drm/i915/gt: Replace hangcheck by heartbeats

Replace sampling the engine state every so often with a periodic
heartbeat request to measure the health of an engine. This is coupled
with the forced-preemption to allow long running requests to survive so
long as they do not block other users.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Jon Bloomfield <jon.bloomfield@intel.com>
Reviewed-by: Jon Bloomfield <jon.bloomfield@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191023133108.21401-5-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/Kconfig.profile

@@ -12,6 +12,17 @@ config DRM_I915_USERFAULT_AUTOSUSPEND
 	  May be 0 to disable the extra delay and solely use the device level
 	  runtime pm autosuspend delay tunable.
 
+config DRM_I915_HEARTBEAT_INTERVAL
+	int "Interval between heartbeat pulses (ms)"
+	default 2500 # milliseconds
+	help
+	  The driver sends a periodic heartbeat down all active engines to
+	  check the health of the GPU and undertake regular house-keeping of
+	  internal driver state.
+
+	  May be 0 to disable heartbeats and therefore disable automatic GPU
+	  hang detection.
+
 config DRM_I915_PREEMPT_TIMEOUT
 	int "Preempt timeout (ms, jiffy granularity)"
 	default 100 # milliseconds

drivers/gpu/drm/i915/Makefile

@@ -87,7 +87,6 @@ gt-y += \
 	gt/intel_gt_pm.o \
 	gt/intel_gt_pm_irq.o \
 	gt/intel_gt_requests.o \
-	gt/intel_hangcheck.o \
 	gt/intel_llc.o \
 	gt/intel_lrc.o \
 	gt/intel_rc6.o \

drivers/gpu/drm/i915/display/intel_display.c

@@ -14863,7 +14863,7 @@ static void intel_plane_unpin_fb(struct intel_plane_state *old_plane_state)
 static void fb_obj_bump_render_priority(struct drm_i915_gem_object *obj)
 {
 	struct i915_sched_attr attr = {
-		.priority = I915_PRIORITY_DISPLAY,
+		.priority = I915_USER_PRIORITY(I915_PRIORITY_DISPLAY),
 	};
 
 	i915_gem_object_wait_priority(obj, 0, &attr);

drivers/gpu/drm/i915/gem/i915_gem_object.h

@@ -462,6 +462,5 @@ int i915_gem_object_wait(struct drm_i915_gem_object *obj,
 int i915_gem_object_wait_priority(struct drm_i915_gem_object *obj,
 				  unsigned int flags,
 				  const struct i915_sched_attr *attr);
-#define I915_PRIORITY_DISPLAY I915_USER_PRIORITY(I915_PRIORITY_MAX)
 
 #endif

drivers/gpu/drm/i915/gem/i915_gem_pm.c

@@ -76,8 +76,6 @@ void i915_gem_suspend(struct drm_i915_private *i915)
 	intel_gt_suspend(&i915->gt);
 	intel_uc_suspend(&i915->gt.uc);
 
-	cancel_delayed_work_sync(&i915->gt.hangcheck.work);
-
 	i915_gem_drain_freed_objects(i915);
 }
 

drivers/gpu/drm/i915/gt/intel_engine.h

@@ -91,38 +91,6 @@ struct intel_gt;
 /* seqno size is actually only a uint32, but since we plan to use MI_FLUSH_DW to
  * do the writes, and that must have qw aligned offsets, simply pretend it's 8b.
  */
-enum intel_engine_hangcheck_action {
-	ENGINE_IDLE = 0,
-	ENGINE_WAIT,
-	ENGINE_ACTIVE_SEQNO,
-	ENGINE_ACTIVE_HEAD,
-	ENGINE_ACTIVE_SUBUNITS,
-	ENGINE_WAIT_KICK,
-	ENGINE_DEAD,
-};
-
-static inline const char *
-hangcheck_action_to_str(const enum intel_engine_hangcheck_action a)
-{
-	switch (a) {
-	case ENGINE_IDLE:
-		return "idle";
-	case ENGINE_WAIT:
-		return "wait";
-	case ENGINE_ACTIVE_SEQNO:
-		return "active seqno";
-	case ENGINE_ACTIVE_HEAD:
-		return "active head";
-	case ENGINE_ACTIVE_SUBUNITS:
-		return "active subunits";
-	case ENGINE_WAIT_KICK:
-		return "wait kick";
-	case ENGINE_DEAD:
-		return "dead";
-	}
-
-	return "unknown";
-}
 
 static inline unsigned int
 execlists_num_ports(const struct intel_engine_execlists * const execlists)

drivers/gpu/drm/i915/gt/intel_engine_cs.c

@@ -308,6 +308,8 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id)
 	engine->instance = info->instance;
 	__sprint_engine_name(engine);
 
