[PATCH 020/262] drm/i915/execlists: Force preemption via reset on timeout

[Chris Wilson <chris@chris-wilson.co.uk>]

Install a timer when trying to preempt on behalf of an important
context such that if the active context does not honour the preemption
request within the desired timeout, then we reset the GPU to allow the
important context to run.

v2: Install the timer on scheduling the preempt request; long before we
even try to inject preemption into the ELSP, as the tasklet/injection
may itself be blocked.
v3: Update the guc to handle the preemption/tasklet timer.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/intel_engine_cs.c      |  4 +
 drivers/gpu/drm/i915/intel_guc_submission.c |  1 +
 drivers/gpu/drm/i915/intel_lrc.c            | 87 +++++++++++++++++++--
 drivers/gpu/drm/i915/intel_ringbuffer.h     |  8 +-
 drivers/gpu/drm/i915/selftests/intel_lrc.c  | 65 +++++++++++++++
 5 files changed, 157 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
index 7165a3d21443..3ec79bd06b83 100644
--- a/drivers/gpu/drm/i915/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/intel_engine_cs.c
@@ -466,6 +466,9 @@ static void intel_engine_init_execlist(struct intel_engine_cs *engine)
 	execlists->queue_priority = INT_MIN;
 	execlists->queue = RB_ROOT;
 	execlists->first = NULL;
+
+	hrtimer_init(&execlists->preempt_timer,
+		     CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 }
 
 /**
@@ -1047,6 +1050,7 @@ void intel_engines_park(struct drm_i915_private *i915)
 
 	for_each_engine(engine, i915, id) {
 		/* Flush the residual irq tasklets first. */
+		hrtimer_cancel(&engine->execlists.preempt_timer);
 		intel_engine_disarm_breadcrumbs(engine);
 		tasklet_kill(&engine->execlists.tasklet);
 
diff --git a/drivers/gpu/drm/i915/intel_guc_submission.c b/drivers/gpu/drm/i915/intel_guc_submission.c
index d9fcd5db4ea4..48c4ae2be8f1 100644
--- a/drivers/gpu/drm/i915/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/intel_guc_submission.c
@@ -751,6 +751,7 @@ static bool __guc_dequeue(struct intel_engine_cs *engine)
 			kmem_cache_free(engine->i915->priorities, p);
 	}
 done:
+	execlists_clear_active(execlists, EXECLISTS_ACTIVE_PREEMPT_TIMEOUT);
 	execlists->queue_priority = rb ? to_priolist(rb)->priority : INT_MIN;
 	execlists->first = rb;
 	if (submit)
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 37839d89e03a..9b015ec236a3 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -563,6 +563,52 @@ static void inject_preempt_context(struct intel_engine_cs *engine)
 	execlists_set_active(execlists, EXECLISTS_ACTIVE_PREEMPT);
 }
 
+static enum hrtimer_restart preempt_timeout(struct hrtimer *hrtimer)
+{
+	struct intel_engine_execlists *execlists =
+		container_of(hrtimer, typeof(*execlists), preempt_timer);
+
+	GEM_TRACE("%s active=%x\n",
+		  container_of(execlists,
+			       struct intel_engine_cs,
+			       execlists)->name,
+		  execlists->active);
+
+	if (!execlists_is_active(execlists, EXECLISTS_ACTIVE_PREEMPT_TIMEOUT))
+		return HRTIMER_NORESTART;
+
+	if (GEM_SHOW_DEBUG()) {
+		struct intel_engine_cs *engine =
+			container_of(execlists, typeof(*engine), execlists);
+		struct drm_printer p = drm_debug_printer(__func__);
+
+		intel_engine_dump(engine, &p, "%s\n", engine->name);
+	}
+
+	queue_work(system_highpri_wq, &execlists->preempt_reset);
+
+	return HRTIMER_NORESTART;
+}
+
+static void preempt_reset(struct work_struct *work)
+{
+	struct intel_engine_execlists *execlists =
+		container_of(work, typeof(*execlists), preempt_reset);
+	struct intel_engine_cs *engine =
+		  container_of(execlists, struct intel_engine_cs, execlists);
+
+	GEM_TRACE("%s\n", engine->name);
+
+	tasklet_disable(&execlists->tasklet);
+
+	execlists->tasklet.func(execlists->tasklet.data);
+	if (execlists_is_active(execlists, EXECLISTS_ACTIVE_PREEMPT_TIMEOUT))
+		i915_handle_error(engine->i915, BIT(engine->id), 0,
+				  "preemption time out on %s", engine->name);
+
+	tasklet_enable(&execlists->tasklet);
+}
+
 static void complete_preempt_context(struct intel_engine_execlists *execlists)
 {
 	GEM_BUG_ON(!execlists_is_active(execlists, EXECLISTS_ACTIVE_PREEMPT));
@@ -663,7 +709,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 		 * priorities of the ports haven't been switch.
 		 */
 		if (port_count(&port[1]))
-			return;
+			goto clear_preempt_timeout;
 
