drm/nouveau: delay busy bo vma removal until fence signals

As opposed to an explicit wait.  Allows userspace to not stall waiting
on buffer deletion.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drivers/gpu/drm/nouveau/nouveau_bo.c

@@ -1550,13 +1550,8 @@ void
 nouveau_bo_vma_del(struct nouveau_bo *nvbo, struct nouveau_vma *vma)
 {
 	if (vma->node) {
-		if (nvbo->bo.mem.mem_type != TTM_PL_SYSTEM) {
-			spin_lock(&nvbo->bo.bdev->fence_lock);
-			ttm_bo_wait(&nvbo->bo, false, false, false);
-			spin_unlock(&nvbo->bo.bdev->fence_lock);
+		if (nvbo->bo.mem.mem_type != TTM_PL_SYSTEM)
 			nouveau_vm_unmap(vma);
-		}
-
 		nouveau_vm_put(vma);
 		list_del(&vma->head);
 	}

drivers/gpu/drm/nouveau/nouveau_fence.c

@@ -35,15 +35,34 @@
 
 #include <engine/fifo.h>
 
+struct fence_work {
+	struct work_struct base;
+	struct list_head head;
+	void (*func)(void *);
+	void *data;
+};
+
+static void
+nouveau_fence_signal(struct nouveau_fence *fence)
+{
+	struct fence_work *work, *temp;
+
+	list_for_each_entry_safe(work, temp, &fence->work, head) {
+		schedule_work(&work->base);
+		list_del(&work->head);
+	}
+
+	fence->channel = NULL;
+	list_del(&fence->head);
+}
+
 void
 nouveau_fence_context_del(struct nouveau_fence_chan *fctx)
 {
 	struct nouveau_fence *fence, *fnext;
 	spin_lock(&fctx->lock);
 	list_for_each_entry_safe(fence, fnext, &fctx->pending, head) {
-		fence->channel = NULL;
-		list_del(&fence->head);
-		nouveau_fence_unref(&fence);
+		nouveau_fence_signal(fence);
 	}
 	spin_unlock(&fctx->lock);
 }
@@ -56,6 +75,50 @@ nouveau_fence_context_new(struct nouveau_fence_chan *fctx)
 	spin_lock_init(&fctx->lock);
 }
 
+static void
+nouveau_fence_work_handler(struct work_struct *kwork)
+{
+	struct fence_work *work = container_of(kwork, typeof(*work), base);
+	work->func(work->data);
+	kfree(work);
+}
+
+void
+nouveau_fence_work(struct nouveau_fence *fence,
+		   void (*func)(void *), void *data)
+{
+	struct nouveau_channel *chan = fence->channel;
+	struct nouveau_fence_chan *fctx;
+	struct fence_work *work = NULL;
+
+	if (nouveau_fence_done(fence)) {
+		func(data);
+		return;
+	}
+
+	fctx = chan->fence;
+	work = kmalloc(sizeof(*work), GFP_KERNEL);
+	if (!work) {
+		WARN_ON(nouveau_fence_wait(fence, false, false));
+		func(data);
+		return;
+	}
+
+	spin_lock(&fctx->lock);
+	if (!fence->channel) {
+		spin_unlock(&fctx->lock);
+		kfree(work);
+		func(data);
+		return;
+	}
+
+	INIT_WORK(&work->base, nouveau_fence_work_handler);
+	work->func = func;
+	work->data = data;
+	list_add(&work->head, &fence->work);
+	spin_unlock(&fctx->lock);
+}
+
 static void
 nouveau_fence_update(struct nouveau_channel *chan)
 {
@@ -67,8 +130,7 @@ nouveau_fence_update(struct nouveau_channel *chan)
 		if (fctx->read(chan) < fence->sequence)
 			break;
 
-		fence->channel = NULL;
-		list_del(&fence->head);
+		nouveau_fence_signal(fence);
 		nouveau_fence_unref(&fence);
 	}
 	spin_unlock(&fctx->lock);
@@ -265,6 +327,7 @@ nouveau_fence_new(struct nouveau_channel *chan, bool sysmem,
 	if (!fence)
 		return -ENOMEM;
 
+	INIT_LIST_HEAD(&fence->work);
 	fence->sysmem = sysmem;
 	kref_init(&fence->kref);
 

drivers/gpu/drm/nouveau/nouveau_fence.h

@@ -5,6 +5,7 @@ struct nouveau_drm;
 
 struct nouveau_fence {
 	struct list_head head;
+	struct list_head work;
 	struct kref kref;
 
 	bool sysmem;
@@ -22,6 +23,7 @@ void nouveau_fence_unref(struct nouveau_fence **);
 
 int  nouveau_fence_emit(struct nouveau_fence *, struct nouveau_channel *);
 bool nouveau_fence_done(struct nouveau_fence *);
+void nouveau_fence_work(struct nouveau_fence *, void (*)(void *), void *);
 int  nouveau_fence_wait(struct nouveau_fence *, bool lazy, bool intr);
 int  nouveau_fence_sync(struct nouveau_fence *, struct nouveau_channel *);
 

drivers/gpu/drm/nouveau/nouveau_gem.c

@@ -101,6 +101,41 @@ out:
 	return ret;
 }
 
+static void
+nouveau_gem_object_delete(void *data)
+{
+	struct nouveau_vma *vma = data;
+	nouveau_vm_unmap(vma);
+	nouveau_vm_put(vma);
+	kfree(vma);
+}
+
+static void
+nouveau_gem_object_unmap(struct nouveau_bo *nvbo, struct nouveau_vma *vma)
+{
+	const bool mapped = nvbo->bo.mem.mem_type != TTM_PL_SYSTEM;
+	struct nouveau_fence *fence = NULL;
+
+	list_del(&vma->head);
+
+	if (mapped) {
+		spin_lock(&nvbo->bo.bdev->fence_lock);
+		if (nvbo->bo.sync_obj)
+			fence = nouveau_fence_ref(nvbo->bo.sync_obj);
+		spin_unlock(&nvbo->bo.bdev->fence_lock);
+	}
+
+	if (fence) {
+		nouveau_fence_work(fence, nouveau_gem_object_delete, vma);
+	} else {
+		if (mapped)
+			nouveau_vm_unmap(vma);
+		nouveau_vm_put(vma);
+		kfree(vma);
+	}
+	nouveau_fence_unref(&fence);
+}
+
 void
 nouveau_gem_object_close(struct drm_gem_object *gem, struct drm_file *file_priv)
 {
@@ -118,10 +153,8 @@ nouveau_gem_object_close(struct drm_gem_object *gem, struct drm_file *file_priv)
 
 	vma = nouveau_bo_vma_find(nvbo, cli->base.vm);
 	if (vma) {
-		if (--vma->refcount == 0) {
-			nouveau_bo_vma_del(nvbo, vma);
-			kfree(vma);
-		}
+		if (--vma->refcount == 0)
+			nouveau_gem_object_unmap(nvbo, vma);
 	}
 	ttm_bo_unreserve(&nvbo->bo);
 }