drm/nouveau: new gem pushbuf interface, bump to 0.0.16

This commit breaks the userspace interface, and requires a new libdrm for
nouveau to operate again.

The multiple GEM_PUSHBUF ioctls that were present in 0.0.15 for
compatibility purposes are now gone, and replaced with the new ioctl which
allows for multiple push buffers to be submitted (necessary for hw index
buffers in the nv50 3d driver) and relocations to be applied on any buffer.

A number of other ioctls (CARD_INIT, GEM_PIN, GEM_UNPIN) that were needed
for userspace modesetting have also been removed.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>

7 files changed, 208 insertions, 404 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_channel.c b/drivers/gpu/drm/nouveau/nouveau_channel.c
index ceb83961b16f..6dfb425cbae9 100644
--- a/drivers/gpu/drm/nouveau/nouveau_channel.c
+++ b/drivers/gpu/drm/nouveau/nouveau_channel.c
@@ -385,6 +385,14 @@ nouveau_ioctl_fifo_alloc(struct drm_device *dev, void *data,
 		return ret;
 	init->channel  = chan->id;
 
+	if (chan->dma.ib_max)
+		init->pushbuf_domains = NOUVEAU_GEM_DOMAIN_VRAM |
+					NOUVEAU_GEM_DOMAIN_GART;
+	else if (chan->pushbuf_bo->bo.mem.mem_type == TTM_PL_VRAM)
+		init->pushbuf_domains = NOUVEAU_GEM_DOMAIN_VRAM;
+	else
+		init->pushbuf_domains = NOUVEAU_GEM_DOMAIN_GART;
+
 	init->subchan[0].handle = NvM2MF;
 	if (dev_priv->card_type < NV_50)
 		init->subchan[0].grclass = 0x0039;
@@ -424,7 +432,6 @@ nouveau_ioctl_fifo_free(struct drm_device *dev, void *data,
  ***********************************/
 
 struct drm_ioctl_desc nouveau_ioctls[] = {
-	DRM_IOCTL_DEF(DRM_NOUVEAU_CARD_INIT, nouveau_ioctl_card_init, DRM_AUTH),
 	DRM_IOCTL_DEF(DRM_NOUVEAU_GETPARAM, nouveau_ioctl_getparam, DRM_AUTH),
 	DRM_IOCTL_DEF(DRM_NOUVEAU_SETPARAM, nouveau_ioctl_setparam, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
 	DRM_IOCTL_DEF(DRM_NOUVEAU_CHANNEL_ALLOC, nouveau_ioctl_fifo_alloc, DRM_AUTH),
@@ -434,13 +441,9 @@ struct drm_ioctl_desc nouveau_ioctls[] = {
 	DRM_IOCTL_DEF(DRM_NOUVEAU_GPUOBJ_FREE, nouveau_ioctl_gpuobj_free, DRM_AUTH),
 	DRM_IOCTL_DEF(DRM_NOUVEAU_GEM_NEW, nouveau_gem_ioctl_new, DRM_AUTH),
 	DRM_IOCTL_DEF(DRM_NOUVEAU_GEM_PUSHBUF, nouveau_gem_ioctl_pushbuf, DRM_AUTH),
-	DRM_IOCTL_DEF(DRM_NOUVEAU_GEM_PUSHBUF_CALL, nouveau_gem_ioctl_pushbuf_call, DRM_AUTH),
-	DRM_IOCTL_DEF(DRM_NOUVEAU_GEM_PIN, nouveau_gem_ioctl_pin, DRM_AUTH),
-	DRM_IOCTL_DEF(DRM_NOUVEAU_GEM_UNPIN, nouveau_gem_ioctl_unpin, DRM_AUTH),
 	DRM_IOCTL_DEF(DRM_NOUVEAU_GEM_CPU_PREP, nouveau_gem_ioctl_cpu_prep, DRM_AUTH),
 	DRM_IOCTL_DEF(DRM_NOUVEAU_GEM_CPU_FINI, nouveau_gem_ioctl_cpu_fini, DRM_AUTH),
 	DRM_IOCTL_DEF(DRM_NOUVEAU_GEM_INFO, nouveau_gem_ioctl_info, DRM_AUTH),
-	DRM_IOCTL_DEF(DRM_NOUVEAU_GEM_PUSHBUF_CALL2, nouveau_gem_ioctl_pushbuf_call2, DRM_AUTH),
 };
 
