i915/gt/intel_engine_heartbeat.c

17  * While the engine is active, we send a periodic pulse along the engine
19  * is stuck, and we fail to preempt it, we declare the engine hung and
23 static bool next_heartbeat(struct intel_engine_cs *engine)
28 	delay = READ_ONCE(engine->props.heartbeat_interval_ms);
30 	rq = engine->heartbeat.systole;
42 	    delay == engine->defaults.heartbeat_interval_ms) {
50 		longer = READ_ONCE(engine->props.preempt_timeout_ms) * 2;
51 		longer = intel_clamp_heartbeat_interval_ms(engine, longer);
62 	mod_delayed_work(system_highpri_wq, &engine->heartbeat.work, delay + 1);
79 static void idle_pulse(struct intel_engine_cs *engine, struct i915_request *rq)
81 	engine->wakeref_serial = READ_ONCE(engine->serial) + 1;
83 	if (!engine->heartbeat.systole && intel_engine_has_heartbeat(engine))
84 		engine->heartbeat.systole = i915_request_get(rq);
90 	idle_pulse(rq->engine, rq);
97 			   struct intel_engine_cs *engine)
100 		drm_dbg_printer(&engine->i915->drm, DRM_UT_DRIVER, "heartbeat");
103 		intel_engine_dump(engine, &p,
105 				  engine->name);
107 		intel_engine_dump(engine, &p,
109 				  engine->name,
117 reset_engine(struct intel_engine_cs *engine, struct i915_request *rq)
120 		show_heartbeat(rq, engine);
122 	if (intel_engine_uses_guc(engine))
128 		intel_guc_find_hung_context(engine);
130 	intel_gt_handle_error(engine->gt, engine->mask,
133 			      engine->name);
139 	struct intel_engine_cs *engine =
140 		container_of(wrk, typeof(*engine), heartbeat.work.work);
141 	struct intel_context *ce = engine->kernel_context;
146 	intel_engine_flush_submission(engine);
148 	rq = engine->heartbeat.systole;
151 		engine->heartbeat.systole = NULL;
154 	if (!intel_engine_pm_get_if_awake(engine))
157 	if (intel_gt_is_wedged(engine->gt))
160 	if (i915_sched_engine_disabled(engine->sched_engine)) {
161 		reset_engine(engine, engine->heartbeat.systole);
165 	if (engine->heartbeat.systole) {
166 		long delay = READ_ONCE(engine->props.heartbeat_interval_ms);
184 		} else if (engine->sched_engine->schedule &&
199 			engine->sched_engine->schedule(rq, &attr);
202 			reset_engine(engine, rq);
209 	serial = READ_ONCE(engine->serial);
210 	if (engine->wakeref_serial == serial)
214 		/* Unable to lock the kernel timeline, is the engine stuck? */
215 		if (xchg(&engine->heartbeat.blocked, serial) == serial)
216 			intel_gt_handle_error(engine->gt, engine->mask,
219 					      engine->name);
232 	if (!engine->i915->params.enable_hangcheck || !next_heartbeat(engine))
233 		i915_request_put(fetch_and_zero(&engine->heartbeat.systole));
234 	intel_engine_pm_put(engine);
237 void intel_engine_unpark_heartbeat(struct intel_engine_cs *engine)
242 	next_heartbeat(engine);
245 void intel_engine_park_heartbeat(struct intel_engine_cs *engine)
247 	if (cancel_delayed_work(&engine->heartbeat.work))
248 		i915_request_put(fetch_and_zero(&engine->heartbeat.systole));
253 	struct intel_engine_cs *engine;
256 	for_each_engine(engine, gt, id)
257 		if (intel_engine_pm_is_awake(engine))
258 			intel_engine_unpark_heartbeat(engine);
263 	struct intel_engine_cs *engine;
266 	for_each_engine(engine, gt, id)
267 		intel_engine_park_heartbeat(engine);
270 void intel_engine_init_heartbeat(struct intel_engine_cs *engine)
272 	INIT_DELAYED_WORK(&engine->heartbeat.work, heartbeat);
275 static int __intel_engine_pulse(struct intel_engine_cs *engine)
278 	struct intel_context *ce = engine->kernel_context;
282 	GEM_BUG_ON(!intel_engine_has_preemption(engine));
283 	GEM_BUG_ON(!intel_engine_pm_is_awake(engine));
295 	next_heartbeat(engine);
300 static unsigned long set_heartbeat(struct intel_engine_cs *engine,
305 	old = xchg(&engine->props.heartbeat_interval_ms, delay);
307 		intel_engine_unpark_heartbeat(engine);
309 		intel_engine_park_heartbeat(engine);
314 int intel_engine_set_heartbeat(struct intel_engine_cs *engine,
317 	struct intel_context *ce = engine->kernel_context;
320 	if (!delay && !intel_engine_has_preempt_reset(engine))
324 	if (delay != engine->defaults.heartbeat_interval_ms &&
325 	    delay < 2 * engine->props.preempt_timeout_ms) {
326 		if (intel_engine_uses_guc(engine))
327 			drm_notice(&engine->i915->drm, "%s heartbeat interval adjusted to a non-default value which may downgrade individual engine resets to full GPU resets!\n",
328 				   engine->name);
330 			drm_notice(&engine->i915->drm, "%s heartbeat interval adjusted to a non-default value which may cause engine resets to target innocent contexts!\n",
331 				   engine->name);
334 	intel_engine_pm_get(engine);
340 	if (delay != engine->props.heartbeat_interval_ms) {
341 		unsigned long saved = set_heartbeat(engine, delay);
344 		if (intel_engine_has_preemption(engine)) {
345 			err = __intel_engine_pulse(engine);
347 				set_heartbeat(engine, saved);
354 	intel_engine_pm_put(engine);
358 int intel_engine_pulse(struct intel_engine_cs *engine)
360 	struct intel_context *ce = engine->kernel_context;
363 	if (!intel_engine_has_preemption(engine))
366 	if (!intel_engine_pm_get_if_awake(engine))
371 		err = __intel_engine_pulse(engine);
375 	intel_engine_flush_submission(engine);
376 	intel_engine_pm_put(engine);
380 int intel_engine_flush_barriers(struct intel_engine_cs *engine)
383 	struct intel_context *ce = engine->kernel_context;
387 	if (llist_empty(&engine->barrier_tasks))
390 	if (!intel_engine_pm_get_if_awake(engine))
410 	intel_engine_pm_put(engine);