From patchwork Sat Nov 16 17:51:39 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Chris Wilson X-Patchwork-Id: 11247651 Return-Path: Received: from mail.kernel.org (pdx-korg-mail-1.web.codeaurora.org [172.30.200.123]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id 2AF201393 for ; Sat, 16 Nov 2019 17:52:08 +0000 (UTC) Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by mail.kernel.org (Postfix) with ESMTPS id 12D5821843 for ; Sat, 16 Nov 2019 17:52:08 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org 12D5821843 Authentication-Results: mail.kernel.org; dmarc=none (p=none dis=none) header.from=chris-wilson.co.uk Authentication-Results: mail.kernel.org; spf=none smtp.mailfrom=intel-gfx-bounces@lists.freedesktop.org Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 53A1D6E130; Sat, 16 Nov 2019 17:52:07 +0000 (UTC) X-Original-To: intel-gfx@lists.freedesktop.org Delivered-To: intel-gfx@lists.freedesktop.org Received: from fireflyinternet.com (mail.fireflyinternet.com [109.228.58.192]) by gabe.freedesktop.org (Postfix) with ESMTPS id 09C186E130 for ; Sat, 16 Nov 2019 17:51:52 +0000 (UTC) X-Default-Received-SPF: pass (skip=forwardok (res=PASS)) x-ip-name=78.156.65.138; Received: from haswell.alporthouse.com (unverified [78.156.65.138]) by fireflyinternet.com (Firefly Internet (M1)) with ESMTP id 19226428-1500050 for multiple; Sat, 16 Nov 2019 17:51:43 +0000 From: Chris Wilson To: intel-gfx@lists.freedesktop.org Date: Sat, 16 Nov 2019 17:51:39 +0000 Message-Id: <20191116175139.1790440-11-chris@chris-wilson.co.uk> X-Mailer: git-send-email 2.24.0 In-Reply-To: <20191116175139.1790440-1-chris@chris-wilson.co.uk> References: <20191116175139.1790440-1-chris@chris-wilson.co.uk> MIME-Version: 1.0 Subject: [Intel-gfx] [PATCH 11/11] drm/i915/gt: Track engine round-trip times X-BeenThere: intel-gfx@lists.freedesktop.org X-Mailman-Version: 2.1.23 Precedence: list List-Id: Intel graphics driver community testing & development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: intel-gfx-bounces@lists.freedesktop.org Sender: "Intel-gfx" Knowing the round trip time of an engine is useful for tracking the health of the system as well as providing a metric for the baseline responsiveness of the engine. We can use the latter metric for automatically tuning our waits in selftests and when idling so we don't confuse a slower system with a dead one. Upon idling the engine, we send one last pulse to switch the context away from precious user state to the volatile kernel context. We know the engine is idle at this point, and the pulse is non-preemptable, so this provides us with a good measurement of the round trip time. A secondary effect is that by installing an interrupt onto the pulse, we can flush the engine immediately upon completion, curtailing the background flush and entering powersaving immediately. v2: Manage pm wakerefs to avoid too early module unload References: 7e34f4e4aad3 ("drm/i915/gen8+: Add RC6 CTX corruption WA") Signed-off-by: Chris Wilson Cc: Tvrtko Ursulin Cc: Stuart Summers --- drivers/gpu/drm/i915/gt/intel_engine_cs.c | 2 ++ drivers/gpu/drm/i915/gt/intel_engine_pm.c | 36 +++++++++++++++++++- drivers/gpu/drm/i915/gt/intel_engine_types.h | 11 ++++++ drivers/gpu/drm/i915/gt/intel_gt_pm.h | 5 +++ drivers/gpu/drm/i915/intel_wakeref.h | 13 +++++++ 5 files changed, 66 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c index b9613d044393..2d11db13dc89 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c +++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c @@ -334,6 +334,7 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id) /* Nothing to do here, execute in order of dependencies */ engine->schedule = NULL; + ewma_delay_init(&engine->delay); seqlock_init(&engine->stats.lock); ATOMIC_INIT_NOTIFIER_HEAD(&engine->context_status_notifier); @@ -1477,6 +1478,7 @@ 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, "\tDelay: %luus\n", ewma_delay_read(&engine->delay)); rcu_read_lock(); rq = READ_ONCE(engine->heartbeat.systole); diff --git a/drivers/gpu/drm/i915/gt/intel_engine_pm.c