* [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets
@ 2019-10-23 13:31 Chris Wilson
2019-10-23 13:31 ` [Intel-gfx] " Chris Wilson
` (6 more replies)
0 siblings, 7 replies; 18+ messages in thread
From: Chris Wilson @ 2019-10-23 13:31 UTC (permalink / raw)
To: intel-gfx
If we are doing a normal GPU reset triggered after detecting a long
period of stalled work, we can take our time and allow the engines to
quiesce. Since we've stopped submission to the engine, and if we wait
long enough an innocent context should complete, leaving the engine idle.
So by waiting a short amount of time, we should prevent clobbering other
users when resetting a stuck context.
Suggested-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Suggested-by: Jon Bloomfield <jon.bloomfield@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
---
drivers/gpu/drm/i915/Kconfig.profile | 11 +++++++++++
drivers/gpu/drm/i915/gt/intel_engine_cs.c | 20 +++++++++++++++++++-
drivers/gpu/drm/i915/gt/intel_engine_types.h | 4 ++++
3 files changed, 34 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile
index 48df8889a88a..3a3881d5e44b 100644
--- a/drivers/gpu/drm/i915/Kconfig.profile
+++ b/drivers/gpu/drm/i915/Kconfig.profile
@@ -25,3 +25,14 @@ config DRM_I915_SPIN_REQUEST
May be 0 to disable the initial spin. In practice, we estimate
the cost of enabling the interrupt (if currently disabled) to be
a few microseconds.
+
+config DRM_I915_STOP_TIMEOUT
+ int "How long to wait for an engine to quiesce gracefully before reset (ms)"
+ default 100 # milliseconds
+ help
+ By stopping submission and sleeping for a short time before resetting
+ the GPU, we allow the innocent contexts also on the system to quiesce.
+ It is then less likely for a hanging context to cause collateral
+ damage as the system is reset in order to recover. The corollary is
+ that the reset itself may take longer and so be more disruptive to
+ interactive or low latency workloads.
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
index 0e20713603ec..e4203eb44139 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
@@ -308,6 +308,9 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id)
engine->instance = info->instance;
__sprint_engine_name(engine);
+ engine->props.stop_timeout_ms =
+ CONFIG_DRM_I915_STOP_TIMEOUT;
+
/*
* To be overridden by the backend on setup. However to facilitate
* cleanup on error during setup, we always provide the destroy vfunc.
@@ -875,6 +878,21 @@ u64 intel_engine_get_last_batch_head(const struct intel_engine_cs *engine)
return bbaddr;
}
+static unsigned long stop_timeout(const struct intel_engine_cs *engine)
+{
+ if (in_atomic() || irqs_disabled()) /* inside atomic preempt-reset? */
+ return 0;
+
+ /*
+ * If we are doing a normal GPU reset, we can take our time and allow
+ * the engine to quiesce. We've stopped submission to the engine, and
+ * if we wait long enough an innocent context should complete and
+ * leave the engine idle. So they should not be caught unaware by
+ * the forthcoming GPU reset (which usually follows the stop_cs)!
+ */
+ return READ_ONCE(engine->props.stop_timeout_ms);
+}
+
int intel_engine_stop_cs(struct intel_engine_cs *engine)
{
struct intel_uncore *uncore = engine->uncore;
@@ -892,7 +910,7 @@ int intel_engine_stop_cs(struct intel_engine_cs *engine)
err = 0;
if (__intel_wait_for_register_fw(uncore,
mode, MODE_IDLE, MODE_IDLE,
- 1000, 0,
+ 1000, stop_timeout(engine),
NULL)) {
GEM_TRACE("%s: timed out on STOP_RING -> IDLE\n", engine->name);
err = -ETIMEDOUT;
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h
index 3451be034caf..87d5c4ef3ae7 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h
@@ -542,6 +542,10 @@ struct intel_engine_cs {
*/
ktime_t total;
} stats;
+
+ struct {
+ unsigned long stop_timeout_ms;
+ } props;
};
static inline bool
--
2.24.0.rc0
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
^ permalink raw reply related [flat|nested] 18+ messages in thread* [Intel-gfx] [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets 2019-10-23 13:31 [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets Chris Wilson @ 2019-10-23 13:31 ` Chris Wilson 2019-10-23 13:31 ` [CI 2/5] drm/i915/execlists: Force preemption Chris Wilson ` (5 subsequent siblings) 6 siblings, 0 replies; 18+ messages in thread From: Chris Wilson @ 2019-10-23 13:31 UTC (permalink / raw) To: intel-gfx If we are doing a normal GPU reset triggered after detecting a long period of stalled work, we can take our time and allow the engines to quiesce. Since we've stopped submission to the engine, and if we wait long enough an innocent context should complete, leaving the engine idle. So by waiting a short amount of time, we should prevent clobbering other users when resetting a stuck context. Suggested-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Suggested-by: Jon Bloomfield <jon.bloomfield@intel.com> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com> --- drivers/gpu/drm/i915/Kconfig.profile | 11 +++++++++++ drivers/gpu/drm/i915/gt/intel_engine_cs.c | 20 +++++++++++++++++++- drivers/gpu/drm/i915/gt/intel_engine_types.h | 4 ++++ 3 files changed, 34 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile index 48df8889a88a..3a3881d5e44b 100644 --- a/drivers/gpu/drm/i915/Kconfig.profile +++ b/drivers/gpu/drm/i915/Kconfig.profile @@ -25,3 +25,14 @@ config DRM_I915_SPIN_REQUEST May be 0 to disable the initial spin. In practice, we estimate the cost of enabling the interrupt (if currently disabled) to be a few microseconds. + +config DRM_I915_STOP_TIMEOUT + int "How long to wait for an engine to quiesce gracefully before reset (ms)" + default 100 # milliseconds + help + By stopping submission and sleeping for a short time before resetting + the GPU, we allow the innocent contexts also on the system to quiesce. + It is then less likely for a hanging context to cause collateral + damage as the system is reset in order to recover. The corollary is + that the reset itself may take longer and so be more disruptive to + interactive or low latency workloads. diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c index 0e20713603ec..e4203eb44139 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c +++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c @@ -308,6 +308,9 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id) engine->instance = info->instance; __sprint_engine_name(engine); + engine->props.stop_timeout_ms = + CONFIG_DRM_I915_STOP_TIMEOUT; + /* * To be overridden by the backend on setup. However to facilitate * cleanup on error during setup, we always provide the destroy vfunc. @@ -875,6 +878,21 @@ u64 intel_engine_get_last_batch_head(const struct intel_engine_cs *engine) return bbaddr; } +static unsigned long stop_timeout(const struct intel_engine_cs *engine) +{ + if (in_atomic() || irqs_disabled()) /* inside atomic preempt-reset? */ + return 0; + + /* + * If we are doing a normal GPU reset, we can take our time and allow + * the engine to quiesce. We've stopped submission to the engine, and + * if we wait long enough an innocent context should complete and + * leave the engine idle. So they should not be caught unaware by + * the forthcoming GPU reset (which usually follows the stop_cs)! + */ + return READ_ONCE(engine->props.stop_timeout_ms); +} + int intel_engine_stop_cs(struct intel_engine_cs *engine) { struct intel_uncore *uncore = engine->uncore; @@ -892,7 +910,7 @@ int intel_engine_stop_cs(struct intel_engine_cs *engine) err = 0; if (__intel_wait_for_register_fw(uncore, mode, MODE_IDLE, MODE_IDLE, - 1000, 0, + 1000, stop_timeout(engine), NULL)) { GEM_TRACE("%s: timed out on STOP_RING -> IDLE\n", engine->name); err = -ETIMEDOUT; diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h index 3451be034caf..87d5c4ef3ae7 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_types.h +++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h @@ -542,6 +542,10 @@ struct intel_engine_cs { */ ktime_t total; } stats; + + struct { + unsigned long stop_timeout_ms; + } props; }; static inline bool -- 2.24.0.rc0 _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply related [flat|nested] 18+ messages in thread
* [CI 2/5] drm/i915/execlists: Force preemption 2019-10-23 13:31 [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets Chris Wilson 2019-10-23 13:31 ` [Intel-gfx] " Chris Wilson @ 2019-10-23 13:31 ` Chris Wilson 2019-10-23 13:31 ` [Intel-gfx] " Chris Wilson 2019-10-23 13:31 ` [CI 3/5] drm/i915/execlists: Cancel banned contexts on schedule-out Chris Wilson ` (4 subsequent siblings) 6 siblings, 1 reply; 18+ messages in thread From: Chris Wilson @ 2019-10-23 13:31 UTC (permalink / raw) To: intel-gfx If the preempted context takes too long to relinquish control, e.g. it is stuck inside a shader with arbitration disabled, evict that context with an engine reset. This ensures that preemptions are reasonably responsive, providing a tighter QoS for the more important context at the cost of flagging unresponsive contexts more frequently (i.e. instead of using an ~10s hangcheck, we now evict at ~100ms). The challenge of lies in picking a timeout that can be reasonably serviced by HW for typical workloads, balancing the existing clients against the needs for responsiveness. Note that coupled with timeslicing, this will lead to rapid GPU "hang" detection with multiple active contexts vying for GPU time. The forced preemption mechanism can be compiled out with ./scripts/config --set-val DRM_I915_PREEMPT_TIMEOUT 0 Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com> --- drivers/gpu/drm/i915/Kconfig.profile | 12 +++ drivers/gpu/drm/i915/gt/intel_engine.h | 9 ++ drivers/gpu/drm/i915/gt/intel_engine_cs.c | 5 +- drivers/gpu/drm/i915/gt/intel_engine_types.h | 6 ++ drivers/gpu/drm/i915/gt/intel_lrc.c | 93 +++++++++++++++-- drivers/gpu/drm/i915/gt/selftest_lrc.c | 100 +++++++++++++++++++ drivers/gpu/drm/i915/i915_gem.h | 14 --- drivers/gpu/drm/i915/i915_params.h | 2 +- drivers/gpu/drm/i915/i915_utils.c | 29 ++++++ drivers/gpu/drm/i915/i915_utils.h | 9 ++ 10 files changed, 255 insertions(+), 24 deletions(-) diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile index 3a3881d5e44b..b071b6024152 100644 --- a/drivers/gpu/drm/i915/Kconfig.profile +++ b/drivers/gpu/drm/i915/Kconfig.profile @@ -12,6 +12,18 @@ 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_PREEMPT_TIMEOUT + int "Preempt timeout (ms, jiffy granularity)" + default 100 # milliseconds + help + How long to wait (in milliseconds) for a preemption event to occur + when submitting a new context via execlists. If the current context + does not hit an arbitration point and yield to HW before the timer + expires, the HW will be reset to allow the more important context + to execute. + + May be 0 to disable the timeout. + config DRM_I915_SPIN_REQUEST int "Busywait for request completion (us)" default 5 # microseconds diff --git a/drivers/gpu/drm/i915/gt/intel_engine.h b/drivers/gpu/drm/i915/gt/intel_engine.h index c2d9d67c63d9..9409b7856299 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine.h +++ b/drivers/gpu/drm/i915/gt/intel_engine.h @@ -527,4 +527,13 @@ void intel_engine_init_active(struct intel_engine_cs *engine, #define ENGINE_MOCK 1 #define ENGINE_VIRTUAL 2 +static inline bool +intel_engine_has_preempt_reset(const struct intel_engine_cs *engine) +{ + if (!CONFIG_DRM_I915_PREEMPT_TIMEOUT) + return 0; + + return intel_engine_has_preemption(engine); +} + #endif /* _INTEL_RINGBUFFER_H_ */ diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c index e4203eb44139..b91ea07f4819 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c +++ b/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.preempt_timeout_ms = + CONFIG_DRM_I915_PREEMPT_TIMEOUT; engine->props.stop_timeout_ms = CONFIG_DRM_I915_STOP_TIMEOUT; @@ -1338,10 +1340,11 @@ static void intel_engine_print_registers(struct intel_engine_cs *engine, unsigned int idx; u8 read, write; - drm_printf(m, "\tExeclist tasklet queued? %s (%s), timeslice? %s\n", + drm_printf(m, "\tExeclist tasklet queued? %s (%s), preempt? %s, timeslice? %s\n", yesno(test_bit(TASKLET_STATE_SCHED, &engine->execlists.tasklet.state)), enableddisabled(!atomic_read(&engine->execlists.tasklet.count)), + repr_timer(&engine->execlists.preempt), repr_timer(&engine->execlists.timer)); read = execlists->csb_head; diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h index 87d5c4ef3ae7..1251dac91f31 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_types.h +++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h @@ -174,6 +174,11 @@ struct intel_engine_execlists { */ struct timer_list timer; + /** + * @preempt: reset the current context if it fails to give way + */ + struct timer_list preempt; + /** * @default_priolist: priority list for I915_PRIORITY_NORMAL */ @@ -544,6 +549,7 @@ struct intel_engine_cs { } stats; struct { + unsigned long preempt_timeout_ms; unsigned long stop_timeout_ms; } props; }; diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c index f9f3e985bb79..ff0dd297e782 100644 --- a/drivers/gpu/drm/i915/gt/intel_lrc.c +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c @@ -1372,6 +1372,26 @@ static void record_preemption(struct intel_engine_execlists *execlists) (void)I915_SELFTEST_ONLY(execlists->preempt_hang.count++); } +static unsigned long active_preempt_timeout(struct intel_engine_cs *engine) +{ + struct i915_request *rq; + + rq = last_active(&engine->execlists); + if (!rq) + return 0; + + return READ_ONCE(engine->props.preempt_timeout_ms); +} + +static void set_preempt_timeout(struct intel_engine_cs *engine) +{ + if (!intel_engine_has_preempt_reset(engine)) + return; + + set_timer_ms(&engine->execlists.preempt, + active_preempt_timeout(engine)); +} + static void execlists_dequeue(struct intel_engine_cs *engine) { struct intel_engine_execlists * const execlists = &engine->execlists; @@ -1747,6 +1767,8 @@ static void execlists_dequeue(struct intel_engine_cs *engine) memset(port + 1, 0, (last_port - port) * sizeof(*port)); execlists_submit_ports(engine); + + set_preempt_timeout(engine); } else { skip_submit: ring_set_paused(engine, 0); @@ -1987,6 +2009,43 @@ static void __execlists_submission_tasklet(struct intel_engine_cs *const engine) } } +static noinline void preempt_reset(struct intel_engine_cs *engine) +{ + const unsigned int bit = I915_RESET_ENGINE + engine->id; + unsigned long *lock = &engine->gt->reset.flags; + + if (i915_modparams.reset < 3) + return; + + if (test_and_set_bit(bit, lock)) + return; + + /* Mark this tasklet as disabled to avoid waiting for it to complete */ + tasklet_disable_nosync(&engine->execlists.tasklet); + + GEM_TRACE("%s: preempt timeout %lu+%ums\n", + engine->name, + READ_ONCE(engine->props.preempt_timeout_ms), + jiffies_to_msecs(jiffies - engine->execlists.preempt.expires)); + intel_engine_reset(engine, "preemption time out"); + + tasklet_enable(&engine->execlists.tasklet); + clear_and_wake_up_bit(bit, lock); +} + +static bool preempt_timeout(const struct intel_engine_cs *const engine) +{ + const struct timer_list *t = &engine->execlists.preempt; + + if (!CONFIG_DRM_I915_PREEMPT_TIMEOUT) + return false; + + if (!timer_expired(t)) + return false; + + return READ_ONCE(engine->execlists.pending[0]); +} + /* * Check the unread Context Status Buffers and manage the submission of new * contexts to the ELSP accordingly. @@ -1994,23 +2053,39 @@ static void __execlists_submission_tasklet(struct intel_engine_cs *const engine) static void execlists_submission_tasklet(unsigned long data) { struct intel_engine_cs * const engine = (struct intel_engine_cs *)data; - unsigned long flags; + bool timeout = preempt_timeout(engine); process_csb(engine); - if (!READ_ONCE(engine->execlists.pending[0])) { + if (!READ_ONCE(engine->execlists.pending[0]) || timeout) { + unsigned long flags; + spin_lock_irqsave(&engine->active.lock, flags); __execlists_submission_tasklet(engine); spin_unlock_irqrestore(&engine->active.lock, flags); + + /* Recheck after serialising with direct-submission */ + if (timeout && preempt_timeout(engine)) + preempt_reset(engine); } } -static void execlists_submission_timer(struct timer_list *timer) +static void __execlists_kick(struct intel_engine_execlists *execlists) { - struct intel_engine_cs *engine = - from_timer(engine, timer, execlists.timer); - /* Kick the tasklet for some interrupt coalescing and reset handling */ - tasklet_hi_schedule(&engine->execlists.tasklet); + tasklet_hi_schedule(&execlists->tasklet); +} + +#define execlists_kick(t, member) \ + __execlists_kick(container_of(t, struct intel_engine_execlists, member)) + +static void execlists_timeslice(struct timer_list *timer) +{ + execlists_kick(timer, timer); +} + +static void execlists_preempt(struct timer_list *timer) +{ + execlists_kick(timer, preempt); } static void queue_request(struct intel_engine_cs *engine, @@ -3455,6 +3530,7 @@ gen12_emit_fini_breadcrumb_rcs(struct i915_request *request, u32 *cs) static void execlists_park(struct intel_engine_cs *engine) { cancel_timer(&engine->execlists.timer); + cancel_timer(&engine->execlists.preempt); } void intel_execlists_set_default_submission(struct intel_engine_cs *engine) @@ -3572,7 +3648,8 @@ int intel_execlists_submission_setup(struct intel_engine_cs *engine) { tasklet_init(&engine->execlists.tasklet, execlists_submission_tasklet, (unsigned long)engine); - timer_setup(&engine->execlists.timer, execlists_submission_timer, 0); + timer_setup(&engine->execlists.timer, execlists_timeslice, 0); + timer_setup(&engine->execlists.preempt, execlists_preempt, 0); logical_ring_default_vfuncs(engine); logical_ring_default_irqs(engine); diff --git a/drivers/gpu/drm/i915/gt/selftest_lrc.c b/drivers/gpu/drm/i915/gt/selftest_lrc.c index 7516d1c90925..b6352671c5a0 100644 --- a/drivers/gpu/drm/i915/gt/selftest_lrc.c +++ b/drivers/gpu/drm/i915/gt/selftest_lrc.c @@ -1697,6 +1697,105 @@ static int live_preempt_hang(void *arg) return err; } +static int live_preempt_timeout(void *arg) +{ + struct intel_gt *gt = arg; + struct i915_gem_context *ctx_hi, *ctx_lo; + struct igt_spinner spin_lo; + struct intel_engine_cs *engine; + enum intel_engine_id id; + int err = -ENOMEM; + + /* + * Check that we force preemption to occur by cancelling the previous + * context if it refuses to