Merge tag 'drm-for-v4.14' of git://people.freedesktop.org/~airlied/linux

Pull drm updates from Dave Airlie:
 "This is the main drm pull request for 4.14 merge window.

  I'm sending this early, as my continuing journey into fatherhood is
  occurring really soon now, I'm going to be mostly useless for the next
  couple of weeks, though I may be able to read email, I doubt I'll be
  doing much patch applications or git sending. If anything urgent pops
  up I've asked Daniel/Jani/Alex/Sean to try and direct stuff towards
  you.

  Outside drm changes:

  Some rcar-du updates that touch the V4L tree, all acks should be in
  place. It adds one export to the radix tree code for new i915 use
  case. There are some minor AGP cleanups (don't see that too often).
  Changes to the vbox driver in staging to avoid breaking compilation.

  Summary:

  core:
   - Atomic helper fixes
   - Atomic UAPI fixes
   - Add YCBCR 4:2:0 support
   - Drop set_busid hook
   - Refactor fb_helper locking
   - Remove a bunch of internal APIs
   - Add a bunch of better default handlers
   - Format modifier/blob plane property added
   - More internal header refactoring
   - Make more internal API names consistent
   - Enhanced syncobj APIs (wait/signal/reset/create signalled)

  bridge:
   - Add Synopsys Designware MIPI DSI host bridge driver

  tiny:
   - Add Pervasive Displays RePaper displays
   - Add support for LEGO MINDSTORMS EV3 LCD

  i915:
   - Lots of GEN10/CNL  support patches
   - drm syncobj support
   - Skylake+ watermark refactoring
   - GVT vGPU 48-bit ppgtt support
   - GVT performance improvements
   - NOA change ioctl
   - CCS (color compression) scanout support
   - GPU reset improvements

  amdgpu:
   - Initial hugepage support
   - BO migration logic rework
   - Vega10 improvements
   - Powerplay fixes
   - Stop reprogramming the MC
   - Fixes for ACP audio on stoney
   - SR-IOV fixes/improvements
   - Command submission overhead improvements

  amdkfd:
   - Non-dGPU upstreaming patches
   - Scratch VA ioctl
   - Image tiling modes
   - Update PM4 headers for new firmware
   - Drop all BUG_ONs.

  nouveau:
   - GP108 modesetting support.
   - Disable MSI on big endian.

  vmwgfx:
   - Add fence fd support.

  msm:
   - Runtime PM improvements

  exynos:
   - NV12MT support
   - Refactor KMS drivers

  imx-drm:
   - Lock scanout channel to improve memory bw
   - Cleanups

  etnaviv:
   - GEM object population fixes

  tegra:
   - Prep work for Tegra186 support
   - PRIME mmap support

  sunxi:
   - HDMI support improvements
   - HDMI CEC support

  omapdrm:
   - HDMI hotplug IRQ support
   - Big driver cleanup
   - OMAP5 DSI support

  rcar-du:
   - vblank fixes
   - VSP1 updates

  arcgpu:
   - Minor fixes

  stm:
   - Add STM32 DSI controller driver

  dw_hdmi:
   - Add support for Rockchip RK3399
   - HDMI CEC support

  atmel-hlcdc:
   - Add 8-bit color support

  vc4:
   - Atomic fixes
   - New ioctl to attach a label to a buffer object
   - HDMI CEC support
   - Allow userspace to dictate rendering order on submit ioctl"

* tag 'drm-for-v4.14' of git://people.freedesktop.org/~airlied/linux: (1074 commits)
  drm/syncobj: Add a signal ioctl (v3)
  drm/syncobj: Add a reset ioctl (v3)
  drm/syncobj: Add a syncobj_array_find helper
  drm/syncobj: Allow wait for submit and signal behavior (v5)
  drm/syncobj: Add a CREATE_SIGNALED flag
  drm/syncobj: Add a callback mechanism for replace_fence (v3)
  drm/syncobj: add sync obj wait interface. (v8)
  i915: Use drm_syncobj_fence_get
  drm/syncobj: Add a race-free drm_syncobj_fence_get helper (v2)
  drm/syncobj: Rename fence_get to find_fence
  drm: kirin: Add mode_valid logic to avoid mode clocks we can't generate
  drm/vmwgfx: Bump the version for fence FD support
  drm/vmwgfx: Add export fence to file descriptor support
  drm/vmwgfx: Add support for imported Fence File Descriptor
  drm/vmwgfx: Prepare to support fence fd
  drm/vmwgfx: Fix incorrect command header offset at restart
  drm/vmwgfx: Support the NOP_ERROR command
  drm/vmwgfx: Restart command buffers after errors
  drm/vmwgfx: Move irq bottom half processing to threads
  drm/vmwgfx: Don't use drm_irq_[un]install
  ...

