From patchwork Fri May 24 15:33:38 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Chris Wilson X-Patchwork-Id: 10960031 Return-Path: Received: from mail.wl.linuxfoundation.org (pdx-wl-mail.web.codeaurora.org [172.30.200.125]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id 707F31390 for ; Fri, 24 May 2019 15:34:13 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 601EE287CE for ; Fri, 24 May 2019 15:34:13 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id 5483F288DB; Fri, 24 May 2019 15:34:13 +0000 (UTC) X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on pdx-wl-mail.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-5.2 required=2.0 tests=BAYES_00,MAILING_LIST_MULTI, RCVD_IN_DNSWL_MED autolearn=ham version=3.3.1 Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher DHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by mail.wl.linuxfoundation.org (Postfix) with ESMTPS id AB08D28908 for ; Fri, 24 May 2019 15:34:12 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id EA61C6E0B7; Fri, 24 May 2019 15:34:11 +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 5FFA66E0B7 for ; Fri, 24 May 2019 15:34:10 +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 16670817-1500050 for multiple; Fri, 24 May 2019 16:33:41 +0100 From: Chris Wilson To: intel-gfx@lists.freedesktop.org Date: Fri, 24 May 2019 16:33:38 +0100 Message-Id: <20190524153338.8877-1-chris@chris-wilson.co.uk> X-Mailer: git-send-email 2.20.1 In-Reply-To: <20190524151223.17879-1-chris@chris-wilson.co.uk> References: <20190524151223.17879-1-chris@chris-wilson.co.uk> MIME-Version: 1.0 Subject: [Intel-gfx] [PATCH] drm/i915: Keep user GGTT alive for a minimum of 250ms 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" X-Virus-Scanned: ClamAV using ClamSMTP Do not allow runtime pm autosuspend to remove userspace GGTT mmaps too quickly. For example, igt sets the autosuspend delay to 0, and so we immediately attempt to perform runtime suspend upon releasing the wakeref. Unfortunately, that involves tearing down GGTT mmaps as they require an active device. Override the autosuspend for GGTT mmaps, by keeping the wakeref around for 250ms after populating the PTE for a fresh mmap. v2: Prefer refcount_t for its under/overflow error detection Signed-off-by: Chris Wilson Cc: Mika Kuoppala Reviewed-by: Mika Kuoppala --- drivers/gpu/drm/i915/Kconfig.profile | 14 +++++++ drivers/gpu/drm/i915/i915_drv.h | 3 ++ drivers/gpu/drm/i915/i915_gem.c | 7 ++++ drivers/gpu/drm/i915/intel_wakeref.c | 58 ++++++++++++++++++++++++++++ drivers/gpu/drm/i915/intel_wakeref.h | 28 ++++++++++++++ 5 files changed, 110 insertions(+) diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile index 0e5db98da8f3..4fd1ea639d0f 100644 --- a/drivers/gpu/drm/i915/Kconfig.profile +++ b/drivers/gpu/drm/i915/Kconfig.profile @@ -1,3 +1,17 @@ +config DRM_I915_USERFAULT_AUTOSUSPEND + int "Runtime autosuspend delay for userspace GGTT mmaps (ms)" + default 250 # milliseconds + help + On runtime suspend, as we suspend the device, we have to revoke + userspace GGTT mmaps and force userspace to take a pagefault on + their next access. The revocation and subsequent recreation of + the GGTT mmap can be very slow and so we impose a small hysteris + that complements the runtime-pm autosuspend and provides a lower + floor on the autosuspend delay. + + May be 0 to disable the extra delay and solely use the device level + runtime pm autosuspend delay tunable. + config DRM_I915_SPIN_REQUEST int default 5 # microseconds diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index 311e19154672..41dad8889eaa 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -874,6 +874,9 @@ struct i915_gem_mm { */ struct list_head userfault_list; + /* Manual runtime pm autosuspend delay for user GGTT mmaps */ + struct intel_wakeref_auto userfault_wakeref; + /** * List of objects which are pending destruction. */ diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index d3b7dac527dc..cb786582e2fe 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -1834,6 +1834,9 @@ vm_fault_t i915_gem_fault(struct vm_fault *vmf) assert_rpm_wakelock_held(dev_priv); if (!i915_vma_set_userfault(vma) && !obj->userfault_count++) list_add(&obj->userfault_link, &dev_priv->mm.userfault_list); + if (CONFIG_DRM_I915_USERFAULT_AUTOSUSPEND) + intel_wakeref_auto(&dev_priv->mm.userfault_wakeref, + msecs_to_jiffies_timeout(CONFIG_DRM_I915_USERFAULT_AUTOSUSPEND)); GEM_BUG_ON(!obj->userfault_count); i915_vma_set_ggtt_write(vma); @@ -4671,6 +4674,8 @@ void i915_gem_fini(struct drm_i915_private *dev_priv) { GEM_BUG_ON(dev_priv->gt.awake); + intel_wakeref_auto_fini(&dev_priv->mm.userfault_wakeref); + i915_gem_suspend_late(dev_priv); intel_disable_gt_powersave(dev_priv); @@ -4748,6 +4753,8 @@ static void i915_gem_init__mm(struct drm_i915_private *i915) INIT_LIST_HEAD(&i915->mm.fence_list); INIT_LIST_HEAD(&i915->mm.userfault_list); + intel_wakeref_auto_init(&i915->mm.userfault_wakeref, i915); + INIT_WORK(&i915->mm.free_work, __i915_gem_free_work); } diff --git a/drivers/gpu/drm/i915/intel_wakeref.c b/drivers/gpu/drm/i915/intel_wakeref.c index 91196d9612bb..78b624c1ed59 100644 --- a/drivers/gpu/drm/i915/intel_wakeref.c +++ b/drivers/gpu/drm/i915/intel_wakeref.c @@ -73,3 +73,61 @@ void __intel_wakeref_init(struct intel_wakeref *wf, struct lock_class_key *key) atomic_set(&wf->count, 0); wf->wakeref = 0; } + +static void wakeref_auto_timeout(struct timer_list *t) +{ + struct intel_wakeref_auto *wf = from_timer(wf, t, timer); + intel_wakeref_t wakeref; + unsigned long flags; + + if (!refcount_dec_and_lock_irqsave(&wf->count, &wf->lock, &flags)) + return; + + wakeref = fetch_and_zero(&wf->wakeref); + spin_unlock_irqrestore(&wf->lock, flags); + + intel_runtime_pm_put(wf->i915, wakeref); +} + +void intel_wakeref_auto_init(struct intel_wakeref_auto *wf, + struct drm_i915_private *i915) +{ + spin_lock_init(&wf->lock); + timer_setup(&wf->timer, wakeref_auto_timeout, 0); + refcount_set(&wf->count, 0); + wf->wakeref = 0; + wf->i915 = i915; +} + +void intel_wakeref_auto(struct intel_wakeref_auto *wf, long timeout) +{ + unsigned long flags; + + assert_rpm_wakelock_held(wf->i915); + + if (!refcount_inc_not_zero(&wf->count)) { + spin_lock_irqsave(&wf->lock, flags); + if (!refcount_read(&wf->count)) { + GEM_BUG_ON(wf->wakeref); + wf->wakeref = intel_runtime_pm_get_if_in_use(wf->i915); + } + refcount_inc(&wf->count); + spin_unlock_irqrestore(&wf->lock, flags); + } + + /* + * If we extend a pending timer, we will only get a single timer + * callback and so need to cancel the local inc by running the + * elided callback to keep the wf->count balanced. + */ + if (mod_timer(&wf->timer, jiffies + timeout)) + wakeref_auto_timeout(&wf->timer); +} + +void intel_wakeref_auto_fini(struct intel_wakeref_auto *wf) +{ + if (del_timer_sync(&wf->timer)) + wakeref_auto_timeout(&wf->timer); + + GEM_BUG_ON(wf->wakeref); +} diff --git a/drivers/gpu/drm/i915/intel_wakeref.h b/drivers/gpu/drm/i915/intel_wakeref.h index db742291211c..221bb237efd8 100644 --- a/drivers/gpu/drm/i915/intel_wakeref.h +++ b/drivers/gpu/drm/i915/intel_wakeref.h @@ -9,7 +9,9 @@ #include #include +#include #include +#include struct drm_i915_private; @@ -130,4 +132,30 @@ intel_wakeref_active(struct intel_wakeref *wf) return READ_ONCE(wf->wakeref); } +struct intel_wakeref_auto { + struct drm_i915_private *i915; + struct timer_list timer; + intel_wakeref_t wakeref; + spinlock_t lock; + refcount_t count; +}; + +/** + * intel_wakeref_auto: Delay the runtime-pm autosuspend + * @wf: the wakeref + * @timeout: relative timeout in jiffies + * + * The runtime-pm core uses a suspend delay after the last wakeref + * is released before triggering runtime suspend of the device. That + * delay is configurable via sysfs with little regard to the device + * characteristics. Instead, we want to tune the autosuspend based on our + * HW knowledge. intel_wakeref_auto() delays the sleep by the supplied + * timeout. + */ +void intel_wakeref_auto(struct intel_wakeref_auto *wf, long timeout); + +void intel_wakeref_auto_init(struct intel_wakeref_auto *wf, + struct drm_i915_private *i915); +void intel_wakeref_auto_fini(struct intel_wakeref_auto *wf); + #endif /* INTEL_WAKEREF_H */