+	engine->props.heartbeat_interval_ms =
+		CONFIG_DRM_I915_HEARTBEAT_INTERVAL;
 	engine->props.preempt_timeout_ms =
 		CONFIG_DRM_I915_PREEMPT_TIMEOUT;
 	engine->props.stop_timeout_ms =
@@ -609,7 +611,6 @@ static int intel_engine_setup_common(struct intel_engine_cs *engine)
 	intel_engine_init_active(engine, ENGINE_PHYSICAL);
 	intel_engine_init_breadcrumbs(engine);
 	intel_engine_init_execlists(engine);
-	intel_engine_init_hangcheck(engine);
 	intel_engine_init_cmd_parser(engine);
 	intel_engine_init__pm(engine);
 
@@ -1470,8 +1471,13 @@ void intel_engine_dump(struct intel_engine_cs *engine,
 		drm_printf(m, "*** WEDGED ***\n");
 
 	drm_printf(m, "\tAwake? %d\n", atomic_read(&engine->wakeref.count));
-	drm_printf(m, "\tHangcheck: %d ms ago\n",
-		   jiffies_to_msecs(jiffies - engine->hangcheck.action_timestamp));
+
+	rcu_read_lock();
+	rq = READ_ONCE(engine->heartbeat.systole);
+	if (rq)
+		drm_printf(m, "\tHeartbeat: %d ms ago\n",
+			   jiffies_to_msecs(jiffies - rq->emitted_jiffies));
+	rcu_read_unlock();
 	drm_printf(m, "\tReset count: %d (global %d)\n",
 		   i915_reset_engine_count(error, engine),
 		   i915_reset_count(error));

drivers/gpu/drm/i915/gt/intel_engine_heartbeat.c

@@ -11,6 +11,30 @@
 #include "intel_engine_pm.h"
 #include "intel_engine.h"
 #include "intel_gt.h"
+#include "intel_reset.h"
+
+/*
+ * While the engine is active, we send a periodic pulse along the engine
+ * to check on its health and to flush any idle-barriers. If that request
+ * is stuck, and we fail to preempt it, we declare the engine hung and
+ * issue a reset -- in the hope that restores progress.
+ */
+
+static bool next_heartbeat(struct intel_engine_cs *engine)
+{
+	long delay;
+
+	delay = READ_ONCE(engine->props.heartbeat_interval_ms);
+	if (!delay)
+		return false;
+
+	delay = msecs_to_jiffies_timeout(delay);
+	if (delay >= HZ)
+		delay = round_jiffies_up_relative(delay);
+	schedule_delayed_work(&engine->heartbeat.work, delay);
+
+	return true;
+}
 
 static void idle_pulse(struct intel_engine_cs *engine, struct i915_request *rq)
 {
@@ -18,6 +42,139 @@ static void idle_pulse(struct intel_engine_cs *engine, struct i915_request *rq)
 	i915_request_add_active_barriers(rq);
 }
 
