Skip to contentSkip to conversation

drm: add syncobj timeline support v5

RFC PATCH In review Chunming Zhou 9 participants 12 messages Sep 14, 2018 – Sep 19, 2018 5 messages not in the archive

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 drm: add syncobj timeline support v5

This patch is for VK_KHR_timeline_semaphore extension, semaphore is called syncobj in kernel side:
This extension introduces a new type of syncobj that has an integer payload identifying a point in a timeline. Such timeline syncobjs support the
following operations:
* CPU query - A host operation that allows querying the payload of the
timeline syncobj.
* CPU wait - A host operation that allows a blocking wait for a
timeline syncobj to reach a specified value.
* Device wait - A device operation that allows waiting for a
timeline syncobj to reach a specified value.
* Device signal - A device operation that allows advancing the
timeline syncobj to a specified value.

Since it's a timeline, that means the front time point(PT) always is signaled before the late PT. a. signal PT design:
Signal PT fence N depends on PT[N-1] fence and signal opertion fence, when PT[N] fence is signaled, the timeline will increase to value of PT[N].
b. wait PT design:
Wait PT fence is signaled by reaching timeline point value, when timeline is increasing, will compare wait PTs value with new timeline value, if PT value is lower than timeline value, then wait PT will be signaled, otherwise keep in list. syncobj wait operation can wait on any point of timeline, so need a RB tree to order them. And wait PT could ahead of signal PT, we need a sumission fence to perform that.

v2:
1. remove unused DRM_SYNCOBJ_CREATE_TYPE_NORMAL. (Christian)
2. move unexposed denitions to .c file. (Daniel Vetter)
3. split up the change to drm_syncobj_find_fence() in a separate patch. (Christian)
4. split up the change to drm_syncobj_replace_fence() in a separate patch.
5. drop the submission_fence implementation and instead use wait_event() for that. (Christian)
6. WARN_ON(point != 0) for NORMAL type syncobj case. (Daniel Vetter)

v3:
1. replace normal syncobj with timeline implemenation. (Vetter and Christian)
a. normal syncobj signal op will create a signal PT to tail of signal pt list.
b. normal syncobj wait op will create a wait pt with last signal point, and this wait PT is only signaled by related signal point PT.
2. many bug fix and clean up
3. stub fence moving is moved to other patch.

v4:
1. fix RB tree loop with while(node=rb_first(...)). (Christian)
2. fix syncobj lifecycle. (Christian)
3. only enable_signaling when there is wait_pt. (Christian)
4. fix timeline path issues.
5. write a timeline test in libdrm

v5: (Christian)
1. semaphore is called syncobj in kernel side.
2. don't need 'timeline' characters in some function name.
3. keep syncobj cb

normal syncobj is tested by ./deqp-vk -n dEQP-VK*semaphore* timeline syncobj is tested by ./amdgpu_test -s 9

Cc Christian Konig <[email protected]>
Cc Dave Airlie <[email protected]>
Cc Daniel Rakos <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/drm_syncobj.c              | 294 ++++++++++++++++++---
 drivers/gpu/drm/i915/i915_gem_execbuffer.c |   4 +-
 include/drm/drm_syncobj.h                  |  62 +++--
 include/uapi/drm/drm.h                     |   1 +
 4 files changed, 292 insertions(+), 69 deletions(-)
drivers/gpu/drm/drm_syncobj.c +258 -36
index e9ce623d049e..e78d076f2703 100644
@@ -56,6 +56,9 @@
 #include "drm_internal.h"
 #include <drm/drm_syncobj.h>
 
+/* merge normal syncobj to timeline syncobj, the point interval is 1 */
+#define DRM_SYNCOBJ_NORMAL_POINT 1
+
 struct drm_syncobj_stub_fence {
 	struct dma_fence base;
 	spinlock_t lock;
@@ -82,6 +85,11 @@ static const struct dma_fence_ops drm_syncobj_stub_fence_ops = {
 	.release = drm_syncobj_stub_fence_release,
 };
 
+struct drm_syncobj_signal_pt {
+	struct dma_fence_array *base;
+	u64    value;
+	struct list_head list;
+};
 
 /**
  * drm_syncobj_find - lookup and reference a sync object.
@@ -124,7 +132,7 @@ static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
 {
 	int ret;
 
-	*fence = drm_syncobj_fence_get(syncobj);
+	ret = drm_syncobj_search_fence(syncobj, 0, 0, fence);
 	if (*fence)
 		return 1;
 
@@ -133,10 +141,10 @@ static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
 	 * have the lock, try one more time just to be sure we don't add a
 	 * callback when a fence has already been set.
 	 */
-	if (syncobj->fence) {
-		*fence = dma_fence_get(rcu_dereference_protected(syncobj->fence,
-								 lockdep_is_held(&syncobj->lock)));
-		ret = 1;
+	if (fence) {
+		drm_syncobj_search_fence(syncobj, 0, 0, fence);
+		if (*fence)
+			ret = 1;
 	} else {
 		*fence = NULL;
 		drm_syncobj_add_callback_locked(syncobj, cb, func);
@@ -164,6 +172,151 @@ void drm_syncobj_remove_callback(struct drm_syncobj *syncobj,
 	spin_unlock(&syncobj->lock);
 }
 
+static void drm_syncobj_timeline_init(struct drm_syncobj *syncobj,
+				      struct drm_syncobj_timeline *syncobj_timeline)
+{
+	spin_lock(&syncobj->lock);
+	syncobj_timeline->timeline_context = dma_fence_context_alloc(1);
+	syncobj_timeline->timeline = 0;
+	syncobj_timeline->signal_point = 0;
+	init_waitqueue_head(&syncobj_timeline->wq);
+
+	INIT_LIST_HEAD(&syncobj_timeline->signal_pt_list);
+	spin_unlock(&syncobj->lock);
+}
+
+static void drm_syncobj_timeline_fini(struct drm_syncobj *syncobj,
+				      struct drm_syncobj_timeline *syncobj_timeline)
+{
+	struct drm_syncobj_signal_pt *signal_pt = NULL, *tmp;
+
+	spin_lock(&syncobj->lock);
+	list_for_each_entry_safe(signal_pt, tmp,
+				 &syncobj_timeline->signal_pt_list, list) {
+		list_del(&signal_pt->list);
+		dma_fence_put(&signal_pt->base->base);
+		kfree(signal_pt);
+	}
+	spin_unlock(&syncobj->lock);
+}
+
+static struct dma_fence
+*drm_syncobj_find_signal_pt_for_point(struct drm_syncobj *syncobj,
+				      uint64_t point)
+{
+	struct drm_syncobj_timeline *timeline = &syncobj->syncobj_timeline;
+	struct drm_syncobj_signal_pt *signal_pt;
+
+	if ((syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) &&
+	    (point <= timeline->timeline)) {
+		struct drm_syncobj_stub_fence *fence =
+			kzalloc(sizeof(struct drm_syncobj_stub_fence),
+				GFP_KERNEL);
+
+		if (!fence)
+			return NULL;
+		spin_lock_init(&fence->lock);
+		dma_fence_init(&fence->base,
+			       &drm_syncobj_stub_fence_ops,
+			       &fence->lock,
+			       syncobj->syncobj_timeline.timeline_context,
+			       point);
+
+		dma_fence_signal(&fence->base);
+		return &fence->base;
+	}
+
+	list_for_each_entry(signal_pt, &timeline->signal_pt_list, list) {
+		if (point > signal_pt->value)
+			continue;
+		if ((syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) &&
+		    (point != signal_pt->value))
+			continue;
+		return dma_fence_get(&signal_pt->base->base);
+	}
+	return NULL;
+}
+
+static int drm_syncobj_create_signal_pt(struct drm_syncobj *syncobj,
+					struct dma_fence *fence,
+					u64 point)
+{
+	struct drm_syncobj_signal_pt *signal_pt =
+		kzalloc(sizeof(struct drm_syncobj_signal_pt), GFP_KERNEL);
+	struct drm_syncobj_signal_pt *tail_pt;
+	struct dma_fence **fences;
+	int num_fences = 0;
+	int ret = 0, i;
+
+	if (!signal_pt)
+		return -ENOMEM;
+	if (syncobj->syncobj_timeline.signal_point >= point) {
+		DRM_WARN("A later signal is ready!");
+		goto out;
+	}
+	if (!fence)
+		goto out;
+
+	fences = kmalloc_array(sizeof(void *), 2, GFP_KERNEL);
+	if (!fences)
+		goto out;
+	fences[num_fences++] = dma_fence_get(fence);
+	/* timeline syncobj must take this dependency */
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		spin_lock(&syncobj->lock);
+		if (!list_empty(&syncobj->syncobj_timeline.signal_pt_list)) {
+			tail_pt = list_last_entry(&syncobj->syncobj_timeline.signal_pt_list,
+						  struct drm_syncobj_signal_pt, list);
+			fences[num_fences++] = dma_fence_get(&tail_pt->base->base);
+		}
+		spin_unlock(&syncobj->lock);
+	}
+	signal_pt->base = dma_fence_array_create(num_fences, fences,
+						 syncobj->syncobj_timeline.timeline_context,
+						 point, false);
+	if (!signal_pt->base)
+		goto fail;
+
+	spin_lock(&syncobj->lock);
+	signal_pt->value = point;
+	INIT_LIST_HEAD(&signal_pt->list);
+	list_add_tail(&signal_pt->list, &syncobj->syncobj_timeline.signal_pt_list);
+	syncobj->syncobj_timeline.signal_point = point;
+	spin_unlock(&syncobj->lock);
+	wake_up_all(&syncobj->syncobj_timeline.wq);
+
+	return 0;
+fail:
+	for (i = 0; i < num_fences; i++)
+		dma_fence_put(fences[i]);
+	kfree(fences);
+out:
+	kfree(signal_pt);
+	return ret;
+}
+
+static void drm_syncobj_garbage_collection(struct drm_syncobj *syncobj)
+{
+	struct drm_syncobj_timeline *timeline = &syncobj->syncobj_timeline;
+	struct drm_syncobj_signal_pt *signal_pt, *tmp;
+
+	spin_lock(&syncobj->lock);
+	list_for_each_entry_safe(signal_pt, tmp,
+				 &timeline->signal_pt_list, list) {
+		if (dma_fence_is_signaled(&signal_pt->base->base)) {
+			timeline->timeline = signal_pt->value;
+			list_del(&signal_pt->list);
+			dma_fence_put(&signal_pt->base->base);
+			kfree(signal_pt);
+		} else {
+			/*signal_pt is in order in list, from small to big, so
+			 * the later must not be signal either */
+			break;
+		}
+	}
+
+	spin_unlock(&syncobj->lock);
+}
 /**
  * drm_syncobj_replace_fence - replace fence in a sync object.
  * @syncobj: Sync object to replace fence in
@@ -176,28 +329,37 @@ void drm_syncobj_replace_fence(struct drm_syncobj *syncobj,
 			       u64 point,
 			       struct dma_fence *fence)
 {
-	struct dma_fence *old_fence;
-	struct drm_syncobj_cb *cur, *tmp;
-
-	if (fence)
-		dma_fence_get(fence);
-
-	spin_lock(&syncobj->lock);
-
-	old_fence = rcu_dereference_protected(syncobj->fence,
-					      lockdep_is_held(&syncobj->lock));
-	rcu_assign_pointer(syncobj->fence, fence);
+	drm_syncobj_garbage_collection(syncobj);
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		if (fence)
+			drm_syncobj_create_signal_pt(syncobj, fence, point);
+	} else if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+		u64 pt_value;
+
+		if (!fence) {
+			drm_syncobj_timeline_fini(syncobj,
+						  &syncobj->syncobj_timeline);
+			drm_syncobj_timeline_init(syncobj,
+						  &syncobj->syncobj_timeline);
+			return;
+		}
+		pt_value = syncobj->syncobj_timeline.signal_point +
+			DRM_SYNCOBJ_NORMAL_POINT;
+		drm_syncobj_create_signal_pt(syncobj, fence, pt_value);
+	} else {
+		DRM_ERROR("the syncobj type isn't support\n");
+		return;
+	}
+	if (fence) {
+		struct drm_syncobj_cb *cur, *tmp;
 
-	if (fence != old_fence) {
+		spin_lock(&syncobj->lock);
 		list_for_each_entry_safe(cur, tmp, &syncobj->cb_list, node) {
 			list_del_init(&cur->node);
 			cur->func(syncobj, cur);
 		}
+		spin_unlock(&syncobj->lock);
 	}
-
-	spin_unlock(&syncobj->lock);
-
-	dma_fence_put(old_fence);
 }
 EXPORT_SYMBOL(drm_syncobj_replace_fence);
 
@@ -220,6 +382,25 @@ static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)
 	return 0;
 }
 
+static struct dma_fence *
+drm_syncobj_point_get(struct drm_syncobj *syncobj, u64 point, u64 flags)
+{
+	struct dma_fence *fence;
+	int ret = 0;
+
+	if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
+		ret = wait_event_timeout(syncobj->syncobj_timeline.wq,
+					point <= syncobj->syncobj_timeline.signal_point,
+					msecs_to_jiffies(10000)); /* wait 10s */
+		if (ret <= 0)
+			return NULL;
+	}
+	spin_lock(&syncobj->lock);
+	fence = drm_syncobj_find_signal_pt_for_point(syncobj, point);
+	spin_unlock(&syncobj->lock);
+	return fence;
+}
+
 /**
  * drm_syncobj_find_fence - lookup and reference the fence in a sync object
  * @file_private: drm file private pointer
@@ -234,20 +415,46 @@ static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)
  * contains a reference to the fence, which must be released by calling
  * dma_fence_put().
  */
-int drm_syncobj_find_fence(struct drm_file *file_private,
-			   u32 handle, u64 point,
-			   struct dma_fence **fence)
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+			     u64 flags, struct dma_fence **fence)
 {
-	struct drm_syncobj *syncobj = drm_syncobj_find(file_private, handle);
 	int ret = 0;
 
 	if (!syncobj)
 		return -ENOENT;
 
-	*fence = drm_syncobj_fence_get(syncobj);
+	drm_syncobj_garbage_collection(syncobj);
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+		/*NORMAL syncobj always wait on last pt */
+		u64 tail_pt_value = syncobj->syncobj_timeline.signal_point;
+
+		if (tail_pt_value == 0)
+			tail_pt_value += DRM_SYNCOBJ_NORMAL_POINT;
+		/* NORMAL syncobj doesn't care point value */
+		WARN_ON(point != 0);
+		*fence = drm_syncobj_point_get(syncobj, tail_pt_value,
+							flags);
+	} else if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		*fence = drm_syncobj_point_get(syncobj, point,
+							flags);
+	} else {
+		DRM_ERROR("Don't support this type syncobj\n");
+		*fence = NULL;
+	}
 	if (!*fence) {
 		ret = -EINVAL;
 	}
+	return ret;
+}
+EXPORT_SYMBOL(drm_syncobj_search_fence);
+int drm_syncobj_find_fence(struct drm_file *file_private,
+			   u32 handle, u64 point,
+			   struct dma_fence **fence) {
+	struct drm_syncobj *syncobj = drm_syncobj_find(file_private, handle);
+
+	int ret = drm_syncobj_search_fence(syncobj, point,
+					DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
+					fence);
 	drm_syncobj_put(syncobj);
 	return ret;
 }
@@ -264,7 +471,7 @@ void drm_syncobj_free(struct kref *kref)
 	struct drm_syncobj *syncobj = container_of(kref,
 						   struct drm_syncobj,
 						   refcount);
-	drm_syncobj_replace_fence(syncobj, 0, NULL);
+	drm_syncobj_timeline_fini(syncobj, &syncobj->syncobj_timeline);
 	kfree(syncobj);
 }
 EXPORT_SYMBOL(drm_syncobj_free);
@@ -294,6 +501,11 @@ int drm_syncobj_create(struct drm_syncobj **out_syncobj, uint32_t flags,
 	kref_init(&syncobj->refcount);
 	INIT_LIST_HEAD(&syncobj->cb_list);
 	spin_lock_init(&syncobj->lock);
+	if (flags & DRM_SYNCOBJ_CREATE_TYPE_TIMELINE)
+		syncobj->type = DRM_SYNCOBJ_TYPE_TIMELINE;
+	else
+		syncobj->type = DRM_SYNCOBJ_TYPE_NORMAL;
+	drm_syncobj_timeline_init(syncobj, &syncobj->syncobj_timeline);
 
 	if (flags & DRM_SYNCOBJ_CREATE_SIGNALED) {
 		ret = drm_syncobj_assign_null_handle(syncobj);
@@ -576,7 +788,8 @@ drm_syncobj_create_ioctl(struct drm_device *dev, void *data,
 		return -ENODEV;
 
 	/* no valid flags yet */
-	if (args->flags & ~DRM_SYNCOBJ_CREATE_SIGNALED)
+	if (args->flags & ~(DRM_SYNCOBJ_CREATE_SIGNALED |
+			    DRM_SYNCOBJ_CREATE_TYPE_TIMELINE))
 		return -EINVAL;
 
 	return drm_syncobj_create_as_handle(file_private,
@@ -669,9 +882,8 @@ static void syncobj_wait_syncobj_func(struct drm_syncobj *syncobj,
 	struct syncobj_wait_entry *wait =
 		container_of(cb, struct syncobj_wait_entry, syncobj_cb);
 
-	/* This happens inside the syncobj lock */
-	wait->fence = dma_fence_get(rcu_dereference_protected(syncobj->fence,
-							      lockdep_is_held(&syncobj->lock)));
+	drm_syncobj_search_fence(syncobj, 0, 0, &wait->fence);
+
 	wake_up_process(wait->task);
 }
 
@@ -698,7 +910,8 @@ static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs,
 	signaled_count = 0;
 	for (i = 0; i < count; ++i) {
 		entries[i].task = current;
-		entries[i].fence = drm_syncobj_fence_get(syncobjs[i]);
+		ret = drm_syncobj_search_fence(syncobjs[i], 0, 0,
+					       &entries[i].fence);
 		if (!entries[i].fence) {
 			if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
 				continue;
@@ -970,12 +1183,21 @@ drm_syncobj_reset_ioctl(struct drm_device *dev, void *data,
 	if (ret < 0)
 		return ret;
 
-	for (i = 0; i < args->count_handles; i++)
-		drm_syncobj_replace_fence(syncobjs[i], 0, NULL);
-
+	for (i = 0; i < args->count_handles; i++) {
+		if (syncobjs[i]->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+			DRM_ERROR("timeline syncobj cannot reset!\n");
+			ret = -EINVAL;
+			goto out;
+		}
+		drm_syncobj_timeline_fini(syncobjs[i],
+					  &syncobjs[i]->syncobj_timeline);
+		drm_syncobj_timeline_init(syncobjs[i],
+					  &syncobjs[i]->syncobj_timeline);
+	}
+out:
 	drm_syncobj_array_free(syncobjs, args->count_handles);
 
-	return 0;
+	return ret;
 }
 
 int
drivers/gpu/drm/i915/i915_gem_execbuffer.c +3 -1
index 0a8d2d64f380..579e91a5858b 100644
@@ -2137,7 +2137,9 @@ await_fence_array(struct i915_execbuffer *eb,
 		if (!(flags & I915_EXEC_FENCE_WAIT))
 			continue;
 
-		fence = drm_syncobj_fence_get(syncobj);
+		drm_syncobj_search_fence(syncobj, 0,
+					 DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
+					 &fence);
 		if (!fence)
 			return -EINVAL;
 
include/drm/drm_syncobj.h +30 -32

96 lines not on this page · raw message

include/uapi/drm/drm.h +1 -0
index 300f336633f2..cebdb2541eb7 100644
@@ -717,6 +717,7 @@ struct drm_prime_handle {
 struct drm_syncobj_create {
 	__u32 handle;
 #define DRM_SYNCOBJ_CREATE_SIGNALED (1 << 0)
+#define DRM_SYNCOBJ_CREATE_TYPE_TIMELINE (1 << 1)
 	__u32 flags;
 };
 
Signature · 6 lines
2.17.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#1 lore raw

Am 14.09.2018 um 12:37 schrieb Chunming Zhou:

Quoted 51 lines

This patch is for VK_KHR_timeline_semaphore extension, semaphore is called syncobj in kernel side:
This extension introduces a new type of syncobj that has an integer payload identifying a point in a timeline. Such timeline syncobjs support the
following operations:
* CPU query - A host operation that allows querying the payload of the timeline syncobj.
* CPU wait - A host operation that allows a blocking wait for a
timeline syncobj to reach a specified value.
* Device wait - A device operation that allows waiting for a
timeline syncobj to reach a specified value.
* Device signal - A device operation that allows advancing the
timeline syncobj to a specified value.

Since it's a timeline, that means the front time point(PT) always is signaled before the late PT. a. signal PT design:
Signal PT fence N depends on PT[N-1] fence and signal opertion fence, when PT[N] fence is signaled, the timeline will increase to value of PT[N].
b. wait PT design:
Wait PT fence is signaled by reaching timeline point value, when timeline is increasing, will compare wait PTs value with new timeline value, if PT value is lower than timeline value, then wait PT will be signaled, otherwise keep in list. syncobj wait operation can wait on any point of timeline, so need a RB tree to order them. And wait PT could ahead of signal PT, we need a sumission fence to perform that.

v2:
1. remove unused DRM_SYNCOBJ_CREATE_TYPE_NORMAL. (Christian)
2. move unexposed denitions to .c file. (Daniel Vetter)
3. split up the change to drm_syncobj_find_fence() in a separate patch. (Christian)
4. split up the change to drm_syncobj_replace_fence() in a separate patch.
5. drop the submission_fence implementation and instead use wait_event() for that. (Christian)
6. WARN_ON(point != 0) for NORMAL type syncobj case. (Daniel Vetter)

v3:
1. replace normal syncobj with timeline implemenation. (Vetter and Christian)
a. normal syncobj signal op will create a signal PT to tail of signal pt list.
b. normal syncobj wait op will create a wait pt with last signal point, and this wait PT is only signaled by related signal point PT.
2. many bug fix and clean up
3. stub fence moving is moved to other patch.

v4:
1. fix RB tree loop with while(node=rb_first(...)). (Christian)
2. fix syncobj lifecycle. (Christian)
3. only enable_signaling when there is wait_pt. (Christian)
4. fix timeline path issues.
5. write a timeline test in libdrm

v5: (Christian)
1. semaphore is called syncobj in kernel side.
2. don't need 'timeline' characters in some function name.
3. keep syncobj cb

Quoted 8 lines

normal syncobj is tested by ./deqp-vk -n dEQP-VK*semaphore* timeline syncobj is tested by ./amdgpu_test -s 9

Cc Christian Konig <[email protected]>
Cc Dave Airlie <[email protected]>
Cc Daniel Rakos <[email protected]>
Cc Daniel Vetter <[email protected]>

At least on first glance that looks like it should work, going to do a detailed review on Monday.

Christian.

Quoted 545 lines
---
  drivers/gpu/drm/drm_syncobj.c              | 294 ++++++++++++++++++---
  drivers/gpu/drm/i915/i915_gem_execbuffer.c |   4 +-
  include/drm/drm_syncobj.h                  |  62 +++--
  include/uapi/drm/drm.h                     |   1 +
  4 files changed, 292 insertions(+), 69 deletions(-)
drivers/gpu/drm/drm_syncobj.c +258 -36
index e9ce623d049e..e78d076f2703 100644
@@ -56,6 +56,9 @@either
  #include "drm_internal.h"
  #include <drm/drm_syncobj.h>
 
+/* merge normal syncobj to timeline syncobj, the point interval is 1 */
+#define DRM_SYNCOBJ_NORMAL_POINT 1
+
  struct drm_syncobj_stub_fence {
  	struct dma_fence base;
  	spinlock_t lock;
@@ -82,6 +85,11 @@ static const struct dma_fence_ops drm_syncobj_stub_fence_ops = {
  	.release = drm_syncobj_stub_fence_release,
  };
 
+struct drm_syncobj_signal_pt {
+	struct dma_fence_array *base;
+	u64    value;
+	struct list_head list;
+};
 
  /**
   * drm_syncobj_find - lookup and reference a sync object.
@@ -124,7 +132,7 @@ static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
  {
  	int ret;
 
-	*fence = drm_syncobj_fence_get(syncobj);
+	ret = drm_syncobj_search_fence(syncobj, 0, 0, fence);
  	if (*fence)
  		return 1;
 
@@ -133,10 +141,10 @@ static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
  	 * have the lock, try one more time just to be sure we don't add a
  	 * callback when a fence has already been set.
  	 */
-	if (syncobj->fence) {
-		*fence = dma_fence_get(rcu_dereference_protected(syncobj->fence,
-								 lockdep_is_held(&syncobj->lock)));
-		ret = 1;
+	if (fence) {
+		drm_syncobj_search_fence(syncobj, 0, 0, fence);
+		if (*fence)
+			ret = 1;
  	} else {
  		*fence = NULL;
  		drm_syncobj_add_callback_locked(syncobj, cb, func);
@@ -164,6 +172,151 @@ void drm_syncobj_remove_callback(struct drm_syncobj *syncobj,
  	spin_unlock(&syncobj->lock);
  }
 
+static void drm_syncobj_timeline_init(struct drm_syncobj *syncobj,
+				      struct drm_syncobj_timeline *syncobj_timeline)
+{
+	spin_lock(&syncobj->lock);
+	syncobj_timeline->timeline_context = dma_fence_context_alloc(1);
+	syncobj_timeline->timeline = 0;
+	syncobj_timeline->signal_point = 0;
+	init_waitqueue_head(&syncobj_timeline->wq);
+
+	INIT_LIST_HEAD(&syncobj_timeline->signal_pt_list);
+	spin_unlock(&syncobj->lock);
+}
+
+static void drm_syncobj_timeline_fini(struct drm_syncobj *syncobj,
+				      struct drm_syncobj_timeline *syncobj_timeline)
+{
+	struct drm_syncobj_signal_pt *signal_pt = NULL, *tmp;
+
+	spin_lock(&syncobj->lock);
+	list_for_each_entry_safe(signal_pt, tmp,
+				 &syncobj_timeline->signal_pt_list, list) {
+		list_del(&signal_pt->list);
+		dma_fence_put(&signal_pt->base->base);
+		kfree(signal_pt);
+	}
+	spin_unlock(&syncobj->lock);
+}
+
+static struct dma_fence
+*drm_syncobj_find_signal_pt_for_point(struct drm_syncobj *syncobj,
+				      uint64_t point)
+{
+	struct drm_syncobj_timeline *timeline = &syncobj->syncobj_timeline;
+	struct drm_syncobj_signal_pt *signal_pt;
+
+	if ((syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) &&
+	    (point <= timeline->timeline)) {
+		struct drm_syncobj_stub_fence *fence =
+			kzalloc(sizeof(struct drm_syncobj_stub_fence),
+				GFP_KERNEL);
+
+		if (!fence)
+			return NULL;
+		spin_lock_init(&fence->lock);
+		dma_fence_init(&fence->base,
+			       &drm_syncobj_stub_fence_ops,
+			       &fence->lock,
+			       syncobj->syncobj_timeline.timeline_context,
+			       point);
+
+		dma_fence_signal(&fence->base);
+		return &fence->base;
+	}
+
+	list_for_each_entry(signal_pt, &timeline->signal_pt_list, list) {
+		if (point > signal_pt->value)
+			continue;
+		if ((syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) &&
+		    (point != signal_pt->value))
+			continue;
+		return dma_fence_get(&signal_pt->base->base);
+	}
+	return NULL;
+}
+
+static int drm_syncobj_create_signal_pt(struct drm_syncobj *syncobj,
+					struct dma_fence *fence,
+					u64 point)
+{
+	struct drm_syncobj_signal_pt *signal_pt =
+		kzalloc(sizeof(struct drm_syncobj_signal_pt), GFP_KERNEL);
+	struct drm_syncobj_signal_pt *tail_pt;
+	struct dma_fence **fences;
+	int num_fences = 0;
+	int ret = 0, i;
+
+	if (!signal_pt)
+		return -ENOMEM;
+	if (syncobj->syncobj_timeline.signal_point >= point) {
+		DRM_WARN("A later signal is ready!");
+		goto out;
+	}
+	if (!fence)
+		goto out;
+
+	fences = kmalloc_array(sizeof(void *), 2, GFP_KERNEL);
+	if (!fences)
+		goto out;
+	fences[num_fences++] = dma_fence_get(fence);
+	/* timeline syncobj must take this dependency */
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		spin_lock(&syncobj->lock);
+		if (!list_empty(&syncobj->syncobj_timeline.signal_pt_list)) {
+			tail_pt = list_last_entry(&syncobj->syncobj_timeline.signal_pt_list,
+						  struct drm_syncobj_signal_pt, list);
+			fences[num_fences++] = dma_fence_get(&tail_pt->base->base);
+		}
+		spin_unlock(&syncobj->lock);
+	}
+	signal_pt->base = dma_fence_array_create(num_fences, fences,
+						 syncobj->syncobj_timeline.timeline_context,
+						 point, false);
+	if (!signal_pt->base)
+		goto fail;
+
+	spin_lock(&syncobj->lock);
+	signal_pt->value = point;
+	INIT_LIST_HEAD(&signal_pt->list);
+	list_add_tail(&signal_pt->list, &syncobj->syncobj_timeline.signal_pt_list);
+	syncobj->syncobj_timeline.signal_point = point;
+	spin_unlock(&syncobj->lock);
+	wake_up_all(&syncobj->syncobj_timeline.wq);
+
+	return 0;
+fail:
+	for (i = 0; i < num_fences; i++)
+		dma_fence_put(fences[i]);
+	kfree(fences);
+out:
+	kfree(signal_pt);
+	return ret;
+}
+
+static void drm_syncobj_garbage_collection(struct drm_syncobj *syncobj)
+{
+	struct drm_syncobj_timeline *timeline = &syncobj->syncobj_timeline;
+	struct drm_syncobj_signal_pt *signal_pt, *tmp;
+
+	spin_lock(&syncobj->lock);
+	list_for_each_entry_safe(signal_pt, tmp,
+				 &timeline->signal_pt_list, list) {
+		if (dma_fence_is_signaled(&signal_pt->base->base)) {
+			timeline->timeline = signal_pt->value;
+			list_del(&signal_pt->list);
+			dma_fence_put(&signal_pt->base->base);
+			kfree(signal_pt);
+		} else {
+			/*signal_pt is in order in list, from small to big, so
+			 * the later must not be signal either */
+			break;
+		}
+	}
+
+	spin_unlock(&syncobj->lock);
+}
  /**
   * drm_syncobj_replace_fence - replace fence in a sync object.
   * @syncobj: Sync object to replace fence in
@@ -176,28 +329,37 @@ void drm_syncobj_replace_fence(struct drm_syncobj *syncobj,
  			       u64 point,
  			       struct dma_fence *fence)
  {
-	struct dma_fence *old_fence;
-	struct drm_syncobj_cb *cur, *tmp;
-
-	if (fence)
-		dma_fence_get(fence);
-
-	spin_lock(&syncobj->lock);
-
-	old_fence = rcu_dereference_protected(syncobj->fence,
-					      lockdep_is_held(&syncobj->lock));
-	rcu_assign_pointer(syncobj->fence, fence);
+	drm_syncobj_garbage_collection(syncobj);
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		if (fence)
+			drm_syncobj_create_signal_pt(syncobj, fence, point);
+	} else if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+		u64 pt_value;
+
+		if (!fence) {
+			drm_syncobj_timeline_fini(syncobj,
+						  &syncobj->syncobj_timeline);
+			drm_syncobj_timeline_init(syncobj,
+						  &syncobj->syncobj_timeline);
+			return;
+		}
+		pt_value = syncobj->syncobj_timeline.signal_point +
+			DRM_SYNCOBJ_NORMAL_POINT;
+		drm_syncobj_create_signal_pt(syncobj, fence, pt_value);
+	} else {
+		DRM_ERROR("the syncobj type isn't support\n");
+		return;
+	}
+	if (fence) {
+		struct drm_syncobj_cb *cur, *tmp;
 