yield the GPU. + */ + if (!CONFIG_DRM_I915_PREEMPT_TIMEOUT) + return 0; + + if (!HAS_LOGICAL_RING_PREEMPTION(gt->i915)) + return 0; + + if (!intel_has_reset_engine(gt)) + return 0; + + if (igt_spinner_init(&spin_lo, gt)) + return -ENOMEM; + + ctx_hi = kernel_context(gt->i915); + if (!ctx_hi) + goto err_spin_lo; + ctx_hi->sched.priority = + I915_USER_PRIORITY(I915_CONTEXT_MAX_USER_PRIORITY); + + ctx_lo = kernel_context(gt->i915); + if (!ctx_lo) + goto err_ctx_hi; + ctx_lo->sched.priority = + I915_USER_PRIORITY(I915_CONTEXT_MIN_USER_PRIORITY); + + for_each_engine(engine, gt, id) { + unsigned long saved_timeout; + struct i915_request *rq; + + if (!intel_engine_has_preemption(engine)) + continue; + + rq = spinner_create_request(&spin_lo, ctx_lo, engine, + MI_NOOP); /* preemption disabled */ + if (IS_ERR(rq)) { + err = PTR_ERR(rq); + goto err_ctx_lo; + } + + i915_request_add(rq); + if (!igt_wait_for_spinner(&spin_lo, rq)) { + intel_gt_set_wedged(gt); + err = -EIO; + goto err_ctx_lo; + } + + rq = igt_request_alloc(ctx_hi, engine); + if (IS_ERR(rq)) { + igt_spinner_end(&spin_lo); + err = PTR_ERR(rq); + goto err_ctx_lo; + } + + /* Flush the previous CS ack before changing timeouts */ + while (READ_ONCE(engine->execlists.pending[0])) + cpu_relax(); + + saved_timeout = engine->props.preempt_timeout_ms; + engine->props.preempt_timeout_ms = 1; /* in ms, -> 1 jiffie */ + + i915_request_get(rq); + i915_request_add(rq); + + intel_engine_flush_submission(engine); + engine->props.preempt_timeout_ms = saved_timeout; + + if (i915_request_wait(rq, 0, HZ / 10) < 0) { + intel_gt_set_wedged(gt); + i915_request_put(rq); + err = -ETIME; + goto err_ctx_lo; + } + + igt_spinner_end(&spin_lo); + i915_request_put(rq); + } + + err = 0; +err_ctx_lo: + kernel_context_close(ctx_lo); +err_ctx_hi: + kernel_context_close(ctx_hi); +err_spin_lo: + igt_spinner_fini(&spin_lo); + return err; +} + static int random_range(struct rnd_state *rnd, int min, int max) { return i915_prandom_u32_max_state(max - min, rnd) + min; @@ -2598,6 +2697,7 @@ int intel_execlists_live_selftests(struct drm_i915_private *i915) SUBTEST(live_suppress_wait_preempt), SUBTEST(live_chain_preempt), SUBTEST(live_preempt_hang), + SUBTEST(live_preempt_timeout), SUBTEST(live_preempt_smoke), SUBTEST(live_virtual_engine), SUBTEST(live_virtual_mask), diff --git a/drivers/gpu/drm/i915/i915_gem.h b/drivers/gpu/drm/i915/i915_gem.h index 2011f8e9a9f1..f6f9675848b8 100644 --- a/drivers/gpu/drm/i915/i915_gem.h +++ b/drivers/gpu/drm/i915/i915_gem.h @@ -112,18 +112,4 @@ static inline bool __tasklet_is_scheduled(struct tasklet_struct *t) return test_bit(TASKLET_STATE_SCHED, &t->state); } -static inline void cancel_timer(struct timer_list *t) -{ - if (!READ_ONCE(t->expires)) - return; - - del_timer(t); - WRITE_ONCE(t->expires, 0); -} - -static inline bool timer_expired(const struct timer_list *t) -{ - return READ_ONCE(t->expires) && !timer_pending(t); -} - #endif /* __I915_GEM_H__ */ diff --git a/drivers/gpu/drm/i915/i915_params.h b/drivers/gpu/drm/i915/i915_params.h index d29ade3b7de6..56058978bb27 100644 --- a/drivers/gpu/drm/i915/i915_params.h +++ b/drivers/gpu/drm/i915/i915_params.h @@ -61,7 +61,7 @@ struct drm_printer; param(char *, dmc_firmware_path, NULL) \ param(int, mmio_debug, -IS_ENABLED(CONFIG_DRM_I915_DEBUG_MMIO)) \ param(int, edp_vswing, 0) \ - param(int, reset, 2) \ + param(int, reset, 3) \ param(unsigned int, inject_load_failure, 0) \ param(int, fastboot, -1) \ param(int, enable_dpcd_backlight, 0) \ diff --git a/drivers/gpu/drm/i915/i915_utils.c b/drivers/gpu/drm/i915/i915_utils.c index 16acdf7bdbe6..02e969b64505 100644 --- a/drivers/gpu/drm/i915/i915_utils.c +++ b/drivers/gpu/drm/i915/i915_utils.c @@ -76,3 +76,32 @@ bool i915_error_injected(void) } #endif + +void cancel_timer(struct timer_list *t) +{ + if (!READ_ONCE(t->expires)) + return; + + del_timer(t); + WRITE_ONCE(t->expires, 0); +} + +void set_timer_ms(struct timer_list *t, unsigned long timeout) +{ + if (!timeout) { + cancel_timer(t); + return; + } + + timeout = msecs_to_jiffies_timeout(timeout); + + /* + * Paranoia to make sure the compiler computes the timeout before + * loading 'jiffies' as jiffies is volatile and may be updated in + * the background by a timer tick. All to reduce the complexity + * of the addition and reduce the risk of losing a jiffie. + */ + barrier(); + + mod_timer(t, jiffies + timeout); +} diff --git a/drivers/gpu/drm/i915/i915_utils.h b/drivers/gpu/drm/i915/i915_utils.h index 562f756da421..94f136d8a5fd 100644 --- a/drivers/gpu/drm/i915/i915_utils.h +++ b/drivers/gpu/drm/i915/i915_utils.h @@ -32,6 +32,7 @@ #include <linux/workqueue.h> struct drm_i915_private; +struct timer_list; #undef WARN_ON /* Many gcc seem to no see through this and fall over :( */ @@ -421,4 +422,12 @@ static inline void add_taint_for_CI(unsigned int taint) add_taint(taint, LOCKDEP_STILL_OK); } +void cancel_timer(struct timer_list *t); +void set_timer_ms(struct timer_list *t, unsigned long timeout); + +static inline bool timer_expired(const struct timer_list *t) +{ + return READ_ONCE(t->expires) && !timer_pending(t); +} + #endif /* !__I915_UTILS_H */ -- 2.24.0.rc0 _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply related [flat|nested] 18+ messages in thread
* [Intel-gfx] [CI 2/5] drm/i915/execlists: Force preemption 2019-10-23 13:31 ` [CI 2/5] drm/i915/execlists: Force preemption Chris Wilson @ 2019-10-23 13:31 ` Chris Wilson 0 siblings, 0 replies; 18+ messages in thread From: Chris Wilson @ 2019-10-23 13:31 UTC (permalink / raw) To: intel-gfx If the preempted context takes too long to relinquish control, e.g. it is stuck inside a shader with arbitration disabled, evict that context with an engine reset. This ensures that preemptions are reasonably responsive, providing a tighter QoS for the more important context at the cost of flagging unresponsive contexts more frequently (i.e. instead of using an ~10s hangcheck, we now evict at ~100ms). The challenge of lies in picking a timeout that can be reasonably serviced by HW for typical workloads, balancing the existing clients against the needs for responsiveness. Note that coupled with timeslicing, this will lead to rapid GPU "hang" detection with multiple active contexts vying for GPU time. The forced preemption mechanism can be compiled out with ./scripts/config --set-val DRM_I915_PREEMPT_TIMEOUT 0 Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com> --- drivers/gpu/drm/i915/Kconfig.profile | 12 +++ drivers/gpu/drm/i915/gt/intel_engine.h | 9 ++ drivers/gpu/drm/i915/gt/intel_engine_cs.c | 5 +- drivers/gpu/drm/i915/gt/intel_engine_types.h | 6 ++ drivers/gpu/drm/i915/gt/intel_lrc.c | 93 +++++++++++++++-- drivers/gpu/drm/i915/gt/selftest_lrc.c | 100 +++++++++++++++++++ drivers/gpu/drm/i915/i915_gem.h | 14 --- drivers/gpu/drm/i915/i915_params.h | 2 +- drivers/gpu/drm/i915/i915_utils.c | 29 ++++++ drivers/gpu/drm/i915/i915_utils.h | 9 ++ 10 files changed, 255 insertions(+), 24 deletions(-) diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile index 3a3881d5e44b..b071b6024152 100644 --- a/drivers/gpu/drm/i915/Kconfig.profile +++ b/drivers/gpu/drm/i915/Kconfig.profile @@ -12,6 +12,18 @@ 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_PREEMPT_TIMEOUT + int "Preempt timeout (ms, jiffy granularity)" + default 100 # milliseconds + help + How long to wait (in milliseconds) for a preemption event to occur + when submitting a new context via execlists. If the current context + does not hit an arbitration point and yield to HW before the timer + expires, the HW will be reset to allow the more important context + to execute. + + May be 0 to disable the timeout. + config DRM_I915_SPIN_REQUEST int "Busywait for request completion (us)" default 5 # microseconds diff --git a/drivers/gpu/drm/i915/gt/intel_engine.h b/drivers/gpu/drm/i915/gt/intel_engine.h index c2d9d67c63d9..9409b7856299 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine.h +++ b/drivers/gpu/drm/i915/gt/intel_engine.h @@ -527,4 +527,13 @@ void intel_engine_init_active(struct intel_engine_cs *engine, #define ENGINE_MOCK 1 #define ENGINE_VIRTUAL 2 +static inline bool +intel_engine_has_preempt_reset(const struct intel_engine_cs *engine) +{ + if (!CONFIG_DRM_I915_PREEMPT_TIMEOUT) + return 0; + + return intel_engine_has_preemption(engine); +} + #endif /* _INTEL_RINGBUFFER_H_ */ diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c index e4203eb44139..b91ea07f4819 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c +++ b/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.preempt_timeout_ms = + CONFIG_DRM_I915_PREEMPT_TIMEOUT; engine->props.stop_timeout_ms = CONFIG_DRM_I915_STOP_TIMEOUT; @@ -1338,10 +1340,11 @@ static void intel_engine_print_registers(struct intel_engine_cs *engine, unsigned int idx; u8 read, write; - drm_printf(m, "\tExeclist tasklet queued? %s (%s), timeslice? %s\n", + drm_printf(m, "\tExeclist tasklet queued? %s (%s), preempt? %s, timeslice? %s\n", yesno(test_bit(TASKLET_STATE_SCHED, &engine->execlists.tasklet.state)), enableddisabled(!atomic_read(&engine->execlists.tasklet.count)), + repr_timer(&engine->execlists.preempt), repr_timer(&engine->execlists.timer)); read = execlists->csb_head; diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h index 87d5c4ef3ae7..1251dac91f31 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_types.h +++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h @@ -174,6 +174,11 @@ struct intel_engine_execlists { */ struct timer_list timer; + /** + * @preempt: reset the current context if it fails to give way + */ + struct timer_list preempt; + /** * @default_priolist: priority list for I915_PRIORITY_NORMAL */ @@ -544,6 +549,7 @@ struct intel_engine_cs { } stats; struct { + unsigned long preempt_timeout_ms; unsigned long stop_timeout_ms; } props; }; diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c index f9f3e985bb79..ff0dd297e782 100644 --- a/drivers/gpu/drm/i915/gt/intel_lrc.c +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c @@ -1372,6 +1372,26 @@ static void record_preemption(struct intel_engine_execlists *execlists) (void)I915_SELFTEST_ONLY(execlists->preempt_hang.count++); } +static unsigned long active_preempt_timeout(struct intel_engine_cs *engine) +{ + struct i915_request *rq; + + rq = last_active(&engine->execlists); + if (!rq) + return 0; + + return READ_ONCE(engine->props.preempt_timeout_ms); +} + +static void set_preempt_timeout(struct intel_engine_cs *engine) +{ + if (!intel_engine_has_preempt_reset(engine)) + return; + + set_timer_ms(&engine->execlists.preempt, + active_preempt_timeout(engine)); +} + static void execlists_dequeue(struct intel_engine_cs *engine) { struct intel_engine_execlists * const execlists = &engine->execlists; @@ -1747,6 +1767,8 @@ static void execlists_dequeue(struct intel_engine_cs *engine) memset(port + 1, 0, (last_port - port) * sizeof(*port)); execlists_submit_ports(engine); + + set_preempt_timeout(engine); } else { skip_submit: ring_set_paused(engine, 0); @@ -1987,6 +2009,43 @@ static void __execlists_submission_tasklet(struct intel_engine_cs *const engine) } } +static noinline void preempt_reset(struct intel_engine_cs *engine) +{ + const unsigned int bit = I915_RESET_ENGINE + engine->id; + unsigned long *lock = &engine->gt->reset.flags; + + if (i915_modparams.reset < 3) + return; + + if (test_and_set_bit(bit, lock)) + return; + + /* Mark this tasklet as disabled to avoid waiting for it to complete */ + tasklet_disable_nosync(&engine->execlists.tasklet); + + GEM_TRACE("%s: preempt timeout %lu+%ums\n", + engine->name, + READ_ONCE(engine->props.preempt_timeout_ms), + jiffies_to_msecs(jiffies - engine->execlists.preempt.expires)); + intel_engine_reset(engine, "preemption time out"); + + tasklet_enable(&engine->execlists.tasklet); + clear_and_wake_up_bit(bit, lock); +} + +static bool preempt_timeout(const struct intel_engine_cs *const engine) +{ + const struct timer_list *t = &engine->execlists.preempt; + + if (!CONFIG_DRM_I915_PREEMPT_TIMEOUT) + return false; + + if (!timer_expired(t)) + return false; + + return READ_ONCE(engine->execlists.pending[0]); +} + /* * Check the unread Context Status Buffers and manage the submission of new * contexts to the ELSP accordingly. @@ -1994,23 +2053,39 @@ static void __execlists_submission_tasklet(struct intel_engine_cs *const engine) static void execlists_submission_tasklet(unsigned long data) { struct intel_engine_cs * const engine = (struct intel_engine_cs *)data; - unsigned long flags; + bool timeout = preempt_timeout(engine); process_csb(engine); - if (!READ_ONCE(engine->execlists.pending[0])) { + if (!READ_ONCE(engine->execlists.pending[0]) || timeout) { + unsigned long flags; + spin_lock_irqsave(&engine->active.lock, flags); __execlists_submission_tasklet(engine); spin_unlock_irqrestore(&engine->active.lock, flags); + + /* Recheck after serialising with direct-submission */ + if (timeout && preempt_timeout(engine)) + preempt_reset(engine); } } -static void execlists_submission_timer(struct timer_list *timer) +static void __execlists_kick(struct intel_engine_execlists *execlists) { - struct intel_engine_cs *engine = - from_timer(engine, timer, execlists.timer); - /* Kick the tasklet for some interrupt coalescing and reset handling */ - tasklet_hi_schedule(&engine->execlists.tasklet); + tasklet_hi_schedule(&execlists->tasklet); +} + +#define execlists_kick(t, member) \ + __execlists_kick(container_of(t, struct intel_engine_execlists, member)) + +static void execlists_timeslice(struct timer_list *timer) +{ + execlists_kick(timer, timer); +} + +static void execlists_preempt(struct timer_list *timer) +{ + execlists_kick(timer, preempt); } static void queue_request(struct intel_engine_cs *engine, @@ -3455,6 +3530,7 @@ gen12_emit_fini_breadcrumb_rcs(struct i915_request *request, u32 *cs) static void execlists_park(struct intel_engine_cs *engine) { cancel_timer(&engine->execlists.timer); + cancel_timer(&engine->execlists.preempt); } void intel_execlists_set_default_submission(struct intel_engine_cs *engine) @@ -3572,7 +3648,8 @@ int intel_execlists_submission_setup(struct intel_engine_cs *engine) { tasklet_init(&engine->execlists.tasklet, execlists_submission_tasklet, (unsigned long)engine); - timer_setup(&engine->execlists.timer, execlists_submission_timer, 0); + timer_setup(&engine->execlists.timer, execlists_timeslice, 0); + timer_setup(&engine->execlists.preempt, execlists_preempt, 0); logical_ring_default_vfuncs(engine); logical_ring_default_irqs(engine); diff --git a/drivers/gpu/drm/i915/gt/selftest_lrc.c b/drivers/gpu/drm/i915/gt/selftest_lrc.c index 7516d1c90925..b6352671c5a0 100644 --- a/drivers/gpu/drm/i915/gt/selftest_lrc.c +++ b/drivers/gpu/drm/i915/gt/selftest_lrc.c @@ -1697,6 +1697,105 @@ static int live_preempt_hang(void *arg) return err; } +static int live_preempt_timeout(void *arg) +{ + struct intel_gt *gt = arg; + struct i915_gem_context *ctx_hi, *ctx_lo; + struct igt_spinner spin_lo; + struct intel_engine_cs *engine; + enum intel_engine_id id; + int err = -ENOMEM; + + /* + * Check that we force preemption to occur by cancelling the previous + * context if it refuses to yield the GPU. + */ + if (!CONFIG_DRM_I915_PREEMPT_TIMEOUT) + return 0; + + if (!HAS_LOGICAL_RING_PREEMPTION(gt->i915)) + return 0; + + if (!intel_has_reset_engine(gt)) + return 0; + + if (igt_spinner_init(&spin_lo, gt)) + return -ENOMEM; + + ctx_hi = kernel_context(gt->i915); + if (!ctx_hi) + goto err_spin_lo; + ctx_hi->sched.priority = + I915_USER_PRIORITY(I915_CONTEXT_MAX_USER_PRIORITY); + + ctx_lo = kernel_context(gt->i915); + if (!ctx_lo) + goto err_ctx_hi; + ctx_lo->sched.priority = + I915_USER_PRIORITY(I915_CONTEXT_MIN_USER_PRIORITY); + + for_each_engine(engine, gt, id) { + unsigned long saved_timeout; + struct i915_request *rq; + + if (!intel_engine_has_preemption(engine)) + continue; + + rq = spinner_create_request(&spin_lo, ctx_lo, engine, + MI_NOOP); /* preemption disabled */ + if (IS_ERR(rq)) { + err = PTR_ERR(rq); + goto err_ctx_lo; + } + + i915_request_add(rq); + if (!igt_wait_for_spinner(&spin_lo, rq)) { + intel_gt_set_wedged(gt); + err = -EIO; + goto err_ctx_lo; + } + + rq = igt_request_alloc(ctx_hi, engine); + if (IS_ERR(rq)) { + igt_spinner_end(&spin_lo); + err = PTR_ERR(rq); + goto err_ctx_lo; + } + + /* Flush the previous CS ack before changing timeouts */ + while (READ_ONCE(engine->execlists.pending[0])) + cpu_relax(); + + saved_timeout = engine->props.preempt_timeout_ms; + engine->props.preempt_timeout_ms = 1; /* in ms, -> 1 jiffie */ + + i915_request_get(rq); + i915_request_add(rq); + + intel_engine_flush_submission(engine); + engine->props.preempt_timeout_ms = saved_timeout; + + if (i915_request_wait(rq, 0, HZ / 10) < 0) { + intel_gt_set_wedged(gt); + i915_request_put(rq); + err = -ETIME; + goto err_ctx_lo; + } + + igt_spinner_end(&spin_lo); + i915_request_put(rq); + } + + err = 0; +err_ctx_lo: + kernel_context_close(ctx_lo); +err_ctx_hi: + kernel_context_close(ctx_hi); +err_spin_lo: + igt_spinner_fini(&spin_lo); + return err; +} + static int random_range(struct rnd_state *rnd, int min, int max) { return i915_prandom_u32_max_state(max - min, rnd) + min; @@ -2598,6 +2697,7 @@ int intel_execlists_live_selftests(struct drm_i915_private *i915) SUBTEST(live_suppress_wait_preempt), SUBTEST(live_chain_preempt), SUBTEST(live_preempt_hang), + SUBTEST(live_preempt_timeout), SUBTEST(live_preempt_smoke), SUBTEST(live_virtual_engine), SUBTEST(live_virtual_mask), diff --git a/drivers/gpu/drm/i915/i915_gem.h b/drivers/gpu/drm/i915/i915_gem.h index 2011f8e9a9f1..f6f9675848b8 100644 --- a/drivers/gpu/drm/i915/i915_gem.h +++ b/drivers/gpu/drm/i915/i915_gem.h @@ -112,18 +112,4 @@ static inline bool __tasklet_is_scheduled(struct tasklet_struct *t) return test_bit(TASKLET_STATE_SCHED, &t->state); } -static inline void cancel_timer(struct timer_list *t) -{ - if (!READ_ONCE(t->expires)) - return; - - del_timer(t); - WRITE_ONCE(t->expires, 0); -} - -static inline bool timer_expired(const struct timer_list *t) -{ - return READ_ONCE(t->expires) && !timer_pending(t); -} - #endif /* __I915_GEM_H__ */ diff --git a/drivers/gpu/drm/i915/i915_params.h b/drivers/gpu/drm/i915/i915_params.h index d29ade3b7de6..56058978bb27 100644 --- a/drivers/gpu/drm/i915/i915_params.h +++ b/drivers/gpu/drm/i915/i915_params.h @@ -61,7 +61,7 @@ struct drm_printer; param(char *, dmc_firmware_path, NULL) \ param(int, mmio_debug, -IS_ENABLED(CONFIG_DRM_I915_DEBUG_MMIO)) \ param(int, edp_vswing, 0) \ - param(int, reset, 2) \ + param(int, reset, 3) \ param(unsigned int, inject_load_failure, 0) \ param(int, fastboot, -1) \ param(int, enable_dpcd_backlight, 0) \ diff --git a/drivers/gpu/drm/i915/i915_utils.c b/drivers/gpu/drm/i915/i915_utils.c index 16acdf7bdbe6..02e969b64505 100644 --- a/drivers/gpu/drm/i915/i915_utils.c +++ b/drivers/gpu/drm/i915/i915_utils.c @@ -76,3 +76,32 @@ bool i915_error_injected(void) } #endif + +void cancel_timer(struct timer_list *t) +{ + if (!READ_ONCE(t->expires)) + return; + + del_timer(t); + WRITE_ONCE(t->expires, 0); +} + +void set_timer_ms(struct timer_list *t, unsigned long timeout) +{ + if (!timeout) { + cancel_timer(t); + return; + } + + timeout = msecs_to_jiffies_timeout(timeout); + + /* + * Paranoia to make sure the compiler computes the timeout before + * loading 'jiffies' as jiffies is volatile and may be updated in + * the background by a timer tick. All to reduce the complexity + * of the addition and reduce the risk of losing a jiffie. + */ + barrier(); + + mod_timer(t, jiffies + timeout); +} diff --git a/drivers/gpu/drm/i915/i915_utils.h b/drivers/gpu/drm/i915/i915_utils.h index 562f756da421..94f136d8a5fd 100644 --- a/drivers/gpu/drm/i915/i915_utils.h +++ b/drivers/gpu/drm/i915/i915_utils.h @@ -32,6 +32,7 @@ #include <linux/workqueue.h> struct drm_i915_private; +struct timer_list; #undef WARN_ON /* Many gcc seem to no see through this and fall over :( */ @@ -421,4 +422,12 @@ static inline void add_taint_for_CI(unsigned int taint) add_taint(taint, LOCKDEP_STILL_OK); } +void cancel_timer(struct timer_list *t); +void set_timer_ms(struct timer_list *t, unsigned long timeout); + +static inline bool timer_expired(const struct timer_list *t) +{ + return READ_ONCE(t->expires) && !timer_pending(t); +} + #endif /* !__I915_UTILS_H */ -- 2.24.0.rc0 _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply related [flat|nested] 18+ messages in thread