+static void show_heartbeat(const struct i915_request *rq,
+			   struct intel_engine_cs *engine)
+{
+	struct drm_printer p = drm_debug_printer("heartbeat");
+
+	intel_engine_dump(engine, &p,
+			  "%s heartbeat {prio:%d} not ticking\n",
+			  engine->name,
+			  rq->sched.attr.priority);
+}
+
+static void heartbeat(struct work_struct *wrk)
+{
+	struct i915_sched_attr attr = {
+		.priority = I915_USER_PRIORITY(I915_PRIORITY_MIN),
+	};
+	struct intel_engine_cs *engine =
+		container_of(wrk, typeof(*engine), heartbeat.work.work);
+	struct intel_context *ce = engine->kernel_context;
+	struct i915_request *rq;
+
+	if (!intel_engine_pm_get_if_awake(engine))
+		return;
+
+	rq = engine->heartbeat.systole;
+	if (rq && i915_request_completed(rq)) {
+		i915_request_put(rq);
+		engine->heartbeat.systole = NULL;
+	}
+
+	if (intel_gt_is_wedged(engine->gt))
+		goto out;
+
+	if (engine->heartbeat.systole) {
+		if (engine->schedule &&
+		    rq->sched.attr.priority < I915_PRIORITY_BARRIER) {
+			/*
+			 * Gradually raise the priority of the heartbeat to
+			 * give high priority work [which presumably desires
+			 * low latency and no jitter] the chance to naturally
+			 * complete before being preempted.
+			 */
+			attr.priority = I915_PRIORITY_MASK;
+			if (rq->sched.attr.priority >= attr.priority)
+				attr.priority |= I915_USER_PRIORITY(I915_PRIORITY_HEARTBEAT);
+			if (rq->sched.attr.priority >= attr.priority)
+				attr.priority = I915_PRIORITY_BARRIER;
+
+			local_bh_disable();
+			engine->schedule(rq, &attr);
+			local_bh_enable();
+		} else {
+			if (IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM))
+				show_heartbeat(rq, engine);
+
+			intel_gt_handle_error(engine->gt, engine->mask,
+					      I915_ERROR_CAPTURE,
+					      "stopped heartbeat on %s",
+					      engine->name);
+		}
+		goto out;
+	}
+
+	if (engine->wakeref_serial == engine->serial)
+		goto out;
+
+	mutex_lock(&ce->timeline->mutex);
+
+	intel_context_enter(ce);
+	rq = __i915_request_create(ce, GFP_NOWAIT | __GFP_NOWARN);
+	intel_context_exit(ce);
+	if (IS_ERR(rq))
+		goto unlock;
+
+	idle_pulse(engine, rq);
+	if (i915_modparams.enable_hangcheck)
+		engine->heartbeat.systole = i915_request_get(rq);
+
+	__i915_request_commit(rq);
+	__i915_request_queue(rq, &attr);
+
+unlock:
+	mutex_unlock(&ce->timeline->mutex);
+out:
+	if (!next_heartbeat(engine))
+		i915_request_put(fetch_and_zero(&engine->heartbeat.systole));
+	intel_engine_pm_put(engine);
+}
+
+void intel_engine_unpark_heartbeat(struct intel_engine_cs *engine)
+{
+	if (!CONFIG_DRM_I915_HEARTBEAT_INTERVAL)
+		return;
+
+	next_heartbeat(engine);
+}
+
+void intel_engine_park_heartbeat(struct intel_engine_cs *engine)
+{
+	cancel_delayed_work(&engine->heartbeat.work);
+	i915_request_put(fetch_and_zero(&engine->heartbeat.systole));
+}
+
+void intel_engine_init_heartbeat(struct intel_engine_cs *engine)
+{
+	INIT_DELAYED_WORK(&engine->heartbeat.work, heartbeat);
+}
+
+int intel_engine_set_heartbeat(struct intel_engine_cs *engine,
+			       unsigned long delay)
+{
+	int err;
+
+	/* Send one last pulse before to cleanup persistent hogs */
+	if (!delay && CONFIG_DRM_I915_PREEMPT_TIMEOUT) {
+		err = intel_engine_pulse(engine);
+		if (err)
+			return err;
+	}
+
+	WRITE_ONCE(engine->props.heartbeat_interval_ms, delay);
+
+	if (intel_engine_pm_get_if_awake(engine)) {
+		if (delay)
+			intel_engine_unpark_heartbeat(engine);
+		else
+			intel_engine_park_heartbeat(engine);
+		intel_engine_pm_put(engine);
+	}
+
+	return 0;
+}
+
 int intel_engine_pulse(struct intel_engine_cs *engine)
 {
 	struct i915_sched_attr attr = { .priority = I915_PRIORITY_BARRIER };

drivers/gpu/drm/i915/gt/intel_engine_heartbeat.h

@@ -9,6 +9,14 @@
 
 struct intel_engine_cs;
 
+void intel_engine_init_heartbeat(struct intel_engine_cs *engine);
+
+int intel_engine_set_heartbeat(struct intel_engine_cs *engine,
+			       unsigned long delay);
+
+void intel_engine_park_heartbeat(struct intel_engine_cs *engine);
+void intel_engine_unpark_heartbeat(struct intel_engine_cs *engine);
+
 int intel_engine_pulse(struct intel_engine_cs *engine);
 int intel_engine_flush_barriers(struct intel_engine_cs *engine);
 

drivers/gpu/drm/i915/gt/intel_engine_pm.c

@@ -7,6 +7,7 @@
 #include "i915_drv.h"
 
 #include "intel_engine.h"
+#include "intel_engine_heartbeat.h"
 #include "intel_engine_pm.h"
 #include "intel_engine_pool.h"
 #include "intel_gt.h"
@@ -34,7 +35,7 @@ static int __engine_unpark(struct intel_wakeref *wf)
 	if (engine->unpark)
 		engine->unpark(engine);
 
-	intel_engine_init_hangcheck(engine);
+	intel_engine_unpark_heartbeat(engine);
 	return 0;
 }
 
@@ -158,6 +159,7 @@ static int __engine_park(struct intel_wakeref *wf)
 
 	call_idle_barriers(engine); /* cleanup after wedging */
 
+	intel_engine_park_heartbeat(engine);
 	intel_engine_disarm_breadcrumbs(engine);
 	intel_engine_pool_park(&engine->pool);
 
