From patchwork Fri Jul 3 11:03:17 2015 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Chris Wilson X-Patchwork-Id: 6714121 Return-Path: X-Original-To: patchwork-intel-gfx@patchwork.kernel.org Delivered-To: patchwork-parsemail@patchwork1.web.kernel.org Received: from mail.kernel.org (mail.kernel.org [198.145.29.136]) by patchwork1.web.kernel.org (Postfix) with ESMTP id 300F29F380 for ; Fri, 3 Jul 2015 11:03:49 +0000 (UTC) Received: from mail.kernel.org (localhost [127.0.0.1]) by mail.kernel.org (Postfix) with ESMTP id 01DCB206FB for ; Fri, 3 Jul 2015 11:03:48 +0000 (UTC) Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) by mail.kernel.org (Postfix) with ESMTP id AB023206D4 for ; Fri, 3 Jul 2015 11:03:46 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 023AB6E317; Fri, 3 Jul 2015 04:03:46 -0700 (PDT) X-Original-To: intel-gfx@lists.freedesktop.org Delivered-To: intel-gfx@lists.freedesktop.org Received: from fireflyinternet.com (mail.fireflyinternet.com [87.106.93.118]) by gabe.freedesktop.org (Postfix) with ESMTP id 3198F6E317 for ; Fri, 3 Jul 2015 04:03:44 -0700 (PDT) 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 41790858-1500048 for multiple; Fri, 03 Jul 2015 12:03:42 +0100 Received: by haswell.alporthouse.com (sSMTP sendmail emulation); Fri, 03 Jul 2015 12:03:20 +0100 From: Chris Wilson To: intel-gfx@lists.freedesktop.org Date: Fri, 3 Jul 2015 12:03:17 +0100 Message-Id: <1435921397-10251-1-git-send-email-chris@chris-wilson.co.uk> X-Mailer: git-send-email 2.1.4 In-Reply-To: <1435683333-17844-3-git-send-email-chris@chris-wilson.co.uk> References: <1435683333-17844-3-git-send-email-chris@chris-wilson.co.uk> MIME-Version: 1.0 X-Originating-IP: 78.156.65.138 X-Country: code=GB country="United Kingdom" ip=78.156.65.138 Subject: [Intel-gfx] [PATCH] drm/i915: Use a task to cancel the userptr on invalidate_range X-BeenThere: intel-gfx@lists.freedesktop.org X-Mailman-Version: 2.1.18 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-Spam-Status: No, score=-4.8 required=5.0 tests=BAYES_00, RCVD_IN_DNSWL_MED, RP_MATCHES_RCVD, UNPARSEABLE_RELAY autolearn=unavailable version=3.3.1 X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on mail.kernel.org X-Virus-Scanned: ClamAV using ClamSMTP Whilst discussing possible ways to trigger an invalidate_range on a userptr with an aliased GGTT mmapping (and so cause a struct_mutex deadlock), the conclusion is that we can, and we must, prevent any possible deadlock by avoiding taking the mutex at all during invalidate_range. This has numerous advantages all of which stem from avoid the sleeping function from inside the unknown context. In particular, it simplifies the invalidate_range because we no longer have to juggle the spinlock/mutex and can just hold the spinlock for the entire walk. To compensate, we have to make get_pages a bit more complicated in order to serialise with a pending cancel_userptr worker. As we hold the struct_mutex, we have no choice but to return EAGAIN and hope that the worker is then flushed before we retry after reacquiring the struct_mutex. The important caveat is that the invalidate_range itself is no longer synchronous. There exists a small but definite period in time in which the old PTE's page remain accessible via the GPU. Note however that the physical pages themselves are not invalidated by the mmu_notifier, just the CPU view of the address space. The impact should be limited to a delay in pages being flushed, rather than a possibility of writing to the wrong pages. The only race condition that this worsens is remapping an userptr active on the GPU where fresh work may still reference the old pages due to struct_mutex