* [CI 3/5] drm/i915/execlists: Cancel banned contexts on schedule-out 2019-10-23 13:31 [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets Chris Wilson 2019-10-23 13:31 ` [Intel-gfx] " Chris Wilson 2019-10-23 13:31 ` [CI 2/5] drm/i915/execlists: Force preemption Chris Wilson @ 2019-10-23 13:31 ` Chris Wilson 2019-10-23 13:31 ` [Intel-gfx] " Chris Wilson 2019-10-23 13:31 ` [CI 4/5] drm/i915/gem: Cancel contexts when hangchecking is disabled Chris Wilson ` (3 subsequent siblings) 6 siblings, 1 reply; 18+ messages in thread From: Chris Wilson @ 2019-10-23 13:31 UTC (permalink / raw) To: intel-gfx On schedule-out (CS completion) of a banned context, scrub the context image so that we do not replay the active payload. The intent is that we skip banned payloads on request submission so that the timeline advancement continues on in the background. However, if we are returning to a preempted request, i915_request_skip() is ineffective and instead we need to patch up the context image so that it continues from the start of the next request. v2: Fixup cancellation so that we only scrub the payload of the active request and do not short-circuit the breadcrumbs (which might cause other contexts to execute out of order). v3: Grammar pass Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> --- drivers/gpu/drm/i915/gt/intel_lrc.c | 131 ++++++---- drivers/gpu/drm/i915/gt/selftest_lrc.c | 321 +++++++++++++++++++++++++ 2 files changed, 411 insertions(+), 41 deletions(-) diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c index ff0dd297e782..651d5dd39464 100644 --- a/drivers/gpu/drm/i915/gt/intel_lrc.c +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c @@ -234,6 +234,9 @@ static void execlists_init_reg_state(u32 *reg_state, const struct intel_engine_cs *engine, const struct intel_ring *ring, bool close); +static void +__execlists_update_reg_state(const struct intel_context *ce, + const struct intel_engine_cs *engine); static void mark_eio(struct i915_request *rq) { @@ -246,6 +249,31 @@ static void mark_eio(struct i915_request *rq) i915_request_mark_complete(rq); } +static struct i915_request *active_request(struct i915_request *rq) +{ + const struct intel_context * const ce = rq->hw_context; + struct i915_request *active = NULL; + struct list_head *list; + + if (!i915_request_is_active(rq)) /* unwound, but incomplete! */ + return rq; + + rcu_read_lock(); + list = &rcu_dereference(rq->timeline)->requests; + list_for_each_entry_from_reverse(rq, list, link) { + if (i915_request_completed(rq)) + break; + + if (rq->hw_context != ce) + break; + + active = rq; + } + rcu_read_unlock(); + + return active; +} + static inline u32 intel_hws_preempt_address(struct intel_engine_cs *engine) { return (i915_ggtt_offset(engine->status_page.vma) + @@ -972,6 +1000,58 @@ static void kick_siblings(struct i915_request *rq, struct intel_context *ce) tasklet_schedule(&ve->base.execlists.tasklet); } +static void restore_default_state(struct intel_context *ce, + struct intel_engine_cs *engine) +{ + u32 *regs = ce->lrc_reg_state; + + if (engine->pinned_default_state) + memcpy(regs, /* skip restoring the vanilla PPHWSP */ + engine->pinned_default_state + LRC_STATE_PN * PAGE_SIZE, + engine->context_size - PAGE_SIZE); + + execlists_init_reg_state(regs, ce, engine, ce->ring, false); +} + +static void reset_active(struct i915_request *rq, + struct intel_engine_cs *engine) +{ + struct intel_context * const ce = rq->hw_context; + + /* + * The executing context has been cancelled. We want to prevent + * further execution along this context and propagate the error on + * to anything depending on its results. + * + * In __i915_request_submit(), we apply the -EIO and remove the + * requests' payloads for any banned requests. But first, we must + * rewind the context back to the start of the incomplete request so + * that we do not jump back into the middle of the batch. + * + * We preserve the breadcrumbs and semaphores of the incomplete + * requests so that inter-timeline dependencies (i.e other timelines) + * remain correctly ordered. And we defer to __i915_request_submit() + * so that all asynchronous waits are correctly handled. + */ + GEM_TRACE("%s(%s): { rq=%llx:%lld }\n", + __func__, engine->name, rq->fence.context, rq->fence.seqno); + + /* On resubmission of the active request, payload will be scrubbed */ + rq = active_request(rq); + if (rq) + ce->ring->head = intel_ring_wrap(ce->ring, rq->head); + else + ce->ring->head = ce->ring->tail; + intel_ring_update_space(ce->ring); + + /* Scrub the context image to prevent replaying the previous batch */ + restore_default_state(ce, engine); + __execlists_update_reg_state(ce, engine); + + /* We've switched away, so this should be a no-op, but intent matters */ + ce->lrc_desc |= CTX_DESC_FORCE_RESTORE; +} + static inline void __execlists_schedule_out(struct i915_request *rq, struct intel_engine_cs * const engine) @@ -982,6 +1062,9 @@ __execlists_schedule_out(struct i915_request *rq, execlists_context_status_change(rq, INTEL_CONTEXT_SCHEDULE_OUT); intel_gt_pm_put(engine->gt); + if (unlikely(i915_gem_context_is_banned(ce->gem_context))) + reset_active(rq, engine); + /* * If this is part of a virtual engine, its next request may * have been blocked waiting for access to the active context. @@ -1380,6 +1463,10 @@ static unsigned long active_preempt_timeout(struct intel_engine_cs *engine) if (!rq) return 0; + /* Force a fast reset for terminated contexts (ignoring sysfs!) */ + if (unlikely(i915_gem_context_is_banned(rq->gem_context))) + return 1; + return READ_ONCE(engine->props.preempt_timeout_ms); } @@ -2792,29 +2879,6 @@ static void reset_csb_pointers(struct intel_engine_cs *engine) &execlists->csb_status[reset_value]); } -static struct i915_request *active_request(struct i915_request *rq) -{ - const struct intel_context * const ce = rq->hw_context; - struct i915_request *active = NULL; - struct list_head *list; - - if (!i915_request_is_active(rq)) /* unwound, but incomplete! */ - return rq; - - list = &i915_request_active_timeline(rq)->requests; - list_for_each_entry_from_reverse(rq, list, link) { - if (i915_request_completed(rq)) - break; - - if (rq->hw_context != ce) - break; - - active = rq; - } - - return active; -} - static void __execlists_reset_reg_state(const struct intel_context *ce, const struct intel_engine_cs *engine) { @@ -2831,7 +2895,6 @@ static void __execlists_reset(struct intel_engine_cs *engine, bool stalled) struct intel_engine_execlists * const execlists = &engine->execlists; struct intel_context *ce; struct i915_request *rq; - u32 *regs; mb(); /* paranoia: read the CSB pointers from after the reset */ clflush(execlists->csb_write); @@ -2907,13 +2970,7 @@ static void __execlists_reset(struct intel_engine_cs *engine, bool stalled) * to recreate its own state. */ GEM_BUG_ON(!intel_context_is_pinned(ce)); - regs = ce->lrc_reg_state; - if (engine->pinned_default_state) { - memcpy(regs, /* skip restoring the vanilla PPHWSP */ - engine->pinned_default_state + LRC_STATE_PN * PAGE_SIZE, - engine->context_size - PAGE_SIZE); - } - execlists_init_reg_state(regs, ce, engine, ce->ring, false); + restore_default_state(ce, engine); out_replay: GEM_TRACE("%s replay {head:%04x, tail:%04x\n", @@ -4574,16 +4631,8 @@ void intel_lr_context_reset(struct intel_engine_cs *engine, * future request will be after userspace has had the opportunity * to recreate its own state. */ - if (scrub) { - u32 *regs = ce->lrc_reg_state; - - if (engine->pinned_default_state) { - memcpy(regs, /* skip restoring the vanilla PPHWSP */ - engine->pinned_default_state + LRC_STATE_PN * PAGE_SIZE, - engine->context_size - PAGE_SIZE); - } - execlists_init_reg_state(regs, ce, engine, ce->ring, false); - } + if (scrub) + restore_default_state(ce, engine); /* Rerun the request; its payload has been neutered (if guilty). */ ce->ring->head = head; diff --git a/drivers/gpu/drm/i915/gt/selftest_lrc.c b/drivers/gpu/drm/i915/gt/selftest_lrc.c index b6352671c5a0..d5d268be554e 100644 --- a/drivers/gpu/drm/i915/gt/selftest_lrc.c +++ b/drivers/gpu/drm/i915/gt/selftest_lrc.c @@ -7,6 +7,7 @@ #include <linux/prime_numbers.h> #include "gem/i915_gem_pm.h" +#include "gt/intel_engine_heartbeat.h" #include "gt/intel_reset.h" #include "i915_selftest.h" @@ -1160,6 +1161,325 @@ static int live_nopreempt(void *arg) goto err_client_b; } +struct live_preempt_cancel { + struct intel_engine_cs *engine; + struct preempt_client a, b; +}; + +static int __cancel_active0(struct live_preempt_cancel *arg) +{ + struct i915_request *rq; + struct igt_live_test t; + int err; + + /* Preempt cancel of ELSP0 */ + GEM_TRACE("%s(%s)\n", __func__, arg->engine->name); + if (igt_live_test_begin(&t, arg->engine->i915, + __func__, arg->engine->name)) + return -EIO; + + clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags); + rq = spinner_create_request(&arg->a.spin, + arg->a.ctx, arg->engine, + MI_ARB_CHECK); + if (IS_ERR(rq)) + return PTR_ERR(rq); + + i915_request_get(rq); + i915_request_add(rq); + if (!igt_wait_for_spinner(&arg->a.spin, rq)) { + err = -EIO; + goto out; + } + + i915_gem_context_set_banned(arg->a.ctx); + err = intel_engine_pulse(arg->engine); + if (err) + goto out; + + if (i915_request_wait(rq, 0, HZ / 5) < 0) { + err = -EIO; + goto out; + } + + if (rq->fence.error != -EIO) { + pr_err("Cancelled inflight0 request did not report -EIO\n"); + err = -EINVAL; + goto out; + } + +out: + i915_request_put(rq); + if (igt_live_test_end(&t)) + err = -EIO; + return err; +} + +static int __cancel_active1(struct live_preempt_cancel *arg) +{ + struct i915_request *rq[2] = {}; + struct igt_live_test t; + int err; + + /* Preempt cancel of ELSP1 */ + GEM_TRACE("%s(%s)\n", __func__, arg->engine->name); + if (igt_live_test_begin(&t, arg->engine->i915, + __func__, arg->engine->name)) + return -EIO; + + clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags); + rq[0] = spinner_create_request(&arg->a.spin, + arg->a.ctx, arg->engine, + MI_NOOP); /* no preemption */ + if (IS_ERR(rq[0])) + return PTR_ERR(rq[0]); + + i915_request_get(rq[0]); + i915_request_add(rq[0]); + if (!igt_wait_for_spinner(&arg->a.spin, rq[0])) { + err = -EIO; + goto out; + } + + clear_bit(CONTEXT_BANNED, &arg->b.ctx->flags); + rq[1] = spinner_create_request(&arg->b.spin, + arg->b.ctx, arg->engine, + MI_ARB_CHECK); + if (IS_ERR(rq[1])) { + err = PTR_ERR(rq[1]); + goto out; + } + + i915_request_get(rq[1]); + err = i915_request_await_dma_fence(rq[1], &rq[0]->fence); + i915_request_add(rq[1]); + if (err) + goto out; + + i915_gem_context_set_banned(arg->b.ctx); + err = intel_engine_pulse(arg->engine); + if (err) + goto out; + + igt_spinner_end(&arg->a.spin); + if (i915_request_wait(rq[1], 0, HZ / 5) < 0) { + err = -EIO; + goto out; + } + + if (rq[0]->fence.error != 0) { + pr_err("Normal inflight0 request did not complete\n"); + err = -EINVAL; + goto out; + } + + if (rq[1]->fence.error != -EIO) { + pr_err("Cancelled inflight1 request did not report -EIO\n"); + err = -EINVAL; + goto out; + } + +out: + i915_request_put(rq[1]); + i915_request_put(rq[0]); + if (igt_live_test_end(&t)) + err = -EIO; + return err; +} + +static int __cancel_queued(struct live_preempt_cancel *arg) +{ + struct i915_request *rq[3] = {}; + struct igt_live_test t; + int err; + + /* Full ELSP and one in the wings */ + GEM_TRACE("%s(%s)\n", __func__, arg->engine->name); + if (igt_live_test_begin(&t, arg->engine->i915, + __func__, arg->engine->name)) + return -EIO; + + clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags); + rq[0] = spinner_create_request(&arg->a.spin, + arg->a.ctx, arg->engine, + MI_ARB_CHECK); + if (IS_ERR(rq[0])) + return PTR_ERR(rq[0]); + + i915_request_get(rq[0]); + i915_request_add(rq[0]); + if (!igt_wait_for_spinner(&arg->a.spin, rq[0])) { + err = -EIO; + goto out; + } + + clear_bit(CONTEXT_BANNED, &arg->b.ctx->flags); + rq[1] = igt_request_alloc(arg->b.ctx, arg->engine); + if (IS_ERR(rq[1])) { + err = PTR_ERR(rq[1]); + goto out; + } + + i915_request_get(rq[1]); + err = i915_request_await_dma_fence(rq[1], &rq[0]->fence); + i915_request_add(rq[1]); + if (err) + goto out; + + rq[2] = spinner_create_request(&arg->b.spin, + arg->a.ctx, arg->engine, + MI_ARB_CHECK); + if (IS_ERR(rq[2])) { + err = PTR_ERR(rq[2]); + goto out; + } + + i915_request_get(rq[2]); + err = i915_request_await_dma_fence(rq[2], &rq[1]->fence); + i915_request_add(rq[2]); + if (err) + goto out; + + i915_gem_context_set_banned(arg->a.ctx); + err = intel_engine_pulse(arg->engine); + if (err) + goto out; + + if (i915_request_wait(rq[2], 0, HZ / 5) < 0) { + err = -EIO; + goto out; + } + + if (rq[0]->fence.error != -EIO) { + pr_err("Cancelled inflight0 request did not report -EIO\n"); + err = -EINVAL; + goto out; + } + + if (rq[1]->fence.error != 0) { + pr_err("Normal inflight1 request did not complete\n"); + err = -EINVAL; + goto out; + } + + if (rq[2]->fence.error != -EIO) { + pr_err("Cancelled queued request did not report -EIO\n"); + err = -EINVAL; + goto out; + } + +out: + i915_request_put(rq[2]); + i915_request_put(rq[1]); + i915_request_put(rq[0]); + if (igt_live_test_end(&t)) + err = -EIO; + return err; +} + +static int __cancel_hostile(struct live_preempt_cancel *arg) +{ + struct i915_request *rq; + int err; + + /* Preempt cancel non-preemptible spinner in ELSP0 */ + if (!CONFIG_DRM_I915_PREEMPT_TIMEOUT) + return 0; + + GEM_TRACE("%s(%s)\n", __func__, arg->engine->name); + clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags); + rq = spinner_create_request(&arg->a.spin, + arg->a.ctx, arg->engine, + MI_NOOP); /* preemption disabled */ + if (IS_ERR(rq)) + return PTR_ERR(rq); + + i915_request_get(rq); + i915_request_add(rq); + if (!igt_wait_for_spinner(&arg->a.spin, rq)) { + err = -EIO; + goto out; + } + + i915_gem_context_set_banned(arg->a.ctx); + err = intel_engine_pulse(arg->engine); /* force reset */ + if (err) + goto out; + + if (i915_request_wait(rq, 0, HZ / 5) < 0) { + err = -EIO; + goto out; + } + + if (rq->fence.error != -EIO) { + pr_err("Cancelled inflight0 request did not report -EIO\n"); + err = -EINVAL; + goto out; + } + +out: + i915_request_put(rq); + if (igt_flush_test(arg->engine->i915)) + err = -EIO; + return err; +} + +static int live_preempt_cancel(void *arg) +{ + struct intel_gt *gt = arg; + struct live_preempt_cancel data; + enum intel_engine_id id; + int err = -ENOMEM; + + /* + * To cancel an inflight context, we need to first remove it from the + * GPU. That sounds like preemption! Plus a little bit of bookkeeping. + */ + + if (!HAS_LOGICAL_RING_PREEMPTION(gt->i915)) + return 0; + + if (preempt_client_init(gt, &data.a)) + return -ENOMEM; + if (preempt_client_init(gt, &data.b)) + goto err_client_a; + + for_each_engine(data.engine, gt, id) { + if (!intel_engine_has_preemption(data.engine)) + continue; + + err = __cancel_active0(&data); + if (err) + goto err_wedged; + + err = __cancel_active1(&data); + if (err) + goto err_wedged; + + err = __cancel_queued(&data); + if (err) + goto err_wedged; + + err = __cancel_hostile(&data); + if (err) + goto err_wedged; + } + + err = 0; +err_client_b: + preempt_client_fini(&data.b); +err_client_a: + preempt_client_fini(&data.a); + return err; + +err_wedged: + GEM_TRACE_DUMP(); + igt_spinner_end(&data.b.spin); + igt_spinner_end(&data.a.spin); + intel_gt_set_wedged(gt); + goto err_client_b; +} + static int live_suppress_self_preempt(void *arg) { struct intel_gt *gt = arg; @@ -2693,6 +3013,7 @@ int intel_execlists_live_selftests(struct drm_i915_private *i915) SUBTEST(live_preempt), SUBTEST(live_late_preempt), SUBTEST(live_nopreempt), + SUBTEST(live_preempt_cancel), SUBTEST(live_suppress_self_preempt), SUBTEST(live_suppress_wait_preempt), SUBTEST(live_chain_preempt), -- 2.24.0.rc0 _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply related [flat|nested] 18+ messages in thread
