Blob Blame History Raw
From 9e17b53aa1ca0fb47b21c32d0e5e3d29004036ac Mon Sep 17 00:00:00 2001
From: Chris Wilson <chris@chris-wilson.co.uk>
Date: Mon, 23 Mar 2020 09:28:35 +0000
Subject: drm/i915: Avoid live-lock with i915_vma_parked()
Git-commit: b0647a5e79b19753faac17e6cc0a109166d22693
Patch-mainline: v5.7-rc1
References: jsc#SLE-12680, jsc#SLE-12880, jsc#SLE-12882, jsc#SLE-12883, jsc#SLE-13496, jsc#SLE-15322

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>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200323092841.22240-2-chris@chris-wilson.co.uk
(cherry picked from commit 3447c4c55d0edc95742fdcd91c3efb050546b907)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Patrik Jakobsson <pjakobsson@suse.de>
---
 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)
-- 
2.28.0