-	if (fence != old_fence) {
+		spin_lock(&syncobj->lock);
  		list_for_each_entry_safe(cur, tmp, &syncobj->cb_list, node) {
  			list_del_init(&cur->node);
  			cur->func(syncobj, cur);
  		}
+		spin_unlock(&syncobj->lock);
  	}
-
-	spin_unlock(&syncobj->lock);
-
-	dma_fence_put(old_fence);
  }
  EXPORT_SYMBOL(drm_syncobj_replace_fence);
 
@@ -220,6 +382,25 @@ static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)
  	return 0;
  }
 
+static struct dma_fence *
+drm_syncobj_point_get(struct drm_syncobj *syncobj, u64 point, u64 flags)
+{
+	struct dma_fence *fence;
+	int ret = 0;
+
+	if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
+		ret = wait_event_timeout(syncobj->syncobj_timeline.wq,
+					point <= syncobj->syncobj_timeline.signal_point,
+					msecs_to_jiffies(10000)); /* wait 10s */
+		if (ret <= 0)
+			return NULL;
+	}
+	spin_lock(&syncobj->lock);
+	fence = drm_syncobj_find_signal_pt_for_point(syncobj, point);
+	spin_unlock(&syncobj->lock);
+	return fence;
+}
+
  /**
   * drm_syncobj_find_fence - lookup and reference the fence in a sync object
   * @file_private: drm file private pointer
@@ -234,20 +415,46 @@ static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)
   * contains a reference to the fence, which must be released by calling
   * dma_fence_put().
   */
-int drm_syncobj_find_fence(struct drm_file *file_private,
-			   u32 handle, u64 point,
-			   struct dma_fence **fence)
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+			     u64 flags, struct dma_fence **fence)
  {
-	struct drm_syncobj *syncobj = drm_syncobj_find(file_private, handle);
  	int ret = 0;
 
  	if (!syncobj)
  		return -ENOENT;
 
-	*fence = drm_syncobj_fence_get(syncobj);
+	drm_syncobj_garbage_collection(syncobj);
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+		/*NORMAL syncobj always wait on last pt */
+		u64 tail_pt_value = syncobj->syncobj_timeline.signal_point;
+
+		if (tail_pt_value == 0)
+			tail_pt_value += DRM_SYNCOBJ_NORMAL_POINT;
+		/* NORMAL syncobj doesn't care point value */
+		WARN_ON(point != 0);
+		*fence = drm_syncobj_point_get(syncobj, tail_pt_value,
+							flags);
+	} else if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		*fence = drm_syncobj_point_get(syncobj, point,
+							flags);
+	} else {
+		DRM_ERROR("Don't support this type syncobj\n");
+		*fence = NULL;
+	}
  	if (!*fence) {
  		ret = -EINVAL;
  	}
+	return ret;
+}
+EXPORT_SYMBOL(drm_syncobj_search_fence);
+int drm_syncobj_find_fence(struct drm_file *file_private,
+			   u32 handle, u64 point,
+			   struct dma_fence **fence) {
+	struct drm_syncobj *syncobj = drm_syncobj_find(file_private, handle);
+
+	int ret = drm_syncobj_search_fence(syncobj, point,
+					DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
+					fence);
  	drm_syncobj_put(syncobj);
  	return ret;
  }
@@ -264,7 +471,7 @@ void drm_syncobj_free(struct kref *kref)
  	struct drm_syncobj *syncobj = container_of(kref,
  						   struct drm_syncobj,
  						   refcount);
-	drm_syncobj_replace_fence(syncobj, 0, NULL);
+	drm_syncobj_timeline_fini(syncobj, &syncobj->syncobj_timeline);
  	kfree(syncobj);
  }
  EXPORT_SYMBOL(drm_syncobj_free);
@@ -294,6 +501,11 @@ int drm_syncobj_create(struct drm_syncobj **out_syncobj, uint32_t flags,
  	kref_init(&syncobj->refcount);
  	INIT_LIST_HEAD(&syncobj->cb_list);
  	spin_lock_init(&syncobj->lock);
+	if (flags & DRM_SYNCOBJ_CREATE_TYPE_TIMELINE)
+		syncobj->type = DRM_SYNCOBJ_TYPE_TIMELINE;
+	else
+		syncobj->type = DRM_SYNCOBJ_TYPE_NORMAL;
+	drm_syncobj_timeline_init(syncobj, &syncobj->syncobj_timeline);
 
  	if (flags & DRM_SYNCOBJ_CREATE_SIGNALED) {
  		ret = drm_syncobj_assign_null_handle(syncobj);
@@ -576,7 +788,8 @@ drm_syncobj_create_ioctl(struct drm_device *dev, void *data,
  		return -ENODEV;
 
  	/* no valid flags yet */
-	if (args->flags & ~DRM_SYNCOBJ_CREATE_SIGNALED)
+	if (args->flags & ~(DRM_SYNCOBJ_CREATE_SIGNALED |
+			    DRM_SYNCOBJ_CREATE_TYPE_TIMELINE))
  		return -EINVAL;
 
  	return drm_syncobj_create_as_handle(file_private,
@@ -669,9 +882,8 @@ static void syncobj_wait_syncobj_func(struct drm_syncobj *syncobj,
  	struct syncobj_wait_entry *wait =
  		container_of(cb, struct syncobj_wait_entry, syncobj_cb);
 
-	/* This happens inside the syncobj lock */
-	wait->fence = dma_fence_get(rcu_dereference_protected(syncobj->fence,
-							      lockdep_is_held(&syncobj->lock)));
+	drm_syncobj_search_fence(syncobj, 0, 0, &wait->fence);
+
  	wake_up_process(wait->task);
  }
 
@@ -698,7 +910,8 @@ static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs,
  	signaled_count = 0;
  	for (i = 0; i < count; ++i) {
  		entries[i].task = current;
-		entries[i].fence = drm_syncobj_fence_get(syncobjs[i]);
+		ret = drm_syncobj_search_fence(syncobjs[i], 0, 0,
+					       &entries[i].fence);
  		if (!entries[i].fence) {
  			if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
  				continue;
@@ -970,12 +1183,21 @@ drm_syncobj_reset_ioctl(struct drm_device *dev, void *data,
  	if (ret < 0)
  		return ret;
 
-	for (i = 0; i < args->count_handles; i++)
-		drm_syncobj_replace_fence(syncobjs[i], 0, NULL);
-
+	for (i = 0; i < args->count_handles; i++) {
+		if (syncobjs[i]->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+			DRM_ERROR("timeline syncobj cannot reset!\n");
+			ret = -EINVAL;
+			goto out;
+		}
+		drm_syncobj_timeline_fini(syncobjs[i],
+					  &syncobjs[i]->syncobj_timeline);
+		drm_syncobj_timeline_init(syncobjs[i],
+					  &syncobjs[i]->syncobj_timeline);
+	}
+out:
  	drm_syncobj_array_free(syncobjs, args->count_handles);
 
-	return 0;
+	return ret;
  }
 
  int
drivers/gpu/drm/i915/i915_gem_execbuffer.c +3 -1
index 0a8d2d64f380..579e91a5858b 100644
@@ -2137,7 +2137,9 @@ await_fence_array(struct i915_execbuffer *eb,
  		if (!(flags & I915_EXEC_FENCE_WAIT))
  			continue;
 
-		fence = drm_syncobj_fence_get(syncobj);
+		drm_syncobj_search_fence(syncobj, 0,
+					 DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
+					 &fence);
  		if (!fence)
  			return -EINVAL;
 
include/drm/drm_syncobj.h +30 -32
index 425432b85a87..9535521e6623 100644
@@ -30,6 +30,25 @@
 
  struct drm_syncobj_cb;
 
+enum drm_syncobj_type {
+	DRM_SYNCOBJ_TYPE_NORMAL,
+	DRM_SYNCOBJ_TYPE_TIMELINE
+};
+
+struct drm_syncobj_timeline {
+	wait_queue_head_t	wq;
+	u64 timeline_context;
+	/**
+	 * @timeline: syncobj timeline
+	 */
+	u64 timeline;
+	u64 signal_point;
+
+
+	struct rb_root wait_pt_tree;
+	struct list_head signal_pt_list;
+};
+
  /**
   * struct drm_syncobj - sync object.
   *
@@ -41,19 +60,19 @@ struct drm_syncobj {
  	 */
  	struct kref refcount;
  	/**
-	 * @fence:
-	 * NULL or a pointer to the fence bound to this object.
-	 *
-	 * This field should not be used directly. Use drm_syncobj_fence_get()
-	 * and drm_syncobj_replace_fence() instead.
+	 * @type: indicate syncobj type
  	 */
-	struct dma_fence __rcu *fence;
+	enum drm_syncobj_type type;
  	/**
-	 * @cb_list: List of callbacks to call when the &fence gets replaced.
+	 * @syncobj_timeline: timeline
  	 */
+	struct drm_syncobj_timeline syncobj_timeline;
+	/**
+         * @cb_list: List of callbacks to call when the &fence gets replaced.
+         */
  	struct list_head cb_list;
  	/**
-	 * @lock: Protects &cb_list and write-locks &fence.
+	 * @lock: Protects syncobj list and write-locks &fence.
  	 */
  	spinlock_t lock;
  	/**
@@ -68,7 +87,7 @@ typedef void (*drm_syncobj_func_t)(struct drm_syncobj *syncobj,
  /**
   * struct drm_syncobj_cb - callback for drm_syncobj_add_callback
   * @node: used by drm_syncob_add_callback to append this struct to
- *	  &drm_syncobj.cb_list
+ *       &drm_syncobj.cb_list
   * @func: drm_syncobj_func_t to call
   *
   * This struct will be initialized by drm_syncobj_add_callback, additional
@@ -106,29 +125,6 @@ drm_syncobj_put(struct drm_syncobj *obj)
  	kref_put(&obj->refcount, drm_syncobj_free);
  }
 
-/**
- * drm_syncobj_fence_get - get a reference to a fence in a sync object
- * @syncobj: sync object.
- *
- * This acquires additional reference to &drm_syncobj.fence contained in @obj,
- * if not NULL. It is illegal to call this without already holding a reference.
- * No locks required.
- *
- * Returns:
- * Either the fence of @obj or NULL if there's none.
- */
-static inline struct dma_fence *
-drm_syncobj_fence_get(struct drm_syncobj *syncobj)
-{
-	struct dma_fence *fence;
-
-	rcu_read_lock();
-	fence = dma_fence_get_rcu_safe(&syncobj->fence);
-	rcu_read_unlock();
-
-	return fence;
-}
-
  struct drm_syncobj *drm_syncobj_find(struct drm_file *file_private,
  				     u32 handle);
  void drm_syncobj_replace_fence(struct drm_syncobj *syncobj, u64 point,
@@ -142,5 +138,7 @@ int drm_syncobj_create(struct drm_syncobj **out_syncobj, uint32_t flags,
  int drm_syncobj_get_handle(struct drm_file *file_private,
  			   struct drm_syncobj *syncobj, u32 *handle);
  int drm_syncobj_get_fd(struct drm_syncobj *syncobj, int *p_fd);
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+			     u64 flags, struct dma_fence **fence);
 
  #endif
include/uapi/drm/drm.h +0 -0
index 300f336633f2..cebdb2541eb7 100644

include/uapi/drm/drm.h struct drm_prime_handle { · line 723

Quoted 7 lines
include/uapi/drm/drm.h +1 -0
@@ -717,6 +717,7 @@ struct drm_prime_handle {
  struct drm_syncobj_create {
  	__u32 handle;
  #define DRM_SYNCOBJ_CREATE_SIGNALED (1 << 0)
+#define DRM_SYNCOBJ_CREATE_TYPE_TIMELINE (1 << 1)
  	__u32 flags;
  };
 

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

A message that is not in the archive (sent off-list or not yet received): 9a07a371-8de1-2ebf-66e7-23b93c1bc1c3-5C7GfCeVMHo@public.gmane.org

#3 lore raw
Quoted 55 lines · Christian König

Am 14.09.2018 um 12:37 schrieb Chunming Zhou:

Quoted 54 lines

This patch is for VK_KHR_timeline_semaphore extension, semaphore is called syncobj in kernel side:
This extension introduces a new type of syncobj that has an integer payload identifying a point in a timeline. Such timeline syncobjs support the
following operations:
* CPU query - A host operation that allows querying the payload of the timeline syncobj.
* CPU wait - A host operation that allows a blocking wait for a
timeline syncobj to reach a specified value.
* Device wait - A device operation that allows waiting for a
timeline syncobj to reach a specified value.
* Device signal - A device operation that allows advancing the
timeline syncobj to a specified value.

Since it's a timeline, that means the front time point(PT) always is signaled before the late PT. a. signal PT design:
Signal PT fence N depends on PT[N-1] fence and signal opertion fence, when PT[N] fence is signaled, the timeline will increase to value of PT[N].
b. wait PT design:
Wait PT fence is signaled by reaching timeline point value, when timeline is increasing, will compare wait PTs value with new timeline value, if PT value is lower than timeline value, then wait PT will be signaled, otherwise keep in list. syncobj wait operation can wait on any point of timeline, so need a RB tree to order them. And wait PT could ahead of signal PT, we need a sumission fence to perform that.

v2:
1. remove unused DRM_SYNCOBJ_CREATE_TYPE_NORMAL. (Christian)
2. move unexposed denitions to .c file. (Daniel Vetter)
3. split up the change to drm_syncobj_find_fence() in a separate patch. (Christian)
4. split up the change to drm_syncobj_replace_fence() in a separate patch.
5. drop the submission_fence implementation and instead use wait_event() for that. (Christian)
6. WARN_ON(point != 0) for NORMAL type syncobj case. (Daniel Vetter)

v3:
1. replace normal syncobj with timeline implemenation. (Vetter and Christian)
a. normal syncobj signal op will create a signal PT to tail of signal pt list.
b. normal syncobj wait op will create a wait pt with last signal point, and this wait PT is only signaled by related signal point PT.
2. many bug fix and clean up
3. stub fence moving is moved to other patch.

v4:
1. fix RB tree loop with while(node=rb_first(...)). (Christian)
2. fix syncobj lifecycle. (Christian)
3. only enable_signaling when there is wait_pt. (Christian)
4. fix timeline path issues.
5. write a timeline test in libdrm

v5: (Christian)
1. semaphore is called syncobj in kernel side.
2. don't need 'timeline' characters in some function name.
3. keep syncobj cb

normal syncobj is tested by ./deqp-vk -n dEQP-VK*semaphore* timeline syncobj is tested by ./amdgpu_test -s 9

Quoted 8 lines
Quoted 5 lines
Cc Christian Konig <[email protected]>
Cc Dave Airlie <[email protected]>
Cc Daniel Rakos <[email protected]>
Cc Daniel Vetter <[email protected]>

At least on first glance that looks like it should work, going to do a detailed review on Monday.

Just for my understanding, it's all condensed down to 1 patch now? I kinda didn't follow the detailed discussion last few days at all :-/

Also, is there a testcase, igt highly preferred (because then we'll run it in our intel-gfx CI, and a bunch of people outside of intel have already discovered that and are using it).

Thanks, Daniel

Quoted 530 lines

Christian.

Quoted 526 lines
---
  drivers/gpu/drm/drm_syncobj.c              | 294 ++++++++++++++++++---
  drivers/gpu/drm/i915/i915_gem_execbuffer.c |   4 +-
  include/drm/drm_syncobj.h                  |  62 +++--
  include/uapi/drm/drm.h                     |   1 +
  4 files changed, 292 insertions(+), 69 deletions(-)
drivers/gpu/drm/drm_syncobj.c +258 -36
index e9ce623d049e..e78d076f2703 100644
@@ -56,6 +56,9 @@either
  #include "drm_internal.h"
  #include <drm/drm_syncobj.h>
+/* merge normal syncobj to timeline syncobj, the point interval is 1 */
+#define DRM_SYNCOBJ_NORMAL_POINT 1
+
  struct drm_syncobj_stub_fence {
  	struct dma_fence base;
  	spinlock_t lock;
@@ -82,6 +85,11 @@ static const struct dma_fence_ops drm_syncobj_stub_fence_ops = {
  	.release = drm_syncobj_stub_fence_release,
  };
+struct drm_syncobj_signal_pt {
+	struct dma_fence_array *base;
+	u64    value;
+	struct list_head list;
+};
  /**
   * drm_syncobj_find - lookup and reference a sync object.
@@ -124,7 +132,7 @@ static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
  {
  	int ret;
-	*fence = drm_syncobj_fence_get(syncobj);
+	ret = drm_syncobj_search_fence(syncobj, 0, 0, fence);
  	if (*fence)
  		return 1;
@@ -133,10 +141,10 @@ static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
  	 * have the lock, try one more time just to be sure we don't add a
  	 * callback when a fence has already been set.
  	 */
-	if (syncobj->fence) {
-		*fence = dma_fence_get(rcu_dereference_protected(syncobj->fence,
-								 lockdep_is_held(&syncobj->lock)));
-		ret = 1;
+	if (fence) {
+		drm_syncobj_search_fence(syncobj, 0, 0, fence);
+		if (*fence)
+			ret = 1;
  	} else {
  		*fence = NULL;
  		drm_syncobj_add_callback_locked(syncobj, cb, func);
@@ -164,6 +172,151 @@ void drm_syncobj_remove_callback(struct drm_syncobj *syncobj,
  	spin_unlock(&syncobj->lock);
  }
+static void drm_syncobj_timeline_init(struct drm_syncobj *syncobj,
+				      struct drm_syncobj_timeline *syncobj_timeline)
+{
+	spin_lock(&syncobj->lock);
+	syncobj_timeline->timeline_context = dma_fence_context_alloc(1);
+	syncobj_timeline->timeline = 0;
+	syncobj_timeline->signal_point = 0;
+	init_waitqueue_head(&syncobj_timeline->wq);
+
+	INIT_LIST_HEAD(&syncobj_timeline->signal_pt_list);
+	spin_unlock(&syncobj->lock);
+}
+
+static void drm_syncobj_timeline_fini(struct drm_syncobj *syncobj,
+				      struct drm_syncobj_timeline *syncobj_timeline)
+{
+	struct drm_syncobj_signal_pt *signal_pt = NULL, *tmp;
+
+	spin_lock(&syncobj->lock);
+	list_for_each_entry_safe(signal_pt, tmp,
+				 &syncobj_timeline->signal_pt_list, list) {
+		list_del(&signal_pt->list);
+		dma_fence_put(&signal_pt->base->base);
+		kfree(signal_pt);
+	}
+	spin_unlock(&syncobj->lock);
+}
+
+static struct dma_fence
+*drm_syncobj_find_signal_pt_for_point(struct drm_syncobj *syncobj,
+				      uint64_t point)
+{
+	struct drm_syncobj_timeline *timeline = &syncobj->syncobj_timeline;
+	struct drm_syncobj_signal_pt *signal_pt;
+
+	if ((syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) &&
+	    (point <= timeline->timeline)) {
+		struct drm_syncobj_stub_fence *fence =
+			kzalloc(sizeof(struct drm_syncobj_stub_fence),
+				GFP_KERNEL);
+
+		if (!fence)
+			return NULL;
+		spin_lock_init(&fence->lock);
+		dma_fence_init(&fence->base,
+			       &drm_syncobj_stub_fence_ops,
+			       &fence->lock,
+			       syncobj->syncobj_timeline.timeline_context,
+			       point);
+
+		dma_fence_signal(&fence->base);
+		return &fence->base;
+	}
+
+	list_for_each_entry(signal_pt, &timeline->signal_pt_list, list) {
+		if (point > signal_pt->value)
+			continue;
+		if ((syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) &&
+		    (point != signal_pt->value))
+			continue;
+		return dma_fence_get(&signal_pt->base->base);
+	}
+	return NULL;
+}
+
+static int drm_syncobj_create_signal_pt(struct drm_syncobj *syncobj,
+					struct dma_fence *fence,
+					u64 point)
+{
+	struct drm_syncobj_signal_pt *signal_pt =
+		kzalloc(sizeof(struct drm_syncobj_signal_pt), GFP_KERNEL);
+	struct drm_syncobj_signal_pt *tail_pt;
+	struct dma_fence **fences;
+	int num_fences = 0;
+	int ret = 0, i;
+
+	if (!signal_pt)
+		return -ENOMEM;
+	if (syncobj->syncobj_timeline.signal_point >= point) {
+		DRM_WARN("A later signal is ready!");
+		goto out;
+	}
+	if (!fence)
+		goto out;
+
+	fences = kmalloc_array(sizeof(void *), 2, GFP_KERNEL);
+	if (!fences)
+		goto out;
+	fences[num_fences++] = dma_fence_get(fence);
+	/* timeline syncobj must take this dependency */
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		spin_lock(&syncobj->lock);
+		if (!list_empty(&syncobj->syncobj_timeline.signal_pt_list)) {
+			tail_pt = list_last_entry(&syncobj->syncobj_timeline.signal_pt_list,
+						  struct drm_syncobj_signal_pt, list);
+			fences[num_fences++] = dma_fence_get(&tail_pt->base->base);
+		}
+		spin_unlock(&syncobj->lock);
+	}
+	signal_pt->base = dma_fence_array_create(num_fences, fences,
+						 syncobj->syncobj_timeline.timeline_context,
+						 point, false);
+	if (!signal_pt->base)
+		goto fail;
+
+	spin_lock(&syncobj->lock);
+	signal_pt->value = point;
+	INIT_LIST_HEAD(&signal_pt->list);
+	list_add_tail(&signal_pt->list, &syncobj->syncobj_timeline.signal_pt_list);
+	syncobj->syncobj_timeline.signal_point = point;
+	spin_unlock(&syncobj->lock);
+	wake_up_all(&syncobj->syncobj_timeline.wq);
+
+	return 0;
+fail:
+	for (i = 0; i < num_fences; i++)
+		dma_fence_put(fences[i]);
+	kfree(fences);
+out:
+	kfree(signal_pt);
+	return ret;
+}
+
+static void drm_syncobj_garbage_collection(struct drm_syncobj *syncobj)
+{
+	struct drm_syncobj_timeline *timeline = &syncobj->syncobj_timeline;
+	struct drm_syncobj_signal_pt *signal_pt, *tmp;
+
+	spin_lock(&syncobj->lock);
+	list_for_each_entry_safe(signal_pt, tmp,
+				 &timeline->signal_pt_list, list) {
+		if (dma_fence_is_signaled(&signal_pt->base->base)) {
+			timeline->timeline = signal_pt->value;
+			list_del(&signal_pt->list);
+			dma_fence_put(&signal_pt->base->base);
+			kfree(signal_pt);
+		} else {
+			/*signal_pt is in order in list, from small to big, so
+			 * the later must not be signal either */
+			break;
+		}
+	}
+
+	spin_unlock(&syncobj->lock);
+}
  /**
   * drm_syncobj_replace_fence - replace fence in a sync object.
   * @syncobj: Sync object to replace fence in
@@ -176,28 +329,37 @@ void drm_syncobj_replace_fence(struct drm_syncobj *syncobj,
  			       u64 point,
  			       struct dma_fence *fence)
  {
-	struct dma_fence *old_fence;
-	struct drm_syncobj_cb *cur, *tmp;
-
-	if (fence)
-		dma_fence_get(fence);
-
-	spin_lock(&syncobj->lock);
-
-	old_fence = rcu_dereference_protected(syncobj->fence,
-					      lockdep_is_held(&syncobj->lock));
-	rcu_assign_pointer(syncobj->fence, fence);
+	drm_syncobj_garbage_collection(syncobj);
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		if (fence)
+			drm_syncobj_create_signal_pt(syncobj, fence, point);
+	} else if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+		u64 pt_value;
+
+		if (!fence) {
+			drm_syncobj_timeline_fini(syncobj,
+						  &syncobj->syncobj_timeline);
+			drm_syncobj_timeline_init(syncobj,
+						  &syncobj->syncobj_timeline);
+			return;
+		}
+		pt_value = syncobj->syncobj_timeline.signal_point +
+			DRM_SYNCOBJ_NORMAL_POINT;
+		drm_syncobj_create_signal_pt(syncobj, fence, pt_value);
+	} else {
+		DRM_ERROR("the syncobj type isn't support\n");
+		return;
+	}
+	if (fence) {
+		struct drm_syncobj_cb *cur, *tmp;
-	if (fence != old_fence) {
+		spin_lock(&syncobj->lock);
  		list_for_each_entry_safe(cur, tmp, &syncobj->cb_list, node) {
  			list_del_init(&cur->node);
  			cur->func(syncobj, cur);
  		}
+		spin_unlock(&syncobj->lock);
  	}
-
-	spin_unlock(&syncobj->lock);
-
-	dma_fence_put(old_fence);
  }
  EXPORT_SYMBOL(drm_syncobj_replace_fence);
@@ -220,6 +382,25 @@ static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)
  	return 0;
  }
+static struct dma_fence *
+drm_syncobj_point_get(struct drm_syncobj *syncobj, u64 point, u64 flags)
+{
+	struct dma_fence *fence;
+	int ret = 0;
+
+	if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
+		ret = wait_event_timeout(syncobj->syncobj_timeline.wq,
+					point <= syncobj->syncobj_timeline.signal_point,
+					msecs_to_jiffies(10000)); /* wait 10s */
+		if (ret <= 0)
+			return NULL;
+	}
+	spin_lock(&syncobj->lock);
+	fence = drm_syncobj_find_signal_pt_for_point(syncobj, point);
+	spin_unlock(&syncobj->lock);
+	return fence;
+}
+
  /**
   * drm_syncobj_find_fence - lookup and reference the fence in a sync object
   * @file_private: drm file private pointer
@@ -234,20 +415,46 @@ static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)
   * contains a reference to the fence, which must be released by calling
   * dma_fence_put().
   */
-int drm_syncobj_find_fence(struct drm_file *file_private,
-			   u32 handle, u64 point,
-			   struct dma_fence **fence)
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+			     u64 flags, struct dma_fence **fence)
  {
-	struct drm_syncobj *syncobj = drm_syncobj_find(file_private, handle);
  	int ret = 0;
  	if (!syncobj)
  		return -ENOENT;
-	*fence = drm_syncobj_fence_get(syncobj);
+	drm_syncobj_garbage_collection(syncobj);
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+		/*NORMAL syncobj always wait on last pt */
+		u64 tail_pt_value = syncobj->syncobj_timeline.signal_point;
+
+		if (tail_pt_value == 0)
+			tail_pt_value += DRM_SYNCOBJ_NORMAL_POINT;
+		/* NORMAL syncobj doesn't care point value */
+		WARN_ON(point != 0);
+		*fence = drm_syncobj_point_get(syncobj, tail_pt_value,
+							flags);
+	} else if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		*fence = drm_syncobj_point_get(syncobj, point,
+							flags);
+	} else {
+		DRM_ERROR("Don't support this type syncobj\n");
+		*fence = NULL;
+	}
  	if (!*fence) {
  		ret = -EINVAL;
  	}
+	return ret;
+}
+EXPORT_SYMBOL(drm_syncobj_search_fence);
+int drm_syncobj_find_fence(struct drm_file *file_private,
+			   u32 handle, u64 point,
+			   struct dma_fence **fence) {
+	struct drm_syncobj *syncobj = drm_syncobj_find(file_private, handle);
+
+	int ret = drm_syncobj_search_fence(syncobj, point,
+					DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
+					fence);
  	drm_syncobj_put(syncobj);
  	return ret;
  }
@@ -264,7 +471,7 @@ void drm_syncobj_free(struct kref *kref)
  	struct drm_syncobj *syncobj = container_of(kref,
  						   struct drm_syncobj,
  						   refcount);
-	drm_syncobj_replace_fence(syncobj, 0, NULL);
+	drm_syncobj_timeline_fini(syncobj, &syncobj->syncobj_timeline);
  	kfree(syncobj);
  }
  EXPORT_SYMBOL(drm_syncobj_free);
@@ -294,6 +501,11 @@ int drm_syncobj_create(struct drm_syncobj **out_syncobj, uint32_t flags,
  	kref_init(&syncobj->refcount);
  	INIT_LIST_HEAD(&syncobj->cb_list);
  	spin_lock_init(&syncobj->lock);
+	if (flags & DRM_SYNCOBJ_CREATE_TYPE_TIMELINE)
+		syncobj->type = DRM_SYNCOBJ_TYPE_TIMELINE;
+	else
+		syncobj->type = DRM_SYNCOBJ_TYPE_NORMAL;
+	drm_syncobj_timeline_init(syncobj, &syncobj->syncobj_timeline);
  	if (flags & DRM_SYNCOBJ_CREATE_SIGNALED) {
  		ret = drm_syncobj_assign_null_handle(syncobj);
@@ -576,7 +788,8 @@ drm_syncobj_create_ioctl(struct drm_device *dev, void *data,
  		return -ENODEV;
  	/* no valid flags yet */
-	if (args->flags & ~DRM_SYNCOBJ_CREATE_SIGNALED)
+	if (args->flags & ~(DRM_SYNCOBJ_CREATE_SIGNALED |
+			    DRM_SYNCOBJ_CREATE_TYPE_TIMELINE))
  		return -EINVAL;
  	return drm_syncobj_create_as_handle(file_private,
@@ -669,9 +882,8 @@ static void syncobj_wait_syncobj_func(struct drm_syncobj *syncobj,
  	struct syncobj_wait_entry *wait =
  		container_of(cb, struct syncobj_wait_entry, syncobj_cb);
-	/* This happens inside the syncobj lock */
-	wait->fence = dma_fence_get(rcu_dereference_protected(syncobj->fence,
-							      lockdep_is_held(&syncobj->lock)));
+	drm_syncobj_search_fence(syncobj, 0, 0, &wait->fence);
+
  	wake_up_process(wait->task);
  }
@@ -698,7 +910,8 @@ static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs,
  	signaled_count = 0;
  	for (i = 0; i < count; ++i) {
  		entries[i].task = current;
-		entries[i].fence = drm_syncobj_fence_get(syncobjs[i]);
+		ret = drm_syncobj_search_fence(syncobjs[i], 0, 0,
+					       &entries[i].fence);
  		if (!entries[i].fence) {
  			if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
  				continue;
@@ -970,12 +1183,21 @@ drm_syncobj_reset_ioctl(struct drm_device *dev, void *data,
  	if (ret < 0)
  		return ret;
-	for (i = 0; i < args->count_handles; i++)
-		drm_syncobj_replace_fence(syncobjs[i], 0, NULL);
-
+	for (i = 0; i < args->count_handles; i++) {
+		if (syncobjs[i]->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+			DRM_ERROR("timeline syncobj cannot reset!\n");
+			ret = -EINVAL;
+			goto out;
+		}
+		drm_syncobj_timeline_fini(syncobjs[i],
+					  &syncobjs[i]->syncobj_timeline);
+		drm_syncobj_timeline_init(syncobjs[i],
+					  &syncobjs[i]->syncobj_timeline);
+	}
+out:
  	drm_syncobj_array_free(syncobjs, args->count_handles);
-	return 0;
+	return ret;
  }
  int
drivers/gpu/drm/i915/i915_gem_execbuffer.c +3 -1
index 0a8d2d64f380..579e91a5858b 100644
@@ -2137,7 +2137,9 @@ await_fence_array(struct i915_execbuffer *eb,
  		if (!(flags & I915_EXEC_FENCE_WAIT))
  			continue;
-		fence = drm_syncobj_fence_get(syncobj);
+		drm_syncobj_search_fence(syncobj, 0,
+					 DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
+					 &fence);
  		if (!fence)
  			return -EINVAL;
include/drm/drm_syncobj.h +30 -32
index 425432b85a87..9535521e6623 100644
@@ -30,6 +30,25 @@
  struct drm_syncobj_cb;
+enum drm_syncobj_type {
+	DRM_SYNCOBJ_TYPE_NORMAL,
+	DRM_SYNCOBJ_TYPE_TIMELINE
+};
+
+struct drm_syncobj_timeline {
+	wait_queue_head_t	wq;
+	u64 timeline_context;
+	/**
+	 * @timeline: syncobj timeline
+	 */
+	u64 timeline;
+	u64 signal_point;
+
+
+	struct rb_root wait_pt_tree;
+	struct list_head signal_pt_list;
+};
+
  /**
   * struct drm_syncobj - sync object.
   *
@@ -41,19 +60,19 @@ struct drm_syncobj {
  	 */
  	struct kref refcount;
  	/**
-	 * @fence:
-	 * NULL or a pointer to the fence bound to this object.
-	 *
-	 * This field should not be used directly. Use drm_syncobj_fence_get()
-	 * and drm_syncobj_replace_fence() instead.
+	 * @type: indicate syncobj type
  	 */
-	struct dma_fence __rcu *fence;
+	enum drm_syncobj_type type;
  	/**
-	 * @cb_list: List of callbacks to call when the &fence gets replaced.
+	 * @syncobj_timeline: timeline
  	 */
+	struct drm_syncobj_timeline syncobj_timeline;
+	/**
+         * @cb_list: List of callbacks to call when the &fence gets replaced.
+         */
  	struct list_head cb_list;
  	/**
-	 * @lock: Protects &cb_list and write-locks &fence.
+	 * @lock: Protects syncobj list and write-locks &fence.
  	 */
  	spinlock_t lock;
  	/**
@@ -68,7 +87,7 @@ typedef void (*drm_syncobj_func_t)(struct drm_syncobj *syncobj,
  /**
   * struct drm_syncobj_cb - callback for drm_syncobj_add_callback
   * @node: used by drm_syncob_add_callback to append this struct to
- *	  &drm_syncobj.cb_list
+ *       &drm_syncobj.cb_list
   * @func: drm_syncobj_func_t to call
   *
   * This struct will be initialized by drm_syncobj_add_callback, additional
@@ -106,29 +125,6 @@ drm_syncobj_put(struct drm_syncobj *obj)
  	kref_put(&obj->refcount, drm_syncobj_free);
  }
-/**
- * drm_syncobj_fence_get - get a reference to a fence in a sync object
- * @syncobj: sync object.
- *
- * This acquires additional reference to &drm_syncobj.fence contained in @obj,
- * if not NULL. It is illegal to call this without already holding a reference.
- * No locks required.
- *
- * Returns:
- * Either the fence of @obj or NULL if there's none.
- */
-static inline struct dma_fence *
-drm_syncobj_fence_get(struct drm_syncobj *syncobj)
-{
-	struct dma_fence *fence;
-
-	rcu_read_lock();
-	fence = dma_fence_get_rcu_safe(&syncobj->fence);
-	rcu_read_unlock();
-
-	return fence;
-}
-
  struct drm_syncobj *drm_syncobj_find(struct drm_file *file_private,
  				     u32 handle);
  void drm_syncobj_replace_fence(struct drm_syncobj *syncobj, u64 point,
@@ -142,5 +138,7 @@ int drm_syncobj_create(struct drm_syncobj **out_syncobj, uint32_t flags,
  int drm_syncobj_get_handle(struct drm_file *file_private,
  			   struct drm_syncobj *syncobj, u32 *handle);
  int drm_syncobj_get_fd(struct drm_syncobj *syncobj, int *p_fd);
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+			     u64 flags, struct dma_fence **fence);
  #endif
include/uapi/drm/drm.h +1 -0
index 300f336633f2..cebdb2541eb7 100644
@@ -717,6 +717,7 @@ struct drm_prime_handle {
  struct drm_syncobj_create {
  	__u32 handle;
  #define DRM_SYNCOBJ_CREATE_SIGNALED (1 << 0)
+#define DRM_SYNCOBJ_CREATE_TYPE_TIMELINE (1 << 1)
  	__u32 flags;
  };
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 20180914161059.GR11082-dv86pmgwkMBes7Z6vYuT8azUEOm+Xw19@public.gmane.org