contention. Given that userspace is racing with the GPU, it is fair to say that the results are undefined. v2: Only queue (and importantly only take one refcnt) the worker once. Signed-off-by: Chris Wilson Cc: Micha? Winiarski Cc: Tvrtko Ursulin --- drivers/gpu/drm/i915/i915_gem_userptr.c | 142 +++++++++++++------------------- 1 file changed, 57 insertions(+), 85 deletions(-) diff --git a/drivers/gpu/drm/i915/i915_gem_userptr.c b/drivers/gpu/drm/i915/i915_gem_userptr.c index d3213fdefafc..6f55c874e65b 100644 --- a/drivers/gpu/drm/i915/i915_gem_userptr.c +++ b/drivers/gpu/drm/i915/i915_gem_userptr.c @@ -50,7 +50,6 @@ struct i915_mmu_notifier { struct mmu_notifier mn; struct rb_root objects; struct list_head linear; - unsigned long serial; bool has_linear; }; @@ -59,14 +58,16 @@ struct i915_mmu_object { struct interval_tree_node it; struct list_head link; struct drm_i915_gem_object *obj; + struct work_struct work; bool active; bool is_linear; }; -static unsigned long cancel_userptr(struct drm_i915_gem_object *obj) +static void __cancel_userptr__worker(struct work_struct *work) { + struct i915_mmu_object *mo = container_of(work, typeof(*mo), work); + struct drm_i915_gem_object *obj = mo->obj; struct drm_device *dev = obj->base.dev; - unsigned long end; mutex_lock(&dev->struct_mutex); /* Cancel any active worker and force us to re-evaluate gup */ @@ -89,46 +90,28 @@ static unsigned long cancel_userptr(struct drm_i915_gem_object *obj) dev_priv->mm.interruptible = was_interruptible; } - end = obj->userptr.ptr + obj->base.size; - drm_gem_object_unreference(&obj->base); mutex_unlock(&dev->struct_mutex); - - return end; } -static void *invalidate_range__linear(struct i915_mmu_notifier *mn, - struct mm_struct *mm, - unsigned long start, - unsigned long end) +static unsigned long cancel_userptr(struct i915_mmu_object *mo) { - struct i915_mmu_object *mo; - unsigned long serial; - -restart: - serial = mn->serial; - list_for_each_entry(mo, &mn->linear, link) { - struct drm_i915_gem_object *obj; - - if (mo->it.last < start || mo->it.start > end) - continue; - - obj = mo->obj; - - if (!mo->active || - !kref_get_unless_zero(&obj->base.refcount)) - continue; - - spin_unlock(&mn->lock); - - cancel_userptr(obj); - - spin_lock(&mn->lock); - if (serial != mn->serial) - goto restart; + unsigned long end = mo->obj->userptr.ptr + mo->obj->base.size; + + /* The mmu_object is released late when destroying the + * GEM object so it is entirely possible to gain a + * reference on an object in the process of being freed + * since our serialisation is via the spinlock and not + * the struct_mutex - and consequently use it after it + * is freed and then double free it. + */ + if (mo->active && kref_get_unless_zero(&mo->obj->base.refcount)) { + schedule_work(&mo->work); + /* only schedule one work packet to avoid the refleak */ + mo->active = false; } - return NULL; + return end; } static void i915_gem_userptr_mn_invalidate_range_start(struct mmu_notifier *_mn, @@ -136,45 +119,32 @@ static void i915_gem_userptr_mn_invalidate_range_start(struct mmu_notifier *_mn, unsigned long start, unsigned long end) { - struct i915_mmu_notifier *mn = container_of(_mn, struct i915_mmu_notifier, mn); - struct interval_tree_node *it = NULL; - unsigned long next = start; - unsigned long serial = 0; - - end--; /* interval ranges are inclusive, but invalidate range is exclusive */ - while (next < end) { - struct drm_i915_gem_object *obj = NULL; - - spin_lock(&mn->lock); - if (mn->has_linear) - it = invalidate_range__linear(mn, mm, start, end); - else if (serial == mn->serial) - it = interval_tree_iter_next(it, next, end); - else - it = interval_tree_iter_first(&mn->objects, start, end); - if (it != NULL) { - struct i915_mmu_object *mo = - container_of(it, struct i915_mmu_object, it); - - /* The mmu_object is released late when destroying the - * GEM object so it is entirely possible to gain a - * reference on an object in the process of being freed - * since our serialisation is via the spinlock and not - * the struct_mutex - and consequently use it after it - * is freed and then double free it. - */ - if (mo->active && - kref_get_unless_zero(&mo->obj->base.refcount)) - obj = mo->obj; + struct i915_mmu_notifier *mn = + container_of(_mn, struct i915_mmu_notifier, mn); + struct i915_mmu_object *mo; - serial = mn->serial; + /* interval ranges are inclusive, but invalidate range is exclusive */ + end--; + + spin_lock(&mn->lock); + if (mn->has_linear) { + list_for_each_entry(mo, &mn->linear, link) { + if (mo->it.last < start || mo->it.start > end) + continue; + + cancel_userptr(mo); } - spin_unlock(&mn->lock); - if (obj == NULL) - return; + } else { + struct interval_tree_node *it; - next = cancel_userptr(obj); + it = interval_tree_iter_first(&mn->objects, start, end); + while (it) { + mo = container_of(it, struct i915_mmu_object, it); + start = cancel_userptr(mo); + it = interval_tree_iter_next(it, start, end); + } } + spin_unlock(&mn->lock); } static const struct mmu_notifier_ops i915_gem_userptr_notifier = { @@ -194,7 +164,6 @@ i915_mmu_notifier_create(struct mm_struct *mm) spin_lock_init(&mn->lock); mn->mn.ops = &i915_gem_userptr_notifier; mn->objects = RB_ROOT; - mn->serial = 1; INIT_LIST_HEAD(&mn->linear); mn->has_linear = false; @@ -208,12 +177,6 @@ i915_mmu_notifier_create(struct mm_struct *mm) return mn; } -static void __i915_mmu_notifier_update_serial(struct i915_mmu_notifier *mn) -{ - if (++mn->serial == 0) - mn->serial = 1; -} - static int i915_mmu_notifier_add(struct drm_device *dev, struct i915_mmu_notifier *mn, @@ -260,10 +223,9 @@ i915_mmu_notifier_add(struct drm_device *dev, } else interval_tree_insert(&mo->it, &mn->objects); - if (ret == 0) { + if (ret == 0) list_add(&mo->link, &mn->linear); - __i915_mmu_notifier_update_serial(mn); - } + spin_unlock(&mn->lock); mutex_unlock(&dev->struct_mutex); @@ -291,7 +253,6 @@ i915_mmu_notifier_del(struct i915_mmu_notifier *mn, mn->has_linear = i915_mmu_notifier_has_linear(mn); else interval_tree_remove(&mo->it, &mn->objects); - __i915_mmu_notifier_update_serial(mn); spin_unlock(&mn->lock); } @@ -358,6 +319,7 @@ i915_gem_userptr_init__mmu_notifier(struct drm_i915_gem_object *obj, mo->it.start = obj->userptr.ptr; mo->it.last = mo->it.start + obj->base.size - 1; mo->obj = obj; + INIT_WORK(&mo->work, __cancel_userptr__worker); ret = i915_mmu_notifier_add(obj->base.dev, mn, mo); if (ret) { @@ -546,10 +508,12 @@ err: return ret; } -static void +static int __i915_gem_userptr_set_active(struct drm_i915_gem_object *obj, bool value) { + int ret = 0; + /* During mm_invalidate_range we need to cancel any userptr that * overlaps the range being invalidated. Doing so requires the * struct_mutex, and that risks recursion. In order to cause @@ -562,12 +526,20 @@ __i915_gem_userptr_set_active(struct drm_i915_gem_object *obj, */ #if defined(CONFIG_MMU_NOTIFIER) if (obj->userptr.mmu_object == NULL) - return; + return 0; spin_lock(&obj->userptr.mmu_object->mn->lock); - obj->userptr.mmu_object->active = value; + /* In order to serialise get_pages with an outstanding + * cancel_userptr, we must drop the struct_mutex and try again. + */ + if (!value || !work_pending(&obj->userptr.mmu_object->work)) + obj->userptr.mmu_object->active = value; + else + ret = -EAGAIN; spin_unlock(&obj->userptr.mmu_object->mn->lock); #endif + + return ret; } static void