1 files changed, 121 insertions, 80 deletions

diff --git a/drivers/dma-buf/sw_sync.c b/drivers/dma-buf/sw_sync.c
index 69c5ff36e2f9..38cc7389a6c1 100644
--- a/drivers/dma-buf/sw_sync.c
+++ b/drivers/dma-buf/sw_sync.c
@@ -96,9 +96,9 @@ static struct sync_timeline *sync_timeline_create(const char *name)
 	obj->context = dma_fence_context_alloc(1);
 	strlcpy(obj->name, name, sizeof(obj->name));
 
-	INIT_LIST_HEAD(&obj->child_list_head);
-	INIT_LIST_HEAD(&obj->active_list_head);
-	spin_lock_init(&obj->child_list_lock);
+	obj->pt_tree = RB_ROOT;
+	INIT_LIST_HEAD(&obj->pt_list);
+	spin_lock_init(&obj->lock);
 
 	sync_timeline_debug_add(obj);
 
@@ -125,68 +125,6 @@ static void sync_timeline_put(struct sync_timeline *obj)
 	kref_put(&obj->kref, sync_timeline_free);
 }
 
-/**
- * sync_timeline_signal() - signal a status change on a sync_timeline
- * @obj:	sync_timeline to signal
- * @inc:	num to increment on timeline->value
- *
- * A sync implementation should call this any time one of it's fences
- * has signaled or has an error condition.
- */
-static void sync_timeline_signal(struct sync_timeline *obj, unsigned int inc)
-{
-	unsigned long flags;
-	struct sync_pt *pt, *next;
-
-	trace_sync_timeline(obj);
-
-	spin_lock_irqsave(&obj->child_list_lock, flags);
-
-	obj->value += inc;
-
-	list_for_each_entry_safe(pt, next, &obj->active_list_head,
-				 active_list) {
-		if (dma_fence_is_signaled_locked(&pt->base))
-			list_del_init(&pt->active_list);
-	}
-
-	spin_unlock_irqrestore(&obj->child_list_lock, flags);
-}
-
-/**
- * sync_pt_create() - creates a sync pt
- * @parent:	fence's parent sync_timeline
- * @size:	size to allocate for this pt
- * @inc:	value of the fence
- *
- * Creates a new sync_pt as a child of @parent.  @size bytes will be
- * allocated allowing for implementation specific data to be kept after
- * the generic sync_timeline struct. Returns the sync_pt object or
- * NULL in case of error.
- */
-static struct sync_pt *sync_pt_create(struct sync_timeline *obj, int size,
-			     unsigned int value)
-{
-	unsigned long flags;
-	struct sync_pt *pt;
-
-	if (size < sizeof(*pt))
-		return NULL;
-
-	pt = kzalloc(size, GFP_KERNEL);
-	if (!pt)
-		return NULL;
-
-	spin_lock_irqsave(&obj->child_list_lock, flags);
-	sync_timeline_get(obj);
-	dma_fence_init(&pt->base, &timeline_fence_ops, &obj->child_list_lock,
-		       obj->context, value);
-	list_add_tail(&pt->child_list, &obj->child_list_head);
-	INIT_LIST_HEAD(&pt->active_list);
-	spin_unlock_irqrestore(&obj->child_list_lock, flags);
-	return pt;
-}
-
 static const char *timeline_fence_get_driver_name(struct dma_fence *fence)
 {
 	return "sw_sync";
@@ -203,13 +141,17 @@ static void timeline_fence_release(struct dma_fence *fence)
 {
 	struct sync_pt *pt = dma_fence_to_sync_pt(fence);
 	struct sync_timeline *parent = dma_fence_parent(fence);
-	unsigned long flags;
 
-	spin_lock_irqsave(fence->lock, flags);
-	list_del(&pt->child_list);
-	if (!list_empty(&pt->active_list))
-		list_del(&pt->active_list);
-	spin_unlock_irqrestore(fence->lock, flags);
+	if (!list_empty(&pt->link)) {
+		unsigned long flags;
+
+		spin_lock_irqsave(fence->lock, flags);
+		if (!list_empty(&pt->link)) {
+			list_del(&pt->link);
+			rb_erase(&pt->node, &parent->pt_tree);
+		}
+		spin_unlock_irqrestore(fence->lock, flags);
+	}
 
 	sync_timeline_put(parent);
 	dma_fence_free(fence);
@@ -219,18 +161,11 @@ static bool timeline_fence_signaled(struct dma_fence *fence)
 {
 	struct sync_timeline *parent = dma_fence_parent(fence);
 