 		/*
 		 * WaIdleLiteRestore:bdw,skl
@@ -771,6 +817,9 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 	/* We must always keep the beast fed if we have work piled up */
 	GEM_BUG_ON(execlists->first && !port_isset(execlists->port));
 
+clear_preempt_timeout:
+	execlists_clear_active(execlists, EXECLISTS_ACTIVE_PREEMPT_TIMEOUT);
+
 	/* Re-evaluate the executing context setup after each preemptive kick */
 	if (last)
 		execlists_user_begin(execlists, execlists->port);
@@ -1131,15 +1180,38 @@ static void queue_request(struct intel_engine_cs *engine,
 		      &lookup_priolist(engine, prio)->requests);
 }
 
-static void __update_queue(struct intel_engine_cs *engine, int prio)
+static void __update_queue(struct intel_engine_cs *engine,
+			   int prio, unsigned int timeout)
 {
+	struct intel_engine_execlists * const execlists = &engine->execlists;
+
+	GEM_TRACE("%s prio=%d (previous=%d)\n",
+		  engine->name, prio, execlists->queue_priority);
+
+	if (unlikely(execlists_is_active(execlists,
+					 EXECLISTS_ACTIVE_PREEMPT_TIMEOUT)))
+		hrtimer_cancel(&execlists->preempt_timer);
+
+	/* Set a timer to force preemption vs hostile userspace */
+	if (timeout &&
+	    __execlists_need_preempt(prio, execlists->queue_priority)) {
+		GEM_TRACE("%s preempt timeout=%uns\n", engine->name, timeout);
+
+		execlists_set_active(execlists,
+				     EXECLISTS_ACTIVE_PREEMPT_TIMEOUT);
+		hrtimer_start(&execlists->preempt_timer,
+			      ns_to_ktime(timeout),
+			      HRTIMER_MODE_REL);
+	}
+
 	engine->execlists.queue_priority = prio;
 }
 
-static void submit_queue(struct intel_engine_cs *engine, int prio)
+static void submit_queue(struct intel_engine_cs *engine,
+			 int prio, unsigned int timeout)
 {
 	if (prio > engine->execlists.queue_priority) {
-		__update_queue(engine, prio);
+		__update_queue(engine, prio, timeout);
 		if (!intel_engine_uses_guc(engine))
 			execlists_dequeue(engine);
 		else
@@ -1160,7 +1232,7 @@ static void execlists_submit_request(struct i915_request *request)
 	GEM_BUG_ON(!engine->execlists.first);
 	GEM_BUG_ON(list_empty(&request->sched.link));
 
-	submit_queue(engine, rq_prio(request));
+	submit_queue(engine, rq_prio(request), 0);
 
 	spin_unlock_irqrestore(&engine->timeline.lock, flags);
 }
@@ -1289,7 +1361,7 @@ static void execlists_schedule(struct i915_request *request,
 
 		if (prio > engine->execlists.queue_priority &&
 		    i915_sw_fence_done(&sched_to_request(node)->submit)) {
-			__update_queue(engine, prio);
+			__update_queue(engine, prio, 0);
 			tasklet_hi_schedule(&engine->execlists.tasklet);
 		}
 	}
@@ -2345,6 +2417,9 @@ logical_ring_setup(struct intel_engine_cs *engine)
 	tasklet_init(&engine->execlists.tasklet,
 		     execlists_submission_tasklet, (unsigned long)engine);
 
