All of lore.kernel.org
 help / color / mirror / Atom feed
From: Chris Wilson <chris@chris-wilson.co.uk>
To: intel-gfx@lists.freedesktop.org
Subject: [Intel-gfx] [PATCH 2/8] drm/i915: Avoid live-lock with i915_vma_parked()
Date: Mon, 23 Mar 2020 09:28:35 +0000	[thread overview]
Message-ID: <20200323092841.22240-2-chris@chris-wilson.co.uk> (raw)
In-Reply-To: <20200323092841.22240-1-chris@chris-wilson.co.uk>

Abuse^W Take advantage that we know we are inside the GT wakeref and
that prevents any client execbuf from reopening the i915_vma in order to
claim all the vma to close without having to drop the spinlock to free
each one individually. By keeping the spinlock, we do not have to
restart if we run concurrently with i915_gem_free_objects -- which
causes them both to restart continually and make very very slow
progress.

Closes: https://gitlab.freedesktop.org/drm/intel/issues/1361
Fixes: 77853186e547 ("drm/i915: Claim vma while under closed_lock in i915_vma_parked()")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_vma.c | 29 ++++++++++++++---------------
 1 file changed, 14 insertions(+), 15 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 5b3efb43a8ef..08699fa069aa 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -1097,6 +1097,7 @@ void i915_vma_release(struct kref *ref)
 void i915_vma_parked(struct intel_gt *gt)
 {
 	struct i915_vma *vma, *next;
+	LIST_HEAD(closed);
 
 	spin_lock_irq(&gt->closed_lock);
 	list_for_each_entry_safe(vma, next, &gt->closed_vma, closed_link) {
@@ -1108,28 +1109,26 @@ void i915_vma_parked(struct intel_gt *gt)
 		if (!kref_get_unless_zero(&obj->base.refcount))
 			continue;
 
-		if (i915_vm_tryopen(vm)) {
-			list_del_init(&vma->closed_link);
-		} else {
+		if (!i915_vm_tryopen(vm)) {
 			i915_gem_object_put(obj);
-			obj = NULL;
+			continue;
 		}
 
-		spin_unlock_irq(&gt->closed_lock);
+		list_move(&vma->closed_link, &closed);
+	}
+	spin_unlock_irq(&gt->closed_lock);
 
-		if (obj) {
-			__i915_vma_put(vma);
-			i915_gem_object_put(obj);
-		}
+	/* As the GT is held idle, no vma can be reopened as we destroy them */
+	list_for_each_entry_safe(vma, next, &closed, closed_link) {
+		struct drm_i915_gem_object *obj = vma->obj;
+		struct i915_address_space *vm = vma->vm;
 
-		i915_vm_close(vm);
+		INIT_LIST_HEAD(&vma->closed_link);
+		__i915_vma_put(vma);
 
-		/* Restart after dropping lock */
-		spin_lock_irq(&gt->closed_lock);
-		next = list_first_entry(&gt->closed_vma,
-					typeof(*next), closed_link);
+		i915_gem_object_put(obj);
+		i915_vm_close(vm);
 	}
-	spin_unlock_irq(&gt->closed_lock);
 }
 
 static void __i915_vma_iounmap(struct i915_vma *vma)
-- 
2.20.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

  reply	other threads:[~2020-03-23  9:29 UTC|newest]

Thread overview: 22+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-03-23  9:28 [Intel-gfx] [PATCH 1/8] drm/i915/gt: Mark timeline->cacheline as destroyed after rcu grace period Chris Wilson
2020-03-23  9:28 ` Chris Wilson [this message]
2020-03-23 10:09   ` [Intel-gfx] [PATCH 2/8] drm/i915: Avoid live-lock with i915_vma_parked() Tvrtko Ursulin
2020-03-23  9:28 ` [Intel-gfx] [PATCH 3/8] drm/i915: Extend intel_wakeref to support delayed puts Chris Wilson
2020-03-23 10:13   ` Tvrtko Ursulin
2020-03-23 10:32   ` [Intel-gfx] [PATCH] " Chris Wilson
2020-03-23 12:24     ` Tvrtko Ursulin
2020-03-23  9:28 ` [Intel-gfx] [PATCH 4/8] drm/i915/gt: Delay release of engine-pm after last retirement Chris Wilson
2020-03-23 10:14   ` Tvrtko Ursulin
2020-03-23  9:28 ` [Intel-gfx] [PATCH 5/8] drm/i915: Rely on direct submission to the queue Chris Wilson
2020-03-23 10:29   ` Tvrtko Ursulin
2020-03-23  9:28 ` [Intel-gfx] [PATCH 6/8] drm/i915/execlists: Pull tasklet interrupt-bh local to direct submission Chris Wilson
2020-03-23  9:28 ` [Intel-gfx] [PATCH 7/8] drm/i915: Immediately execute the fenced work Chris Wilson
2020-03-23 10:37   ` Tvrtko Ursulin
2020-03-23 10:46     ` Chris Wilson
2020-03-24 16:13       ` Tvrtko Ursulin
2020-03-24 16:19         ` Chris Wilson
2020-03-23 11:04   ` [Intel-gfx] [PATCH] " Chris Wilson
2020-03-23  9:28 ` [Intel-gfx] [PATCH 8/8] drm/i915/gem: Avoid gem_context->mutex for simple vma lookup Chris Wilson
2020-03-23 10:38   ` Tvrtko Ursulin
2020-03-23  9:43 ` [Intel-gfx] [PATCH 1/8] drm/i915/gt: Mark timeline->cacheline as destroyed after rcu grace period Matthew Auld
2020-03-23 13:06 ` [Intel-gfx] ✗ Fi.CI.BUILD: failure for series starting with [1/8] drm/i915/gt: Mark timeline->cacheline as destroyed after rcu grace period (rev3) Patchwork

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200323092841.22240-2-chris@chris-wilson.co.uk \
    --to=chris@chris-wilson.co.uk \
    --cc=intel-gfx@lists.freedesktop.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.