-	return (fence->seqno > parent->value) ? false : true;
+	return !__dma_fence_is_later(fence->seqno, parent->value);
 }
 
 static bool timeline_fence_enable_signaling(struct dma_fence *fence)
 {
-	struct sync_pt *pt = dma_fence_to_sync_pt(fence);
-	struct sync_timeline *parent = dma_fence_parent(fence);
-
-	if (timeline_fence_signaled(fence))
-		return false;
-
-	list_add_tail(&pt->active_list, &parent->active_list_head);
 	return true;
 }
 
@@ -259,6 +194,107 @@ static const struct dma_fence_ops timeline_fence_ops = {
 	.timeline_value_str = timeline_fence_timeline_value_str,
 };
 
+/**
+ * sync_timeline_signal() - signal a status change on a sync_timeline
+ * @obj:	sync_timeline to signal
+ * @inc:	num to increment on timeline->value
+ *
+ * A sync implementation should call this any time one of it's fences
+ * has signaled or has an error condition.
+ */
+static void sync_timeline_signal(struct sync_timeline *obj, unsigned int inc)
+{
+	struct sync_pt *pt, *next;
+
+	trace_sync_timeline(obj);
+
+	spin_lock_irq(&obj->lock);
+
+	obj->value += inc;
+
+	list_for_each_entry_safe(pt, next, &obj->pt_list, link) {
+		if (!timeline_fence_signaled(&pt->base))
+			break;
+
+		list_del_init(&pt->link);
+		rb_erase(&pt->node, &obj->pt_tree);
+
+		/*
+		 * A signal callback may release the last reference to this
+		 * fence, causing it to be freed. That operation has to be
+		 * last to avoid a use after free inside this loop, and must
+		 * be after we remove the fence from the timeline in order to
+		 * prevent deadlocking on timeline->lock inside
+		 * timeline_fence_release().
+		 */
+		dma_fence_signal_locked(&pt->base);
+	}
+
+	spin_unlock_irq(&obj->lock);
+}
+
+/**
+ * sync_pt_create() - creates a sync pt
+ * @parent:	fence's parent sync_timeline
+ * @inc:	value of the fence
+ *
+ * Creates a new sync_pt as a child of @parent.  @size bytes will be
+ * allocated allowing for implementation specific data to be kept after
+ * the generic sync_timeline struct. Returns the sync_pt object or
+ * NULL in case of error.
+ */
+static struct sync_pt *sync_pt_create(struct sync_timeline *obj,
+				      unsigned int value)
+{
+	struct sync_pt *pt;
+
+	pt = kzalloc(sizeof(*pt), GFP_KERNEL);
+	if (!pt)
+		return NULL;
+
+	sync_timeline_get(obj);
+	dma_fence_init(&pt->base, &timeline_fence_ops, &obj->lock,
+		       obj->context, value);
+	INIT_LIST_HEAD(&pt->link);
+
+	spin_lock_irq(&obj->lock);
+	if (!dma_fence_is_signaled_locked(&pt->base)) {
+		struct rb_node **p = &obj->pt_tree.rb_node;
+		struct rb_node *parent = NULL;
+
+		while (*p) {
+			struct sync_pt *other;
+			int cmp;
+
+			parent = *p;
+			other = rb_entry(parent, typeof(*pt), node);
+			cmp = value - other->base.seqno;
+			if (cmp > 0) {
+				p = &parent->rb_right;
+			} else if (cmp < 0) {
+				p = &parent->rb_left;
+			} else {
+				if (dma_fence_get_rcu(&other->base)) {
+					dma_fence_put(&pt->base);
+					pt = other;
+					goto unlock;
+				}
+				p = &parent->rb_left;
+			}
+		}
+		rb_link_node(&pt->node, parent, p);
+		rb_insert_color(&pt->node, &obj->pt_tree);
+
+		parent = rb_next(&pt->node);
+		list_add_tail(&pt->link,
+			      parent ? &rb_entry(parent, typeof(*pt), node)->link : &obj->pt_list);
+	}
+unlock:
+	spin_unlock_irq(&obj->lock);
+
+	return pt;
+}
+
 /*
  * *WARNING*
  *
@@ -309,7 +345,7 @@ static long sw_sync_ioctl_create_fence(struct sync_timeline *obj,
 		goto err;
 	}
 
-	pt = sync_pt_create(obj, sizeof(*pt), data.value);
+	pt = sync_pt_create(obj, data.value);
 	if (!pt) {
 		err = -ENOMEM;
 		goto err;
@@ -345,6 +381,11 @@ static long sw_sync_ioctl_inc(struct sync_timeline *obj, unsigned long arg)
 	if (copy_from_user(&value, (void __user *)arg, sizeof(value)))
 		return -EFAULT;
 
+	while (value > INT_MAX)  {
+		sync_timeline_signal(obj, INT_MAX);
+		value -= INT_MAX;
+	}
+
 	sync_timeline_signal(obj, value);
 
 	return 0;