CHROMIUM: drm/nouveau/bo: fix warning while enabling lock debugging

We hit the warning below when the lock debugging is enabled. The
nouveau_fence_sync holds the RCU read lock and then calls into mutex_lock.
To get rid of this warning, we don't hold the vma_list lock if that's not
necessary.

BUG: sleeping function called from invalid context at kernel/locking/mutex.c:97
in_atomic(): 0, irqs_disabled(): 0, pid: 7064, name: nouveau_pushbuf
Preemption disabled at:[<ffffffc000a962a8>] printk+0x6c/0x78
CPU: 3 PID: 7064 Comm: nouveau_pushbuf Tainted: G     U         3.18.0-00019-gf7ed90de5956 #421
Hardware name: Google Tegra210 Smaug Rev 1,3+ (DT)
Call trace:
[<ffffffc0002073f8>] dump_backtrace+0x0/0x10c
[<ffffffc000207514>] show_stack+0x10/0x1c
[<ffffffc000a96f9c>] dump_stack+0x74/0xb8
[<ffffffc0002449ac>] __might_sleep+0x158/0x168
[<ffffffc000a9cc18>] mutex_lock+0x20/0x48
[<ffffffc0005f1f1c>] nouveau_bo_vma_find+0x24/0x80
[<ffffffc000603c6c>] nv50_dma_push_bo+0x3c/0xa0
[<ffffffc000606afc>] nvc0_fence_sync32+0xec/0x158
[<ffffffc000605f88>] nv84_fence_sync+0x40/0x4c
[<ffffffc000604f64>] nouveau_fence_sync+0xc8/0xfc
[<ffffffc0005f4644>] nouveau_gem_pushbuf_queue_kthread_fn+0x2b0/0x554
[<ffffffc000237790>] kthread+0xdc/0xe8

BUG=chrome-os-partner:54432,chromium:635609
TEST=Boot to UI

Change-Id: I65b693d0f115808486d744aa06bf7df1358cf737
Signed-off-by: Vince Hsu <vince.h@nvidia.com>
Reviewed-on: https://chromium-review.googlesource.com/356610
Commit-Ready: Alexander Potapenko <glider@chromium.org>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Reviewed-by: Alexander Potapenko <glider@chromium.org>

10 files changed, 66 insertions, 37 deletions

diff --git a/drivers/gpu/drm/nouveau/dispnv04/crtc.c b/drivers/gpu/drm/nouveau/dispnv04/crtc.c
index 15b123e9aaa4..14df584c8350 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/crtc.c
+++ b/drivers/gpu/drm/nouveau/dispnv04/crtc.c
@@ -1128,7 +1128,7 @@ nv04_crtc_create(struct drm_device *dev, int crtc_num)
 	drm_mode_crtc_set_gamma_size(&nv_crtc->base, 256);
 
 	ret = nouveau_bo_new(dev, 64*64*4, 0x100, TTM_PL_FLAG_VRAM,
-			     0, 0x0000, NULL, NULL, &nv_crtc->cursor.nvbo);
+			     0, 0x0000, NULL, NULL, &nv_crtc->cursor.nvbo, true);
 	if (!ret) {
 		ret = nouveau_bo_pin(nv_crtc->cursor.nvbo, TTM_PL_FLAG_VRAM, false);
 		if (!ret) {
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 0585a86e484c..3758635204fd 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -154,7 +154,7 @@ nouveau_bo_del_ttm(struct ttm_buffer_object *bo)
 	 * are still in the list at this point are really leaked and can be
 	 * deleted safely.
 	 */
-	mutex_lock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_lock(nvbo);
 	list_for_each_entry(vma, &nvbo->vma_list, head) {
 		DRM_INFO("Cleaning up leaked mapping offset 0x%llx\n",
 			 vma->offset);
@@ -163,7 +163,7 @@ nouveau_bo_del_ttm(struct ttm_buffer_object *bo)
 		nvkm_vm_put(vma);
 		kfree(vma);
 	}
-	mutex_unlock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_unlock(nvbo);
 