 int nouveau_max_ioctl = DRM_ARRAY_SIZE(nouveau_ioctls);
diff --git a/drivers/gpu/drm/nouveau/nouveau_dma.c b/drivers/gpu/drm/nouveau/nouveau_dma.c
index 679b03c28df1..c8482a108a78 100644
--- a/drivers/gpu/drm/nouveau/nouveau_dma.c
+++ b/drivers/gpu/drm/nouveau/nouveau_dma.c
@@ -179,7 +179,7 @@ READ_GET(struct nouveau_channel *chan, uint32_t *prev_get, uint32_t *timeout)
 
 void
 nv50_dma_push(struct nouveau_channel *chan, struct nouveau_bo *bo,
-	      int delta, int dwords)
+	      int delta, int length)
 {
 	struct nouveau_bo *pb = chan->pushbuf_bo;
 	uint64_t offset = bo->bo.offset + delta;
@@ -187,7 +187,7 @@ nv50_dma_push(struct nouveau_channel *chan, struct nouveau_bo *bo,
 
 	BUG_ON(chan->dma.ib_free < 1);
 	nouveau_bo_wr32(pb, ip++, lower_32_bits(offset));
-	nouveau_bo_wr32(pb, ip++, upper_32_bits(offset) | dwords << 10);
+	nouveau_bo_wr32(pb, ip++, upper_32_bits(offset) | length << 8);
 
 	chan->dma.ib_put = (chan->dma.ib_put + 1) & chan->dma.ib_max;
 	nvchan_wr32(chan, 0x8c, chan->dma.ib_put);
diff --git a/drivers/gpu/drm/nouveau/nouveau_dma.h b/drivers/gpu/drm/nouveau/nouveau_dma.h
index da6e16dafa4d..8b05c15866d5 100644
--- a/drivers/gpu/drm/nouveau/nouveau_dma.h
+++ b/drivers/gpu/drm/nouveau/nouveau_dma.h
@@ -32,7 +32,7 @@
 #endif
 
 void nv50_dma_push(struct nouveau_channel *, struct nouveau_bo *,
-		   int delta, int dwords);
+		   int delta, int length);
 
 /*
  * There's a hw race condition where you can't jump to your PUT offset,
@@ -149,7 +149,7 @@ FIRE_RING(struct nouveau_channel *chan)
 
 	if (chan->dma.ib_max) {
 		nv50_dma_push(chan, chan->pushbuf_bo, chan->dma.put << 2,
-			      chan->dma.cur - chan->dma.put);
+			      (chan->dma.cur - chan->dma.put) << 2);
 	} else {
 		WRITE_PUT(chan->dma.cur);
 	}
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index d221044e0793..a33423622860 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -34,7 +34,7 @@
 
 #define DRIVER_MAJOR		0
 #define DRIVER_MINOR		0
-#define DRIVER_PATCHLEVEL	15
+#define DRIVER_PATCHLEVEL	16
 
 #define NOUVEAU_FAMILY   0x0000FFFF
 #define NOUVEAU_FLAGS    0xFFFF0000
@@ -83,6 +83,7 @@ struct nouveau_bo {
 	struct drm_file *reserved_by;
 	struct list_head entry;
 	int pbbo_index;
+	bool validate_mapped;
 
 	struct nouveau_channel *channel;
 
