915_scheduler.h index bb3d496d4c49..7d4a49750d92 100644 --- a/drivers/gpu/drm/i915/i915_scheduler.h +++ b/drivers/gpu/drm/i915/i915_scheduler.h @@ -33,6 +33,8 @@ enum { #define I915_PRIORITY_WAIT ((u8)BIT(0)) #define I915_PRIORITY_NEWCLIENT ((u8)BIT(1)) +#define __NO_PREEMPTION (I915_PRIORITY_WAIT) + struct i915_sched_attr { /** * @priority: execution and service priority diff --git a/drivers/gpu/drm/i915/intel_guc_submission.c b/drivers/gpu/drm/i915/intel_guc_submission.c index 4366db7978a8..56ba2fcbabe6 100644 --- a/drivers/gpu/drm/i915/intel_guc_submission.c +++ b/drivers/gpu/drm/i915/intel_guc_submission.c @@ -720,7 +720,7 @@ static inline int rq_prio(const struct i915_request *rq) static inline int port_prio(const struct execlist_port *port) { - return rq_prio(port_request(port)); + return rq_prio(port_request(port)) | __NO_PREEMPTION; } static bool __guc_dequeue(struct intel_engine_cs *engine) diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c index 661d2f6da84f..4f2187aa44e4 100644 --- a/drivers/gpu/drm/i915/intel_lrc.c +++ b/drivers/gpu/drm/i915/intel_lrc.c @@ -188,6 +188,12 @@ static inline int rq_prio(const struct i915_request *rq) return rq->sched.attr.priority; } +static int effective_prio(const struct i915_request *rq) +{ + /* Restrict mere WAIT boosts from triggering preemption */ + return rq_prio(rq) | __NO_PREEMPTION; +} + static int queue_prio(const struct intel_engine_execlists *execlists) { struct i915_priolist *p; @@ -208,7 +214,7 @@ static int queue_prio(const struct intel_engine_execlists *execlists) static inline bool need_preempt(const struct intel_engine_cs *engine, const struct i915_request *rq) { - const int last_prio = rq_prio(rq); + int last_prio; if (!intel_engine_has_preemption(engine)) return false; @@ -228,6 +234,7 @@ static inline bool need_preempt(const struct intel_engine_cs *engine, * preempt. If that hint is stale or we may be trying to preempt * ourselves, ignore the request. */ + last_prio = effective_prio(rq); if (!__execlists_need_preempt(engine->execlists.queue_priority_hint, last_prio)) return false; diff --git a/drivers/gpu/drm/i915/selftests/intel_lrc.c b/drivers/gpu/drm/i915/selftests/intel_lrc.c index 0f7a5bf69646..0677038a5466 100644 --- a/drivers/gpu/drm/i915/selftests/intel_lrc.c +++ b/drivers/gpu/drm/i915/selftests/intel_lrc.c @@ -407,6 +407,168 @@ err_wedged: goto err_client_b; } +static int __i915_sw_fence_call +dummy_notify(struct i915_sw_fence *fence, enum i915_sw_fence_notify state) +{ + return NOTIFY_DONE; +} + +static struct i915_request *dummy_request(struct intel_engine_cs *engine) +{ + struct i915_request *rq; + + rq = kzalloc(sizeof(*rq), GFP_KERNEL); + if (!rq) + return NULL; + + INIT_LIST_HEAD(&rq->active_list); + rq->engine = engine; + + i915_sched_node_init(&rq->sched); + + /* mark this request as permanently incomplete */ + rq->fence.seqno = 1; + BUILD_BUG_ON(sizeof(rq->fence.seqno) != 8); /* upper 32b == 0 */ + rq->hwsp_seqno = (u32 *)&rq->fence.seqno + 1; + GEM_BUG_ON(i915_request_completed(rq)); + + i915_sw_fence_init(&rq->submit, dummy_notify); + i915_sw_fence_commit(&rq->submit); + + return rq; +} + +static void dummy_request_free(struct i915_request *dummy) +{ + i915_request_mark_complete(dummy); + i915_sched_node_fini(&dummy->sched); + i915_sw_fence_fini(&dummy->submit); + + dma_fence_free(&dummy->fence); +} + +static int