@@ -188,6 +190,7 @@ void intel_engine_init__pm(struct intel_engine_cs *engine)
 	struct intel_runtime_pm *rpm = engine->uncore->rpm;
 
 	intel_wakeref_init(&engine->wakeref, rpm, &wf_ops);
+	intel_engine_init_heartbeat(engine);
 }
 
 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)

drivers/gpu/drm/i915/gt/intel_engine_types.h

@@ -15,6 +15,7 @@
 #include <linux/rbtree.h>
 #include <linux/timer.h>
 #include <linux/types.h>
+#include <linux/workqueue.h>
 
 #include "i915_gem.h"
 #include "i915_pmu.h"
@@ -76,14 +77,6 @@ struct intel_instdone {
 	u32 row[I915_MAX_SLICES][I915_MAX_SUBSLICES];
 };
 
-struct intel_engine_hangcheck {
-	u64 acthd;
-	u32 last_ring;
-	u32 last_head;
-	unsigned long action_timestamp;
-	struct intel_instdone instdone;
-};
-
 struct intel_ring {
 	struct kref ref;
 	struct i915_vma *vma;
@@ -331,6 +324,11 @@ struct intel_engine_cs {
 
 	intel_engine_mask_t saturated; /* submitting semaphores too late? */
 
+	struct {
+		struct delayed_work work;
+		struct i915_request *systole;
+	} heartbeat;
+
 	unsigned long serial;
 
 	unsigned long wakeref_serial;
@@ -481,8 +479,6 @@ struct intel_engine_cs {
 	/* status_notifier: list of callbacks for context-switch changes */
 	struct atomic_notifier_head context_status_notifier;
 
-	struct intel_engine_hangcheck hangcheck;
-
 #define I915_ENGINE_NEEDS_CMD_PARSER BIT(0)
 #define I915_ENGINE_SUPPORTS_STATS   BIT(1)
 #define I915_ENGINE_HAS_PREEMPTION   BIT(2)
@@ -549,6 +545,7 @@ struct intel_engine_cs {
 	} stats;
 
 	struct {
+		unsigned long heartbeat_interval_ms;
 		unsigned long preempt_timeout_ms;
 		unsigned long stop_timeout_ms;
 	} props;

drivers/gpu/drm/i915/gt/intel_gt.c

@@ -22,7 +22,6 @@ void intel_gt_init_early(struct intel_gt *gt, struct drm_i915_private *i915)
 	INIT_LIST_HEAD(&gt->closed_vma);
 	spin_lock_init(&gt->closed_lock);
 
-	intel_gt_init_hangcheck(gt);
 	intel_gt_init_reset(gt);
 	intel_gt_init_requests(gt);
 	intel_gt_pm_init_early(gt);

drivers/gpu/drm/i915/gt/intel_gt.h

@@ -46,8 +46,6 @@ void intel_gt_clear_error_registers(struct intel_gt *gt,
 void intel_gt_flush_ggtt_writes(struct intel_gt *gt);
 void intel_gt_chipset_flush(struct intel_gt *gt);
 
-void intel_gt_init_hangcheck(struct intel_gt *gt);
-
 static inline u32 intel_gt_scratch_offset(const struct intel_gt *gt,
 					  enum intel_gt_scratch_field field)
 {
@@ -59,6 +57,4 @@ static inline bool intel_gt_is_wedged(struct intel_gt *gt)
 	return __intel_reset_failed(&gt->reset);
 }
 
-void intel_gt_queue_hangcheck(struct intel_gt *gt);
-
 #endif /* __INTEL_GT_H__ */

drivers/gpu/drm/i915/gt/intel_gt_pm.c

@@ -47,7 +47,6 @@ static int __gt_unpark(struct intel_wakeref *wf)
 
 	i915_pmu_gt_unparked(i915);
 
-	intel_gt_queue_hangcheck(gt);
 	intel_gt_unpark_requests(gt);
 
 	return 0;

drivers/gpu/drm/i915/gt/intel_gt_types.h

@@ -27,14 +27,6 @@ struct i915_ggtt;
 struct intel_engine_cs;
 struct intel_uncore;
 
-struct intel_hangcheck {
-	/* For hangcheck timer */
-#define DRM_I915_HANGCHECK_PERIOD 1500 /* in ms */
-#define DRM_I915_HANGCHECK_JIFFIES msecs_to_jiffies(DRM_I915_HANGCHECK_PERIOD)
-
-	struct delayed_work work;
-};
-
 struct intel_gt {
 	struct drm_i915_private *i915;
 	struct intel_uncore *uncore;
@@ -68,7 +60,6 @@ struct intel_gt {
 	struct list_head closed_vma;
 	spinlock_t closed_lock; /* guards the list of closed_vma */
 