 	nv10_bo_put_tile_region(dev, nvbo->tile, NULL);
 	kfree(nvbo);
@@ -207,7 +207,7 @@ int
 nouveau_bo_new(struct drm_device *dev, int size, int align,
 	       uint32_t flags, uint32_t tile_mode, uint32_t tile_flags,
 	       struct sg_table *sg, struct reservation_object *robj,
-	       struct nouveau_bo **pnvbo)
+	       struct nouveau_bo **pnvbo, bool unchanged_vma_list)
 {
 	struct nouveau_drm *drm = nouveau_drm(dev);
 	struct nouveau_bo *nvbo;
@@ -236,6 +236,7 @@ nouveau_bo_new(struct drm_device *dev, int size, int align,
 	INIT_LIST_HEAD(&nvbo->entry);
 	INIT_LIST_HEAD(&nvbo->vma_list);
 	mutex_init(&nvbo->vma_list_lock);
+	nvbo->vma_immutable = unchanged_vma_list;
 	nvbo->tile_mode = tile_mode;
 	nvbo->tile_flags = tile_flags;
 	nvbo->gpu_cacheable = !(flags & TTM_PL_FLAG_UNCACHED);
@@ -601,6 +602,20 @@ nouveau_bo_validate(struct nouveau_bo *nvbo, bool interruptible,
 	return 0;
 }
 
+void
+nouveau_bo_vma_list_lock(struct nouveau_bo *nvbo)
+{
+	if (!nvbo->vma_immutable)
+		mutex_lock(&nvbo->vma_list_lock);
+}
+
+void
+nouveau_bo_vma_list_unlock(struct nouveau_bo *nvbo)
+{
+	if (!nvbo->vma_immutable)
+		mutex_unlock(&nvbo->vma_list_lock);
+}
+
 static inline void *
 _nouveau_bo_mem_index(struct nouveau_bo *nvbo, unsigned index, void *mem, u8 sz)
 {
@@ -1338,7 +1353,7 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem)
 	if (bo->destroy != nouveau_bo_del_ttm)
 		return;
 
-	mutex_lock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_lock(nvbo);
 
 	list_for_each_entry(vma, &nvbo->vma_list, head) {
 		if (new_mem && new_mem->mem_type != TTM_PL_SYSTEM &&
@@ -1350,7 +1365,7 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem)
 		}
 	}
 
-	mutex_unlock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_unlock(nvbo);
 }
 
 static int
@@ -1723,17 +1738,17 @@ nouveau_bo_vma_find(struct nouveau_bo *nvbo, struct nvkm_vm *vm)
 {
 	struct nvkm_vma *vma;
 
-	mutex_lock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_lock(nvbo);
 
 	list_for_each_entry(vma, &nvbo->vma_list, head) {
 		if (vma->implicit &&
 		    (vma->vm == vm)) {
-			mutex_unlock(&nvbo->vma_list_lock);
+			nouveau_bo_vma_list_unlock(nvbo);
 			return vma;
 		}
 	}
 
-	mutex_unlock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_unlock(nvbo);
 
 	return NULL;
 }
@@ -1744,7 +1759,7 @@ nouveau_bo_subvma_find(struct nouveau_bo *nvbo, struct nvkm_vm *vm, u64 offset,
 {
 	struct nvkm_vma *vma;
 
-	mutex_lock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_lock(nvbo);
 
 	/*
 	 * Look for an existing subvma that we can reuse.  The delta and length
@@ -1760,11 +1775,11 @@ nouveau_bo_subvma_find(struct nouveau_bo *nvbo, struct nvkm_vm *vm, u64 offset,
 		    (vma->length == length) &&
 		    (!offset || (vma->offset == offset)) &&
 		    !vma->unmap_pending) {
-			mutex_unlock(&nvbo->vma_list_lock);
+			nouveau_bo_vma_list_unlock(nvbo);
 			return vma;
 		}
 
-	mutex_unlock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_unlock(nvbo);
 
 	return NULL;
 }
@@ -1775,18 +1790,18 @@ nouveau_bo_subvma_find_offset(struct nouveau_bo *nvbo, struct nvkm_vm *vm,
 {
 	struct nvkm_vma *vma;
 
-	mutex_lock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_lock(nvbo);
 
 	list_for_each_entry(vma, &nvbo->vma_list, head)
 		if (!vma->implicit &&
 		    (vma->vm == vm) &&
 		    (vma->offset == offset) &&
 		    !vma->unmap_pending) {
-			mutex_unlock(&nvbo->vma_list_lock);
+			nouveau_bo_vma_list_unlock(nvbo);
 			return vma;
 		}
 
-	mutex_unlock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_unlock(nvbo);
 
 	return NULL;
 }
@@ -1876,9 +1891,9 @@ nouveau_bo_vma_add_offset(struct nouveau_bo *nvbo, struct nvkm_vm *vm,
 			nvkm_vm_map(vma, nvbo->bo.mem.mm_node);
 	}
 