@@ -704,12 +705,6 @@ extern bool nouveau_wait_until(struct drm_device *, uint64_t timeout,
 			       uint32_t reg, uint32_t mask, uint32_t val);
 extern bool nouveau_wait_for_idle(struct drm_device *);
 extern int  nouveau_card_init(struct drm_device *);
-extern int  nouveau_ioctl_card_init(struct drm_device *, void *data,
-				    struct drm_file *);
-extern int  nouveau_ioctl_suspend(struct drm_device *, void *data,
-				  struct drm_file *);
-extern int  nouveau_ioctl_resume(struct drm_device *, void *data,
-				 struct drm_file *);
 
 /* nouveau_mem.c */
 extern int  nouveau_mem_init_heap(struct mem_block **, uint64_t start,
@@ -1160,16 +1155,6 @@ extern int nouveau_gem_ioctl_new(struct drm_device *, void *,
 				 struct drm_file *);
 extern int nouveau_gem_ioctl_pushbuf(struct drm_device *, void *,
 				     struct drm_file *);
-extern int nouveau_gem_ioctl_pushbuf_call(struct drm_device *, void *,
-					  struct drm_file *);
-extern int nouveau_gem_ioctl_pushbuf_call2(struct drm_device *, void *,
-					   struct drm_file *);
-extern int nouveau_gem_ioctl_pin(struct drm_device *, void *,
-				 struct drm_file *);
-extern int nouveau_gem_ioctl_unpin(struct drm_device *, void *,
-				   struct drm_file *);
-extern int nouveau_gem_ioctl_tile(struct drm_device *, void *,
-				  struct drm_file *);
 extern int nouveau_gem_ioctl_cpu_prep(struct drm_device *, void *,
 				      struct drm_file *);
 extern int nouveau_gem_ioctl_cpu_fini(struct drm_device *, void *,
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
index df72cd847025..fee959c72f40 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -243,6 +243,11 @@ 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 +307,14 @@ 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 +344,10 @@ 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 +361,7 @@ 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 +373,46 @@ 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 +426,7 @@ 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 +437,14 @@ 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 +452,7 @@ 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 +460,7 @@ 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 +489,82 @@ 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);
+		if (ret) {
+			NV_ERROR(dev, "reloc wait_idle failed: %d\n", ret);
+			break;
+		}
+		spin_unlock(&nvbo->bo.lock);
+
+		nouveau_bo_wr32(nvbo, r->reloc_bo_offset >> 2, data);
 	}
 
 	kfree(reloc);
@@ -528,125 +575,50 @@ 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;
-}
-
-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);
 
@@ -658,94 +630,9 @@ 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 (dev_priv->card_type < NV_20) {
-		uint32_t retaddy;
-
-		if (chan->dma.free < 4 + NOUVEAU_DMA_SKIPS) {
-			ret = nouveau_dma_wait(chan, 0, 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);
-		if (ret) {
-			NV_ERROR(dev, "kmap pb: %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 (dev_priv->card_type < NV_20) {
-			nouveau_bo_wr32(pbbo,
-					req->offset / 4 + req->nr_dwords - 2,
-					req->suffix0);
-		}
-
-		ttm_bo_kunmap(&pbbo->kmap);
+		ret = nouveau_gem_pushbuf_reloc_apply(dev, req, bo);
 		if (ret) {
 			NV_ERROR(dev, "reloc apply: %d\n", ret);
 			goto out;
@@ -753,36 +640,74 @@ nouveau_gem_ioctl_pushbuf_call(struct drm_device *dev, void *data,
 	}
 
 	if (chan->dma.ib_max) {
-		ret = nouveau_dma_wait(chan, 2, 6);
+		ret = nouveau_dma_wait(chan, req->nr_push + 1, 6);
 		if (ret) {
 			NV_INFO(dev, "nv50cal_space: %d\n", ret);
 			goto out;
 		}
 