-	struct intel_hangcheck hangcheck;
 	struct intel_reset reset;
 
 	/**

drivers/gpu/drm/i915/gt/intel_reset.c

@@ -1024,8 +1024,6 @@ void intel_gt_reset(struct intel_gt *gt,
 	if (ret)
 		goto taint;
 
-	intel_gt_queue_hangcheck(gt);
-
 finish:
 	reset_finish(gt, awake);
 unlock:
@@ -1353,4 +1351,5 @@ void __intel_fini_wedge(struct intel_wedge_me *w)
 
 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
 #include "selftest_reset.c"
+#include "selftest_hangcheck.c"
 #endif

drivers/gpu/drm/i915/gt/selftest_engine_heartbeat.c

@@ -4,6 +4,8 @@
  * Copyright © 2018 Intel Corporation
  */
 
+#include <linux/sort.h>
+
 #include "i915_drv.h"
 
 #include "intel_gt_requests.h"
@@ -150,11 +152,180 @@ static int live_idle_pulse(void *arg)
 	return err;
 }
 
+static int cmp_u32(const void *_a, const void *_b)
+{
+	const u32 *a = _a, *b = _b;
+
+	return *a - *b;
+}
+
+static int __live_heartbeat_fast(struct intel_engine_cs *engine)
+{
+	struct intel_context *ce;
+	struct i915_request *rq;
+	ktime_t t0, t1;
+	u32 times[5];
+	int err;
+	int i;
+
+	ce = intel_context_create(engine->kernel_context->gem_context,
+				  engine);
+	if (IS_ERR(ce))
+		return PTR_ERR(ce);
+
+	intel_engine_pm_get(engine);
+
+	err = intel_engine_set_heartbeat(engine, 1);
+	if (err)
+		goto err_pm;
+
+	for (i = 0; i < ARRAY_SIZE(times); i++) {
+		/* Manufacture a tick */
+		do {
+			while (READ_ONCE(engine->heartbeat.systole))
+				flush_delayed_work(&engine->heartbeat.work);
+
+			engine->serial++; /* quick, pretend we are not idle! */
+			flush_delayed_work(&engine->heartbeat.work);
+			if (!delayed_work_pending(&engine->heartbeat.work)) {
+				pr_err("%s: heartbeat did not start\n",
+				       engine->name);
+				err = -EINVAL;
+				goto err_pm;
+			}
+
+			rcu_read_lock();
+			rq = READ_ONCE(engine->heartbeat.systole);
+			if (rq)
+				rq = i915_request_get_rcu(rq);
+			rcu_read_unlock();
+		} while (!rq);
+
+		t0 = ktime_get();
+		while (rq == READ_ONCE(engine->heartbeat.systole))
+			yield(); /* work is on the local cpu! */
+		t1 = ktime_get();
+
+		i915_request_put(rq);
+		times[i] = ktime_us_delta(t1, t0);
+	}
+
+	sort(times, ARRAY_SIZE(times), sizeof(times[0]), cmp_u32, NULL);
+
+	pr_info("%s: Heartbeat delay: %uus [%u, %u]\n",
+		engine->name,
+		times[ARRAY_SIZE(times) / 2],
+		times[0],
+		times[ARRAY_SIZE(times) - 1]);
+
+	/* Min work delay is 2 * 2 (worst), +1 for scheduling, +1 for slack */
+	if (times[ARRAY_SIZE(times) / 2] > jiffies_to_usecs(6)) {
+		pr_err("%s: Heartbeat delay was %uus, expected less than %dus\n",
+		       engine->name,
+		       times[ARRAY_SIZE(times) / 2],
+		       jiffies_to_usecs(6));
+		err = -EINVAL;
+	}
+
+	intel_engine_set_heartbeat(engine, CONFIG_DRM_I915_HEARTBEAT_INTERVAL);
+err_pm:
+	intel_engine_pm_put(engine);
+	intel_context_put(ce);
+	return err;
+}
+
+static int live_heartbeat_fast(void *arg)
+{
+	struct intel_gt *gt = arg;
+	struct intel_engine_cs *engine;
+	enum intel_engine_id id;
+	int err;
+
+	/* Check that the heartbeat ticks at the desired rate. */
+	if (!CONFIG_DRM_I915_HEARTBEAT_INTERVAL)
+		return 0;
+
+	for_each_engine(engine, gt, id) {
+		err = __live_heartbeat_fast(engine);
+		if (err)
+			break;
+	}
+
+	return err;
+}
+
+static int __live_heartbeat_off(struct intel_engine_cs *engine)
+{
+	int err;
+
+	intel_engine_pm_get(engine);
+
+	engine->serial++;
+	flush_delayed_work(&engine->heartbeat.work);
+	if (!delayed_work_pending(&engine->heartbeat.work)) {
+		pr_err("%s: heartbeat not running\n",
+		       engine->name);
+		err = -EINVAL;
+		goto err_pm;
+	}
+
+	err = intel_engine_set_heartbeat(engine, 0);
+	if (err)
+		goto err_pm;
+
+	engine->serial++;
+	flush_delayed_work(&engine->heartbeat.work);
+	if (delayed_work_pending(&engine->heartbeat.work)) {
+		pr_err("%s: heartbeat still running\n",
+		       engine->name);
+		err = -EINVAL;
+		goto err_beat;
+	}
+
+	if (READ_ONCE(engine->heartbeat.systole)) {
+		pr_err("%s: heartbeat still allocated\n",
+		       engine->name);
+		err = -EINVAL;
+		goto err_beat;
+	}
+
+err_beat:
+	intel_engine_set_heartbeat(engine, CONFIG_DRM_I915_HEARTBEAT_INTERVAL);
+err_pm:
+	intel_engine_pm_put(engine);
+	return err;
+}
+
+static int live_heartbeat_off(void *arg)
+{
+	struct intel_gt *gt = arg;
+	struct intel_engine_cs *engine;
+	enum intel_engine_id id;
+	int err;
+
+	/* Check that we can turn off heartbeat and not interrupt VIP */
+	if (!CONFIG_DRM_I915_HEARTBEAT_INTERVAL)
+		return 0;
+
+	for_each_engine(engine, gt, id) {
+		if (!intel_engine_has_preemption(engine))
+			continue;
+
+		err = __live_heartbeat_off(engine);
+		if (err)
+			break;
+	}
+
+	return err;
+}
+
 int intel_heartbeat_live_selftests(struct drm_i915_private *i915)
 {
 	static const struct i915_subtest tests[] = {
 		SUBTEST(live_idle_flush),
 		SUBTEST(live_idle_pulse),
+		SUBTEST(live_heartbeat_fast),
+		SUBTEST(live_heartbeat_off),
 	};
 	int saved_hangcheck;
 	int err;