#5 lore raw

Am 14.09.2018 um 18:10 schrieb Daniel Vetter:

Quoted 61 lines · Christian König
Quoted 60 lines

Am 14.09.2018 um 12:37 schrieb Chunming Zhou:

Quoted 59 lines

This patch is for VK_KHR_timeline_semaphore extension, semaphore is called syncobj in kernel side:
This extension introduces a new type of syncobj that has an integer payload identifying a point in a timeline. Such timeline syncobjs support the
following operations:
* CPU query - A host operation that allows querying the payload of the timeline syncobj.
* CPU wait - A host operation that allows a blocking wait for a
timeline syncobj to reach a specified value.
* Device wait - A device operation that allows waiting for a
timeline syncobj to reach a specified value.
* Device signal - A device operation that allows advancing the
timeline syncobj to a specified value.

Since it's a timeline, that means the front time point(PT) always is signaled before the late PT. a. signal PT design:
Signal PT fence N depends on PT[N-1] fence and signal opertion fence, when PT[N] fence is signaled, the timeline will increase to value of PT[N].
b. wait PT design:
Wait PT fence is signaled by reaching timeline point value, when timeline is increasing, will compare wait PTs value with new timeline value, if PT value is lower than timeline value, then wait PT will be signaled, otherwise keep in list. syncobj wait operation can wait on any point of timeline, so need a RB tree to order them. And wait PT could ahead of signal PT, we need a sumission fence to perform that.

v2:
1. remove unused DRM_SYNCOBJ_CREATE_TYPE_NORMAL. (Christian)
2. move unexposed denitions to .c file. (Daniel Vetter)
3. split up the change to drm_syncobj_find_fence() in a separate patch. (Christian)
4. split up the change to drm_syncobj_replace_fence() in a separate patch.
5. drop the submission_fence implementation and instead use wait_event() for that. (Christian)
6. WARN_ON(point != 0) for NORMAL type syncobj case. (Daniel Vetter)

v3:
1. replace normal syncobj with timeline implemenation. (Vetter and Christian)
a. normal syncobj signal op will create a signal PT to tail of signal pt list.
b. normal syncobj wait op will create a wait pt with last signal point, and this wait PT is only signaled by related signal point PT.
2. many bug fix and clean up
3. stub fence moving is moved to other patch.

v4:
1. fix RB tree loop with while(node=rb_first(...)). (Christian)
2. fix syncobj lifecycle. (Christian)
3. only enable_signaling when there is wait_pt. (Christian)
4. fix timeline path issues.
5. write a timeline test in libdrm

v5: (Christian)
1. semaphore is called syncobj in kernel side.
2. don't need 'timeline' characters in some function name.
3. keep syncobj cb

normal syncobj is tested by ./deqp-vk -n dEQP-VK*semaphore* timeline syncobj is tested by ./amdgpu_test -s 9

Cc Christian Konig <[email protected]>
Cc Dave Airlie <[email protected]>
Cc Daniel Rakos <[email protected]>
Cc Daniel Vetter <[email protected]>
Quoted 4 lines
Quoted 2 lines

At least on first glance that looks like it should work, going to do a detailed review on Monday.

Just for my understanding, it's all condensed down to 1 patch now? I kinda didn't follow the detailed discussion last few days at all :-/

I've already committed all the cleanup/fix prerequisites to drm-misc-next.

The driver specific implementation needs to come on top and maybe a new CPU wait IOCTL.

But essentially this patch is just the core of the kernel implementation.

Quoted 3 lines

Also, is there a testcase, igt highly preferred (because then we'll run it in our intel-gfx CI, and a bunch of people outside of intel have already discovered that and are using it).

libdrm patches and I think amdgpu based test cases where already published as well.

Not sure about igt testcases.

Christian.

Quoted 523 lines

Thanks, Daniel

Quoted 520 lines

Christian.

Quoted 518 lines
---
   drivers/gpu/drm/drm_syncobj.c              | 294 ++++++++++++++++++---
   drivers/gpu/drm/i915/i915_gem_execbuffer.c |   4 +-
   include/drm/drm_syncobj.h                  |  62 +++--
   include/uapi/drm/drm.h                     |   1 +
   4 files changed, 292 insertions(+), 69 deletions(-)
drivers/gpu/drm/drm_syncobj.c +258 -36
index e9ce623d049e..e78d076f2703 100644
@@ -56,6 +56,9 @@either
   #include "drm_internal.h"
   #include <drm/drm_syncobj.h>
+/* merge normal syncobj to timeline syncobj, the point interval is 1 */
+#define DRM_SYNCOBJ_NORMAL_POINT 1
+
   struct drm_syncobj_stub_fence {
   	struct dma_fence base;
   	spinlock_t lock;
@@ -82,6 +85,11 @@ static const struct dma_fence_ops drm_syncobj_stub_fence_ops = {
   	.release = drm_syncobj_stub_fence_release,
   };
+struct drm_syncobj_signal_pt {
+	struct dma_fence_array *base;
+	u64    value;
+	struct list_head list;
+};
   /**
    * drm_syncobj_find - lookup and reference a sync object.
@@ -124,7 +132,7 @@ static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
   {
   	int ret;
-	*fence = drm_syncobj_fence_get(syncobj);
+	ret = drm_syncobj_search_fence(syncobj, 0, 0, fence);
   	if (*fence)
   		return 1;
@@ -133,10 +141,10 @@ static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
   	 * have the lock, try one more time just to be sure we don't add a
   	 * callback when a fence has already been set.
   	 */
-	if (syncobj->fence) {
-		*fence = dma_fence_get(rcu_dereference_protected(syncobj->fence,
-								 lockdep_is_held(&syncobj->lock)));
-		ret = 1;
+	if (fence) {
+		drm_syncobj_search_fence(syncobj, 0, 0, fence);
+		if (*fence)
+			ret = 1;
   	} else {
   		*fence = NULL;
   		drm_syncobj_add_callback_locked(syncobj, cb, func);
@@ -164,6 +172,151 @@ void drm_syncobj_remove_callback(struct drm_syncobj *syncobj,
   	spin_unlock(&syncobj->lock);
   }
+static void drm_syncobj_timeline_init(struct drm_syncobj *syncobj,
+				      struct drm_syncobj_timeline *syncobj_timeline)
+{
+	spin_lock(&syncobj->lock);
+	syncobj_timeline->timeline_context = dma_fence_context_alloc(1);
+	syncobj_timeline->timeline = 0;
+	syncobj_timeline->signal_point = 0;
+	init_waitqueue_head(&syncobj_timeline->wq);
+
+	INIT_LIST_HEAD(&syncobj_timeline->signal_pt_list);
+	spin_unlock(&syncobj->lock);
+}
+
+static void drm_syncobj_timeline_fini(struct drm_syncobj *syncobj,
+				      struct drm_syncobj_timeline *syncobj_timeline)
+{
+	struct drm_syncobj_signal_pt *signal_pt = NULL, *tmp;
+
+	spin_lock(&syncobj->lock);
+	list_for_each_entry_safe(signal_pt, tmp,
+				 &syncobj_timeline->signal_pt_list, list) {
+		list_del(&signal_pt->list);
+		dma_fence_put(&signal_pt->base->base);
+		kfree(signal_pt);
+	}
+	spin_unlock(&syncobj->lock);
+}
+
+static struct dma_fence
+*drm_syncobj_find_signal_pt_for_point(struct drm_syncobj *syncobj,
+				      uint64_t point)
+{
+	struct drm_syncobj_timeline *timeline = &syncobj->syncobj_timeline;
+	struct drm_syncobj_signal_pt *signal_pt;
+
+	if ((syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) &&
+	    (point <= timeline->timeline)) {
+		struct drm_syncobj_stub_fence *fence =
+			kzalloc(sizeof(struct drm_syncobj_stub_fence),
+				GFP_KERNEL);
+
+		if (!fence)
+			return NULL;
+		spin_lock_init(&fence->lock);
+		dma_fence_init(&fence->base,
+			       &drm_syncobj_stub_fence_ops,
+			       &fence->lock,
+			       syncobj->syncobj_timeline.timeline_context,
+			       point);
+
+		dma_fence_signal(&fence->base);
+		return &fence->base;
+	}
+
+	list_for_each_entry(signal_pt, &timeline->signal_pt_list, list) {
+		if (point > signal_pt->value)
+			continue;
+		if ((syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) &&
+		    (point != signal_pt->value))
+			continue;
+		return dma_fence_get(&signal_pt->base->base);
+	}
+	return NULL;
+}
+
+static int drm_syncobj_create_signal_pt(struct drm_syncobj *syncobj,
+					struct dma_fence *fence,
+					u64 point)
+{
+	struct drm_syncobj_signal_pt *signal_pt =
+		kzalloc(sizeof(struct drm_syncobj_signal_pt), GFP_KERNEL);
+	struct drm_syncobj_signal_pt *tail_pt;
+	struct dma_fence **fences;
+	int num_fences = 0;
+	int ret = 0, i;
+
+	if (!signal_pt)
+		return -ENOMEM;
+	if (syncobj->syncobj_timeline.signal_point >= point) {
+		DRM_WARN("A later signal is ready!");
+		goto out;
+	}
+	if (!fence)
+		goto out;
+
+	fences = kmalloc_array(sizeof(void *), 2, GFP_KERNEL);
+	if (!fences)
+		goto out;
+	fences[num_fences++] = dma_fence_get(fence);
+	/* timeline syncobj must take this dependency */
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		spin_lock(&syncobj->lock);
+		if (!list_empty(&syncobj->syncobj_timeline.signal_pt_list)) {
+			tail_pt = list_last_entry(&syncobj->syncobj_timeline.signal_pt_list,
+						  struct drm_syncobj_signal_pt, list);
+			fences[num_fences++] = dma_fence_get(&tail_pt->base->base);
+		}
+		spin_unlock(&syncobj->lock);
+	}
+	signal_pt->base = dma_fence_array_create(num_fences, fences,
+						 syncobj->syncobj_timeline.timeline_context,
+						 point, false);
+	if (!signal_pt->base)
+		goto fail;
+
+	spin_lock(&syncobj->lock);
+	signal_pt->value = point;
+	INIT_LIST_HEAD(&signal_pt->list);
+	list_add_tail(&signal_pt->list, &syncobj->syncobj_timeline.signal_pt_list);
+	syncobj->syncobj_timeline.signal_point = point;
+	spin_unlock(&syncobj->lock);
+	wake_up_all(&syncobj->syncobj_timeline.wq);
+
+	return 0;
+fail:
+	for (i = 0; i < num_fences; i++)
+		dma_fence_put(fences[i]);
+	kfree(fences);
+out:
+	kfree(signal_pt);
+	return ret;
+}
+
+static void drm_syncobj_garbage_collection(struct drm_syncobj *syncobj)
+{
+	struct drm_syncobj_timeline *timeline = &syncobj->syncobj_timeline;
+	struct drm_syncobj_signal_pt *signal_pt, *tmp;
+
+	spin_lock(&syncobj->lock);
+	list_for_each_entry_safe(signal_pt, tmp,
+				 &timeline->signal_pt_list, list) {
+		if (dma_fence_is_signaled(&signal_pt->base->base)) {
+			timeline->timeline = signal_pt->value;
+			list_del(&signal_pt->list);
+			dma_fence_put(&signal_pt->base->base);
+			kfree(signal_pt);
+		} else {
+			/*signal_pt is in order in list, from small to big, so
+			 * the later must not be signal either */
+			break;
+		}
+	}
+
+	spin_unlock(&syncobj->lock);
+}
   /**
    * drm_syncobj_replace_fence - replace fence in a sync object.
    * @syncobj: Sync object to replace fence in
@@ -176,28 +329,37 @@ void drm_syncobj_replace_fence(struct drm_syncobj *syncobj,
   			       u64 point,
   			       struct dma_fence *fence)
   {
-	struct dma_fence *old_fence;
-	struct drm_syncobj_cb *cur, *tmp;
-
-	if (fence)
-		dma_fence_get(fence);
-
-	spin_lock(&syncobj->lock);
-
-	old_fence = rcu_dereference_protected(syncobj->fence,
-					      lockdep_is_held(&syncobj->lock));
-	rcu_assign_pointer(syncobj->fence, fence);
+	drm_syncobj_garbage_collection(syncobj);
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		if (fence)
+			drm_syncobj_create_signal_pt(syncobj, fence, point);
+	} else if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+		u64 pt_value;
+
+		if (!fence) {
+			drm_syncobj_timeline_fini(syncobj,
+						  &syncobj->syncobj_timeline);
+			drm_syncobj_timeline_init(syncobj,
+						  &syncobj->syncobj_timeline);
+			return;
+		}
+		pt_value = syncobj->syncobj_timeline.signal_point +
+			DRM_SYNCOBJ_NORMAL_POINT;
+		drm_syncobj_create_signal_pt(syncobj, fence, pt_value);
+	} else {
+		DRM_ERROR("the syncobj type isn't support\n");
+		return;
+	}
+	if (fence) {
+		struct drm_syncobj_cb *cur, *tmp;
-	if (fence != old_fence) {
+		spin_lock(&syncobj->lock);
   		list_for_each_entry_safe(cur, tmp, &syncobj->cb_list, node) {
   			list_del_init(&cur->node);
   			cur->func(syncobj, cur);
   		}
+		spin_unlock(&syncobj->lock);
   	}
-
-	spin_unlock(&syncobj->lock);
-
-	dma_fence_put(old_fence);
   }
   EXPORT_SYMBOL(drm_syncobj_replace_fence);
@@ -220,6 +382,25 @@ static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)
   	return 0;
   }
+static struct dma_fence *
+drm_syncobj_point_get(struct drm_syncobj *syncobj, u64 point, u64 flags)
+{
+	struct dma_fence *fence;
+	int ret = 0;
+
+	if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
+		ret = wait_event_timeout(syncobj->syncobj_timeline.wq,
+					point <= syncobj->syncobj_timeline.signal_point,
+					msecs_to_jiffies(10000)); /* wait 10s */
+		if (ret <= 0)
+			return NULL;
+	}
+	spin_lock(&syncobj->lock);
+	fence = drm_syncobj_find_signal_pt_for_point(syncobj, point);
+	spin_unlock(&syncobj->lock);
+	return fence;
+}
+
   /**
    * drm_syncobj_find_fence - lookup and reference the fence in a sync object
    * @file_private: drm file private pointer
@@ -234,20 +415,46 @@ static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)
    * contains a reference to the fence, which must be released by calling
    * dma_fence_put().
    */
-int drm_syncobj_find_fence(struct drm_file *file_private,
-			   u32 handle, u64 point,
-			   struct dma_fence **fence)
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+			     u64 flags, struct dma_fence **fence)
   {
-	struct drm_syncobj *syncobj = drm_syncobj_find(file_private, handle);
   	int ret = 0;
   	if (!syncobj)
   		return -ENOENT;
-	*fence = drm_syncobj_fence_get(syncobj);
+	drm_syncobj_garbage_collection(syncobj);
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+		/*NORMAL syncobj always wait on last pt */
+		u64 tail_pt_value = syncobj->syncobj_timeline.signal_point;
+
+		if (tail_pt_value == 0)
+			tail_pt_value += DRM_SYNCOBJ_NORMAL_POINT;
+		/* NORMAL syncobj doesn't care point value */
+		WARN_ON(point != 0);
+		*fence = drm_syncobj_point_get(syncobj, tail_pt_value,
+							flags);
+	} else if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		*fence = drm_syncobj_point_get(syncobj, point,
+							flags);
+	} else {
+		DRM_ERROR("Don't support this type syncobj\n");
+		*fence = NULL;
+	}
   	if (!*fence) {
   		ret = -EINVAL;
   	}
+	return ret;
+}
+EXPORT_SYMBOL(drm_syncobj_search_fence);
+int drm_syncobj_find_fence(struct drm_file *file_private,
+			   u32 handle, u64 point,
+			   struct dma_fence **fence) {
+	struct drm_syncobj *syncobj = drm_syncobj_find(file_private, handle);
+
+	int ret = drm_syncobj_search_fence(syncobj, point,
+					DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
+					fence);
   	drm_syncobj_put(syncobj);
   	return ret;
   }
@@ -264,7 +471,7 @@ void drm_syncobj_free(struct kref *kref)
   	struct drm_syncobj *syncobj = container_of(kref,
   						   struct drm_syncobj,
   						   refcount);
-	drm_syncobj_replace_fence(syncobj, 0, NULL);
+	drm_syncobj_timeline_fini(syncobj, &syncobj->syncobj_timeline);
   	kfree(syncobj);
   }
   EXPORT_SYMBOL(drm_syncobj_free);
@@ -294,6 +501,11 @@ int drm_syncobj_create(struct drm_syncobj **out_syncobj, uint32_t flags,
   	kref_init(&syncobj->refcount);
   	INIT_LIST_HEAD(&syncobj->cb_list);
   	spin_lock_init(&syncobj->lock);
+	if (flags & DRM_SYNCOBJ_CREATE_TYPE_TIMELINE)
+		syncobj->type = DRM_SYNCOBJ_TYPE_TIMELINE;
+	else
+		syncobj->type = DRM_SYNCOBJ_TYPE_NORMAL;
+	drm_syncobj_timeline_init(syncobj, &syncobj->syncobj_timeline);
   	if (flags & DRM_SYNCOBJ_CREATE_SIGNALED) {
   		ret = drm_syncobj_assign_null_handle(syncobj);
@@ -576,7 +788,8 @@ drm_syncobj_create_ioctl(struct drm_device *dev, void *data,
   		return -ENODEV;
   	/* no valid flags yet */
-	if (args->flags & ~DRM_SYNCOBJ_CREATE_SIGNALED)
+	if (args->flags & ~(DRM_SYNCOBJ_CREATE_SIGNALED |
+			    DRM_SYNCOBJ_CREATE_TYPE_TIMELINE))
   		return -EINVAL;
   	return drm_syncobj_create_as_handle(file_private,
@@ -669,9 +882,8 @@ static void syncobj_wait_syncobj_func(struct drm_syncobj *syncobj,
   	struct syncobj_wait_entry *wait =
   		container_of(cb, struct syncobj_wait_entry, syncobj_cb);
-	/* This happens inside the syncobj lock */
-	wait->fence = dma_fence_get(rcu_dereference_protected(syncobj->fence,
-							      lockdep_is_held(&syncobj->lock)));
+	drm_syncobj_search_fence(syncobj, 0, 0, &wait->fence);
+
   	wake_up_process(wait->task);
   }
@@ -698,7 +910,8 @@ static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs,
   	signaled_count = 0;
   	for (i = 0; i < count; ++i) {
   		entries[i].task = current;
-		entries[i].fence = drm_syncobj_fence_get(syncobjs[i]);
+		ret = drm_syncobj_search_fence(syncobjs[i], 0, 0,
+					       &entries[i].fence);
   		if (!entries[i].fence) {
   			if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
   				continue;
@@ -970,12 +1183,21 @@ drm_syncobj_reset_ioctl(struct drm_device *dev, void *data,
   	if (ret < 0)
   		return ret;
-	for (i = 0; i < args->count_handles; i++)
-		drm_syncobj_replace_fence(syncobjs[i], 0, NULL);
-
+	for (i = 0; i < args->count_handles; i++) {
+		if (syncobjs[i]->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+			DRM_ERROR("timeline syncobj cannot reset!\n");
+			ret = -EINVAL;
+			goto out;
+		}
+		drm_syncobj_timeline_fini(syncobjs[i],
+					  &syncobjs[i]->syncobj_timeline);
+		drm_syncobj_timeline_init(syncobjs[i],
+					  &syncobjs[i]->syncobj_timeline);
+	}
+out:
   	drm_syncobj_array_free(syncobjs, args->count_handles);
-	return 0;
+	return ret;
   }
   int
drivers/gpu/drm/i915/i915_gem_execbuffer.c +3 -1
index 0a8d2d64f380..579e91a5858b 100644
@@ -2137,7 +2137,9 @@ await_fence_array(struct i915_execbuffer *eb,
   		if (!(flags & I915_EXEC_FENCE_WAIT))
   			continue;
-		fence = drm_syncobj_fence_get(syncobj);
+		drm_syncobj_search_fence(syncobj, 0,
+					 DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
+					 &fence);
   		if (!fence)
   			return -EINVAL;
include/drm/drm_syncobj.h +30 -32
index 425432b85a87..9535521e6623 100644
@@ -30,6 +30,25 @@
   struct drm_syncobj_cb;
+enum drm_syncobj_type {
+	DRM_SYNCOBJ_TYPE_NORMAL,
+	DRM_SYNCOBJ_TYPE_TIMELINE
+};
+
+struct drm_syncobj_timeline {
+	wait_queue_head_t	wq;
+	u64 timeline_context;
+	/**
+	 * @timeline: syncobj timeline
+	 */
+	u64 timeline;
+	u64 signal_point;
+
+
+	struct rb_root wait_pt_tree;
+	struct list_head signal_pt_list;
+};
+
   /**
    * struct drm_syncobj - sync object.
    *
@@ -41,19 +60,19 @@ struct drm_syncobj {
   	 */
   	struct kref refcount;
   	/**
-	 * @fence:
-	 * NULL or a pointer to the fence bound to this object.
-	 *
-	 * This field should not be used directly. Use drm_syncobj_fence_get()
-	 * and drm_syncobj_replace_fence() instead.
+	 * @type: indicate syncobj type
   	 */
-	struct dma_fence __rcu *fence;
+	enum drm_syncobj_type type;
   	/**
-	 * @cb_list: List of callbacks to call when the &fence gets replaced.
+	 * @syncobj_timeline: timeline
   	 */
+	struct drm_syncobj_timeline syncobj_timeline;
+	/**
+         * @cb_list: List of callbacks to call when the &fence gets replaced.
+         */
   	struct list_head cb_list;
   	/**
-	 * @lock: Protects &cb_list and write-locks &fence.
+	 * @lock: Protects syncobj list and write-locks &fence.
   	 */
   	spinlock_t lock;
   	/**
@@ -68,7 +87,7 @@ typedef void (*drm_syncobj_func_t)(struct drm_syncobj *syncobj,
   /**
    * struct drm_syncobj_cb - callback for drm_syncobj_add_callback
    * @node: used by drm_syncob_add_callback to append this struct to
- *	  &drm_syncobj.cb_list
+ *       &drm_syncobj.cb_list
    * @func: drm_syncobj_func_t to call
    *
    * This struct will be initialized by drm_syncobj_add_callback, additional
@@ -106,29 +125,6 @@ drm_syncobj_put(struct drm_syncobj *obj)
   	kref_put(&obj->refcount, drm_syncobj_free);
   }
-/**
- * drm_syncobj_fence_get - get a reference to a fence in a sync object
- * @syncobj: sync object.
- *
- * This acquires additional reference to &drm_syncobj.fence contained in @obj,
- * if not NULL. It is illegal to call this without already holding a reference.
- * No locks required.
- *
- * Returns:
- * Either the fence of @obj or NULL if there's none.
- */
-static inline struct dma_fence *
-drm_syncobj_fence_get(struct drm_syncobj *syncobj)
-{
-	struct dma_fence *fence;
-
-	rcu_read_lock();
-	fence = dma_fence_get_rcu_safe(&syncobj->fence);
-	rcu_read_unlock();
-
-	return fence;
-}
-
   struct drm_syncobj *drm_syncobj_find(struct drm_file *file_private,
   				     u32 handle);
   void drm_syncobj_replace_fence(struct drm_syncobj *syncobj, u64 point,
@@ -142,5 +138,7 @@ int drm_syncobj_create(struct drm_syncobj **out_syncobj, uint32_t flags,
   int drm_syncobj_get_handle(struct drm_file *file_private,
   			   struct drm_syncobj *syncobj, u32 *handle);
   int drm_syncobj_get_fd(struct drm_syncobj *syncobj, int *p_fd);
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+			     u64 flags, struct dma_fence **fence);
   #endif
include/uapi/drm/drm.h +0 -1
index 300f336633f2..cebdb2541eb7 100644
--- a/include/uapi/drm/drm.h
Quoted 8 lines
Quoted 8 lines

include/uapi/drm/drm.h struct drm_prime_handle { · line 722

Quoted 8 lines

+++ b/include/uapi/drm/drm.h

include/uapi/drm/drm.h +1 -0
@@ -717,6 +717,7 @@ struct drm_prime_handle {
   struct drm_syncobj_create {
   	__u32 handle;
   #define DRM_SYNCOBJ_CREATE_SIGNALED (1 << 0)
+#define DRM_SYNCOBJ_CREATE_TYPE_TIMELINE (1 << 1)
   	__u32 flags;
   };

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

#6 lore raw
Quoted 88 lines · Christian König

Am 14.09.2018 um 18:10 schrieb Daniel Vetter:

Quoted 80 lines · Christian König
Quoted 78 lines

Am 14.09.2018 um 12:37 schrieb Chunming Zhou:

Quoted 75 lines

This patch is for VK_KHR_timeline_semaphore extension, semaphore is called syncobj in kernel side:
This extension introduces a new type of syncobj that has an integer payload
identifying a point in a timeline. Such timeline syncobjs support the following operations:
* CPU query - A host operation that allows querying the payload of
the
timeline syncobj.
* CPU wait - A host operation that allows a blocking wait for a timeline syncobj to reach a specified value.
* Device wait - A device operation that allows waiting for a timeline syncobj to reach a specified value.
* Device signal - A device operation that allows advancing the timeline syncobj to a specified value.

