On Thu, Aug 26, 2021 at 11:52:14AM +0100, Tvrtko Ursulin wrote:
> From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
>
> When a non-persistent context exits we currently mark it as banned in
> order to trigger fast termination of any outstanding GPU jobs it may have
> left running.
>
> In doing so we apply a very strict 1ms limit in which the left over job
> has to preempt before we issues an engine resets.
>
> Some workloads are not able to cleanly preempt in that time window and it
> can be argued that it would instead be better to give them a bit more
> grace since avoiding engine resets is generally preferrable.
>
> To achieve this the patch splits handling of banned contexts from simply
> closed non-persistent ones and then applies different timeouts for both
> and also extends the criteria which determines if a request should be
> scheduled back in after preemption or not.
>
> 20ms preempt timeout grace is given to exited non-persistent contexts
> which have been empirically tested to satisfy customers requirements
> and still provides reasonably quick cleanup post exit.
>
> v2:
> * Streamline fast path checks.
>
> v3:
> * Simplify by using only schedulable status.
> * Increase timeout to 20ms.
>
> v4:
> * Fix live_execlists selftest.
>
> v5:
> * Fix logic in kill_engines.
>
> v6:
> * Rebase.
>
> v7:
> * Add GuC support.
>
> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
> Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
> Cc: Zhen Han <zhen.han@xxxxxxxxx>
> Cc: Matthew Brost <matthew.brost@xxxxxxxxx>
> ---
> drivers/gpu/drm/i915/gem/i915_gem_context.c | 22 +++++++++++-----
> drivers/gpu/drm/i915/gt/intel_context.c | 25 ++++++++++++++++++
> drivers/gpu/drm/i915/gt/intel_context.h | 26 ++++++++++++++-----
> drivers/gpu/drm/i915/gt/intel_context_types.h | 3 ++-
> .../drm/i915/gt/intel_execlists_submission.c | 13 +++++++---
> .../gpu/drm/i915/gt/intel_ring_submission.c | 7 ++---
> .../gpu/drm/i915/gt/uc/intel_guc_submission.c | 13 ++++++----
> drivers/gpu/drm/i915/i915_request.c | 2 +-
> 8 files changed, 84 insertions(+), 27 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c b/drivers/gpu/drm/i915/gem/i915_gem_context.c
> index fd169cf2f75a..6ae803cb4de3 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_context.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c
> @@ -1072,7 +1072,8 @@ static struct intel_engine_cs *active_engine(struct intel_context *ce)
> return engine;
> }
>
> -static void kill_engines(struct i915_gem_engines *engines, bool ban)
> +static void
> +kill_engines(struct i915_gem_engines *engines, bool ban, bool persistent)
> {
> struct i915_gem_engines_iter it;
> struct intel_context *ce;
> @@ -1086,8 +1087,15 @@ static void kill_engines(struct i915_gem_engines *engines, bool ban)
> */
> for_each_gem_engine(ce, engines, it) {
> struct intel_engine_cs *engine;
> + bool skip = false;
>
> - if (ban && intel_context_ban(ce, NULL))
> + if (ban)
> + skip = intel_context_ban(ce, NULL);
> + else if (!persistent)
> + skip = intel_context_exit_nonpersistent(ce, NULL);
> +
> + /* Already banned or non-persistent closed. */
> + if (skip)
> continue;
>
> /*
> @@ -1100,7 +1108,7 @@ static void kill_engines(struct i915_gem_engines *engines, bool ban)
> engine = active_engine(ce);
>
> /* First attempt to gracefully cancel the context */
> - if (engine && !__cancel_engine(engine) && ban)
> + if (engine && !__cancel_engine(engine) && (ban || !persistent))
> /*
> * If we are unable to send a preemptive pulse to bump
> * the context from the GPU, we have to resort to a full