live_suppress_wait_preempt(void *arg) +{ + struct drm_i915_private *i915 = arg; + struct preempt_client client[4]; + struct intel_engine_cs *engine; + enum intel_engine_id id; + intel_wakeref_t wakeref; + int err = -ENOMEM; + int i; + + /* + * Waiters are given a little priority nudge, but not enough + * to actually cause any preemption. Double check that we do + * not needlessly generate preempt-to-idle cycles. + */ + + if (!HAS_LOGICAL_RING_PREEMPTION(i915)) + return 0; + + mutex_lock(&i915->drm.struct_mutex); + wakeref = intel_runtime_pm_get(i915); + + if (preempt_client_init(i915, &client[0])) /* ELSP[0] */ + goto err_unlock; + if (preempt_client_init(i915, &client[1])) /* ELSP[1] */ + goto err_client_0; + if (preempt_client_init(i915, &client[2])) /* head of queue */ + goto err_client_1; + if (preempt_client_init(i915, &client[3])) /* bystander */ + goto err_client_2; + + for_each_engine(engine, i915, id) { + int depth; + + if (!engine->emit_init_breadcrumb) + continue; + + for (depth = 0; depth < ARRAY_SIZE(client); depth++) { + struct i915_request *rq[ARRAY_SIZE(client)]; + struct i915_request *dummy; + + engine->execlists.preempt_hang.count = 0; + + dummy = dummy_request(engine); + if (!dummy) + goto err_client_3; + + for (i = 0; i < ARRAY_SIZE(client); i++) { + rq[i] = igt_spinner_create_request(&client[i].spin, + client[i].ctx, engine, + MI_NOOP); + if (IS_ERR(rq[i])) { + err = PTR_ERR(rq[i]); + goto err_wedged; + } + + /* Disable NEWCLIENT promotion */ + __i915_active_request_set(&rq[i]->timeline->last_request, + dummy); + i915_request_add(rq[i]); + } + + dummy_request_free(dummy); + + GEM_BUG_ON(i915_request_completed(rq[0])); + if (!igt_wait_for_spinner(&client[0].spin, rq[0])) { + pr_err("%s: First client failed to start\n", + engine->name); + goto err_wedged; + } + GEM_BUG_ON(!i915_request_started(rq[0])); + + if (i915_request_wait(rq[depth], + I915_WAIT_LOCKED | + I915_WAIT_PRIORITY, + 1) != -ETIME) { + pr_err("%s: Waiter depth:%d completed!\n", + engine->name, depth); + goto err_wedged; + } + + for (i = 0; i < ARRAY_SIZE(client); i++) + igt_spinner_end(&client[i].spin); + + if (igt_flush_test(i915, I915_WAIT_LOCKED)) + goto err_wedged; + + if (engine->execlists.preempt_hang.count) { + pr_err("%s: Preemption recorded x%d, depth %d; should have been suppressed!\n", + engine->name, + engine->execlists.preempt_hang.count, + depth); + err = -EINVAL; + goto err_client_3; + } + } + } + + err = 0; +err_client_3: + preempt_client_fini(&client[3]); +err_client_2: + preempt_client_fini(&client[2]); +err_client_1: + preempt_client_fini(&client[1]); +err_client_0: + preempt_client_fini(&client[0]); +err_unlock: + if (igt_flush_test(i915, I915_WAIT_LOCKED)) + err = -EIO; + intel_runtime_pm_put(i915, wakeref); + mutex_unlock(&i915->drm.struct_mutex); + return err; + +err_wedged: + for (i = 0; i < ARRAY_SIZE(client); i++) + igt_spinner_end(&client[i].spin); + i915_gem_set_wedged(i915); + err = -EIO; + goto err_client_3; +} + static int live_chain_preempt(void *arg) { struct drm_i915_private *i915 = arg; @@ -887,6 +1049,7 @@ int intel_execlists_live_selftests(struct drm_i915_private *i915) SUBTEST(live_preempt), SUBTEST(live_late_preempt), SUBTEST(live_suppress_self_preempt), + SUBTEST(live_suppress_wait_preempt), SUBTEST(live_chain_preempt), SUBTEST(live_preempt_hang), SUBTEST(live_preempt_smoke), -- 2.30.2