-	mutex_lock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_lock(nvbo);
 	list_add_tail(&vma->head, &nvbo->vma_list);
-	mutex_unlock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_unlock(nvbo);
 	vma->refcount = 1;
 	vma->implicit = true;
 	return 0;
@@ -1898,9 +1913,9 @@ nouveau_bo_vma_del(struct nouveau_bo *nvbo, struct nvkm_vma *vma)
 		if (vma->mapped && nvbo->bo.mem.mem_type != TTM_PL_SYSTEM)
 			nvkm_vm_unmap(vma);
 		nvkm_vm_put(vma);
-		mutex_lock(&nvbo->vma_list_lock);
+		nouveau_bo_vma_list_lock(nvbo);
 		list_del(&vma->head);
-		mutex_unlock(&nvbo->vma_list_lock);
+		nouveau_bo_vma_list_unlock(nvbo);
 	}
 }
 
@@ -1927,9 +1942,10 @@ nouveau_bo_subvma_add(struct nouveau_bo *nvbo, struct nvkm_vm *vm,
 			nvkm_vm_map(vma, nvbo->bo.mem.mm_node);
 	}
 
-	mutex_lock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_lock(nvbo);
+	WARN_ON(nvbo->vma_immutable);
 	list_add_tail(&vma->head, &nvbo->vma_list);
-	mutex_unlock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_unlock(nvbo);
 	vma->refcount = 1;
 	return 0;
 }
@@ -1937,5 +1953,6 @@ nouveau_bo_subvma_add(struct nouveau_bo *nvbo, struct nvkm_vm *vm,
 void
 nouveau_bo_subvma_del(struct nouveau_bo *nvbo, struct nvkm_vma *vma)
 {
+	WARN_ON(nvbo->vma_immutable);
 	nouveau_bo_vma_del(nvbo, vma);
 }
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h
index 8dc5096bc9b7..307c5a848e89 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.h
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.h
@@ -26,6 +26,7 @@ struct nouveau_bo {
 
 	struct list_head vma_list;
 	struct mutex vma_list_lock;
+	bool vma_immutable;
 	unsigned page_shift;
 
 	u32 tile_mode;
@@ -74,7 +75,7 @@ void nouveau_bo_move_init(struct nouveau_drm *);
 int  nouveau_bo_new(struct drm_device *, int size, int align, u32 flags,
 		    u32 tile_mode, u32 tile_flags, struct sg_table *sg,
 		    struct reservation_object *robj,
-		    struct nouveau_bo **);
+		    struct nouveau_bo **, bool unchanged_vma_list);
 int  nouveau_bo_pin(struct nouveau_bo *, u32 flags, bool contig);
 int  nouveau_bo_unpin(struct nouveau_bo *);
 int  nouveau_bo_map(struct nouveau_bo *);
@@ -114,6 +115,8 @@ int  nouveau_bo_subvma_add(struct nouveau_bo *, struct nvkm_vm *,
 void nouveau_bo_subvma_del(struct nouveau_bo *, struct nvkm_vma *);
 void nouveau_defer_vm_map(struct nvkm_vma *vma, struct nouveau_bo *nvbo);
 void nouveau_cancel_defer_vm_map(struct nvkm_vma *vma, struct nouveau_bo *nvbo);
+void nouveau_bo_vma_list_lock(struct nouveau_bo *);
+void nouveau_bo_vma_list_unlock(struct nouveau_bo *);
 
 /* TODO: submit equivalent to TTM generic API upstream? */
 static inline void __iomem *
diff --git a/drivers/gpu/drm/nouveau/nouveau_chan.c b/drivers/gpu/drm/nouveau/nouveau_chan.c
index d93438f69000..bdd49985a563 100644
--- a/drivers/gpu/drm/nouveau/nouveau_chan.c
+++ b/drivers/gpu/drm/nouveau/nouveau_chan.c
@@ -201,7 +201,7 @@ nouveau_channel_prep(struct nouveau_drm *drm, struct nvif_device *device,
 		target = TTM_PL_FLAG_VRAM;
 
 	ret = nouveau_bo_new(drm->dev, size, 0, target, 0, 0, NULL, NULL,
-			    &chan->push.buffer);
+			    &chan->push.buffer, true);
 	if (ret == 0) {
 		ret = nouveau_bo_pin(chan->push.buffer, target, false);
 		if (ret == 0)
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
index 519b901b7d86..c3912823e960 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -264,9 +264,10 @@ static void gem_unmap_work(struct work_struct *__work)
 
 	fobj = reservation_object_get_list(resv);
 