* [Intel-gfx] [CI 3/5] drm/i915/execlists: Cancel banned contexts on schedule-out 2019-10-23 13:31 ` [CI 3/5] drm/i915/execlists: Cancel banned contexts on schedule-out Chris Wilson @ 2019-10-23 13:31 ` Chris Wilson 0 siblings, 0 replies; 18+ messages in thread From: Chris Wilson @ 2019-10-23 13:31 UTC (permalink / raw) To: intel-gfx On schedule-out (CS completion) of a banned context, scrub the context image so that we do not replay the active payload. The intent is that we skip banned payloads on request submission so that the timeline advancement continues on in the background. However, if we are returning to a preempted request, i915_request_skip() is ineffective and instead we need to patch up the context image so that it continues from the start of the next request. v2: Fixup cancellation so that we only scrub the payload of the active request and do not short-circuit the breadcrumbs (which might cause other contexts to execute out of order). v3: Grammar pass Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> --- drivers/gpu/drm/i915/gt/intel_lrc.c | 131 ++++++---- drivers/gpu/drm/i915/gt/selftest_lrc.c | 321 +++++++++++++++++++++++++ 2 files changed, 411 insertions(+), 41 deletions(-) diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c index ff0dd297e782..651d5dd39464 100644 --- a/drivers/gpu/drm/i915/gt/intel_lrc.c +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c @@ -234,6 +234,9 @@ static void execlists_init_reg_state(u32 *reg_state, const struct intel_engine_cs *engine, const struct intel_ring *ring, bool close); +static void +__execlists_update_reg_state(const struct intel_context *ce, + const struct intel_engine_cs *engine); static void mark_eio(struct i915_request *rq) { @@ -246,6 +249,31 @@ static void mark_eio(struct i915_request *rq) i915_request_mark_complete(rq); } +static struct i915_request *active_request(struct i915_request *rq) +{ + const struct intel_context * const ce = rq->hw_context; + struct i915_request *active = NULL; + struct list_head *list; + + if (!i915_request_is_active(rq)) /* unwound, but incomplete! */ + return rq; + + rcu_read_lock(); + list = &rcu_dereference(rq->timeline)->requests; + list_for_each_entry_from_reverse(rq, list, link) { + if (i915_request_completed(rq)) + break; + + if (rq->hw_context != ce) + break; + + active = rq; + } + rcu_read_unlock(); + + return active; +} + static inline u32 intel_hws_preempt_address(struct intel_engine_cs *engine) { return (i915_ggtt_offset(engine->status_page.vma) + @@ -972,6 +1000,58 @@ static void kick_siblings(struct i915_request *rq, struct intel_context *ce) tasklet_schedule(&ve->base.execlists.tasklet); } +static void restore_default_state(struct intel_context *ce, + struct intel_engine_cs *engine) +{ + u32 *regs = ce->lrc_reg_state; + + if (engine->pinned_default_state) + memcpy(regs, /* skip restoring the vanilla PPHWSP */ + engine->pinned_default_state + LRC_STATE_PN * PAGE_SIZE, + engine->context_size - PAGE_SIZE); + + execlists_init_reg_state(regs, ce, engine, ce->ring, false); +} + +static void reset_active(struct i915_request *rq, + struct intel_engine_cs *engine) +{ + struct intel_context * const ce = rq->hw_context; + + /* + * The executing context has been cancelled. We want to prevent + * further execution along this context and propagate the error on + * to anything depending on its results. + * + * In __i915_request_submit(), we apply the -EIO and remove the + * requests' payloads for any banned requests. But first, we must + * rewind the context back to the start of the incomplete request so + * that we do not jump back into the middle of the batch. + * + * We preserve the breadcrumbs and semaphores of the incomplete + * requests so that inter-timeline dependencies (i.e other timelines) + * remain correctly ordered. And we defer to __i915_request_submit() + * so that all asynchronous waits are correctly handled. + */ + GEM_TRACE("%s(%s): { rq=%llx:%lld }\n", + __func__, engine->name, rq->fence.context, rq->fence.seqno); + + /* On resubmission of the active request, payload will be scrubbed */ + rq = active_request(rq); + if (rq) + ce->ring->head = intel_ring_wrap(ce->ring, rq->head); + else + ce->ring->head = ce->ring->tail; + intel_ring_update_space(ce->ring); + + /* Scrub the context image to prevent replaying the previous batch */ + restore_default_state(ce, engine); + __execlists_update_reg_state(ce, engine); + + /* We've switched away, so this should be a no-op, but intent matters */ + ce->lrc_desc |= CTX_DESC_FORCE_RESTORE; +} + static inline void __execlists_schedule_out(struct i915_request *rq, struct intel_engine_cs * const engine) @@ -982,6 +1062,9 @@ __execlists_schedule_out(struct i915_request *rq, execlists_context_status_change(rq, INTEL_CONTEXT_SCHEDULE_OUT); intel_gt_pm_put(engine->gt); + if (unlikely(i915_gem_context_is_banned(ce->gem_context))) + reset_active(rq, engine); + /* * If this is part of a virtual engine, its next request may * have been blocked waiting for access to the active context. @@ -1380,6 +1463,10 @@ static unsigned long active_preempt_timeout(struct intel_engine_cs *engine) if (!rq) return 0; + /* Force a fast reset for terminated contexts (ignoring sysfs!) */ + if (unlikely(i915_gem_context_is_banned(rq->gem_context))) + return 1; + return READ_ONCE(engine->props.preempt_timeout_ms); } @@ -2792,29 +2879,6 @@ static void reset_csb_pointers(struct intel_engine_cs *engine) &execlists->csb_status[reset_value]); } -static struct i915_request *active_request(struct i915_request *rq) -{ - const struct intel_context * const ce = rq->hw_context; - struct i915_request *active = NULL; - struct list_head *list; - - if (!i915_request_is_active(rq)) /* unwound, but incomplete! */ - return rq; - - list = &i915_request_active_timeline(rq)->requests; - list_for_each_entry_from_reverse(rq, list, link) { - if (i915_request_completed(rq)) - break; - - if (rq->hw_context != ce) - break; - - active = rq; - } - - return active; -} - static void __execlists_reset_reg_state(const struct intel_context *ce, const struct intel_engine_cs *engine) { @@ -2831,7 +2895,6 @@ static void __execlists_reset(struct intel_engine_cs *engine, bool stalled) struct intel_engine_execlists * const execlists = &engine->execlists; struct intel_context *ce; struct i915_request *rq; - u32 *regs; mb(); /* paranoia: read the CSB pointers from after the reset */ clflush(execlists->csb_write); @@ -2907,13 +2970,7 @@ static void __execlists_reset(struct intel_engine_cs *engine, bool stalled) * to recreate its own state. */ GEM_BUG_ON(!intel_context_is_pinned(ce)); - regs = ce->lrc_reg_state; - if (engine->pinned_default_state) { - memcpy(regs, /* skip restoring the vanilla PPHWSP */ - engine->pinned_default_state + LRC_STATE_PN * PAGE_SIZE, - engine->context_size - PAGE_SIZE); - } - execlists_init_reg_state(regs, ce, engine, ce->ring, false); + restore_default_state(ce, engine); out_replay: GEM_TRACE("%s replay {head:%04x, tail:%04x\n", @@ -4574,16 +4631,8 @@ void intel_lr_context_reset(struct intel_engine_cs *engine, * future request will be after userspace has had the opportunity * to recreate its own state. */ - if (scrub) { - u32 *regs = ce->lrc_reg_state; - - if (engine->pinned_default_state) { - memcpy(regs, /* skip restoring the vanilla PPHWSP */ - engine->pinned_default_state + LRC_STATE_PN * PAGE_SIZE, - engine->context_size - PAGE_SIZE); - } - execlists_init_reg_state(regs, ce, engine, ce->ring, false); - } + if (scrub) + restore_default_state(ce, engine); /* Rerun the request; its payload has been neutered (if guilty). */ ce->ring->head = head; diff --git a/drivers/gpu/drm/i915/gt/selftest_lrc.c b/drivers/gpu/drm/i915/gt/selftest_lrc.c index b6352671c5a0..d5d268be554e 100644 --- a/drivers/gpu/drm/i915/gt/selftest_lrc.c +++ b/drivers/gpu/drm/i915/gt/selftest_lrc.c @@ -7,6 +7,7 @@ #include <linux/prime_numbers.h> #include "gem/i915_gem_pm.h" +#include "gt/intel_engine_heartbeat.h" #include "gt/intel_reset.h" #include "i915_selftest.h" @@ -1160,6 +1161,325 @@ static int live_nopreempt(void *arg) goto err_client_b; } +struct live_preempt_cancel { + struct intel_engine_cs *engine; + struct preempt_client a, b; +}; + +static int __cancel_active0(struct live_preempt_cancel *arg) +{ + struct i915_request *rq; + struct igt_live_test t; + int err; + + /* Preempt cancel of ELSP0 */ + GEM_TRACE("%s(%s)\n", __func__, arg->engine->name); + if (igt_live_test_begin(&t, arg->engine->i915, + __func__, arg->engine->name)) + return -EIO; + + clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags); + rq = spinner_create_request(&arg->a.spin, + arg->a.ctx, arg->engine, + MI_ARB_CHECK); + if (IS_ERR(rq)) + return PTR_ERR(rq); + + i915_request_get(rq); + i915_request_add(rq); + if (!igt_wait_for_spinner(&arg->a.spin, rq)) { + err = -EIO; + goto out; + } + + i915_gem_context_set_banned(arg->a.ctx); + err = intel_engine_pulse(arg->engine); + if (err) + goto out; + + if (i915_request_wait(rq, 0, HZ / 5) < 0) { + err = -EIO; + goto out; + } + + if (rq->fence.error != -EIO) { + pr_err("Cancelled inflight0 request did not report -EIO\n"); + err = -EINVAL; + goto out; + } + +out: + i915_request_put(rq); + if (igt_live_test_end(&t)) + err = -EIO; + return err; +} + +static int __cancel_active1(struct live_preempt_cancel *arg) +{ + struct i915_request *rq[2] = {}; + struct igt_live_test t; + int err; + + /* Preempt cancel of ELSP1 */ + GEM_TRACE("%s(%s)\n", __func__, arg->engine->name); + if (igt_live_test_begin(&t, arg->engine->i915, + __func__, arg->engine->name)) + return -EIO; + + clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags); + rq[0] = spinner_create_request(&arg->a.spin, + arg->a.ctx, arg->engine, + MI_NOOP); /* no preemption */ + if (IS_ERR(rq[0])) + return PTR_ERR(rq[0]); + + i915_request_get(rq[0]); + i915_request_add(rq[0]); + if (!igt_wait_for_spinner(&arg->a.spin, rq[0])) { + err = -EIO; + goto out; + } + + clear_bit(CONTEXT_BANNED, &arg->b.ctx->flags); + rq[1] = spinner_create_request(&arg->b.spin, + arg->b.ctx, arg->engine, + MI_ARB_CHECK); + if (IS_ERR(rq[1])) { + err = PTR_ERR(rq[1]); + goto out; + } + + i915_request_get(rq[1]); + err = i915_request_await_dma_fence(rq[1], &rq[0]->fence); + i915_request_add(rq[1]); + if (err) + goto out; + + i915_gem_context_set_banned(arg->b.ctx); + err = intel_engine_pulse(arg->engine); + if (err) + goto out; + + igt_spinner_end(&arg->a.spin); + if (i915_request_wait(rq[1], 0, HZ / 5) < 0) { + err = -EIO; + goto out; + } + + if (rq[0]->fence.error != 0) { + pr_err("Normal inflight0 request did not complete\n"); + err = -EINVAL; + goto out; + } + + if (rq[1]->fence.error != -EIO) { + pr_err("Cancelled inflight1 request did not report -EIO\n"); + err = -EINVAL; + goto out; + } + +out: + i915_request_put(rq[1]); + i915_request_put(rq[0]); + if (igt_live_test_end(&t)) + err = -EIO; + return err; +} + +static int __cancel_queued(struct live_preempt_cancel *arg) +{ + struct i915_request *rq[3] = {}; + struct igt_live_test t; + int err; + + /* Full ELSP and one in the wings */ + GEM_TRACE("%s(%s)\n", __func__, arg->engine->name); + if (igt_live_test_begin(&t, arg->engine->i915, + __func__, arg->engine->name)) + return -EIO; + + clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags); + rq[0] = spinner_create_request(&arg->a.spin, + arg->a.ctx, arg->engine, + MI_ARB_CHECK); + if (IS_ERR(rq[0])) + return PTR_ERR(rq[0]); + + i915_request_get(rq[0]); + i915_request_add(rq[0]); + if (!igt_wait_for_spinner(&arg->a.spin, rq[0])) { + err = -EIO; + goto out; + } + + clear_bit(CONTEXT_BANNED, &arg->b.ctx->flags); + rq[1] = igt_request_alloc(arg->b.ctx, arg->engine); + if (IS_ERR(rq[1])) { + err = PTR_ERR(rq[1]); + goto out; + } + + i915_request_get(rq[1]); + err = i915_request_await_dma_fence(rq[1], &rq[0]->fence); + i915_request_add(rq[1]); + if (err) + goto out; + + rq[2] = spinner_create_request(&arg->b.spin, + arg->a.ctx, arg->engine, + MI_ARB_CHECK); + if (IS_ERR(rq[2])) { + err = PTR_ERR(rq[2]); + goto out; + } + + i915_request_get(rq[2]); + err = i915_request_await_dma_fence(rq[2], &rq[1]->fence); + i915_request_add(rq[2]); + if (err) + goto out; + + i915_gem_context_set_banned(arg->a.ctx); + err = intel_engine_pulse(arg->engine); + if (err) + goto out; + + if (i915_request_wait(rq[2], 0, HZ / 5) < 0) { + err = -EIO; + goto out; + } + + if (rq[0]->fence.error != -EIO) { + pr_err("Cancelled inflight0 request did not report -EIO\n"); + err = -EINVAL; + goto out; + } + + if (rq[1]->fence.error != 0) { + pr_err("Normal inflight1 request did not complete\n"); + err = -EINVAL; + goto out; + } + + if (rq[2]->fence.error != -EIO) { + pr_err("Cancelled queued request did not report -EIO\n"); + err = -EINVAL; + goto out; + } + +out: + i915_request_put(rq[2]); + i915_request_put(rq[1]); + i915_request_put(rq[0]); + if (igt_live_test_end(&t)) + err = -EIO; + return err; +} + +static int __cancel_hostile(struct live_preempt_cancel *arg) +{ + struct i915_request *rq; + int err; + + /* Preempt cancel non-preemptible spinner in ELSP0 */ + if (!CONFIG_DRM_I915_PREEMPT_TIMEOUT) + return 0; + + GEM_TRACE("%s(%s)\n", __func__, arg->engine->name); + clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags); + rq = spinner_create_request(&arg->a.spin, + arg->a.ctx, arg->engine, + MI_NOOP); /* preemption disabled */ + if (IS_ERR(rq)) + return PTR_ERR(rq); + + i915_request_get(rq); + i915_request_add(rq); + if (!igt_wait_for_spinner(&arg->a.spin, rq)) { + err = -EIO; + goto out; + } + + i915_gem_context_set_banned(arg->a.ctx); + err = intel_engine_pulse(arg->engine); /* force reset */ + if (err) + goto out; + + if (i915_request_wait(rq, 0, HZ / 5) < 0) { + err = -EIO; + goto out; + } + + if (rq->fence.error != -EIO) { + pr_err("Cancelled inflight0 request did not report -EIO\n"); + err = -EINVAL; + goto out; + } + +out: + i915_request_put(rq); + if (igt_flush_test(arg->engine->i915)) + err = -EIO; + return err; +} + +static int live_preempt_cancel(void *arg) +{ + struct intel_gt *gt = arg; + struct live_preempt_cancel data; + enum intel_engine_id id; + int err = -ENOMEM; + + /* + * To cancel an inflight context, we need to first remove it from the + * GPU. That sounds like preemption! Plus a little bit of bookkeeping. + */ + + if (!HAS_LOGICAL_RING_PREEMPTION(gt->i915)) + return 0; + + if (preempt_client_init(gt, &data.a)) + return -ENOMEM; + if (preempt_client_init(gt, &data.b)) + goto err_client_a; + + for_each_engine(data.engine, gt, id) { + if (!intel_engine_has_preemption(data.engine)) + continue; + + err = __cancel_active0(&data); + if (err) + goto err_wedged; + + err = __cancel_active1(&data); + if (err) + goto err_wedged; + + err = __cancel_queued(&data); + if (err) + goto err_wedged; + + err = __cancel_hostile(&data); + if (err) + goto err_wedged; + } + + err = 0; +err_client_b: + preempt_client_fini(&data.b); +err_client_a: + preempt_client_fini(&data.a); + return err; + +err_wedged: + GEM_TRACE_DUMP(); + igt_spinner_end(&data.b.spin); + igt_spinner_end(&data.a.spin); + intel_gt_set_wedged(gt); + goto err_client_b; +} + static int live_suppress_self_preempt(void *arg) { struct intel_gt *gt = arg; @@ -2693,6 +3013,7 @@ int intel_execlists_live_selftests(struct drm_i915_private *i915) SUBTEST(live_preempt), SUBTEST(live_late_preempt), SUBTEST(live_nopreempt), + SUBTEST(live_preempt_cancel), SUBTEST(live_suppress_self_preempt), SUBTEST(live_suppress_wait_preempt), SUBTEST(live_chain_preempt), -- 2.24.0.rc0 _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply related [flat|nested] 18+ messages in thread
* [CI 4/5] drm/i915/gem: Cancel contexts when hangchecking is disabled 2019-10-23 13:31 [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets Chris Wilson ` (2 preceding siblings ...) 2019-10-23 13:31 ` [CI 3/5] drm/i915/execlists: Cancel banned contexts on schedule-out Chris Wilson @ 2019-10-23 13:31 ` Chris Wilson 2019-10-23 13:31 ` [Intel-gfx] " Chris Wilson 2019-10-23 13:31 ` [CI 5/5] drm/i915/gt: Replace hangcheck by heartbeats Chris Wilson ` (2 subsequent siblings) 6 siblings, 1 reply; 18+ messages in thread From: Chris Wilson @ 2019-10-23 13:31 UTC (permalink / raw) To: intel-gfx Normally, we rely on our hangcheck to prevent persistent batches from hogging the GPU. However, if the user disables hangcheck, this mechanism breaks down. Despite our insistence that this is unsafe, the users are equally insistent that they want to use endless batches and will disable the hangcheck mechanism. We are looking at replacing hangcheck, in the next patch, with a softer mechanism, that sends a pulse down the engine to check if it is well. We can use the same preemptive pulse to flush an active context off the GPU upon context close, preventing resources being lost and unkillable requests remaining on the GPU after process termination. Testcase: igt/gem_ctx_exec/basic-nohangcheck Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Cc: Michał Winiarski <michal.winiarski@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> --- drivers/gpu/drm/i915/gem/i915_gem_context.c | 141 ++++++++++++++++++++ 1 file changed, 141 insertions(+) diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c b/drivers/gpu/drm/i915/gem/i915_gem_context.c index 7b01f4605f21..b2f042d87be0 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_context.