drivers/gpu/drm/i915/gt/selftest_hangcheck.c

@@ -1686,7 +1686,6 @@ int intel_hangcheck_live_selftests(struct drm_i915_private *i915)
 	};
 	struct intel_gt *gt = &i915->gt;
 	intel_wakeref_t wakeref;
-	bool saved_hangcheck;
 	int err;
 
 	if (!intel_has_gpu_reset(gt))
@@ -1696,12 +1695,9 @@ int intel_hangcheck_live_selftests(struct drm_i915_private *i915)
 		return -EIO; /* we're long past hope of a successful reset */
 
 	wakeref = intel_runtime_pm_get(gt->uncore->rpm);
-	saved_hangcheck = fetch_and_zero(&i915_modparams.enable_hangcheck);
-	drain_delayed_work(&gt->hangcheck.work); /* flush param */
 
 	err = intel_gt_live_subtests(tests, gt);
 
-	i915_modparams.enable_hangcheck = saved_hangcheck;
 	intel_runtime_pm_put(gt->uncore->rpm, wakeref);
 
 	return err;

drivers/gpu/drm/i915/i915_debugfs.c

@@ -1011,92 +1011,6 @@ static int i915_frequency_info(struct seq_file *m, void *unused)
 	return ret;
 }
 
-static void i915_instdone_info(struct drm_i915_private *dev_priv,
-			       struct seq_file *m,
-			       struct intel_instdone *instdone)
-{
-	const struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu;
-	int slice;
-	int subslice;
-
-	seq_printf(m, "\t\tINSTDONE: 0x%08x\n",
-		   instdone->instdone);
-
-	if (INTEL_GEN(dev_priv) <= 3)
-		return;
-
-	seq_printf(m, "\t\tSC_INSTDONE: 0x%08x\n",
-		   instdone->slice_common);
-
-	if (INTEL_GEN(dev_priv) <= 6)
-		return;
-
-	for_each_instdone_slice_subslice(dev_priv, sseu, slice, subslice)
-		seq_printf(m, "\t\tSAMPLER_INSTDONE[%d][%d]: 0x%08x\n",
-			   slice, subslice, instdone->sampler[slice][subslice]);
-
-	for_each_instdone_slice_subslice(dev_priv, sseu, slice, subslice)
-		seq_printf(m, "\t\tROW_INSTDONE[%d][%d]: 0x%08x\n",
-			   slice, subslice, instdone->row[slice][subslice]);
-}
-
-static int i915_hangcheck_info(struct seq_file *m, void *unused)
-{
-	struct drm_i915_private *i915 = node_to_i915(m->private);
-	struct intel_gt *gt = &i915->gt;
-	struct intel_engine_cs *engine;
-	intel_wakeref_t wakeref;
-	enum intel_engine_id id;
-
-	seq_printf(m, "Reset flags: %lx\n", gt->reset.flags);
-	if (test_bit(I915_WEDGED, &gt->reset.flags))
-		seq_puts(m, "\tWedged\n");
-	if (test_bit(I915_RESET_BACKOFF, &gt->reset.flags))
-		seq_puts(m, "\tDevice (global) reset in progress\n");
-
-	if (!i915_modparams.enable_hangcheck) {
-		seq_puts(m, "Hangcheck disabled\n");
-		return 0;
-	}
-
-	if (timer_pending(&gt->hangcheck.work.timer))
-		seq_printf(m, "Hangcheck active, timer fires in %dms\n",
-			   jiffies_to_msecs(gt->hangcheck.work.timer.expires -
-					    jiffies));
-	else if (delayed_work_pending(&gt->hangcheck.work))
-		seq_puts(m, "Hangcheck active, work pending\n");
-	else
-		seq_puts(m, "Hangcheck inactive\n");