-	mutex_lock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_lock(nvbo);
+	WARN_ON(nvbo->vma_immutable);
 	list_del(&vma->head);
-	mutex_unlock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_unlock(nvbo);
 
 	if (fobj && fobj->shared_count > 1)
 		ttm_bo_wait(&nvbo->bo, true, false, false);
@@ -363,7 +364,7 @@ nouveau_gem_object_close(struct drm_gem_object *gem, struct drm_file *file_priv)
 		}
 	}
 
-	mutex_lock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_lock(nvbo);
 	list_for_each_entry(vma, &nvbo->vma_list, head)
 		if (!vma->implicit &&
 		    (vma->vm == cli->vm) &&
@@ -371,7 +372,7 @@ nouveau_gem_object_close(struct drm_gem_object *gem, struct drm_file *file_priv)
 			nouveau_gem_object_unmap(nvbo, vma);
 			drm_gem_object_unreference(&nvbo->gem);
 		}
-	mutex_unlock(&nvbo->vma_list_lock);
+	nouveau_bo_vma_list_unlock(nvbo);
 
 	ttm_bo_unreserve(&nvbo->bo);
 }
@@ -458,7 +459,7 @@ nouveau_gem_new(struct drm_device *dev, int size, int align, uint32_t domain,
 		flags |= TTM_PL_FLAG_UNCACHED;
 
 	ret = nouveau_bo_new(dev, size, align, flags, tile_mode,
-			     tile_flags, NULL, NULL, pnvbo);
+			     tile_flags, NULL, NULL, pnvbo, true);
 	if (ret)
 		return ret;
 	nvbo = *pnvbo;
@@ -526,7 +527,11 @@ static int nouveau_gem_remap(struct nouveau_drm *drm, struct nvkm_vm *vm,
 
 	/* Unmap the old vma. */
 	nouveau_cancel_defer_vm_map(vma, nvbo);
-	nouveau_bo_vma_del(nvbo, vma);
+	/*
+	 * Calling nouveau_bo_subvma_del instead of nouveau_bo_vma_del to
+	 * reflect the fact that we're changing the bo's vma_list.
+	 */
+	nouveau_bo_subvma_del(nvbo, vma);
 
 	/*
 	 * If this offset falls within an address space allocation, then honor
@@ -1920,7 +1925,11 @@ success:
 	vma = nouveau_bo_vma_find(nvbo, cli->vm);
 	if (vma) {
 		nouveau_cancel_defer_vm_map(vma, nvbo);
-		nouveau_bo_vma_del(nvbo, vma);
+		/*
+		 * Calling nouveau_bo_subvma_del instead of nouveau_bo_vma_del to
+		 * reflect the fact that we're changing the bo's vma_list.
+		 */
+		nouveau_bo_subvma_del(nvbo, vma);
 		kfree(vma);
 	}
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_prime.c b/drivers/gpu/drm/nouveau/nouveau_prime.c
index dd32ad6db53d..44d679a7f99e 100644
--- a/drivers/gpu/drm/nouveau/nouveau_prime.c
+++ b/drivers/gpu/drm/nouveau/nouveau_prime.c
@@ -69,7 +69,7 @@ struct drm_gem_object *nouveau_gem_prime_import_sg_table(struct drm_device *dev,
 
 	ww_mutex_lock(&robj->lock, NULL);
 	ret = nouveau_bo_new(dev, attach->dmabuf->size, 0, flags, 0, 0,
-			     sg, robj, &nvbo);
+			     sg, robj, &nvbo, false);
 	ww_mutex_unlock(&robj->lock);
 	if (ret)
 		return ERR_PTR(ret);
diff --git a/drivers/gpu/drm/nouveau/nv17_fence.c b/drivers/gpu/drm/nouveau/nv17_fence.c
index 57860cfa1de5..d6074c321c2e 100644
--- a/drivers/gpu/drm/nouveau/nv17_fence.c
+++ b/drivers/gpu/drm/nouveau/nv17_fence.c
@@ -129,7 +129,7 @@ nv17_fence_create(struct nouveau_drm *drm)
 	spin_lock_init(&priv->lock);
 