Since it's a timeline, that means the front time point(PT) always is signaled before the late PT.
a. signal PT design:
Signal PT fence N depends on PT[N-1] fence and signal opertion fence, when PT[N] fence is signaled,
the timeline will increase to value of PT[N].
b. wait PT design:
Wait PT fence is signaled by reaching timeline point value, when timeline is increasing, will compare
wait PTs value with new timeline value, if PT value is lower than timeline value, then wait PT will be
signaled, otherwise keep in list. syncobj wait operation can wait on any point of timeline,
so need a RB tree to order them. And wait PT could ahead of signal PT, we need a sumission fence to
perform that.

v2:
1. remove unused DRM_SYNCOBJ_CREATE_TYPE_NORMAL. (Christian)
2. move unexposed denitions to .c file. (Daniel Vetter)
3. split up the change to drm_syncobj_find_fence() in a separate patch. (Christian)
4. split up the change to drm_syncobj_replace_fence() in a separate patch.
5. drop the submission_fence implementation and instead use wait_event() for that. (Christian)
6. WARN_ON(point != 0) for NORMAL type syncobj case. (Daniel Vetter)

v3:
1. replace normal syncobj with timeline implemenation. (Vetter and Christian)
a. normal syncobj signal op will create a signal PT to tail of
signal pt list.
b. normal syncobj wait op will create a wait pt with last signal
point, and this wait PT is only signaled by related signal point PT.
2. many bug fix and clean up
3. stub fence moving is moved to other patch.

v4:
1. fix RB tree loop with while(node=rb_first(...)). (Christian)
2. fix syncobj lifecycle. (Christian)
3. only enable_signaling when there is wait_pt. (Christian)
4. fix timeline path issues.
5. write a timeline test in libdrm

v5: (Christian)
1. semaphore is called syncobj in kernel side.
2. don't need 'timeline' characters in some function name.
3. keep syncobj cb

normal syncobj is tested by ./deqp-vk -n dEQP-VK*semaphore* timeline syncobj is tested by ./amdgpu_test -s 9

Cc Christian Konig <[email protected]>
Cc Dave Airlie <[email protected]>
Cc Daniel Rakos <[email protected]>
Cc Daniel Vetter <[email protected]>
Quoted 5 lines
Quoted 2 lines

At least on first glance that looks like it should work, going to do a detailed review on Monday.

Just for my understanding, it's all condensed down to 1 patch now? I kinda didn't follow the detailed discussion last few days at all :-/

Quoted 6 lines

I've already committed all the cleanup/fix prerequisites to drm-misc-next.

The driver specific implementation needs to come on top and maybe a new CPU wait IOCTL.

But essentially this patch is just the core of the kernel implementation.

Ah cool, missed that.

Quoted 9 lines
Quoted 3 lines

Also, is there a testcase, igt highly preferred (because then we'll run it in our intel-gfx CI, and a bunch of people outside of intel have already discovered that and are using it).

libdrm patches and I think amdgpu based test cases where already published as well.

Not sure about igt testcases.

I guess we can write them when the intel implementation shows up. Just kinda still hoping that we'd have a more unfified test suite. And not really well-kept secret: We do have an amdgpu in our CI, in the form of kbl-g :-) But unfortunately it's not running the full test set for patches (only for drm-tip). But we could perhaps run more of the amdgpu tests somehow, if there's serious interest.

Cheers, Daniel

Quoted 598 lines

Christian.

Quoted 593 lines

Thanks, Daniel

Quoted 590 lines

Christian.

Quoted 588 lines
---
   drivers/gpu/drm/drm_syncobj.c              | 294
++++++++++++++++++---
   drivers/gpu/drm/i915/i915_gem_execbuffer.c |   4 +-
   include/drm/drm_syncobj.h                  |  62 +++--
   include/uapi/drm/drm.h                     |   1 +
   4 files changed, 292 insertions(+), 69 deletions(-)
drivers/gpu/drm/drm_syncobj.c +9 -1
index e9ce623d049e..e78d076f2703 100644
@@ -56,6 +56,9 @@either
   #include "drm_internal.h"
   #include <drm/drm_syncobj.h>
+/* merge normal syncobj to timeline syncobj, the point interval is 1 */
+#define DRM_SYNCOBJ_NORMAL_POINT 1
+
   struct drm_syncobj_stub_fence {
        struct dma_fence base;
        spinlock_t lock;
@@ -82,6 +85,11 @@ static const struct dma_fence_ops
 drm_syncobj_stub_fence_ops = {
        .release = drm_syncobj_stub_fence_release,
   };
+struct drm_syncobj_signal_pt {
+       struct dma_fence_array *base;
+       u64    value;
+       struct list_head list;
+};
   /**
    * drm_syncobj_find - lookup and reference a sync object.
@@ -124,7 +132,7 @@ static int
 drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
   {
        int ret;
-       *fence = drm_syncobj_fence_get(syncobj);
+       ret = drm_syncobj_search_fence(syncobj, 0, 0, fence);
        if (*fence)
                return 1;
@@ -133,10 +141,10 @@ static int
 drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
         * have the lock, try one more time just to be sure we don't add

a
* callback when a fence has already been set. */

drivers/gpu/drm/drm_syncobj.c +0 -2
-       if (syncobj->fence) {
-               *fence =

dma_fence_get(rcu_dereference_protected(syncobj->fence, -
lockdep_is_held(&syncobj->lock)));

drivers/gpu/drm/drm_syncobj.c +37 -1
-               ret = 1;
+       if (fence) {
+               drm_syncobj_search_fence(syncobj, 0, 0, fence);
+               if (*fence)
+                       ret = 1;
        } else {
                *fence = NULL;
                drm_syncobj_add_callback_locked(syncobj, cb, func);
@@ -164,6 +172,151 @@ void drm_syncobj_remove_callback(struct
 drm_syncobj *syncobj,
        spin_unlock(&syncobj->lock);
   }
+static void drm_syncobj_timeline_init(struct drm_syncobj *syncobj,
+                                     struct drm_syncobj_timeline
 *syncobj_timeline)
+{
+       spin_lock(&syncobj->lock);
+       syncobj_timeline->timeline_context = dma_fence_context_alloc(1);
+       syncobj_timeline->timeline = 0;
+       syncobj_timeline->signal_point = 0;
+       init_waitqueue_head(&syncobj_timeline->wq);
+
+       INIT_LIST_HEAD(&syncobj_timeline->signal_pt_list);
+       spin_unlock(&syncobj->lock);
+}
+
+static void drm_syncobj_timeline_fini(struct drm_syncobj *syncobj,
+                                     struct drm_syncobj_timeline
 *syncobj_timeline)
+{
+       struct drm_syncobj_signal_pt *signal_pt = NULL, *tmp;
+
+       spin_lock(&syncobj->lock);
+       list_for_each_entry_safe(signal_pt, tmp,
+                                &syncobj_timeline->signal_pt_list,
 list) {
+               list_del(&signal_pt->list);
+               dma_fence_put(&signal_pt->base->base);
+               kfree(signal_pt);
+       }
+       spin_unlock(&syncobj->lock);
+}
+
+static struct dma_fence
+*drm_syncobj_find_signal_pt_for_point(struct drm_syncobj *syncobj,
+                                     uint64_t point)
+{
+       struct drm_syncobj_timeline *timeline =

&syncobj->syncobj_timeline;

drivers/gpu/drm/drm_syncobj.c +15 -0
+       struct drm_syncobj_signal_pt *signal_pt;
+
+       if ((syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) &&
+           (point <= timeline->timeline)) {
+               struct drm_syncobj_stub_fence *fence =
+                       kzalloc(sizeof(struct drm_syncobj_stub_fence),
+                               GFP_KERNEL);
+
+               if (!fence)
+                       return NULL;
+               spin_lock_init(&fence->lock);
+               dma_fence_init(&fence->base,
+                              &drm_syncobj_stub_fence_ops,
+                              &fence->lock,
+

syncobj->syncobj_timeline.timeline_context,

drivers/gpu/drm/drm_syncobj.c +7 -0
+                              point);
+
+               dma_fence_signal(&fence->base);
+               return &fence->base;
+       }
+
+       list_for_each_entry(signal_pt, &timeline->signal_pt_list, list)

{

drivers/gpu/drm/drm_syncobj.c +16 -0
+               if (point > signal_pt->value)
+                       continue;
+               if ((syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) &&
+                   (point != signal_pt->value))
+                       continue;
+               return dma_fence_get(&signal_pt->base->base);
+       }
+       return NULL;
+}
+
+static int drm_syncobj_create_signal_pt(struct drm_syncobj *syncobj,
+                                       struct dma_fence *fence,
+                                       u64 point)
+{
+       struct drm_syncobj_signal_pt *signal_pt =
+               kzalloc(sizeof(struct drm_syncobj_signal_pt),

GFP_KERNEL);

drivers/gpu/drm/drm_syncobj.c +22 -0
+       struct drm_syncobj_signal_pt *tail_pt;
+       struct dma_fence **fences;
+       int num_fences = 0;
+       int ret = 0, i;
+
+       if (!signal_pt)
+               return -ENOMEM;
+       if (syncobj->syncobj_timeline.signal_point >= point) {
+               DRM_WARN("A later signal is ready!");
+               goto out;
+       }
+       if (!fence)
+               goto out;
+
+       fences = kmalloc_array(sizeof(void *), 2, GFP_KERNEL);
+       if (!fences)
+               goto out;
+       fences[num_fences++] = dma_fence_get(fence);
+       /* timeline syncobj must take this dependency */
+       if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+               spin_lock(&syncobj->lock);
+               if

(!list_empty(&syncobj->syncobj_timeline.signal_pt_list)) {
+ tail_pt =
list_last_entry(&syncobj->syncobj_timeline.signal_pt_list,
+ struct
drm_syncobj_signal_pt, list);
+ fences[num_fences++] =
dma_fence_get(&tail_pt->base->base);

drivers/gpu/drm/drm_syncobj.c +5 -0
+               }
+               spin_unlock(&syncobj->lock);
+       }
+       signal_pt->base = dma_fence_array_create(num_fences, fences,
+

syncobj->syncobj_timeline.timeline_context,

drivers/gpu/drm/drm_syncobj.c +8 -0
+                                                point, false);
+       if (!signal_pt->base)
+               goto fail;
+
+       spin_lock(&syncobj->lock);
+       signal_pt->value = point;
+       INIT_LIST_HEAD(&signal_pt->list);
+       list_add_tail(&signal_pt->list,

&syncobj->syncobj_timeline.signal_pt_list);

drivers/gpu/drm/drm_syncobj.c +17 -0
+       syncobj->syncobj_timeline.signal_point = point;
+       spin_unlock(&syncobj->lock);
+       wake_up_all(&syncobj->syncobj_timeline.wq);
+
+       return 0;
+fail:
+       for (i = 0; i < num_fences; i++)
+               dma_fence_put(fences[i]);
+       kfree(fences);
+out:
+       kfree(signal_pt);
+       return ret;
+}
+
+static void drm_syncobj_garbage_collection(struct drm_syncobj *syncobj)
+{
+       struct drm_syncobj_timeline *timeline =

&syncobj->syncobj_timeline;

drivers/gpu/drm/drm_syncobj.c +12 -0
+       struct drm_syncobj_signal_pt *signal_pt, *tmp;
+
+       spin_lock(&syncobj->lock);
+       list_for_each_entry_safe(signal_pt, tmp,
+                                &timeline->signal_pt_list, list) {
+               if (dma_fence_is_signaled(&signal_pt->base->base)) {
+                       timeline->timeline = signal_pt->value;
+                       list_del(&signal_pt->list);
+                       dma_fence_put(&signal_pt->base->base);
+                       kfree(signal_pt);
+               } else {
+                       /*signal_pt is in order in list, from small to

big, so

drivers/gpu/drm/drm_syncobj.c +42 -16
+                        * the later must not be signal either */
+                       break;
+               }
+       }
+
+       spin_unlock(&syncobj->lock);
+}
   /**
    * drm_syncobj_replace_fence - replace fence in a sync object.
    * @syncobj: Sync object to replace fence in
@@ -176,28 +329,37 @@ void drm_syncobj_replace_fence(struct drm_syncobj
 *syncobj,
                               u64 point,
                               struct dma_fence *fence)
   {
-       struct dma_fence *old_fence;
-       struct drm_syncobj_cb *cur, *tmp;
-
-       if (fence)
-               dma_fence_get(fence);
-
-       spin_lock(&syncobj->lock);
-
-       old_fence = rcu_dereference_protected(syncobj->fence,
-
 lockdep_is_held(&syncobj->lock));
-       rcu_assign_pointer(syncobj->fence, fence);
+       drm_syncobj_garbage_collection(syncobj);
+       if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+               if (fence)
+                       drm_syncobj_create_signal_pt(syncobj, fence,
 point);
+       } else if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+               u64 pt_value;
+
+               if (!fence) {
+                       drm_syncobj_timeline_fini(syncobj,
+
 &syncobj->syncobj_timeline);
+                       drm_syncobj_timeline_init(syncobj,
+
 &syncobj->syncobj_timeline);
+                       return;
+               }
+               pt_value = syncobj->syncobj_timeline.signal_point +
+                       DRM_SYNCOBJ_NORMAL_POINT;
+               drm_syncobj_create_signal_pt(syncobj, fence, pt_value);
+       } else {
+               DRM_ERROR("the syncobj type isn't support\n");
+               return;
+       }
+       if (fence) {
+               struct drm_syncobj_cb *cur, *tmp;
-       if (fence != old_fence) {
+               spin_lock(&syncobj->lock);
                list_for_each_entry_safe(cur, tmp, &syncobj->cb_list,
 node) {
                        list_del_init(&cur->node);
                        cur->func(syncobj, cur);
                }
+               spin_unlock(&syncobj->lock);
        }
-
-       spin_unlock(&syncobj->lock);
-
-       dma_fence_put(old_fence);
   }
   EXPORT_SYMBOL(drm_syncobj_replace_fence);
@@ -220,6 +382,25 @@ static int drm_syncobj_assign_null_handle(struct
 drm_syncobj *syncobj)
        return 0;
   }
+static struct dma_fence *
+drm_syncobj_point_get(struct drm_syncobj *syncobj, u64 point, u64
 flags)
+{
+       struct dma_fence *fence;
+       int ret = 0;
+
+       if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
+               ret = wait_event_timeout(syncobj->syncobj_timeline.wq,
+                                       point <=
 syncobj->syncobj_timeline.signal_point,
+                                       msecs_to_jiffies(10000)); /*

wait 10s */

drivers/gpu/drm/drm_syncobj.c +9 -0
+               if (ret <= 0)
+                       return NULL;
+       }
+       spin_lock(&syncobj->lock);
+       fence = drm_syncobj_find_signal_pt_for_point(syncobj, point);
+       spin_unlock(&syncobj->lock);
+       return fence;
+}
+
   /**
    * drm_syncobj_find_fence - lookup and reference the fence in a sync

object
* @file_private: drm file private pointer

drivers/gpu/drm/drm_syncobj.c +0 -0
@@ -234,20 +415,46 @@ static int drm_syncobj_assign_null_handle(struct
 drm_syncobj *syncobj)
    * contains a reference to the fence, which must be released by

calling
* dma_fence_put().
*/

drivers/gpu/drm/drm_syncobj.c +2 -4
-int drm_syncobj_find_fence(struct drm_file *file_private,
-                          u32 handle, u64 point,
-                          struct dma_fence **fence)
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+                            u64 flags, struct dma_fence **fence)
   {
-       struct drm_syncobj *syncobj = drm_syncobj_find(file_private,
handle);
        int ret = 0;
        if (!syncobj)
                return -ENOENT;
drivers/gpu/drm/drm_syncobj.c +4 -1
-       *fence = drm_syncobj_fence_get(syncobj);
+       drm_syncobj_garbage_collection(syncobj);
+       if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+               /*NORMAL syncobj always wait on last pt */
+               u64 tail_pt_value =

syncobj->syncobj_timeline.signal_point;

drivers/gpu/drm/drm_syncobj.c +21 -0
+
+               if (tail_pt_value == 0)
+                       tail_pt_value += DRM_SYNCOBJ_NORMAL_POINT;
+               /* NORMAL syncobj doesn't care point value */
+               WARN_ON(point != 0);
+               *fence = drm_syncobj_point_get(syncobj, tail_pt_value,
+                                                       flags);
+       } else if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+               *fence = drm_syncobj_point_get(syncobj, point,
+                                                       flags);
+       } else {
+               DRM_ERROR("Don't support this type syncobj\n");
+               *fence = NULL;
+       }
        if (!*fence) {
                ret = -EINVAL;
        }
+       return ret;
+}
+EXPORT_SYMBOL(drm_syncobj_search_fence);
+int drm_syncobj_find_fence(struct drm_file *file_private,
+                          u32 handle, u64 point,
+                          struct dma_fence **fence) {
+       struct drm_syncobj *syncobj = drm_syncobj_find(file_private,

handle);

drivers/gpu/drm/drm_syncobj.c +3 -0
+
+       int ret = drm_syncobj_search_fence(syncobj, point,
+

DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,

drivers/gpu/drm/drm_syncobj.c +11 -5
+                                       fence);
        drm_syncobj_put(syncobj);
        return ret;
   }
@@ -264,7 +471,7 @@ void drm_syncobj_free(struct kref *kref)
        struct drm_syncobj *syncobj = container_of(kref,
                                                   struct drm_syncobj,
                                                   refcount);
-       drm_syncobj_replace_fence(syncobj, 0, NULL);
+       drm_syncobj_timeline_fini(syncobj, &syncobj->syncobj_timeline);
        kfree(syncobj);
   }
   EXPORT_SYMBOL(drm_syncobj_free);
@@ -294,6 +501,11 @@ int drm_syncobj_create(struct drm_syncobj
 **out_syncobj, uint32_t flags,
        kref_init(&syncobj->refcount);
        INIT_LIST_HEAD(&syncobj->cb_list);
        spin_lock_init(&syncobj->lock);
+       if (flags & DRM_SYNCOBJ_CREATE_TYPE_TIMELINE)
+               syncobj->type = DRM_SYNCOBJ_TYPE_TIMELINE;
+       else
+               syncobj->type = DRM_SYNCOBJ_TYPE_NORMAL;
+       drm_syncobj_timeline_init(syncobj, &syncobj->syncobj_timeline);
        if (flags & DRM_SYNCOBJ_CREATE_SIGNALED) {
                ret = drm_syncobj_assign_null_handle(syncobj);
@@ -576,7 +788,8 @@ drm_syncobj_create_ioctl(struct drm_device *dev,
 void *data,
                return -ENODEV;
        /* no valid flags yet */
-       if (args->flags & ~DRM_SYNCOBJ_CREATE_SIGNALED)
+       if (args->flags & ~(DRM_SYNCOBJ_CREATE_SIGNALED |
+                           DRM_SYNCOBJ_CREATE_TYPE_TIMELINE))
                return -EINVAL;
        return drm_syncobj_create_as_handle(file_private,
@@ -669,9 +882,8 @@ static void syncobj_wait_syncobj_func(struct
 drm_syncobj *syncobj,
        struct syncobj_wait_entry *wait =
                container_of(cb, struct syncobj_wait_entry, syncobj_cb);
-       /* This happens inside the syncobj lock */
-       wait->fence =
 dma_fence_get(rcu_dereference_protected(syncobj->fence,
-
 lockdep_is_held(&syncobj->lock)));
+       drm_syncobj_search_fence(syncobj, 0, 0, &wait->fence);
+
        wake_up_process(wait->task);
   }
drivers/gpu/drm/drm_syncobj.c +2 -1
@@ -698,7 +910,8 @@ static signed long
 drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs,
        signaled_count = 0;
        for (i = 0; i < count; ++i) {
                entries[i].task = current;
-               entries[i].fence = drm_syncobj_fence_get(syncobjs[i]);
+               ret = drm_syncobj_search_fence(syncobjs[i], 0, 0,
+                                              &entries[i].fence);
                if (!entries[i].fence) {
                        if (flags &
DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
                                continue;
drivers/gpu/drm/drm_syncobj.c +13 -4
@@ -970,12 +1183,21 @@ drm_syncobj_reset_ioctl(struct drm_device *dev,
 void *data,
        if (ret < 0)
                return ret;
-       for (i = 0; i < args->count_handles; i++)
-               drm_syncobj_replace_fence(syncobjs[i], 0, NULL);
-
+       for (i = 0; i < args->count_handles; i++) {
+               if (syncobjs[i]->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+                       DRM_ERROR("timeline syncobj cannot reset!\n");
+                       ret = -EINVAL;
+                       goto out;
+               }
+               drm_syncobj_timeline_fini(syncobjs[i],
+
 &syncobjs[i]->syncobj_timeline);
+               drm_syncobj_timeline_init(syncobjs[i],
+
 &syncobjs[i]->syncobj_timeline);
+       }
+out:
        drm_syncobj_array_free(syncobjs, args->count_handles);
-       return 0;
+       return ret;
   }
   int
drivers/gpu/drm/i915/i915_gem_execbuffer.c +3 -1
index 0a8d2d64f380..579e91a5858b 100644
@@ -2137,7 +2137,9 @@ await_fence_array(struct i915_execbuffer *eb,
                if (!(flags & I915_EXEC_FENCE_WAIT))
                        continue;
-               fence = drm_syncobj_fence_get(syncobj);
+               drm_syncobj_search_fence(syncobj, 0,
+
 DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
+                                        &fence);
                if (!fence)
                        return -EINVAL;
include/drm/drm_syncobj.h +21 -7
index 425432b85a87..9535521e6623 100644
@@ -30,6 +30,25 @@
   struct drm_syncobj_cb;
+enum drm_syncobj_type {
+       DRM_SYNCOBJ_TYPE_NORMAL,
+       DRM_SYNCOBJ_TYPE_TIMELINE
+};
+
+struct drm_syncobj_timeline {
+       wait_queue_head_t       wq;
+       u64 timeline_context;
+       /**
+        * @timeline: syncobj timeline
+        */
+       u64 timeline;
+       u64 signal_point;
+
+
+       struct rb_root wait_pt_tree;
+       struct list_head signal_pt_list;
+};
+
   /**
    * struct drm_syncobj - sync object.
    *
@@ -41,19 +60,19 @@ struct drm_syncobj {
         */
        struct kref refcount;
        /**
-        * @fence:
-        * NULL or a pointer to the fence bound to this object.
-        *
-        * This field should not be used directly. Use
 drm_syncobj_fence_get()
-        * and drm_syncobj_replace_fence() instead.
+        * @type: indicate syncobj type
         */
-       struct dma_fence __rcu *fence;
+       enum drm_syncobj_type type;
        /**
-        * @cb_list: List of callbacks to call when the &fence gets

replaced.

include/drm/drm_syncobj.h +4 -0
+        * @syncobj_timeline: timeline
         */
+       struct drm_syncobj_timeline syncobj_timeline;
+       /**
+         * @cb_list: List of callbacks to call when the &fence gets

replaced.

include/drm/drm_syncobj.h +3 -2
+         */
        struct list_head cb_list;
        /**
-        * @lock: Protects &cb_list and write-locks &fence.
+        * @lock: Protects syncobj list and write-locks &fence.
         */
        spinlock_t lock;
        /**
@@ -68,7 +87,7 @@ typedef void (*drm_syncobj_func_t)(struct drm_syncobj
 *syncobj,
   /**
    * struct drm_syncobj_cb - callback for drm_syncobj_add_callback
    * @node: used by drm_syncob_add_callback to append this struct to
- *       &drm_syncobj.cb_list
+ *       &drm_syncobj.cb_list
    * @func: drm_syncobj_func_t to call
    *

* This struct will be initialized by drm_syncobj_add_callback,
additional

include/drm/drm_syncobj.h +0 -6
@@ -106,29 +125,6 @@ drm_syncobj_put(struct drm_syncobj *obj)
        kref_put(&obj->refcount, drm_syncobj_free);
   }
-/**
- * drm_syncobj_fence_get - get a reference to a fence in a sync object
- * @syncobj: sync object.
- *
- * This acquires additional reference to &drm_syncobj.fence contained
 in @obj,
- * if not NULL. It is illegal to call this without already holding a

reference.
- * No locks required.
- *
- * Returns:
- * Either the fence of @obj or NULL if there's none.
- */

include/drm/drm_syncobj.h +0 -12
-static inline struct dma_fence *
-drm_syncobj_fence_get(struct drm_syncobj *syncobj)
-{
-       struct dma_fence *fence;
-
-       rcu_read_lock();
-       fence = dma_fence_get_rcu_safe(&syncobj->fence);
-       rcu_read_unlock();
-
-       return fence;
-}
-
   struct drm_syncobj *drm_syncobj_find(struct drm_file *file_private,
                                     u32 handle);
   void drm_syncobj_replace_fence(struct drm_syncobj *syncobj, u64

point,

include/drm/drm_syncobj.h +2 -0
@@ -142,5 +138,7 @@ int drm_syncobj_create(struct drm_syncobj
 **out_syncobj, uint32_t flags,
   int drm_syncobj_get_handle(struct drm_file *file_private,
                           struct drm_syncobj *syncobj, u32 *handle);
   int drm_syncobj_get_fd(struct drm_syncobj *syncobj, int *p_fd);
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+                            u64 flags, struct dma_fence **fence);
   #endif
include/uapi/drm/drm.h +1 -0
index 300f336633f2..cebdb2541eb7 100644
@@ -717,6 +717,7 @@ struct drm_prime_handle {
   struct drm_syncobj_create {
        __u32 handle;
   #define DRM_SYNCOBJ_CREATE_SIGNALED (1 << 0)
+#define DRM_SYNCOBJ_CREATE_TYPE_TIMELINE (1 << 1)
        __u32 flags;
   };
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

A message that is not in the archive (sent off-list or not yet received): CAKMK7uH9jLO0bJe8ek-JyK3u1jbKY6Jk6xgT9ifgcko7Hi2URA-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org

#8 lore raw

Am 14.09.2018 um 20:24 schrieb Daniel Vetter:

Quoted 99 lines
Quoted 82 lines · Christian König

Am 14.09.2018 um 18:10 schrieb Daniel Vetter:

Quoted 76 lines · Christian König
Quoted 75 lines

Am 14.09.2018 um 12:37 schrieb Chunming Zhou:

Quoted 74 lines

This patch is for VK_KHR_timeline_semaphore extension, semaphore is called syncobj in kernel side:
This extension introduces a new type of syncobj that has an integer payload
identifying a point in a timeline. Such timeline syncobjs support the following operations:
* CPU query - A host operation that allows querying the payload of
the
timeline syncobj.
* CPU wait - A host operation that allows a blocking wait for a timeline syncobj to reach a specified value.
* Device wait - A device operation that allows waiting for a timeline syncobj to reach a specified value.
* Device signal - A device operation that allows advancing the timeline syncobj to a specified value.