-
-	seq_printf(m, "GT active? %s\n", yesno(gt->awake));
-
-	with_intel_runtime_pm(&i915->runtime_pm, wakeref) {
-		for_each_engine(engine, i915, id) {
-			struct intel_instdone instdone;
-
-			seq_printf(m, "%s: %d ms ago\n",
-				   engine->name,
-				   jiffies_to_msecs(jiffies -
-						    engine->hangcheck.action_timestamp));
-
-			seq_printf(m, "\tACTHD = 0x%08llx [current 0x%08llx]\n",
-				   (long long)engine->hangcheck.acthd,
-				   intel_engine_get_active_head(engine));
-
-			intel_engine_get_instdone(engine, &instdone);
-
-			seq_puts(m, "\tinstdone read =\n");
-			i915_instdone_info(i915, m, &instdone);
-
-			seq_puts(m, "\tinstdone accu =\n");
-			i915_instdone_info(i915, m,
-					   &engine->hangcheck.instdone);
-		}
-	}
-
-	return 0;
-}
-
 static int ironlake_drpc_info(struct seq_file *m)
 {
 	struct drm_i915_private *i915 = node_to_i915(m->private);
@@ -4363,7 +4277,6 @@ static const struct drm_info_list i915_debugfs_list[] = {
 	{"i915_guc_stage_pool", i915_guc_stage_pool, 0},
 	{"i915_huc_load_status", i915_huc_load_status_info, 0},
 	{"i915_frequency_info", i915_frequency_info, 0},
-	{"i915_hangcheck_info", i915_hangcheck_info, 0},
 	{"i915_drpc_info", i915_drpc_info, 0},
 	{"i915_ring_freq_table", i915_ring_freq_table, 0},
 	{"i915_frontbuffer_tracking", i915_frontbuffer_tracking, 0},

drivers/gpu/drm/i915/i915_drv.c

@@ -1548,10 +1548,7 @@ void i915_driver_remove(struct drm_i915_private *i915)
 
 	i915_driver_modeset_remove(i915);
 
-	/* Free error state after interrupts are fully disabled. */
-	cancel_delayed_work_sync(&i915->gt.hangcheck.work);
 	i915_reset_error_state(i915);
-
 	i915_gem_driver_remove(i915);
 
 	intel_power_domains_driver_remove(i915);

drivers/gpu/drm/i915/i915_drv.h

@@ -1846,7 +1846,6 @@ void i915_driver_remove(struct drm_i915_private *i915);
 int i915_resume_switcheroo(struct drm_i915_private *i915);
 int i915_suspend_switcheroo(struct drm_i915_private *i915, pm_message_t state);
 
-void intel_engine_init_hangcheck(struct intel_engine_cs *engine);
 int vlv_force_gfx_clock(struct drm_i915_private *dev_priv, bool on);
 
 static inline bool intel_gvt_active(struct drm_i915_private *dev_priv)

drivers/gpu/drm/i915/i915_gpu_error.c

@@ -534,10 +534,6 @@ static void error_print_engine(struct drm_i915_error_state_buf *m,
 	}
 	err_printf(m, "  ring->head: 0x%08x\n", ee->cpu_ring_head);
 	err_printf(m, "  ring->tail: 0x%08x\n", ee->cpu_ring_tail);
-	err_printf(m, "  hangcheck timestamp: %dms (%lu%s)\n",
-		   jiffies_to_msecs(ee->hangcheck_timestamp - epoch),
-		   ee->hangcheck_timestamp,
-		   ee->hangcheck_timestamp == epoch ? "; epoch" : "");
 	err_printf(m, "  engine reset count: %u\n", ee->reset_count);
 