 	ret = nouveau_bo_new(drm->dev, 4096, 0x1000, TTM_PL_FLAG_VRAM,
-			     0, 0x0000, NULL, NULL, &priv->bo);
+			     0, 0x0000, NULL, NULL, &priv->bo, true);
 	if (!ret) {
 		ret = nouveau_bo_pin(priv->bo, TTM_PL_FLAG_VRAM, false);
 		if (!ret) {
diff --git a/drivers/gpu/drm/nouveau/nv50_display.c b/drivers/gpu/drm/nouveau/nv50_display.c
index bff804189d27..ac42c1c3c668 100644
--- a/drivers/gpu/drm/nouveau/nv50_display.c
+++ b/drivers/gpu/drm/nouveau/nv50_display.c
@@ -1444,7 +1444,7 @@ nv50_crtc_create(struct drm_device *dev, int index)
 	drm_mode_crtc_set_gamma_size(crtc, 256);
 
 	ret = nouveau_bo_new(dev, 8192, 0x100, TTM_PL_FLAG_VRAM,
-			     0, 0x0000, NULL, NULL, &head->base.lut.nvbo);
+			     0, 0x0000, NULL, NULL, &head->base.lut.nvbo, true);
 	if (!ret) {
 		ret = nouveau_bo_pin(head->base.lut.nvbo, TTM_PL_FLAG_VRAM, true);
 		if (!ret) {
@@ -2494,7 +2494,7 @@ nv50_display_create(struct drm_device *dev)
 
 	/* small shared memory area we use for notifiers and semaphores */
 	ret = nouveau_bo_new(dev, 4096, 0x1000, TTM_PL_FLAG_VRAM,
-			     0, 0x0000, NULL, NULL, &disp->sync);
+			     0, 0x0000, NULL, NULL, &disp->sync, true);
 	if (!ret) {
 		ret = nouveau_bo_pin(disp->sync, TTM_PL_FLAG_VRAM, true);
 		if (!ret) {
diff --git a/drivers/gpu/drm/nouveau/nv50_fence.c b/drivers/gpu/drm/nouveau/nv50_fence.c
index a82d9ea7c6fd..dc34b176f377 100644
--- a/drivers/gpu/drm/nouveau/nv50_fence.c
+++ b/drivers/gpu/drm/nouveau/nv50_fence.c
@@ -100,7 +100,7 @@ nv50_fence_create(struct nouveau_drm *drm)
 	spin_lock_init(&priv->lock);
 
 	ret = nouveau_bo_new(drm->dev, 4096, 0x1000, TTM_PL_FLAG_VRAM,
-			     0, 0x0000, NULL, NULL, &priv->bo);
+			     0, 0x0000, NULL, NULL, &priv->bo, true);
 	if (!ret) {
 		ret = nouveau_bo_pin(priv->bo, TTM_PL_FLAG_VRAM, false);
 		if (!ret) {
diff --git a/drivers/gpu/drm/nouveau/nv84_fence.c b/drivers/gpu/drm/nouveau/nv84_fence.c
index 3e5c131a14de..b15f3e6a7b2b 100644
--- a/drivers/gpu/drm/nouveau/nv84_fence.c
+++ b/drivers/gpu/drm/nouveau/nv84_fence.c
@@ -241,7 +241,7 @@ nv84_fence_create(struct nouveau_drm *drm)
 			  */
 			 TTM_PL_FLAG_TT | TTM_PL_FLAG_UNCACHED;
 	ret = nouveau_bo_new(drm->dev, 16 * priv->base.contexts, 0, domain, 0,
-			     0, NULL, NULL, &priv->bo);
+			     0, NULL, NULL, &priv->bo, true);
 	if (ret == 0) {
 		ret = nouveau_bo_pin(priv->bo, domain, false);
 		if (ret == 0) {
@@ -256,7 +256,7 @@ nv84_fence_create(struct nouveau_drm *drm)
 	if (ret == 0)
 		ret = nouveau_bo_new(drm->dev, 16 * priv->base.contexts, 0,
 				     TTM_PL_FLAG_TT | TTM_PL_FLAG_UNCACHED, 0,
-				     0, NULL, NULL, &priv->bo_gart);
+				     0, NULL, NULL, &priv->bo_gart, true);
 	if (ret == 0) {
 		ret = nouveau_bo_pin(priv->bo_gart, TTM_PL_FLAG_TT, false);
 		if (ret == 0) {