Since it's a timeline, that means the front time point(PT) always is signaled before the late PT.
a. signal PT design:
Signal PT fence N depends on PT[N-1] fence and signal opertion fence, when PT[N] fence is signaled,
the timeline will increase to value of PT[N].
b. wait PT design:
Wait PT fence is signaled by reaching timeline point value, when timeline is increasing, will compare
wait PTs value with new timeline value, if PT value is lower than timeline value, then wait PT will be
signaled, otherwise keep in list. syncobj wait operation can wait on any point of timeline,
so need a RB tree to order them. And wait PT could ahead of signal PT, we need a sumission fence to
perform that.

v2:
1. remove unused DRM_SYNCOBJ_CREATE_TYPE_NORMAL. (Christian)
2. move unexposed denitions to .c file. (Daniel Vetter)
3. split up the change to drm_syncobj_find_fence() in a separate patch. (Christian)
4. split up the change to drm_syncobj_replace_fence() in a separate patch.
5. drop the submission_fence implementation and instead use wait_event() for that. (Christian)
6. WARN_ON(point != 0) for NORMAL type syncobj case. (Daniel Vetter)

v3:
1. replace normal syncobj with timeline implemenation. (Vetter and Christian)
a. normal syncobj signal op will create a signal PT to tail of
signal pt list.
b. normal syncobj wait op will create a wait pt with last signal
point, and this wait PT is only signaled by related signal point PT.
2. many bug fix and clean up
3. stub fence moving is moved to other patch.

v4:
1. fix RB tree loop with while(node=rb_first(...)). (Christian)
2. fix syncobj lifecycle. (Christian)
3. only enable_signaling when there is wait_pt. (Christian)
4. fix timeline path issues.
5. write a timeline test in libdrm

v5: (Christian)
1. semaphore is called syncobj in kernel side.
2. don't need 'timeline' characters in some function name.
3. keep syncobj cb

normal syncobj is tested by ./deqp-vk -n dEQP-VK*semaphore* timeline syncobj is tested by ./amdgpu_test -s 9

Cc Christian Konig <[email protected]>
Cc Dave Airlie <[email protected]>
Cc Daniel Rakos <[email protected]>
Cc Daniel Vetter <[email protected]>
Quoted 4 lines
Quoted 2 lines

At least on first glance that looks like it should work, going to do a detailed review on Monday.

Just for my understanding, it's all condensed down to 1 patch now? I kinda didn't follow the detailed discussion last few days at all :-/

Quoted 6 lines

I've already committed all the cleanup/fix prerequisites to drm-misc-next.

The driver specific implementation needs to come on top and maybe a new CPU wait IOCTL.

But essentially this patch is just the core of the kernel implementation.

Ah cool, missed that.

Quoted 7 lines
Quoted 3 lines

Also, is there a testcase, igt highly preferred (because then we'll run it in our intel-gfx CI, and a bunch of people outside of intel have already discovered that and are using it).

libdrm patches and I think amdgpu based test cases where already published as well.

Quoted 7 lines
Quoted 1 line

Not sure about igt testcases.

I guess we can write them when the intel implementation shows up. Just kinda still hoping that we'd have a more unfified test suite. And not really well-kept secret: We do have an amdgpu in our CI, in the form of kbl-g :-) But unfortunately it's not running the full test set for patches (only for drm-tip). But we could perhaps run more of the amdgpu tests somehow, if there's serious interest.

Well I wouldn't mind if we sooner or later get rid of the amdgpu unit tests in libdrm.

They are more or less just a really bloody mess.

Christian.

Quoted 602 lines

Cheers, Daniel

Quoted 596 lines

Christian.

Quoted 592 lines

Thanks, Daniel

Quoted 590 lines

Christian.

Quoted 588 lines
---
    drivers/gpu/drm/drm_syncobj.c              | 294
++++++++++++++++++---
    drivers/gpu/drm/i915/i915_gem_execbuffer.c |   4 +-
    include/drm/drm_syncobj.h                  |  62 +++--
    include/uapi/drm/drm.h                     |   1 +
    4 files changed, 292 insertions(+), 69 deletions(-)
drivers/gpu/drm/drm_syncobj.c +9 -1
index e9ce623d049e..e78d076f2703 100644
@@ -56,6 +56,9 @@either
    #include "drm_internal.h"
    #include <drm/drm_syncobj.h>
+/* merge normal syncobj to timeline syncobj, the point interval is 1 */
+#define DRM_SYNCOBJ_NORMAL_POINT 1
+
    struct drm_syncobj_stub_fence {
         struct dma_fence base;
         spinlock_t lock;
@@ -82,6 +85,11 @@ static const struct dma_fence_ops
 drm_syncobj_stub_fence_ops = {
         .release = drm_syncobj_stub_fence_release,
    };
+struct drm_syncobj_signal_pt {
+       struct dma_fence_array *base;
+       u64    value;
+       struct list_head list;
+};
    /**
     * drm_syncobj_find - lookup and reference a sync object.
@@ -124,7 +132,7 @@ static int
 drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
    {
         int ret;
-       *fence = drm_syncobj_fence_get(syncobj);
+       ret = drm_syncobj_search_fence(syncobj, 0, 0, fence);
         if (*fence)
                 return 1;
@@ -133,10 +141,10 @@ static int
 drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
          * have the lock, try one more time just to be sure we don't add

a
* callback when a fence has already been set. */

drivers/gpu/drm/drm_syncobj.c +0 -2
-       if (syncobj->fence) {
-               *fence =

dma_fence_get(rcu_dereference_protected(syncobj->fence, -
lockdep_is_held(&syncobj->lock)));

drivers/gpu/drm/drm_syncobj.c +37 -1
-               ret = 1;
+       if (fence) {
+               drm_syncobj_search_fence(syncobj, 0, 0, fence);
+               if (*fence)
+                       ret = 1;
         } else {
                 *fence = NULL;
                 drm_syncobj_add_callback_locked(syncobj, cb, func);
@@ -164,6 +172,151 @@ void drm_syncobj_remove_callback(struct
 drm_syncobj *syncobj,
         spin_unlock(&syncobj->lock);
    }
+static void drm_syncobj_timeline_init(struct drm_syncobj *syncobj,
+                                     struct drm_syncobj_timeline
 *syncobj_timeline)
+{
+       spin_lock(&syncobj->lock);
+       syncobj_timeline->timeline_context = dma_fence_context_alloc(1);
+       syncobj_timeline->timeline = 0;
+       syncobj_timeline->signal_point = 0;
+       init_waitqueue_head(&syncobj_timeline->wq);
+
+       INIT_LIST_HEAD(&syncobj_timeline->signal_pt_list);
+       spin_unlock(&syncobj->lock);
+}
+
+static void drm_syncobj_timeline_fini(struct drm_syncobj *syncobj,
+                                     struct drm_syncobj_timeline
 *syncobj_timeline)
+{
+       struct drm_syncobj_signal_pt *signal_pt = NULL, *tmp;
+
+       spin_lock(&syncobj->lock);
+       list_for_each_entry_safe(signal_pt, tmp,
+                                &syncobj_timeline->signal_pt_list,
 list) {
+               list_del(&signal_pt->list);
+               dma_fence_put(&signal_pt->base->base);
+               kfree(signal_pt);
+       }
+       spin_unlock(&syncobj->lock);
+}
+
+static struct dma_fence
+*drm_syncobj_find_signal_pt_for_point(struct drm_syncobj *syncobj,
+                                     uint64_t point)
+{
+       struct drm_syncobj_timeline *timeline =

&syncobj->syncobj_timeline;

drivers/gpu/drm/drm_syncobj.c +15 -0
+       struct drm_syncobj_signal_pt *signal_pt;
+
+       if ((syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) &&
+           (point <= timeline->timeline)) {
+               struct drm_syncobj_stub_fence *fence =
+                       kzalloc(sizeof(struct drm_syncobj_stub_fence),
+                               GFP_KERNEL);
+
+               if (!fence)
+                       return NULL;
+               spin_lock_init(&fence->lock);
+               dma_fence_init(&fence->base,
+                              &drm_syncobj_stub_fence_ops,
+                              &fence->lock,
+

syncobj->syncobj_timeline.timeline_context,

drivers/gpu/drm/drm_syncobj.c +7 -0
+                              point);
+
+               dma_fence_signal(&fence->base);
+               return &fence->base;
+       }
+
+       list_for_each_entry(signal_pt, &timeline->signal_pt_list, list)

{

drivers/gpu/drm/drm_syncobj.c +16 -0
+               if (point > signal_pt->value)
+                       continue;
+               if ((syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) &&
+                   (point != signal_pt->value))
+                       continue;
+               return dma_fence_get(&signal_pt->base->base);
+       }
+       return NULL;
+}
+
+static int drm_syncobj_create_signal_pt(struct drm_syncobj *syncobj,
+                                       struct dma_fence *fence,
+                                       u64 point)
+{
+       struct drm_syncobj_signal_pt *signal_pt =
+               kzalloc(sizeof(struct drm_syncobj_signal_pt),

GFP_KERNEL);

drivers/gpu/drm/drm_syncobj.c +22 -0
+       struct drm_syncobj_signal_pt *tail_pt;
+       struct dma_fence **fences;
+       int num_fences = 0;
+       int ret = 0, i;
+
+       if (!signal_pt)
+               return -ENOMEM;
+       if (syncobj->syncobj_timeline.signal_point >= point) {
+               DRM_WARN("A later signal is ready!");
+               goto out;
+       }
+       if (!fence)
+               goto out;
+
+       fences = kmalloc_array(sizeof(void *), 2, GFP_KERNEL);
+       if (!fences)
+               goto out;
+       fences[num_fences++] = dma_fence_get(fence);
+       /* timeline syncobj must take this dependency */
+       if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+               spin_lock(&syncobj->lock);
+               if

(!list_empty(&syncobj->syncobj_timeline.signal_pt_list)) {
+ tail_pt =
list_last_entry(&syncobj->syncobj_timeline.signal_pt_list,
+ struct
drm_syncobj_signal_pt, list);
+ fences[num_fences++] =
dma_fence_get(&tail_pt->base->base);

drivers/gpu/drm/drm_syncobj.c +5 -0
+               }
+               spin_unlock(&syncobj->lock);
+       }
+       signal_pt->base = dma_fence_array_create(num_fences, fences,
+

syncobj->syncobj_timeline.timeline_context,

drivers/gpu/drm/drm_syncobj.c +8 -0
+                                                point, false);
+       if (!signal_pt->base)
+               goto fail;
+
+       spin_lock(&syncobj->lock);
+       signal_pt->value = point;
+       INIT_LIST_HEAD(&signal_pt->list);
+       list_add_tail(&signal_pt->list,

&syncobj->syncobj_timeline.signal_pt_list);

drivers/gpu/drm/drm_syncobj.c +17 -0
+       syncobj->syncobj_timeline.signal_point = point;
+       spin_unlock(&syncobj->lock);
+       wake_up_all(&syncobj->syncobj_timeline.wq);
+
+       return 0;
+fail:
+       for (i = 0; i < num_fences; i++)
+               dma_fence_put(fences[i]);
+       kfree(fences);
+out:
+       kfree(signal_pt);
+       return ret;
+}
+
+static void drm_syncobj_garbage_collection(struct drm_syncobj *syncobj)
+{
+       struct drm_syncobj_timeline *timeline =

&syncobj->syncobj_timeline;

drivers/gpu/drm/drm_syncobj.c +12 -0
+       struct drm_syncobj_signal_pt *signal_pt, *tmp;
+
+       spin_lock(&syncobj->lock);
+       list_for_each_entry_safe(signal_pt, tmp,
+                                &timeline->signal_pt_list, list) {
+               if (dma_fence_is_signaled(&signal_pt->base->base)) {
+                       timeline->timeline = signal_pt->value;
+                       list_del(&signal_pt->list);
+                       dma_fence_put(&signal_pt->base->base);
+                       kfree(signal_pt);
+               } else {
+                       /*signal_pt is in order in list, from small to

big, so

drivers/gpu/drm/drm_syncobj.c +42 -16
+                        * the later must not be signal either */
+                       break;
+               }
+       }
+
+       spin_unlock(&syncobj->lock);
+}
    /**
     * drm_syncobj_replace_fence - replace fence in a sync object.
     * @syncobj: Sync object to replace fence in
@@ -176,28 +329,37 @@ void drm_syncobj_replace_fence(struct drm_syncobj
 *syncobj,
                                u64 point,
                                struct dma_fence *fence)
    {
-       struct dma_fence *old_fence;
-       struct drm_syncobj_cb *cur, *tmp;
-
-       if (fence)
-               dma_fence_get(fence);
-
-       spin_lock(&syncobj->lock);
-
-       old_fence = rcu_dereference_protected(syncobj->fence,
-
 lockdep_is_held(&syncobj->lock));
-       rcu_assign_pointer(syncobj->fence, fence);
+       drm_syncobj_garbage_collection(syncobj);
+       if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+               if (fence)
+                       drm_syncobj_create_signal_pt(syncobj, fence,
 point);
+       } else if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+               u64 pt_value;
+
+               if (!fence) {
+                       drm_syncobj_timeline_fini(syncobj,
+
 &syncobj->syncobj_timeline);
+                       drm_syncobj_timeline_init(syncobj,
+
 &syncobj->syncobj_timeline);
+                       return;
+               }
+               pt_value = syncobj->syncobj_timeline.signal_point +
+                       DRM_SYNCOBJ_NORMAL_POINT;
+               drm_syncobj_create_signal_pt(syncobj, fence, pt_value);
+       } else {
+               DRM_ERROR("the syncobj type isn't support\n");
+               return;
+       }
+       if (fence) {
+               struct drm_syncobj_cb *cur, *tmp;
-       if (fence != old_fence) {
+               spin_lock(&syncobj->lock);
                 list_for_each_entry_safe(cur, tmp, &syncobj->cb_list,
 node) {
                         list_del_init(&cur->node);
                         cur->func(syncobj, cur);
                 }
+               spin_unlock(&syncobj->lock);
         }
-
-       spin_unlock(&syncobj->lock);
-
-       dma_fence_put(old_fence);
    }
    EXPORT_SYMBOL(drm_syncobj_replace_fence);
@@ -220,6 +382,25 @@ static int drm_syncobj_assign_null_handle(struct
 drm_syncobj *syncobj)
         return 0;
    }
+static struct dma_fence *
+drm_syncobj_point_get(struct drm_syncobj *syncobj, u64 point, u64
 flags)
+{
+       struct dma_fence *fence;
+       int ret = 0;
+
+       if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
+               ret = wait_event_timeout(syncobj->syncobj_timeline.wq,
+                                       point <=
 syncobj->syncobj_timeline.signal_point,
+                                       msecs_to_jiffies(10000)); /*

wait 10s */

drivers/gpu/drm/drm_syncobj.c +9 -0
+               if (ret <= 0)
+                       return NULL;
+       }
+       spin_lock(&syncobj->lock);
+       fence = drm_syncobj_find_signal_pt_for_point(syncobj, point);
+       spin_unlock(&syncobj->lock);
+       return fence;
+}
+
    /**
     * drm_syncobj_find_fence - lookup and reference the fence in a sync

object
* @file_private: drm file private pointer

drivers/gpu/drm/drm_syncobj.c +0 -0
@@ -234,20 +415,46 @@ static int drm_syncobj_assign_null_handle(struct
 drm_syncobj *syncobj)
     * contains a reference to the fence, which must be released by

calling
* dma_fence_put().
*/

drivers/gpu/drm/drm_syncobj.c +2 -4
-int drm_syncobj_find_fence(struct drm_file *file_private,
-                          u32 handle, u64 point,
-                          struct dma_fence **fence)
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+                            u64 flags, struct dma_fence **fence)
    {
-       struct drm_syncobj *syncobj = drm_syncobj_find(file_private,
handle);
         int ret = 0;
         if (!syncobj)
                 return -ENOENT;
drivers/gpu/drm/drm_syncobj.c +4 -1
-       *fence = drm_syncobj_fence_get(syncobj);
+       drm_syncobj_garbage_collection(syncobj);
+       if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+               /*NORMAL syncobj always wait on last pt */
+               u64 tail_pt_value =

syncobj->syncobj_timeline.signal_point;

drivers/gpu/drm/drm_syncobj.c +21 -0
+
+               if (tail_pt_value == 0)
+                       tail_pt_value += DRM_SYNCOBJ_NORMAL_POINT;
+               /* NORMAL syncobj doesn't care point value */
+               WARN_ON(point != 0);
+               *fence = drm_syncobj_point_get(syncobj, tail_pt_value,
+                                                       flags);
+       } else if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+               *fence = drm_syncobj_point_get(syncobj, point,
+                                                       flags);
+       } else {
+               DRM_ERROR("Don't support this type syncobj\n");
+               *fence = NULL;
+       }
         if (!*fence) {
                 ret = -EINVAL;
         }
+       return ret;
+}
+EXPORT_SYMBOL(drm_syncobj_search_fence);
+int drm_syncobj_find_fence(struct drm_file *file_private,
+                          u32 handle, u64 point,
+                          struct dma_fence **fence) {
+       struct drm_syncobj *syncobj = drm_syncobj_find(file_private,

handle);

drivers/gpu/drm/drm_syncobj.c +3 -0
+
+       int ret = drm_syncobj_search_fence(syncobj, point,
+

DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,

drivers/gpu/drm/drm_syncobj.c +11 -5
+                                       fence);
         drm_syncobj_put(syncobj);
         return ret;
    }
@@ -264,7 +471,7 @@ void drm_syncobj_free(struct kref *kref)
         struct drm_syncobj *syncobj = container_of(kref,
                                                    struct drm_syncobj,
                                                    refcount);
-       drm_syncobj_replace_fence(syncobj, 0, NULL);
+       drm_syncobj_timeline_fini(syncobj, &syncobj->syncobj_timeline);
         kfree(syncobj);
    }
    EXPORT_SYMBOL(drm_syncobj_free);
@@ -294,6 +501,11 @@ int drm_syncobj_create(struct drm_syncobj
 **out_syncobj, uint32_t flags,
         kref_init(&syncobj->refcount);
         INIT_LIST_HEAD(&syncobj->cb_list);
         spin_lock_init(&syncobj->lock);
+       if (flags & DRM_SYNCOBJ_CREATE_TYPE_TIMELINE)
+               syncobj->type = DRM_SYNCOBJ_TYPE_TIMELINE;
+       else
+               syncobj->type = DRM_SYNCOBJ_TYPE_NORMAL;
+       drm_syncobj_timeline_init(syncobj, &syncobj->syncobj_timeline);
         if (flags & DRM_SYNCOBJ_CREATE_SIGNALED) {
                 ret = drm_syncobj_assign_null_handle(syncobj);
@@ -576,7 +788,8 @@ drm_syncobj_create_ioctl(struct drm_device *dev,
 void *data,
                 return -ENODEV;
         /* no valid flags yet */
-       if (args->flags & ~DRM_SYNCOBJ_CREATE_SIGNALED)
+       if (args->flags & ~(DRM_SYNCOBJ_CREATE_SIGNALED |
+                           DRM_SYNCOBJ_CREATE_TYPE_TIMELINE))
                 return -EINVAL;
         return drm_syncobj_create_as_handle(file_private,
@@ -669,9 +882,8 @@ static void syncobj_wait_syncobj_func(struct
 drm_syncobj *syncobj,
         struct syncobj_wait_entry *wait =
                 container_of(cb, struct syncobj_wait_entry, syncobj_cb);
-       /* This happens inside the syncobj lock */
-       wait->fence =
 dma_fence_get(rcu_dereference_protected(syncobj->fence,
-
 lockdep_is_held(&syncobj->lock)));
+       drm_syncobj_search_fence(syncobj, 0, 0, &wait->fence);
+
         wake_up_process(wait->task);
    }
drivers/gpu/drm/drm_syncobj.c +2 -1
@@ -698,7 +910,8 @@ static signed long
 drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs,
         signaled_count = 0;
         for (i = 0; i < count; ++i) {
                 entries[i].task = current;
-               entries[i].fence = drm_syncobj_fence_get(syncobjs[i]);
+               ret = drm_syncobj_search_fence(syncobjs[i], 0, 0,
+                                              &entries[i].fence);
                 if (!entries[i].fence) {
                         if (flags &
DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
                                 continue;
drivers/gpu/drm/drm_syncobj.c +13 -4
@@ -970,12 +1183,21 @@ drm_syncobj_reset_ioctl(struct drm_device *dev,
 void *data,
         if (ret < 0)
                 return ret;
-       for (i = 0; i < args->count_handles; i++)
-               drm_syncobj_replace_fence(syncobjs[i], 0, NULL);
-
+       for (i = 0; i < args->count_handles; i++) {
+               if (syncobjs[i]->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+                       DRM_ERROR("timeline syncobj cannot reset!\n");
+                       ret = -EINVAL;
+                       goto out;
+               }
+               drm_syncobj_timeline_fini(syncobjs[i],
+
 &syncobjs[i]->syncobj_timeline);
+               drm_syncobj_timeline_init(syncobjs[i],
+
 &syncobjs[i]->syncobj_timeline);
+       }
+out:
         drm_syncobj_array_free(syncobjs, args->count_handles);
-       return 0;
+       return ret;
    }
    int
drivers/gpu/drm/i915/i915_gem_execbuffer.c +3 -1
index 0a8d2d64f380..579e91a5858b 100644
@@ -2137,7 +2137,9 @@ await_fence_array(struct i915_execbuffer *eb,
                 if (!(flags & I915_EXEC_FENCE_WAIT))
                         continue;
-               fence = drm_syncobj_fence_get(syncobj);
+               drm_syncobj_search_fence(syncobj, 0,
+
 DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
+                                        &fence);
                 if (!fence)
                         return -EINVAL;
include/drm/drm_syncobj.h +21 -7
index 425432b85a87..9535521e6623 100644
@@ -30,6 +30,25 @@
    struct drm_syncobj_cb;
+enum drm_syncobj_type {
+       DRM_SYNCOBJ_TYPE_NORMAL,
+       DRM_SYNCOBJ_TYPE_TIMELINE
+};
+
+struct drm_syncobj_timeline {
+       wait_queue_head_t       wq;
+       u64 timeline_context;
+       /**
+        * @timeline: syncobj timeline
+        */
+       u64 timeline;
+       u64 signal_point;
+
+
+       struct rb_root wait_pt_tree;
+       struct list_head signal_pt_list;
+};
+
    /**
     * struct drm_syncobj - sync object.
     *
@@ -41,19 +60,19 @@ struct drm_syncobj {
          */
         struct kref refcount;
         /**
-        * @fence:
-        * NULL or a pointer to the fence bound to this object.
-        *
-        * This field should not be used directly. Use
 drm_syncobj_fence_get()
-        * and drm_syncobj_replace_fence() instead.
+        * @type: indicate syncobj type
          */
-       struct dma_fence __rcu *fence;
+       enum drm_syncobj_type type;
         /**
-        * @cb_list: List of callbacks to call when the &fence gets

replaced.

include/drm/drm_syncobj.h +4 -0
+        * @syncobj_timeline: timeline
          */
+       struct drm_syncobj_timeline syncobj_timeline;
+       /**
+         * @cb_list: List of callbacks to call when the &fence gets

replaced.

include/drm/drm_syncobj.h +3 -2
+         */
         struct list_head cb_list;
         /**
-        * @lock: Protects &cb_list and write-locks &fence.
+        * @lock: Protects syncobj list and write-locks &fence.
          */
         spinlock_t lock;
         /**
@@ -68,7 +87,7 @@ typedef void (*drm_syncobj_func_t)(struct drm_syncobj
 *syncobj,
    /**
     * struct drm_syncobj_cb - callback for drm_syncobj_add_callback
     * @node: used by drm_syncob_add_callback to append this struct to
- *       &drm_syncobj.cb_list
+ *       &drm_syncobj.cb_list
     * @func: drm_syncobj_func_t to call
     *

* This struct will be initialized by drm_syncobj_add_callback,
additional

include/drm/drm_syncobj.h +0 -6
@@ -106,29 +125,6 @@ drm_syncobj_put(struct drm_syncobj *obj)
         kref_put(&obj->refcount, drm_syncobj_free);
    }
-/**
- * drm_syncobj_fence_get - get a reference to a fence in a sync object
- * @syncobj: sync object.
- *
- * This acquires additional reference to &drm_syncobj.fence contained
 in @obj,
- * if not NULL. It is illegal to call this without already holding a

reference.
- * No locks required.
- *
- * Returns:
- * Either the fence of @obj or NULL if there's none.
- */

include/drm/drm_syncobj.h +0 -12
-static inline struct dma_fence *
-drm_syncobj_fence_get(struct drm_syncobj *syncobj)
-{
-       struct dma_fence *fence;
-
-       rcu_read_lock();
-       fence = dma_fence_get_rcu_safe(&syncobj->fence);
-       rcu_read_unlock();
-
-       return fence;
-}
-
    struct drm_syncobj *drm_syncobj_find(struct drm_file *file_private,
                                      u32 handle);
    void drm_syncobj_replace_fence(struct drm_syncobj *syncobj, u64

point,

include/drm/drm_syncobj.h +2 -0
@@ -142,5 +138,7 @@ int drm_syncobj_create(struct drm_syncobj
 **out_syncobj, uint32_t flags,
    int drm_syncobj_get_handle(struct drm_file *file_private,
                            struct drm_syncobj *syncobj, u32 *handle);
    int drm_syncobj_get_fd(struct drm_syncobj *syncobj, int *p_fd);
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+                            u64 flags, struct dma_fence **fence);
    #endif
include/uapi/drm/drm.h +1 -0
index 300f336633f2..cebdb2541eb7 100644
@@ -717,6 +717,7 @@ struct drm_prime_handle {
    struct drm_syncobj_create {
         __u32 handle;
    #define DRM_SYNCOBJ_CREATE_SIGNALED (1 << 0)
+#define DRM_SYNCOBJ_CREATE_TYPE_TIMELINE (1 << 1)
         __u32 flags;
    };

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

#9 lore raw
In reply to a message not in the archive
Quoted 82 lines

-----Original Message-----
From: Daniel Vetter <[email protected]> On Behalf Of Daniel Vetter
Sent: Saturday, September 15, 2018 12:11 AM
To: Koenig, Christian <[email protected]>
Cc: Zhou, David(ChunMing) <[email protected]>; dri- [email protected]; [email protected]; Dave Airlie <[email protected]>; Rakos, Daniel <[email protected]>; Daniel Vetter <[email protected]>
Subject: Re: [PATCH] [RFC]drm: add syncobj timeline support v5

Quoted 2 lines · Christian König

Am 14.09.2018 um 12:37 schrieb Chunming Zhou:

Quoted 1 line

This patch is for VK_KHR_timeline_semaphore extension, semaphore is

called syncobj in kernel side:

Quoted 12 lines
Quoted 12 lines

This extension introduces a new type of syncobj that has an integer payload identifying a point in a timeline. Such timeline syncobjs support the following operations:
* CPU query - A host operation that allows querying the payload of the timeline syncobj.
* CPU wait - A host operation that allows a blocking wait for a timeline syncobj to reach a specified value.
* Device wait - A device operation that allows waiting for a timeline syncobj to reach a specified value.
* Device signal - A device operation that allows advancing the timeline syncobj to a specified value.

Quoted 1 line
Quoted 1 line

Since it's a timeline, that means the front time point(PT) always is

signaled before the late PT.

Quoted 3 lines
Quoted 3 lines

a. signal PT design:
Signal PT fence N depends on PT[N-1] fence and signal opertion fence, when PT[N] fence is signaled, the timeline will increase to value of

PT[N].

Quoted 6 lines
Quoted 6 lines

b. wait PT design:
Wait PT fence is signaled by reaching timeline point value, when timeline is increasing, will compare wait PTs value with new timeline value, if PT value is lower than timeline value, then wait PT will be signaled, otherwise keep in list. syncobj wait operation can wait on any point of timeline, so need a RB tree to order them. And

wait PT could ahead of signal PT, we need a sumission fence to perform that.

Quoted 3 lines
Quoted 3 lines

v2:
1. remove unused DRM_SYNCOBJ_CREATE_TYPE_NORMAL. (Christian) 2.

move

Quoted 3 lines
Quoted 3 lines

unexposed denitions to .c file. (Daniel Vetter) 3. split up the change to drm_syncobj_find_fence() in a separate patch. (Christian)
4. split up the change to drm_syncobj_replace_fence() in a separate

patch.

Quoted 6 lines
Quoted 6 lines

5. drop the submission_fence implementation and instead use wait_event() for that. (Christian) 6. WARN_ON(point != 0) for NORMAL type syncobj case. (Daniel Vetter)

v3:
1. replace normal syncobj with timeline implemenation. (Vetter and

Christian)

Quoted 2 lines
Quoted 2 lines

a. normal syncobj signal op will create a signal PT to tail of signal pt list. b. normal syncobj wait op will create a wait pt with last signal point, and

this wait PT is only signaled by related signal point PT.

Quoted 23 lines
Quoted 22 lines

2. many bug fix and clean up
3. stub fence moving is moved to other patch.

v4:
1. fix RB tree loop with while(node=rb_first(...)). (Christian) 2. fix syncobj lifecycle. (Christian) 3. only enable_signaling when there is wait_pt. (Christian) 4. fix timeline path issues.
5. write a timeline test in libdrm

v5: (Christian)
1. semaphore is called syncobj in kernel side.
2. don't need 'timeline' characters in some function name.
3. keep syncobj cb

normal syncobj is tested by ./deqp-vk -n dEQP-VK*semaphore* timeline syncobj is tested by ./amdgpu_test -s 9

Cc Christian Konig <[email protected]>
Cc Dave Airlie <[email protected]>
Cc Daniel Rakos <[email protected]>
Cc Daniel Vetter <[email protected]>
Quoted 4 lines
Quoted 2 lines

At least on first glance that looks like it should work, going to do a detailed review on Monday.

Just for my understanding, it's all condensed down to 1 patch now?

Yes, Christian suggest that.

Quoted 6 lines

I kinda
didn't follow the detailed discussion last few days at all :-/

Also, is there a testcase, igt highly preferred (because then we'll run it in our intel-gfx CI, and a bunch of people outside of intel have already discovered that and are using it).

I already wrote the test in libdrm unit test, since I'm not familiar with IGT stuff.

Thanks,
David Zhou

Quoted 575 lines

Thanks, Daniel

Quoted 10 lines

Christian.