b/drivers/gpu/drm/i915/gt/intel_engine_pm.c index 3c0f490ff2c7..e79b14e4599b 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_pm.c +++ b/drivers/gpu/drm/i915/gt/intel_engine_pm.c @@ -73,6 +73,24 @@ static inline void __timeline_mark_unlock(struct intel_context *ce, #endif /* !IS_ENABLED(CONFIG_LOCKDEP) */ +struct duration_cb { + struct dma_fence_cb cb; + ktime_t emitted; +}; + +static void duration_cb(struct dma_fence *fence, struct dma_fence_cb *cb) +{ + struct duration_cb *dcb = container_of(cb, typeof(*dcb), cb); + struct intel_engine_cs *engine = to_request(fence)->engine; + + ewma_delay_add(&engine->delay, + ktime_us_delta(ktime_get(), dcb->emitted)); + + /* Kick retire for quicker powersaving (soft-rc6). */ + mod_delayed_work(system_wq, &engine->gt->requests.retire_work, 0); + intel_gt_pm_put(engine->gt); +} + static bool switch_to_kernel_context(struct intel_engine_cs *engine) { struct i915_request *rq; @@ -114,7 +132,23 @@ static bool switch_to_kernel_context(struct intel_engine_cs *engine) /* Install ourselves as a preemption barrier */ rq->sched.attr.priority = I915_PRIORITY_BARRIER; - __i915_request_commit(rq); + if (likely(!__i915_request_commit(rq))) { /* engine should be idle! */ + struct duration_cb *dcb; + + BUILD_BUG_ON(sizeof(*dcb) > sizeof(rq->submitq)); + dcb = (struct duration_cb *)&rq->submitq; + + /* + * Use an interrupt for precise measurement of duration, + * otherwise we rely on someone else retiring all the requests + * which may delay the signaling (i.e. we will likely wait + * until the background request retirement running every + * second or two). + */ + __intel_gt_pm_get(rq->engine->gt); + dma_fence_add_callback(&rq->fence, &dcb->cb, duration_cb); + dcb->emitted = ktime_get(); + } /* Release our exclusive hold on the engine */ __intel_wakeref_defer_park(&engine->wakeref); diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h index 758f0e8ec672..1de121583482 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_types.h +++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h @@ -7,6 +7,7 @@ #ifndef __INTEL_ENGINE_TYPES__ #define __INTEL_ENGINE_TYPES__ +#include #include #include #include @@ -119,6 +120,9 @@ enum intel_engine_id { #define INVALID_ENGINE ((enum intel_engine_id)-1) }; +/* A simple estimator for the round-trip responsive time of an engine */ +DECLARE_EWMA(delay, 6, 4) + struct st_preempt_hang { struct completion completion; unsigned int count; @@ -316,6 +320,13 @@ struct intel_engine_cs { struct intel_timeline *timeline; } legacy; + /* + * We track the average duration of the idle pulse on parking the + * engine to keep an estimate of the how the fast the engine is + * under ideal conditions. + */ + struct ewma_delay delay; + /* Rather than have every client wait upon all user interrupts, * with the herd waking after every interrupt and each doing the * heavyweight seqno dance, we delegate the task (of being the diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm.h b/drivers/gpu/drm/i915/gt/intel_gt_pm.h index 990efc27a4e4..4a9e48c12bd4 100644 --- a/drivers/gpu/drm/i915/gt/intel_gt_pm.h +++ b/drivers/gpu/drm/i915/gt/intel_gt_pm.h @@ -22,6 +22,11 @@ static inline void intel_gt_pm_get(struct intel_gt *gt) intel_wakeref_get(>->wakeref); } +static inline void __intel_gt_pm_get(struct intel_gt *gt) +{ + __intel_wakeref_get(>->wakeref); +} + static inline bool intel_gt_pm_get_if_awake(struct intel_gt *gt) { return intel_wakeref_get_if_active(>->wakeref); diff --git a/drivers/gpu/drm/i915/intel_wakeref.h b/drivers/gpu/drm/i915/intel_wakeref.h index eeac3e7dfafe..9221067269ce 100644 --- a/drivers/gpu/drm/i915/intel_wakeref.h +++ b/drivers/gpu/drm/i915/intel_wakeref.h @@ -81,6 +81,19 @@ intel_wakeref_get(struct intel_wakeref *wf) return 0; } +/** + * __intel_wakeref_get: Acquire the wakeref, again + * @wf: the wakeref + * + * Acquire a hold on an already active wakeref. + */ +static inline void +__intel_wakeref_get(struct intel_wakeref *wf) +{ + INTEL_WAKEREF_BUG_ON(atomic_read(&wf->count) <= 0); + atomic_inc(&wf->count); +} + /** * intel_wakeref_get_if_in_use: Acquire the wakeref * @wf: the wakeref