-		nv50_dma_push(chan, pbbo, req->offset, req->nr_dwords);
+		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, 2);
+		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);
@@ -797,6 +722,7 @@ out:
 	nouveau_fence_unref((void**)&fence);
 	mutex_unlock(&dev->struct_mutex);
 	kfree(bo);
+	kfree(push);
 
 out_next:
 	if (chan->dma.ib_max) {
@@ -815,19 +741,6 @@ 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)
 {
@@ -842,74 +755,6 @@ 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)
 {
diff --git a/drivers/gpu/drm/nouveau/nouveau_state.c b/drivers/gpu/drm/nouveau/nouveau_state.c
index ed5ac0b9a0ac..516a8d36cb10 100644
--- a/drivers/gpu/drm/nouveau/nouveau_state.c
+++ b/drivers/gpu/drm/nouveau/nouveau_state.c
@@ -777,13 +777,6 @@ int nouveau_unload(struct drm_device *dev)
 	return 0;
 }
 
-int
-nouveau_ioctl_card_init(struct drm_device *dev, void *data,
-			struct drm_file *file_priv)
-{
-	return nouveau_card_init(dev);
-}
-
 int nouveau_ioctl_getparam(struct drm_device *dev, void *data,
 						struct drm_file *file_priv)
 {
diff --git a/include/drm/nouveau_drm.h b/include/drm/nouveau_drm.h
index f745948b61e4..a6a9f4af5ebd 100644
--- a/include/drm/nouveau_drm.h
+++ b/include/drm/nouveau_drm.h
@@ -25,13 +25,14 @@
 #ifndef __NOUVEAU_DRM_H__
 #define __NOUVEAU_DRM_H__
 
-#define NOUVEAU_DRM_HEADER_PATCHLEVEL 15
+#define NOUVEAU_DRM_HEADER_PATCHLEVEL 16
 
 struct drm_nouveau_channel_alloc {
 	uint32_t     fb_ctxdma_handle;
 	uint32_t     tt_ctxdma_handle;
 
 	int          channel;
+	uint32_t     pushbuf_domains;
 
 	/* Notifier memory */
 	uint32_t     notifier_handle;
@@ -109,68 +110,58 @@ struct drm_nouveau_gem_new {
 	uint32_t align;
 };
 
+#define NOUVEAU_GEM_MAX_BUFFERS 1024
+struct drm_nouveau_gem_pushbuf_bo_presumed {
+	uint32_t valid;
+	uint32_t domain;
+	uint64_t offset;
+};
+
 struct drm_nouveau_gem_pushbuf_bo {
 	uint64_t user_priv;
 	uint32_t handle;
 	uint32_t read_domains;
 	uint32_t write_domains;
 	uint32_t valid_domains;
-	uint32_t presumed_ok;
-	uint32_t presumed_domain;
-	uint64_t presumed_offset;
+	struct drm_nouveau_gem_pushbuf_bo_presumed presumed;
 };
 
 #define NOUVEAU_GEM_RELOC_LOW  (1 << 0)
 #define NOUVEAU_GEM_RELOC_HIGH (1 << 1)
 #define NOUVEAU_GEM_RELOC_OR   (1 << 2)
+#define NOUVEAU_GEM_MAX_RELOCS 1024
 struct drm_nouveau_gem_pushbuf_reloc {
+	uint32_t reloc_bo_index;
+	uint32_t reloc_bo_offset;
 	uint32_t bo_index;
-	uint32_t reloc_index;
 	uint32_t flags;
 	uint32_t data;
 	uint32_t vor;
 	uint32_t tor;
 };
 
-#define NOUVEAU_GEM_MAX_BUFFERS 1024
-#define NOUVEAU_GEM_MAX_RELOCS 1024
+#define NOUVEAU_GEM_MAX_PUSH 512
+struct drm_nouveau_gem_pushbuf_push {
+	uint32_t bo_index;
+	uint32_t pad;
+	uint64_t offset;
+	uint64_t length;
+};
 
 struct drm_nouveau_gem_pushbuf {
 	uint32_t channel;
-	uint32_t nr_dwords;
 	uint32_t nr_buffers;
-	uint32_t nr_relocs;
-	uint64_t dwords;
 	uint64_t buffers;
-	uint64_t relocs;
-};
-
-struct drm_nouveau_gem_pushbuf_call {
-	uint32_t channel;
-	uint32_t handle;
-	uint32_t offset;
-	uint32_t nr_buffers;
 	uint32_t nr_relocs;
-	uint32_t nr_dwords;
-	uint64_t buffers;
+	uint32_t nr_push;
 	uint64_t relocs;
+	uint64_t push;
 	uint32_t suffix0;
 	uint32_t suffix1;
-	/* below only accessed for CALL2 */
 	uint64_t vram_available;
 	uint64_t gart_available;
 };
 
-struct drm_nouveau_gem_pin {
-	uint32_t handle;
-	uint32_t domain;
-	uint64_t offset;
-};
-
-struct drm_nouveau_gem_unpin {
-	uint32_t handle;
-};
-
 #define NOUVEAU_GEM_CPU_PREP_NOWAIT                                  0x00000001
 #define NOUVEAU_GEM_CPU_PREP_NOBLOCK                                 0x00000002
 #define NOUVEAU_GEM_CPU_PREP_WRITE                                   0x00000004
@@ -183,14 +174,6 @@ struct drm_nouveau_gem_cpu_fini {
 	uint32_t handle;
 };
 
-struct drm_nouveau_gem_tile {
-	uint32_t handle;
-	uint32_t offset;
-	uint32_t size;
-	uint32_t tile_mode;
-	uint32_t tile_flags;
-};
-
 enum nouveau_bus_type {
 	NV_AGP     = 0,
 	NV_PCI     = 1,
@@ -200,22 +183,17 @@ enum nouveau_bus_type {
 struct drm_nouveau_sarea {
 };
 
-#define DRM_NOUVEAU_CARD_INIT          0x00
-#define DRM_NOUVEAU_GETPARAM           0x01
-#define DRM_NOUVEAU_SETPARAM           0x02
-#define DRM_NOUVEAU_CHANNEL_ALLOC      0x03
-#define DRM_NOUVEAU_CHANNEL_FREE       0x04
-#define DRM_NOUVEAU_GROBJ_ALLOC        0x05
-#define DRM_NOUVEAU_NOTIFIEROBJ_ALLOC  0x06
-#define DRM_NOUVEAU_GPUOBJ_FREE        0x07
+#define DRM_NOUVEAU_GETPARAM           0x00
+#define DRM_NOUVEAU_SETPARAM           0x01
+#define DRM_NOUVEAU_CHANNEL_ALLOC      0x02
+#define DRM_NOUVEAU_CHANNEL_FREE       0x03
+#define DRM_NOUVEAU_GROBJ_ALLOC        0x04
+#define DRM_NOUVEAU_NOTIFIEROBJ_ALLOC  0x05
+#define DRM_NOUVEAU_GPUOBJ_FREE        0x06
 #define DRM_NOUVEAU_GEM_NEW            0x40
 #define DRM_NOUVEAU_GEM_PUSHBUF        0x41
-#define DRM_NOUVEAU_GEM_PUSHBUF_CALL   0x42
-#define DRM_NOUVEAU_GEM_PIN            0x43 /* !KMS only */
-#define DRM_NOUVEAU_GEM_UNPIN          0x44 /* !KMS only */
-#define DRM_NOUVEAU_GEM_CPU_PREP       0x45
-#define DRM_NOUVEAU_GEM_CPU_FINI       0x46
-#define DRM_NOUVEAU_GEM_INFO           0x47
-#define DRM_NOUVEAU_GEM_PUSHBUF_CALL2  0x48
+#define DRM_NOUVEAU_GEM_CPU_PREP       0x42
+#define DRM_NOUVEAU_GEM_CPU_FINI       0x43
+#define DRM_NOUVEAU_GEM_INFO           0x44
 
 #endif /* __NOUVEAU_DRM_H__ */