> @@ -1112,8 +1120,6 @@ static void kill_engines(struct i915_gem_engines *engines, bool ban)
>
> static void kill_context(struct i915_gem_context *ctx)
> {
> - bool ban = (!i915_gem_context_is_persistent(ctx) ||
> - !ctx->i915->params.enable_hangcheck);
> struct i915_gem_engines *pos, *next;
>
> spin_lock_irq(&ctx->stale.lock);
> @@ -1126,7 +1132,8 @@ static void kill_context(struct i915_gem_context *ctx)
>
> spin_unlock_irq(&ctx->stale.lock);
>
> - kill_engines(pos, ban);
> + kill_engines(pos, !ctx->i915->params.enable_hangcheck,
> + i915_gem_context_is_persistent(ctx));
>
> spin_lock_irq(&ctx->stale.lock);
> GEM_BUG_ON(i915_sw_fence_signaled(&pos->fence));
> @@ -1172,7 +1179,8 @@ static void engines_idle_release(struct i915_gem_context *ctx,
>
> kill:
> if (list_empty(&engines->link)) /* raced, already closed */
> - kill_engines(engines, true);
> + kill_engines(engines, true,
> + i915_gem_context_is_persistent(ctx));
>
> i915_sw_fence_commit(&engines->fence);
> }
> diff --git a/drivers/gpu/drm/i915/gt/intel_context.c b/drivers/gpu/drm/i915/gt/intel_context.c
> index 745e84c72c90..b9880ffe5da7 100644
> --- a/drivers/gpu/drm/i915/gt/intel_context.c
> +++ b/drivers/gpu/drm/i915/gt/intel_context.c
> @@ -533,6 +533,31 @@ struct i915_request *intel_context_find_active_request(struct intel_context *ce)
> return active;
> }
>
> +bool intel_context_ban(struct intel_context *ce, struct i915_request *rq)
> +{
> + bool ret = intel_context_set_banned(ce);
> +
> + trace_intel_context_ban(ce);
> +
> + if (ce->ops->revoke)
> + ce->ops->revoke(ce, rq,
> + INTEL_CONTEXT_BANNED_PREEMPT_TIMEOUT_MS);
> +
> + return ret;
> +}
> +
> +bool intel_context_exit_nonpersistent(struct intel_context *ce,
> + struct i915_request *rq)
> +{
> + bool ret = intel_context_set_exiting(ce);
> +
> + if (ce->ops->revoke)
> + ce->ops->revoke(ce, rq,
> + INTEL_CONTEXT_EXITING_PREEMPT_TIMEOUT_MS);
> +
> + return ret;
> +}
> +
> #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
> #include "selftest_context.c"
> #endif
> diff --git a/drivers/gpu/drm/i915/gt/intel_context.h b/drivers/gpu/drm/i915/gt/intel_context.h
> index c41098950746..8b77f3e7449a 100644
> --- a/drivers/gpu/drm/i915/gt/intel_context.h
> +++ b/drivers/gpu/drm/i915/gt/intel_context.h
> @@ -25,6 +25,9 @@
> ##__VA_ARGS__); \
> } while (0)
>
> +#define INTEL_CONTEXT_EXITING_PREEMPT_TIMEOUT_MS (20)
> +#define INTEL_CONTEXT_BANNED_PREEMPT_TIMEOUT_MS (1)
> +
> struct i915_gem_ww_ctx;
>
> void intel_context_init(struct intel_context *ce,
> @@ -254,18 +257,27 @@ static inline bool intel_context_set_banned(struct intel_context *ce)
> return test_and_set_bit(CONTEXT_BANNED, &ce->flags);
> }
>
> -static inline bool intel_context_ban(struct intel_context *ce,
> - struct i915_request *rq)
> +bool intel_context_ban(struct intel_context *ce, struct i915_request *rq);
> +
> +static inline bool intel_context_is_schedulable(const struct intel_context *ce)
> {
> - bool ret = intel_context_set_banned(ce);
> + return !test_bit(CONTEXT_EXITING, &ce->flags) &&
> + !test_bit(CONTEXT_BANNED, &ce->flags);
> +}
>
> - trace_intel_context_ban(ce);
> - if (ce->ops->ban)
> - ce->ops->ban(ce, rq);
> +static inline bool intel_context_is_exiting(const struct intel_context *ce)
> +{
> + return test_bit(CONTEXT_EXITING, &ce->flags);
> +}
>
> - return ret;
> +static inline bool intel_context_set_exiting(struct intel_context *ce)
> +{
> + return test_and_set_bit(CONTEXT_EXITING, &ce->flags);
Uh more unordered atomics without barriers and without comments explaining
the barriers and pointing at the counter-party barrier. This needs to be
fixed.
In general it would be really good if all the context status bits are at
least properly documented. As-is this is an extremely challenging maze,
and I'm not sure it's really correct in all cases.
-Daniel
> }
>
> +bool intel_context_exit_nonpersistent(struct intel_context *ce,
> + struct i915_request *rq);
> +
> static inline bool
> intel_context_force_single_submission(const struct intel_context *ce)
> {
> diff --git a/drivers/gpu/drm/i915/gt/intel_context_types.h b/drivers/gpu/drm/i915/gt/intel_context_types.h
> index e54351a170e2..4a0a1e736972 100644
> --- a/drivers/gpu/drm/i915/gt/intel_context_types.h
> +++ b/drivers/gpu/drm/i915/gt/intel_context_types.h
> @@ -37,7 +37,7 @@ struct intel_context_ops {
>
> int (*alloc)(struct intel_context *ce);
>
> - void (*ban)(struct intel_context *ce, struct i915_request *rq);
> + void (*revoke)(struct intel_context *ce, struct i915_request *rq, unsigned int preempt_timeout_ms);
>
> int (*pre_pin)(struct intel_context *ce, struct i915_gem_ww_ctx *ww, void **vaddr);
> int (*pin)(struct intel_context *ce, void *vaddr);
> @@ -112,6 +112,7 @@ struct intel_context {
> #define CONTEXT_FORCE_SINGLE_SUBMISSION 7
> #define CONTEXT_NOPREEMPT 8
> #define CONTEXT_LRCA_DIRTY 9
> +#define CONTEXT_EXITING 10
>
> struct {
> u64 timeout_us;
> diff --git a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
> index de5f9c86b9a4..43823b9d4013 100644
> --- a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
> +++ b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
> @@ -478,7 +478,7 @@ __execlists_schedule_in(struct i915_request *rq)
> !intel_engine_has_heartbeat(engine)))
> intel_context_set_banned(ce);
>
> - if (unlikely(intel_context_is_banned(ce) || bad_request(rq)))
> + if (unlikely(!intel_context_is_schedulable(ce) || bad_request(rq)))
> reset_active(rq, engine);
>
> if (IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM))
> @@ -1222,12 +1222,19 @@ static void record_preemption(struct intel_engine_execlists *execlists)
> static unsigned long active_preempt_timeout(struct intel_engine_cs *engine,
> const struct i915_request *rq)
> {
> + struct intel_context *ce;
> +
> if (!rq)
> return 0;
>
> + ce = rq->context;
> +
> /* Force a fast reset for terminated contexts (ignoring sysfs!) */
> - if (unlikely(intel_context_is_banned(rq->context) || bad_request(rq)))
> - return 1;
> + if (unlikely(intel_context_is_banned(ce) || bad_request(rq)))
> + return INTEL_CONTEXT_BANNED_PREEMPT_TIMEOUT_MS;
> + /* Longer grace for closed non-persistent contexts to avoid resets. */
> + else if (unlikely(intel_context_is_exiting(ce)))
> + return INTEL_CONTEXT_EXITING_PREEMPT_TIMEOUT_MS;
>
> return READ_ONCE(engine->props.preempt_timeout_ms);
> }
> diff --git a/drivers/gpu/drm/i915/gt/intel_ring_submission.c b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
> index 3c65efcb7bed..a28862bbe51e 100644
> --- a/drivers/gpu/drm/i915/gt/intel_ring_submission.c
> +++ b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
> @@ -587,8 +587,9 @@ static void ring_context_reset(struct intel_context *ce)
> clear_bit(CONTEXT_VALID_BIT, &ce->flags);
> }
>
> -static void ring_context_ban(struct intel_context *ce,
> - struct i915_request *rq)
> +static void ring_context_revoke(struct intel_context *ce,
> + struct i915_request *rq,
> + unsigned int preempt_timeout_ms)
> {
> struct intel_engine_cs *engine;
>
> @@ -623,7 +624,7 @@ static const struct intel_context_ops ring_context_ops = {
>
> .cancel_request = ring_context_cancel_request,
>
> - .ban = ring_context_ban,
> + .revoke = ring_context_revoke,
>
> .pre_pin = ring_context_pre_pin,
> .pin = ring_context_pin,
> diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
> index 87d8dc8f51b9..b0a8629b26e3 100644
> --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
> +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
> @@ -1618,7 +1618,9 @@ static void __guc_context_set_preemption_timeout(struct intel_guc *guc,
> intel_guc_send_busy_loop(guc, action, ARRAY_SIZE(action), 0, true);
> }
>
> -static void guc_context_ban(struct intel_context *ce, struct i915_request *rq)
> +static void
> +guc_context_revoke(struct intel_context *ce, struct i915_request *rq,
> + unsigned int preempt_timeout_ms)
> {
> struct intel_guc *guc = ce_to_guc(ce);
> struct intel_runtime_pm *runtime_pm =
> @@ -1655,7 +1657,8 @@ static void guc_context_ban(struct intel_context *ce, struct i915_request *rq)
> * gets kicked off the HW ASAP.
> */
> with_intel_runtime_pm(runtime_pm, wakeref) {
> - __guc_context_set_preemption_timeout(guc, guc_id, 1);
> + __guc_context_set_preemption_timeout(guc, guc_id,
> + preempt_timeout_ms);
> __guc_context_sched_disable(guc, ce, guc_id);
> }
> } else {
> @@ -1663,7 +1666,7 @@ static void guc_context_ban(struct intel_context *ce, struct i915_request *rq)
> with_intel_runtime_pm(runtime_pm, wakeref)
> __guc_context_set_preemption_timeout(guc,
> ce->guc_id,
> - 1);
> + preempt_timeout_ms);
> spin_unlock_irqrestore(&ce->guc_state.lock, flags);
> }
> }
> @@ -1979,7 +1982,7 @@ static const struct intel_context_ops guc_context_ops = {
> .unpin = guc_context_unpin,
> .post_unpin = guc_context_post_unpin,
>
> - .ban = guc_context_ban,
> + .revoke = guc_context_revoke,
>
> .cancel_request = guc_context_cancel_request,
>
> @@ -2176,7 +2179,7 @@ static const struct intel_context_ops virtual_guc_context_ops = {
> .unpin = guc_context_unpin,
> .post_unpin = guc_context_post_unpin,
>
> - .ban = guc_context_ban,
> + .revoke = guc_context_revoke,
>
> .cancel_request = guc_context_cancel_request,
>
> diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
> index ce446716d092..b1a9bec83339 100644
> --- a/drivers/gpu/drm/i915/i915_request.c
> +++ b/drivers/gpu/drm/i915/i915_request.c
> @@ -583,7 +583,7 @@ bool __i915_request_submit(struct i915_request *request)
> goto active;
> }
>
> - if (unlikely(intel_context_is_banned(request->context)))
> + if (unlikely(!intel_context_is_schedulable(request->context)))
> i915_request_set_error_once(request, -EIO);
>
> if (unlikely(fatal_error(request->fence.error)))
> --
> 2.30.2
>
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