 	for (n = 0; n < ee->num_ports; n++) {
@@ -679,11 +675,8 @@ static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
 	ts = ktime_to_timespec64(error->uptime);
 	err_printf(m, "Uptime: %lld s %ld us\n",
 		   (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
-	err_printf(m, "Epoch: %lu jiffies (%u HZ)\n", error->epoch, HZ);
-	err_printf(m, "Capture: %lu jiffies; %d ms ago, %d ms after epoch\n",
-		   error->capture,
-		   jiffies_to_msecs(jiffies - error->capture),
-		   jiffies_to_msecs(error->capture - error->epoch));
+	err_printf(m, "Capture: %lu jiffies; %d ms ago\n",
+		   error->capture, jiffies_to_msecs(jiffies - error->capture));
 
 	for (ee = error->engine; ee; ee = ee->next)
 		err_printf(m, "Active process (on ring %s): %s [%d]\n",
@@ -742,7 +735,7 @@ static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
 		err_printf(m, "GTT_CACHE_EN: 0x%08x\n", error->gtt_cache);
 
 	for (ee = error->engine; ee; ee = ee->next)
-		error_print_engine(m, ee, error->epoch);
+		error_print_engine(m, ee, error->capture);
 
 	for (ee = error->engine; ee; ee = ee->next) {
 		const struct drm_i915_error_object *obj;
@@ -770,7 +763,7 @@ static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
 			for (j = 0; j < ee->num_requests; j++)
 				error_print_request(m, " ",
 						    &ee->requests[j],
-						    error->epoch);
+						    error->capture);
 		}
 
 		print_error_obj(m, ee->engine, "ringbuffer", ee->ringbuffer);
@@ -1144,8 +1137,6 @@ static void error_record_engine_registers(struct i915_gpu_state *error,
 	}
 
 	ee->idle = intel_engine_is_idle(engine);
-	if (!ee->idle)
-		ee->hangcheck_timestamp = engine->hangcheck.action_timestamp;
 	ee->reset_count = i915_reset_engine_count(&dev_priv->gpu_error,
 						  engine);
 
@@ -1657,20 +1648,6 @@ static void capture_params(struct i915_gpu_state *error)
 	i915_params_copy(&error->params, &i915_modparams);
 }
 
-static unsigned long capture_find_epoch(const struct i915_gpu_state *error)
-{
-	const struct drm_i915_error_engine *ee;
-	unsigned long epoch = error->capture;
-
-	for (ee = error->engine; ee; ee = ee->next) {
-		if (ee->hangcheck_timestamp &&
-		    time_before(ee->hangcheck_timestamp, epoch))
-			epoch = ee->hangcheck_timestamp;
-	}
-
-	return epoch;
-}
-
 static void capture_finish(struct i915_gpu_state *error)
 {
 	struct i915_ggtt *ggtt = &error->i915->ggtt;
@@ -1722,8 +1699,6 @@ i915_capture_gpu_state(struct drm_i915_private *i915)
 	error->overlay = intel_overlay_capture_error_state(i915);
 	error->display = intel_display_capture_error_state(i915);
 
-	error->epoch = capture_find_epoch(error);
-
 	capture_finish(error);
 	compress_fini(&compress);
 

drivers/gpu/drm/i915/i915_gpu_error.h

@@ -34,7 +34,6 @@ struct i915_gpu_state {
 	ktime_t boottime;
 	ktime_t uptime;
 	unsigned long capture;
-	unsigned long epoch;
 
 	struct drm_i915_private *i915;
 
@@ -86,7 +85,6 @@ struct i915_gpu_state {
 
 		/* Software tracked state */
 		bool idle;
-		unsigned long hangcheck_timestamp;
 		int num_requests;
 		u32 reset_count;
 

drivers/gpu/drm/i915/i915_priolist_types.h

@@ -16,6 +16,12 @@ enum {
 	I915_PRIORITY_MIN = I915_CONTEXT_MIN_USER_PRIORITY - 1,
 	I915_PRIORITY_NORMAL = I915_CONTEXT_DEFAULT_PRIORITY,
 	I915_PRIORITY_MAX = I915_CONTEXT_MAX_USER_PRIORITY + 1,
+
+	/* A preemptive pulse used to monitor the health of each engine */
+	I915_PRIORITY_HEARTBEAT,
+
+	/* Interactive workload, scheduled for immediate pageflipping */
+	I915_PRIORITY_DISPLAY,
 };
 
 #define I915_USER_PRIORITY_SHIFT 2