Quoted 7 lines
---
  drivers/gpu/drm/drm_syncobj.c              | 294 ++++++++++++++++++---
  drivers/gpu/drm/i915/i915_gem_execbuffer.c |   4 +-
  include/drm/drm_syncobj.h                  |  62 +++--
  include/uapi/drm/drm.h                     |   1 +
  4 files changed, 292 insertions(+), 69 deletions(-)
Quoted 15 lines

drivers/gpu/drm/drm_syncobj.c static const struct dma_fence_ops

Quoted 15 lines
(quoted diff) +0 -0

b/drivers/gpu/drm/drm_syncobj.c index e9ce623d049e..e78d076f2703 100644

drivers/gpu/drm/drm_syncobj.c +3 -0
@@ -56,6 +56,9 @@either
  #include "drm_internal.h"
  #include <drm/drm_syncobj.h>
+/* merge normal syncobj to timeline syncobj, the point interval is
+1 */ #define DRM_SYNCOBJ_NORMAL_POINT 1
+
  struct drm_syncobj_stub_fence {
  	struct dma_fence base;
  	spinlock_t lock;
@@ -82,6 +85,11 @@ static const struct dma_fence_ops

drm_syncobj_stub_fence_ops = {

Quoted 10 lines

drivers/gpu/drm/drm_syncobj.c static int

Quoted 10 lines
drivers/gpu/drm/drm_syncobj.c +5 -0
  	.release = drm_syncobj_stub_fence_release,
  };
+struct drm_syncobj_signal_pt {
+	struct dma_fence_array *base;
+	u64    value;
+	struct list_head list;
+};
  /**
   * drm_syncobj_find - lookup and reference a sync object.
@@ -124,7 +132,7 @@ static int

drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,

Quoted 7 lines

drivers/gpu/drm/drm_syncobj.c static int

Quoted 7 lines
drivers/gpu/drm/drm_syncobj.c +1 -1
  {
  	int ret;
-	*fence = drm_syncobj_fence_get(syncobj);
+	ret = drm_syncobj_search_fence(syncobj, 0, 0, fence);
  	if (*fence)
  		return 1;
@@ -133,10 +141,10 @@ static int

drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,

Quoted 5 lines

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,

Quoted 5 lines
drivers/gpu/drm/drm_syncobj.c +0 -2
  	 * have the lock, try one more time just to be sure we don't add a
  	 * callback when a fence has already been set.
  	 */
-	if (syncobj->fence) {
-		*fence =

dma_fence_get(rcu_dereference_protected(syncobj->fence,

Quoted 1 line

drivers/gpu/drm/drm_syncobj.c static int · line 138

Quoted 1 line
drivers/gpu/drm/drm_syncobj.c +0 -1
-

lockdep_is_held(&syncobj->lock)));

Quoted 8 lines

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,

Quoted 8 lines
drivers/gpu/drm/drm_syncobj.c +4 -1
-		ret = 1;
+	if (fence) {
+		drm_syncobj_search_fence(syncobj, 0, 0, fence);
+		if (*fence)
+			ret = 1;
  	} else {
  		*fence = NULL;
  		drm_syncobj_add_callback_locked(syncobj, cb, func); @@ -

164,6

Quoted 1 line
Quoted 1 line

+172,151 @@ void drm_syncobj_remove_callback(struct drm_syncobj

*syncobj,

Quoted 4 lines

drivers/gpu/drm/drm_syncobj.c

Quoted 4 lines
  	spin_unlock(&syncobj->lock);
  }
drivers/gpu/drm/drm_syncobj.c +2 -0
+static void drm_syncobj_timeline_init(struct drm_syncobj *syncobj,
+				      struct drm_syncobj_timeline

*syncobj_timeline) {

Quoted 4 lines
Quoted 4 lines
drivers/gpu/drm/drm_syncobj.c +4 -0
+	spin_lock(&syncobj->lock);
+	syncobj_timeline->timeline_context = dma_fence_context_alloc(1);
+	syncobj_timeline->timeline = 0;
+	syncobj_timeline->signal_point = 0;
Quoted 8 lines

drivers/gpu/drm/drm_syncobj.c void drm_syncobj_remove_callback(struct drm_syncobj *syncobj,

Quoted 8 lines
drivers/gpu/drm/drm_syncobj.c +8 -0
+	init_waitqueue_head(&syncobj_timeline->wq);
+
+	INIT_LIST_HEAD(&syncobj_timeline->signal_pt_list);
+	spin_unlock(&syncobj->lock);
+}
+
+static void drm_syncobj_timeline_fini(struct drm_syncobj *syncobj,
+				      struct drm_syncobj_timeline

*syncobj_timeline) {

Quoted 80 lines
Quoted 7 lines
drivers/gpu/drm/drm_syncobj.c +7 -0
+	struct drm_syncobj_signal_pt *signal_pt = NULL, *tmp;
+
+	spin_lock(&syncobj->lock);
+	list_for_each_entry_safe(signal_pt, tmp,
+				 &syncobj_timeline->signal_pt_list, list) {
+		list_del(&signal_pt->list);
+		dma_fence_put(&signal_pt->base->base);

drivers/gpu/drm/drm_syncobj.c void drm_syncobj_remove_callback(struct drm_syncobj *syncobj,

Quoted 10 lines
drivers/gpu/drm/drm_syncobj.c +10 -0
+		kfree(signal_pt);
+	}
+	spin_unlock(&syncobj->lock);
+}
+
+static struct dma_fence
+*drm_syncobj_find_signal_pt_for_point(struct drm_syncobj *syncobj,
+				      uint64_t point)
+{
+	struct drm_syncobj_timeline *timeline = &syncobj-

syncobj_timeline;

Quoted 51 lines
drivers/gpu/drm/drm_syncobj.c +51 -0
+	struct drm_syncobj_signal_pt *signal_pt;
+
+	if ((syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) &&
+	    (point <= timeline->timeline)) {
+		struct drm_syncobj_stub_fence *fence =
+			kzalloc(sizeof(struct drm_syncobj_stub_fence),
+				GFP_KERNEL);
+
+		if (!fence)
+			return NULL;
+		spin_lock_init(&fence->lock);
+		dma_fence_init(&fence->base,
+			       &drm_syncobj_stub_fence_ops,
+			       &fence->lock,
+			       syncobj->syncobj_timeline.timeline_context,
+			       point);
+
+		dma_fence_signal(&fence->base);
+		return &fence->base;
+	}
+
+	list_for_each_entry(signal_pt, &timeline->signal_pt_list, list) {
+		if (point > signal_pt->value)
+			continue;
+		if ((syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) &&
+		    (point != signal_pt->value))
+			continue;
+		return dma_fence_get(&signal_pt->base->base);
+	}
+	return NULL;
+}
+
+static int drm_syncobj_create_signal_pt(struct drm_syncobj *syncobj,
+					struct dma_fence *fence,
+					u64 point)
+{
+	struct drm_syncobj_signal_pt *signal_pt =
+		kzalloc(sizeof(struct drm_syncobj_signal_pt), GFP_KERNEL);
+	struct drm_syncobj_signal_pt *tail_pt;
+	struct dma_fence **fences;
+	int num_fences = 0;
+	int ret = 0, i;
+
+	if (!signal_pt)
+		return -ENOMEM;
+	if (syncobj->syncobj_timeline.signal_point >= point) {
+		DRM_WARN("A later signal is ready!");
+		goto out;
+	}
+	if (!fence)
+		goto out;

drivers/gpu/drm/drm_syncobj.c void drm_syncobj_remove_callback(struct drm_syncobj *syncobj,

Quoted 10 lines
drivers/gpu/drm/drm_syncobj.c +10 -0
+
+	fences = kmalloc_array(sizeof(void *), 2, GFP_KERNEL);
+	if (!fences)
+		goto out;
+	fences[num_fences++] = dma_fence_get(fence);
+	/* timeline syncobj must take this dependency */
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		spin_lock(&syncobj->lock);
+		if (!list_empty(&syncobj->syncobj_timeline.signal_pt_list)) {
+			tail_pt = list_last_entry(&syncobj-

syncobj_timeline.signal_pt_list,

Quoted 1 line

drivers/gpu/drm/drm_syncobj.c

Quoted 1 line
drivers/gpu/drm/drm_syncobj.c +1 -0
+						  struct

drm_syncobj_signal_pt, list);

Quoted 68 lines

drivers/gpu/drm/drm_syncobj.c

Quoted 1 line
drivers/gpu/drm/drm_syncobj.c +1 -0
+			fences[num_fences++] = dma_fence_get(&tail_pt-

base->base);

drivers/gpu/drm/drm_syncobj.c void drm_syncobj_remove_callback(struct drm_syncobj *syncobj,

Quoted 5 lines
drivers/gpu/drm/drm_syncobj.c +5 -0
+		}
+		spin_unlock(&syncobj->lock);
+	}
+	signal_pt->base = dma_fence_array_create(num_fences, fences,
+						 syncobj-

syncobj_timeline.timeline_context,

drivers/gpu/drm/drm_syncobj.c void drm_syncobj_remove_callback(struct drm_syncobj *syncobj,

Quoted 8 lines
drivers/gpu/drm/drm_syncobj.c +8 -0
+						 point, false);
+	if (!signal_pt->base)
+		goto fail;
+
+	spin_lock(&syncobj->lock);
+	signal_pt->value = point;
+	INIT_LIST_HEAD(&signal_pt->list);
+	list_add_tail(&signal_pt->list, &syncobj-

syncobj_timeline.signal_pt_list);

Quoted 7 lines
drivers/gpu/drm/drm_syncobj.c +7 -0
+	syncobj->syncobj_timeline.signal_point = point;
+	spin_unlock(&syncobj->lock);
+	wake_up_all(&syncobj->syncobj_timeline.wq);
+
+	return 0;
+fail:
+	for (i = 0; i < num_fences; i++)

drivers/gpu/drm/drm_syncobj.c void drm_syncobj_remove_callback(struct drm_syncobj *syncobj,

Quoted 10 lines
drivers/gpu/drm/drm_syncobj.c +10 -0
+		dma_fence_put(fences[i]);
+	kfree(fences);
+out:
+	kfree(signal_pt);
+	return ret;
+}
+
+static void drm_syncobj_garbage_collection(struct drm_syncobj
+*syncobj) {
+	struct drm_syncobj_timeline *timeline = &syncobj-

syncobj_timeline;

Quoted 33 lines
drivers/gpu/drm/drm_syncobj.c +19 -0
+	struct drm_syncobj_signal_pt *signal_pt, *tmp;
+
+	spin_lock(&syncobj->lock);
+	list_for_each_entry_safe(signal_pt, tmp,
+				 &timeline->signal_pt_list, list) {
+		if (dma_fence_is_signaled(&signal_pt->base->base)) {
+			timeline->timeline = signal_pt->value;
+			list_del(&signal_pt->list);
+			dma_fence_put(&signal_pt->base->base);
+			kfree(signal_pt);
+		} else {
+			/*signal_pt is in order in list, from small to big, so
+			 * the later must not be signal either */
+			break;
+		}
+	}
+
+	spin_unlock(&syncobj->lock);
+}
  /**
   * drm_syncobj_replace_fence - replace fence in a sync object.
   * @syncobj: Sync object to replace fence in @@ -176,28 +329,37 @@

void drm_syncobj_replace_fence(struct drm_syncobj *syncobj,
u64 point,
struct dma_fence *fence)
{

drivers/gpu/drm/drm_syncobj.c +0 -7
-	struct dma_fence *old_fence;
-	struct drm_syncobj_cb *cur, *tmp;
-
-	if (fence)
-		dma_fence_get(fence);
-
-	spin_lock(&syncobj->lock);
Quoted 8 lines

drivers/gpu/drm/drm_syncobj.c void drm_syncobj_replace_fence(struct drm_syncobj *syncobj,

Quoted 8 lines
drivers/gpu/drm/drm_syncobj.c +4 -4
-
-	old_fence = rcu_dereference_protected(syncobj->fence,
-					      lockdep_is_held(&syncobj->lock));
-	rcu_assign_pointer(syncobj->fence, fence);
+	drm_syncobj_garbage_collection(syncobj);
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		if (fence)
+			drm_syncobj_create_signal_pt(syncobj, fence,

point);

Quoted 10 lines

drivers/gpu/drm/drm_syncobj.c void drm_syncobj_replace_fence(struct drm_syncobj *syncobj,

Quoted 6 lines
drivers/gpu/drm/drm_syncobj.c +6 -0
+	} else if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+		u64 pt_value;
+
+		if (!fence) {
+			drm_syncobj_timeline_fini(syncobj,
+						  &syncobj-

syncobj_timeline);

drivers/gpu/drm/drm_syncobj.c

Quoted 2 lines
drivers/gpu/drm/drm_syncobj.c +2 -0
+			drm_syncobj_timeline_init(syncobj,
+						  &syncobj-

syncobj_timeline);

Quoted 14 lines
Quoted 4 lines
drivers/gpu/drm/drm_syncobj.c +4 -0
+			return;
+		}
+		pt_value = syncobj->syncobj_timeline.signal_point +
+			DRM_SYNCOBJ_NORMAL_POINT;

drivers/gpu/drm/drm_syncobj.c void drm_syncobj_replace_fence(struct drm_syncobj *syncobj,

Quoted 10 lines
drivers/gpu/drm/drm_syncobj.c +8 -1
+		drm_syncobj_create_signal_pt(syncobj, fence, pt_value);
+	} else {
+		DRM_ERROR("the syncobj type isn't support\n");
+		return;
+	}
+	if (fence) {
+		struct drm_syncobj_cb *cur, *tmp;
-	if (fence != old_fence) {
+		spin_lock(&syncobj->lock);
  		list_for_each_entry_safe(cur, tmp, &syncobj->cb_list, node)

{

Quoted 4 lines
Quoted 4 lines
  			list_del_init(&cur->node);
  			cur->func(syncobj, cur);
  		}
drivers/gpu/drm/drm_syncobj.c +1 -0
+		spin_unlock(&syncobj->lock);
Quoted 8 lines

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_assign_null_handle(struct

Quoted 8 lines
  	}
drivers/gpu/drm/drm_syncobj.c +0 -4
-
-	spin_unlock(&syncobj->lock);
-
-	dma_fence_put(old_fence);
  }
  EXPORT_SYMBOL(drm_syncobj_replace_fence);
@@ -220,6 +382,25 @@ static int drm_syncobj_assign_null_handle(struct

drm_syncobj *syncobj)

Quoted 12 lines

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)

Quoted 11 lines
drivers/gpu/drm/drm_syncobj.c +9 -0
  	return 0;
  }
+static struct dma_fence *
+drm_syncobj_point_get(struct drm_syncobj *syncobj, u64 point, u64
+flags) {
+	struct dma_fence *fence;
+	int ret = 0;
+
+	if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
+		ret = wait_event_timeout(syncobj->syncobj_timeline.wq,
+					point <= syncobj-

syncobj_timeline.signal_point,

Quoted 1 line

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_assign_null_handle(struct · line 393

Quoted 1 line
drivers/gpu/drm/drm_syncobj.c +1 -0
+					msecs_to_jiffies(10000)); /* wait 10s

*/

Quoted 11 lines

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)

Quoted 11 lines
drivers/gpu/drm/drm_syncobj.c +9 -0
+		if (ret <= 0)
+			return NULL;
+	}
+	spin_lock(&syncobj->lock);
+	fence = drm_syncobj_find_signal_pt_for_point(syncobj, point);
+	spin_unlock(&syncobj->lock);
+	return fence;
+}
+
  /**
   * drm_syncobj_find_fence - lookup and reference the fence in a sync

object

Quoted 4 lines
Quoted 4 lines
drivers/gpu/drm/drm_syncobj.c +0 -0
   * @file_private: drm file private pointer @@ -234,20 +415,46 @@
 static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)

* contains a reference to the fence, which must be released by calling
* dma_fence_put().

Quoted 8 lines

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)

Quoted 8 lines
   */
drivers/gpu/drm/drm_syncobj.c +2 -4
-int drm_syncobj_find_fence(struct drm_file *file_private,
-			   u32 handle, u64 point,
-			   struct dma_fence **fence)
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+			     u64 flags, struct dma_fence **fence)
  {
-	struct drm_syncobj *syncobj = drm_syncobj_find(file_private,

handle);

Quoted 28 lines
Quoted 28 lines
  	int ret = 0;
  	if (!syncobj)
  		return -ENOENT;
drivers/gpu/drm/drm_syncobj.c +21 -1
-	*fence = drm_syncobj_fence_get(syncobj);
+	drm_syncobj_garbage_collection(syncobj);
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+		/*NORMAL syncobj always wait on last pt */
+		u64 tail_pt_value = syncobj->syncobj_timeline.signal_point;
+
+		if (tail_pt_value == 0)
+			tail_pt_value += DRM_SYNCOBJ_NORMAL_POINT;
+		/* NORMAL syncobj doesn't care point value */
+		WARN_ON(point != 0);
+		*fence = drm_syncobj_point_get(syncobj, tail_pt_value,
+							flags);
+	} else if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		*fence = drm_syncobj_point_get(syncobj, point,
+							flags);
+	} else {
+		DRM_ERROR("Don't support this type syncobj\n");
+		*fence = NULL;
+	}
  	if (!*fence) {
  		ret = -EINVAL;
  	}
+	return ret;
+}
+EXPORT_SYMBOL(drm_syncobj_search_fence);
Quoted 8 lines

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)

Quoted 8 lines
drivers/gpu/drm/drm_syncobj.c +8 -0
+int drm_syncobj_find_fence(struct drm_file *file_private,
+			   u32 handle, u64 point,
+			   struct dma_fence **fence) {
+	struct drm_syncobj *syncobj = drm_syncobj_find(file_private,
+handle);
+
+	int ret = drm_syncobj_search_fence(syncobj, point,
+

DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,

Quoted 6 lines
Quoted 6 lines
drivers/gpu/drm/drm_syncobj.c +1 -0
+					fence);
  	drm_syncobj_put(syncobj);
  	return ret;
  }
@@ -264,7 +471,7 @@ void drm_syncobj_free(struct kref *kref)
  	struct drm_syncobj *syncobj = container_of(kref,
Quoted 8 lines

drivers/gpu/drm/drm_syncobj.c int drm_syncobj_create(struct drm_syncobj

Quoted 8 lines
drivers/gpu/drm/drm_syncobj.c +1 -1
  						   struct drm_syncobj,
  						   refcount);
-	drm_syncobj_replace_fence(syncobj, 0, NULL);
+	drm_syncobj_timeline_fini(syncobj, &syncobj->syncobj_timeline);
  	kfree(syncobj);
  }
  EXPORT_SYMBOL(drm_syncobj_free);
@@ -294,6 +501,11 @@ int drm_syncobj_create(struct drm_syncobj

**out_syncobj, uint32_t flags,

Quoted 11 lines

drivers/gpu/drm/drm_syncobj.c drm_syncobj_create_ioctl(struct drm_device

Quoted 11 lines
drivers/gpu/drm/drm_syncobj.c +5 -0
  	kref_init(&syncobj->refcount);
  	INIT_LIST_HEAD(&syncobj->cb_list);
  	spin_lock_init(&syncobj->lock);
+	if (flags & DRM_SYNCOBJ_CREATE_TYPE_TIMELINE)
+		syncobj->type = DRM_SYNCOBJ_TYPE_TIMELINE;
+	else
+		syncobj->type = DRM_SYNCOBJ_TYPE_NORMAL;
+	drm_syncobj_timeline_init(syncobj, &syncobj->syncobj_timeline);
  	if (flags & DRM_SYNCOBJ_CREATE_SIGNALED) {
  		ret = drm_syncobj_assign_null_handle(syncobj);
@@ -576,7 +788,8 @@ drm_syncobj_create_ioctl(struct drm_device

*dev, void *data,

Quoted 8 lines

drivers/gpu/drm/drm_syncobj.c static void syncobj_wait_syncobj_func(struct

Quoted 8 lines
drivers/gpu/drm/drm_syncobj.c +2 -1
  		return -ENODEV;
  	/* no valid flags yet */
-	if (args->flags & ~DRM_SYNCOBJ_CREATE_SIGNALED)
+	if (args->flags & ~(DRM_SYNCOBJ_CREATE_SIGNALED |
+			    DRM_SYNCOBJ_CREATE_TYPE_TIMELINE))
  		return -EINVAL;
  	return drm_syncobj_create_as_handle(file_private,
@@ -669,9 +882,8 @@ static void syncobj_wait_syncobj_func(struct

drm_syncobj *syncobj,

Quoted 6 lines

drivers/gpu/drm/drm_syncobj.c static void syncobj_wait_syncobj_func(struct drm_syncobj *syncobj,

Quoted 4 lines
drivers/gpu/drm/drm_syncobj.c +0 -2
  	struct syncobj_wait_entry *wait =
  		container_of(cb, struct syncobj_wait_entry, syncobj_cb);
-	/* This happens inside the syncobj lock */
-	wait->fence = dma_fence_get(rcu_dereference_protected(syncobj-

fence,

Quoted 1 line
drivers/gpu/drm/drm_syncobj.c +0 -1
-

lockdep_is_held(&syncobj->lock)));

Quoted 5 lines

drivers/gpu/drm/drm_syncobj.c static signed long

Quoted 5 lines
drivers/gpu/drm/drm_syncobj.c +2 -0
+	drm_syncobj_search_fence(syncobj, 0, 0, &wait->fence);
+
  	wake_up_process(wait->task);
  }
@@ -698,7 +910,8 @@ static signed long

drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs,

Quoted 8 lines

drivers/gpu/drm/drm_syncobj.c static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs,

Quoted 8 lines
drivers/gpu/drm/drm_syncobj.c +2 -1
  	signaled_count = 0;
  	for (i = 0; i < count; ++i) {
  		entries[i].task = current;
-		entries[i].fence = drm_syncobj_fence_get(syncobjs[i]);
+		ret = drm_syncobj_search_fence(syncobjs[i], 0, 0,
+					       &entries[i].fence);
  		if (!entries[i].fence) {
  			if (flags &

DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {

Quoted 2 lines

drivers/gpu/drm/drm_syncobj.c drm_syncobj_reset_ioctl(struct drm_device

Quoted 2 lines
drivers/gpu/drm/drm_syncobj.c +0 -0
  				continue;
@@ -970,12 +1183,21 @@ drm_syncobj_reset_ioctl(struct drm_device

*dev, void *data,

Quoted 25 lines
Quoted 25 lines
drivers/gpu/drm/drm_syncobj.c +13 -4
  	if (ret < 0)
  		return ret;
-	for (i = 0; i < args->count_handles; i++)
-		drm_syncobj_replace_fence(syncobjs[i], 0, NULL);
-
+	for (i = 0; i < args->count_handles; i++) {
+		if (syncobjs[i]->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+			DRM_ERROR("timeline syncobj cannot reset!\n");
+			ret = -EINVAL;
+			goto out;
+		}
+		drm_syncobj_timeline_fini(syncobjs[i],
+					  &syncobjs[i]->syncobj_timeline);
+		drm_syncobj_timeline_init(syncobjs[i],
+					  &syncobjs[i]->syncobj_timeline);
+	}
+out:
  	drm_syncobj_array_free(syncobjs, args->count_handles);
-	return 0;
+	return ret;
  }
  int
drivers/gpu/drm/i915/i915_gem_execbuffer.c +0 -0
index 0a8d2d64f380..579e91a5858b 100644
Quoted 8 lines

drivers/gpu/drm/i915/i915_gem_execbuffer.c await_fence_array(struct i915_execbuffer *eb, · line 2140

Quoted 8 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +2 -1
@@ -2137,7 +2137,9 @@ await_fence_array(struct i915_execbuffer *eb,
  		if (!(flags & I915_EXEC_FENCE_WAIT))
  			continue;
-		fence = drm_syncobj_fence_get(syncobj);
+		drm_syncobj_search_fence(syncobj, 0,
+

DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,

Quoted 31 lines
Quoted 31 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +1 -0
+					 &fence);
  		if (!fence)
  			return -EINVAL;
include/drm/drm_syncobj.h +19 -0
index 425432b85a87..9535521e6623 100644
@@ -30,6 +30,25 @@
  struct drm_syncobj_cb;
+enum drm_syncobj_type {
+	DRM_SYNCOBJ_TYPE_NORMAL,
+	DRM_SYNCOBJ_TYPE_TIMELINE
+};
+
+struct drm_syncobj_timeline {
+	wait_queue_head_t	wq;
+	u64 timeline_context;
+	/**
+	 * @timeline: syncobj timeline
+	 */
+	u64 timeline;
+	u64 signal_point;
+
+
+	struct rb_root wait_pt_tree;
+	struct list_head signal_pt_list;
+};
+
  /**
   * struct drm_syncobj - sync object.
   *
Quoted 8 lines

include/drm/drm_syncobj.h struct drm_syncobj { · line 47

Quoted 8 lines
include/drm/drm_syncobj.h +0 -4
@@ -41,19 +60,19 @@ struct drm_syncobj {
  	 */
  	struct kref refcount;
  	/**
-	 * @fence:
-	 * NULL or a pointer to the fence bound to this object.
-	 *
-	 * This field should not be used directly. Use

drm_syncobj_fence_get()

Quoted 13 lines
Quoted 13 lines
include/drm/drm_syncobj.h +7 -3
-	 * and drm_syncobj_replace_fence() instead.
+	 * @type: indicate syncobj type
  	 */
-	struct dma_fence __rcu *fence;
+	enum drm_syncobj_type type;
  	/**
-	 * @cb_list: List of callbacks to call when the &fence gets replaced.
+	 * @syncobj_timeline: timeline
  	 */
+	struct drm_syncobj_timeline syncobj_timeline;
+	/**
+         * @cb_list: List of callbacks to call when the &fence gets replaced.
+         */
Quoted 8 lines

include/drm/drm_syncobj.h typedef void (*drm_syncobj_func_t)(struct

Quoted 8 lines
include/drm/drm_syncobj.h +1 -1
  	struct list_head cb_list;
  	/**
-	 * @lock: Protects &cb_list and write-locks &fence.
+	 * @lock: Protects syncobj list and write-locks &fence.
  	 */
  	spinlock_t lock;
  	/**
@@ -68,7 +87,7 @@ typedef void (*drm_syncobj_func_t)(struct

drm_syncobj *syncobj,

Quoted 9 lines
Quoted 9 lines
include/drm/drm_syncobj.h +1 -1
  /**
   * struct drm_syncobj_cb - callback for drm_syncobj_add_callback
   * @node: used by drm_syncob_add_callback to append this struct to
- *	  &drm_syncobj.cb_list
+ *       &drm_syncobj.cb_list
   * @func: drm_syncobj_func_t to call
   *
   * This struct will be initialized by drm_syncobj_add_callback,

additional @@ -106,29 +125,6 @@ drm_syncobj_put(struct drm_syncobj

*obj)

Quoted 22 lines
Quoted 22 lines

kref_put(&obj->refcount, drm_syncobj_free);
}
-/**
- * drm_syncobj_fence_get - get a reference to a fence in a sync object
- * @syncobj: sync object.
- *
- * This acquires additional reference to &drm_syncobj.fence contained in @obj,
- * if not NULL. It is illegal to call this without already holding a reference.
- * No locks required.
- *
- * Returns:
- * Either the fence of @obj or NULL if there's none.
- */

include/drm/drm_syncobj.h +0 -7
-static inline struct dma_fence *
-drm_syncobj_fence_get(struct drm_syncobj *syncobj) -{
-	struct dma_fence *fence;
-
-	rcu_read_lock();
-	fence = dma_fence_get_rcu_safe(&syncobj->fence);
-	rcu_read_unlock();
Quoted 8 lines
Quoted 8 lines
include/drm/drm_syncobj.h +0 -4
-
-	return fence;
-}
-
  struct drm_syncobj *drm_syncobj_find(struct drm_file *file_private,
  				     u32 handle);
  void drm_syncobj_replace_fence(struct drm_syncobj *syncobj, u64

point, @@ -142,5 +138,7 @@ int drm_syncobj_create(struct

drm_syncobj **out_syncobj, uint32_t flags,

Quoted 9 lines
Quoted 9 lines

int drm_syncobj_get_handle(struct drm_file *file_private,
struct drm_syncobj *syncobj, u32 *handle);
int drm_syncobj_get_fd(struct drm_syncobj *syncobj, int *p_fd);

include/drm/drm_syncobj.h +2 -0
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+			     u64 flags, struct dma_fence **fence);
  #endif
include/uapi/drm/drm.h index renamed from include/uapi/drm/drm.h +0 -0

300f336633f2..cebdb2541eb7 100644
--- a/include/uapi/drm/drm.h

Quoted 9 lines

include/uapi/drm/drm.h index struct drm_prime_handle { · line 722

Quoted 8 lines

+++ b/include/uapi/drm/drm.h

include/uapi/drm/drm.h index renamed from include/uapi/drm/drm.h +1 -0
@@ -717,6 +717,7 @@ struct drm_prime_handle {
  struct drm_syncobj_create {
  	__u32 handle;
  #define DRM_SYNCOBJ_CREATE_SIGNALED (1 << 0)
+#define DRM_SYNCOBJ_CREATE_TYPE_TIMELINE (1 << 1)
  	__u32 flags;
  };
Quoted 4 lines

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

#10 lore raw
In reply to Chunming Zhou

Am 14.09.2018 um 12:37 schrieb Chunming Zhou:

Quoted 93 lines

This patch is for VK_KHR_timeline_semaphore extension, semaphore is called syncobj in kernel side:
This extension introduces a new type of syncobj that has an integer payload identifying a point in a timeline. Such timeline syncobjs support the
following operations:
* CPU query - A host operation that allows querying the payload of the timeline syncobj.
* CPU wait - A host operation that allows a blocking wait for a
timeline syncobj to reach a specified value.
* Device wait - A device operation that allows waiting for a
timeline syncobj to reach a specified value.
* Device signal - A device operation that allows advancing the
timeline syncobj to a specified value.