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c @@ -70,6 +70,7 @@ #include <drm/i915_drm.h> #include "gt/intel_lrc_reg.h" +#include "gt/intel_engine_heartbeat.h" #include "gt/intel_engine_user.h" #include "i915_gem_context.h" @@ -276,6 +277,135 @@ void i915_gem_context_release(struct kref *ref) schedule_work(&gc->free_work); } +static inline struct i915_gem_engines * +__context_engines_static(const struct i915_gem_context *ctx) +{ + return rcu_dereference_protected(ctx->engines, true); +} + +static bool __reset_engine(struct intel_engine_cs *engine) +{ + struct intel_gt *gt = engine->gt; + bool success = false; + + if (!intel_has_reset_engine(gt)) + return false; + + if (!test_and_set_bit(I915_RESET_ENGINE + engine->id, + >->reset.flags)) { + success = intel_engine_reset(engine, NULL) == 0; + clear_and_wake_up_bit(I915_RESET_ENGINE + engine->id, + >->reset.flags); + } + + return success; +} + +static void __reset_context(struct i915_gem_context *ctx, + struct intel_engine_cs *engine) +{ + intel_gt_handle_error(engine->gt, engine->mask, 0, + "context closure in %s", ctx->name); +} + +static bool __cancel_engine(struct intel_engine_cs *engine) +{ + /* + * Send a "high priority pulse" down the engine to cause the + * current request to be momentarily preempted. (If it fails to + * be preempted, it will be reset). As we have marked our context + * as banned, any incomplete request, including any running, will + * be skipped following the preemption. + * + * If there is no hangchecking (one of the reasons why we try to + * cancel the context) and no forced preemption, there may be no + * means by which we reset the GPU and evict the persistent hog. + * Ergo if we are unable to inject a preemptive pulse that can + * kill the banned context, we fallback to doing a local reset + * instead. + */ + if (CONFIG_DRM_I915_PREEMPT_TIMEOUT && !intel_engine_pulse(engine)) + return true; + + /* If we are unable to send a pulse, try resetting this engine. */ + return __reset_engine(engine); +} + +static struct intel_engine_cs * +active_engine(struct dma_fence *fence, struct intel_context *ce) +{ + struct i915_request *rq = to_request(fence); + struct intel_engine_cs *engine, *locked; + + /* + * Serialise with __i915_request_submit() so that it sees + * is-banned?, or we know the request is already inflight. + */ + locked = READ_ONCE(rq->engine); + spin_lock_irq(&locked->active.lock); + while (unlikely(locked != (engine = READ_ONCE(rq->engine)))) { + spin_unlock(&locked->active.lock); + spin_lock(&engine->active.lock); + locked = engine; + } + + engine = NULL; + if (i915_request_is_active(rq) && !rq->fence.error) + engine = rq->engine; + + spin_unlock_irq(&locked->active.lock); + + return engine; +} + +static void kill_context(struct i915_gem_context *ctx) +{ + struct i915_gem_engines_iter it; + struct intel_context *ce; + + /* + * If we are already banned, it was due to a guilty request causing + * a reset and the entire context being evicted from the GPU. + */ + if (i915_gem_context_is_banned(ctx)) + return; + + i915_gem_context_set_banned(ctx); + + /* + * Map the user's engine back to the actual engines; one virtual + * engine will be mapped to multiple engines, and using ctx->engine[] + * the same engine may be have multiple instances in the user's map. + * However, we only care about pending requests, so only include + * engines on which there are incomplete requests. + */ + for_each_gem_engine(ce, __context_engines_static(ctx), it) { + struct intel_engine_cs *engine; + struct dma_fence *fence; + + if (!ce->timeline) + continue; + + fence = i915_active_fence_get(&ce->timeline->last_request); + if (!fence) + continue; + + /* Check with the backend if the request is still inflight */ + engine = active_engine(fence, ce); + + /* First attempt to gracefully cancel the context */ + if (engine && !__cancel_engine(engine)) + /* + * If we are unable to send a preemptive pulse to bump + * the context from the GPU, we have to resort to a full + * reset. We hope the collateral damage is worth it. + */ + __reset_context(ctx, engine); + + dma_fence_put(fence); + } +} + static void context_close(struct i915_gem_context *ctx) { struct i915_address_space *vm; @@ -298,6 +428,17 @@ static void context_close(struct i915_gem_context *ctx) lut_close(ctx); mutex_unlock(&ctx->mutex); + + /* + * If the user has disabled hangchecking, we can not be sure that + * the batches will ever complete after the context is closed, + * keeping the context and all resources pinned forever. So in this + * case we opt to forcibly kill off all remaining requests on + * context close. + */ + if (!i915_modparams.enable_hangcheck) + kill_context(ctx); + i915_gem_context_put(ctx); } -- 2.24.0.rc0 _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply related [flat|nested] 18+ messages in thread
* [Intel-gfx] [CI 4/5] drm/i915/gem: Cancel contexts when hangchecking is disabled 2019-10-23 13:31 ` [CI 4/5] drm/i915/gem: Cancel contexts when hangchecking is disabled Chris Wilson @ 2019-10-23 13:31 ` Chris Wilson 0 siblings, 0 replies; 18+ messages in thread From: Chris Wilson @ 2019-10-23 13:31 UTC (permalink / raw) To: intel-gfx Normally, we rely on our hangcheck to prevent persistent batches from hogging the GPU. However, if the user disables hangcheck, this mechanism breaks down. Despite our insistence that this is unsafe, the users are equally insistent that they want to use endless batches and will disable the hangcheck mechanism. We are looking at replacing hangcheck, in the next patch, with a softer mechanism, that sends a pulse down the engine to check if it is well. We can use the same preemptive pulse to flush an active context off the GPU upon context close, preventing resources being lost and unkillable requests remaining on the GPU after process termination. Testcase: igt/gem_ctx_exec/basic-nohangcheck Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Cc: Michał Winiarski <michal.winiarski@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> --- drivers/gpu/drm/i915/gem/i915_gem_context.c | 141 ++++++++++++++++++++ 1 file changed, 141 insertions(+) diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c b/drivers/gpu/drm/i915/gem/i915_gem_context.c index 7b01f4605f21..b2f042d87be0 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_context.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c @@ -70,6 +70,7 @@ #include <drm/i915_drm.h> #include "gt/intel_lrc_reg.h" +#include "gt/intel_engine_heartbeat.h" #include "gt/intel_engine_user.h" #include "i915_gem_context.h" @@ -276,6 +277,135 @@ void i915_gem_context_release(struct kref *ref) schedule_work(&gc->free_work); } +static inline struct i915_gem_engines * +__context_engines_static(const struct i915_gem_context *ctx) +{ + return rcu_dereference_protected(ctx->engines, true); +} + +static bool __reset_engine(struct intel_engine_cs *engine) +{ + struct intel_gt *gt = engine->gt; + bool success = false; + + if (!intel_has_reset_engine(gt)) + return false; + + if (!test_and_set_bit(I915_RESET_ENGINE + engine->id, + >->reset.flags)) { + success = intel_engine_reset(engine, NULL) == 0; + clear_and_wake_up_bit(I915_RESET_ENGINE + engine->id, + >->reset.flags); + } + + return success; +} + +static void __reset_context(struct i915_gem_context *ctx, + struct intel_engine_cs *engine) +{ + intel_gt_handle_error(engine->gt, engine->mask, 0, + "context closure in %s", ctx->name); +} + +static bool __cancel_engine(struct intel_engine_cs *engine) +{ + /* + * Send a "high priority pulse" down the engine to cause the + * current request to be momentarily preempted. (If it fails to + * be preempted, it will be reset). As we have marked our context + * as banned, any incomplete request, including any running, will + * be skipped following the preemption. + * + * If there is no hangchecking (one of the reasons why we try to + * cancel the context) and no forced preemption, there may be no + * means by which we reset the GPU and evict the persistent hog. + * Ergo if we are unable to inject a preemptive pulse that can + * kill the banned context, we fallback to doing a local reset + * instead. + */ + if (CONFIG_DRM_I915_PREEMPT_TIMEOUT && !intel_engine_pulse(engine)) + return true; + + /* If we are unable to send a pulse, try resetting this engine. */ + return __reset_engine(engine); +} + +static struct intel_engine_cs * +active_engine(struct dma_fence *fence, struct intel_context *ce) +{ + struct i915_request *rq = to_request(fence); + struct intel_engine_cs *engine, *locked; + + /* + * Serialise with __i915_request_submit() so that it sees + * is-banned?, or we know the request is already inflight. + */ + locked = READ_ONCE(rq->engine); + spin_lock_irq(&locked->active.lock); + while (unlikely(locked != (engine = READ_ONCE(rq->engine)))) { + spin_unlock(&locked->active.lock); + spin_lock(&engine->active.lock); + locked = engine; + } + + engine = NULL; + if (i915_request_is_active(rq) && !rq->fence.error) + engine = rq->engine; + + spin_unlock_irq(&locked->active.lock); + + return engine; +} + +static void kill_context(struct i915_gem_context *ctx) +{ + struct i915_gem_engines_iter it; + struct intel_context *ce; + + /* + * If we are already banned, it was due to a guilty request causing + * a reset and the entire context being evicted from the GPU. + */ + if (i915_gem_context_is_banned(ctx)) + return; + + i915_gem_context_set_banned(ctx); + + /* + * Map the user's engine back to the actual engines; one virtual + * engine will be mapped to multiple engines, and using ctx->engine[] + * the same engine may be have multiple instances in the user's map. + * However, we only care about pending requests, so only include + * engines on which there are incomplete requests. + */ + for_each_gem_engine(ce, __context_engines_static(ctx), it) { + struct intel_engine_cs *engine; + struct dma_fence *fence; + + if (!ce->timeline) + continue; + + fence = i915_active_fence_get(&ce->timeline->last_request); + if (!fence) + continue; + + /* Check with the backend if the request is still inflight */ + engine = active_engine(fence, ce); + + /* First attempt to gracefully cancel the context */ + if (engine && !__cancel_engine(engine)) + /* + * If we are unable to send a preemptive pulse to bump + * the context from the GPU, we have to resort to a full + * reset. We hope the collateral damage is worth it. + */ + __reset_context(ctx, engine); + + dma_fence_put(fence); + } +} + static void context_close(struct i915_gem_context *ctx) { struct i915_address_space *vm; @@ -298,6 +428,17 @@ static void context_close(struct i915_gem_context *ctx) lut_close(ctx); mutex_unlock(&ctx->mutex); + + /* + * If the user has disabled hangchecking, we can not be sure that + * the batches will ever complete after the context is closed, + * keeping the context and all resources pinned forever. So in this + * case we opt to forcibly kill off all remaining requests on + * context close. + */ + if (!i915_modparams.enable_hangcheck) + kill_context(ctx); + i915_gem_context_put(ctx); } -- 2.24.0.rc0 _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply related [flat|nested] 18+ messages in thread
* [CI 5/5] drm/i915/gt: Replace hangcheck by heartbeats 2019-10-23 13:31 [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets Chris Wilson ` (3 preceding siblings ...) 2019-10-23 13:31 ` [CI 4/5] drm/i915/gem: Cancel contexts when hangchecking is disabled Chris Wilson @ 2019-10-23 13:31 ` Chris Wilson 2019-10-23 13:31 ` [Intel-gfx] " Chris Wilson 2019-10-24 0:09 ` ✗ Fi.CI.BUILD: failure for series starting with [CI,1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets Patchwork 2019-10-24 1:32 ` [CI 1/5] " Ramalingam C 6 siblings, 1 reply; 18+ messages in thread From: Chris Wilson @ 2019-10-23 13:31 UTC (permalink / raw) To: intel-gfx 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> --- drivers/gpu/drm/i915/Kconfig.profile | 11 + drivers/gpu/drm/i915/Makefile | 1 - drivers/gpu/drm/i915/display/intel_display.c | 2 +- drivers/gpu/drm/i915/gem/i915_gem_object.h | 1 - drivers/gpu/drm/i915/gem/i915_gem_pm.c | 2 - drivers/gpu/drm/i915/gt/intel_engine.h | 32 -- drivers/gpu/drm/i915/gt/intel_engine_cs.c | 12 +- .../gpu/drm/i915/gt/intel_engine_heartbeat.c | 157 ++++++++ .../gpu/drm/i915/gt/intel_engine_heartbeat.h | 8 + drivers/gpu/drm/i915/gt/intel_engine_pm.c | 5 +- drivers/gpu/drm/i915/gt/intel_engine_types.h | 17 +- drivers/gpu/drm/i915/gt/intel_gt.c | 1 - drivers/gpu/drm/i915/gt/intel_gt.h | 4 - drivers/gpu/drm/i915/gt/intel_gt_pm.c | 1 - drivers/gpu/drm/i915/gt/intel_gt_types.h | 9 - drivers/gpu/drm/i915/gt/intel_hangcheck.c | 361 ------------------ drivers/gpu/drm/i915/gt/intel_reset.c | 3 +- .../drm/i915/gt/selftest_engine_heartbeat.c | 171 +++++++++ drivers/gpu/drm/i915/gt/selftest_hangcheck.c | 4 - drivers/gpu/drm/i915/i915_debugfs.c | 87 ----- drivers/gpu/drm/i915/i915_drv.c | 3 - drivers/gpu/drm/i915/i915_drv.h | 1 - drivers/gpu/drm/i915/i915_gpu_error.c | 33 +- drivers/gpu/drm/i915/i915_gpu_error.h | 2 - drivers/gpu/drm/i915/i915_priolist_types.h | 6 + 25 files changed, 379 insertions(+), 555 deletions(-) delete mode 100644 drivers/gpu/drm/i915/gt/intel_hangcheck.c diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile index b071b6024152..8ab7af5eb311 100644 --- a/drivers/gpu/drm/i915/Kconfig.profile +++ b/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 diff --git a/drivers/gpu/drm/i915/Makefile b/drivers/gpu/drm/i915/Makefile index 2fd4bed188e5..21601bb27b22 100644 --- a/drivers/gpu/drm/i915/Makefile +++ b/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 \ diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c index 236fdf122e47..579655675b08 100644 --- a/drivers/gpu/drm/i915/display/intel_display.c +++ b/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); diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.h b/drivers/gpu/drm/i915/gem/i915_gem_object.h index aead7e6725f9..458cd51331f1 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_object.h +++ b/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 diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pm.c b/drivers/gpu/drm/i915/gem/i915_gem_pm.c index 9bc0cf3139e3..c99bb94fe41e 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_pm.c +++ b/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); } diff --git a/drivers/gpu/drm/i915/gt/intel_engine.h b/drivers/gpu/drm/i915/gt/intel_engine.h index 9409b7856299..ee47444a6ad4 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine.h +++ b/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) diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c index b91ea07f4819..8b5129e0b49d 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c +++ b/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)); diff --git a/drivers/gpu/drm/i915/gt/intel_engine_heartbeat.c b/drivers/gpu/drm/i915/gt/intel_engine_heartbeat.c index 4b9ab7813d54..9977f59f6b53 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_heartbeat.c +++ b/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 }; diff --git a/drivers/gpu/drm/i915/gt/intel_engine_heartbeat.h b/drivers/gpu/drm/i915/gt/intel_engine_heartbeat.h index b334e5aaf78d..a7b8c0f9e005 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_heartbeat.h +++ b/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); diff --git a/drivers/gpu/drm/i915/gt/intel_engine_pm.c b/drivers/gpu/drm/i915/gt/intel_engine_pm.c index 7d76611d9df1..6fbfa2162e54 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_pm.c +++ b/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) diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h index 1251dac91f31..26998901feff 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_types.h +++ b/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; diff --git a/drivers/gpu/drm/i915/gt/intel_gt.c b/drivers/gpu/drm/i915/gt/intel_gt.c index 1c4b6c9642ad..35127699591d 100644 --- a/drivers/gpu/drm/i915/gt/intel_gt.c +++ b/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(>->closed_vma); spin_lock_init(>->closed_lock); - intel_gt_init_hangcheck(gt); intel_gt_init_reset(gt); intel_gt_init_requests(gt); intel_gt_pm_init_early(gt); diff --git a/drivers/gpu/drm/i915/gt/intel_gt.h b/drivers/gpu/drm/i915/gt/intel_gt.h index e6ab0bff0efb..5b6effed3713 100644 --- a/drivers/gpu/drm/i915/gt/intel_gt.h +++ b/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(>->reset); } -void intel_gt_queue_hangcheck(struct intel_gt *gt); - #endif /* __INTEL_GT_H__ */ diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm.c b/drivers/gpu/drm/i915/gt/intel_gt_pm.c index 06e73d56cfcf..aff4eda47819 100644 --- a/drivers/gpu/drm/i915/gt/intel_gt_pm.c +++ b/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; diff --git a/drivers/gpu/drm/i915/gt/intel_gt_types.h b/drivers/gpu/drm/i915/gt/intel_gt_types.h index 980973e66e7f..fcb3a6c9421c 100644 --- a/drivers/gpu/drm/i915/gt/intel_gt_types.h +++ b/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; /** diff --git a/drivers/gpu/drm/i915/gt/intel_hangcheck.c b/drivers/gpu/drm/i915/gt/intel_hangcheck.c deleted file mode 100644 index 0fdef00af9e4..000000000000 --- a/drivers/gpu/drm/i915/gt/intel_hangcheck.c +++ /dev/null @@ -1,361 +0,0 @@ -/* - * Copyright © 2016 Intel Corporation - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS - * IN THE SOFTWARE. - * - */ - -#include "i915_drv.h" -#include "intel_engine.h" -#include "intel_gt.h" -#include "intel_reset.h" - -struct hangcheck { - u64 acthd; - u32 ring; - u32 head; - enum intel_engine_hangcheck_action action; - unsigned long action_timestamp; - int deadlock; - struct intel_instdone instdone; - bool wedged:1; - bool stalled:1; -}; - -static bool instdone_unchanged(u32 current_instdone, u32 *old_instdone) -{ - u32 tmp = current_instdone | *old_instdone; - bool unchanged; - - unchanged = tmp == *old_instdone; - *old_instdone |= tmp; - - return unchanged; -} - -static bool subunits_stuck(struct intel_engine_cs *engine) -{ - struct drm_i915_private *dev_priv = engine->i915; - const struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu; - struct intel_instdone instdone; - struct intel_instdone *accu_instdone = &engine->hangcheck.instdone; - bool stuck; - int slice; - int subslice; - - intel_engine_get_instdone(engine, &instdone); - - /* There might be unstable subunit states even when - * actual head is not moving. Filter out the unstable ones by - * accumulating the undone -> done transitions and only - * consider those as progress. - */ - stuck = instdone_unchanged(instdone.instdone, - &accu_instdone->instdone); - stuck &= instdone_unchanged(instdone.slice_common, - &accu_instdone->slice_common); - - for_each_instdone_slice_subslice(dev_priv, sseu, slice, subslice) { - stuck &= instdone_unchanged(instdone.sampler[slice][subslice], - &accu_instdone->sampler[slice][subslice]); - stuck &= instdone_unchanged(instdone.row[slice][subslice], - &accu_instdone->row[slice][subslice]); - } - - return stuck; -} - -static enum intel_engine_hangcheck_action -head_stuck(struct intel_engine_cs *engine, u64 acthd) -{ - if (acthd != engine->hangcheck.acthd) { - - /* Clear subunit states on head movement */ - memset(&engine->hangcheck.instdone, 0, - sizeof(engine->hangcheck.instdone)); - - return ENGINE_ACTIVE_HEAD; - } - - if (!subunits_stuck(engine)) - return ENGINE_ACTIVE_SUBUNITS; - - return ENGINE_DEAD; -} - -static enum intel_engine_hangcheck_action -engine_stuck(struct intel_engine_cs *engine, u64 acthd) -{ - enum intel_engine_hangcheck_action ha; - u32 tmp; - - ha = head_stuck(engine, acthd); - if (ha != ENGINE_DEAD) - return ha; - - if (IS_GEN(engine->i915, 2)) - return ENGINE_DEAD; - - /* Is the chip hanging on a WAIT_FOR_EVENT? - * If so we can simply poke the RB_WAIT bit - * and break the hang. This should work on - * all but the second generation chipsets. - */ - tmp = ENGINE_READ(engine, RING_CTL); - if (tmp & RING_WAIT) { - intel_gt_handle_error(engine->gt, engine->mask, 0, - "stuck wait on %s", engine->name); - ENGINE_WRITE(engine, RING_CTL, tmp); - return ENGINE_WAIT_KICK; - } - - return ENGINE_DEAD; -} - -static void hangcheck_load_sample(struct intel_engine_cs *engine, - struct hangcheck *hc) -{ - hc->acthd = intel_engine_get_active_head(engine); - hc->ring = ENGINE_READ(engine, RING_START); - hc->head = ENGINE_READ(engine, RING_HEAD); -} - -static void hangcheck_store_sample(struct intel_engine_cs *engine, - const struct hangcheck *hc) -{ - engine->hangcheck.acthd = hc->acthd; - engine->hangcheck.last_ring = hc->ring; - engine->hangcheck.last_head = hc->head; -} - -static enum intel_engine_hangcheck_action -hangcheck_get_action(struct intel_engine_cs *engine, - const struct hangcheck *hc) -{ - if (intel_engine_is_idle(engine)) - return ENGINE_IDLE; - - if (engine->hangcheck.last_ring != hc->ring) - return ENGINE_ACTIVE_SEQNO; - - if (engine->hangcheck.last_head != hc->head) - return ENGINE_ACTIVE_SEQNO; - - return engine_stuck(engine, hc->acthd); -} - -static void hangcheck_accumulate_sample(struct intel_engine_cs *engine, - struct hangcheck *hc) -{ - unsigned long timeout = I915_ENGINE_DEAD_TIMEOUT; - - hc->action = hangcheck_get_action(engine, hc); - - /* We always increment the progress - * if the engine is busy and still processing - * the same request, so that no single request - * can run indefinitely (such as a chain of - * batches). The only time we do not increment - * the hangcheck score on this ring, if this - * engine is in a legitimate wait for another - * engine. In that case the waiting engine is a - * victim and we want to be sure we catch the - * right culprit. Then every time we do kick - * the ring, make it as a progress as the seqno - * advancement might ensure and if not, it - * will catch the hanging engine. - */ - - switch (hc->action) { - case ENGINE_IDLE: - case ENGINE_ACTIVE_SEQNO: - /* Clear head and subunit states on seqno movement */ - hc->acthd = 0; - - memset(&engine->hangcheck.instdone, 0, - sizeof(engine->hangcheck.instdone)); - - /* Intentional fall through */ - case ENGINE_WAIT_KICK: - case ENGINE_WAIT: - engine->hangcheck.action_timestamp = jiffies; - break; - - case ENGINE_ACTIVE_HEAD: - case ENGINE_ACTIVE_SUBUNITS: - /* - * Seqno stuck with still active engine gets leeway, - * in hopes that it is just a long shader. - */ - timeout = I915_SEQNO_DEAD_TIMEOUT; - break; - - case ENGINE_DEAD: - break; - - default: - MISSING_CASE(hc->action); - } - - hc->stalled = time_after(jiffies, - engine->hangcheck.action_timestamp + timeout); - hc->wedged = time_after(jiffies, - engine->hangcheck.action_timestamp + - I915_ENGINE_WEDGED_TIMEOUT); -} - -static void hangcheck_declare_hang(struct intel_gt *gt, - intel_engine_mask_t hung, - intel_engine_mask_t stuck) -{ - struct intel_engine_cs *engine; - intel_engine_mask_t tmp; - char msg[80]; - int len; - - /* If some rings hung but others were still busy, only - * blame the hanging rings in the synopsis. - */ - if (stuck != hung) - hung &= ~stuck; - len = scnprintf(msg, sizeof(msg), - "%s on ", stuck == hung ? "no progress" : "hang"); - for_each_engine_masked(engine, gt, hung, tmp) - len += scnprintf(msg + len, sizeof(msg) - len, - "%s, ", engine->name); - msg[len-2] = '\0'; - - return intel_gt_handle_error(gt, hung, I915_ERROR_CAPTURE, "%s", msg); -} - -/* - * This is called when the chip hasn't reported back with completed - * batchbuffers in a long time. We keep track per ring seqno progress and - * if there are no progress, hangcheck score for that ring is increased. - * Further, acthd is inspected to see if the ring is stuck. On stuck case - * we kick the ring. If we see no progress on three subsequent calls - * we assume chip is wedged and try to fix it by resetting the chip. - */ -static void hangcheck_elapsed(struct work_struct *work) -{ - struct intel_gt *gt = - container_of(work, typeof(*gt), hangcheck.work.work); - intel_engine_mask_t hung = 0, stuck = 0, wedged = 0; - struct intel_engine_cs *engine; - enum intel_engine_id id; - intel_wakeref_t wakeref; - - if (!i915_modparams.enable_hangcheck) - return; - - if (!READ_ONCE(gt->awake)) - return; - - if (intel_gt_is_wedged(gt)) - return; - - wakeref = intel_runtime_pm_get_if_in_use(gt->uncore->rpm); - if (!wakeref) - return; - - /* As enabling the GPU requires fairly extensive mmio access, - * periodically arm the mmio checker to see if we are triggering - * any invalid access. - */ - intel_uncore_arm_unclaimed_mmio_detection(gt->uncore); - - for_each_engine(engine, gt, id) { - struct hangcheck hc; - - intel_engine_breadcrumbs_irq(engine); - - hangcheck_load_sample(engine, &hc); - hangcheck_accumulate_sample(engine, &hc); - hangcheck_store_sample(engine, &hc); - - if (hc.stalled) { - hung |= engine->mask; - if (hc.action != ENGINE_DEAD) - stuck |= engine->mask; - } - - if (hc.wedged) - wedged |= engine->mask; - } - - if (GEM_SHOW_DEBUG() && (hung | stuck)) { - struct drm_printer p = drm_debug_printer("hangcheck"); - - for_each_engine(engine, gt, id) { - if (intel_engine_is_idle(engine)) - continue; - - intel_engine_dump(engine, &p, "%s\n", engine->name); - } - } - - if (wedged) { - dev_err(gt->i915->drm.dev, - "GPU recovery timed out," - " cancelling all in-flight rendering.\n"); - GEM_TRACE_DUMP(); - intel_gt_set_wedged(gt); - } - - if (hung) - hangcheck_declare_hang(gt, hung, stuck); - - intel_runtime_pm_put(gt->uncore->rpm, wakeref); - - /* Reset timer in case GPU hangs without another request being added */ - intel_gt_queue_hangcheck(gt); -} - -void intel_gt_queue_hangcheck(struct intel_gt *gt) -{ - unsigned long delay; - - if (unlikely(!i915_modparams.enable_hangcheck)) - return; - - /* - * Don't continually defer the hangcheck so that it is always run at - * least once after work has been scheduled on any ring. Otherwise, - * we will ignore a hung ring if a second ring is kept busy. - */ - - delay = round_jiffies_up_relative(DRM_I915_HANGCHECK_JIFFIES); - queue_delayed_work(system_long_wq, >->hangcheck.work, delay); -} - -void intel_engine_init_hangcheck(struct intel_engine_cs *engine) -{ - memset(&engine->hangcheck, 0, sizeof(engine->hangcheck)); - engine->hangcheck.action_timestamp = jiffies; -} - -void intel_gt_init_hangcheck(struct intel_gt *gt) -{ - INIT_DELAYED_WORK(>->hangcheck.work, hangcheck_elapsed); -} - -#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST) -#include "selftest_hangcheck.c" -#endif diff --git a/drivers/gpu/drm/i915/gt/intel_reset.c b/drivers/gpu/drm/i915/gt/intel_reset.c index bf8d1ed4b1d8..f03e000051c1 100644 --- a/drivers/gpu/drm/i915/gt/intel_reset.c +++ b/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 diff --git a/drivers/gpu/drm/i915/gt/selftest_engine_heartbeat.c b/drivers/gpu/drm/i915/gt/selftest_engine_heartbeat.c index 5af27c37b65b..9fb5d30c43bb 100644 --- a/drivers/gpu/drm/i915/gt/selftest_engine_heartbeat.c +++ b/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; diff --git a/drivers/gpu/drm/i915/gt/selftest_hangcheck.c b/drivers/gpu/drm/i915/gt/selftest_hangcheck.c index 8e0016464325..b892b47348ab 100644 --- a/drivers/gpu/drm/i915/gt/selftest_hangcheck.c +++ b/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(>->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; diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c index 16211430eb78..70b8ccaafb05 100644 --- a/drivers/gpu/drm/i915/i915_debugfs.c +++ b/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, >->reset.flags)) - seq_puts(m, "\tWedged\n"); - if (test_bit(I915_RESET_BACKOFF, >->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(>->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(>->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); @@ -4349,7 +4263,6 @@ 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}, diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c index 5138d1eed306..24a3988281d2 100644 --- a/drivers/gpu/drm/i915/i915_drv.c +++ b/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); diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index 8882c0908c3b..674e9e921839 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/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) diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c index 5cf4eed5add8..47239df653f2 100644 --- a/drivers/gpu/drm/i915/i915_gpu_error.c +++ b/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); diff --git a/drivers/gpu/drm/i915/i915_gpu_error.h b/drivers/gpu/drm/i915/i915_gpu_error.h index 7f1cd0b1fef7..4dc36d6ee3a2 100644 --- a/drivers/gpu/drm/i915/i915_gpu_error.h +++ b/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; diff --git a/drivers/gpu/drm/i915/i915_priolist_types.h b/drivers/gpu/drm/i915/i915_priolist_types.h index ae8bb3cb627e..732aad148881 100644 --- a/drivers/gpu/drm/i915/i915_priolist_types.h +++ b/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 -- 2.24.0.rc0 _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply related [flat|nested] 18+ messages in thread
* [Intel-gfx] [CI 5/5] drm/i915/gt: Replace hangcheck by heartbeats 2019-10-23 13:31 ` [CI 5/5] drm/i915/gt: Replace hangcheck by heartbeats Chris Wilson @ 2019-10-23 13:31 ` Chris Wilson 0 siblings, 0 replies; 18+ messages in thread From: Chris Wilson @ 2019-10-23 13:31 UTC (permalink / raw) To: intel-gfx 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> --- drivers/gpu/drm/i915/Kconfig.profile | 11 + drivers/gpu/drm/i915/Makefile | 1 - drivers/gpu/drm/i915/display/intel_display.c | 2 +- drivers/gpu/drm/i915/gem/i915_gem_object.h | 1 - drivers/gpu/drm/i915/gem/i915_gem_pm.c | 2 - drivers/gpu/drm/i915/gt/intel_engine.h | 32 -- drivers/gpu/drm/i915/gt/intel_engine_cs.c | 12 +- .../gpu/drm/i915/gt/intel_engine_heartbeat.c | 157 ++++++++ .../gpu/drm/i915/gt/intel_engine_heartbeat.h | 8 + drivers/gpu/drm/i915/gt/intel_engine_pm.c | 5 +- drivers/gpu/drm/i915/gt/intel_engine_types.h | 17 +- drivers/gpu/drm/i915/gt/intel_gt.c | 1 - drivers/gpu/drm/i915/gt/intel_gt.h | 4 - drivers/gpu/drm/i915/gt/intel_gt_pm.c | 1 - drivers/gpu/drm/i915/gt/intel_gt_types.h | 9 - drivers/gpu/drm/i915/gt/intel_hangcheck.c | 361 ------------------ drivers/gpu/drm/i915/gt/intel_reset.c | 3 +- .../drm/i915/gt/selftest_engine_heartbeat.c | 171 +++++++++ drivers/gpu/drm/i915/gt/selftest_hangcheck.c | 4 - drivers/gpu/drm/i915/i915_debugfs.c | 87 ----- drivers/gpu/drm/i915/i915_drv.c | 3 - drivers/gpu/drm/i915/i915_drv.h | 1 - drivers/gpu/drm/i915/i915_gpu_error.c | 33 +- drivers/gpu/drm/i915/i915_gpu_error.h | 2 - drivers/gpu/drm/i915/i915_priolist_types.h | 6 + 25 files changed, 379 insertions(+), 555 deletions(-) delete mode 100644 drivers/gpu/drm/i915/gt/intel_hangcheck.c diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile index b071b6024152..8ab7af5eb311 100644 --- a/drivers/gpu/drm/i915/Kconfig.profile +++ b/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 diff --git a/drivers/gpu/drm/i915/Makefile b/drivers/gpu/drm/i915/Makefile index 2fd4bed188e5..21601bb27b22 100644 --- a/drivers/gpu/drm/i915/Makefile +++ b/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 \ diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c index 236fdf122e47..579655675b08 100644 --- a/drivers/gpu/drm/i915/display/intel_display.c +++ b/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); diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.h b/drivers/gpu/drm/i915/gem/i915_gem_object.h index aead7e6725f9..458cd51331f1 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_object.h +++ b/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 diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pm.c b/drivers/gpu/drm/i915/gem/i915_gem_pm.c index 9bc0cf3139e3..c99bb94fe41e 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_pm.c +++ b/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); } diff --git a/drivers/gpu/drm/i915/gt/intel_engine.h b/drivers/gpu/drm/i915/gt/intel_engine.h index 9409b7856299..ee47444a6ad4 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine.h +++ b/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) diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c index b91ea07f4819..8b5129e0b49d 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c +++ b/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)); diff --git a/drivers/gpu/drm/i915/gt/intel_engine_heartbeat.c b/drivers/gpu/drm/i915/gt/intel_engine_heartbeat.c index 4b9ab7813d54..9977f59f6b53 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_heartbeat.c +++ b/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 }; diff --git a/drivers/gpu/drm/i915/gt/intel_engine_heartbeat.h b/drivers/gpu/drm/i915/gt/intel_engine_heartbeat.h index b334e5aaf78d..a7b8c0f9e005 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_heartbeat.h +++ b/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); diff --git a/drivers/gpu/drm/i915/gt/intel_engine_pm.c b/drivers/gpu/drm/i915/gt/intel_engine_pm.c index 7d76611d9df1..6fbfa2162e54 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_pm.c +++ b/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) diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h index 1251dac91f31..26998901feff 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_types.h +++ b/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; diff --git a/drivers/gpu/drm/i915/gt/intel_gt.c b/drivers/gpu/drm/i915/gt/intel_gt.c index 1c4b6c9642ad..35127699591d 100644 --- a/drivers/gpu/drm/i915/gt/intel_gt.c +++ b/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(>->closed_vma); spin_lock_init(>->closed_lock); - intel_gt_init_hangcheck(gt); intel_gt_init_reset(gt); intel_gt_init_requests(gt); intel_gt_pm_init_early(gt); diff --git a/drivers/gpu/drm/i915/gt/intel_gt.h b/drivers/gpu/drm/i915/gt/intel_gt.h index e6ab0bff0efb..5b6effed3713 100644 --- a/drivers/gpu/drm/i915/gt/intel_gt.h +++ b/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(>->reset); } -void intel_gt_queue_hangcheck(struct intel_gt *gt); - #endif /* __INTEL_GT_H__ */ diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm.c b/drivers/gpu/drm/i915/gt/intel_gt_pm.c index 06e73d56cfcf..aff4eda47819 100644 --- a/drivers/gpu/drm/i915/gt/intel_gt_pm.c +++ b/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; diff --git a/drivers/gpu/drm/i915/gt/intel_gt_types.h b/drivers/gpu/drm/i915/gt/intel_gt_types.h index 980973e66e7f..fcb3a6c9421c 100644 --- a/drivers/gpu/drm/i915/gt/intel_gt_types.h +++ b/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; /** diff --git a/drivers/gpu/drm/i915/gt/intel_hangcheck.c b/drivers/gpu/drm/i915/gt/intel_hangcheck.c deleted file mode 100644 index 0fdef00af9e4..000000000000 --- a/drivers/gpu/drm/i915/gt/intel_hangcheck.c +++ /dev/null @@ -1,361 +0,0 @@ -/* - * Copyright © 2016 Intel Corporation - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS - * IN THE SOFTWARE. - * - */ - -#include "i915_drv.h" -#include "intel_engine.h" -#include "intel_gt.h" -#include "intel_reset.h" - -struct hangcheck { - u64 acthd; - u32 ring; - u32 head; - enum intel_engine_hangcheck_action action; - unsigned long action_timestamp; - int deadlock; - struct intel_instdone instdone; - bool wedged:1; - bool stalled:1; -}; - -static bool instdone_unchanged(u32 current_instdone, u32 *old_instdone) -{ - u32 tmp = current_instdone | *old_instdone; - bool unchanged; - - unchanged = tmp == *old_instdone; - *old_instdone |= tmp; - - return unchanged; -} - -static bool subunits_stuck(struct intel_engine_cs *engine) -{ - struct drm_i915_private *dev_priv = engine->i915; - const struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu; - struct intel_instdone instdone; - struct intel_instdone *accu_instdone = &engine->hangcheck.instdone; - bool stuck; - int slice; - int subslice; - - intel_engine_get_instdone(engine, &instdone); - - /* There might be unstable subunit states even when - * actual head is not moving. Filter out the unstable ones by - * accumulating the undone -> done transitions and only - * consider those as progress. - */ - stuck = instdone_unchanged(instdone.instdone, - &accu_instdone->instdone); - stuck &= instdone_unchanged(instdone.slice_common, - &accu_instdone->slice_common); - - for_each_instdone_slice_subslice(dev_priv, sseu, slice, subslice) { - stuck &= instdone_unchanged(instdone.sampler[slice][subslice], - &accu_instdone->sampler[slice][subslice]); - stuck &= instdone_unchanged(instdone.row[slice][subslice], - &accu_instdone->row[slice][subslice]); - } - - return stuck; -} - -static enum intel_engine_hangcheck_action -head_stuck(struct intel_engine_cs *engine, u64 acthd) -{ - if (acthd != engine->hangcheck.acthd) { - - /* Clear subunit states on head movement */ - memset(&engine->hangcheck.instdone, 0, - sizeof(engine->hangcheck.instdone)); - - return ENGINE_ACTIVE_HEAD; - } - - if (!subunits_stuck(engine)) - return ENGINE_ACTIVE_SUBUNITS; - - return ENGINE_DEAD; -} - -static enum intel_engine_hangcheck_action -engine_stuck(struct intel_engine_cs *engine, u64 acthd) -{ - enum intel_engine_hangcheck_action ha; - u32 tmp; - - ha = head_stuck(engine, acthd); - if (ha != ENGINE_DEAD) - return ha; - - if (IS_GEN(engine->i915, 2)) - return ENGINE_DEAD; - - /* Is the chip hanging on a WAIT_FOR_EVENT? - * If so we can simply poke the RB_WAIT bit - * and break the hang. This should work on - * all but the second generation chipsets. - */ - tmp = ENGINE_READ(engine, RING_CTL); - if (tmp & RING_WAIT) { - intel_gt_handle_error(engine->gt, engine->mask, 0, - "stuck wait on %s", engine->name); - ENGINE_WRITE(engine, RING_CTL, tmp); - return ENGINE_WAIT_KICK; - } - - return ENGINE_DEAD; -} - -static void hangcheck_load_sample(struct intel_engine_cs *engine, - struct hangcheck *hc) -{ - hc->acthd = intel_engine_get_active_head(engine); - hc->ring = ENGINE_READ(engine, RING_START); - hc->head = ENGINE_READ(engine, RING_HEAD); -} - -static void hangcheck_store_sample(struct intel_engine_cs *engine, - const struct hangcheck *hc) -{ - engine->hangcheck.acthd = hc->acthd; - engine->hangcheck.last_ring = hc->ring; - engine->hangcheck.last_head = hc->head; -} - -static enum intel_engine_hangcheck_action -hangcheck_get_action(struct intel_engine_cs *engine, - const struct hangcheck *hc) -{ - if (intel_engine_is_idle(engine)) - return ENGINE_IDLE; - - if (engine->hangcheck.last_ring != hc->ring) - return ENGINE_ACTIVE_SEQNO; - - if (engine->hangcheck.last_head != hc->head) - return ENGINE_ACTIVE_SEQNO; - - return engine_stuck(engine, hc->acthd); -} - -static void hangcheck_accumulate_sample(struct intel_engine_cs *engine, - struct hangcheck *hc) -{ - unsigned long timeout = I915_ENGINE_DEAD_TIMEOUT; - - hc->action = hangcheck_get_action(engine, hc); - - /* We always increment the progress - * if the engine is busy and still processing - * the same request, so that no single request - * can run indefinitely (such as a chain of - * batches). The only time we do not increment - * the hangcheck score on this ring, if this - * engine is in a legitimate wait for another - * engine. In that case the waiting engine is a - * victim and we want to be sure we catch the - * right culprit. Then every time we do kick - * the ring, make it as a progress as the seqno - * advancement might ensure and if not, it - * will catch the hanging engine. - */ - - switch (hc->action) { - case ENGINE_IDLE: - case ENGINE_ACTIVE_SEQNO: - /* Clear head and subunit states on seqno movement */ - hc->acthd = 0; - - memset(&engine->hangcheck.instdone, 0, - sizeof(engine->hangcheck.instdone)); - - /* Intentional fall through */ - case ENGINE_WAIT_KICK: - case ENGINE_WAIT: - engine->hangcheck.action_timestamp = jiffies; - break; - - case ENGINE_ACTIVE_HEAD: - case ENGINE_ACTIVE_SUBUNITS: - /* - * Seqno stuck with still active engine gets leeway, - * in hopes that it is just a long shader. - */ - timeout = I915_SEQNO_DEAD_TIMEOUT; - break; - - case ENGINE_DEAD: - break; - - default: - MISSING_CASE(hc->action); - } - - hc->stalled = time_after(jiffies, - engine->hangcheck.action_timestamp + timeout); - hc->wedged = time_after(jiffies, - engine->hangcheck.action_timestamp + - I915_ENGINE_WEDGED_TIMEOUT); -} - -static void hangcheck_declare_hang(struct intel_gt *gt, - intel_engine_mask_t hung, - intel_engine_mask_t stuck) -{ - struct intel_engine_cs *engine; - intel_engine_mask_t tmp; - char msg[80]; - int len; - - /* If some rings hung but others were still busy, only - * blame the hanging rings in the synopsis. - */ - if (stuck != hung) - hung &= ~stuck; - len = scnprintf(msg, sizeof(msg), - "%s on ", stuck == hung ? "no progress" : "hang"); - for_each_engine_masked(engine, gt, hung, tmp) - len += scnprintf(msg + len, sizeof(msg) - len, - "%s, ", engine->name); - msg[len-2] = '\0'; - - return intel_gt_handle_error(gt, hung, I915_ERROR_CAPTURE, "%s", msg); -} - -/* - * This is called when the chip hasn't reported back with completed - * batchbuffers in a long time. We keep track per ring seqno progress and - * if there are no progress, hangcheck score for that ring is increased. - * Further, acthd is inspected to see if the ring is stuck. On stuck case - * we kick the ring. If we see no progress on three subsequent calls - * we assume chip is wedged and try to fix it by resetting the chip. - */ -static void hangcheck_elapsed(struct work_struct *work) -{ - struct intel_gt *gt = - container_of(work, typeof(*gt), hangcheck.work.work); - intel_engine_mask_t hung = 0, stuck = 0, wedged = 0; - struct intel_engine_cs *engine; - enum intel_engine_id id; - intel_wakeref_t wakeref; - - if (!i915_modparams.enable_hangcheck) - return; - - if (!READ_ONCE(gt->awake)) - return; - - if (intel_gt_is_wedged(gt)) - return; - - wakeref = intel_runtime_pm_get_if_in_use(gt->uncore->rpm); - if (!wakeref) - return; - - /* As enabling the GPU requires fairly extensive mmio access, - * periodically arm the mmio checker to see if we are triggering - * any invalid access. - */ - intel_uncore_arm_unclaimed_mmio_detection(gt->uncore); - - for_each_engine(engine, gt, id) { - struct hangcheck hc; - - intel_engine_breadcrumbs_irq(engine); - - hangcheck_load_sample(engine, &hc); - hangcheck_accumulate_sample(engine, &hc); - hangcheck_store_sample(engine, &hc); - - if (hc.stalled) { - hung |= engine->mask; - if (hc.action != ENGINE_DEAD) - stuck |= engine->mask; - } - - if (hc.wedged) - wedged |= engine->mask; - } - - if (GEM_SHOW_DEBUG() && (hung | stuck)) { - struct drm_printer p = drm_debug_printer("hangcheck"); - - for_each_engine(engine, gt, id) { - if (intel_engine_is_idle(engine)) - continue; - - intel_engine_dump(engine, &p, "%s\n", engine->name); - } - } - - if (wedged) { - dev_err(gt->i915->drm.dev, - "GPU recovery timed out," - " cancelling all in-flight rendering.\n"); - GEM_TRACE_DUMP(); - intel_gt_set_wedged(gt); - } - - if (hung) - hangcheck_declare_hang(gt, hung, stuck); - - intel_runtime_pm_put(gt->uncore->rpm, wakeref); - - /* Reset timer in case GPU hangs without another request being added */ - intel_gt_queue_hangcheck(gt); -} - -void intel_gt_queue_hangcheck(struct intel_gt *gt) -{ - unsigned long delay; - - if (unlikely(!i915_modparams.enable_hangcheck)) - return; - - /* - * Don't continually defer the hangcheck so that it is always run at - * least once after work has been scheduled on any ring. Otherwise, - * we will ignore a hung ring if a second ring is kept busy. - */ - - delay = round_jiffies_up_relative(DRM_I915_HANGCHECK_JIFFIES); - queue_delayed_work(system_long_wq, >->hangcheck.work, delay); -} - -void intel_engine_init_hangcheck(struct intel_engine_cs *engine) -{ - memset(&engine->hangcheck, 0, sizeof(engine->hangcheck)); - engine->hangcheck.action_timestamp = jiffies; -} - -void intel_gt_init_hangcheck(struct intel_gt *gt) -{ - INIT_DELAYED_WORK(>->hangcheck.work, hangcheck_elapsed); -} - -#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST) -#include "selftest_hangcheck.c" -#endif diff --git a/drivers/gpu/drm/i915/gt/intel_reset.c b/drivers/gpu/drm/i915/gt/intel_reset.c index bf8d1ed4b1d8..f03e000051c1 100644 --- a/drivers/gpu/drm/i915/gt/intel_reset.c +++ b/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 diff --git a/drivers/gpu/drm/i915/gt/selftest_engine_heartbeat.c b/drivers/gpu/drm/i915/gt/selftest_engine_heartbeat.c index 5af27c37b65b..9fb5d30c43bb 100644 --- a/drivers/gpu/drm/i915/gt/selftest_engine_heartbeat.c +++ b/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; diff --git a/drivers/gpu/drm/i915/gt/selftest_hangcheck.c b/drivers/gpu/drm/i915/gt/selftest_hangcheck.c index 8e0016464325..b892b47348ab 100644 --- a/drivers/gpu/drm/i915/gt/selftest_hangcheck.c +++ b/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(>->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; diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c index 16211430eb78..70b8ccaafb05 100644 --- a/drivers/gpu/drm/i915/i915_debugfs.c +++ b/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, >->reset.flags)) - seq_puts(m, "\tWedged\n"); - if (test_bit(I915_RESET_BACKOFF, >->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(>->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(>->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); @@ -4349,7 +4263,6 @@ 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}, diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c index 5138d1eed306..24a3988281d2 100644 --- a/drivers/gpu/drm/i915/i915_drv.c +++ b/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); diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index 8882c0908c3b..674e9e921839 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/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) diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c index 5cf4eed5add8..47239df653f2 100644 --- a/drivers/gpu/drm/i915/i915_gpu_error.c +++ b/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); diff --git a/drivers/gpu/drm/i915/i915_gpu_error.h b/drivers/gpu/drm/i915/i915_gpu_error.h index 7f1cd0b1fef7..4dc36d6ee3a2 100644 --- a/drivers/gpu/drm/i915/i915_gpu_error.h +++ b/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; diff --git a/drivers/gpu/drm/i915/i915_priolist_types.h b/drivers/gpu/drm/i915/i915_priolist_types.h index ae8bb3cb627e..732aad148881 100644 --- a/drivers/gpu/drm/i915/i915_priolist_types.h +++ b/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 -- 2.24.0.rc0 _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply related [flat|nested] 18+ messages in thread
* ✗ Fi.CI.BUILD: failure for series starting with [CI,1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets 2019-10-23 13:31 [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets Chris Wilson ` (4 preceding siblings ...) 2019-10-23 13:31 ` [CI 5/5] drm/i915/gt: Replace hangcheck by heartbeats Chris Wilson @ 2019-10-24 0:09 ` Patchwork 2019-10-24 0:09 ` [Intel-gfx] " Patchwork 2019-10-24 1:32 ` [CI 1/5] " Ramalingam C 6 siblings, 1 reply; 18+ messages in thread From: Patchwork @ 2019-10-24 0:09 UTC (permalink / raw) To: Chris Wilson; +Cc: intel-gfx == Series Details == Series: series starting with [CI,1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets URL : https://patchwork.freedesktop.org/series/68463/ State : failure == Summary == Applying: drm/i915/gt: Try to more gracefully quiesce the system before resets Using index info to reconstruct a base tree... M drivers/gpu/drm/i915/Kconfig.profile M drivers/gpu/drm/i915/gt/intel_engine_cs.c M drivers/gpu/drm/i915/gt/intel_engine_types.h Falling back to patching base and 3-way merge... Auto-merging drivers/gpu/drm/i915/gt/intel_engine_types.h CONFLICT (content): Merge conflict in drivers/gpu/drm/i915/gt/intel_engine_types.h Auto-merging drivers/gpu/drm/i915/gt/intel_engine_cs.c CONFLICT (content): Merge conflict in drivers/gpu/drm/i915/gt/intel_engine_cs.c Auto-merging drivers/gpu/drm/i915/Kconfig.profile error: Failed to merge in the changes. hint: Use 'git am --show-current-patch' to see the failed patch Patch failed at 0001 drm/i915/gt: Try to more gracefully quiesce the system before resets When you have resolved this problem, run "git am --continue". If you prefer to skip this patch, run "git am --skip" instead. To restore the original branch and stop patching, run "git am --abort". _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply [flat|nested] 18+ messages in thread
* [Intel-gfx] ✗ Fi.CI.BUILD: failure for series starting with [CI,1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets 2019-10-24 0:09 ` ✗ Fi.CI.BUILD: failure for series starting with [CI,1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets Patchwork @ 2019-10-24 0:09 ` Patchwork 0 siblings, 0 replies; 18+ messages in thread From: Patchwork @ 2019-10-24 0:09 UTC (permalink / raw) To: Chris Wilson; +Cc: intel-gfx == Series Details == Series: series starting with [CI,1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets URL : https://patchwork.freedesktop.org/series/68463/ State : failure == Summary == Applying: drm/i915/gt: Try to more gracefully quiesce the system before resets Using index info to reconstruct a base tree... M drivers/gpu/drm/i915/Kconfig.profile M drivers/gpu/drm/i915/gt/intel_engine_cs.c M drivers/gpu/drm/i915/gt/intel_engine_types.h Falling back to patching base and 3-way merge... Auto-merging drivers/gpu/drm/i915/gt/intel_engine_types.h CONFLICT (content): Merge conflict in drivers/gpu/drm/i915/gt/intel_engine_types.h Auto-merging drivers/gpu/drm/i915/gt/intel_engine_cs.c CONFLICT (content): Merge conflict in drivers/gpu/drm/i915/gt/intel_engine_cs.c Auto-merging drivers/gpu/drm/i915/Kconfig.profile error: Failed to merge in the changes. hint: Use 'git am --show-current-patch' to see the failed patch Patch failed at 0001 drm/i915/gt: Try to more gracefully quiesce the system before resets When you have resolved this problem, run "git am --continue". If you prefer to skip this patch, run "git am --skip" instead. To restore the original branch and stop patching, run "git am --abort". _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets 2019-10-23 13:31 [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets Chris Wilson ` (5 preceding siblings ...) 2019-10-24 0:09 ` ✗ Fi.CI.BUILD: failure for series starting with [CI,1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets Patchwork @ 2019-10-24 1:32 ` Ramalingam C 2019-10-24 1:32 ` [Intel-gfx] " Ramalingam C 2019-10-24 6:14 ` Chris Wilson 6 siblings, 2 replies; 18+ messages in thread From: Ramalingam C @ 2019-10-24 1:32 UTC (permalink / raw) To: Chris Wilson; +Cc: intel-gfx On 2019-10-23 at 14:31:04 +0100, Chris Wilson wrote: > If we are doing a normal GPU reset triggered after detecting a long > period of stalled work, we can take our time and allow the engines to > quiesce. Since we've stopped submission to the engine, and if we wait > long enough an innocent context should complete, leaving the engine idle. > So by waiting a short amount of time, we should prevent clobbering other > users when resetting a stuck context. > > Suggested-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > Suggested-by: Jon Bloomfield <jon.bloomfield@intel.com> > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> > Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com> > --- > drivers/gpu/drm/i915/Kconfig.profile | 11 +++++++++++ > drivers/gpu/drm/i915/gt/intel_engine_cs.c | 20 +++++++++++++++++++- > drivers/gpu/drm/i915/gt/intel_engine_types.h | 4 ++++ > 3 files changed, 34 insertions(+), 1 deletion(-) > > diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile > index 48df8889a88a..3a3881d5e44b 100644 > --- a/drivers/gpu/drm/i915/Kconfig.profile > +++ b/drivers/gpu/drm/i915/Kconfig.profile > @@ -25,3 +25,14 @@ config DRM_I915_SPIN_REQUEST > May be 0 to disable the initial spin. In practice, we estimate > the cost of enabling the interrupt (if currently disabled) to be > a few microseconds. > + > +config DRM_I915_STOP_TIMEOUT > + int "How long to wait for an engine to quiesce gracefully before reset (ms)" > + default 100 # milliseconds > + help > + By stopping submission and sleeping for a short time before resetting > + the GPU, we allow the innocent contexts also on the system to quiesce. > + It is then less likely for a hanging context to cause collateral > + damage as the system is reset in order to recover. The corollary is > + that the reset itself may take longer and so be more disruptive to > + interactive or low latency workloads. > diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c > index 0e20713603ec..e4203eb44139 100644 > --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c > +++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c > @@ -308,6 +308,9 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id) > engine->instance = info->instance; > __sprint_engine_name(engine); > > + engine->props.stop_timeout_ms = > + CONFIG_DRM_I915_STOP_TIMEOUT; Compare to previous version where you used the CONFIG variable directly, what is the benefit of using it through a variable? So that we could alter it in runtime? -Ram > + > /* > * To be overridden by the backend on setup. However to facilitate > * cleanup on error during setup, we always provide the destroy vfunc. > @@ -875,6 +878,21 @@ u64 intel_engine_get_last_batch_head(const struct intel_engine_cs *engine) > return bbaddr; > } > > +static unsigned long stop_timeout(const struct intel_engine_cs *engine) > +{ > + if (in_atomic() || irqs_disabled()) /* inside atomic preempt-reset? */ > + return 0; > + > + /* > + * If we are doing a normal GPU reset, we can take our time and allow > + * the engine to quiesce. We've stopped submission to the engine, and > + * if we wait long enough an innocent context should complete and > + * leave the engine idle. So they should not be caught unaware by > + * the forthcoming GPU reset (which usually follows the stop_cs)! > + */ > + return READ_ONCE(engine->props.stop_timeout_ms); > +} > + > int intel_engine_stop_cs(struct intel_engine_cs *engine) > { > struct intel_uncore *uncore = engine->uncore; > @@ -892,7 +910,7 @@ int intel_engine_stop_cs(struct intel_engine_cs *engine) > err = 0; > if (__intel_wait_for_register_fw(uncore, > mode, MODE_IDLE, MODE_IDLE, > - 1000, 0, > + 1000, stop_timeout(engine), > NULL)) { > GEM_TRACE("%s: timed out on STOP_RING -> IDLE\n", engine->name); > err = -ETIMEDOUT; > diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h > index 3451be034caf..87d5c4ef3ae7 100644 > --- a/drivers/gpu/drm/i915/gt/intel_engine_types.h > +++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h > @@ -542,6 +542,10 @@ struct intel_engine_cs { > */ > ktime_t total; > } stats; > + > + struct { > + unsigned long stop_timeout_ms; > + } props; > }; > > static inline bool > -- > 2.24.0.rc0 > > _______________________________________________ > Intel-gfx mailing list > Intel-gfx@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/intel-gfx _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [Intel-gfx] [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets 2019-10-24 1:32 ` [CI 1/5] " Ramalingam C @ 2019-10-24 1:32 ` Ramalingam C 2019-10-24 6:14 ` Chris Wilson 1 sibling, 0 replies; 18+ messages in thread From: Ramalingam C @ 2019-10-24 1:32 UTC (permalink / raw) To: Chris Wilson; +Cc: intel-gfx On 2019-10-23 at 14:31:04 +0100, Chris Wilson wrote: > If we are doing a normal GPU reset triggered after detecting a long > period of stalled work, we can take our time and allow the engines to > quiesce. Since we've stopped submission to the engine, and if we wait > long enough an innocent context should complete, leaving the engine idle. > So by waiting a short amount of time, we should prevent clobbering other > users when resetting a stuck context. > > Suggested-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > Suggested-by: Jon Bloomfield <jon.bloomfield@intel.com> > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> > Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com> > --- > drivers/gpu/drm/i915/Kconfig.profile | 11 +++++++++++ > drivers/gpu/drm/i915/gt/intel_engine_cs.c | 20 +++++++++++++++++++- > drivers/gpu/drm/i915/gt/intel_engine_types.h | 4 ++++ > 3 files changed, 34 insertions(+), 1 deletion(-) > > diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile > index 48df8889a88a..3a3881d5e44b 100644 > --- a/drivers/gpu/drm/i915/Kconfig.profile > +++ b/drivers/gpu/drm/i915/Kconfig.profile > @@ -25,3 +25,14 @@ config DRM_I915_SPIN_REQUEST > May be 0 to disable the initial spin. In practice, we estimate > the cost of enabling the interrupt (if currently disabled) to be > a few microseconds. > + > +config DRM_I915_STOP_TIMEOUT > + int "How long to wait for an engine to quiesce gracefully before reset (ms)" > + default 100 # milliseconds > + help > + By stopping submission and sleeping for a short time before resetting > + the GPU, we allow the innocent contexts also on the system to quiesce. > + It is then less likely for a hanging context to cause collateral > + damage as the system is reset in order to recover. The corollary is > + that the reset itself may take longer and so be more disruptive to > + interactive or low latency workloads. > diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c > index 0e20713603ec..e4203eb44139 100644 > --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c > +++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c > @@ -308,6 +308,9 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id) > engine->instance = info->instance; > __sprint_engine_name(engine); > > + engine->props.stop_timeout_ms = > + CONFIG_DRM_I915_STOP_TIMEOUT; Compare to previous version where you used the CONFIG variable directly, what is the benefit of using it through a variable? So that we could alter it in runtime? -Ram > + > /* > * To be overridden by the backend on setup. However to facilitate > * cleanup on error during setup, we always provide the destroy vfunc. > @@ -875,6 +878,21 @@ u64 intel_engine_get_last_batch_head(const struct intel_engine_cs *engine) > return bbaddr; > } > > +static unsigned long stop_timeout(const struct intel_engine_cs *engine) > +{ > + if (in_atomic() || irqs_disabled()) /* inside atomic preempt-reset? */ > + return 0; > + > + /* > + * If we are doing a normal GPU reset, we can take our time and allow > + * the engine to quiesce. We've stopped submission to the engine, and > + * if we wait long enough an innocent context should complete and > + * leave the engine idle. So they should not be caught unaware by > + * the forthcoming GPU reset (which usually follows the stop_cs)! > + */ > + return READ_ONCE(engine->props.stop_timeout_ms); > +} > + > int intel_engine_stop_cs(struct intel_engine_cs *engine) > { > struct intel_uncore *uncore = engine->uncore; > @@ -892,7 +910,7 @@ int intel_engine_stop_cs(struct intel_engine_cs *engine) > err = 0; > if (__intel_wait_for_register_fw(uncore, > mode, MODE_IDLE, MODE_IDLE, > - 1000, 0, > + 1000, stop_timeout(engine), > NULL)) { > GEM_TRACE("%s: timed out on STOP_RING -> IDLE\n", engine->name); > err = -ETIMEDOUT; > diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h > index 3451be034caf..87d5c4ef3ae7 100644 > --- a/drivers/gpu/drm/i915/gt/intel_engine_types.h > +++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h > @@ -542,6 +542,10 @@ struct intel_engine_cs { > */ > ktime_t total; > } stats; > + > + struct { > + unsigned long stop_timeout_ms; > + } props; > }; > > static inline bool > -- > 2.24.0.rc0 > > _______________________________________________ > Intel-gfx mailing list > Intel-gfx@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/intel-gfx _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets 2019-10-24 1:32 ` [CI 1/5] " Ramalingam C 2019-10-24 1:32 ` [Intel-gfx] " Ramalingam C @ 2019-10-24 6:14 ` Chris Wilson 2019-10-24 6:14 ` [Intel-gfx] " Chris Wilson 2019-10-24 6:37 ` Ramalingam C 1 sibling, 2 replies; 18+ messages in thread From: Chris Wilson @ 2019-10-24 6:14 UTC (permalink / raw) To: Ramalingam C; +Cc: intel-gfx Quoting Ramalingam C (2019-10-24 02:32:01) > On 2019-10-23 at 14:31:04 +0100, Chris Wilson wrote: > > If we are doing a normal GPU reset triggered after detecting a long > > period of stalled work, we can take our time and allow the engines to > > quiesce. Since we've stopped submission to the engine, and if we wait > > long enough an innocent context should complete, leaving the engine idle. > > So by waiting a short amount of time, we should prevent clobbering other > > users when resetting a stuck context. > > > > Suggested-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > > Suggested-by: Jon Bloomfield <jon.bloomfield@intel.com> > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > > Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> > > Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > > Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com> > > --- > > drivers/gpu/drm/i915/Kconfig.profile | 11 +++++++++++ > > drivers/gpu/drm/i915/gt/intel_engine_cs.c | 20 +++++++++++++++++++- > > drivers/gpu/drm/i915/gt/intel_engine_types.h | 4 ++++ > > 3 files changed, 34 insertions(+), 1 deletion(-) > > > > diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile > > index 48df8889a88a..3a3881d5e44b 100644 > > --- a/drivers/gpu/drm/i915/Kconfig.profile > > +++ b/drivers/gpu/drm/i915/Kconfig.profile > > @@ -25,3 +25,14 @@ config DRM_I915_SPIN_REQUEST > > May be 0 to disable the initial spin. In practice, we estimate > > the cost of enabling the interrupt (if currently disabled) to be > > a few microseconds. > > + > > +config DRM_I915_STOP_TIMEOUT > > + int "How long to wait for an engine to quiesce gracefully before reset (ms)" > > + default 100 # milliseconds > > + help > > + By stopping submission and sleeping for a short time before resetting > > + the GPU, we allow the innocent contexts also on the system to quiesce. > > + It is then less likely for a hanging context to cause collateral > > + damage as the system is reset in order to recover. The corollary is > > + that the reset itself may take longer and so be more disruptive to > > + interactive or low latency workloads. > > diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c > > index 0e20713603ec..e4203eb44139 100644 > > --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c > > +++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c > > @@ -308,6 +308,9 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id) > > engine->instance = info->instance; > > __sprint_engine_name(engine); > > > > + engine->props.stop_timeout_ms = > > + CONFIG_DRM_I915_STOP_TIMEOUT; > Compare to previous version where you used the CONFIG variable directly, > what is the benefit of using it through a variable? So that we could > alter it in runtime? That is what the next series proposes. Also it turns out to be more easily testable if one is able to alter the timeouts for testing. -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [Intel-gfx] [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets 2019-10-24 6:14 ` Chris Wilson @ 2019-10-24 6:14 ` Chris Wilson 2019-10-24 6:37 ` Ramalingam C 1 sibling, 0 replies; 18+ messages in thread From: Chris Wilson @ 2019-10-24 6:14 UTC (permalink / raw) To: Ramalingam C; +Cc: intel-gfx Quoting Ramalingam C (2019-10-24 02:32:01) > On 2019-10-23 at 14:31:04 +0100, Chris Wilson wrote: > > If we are doing a normal GPU reset triggered after detecting a long > > period of stalled work, we can take our time and allow the engines to > > quiesce. Since we've stopped submission to the engine, and if we wait > > long enough an innocent context should complete, leaving the engine idle. > > So by waiting a short amount of time, we should prevent clobbering other > > users when resetting a stuck context. > > > > Suggested-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > > Suggested-by: Jon Bloomfield <jon.bloomfield@intel.com> > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > > Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> > > Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > > Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com> > > --- > > drivers/gpu/drm/i915/Kconfig.profile | 11 +++++++++++ > > drivers/gpu/drm/i915/gt/intel_engine_cs.c | 20 +++++++++++++++++++- > > drivers/gpu/drm/i915/gt/intel_engine_types.h | 4 ++++ > > 3 files changed, 34 insertions(+), 1 deletion(-) > > > > diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile > > index 48df8889a88a..3a3881d5e44b 100644 > > --- a/drivers/gpu/drm/i915/Kconfig.profile > > +++ b/drivers/gpu/drm/i915/Kconfig.profile > > @@ -25,3 +25,14 @@ config DRM_I915_SPIN_REQUEST > > May be 0 to disable the initial spin. In practice, we estimate > > the cost of enabling the interrupt (if currently disabled) to be > > a few microseconds. > > + > > +config DRM_I915_STOP_TIMEOUT > > + int "How long to wait for an engine to quiesce gracefully before reset (ms)" > > + default 100 # milliseconds > > + help > > + By stopping submission and sleeping for a short time before resetting > > + the GPU, we allow the innocent contexts also on the system to quiesce. > > + It is then less likely for a hanging context to cause collateral > > + damage as the system is reset in order to recover. The corollary is > > + that the reset itself may take longer and so be more disruptive to > > + interactive or low latency workloads. > > diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c > > index 0e20713603ec..e4203eb44139 100644 > > --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c > > +++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c > > @@ -308,6 +308,9 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id) > > engine->instance = info->instance; > > __sprint_engine_name(engine); > > > > + engine->props.stop_timeout_ms = > > + CONFIG_DRM_I915_STOP_TIMEOUT; > Compare to previous version where you used the CONFIG variable directly, > what is the benefit of using it through a variable? So that we could > alter it in runtime? That is what the next series proposes. Also it turns out to be more easily testable if one is able to alter the timeouts for testing. -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets 2019-10-24 6:14 ` Chris Wilson 2019-10-24 6:14 ` [Intel-gfx] " Chris Wilson @ 2019-10-24 6:37 ` Ramalingam C 2019-10-24 6:37 ` [Intel-gfx] " Ramalingam C 1 sibling, 1 reply; 18+ messages in thread From: Ramalingam C @ 2019-10-24 6:37 UTC (permalink / raw) To: Chris Wilson; +Cc: intel-gfx On 2019-10-24 at 07:14:56 +0100, Chris Wilson wrote: > Quoting Ramalingam C (2019-10-24 02:32:01) > > On 2019-10-23 at 14:31:04 +0100, Chris Wilson wrote: > > > If we are doing a normal GPU reset triggered after detecting a long > > > period of stalled work, we can take our time and allow the engines to > > > quiesce. Since we've stopped submission to the engine, and if we wait > > > long enough an innocent context should complete, leaving the engine idle. > > > So by waiting a short amount of time, we should prevent clobbering other > > > users when resetting a stuck context. > > > > > > Suggested-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > > > Suggested-by: Jon Bloomfield <jon.bloomfield@intel.com> > > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > > > Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> > > > Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > > > Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com> > > > --- > > > drivers/gpu/drm/i915/Kconfig.profile | 11 +++++++++++ > > > drivers/gpu/drm/i915/gt/intel_engine_cs.c | 20 +++++++++++++++++++- > > > drivers/gpu/drm/i915/gt/intel_engine_types.h | 4 ++++ > > > 3 files changed, 34 insertions(+), 1 deletion(-) > > > > > > diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile > > > index 48df8889a88a..3a3881d5e44b 100644 > > > --- a/drivers/gpu/drm/i915/Kconfig.profile > > > +++ b/drivers/gpu/drm/i915/Kconfig.profile > > > @@ -25,3 +25,14 @@ config DRM_I915_SPIN_REQUEST > > > May be 0 to disable the initial spin. In practice, we estimate > > > the cost of enabling the interrupt (if currently disabled) to be > > > a few microseconds. > > > + > > > +config DRM_I915_STOP_TIMEOUT > > > + int "How long to wait for an engine to quiesce gracefully before reset (ms)" > > > + default 100 # milliseconds > > > + help > > > + By stopping submission and sleeping for a short time before resetting > > > + the GPU, we allow the innocent contexts also on the system to quiesce. > > > + It is then less likely for a hanging context to cause collateral > > > + damage as the system is reset in order to recover. The corollary is > > > + that the reset itself may take longer and so be more disruptive to > > > + interactive or low latency workloads. > > > diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c > > > index 0e20713603ec..e4203eb44139 100644 > > > --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c > > > +++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c > > > @@ -308,6 +308,9 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id) > > > engine->instance = info->instance; > > > __sprint_engine_name(engine); > > > > > > + engine->props.stop_timeout_ms = > > > + CONFIG_DRM_I915_STOP_TIMEOUT; > > Compare to previous version where you used the CONFIG variable directly, > > what is the benefit of using it through a variable? So that we could > > alter it in runtime? > > That is what the next series proposes. Also it turns out to be more > easily testable if one is able to alter the timeouts for testing. Thanks for explaining Chris! -Ram. > -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [Intel-gfx] [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets 2019-10-24 6:37 ` Ramalingam C @ 2019-10-24 6:37 ` Ramalingam C 0 siblings, 0 replies; 18+ messages in thread From: Ramalingam C @ 2019-10-24 6:37 UTC (permalink / raw) To: Chris Wilson; +Cc: intel-gfx On 2019-10-24 at 07:14:56 +0100, Chris Wilson wrote: > Quoting Ramalingam C (2019-10-24 02:32:01) > > On 2019-10-23 at 14:31:04 +0100, Chris Wilson wrote: > > > If we are doing a normal GPU reset triggered after detecting a long > > > period of stalled work, we can take our time and allow the engines to > > > quiesce. Since we've stopped submission to the engine, and if we wait > > > long enough an innocent context should complete, leaving the engine idle. > > > So by waiting a short amount of time, we should prevent clobbering other > > > users when resetting a stuck context. > > > > > > Suggested-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > > > Suggested-by: Jon Bloomfield <jon.bloomfield@intel.com> > > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > > > Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> > > > Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > > > Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com> > > > --- > > > drivers/gpu/drm/i915/Kconfig.profile | 11 +++++++++++ > > > drivers/gpu/drm/i915/gt/intel_engine_cs.c | 20 +++++++++++++++++++- > > > drivers/gpu/drm/i915/gt/intel_engine_types.h | 4 ++++ > > > 3 files changed, 34 insertions(+), 1 deletion(-) > > > > > > diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile > > > index 48df8889a88a..3a3881d5e44b 100644 > > > --- a/drivers/gpu/drm/i915/Kconfig.profile > > > +++ b/drivers/gpu/drm/i915/Kconfig.profile > > > @@ -25,3 +25,14 @@ config DRM_I915_SPIN_REQUEST > > > May be 0 to disable the initial spin. In practice, we estimate > > > the cost of enabling the interrupt (if currently disabled) to be > > > a few microseconds. > > > + > > > +config DRM_I915_STOP_TIMEOUT > > > + int "How long to wait for an engine to quiesce gracefully before reset (ms)" > > > + default 100 # milliseconds > > > + help > > > + By stopping submission and sleeping for a short time before resetting > > > + the GPU, we allow the innocent contexts also on the system to quiesce. > > > + It is then less likely for a hanging context to cause collateral > > > + damage as the system is reset in order to recover. The corollary is > > > + that the reset itself may take longer and so be more disruptive to > > > + interactive or low latency workloads. > > > diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c > > > index 0e20713603ec..e4203eb44139 100644 > > > --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c > > > +++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c > > > @@ -308,6 +308,9 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id) > > > engine->instance = info->instance; > > > __sprint_engine_name(engine); > > > > > > + engine->props.stop_timeout_ms = > > > + CONFIG_DRM_I915_STOP_TIMEOUT; > > Compare to previous version where you used the CONFIG variable directly, > > what is the benefit of using it through a variable? So that we could > > alter it in runtime? > > That is what the next series proposes. Also it turns out to be more > easily testable if one is able to alter the timeouts for testing. Thanks for explaining Chris! -Ram. > -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx ^ permalink raw reply [flat|nested] 18+ messages in thread
end of thread, other threads:[~2019-10-24 6:37 UTC | newest] Thread overview: 18+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2019-10-23 13:31 [CI 1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets Chris Wilson 2019-10-23 13:31 ` [Intel-gfx] " Chris Wilson 2019-10-23 13:31 ` [CI 2/5] drm/i915/execlists: Force preemption Chris Wilson 2019-10-23 13:31 ` [Intel-gfx] " Chris Wilson 2019-10-23 13:31 ` [CI 3/5] drm/i915/execlists: Cancel banned contexts on schedule-out Chris Wilson 2019-10-23 13:31 ` [Intel-gfx] " Chris Wilson 2019-10-23 13:31 ` [CI 4/5] drm/i915/gem: Cancel contexts when hangchecking is disabled Chris Wilson 2019-10-23 13:31 ` [Intel-gfx] " Chris Wilson 2019-10-23 13:31 ` [CI 5/5] drm/i915/gt: Replace hangcheck by heartbeats Chris Wilson 2019-10-23 13:31 ` [Intel-gfx] " Chris Wilson 2019-10-24 0:09 ` ✗ Fi.CI.BUILD: failure for series starting with [CI,1/5] drm/i915/gt: Try to more gracefully quiesce the system before resets Patchwork 2019-10-24 0:09 ` [Intel-gfx] " Patchwork 2019-10-24 1:32 ` [CI 1/5] " Ramalingam C 2019-10-24 1:32 ` [Intel-gfx] " Ramalingam C 2019-10-24 6:14 ` Chris Wilson 2019-10-24 6:14 ` [Intel-gfx] " Chris Wilson 2019-10-24 6:37 ` Ramalingam C 2019-10-24 6:37 ` [Intel-gfx] " Ramalingam C
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).