+	INIT_WORK(&engine->execlists.preempt_reset, preempt_reset);
+	engine->execlists.preempt_timer.function = preempt_timeout;
+
 	logical_ring_default_vfuncs(engine);
 	logical_ring_default_irqs(engine);
 }
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 42a136810e15..495cc2628fa9 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -270,8 +270,9 @@ struct intel_engine_execlists {
 	 */
 	unsigned int active;
 #define EXECLISTS_ACTIVE_USER 0
-#define EXECLISTS_ACTIVE_PREEMPT 1
-#define EXECLISTS_ACTIVE_HWACK 2
+#define EXECLISTS_ACTIVE_HWACK 1
+#define EXECLISTS_ACTIVE_PREEMPT 2
+#define EXECLISTS_ACTIVE_PREEMPT_TIMEOUT 3
 
 	/**
 	 * @port_mask: number of execlist ports - 1
@@ -328,6 +329,9 @@ struct intel_engine_execlists {
 	 * @preempt_complete_status: expected CSB upon completing preemption
 	 */
 	u32 preempt_complete_status;
+
+	struct hrtimer preempt_timer;
+	struct work_struct preempt_reset;
 };
 
 #define INTEL_ENGINE_CS_MAX_NAME 8
diff --git a/drivers/gpu/drm/i915/selftests/intel_lrc.c b/drivers/gpu/drm/i915/selftests/intel_lrc.c
index 68cb9126b3e1..7409c582b478 100644
--- a/drivers/gpu/drm/i915/selftests/intel_lrc.c
+++ b/drivers/gpu/drm/i915/selftests/intel_lrc.c
@@ -444,12 +444,77 @@ static int live_late_preempt(void *arg)
 	goto err_ctx_lo;
 }
 
+static void mark_preemption_hang(struct intel_engine_execlists *execlists)
+{
+	execlists_set_active(execlists, EXECLISTS_ACTIVE_PREEMPT);
+	execlists_set_active(execlists, EXECLISTS_ACTIVE_PREEMPT_TIMEOUT);
+}
+
+static int live_preempt_timeout(void *arg)
+{
+	struct drm_i915_private *i915 = arg;
+	struct intel_engine_cs *engine;
+	struct i915_gem_context *ctx;
+	enum intel_engine_id id;
+	struct spinner spin;
+	int err = -ENOMEM;
+
+	if (!HAS_LOGICAL_RING_PREEMPTION(i915))
+		return 0;
+
+	mutex_lock(&i915->drm.struct_mutex);
+
+	if (spinner_init(&spin, i915))
+		goto err_unlock;
+
+	ctx = kernel_context(i915);
+	if (!ctx)
+		goto err_spin;
+
+	for_each_engine(engine, i915, id) {
+		struct i915_request *rq;
+
+		rq = spinner_create_request(&spin, ctx, engine, MI_NOOP);
+		if (IS_ERR(rq)) {
+			err = PTR_ERR(rq);
+			goto err_ctx;
+		}
+
+		i915_request_add(rq);
+		if (!wait_for_spinner(&spin, rq)) {
+			i915_gem_set_wedged(i915);
+			err = -EIO;
+			goto err_ctx;
+		}
+
+		GEM_TRACE("%s triggering reset\n", engine->name);
+		mark_preemption_hang(&engine->execlists);
+		preempt_reset(&engine->execlists.preempt_reset);
+
+		if (igt_flush_test(i915, I915_WAIT_LOCKED)) {
+			err = -EIO;
+			goto err_ctx;
+		}
+	}
+
+	err = 0;
+err_ctx:
+	kernel_context_close(ctx);
+err_spin:
+	spinner_fini(&spin);
+err_unlock:
+	igt_flush_test(i915, I915_WAIT_LOCKED);
+	mutex_unlock(&i915->drm.struct_mutex);
+	return err;
+}
+
 int intel_execlists_live_selftests(struct drm_i915_private *i915)
 {
 	static const struct i915_subtest tests[] = {
 		SUBTEST(live_sanitycheck),
 		SUBTEST(live_preempt),
 		SUBTEST(live_late_preempt),
+		SUBTEST(live_preempt_timeout),
 	};
 
 	if (!HAS_EXECLISTS(i915))
-- 
2.17.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx