1 files changed, 178 insertions, 330 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
index 70cc30803e3..0d22f66f1c7 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -167,12 +167,10 @@ nouveau_gem_ioctl_new(struct drm_device *dev, void *data,
 
 	ret = drm_gem_handle_create(file_priv, nvbo->gem, &req->info.handle);
 out:
-	mutex_lock(&dev->struct_mutex);
-	drm_gem_object_handle_unreference(nvbo->gem);
-	mutex_unlock(&dev->struct_mutex);
+	drm_gem_object_handle_unreference_unlocked(nvbo->gem);
 
 	if (ret)
-		drm_gem_object_unreference(nvbo->gem);
+		drm_gem_object_unreference_unlocked(nvbo->gem);
 	return ret;
 }
 
@@ -243,6 +241,11 @@ validate_fini_list(struct list_head *list, struct nouveau_fence *fence)
 			nouveau_fence_unref((void *)&prev_fence);
 		}
 
+		if (unlikely(nvbo->validate_mapped)) {
+			ttm_bo_kunmap(&nvbo->kmap);
+			nvbo->validate_mapped = false;
+		}
+
 		list_del(&nvbo->entry);
 		nvbo->reserved_by = NULL;
 		ttm_bo_unreserve(&nvbo->bo);
@@ -302,11 +305,14 @@ retry:
 			if (ret == -EAGAIN)
 				ret = ttm_bo_wait_unreserved(&nvbo->bo, false);
 			drm_gem_object_unreference(gem);
-			if (ret)
+			if (ret) {
+				NV_ERROR(dev, "fail reserve\n");
 				return ret;
+			}
 			goto retry;
 		}
 
+		b->user_priv = (uint64_t)(unsigned long)nvbo;
 		nvbo->reserved_by = file_priv;
 		nvbo->pbbo_index = i;
 		if ((b->valid_domains & NOUVEAU_GEM_DOMAIN_VRAM) &&
@@ -336,8 +342,10 @@ retry:
 			}
 
 			ret = ttm_bo_wait_cpu(&nvbo->bo, false);
-			if (ret)
+			if (ret) {
+				NV_ERROR(dev, "fail wait_cpu\n");
 				return ret;
+			}
 			goto retry;
 		}
 	}
@@ -351,6 +359,7 @@ validate_list(struct nouveau_channel *chan, struct list_head *list,
 {
 	struct drm_nouveau_gem_pushbuf_bo __user *upbbo =
 				(void __force __user *)(uintptr_t)user_pbbo_ptr;
+	struct drm_device *dev = chan->dev;
 	struct nouveau_bo *nvbo;
 	int ret, relocs = 0;
 
@@ -362,39 +371,46 @@ validate_list(struct nouveau_channel *chan, struct list_head *list,
 			spin_lock(&nvbo->bo.lock);
 			ret = ttm_bo_wait(&nvbo->bo, false, false, false);
 			spin_unlock(&nvbo->bo.lock);
-			if (unlikely(ret))
+			if (unlikely(ret)) {
+				NV_ERROR(dev, "fail wait other chan\n");
 				return ret;
+			}
 		}
 
 		ret = nouveau_gem_set_domain(nvbo->gem, b->read_domains,
 					     b->write_domains,
 					     b->valid_domains);
-		if (unlikely(ret))
+		if (unlikely(ret)) {
+			NV_ERROR(dev, "fail set_domain\n");
 			return ret;
+		}
 
 		nvbo->channel = chan;
 		ret = ttm_bo_validate(&nvbo->bo, &nvbo->placement,
 				      false, false);
 		nvbo->channel = NULL;
-		if (unlikely(ret))
+		if (unlikely(ret)) {
+			NV_ERROR(dev, "fail ttm_validate\n");
 			return ret;
+		}
 
-		if (nvbo->bo.offset == b->presumed_offset &&
+		if (nvbo->bo.offset == b->presumed.offset &&
 		    ((nvbo->bo.mem.mem_type == TTM_PL_VRAM &&
-		      b->presumed_domain & NOUVEAU_GEM_DOMAIN_VRAM) ||
+		      b->presumed.domain & NOUVEAU_GEM_DOMAIN_VRAM) ||
 		     (nvbo->bo.mem.mem_type == TTM_PL_TT &&
-		      b->presumed_domain & NOUVEAU_GEM_DOMAIN_GART)))
+		      b->presumed.domain & NOUVEAU_GEM_DOMAIN_GART)))
 			continue;
 
 		if (nvbo->bo.mem.mem_type == TTM_PL_TT)
-			b->presumed_domain = NOUVEAU_GEM_DOMAIN_GART;
+			b->presumed.domain = NOUVEAU_GEM_DOMAIN_GART;
 		else
-			b->presumed_domain = NOUVEAU_GEM_DOMAIN_VRAM;
-		b->presumed_offset = nvbo->bo.offset;
-		b->presumed_ok = 0;
+			b->presumed.domain = NOUVEAU_GEM_DOMAIN_VRAM;
+		b->presumed.offset = nvbo->bo.offset;
+		b->presumed.valid = 0;
 		relocs++;
 
-		if (DRM_COPY_TO_USER(&upbbo[nvbo->pbbo_index], b, sizeof(*b)))
+		if (DRM_COPY_TO_USER(&upbbo[nvbo->pbbo_index].presumed,
+				     &b->presumed, sizeof(b->presumed)))
 			return -EFAULT;
 	}
 
@@ -408,6 +424,7 @@ nouveau_gem_pushbuf_validate(struct nouveau_channel *chan,
 			     uint64_t user_buffers, int nr_buffers,
 			     struct validate_op *op, int *apply_relocs)
 {
+	struct drm_device *dev = chan->dev;
 	int ret, relocs = 0;
 
 	INIT_LIST_HEAD(&op->vram_list);
@@ -418,11 +435,14 @@ nouveau_gem_pushbuf_validate(struct nouveau_channel *chan,
 		return 0;
 
 	ret = validate_init(chan, file_priv, pbbo, nr_buffers, op);
-	if (unlikely(ret))
+	if (unlikely(ret)) {
+		NV_ERROR(dev, "validate_init\n");
 		return ret;
+	}
 
 	ret = validate_list(chan, &op->vram_list, pbbo, user_buffers);
 	if (unlikely(ret < 0)) {
+		NV_ERROR(dev, "validate vram_list\n");
 		validate_fini(op, NULL);
 		return ret;
 	}
@@ -430,6 +450,7 @@ nouveau_gem_pushbuf_validate(struct nouveau_channel *chan,
 
 	ret = validate_list(chan, &op->gart_list, pbbo, user_buffers);
 	if (unlikely(ret < 0)) {
+		NV_ERROR(dev, "validate gart_list\n");
 		validate_fini(op, NULL);
 		return ret;
 	}
@@ -437,6 +458,7 @@ nouveau_gem_pushbuf_validate(struct nouveau_channel *chan,
 
 	ret = validate_list(chan, &op->both_list, pbbo, user_buffers);
 	if (unlikely(ret < 0)) {
+		NV_ERROR(dev, "validate both_list\n");
 		validate_fini(op, NULL);
 		return ret;
 	}
@@ -465,59 +487,82 @@ u_memcpya(uint64_t user, unsigned nmemb, unsigned size)
 }
 
 static int
-nouveau_gem_pushbuf_reloc_apply(struct nouveau_channel *chan, int nr_bo,
-				struct drm_nouveau_gem_pushbuf_bo *bo,
-				unsigned nr_relocs, uint64_t ptr_relocs,
-				unsigned nr_dwords, unsigned first_dword,
-				uint32_t *pushbuf, bool is_iomem)
+nouveau_gem_pushbuf_reloc_apply(struct drm_device *dev,
+				struct drm_nouveau_gem_pushbuf *req,
+				struct drm_nouveau_gem_pushbuf_bo *bo)
 {
 	struct drm_nouveau_gem_pushbuf_reloc *reloc = NULL;
-	struct drm_device *dev = chan->dev;
 	int ret = 0;
 	unsigned i;
 
-	reloc = u_memcpya(ptr_relocs, nr_relocs, sizeof(*reloc));
+	reloc = u_memcpya(req->relocs, req->nr_relocs, sizeof(*reloc));
 	if (IS_ERR(reloc))
 		return PTR_ERR(reloc);
 
-	for (i = 0; i < nr_relocs; i++) {
+	for (i = 0; i < req->nr_relocs; i++) {
 		struct drm_nouveau_gem_pushbuf_reloc *r = &reloc[i];
 		struct drm_nouveau_gem_pushbuf_bo *b;
+		struct nouveau_bo *nvbo;
 		uint32_t data;
 
-		if (r->bo_index >= nr_bo || r->reloc_index < first_dword ||
-		    r->reloc_index >= first_dword + nr_dwords) {
-			NV_ERROR(dev, "Bad relocation %d\n", i);
-			NV_ERROR(dev, "  bo: %d max %d\n", r->bo_index, nr_bo);
-			NV_ERROR(dev, "  id: %d max %d\n", r->reloc_index, nr_dwords);
+		if (unlikely(r->bo_index > req->nr_buffers)) {
+			NV_ERROR(dev, "reloc bo index invalid\n");
 			ret = -EINVAL;
 			break;
 		}
 
 		b = &bo[r->bo_index];
-		if (b->presumed_ok)
+		if (b->presumed.valid)
 			continue;
 
+		if (unlikely(r->reloc_bo_index > req->nr_buffers)) {
+			NV_ERROR(dev, "reloc container bo index invalid\n");
+			ret = -EINVAL;
+			break;
+		}
+		nvbo = (void *)(unsigned long)bo[r->reloc_bo_index].user_priv;
+
+		if (unlikely(r->reloc_bo_offset + 4 >
+			     nvbo->bo.mem.num_pages << PAGE_SHIFT)) {
+			NV_ERROR(dev, "reloc outside of bo\n");
+			ret = -EINVAL;
+			break;
+		}
+
+		if (!nvbo->kmap.virtual) {
+			ret = ttm_bo_kmap(&nvbo->bo, 0, nvbo->bo.mem.num_pages,
+					  &nvbo->kmap);
+			if (ret) {
+				NV_ERROR(dev, "failed kmap for reloc\n");
+				break;
+			}
+			nvbo->validate_mapped = true;
+		}
+
 		if (r->flags & NOUVEAU_GEM_RELOC_LOW)
-			data = b->presumed_offset + r->data;
+			data = b->presumed.offset + r->data;
 		else
 		if (r->flags & NOUVEAU_GEM_RELOC_HIGH)
-			data = (b->presumed_offset + r->data) >> 32;
+			data = (b->presumed.offset + r->data) >> 32;
 		else
 			data = r->data;
 
 		if (r->flags & NOUVEAU_GEM_RELOC_OR) {
-			if (b->presumed_domain == NOUVEAU_GEM_DOMAIN_GART)
+			if (b->presumed.domain == NOUVEAU_GEM_DOMAIN_GART)
 				data |= r->tor;
 			else
 				data |= r->vor;
 		}
 
-		if (is_iomem)
-			iowrite32_native(data, (void __force __iomem *)
-						&pushbuf[r->reloc_index]);
-		else
-			pushbuf[r->reloc_index] = data;
+		spin_lock(&nvbo->bo.lock);
+		ret = ttm_bo_wait(&nvbo->bo, false, false, false);
+		spin_unlock(&nvbo->bo.lock);
+		if (ret) {
+			NV_ERROR(dev, "reloc wait_idle failed: %d\n", ret);
+			break;
+		}
+
+		nouveau_bo_wr32(nvbo, r->reloc_bo_offset >> 2, data);
 	}
 
 	kfree(reloc);
@@ -528,127 +573,50 @@ int
 nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
 			  struct drm_file *file_priv)
 {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct drm_nouveau_gem_pushbuf *req = data;
-	struct drm_nouveau_gem_pushbuf_bo *bo = NULL;
+	struct drm_nouveau_gem_pushbuf_push *push;
+	struct drm_nouveau_gem_pushbuf_bo *bo;
 	struct nouveau_channel *chan;
 	struct validate_op op;
-	struct nouveau_fence* fence = 0;
-	uint32_t *pushbuf = NULL;
-	int ret = 0, do_reloc = 0, i;
+	struct nouveau_fence *fence = 0;
+	int i, j, ret = 0, do_reloc = 0;
 
 	NOUVEAU_CHECK_INITIALISED_WITH_RETURN;
 	NOUVEAU_GET_USER_CHANNEL_WITH_RETURN(req->channel, file_priv, chan);
 
-	if (req->nr_dwords >= chan->dma.max ||
-	    req->nr_buffers > NOUVEAU_GEM_MAX_BUFFERS ||
-	    req->nr_relocs > NOUVEAU_GEM_MAX_RELOCS) {
-		NV_ERROR(dev, "Pushbuf config exceeds limits:\n");
-		NV_ERROR(dev, "  dwords : %d max %d\n", req->nr_dwords,
-			 chan->dma.max - 1);
-		NV_ERROR(dev, "  buffers: %d max %d\n", req->nr_buffers,
-			 NOUVEAU_GEM_MAX_BUFFERS);
-		NV_ERROR(dev, "  relocs : %d max %d\n", req->nr_relocs,
-			 NOUVEAU_GEM_MAX_RELOCS);
-		return -EINVAL;
-	}
-
-	pushbuf = u_memcpya(req->dwords, req->nr_dwords, sizeof(uint32_t));
-	if (IS_ERR(pushbuf))
-		return PTR_ERR(pushbuf);
-
-	bo = u_memcpya(req->buffers, req->nr_buffers, sizeof(*bo));
-	if (IS_ERR(bo)) {
-		kfree(pushbuf);
-		return PTR_ERR(bo);
-	}
-
-	mutex_lock(&dev->struct_mutex);
-
-	/* Validate buffer list */
-	ret = nouveau_gem_pushbuf_validate(chan, file_priv, bo, req->buffers,
-					   req->nr_buffers, &op, &do_reloc);
-	if (ret)
-		goto out;
-
-	/* Apply any relocations that are required */
-	if (do_reloc) {
-		ret = nouveau_gem_pushbuf_reloc_apply(chan, req->nr_buffers,
-						      bo, req->nr_relocs,
-						      req->relocs,
-						      req->nr_dwords, 0,
-						      pushbuf, false);
-		if (ret)
-			goto out;
-	}
-
-	/* Emit push buffer to the hw
-	 */
-	ret = RING_SPACE(chan, req->nr_dwords);
-	if (ret)
-		goto out;
-
-	OUT_RINGp(chan, pushbuf, req->nr_dwords);
+	req->vram_available = dev_priv->fb_aper_free;
+	req->gart_available = dev_priv->gart_info.aper_free;
+	if (unlikely(req->nr_push == 0))
+		goto out_next;
 
-	ret = nouveau_fence_new(chan, &fence, true);
-	if (ret) {
-		NV_ERROR(dev, "error fencing pushbuf: %d\n", ret);
-		WIND_RING(chan);
-		goto out;
+	if (unlikely(req->nr_push > NOUVEAU_GEM_MAX_PUSH)) {
+		NV_ERROR(dev, "pushbuf push count exceeds limit: %d max %d\n",
+			 req->nr_push, NOUVEAU_GEM_MAX_PUSH);
+		return -EINVAL;
 	}
 
-	if (nouveau_gem_pushbuf_sync(chan)) {
-		ret = nouveau_fence_wait(fence, NULL, false, false);
-		if (ret) {
-			for (i = 0; i < req->nr_dwords; i++)
-				NV_ERROR(dev, "0x%08x\n", pushbuf[i]);
-			NV_ERROR(dev, "^^ above push buffer is fail :(\n");
-		}
+	if (unlikely(req->nr_buffers > NOUVEAU_GEM_MAX_BUFFERS)) {
+		NV_ERROR(dev, "pushbuf bo count exceeds limit: %d max %d\n",
+			 req->nr_buffers, NOUVEAU_GEM_MAX_BUFFERS);
+		return -EINVAL;
 	}
 
-out:
-	validate_fini(&op, fence);
-	nouveau_fence_unref((void**)&fence);
-	mutex_unlock(&dev->struct_mutex);
-	kfree(pushbuf);
-	kfree(bo);
-	return ret;
-}
-
-#define PUSHBUF_CAL (dev_priv->card_type >= NV_20)
-
-int
-nouveau_gem_ioctl_pushbuf_call(struct drm_device *dev, void *data,
-			       struct drm_file *file_priv)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct drm_nouveau_gem_pushbuf_call *req = data;
-	struct drm_nouveau_gem_pushbuf_bo *bo = NULL;
-	struct nouveau_channel *chan;
-	struct drm_gem_object *gem;
-	struct nouveau_bo *pbbo;
-	struct validate_op op;
-	struct nouveau_fence* fence = 0;
-	int i, ret = 0, do_reloc = 0;
-
-	NOUVEAU_CHECK_INITIALISED_WITH_RETURN;
-	NOUVEAU_GET_USER_CHANNEL_WITH_RETURN(req->channel, file_priv, chan);
-
-	if (unlikely(req->handle == 0))
-		goto out_next;
-
-	if (req->nr_buffers > NOUVEAU_GEM_MAX_BUFFERS ||
-	    req->nr_relocs > NOUVEAU_GEM_MAX_RELOCS) {
-		NV_ERROR(dev, "Pushbuf config exceeds limits:\n");
-		NV_ERROR(dev, "  buffers: %d max %d\n", req->nr_buffers,
-			 NOUVEAU_GEM_MAX_BUFFERS);
-		NV_ERROR(dev, "  relocs : %d max %d\n", req->nr_relocs,
-			 NOUVEAU_GEM_MAX_RELOCS);
+	if (unlikely(req->nr_relocs > NOUVEAU_GEM_MAX_RELOCS)) {
+		NV_ERROR(dev, "pushbuf reloc count exceeds limit: %d max %d\n",
+			 req->nr_relocs, NOUVEAU_GEM_MAX_RELOCS);
 		return -EINVAL;
 	}
 
+	push = u_memcpya(req->push, req->nr_push, sizeof(*push));
+	if (IS_ERR(push))
+		return PTR_ERR(push);
+
 	bo = u_memcpya(req->buffers, req->nr_buffers, sizeof(*bo));
-	if (IS_ERR(bo))
+	if (IS_ERR(bo)) {
+		kfree(push);
 		return PTR_ERR(bo);
+	}
 
 	mutex_lock(&dev->struct_mutex);
 
@@ -660,122 +628,84 @@ nouveau_gem_ioctl_pushbuf_call(struct drm_device *dev, void *data,
 		goto out;
 	}
 
-	/* Validate DMA push buffer */
-	gem = drm_gem_object_lookup(dev, file_priv, req->handle);
-	if (!gem) {
-		NV_ERROR(dev, "Unknown pb handle 0x%08x\n", req->handle);
-		ret = -EINVAL;
-		goto out;
-	}
-	pbbo = nouveau_gem_object(gem);
-
-	if ((req->offset & 3) || req->nr_dwords < 2 ||
-	    (unsigned long)req->offset > (unsigned long)pbbo->bo.mem.size ||
-	    (unsigned long)req->nr_dwords >
-	     ((unsigned long)(pbbo->bo.mem.size - req->offset ) >> 2)) {
-		NV_ERROR(dev, "pb call misaligned or out of bounds: "
-			      "%d + %d * 4 > %ld\n",
-			 req->offset, req->nr_dwords, pbbo->bo.mem.size);
-		ret = -EINVAL;
-		drm_gem_object_unreference(gem);
-		goto out;
-	}
-
-	ret = ttm_bo_reserve(&pbbo->bo, false, false, true,
-			     chan->fence.sequence);
-	if (ret) {
-		NV_ERROR(dev, "resv pb: %d\n", ret);
-		drm_gem_object_unreference(gem);
-		goto out;
-	}
-
-	nouveau_bo_placement_set(pbbo, 1 << chan->pushbuf_bo->bo.mem.mem_type);
-	ret = ttm_bo_validate(&pbbo->bo, &pbbo->placement, false, false);
-	if (ret) {
-		NV_ERROR(dev, "validate pb: %d\n", ret);
-		ttm_bo_unreserve(&pbbo->bo);
-		drm_gem_object_unreference(gem);
-		goto out;
-	}
-
-	list_add_tail(&pbbo->entry, &op.both_list);
-
-	/* If presumed return address doesn't match, we need to map the
-	 * push buffer and fix it..
-	 */
-	if (!PUSHBUF_CAL) {
-		uint32_t retaddy;
-
-		if (chan->dma.free < 4 + NOUVEAU_DMA_SKIPS) {
-			ret = nouveau_dma_wait(chan, 4 + NOUVEAU_DMA_SKIPS);
-			if (ret) {
-				NV_ERROR(dev, "jmp_space: %d\n", ret);
-				goto out;
-			}
-		}
-
-		retaddy  = chan->pushbuf_base + ((chan->dma.cur + 2) << 2);
-		retaddy |= 0x20000000;
-		if (retaddy != req->suffix0) {
-			req->suffix0 = retaddy;
-			do_reloc = 1;
-		}
-	}
-
 	/* Apply any relocations that are required */
 	if (do_reloc) {
-		void *pbvirt;
-		bool is_iomem;
-		ret = ttm_bo_kmap(&pbbo->bo, 0, pbbo->bo.mem.num_pages,
-				  &pbbo->kmap);
+		ret = nouveau_gem_pushbuf_reloc_apply(dev, req, bo);
 		if (ret) {
-			NV_ERROR(dev, "kmap pb: %d\n", ret);
+			NV_ERROR(dev, "reloc apply: %d\n", ret);
 			goto out;
 		}
+	}
 
-		pbvirt = ttm_kmap_obj_virtual(&pbbo->kmap, &is_iomem);
-		ret = nouveau_gem_pushbuf_reloc_apply(chan, req->nr_buffers, bo,
-						      req->nr_relocs,
-						      req->relocs,
-						      req->nr_dwords,
-						      req->offset / 4,
-						      pbvirt, is_iomem);
-
-		if (!PUSHBUF_CAL) {
-			nouveau_bo_wr32(pbbo,
-					req->offset / 4 + req->nr_dwords - 2,
-					req->suffix0);
-		}
-
-		ttm_bo_kunmap(&pbbo->kmap);
+	if (chan->dma.ib_max) {
+		ret = nouveau_dma_wait(chan, req->nr_push + 1, 6);
 		if (ret) {
-			NV_ERROR(dev, "reloc apply: %d\n", ret);
+			NV_INFO(dev, "nv50cal_space: %d\n", ret);
 			goto out;
 		}
-	}
 
-	if (PUSHBUF_CAL) {
-		ret = RING_SPACE(chan, 2);
+		for (i = 0; i < req->nr_push; i++) {
+			struct nouveau_bo *nvbo = (void *)(unsigned long)
+				bo[push[i].bo_index].user_priv;
+
+			nv50_dma_push(chan, nvbo, push[i].offset,
+				      push[i].length);
+		}
+	} else
+	if (dev_priv->card_type >= NV_20) {
+		ret = RING_SPACE(chan, req->nr_push * 2);
 		if (ret) {
 			NV_ERROR(dev, "cal_space: %d\n", ret);
 			goto out;
 		}
-		OUT_RING(chan, ((pbbo->bo.mem.mm_node->start << PAGE_SHIFT) +
-				  req->offset) | 2);
-		OUT_RING(chan, 0);
+
+		for (i = 0; i < req->nr_push; i++) {
+			struct nouveau_bo *nvbo = (void *)(unsigned long)
+				bo[push[i].bo_index].user_priv;
+			struct drm_mm_node *mem = nvbo->bo.mem.mm_node;
+
+			OUT_RING(chan, ((mem->start << PAGE_SHIFT) +
+					push[i].offset) | 2);
+			OUT_RING(chan, 0);
+		}
 	} else {
-		ret = RING_SPACE(chan, 2 + NOUVEAU_DMA_SKIPS);
+		ret = RING_SPACE(chan, req->nr_push * (2 + NOUVEAU_DMA_SKIPS));
 		if (ret) {
 			NV_ERROR(dev, "jmp_space: %d\n", ret);
 			goto out;
 		}
-		OUT_RING(chan, ((pbbo->bo.mem.mm_node->start << PAGE_SHIFT) +
-				  req->offset) | 0x20000000);
-		OUT_RING(chan, 0);
 
-		/* Space the jumps apart with NOPs. */
-		for (i = 0; i < NOUVEAU_DMA_SKIPS; i++)
+		for (i = 0; i < req->nr_push; i++) {
+			struct nouveau_bo *nvbo = (void *)(unsigned long)
+				bo[push[i].bo_index].user_priv;
+			struct drm_mm_node *mem = nvbo->bo.mem.mm_node;
+			uint32_t cmd;
+
+			cmd = chan->pushbuf_base + ((chan->dma.cur + 2) << 2);
+			cmd |= 0x20000000;
+			if (unlikely(cmd != req->suffix0)) {
+				if (!nvbo->kmap.virtual) {
+					ret = ttm_bo_kmap(&nvbo->bo, 0,
+							  nvbo->bo.mem.
+							  num_pages,
+							  &nvbo->kmap);
+					if (ret) {
+						WIND_RING(chan);
+						goto out;
+					}
+					nvbo->validate_mapped = true;
+				}
+
+				nouveau_bo_wr32(nvbo, (push[i].offset +
+						push[i].length - 8) / 4, cmd);
+			}
+
+			OUT_RING(chan, ((mem->start << PAGE_SHIFT) +
+					push[i].offset) | 0x20000000);
 			OUT_RING(chan, 0);
+			for (j = 0; j < NOUVEAU_DMA_SKIPS; j++)
+				OUT_RING(chan, 0);
+		}
 	}
 
 	ret = nouveau_fence_new(chan, &fence, true);
@@ -790,9 +720,14 @@ out:
 	nouveau_fence_unref((void**)&fence);
 	mutex_unlock(&dev->struct_mutex);
 	kfree(bo);
+	kfree(push);
 
 out_next:
-	if (PUSHBUF_CAL) {
+	if (chan->dma.ib_max) {
+		req->suffix0 = 0x00000000;
+		req->suffix1 = 0x00000000;
+	} else
+	if (dev_priv->card_type >= NV_20) {
 		req->suffix0 = 0x00020000;
 		req->suffix1 = 0x00000000;
 	} else {
@@ -804,19 +739,6 @@ out_next:
 	return ret;
 }
 
-int
-nouveau_gem_ioctl_pushbuf_call2(struct drm_device *dev, void *data,
-				struct drm_file *file_priv)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct drm_nouveau_gem_pushbuf_call *req = data;
-
-	req->vram_available = dev_priv->fb_aper_free;
-	req->gart_available = dev_priv->gart_info.aper_free;
-
-	return nouveau_gem_ioctl_pushbuf_call(dev, data, file_priv);
-}
-
 static inline uint32_t
 domain_to_ttm(struct nouveau_bo *nvbo, uint32_t domain)
 {
@@ -831,74 +753,6 @@ domain_to_ttm(struct nouveau_bo *nvbo, uint32_t domain)
 }
 
 int
-nouveau_gem_ioctl_pin(struct drm_device *dev, void *data,
-		      struct drm_file *file_priv)
-{
-	struct drm_nouveau_gem_pin *req = data;
-	struct drm_gem_object *gem;
-	struct nouveau_bo *nvbo;
-	int ret = 0;
-
-	NOUVEAU_CHECK_INITIALISED_WITH_RETURN;
-
-	if (drm_core_check_feature(dev, DRIVER_MODESET)) {
-		NV_ERROR(dev, "pin only allowed without kernel modesetting\n");
-		return -EINVAL;
-	}
-
-	if (!DRM_SUSER(DRM_CURPROC))
-		return -EPERM;
-
-	gem = drm_gem_object_lookup(dev, file_priv, req->handle);
-	if (!gem)
-		return -EINVAL;
-	nvbo = nouveau_gem_object(gem);
-
-	ret = nouveau_bo_pin(nvbo, domain_to_ttm(nvbo, req->domain));
-	if (ret)
-		goto out;
-
-	req->offset = nvbo->bo.offset;
-	if (nvbo->bo.mem.mem_type == TTM_PL_TT)
-		req->domain = NOUVEAU_GEM_DOMAIN_GART;
-	else
-		req->domain = NOUVEAU_GEM_DOMAIN_VRAM;
-
-out:
-	mutex_lock(&dev->struct_mutex);
-	drm_gem_object_unreference(gem);
-	mutex_unlock(&dev->struct_mutex);
-
-	return ret;
-}
-
-int
-nouveau_gem_ioctl_unpin(struct drm_device *dev, void *data,
-			struct drm_file *file_priv)
-{
-	struct drm_nouveau_gem_pin *req = data;
-	struct drm_gem_object *gem;
-	int ret;
-
-	NOUVEAU_CHECK_INITIALISED_WITH_RETURN;
-
-	if (drm_core_check_feature(dev, DRIVER_MODESET))
-		return -EINVAL;
-
-	gem = drm_gem_object_lookup(dev, file_priv, req->handle);
-	if (!gem)
-		return -EINVAL;
-
-	ret = nouveau_bo_unpin(nouveau_gem_object(gem));
-
-	mutex_lock(&dev->struct_mutex);
-	drm_gem_object_unreference(gem);
-	mutex_unlock(&dev->struct_mutex);
-
-	return ret;
-}
-
-int
 nouveau_gem_ioctl_cpu_prep(struct drm_device *dev, void *data,
 			   struct drm_file *file_priv)
 {
@@ -935,9 +789,7 @@ nouveau_gem_ioctl_cpu_prep(struct drm_device *dev, void *data,
 	}
 
 out:
-	mutex_lock(&dev->struct_mutex);
-	drm_gem_object_unreference(gem);
-	mutex_unlock(&dev->struct_mutex);
+	drm_gem_object_unreference_unlocked(gem);
 	return ret;
 }
 
@@ -965,9 +817,7 @@ nouveau_gem_ioctl_cpu_fini(struct drm_device *dev, void *data,
 	ret = 0;
 
 out:
-	mutex_lock(&dev->struct_mutex);
-	drm_gem_object_unreference(gem);
-	mutex_unlock(&dev->struct_mutex);
+	drm_gem_object_unreference_unlocked(gem);
 	return ret;
 }
 
@@ -986,9 +836,7 @@ nouveau_gem_ioctl_info(struct drm_device *dev, void *data,
 		return -EINVAL;
 
 	ret = nouveau_gem_info(gem, req);
-	mutex_lock(&dev->struct_mutex);
-	drm_gem_object_unreference(gem);
-	mutex_unlock(&dev->struct_mutex);
+	drm_gem_object_unreference_unlocked(gem);
 	return ret;
 }