Since it's a timeline, that means the front time point(PT) always is signaled before the late PT. a. signal PT design:
Signal PT fence N depends on PT[N-1] fence and signal opertion fence, when PT[N] fence is signaled, the timeline will increase to value of PT[N].
b. wait PT design:
Wait PT fence is signaled by reaching timeline point value, when timeline is increasing, will compare wait PTs value with new timeline value, if PT value is lower than timeline value, then wait PT will be signaled, otherwise keep in list. syncobj wait operation can wait on any point of timeline, so need a RB tree to order them. And wait PT could ahead of signal PT, we need a sumission fence to perform that.

v2:
1. remove unused DRM_SYNCOBJ_CREATE_TYPE_NORMAL. (Christian)
2. move unexposed denitions to .c file. (Daniel Vetter)
3. split up the change to drm_syncobj_find_fence() in a separate patch. (Christian)
4. split up the change to drm_syncobj_replace_fence() in a separate patch.
5. drop the submission_fence implementation and instead use wait_event() for that. (Christian)
6. WARN_ON(point != 0) for NORMAL type syncobj case. (Daniel Vetter)

v3:
1. replace normal syncobj with timeline implemenation. (Vetter and Christian)
a. normal syncobj signal op will create a signal PT to tail of signal pt list.
b. normal syncobj wait op will create a wait pt with last signal point, and this wait PT is only signaled by related signal point PT.
2. many bug fix and clean up
3. stub fence moving is moved to other patch.

v4:
1. fix RB tree loop with while(node=rb_first(...)). (Christian)
2. fix syncobj lifecycle. (Christian)
3. only enable_signaling when there is wait_pt. (Christian)
4. fix timeline path issues.
5. write a timeline test in libdrm

v5: (Christian)
1. semaphore is called syncobj in kernel side.
2. don't need 'timeline' characters in some function name.
3. keep syncobj cb

normal syncobj is tested by ./deqp-vk -n dEQP-VK*semaphore* timeline syncobj is tested by ./amdgpu_test -s 9

Cc Christian Konig <[email protected]>
Cc Dave Airlie <[email protected]>
Cc Daniel Rakos <[email protected]>
Cc Daniel Vetter <[email protected]>
---
  drivers/gpu/drm/drm_syncobj.c              | 294 ++++++++++++++++++---
  drivers/gpu/drm/i915/i915_gem_execbuffer.c |   4 +-
  include/drm/drm_syncobj.h                  |  62 +++--
  include/uapi/drm/drm.h                     |   1 +
  4 files changed, 292 insertions(+), 69 deletions(-)
drivers/gpu/drm/drm_syncobj.c +8 -0
index e9ce623d049e..e78d076f2703 100644
@@ -56,6 +56,9 @@
  #include "drm_internal.h"
  #include <drm/drm_syncobj.h>
 
+/* merge normal syncobj to timeline syncobj, the point interval is 1 */
+#define DRM_SYNCOBJ_NORMAL_POINT 1
+
  struct drm_syncobj_stub_fence {
  	struct dma_fence base;
  	spinlock_t lock;
@@ -82,6 +85,11 @@ static const struct dma_fence_ops drm_syncobj_stub_fence_ops = {
  	.release = drm_syncobj_stub_fence_release,
  };
 
+struct drm_syncobj_signal_pt {
+	struct dma_fence_array *base;
+	u64    value;
+	struct list_head list;
+};
 
  /**
   * drm_syncobj_find - lookup and reference a sync object.

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj, · line 136

Quoted 6 lines
drivers/gpu/drm/drm_syncobj.c +1 -1
@@ -124,7 +132,7 @@ static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
  {
  	int ret;
 
-	*fence = drm_syncobj_fence_get(syncobj);
+	ret = drm_syncobj_search_fence(syncobj, 0, 0, fence);
  	if (*fence)

Don't we need to check ret here instead?

Quoted 4 lines
drivers/gpu/drm/drm_syncobj.c +0 -0
  		return 1;
 
@@ -133,10 +141,10 @@ static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
  	 * have the lock, try one more time just to be sure we don't add a

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj, · line 147

Quoted 10 lines
drivers/gpu/drm/drm_syncobj.c +4 -4
  	 * callback when a fence has already been set.
  	 */
-	if (syncobj->fence) {
-		*fence = dma_fence_get(rcu_dereference_protected(syncobj->fence,
-								 lockdep_is_held(&syncobj->lock)));
-		ret = 1;
+	if (fence) {
+		drm_syncobj_search_fence(syncobj, 0, 0, fence);
+		if (*fence)
+			ret = 1;

That doesn't looks correct to me, drm_syncobj_search_fence() would try to grab the lock once more.

That should probably be something like checking the signal_pt list and if it is not empty any more drop the lock and retry.

Quoted 176 lines
drivers/gpu/drm/drm_syncobj.c +151 -11
  	} else {
  		*fence = NULL;
  		drm_syncobj_add_callback_locked(syncobj, cb, func);
@@ -164,6 +172,151 @@ void drm_syncobj_remove_callback(struct drm_syncobj *syncobj,
  	spin_unlock(&syncobj->lock);
  }
 
+static void drm_syncobj_timeline_init(struct drm_syncobj *syncobj,
+				      struct drm_syncobj_timeline *syncobj_timeline)
+{
+	spin_lock(&syncobj->lock);
+	syncobj_timeline->timeline_context = dma_fence_context_alloc(1);
+	syncobj_timeline->timeline = 0;
+	syncobj_timeline->signal_point = 0;
+	init_waitqueue_head(&syncobj_timeline->wq);
+
+	INIT_LIST_HEAD(&syncobj_timeline->signal_pt_list);
+	spin_unlock(&syncobj->lock);
+}
+
+static void drm_syncobj_timeline_fini(struct drm_syncobj *syncobj,
+				      struct drm_syncobj_timeline *syncobj_timeline)
+{
+	struct drm_syncobj_signal_pt *signal_pt = NULL, *tmp;
+
+	spin_lock(&syncobj->lock);
+	list_for_each_entry_safe(signal_pt, tmp,
+				 &syncobj_timeline->signal_pt_list, list) {
+		list_del(&signal_pt->list);
+		dma_fence_put(&signal_pt->base->base);
+		kfree(signal_pt);
+	}
+	spin_unlock(&syncobj->lock);
+}
+
+static struct dma_fence
+*drm_syncobj_find_signal_pt_for_point(struct drm_syncobj *syncobj,
+				      uint64_t point)
+{
+	struct drm_syncobj_timeline *timeline = &syncobj->syncobj_timeline;
+	struct drm_syncobj_signal_pt *signal_pt;
+
+	if ((syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) &&
+	    (point <= timeline->timeline)) {
+		struct drm_syncobj_stub_fence *fence =
+			kzalloc(sizeof(struct drm_syncobj_stub_fence),
+				GFP_KERNEL);
+
+		if (!fence)
+			return NULL;
+		spin_lock_init(&fence->lock);
+		dma_fence_init(&fence->base,
+			       &drm_syncobj_stub_fence_ops,
+			       &fence->lock,
+			       syncobj->syncobj_timeline.timeline_context,
+			       point);
+
+		dma_fence_signal(&fence->base);
+		return &fence->base;
+	}
+
+	list_for_each_entry(signal_pt, &timeline->signal_pt_list, list) {
+		if (point > signal_pt->value)
+			continue;
+		if ((syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) &&
+		    (point != signal_pt->value))
+			continue;
+		return dma_fence_get(&signal_pt->base->base);
+	}
+	return NULL;
+}
+
+static int drm_syncobj_create_signal_pt(struct drm_syncobj *syncobj,
+					struct dma_fence *fence,
+					u64 point)
+{
+	struct drm_syncobj_signal_pt *signal_pt =
+		kzalloc(sizeof(struct drm_syncobj_signal_pt), GFP_KERNEL);
+	struct drm_syncobj_signal_pt *tail_pt;
+	struct dma_fence **fences;
+	int num_fences = 0;
+	int ret = 0, i;
+
+	if (!signal_pt)
+		return -ENOMEM;
+	if (syncobj->syncobj_timeline.signal_point >= point) {
+		DRM_WARN("A later signal is ready!");
+		goto out;
+	}
+	if (!fence)
+		goto out;
+
+	fences = kmalloc_array(sizeof(void *), 2, GFP_KERNEL);
+	if (!fences)
+		goto out;
+	fences[num_fences++] = dma_fence_get(fence);
+	/* timeline syncobj must take this dependency */
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		spin_lock(&syncobj->lock);
+		if (!list_empty(&syncobj->syncobj_timeline.signal_pt_list)) {
+			tail_pt = list_last_entry(&syncobj->syncobj_timeline.signal_pt_list,
+						  struct drm_syncobj_signal_pt, list);
+			fences[num_fences++] = dma_fence_get(&tail_pt->base->base);
+		}
+		spin_unlock(&syncobj->lock);
+	}
+	signal_pt->base = dma_fence_array_create(num_fences, fences,
+						 syncobj->syncobj_timeline.timeline_context,
+						 point, false);
+	if (!signal_pt->base)
+		goto fail;
+
+	spin_lock(&syncobj->lock);
+	signal_pt->value = point;
+	INIT_LIST_HEAD(&signal_pt->list);
+	list_add_tail(&signal_pt->list, &syncobj->syncobj_timeline.signal_pt_list);
+	syncobj->syncobj_timeline.signal_point = point;
+	spin_unlock(&syncobj->lock);
+	wake_up_all(&syncobj->syncobj_timeline.wq);
+
+	return 0;
+fail:
+	for (i = 0; i < num_fences; i++)
+		dma_fence_put(fences[i]);
+	kfree(fences);
+out:
+	kfree(signal_pt);
+	return ret;
+}
+
+static void drm_syncobj_garbage_collection(struct drm_syncobj *syncobj)
+{
+	struct drm_syncobj_timeline *timeline = &syncobj->syncobj_timeline;
+	struct drm_syncobj_signal_pt *signal_pt, *tmp;
+
+	spin_lock(&syncobj->lock);
+	list_for_each_entry_safe(signal_pt, tmp,
+				 &timeline->signal_pt_list, list) {
+		if (dma_fence_is_signaled(&signal_pt->base->base)) {
+			timeline->timeline = signal_pt->value;
+			list_del(&signal_pt->list);
+			dma_fence_put(&signal_pt->base->base);
+			kfree(signal_pt);
+		} else {
+			/*signal_pt is in order in list, from small to big, so
+			 * the later must not be signal either */
+			break;
+		}
+	}
+
+	spin_unlock(&syncobj->lock);
+}
  /**
   * drm_syncobj_replace_fence - replace fence in a sync object.
   * @syncobj: Sync object to replace fence in
@@ -176,28 +329,37 @@ void drm_syncobj_replace_fence(struct drm_syncobj *syncobj,
  			       u64 point,
  			       struct dma_fence *fence)
  {
-	struct dma_fence *old_fence;
-	struct drm_syncobj_cb *cur, *tmp;
-
-	if (fence)
-		dma_fence_get(fence);
-
-	spin_lock(&syncobj->lock);
-
-	old_fence = rcu_dereference_protected(syncobj->fence,
-					      lockdep_is_held(&syncobj->lock));
-	rcu_assign_pointer(syncobj->fence, fence);
+	drm_syncobj_garbage_collection(syncobj);
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		if (fence)
+			drm_syncobj_create_signal_pt(syncobj, fence, point);
+	} else if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+		u64 pt_value;

drivers/gpu/drm/drm_syncobj.c void drm_syncobj_replace_fence(struct drm_syncobj *syncobj, · line 347

Quoted 10 lines
drivers/gpu/drm/drm_syncobj.c +10 -0
+
+		if (!fence) {
+			drm_syncobj_timeline_fini(syncobj,
+						  &syncobj->syncobj_timeline);
+			drm_syncobj_timeline_init(syncobj,
+						  &syncobj->syncobj_timeline);
+			return;
+		}
+		pt_value = syncobj->syncobj_timeline.signal_point +
+			DRM_SYNCOBJ_NORMAL_POINT;

Looks like this could be simplified. If we don't have a timeline syncobj we just finish and reinitialize it.

Adding the fence then becomes common for both code paths.

Quoted 27 lines
drivers/gpu/drm/drm_syncobj.c +9 -5
+		drm_syncobj_create_signal_pt(syncobj, fence, pt_value);
+	} else {
+		DRM_ERROR("the syncobj type isn't support\n");
+		return;
+	}
+	if (fence) {
+		struct drm_syncobj_cb *cur, *tmp;
 
-	if (fence != old_fence) {
+		spin_lock(&syncobj->lock);
  		list_for_each_entry_safe(cur, tmp, &syncobj->cb_list, node) {
  			list_del_init(&cur->node);
  			cur->func(syncobj, cur);
  		}
+		spin_unlock(&syncobj->lock);
  	}
-
-	spin_unlock(&syncobj->lock);
-
-	dma_fence_put(old_fence);
  }
  EXPORT_SYMBOL(drm_syncobj_replace_fence);
 
@@ -220,6 +382,25 @@ static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)
  	return 0;
  }
 

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj) · line 394

Quoted 10 lines
drivers/gpu/drm/drm_syncobj.c +10 -0
+static struct dma_fence *
+drm_syncobj_point_get(struct drm_syncobj *syncobj, u64 point, u64 flags)
+{
+	struct dma_fence *fence;
+	int ret = 0;
+
+	if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
+		ret = wait_event_timeout(syncobj->syncobj_timeline.wq,
+					point <= syncobj->syncobj_timeline.signal_point,
+					msecs_to_jiffies(10000)); /* wait 10s */

Either don't use a timeout or provide the timeout explicitely, but don't hard code it here.

Additional to that we probably need an incorruptible wait.

Quoted 33 lines
drivers/gpu/drm/drm_syncobj.c +15 -5
+		if (ret <= 0)
+			return NULL;
+	}
+	spin_lock(&syncobj->lock);
+	fence = drm_syncobj_find_signal_pt_for_point(syncobj, point);
+	spin_unlock(&syncobj->lock);
+	return fence;
+}
+
  /**
   * drm_syncobj_find_fence - lookup and reference the fence in a sync object
   * @file_private: drm file private pointer
@@ -234,20 +415,46 @@ static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)
   * contains a reference to the fence, which must be released by calling
   * dma_fence_put().
   */
-int drm_syncobj_find_fence(struct drm_file *file_private,
-			   u32 handle, u64 point,
-			   struct dma_fence **fence)
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+			     u64 flags, struct dma_fence **fence)
  {
-	struct drm_syncobj *syncobj = drm_syncobj_find(file_private, handle);
  	int ret = 0;
 
  	if (!syncobj)
  		return -ENOENT;
 
-	*fence = drm_syncobj_fence_get(syncobj);
+	drm_syncobj_garbage_collection(syncobj);
+	if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+		/*NORMAL syncobj always wait on last pt */
+		u64 tail_pt_value = syncobj->syncobj_timeline.signal_point;

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj) · line 439

Quoted 10 lines
drivers/gpu/drm/drm_syncobj.c +10 -0
+
+		if (tail_pt_value == 0)
+			tail_pt_value += DRM_SYNCOBJ_NORMAL_POINT;
+		/* NORMAL syncobj doesn't care point value */
+		WARN_ON(point != 0);
+		*fence = drm_syncobj_point_get(syncobj, tail_pt_value,
+							flags);
+	} else if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+		*fence = drm_syncobj_point_get(syncobj, point,
+							flags);

Finding the fence is common for normal and timeline fence, only figuring out the sequence number is different.

So that can probably be simplified further here.

Quoted 75 lines
drivers/gpu/drm/drm_syncobj.c +27 -6
+	} else {
+		DRM_ERROR("Don't support this type syncobj\n");
+		*fence = NULL;
+	}
  	if (!*fence) {
  		ret = -EINVAL;
  	}
+	return ret;
+}
+EXPORT_SYMBOL(drm_syncobj_search_fence);
+int drm_syncobj_find_fence(struct drm_file *file_private,
+			   u32 handle, u64 point,
+			   struct dma_fence **fence) {
+	struct drm_syncobj *syncobj = drm_syncobj_find(file_private, handle);
+
+	int ret = drm_syncobj_search_fence(syncobj, point,
+					DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
+					fence);
  	drm_syncobj_put(syncobj);
  	return ret;
  }
@@ -264,7 +471,7 @@ void drm_syncobj_free(struct kref *kref)
  	struct drm_syncobj *syncobj = container_of(kref,
  						   struct drm_syncobj,
  						   refcount);
-	drm_syncobj_replace_fence(syncobj, 0, NULL);
+	drm_syncobj_timeline_fini(syncobj, &syncobj->syncobj_timeline);
  	kfree(syncobj);
  }
  EXPORT_SYMBOL(drm_syncobj_free);
@@ -294,6 +501,11 @@ int drm_syncobj_create(struct drm_syncobj **out_syncobj, uint32_t flags,
  	kref_init(&syncobj->refcount);
  	INIT_LIST_HEAD(&syncobj->cb_list);
  	spin_lock_init(&syncobj->lock);
+	if (flags & DRM_SYNCOBJ_CREATE_TYPE_TIMELINE)
+		syncobj->type = DRM_SYNCOBJ_TYPE_TIMELINE;
+	else
+		syncobj->type = DRM_SYNCOBJ_TYPE_NORMAL;
+	drm_syncobj_timeline_init(syncobj, &syncobj->syncobj_timeline);
 
  	if (flags & DRM_SYNCOBJ_CREATE_SIGNALED) {
  		ret = drm_syncobj_assign_null_handle(syncobj);
@@ -576,7 +788,8 @@ drm_syncobj_create_ioctl(struct drm_device *dev, void *data,
  		return -ENODEV;
 
  	/* no valid flags yet */
-	if (args->flags & ~DRM_SYNCOBJ_CREATE_SIGNALED)
+	if (args->flags & ~(DRM_SYNCOBJ_CREATE_SIGNALED |
+			    DRM_SYNCOBJ_CREATE_TYPE_TIMELINE))
  		return -EINVAL;
 
  	return drm_syncobj_create_as_handle(file_private,
@@ -669,9 +882,8 @@ static void syncobj_wait_syncobj_func(struct drm_syncobj *syncobj,
  	struct syncobj_wait_entry *wait =
  		container_of(cb, struct syncobj_wait_entry, syncobj_cb);
 
-	/* This happens inside the syncobj lock */
-	wait->fence = dma_fence_get(rcu_dereference_protected(syncobj->fence,
-							      lockdep_is_held(&syncobj->lock)));
+	drm_syncobj_search_fence(syncobj, 0, 0, &wait->fence);
+
  	wake_up_process(wait->task);
  }
 
@@ -698,7 +910,8 @@ static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs,
  	signaled_count = 0;
  	for (i = 0; i < count; ++i) {
  		entries[i].task = current;
-		entries[i].fence = drm_syncobj_fence_get(syncobjs[i]);
+		ret = drm_syncobj_search_fence(syncobjs[i], 0, 0,
+					       &entries[i].fence);
  		if (!entries[i].fence) {
  			if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
  				continue;

drivers/gpu/drm/drm_syncobj.c drm_syncobj_reset_ioctl(struct drm_device *dev, void *data, · line 1188

Quoted 9 lines
drivers/gpu/drm/drm_syncobj.c +3 -3
@@ -970,12 +1183,21 @@ drm_syncobj_reset_ioctl(struct drm_device *dev, void *data,
  	if (ret < 0)
  		return ret;
 
-	for (i = 0; i < args->count_handles; i++)
-		drm_syncobj_replace_fence(syncobjs[i], 0, NULL);
-
+	for (i = 0; i < args->count_handles; i++) {
+		if (syncobjs[i]->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+			DRM_ERROR("timeline syncobj cannot reset!\n");

Why not? I mean that should still work or do I miss anything?

Quoted 21 lines
drivers/gpu/drm/drm_syncobj.c +10 -1
+			ret = -EINVAL;
+			goto out;
+		}
+		drm_syncobj_timeline_fini(syncobjs[i],
+					  &syncobjs[i]->syncobj_timeline);
+		drm_syncobj_timeline_init(syncobjs[i],
+					  &syncobjs[i]->syncobj_timeline);
+	}
+out:
  	drm_syncobj_array_free(syncobjs, args->count_handles);
 
-	return 0;
+	return ret;
  }
 
  int
drivers/gpu/drm/i915/i915_gem_execbuffer.c +0 -0
index 0a8d2d64f380..579e91a5858b 100644

drivers/gpu/drm/i915/i915_gem_execbuffer.c await_fence_array(struct i915_execbuffer *eb, · line 2141

Quoted 6 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +2 -1
@@ -2137,7 +2137,9 @@ await_fence_array(struct i915_execbuffer *eb,
  		if (!(flags & I915_EXEC_FENCE_WAIT))
  			continue;
 
-		fence = drm_syncobj_fence_get(syncobj);
+		drm_syncobj_search_fence(syncobj, 0,
+					 DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,

Not sure if we can make DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT the default here.

Maybe better if drivers explicitly need to ask for it?

Quoted 9 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +1 -0
+					 &fence);
  		if (!fence)
  			return -EINVAL;
 
include/drm/drm_syncobj.h +0 -0
index 425432b85a87..9535521e6623 100644

include/drm/drm_syncobj.h line 34

Quoted 5 lines
include/drm/drm_syncobj.h +2 -0
@@ -30,6 +30,25 @@
 
  struct drm_syncobj_cb;
 
+enum drm_syncobj_type {
+	DRM_SYNCOBJ_TYPE_NORMAL,

Does anybody have a better suggestion for _TYPE_NORMAL? That sounds like timeline would be special.

Quoted 4 lines
include/drm/drm_syncobj.h +4 -0
+	DRM_SYNCOBJ_TYPE_TIMELINE
+};
+
+struct drm_syncobj_timeline {

include/drm/drm_syncobj.h line 48

Quoted 10 lines
include/drm/drm_syncobj.h +10 -0
+	wait_queue_head_t	wq;
+	u64 timeline_context;
+	/**
+	 * @timeline: syncobj timeline
+	 */
+	u64 timeline;
+	u64 signal_point;
+
+
+	struct rb_root wait_pt_tree;

Not used any more?

Quoted 14 lines
include/drm/drm_syncobj.h +3 -4
+	struct list_head signal_pt_list;
+};
+
  /**
   * struct drm_syncobj - sync object.
   *
@@ -41,19 +60,19 @@ struct drm_syncobj {
  	 */
  	struct kref refcount;
  	/**
-	 * @fence:
-	 * NULL or a pointer to the fence bound to this object.
-	 *
-	 * This field should not be used directly. Use drm_syncobj_fence_get()

include/drm/drm_syncobj.h struct drm_syncobj { · line 69

Quoted 10 lines
include/drm/drm_syncobj.h +4 -3
-	 * and drm_syncobj_replace_fence() instead.
+	 * @type: indicate syncobj type
  	 */
-	struct dma_fence __rcu *fence;
+	enum drm_syncobj_type type;
  	/**
-	 * @cb_list: List of callbacks to call when the &fence gets replaced.
+	 * @syncobj_timeline: timeline
  	 */
+	struct drm_syncobj_timeline syncobj_timeline;

Looks like that can now be embedded into drm_syncobj.

Regards,
Christian.

Quoted 62 lines
include/drm/drm_syncobj.h +7 -25
+	/**
+         * @cb_list: List of callbacks to call when the &fence gets replaced.
+         */
  	struct list_head cb_list;
  	/**
-	 * @lock: Protects &cb_list and write-locks &fence.
+	 * @lock: Protects syncobj list and write-locks &fence.
  	 */
  	spinlock_t lock;
  	/**
@@ -68,7 +87,7 @@ typedef void (*drm_syncobj_func_t)(struct drm_syncobj *syncobj,
  /**
   * struct drm_syncobj_cb - callback for drm_syncobj_add_callback
   * @node: used by drm_syncob_add_callback to append this struct to
- *	  &drm_syncobj.cb_list
+ *       &drm_syncobj.cb_list
   * @func: drm_syncobj_func_t to call
   *
   * This struct will be initialized by drm_syncobj_add_callback, additional
@@ -106,29 +125,6 @@ drm_syncobj_put(struct drm_syncobj *obj)
  	kref_put(&obj->refcount, drm_syncobj_free);
  }
 
-/**
- * drm_syncobj_fence_get - get a reference to a fence in a sync object
- * @syncobj: sync object.
- *
- * This acquires additional reference to &drm_syncobj.fence contained in @obj,
- * if not NULL. It is illegal to call this without already holding a reference.
- * No locks required.
- *
- * Returns:
- * Either the fence of @obj or NULL if there's none.
- */
-static inline struct dma_fence *
-drm_syncobj_fence_get(struct drm_syncobj *syncobj)
-{
-	struct dma_fence *fence;
-
-	rcu_read_lock();
-	fence = dma_fence_get_rcu_safe(&syncobj->fence);
-	rcu_read_unlock();
-
-	return fence;
-}
-
  struct drm_syncobj *drm_syncobj_find(struct drm_file *file_private,
  				     u32 handle);
  void drm_syncobj_replace_fence(struct drm_syncobj *syncobj, u64 point,
@@ -142,5 +138,7 @@ int drm_syncobj_create(struct drm_syncobj **out_syncobj, uint32_t flags,
  int drm_syncobj_get_handle(struct drm_file *file_private,
  			   struct drm_syncobj *syncobj, u32 *handle);
  int drm_syncobj_get_fd(struct drm_syncobj *syncobj, int *p_fd);
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+			     u64 flags, struct dma_fence **fence);
 
  #endif
include/uapi/drm/drm.h +0 -0
index 300f336633f2..cebdb2541eb7 100644

include/uapi/drm/drm.h struct drm_prime_handle { · line 723

Quoted 7 lines
include/uapi/drm/drm.h +1 -0
@@ -717,6 +717,7 @@ struct drm_prime_handle {
  struct drm_syncobj_create {
  	__u32 handle;
  #define DRM_SYNCOBJ_CREATE_SIGNALED (1 << 0)
+#define DRM_SYNCOBJ_CREATE_TYPE_TIMELINE (1 << 1)
  	__u32 flags;
  };
 

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

A message that is not in the archive (sent off-list or not yet received): a5d2d299-4360-4267-046e-40b550239524-5C7GfCeVMHo@public.gmane.org

#12 lore raw
Quoted 92 lines · Christian König

Am 14.09.2018 um 12:37 schrieb Chunming Zhou:

Quoted 91 lines

This patch is for VK_KHR_timeline_semaphore extension, semaphore is called syncobj in kernel side:
This extension introduces a new type of syncobj that has an integer payload identifying a point in a timeline. Such timeline syncobjs support the
following operations:
    * CPU query - A host operation that allows querying the payload of the timeline syncobj.
    * CPU wait - A host operation that allows a blocking wait for a
      timeline syncobj to reach a specified value.
    * Device wait - A device operation that allows waiting for a
      timeline syncobj to reach a specified value.
    * Device signal - A device operation that allows advancing the
      timeline syncobj to a specified value.

Since it's a timeline, that means the front time point(PT) always is signaled before the late PT. a. signal PT design:
Signal PT fence N depends on PT[N-1] fence and signal opertion fence, when PT[N] fence is signaled, the timeline will increase to value of PT[N].
b. wait PT design:
Wait PT fence is signaled by reaching timeline point value, when timeline is increasing, will compare wait PTs value with new timeline value, if PT value is lower than timeline value, then wait PT will be signaled, otherwise keep in list. syncobj wait operation can wait on any point of timeline, so need a RB tree to order them. And wait PT could ahead of signal PT, we need a sumission fence to perform that.

v2:
1. remove unused DRM_SYNCOBJ_CREATE_TYPE_NORMAL. (Christian)
2. move unexposed denitions to .c file. (Daniel Vetter)
3. split up the change to drm_syncobj_find_fence() in a separate patch. (Christian)
4. split up the change to drm_syncobj_replace_fence() in a separate patch.
5. drop the submission_fence implementation and instead use wait_event() for that. (Christian)
6. WARN_ON(point != 0) for NORMAL type syncobj case. (Daniel Vetter)

v3:
1. replace normal syncobj with timeline implemenation. (Vetter and Christian)
     a. normal syncobj signal op will create a signal PT to tail of signal pt list.
     b. normal syncobj wait op will create a wait pt with last signal point, and this wait PT is only signaled by related signal point PT.
2. many bug fix and clean up
3. stub fence moving is moved to other patch.

v4:
1. fix RB tree loop with while(node=rb_first(...)). (Christian)
2. fix syncobj lifecycle. (Christian)
3. only enable_signaling when there is wait_pt. (Christian)
4. fix timeline path issues.
5. write a timeline test in libdrm

v5: (Christian)
1. semaphore is called syncobj in kernel side.
2. don't need 'timeline' characters in some function name.
3. keep syncobj cb

normal syncobj is tested by ./deqp-vk -n dEQP-VK*semaphore* timeline syncobj is tested by ./amdgpu_test -s 9

Cc Christian Konig <[email protected]>
Cc Dave Airlie <[email protected]>
Cc Daniel Rakos <[email protected]>
Cc Daniel Vetter <[email protected]>

---
  drivers/gpu/drm/drm_syncobj.c              | 294 ++++++++++++++++++--- drivers/gpu/drm/i915/i915_gem_execbuffer.c |   4 +- include/drm/drm_syncobj.h                  |  62 +++-- include/uapi/drm/drm.h                     |   1 +

  4 files changed, 292 insertions(+), 69 deletions(-)
drivers/gpu/drm/drm_syncobj.c +0 -0
index e9ce623d049e..e78d076f2703 100644
@@ -56,6 +56,9 @@
   #include "drm_internal.h"
   #include <drm/drm_syncobj.h>

  +/* merge normal syncobj to timeline syncobj, the point interval is 1 */

drivers/gpu/drm/drm_syncobj.c +2 -0
+#define DRM_SYNCOBJ_NORMAL_POINT 1
+

  struct drm_syncobj_stub_fence {
      struct dma_fence base;
      spinlock_t lock;

drivers/gpu/drm/drm_syncobj.c +4 -0
@@ -82,6 +85,11 @@ static const struct dma_fence_ops 
 drm_syncobj_stub_fence_ops = {
       .release = drm_syncobj_stub_fence_release,
   };
   +struct drm_syncobj_signal_pt {
+    struct dma_fence_array *base;
+    u64    value;
+    struct list_head list;
+};
     /**
    * drm_syncobj_find - lookup and reference a sync object.
@@ -124,7 +132,7 @@ static int 
Quoted 8 lines

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj, · line 136

Quoted 6 lines
drivers/gpu/drm/drm_syncobj.c +1 -0
 drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
   {
       int ret;
   -    *fence = drm_syncobj_fence_get(syncobj);
+    ret = drm_syncobj_search_fence(syncobj, 0, 0, fence);
       if (*fence)

Don't we need to check ret here instead?

both ok, if you like check ret, will update it in v6.

Quoted 16 lines
Quoted 7 lines
drivers/gpu/drm/drm_syncobj.c +0 -0
           return 1;

  @@ -133,10 +141,10 @@ static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
       * have the lock, try one more time just to be sure we don't add a
       * callback when a fence has already been set.
       */
-    if (syncobj->fence) {

Quoted 8 lines

-        *fence = dma_fence_get(rcu_dereference_protected(syncobj->fence,
- lockdep_is_held(&syncobj->lock)));
-        ret = 1;
+    if (fence) {
+        drm_syncobj_search_fence(syncobj, 0, 0, fence); +        if (*fence)
+            ret = 1;

Quoted 3 lines

That doesn't looks correct to me, drm_syncobj_search_fence() would try to grab the lock once more.

That should probably be something like checking the signal_pt list and if it is not empty any more drop the lock and retry.

Good catch, will change that.

Quoted 233 lines
Quoted 180 lines

      } else {
          *fence = NULL; drm_syncobj_add_callback_locked(syncobj, cb, func);

drivers/gpu/drm/drm_syncobj.c +144 -0
@@ -164,6 +172,151 @@ void drm_syncobj_remove_callback(struct 
 drm_syncobj *syncobj,
       spin_unlock(&syncobj->lock);
   }
   +static void drm_syncobj_timeline_init(struct drm_syncobj *syncobj,
+                      struct drm_syncobj_timeline *syncobj_timeline)
+{
+    spin_lock(&syncobj->lock);
+    syncobj_timeline->timeline_context = dma_fence_context_alloc(1);
+    syncobj_timeline->timeline = 0;
+    syncobj_timeline->signal_point = 0;
+    init_waitqueue_head(&syncobj_timeline->wq);
+
+    INIT_LIST_HEAD(&syncobj_timeline->signal_pt_list);
+    spin_unlock(&syncobj->lock);
+}
+
+static void drm_syncobj_timeline_fini(struct drm_syncobj *syncobj,
+                      struct drm_syncobj_timeline *syncobj_timeline)
+{
+    struct drm_syncobj_signal_pt *signal_pt = NULL, *tmp;
+
+    spin_lock(&syncobj->lock);
+    list_for_each_entry_safe(signal_pt, tmp,
+                 &syncobj_timeline->signal_pt_list, list) {
+        list_del(&signal_pt->list);
+        dma_fence_put(&signal_pt->base->base);
+        kfree(signal_pt);
+    }
+    spin_unlock(&syncobj->lock);
+}
+
+static struct dma_fence
+*drm_syncobj_find_signal_pt_for_point(struct drm_syncobj *syncobj,
+                      uint64_t point)
+{
+    struct drm_syncobj_timeline *timeline = &syncobj->syncobj_timeline;
+    struct drm_syncobj_signal_pt *signal_pt;
+
+    if ((syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) &&
+        (point <= timeline->timeline)) {
+        struct drm_syncobj_stub_fence *fence =
+            kzalloc(sizeof(struct drm_syncobj_stub_fence),
+                GFP_KERNEL);
+
+        if (!fence)
+            return NULL;
+        spin_lock_init(&fence->lock);
+        dma_fence_init(&fence->base,
+                   &drm_syncobj_stub_fence_ops,
+                   &fence->lock,
+ syncobj->syncobj_timeline.timeline_context,
+                   point);
+
+        dma_fence_signal(&fence->base);
+        return &fence->base;
+    }
+
+    list_for_each_entry(signal_pt, &timeline->signal_pt_list, list) {
+        if (point > signal_pt->value)
+            continue;
+        if ((syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) &&
+            (point != signal_pt->value))
+            continue;
+        return dma_fence_get(&signal_pt->base->base);
+    }
+    return NULL;
+}
+
+static int drm_syncobj_create_signal_pt(struct drm_syncobj *syncobj,
+                    struct dma_fence *fence,
+                    u64 point)
+{
+    struct drm_syncobj_signal_pt *signal_pt =
+        kzalloc(sizeof(struct drm_syncobj_signal_pt), GFP_KERNEL);
+    struct drm_syncobj_signal_pt *tail_pt;
+    struct dma_fence **fences;
+    int num_fences = 0;
+    int ret = 0, i;
+
+    if (!signal_pt)
+        return -ENOMEM;
+    if (syncobj->syncobj_timeline.signal_point >= point) {
+        DRM_WARN("A later signal is ready!");
+        goto out;
+    }
+    if (!fence)
+        goto out;
+
+    fences = kmalloc_array(sizeof(void *), 2, GFP_KERNEL);
+    if (!fences)
+        goto out;
+    fences[num_fences++] = dma_fence_get(fence);
+    /* timeline syncobj must take this dependency */
+    if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+        spin_lock(&syncobj->lock);
+        if (!list_empty(&syncobj->syncobj_timeline.signal_pt_list)) {
+            tail_pt = 
 list_last_entry(&syncobj->syncobj_timeline.signal_pt_list,
+                          struct drm_syncobj_signal_pt, list);
+            fences[num_fences++] = dma_fence_get(&tail_pt->base->base);
+        }
+        spin_unlock(&syncobj->lock);
+    }
+    signal_pt->base = dma_fence_array_create(num_fences, fences,
+ syncobj->syncobj_timeline.timeline_context,
+                         point, false);
+    if (!signal_pt->base)
+        goto fail;
+
+    spin_lock(&syncobj->lock);
+    signal_pt->value = point;
+    INIT_LIST_HEAD(&signal_pt->list);
+    list_add_tail(&signal_pt->list, 
 &syncobj->syncobj_timeline.signal_pt_list);
+    syncobj->syncobj_timeline.signal_point = point;
+    spin_unlock(&syncobj->lock);
+    wake_up_all(&syncobj->syncobj_timeline.wq);
+
+    return 0;
+fail:
+    for (i = 0; i < num_fences; i++)
+        dma_fence_put(fences[i]);
+    kfree(fences);
+out:
+    kfree(signal_pt);
+    return ret;
+}
+
+static void drm_syncobj_garbage_collection(struct drm_syncobj *syncobj)
+{
+    struct drm_syncobj_timeline *timeline = &syncobj->syncobj_timeline;
+    struct drm_syncobj_signal_pt *signal_pt, *tmp;
+
+    spin_lock(&syncobj->lock);
+    list_for_each_entry_safe(signal_pt, tmp,
+                 &timeline->signal_pt_list, list) {
+        if (dma_fence_is_signaled(&signal_pt->base->base)) {
+            timeline->timeline = signal_pt->value;
+            list_del(&signal_pt->list);
+            dma_fence_put(&signal_pt->base->base);
+            kfree(signal_pt);
+        } else {
+            /*signal_pt is in order in list, from small to big, so
+             * the later must not be signal either */
+            break;
+        }
+    }
+
+    spin_unlock(&syncobj->lock);
+}

  /**
   * drm_syncobj_replace_fence - replace fence in a sync object.
   * @syncobj: Sync object to replace fence in

drivers/gpu/drm/drm_syncobj.c +6 -11
@@ -176,28 +329,37 @@ void drm_syncobj_replace_fence(struct 
 drm_syncobj *syncobj,
                      u64 point,
                      struct dma_fence *fence)
   {
-    struct dma_fence *old_fence;
-    struct drm_syncobj_cb *cur, *tmp;
-
-    if (fence)
-        dma_fence_get(fence);
-
-    spin_lock(&syncobj->lock);
-
-    old_fence = rcu_dereference_protected(syncobj->fence,
- lockdep_is_held(&syncobj->lock));
-    rcu_assign_pointer(syncobj->fence, fence);
+    drm_syncobj_garbage_collection(syncobj);
+    if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+        if (fence)
+            drm_syncobj_create_signal_pt(syncobj, fence, point);
+    } else if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+        u64 pt_value;

drivers/gpu/drm/drm_syncobj.c void drm_syncobj_replace_fence(struct · line 348

Quoted 10 lines
drivers/gpu/drm/drm_syncobj.c +10 -0
+
+        if (!fence) {
+            drm_syncobj_timeline_fini(syncobj,
+                          &syncobj->syncobj_timeline);
+            drm_syncobj_timeline_init(syncobj,
+                          &syncobj->syncobj_timeline);
+            return;
+        }
+        pt_value = syncobj->syncobj_timeline.signal_point +
+            DRM_SYNCOBJ_NORMAL_POINT;

Looks like this could be simplified. If we don't have a timeline syncobj we just finish and reinitialize it.

Adding the fence then becomes common for both code paths.

Quoted 27 lines
drivers/gpu/drm/drm_syncobj.c +9 -4
+ drm_syncobj_create_signal_pt(syncobj, fence, pt_value);
+    } else {
+        DRM_ERROR("the syncobj type isn't support\n");
+        return;
+    }
+    if (fence) {
+        struct drm_syncobj_cb *cur, *tmp;
   -    if (fence != old_fence) {
+        spin_lock(&syncobj->lock);
           list_for_each_entry_safe(cur, tmp, &syncobj->cb_list, node) {
               list_del_init(&cur->node);
               cur->func(syncobj, cur);
           }
+        spin_unlock(&syncobj->lock);
       }
-
-    spin_unlock(&syncobj->lock);
-
-    dma_fence_put(old_fence);
   }
   EXPORT_SYMBOL(drm_syncobj_replace_fence);

  @@ -220,6 +382,25 @@ static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj) return 0;
  }
  +static struct dma_fence *
+drm_syncobj_point_get(struct drm_syncobj *syncobj, u64 point, u64
flags)

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj) · line 394

Quoted 8 lines
drivers/gpu/drm/drm_syncobj.c +8 -0
+{
+    struct dma_fence *fence;
+    int ret = 0;
+
+    if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
+        ret = wait_event_timeout(syncobj->syncobj_timeline.wq,
+                    point <= syncobj->syncobj_timeline.signal_point,
+                    msecs_to_jiffies(10000)); /* wait 10s */

Either don't use a timeout or provide the timeout explicitely, but don't hard code it here.

Quoted 1 line

Additional to that we probably need an incorruptible wait.

Initially, I used interruptible wait, I found it sometimes often return -ERESTARTSYS without waiting any second when I wrote unit test. Any ideas for that? still need to interruptible wait?
if no interruptible, we have to use a timeout.

Quoted 47 lines
Quoted 34 lines

+        if (ret <= 0)
+            return NULL;
+    }
+    spin_lock(&syncobj->lock);
+    fence = drm_syncobj_find_signal_pt_for_point(syncobj, point); +    spin_unlock(&syncobj->lock);
+    return fence;

drivers/gpu/drm/drm_syncobj.c +2 -0
+}
+

  /**
   * drm_syncobj_find_fence - lookup and reference the fence in a sync object
   * @file_private: drm file private pointer

drivers/gpu/drm/drm_syncobj.c +6 -4
@@ -234,20 +415,46 @@ static int 
 drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)
    * contains a reference to the fence, which must be released by calling
    * dma_fence_put().
    */
-int drm_syncobj_find_fence(struct drm_file *file_private,
-               u32 handle, u64 point,
-               struct dma_fence **fence)
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+                 u64 flags, struct dma_fence **fence)
   {
-    struct drm_syncobj *syncobj = drm_syncobj_find(file_private, 
 handle);
       int ret = 0;
         if (!syncobj)
           return -ENOENT;
   -    *fence = drm_syncobj_fence_get(syncobj);
+    drm_syncobj_garbage_collection(syncobj);
+    if (syncobj->type == DRM_SYNCOBJ_TYPE_NORMAL) {
+        /*NORMAL syncobj always wait on last pt */
+        u64 tail_pt_value = syncobj->syncobj_timeline.signal_point;

drivers/gpu/drm/drm_syncobj.c static int · line 440

Quoted 10 lines
drivers/gpu/drm/drm_syncobj.c +10 -0
+
+        if (tail_pt_value == 0)
+            tail_pt_value += DRM_SYNCOBJ_NORMAL_POINT;
+        /* NORMAL syncobj doesn't care point value */
+        WARN_ON(point != 0);
+        *fence = drm_syncobj_point_get(syncobj, tail_pt_value,
+                            flags);
+    } else if (syncobj->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+        *fence = drm_syncobj_point_get(syncobj, point,
+                            flags);

Finding the fence is common for normal and timeline fence, only figuring out the sequence number is different.

Quoted 1 line

So that can probably be simplified further here.

will do.

Quoted 79 lines
Quoted 79 lines
drivers/gpu/drm/drm_syncobj.c +15 -0
+    } else {
+        DRM_ERROR("Don't support this type syncobj\n");
+        *fence = NULL;
+    }
       if (!*fence) {
           ret = -EINVAL;
       }
+    return ret;
+}
+EXPORT_SYMBOL(drm_syncobj_search_fence);
+int drm_syncobj_find_fence(struct drm_file *file_private,
+               u32 handle, u64 point,
+               struct dma_fence **fence) {
+    struct drm_syncobj *syncobj = drm_syncobj_find(file_private, 
 handle);
+
+    int ret = drm_syncobj_search_fence(syncobj, point,
+                    DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
+                    fence);
       drm_syncobj_put(syncobj);

      return ret;
  }

drivers/gpu/drm/drm_syncobj.c +8 -2
@@ -264,7 +471,7 @@ void drm_syncobj_free(struct kref *kref)
       struct drm_syncobj *syncobj = container_of(kref,
                              struct drm_syncobj,
                              refcount);
-    drm_syncobj_replace_fence(syncobj, 0, NULL);
+    drm_syncobj_timeline_fini(syncobj, &syncobj->syncobj_timeline);
       kfree(syncobj);
   }
   EXPORT_SYMBOL(drm_syncobj_free);
@@ -294,6 +501,11 @@ int drm_syncobj_create(struct drm_syncobj 
 **out_syncobj, uint32_t flags,
       kref_init(&syncobj->refcount);
       INIT_LIST_HEAD(&syncobj->cb_list);
       spin_lock_init(&syncobj->lock);
+    if (flags & DRM_SYNCOBJ_CREATE_TYPE_TIMELINE)
+        syncobj->type = DRM_SYNCOBJ_TYPE_TIMELINE;
+    else
+        syncobj->type = DRM_SYNCOBJ_TYPE_NORMAL;
+    drm_syncobj_timeline_init(syncobj, &syncobj->syncobj_timeline);
         if (flags & DRM_SYNCOBJ_CREATE_SIGNALED) {
           ret = drm_syncobj_assign_null_handle(syncobj);
@@ -576,7 +788,8 @@ drm_syncobj_create_ioctl(struct drm_device *dev, 
 void *data,
           return -ENODEV;
         /* no valid flags yet */
-    if (args->flags & ~DRM_SYNCOBJ_CREATE_SIGNALED)
+    if (args->flags & ~(DRM_SYNCOBJ_CREATE_SIGNALED |
+                DRM_SYNCOBJ_CREATE_TYPE_TIMELINE))
           return -EINVAL;
         return drm_syncobj_create_as_handle(file_private,
@@ -669,9 +882,8 @@ static void syncobj_wait_syncobj_func(struct 
 drm_syncobj *syncobj,

      struct syncobj_wait_entry *wait =
          container_of(cb, struct syncobj_wait_entry, syncobj_cb);
  -    /* This happens inside the syncobj lock */
-    wait->fence =
dma_fence_get(rcu_dereference_protected(syncobj->fence,
- lockdep_is_held(&syncobj->lock)));
+    drm_syncobj_search_fence(syncobj, 0, 0, &wait->fence);
+
      wake_up_process(wait->task);
  }
  @@ -698,7 +910,8 @@ static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs, signaled_count = 0;
      for (i = 0; i < count; ++i) {
          entries[i].task = current;
-        entries[i].fence = drm_syncobj_fence_get(syncobjs[i]);
+        ret = drm_syncobj_search_fence(syncobjs[i], 0, 0,
+                           &entries[i].fence);
          if (!entries[i].fence) {
              if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
                  continue;

drivers/gpu/drm/drm_syncobj.c +0 -0
@@ -970,12 +1183,21 @@ drm_syncobj_reset_ioctl(struct drm_device 
 *dev, void *data,
       if (ret < 0)
           return ret;
Quoted 8 lines

drivers/gpu/drm/drm_syncobj.c drm_syncobj_reset_ioctl(struct drm_device *dev, void *data, · line 1188

Quoted 6 lines
drivers/gpu/drm/drm_syncobj.c +3 -2
   -    for (i = 0; i < args->count_handles; i++)
-        drm_syncobj_replace_fence(syncobjs[i], 0, NULL);
-
+    for (i = 0; i < args->count_handles; i++) {
+        if (syncobjs[i]->type == DRM_SYNCOBJ_TYPE_TIMELINE) {
+            DRM_ERROR("timeline syncobj cannot reset!\n");

Why not? I mean that should still work or do I miss anything?

timeline semaphore spec doesn't require reset interface, it says the timeline value only can be changed by signal operations.

Quoted 28 lines
Quoted 20 lines
drivers/gpu/drm/drm_syncobj.c +10 -0
+            ret = -EINVAL;
+            goto out;
+        }
+        drm_syncobj_timeline_fini(syncobjs[i],
+                      &syncobjs[i]->syncobj_timeline);
+        drm_syncobj_timeline_init(syncobjs[i],
+                      &syncobjs[i]->syncobj_timeline);
+    }
+out:
       drm_syncobj_array_free(syncobjs, args->count_handles);
   -    return 0;
+    return ret;
   }

    int

drivers/gpu/drm/i915/i915_gem_execbuffer.c +0 -0
index 0a8d2d64f380..579e91a5858b 100644

drivers/gpu/drm/i915/i915_gem_execbuffer.c await_fence_array(struct i915_execbuffer *eb, · line 2141

Quoted 5 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +2 -0
@@ -2137,7 +2137,9 @@ await_fence_array(struct i915_execbuffer *eb,
           if (!(flags & I915_EXEC_FENCE_WAIT))
               continue;
   -        fence = drm_syncobj_fence_get(syncobj);
+        drm_syncobj_search_fence(syncobj, 0,
+                     DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,

Not sure if we can make DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT the default here.

Quoted 1 line

Maybe better if drivers explicitly need to ask for it?

if you don't like the flag used in specific driver, I can wrap it to a new function.

Quoted 14 lines

include/drm/drm_syncobj.h line 32

Quoted 12 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +1 -0
+                     &fence);
           if (!fence)
               return -EINVAL;

  diff --git a/include/drm/drm_syncobj.h b/include/drm/drm_syncobj.h index 425432b85a87..9535521e6623 100644

include/drm/drm_syncobj.h +1 -0
@@ -30,6 +30,25 @@
     struct drm_syncobj_cb;
   +enum drm_syncobj_type {
+    DRM_SYNCOBJ_TYPE_NORMAL,

Does anybody have a better suggestion for _TYPE_NORMAL? That sounds like timeline would be special.

How about _TYPE_INDIVIDUAL?

Quoted 17 lines
Quoted 5 lines
include/drm/drm_syncobj.h +4 -0
+    DRM_SYNCOBJ_TYPE_TIMELINE
+};
+
+struct drm_syncobj_timeline {

include/drm/drm_syncobj.h line 48

Quoted 10 lines
include/drm/drm_syncobj.h +10 -0
+    wait_queue_head_t    wq;
+    u64 timeline_context;
+    /**
+     * @timeline: syncobj timeline
+     */
+    u64 timeline;
+    u64 signal_point;
+
+
+    struct rb_root wait_pt_tree;

Not used any more?

will clean up.

Quoted 21 lines
Quoted 21 lines
include/drm/drm_syncobj.h +5 -6
+    struct list_head signal_pt_list;
+};
+
   /**
    * struct drm_syncobj - sync object.
    *
@@ -41,19 +60,19 @@ struct drm_syncobj {
        */
       struct kref refcount;
       /**
-     * @fence:
-     * NULL or a pointer to the fence bound to this object.
-     *
-     * This field should not be used directly. Use 
 drm_syncobj_fence_get()
-     * and drm_syncobj_replace_fence() instead.
+     * @type: indicate syncobj type
        */
-    struct dma_fence __rcu *fence;
+    enum drm_syncobj_type type;
Quoted 8 lines
Quoted 6 lines
include/drm/drm_syncobj.h +0 -1
       /**
-     * @cb_list: List of callbacks to call when the &fence gets 

replaced.
+     * @syncobj_timeline: timeline
       */
+    struct drm_syncobj_timeline syncobj_timeline;

Looks like that can now be embedded into drm_syncobj.

will try.

btw: I encounter a problem after removing advanced wait pt, I debug it two days, but still don't find reason, from log, it's timeount when wait_event_timeout, that means it don't receive signal operation.

"./deqp-vk -n dEQP-VK*semaphore*" will fail on 'dEQP-VK.synchronization.basic.semaphore.chain' case, but can pass if only run that one case like "./deqp-vk -n dEQP-VK.synchronization.basic.semaphore.chain".

Both old and my previous implementation can pass and have no this problem.

Thanks,
David Zhou

Quoted 67 lines

Regards,
Christian.

Quoted 63 lines

+    /**
+         * @cb_list: List of callbacks to call when the &fence gets replaced.
+         */
      struct list_head cb_list;
      /**
-     * @lock: Protects &cb_list and write-locks &fence. +     * @lock: Protects syncobj list and write-locks &fence. */
      spinlock_t lock;
      /**

include/drm/drm_syncobj.h +1 -1
@@ -68,7 +87,7 @@ typedef void (*drm_syncobj_func_t)(struct 
 drm_syncobj *syncobj,
   /**
    * struct drm_syncobj_cb - callback for drm_syncobj_add_callback
    * @node: used by drm_syncob_add_callback to append this struct to
- *      &drm_syncobj.cb_list
+ *       &drm_syncobj.cb_list
    * @func: drm_syncobj_func_t to call
    *

   * This struct will be initialized by drm_syncobj_add_callback, additional

include/drm/drm_syncobj.h +0 -0
@@ -106,29 +125,6 @@ drm_syncobj_put(struct drm_syncobj *obj)
       kref_put(&obj->refcount, drm_syncobj_free);
   }

  -/**
- * drm_syncobj_fence_get - get a reference to a fence in a sync object
- * @syncobj: sync object.
- *
- * This acquires additional reference to &drm_syncobj.fence contained in @obj,
- * if not NULL. It is illegal to call this without already holding a reference.
- * No locks required.
- *
- * Returns:
- * Either the fence of @obj or NULL if there's none.
- */

include/drm/drm_syncobj.h +0 -12
-static inline struct dma_fence *
-drm_syncobj_fence_get(struct drm_syncobj *syncobj)
-{
-    struct dma_fence *fence;
-
-    rcu_read_lock();
-    fence = dma_fence_get_rcu_safe(&syncobj->fence);
-    rcu_read_unlock();
-
-    return fence;
-}
-

  struct drm_syncobj *drm_syncobj_find(struct drm_file *file_private, u32 handle);
  void drm_syncobj_replace_fence(struct drm_syncobj *syncobj, u64 point,

include/drm/drm_syncobj.h +2 -0
@@ -142,5 +138,7 @@ int drm_syncobj_create(struct drm_syncobj 
 **out_syncobj, uint32_t flags,
   int drm_syncobj_get_handle(struct drm_file *file_private,
                  struct drm_syncobj *syncobj, u32 *handle);
   int drm_syncobj_get_fd(struct drm_syncobj *syncobj, int *p_fd);
+int drm_syncobj_search_fence(struct drm_syncobj *syncobj, u64 point,
+                 u64 flags, struct dma_fence **fence);
     #endif
include/uapi/drm/drm.h +0 -1
index 300f336633f2..cebdb2541eb7 100644
--- a/include/uapi/drm/drm.h
Quoted 9 lines

include/uapi/drm/drm.h struct drm_prime_handle { · line 722

Quoted 8 lines

+++ b/include/uapi/drm/drm.h

include/uapi/drm/drm.h +1 -0
@@ -717,6 +717,7 @@ struct drm_prime_handle {
   struct drm_syncobj_create {
       __u32 handle;
   #define DRM_SYNCOBJ_CREATE_SIGNALED (1 << 0)
+#define DRM_SYNCOBJ_CREATE_TYPE_TIMELINE (1 << 1)
       __u32 flags;
   };

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

#13 lore raw

Am 17.09.2018 um 11:33 schrieb zhoucm1:

Quoted 15 lines

[SNIP]

Quoted 14 lines

drivers/gpu/drm/drm_syncobj.c static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj) · line 394

Quoted 11 lines

  +static struct dma_fence * +drm_syncobj_point_get(struct drm_syncobj *syncobj, u64 point, u64 flags)

drivers/gpu/drm/drm_syncobj.c +8 -0
+{
+    struct dma_fence *fence;
+    int ret = 0;
+
+    if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) {
+        ret = wait_event_timeout(syncobj->syncobj_timeline.wq,
+                    point <= syncobj->syncobj_timeline.signal_point,
+                    msecs_to_jiffies(10000)); /* wait 10s */

Either don't use a timeout or provide the timeout explicitely, but don't hard code it here.

Quoted 3 lines
Quoted 1 line

Additional to that we probably need an incorruptible wait.

Initially, I used interruptible wait, I found it sometimes often return -ERESTARTSYS without waiting any second when I wrote unit test. Any ideas for that?

Yeah, that is perfectly normal. The application just received a signal and we need to abort the IOCTL ASAP.

Quoted 1 line

still need to interruptible wait?

Yes, that should certainly be an interruptible wait.

Otherwise an application could block on this forever when somebody forgets to supply a signal point.

Quoted 9 lines
Quoted 8 lines
Quoted 6 lines

  -    for (i = 0; i < args->count_handles; i++)
-        drm_syncobj_replace_fence(syncobjs[i], 0, NULL); -
+    for (i = 0; i < args->count_handles; i++) {
+        if (syncobjs[i]->type == DRM_SYNCOBJ_TYPE_TIMELINE) { +            DRM_ERROR("timeline syncobj cannot reset!\n");

Why not? I mean that should still work or do I miss anything?

timeline semaphore spec doesn't require reset interface, it says the timeline value only can be changed by signal operations.

Yeah, but we don't care about the timeline spec in the kernel.

Question is rather if that still makes sense to support that and as far as I can see it should be trivial to reinitialize the object.

Quoted 27 lines
Quoted 27 lines
Quoted 19 lines

+            ret = -EINVAL;
+            goto out;
+        }
+        drm_syncobj_timeline_fini(syncobjs[i],
+                      &syncobjs[i]->syncobj_timeline); +        drm_syncobj_timeline_init(syncobjs[i],
+                      &syncobjs[i]->syncobj_timeline); +    }
+out:
      drm_syncobj_array_free(syncobjs, args->count_handles); -    return 0;
+    return ret;
  }
    int

drivers/gpu/drm/i915/i915_gem_execbuffer.c +0 -0
index 0a8d2d64f380..579e91a5858b 100644

drivers/gpu/drm/i915/i915_gem_execbuffer.c await_fence_array(struct i915_execbuffer *eb, · line 2141

Quoted 5 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +2 -0
@@ -2137,7 +2137,9 @@ await_fence_array(struct i915_execbuffer *eb,
           if (!(flags & I915_EXEC_FENCE_WAIT))
               continue;
   -        fence = drm_syncobj_fence_get(syncobj);
+        drm_syncobj_search_fence(syncobj, 0,
+                     DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,

Not sure if we can make DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT the default here.

Quoted 2 lines
Quoted 1 line

Maybe better if drivers explicitly need to ask for it?

if you don't like the flag used in specific driver, I can wrap it to a new function.

Yeah, that is probably a good idea.

Quoted 14 lines
Quoted 14 lines

include/drm/drm_syncobj.h line 32

Quoted 12 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +1 -0
+                     &fence);
           if (!fence)
               return -EINVAL;

  diff --git a/include/drm/drm_syncobj.h b/include/drm/drm_syncobj.h index 425432b85a87..9535521e6623 100644

include/drm/drm_syncobj.h +1 -0
@@ -30,6 +30,25 @@
     struct drm_syncobj_cb;
   +enum drm_syncobj_type {
+    DRM_SYNCOBJ_TYPE_NORMAL,

Does anybody have a better suggestion for _TYPE_NORMAL? That sounds like timeline would be special.

Quoted 1 line

How about _TYPE_INDIVIDUAL?

Oh, really good idea.

Quoted 1 line

btw: I encounter a problem after removing advanced wait pt, I debug it two days, but still don't find reason, from log, it's timeount when wait_event_timeout, that means it don't receive signal operation.

Good question. Maybe a race condition? Have you tried to trace both threads to the trace buffer, e.g. use trace_printk?

Quoted 3 lines

"./deqp-vk -n dEQP-VK*semaphore*" will fail on 'dEQP-VK.synchronization.basic.semaphore.chain' case, but can pass if only run that one case like "./deqp-vk -n dEQP-VK.synchronization.basic.semaphore.chain".

Both old and my previous implementation can pass and have no this problem.

Can you reproduce that as well with the unit tests in libdrm?

Christian.
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#14 lore raw
Quoted 10 lines · Christian König
Quoted 6 lines
Quoted 5 lines
Quoted 3 lines

+    for (i = 0; i < args->count_handles; i++) {
+        if (syncobjs[i]->type == DRM_SYNCOBJ_TYPE_TIMELINE) { +            DRM_ERROR("timeline syncobj cannot reset!\n");

Why not? I mean that should still work or do I miss anything?

timeline semaphore spec doesn't require reset interface, it says the timeline value only can be changed by signal operations.

Yeah, but we don't care about the timeline spec in the kernel.

Question is rather if that still makes sense to support that and as far as I can see it should be trivial to reinitialize the object.

Hi Daniel Rakos,

Could you give a comment on this question? Is it necessary to support timeline reset interface?  I only see the timeline value can be changed by signal operations in Spec.

Thanks,
David Zhou

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

A message that is not in the archive (sent off-list or not yet received): 0231250d-0ed0-e99b-84c6-166b47a0b9cc-5C7GfCeVMHo@public.gmane.org

#16 lore raw

Am 19.09.2018 um 05:20 schrieb zhoucm1:

Quoted 9 lines · Christian König
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 3 lines

+    for (i = 0; i < args->count_handles; i++) {
+        if (syncobjs[i]->type == DRM_SYNCOBJ_TYPE_TIMELINE) { +            DRM_ERROR("timeline syncobj cannot reset!\n");

Why not? I mean that should still work or do I miss anything?

Quoted 8 lines
Quoted 5 lines
Quoted 1 line

timeline semaphore spec doesn't require reset interface, it says the timeline value only can be changed by signal operations.

Yeah, but we don't care about the timeline spec in the kernel.

Question is rather if that still makes sense to support that and as far as I can see it should be trivial to reinitialize the object.

Hi Daniel Rakos,

Could you give a comment on this question? Is it necessary to support timeline reset interface?  I only see the timeline value can be changed by signal operations in Spec.

I think the Vulkan spec is rather irrelevant here. We added the reset operation because of drivers which wanted to re-use an existing syncobj.

That is still a valid use case even when Vulkan doesn't have an interface for it.

Christian.

Quoted 10 lines

Thanks,
David Zhou

_______________________________________________
amd-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx