Merge branch 'drm-nouveau-next' of...

Merge branch 'drm-nouveau-next' of git://anongit.freedesktop.org/git/nouveau/linux-2.6 into drm-core-next

* 'drm-nouveau-next' of git://anongit.freedesktop.org/git/nouveau/linux-2.6: (58 commits)
  drm/nouveau: fix off-by-one
  drm/nouveau/temp: Add default calibration values for nv67
  drm/nouveau/temp: Fix signed/unsigned int logic
  drm/nvc0: push prunk140 irq messages to debug loglevel
  drm/nouveau: un-blacklist nvce accel
  drm/nouveau: fix null pointer deref on pre-nv50 chipsets
  drm/nouveau: rework vram init/fini ordering a little
  drm/nouveau: shut lockdep up if last vm ref needs to destroy pgd
  drm/nouveau: fix display takedown order to match reverse init order
  drm/nvc0: enable per-client address spaces
  drm/nouveau: add some debug output if nouveau_mm busy at destroy time
  drm/nv50: enable use of per-client gpu address space
  drm/nouveau: remove implicit mapping of every bo into chan_vm
  drm/nouveau: remove 'chan' argument from nouveau_bo_new
  drm/nouveau: fixup gem_info ioctl to return client-specific bo virtual
  drm/nvc0: explicitly map PDISP semaphore buffer into each channel's vm
  drm/nv50-nvc0: lookup pushbuf virtual address on dma_push
  drm/nv84-nvc0: explicitly map semaphore buffer into channel vm
  drm/nv50-nvc0: explicitly map pushbuf bo into channel vm
  drm/nv50-nvc0: explicitly map notifier bo into channel vm
  ...

drivers/gpu/drm/nouveau/nouveau_bios.c

@@ -5186,7 +5186,7 @@ static int parse_bit_A_tbl_entry(struct drm_device *dev, struct nvbios *bios, st
 	load_table_ptr = ROM16(bios->data[bitentry->offset]);
 
 	if (load_table_ptr == 0x0) {
-		NV_ERROR(dev, "Pointer to BIT loadval table invalid\n");
+		NV_DEBUG(dev, "Pointer to BIT loadval table invalid\n");
 		return -EINVAL;
 	}
 
@@ -6377,6 +6377,37 @@ apply_dcb_encoder_quirks(struct drm_device *dev, int idx, u32 *conn, u32 *conf)
 		}
 	}
 
+	/* Some other twisted XFX board (rhbz#694914)
+	 *
+	 * The DVI/VGA encoder combo that's supposed to represent the
+	 * DVI-I connector actually point at two different ones, and
+	 * the HDMI connector ends up paired with the VGA instead.
+	 *
+	 * Connector table is missing anything for VGA at all, pointing it
+	 * an invalid conntab entry 2 so we figure it out ourself.
+	 */
+	if (nv_match_device(dev, 0x0615, 0x1682, 0x2605)) {
+		if (idx == 0) {
+			*conn = 0x02002300; /* VGA, connector 2 */
+			*conf = 0x00000028;
+		} else
+		if (idx == 1) {
+			*conn = 0x01010312; /* DVI, connector 0 */
+			*conf = 0x00020030;
+		} else
+		if (idx == 2) {
+			*conn = 0x04020310; /* VGA, connector 0 */
+			*conf = 0x00000028;
+		} else
+		if (idx == 3) {
+			*conn = 0x02021322; /* HDMI, connector 1 */
+			*conf = 0x00020010;
+		} else {
+			*conn = 0x0000000e; /* EOL */
+			*conf = 0x00000000;
+		}
+	}
+
 	return true;
 }
 

drivers/gpu/drm/nouveau/nouveau_bo.c

@@ -49,16 +49,12 @@ nouveau_bo_del_ttm(struct ttm_buffer_object *bo)
 		DRM_ERROR("bo %p still attached to GEM object\n", bo);
 
 	nv10_mem_put_tile_region(dev, nvbo->tile, NULL);
-	if (nvbo->vma.node) {
-		nouveau_vm_unmap(&nvbo->vma);
-		nouveau_vm_put(&nvbo->vma);
-	}
 	kfree(nvbo);
 }
 
 static void
 nouveau_bo_fixup_align(struct nouveau_bo *nvbo, u32 flags,
-		       int *align, int *size, int *page_shift)
+		       int *align, int *size)
 {
 	struct drm_nouveau_private *dev_priv = nouveau_bdev(nvbo->bo.bdev);
 
@@ -82,67 +78,51 @@ nouveau_bo_fixup_align(struct nouveau_bo *nvbo, u32 flags,
 			}
 		}
 	} else {
-		if (likely(dev_priv->chan_vm)) {
-			if (!(flags & TTM_PL_FLAG_TT) &&  *size > 256 * 1024)
-				*page_shift = dev_priv->chan_vm->lpg_shift;
-			else
-				*page_shift = dev_priv->chan_vm->spg_shift;
-		} else {
-			*page_shift = 12;
-		}
-
-		*size = roundup(*size, (1 << *page_shift));
-		*align = max((1 << *page_shift), *align);
+		*size = roundup(*size, (1 << nvbo->page_shift));
+		*align = max((1 <<  nvbo->page_shift), *align);
 	}
 
 	*size = roundup(*size, PAGE_SIZE);
 }
 
 int
-nouveau_bo_new(struct drm_device *dev, struct nouveau_channel *chan,
-	       int size, int align, uint32_t flags, uint32_t tile_mode,
-	       uint32_t tile_flags, struct nouveau_bo **pnvbo)
+nouveau_bo_new(struct drm_device *dev, int size, int align,
+	       uint32_t flags, uint32_t tile_mode, uint32_t tile_flags,
+	       struct nouveau_bo **pnvbo)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_bo *nvbo;
-	int ret = 0, page_shift = 0;
+	int ret;
 
 	nvbo = kzalloc(sizeof(struct nouveau_bo), GFP_KERNEL);
 	if (!nvbo)
 		return -ENOMEM;
 	INIT_LIST_HEAD(&nvbo->head);
 	INIT_LIST_HEAD(&nvbo->entry);
+	INIT_LIST_HEAD(&nvbo->vma_list);
 	nvbo->tile_mode = tile_mode;
 	nvbo->tile_flags = tile_flags;
 	nvbo->bo.bdev = &dev_priv->ttm.bdev;
 
-	nouveau_bo_fixup_align(nvbo, flags, &align, &size, &page_shift);
-	align >>= PAGE_SHIFT;
-
-	if (dev_priv->chan_vm) {
-		ret = nouveau_vm_get(dev_priv->chan_vm, size, page_shift,
-				     NV_MEM_ACCESS_RW, &nvbo->vma);
-		if (ret) {
-			kfree(nvbo);
-			return ret;
-		}
+	nvbo->page_shift = 12;
+	if (dev_priv->bar1_vm) {
+		if (!(flags & TTM_PL_FLAG_TT) && size > 256 * 1024)
+			nvbo->page_shift = dev_priv->bar1_vm->lpg_shift;
 	}
 
+	nouveau_bo_fixup_align(nvbo, flags, &align, &size);
 	nvbo->bo.mem.num_pages = size >> PAGE_SHIFT;
 	nouveau_bo_placement_set(nvbo, flags, 0);
 
-	nvbo->channel = chan;
 	ret = ttm_bo_init(&dev_priv->ttm.bdev, &nvbo->bo, size,
-			  ttm_bo_type_device, &nvbo->placement, align, 0,
-			  false, NULL, size, nouveau_bo_del_ttm);
+			  ttm_bo_type_device, &nvbo->placement,
+			  align >> PAGE_SHIFT, 0, false, NULL, size,
+			  nouveau_bo_del_ttm);
 	if (ret) {
 		/* ttm will call nouveau_bo_del_ttm if it fails.. */
 		return ret;
 	}
-	nvbo->channel = NULL;
 
-	if (nvbo->vma.node)
-		nvbo->bo.offset = nvbo->vma.offset;
 	*pnvbo = nvbo;
 	return 0;
 }
@@ -312,8 +292,6 @@ nouveau_bo_validate(struct nouveau_bo *nvbo, bool interruptible,
 	if (ret)
 		return ret;
 
-	if (nvbo->vma.node)
-		nvbo->bo.offset = nvbo->vma.offset;
 	return 0;
 }
 
@@ -440,7 +418,6 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 				     TTM_MEMTYPE_FLAG_CMA;
 			man->available_caching = TTM_PL_MASK_CACHING;
 			man->default_caching = TTM_PL_FLAG_CACHED;
-			man->gpu_offset = dev_priv->gart_info.aper_base;
 			break;
 		default:
 			NV_ERROR(dev, "Unknown GART type: %d\n",
@@ -501,19 +478,12 @@ static int
 nvc0_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 		  struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
 {
-	struct nouveau_mem *old_node = old_mem->mm_node;
-	struct nouveau_mem *new_node = new_mem->mm_node;
-	struct nouveau_bo *nvbo = nouveau_bo(bo);
+	struct nouveau_mem *node = old_mem->mm_node;
+	u64 src_offset = node->vma[0].offset;
+	u64 dst_offset = node->vma[1].offset;
 	u32 page_count = new_mem->num_pages;
-	u64 src_offset, dst_offset;
 	int ret;
 
-	src_offset = old_node->tmp_vma.offset;
-	if (new_node->tmp_vma.node)
-		dst_offset = new_node->tmp_vma.offset;
-	else
-		dst_offset = nvbo->vma.offset;
-
 	page_count = new_mem->num_pages;
 	while (page_count) {
 		int line_count = (page_count > 2047) ? 2047 : page_count;
@@ -547,19 +517,13 @@ static int
 nv50_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 		  struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
 {
-	struct nouveau_mem *old_node = old_mem->mm_node;
-	struct nouveau_mem *new_node = new_mem->mm_node;
+	struct nouveau_mem *node = old_mem->mm_node;
 	struct nouveau_bo *nvbo = nouveau_bo(bo);
 	u64 length = (new_mem->num_pages << PAGE_SHIFT);
-	u64 src_offset, dst_offset;
+	u64 src_offset = node->vma[0].offset;
+	u64 dst_offset = node->vma[1].offset;
 	int ret;
 
-	src_offset = old_node->tmp_vma.offset;
-	if (new_node->tmp_vma.node)
-		dst_offset = new_node->tmp_vma.offset;
-	else
-		dst_offset = nvbo->vma.offset;
-
 	while (length) {
 		u32 amount, stride, height;
 
@@ -694,6 +658,27 @@ nv04_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 	return 0;
 }
 
+static int
+nouveau_vma_getmap(struct nouveau_channel *chan, struct nouveau_bo *nvbo,
+		   struct ttm_mem_reg *mem, struct nouveau_vma *vma)
+{
+	struct nouveau_mem *node = mem->mm_node;
+	int ret;
+
+	ret = nouveau_vm_get(chan->vm, mem->num_pages << PAGE_SHIFT,
+			     node->page_shift, NV_MEM_ACCESS_RO, vma);
+	if (ret)
+		return ret;
+
+	if (mem->mem_type == TTM_PL_VRAM)
+		nouveau_vm_map(vma, node);
+	else
+		nouveau_vm_map_sg(vma, 0, mem->num_pages << PAGE_SHIFT,
+				  node, node->pages);
+
+	return 0;
+}
+
 static int
 nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr,
 		     bool no_wait_reserve, bool no_wait_gpu,
@@ -711,31 +696,20 @@ nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr,
 		mutex_lock_nested(&chan->mutex, NOUVEAU_KCHANNEL_MUTEX);
 	}
 
-	/* create temporary vma for old memory, this will get cleaned
-	 * up after ttm destroys the ttm_mem_reg
+	/* create temporary vmas for the transfer and attach them to the
+	 * old nouveau_mem node, these will get cleaned up after ttm has
+	 * destroyed the ttm_mem_reg
 	 */
 	if (dev_priv->card_type >= NV_50) {
 		struct nouveau_mem *node = old_mem->mm_node;
-		if (!node->tmp_vma.node) {
-			u32 page_shift = nvbo->vma.node->type;
-			if (old_mem->mem_type == TTM_PL_TT)
-				page_shift = nvbo->vma.vm->spg_shift;
-
-			ret = nouveau_vm_get(chan->vm,
-					     old_mem->num_pages << PAGE_SHIFT,
-					     page_shift, NV_MEM_ACCESS_RO,
-					     &node->tmp_vma);
+
+		ret = nouveau_vma_getmap(chan, nvbo, old_mem, &node->vma[0]);
 		if (ret)
 			goto out;
-		}
 
-		if (old_mem->mem_type == TTM_PL_VRAM)
-			nouveau_vm_map(&node->tmp_vma, node);
-		else {
-			nouveau_vm_map_sg(&node->tmp_vma, 0,
-					  old_mem->num_pages << PAGE_SHIFT,
-					  node, node->pages);
-		}
+		ret = nouveau_vma_getmap(chan, nvbo, new_mem, &node->vma[1]);
+		if (ret)
+			goto out;
 	}
 
 	if (dev_priv->card_type < NV_50)
@@ -762,7 +736,6 @@ nouveau_bo_move_flipd(struct ttm_buffer_object *bo, bool evict, bool intr,
 		      bool no_wait_reserve, bool no_wait_gpu,
 		      struct ttm_mem_reg *new_mem)
 {
-	struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev);
 	u32 placement_memtype = TTM_PL_FLAG_TT | TTM_PL_MASK_CACHING;
 	struct ttm_placement placement;
 	struct ttm_mem_reg tmp_mem;
@@ -782,23 +755,7 @@ nouveau_bo_move_flipd(struct ttm_buffer_object *bo, bool evict, bool intr,
 	if (ret)
 		goto out;
 
-	if (dev_priv->card_type >= NV_50) {
-		struct nouveau_bo *nvbo = nouveau_bo(bo);
-		struct nouveau_mem *node = tmp_mem.mm_node;
-		struct nouveau_vma *vma = &nvbo->vma;
-		if (vma->node->type != vma->vm->spg_shift)
-			vma = &node->tmp_vma;
-		nouveau_vm_map_sg(vma, 0, tmp_mem.num_pages << PAGE_SHIFT,
-				  node, node->pages);
-	}
-
 	ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_reserve, no_wait_gpu, &tmp_mem);
-
-	if (dev_priv->card_type >= NV_50) {
-		struct nouveau_bo *nvbo = nouveau_bo(bo);
-		nouveau_vm_unmap(&nvbo->vma);
-	}
-
 	if (ret)
 		goto out;
 
@@ -844,30 +801,22 @@ nouveau_bo_move_flips(struct ttm_buffer_object *bo, bool evict, bool intr,
 static void
 nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem)
 {
-	struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev);
 	struct nouveau_mem *node = new_mem->mm_node;
 	struct nouveau_bo *nvbo = nouveau_bo(bo);
-	struct nouveau_vma *vma = &nvbo->vma;
-	struct nouveau_vm *vm = vma->vm;
-
-	if (dev_priv->card_type < NV_50)
-		return;
-
-	switch (new_mem->mem_type) {
-	case TTM_PL_VRAM:
-		nouveau_vm_map(vma, node);
-		break;
-	case TTM_PL_TT:
-		if (vma->node->type != vm->spg_shift) {
+	struct nouveau_vma *vma;
+
+	list_for_each_entry(vma, &nvbo->vma_list, head) {
+		if (new_mem->mem_type == TTM_PL_VRAM) {
+			nouveau_vm_map(vma, new_mem->mm_node);
+		} else
+		if (new_mem->mem_type == TTM_PL_TT &&
+		    nvbo->page_shift == vma->vm->spg_shift) {
+			nouveau_vm_map_sg(vma, 0, new_mem->
+					  num_pages << PAGE_SHIFT,
+					  node, node->pages);
+		} else {
 			nouveau_vm_unmap(vma);
-			vma = &node->tmp_vma;
 		}
-		nouveau_vm_map_sg(vma, 0, new_mem->num_pages << PAGE_SHIFT,
-				  node, node->pages);
-		break;
-	default:
-		nouveau_vm_unmap(&nvbo->vma);
-		break;
 	}
 }
 
@@ -1113,3 +1062,54 @@ struct ttm_bo_driver nouveau_bo_driver = {
 	.io_mem_free = &nouveau_ttm_io_mem_free,
 };
 
+struct nouveau_vma *
+nouveau_bo_vma_find(struct nouveau_bo *nvbo, struct nouveau_vm *vm)
+{
+	struct nouveau_vma *vma;
+	list_for_each_entry(vma, &nvbo->vma_list, head) {
+		if (vma->vm == vm)
+			return vma;
+	}
+
+	return NULL;
+}
+
+int
+nouveau_bo_vma_add(struct nouveau_bo *nvbo, struct nouveau_vm *vm,
+		   struct nouveau_vma *vma)
+{
+	const u32 size = nvbo->bo.mem.num_pages << PAGE_SHIFT;
+	struct nouveau_mem *node = nvbo->bo.mem.mm_node;
+	int ret;
+
+	ret = nouveau_vm_get(vm, size, nvbo->page_shift,
+			     NV_MEM_ACCESS_RW, vma);
+	if (ret)
+		return ret;
+
+	if (nvbo->bo.mem.mem_type == TTM_PL_VRAM)
+		nouveau_vm_map(vma, nvbo->bo.mem.mm_node);
+	else
+	if (nvbo->bo.mem.mem_type == TTM_PL_TT)
+		nouveau_vm_map_sg(vma, 0, size, node, node->pages);
+
+	list_add_tail(&vma->head, &nvbo->vma_list);
+	vma->refcount = 1;
+	return 0;
+}
+
+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);
+			nouveau_vm_unmap(vma);
+		}
+
+		nouveau_vm_put(vma);
+		list_del(&vma->head);
+	}
+}

drivers/gpu/drm/nouveau/nouveau_channel.c

@@ -27,40 +27,63 @@
 #include "nouveau_drv.h"
 #include "nouveau_drm.h"
 #include "nouveau_dma.h"
+#include "nouveau_ramht.h"
 
 static int
-nouveau_channel_pushbuf_ctxdma_init(struct nouveau_channel *chan)
+nouveau_channel_pushbuf_init(struct nouveau_channel *chan)
 {
+	u32 mem = nouveau_vram_pushbuf ? TTM_PL_FLAG_VRAM : TTM_PL_FLAG_TT;
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_bo *pb = chan->pushbuf_bo;
-	struct nouveau_gpuobj *pushbuf = NULL;
-	int ret = 0;
+	int ret;
+
+	/* allocate buffer object */
+	ret = nouveau_bo_new(dev, 65536, 0, mem, 0, 0, &chan->pushbuf_bo);
+	if (ret)
+		goto out;
+
+	ret = nouveau_bo_pin(chan->pushbuf_bo, mem);
+	if (ret)
+		goto out;
 
+	ret = nouveau_bo_map(chan->pushbuf_bo);
+	if (ret)
+		goto out;
+
+	/* create DMA object covering the entire memtype where the push
+	 * buffer resides, userspace can submit its own push buffers from
+	 * anywhere within the same memtype.
+	 */
+	chan->pushbuf_base = chan->pushbuf_bo->bo.offset;
 	if (dev_priv->card_type >= NV_50) {
+		ret = nouveau_bo_vma_add(chan->pushbuf_bo, chan->vm,
+					 &chan->pushbuf_vma);
+		if (ret)
+			goto out;
+
 		if (dev_priv->card_type < NV_C0) {
 			ret = nouveau_gpuobj_dma_new(chan,
 						     NV_CLASS_DMA_IN_MEMORY, 0,
 						     (1ULL << 40),
 						     NV_MEM_ACCESS_RO,
 						     NV_MEM_TARGET_VM,
-						     &pushbuf);
+						     &chan->pushbuf);
 		}
-		chan->pushbuf_base = pb->bo.offset;
+		chan->pushbuf_base = chan->pushbuf_vma.offset;
 	} else
-	if (pb->bo.mem.mem_type == TTM_PL_TT) {
+	if (chan->pushbuf_bo->bo.mem.mem_type == TTM_PL_TT) {
 		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY, 0,
 					     dev_priv->gart_info.aper_size,
 					     NV_MEM_ACCESS_RO,
-					     NV_MEM_TARGET_GART, &pushbuf);
-		chan->pushbuf_base = pb->bo.mem.start << PAGE_SHIFT;
+					     NV_MEM_TARGET_GART,
+					     &chan->pushbuf);
 	} else
 	if (dev_priv->card_type != NV_04) {
 		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY, 0,
 					     dev_priv->fb_available_size,
 					     NV_MEM_ACCESS_RO,
-					     NV_MEM_TARGET_VRAM, &pushbuf);
-		chan->pushbuf_base = pb->bo.mem.start << PAGE_SHIFT;
+					     NV_MEM_TARGET_VRAM,
+					     &chan->pushbuf);
 	} else {
 		/* NV04 cmdbuf hack, from original ddx.. not sure of it's
 		 * exact reason for existing :)  PCI access to cmdbuf in
@@ -70,47 +93,22 @@ nouveau_channel_pushbuf_ctxdma_init(struct nouveau_channel *chan)
 					     pci_resource_start(dev->pdev, 1),
 					     dev_priv->fb_available_size,
 					     NV_MEM_ACCESS_RO,
-					     NV_MEM_TARGET_PCI, &pushbuf);
-		chan->pushbuf_base = pb->bo.mem.start << PAGE_SHIFT;
+					     NV_MEM_TARGET_PCI,
+					     &chan->pushbuf);
 	}
 
-	nouveau_gpuobj_ref(pushbuf, &chan->pushbuf);
-	nouveau_gpuobj_ref(NULL, &pushbuf);
-	return ret;
-}
-
-static struct nouveau_bo *
-nouveau_channel_user_pushbuf_alloc(struct drm_device *dev)
-{
-	struct nouveau_bo *pushbuf = NULL;
-	int location, ret;
-
-	if (nouveau_vram_pushbuf)
-		location = TTM_PL_FLAG_VRAM;
-	else
-		location = TTM_PL_FLAG_TT;
-
-	ret = nouveau_bo_new(dev, NULL, 65536, 0, location, 0, 0x0000, &pushbuf);
+out:
 	if (ret) {
-		NV_ERROR(dev, "error allocating DMA push buffer: %d\n", ret);
-		return NULL;
-	}
-
-	ret = nouveau_bo_pin(pushbuf, location);
-	if (ret) {
-		NV_ERROR(dev, "error pinning DMA push buffer: %d\n", ret);
-		nouveau_bo_ref(NULL, &pushbuf);
-		return NULL;
+		NV_ERROR(dev, "error initialising pushbuf: %d\n", ret);
+		nouveau_bo_vma_del(chan->pushbuf_bo, &chan->pushbuf_vma);
+		nouveau_gpuobj_ref(NULL, &chan->pushbuf);
+		if (chan->pushbuf_bo) {
+			nouveau_bo_unmap(chan->pushbuf_bo);
+			nouveau_bo_ref(NULL, &chan->pushbuf_bo);
 		}
-
-	ret = nouveau_bo_map(pushbuf);
-	if (ret) {
-		nouveau_bo_unpin(pushbuf);
-		nouveau_bo_ref(NULL, &pushbuf);
-		return NULL;
 	}
 
-	return pushbuf;
+	return 0;
 }
 
 /* allocates and initializes a fifo for user space consumption */
@@ -121,6 +119,7 @@ nouveau_channel_alloc(struct drm_device *dev, struct nouveau_channel **chan_ret,
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
+	struct nouveau_fpriv *fpriv = nouveau_fpriv(file_priv);
 	struct nouveau_channel *chan;
 	unsigned long flags;
 	int ret;
@@ -160,19 +159,14 @@ nouveau_channel_alloc(struct drm_device *dev, struct nouveau_channel **chan_ret,
 	INIT_LIST_HEAD(&chan->nvsw.flip);
 	INIT_LIST_HEAD(&chan->fence.pending);
 
-	/* Allocate DMA push buffer */
-	chan->pushbuf_bo = nouveau_channel_user_pushbuf_alloc(dev);
-	if (!chan->pushbuf_bo) {
-		ret = -ENOMEM;
-		NV_ERROR(dev, "pushbuf %d\n", ret);
+	/* setup channel's memory and vm */
+	ret = nouveau_gpuobj_channel_init(chan, vram_handle, gart_handle);
+	if (ret) {
+		NV_ERROR(dev, "gpuobj %d\n", ret);
 		nouveau_channel_put(&chan);
 		return ret;
 	}
 
-	nouveau_dma_pre_init(chan);
-	chan->user_put = 0x40;
-	chan->user_get = 0x44;
-
 	/* Allocate space for per-channel fixed notifier memory */
 	ret = nouveau_notifier_init_channel(chan);
 	if (ret) {
@@ -181,21 +175,17 @@ nouveau_channel_alloc(struct drm_device *dev, struct nouveau_channel **chan_ret,
 		return ret;
 	}
 
-	/* Setup channel's default objects */
-	ret = nouveau_gpuobj_channel_init(chan, vram_handle, gart_handle);
+	/* Allocate DMA push buffer */
+	ret = nouveau_channel_pushbuf_init(chan);
 	if (ret) {
-		NV_ERROR(dev, "gpuobj %d\n", ret);
+		NV_ERROR(dev, "pushbuf %d\n", ret);
 		nouveau_channel_put(&chan);
 		return ret;
 	}
 
-	/* Create a dma object for the push buffer */
-	ret = nouveau_channel_pushbuf_ctxdma_init(chan);
-	if (ret) {
-		NV_ERROR(dev, "pbctxdma %d\n", ret);
-		nouveau_channel_put(&chan);
-		return ret;
-	}
+	nouveau_dma_pre_init(chan);
+	chan->user_put = 0x40;
+	chan->user_get = 0x44;
 
 	/* disable the fifo caches */
 	pfifo->reassign(dev, false);
@@ -220,6 +210,11 @@ nouveau_channel_alloc(struct drm_device *dev, struct nouveau_channel **chan_ret,
 	nouveau_debugfs_channel_init(chan);
 
 	NV_DEBUG(dev, "channel %d initialised\n", chan->id);
+	if (fpriv) {
+		spin_lock(&fpriv->lock);
+		list_add(&chan->list, &fpriv->channels);
+		spin_unlock(&fpriv->lock);
+	}
 	*chan_ret = chan;
 	return 0;
 }
@@ -236,29 +231,23 @@ nouveau_channel_get_unlocked(struct nouveau_channel *ref)
 }
 
 struct nouveau_channel *
-nouveau_channel_get(struct drm_device *dev, struct drm_file *file_priv, int id)
+nouveau_channel_get(struct drm_file *file_priv, int id)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_fpriv *fpriv = nouveau_fpriv(file_priv);
 	struct nouveau_channel *chan;
-	unsigned long flags;
-
-	if (unlikely(id < 0 || id >= NOUVEAU_MAX_CHANNEL_NR))
-		return ERR_PTR(-EINVAL);
-
-	spin_lock_irqsave(&dev_priv->channels.lock, flags);
-	chan = nouveau_channel_get_unlocked(dev_priv->channels.ptr[id]);
-	spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
-
-	if (unlikely(!chan))
-		return ERR_PTR(-EINVAL);
-
-	if (unlikely(file_priv && chan->file_priv != file_priv)) {
-		nouveau_channel_put_unlocked(&chan);
-		return ERR_PTR(-EINVAL);
-	}
 
+	spin_lock(&fpriv->lock);
+	list_for_each_entry(chan, &fpriv->channels, list) {
+		if (chan->id == id) {
+			chan = nouveau_channel_get_unlocked(chan);
+			spin_unlock(&fpriv->lock);
 			mutex_lock(&chan->mutex);
 			return chan;
+		}
+	}
+	spin_unlock(&fpriv->lock);
+
+	return ERR_PTR(-EINVAL);
 }
 
 void
@@ -312,12 +301,14 @@ nouveau_channel_put_unlocked(struct nouveau_channel **pchan)
 	/* destroy any resources the channel owned */
 	nouveau_gpuobj_ref(NULL, &chan->pushbuf);
 	if (chan->pushbuf_bo) {
+		nouveau_bo_vma_del(chan->pushbuf_bo, &chan->pushbuf_vma);
 		nouveau_bo_unmap(chan->pushbuf_bo);
 		nouveau_bo_unpin(chan->pushbuf_bo);
 		nouveau_bo_ref(NULL, &chan->pushbuf_bo);
 	}
-	nouveau_gpuobj_channel_takedown(chan);
+	nouveau_ramht_ref(NULL, &chan->ramht, chan);
 	nouveau_notifier_takedown_channel(chan);
+	nouveau_gpuobj_channel_takedown(chan);
 
 	nouveau_channel_ref(NULL, pchan);
 }
@@ -383,10 +374,11 @@ nouveau_channel_cleanup(struct drm_device *dev, struct drm_file *file_priv)
 
 	NV_DEBUG(dev, "clearing FIFO enables from file_priv\n");
 	for (i = 0; i < engine->fifo.channels; i++) {
-		chan = nouveau_channel_get(dev, file_priv, i);
+		chan = nouveau_channel_get(file_priv, i);
 		if (IS_ERR(chan))
 			continue;
 
+		list_del(&chan->list);
 		atomic_dec(&chan->users);
 		nouveau_channel_put(&chan);
 	}
@@ -459,10 +451,11 @@ nouveau_ioctl_fifo_free(struct drm_device *dev, void *data,
 	struct drm_nouveau_channel_free *req = data;
 	struct nouveau_channel *chan;
 
-	chan = nouveau_channel_get(dev, file_priv, req->channel);
+	chan = nouveau_channel_get(file_priv, req->channel);
 	if (IS_ERR(chan))
 		return PTR_ERR(chan);
 
+	list_del(&chan->list);
 	atomic_dec(&chan->users);
 	nouveau_channel_put(&chan);
 	return 0;

drivers/gpu/drm/nouveau/nouveau_dma.c

@@ -167,8 +167,13 @@ nv50_dma_push(struct nouveau_channel *chan, struct nouveau_bo *bo,
 	      int delta, int length)
 {
 	struct nouveau_bo *pb = chan->pushbuf_bo;
-	uint64_t offset = bo->bo.offset + delta;
+	struct nouveau_vma *vma;
 	int ip = (chan->dma.ib_put * 2) + chan->dma.ib_base;
+	u64 offset;
+
+	vma = nouveau_bo_vma_find(bo, chan->vm);
+	BUG_ON(!vma);
+	offset = vma->offset + delta;
 
 	BUG_ON(chan->dma.ib_free < 1);
 	nouveau_bo_wr32(pb, ip++, lower_32_bits(offset));

drivers/gpu/drm/nouveau/nouveau_drv.c

@@ -73,7 +73,7 @@ int nouveau_ignorelid = 0;
 module_param_named(ignorelid, nouveau_ignorelid, int, 0400);
 
 MODULE_PARM_DESC(noaccel, "Disable all acceleration");
-int nouveau_noaccel = 0;
+int nouveau_noaccel = -1;
 module_param_named(noaccel, nouveau_noaccel, int, 0400);
 
 MODULE_PARM_DESC(nofbaccel, "Disable fbcon acceleration");
@@ -119,6 +119,10 @@ MODULE_PARM_DESC(msi, "Enable MSI (default: off)\n");
 int nouveau_msi;
 module_param_named(msi, nouveau_msi, int, 0400);
 
+MODULE_PARM_DESC(ctxfw, "Use external HUB/GPC ucode (fermi)\n");
+int nouveau_ctxfw;
+module_param_named(ctxfw, nouveau_ctxfw, int, 0400);
+
 int nouveau_fbpercrtc;
 #if 0
 module_param_named(fbpercrtc, nouveau_fbpercrtc, int, 0400);
@@ -354,7 +358,7 @@ nouveau_pci_resume(struct pci_dev *pdev)
 
 	list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
 		struct nouveau_crtc *nv_crtc = nouveau_crtc(crtc);
-		u32 offset = nv_crtc->cursor.nvbo->bo.mem.start << PAGE_SHIFT;
+		u32 offset = nv_crtc->cursor.nvbo->bo.offset;
 
 		nv_crtc->cursor.set_offset(nv_crtc, offset);
 		nv_crtc->cursor.set_pos(nv_crtc, nv_crtc->cursor_saved_x,
@@ -389,7 +393,9 @@ static struct drm_driver driver = {
 	.firstopen = nouveau_firstopen,
 	.lastclose = nouveau_lastclose,
 	.unload = nouveau_unload,
+	.open = nouveau_open,
 	.preclose = nouveau_preclose,
+	.postclose = nouveau_postclose,
 #if defined(CONFIG_DRM_NOUVEAU_DEBUG)
 	.debugfs_init = nouveau_debugfs_init,
 	.debugfs_cleanup = nouveau_debugfs_takedown,
@@ -420,6 +426,8 @@ static struct drm_driver driver = {
 
 	.gem_init_object = nouveau_gem_object_new,
 	.gem_free_object = nouveau_gem_object_del,
+	.gem_open_object = nouveau_gem_object_open,
+	.gem_close_object = nouveau_gem_object_close,
 
 	.name = DRIVER_NAME,
 	.desc = DRIVER_DESC,

drivers/gpu/drm/nouveau/nouveau_drv.h

@@ -46,9 +46,17 @@
 #include "ttm/ttm_module.h"
 
 struct nouveau_fpriv {
-	struct ttm_object_file *tfile;
+	spinlock_t lock;
+	struct list_head channels;
+	struct nouveau_vm *vm;
 };
 
+static inline struct nouveau_fpriv *
+nouveau_fpriv(struct drm_file *file_priv)
+{
+	return file_priv ? file_priv->driver_priv : NULL;
+}
+
 #define DRM_FILE_PAGE_OFFSET (0x100000000ULL >> PAGE_SHIFT)
 
 #include "nouveau_drm.h"
@@ -69,7 +77,7 @@ struct nouveau_mem {
 	struct drm_device *dev;
 
 	struct nouveau_vma bar_vma;
-	struct nouveau_vma tmp_vma;
+	struct nouveau_vma vma[2];
 	u8  page_shift;
 
 	struct drm_mm_node *tag;
@@ -107,7 +115,8 @@ struct nouveau_bo {
 
 	struct nouveau_channel *channel;
 
-	struct nouveau_vma vma;
+	struct list_head vma_list;
+	unsigned page_shift;
 
 	uint32_t tile_mode;
 	uint32_t tile_flags;
@@ -176,9 +185,10 @@ struct nouveau_gpuobj {
 	uint32_t flags;
 
 	u32 size;
-	u32 pinst;
-	u32 cinst;
-	u64 vinst;
+	u32 pinst;	/* PRAMIN BAR offset */
+	u32 cinst;	/* Channel offset */
+	u64 vinst;	/* VRAM address */
+	u64 linst;	/* VM address */
 
 	uint32_t engine;
 	uint32_t class;
@@ -201,6 +211,7 @@ enum nouveau_channel_mutex_class {
 
 struct nouveau_channel {
 	struct drm_device *dev;
+	struct list_head list;
 	int id;
 
 	/* references to the channel data structure */
@@ -228,15 +239,18 @@ struct nouveau_channel {
 		uint32_t sequence;
 		uint32_t sequence_ack;
 		atomic_t last_sequence_irq;
+		struct nouveau_vma vma;
 	} fence;
 
 	/* DMA push buffer */
 	struct nouveau_gpuobj *pushbuf;
 	struct nouveau_bo     *pushbuf_bo;
+	struct nouveau_vma     pushbuf_vma;
 	uint32_t               pushbuf_base;
 
 	/* Notifier memory */
 	struct nouveau_bo *notifier_bo;
+	struct nouveau_vma notifier_vma;
 	struct drm_mm notifier_heap;
 
 	/* PFIFO context */
@@ -278,6 +292,7 @@ struct nouveau_channel {
 
 	uint32_t sw_subchannel[8];
 
+	struct nouveau_vma dispc_vma[2];
 	struct {
 		struct nouveau_gpuobj *vblsem;
 		uint32_t vblsem_head;
@@ -314,7 +329,8 @@ struct nouveau_instmem_engine {
 	int	(*suspend)(struct drm_device *dev);
 	void	(*resume)(struct drm_device *dev);
 
-	int	(*get)(struct nouveau_gpuobj *, u32 size, u32 align);
+	int	(*get)(struct nouveau_gpuobj *, struct nouveau_channel *,
+		       u32 size, u32 align);
 	void	(*put)(struct nouveau_gpuobj *);
 	int	(*map)(struct nouveau_gpuobj *);
 	void	(*unmap)(struct nouveau_gpuobj *);
@@ -445,9 +461,9 @@ struct nouveau_pm_level {
 struct nouveau_pm_temp_sensor_constants {
 	u16 offset_constant;
 	s16 offset_mult;
-	u16 offset_div;
-	u16 slope_mult;
-	u16 slope_div;
+	s16 offset_div;
+	s16 slope_mult;
+	s16 slope_div;
 };
 
 struct nouveau_pm_threshold_temp {
@@ -488,7 +504,10 @@ struct nouveau_pm_engine {
 };
 
 struct nouveau_vram_engine {
+	struct nouveau_mm *mm;
+
 	int  (*init)(struct drm_device *);
+	void (*takedown)(struct drm_device *dev);
 	int  (*get)(struct drm_device *, u64, u32 align, u32 size_nc,
 		    u32 type, struct nouveau_mem **);
 	void (*put)(struct drm_device *, struct nouveau_mem **);
@@ -608,6 +627,7 @@ enum nouveau_card_type {
 
 struct drm_nouveau_private {
 	struct drm_device *dev;
+	bool noaccel;
 
 	/* the card type, takes NV_* as values */
 	enum nouveau_card_type card_type;
@@ -700,7 +720,6 @@ struct drm_nouveau_private {
 	/* VRAM/fb configuration */
 	uint64_t vram_size;
 	uint64_t vram_sys_base;
-	u32 vram_rblock_size;
 
 	uint64_t fb_phys;
 	uint64_t fb_available_size;
@@ -784,12 +803,15 @@ extern int nouveau_override_conntype;
 extern char *nouveau_perflvl;
 extern int nouveau_perflvl_wr;
 extern int nouveau_msi;
+extern int nouveau_ctxfw;
 
 extern int nouveau_pci_suspend(struct pci_dev *pdev, pm_message_t pm_state);
 extern int nouveau_pci_resume(struct pci_dev *pdev);
 
 /* nouveau_state.c */
+extern int  nouveau_open(struct drm_device *, struct drm_file *);
 extern void nouveau_preclose(struct drm_device *dev, struct drm_file *);
+extern void nouveau_postclose(struct drm_device *, struct drm_file *);
 extern int  nouveau_load(struct drm_device *, unsigned long flags);
 extern int  nouveau_firstopen(struct drm_device *);
 extern void nouveau_lastclose(struct drm_device *);
@@ -847,7 +869,7 @@ extern int  nouveau_channel_alloc(struct drm_device *dev,
 extern struct nouveau_channel *
 nouveau_channel_get_unlocked(struct nouveau_channel *);
 extern struct nouveau_channel *
-nouveau_channel_get(struct drm_device *, struct drm_file *, int id);
+nouveau_channel_get(struct drm_file *, int id);
 extern void nouveau_channel_put_unlocked(struct nouveau_channel **);
 extern void nouveau_channel_put(struct nouveau_channel **);
 extern void nouveau_channel_ref(struct nouveau_channel *chan,
@@ -1169,7 +1191,8 @@ extern int  nv04_instmem_init(struct drm_device *);
 extern void nv04_instmem_takedown(struct drm_device *);
 extern int  nv04_instmem_suspend(struct drm_device *);
 extern void nv04_instmem_resume(struct drm_device *);
-extern int  nv04_instmem_get(struct nouveau_gpuobj *, u32 size, u32 align);
+extern int  nv04_instmem_get(struct nouveau_gpuobj *, struct nouveau_channel *,
+			     u32 size, u32 align);
 extern void nv04_instmem_put(struct nouveau_gpuobj *);
 extern int  nv04_instmem_map(struct nouveau_gpuobj *);
 extern void nv04_instmem_unmap(struct nouveau_gpuobj *);
@@ -1180,7 +1203,8 @@ extern int  nv50_instmem_init(struct drm_device *);
 extern void nv50_instmem_takedown(struct drm_device *);
 extern int  nv50_instmem_suspend(struct drm_device *);
 extern void nv50_instmem_resume(struct drm_device *);
-extern int  nv50_instmem_get(struct nouveau_gpuobj *, u32 size, u32 align);
+extern int  nv50_instmem_get(struct nouveau_gpuobj *, struct nouveau_channel *,
+			     u32 size, u32 align);
 extern void nv50_instmem_put(struct nouveau_gpuobj *);
 extern int  nv50_instmem_map(struct nouveau_gpuobj *);
 extern void nv50_instmem_unmap(struct nouveau_gpuobj *);
@@ -1247,10 +1271,9 @@ extern int nv04_crtc_create(struct drm_device *, int index);
 
 /* nouveau_bo.c */
 extern struct ttm_bo_driver nouveau_bo_driver;
-extern int nouveau_bo_new(struct drm_device *, struct nouveau_channel *,
-			  int size, int align, uint32_t flags,
-			  uint32_t tile_mode, uint32_t tile_flags,
-			  struct nouveau_bo **);
+extern int nouveau_bo_new(struct drm_device *, int size, int align,
+			  uint32_t flags, uint32_t tile_mode,
+			  uint32_t tile_flags, struct nouveau_bo **);
 extern int nouveau_bo_pin(struct nouveau_bo *, uint32_t flags);
 extern int nouveau_bo_unpin(struct nouveau_bo *);
 extern int nouveau_bo_map(struct nouveau_bo *);
@@ -1265,6 +1288,12 @@ extern void nouveau_bo_fence(struct nouveau_bo *, struct nouveau_fence *);
 extern int nouveau_bo_validate(struct nouveau_bo *, bool interruptible,
 			       bool no_wait_reserve, bool no_wait_gpu);
 
+extern struct nouveau_vma *
+nouveau_bo_vma_find(struct nouveau_bo *, struct nouveau_vm *);
+extern int  nouveau_bo_vma_add(struct nouveau_bo *, struct nouveau_vm *,
+			       struct nouveau_vma *);
+extern void nouveau_bo_vma_del(struct nouveau_bo *, struct nouveau_vma *);
+
 /* nouveau_fence.c */
 struct nouveau_fence;
 extern int nouveau_fence_init(struct drm_device *);
@@ -1310,12 +1339,14 @@ static inline struct nouveau_fence *nouveau_fence_ref(struct nouveau_fence *obj)
 }
 
 /* nouveau_gem.c */
-extern int nouveau_gem_new(struct drm_device *, struct nouveau_channel *,
-			   int size, int align, uint32_t domain,
-			   uint32_t tile_mode, uint32_t tile_flags,
-			   struct nouveau_bo **);
+extern int nouveau_gem_new(struct drm_device *, int size, int align,
+			   uint32_t domain, uint32_t tile_mode,
+			   uint32_t tile_flags, struct nouveau_bo **);
 extern int nouveau_gem_object_new(struct drm_gem_object *);
 extern void nouveau_gem_object_del(struct drm_gem_object *);
+extern int nouveau_gem_object_open(struct drm_gem_object *, struct drm_file *);
+extern void nouveau_gem_object_close(struct drm_gem_object *,
+				     struct drm_file *);
 extern int nouveau_gem_ioctl_new(struct drm_device *, void *,
 				 struct drm_file *);
 extern int nouveau_gem_ioctl_pushbuf(struct drm_device *, void *,

drivers/gpu/drm/nouveau/nouveau_fb.h

@@ -30,6 +30,7 @@
 struct nouveau_framebuffer {
 	struct drm_framebuffer base;
 	struct nouveau_bo *nvbo;
+	struct nouveau_vma vma;
 	u32 r_dma;
 	u32 r_format;
 	u32 r_pitch;

drivers/gpu/drm/nouveau/nouveau_fbcon.c

@@ -279,6 +279,7 @@ nouveau_fbcon_create(struct nouveau_fbdev *nfbdev,
 	struct fb_info *info;
 	struct drm_framebuffer *fb;
 	struct nouveau_framebuffer *nouveau_fb;
+	struct nouveau_channel *chan;
 	struct nouveau_bo *nvbo;
 	struct drm_mode_fb_cmd mode_cmd;
 	struct pci_dev *pdev = dev->pdev;
@@ -296,8 +297,8 @@ nouveau_fbcon_create(struct nouveau_fbdev *nfbdev,
 	size = mode_cmd.pitch * mode_cmd.height;
 	size = roundup(size, PAGE_SIZE);
 
-	ret = nouveau_gem_new(dev, dev_priv->channel, size, 0,
-			      NOUVEAU_GEM_DOMAIN_VRAM, 0, 0x0000, &nvbo);
+	ret = nouveau_gem_new(dev, size, 0, NOUVEAU_GEM_DOMAIN_VRAM,
+			      0, 0x0000, &nvbo);
 	if (ret) {
 		NV_ERROR(dev, "failed to allocate framebuffer\n");
 		goto out;
@@ -318,6 +319,15 @@ nouveau_fbcon_create(struct nouveau_fbdev *nfbdev,
 		goto out;
 	}
 
+	chan = nouveau_nofbaccel ? NULL : dev_priv->channel;
+	if (chan && dev_priv->card_type >= NV_50) {
+		ret = nouveau_bo_vma_add(nvbo, chan->vm, &nfbdev->nouveau_fb.vma);
+		if (ret) {
+			NV_ERROR(dev, "failed to map fb into chan: %d\n", ret);
+			chan = NULL;
+		}
+	}
+
 	mutex_lock(&dev->struct_mutex);
 
 	info = framebuffer_alloc(0, device);
@@ -448,6 +458,7 @@ nouveau_fbcon_destroy(struct drm_device *dev, struct nouveau_fbdev *nfbdev)
 
 	if (nouveau_fb->nvbo) {
 		nouveau_bo_unmap(nouveau_fb->nvbo);
+		nouveau_bo_vma_del(nouveau_fb->nvbo, &nouveau_fb->vma);
 		drm_gem_object_unreference_unlocked(nouveau_fb->nvbo->gem);
 		nouveau_fb->nvbo = NULL;
 	}

drivers/gpu/drm/nouveau/nouveau_fence.c

@@ -336,6 +336,7 @@ semaphore_acquire(struct nouveau_channel *chan, struct nouveau_semaphore *sema)
 {
 	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
 	struct nouveau_fence *fence = NULL;
+	u64 offset = chan->fence.vma.offset + sema->mem->start;
 	int ret;
 
 	if (dev_priv->chipset < 0x84) {
@@ -345,13 +346,10 @@ semaphore_acquire(struct nouveau_channel *chan, struct nouveau_semaphore *sema)
 
 		BEGIN_RING(chan, NvSubSw, NV_SW_DMA_SEMAPHORE, 3);
 		OUT_RING  (chan, NvSema);
-		OUT_RING  (chan, sema->mem->start);
+		OUT_RING  (chan, offset);
 		OUT_RING  (chan, 1);
 	} else
 	if (dev_priv->chipset < 0xc0) {
-		struct nouveau_vma *vma = &dev_priv->fence.bo->vma;
-		u64 offset = vma->offset + sema->mem->start;
-
 		ret = RING_SPACE(chan, 7);
 		if (ret)
 			return ret;
@@ -364,9 +362,6 @@ semaphore_acquire(struct nouveau_channel *chan, struct nouveau_semaphore *sema)
 		OUT_RING  (chan, 1);
 		OUT_RING  (chan, 1); /* ACQUIRE_EQ */
 	} else {
-		struct nouveau_vma *vma = &dev_priv->fence.bo->vma;
-		u64 offset = vma->offset + sema->mem->start;
-
 		ret = RING_SPACE(chan, 5);
 		if (ret)
 			return ret;
@@ -394,6 +389,7 @@ semaphore_release(struct nouveau_channel *chan, struct nouveau_semaphore *sema)
 {
 	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
 	struct nouveau_fence *fence = NULL;
+	u64 offset = chan->fence.vma.offset + sema->mem->start;
 	int ret;
 
 	if (dev_priv->chipset < 0x84) {
@@ -403,14 +399,11 @@ semaphore_release(struct nouveau_channel *chan, struct nouveau_semaphore *sema)
 
 		BEGIN_RING(chan, NvSubSw, NV_SW_DMA_SEMAPHORE, 2);
 		OUT_RING  (chan, NvSema);
-		OUT_RING  (chan, sema->mem->start);
+		OUT_RING  (chan, offset);
 		BEGIN_RING(chan, NvSubSw, NV_SW_SEMAPHORE_RELEASE, 1);
 		OUT_RING  (chan, 1);
 	} else
 	if (dev_priv->chipset < 0xc0) {
-		struct nouveau_vma *vma = &dev_priv->fence.bo->vma;
-		u64 offset = vma->offset + sema->mem->start;
-
 		ret = RING_SPACE(chan, 7);
 		if (ret)
 			return ret;
@@ -423,9 +416,6 @@ semaphore_release(struct nouveau_channel *chan, struct nouveau_semaphore *sema)
 		OUT_RING  (chan, 1);
 		OUT_RING  (chan, 2); /* RELEASE */
 	} else {
-		struct nouveau_vma *vma = &dev_priv->fence.bo->vma;
-		u64 offset = vma->offset + sema->mem->start;
-
 		ret = RING_SPACE(chan, 5);
 		if (ret)
 			return ret;
@@ -540,6 +530,12 @@ nouveau_fence_channel_init(struct nouveau_channel *chan)
 		nouveau_gpuobj_ref(NULL, &obj);
 		if (ret)
 			return ret;
+	} else {
+		/* map fence bo into channel's vm */
+		ret = nouveau_bo_vma_add(dev_priv->fence.bo, chan->vm,
+					 &chan->fence.vma);
+		if (ret)
+			return ret;
 	}
 
 	INIT_LIST_HEAD(&chan->fence.pending);
@@ -551,10 +547,10 @@ nouveau_fence_channel_init(struct nouveau_channel *chan)
 void
 nouveau_fence_channel_fini(struct nouveau_channel *chan)
 {
+	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
 	struct nouveau_fence *tmp, *fence;
 
 	spin_lock(&chan->fence.lock);
-
 	list_for_each_entry_safe(fence, tmp, &chan->fence.pending, entry) {
 		fence->signalled = true;
 		list_del(&fence->entry);
@@ -564,8 +560,9 @@ nouveau_fence_channel_fini(struct nouveau_channel *chan)
 
 		kref_put(&fence->refcount, nouveau_fence_del);
 	}
-
 	spin_unlock(&chan->fence.lock);
+
+	nouveau_bo_vma_del(dev_priv->fence.bo, &chan->fence.vma);
 }
 
 int
@@ -577,7 +574,7 @@ nouveau_fence_init(struct drm_device *dev)
 
 	/* Create a shared VRAM heap for cross-channel sync. */
 	if (USE_SEMA(dev)) {
-		ret = nouveau_bo_new(dev, NULL, size, 0, TTM_PL_FLAG_VRAM,
+		ret = nouveau_bo_new(dev, size, 0, TTM_PL_FLAG_VRAM,
 				     0, 0, &dev_priv->fence.bo);
 		if (ret)
 			return ret;

drivers/gpu/drm/nouveau/nouveau_gem.c

@@ -60,9 +60,69 @@ nouveau_gem_object_del(struct drm_gem_object *gem)
 }
 
 int
-nouveau_gem_new(struct drm_device *dev, struct nouveau_channel *chan,
-		int size, int align, uint32_t domain, uint32_t tile_mode,
-		uint32_t tile_flags, struct nouveau_bo **pnvbo)
+nouveau_gem_object_open(struct drm_gem_object *gem, struct drm_file *file_priv)
+{
+	struct nouveau_fpriv *fpriv = nouveau_fpriv(file_priv);
+	struct nouveau_bo *nvbo = nouveau_gem_object(gem);
+	struct nouveau_vma *vma;
+	int ret;
+
+	if (!fpriv->vm)
+		return 0;
+
+	ret = ttm_bo_reserve(&nvbo->bo, false, false, false, 0);
+	if (ret)
+		return ret;
+
+	vma = nouveau_bo_vma_find(nvbo, fpriv->vm);
+	if (!vma) {
+		vma = kzalloc(sizeof(*vma), GFP_KERNEL);
+		if (!vma) {
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		ret = nouveau_bo_vma_add(nvbo, fpriv->vm, vma);
+		if (ret) {
+			kfree(vma);
+			goto out;
+		}
+	} else {
+		vma->refcount++;
+	}
+
+out:
+	ttm_bo_unreserve(&nvbo->bo);
+	return ret;
+}
+
+void
+nouveau_gem_object_close(struct drm_gem_object *gem, struct drm_file *file_priv)
+{
+	struct nouveau_fpriv *fpriv = nouveau_fpriv(file_priv);
+	struct nouveau_bo *nvbo = nouveau_gem_object(gem);
+	struct nouveau_vma *vma;
+	int ret;
+
+	if (!fpriv->vm)
+		return;
+
+	ret = ttm_bo_reserve(&nvbo->bo, false, false, false, 0);
+	if (ret)
+		return;
+
+	vma = nouveau_bo_vma_find(nvbo, fpriv->vm);
+	if (vma) {
+		if (--vma->refcount == 0)
+			nouveau_bo_vma_del(nvbo, vma);
+	}
+	ttm_bo_unreserve(&nvbo->bo);
+}
+
+int
+nouveau_gem_new(struct drm_device *dev, int size, int align, uint32_t domain,
+		uint32_t tile_mode, uint32_t tile_flags,
+		struct nouveau_bo **pnvbo)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_bo *nvbo;
@@ -76,7 +136,7 @@ nouveau_gem_new(struct drm_device *dev, struct nouveau_channel *chan,
 	if (!flags || domain & NOUVEAU_GEM_DOMAIN_CPU)
 		flags |= TTM_PL_FLAG_SYSTEM;
 
-	ret = nouveau_bo_new(dev, chan, size, align, flags, tile_mode,
+	ret = nouveau_bo_new(dev, size, align, flags, tile_mode,
 			     tile_flags, pnvbo);
 	if (ret)
 		return ret;
@@ -103,17 +163,28 @@ nouveau_gem_new(struct drm_device *dev, struct nouveau_channel *chan,
 }
 
 static int
-nouveau_gem_info(struct drm_gem_object *gem, struct drm_nouveau_gem_info *rep)
+nouveau_gem_info(struct drm_file *file_priv, struct drm_gem_object *gem,
+		 struct drm_nouveau_gem_info *rep)
 {
+	struct nouveau_fpriv *fpriv = nouveau_fpriv(file_priv);
 	struct nouveau_bo *nvbo = nouveau_gem_object(gem);
+	struct nouveau_vma *vma;
 
 	if (nvbo->bo.mem.mem_type == TTM_PL_TT)
 		rep->domain = NOUVEAU_GEM_DOMAIN_GART;
 	else
 		rep->domain = NOUVEAU_GEM_DOMAIN_VRAM;
 
-	rep->size = nvbo->bo.mem.num_pages << PAGE_SHIFT;
 	rep->offset = nvbo->bo.offset;
+	if (fpriv->vm) {
+		vma = nouveau_bo_vma_find(nvbo, fpriv->vm);
+		if (!vma)
+			return -EINVAL;
+
+		rep->offset = vma->offset;
+	}
+
+	rep->size = nvbo->bo.mem.num_pages << PAGE_SHIFT;
 	rep->map_handle = nvbo->bo.addr_space_offset;
 	rep->tile_mode = nvbo->tile_mode;
 	rep->tile_flags = nvbo->tile_flags;
@@ -127,7 +198,6 @@ nouveau_gem_ioctl_new(struct drm_device *dev, void *data,
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct drm_nouveau_gem_new *req = data;
 	struct nouveau_bo *nvbo = NULL;
-	struct nouveau_channel *chan = NULL;
 	int ret = 0;
 
 	if (unlikely(dev_priv->ttm.bdev.dev_mapping == NULL))
@@ -138,28 +208,21 @@ nouveau_gem_ioctl_new(struct drm_device *dev, void *data,
 		return -EINVAL;
 	}
 
-	if (req->channel_hint) {
-		chan = nouveau_channel_get(dev, file_priv, req->channel_hint);
-		if (IS_ERR(chan))
-			return PTR_ERR(chan);
-	}
-
-	ret = nouveau_gem_new(dev, chan, req->info.size, req->align,
+	ret = nouveau_gem_new(dev, req->info.size, req->align,
 			      req->info.domain, req->info.tile_mode,
 			      req->info.tile_flags, &nvbo);
-	if (chan)
-		nouveau_channel_put(&chan);
 	if (ret)
 		return ret;
 
-	ret = nouveau_gem_info(nvbo->gem, &req->info);
+	ret = drm_gem_handle_create(file_priv, nvbo->gem, &req->info.handle);
+	if (ret == 0) {
+		ret = nouveau_gem_info(file_priv, nvbo->gem, &req->info);
 		if (ret)
-		goto out;
+			drm_gem_handle_delete(file_priv, req->info.handle);
+	}
 
-	ret = drm_gem_handle_create(file_priv, nvbo->gem, &req->info.handle);
 	/* drop reference from allocate - handle holds it now */
 	drm_gem_object_unreference_unlocked(nvbo->gem);
-out:
 	return ret;
 }
 
@@ -318,6 +381,7 @@ static int
 validate_list(struct nouveau_channel *chan, struct list_head *list,
 	      struct drm_nouveau_gem_pushbuf_bo *pbbo, uint64_t user_pbbo_ptr)
 {
+	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
 	struct drm_nouveau_gem_pushbuf_bo __user *upbbo =
 				(void __force __user *)(uintptr_t)user_pbbo_ptr;
 	struct drm_device *dev = chan->dev;
@@ -356,6 +420,7 @@ validate_list(struct nouveau_channel *chan, struct list_head *list,
 			return ret;
 		}
 
+		if (dev_priv->card_type < NV_50) {
 			if (nvbo->bo.offset == b->presumed.offset &&
 			    ((nvbo->bo.mem.mem_type == TTM_PL_VRAM &&
 			      b->presumed.domain & NOUVEAU_GEM_DOMAIN_VRAM) ||
@@ -375,6 +440,7 @@ validate_list(struct nouveau_channel *chan, struct list_head *list,
 					     &b->presumed, sizeof(b->presumed)))
 				return -EFAULT;
 		}
+	}
 
 	return relocs;
 }
@@ -548,7 +614,7 @@ nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
 	struct nouveau_fence *fence = NULL;
 	int i, j, ret = 0, do_reloc = 0;
 
-	chan = nouveau_channel_get(dev, file_priv, req->channel);
+	chan = nouveau_channel_get(file_priv, req->channel);
 	if (IS_ERR(chan))
 		return PTR_ERR(chan);
 
@@ -782,7 +848,7 @@ nouveau_gem_ioctl_info(struct drm_device *dev, void *data,
 	if (!gem)
 		return -ENOENT;
 
-	ret = nouveau_gem_info(gem, req);
+	ret = nouveau_gem_info(file_priv, gem, req);
 	drm_gem_object_unreference_unlocked(gem);
 	return ret;
 }

drivers/gpu/drm/nouveau/nouveau_mem.c

@@ -451,10 +451,6 @@ nouveau_mem_vram_init(struct drm_device *dev)
 		dev_priv->ramin_rsvd_vram = 512 * 1024;
 	}
 
-	ret = dev_priv->engine.vram.init(dev);
-	if (ret)
-		return ret;
-
 	NV_INFO(dev, "Detected %dMiB VRAM\n", (int)(dev_priv->vram_size >> 20));
 	if (dev_priv->vram_sys_base) {
 		NV_INFO(dev, "Stolen system memory at: 0x%010llx\n",
@@ -479,7 +475,7 @@ nouveau_mem_vram_init(struct drm_device *dev)
 	}
 
 	if (dev_priv->card_type < NV_50) {
-		ret = nouveau_bo_new(dev, NULL, 256*1024, 0, TTM_PL_FLAG_VRAM,
+		ret = nouveau_bo_new(dev, 256*1024, 0, TTM_PL_FLAG_VRAM,
 				     0, 0, &dev_priv->vga_ram);
 		if (ret == 0)
 			ret = nouveau_bo_pin(dev_priv->vga_ram,
@@ -729,37 +725,31 @@ nouveau_mem_timing_fini(struct drm_device *dev)
 }
 
 static int
-nouveau_vram_manager_init(struct ttm_mem_type_manager *man, unsigned long p_size)
+nouveau_vram_manager_init(struct ttm_mem_type_manager *man, unsigned long psize)
 {
-	struct drm_nouveau_private *dev_priv = nouveau_bdev(man->bdev);
-	struct nouveau_mm *mm;
-	u64 size, block, rsvd;
-	int ret;
-
-	rsvd  = (256 * 1024); /* vga memory */
-	size  = (p_size << PAGE_SHIFT) - rsvd;
-	block = dev_priv->vram_rblock_size;
-
-	ret = nouveau_mm_init(&mm, rsvd >> 12, size >> 12, block >> 12);
-	if (ret)
-		return ret;
-
-	man->priv = mm;
+	/* nothing to do */
 	return 0;
 }
 
 static int
 nouveau_vram_manager_fini(struct ttm_mem_type_manager *man)
 {
-	struct nouveau_mm *mm = man->priv;
-	int ret;
+	/* nothing to do */
+	return 0;
+}
 
-	ret = nouveau_mm_fini(&mm);
-	if (ret)
-		return ret;
+static inline void
+nouveau_mem_node_cleanup(struct nouveau_mem *node)
+{
+	if (node->vma[0].node) {
+		nouveau_vm_unmap(&node->vma[0]);
+		nouveau_vm_put(&node->vma[0]);
+	}
 
-	man->priv = NULL;
-	return 0;
+	if (node->vma[1].node) {
+		nouveau_vm_unmap(&node->vma[1]);
+		nouveau_vm_put(&node->vma[1]);
+	}
 }
 
 static void
@@ -768,14 +758,9 @@ nouveau_vram_manager_del(struct ttm_mem_type_manager *man,
 {
 	struct drm_nouveau_private *dev_priv = nouveau_bdev(man->bdev);
 	struct nouveau_vram_engine *vram = &dev_priv->engine.vram;
-	struct nouveau_mem *node = mem->mm_node;
 	struct drm_device *dev = dev_priv->dev;
 
-	if (node->tmp_vma.node) {
-		nouveau_vm_unmap(&node->tmp_vma);
-		nouveau_vm_put(&node->tmp_vma);
-	}
-
+	nouveau_mem_node_cleanup(mem->mm_node);
 	vram->put(dev, (struct nouveau_mem **)&mem->mm_node);
 }
 
@@ -794,7 +779,7 @@ nouveau_vram_manager_new(struct ttm_mem_type_manager *man,
 	int ret;
 
 	if (nvbo->tile_flags & NOUVEAU_GEM_TILE_NONCONTIG)
-		size_nc = 1 << nvbo->vma.node->type;
+		size_nc = 1 << nvbo->page_shift;
 
 	ret = vram->get(dev, mem->num_pages << PAGE_SHIFT,
 			mem->page_alignment << PAGE_SHIFT, size_nc,
@@ -804,9 +789,7 @@ nouveau_vram_manager_new(struct ttm_mem_type_manager *man,
 		return (ret == -ENOSPC) ? 0 : ret;
 	}
 
-	node->page_shift = 12;
-	if (nvbo->vma.node)
-		node->page_shift = nvbo->vma.node->type;
+	node->page_shift = nvbo->page_shift;
 
 	mem->mm_node = node;
 	mem->start   = node->offset >> PAGE_SHIFT;
@@ -862,15 +845,9 @@ static void
 nouveau_gart_manager_del(struct ttm_mem_type_manager *man,
 			 struct ttm_mem_reg *mem)
 {
-	struct nouveau_mem *node = mem->mm_node;
-
-	if (node->tmp_vma.node) {
-		nouveau_vm_unmap(&node->tmp_vma);
-		nouveau_vm_put(&node->tmp_vma);
-	}
-
+	nouveau_mem_node_cleanup(mem->mm_node);
 	mem->mm_node = NULL;
-	kfree(node);
+	kfree(mem->mm_node);
 }
 
 static int
@@ -880,11 +857,7 @@ nouveau_gart_manager_new(struct ttm_mem_type_manager *man,
 			 struct ttm_mem_reg *mem)
 {
 	struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev);
-	struct nouveau_bo *nvbo = nouveau_bo(bo);
-	struct nouveau_vma *vma = &nvbo->vma;
-	struct nouveau_vm *vm = vma->vm;
 	struct nouveau_mem *node;
-	int ret;
 
 	if (unlikely((mem->num_pages << PAGE_SHIFT) >=
 		     dev_priv->gart_info.aper_size))
@@ -893,24 +866,8 @@ nouveau_gart_manager_new(struct ttm_mem_type_manager *man,
 	node = kzalloc(sizeof(*node), GFP_KERNEL);
 	if (!node)
 		return -ENOMEM;
+	node->page_shift = 12;
 
-	/* This node must be for evicting large-paged VRAM
-	 * to system memory.  Due to a nv50 limitation of
-	 * not being able to mix large/small pages within
-	 * the same PDE, we need to create a temporary
-	 * small-paged VMA for the eviction.
-	 */
-	if (vma->node->type != vm->spg_shift) {
-		ret = nouveau_vm_get(vm, (u64)vma->node->length << 12,
-				     vm->spg_shift, NV_MEM_ACCESS_RW,
-				     &node->tmp_vma);
-		if (ret) {
-			kfree(node);
-			return ret;
-		}
-	}
-
-	node->page_shift = nvbo->vma.node->type;
 	mem->mm_node = node;
 	mem->start   = 0;
 	return 0;

drivers/gpu/drm/nouveau/nouveau_mm.c

@@ -158,11 +158,18 @@ int
 nouveau_mm_fini(struct nouveau_mm **prmm)
 {
 	struct nouveau_mm *rmm = *prmm;
-	struct nouveau_mm_node *heap =
+	struct nouveau_mm_node *node, *heap =
 		list_first_entry(&rmm->nodes, struct nouveau_mm_node, nl_entry);
 
-	if (!list_is_singular(&rmm->nodes))
+	if (!list_is_singular(&rmm->nodes)) {
+		printk(KERN_ERR "nouveau_mm not empty at destroy time!\n");
+		list_for_each_entry(node, &rmm->nodes, nl_entry) {
+			printk(KERN_ERR "0x%02x: 0x%08x 0x%08x\n",
+			       node->type, node->offset, node->length);
+		}
+		WARN_ON(1);
 		return -EBUSY;
+	}
 
 	kfree(heap);
 	kfree(rmm);

drivers/gpu/drm/nouveau/nouveau_mm.h

@@ -52,6 +52,7 @@ int  nouveau_mm_get(struct nouveau_mm *, int type, u32 size, u32 size_nc,
 void nouveau_mm_put(struct nouveau_mm *, struct nouveau_mm_node *);
 
 int  nv50_vram_init(struct drm_device *);
+void nv50_vram_fini(struct drm_device *);
 int  nv50_vram_new(struct drm_device *, u64 size, u32 align, u32 size_nc,
 		    u32 memtype, struct nouveau_mem **);
 void nv50_vram_del(struct drm_device *, struct nouveau_mem **);

drivers/gpu/drm/nouveau/nouveau_notifier.c

@@ -34,6 +34,7 @@ int
 nouveau_notifier_init_channel(struct nouveau_channel *chan)
 {
 	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_bo *ntfy = NULL;
 	uint32_t flags, ttmpl;
 	int ret;
@@ -46,7 +47,7 @@ nouveau_notifier_init_channel(struct nouveau_channel *chan)
 		ttmpl = TTM_PL_FLAG_TT;
 	}
 
-	ret = nouveau_gem_new(dev, NULL, PAGE_SIZE, 0, flags, 0, 0, &ntfy);
+	ret = nouveau_gem_new(dev, PAGE_SIZE, 0, flags, 0, 0, &ntfy);
 	if (ret)
 		return ret;
 
@@ -58,14 +59,22 @@ nouveau_notifier_init_channel(struct nouveau_channel *chan)
 	if (ret)
 		goto out_err;
 
+	if (dev_priv->card_type >= NV_50) {
+		ret = nouveau_bo_vma_add(ntfy, chan->vm, &chan->notifier_vma);
+		if (ret)
+			goto out_err;
+	}
+
 	ret = drm_mm_init(&chan->notifier_heap, 0, ntfy->bo.mem.size);
 	if (ret)
 		goto out_err;
 
 	chan->notifier_bo = ntfy;
 out_err:
-	if (ret)
+	if (ret) {
+		nouveau_bo_vma_del(ntfy, &chan->notifier_vma);
 		drm_gem_object_unreference_unlocked(ntfy->gem);
+	}
 
 	return ret;
 }
@@ -78,6 +87,7 @@ nouveau_notifier_takedown_channel(struct nouveau_channel *chan)
 	if (!chan->notifier_bo)
 		return;
 
+	nouveau_bo_vma_del(chan->notifier_bo, &chan->notifier_vma);
 	nouveau_bo_unmap(chan->notifier_bo);
 	mutex_lock(&dev->struct_mutex);
 	nouveau_bo_unpin(chan->notifier_bo);
@@ -122,10 +132,10 @@ nouveau_notifier_alloc(struct nouveau_channel *chan, uint32_t handle,
 			target = NV_MEM_TARGET_VRAM;
 		else
 			target = NV_MEM_TARGET_GART;
-		offset  = chan->notifier_bo->bo.mem.start << PAGE_SHIFT;
+		offset  = chan->notifier_bo->bo.offset;
 	} else {
 		target = NV_MEM_TARGET_VM;
-		offset = chan->notifier_bo->vma.offset;
+		offset = chan->notifier_vma.offset;
 	}
 	offset += mem->start;
 
@@ -183,7 +193,7 @@ nouveau_ioctl_notifier_alloc(struct drm_device *dev, void *data,
 	if (unlikely(dev_priv->card_type >= NV_C0))
 		return -EINVAL;
 
-	chan = nouveau_channel_get(dev, file_priv, na->channel);
+	chan = nouveau_channel_get(file_priv, na->channel);
 	if (IS_ERR(chan))
 		return PTR_ERR(chan);
 

drivers/gpu/drm/nouveau/nouveau_object.c

@@ -125,7 +125,7 @@ nouveau_gpuobj_mthd_call2(struct drm_device *dev, int chid,
 	int ret = -EINVAL;
 
 	spin_lock_irqsave(&dev_priv->channels.lock, flags);
-	if (chid > 0 && chid < dev_priv->engine.fifo.channels)
+	if (chid >= 0 && chid < dev_priv->engine.fifo.channels)
 		chan = dev_priv->channels.ptr[chid];
 	if (chan)
 		ret = nouveau_gpuobj_mthd_call(chan, class, mthd, data);
@@ -191,7 +191,7 @@ nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan,
 	list_add_tail(&gpuobj->list, &dev_priv->gpuobj_list);
 	spin_unlock(&dev_priv->ramin_lock);
 
-	if (chan) {
+	if (!(flags & NVOBJ_FLAG_VM) && chan) {
 		ramin = drm_mm_search_free(&chan->ramin_heap, size, align, 0);
 		if (ramin)
 			ramin = drm_mm_get_block(ramin, size, align);
@@ -208,7 +208,7 @@ nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan,
 		gpuobj->vinst = ramin->start + chan->ramin->vinst;
 		gpuobj->node  = ramin;
 	} else {
-		ret = instmem->get(gpuobj, size, align);
+		ret = instmem->get(gpuobj, chan, size, align);
 		if (ret) {
 			nouveau_gpuobj_ref(NULL, &gpuobj);
 			return ret;
@@ -690,36 +690,65 @@ nouveau_gpuobj_channel_init_pramin(struct nouveau_channel *chan)
 	return 0;
 }
 
-int
-nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
-			    uint32_t vram_h, uint32_t tt_h)
+static int
+nvc0_gpuobj_channel_init(struct nouveau_channel *chan, struct nouveau_vm *vm)
 {
 	struct drm_device *dev = chan->dev;
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj *vram = NULL, *tt = NULL;
-	int ret, i;
-
-	NV_DEBUG(dev, "ch%d vram=0x%08x tt=0x%08x\n", chan->id, vram_h, tt_h);
-
-	if (dev_priv->card_type == NV_C0) {
-		struct nouveau_vm *vm = dev_priv->chan_vm;
+	struct nouveau_gpuobj *pgd = NULL;
 	struct nouveau_vm_pgd *vpgd;
+	int ret, i;
 
-		ret = nouveau_gpuobj_new(dev, NULL, 4096, 0x1000, 0,
-					 &chan->ramin);
+	ret = nouveau_gpuobj_new(dev, NULL, 4096, 0x1000, 0, &chan->ramin);
 	if (ret)
 		return ret;
 
-		nouveau_vm_ref(vm, &chan->vm, NULL);
+	/* create page directory for this vm if none currently exists,
+	 * will be destroyed automagically when last reference to the
+	 * vm is removed
+	 */
+	if (list_empty(&vm->pgd_list)) {
+		ret = nouveau_gpuobj_new(dev, NULL, 65536, 0x1000, 0, &pgd);
+		if (ret)
+			return ret;
+	}
+	nouveau_vm_ref(vm, &chan->vm, pgd);
+	nouveau_gpuobj_ref(NULL, &pgd);
 
+	/* point channel at vm's page directory */
 	vpgd = list_first_entry(&vm->pgd_list, struct nouveau_vm_pgd, head);
 	nv_wo32(chan->ramin, 0x0200, lower_32_bits(vpgd->obj->vinst));
 	nv_wo32(chan->ramin, 0x0204, upper_32_bits(vpgd->obj->vinst));
 	nv_wo32(chan->ramin, 0x0208, 0xffffffff);
 	nv_wo32(chan->ramin, 0x020c, 0x000000ff);
-		return 0;
+
+	/* map display semaphore buffers into channel's vm */
+	for (i = 0; i < 2; i++) {
+		struct nv50_display_crtc *dispc = &nv50_display(dev)->crtc[i];
+
+		ret = nouveau_bo_vma_add(dispc->sem.bo, chan->vm,
+					 &chan->dispc_vma[i]);
+		if (ret)
+			return ret;
 	}
 
+	return 0;
+}
+
+int
+nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
+			    uint32_t vram_h, uint32_t tt_h)
+{
+	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_fpriv *fpriv = nouveau_fpriv(chan->file_priv);
+	struct nouveau_vm *vm = fpriv ? fpriv->vm : dev_priv->chan_vm;
+	struct nouveau_gpuobj *vram = NULL, *tt = NULL;
+	int ret, i;
+
+	NV_DEBUG(dev, "ch%d vram=0x%08x tt=0x%08x\n", chan->id, vram_h, tt_h);
+	if (dev_priv->card_type == NV_C0)
+		return nvc0_gpuobj_channel_init(chan, vm);
+
 	/* Allocate a chunk of memory for per-channel object storage */
 	ret = nouveau_gpuobj_channel_init_pramin(chan);
 	if (ret) {
@@ -731,7 +760,7 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 	 *  - Allocate per-channel page-directory
 	 *  - Link with shared channel VM
 	 */
-	if (dev_priv->chan_vm) {
+	if (vm) {
 		u32 pgd_offs = (dev_priv->chipset == 0x50) ? 0x1400 : 0x0200;
 		u64 vm_vinst = chan->ramin->vinst + pgd_offs;
 		u32 vm_pinst = chan->ramin->pinst;
@@ -744,7 +773,7 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 		if (ret)
 			return ret;
 
-		nouveau_vm_ref(dev_priv->chan_vm, &chan->vm, chan->vm_pd);
+		nouveau_vm_ref(vm, &chan->vm, chan->vm_pd);
 	}
 
 	/* RAMHT */
@@ -768,7 +797,7 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 			struct nouveau_gpuobj *sem = NULL;
 			struct nv50_display_crtc *dispc =
 				&nv50_display(dev)->crtc[i];
-			u64 offset = dispc->sem.bo->bo.mem.start << PAGE_SHIFT;
+			u64 offset = dispc->sem.bo->bo.offset;
 
 			ret = nouveau_gpuobj_dma_new(chan, 0x3d, offset, 0xfff,
 						     NV_MEM_ACCESS_RW,
@@ -841,13 +870,22 @@ void
 nouveau_gpuobj_channel_takedown(struct nouveau_channel *chan)
 {
 	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	int i;
 
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
-	nouveau_ramht_ref(NULL, &chan->ramht, chan);
+	if (dev_priv->card_type >= NV_50) {
+		struct nv50_display *disp = nv50_display(dev);
+
+		for (i = 0; i < 2; i++) {
+			struct nv50_display_crtc *dispc = &disp->crtc[i];
+			nouveau_bo_vma_del(dispc->sem.bo, &chan->dispc_vma[i]);
+		}
 
 		nouveau_vm_ref(NULL, &chan->vm, chan->vm_pd);
 		nouveau_gpuobj_ref(NULL, &chan->vm_pd);
+	}
 
 	if (drm_mm_initialized(&chan->ramin_heap))
 		drm_mm_takedown(&chan->ramin_heap);
@@ -909,7 +947,7 @@ int nouveau_ioctl_grobj_alloc(struct drm_device *dev, void *data,
 	if (init->handle == ~0)
 		return -EINVAL;
 
-	chan = nouveau_channel_get(dev, file_priv, init->channel);
+	chan = nouveau_channel_get(file_priv, init->channel);
 	if (IS_ERR(chan))
 		return PTR_ERR(chan);
 
@@ -936,7 +974,7 @@ int nouveau_ioctl_gpuobj_free(struct drm_device *dev, void *data,
 	struct nouveau_channel *chan;
 	int ret;
 
-	chan = nouveau_channel_get(dev, file_priv, objfree->channel);
+	chan = nouveau_channel_get(file_priv, objfree->channel);
 	if (IS_ERR(chan))
 		return PTR_ERR(chan);
 

drivers/gpu/drm/nouveau/nouveau_state.c

@@ -91,6 +91,7 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clock_pre		= nv04_pm_clock_pre;
 		engine->pm.clock_set		= nv04_pm_clock_set;
 		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.takedown		= nouveau_stub_takedown;
 		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
 	case 0x10:
@@ -139,6 +140,7 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clock_pre		= nv04_pm_clock_pre;
 		engine->pm.clock_set		= nv04_pm_clock_set;
 		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.takedown		= nouveau_stub_takedown;
 		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
 	case 0x20:
@@ -187,6 +189,7 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clock_pre		= nv04_pm_clock_pre;
 		engine->pm.clock_set		= nv04_pm_clock_set;
 		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.takedown		= nouveau_stub_takedown;
 		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
 	case 0x30:
@@ -237,6 +240,7 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.voltage_get		= nouveau_voltage_gpio_get;
 		engine->pm.voltage_set		= nouveau_voltage_gpio_set;
 		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.takedown		= nouveau_stub_takedown;
 		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
 	case 0x40:
@@ -289,6 +293,7 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.voltage_set		= nouveau_voltage_gpio_set;
 		engine->pm.temp_get		= nv40_temp_get;
 		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.takedown		= nouveau_stub_takedown;
 		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
 	case 0x50:
@@ -366,6 +371,7 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		else
 			engine->pm.temp_get	= nv40_temp_get;
 		engine->vram.init		= nv50_vram_init;
+		engine->vram.takedown		= nv50_vram_fini;
 		engine->vram.get		= nv50_vram_new;
 		engine->vram.put		= nv50_vram_del;
 		engine->vram.flags_valid	= nv50_vram_flags_valid;
@@ -412,9 +418,11 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->gpio.irq_unregister	= nv50_gpio_irq_unregister;
 		engine->gpio.irq_enable		= nv50_gpio_irq_enable;
 		engine->vram.init		= nvc0_vram_init;
+		engine->vram.takedown		= nv50_vram_fini;
 		engine->vram.get		= nvc0_vram_new;
 		engine->vram.put		= nv50_vram_del;
 		engine->vram.flags_valid	= nvc0_vram_flags_valid;
+		engine->pm.temp_get		= nv84_temp_get;
 		break;
 	default:
 		NV_ERROR(dev, "NV%02x unsupported\n", dev_priv->chipset);
@@ -448,8 +456,8 @@ nouveau_card_init_channel(struct drm_device *dev)
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	int ret;
 
-	ret = nouveau_channel_alloc(dev, &dev_priv->channel,
-				    (struct drm_file *)-2, NvDmaFB, NvDmaTT);
+	ret = nouveau_channel_alloc(dev, &dev_priv->channel, NULL,
+				    NvDmaFB, NvDmaTT);
 	if (ret)
 		return ret;
 
@@ -528,7 +536,7 @@ nouveau_card_init(struct drm_device *dev)
 
 	nouveau_pm_init(dev);
 
-	ret = nouveau_mem_vram_init(dev);
+	ret = engine->vram.init(dev);
 	if (ret)
 		goto out_bios;
 
@@ -540,10 +548,14 @@ nouveau_card_init(struct drm_device *dev)
 	if (ret)
 		goto out_gpuobj;
 
-	ret = nouveau_mem_gart_init(dev);
+	ret = nouveau_mem_vram_init(dev);
 	if (ret)
 		goto out_instmem;
 
+	ret = nouveau_mem_gart_init(dev);
+	if (ret)
+		goto out_ttmvram;
+
 	/* PMC */
 	ret = engine->mc.init(dev);
 	if (ret)
@@ -564,7 +576,7 @@ nouveau_card_init(struct drm_device *dev)
 	if (ret)
 		goto out_timer;
 
-	if (!nouveau_noaccel) {
+	if (!dev_priv->noaccel) {
 		switch (dev_priv->card_type) {
 		case NV_04:
 			nv04_graph_create(dev);
@@ -676,10 +688,10 @@ nouveau_card_init(struct drm_device *dev)
 	drm_vblank_cleanup(dev);
 	engine->display.destroy(dev);
 out_fifo:
-	if (!nouveau_noaccel)
+	if (!dev_priv->noaccel)
 		engine->fifo.takedown(dev);
 out_engine:
-	if (!nouveau_noaccel) {
+	if (!dev_priv->noaccel) {
 		for (e = e - 1; e >= 0; e--) {
 			if (!dev_priv->eng[e])
 				continue;
@@ -697,12 +709,14 @@ nouveau_card_init(struct drm_device *dev)
 	engine->mc.takedown(dev);
 out_gart:
 	nouveau_mem_gart_fini(dev);
+out_ttmvram:
+	nouveau_mem_vram_fini(dev);
 out_instmem:
 	engine->instmem.takedown(dev);
 out_gpuobj:
 	nouveau_gpuobj_takedown(dev);
 out_vram:
-	nouveau_mem_vram_fini(dev);
+	engine->vram.takedown(dev);
 out_bios:
 	nouveau_pm_fini(dev);
 	nouveau_bios_takedown(dev);
@@ -719,12 +733,17 @@ static void nouveau_card_takedown(struct drm_device *dev)
 	struct nouveau_engine *engine = &dev_priv->engine;
 	int e;
 
+	drm_kms_helper_poll_fini(dev);
+	nouveau_fbcon_fini(dev);
+
 	if (dev_priv->channel) {
-		nouveau_fence_fini(dev);
 		nouveau_channel_put_unlocked(&dev_priv->channel);
+		nouveau_fence_fini(dev);
 	}
 
-	if (!nouveau_noaccel) {
+	engine->display.destroy(dev);
+
+	if (!dev_priv->noaccel) {
 		engine->fifo.takedown(dev);
 		for (e = NVOBJ_ENGINE_NR - 1; e >= 0; e--) {
 			if (dev_priv->eng[e]) {
@@ -749,10 +768,11 @@ static void nouveau_card_takedown(struct drm_device *dev)
 	ttm_bo_clean_mm(&dev_priv->ttm.bdev, TTM_PL_TT);
 	mutex_unlock(&dev->struct_mutex);
 	nouveau_mem_gart_fini(dev);
+	nouveau_mem_vram_fini(dev);
 
 	engine->instmem.takedown(dev);
 	nouveau_gpuobj_takedown(dev);
-	nouveau_mem_vram_fini(dev);
+	engine->vram.takedown(dev);
 
 	nouveau_irq_fini(dev);
 	drm_vblank_cleanup(dev);
@@ -763,6 +783,41 @@ static void nouveau_card_takedown(struct drm_device *dev)
 	vga_client_register(dev->pdev, NULL, NULL, NULL);
 }
 
+int
+nouveau_open(struct drm_device *dev, struct drm_file *file_priv)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_fpriv *fpriv;
+	int ret;
+
+	fpriv = kzalloc(sizeof(*fpriv), GFP_KERNEL);
+	if (unlikely(!fpriv))
+		return -ENOMEM;
+
+	spin_lock_init(&fpriv->lock);
+	INIT_LIST_HEAD(&fpriv->channels);
+
+	if (dev_priv->card_type == NV_50) {
+		ret = nouveau_vm_new(dev, 0, (1ULL << 40), 0x0020000000ULL,
+				     &fpriv->vm);
+		if (ret) {
+			kfree(fpriv);
+			return ret;
+		}
+	} else
+	if (dev_priv->card_type >= NV_C0) {
+		ret = nouveau_vm_new(dev, 0, (1ULL << 40), 0x0008000000ULL,
+				     &fpriv->vm);
+		if (ret) {
+			kfree(fpriv);
+			return ret;
+		}
+	}
+
+	file_priv->driver_priv = fpriv;
+	return 0;
+}
+
 /* here a client dies, release the stuff that was allocated for its
  * file_priv */
 void nouveau_preclose(struct drm_device *dev, struct drm_file *file_priv)
@@ -770,6 +825,14 @@ void nouveau_preclose(struct drm_device *dev, struct drm_file *file_priv)
 	nouveau_channel_cleanup(dev, file_priv);
 }
 
+void
+nouveau_postclose(struct drm_device *dev, struct drm_file *file_priv)
+{
+	struct nouveau_fpriv *fpriv = nouveau_fpriv(file_priv);
+	nouveau_vm_ref(NULL, &fpriv->vm, NULL);
+	kfree(fpriv);
+}
+
 /* first module load, setup the mmio/fb mapping */
 /* KMS: we need mmio at load time, not when the first drm client opens. */
 int nouveau_firstopen(struct drm_device *dev)
@@ -935,6 +998,25 @@ int nouveau_load(struct drm_device *dev, unsigned long flags)
 	NV_INFO(dev, "Detected an NV%2x generation card (0x%08x)\n",
 		dev_priv->card_type, reg0);
 
+	/* Determine whether we'll attempt acceleration or not, some
+	 * cards are disabled by default here due to them being known
+	 * non-functional, or never been tested due to lack of hw.
+	 */
+	dev_priv->noaccel = !!nouveau_noaccel;
+	if (nouveau_noaccel == -1) {
+		switch (dev_priv->chipset) {
+		case 0xc1: /* known broken */
+		case 0xc8: /* never tested */
+			NV_INFO(dev, "acceleration disabled by default, pass "
+				     "noaccel=0 to force enable\n");
+			dev_priv->noaccel = true;
+			break;
+		default:
+			dev_priv->noaccel = false;
+			break;
+		}
+	}
+
 	ret = nouveau_remove_conflicting_drivers(dev);
 	if (ret)
 		goto err_mmio;
@@ -999,11 +1081,7 @@ void nouveau_lastclose(struct drm_device *dev)
 int nouveau_unload(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_engine *engine = &dev_priv->engine;
 
-	drm_kms_helper_poll_fini(dev);
-	nouveau_fbcon_fini(dev);
-	engine->display.destroy(dev);
 	nouveau_card_takedown(dev);
 
 	iounmap(dev_priv->mmio);

drivers/gpu/drm/nouveau/nouveau_temp.c

@@ -43,7 +43,7 @@ nouveau_temp_vbios_parse(struct drm_device *dev, u8 *temp)
 
 	/* Set the default sensor's contants */
 	sensor->offset_constant = 0;
-	sensor->offset_mult = 1;
+	sensor->offset_mult = 0;
 	sensor->offset_div = 1;
 	sensor->slope_mult = 1;
 	sensor->slope_div = 1;
@@ -99,6 +99,13 @@ nouveau_temp_vbios_parse(struct drm_device *dev, u8 *temp)
 			sensor->slope_mult = 431;
 			sensor->slope_div = 10000;
 			break;
+
+		case 0x67:
+			sensor->offset_mult = -26149;
+			sensor->offset_div = 100;
+			sensor->slope_mult = 484;
+			sensor->slope_div = 10000;
+			break;
 		}
 	}
 
@@ -109,7 +116,7 @@ nouveau_temp_vbios_parse(struct drm_device *dev, u8 *temp)
 
 	/* Read the entries from the table */
 	for (i = 0; i < entries; i++) {
-		u16 value = ROM16(temp[1]);
+		s16 value = ROM16(temp[1]);
 
 		switch (temp[0]) {
 		case 0x01:
@@ -160,8 +167,8 @@ nv40_sensor_setup(struct drm_device *dev)
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_pm_engine *pm = &dev_priv->engine.pm;
 	struct nouveau_pm_temp_sensor_constants *sensor = &pm->sensor_constants;
-	u32 offset = sensor->offset_mult / sensor->offset_div;
-	u32 sensor_calibration;
+	s32 offset = sensor->offset_mult / sensor->offset_div;
+	s32 sensor_calibration;
 
 	/* set up the sensors */
 	sensor_calibration = 120 - offset - sensor->offset_constant;

drivers/gpu/drm/nouveau/nouveau_vm.c

@@ -369,23 +369,26 @@ nouveau_vm_link(struct nouveau_vm *vm, struct nouveau_gpuobj *pgd)
 }
 
 static void
-nouveau_vm_unlink(struct nouveau_vm *vm, struct nouveau_gpuobj *pgd)
+nouveau_vm_unlink(struct nouveau_vm *vm, struct nouveau_gpuobj *mpgd)
 {
 	struct nouveau_vm_pgd *vpgd, *tmp;
+	struct nouveau_gpuobj *pgd = NULL;
 
-	if (!pgd)
+	if (!mpgd)
 		return;
 
 	mutex_lock(&vm->mm->mutex);
 	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
-		if (vpgd->obj != pgd)
-			continue;
-
+		if (vpgd->obj == mpgd) {
+			pgd = vpgd->obj;
 			list_del(&vpgd->head);
-		nouveau_gpuobj_ref(NULL, &vpgd->obj);
 			kfree(vpgd);
+			break;
+		}
 	}
 	mutex_unlock(&vm->mm->mutex);
+
+	nouveau_gpuobj_ref(NULL, &pgd);
 }
 
 static void
@@ -396,8 +399,8 @@ nouveau_vm_del(struct nouveau_vm *vm)
 	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
 		nouveau_vm_unlink(vm, vpgd->obj);
 	}
-	WARN_ON(nouveau_mm_fini(&vm->mm) != 0);
 
+	nouveau_mm_fini(&vm->mm);
 	kfree(vm->pgt);
 	kfree(vm);
 }

drivers/gpu/drm/nouveau/nouveau_vm.h

@@ -41,6 +41,8 @@ struct nouveau_vm_pgd {
 };
 
 struct nouveau_vma {
+	struct list_head head;
+	int refcount;
 	struct nouveau_vm *vm;
 	struct nouveau_mm_node *node;
 	u64 offset;

drivers/gpu/drm/nouveau/nv04_crtc.c

@@ -1035,7 +1035,7 @@ nv04_crtc_create(struct drm_device *dev, int crtc_num)
 	drm_crtc_helper_add(&nv_crtc->base, &nv04_crtc_helper_funcs);
 	drm_mode_crtc_set_gamma_size(&nv_crtc->base, 256);
 
-	ret = nouveau_bo_new(dev, NULL, 64*64*4, 0x100, TTM_PL_FLAG_VRAM,
+	ret = nouveau_bo_new(dev, 64*64*4, 0x100, TTM_PL_FLAG_VRAM,
 			     0, 0x0000, &nv_crtc->cursor.nvbo);
 	if (!ret) {
 		ret = nouveau_bo_pin(nv_crtc->cursor.nvbo, TTM_PL_FLAG_VRAM);

drivers/gpu/drm/nouveau/nv04_instmem.c

@@ -112,7 +112,8 @@ nv04_instmem_resume(struct drm_device *dev)
 }
 
 int
-nv04_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align)
+nv04_instmem_get(struct nouveau_gpuobj *gpuobj, struct nouveau_channel *chan,
+		 u32 size, u32 align)
 {
 	struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
 	struct drm_mm_node *ramin = NULL;

drivers/gpu/drm/nouveau/nv50_crtc.c

@@ -104,7 +104,7 @@ nv50_crtc_blank(struct nouveau_crtc *nv_crtc, bool blanked)
 		OUT_RING(evo, nv_crtc->lut.depth == 8 ?
 				NV50_EVO_CRTC_CLUT_MODE_OFF :
 				NV50_EVO_CRTC_CLUT_MODE_ON);
-		OUT_RING(evo, (nv_crtc->lut.nvbo->bo.mem.start << PAGE_SHIFT) >> 8);
+		OUT_RING(evo, nv_crtc->lut.nvbo->bo.offset >> 8);
 		if (dev_priv->chipset != 0x50) {
 			BEGIN_RING(evo, 0, NV84_EVO_CRTC(index, CLUT_DMA), 1);
 			OUT_RING(evo, NvEvoVRAM);
@@ -372,7 +372,7 @@ nv50_crtc_cursor_set(struct drm_crtc *crtc, struct drm_file *file_priv,
 
 	nouveau_bo_unmap(cursor);
 
-	nv_crtc->cursor.set_offset(nv_crtc, nv_crtc->cursor.nvbo->bo.mem.start << PAGE_SHIFT);
+	nv_crtc->cursor.set_offset(nv_crtc, nv_crtc->cursor.nvbo->bo.offset);
 	nv_crtc->cursor.show(nv_crtc, true);
 
 out:
@@ -546,7 +546,7 @@ nv50_crtc_do_mode_set_base(struct drm_crtc *crtc,
 		}
 	}
 
-	nv_crtc->fb.offset = fb->nvbo->bo.mem.start << PAGE_SHIFT;
+	nv_crtc->fb.offset = fb->nvbo->bo.offset;
 	nv_crtc->fb.tile_flags = nouveau_bo_tile_layout(fb->nvbo);
 	nv_crtc->fb.cpp = drm_fb->bits_per_pixel / 8;
 	if (!nv_crtc->fb.blanked && dev_priv->chipset != 0x50) {
@@ -747,7 +747,7 @@ nv50_crtc_create(struct drm_device *dev, int index)
 	}
 	nv_crtc->lut.depth = 0;
 
-	ret = nouveau_bo_new(dev, NULL, 4096, 0x100, TTM_PL_FLAG_VRAM,
+	ret = nouveau_bo_new(dev, 4096, 0x100, TTM_PL_FLAG_VRAM,
 			     0, 0x0000, &nv_crtc->lut.nvbo);
 	if (!ret) {
 		ret = nouveau_bo_pin(nv_crtc->lut.nvbo, TTM_PL_FLAG_VRAM);
@@ -773,7 +773,7 @@ nv50_crtc_create(struct drm_device *dev, int index)
 	drm_crtc_helper_add(&nv_crtc->base, &nv50_crtc_helper_funcs);
 	drm_mode_crtc_set_gamma_size(&nv_crtc->base, 256);
 
-	ret = nouveau_bo_new(dev, NULL, 64*64*4, 0x100, TTM_PL_FLAG_VRAM,
+	ret = nouveau_bo_new(dev, 64*64*4, 0x100, TTM_PL_FLAG_VRAM,
 			     0, 0x0000, &nv_crtc->cursor.nvbo);
 	if (!ret) {
 		ret = nouveau_bo_pin(nv_crtc->cursor.nvbo, TTM_PL_FLAG_VRAM);

drivers/gpu/drm/nouveau/nv50_display.c

@@ -415,8 +415,6 @@ nv50_display_flip_next(struct drm_crtc *crtc, struct drm_framebuffer *fb,
 
 	/* synchronise with the rendering channel, if necessary */
 	if (likely(chan)) {
-		u64 offset = dispc->sem.bo->vma.offset + dispc->sem.offset;
-
 		ret = RING_SPACE(chan, 10);
 		if (ret) {
 			WIND_RING(evo);
@@ -438,6 +436,8 @@ nv50_display_flip_next(struct drm_crtc *crtc, struct drm_framebuffer *fb,
 			else
 				OUT_RING  (chan, chan->vram_handle);
 		} else {
+			u64 offset = chan->dispc_vma[nv_crtc->index].offset;
+			offset += dispc->sem.offset;
 			BEGIN_NVC0(chan, 2, NvSubM2MF, 0x0010, 4);
 			OUT_RING  (chan, upper_32_bits(offset));
 			OUT_RING  (chan, lower_32_bits(offset));
@@ -484,7 +484,7 @@ nv50_display_flip_next(struct drm_crtc *crtc, struct drm_framebuffer *fb,
 	OUT_RING  (evo, 0x00000000);
 	OUT_RING  (evo, 0x00000000);
 	BEGIN_RING(evo, 0, 0x0800, 5);
-	OUT_RING  (evo, (nv_fb->nvbo->bo.mem.start << PAGE_SHIFT) >> 8);
+	OUT_RING  (evo, nv_fb->nvbo->bo.offset >> 8);
 	OUT_RING  (evo, 0);
 	OUT_RING  (evo, (fb->height << 16) | fb->width);
 	OUT_RING  (evo, nv_fb->r_pitch);

drivers/gpu/drm/nouveau/nv50_evo.c

@@ -38,6 +38,7 @@ nv50_evo_channel_del(struct nouveau_channel **pevo)
 		return;
 	*pevo = NULL;
 
+	nouveau_ramht_ref(NULL, &evo->ramht, evo);
 	nouveau_gpuobj_channel_takedown(evo);
 	nouveau_bo_unmap(evo->pushbuf_bo);
 	nouveau_bo_ref(NULL, &evo->pushbuf_bo);
@@ -116,7 +117,7 @@ nv50_evo_channel_new(struct drm_device *dev, int chid,
 	evo->user_get = 4;
 	evo->user_put = 0;
 
-	ret = nouveau_bo_new(dev, NULL, 4096, 0, TTM_PL_FLAG_VRAM, 0, 0,
+	ret = nouveau_bo_new(dev, 4096, 0, TTM_PL_FLAG_VRAM, 0, 0,
 			     &evo->pushbuf_bo);
 	if (ret == 0)
 		ret = nouveau_bo_pin(evo->pushbuf_bo, TTM_PL_FLAG_VRAM);
@@ -153,7 +154,7 @@ nv50_evo_channel_init(struct nouveau_channel *evo)
 {
 	struct drm_device *dev = evo->dev;
 	int id = evo->id, ret, i;
-	u64 pushbuf = evo->pushbuf_bo->bo.mem.start << PAGE_SHIFT;
+	u64 pushbuf = evo->pushbuf_bo->bo.offset;
 	u32 tmp;
 
 	tmp = nv_rd32(dev, NV50_PDISPLAY_EVO_CTRL(id));
@@ -331,16 +332,15 @@ nv50_evo_create(struct drm_device *dev)
 		if (ret)
 			goto err;
 
-		ret = nouveau_bo_new(dev, NULL, 4096, 0x1000, TTM_PL_FLAG_VRAM,
+		ret = nouveau_bo_new(dev, 4096, 0x1000, TTM_PL_FLAG_VRAM,
 				     0, 0x0000, &dispc->sem.bo);
 		if (!ret) {
-			offset = dispc->sem.bo->bo.mem.start << PAGE_SHIFT;
-
 			ret = nouveau_bo_pin(dispc->sem.bo, TTM_PL_FLAG_VRAM);
 			if (!ret)
 				ret = nouveau_bo_map(dispc->sem.bo);
 			if (ret)
 				nouveau_bo_ref(NULL, &dispc->sem.bo);
+			offset = dispc->sem.bo->bo.offset;
 		}
 
 		if (ret)

drivers/gpu/drm/nouveau/nv50_fbcon.c

@@ -159,7 +159,7 @@ nv50_fbcon_accel_init(struct fb_info *info)
 	struct drm_device *dev = nfbdev->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_channel *chan = dev_priv->channel;
-	struct nouveau_bo *nvbo = nfbdev->nouveau_fb.nvbo;
+	struct nouveau_framebuffer *fb = &nfbdev->nouveau_fb;
 	int ret, format;
 
 	switch (info->var.bits_per_pixel) {
@@ -247,8 +247,8 @@ nv50_fbcon_accel_init(struct fb_info *info)
 	OUT_RING(chan, info->fix.line_length);
 	OUT_RING(chan, info->var.xres_virtual);
 	OUT_RING(chan, info->var.yres_virtual);
-	OUT_RING(chan, upper_32_bits(nvbo->vma.offset));
-	OUT_RING(chan, lower_32_bits(nvbo->vma.offset));
+	OUT_RING(chan, upper_32_bits(fb->vma.offset));
+	OUT_RING(chan, lower_32_bits(fb->vma.offset));
 	BEGIN_RING(chan, NvSub2D, 0x0230, 2);
 	OUT_RING(chan, format);
 	OUT_RING(chan, 1);
@@ -256,8 +256,8 @@ nv50_fbcon_accel_init(struct fb_info *info)
 	OUT_RING(chan, info->fix.line_length);
 	OUT_RING(chan, info->var.xres_virtual);
 	OUT_RING(chan, info->var.yres_virtual);
-	OUT_RING(chan, upper_32_bits(nvbo->vma.offset));
-	OUT_RING(chan, lower_32_bits(nvbo->vma.offset));
+	OUT_RING(chan, upper_32_bits(fb->vma.offset));
+	OUT_RING(chan, lower_32_bits(fb->vma.offset));
 
 	return 0;
 }

drivers/gpu/drm/nouveau/nv50_instmem.c

@@ -305,9 +305,9 @@ struct nv50_gpuobj_node {
 	u32 align;
 };
 
-
 int
-nv50_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align)
+nv50_instmem_get(struct nouveau_gpuobj *gpuobj, struct nouveau_channel *chan,
+		 u32 size, u32 align)
 {
 	struct drm_device *dev = gpuobj->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
@@ -336,7 +336,7 @@ nv50_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align)
 		if (!(gpuobj->flags & NVOBJ_FLAG_VM_USER))
 			flags |= NV_MEM_ACCESS_SYS;
 
-		ret = nouveau_vm_get(dev_priv->chan_vm, size, 12, flags,
+		ret = nouveau_vm_get(chan->vm, size, 12, flags,
 				     &node->chan_vma);
 		if (ret) {
 			vram->put(dev, &node->vram);
@@ -345,7 +345,7 @@ nv50_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align)
 		}
 
 		nouveau_vm_map(&node->chan_vma, node->vram);
-		gpuobj->vinst = node->chan_vma.offset;
+		gpuobj->linst = node->chan_vma.offset;
 	}
 
 	gpuobj->size = size;

drivers/gpu/drm/nouveau/nv50_vm.c

@@ -156,7 +156,7 @@ nv50_vm_flush(struct nouveau_vm *vm)
 	pinstmem->flush(vm->dev);
 
 	/* BAR */
-	if (vm != dev_priv->chan_vm) {
+	if (vm == dev_priv->bar1_vm || vm == dev_priv->bar3_vm) {
 		nv50_vm_flush_engine(vm->dev, 6);
 		return;
 	}

drivers/gpu/drm/nouveau/nv50_vram.c

@@ -51,9 +51,7 @@ void
 nv50_vram_del(struct drm_device *dev, struct nouveau_mem **pmem)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct ttm_bo_device *bdev = &dev_priv->ttm.bdev;
-	struct ttm_mem_type_manager *man = &bdev->man[TTM_PL_VRAM];
-	struct nouveau_mm *mm = man->priv;
+	struct nouveau_mm *mm = dev_priv->engine.vram.mm;
 	struct nouveau_mm_node *this;
 	struct nouveau_mem *mem;
 
@@ -84,9 +82,7 @@ nv50_vram_new(struct drm_device *dev, u64 size, u32 align, u32 size_nc,
 	      u32 memtype, struct nouveau_mem **pmem)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct ttm_bo_device *bdev = &dev_priv->ttm.bdev;
-	struct ttm_mem_type_manager *man = &bdev->man[TTM_PL_VRAM];
-	struct nouveau_mm *mm = man->priv;
+	struct nouveau_mm *mm = dev_priv->engine.vram.mm;
 	struct nouveau_mm_node *r;
 	struct nouveau_mem *mem;
 	int comp = (memtype & 0x300) >> 8;
@@ -190,22 +186,35 @@ int
 nv50_vram_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_vram_engine *vram = &dev_priv->engine.vram;
+	const u32 rsvd_head = ( 256 * 1024) >> 12; /* vga memory */
+	const u32 rsvd_tail = (1024 * 1024) >> 12; /* vbios etc */
+	u32 rblock, length;
 
 	dev_priv->vram_size  = nv_rd32(dev, 0x10020c);
 	dev_priv->vram_size |= (dev_priv->vram_size & 0xff) << 32;
 	dev_priv->vram_size &= 0xffffffff00ULL;
 
-	switch (dev_priv->chipset) {
-	case 0xaa:
-	case 0xac:
-	case 0xaf:
+	/* IGPs, no funky reordering happens here, they don't have VRAM */
+	if (dev_priv->chipset == 0xaa ||
+	    dev_priv->chipset == 0xac ||
+	    dev_priv->chipset == 0xaf) {
 		dev_priv->vram_sys_base = (u64)nv_rd32(dev, 0x100e10) << 12;
-		dev_priv->vram_rblock_size = 4096;
-		break;
-	default:
-		dev_priv->vram_rblock_size = nv50_vram_rblock(dev);
-		break;
+		rblock = 4096 >> 12;
+	} else {
+		rblock = nv50_vram_rblock(dev) >> 12;
 	}
 
-	return 0;
+	length = (dev_priv->vram_size >> 12) - rsvd_head - rsvd_tail;
+
+	return nouveau_mm_init(&vram->mm, rsvd_head, length, rblock);
+}
+
+void
+nv50_vram_fini(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_vram_engine *vram = &dev_priv->engine.vram;
+
+	nouveau_mm_fini(&vram->mm);
 }

drivers/gpu/drm/nouveau/nvc0_copy.c

@@ -48,14 +48,14 @@ nvc0_copy_context_new(struct nouveau_channel *chan, int engine)
 	struct nouveau_gpuobj *ctx = NULL;
 	int ret;
 
-	ret = nouveau_gpuobj_new(dev, NULL, 256, 256,
+	ret = nouveau_gpuobj_new(dev, chan, 256, 256,
 				 NVOBJ_FLAG_VM | NVOBJ_FLAG_VM_USER |
 				 NVOBJ_FLAG_ZERO_ALLOC, &ctx);
 	if (ret)
 		return ret;
 
-	nv_wo32(ramin, pcopy->ctx + 0, lower_32_bits(ctx->vinst));
-	nv_wo32(ramin, pcopy->ctx + 4, upper_32_bits(ctx->vinst));
+	nv_wo32(ramin, pcopy->ctx + 0, lower_32_bits(ctx->linst));
+	nv_wo32(ramin, pcopy->ctx + 4, upper_32_bits(ctx->linst));
 	dev_priv->engine.instmem.flush(dev);
 
 	chan->engctx[engine] = ctx;

drivers/gpu/drm/nouveau/nvc0_fb.c

 /*
- * Copyright 2010 Red Hat Inc.
+ * Copyright 2011 Red Hat Inc.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -23,16 +23,80 @@
  */
 
 #include "drmP.h"
-
+#include "drm.h"
 #include "nouveau_drv.h"
+#include "nouveau_drm.h"
+
+struct nvc0_fb_priv {
+	struct page *r100c10_page;
+	dma_addr_t r100c10;
+};
+
+static void
+nvc0_fb_destroy(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
+	struct nvc0_fb_priv *priv = pfb->priv;
+
+	if (priv->r100c10_page) {
+		pci_unmap_page(dev->pdev, priv->r100c10, PAGE_SIZE,
+			       PCI_DMA_BIDIRECTIONAL);
+		__free_page(priv->r100c10_page);
+	}
+
+	kfree(priv);
+	pfb->priv = NULL;
+}
+
+static int
+nvc0_fb_create(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
+	struct nvc0_fb_priv *priv;
+
+	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
+	if (!priv)
+		return -ENOMEM;
+	pfb->priv = priv;
+
+	priv->r100c10_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
+	if (!priv->r100c10_page) {
+		nvc0_fb_destroy(dev);
+		return -ENOMEM;
+	}
+
+	priv->r100c10 = pci_map_page(dev->pdev, priv->r100c10_page, 0,
+				     PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
+	if (pci_dma_mapping_error(dev->pdev, priv->r100c10)) {
+		nvc0_fb_destroy(dev);
+		return -EFAULT;
+	}
+
+	return 0;
+}
 
 int
 nvc0_fb_init(struct drm_device *dev)
 {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nvc0_fb_priv *priv;
+	int ret;
+
+	if (!dev_priv->engine.fb.priv) {
+		ret = nvc0_fb_create(dev);
+		if (ret)
+			return ret;
+	}
+	priv = dev_priv->engine.fb.priv;
+
+	nv_wr32(dev, 0x100c10, priv->r100c10 >> 8);
 	return 0;
 }
 
 void
 nvc0_fb_takedown(struct drm_device *dev)
 {
+	nvc0_fb_destroy(dev);
 }

drivers/gpu/drm/nouveau/nvc0_fbcon.c

@@ -159,7 +159,7 @@ nvc0_fbcon_accel_init(struct fb_info *info)
 	struct drm_device *dev = nfbdev->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_channel *chan = dev_priv->channel;
-	struct nouveau_bo *nvbo = nfbdev->nouveau_fb.nvbo;
+	struct nouveau_framebuffer *fb = &nfbdev->nouveau_fb;
 	int ret, format;
 
 	ret = nouveau_gpuobj_gr_new(chan, 0x902d, 0x902d);
@@ -203,8 +203,8 @@ nvc0_fbcon_accel_init(struct fb_info *info)
 	BEGIN_NVC0(chan, 2, NvSub2D, 0x0000, 1);
 	OUT_RING  (chan, 0x0000902d);
 	BEGIN_NVC0(chan, 2, NvSub2D, 0x0104, 2);
-	OUT_RING  (chan, upper_32_bits(chan->notifier_bo->bo.offset));
-	OUT_RING  (chan, lower_32_bits(chan->notifier_bo->bo.offset));
+	OUT_RING  (chan, upper_32_bits(chan->notifier_vma.offset));
+	OUT_RING  (chan, lower_32_bits(chan->notifier_vma.offset));
 	BEGIN_NVC0(chan, 2, NvSub2D, 0x0290, 1);
 	OUT_RING  (chan, 0);
 	BEGIN_NVC0(chan, 2, NvSub2D, 0x0888, 1);
@@ -249,8 +249,8 @@ nvc0_fbcon_accel_init(struct fb_info *info)
 	OUT_RING  (chan, info->fix.line_length);
 	OUT_RING  (chan, info->var.xres_virtual);
 	OUT_RING  (chan, info->var.yres_virtual);
-	OUT_RING  (chan, upper_32_bits(nvbo->vma.offset));
-	OUT_RING  (chan, lower_32_bits(nvbo->vma.offset));
+	OUT_RING  (chan, upper_32_bits(fb->vma.offset));
+	OUT_RING  (chan, lower_32_bits(fb->vma.offset));
 	BEGIN_NVC0(chan, 2, NvSub2D, 0x0230, 10);
 	OUT_RING  (chan, format);
 	OUT_RING  (chan, 1);
@@ -260,8 +260,8 @@ nvc0_fbcon_accel_init(struct fb_info *info)
 	OUT_RING  (chan, info->fix.line_length);
 	OUT_RING  (chan, info->var.xres_virtual);
 	OUT_RING  (chan, info->var.yres_virtual);
-	OUT_RING  (chan, upper_32_bits(nvbo->vma.offset));
-	OUT_RING  (chan, lower_32_bits(nvbo->vma.offset));
+	OUT_RING  (chan, upper_32_bits(fb->vma.offset));
+	OUT_RING  (chan, lower_32_bits(fb->vma.offset));
 	FIRE_RING (chan);
 
 	return 0;

drivers/gpu/drm/nouveau/nvc0_fifo.c

@@ -210,10 +210,10 @@ nvc0_fifo_unload_context(struct drm_device *dev)
 	int i;
 
 	for (i = 0; i < 128; i++) {
-		if (!(nv_rd32(dev, 0x003004 + (i * 4)) & 1))
+		if (!(nv_rd32(dev, 0x003004 + (i * 8)) & 1))
 			continue;
 
-		nv_mask(dev, 0x003004 + (i * 4), 0x00000001, 0x00000000);
+		nv_mask(dev, 0x003004 + (i * 8), 0x00000001, 0x00000000);
 		nv_wr32(dev, 0x002634, i);
 		if (!nv_wait(dev, 0x002634, 0xffffffff, i)) {
 			NV_INFO(dev, "PFIFO: kick ch %d failed: 0x%08x\n",

drivers/gpu/drm/nouveau/nvc0_graph.c

@@ -28,7 +28,34 @@
 
 #include "nouveau_drv.h"
 #include "nouveau_mm.h"
+
 #include "nvc0_graph.h"
+#include "nvc0_grhub.fuc.h"
+#include "nvc0_grgpc.fuc.h"
+
+static void
+nvc0_graph_ctxctl_debug_unit(struct drm_device *dev, u32 base)
+{
+	NV_INFO(dev, "PGRAPH: %06x - done 0x%08x\n", base,
+		nv_rd32(dev, base + 0x400));
+	NV_INFO(dev, "PGRAPH: %06x - stat 0x%08x 0x%08x 0x%08x 0x%08x\n", base,
+		nv_rd32(dev, base + 0x800), nv_rd32(dev, base + 0x804),
+		nv_rd32(dev, base + 0x808), nv_rd32(dev, base + 0x80c));
+	NV_INFO(dev, "PGRAPH: %06x - stat 0x%08x 0x%08x 0x%08x 0x%08x\n", base,
+		nv_rd32(dev, base + 0x810), nv_rd32(dev, base + 0x814),
+		nv_rd32(dev, base + 0x818), nv_rd32(dev, base + 0x81c));
+}
+
+static void
+nvc0_graph_ctxctl_debug(struct drm_device *dev)
+{
+	u32 gpcnr = nv_rd32(dev, 0x409604) & 0xffff;
+	u32 gpc;
+
+	nvc0_graph_ctxctl_debug_unit(dev, 0x409000);
+	for (gpc = 0; gpc < gpcnr; gpc++)
+		nvc0_graph_ctxctl_debug_unit(dev, 0x502000 + (gpc * 0x8000));
+}
 
 static int
 nvc0_graph_load_context(struct nouveau_channel *chan)
@@ -72,6 +99,16 @@ nvc0_graph_construct_context(struct nouveau_channel *chan)
 	if (!ctx)
 		return -ENOMEM;
 
+	if (!nouveau_ctxfw) {
+		nv_wr32(dev, 0x409840, 0x80000000);
+		nv_wr32(dev, 0x409500, 0x80000000 | chan->ramin->vinst >> 12);
+		nv_wr32(dev, 0x409504, 0x00000001);
+		if (!nv_wait(dev, 0x409800, 0x80000000, 0x80000000)) {
+			NV_ERROR(dev, "PGRAPH: HUB_SET_CHAN timeout\n");
+			nvc0_graph_ctxctl_debug(dev);
+			return -EBUSY;
+		}
+	} else {
 		nvc0_graph_load_context(chan);
 
 		nv_wo32(grch->grctx, 0x1c, 1);
@@ -79,6 +116,7 @@ nvc0_graph_construct_context(struct nouveau_channel *chan)
 		nv_wo32(grch->grctx, 0x28, 0);
 		nv_wo32(grch->grctx, 0x2c, 0);
 		dev_priv->engine.instmem.flush(dev);
+	}
 
 	ret = nvc0_grctx_generate(chan);
 	if (ret) {
@@ -86,11 +124,22 @@ nvc0_graph_construct_context(struct nouveau_channel *chan)
 		return ret;
 	}
 
+	if (!nouveau_ctxfw) {
+		nv_wr32(dev, 0x409840, 0x80000000);
+		nv_wr32(dev, 0x409500, 0x80000000 | chan->ramin->vinst >> 12);
+		nv_wr32(dev, 0x409504, 0x00000002);
+		if (!nv_wait(dev, 0x409800, 0x80000000, 0x80000000)) {
+			NV_ERROR(dev, "PGRAPH: HUB_CTX_SAVE timeout\n");
+			nvc0_graph_ctxctl_debug(dev);
+			return -EBUSY;
+		}
+	} else {
 		ret = nvc0_graph_unload_context_to(dev, chan->ramin->vinst);
 		if (ret) {
 			kfree(ctx);
 			return ret;
 		}
+	}
 
 	for (i = 0; i < priv->grctx_size; i += 4)
 		ctx[i / 4] = nv_ro32(grch->grctx, i);
@@ -108,50 +157,50 @@ nvc0_graph_create_context_mmio_list(struct nouveau_channel *chan)
 	int i = 0, gpc, tp, ret;
 	u32 magic;
 
-	ret = nouveau_gpuobj_new(dev, NULL, 0x2000, 256, NVOBJ_FLAG_VM,
+	ret = nouveau_gpuobj_new(dev, chan, 0x2000, 256, NVOBJ_FLAG_VM,
 				 &grch->unk408004);
 	if (ret)
 		return ret;
 
-	ret = nouveau_gpuobj_new(dev, NULL, 0x8000, 256, NVOBJ_FLAG_VM,
+	ret = nouveau_gpuobj_new(dev, chan, 0x8000, 256, NVOBJ_FLAG_VM,
 				 &grch->unk40800c);
 	if (ret)
 		return ret;
 
-	ret = nouveau_gpuobj_new(dev, NULL, 384 * 1024, 4096,
+	ret = nouveau_gpuobj_new(dev, chan, 384 * 1024, 4096,
 				 NVOBJ_FLAG_VM | NVOBJ_FLAG_VM_USER,
 				 &grch->unk418810);
 	if (ret)
 		return ret;
 
-	ret = nouveau_gpuobj_new(dev, NULL, 0x1000, 0, NVOBJ_FLAG_VM,
+	ret = nouveau_gpuobj_new(dev, chan, 0x1000, 0, NVOBJ_FLAG_VM,
 				 &grch->mmio);
 	if (ret)
 		return ret;
 
 
 	nv_wo32(grch->mmio, i++ * 4, 0x00408004);
-	nv_wo32(grch->mmio, i++ * 4, grch->unk408004->vinst >> 8);
+	nv_wo32(grch->mmio, i++ * 4, grch->unk408004->linst >> 8);
 	nv_wo32(grch->mmio, i++ * 4, 0x00408008);
 	nv_wo32(grch->mmio, i++ * 4, 0x80000018);
 
 	nv_wo32(grch->mmio, i++ * 4, 0x0040800c);
-	nv_wo32(grch->mmio, i++ * 4, grch->unk40800c->vinst >> 8);
+	nv_wo32(grch->mmio, i++ * 4, grch->unk40800c->linst >> 8);
 	nv_wo32(grch->mmio, i++ * 4, 0x00408010);
 	nv_wo32(grch->mmio, i++ * 4, 0x80000000);
 
 	nv_wo32(grch->mmio, i++ * 4, 0x00418810);
-	nv_wo32(grch->mmio, i++ * 4, 0x80000000 | grch->unk418810->vinst >> 12);
+	nv_wo32(grch->mmio, i++ * 4, 0x80000000 | grch->unk418810->linst >> 12);
 	nv_wo32(grch->mmio, i++ * 4, 0x00419848);
-	nv_wo32(grch->mmio, i++ * 4, 0x10000000 | grch->unk418810->vinst >> 12);
+	nv_wo32(grch->mmio, i++ * 4, 0x10000000 | grch->unk418810->linst >> 12);
 
 	nv_wo32(grch->mmio, i++ * 4, 0x00419004);
-	nv_wo32(grch->mmio, i++ * 4, grch->unk40800c->vinst >> 8);
+	nv_wo32(grch->mmio, i++ * 4, grch->unk40800c->linst >> 8);
 	nv_wo32(grch->mmio, i++ * 4, 0x00419008);
 	nv_wo32(grch->mmio, i++ * 4, 0x00000000);
 
 	nv_wo32(grch->mmio, i++ * 4, 0x00418808);
-	nv_wo32(grch->mmio, i++ * 4, grch->unk408004->vinst >> 8);
+	nv_wo32(grch->mmio, i++ * 4, grch->unk408004->linst >> 8);
 	nv_wo32(grch->mmio, i++ * 4, 0x0041880c);
 	nv_wo32(grch->mmio, i++ * 4, 0x80000018);
 
@@ -159,7 +208,7 @@ nvc0_graph_create_context_mmio_list(struct nouveau_channel *chan)
 	nv_wo32(grch->mmio, i++ * 4, 0x00405830);
 	nv_wo32(grch->mmio, i++ * 4, magic);
 	for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
-		for (tp = 0; tp < priv->tp_nr[gpc]; tp++, magic += 0x02fc) {
+		for (tp = 0; tp < priv->tp_nr[gpc]; tp++, magic += 0x0324) {
 			u32 reg = 0x504520 + (gpc * 0x8000) + (tp * 0x0800);
 			nv_wo32(grch->mmio, i++ * 4, reg);
 			nv_wo32(grch->mmio, i++ * 4, magic);
@@ -186,7 +235,7 @@ nvc0_graph_context_new(struct nouveau_channel *chan, int engine)
 		return -ENOMEM;
 	chan->engctx[NVOBJ_ENGINE_GR] = grch;
 
-	ret = nouveau_gpuobj_new(dev, NULL, priv->grctx_size, 256,
+	ret = nouveau_gpuobj_new(dev, chan, priv->grctx_size, 256,
 				 NVOBJ_FLAG_VM | NVOBJ_FLAG_ZERO_ALLOC,
 				 &grch->grctx);
 	if (ret)
@@ -197,8 +246,8 @@ nvc0_graph_context_new(struct nouveau_channel *chan, int engine)
 	if (ret)
 		goto error;
 
-	nv_wo32(chan->ramin, 0x0210, lower_32_bits(grctx->vinst) | 4);
-	nv_wo32(chan->ramin, 0x0214, upper_32_bits(grctx->vinst));
+	nv_wo32(chan->ramin, 0x0210, lower_32_bits(grctx->linst) | 4);
+	nv_wo32(chan->ramin, 0x0214, upper_32_bits(grctx->linst));
 	pinstmem->flush(dev);
 
 	if (!priv->grctx_vals) {
@@ -210,15 +259,20 @@ nvc0_graph_context_new(struct nouveau_channel *chan, int engine)
 	for (i = 0; i < priv->grctx_size; i += 4)
 		nv_wo32(grctx, i, priv->grctx_vals[i / 4]);
 
+	if (!nouveau_ctxfw) {
+		nv_wo32(grctx, 0x00, grch->mmio_nr);
+		nv_wo32(grctx, 0x04, grch->mmio->linst >> 8);
+	} else {
 		nv_wo32(grctx, 0xf4, 0);
 		nv_wo32(grctx, 0xf8, 0);
 		nv_wo32(grctx, 0x10, grch->mmio_nr);
-	nv_wo32(grctx, 0x14, lower_32_bits(grch->mmio->vinst));
-	nv_wo32(grctx, 0x18, upper_32_bits(grch->mmio->vinst));
+		nv_wo32(grctx, 0x14, lower_32_bits(grch->mmio->linst));
+		nv_wo32(grctx, 0x18, upper_32_bits(grch->mmio->linst));
 		nv_wo32(grctx, 0x1c, 1);
 		nv_wo32(grctx, 0x20, 0);
 		nv_wo32(grctx, 0x28, 0);
 		nv_wo32(grctx, 0x2c, 0);
+	}
 	pinstmem->flush(dev);
 	return 0;
 
@@ -296,6 +350,7 @@ static void
 nvc0_graph_init_gpc_0(struct drm_device *dev)
 {
 	struct nvc0_graph_priv *priv = nv_engine(dev, NVOBJ_ENGINE_GR);
+	const u32 magicgpc918 = DIV_ROUND_UP(0x00800000, priv->tp_total);
 	u32 data[TP_MAX / 8];
 	u8  tpnr[GPC_MAX];
 	int i, gpc, tpc;
@@ -307,13 +362,6 @@ nvc0_graph_init_gpc_0(struct drm_device *dev)
 	 * 465: 3/4/4/0 4        7
 	 * 470: 3/3/4/4 5        5
 	 * 480: 3/4/4/4 6        6
-	 *
-	 * magicgpc918
-	 * 450: 00200000 00000000001000000000000000000000
-	 * 460: 00124925 00000000000100100100100100100101
-	 * 465: 000ba2e9 00000000000010111010001011101001
-	 * 470: 00092493 00000000000010010010010010010011
-	 * 480: 00088889 00000000000010001000100010001001
 	 */
 
 	memset(data, 0x00, sizeof(data));
@@ -336,10 +384,10 @@ nvc0_graph_init_gpc_0(struct drm_device *dev)
 		nv_wr32(dev, GPC_UNIT(gpc, 0x0914), priv->magic_not_rop_nr << 8 |
 						  priv->tp_nr[gpc]);
 		nv_wr32(dev, GPC_UNIT(gpc, 0x0910), 0x00040000 | priv->tp_total);
-		nv_wr32(dev, GPC_UNIT(gpc, 0x0918), priv->magicgpc918);
+		nv_wr32(dev, GPC_UNIT(gpc, 0x0918), magicgpc918);
 	}
 
-	nv_wr32(dev, GPC_BCAST(0x1bd4), priv->magicgpc918);
+	nv_wr32(dev, GPC_BCAST(0x1bd4), magicgpc918);
 	nv_wr32(dev, GPC_BCAST(0x08ac), priv->rop_nr);
 }
 
@@ -419,8 +467,51 @@ nvc0_graph_init_fuc(struct drm_device *dev, u32 fuc_base,
 static int
 nvc0_graph_init_ctxctl(struct drm_device *dev)
 {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nvc0_graph_priv *priv = nv_engine(dev, NVOBJ_ENGINE_GR);
 	u32 r000260;
+	int i;
+
+	if (!nouveau_ctxfw) {
+		/* load HUB microcode */
+		r000260 = nv_mask(dev, 0x000260, 0x00000001, 0x00000000);
+		nv_wr32(dev, 0x4091c0, 0x01000000);
+		for (i = 0; i < sizeof(nvc0_grhub_data) / 4; i++)
+			nv_wr32(dev, 0x4091c4, nvc0_grhub_data[i]);
+
+		nv_wr32(dev, 0x409180, 0x01000000);
+		for (i = 0; i < sizeof(nvc0_grhub_code) / 4; i++) {
+			if ((i & 0x3f) == 0)
+				nv_wr32(dev, 0x409188, i >> 6);
+			nv_wr32(dev, 0x409184, nvc0_grhub_code[i]);
+		}
+
+		/* load GPC microcode */
+		nv_wr32(dev, 0x41a1c0, 0x01000000);
+		for (i = 0; i < sizeof(nvc0_grgpc_data) / 4; i++)
+			nv_wr32(dev, 0x41a1c4, nvc0_grgpc_data[i]);
+
+		nv_wr32(dev, 0x41a180, 0x01000000);
+		for (i = 0; i < sizeof(nvc0_grgpc_code) / 4; i++) {
+			if ((i & 0x3f) == 0)
+				nv_wr32(dev, 0x41a188, i >> 6);
+			nv_wr32(dev, 0x41a184, nvc0_grgpc_code[i]);
+		}
+		nv_wr32(dev, 0x000260, r000260);
+
+		/* start HUB ucode running, it'll init the GPCs */
+		nv_wr32(dev, 0x409800, dev_priv->chipset);
+		nv_wr32(dev, 0x40910c, 0x00000000);
+		nv_wr32(dev, 0x409100, 0x00000002);
+		if (!nv_wait(dev, 0x409800, 0x80000000, 0x80000000)) {
+			NV_ERROR(dev, "PGRAPH: HUB_INIT timed out\n");
+			nvc0_graph_ctxctl_debug(dev);
+			return -EBUSY;
+		}
+
+		priv->grctx_size = nv_rd32(dev, 0x409804);
+		return 0;
+	}
 
 	/* load fuc microcode */
 	r000260 = nv_mask(dev, 0x000260, 0x00000001, 0x00000000);
@@ -527,6 +618,22 @@ nvc0_graph_isr_chid(struct drm_device *dev, u64 inst)
 	return i;
 }
 
+static void
+nvc0_graph_ctxctl_isr(struct drm_device *dev)
+{
+	u32 ustat = nv_rd32(dev, 0x409c18);
+
+	if (ustat & 0x00000001)
+		NV_INFO(dev, "PGRAPH: CTXCTRL ucode error\n");
+	if (ustat & 0x00080000)
+		NV_INFO(dev, "PGRAPH: CTXCTRL watchdog timeout\n");
+	if (ustat & ~0x00080001)
+		NV_INFO(dev, "PGRAPH: CTXCTRL 0x%08x\n", ustat);
+
+	nvc0_graph_ctxctl_debug(dev);
+	nv_wr32(dev, 0x409c20, ustat);
+}
+
 static void
 nvc0_graph_isr(struct drm_device *dev)
 {
@@ -578,11 +685,7 @@ nvc0_graph_isr(struct drm_device *dev)
 	}
 
 	if (stat & 0x00080000) {
-		u32 ustat = nv_rd32(dev, 0x409c18);
-
-		NV_INFO(dev, "PGRAPH: CTXCTRL ustat 0x%08x\n", ustat);
-
-		nv_wr32(dev, 0x409c20, ustat);
+		nvc0_graph_ctxctl_isr(dev);
 		nv_wr32(dev, 0x400100, 0x00080000);
 		stat &= ~0x00080000;
 	}
@@ -606,7 +709,7 @@ nvc0_runk140_isr(struct drm_device *dev)
 		u32 st0 = nv_mask(dev, reg + 0x1020, 0, 0);
 		u32 st1 = nv_mask(dev, reg + 0x1420, 0, 0);
 
-		NV_INFO(dev, "PRUNK140: %d 0x%08x 0x%08x\n", unit, st0, st1);
+		NV_DEBUG(dev, "PRUNK140: %d 0x%08x 0x%08x\n", unit, st0, st1);
 		units &= ~(1 << unit);
 	}
 }
@@ -651,10 +754,12 @@ nvc0_graph_destroy(struct drm_device *dev, int engine)
 {
 	struct nvc0_graph_priv *priv = nv_engine(dev, engine);
 
+	if (nouveau_ctxfw) {
 		nvc0_graph_destroy_fw(&priv->fuc409c);
 		nvc0_graph_destroy_fw(&priv->fuc409d);
 		nvc0_graph_destroy_fw(&priv->fuc41ac);
 		nvc0_graph_destroy_fw(&priv->fuc41ad);
+	}
 
 	nouveau_irq_unregister(dev, 12);
 	nouveau_irq_unregister(dev, 25);
@@ -675,13 +780,10 @@ nvc0_graph_create(struct drm_device *dev)
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nvc0_graph_priv *priv;
 	int ret, gpc, i;
+	u32 fermi;
 
-	switch (dev_priv->chipset) {
-	case 0xc0:
-	case 0xc3:
-	case 0xc4:
-		break;
-	default:
+	fermi = nvc0_graph_class(dev);
+	if (!fermi) {
 		NV_ERROR(dev, "PGRAPH: unsupported chipset, please report!\n");
 		return 0;
 	}
@@ -701,6 +803,8 @@ nvc0_graph_create(struct drm_device *dev)
 	nouveau_irq_register(dev, 12, nvc0_graph_isr);
 	nouveau_irq_register(dev, 25, nvc0_runk140_isr);
 
+	if (nouveau_ctxfw) {
+		NV_INFO(dev, "PGRAPH: using external firmware\n");
 		if (nvc0_graph_create_fw(dev, "fuc409c", &priv->fuc409c) ||
 		    nvc0_graph_create_fw(dev, "fuc409d", &priv->fuc409d) ||
 		    nvc0_graph_create_fw(dev, "fuc41ac", &priv->fuc41ac) ||
@@ -708,7 +812,7 @@ nvc0_graph_create(struct drm_device *dev)
 			ret = 0;
 			goto error;
 		}
-
+	}
 
 	ret = nouveau_gpuobj_new(dev, NULL, 0x1000, 256, 0, &priv->unk4188b4);
 	if (ret)
@@ -735,25 +839,28 @@ nvc0_graph_create(struct drm_device *dev)
 	case 0xc0:
 		if (priv->tp_total == 11) { /* 465, 3/4/4/0, 4 */
 			priv->magic_not_rop_nr = 0x07;
-			/* filled values up to tp_total, the rest 0 */
-			priv->magicgpc918      = 0x000ba2e9;
 		} else
 		if (priv->tp_total == 14) { /* 470, 3/3/4/4, 5 */
 			priv->magic_not_rop_nr = 0x05;
-			priv->magicgpc918      = 0x00092493;
 		} else
 		if (priv->tp_total == 15) { /* 480, 3/4/4/4, 6 */
 			priv->magic_not_rop_nr = 0x06;
-			priv->magicgpc918      = 0x00088889;
 		}
 		break;
 	case 0xc3: /* 450, 4/0/0/0, 2 */
 		priv->magic_not_rop_nr = 0x03;
-		priv->magicgpc918      = 0x00200000;
 		break;
 	case 0xc4: /* 460, 3/4/0/0, 4 */
 		priv->magic_not_rop_nr = 0x01;
-		priv->magicgpc918      = 0x00124925;
+		break;
+	case 0xc1: /* 2/0/0/0, 1 */
+		priv->magic_not_rop_nr = 0x01;
+		break;
+	case 0xc8: /* 4/4/3/4, 5 */
+		priv->magic_not_rop_nr = 0x06;
+		break;
+	case 0xce: /* 4/4/0/0, 4 */
+		priv->magic_not_rop_nr = 0x03;
 		break;
 	}
 
@@ -763,13 +870,16 @@ nvc0_graph_create(struct drm_device *dev)
 			 priv->tp_nr[3], priv->rop_nr);
 		/* use 0xc3's values... */
 		priv->magic_not_rop_nr = 0x03;
-		priv->magicgpc918      = 0x00200000;
 	}
 
 	NVOBJ_CLASS(dev, 0x902d, GR); /* 2D */
 	NVOBJ_CLASS(dev, 0x9039, GR); /* M2MF */
 	NVOBJ_MTHD (dev, 0x9039, 0x0500, nvc0_graph_mthd_page_flip);
 	NVOBJ_CLASS(dev, 0x9097, GR); /* 3D */
+	if (fermi >= 0x9197)
+		NVOBJ_CLASS(dev, 0x9197, GR); /* 3D (NVC1-) */
+	if (fermi >= 0x9297)
+		NVOBJ_CLASS(dev, 0x9297, GR); /* 3D (NVC8-) */
 	NVOBJ_CLASS(dev, 0x90c0, GR); /* COMPUTE */
 	return 0;
 

drivers/gpu/drm/nouveau/nvc0_graph.fuc

+/* fuc microcode util functions for nvc0 PGRAPH
+ *
+ * Copyright 2011 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+define(`mmctx_data', `.b32 eval((($2 - 1) << 26) | $1)')
+define(`queue_init', `.skip eval((2 * 4) + ((8 * 4) * 2))')
+
+ifdef(`include_code', `
+// Error codes
+define(`E_BAD_COMMAND', 0x01)
+define(`E_CMD_OVERFLOW', 0x02)
+
+// Util macros to help with debugging ucode hangs etc
+define(`T_WAIT', 0)
+define(`T_MMCTX', 1)
+define(`T_STRWAIT', 2)
+define(`T_STRINIT', 3)
+define(`T_AUTO', 4)
+define(`T_CHAN', 5)
+define(`T_LOAD', 6)
+define(`T_SAVE', 7)
+define(`T_LCHAN', 8)
+define(`T_LCTXH', 9)
+
+define(`trace_set', `
+	mov $r8 0x83c
+	shl b32 $r8 6
+	clear b32 $r9
+	bset $r9 $1
+	iowr I[$r8 + 0x000] $r9		// CC_SCRATCH[7]
+')
+
+define(`trace_clr', `
+	mov $r8 0x85c
+	shl b32 $r8 6
+	clear b32 $r9
+	bset $r9 $1
+	iowr I[$r8 + 0x000] $r9		// CC_SCRATCH[7]
+')
+
+// queue_put - add request to queue
+//
+// In : $r13 queue pointer
+//	$r14 command
+//	$r15 data
+//
+queue_put:
+	// make sure we have space..
+	ld b32 $r8 D[$r13 + 0x0]	// GET
+	ld b32 $r9 D[$r13 + 0x4]	// PUT
+	xor $r8 8
+	cmpu b32 $r8 $r9
+	bra ne queue_put_next
+		mov $r15 E_CMD_OVERFLOW
+		call error
+		ret
+
+	// store cmd/data on queue
+	queue_put_next:
+	and $r8 $r9 7
+	shl b32 $r8 3
+	add b32 $r8 $r13
+	add b32 $r8 8
+	st b32 D[$r8 + 0x0] $r14
+	st b32 D[$r8 + 0x4] $r15
+
+	// update PUT
+	add b32 $r9 1
+	and $r9 0xf
+	st b32 D[$r13 + 0x4] $r9
+	ret
+
+// queue_get - fetch request from queue
+//
+// In : $r13 queue pointer
+//
+// Out:	$p1  clear on success (data available)
+//	$r14 command
+// 	$r15 data
+//
+queue_get:
+	bset $flags $p1
+	ld b32 $r8 D[$r13 + 0x0]	// GET
+	ld b32 $r9 D[$r13 + 0x4]	// PUT
+	cmpu b32 $r8 $r9
+	bra e queue_get_done
+		// fetch first cmd/data pair
+		and $r9 $r8 7
+		shl b32 $r9 3
+		add b32 $r9 $r13
+		add b32 $r9 8
+		ld b32 $r14 D[$r9 + 0x0]
+		ld b32 $r15 D[$r9 + 0x4]
+
+		// update GET
+		add b32 $r8 1
+		and $r8 0xf
+		st b32 D[$r13 + 0x0] $r8
+		bclr $flags $p1
+queue_get_done:
+	ret
+
+// nv_rd32 - read 32-bit value from nv register
+//
+// In : $r14 register
+// Out: $r15 value
+//
+nv_rd32:
+	mov $r11 0x728
+	shl b32 $r11 6
+	mov b32 $r12 $r14
+	bset $r12 31			// MMIO_CTRL_PENDING
+	iowr I[$r11 + 0x000] $r12	// MMIO_CTRL
+	nv_rd32_wait:
+		iord $r12 I[$r11 + 0x000]
+		xbit $r12 $r12 31
+		bra ne nv_rd32_wait
+	mov $r10 6			// DONE_MMIO_RD
+	call wait_doneo
+	iord $r15 I[$r11 + 0x100]	// MMIO_RDVAL
+	ret
+
+// nv_wr32 - write 32-bit value to nv register
+//
+// In : $r14 register
+//      $r15 value
+//
+nv_wr32:
+	mov $r11 0x728
+	shl b32 $r11 6
+	iowr I[$r11 + 0x200] $r15	// MMIO_WRVAL
+	mov b32 $r12 $r14
+	bset $r12 31			// MMIO_CTRL_PENDING
+	bset $r12 30			// MMIO_CTRL_WRITE
+	iowr I[$r11 + 0x000] $r12	// MMIO_CTRL
+	nv_wr32_wait:
+		iord $r12 I[$r11 + 0x000]
+		xbit $r12 $r12 31
+		bra ne nv_wr32_wait
+	ret
+
+// (re)set watchdog timer
+//
+// In : $r15 timeout
+//
+watchdog_reset:
+	mov $r8 0x430
+	shl b32 $r8 6
+	bset $r15 31
+	iowr I[$r8 + 0x000] $r15
+	ret
+
+// clear watchdog timer
+watchdog_clear:
+	mov $r8 0x430
+	shl b32 $r8 6
+	iowr I[$r8 + 0x000] $r0
+	ret
+
+// wait_done{z,o} - wait on FUC_DONE bit to become clear/set
+//
+// In : $r10 bit to wait on
+//
+define(`wait_done', `
+$1:
+	trace_set(T_WAIT);
+	mov $r8 0x818
+	shl b32 $r8 6
+	iowr I[$r8 + 0x000] $r10	// CC_SCRATCH[6] = wait bit
+	wait_done_$1:
+		mov $r8 0x400
+		shl b32 $r8 6
+		iord $r8 I[$r8 + 0x000]	// DONE
+		xbit $r8 $r8 $r10
+		bra $2 wait_done_$1
+	trace_clr(T_WAIT)
+	ret
+')
+wait_done(wait_donez, ne)
+wait_done(wait_doneo, e)
+
+// mmctx_size - determine size of a mmio list transfer
+//
+// In : $r14 mmio list head
+//      $r15 mmio list tail
+// Out: $r15 transfer size (in bytes)
+//
+mmctx_size:
+	clear b32 $r9
+	nv_mmctx_size_loop:
+		ld b32 $r8 D[$r14]
+		shr b32 $r8 26
+		add b32 $r8 1
+		shl b32 $r8 2
+		add b32 $r9 $r8
+		add b32 $r14 4
+		cmpu b32 $r14 $r15
+		bra ne nv_mmctx_size_loop
+	mov b32 $r15 $r9
+	ret
+
+// mmctx_xfer - execute a list of mmio transfers
+//
+// In : $r10 flags
+//		bit 0: direction (0 = save, 1 = load)
+//		bit 1: set if first transfer
+//		bit 2: set if last transfer
+//	$r11 base
+//	$r12 mmio list head
+//	$r13 mmio list tail
+//	$r14 multi_stride
+//	$r15 multi_mask
+//
+mmctx_xfer:
+	trace_set(T_MMCTX)
+	mov $r8 0x710
+	shl b32 $r8 6
+	clear b32 $r9
+	or $r11 $r11
+	bra e mmctx_base_disabled
+		iowr I[$r8 + 0x000] $r11	// MMCTX_BASE
+		bset $r9 0			// BASE_EN
+	mmctx_base_disabled:
+	or $r14 $r14
+	bra e mmctx_multi_disabled
+		iowr I[$r8 + 0x200] $r14 	// MMCTX_MULTI_STRIDE
+		iowr I[$r8 + 0x300] $r15 	// MMCTX_MULTI_MASK
+		bset $r9 1			// MULTI_EN
+	mmctx_multi_disabled:
+	add b32 $r8 0x100
+
+	xbit $r11 $r10 0
+	shl b32 $r11 16			// DIR
+	bset $r11 12			// QLIMIT = 0x10
+	xbit $r14 $r10 1
+	shl b32 $r14 17
+	or $r11 $r14			// START_TRIGGER
+	iowr I[$r8 + 0x000] $r11	// MMCTX_CTRL
+
+	// loop over the mmio list, and send requests to the hw
+	mmctx_exec_loop:
+		// wait for space in mmctx queue
+		mmctx_wait_free:
+			iord $r14 I[$r8 + 0x000] // MMCTX_CTRL
+			and $r14 0x1f
+			bra e mmctx_wait_free
+
+		// queue up an entry
+		ld b32 $r14 D[$r12]
+		or $r14 $r9
+		iowr I[$r8 + 0x300] $r14
+		add b32 $r12 4
+		cmpu b32 $r12 $r13
+		bra ne mmctx_exec_loop
+
+	xbit $r11 $r10 2
+	bra ne mmctx_stop
+		// wait for queue to empty
+		mmctx_fini_wait:
+			iord $r11 I[$r8 + 0x000]	// MMCTX_CTRL
+			and $r11 0x1f
+			cmpu b32 $r11 0x10
+			bra ne mmctx_fini_wait
+		mov $r10 2				// DONE_MMCTX
+		call wait_donez
+		bra mmctx_done
+	mmctx_stop:
+		xbit $r11 $r10 0
+		shl b32 $r11 16			// DIR
+		bset $r11 12			// QLIMIT = 0x10
+		bset $r11 18			// STOP_TRIGGER
+		iowr I[$r8 + 0x000] $r11	// MMCTX_CTRL
+		mmctx_stop_wait:
+			// wait for STOP_TRIGGER to clear
+			iord $r11 I[$r8 + 0x000] // MMCTX_CTRL
+			xbit $r11 $r11 18
+			bra ne mmctx_stop_wait
+	mmctx_done:
+	trace_clr(T_MMCTX)
+	ret
+
+// Wait for DONE_STRAND
+//
+strand_wait:
+	push $r10
+	mov $r10 2
+	call wait_donez
+	pop $r10
+	ret
+
+// unknown - call before issuing strand commands
+//
+strand_pre:
+	mov $r8 0x4afc
+	sethi $r8 0x20000
+	mov $r9 0xc
+	iowr I[$r8] $r9
+	call strand_wait
+	ret
+
+// unknown - call after issuing strand commands
+//
+strand_post:
+	mov $r8 0x4afc
+	sethi $r8 0x20000
+	mov $r9 0xd
+	iowr I[$r8] $r9
+	call strand_wait
+	ret
+
+// Selects strand set?!
+//
+// In: $r14 id
+//
+strand_set:
+	mov $r10 0x4ffc
+	sethi $r10 0x20000
+	sub b32 $r11 $r10 0x500
+	mov $r12 0xf
+	iowr I[$r10 + 0x000] $r12		// 0x93c = 0xf
+	mov $r12 0xb
+	iowr I[$r11 + 0x000] $r12		// 0x928 = 0xb
+	call strand_wait
+	iowr I[$r10 + 0x000] $r14		// 0x93c = <id>
+	mov $r12 0xa
+	iowr I[$r11 + 0x000] $r12		// 0x928 = 0xa
+	call strand_wait
+	ret
+
+// Initialise strand context data
+//
+// In : $r15 context base
+// Out: $r15 context size (in bytes)
+//
+// Strandset(?) 3 hardcoded currently
+//
+strand_ctx_init:
+	trace_set(T_STRINIT)
+	call strand_pre
+	mov $r14 3
+	call strand_set
+	mov $r10 0x46fc
+	sethi $r10 0x20000
+	add b32 $r11 $r10 0x400
+	iowr I[$r10 + 0x100] $r0	// STRAND_FIRST_GENE = 0
+	mov $r12 1
+	iowr I[$r11 + 0x000] $r12	// STRAND_CMD = LATCH_FIRST_GENE
+	call strand_wait
+	sub b32 $r12 $r0 1
+	iowr I[$r10 + 0x000] $r12	// STRAND_GENE_CNT = 0xffffffff
+	mov $r12 2
+	iowr I[$r11 + 0x000] $r12	// STRAND_CMD = LATCH_GENE_CNT
+	call strand_wait
+	call strand_post
+
+	// read the size of each strand, poke the context offset of
+	// each into STRAND_{SAVE,LOAD}_SWBASE now, no need to worry
+	// about it later then.
+	mov $r8 0x880
+	shl b32 $r8 6
+	iord $r9 I[$r8 + 0x000]		// STRANDS
+	add b32 $r8 0x2200
+	shr b32 $r14 $r15 8
+	ctx_init_strand_loop:
+		iowr I[$r8 + 0x000] $r14	// STRAND_SAVE_SWBASE
+		iowr I[$r8 + 0x100] $r14	// STRAND_LOAD_SWBASE
+		iord $r10 I[$r8 + 0x200]	// STRAND_SIZE
+		shr b32 $r10 6
+		add b32 $r10 1
+		add b32 $r14 $r10
+		add b32 $r8 4
+		sub b32 $r9 1
+		bra ne ctx_init_strand_loop
+
+	shl b32 $r14 8
+	sub b32 $r15 $r14 $r15
+	trace_clr(T_STRINIT)
+	ret
+')

drivers/gpu/drm/nouveau/nvc0_graph.h

@@ -58,7 +58,6 @@ struct nvc0_graph_priv {
 	struct nouveau_gpuobj *unk4188b8;
 
 	u8 magic_not_rop_nr;
-	u32 magicgpc918;
 };
 
 struct nvc0_graph_chan {
@@ -72,4 +71,25 @@ struct nvc0_graph_chan {
 
 int nvc0_grctx_generate(struct nouveau_channel *);
 
+/* nvc0_graph.c uses this also to determine supported chipsets */
+static inline u32
+nvc0_graph_class(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	switch (dev_priv->chipset) {
+	case 0xc0:
+	case 0xc3:
+	case 0xc4:
+	case 0xce: /* guess, mmio trace shows only 0x9097 state */
+		return 0x9097;
+	case 0xc1:
+		return 0x9197;
+	case 0xc8:
+		return 0x9297;
+	default:
+		return 0;
+	}
+}
+
 #endif

drivers/gpu/drm/nouveau/nvc0_grctx.c

@@ -45,6 +45,9 @@ nv_mthd(struct drm_device *dev, u32 class, u32 mthd, u32 data)
 static void
 nvc0_grctx_generate_9097(struct drm_device *dev)
 {
+	u32 fermi = nvc0_graph_class(dev);
+	u32 mthd;
+
 	nv_mthd(dev, 0x9097, 0x0800, 0x00000000);
 	nv_mthd(dev, 0x9097, 0x0840, 0x00000000);
 	nv_mthd(dev, 0x9097, 0x0880, 0x00000000);
@@ -824,134 +827,10 @@ nvc0_grctx_generate_9097(struct drm_device *dev)
 	nv_mthd(dev, 0x9097, 0x1eb8, 0x00000001);
 	nv_mthd(dev, 0x9097, 0x1ed8, 0x00000001);
 	nv_mthd(dev, 0x9097, 0x1ef8, 0x00000001);
-	nv_mthd(dev, 0x9097, 0x3400, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3404, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3408, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x340c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3410, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3414, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3418, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x341c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3420, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3424, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3428, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x342c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3430, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3434, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3438, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x343c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3440, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3444, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3448, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x344c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3450, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3454, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3458, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x345c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3460, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3464, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3468, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x346c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3470, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3474, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3478, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x347c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3480, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3484, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3488, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x348c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3490, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3494, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3498, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x349c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34a0, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34a4, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34a8, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34ac, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34b0, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34b4, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34b8, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34bc, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34c0, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34c4, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34c8, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34cc, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34d0, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34d4, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34d8, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34dc, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34e0, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34e4, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34e8, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34ec, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34f0, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34f4, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34f8, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x34fc, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3500, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3504, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3508, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x350c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3510, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3514, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3518, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x351c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3520, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3524, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3528, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x352c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3530, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3534, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3538, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x353c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3540, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3544, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3548, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x354c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3550, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3554, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3558, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x355c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3560, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3564, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3568, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x356c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3570, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3574, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3578, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x357c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3580, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3584, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3588, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x358c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3590, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3594, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x3598, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x359c, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35a0, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35a4, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35a8, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35ac, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35b0, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35b4, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35b8, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35bc, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35c0, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35c4, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35c8, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35cc, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35d0, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35d4, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35d8, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35dc, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35e0, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35e4, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35e8, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35ec, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35f0, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35f4, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35f8, 0x00000000);
-	nv_mthd(dev, 0x9097, 0x35fc, 0x00000000);
+	if (fermi == 0x9097) {
+		for (mthd = 0x3400; mthd <= 0x35fc; mthd += 4)
+			nv_mthd(dev, 0x9097, mthd, 0x00000000);
+	}
 	nv_mthd(dev, 0x9097, 0x030c, 0x00000001);
 	nv_mthd(dev, 0x9097, 0x1944, 0x00000000);
 	nv_mthd(dev, 0x9097, 0x1514, 0x00000000);
@@ -1320,6 +1199,37 @@ nvc0_grctx_generate_9097(struct drm_device *dev)
 	nv_mthd(dev, 0x9097, 0x3410, 0x80002006);
 }
 
+static void
+nvc0_grctx_generate_9197(struct drm_device *dev)
+{
+	u32 fermi = nvc0_graph_class(dev);
+	u32 mthd;
+
+	if (fermi == 0x9197) {
+		for (mthd = 0x3400; mthd <= 0x35fc; mthd += 4)
+			nv_mthd(dev, 0x9197, mthd, 0x00000000);
+	}
+	nv_mthd(dev, 0x9197, 0x02e4, 0x0000b001);
+}
+
+static void
+nvc0_grctx_generate_9297(struct drm_device *dev)
+{
+	u32 fermi = nvc0_graph_class(dev);
+	u32 mthd;
+
+	if (fermi == 0x9297) {
+		for (mthd = 0x3400; mthd <= 0x35fc; mthd += 4)
+			nv_mthd(dev, 0x9297, mthd, 0x00000000);
+	}
+	nv_mthd(dev, 0x9297, 0x036c, 0x00000000);
+	nv_mthd(dev, 0x9297, 0x0370, 0x00000000);
+	nv_mthd(dev, 0x9297, 0x07a4, 0x00000000);
+	nv_mthd(dev, 0x9297, 0x07a8, 0x00000000);
+	nv_mthd(dev, 0x9297, 0x0374, 0x00000000);
+	nv_mthd(dev, 0x9297, 0x0378, 0x00000020);
+}
+
 static void
 nvc0_grctx_generate_902d(struct drm_device *dev)
 {
@@ -1559,8 +1469,15 @@ nvc0_grctx_generate_unk47xx(struct drm_device *dev)
 static void
 nvc0_grctx_generate_shaders(struct drm_device *dev)
 {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	if (dev_priv->chipset != 0xc1) {
 		nv_wr32(dev, 0x405800, 0x078000bf);
 		nv_wr32(dev, 0x405830, 0x02180000);
+	} else {
+		nv_wr32(dev, 0x405800, 0x0f8000bf);
+		nv_wr32(dev, 0x405830, 0x02180218);
+	}
 	nv_wr32(dev, 0x405834, 0x00000000);
 	nv_wr32(dev, 0x405838, 0x00000000);
 	nv_wr32(dev, 0x405854, 0x00000000);
@@ -1586,10 +1503,16 @@ nvc0_grctx_generate_unk60xx(struct drm_device *dev)
 static void
 nvc0_grctx_generate_unk64xx(struct drm_device *dev)
 {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
 	nv_wr32(dev, 0x4064a8, 0x00000000);
 	nv_wr32(dev, 0x4064ac, 0x00003fff);
 	nv_wr32(dev, 0x4064b4, 0x00000000);
 	nv_wr32(dev, 0x4064b8, 0x00000000);
+	if (dev_priv->chipset == 0xc1) {
+		nv_wr32(dev, 0x4064c0, 0x80140078);
+		nv_wr32(dev, 0x4064c4, 0x0086ffff);
+	}
 }
 
 static void
@@ -1622,21 +1545,14 @@ static void
 nvc0_grctx_generate_rop(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	int chipset = dev_priv->chipset;
 
 	/* ROPC_BROADCAST */
 	nv_wr32(dev, 0x408800, 0x02802a3c);
 	nv_wr32(dev, 0x408804, 0x00000040);
-	nv_wr32(dev, 0x408808, 0x0003e00d);
-	switch (dev_priv->chipset) {
-	case 0xc0:
-		nv_wr32(dev, 0x408900, 0x0080b801);
-		break;
-	case 0xc3:
-	case 0xc4:
+	nv_wr32(dev, 0x408808, chipset != 0xc1 ? 0x0003e00d : 0x1003e005);
 	nv_wr32(dev, 0x408900, 0x3080b801);
-		break;
-	}
-	nv_wr32(dev, 0x408904, 0x02000001);
+	nv_wr32(dev, 0x408904, chipset != 0xc1 ? 0x02000001 : 0x62000001);
 	nv_wr32(dev, 0x408908, 0x00c80929);
 	nv_wr32(dev, 0x40890c, 0x00000000);
 	nv_wr32(dev, 0x408980, 0x0000011d);
@@ -1645,6 +1561,8 @@ nvc0_grctx_generate_rop(struct drm_device *dev)
 static void
 nvc0_grctx_generate_gpc(struct drm_device *dev)
 {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	int chipset = dev_priv->chipset;
 	int i;
 
 	/* GPC_BROADCAST */
@@ -1676,7 +1594,7 @@ nvc0_grctx_generate_gpc(struct drm_device *dev)
 	nv_wr32(dev, 0x41880c, 0x00000000);
 	nv_wr32(dev, 0x418810, 0x00000000);
 	nv_wr32(dev, 0x418828, 0x00008442);
-	nv_wr32(dev, 0x418830, 0x00000001);
+	nv_wr32(dev, 0x418830, chipset != 0xc1 ? 0x00000001 : 0x10000001);
 	nv_wr32(dev, 0x4188d8, 0x00000008);
 	nv_wr32(dev, 0x4188e0, 0x01000000);
 	nv_wr32(dev, 0x4188e8, 0x00000000);
@@ -1684,7 +1602,7 @@ nvc0_grctx_generate_gpc(struct drm_device *dev)
 	nv_wr32(dev, 0x4188f0, 0x00000000);
 	nv_wr32(dev, 0x4188f4, 0x00000000);
 	nv_wr32(dev, 0x4188f8, 0x00000000);
-	nv_wr32(dev, 0x4188fc, 0x00100000);
+	nv_wr32(dev, 0x4188fc, chipset != 0xc1 ? 0x00100000 : 0x00100018);
 	nv_wr32(dev, 0x41891c, 0x00ff00ff);
 	nv_wr32(dev, 0x418924, 0x00000000);
 	nv_wr32(dev, 0x418928, 0x00ffff00);
@@ -1715,6 +1633,8 @@ nvc0_grctx_generate_gpc(struct drm_device *dev)
 	nv_wr32(dev, 0x418c24, 0x00000000);
 	nv_wr32(dev, 0x418c28, 0x00000000);
 	nv_wr32(dev, 0x418c2c, 0x00000000);
+	if (chipset == 0xc1)
+		nv_wr32(dev, 0x418c6c, 0x00000001);
 	nv_wr32(dev, 0x418c80, 0x20200004);
 	nv_wr32(dev, 0x418c8c, 0x00000001);
 	nv_wr32(dev, 0x419000, 0x00000780);
@@ -1727,10 +1647,13 @@ static void
 nvc0_grctx_generate_tp(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	int chipset = dev_priv->chipset;
 
 	/* GPC_BROADCAST.TP_BROADCAST */
+	nv_wr32(dev, 0x419818, 0x00000000);
+	nv_wr32(dev, 0x41983c, 0x00038bc7);
 	nv_wr32(dev, 0x419848, 0x00000000);
-	nv_wr32(dev, 0x419864, 0x0000012a);
+	nv_wr32(dev, 0x419864, chipset != 0xc1 ? 0x0000012a : 0x00000129);
 	nv_wr32(dev, 0x419888, 0x00000000);
 	nv_wr32(dev, 0x419a00, 0x000001f0);
 	nv_wr32(dev, 0x419a04, 0x00000001);
@@ -1740,8 +1663,8 @@ nvc0_grctx_generate_tp(struct drm_device *dev)
 	nv_wr32(dev, 0x419a14, 0x00000200);
 	nv_wr32(dev, 0x419a1c, 0x00000000);
 	nv_wr32(dev, 0x419a20, 0x00000800);
-	if (dev_priv->chipset != 0xc0)
-		nv_wr32(dev, 0x00419ac4, 0x0007f440); /* 0xc3 */
+	if (chipset != 0xc0 && chipset != 0xc8)
+		nv_wr32(dev, 0x00419ac4, 0x0007f440);
 	nv_wr32(dev, 0x419b00, 0x0a418820);
 	nv_wr32(dev, 0x419b04, 0x062080e6);
 	nv_wr32(dev, 0x419b08, 0x020398a4);
@@ -1749,17 +1672,19 @@ nvc0_grctx_generate_tp(struct drm_device *dev)
 	nv_wr32(dev, 0x419b10, 0x0a418820);
 	nv_wr32(dev, 0x419b14, 0x000000e6);
 	nv_wr32(dev, 0x419bd0, 0x00900103);
-	nv_wr32(dev, 0x419be0, 0x00000001);
+	nv_wr32(dev, 0x419be0, chipset != 0xc1 ? 0x00000001 : 0x00400001);
 	nv_wr32(dev, 0x419be4, 0x00000000);
 	nv_wr32(dev, 0x419c00, 0x00000002);
 	nv_wr32(dev, 0x419c04, 0x00000006);
 	nv_wr32(dev, 0x419c08, 0x00000002);
 	nv_wr32(dev, 0x419c20, 0x00000000);
-	nv_wr32(dev, 0x419cbc, 0x28137606);
+	nv_wr32(dev, 0x419cb0, 0x00060048); //XXX: 0xce 0x00020048
 	nv_wr32(dev, 0x419ce8, 0x00000000);
 	nv_wr32(dev, 0x419cf4, 0x00000183);
-	nv_wr32(dev, 0x419d20, 0x02180000);
+	nv_wr32(dev, 0x419d20, chipset != 0xc1 ? 0x02180000 : 0x12180000);
 	nv_wr32(dev, 0x419d24, 0x00001fff);
+	if (chipset == 0xc1)
+		nv_wr32(dev, 0x419d44, 0x02180218);
 	nv_wr32(dev, 0x419e04, 0x00000000);
 	nv_wr32(dev, 0x419e08, 0x00000000);
 	nv_wr32(dev, 0x419e0c, 0x00000000);
@@ -1785,11 +1710,11 @@ nvc0_grctx_generate_tp(struct drm_device *dev)
 	nv_wr32(dev, 0x419e8c, 0x00000000);
 	nv_wr32(dev, 0x419e90, 0x00000000);
 	nv_wr32(dev, 0x419e98, 0x00000000);
-	if (dev_priv->chipset != 0xc0)
+	if (chipset != 0xc0 && chipset != 0xc8)
 		nv_wr32(dev, 0x419ee0, 0x00011110);
 	nv_wr32(dev, 0x419f50, 0x00000000);
 	nv_wr32(dev, 0x419f54, 0x00000000);
-	if (dev_priv->chipset != 0xc0)
+	if (chipset != 0xc0 && chipset != 0xc8)
 		nv_wr32(dev, 0x419f58, 0x00000000);
 }
 
@@ -1801,6 +1726,7 @@ nvc0_grctx_generate(struct nouveau_channel *chan)
 	struct nvc0_graph_chan *grch = chan->engctx[NVOBJ_ENGINE_GR];
 	struct drm_device *dev = chan->dev;
 	int i, gpc, tp, id;
+	u32 fermi = nvc0_graph_class(dev);
 	u32 r000260, tmp;
 
 	r000260 = nv_rd32(dev, 0x000260);
@@ -1857,10 +1783,11 @@ nvc0_grctx_generate(struct nouveau_channel *chan)
 	nv_wr32(dev, 0x40587c, 0x00000000);
 
 	if (1) {
-		const u8 chipset_tp_max[] = { 16, 0, 0, 4, 8 };
+		const u8 chipset_tp_max[] = { 16, 4, 0, 4, 8, 0, 0, 0,
+					      16, 0, 0, 0, 0, 0, 8, 0 };
 		u8 max = chipset_tp_max[dev_priv->chipset & 0x0f];
 		u8 tpnr[GPC_MAX];
-		u8 data[32];
+		u8 data[TP_MAX];
 
 		memcpy(tpnr, priv->tp_nr, sizeof(priv->tp_nr));
 		memset(data, 0x1f, sizeof(data));
@@ -2633,6 +2560,8 @@ nvc0_grctx_generate(struct nouveau_channel *chan)
 	nv_icmd(dev, 0x0000053f, 0xffff0000);
 	nv_icmd(dev, 0x00000585, 0x0000003f);
 	nv_icmd(dev, 0x00000576, 0x00000003);
+	if (dev_priv->chipset == 0xc1)
+		nv_icmd(dev, 0x0000057b, 0x00000059);
 	nv_icmd(dev, 0x00000586, 0x00000040);
 	nv_icmd(dev, 0x00000582, 0x00000080);
 	nv_icmd(dev, 0x00000583, 0x00000080);
@@ -2865,6 +2794,10 @@ nvc0_grctx_generate(struct nouveau_channel *chan)
 	nv_wr32(dev, 0x404154, 0x00000400);
 
 	nvc0_grctx_generate_9097(dev);
+	if (fermi >= 0x9197)
+		nvc0_grctx_generate_9197(dev);
+	if (fermi >= 0x9297)
+		nvc0_grctx_generate_9297(dev);
 	nvc0_grctx_generate_902d(dev);
 	nvc0_grctx_generate_9039(dev);
 	nvc0_grctx_generate_90c0(dev);

drivers/gpu/drm/nouveau/nvc0_grgpc.fuc

+/* fuc microcode for nvc0 PGRAPH/GPC
+ *
+ * Copyright 2011 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+/* To build:
+ *    m4 nvc0_grgpc.fuc | envyas -a -w -m fuc -V nva3 -o nvc0_grgpc.fuc.h
+ */
+
+/* TODO
+ * - bracket certain functions with scratch writes, useful for debugging
+ * - watchdog timer around ctx operations
+ */
+
+.section nvc0_grgpc_data
+include(`nvc0_graph.fuc')
+gpc_id:			.b32 0
+gpc_mmio_list_head:	.b32 0
+gpc_mmio_list_tail:	.b32 0
+
+tpc_count:		.b32 0
+tpc_mask:		.b32 0
+tpc_mmio_list_head:	.b32 0
+tpc_mmio_list_tail:	.b32 0
+
+cmd_queue:		queue_init
+
+// chipset descriptions
+chipsets:
+.b8  0xc0 0 0 0
+.b16 nvc0_gpc_mmio_head
+.b16 nvc0_gpc_mmio_tail
+.b16 nvc0_tpc_mmio_head
+.b16 nvc0_tpc_mmio_tail
+.b8  0xc1 0 0 0
+.b16 nvc0_gpc_mmio_head
+.b16 nvc1_gpc_mmio_tail
+.b16 nvc0_tpc_mmio_head
+.b16 nvc1_tpc_mmio_tail
+.b8  0xc3 0 0 0
+.b16 nvc0_gpc_mmio_head
+.b16 nvc0_gpc_mmio_tail
+.b16 nvc0_tpc_mmio_head
+.b16 nvc3_tpc_mmio_tail
+.b8  0xc4 0 0 0
+.b16 nvc0_gpc_mmio_head
+.b16 nvc0_gpc_mmio_tail
+.b16 nvc0_tpc_mmio_head
+.b16 nvc3_tpc_mmio_tail
+.b8  0xc8 0 0 0
+.b16 nvc0_gpc_mmio_head
+.b16 nvc0_gpc_mmio_tail
+.b16 nvc0_tpc_mmio_head
+.b16 nvc0_tpc_mmio_tail
+.b8  0xce 0 0 0
+.b16 nvc0_gpc_mmio_head
+.b16 nvc0_gpc_mmio_tail
+.b16 nvc0_tpc_mmio_head
+.b16 nvc3_tpc_mmio_tail
+.b8  0 0 0 0
+
+// GPC mmio lists
+nvc0_gpc_mmio_head:
+mmctx_data(0x000380, 1)
+mmctx_data(0x000400, 6)
+mmctx_data(0x000450, 9)
+mmctx_data(0x000600, 1)
+mmctx_data(0x000684, 1)
+mmctx_data(0x000700, 5)
+mmctx_data(0x000800, 1)
+mmctx_data(0x000808, 3)
+mmctx_data(0x000828, 1)
+mmctx_data(0x000830, 1)
+mmctx_data(0x0008d8, 1)
+mmctx_data(0x0008e0, 1)
+mmctx_data(0x0008e8, 6)
+mmctx_data(0x00091c, 1)
+mmctx_data(0x000924, 3)
+mmctx_data(0x000b00, 1)
+mmctx_data(0x000b08, 6)
+mmctx_data(0x000bb8, 1)
+mmctx_data(0x000c08, 1)
+mmctx_data(0x000c10, 8)
+mmctx_data(0x000c80, 1)
+mmctx_data(0x000c8c, 1)
+mmctx_data(0x001000, 3)
+mmctx_data(0x001014, 1)
+nvc0_gpc_mmio_tail:
+mmctx_data(0x000c6c, 1);
+nvc1_gpc_mmio_tail:
+
+// TPC mmio lists
+nvc0_tpc_mmio_head:
+mmctx_data(0x000018, 1)
+mmctx_data(0x00003c, 1)
+mmctx_data(0x000048, 1)
+mmctx_data(0x000064, 1)
+mmctx_data(0x000088, 1)
+mmctx_data(0x000200, 6)
+mmctx_data(0x00021c, 2)
+mmctx_data(0x000300, 6)
+mmctx_data(0x0003d0, 1)
+mmctx_data(0x0003e0, 2)
+mmctx_data(0x000400, 3)
+mmctx_data(0x000420, 1)
+mmctx_data(0x0004b0, 1)
+mmctx_data(0x0004e8, 1)
+mmctx_data(0x0004f4, 1)
+mmctx_data(0x000520, 2)
+mmctx_data(0x000604, 4)
+mmctx_data(0x000644, 20)
+mmctx_data(0x000698, 1)
+mmctx_data(0x000750, 2)
+nvc0_tpc_mmio_tail:
+mmctx_data(0x000758, 1)
+mmctx_data(0x0002c4, 1)
+mmctx_data(0x0004bc, 1)
+mmctx_data(0x0006e0, 1)
+nvc3_tpc_mmio_tail:
+mmctx_data(0x000544, 1)
+nvc1_tpc_mmio_tail:
+
+
+.section nvc0_grgpc_code
+bra init
+define(`include_code')
+include(`nvc0_graph.fuc')
+
+// reports an exception to the host
+//
+// In: $r15 error code (see nvc0_graph.fuc)
+//
+error:
+	push $r14
+	mov $r14 -0x67ec 	// 0x9814
+	sethi $r14 0x400000
+	call nv_wr32		// HUB_CTXCTL_CC_SCRATCH[5] = error code
+	add b32 $r14 0x41c
+	mov $r15 1
+	call nv_wr32		// HUB_CTXCTL_INTR_UP_SET
+	pop $r14
+	ret
+
+// GPC fuc initialisation, executed by triggering ucode start, will
+// fall through to main loop after completion.
+//
+// Input:
+//   CC_SCRATCH[0]: chipset (PMC_BOOT_0 read returns 0x0bad0bad... sigh)
+//   CC_SCRATCH[1]: context base
+//
+// Output:
+//   CC_SCRATCH[0]:
+//	     31:31: set to signal completion
+//   CC_SCRATCH[1]:
+//	      31:0: GPC context size
+//
+init:
+	clear b32 $r0
+	mov $sp $r0
+
+	// enable fifo access
+	mov $r1 0x1200
+	mov $r2 2
+	iowr I[$r1 + 0x000] $r2		// FIFO_ENABLE
+
+	// setup i0 handler, and route all interrupts to it
+	mov $r1 ih
+	mov $iv0 $r1
+	mov $r1 0x400
+	iowr I[$r1 + 0x300] $r0		// INTR_DISPATCH
+
+	// enable fifo interrupt
+	mov $r2 4
+	iowr I[$r1 + 0x000] $r2		// INTR_EN_SET
+
+	// enable interrupts
+	bset $flags ie0
+
+	// figure out which GPC we are, and how many TPCs we have
+	mov $r1 0x608
+	shl b32 $r1 6
+	iord $r2 I[$r1 + 0x000]		// UNITS
+	mov $r3 1
+	and $r2 0x1f
+	shl b32 $r3 $r2
+	sub b32 $r3 1
+	st b32 D[$r0 + tpc_count] $r2
+	st b32 D[$r0 + tpc_mask] $r3
+	add b32 $r1 0x400
+	iord $r2 I[$r1 + 0x000]		// MYINDEX
+	st b32 D[$r0 + gpc_id] $r2
+
+	// find context data for this chipset
+	mov $r2 0x800
+	shl b32 $r2 6
+	iord $r2 I[$r2 + 0x000]		// CC_SCRATCH[0]
+	mov $r1 chipsets - 12
+	init_find_chipset:
+		add b32 $r1 12
+		ld b32 $r3 D[$r1 + 0x00]
+		cmpu b32 $r3 $r2
+		bra e init_context
+		cmpu b32 $r3 0
+		bra ne init_find_chipset
+		// unknown chipset
+		ret
+
+	// initialise context base, and size tracking
+	init_context:
+	mov $r2 0x800
+	shl b32 $r2 6
+	iord $r2 I[$r2 + 0x100]	// CC_SCRATCH[1], initial base
+	clear b32 $r3		// track GPC context size here
+
+	// set mmctx base addresses now so we don't have to do it later,
+	// they don't currently ever change
+	mov $r4 0x700
+	shl b32 $r4 6
+	shr b32 $r5 $r2 8
+	iowr I[$r4 + 0x000] $r5		// MMCTX_SAVE_SWBASE
+	iowr I[$r4 + 0x100] $r5		// MMCTX_LOAD_SWBASE
+
+	// calculate GPC mmio context size, store the chipset-specific
+	// mmio list pointers somewhere we can get at them later without
+	// re-parsing the chipset list
+	clear b32 $r14
+	clear b32 $r15
+	ld b16 $r14 D[$r1 + 4]
+	ld b16 $r15 D[$r1 + 6]
+	st b16 D[$r0 + gpc_mmio_list_head] $r14
+	st b16 D[$r0 + gpc_mmio_list_tail] $r15
+	call mmctx_size
+	add b32 $r2 $r15
+	add b32 $r3 $r15
+
+	// calculate per-TPC mmio context size, store the list pointers
+	ld b16 $r14 D[$r1 + 8]
+	ld b16 $r15 D[$r1 + 10]
+	st b16 D[$r0 + tpc_mmio_list_head] $r14
+	st b16 D[$r0 + tpc_mmio_list_tail] $r15
+	call mmctx_size
+	ld b32 $r14 D[$r0 + tpc_count]
+	mulu $r14 $r15
+	add b32 $r2 $r14
+	add b32 $r3 $r14
+
+	// round up base/size to 256 byte boundary (for strand SWBASE)
+	add b32 $r4 0x1300
+	shr b32 $r3 2
+	iowr I[$r4 + 0x000] $r3		// MMCTX_LOAD_COUNT, wtf for?!?
+	shr b32 $r2 8
+	shr b32 $r3 6
+	add b32 $r2 1
+	add b32 $r3 1
+	shl b32 $r2 8
+	shl b32 $r3 8
+
+	// calculate size of strand context data
+	mov b32 $r15 $r2
+	call strand_ctx_init
+	add b32 $r3 $r15
+
+	// save context size, and tell HUB we're done
+	mov $r1 0x800
+	shl b32 $r1 6
+	iowr I[$r1 + 0x100] $r3		// CC_SCRATCH[1]  = context size
+	add b32 $r1 0x800
+	clear b32 $r2
+	bset $r2 31
+	iowr I[$r1 + 0x000] $r2		// CC_SCRATCH[0] |= 0x80000000
+
+// Main program loop, very simple, sleeps until woken up by the interrupt
+// handler, pulls a command from the queue and executes its handler
+//
+main:
+	bset $flags $p0
+	sleep $p0
+	mov $r13 cmd_queue
+	call queue_get
+	bra $p1 main
+
+	// 0x0000-0x0003 are all context transfers
+	cmpu b32 $r14 0x04
+	bra nc main_not_ctx_xfer
+		// fetch $flags and mask off $p1/$p2
+		mov $r1 $flags
+		mov $r2 0x0006
+		not b32 $r2
+		and $r1 $r2
+		// set $p1/$p2 according to transfer type
+		shl b32 $r14 1
+		or $r1 $r14
+		mov $flags $r1
+		// transfer context data
+		call ctx_xfer
+		bra main
+
+	main_not_ctx_xfer:
+	shl b32 $r15 $r14 16
+	or $r15 E_BAD_COMMAND
+	call error
+	bra main
+
+// interrupt handler
+ih:
+	push $r8
+	mov $r8 $flags
+	push $r8
+	push $r9
+	push $r10
+	push $r11
+	push $r13
+	push $r14
+	push $r15
+
+	// incoming fifo command?
+	iord $r10 I[$r0 + 0x200]	// INTR
+	and $r11 $r10 0x00000004
+	bra e ih_no_fifo
+		// queue incoming fifo command for later processing
+		mov $r11 0x1900
+		mov $r13 cmd_queue
+		iord $r14 I[$r11 + 0x100]	// FIFO_CMD
+		iord $r15 I[$r11 + 0x000]	// FIFO_DATA
+		call queue_put
+		add b32 $r11 0x400
+		mov $r14 1
+		iowr I[$r11 + 0x000] $r14	// FIFO_ACK
+
+	// ack, and wake up main()
+	ih_no_fifo:
+	iowr I[$r0 + 0x100] $r10	// INTR_ACK
+
+	pop $r15
+	pop $r14
+	pop $r13
+	pop $r11
+	pop $r10
+	pop $r9
+	pop $r8
+	mov $flags $r8
+	pop $r8
+	bclr $flags $p0
+	iret
+
+// Set this GPC's bit in HUB_BAR, used to signal completion of various
+// activities to the HUB fuc
+//
+hub_barrier_done:
+	mov $r15 1
+	ld b32 $r14 D[$r0 + gpc_id]
+	shl b32 $r15 $r14
+	mov $r14 -0x6be8 	// 0x409418 - HUB_BAR_SET
+	sethi $r14 0x400000
+	call nv_wr32
+	ret
+
+// Disables various things, waits a bit, and re-enables them..
+//
+// Not sure how exactly this helps, perhaps "ENABLE" is not such a
+// good description for the bits we turn off?  Anyways, without this,
+// funny things happen.
+//
+ctx_redswitch:
+	mov $r14 0x614
+	shl b32 $r14 6
+	mov $r15 0x020
+	iowr I[$r14] $r15	// GPC_RED_SWITCH = POWER
+	mov $r15 8
+	ctx_redswitch_delay:
+		sub b32 $r15 1
+		bra ne ctx_redswitch_delay
+	mov $r15 0xa20
+	iowr I[$r14] $r15	// GPC_RED_SWITCH = UNK11, ENABLE, POWER
+	ret
+
+// Transfer GPC context data between GPU and storage area
+//
+// In: $r15 context base address
+//     $p1 clear on save, set on load
+//     $p2 set if opposite direction done/will be done, so:
+//		on save it means: "a load will follow this save"
+//		on load it means: "a save preceeded this load"
+//
+ctx_xfer:
+	// set context base address
+	mov $r1 0xa04
+	shl b32 $r1 6
+	iowr I[$r1 + 0x000] $r15// MEM_BASE
+	bra not $p1 ctx_xfer_not_load
+		call ctx_redswitch
+	ctx_xfer_not_load:
+
+	// strands
+	mov $r1 0x4afc
+	sethi $r1 0x20000
+	mov $r2 0xc
+	iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x0c
+	call strand_wait
+	mov $r2 0x47fc
+	sethi $r2 0x20000
+	iowr I[$r2] $r0		// STRAND_FIRST_GENE(0x3f) = 0x00
+	xbit $r2 $flags $p1
+	add b32 $r2 3
+	iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x03/0x04 (SAVE/LOAD)
+
+	// mmio context
+	xbit $r10 $flags $p1	// direction
+	or $r10 2		// first
+	mov $r11 0x0000
+	sethi $r11 0x500000
+	ld b32 $r12 D[$r0 + gpc_id]
+	shl b32 $r12 15
+	add b32 $r11 $r12	// base = NV_PGRAPH_GPCn
+	ld b32 $r12 D[$r0 + gpc_mmio_list_head]
+	ld b32 $r13 D[$r0 + gpc_mmio_list_tail]
+	mov $r14 0		// not multi
+	call mmctx_xfer
+
+	// per-TPC mmio context
+	xbit $r10 $flags $p1	// direction
+	or $r10 4		// last
+	mov $r11 0x4000
+	sethi $r11 0x500000	// base = NV_PGRAPH_GPC0_TPC0
+	ld b32 $r12 D[$r0 + gpc_id]
+	shl b32 $r12 15
+	add b32 $r11 $r12	// base = NV_PGRAPH_GPCn_TPC0
+	ld b32 $r12 D[$r0 + tpc_mmio_list_head]
+	ld b32 $r13 D[$r0 + tpc_mmio_list_tail]
+	ld b32 $r15 D[$r0 + tpc_mask]
+	mov $r14 0x800		// stride = 0x800
+	call mmctx_xfer
+
+	// wait for strands to finish
+	call strand_wait
+
+	// if load, or a save without a load following, do some
+	// unknown stuff that's done after finishing a block of
+	// strand commands
+	bra $p1 ctx_xfer_post
+	bra not $p2 ctx_xfer_done
+	ctx_xfer_post:
+		mov $r1 0x4afc
+		sethi $r1 0x20000
+		mov $r2 0xd
+		iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x0d
+		call strand_wait
+
+	// mark completion in HUB's barrier
+	ctx_xfer_done:
+	call hub_barrier_done
+	ret
+
+.align 256

drivers/gpu/drm/nouveau/nvc0_grgpc.fuc.h

+uint32_t nvc0_grgpc_data[] = {
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x000000c0,
+	0x011000b0,
+	0x01640114,
+	0x000000c1,
+	0x011400b0,
+	0x01780114,
+	0x000000c3,
+	0x011000b0,
+	0x01740114,
+	0x000000c4,
+	0x011000b0,
+	0x01740114,
+	0x000000c8,
+	0x011000b0,
+	0x01640114,
+	0x000000ce,
+	0x011000b0,
+	0x01740114,
+	0x00000000,
+	0x00000380,
+	0x14000400,
+	0x20000450,
+	0x00000600,
+	0x00000684,
+	0x10000700,
+	0x00000800,
+	0x08000808,
+	0x00000828,
+	0x00000830,
+	0x000008d8,
+	0x000008e0,
+	0x140008e8,
+	0x0000091c,
+	0x08000924,
+	0x00000b00,
+	0x14000b08,
+	0x00000bb8,
+	0x00000c08,
+	0x1c000c10,
+	0x00000c80,
+	0x00000c8c,
+	0x08001000,
+	0x00001014,
+	0x00000c6c,
+	0x00000018,
+	0x0000003c,
+	0x00000048,
+	0x00000064,
+	0x00000088,
+	0x14000200,
+	0x0400021c,
+	0x14000300,
+	0x000003d0,
+	0x040003e0,
+	0x08000400,
+	0x00000420,
+	0x000004b0,
+	0x000004e8,
+	0x000004f4,
+	0x04000520,
+	0x0c000604,
+	0x4c000644,
+	0x00000698,
+	0x04000750,
+	0x00000758,
+	0x000002c4,
+	0x000004bc,
+	0x000006e0,
+	0x00000544,
+};
+
+uint32_t nvc0_grgpc_code[] = {
+	0x03060ef5,
+	0x9800d898,
+	0x86f001d9,
+	0x0489b808,
+	0xf00c1bf4,
+	0x21f502f7,
+	0x00f802ec,
+	0xb60798c4,
+	0x8dbb0384,
+	0x0880b600,
+	0x80008e80,
+	0x90b6018f,
+	0x0f94f001,
+	0xf801d980,
+	0x0131f400,
+	0x9800d898,
+	0x89b801d9,
+	0x210bf404,
+	0xb60789c4,
+	0x9dbb0394,
+	0x0890b600,
+	0x98009e98,
+	0x80b6019f,
+	0x0f84f001,
+	0xf400d880,
+	0x00f80132,
+	0x0728b7f1,
+	0xb906b4b6,
+	0xc9f002ec,
+	0x00bcd01f,
+	0xc800bccf,
+	0x1bf41fcc,
+	0x06a7f0fa,
+	0x010321f5,
+	0xf840bfcf,
+	0x28b7f100,
+	0x06b4b607,
+	0xb980bfd0,
+	0xc9f002ec,
+	0x1ec9f01f,
+	0xcf00bcd0,
+	0xccc800bc,
+	0xfa1bf41f,
+	0x87f100f8,
+	0x84b60430,
+	0x1ff9f006,
+	0xf8008fd0,
+	0x3087f100,
+	0x0684b604,
+	0xf80080d0,
+	0x3c87f100,
+	0x0684b608,
+	0x99f094bd,
+	0x0089d000,
+	0x081887f1,
+	0xd00684b6,
+	0x87f1008a,
+	0x84b60400,
+	0x0088cf06,
+	0xf4888aff,
+	0x87f1f31b,
+	0x84b6085c,
+	0xf094bd06,
+	0x89d00099,
+	0xf100f800,
+	0xb6083c87,
+	0x94bd0684,
+	0xd00099f0,
+	0x87f10089,
+	0x84b60818,
+	0x008ad006,
+	0x040087f1,
+	0xcf0684b6,
+	0x8aff0088,
+	0xf30bf488,
+	0x085c87f1,
+	0xbd0684b6,
+	0x0099f094,
+	0xf80089d0,
+	0x9894bd00,
+	0x85b600e8,
+	0x0180b61a,
+	0xbb0284b6,
+	0xe0b60098,
+	0x04efb804,
+	0xb9eb1bf4,
+	0x00f8029f,
+	0x083c87f1,
+	0xbd0684b6,
+	0x0199f094,
+	0xf10089d0,
+	0xb6071087,
+	0x94bd0684,
+	0xf405bbfd,
+	0x8bd0090b,
+	0x0099f000,
+	0xf405eefd,
+	0x8ed00c0b,
+	0xc08fd080,
+	0xb70199f0,
+	0xc8010080,
+	0xb4b600ab,
+	0x0cb9f010,
+	0xb601aec8,
+	0xbefd11e4,
+	0x008bd005,
+	0xf0008ecf,
+	0x0bf41fe4,
+	0x00ce98fa,
+	0xd005e9fd,
+	0xc0b6c08e,
+	0x04cdb804,
+	0xc8e81bf4,
+	0x1bf402ab,
+	0x008bcf18,
+	0xb01fb4f0,
+	0x1bf410b4,
+	0x02a7f0f7,
+	0xf4c921f4,
+	0xabc81b0e,
+	0x10b4b600,
+	0xf00cb9f0,
+	0x8bd012b9,
+	0x008bcf00,
+	0xf412bbc8,
+	0x87f1fa1b,
+	0x84b6085c,
+	0xf094bd06,
+	0x89d00199,
+	0xf900f800,
+	0x02a7f0a0,
+	0xfcc921f4,
+	0xf100f8a0,
+	0xf04afc87,
+	0x97f00283,
+	0x0089d00c,
+	0x020721f5,
+	0x87f100f8,
+	0x83f04afc,
+	0x0d97f002,
+	0xf50089d0,
+	0xf8020721,
+	0xfca7f100,
+	0x02a3f04f,
+	0x0500aba2,
+	0xd00fc7f0,
+	0xc7f000ac,
+	0x00bcd00b,
+	0x020721f5,
+	0xf000aed0,
+	0xbcd00ac7,
+	0x0721f500,
+	0xf100f802,
+	0xb6083c87,
+	0x94bd0684,
+	0xd00399f0,
+	0x21f50089,
+	0xe7f00213,
+	0x3921f503,
+	0xfca7f102,
+	0x02a3f046,
+	0x0400aba0,
+	0xf040a0d0,
+	0xbcd001c7,
+	0x0721f500,
+	0x010c9202,
+	0xf000acd0,
+	0xbcd002c7,
+	0x0721f500,
+	0x2621f502,
+	0x8087f102,
+	0x0684b608,
+	0xb70089cf,
+	0x95220080,
+	0x8ed008fe,
+	0x408ed000,
+	0xb6808acf,
+	0xa0b606a5,
+	0x00eabb01,
+	0xb60480b6,
+	0x1bf40192,
+	0x08e4b6e8,
+	0xf1f2efbc,
+	0xb6085c87,
+	0x94bd0684,
+	0xd00399f0,
+	0x00f80089,
+	0xe7f1e0f9,
+	0xe3f09814,
+	0x8d21f440,
+	0x041ce0b7,
+	0xf401f7f0,
+	0xe0fc8d21,
+	0x04bd00f8,
+	0xf10004fe,
+	0xf0120017,
+	0x12d00227,
+	0x3e17f100,
+	0x0010fe04,
+	0x040017f1,
+	0xf0c010d0,
+	0x12d00427,
+	0x1031f400,
+	0x060817f1,
+	0xcf0614b6,
+	0x37f00012,
+	0x1f24f001,
+	0xb60432bb,
+	0x02800132,
+	0x04038003,
+	0x040010b7,
+	0x800012cf,
+	0x27f10002,
+	0x24b60800,
+	0x0022cf06,
+	0xb65817f0,
+	0x13980c10,
+	0x0432b800,
+	0xb00b0bf4,
+	0x1bf40034,
+	0xf100f8f1,
+	0xb6080027,
+	0x22cf0624,
+	0xf134bd40,
+	0xb6070047,
+	0x25950644,
+	0x0045d008,
+	0xbd4045d0,
+	0x58f4bde4,
+	0x1f58021e,
+	0x020e4003,
+	0xf5040f40,
+	0xbb013d21,
+	0x3fbb002f,
+	0x041e5800,
+	0x40051f58,
+	0x0f400a0e,
+	0x3d21f50c,
+	0x030e9801,
+	0xbb00effd,
+	0x3ebb002e,
+	0x0040b700,
+	0x0235b613,
+	0xb60043d0,
+	0x35b60825,
+	0x0120b606,
+	0xb60130b6,
+	0x34b60824,
+	0x022fb908,
+	0x026321f5,
+	0xf1003fbb,
+	0xb6080017,
+	0x13d00614,
+	0x0010b740,
+	0xf024bd08,
+	0x12d01f29,
+	0x0031f400,
+	0xf00028f4,
+	0x21f41cd7,
+	0xf401f439,
+	0xf404e4b0,
+	0x81fe1e18,
+	0x0627f001,
+	0x12fd20bd,
+	0x01e4b604,
+	0xfe051efd,
+	0x21f50018,
+	0x0ef404c3,
+	0x10ef94d3,
+	0xf501f5f0,
+	0xf402ec21,
+	0x80f9c60e,
+	0xf90188fe,
+	0xf990f980,
+	0xf9b0f9a0,
+	0xf9e0f9d0,
+	0x800acff0,
+	0xf404abc4,
+	0xb7f11d0b,
+	0xd7f01900,
+	0x40becf1c,
+	0xf400bfcf,
+	0xb0b70421,
+	0xe7f00400,
+	0x00bed001,
+	0xfc400ad0,
+	0xfce0fcf0,
+	0xfcb0fcd0,
+	0xfc90fca0,
+	0x0088fe80,
+	0x32f480fc,
+	0xf001f800,
+	0x0e9801f7,
+	0x04febb00,
+	0x9418e7f1,
+	0xf440e3f0,
+	0x00f88d21,
+	0x0614e7f1,
+	0xf006e4b6,
+	0xefd020f7,
+	0x08f7f000,
+	0xf401f2b6,
+	0xf7f1fd1b,
+	0xefd00a20,
+	0xf100f800,
+	0xb60a0417,
+	0x1fd00614,
+	0x0711f400,
+	0x04a421f5,
+	0x4afc17f1,
+	0xf00213f0,
+	0x12d00c27,
+	0x0721f500,
+	0xfc27f102,
+	0x0223f047,
+	0xf00020d0,
+	0x20b6012c,
+	0x0012d003,
+	0xf001acf0,
+	0xb7f002a5,
+	0x50b3f000,
+	0xb6000c98,
+	0xbcbb0fc4,
+	0x010c9800,
+	0xf0020d98,
+	0x21f500e7,
+	0xacf0015c,
+	0x04a5f001,
+	0x4000b7f1,
+	0x9850b3f0,
+	0xc4b6000c,
+	0x00bcbb0f,
+	0x98050c98,
+	0x0f98060d,
+	0x00e7f104,
+	0x5c21f508,
+	0x0721f501,
+	0x0601f402,
+	0xf11412f4,
+	0xf04afc17,
+	0x27f00213,
+	0x0012d00d,
+	0x020721f5,
+	0x048f21f5,
+	0x000000f8,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+};

drivers/gpu/drm/nouveau/nvc0_grhub.fuc

+/* fuc microcode for nvc0 PGRAPH/HUB
+ *
+ * Copyright 2011 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+/* To build:
+ *    m4 nvc0_grhub.fuc | envyas -a -w -m fuc -V nva3 -o nvc0_grhub.fuc.h
+ */
+
+.section nvc0_grhub_data
+include(`nvc0_graph.fuc')
+gpc_count:		.b32 0
+rop_count:		.b32 0
+cmd_queue:		queue_init
+hub_mmio_list_head:	.b32 0
+hub_mmio_list_tail:	.b32 0
+
+ctx_current:		.b32 0
+
+chipsets:
+.b8  0xc0 0 0 0
+.b16 nvc0_hub_mmio_head
+.b16 nvc0_hub_mmio_tail
+.b8  0xc1 0 0 0
+.b16 nvc0_hub_mmio_head
+.b16 nvc1_hub_mmio_tail
+.b8  0xc3 0 0 0
+.b16 nvc0_hub_mmio_head
+.b16 nvc0_hub_mmio_tail
+.b8  0xc4 0 0 0
+.b16 nvc0_hub_mmio_head
+.b16 nvc0_hub_mmio_tail
+.b8  0xc8 0 0 0
+.b16 nvc0_hub_mmio_head
+.b16 nvc0_hub_mmio_tail
+.b8  0xce 0 0 0
+.b16 nvc0_hub_mmio_head
+.b16 nvc0_hub_mmio_tail
+.b8  0 0 0 0
+
+nvc0_hub_mmio_head:
+mmctx_data(0x17e91c, 2)
+mmctx_data(0x400204, 2)
+mmctx_data(0x404004, 11)
+mmctx_data(0x404044, 1)
+mmctx_data(0x404094, 14)
+mmctx_data(0x4040d0, 7)
+mmctx_data(0x4040f8, 1)
+mmctx_data(0x404130, 3)
+mmctx_data(0x404150, 3)
+mmctx_data(0x404164, 2)
+mmctx_data(0x404174, 3)
+mmctx_data(0x404200, 8)
+mmctx_data(0x404404, 14)
+mmctx_data(0x404460, 4)
+mmctx_data(0x404480, 1)
+mmctx_data(0x404498, 1)
+mmctx_data(0x404604, 4)
+mmctx_data(0x404618, 32)
+mmctx_data(0x404698, 21)
+mmctx_data(0x4046f0, 2)
+mmctx_data(0x404700, 22)
+mmctx_data(0x405800, 1)
+mmctx_data(0x405830, 3)
+mmctx_data(0x405854, 1)
+mmctx_data(0x405870, 4)
+mmctx_data(0x405a00, 2)
+mmctx_data(0x405a18, 1)
+mmctx_data(0x406020, 1)
+mmctx_data(0x406028, 4)
+mmctx_data(0x4064a8, 2)
+mmctx_data(0x4064b4, 2)
+mmctx_data(0x407804, 1)
+mmctx_data(0x40780c, 6)
+mmctx_data(0x4078bc, 1)
+mmctx_data(0x408000, 7)
+mmctx_data(0x408064, 1)
+mmctx_data(0x408800, 3)
+mmctx_data(0x408900, 4)
+mmctx_data(0x408980, 1)
+nvc0_hub_mmio_tail:
+mmctx_data(0x4064c0, 2)
+nvc1_hub_mmio_tail:
+
+.align 256
+chan_data:
+chan_mmio_count:	.b32 0
+chan_mmio_address:	.b32 0
+
+.align 256
+xfer_data: 		.b32 0
+
+.section nvc0_grhub_code
+bra init
+define(`include_code')
+include(`nvc0_graph.fuc')
+
+// reports an exception to the host
+//
+// In: $r15 error code (see nvc0_graph.fuc)
+//
+error:
+	push $r14
+	mov $r14 0x814
+	shl b32 $r14 6
+	iowr I[$r14 + 0x000] $r15	// CC_SCRATCH[5] = error code
+	mov $r14 0xc1c
+	shl b32 $r14 6
+	mov $r15 1
+	iowr I[$r14 + 0x000] $r15	// INTR_UP_SET
+	pop $r14
+	ret
+
+// HUB fuc initialisation, executed by triggering ucode start, will
+// fall through to main loop after completion.
+//
+// Input:
+//   CC_SCRATCH[0]: chipset (PMC_BOOT_0 read returns 0x0bad0bad... sigh)
+//
+// Output:
+//   CC_SCRATCH[0]:
+//	     31:31: set to signal completion
+//   CC_SCRATCH[1]:
+//	      31:0: total PGRAPH context size
+//
+init:
+	clear b32 $r0
+	mov $sp $r0
+	mov $xdbase $r0
+
+	// enable fifo access
+	mov $r1 0x1200
+	mov $r2 2
+	iowr I[$r1 + 0x000] $r2	// FIFO_ENABLE
+
+	// setup i0 handler, and route all interrupts to it
+	mov $r1 ih
+	mov $iv0 $r1
+	mov $r1 0x400
+	iowr I[$r1 + 0x300] $r0	// INTR_DISPATCH
+
+	// route HUB_CHANNEL_SWITCH to fuc interrupt 8
+	mov $r3 0x404
+	shl b32 $r3 6
+	mov $r2 0x2003		// { HUB_CHANNEL_SWITCH, ZERO } -> intr 8
+	iowr I[$r3 + 0x000] $r2
+
+	// not sure what these are, route them because NVIDIA does, and
+	// the IRQ handler will signal the host if we ever get one.. we
+	// may find out if/why we need to handle these if so..
+	//
+	mov $r2 0x2004
+	iowr I[$r3 + 0x004] $r2 // { 0x04, ZERO } -> intr 9
+	mov $r2 0x200b
+	iowr I[$r3 + 0x008] $r2 // { 0x0b, ZERO } -> intr 10
+	mov $r2 0x200c
+	iowr I[$r3 + 0x01c] $r2 // { 0x0c, ZERO } -> intr 15
+
+	// enable all INTR_UP interrupts
+	mov $r2 0xc24
+	shl b32 $r2 6
+	not b32 $r3 $r0
+	iowr I[$r2] $r3
+
+	// enable fifo, ctxsw, 9, 10, 15 interrupts
+	mov $r2 -0x78fc		// 0x8704
+	sethi $r2 0
+	iowr I[$r1 + 0x000] $r2	// INTR_EN_SET
+
+	// fifo level triggered, rest edge
+	sub b32 $r1 0x100
+	mov $r2 4
+	iowr I[$r1] $r2
+
+	// enable interrupts
+	bset $flags ie0
+
+	// fetch enabled GPC/ROP counts
+	mov $r14 -0x69fc	// 0x409604
+	sethi $r14 0x400000
+	call nv_rd32
+	extr $r1 $r15 16:20
+	st b32 D[$r0 + rop_count] $r1
+	and $r15 0x1f
+	st b32 D[$r0 + gpc_count] $r15
+
+	// set BAR_REQMASK to GPC mask
+	mov $r1 1
+	shl b32 $r1 $r15
+	sub b32 $r1 1
+	mov $r2 0x40c
+	shl b32 $r2 6
+	iowr I[$r2 + 0x000] $r1
+	iowr I[$r2 + 0x100] $r1
+
+	// find context data for this chipset
+	mov $r2 0x800
+	shl b32 $r2 6
+	iord $r2 I[$r2 + 0x000]		// CC_SCRATCH[0]
+	mov $r15 chipsets - 8
+	init_find_chipset:
+		add b32 $r15 8
+		ld b32 $r3 D[$r15 + 0x00]
+		cmpu b32 $r3 $r2
+		bra e init_context
+		cmpu b32 $r3 0
+		bra ne init_find_chipset
+		// unknown chipset
+		ret
+
+	// context size calculation, reserve first 256 bytes for use by fuc
+	init_context:
+	mov $r1 256
+
+	// calculate size of mmio context data
+	ld b16 $r14 D[$r15 + 4]
+	ld b16 $r15 D[$r15 + 6]
+	sethi $r14 0
+	st b32 D[$r0 + hub_mmio_list_head] $r14
+	st b32 D[$r0 + hub_mmio_list_tail] $r15
+	call mmctx_size
+
+	// set mmctx base addresses now so we don't have to do it later,
+	// they don't (currently) ever change
+	mov $r3 0x700
+	shl b32 $r3 6
+	shr b32 $r4 $r1 8
+	iowr I[$r3 + 0x000] $r4		// MMCTX_SAVE_SWBASE
+	iowr I[$r3 + 0x100] $r4		// MMCTX_LOAD_SWBASE
+	add b32 $r3 0x1300
+	add b32 $r1 $r15
+	shr b32 $r15 2
+	iowr I[$r3 + 0x000] $r15	// MMCTX_LOAD_COUNT, wtf for?!?
+
+	// strands, base offset needs to be aligned to 256 bytes
+	shr b32 $r1 8
+	add b32 $r1 1
+	shl b32 $r1 8
+	mov b32 $r15 $r1
+	call strand_ctx_init
+	add b32 $r1 $r15
+
+	// initialise each GPC in sequence by passing in the offset of its
+	// context data in GPCn_CC_SCRATCH[1], and starting its FUC (which
+	// has previously been uploaded by the host) running.
+	//
+	// the GPC fuc init sequence will set GPCn_CC_SCRATCH[0] bit 31
+	// when it has completed, and return the size of its context data
+	// in GPCn_CC_SCRATCH[1]
+	//
+	ld b32 $r3 D[$r0 + gpc_count]
+	mov $r4 0x2000
+	sethi $r4 0x500000
+	init_gpc:
+		// setup, and start GPC ucode running
+		add b32 $r14 $r4 0x804
+		mov b32 $r15 $r1
+		call nv_wr32			// CC_SCRATCH[1] = ctx offset
+		add b32 $r14 $r4 0x800
+		mov b32 $r15 $r2
+		call nv_wr32			// CC_SCRATCH[0] = chipset
+		add b32 $r14 $r4 0x10c
+		clear b32 $r15
+		call nv_wr32
+		add b32 $r14 $r4 0x104
+		call nv_wr32			// ENTRY
+		add b32 $r14 $r4 0x100
+		mov $r15 2			// CTRL_START_TRIGGER
+		call nv_wr32			// CTRL
+
+		// wait for it to complete, and adjust context size
+		add b32 $r14 $r4 0x800
+		init_gpc_wait:
+			call nv_rd32
+			xbit $r15 $r15 31
+			bra e init_gpc_wait
+		add b32 $r14 $r4 0x804
+		call nv_rd32
+		add b32 $r1 $r15
+
+		// next!
+		add b32 $r4 0x8000
+		sub b32 $r3 1
+		bra ne init_gpc
+
+	// save context size, and tell host we're ready
+	mov $r2 0x800
+	shl b32 $r2 6
+	iowr I[$r2 + 0x100] $r1		// CC_SCRATCH[1]  = context size
+	add b32 $r2 0x800
+	clear b32 $r1
+	bset $r1 31
+	iowr I[$r2 + 0x000] $r1		// CC_SCRATCH[0] |= 0x80000000
+
+// Main program loop, very simple, sleeps until woken up by the interrupt
+// handler, pulls a command from the queue and executes its handler
+//
+main:
+	// sleep until we have something to do
+	bset $flags $p0
+	sleep $p0
+	mov $r13 cmd_queue
+	call queue_get
+	bra $p1 main
+
+	// context switch, requested by GPU?
+	cmpu b32 $r14 0x4001
+	bra ne main_not_ctx_switch
+		trace_set(T_AUTO)
+		mov $r1 0xb00
+		shl b32 $r1 6
+		iord $r2 I[$r1 + 0x100]		// CHAN_NEXT
+		iord $r1 I[$r1 + 0x000]		// CHAN_CUR
+
+		xbit $r3 $r1 31
+		bra e chsw_no_prev
+			xbit $r3 $r2 31
+			bra e chsw_prev_no_next
+				push $r2
+				mov b32 $r2 $r1
+				trace_set(T_SAVE)
+				bclr $flags $p1
+				bset $flags $p2
+				call ctx_xfer
+				trace_clr(T_SAVE);
+				pop $r2
+				trace_set(T_LOAD);
+				bset $flags $p1
+				call ctx_xfer
+				trace_clr(T_LOAD);
+				bra chsw_done
+			chsw_prev_no_next:
+				push $r2
+				mov b32 $r2 $r1
+				bclr $flags $p1
+				bclr $flags $p2
+				call ctx_xfer
+				pop $r2
+				mov $r1 0xb00
+				shl b32 $r1 6
+				iowr I[$r1] $r2
+				bra chsw_done
+		chsw_no_prev:
+			xbit $r3 $r2 31
+			bra e chsw_done
+				bset $flags $p1
+				bclr $flags $p2
+				call ctx_xfer
+
+		// ack the context switch request
+		chsw_done:
+		mov $r1 0xb0c
+		shl b32 $r1 6
+		mov $r2 1
+		iowr I[$r1 + 0x000] $r2		// 0x409b0c
+		trace_clr(T_AUTO)
+		bra main
+
+	// request to set current channel? (*not* a context switch)
+	main_not_ctx_switch:
+	cmpu b32 $r14 0x0001
+	bra ne main_not_ctx_chan
+		mov b32 $r2 $r15
+		call ctx_chan
+		bra main_done
+
+	// request to store current channel context?
+	main_not_ctx_chan:
+	cmpu b32 $r14 0x0002
+	bra ne main_not_ctx_save
+		trace_set(T_SAVE)
+		bclr $flags $p1
+		bclr $flags $p2
+		call ctx_xfer
+		trace_clr(T_SAVE)
+		bra main_done
+
+	main_not_ctx_save:
+		shl b32 $r15 $r14 16
+		or $r15 E_BAD_COMMAND
+		call error
+		bra main
+
+	main_done:
+	mov $r1 0x820
+	shl b32 $r1 6
+	clear b32 $r2
+	bset $r2 31
+	iowr I[$r1 + 0x000] $r2		// CC_SCRATCH[0] |= 0x80000000
+	bra main
+
+// interrupt handler
+ih:
+	push $r8
+	mov $r8 $flags
+	push $r8
+	push $r9
+	push $r10
+	push $r11
+	push $r13
+	push $r14
+	push $r15
+
+	// incoming fifo command?
+	iord $r10 I[$r0 + 0x200]	// INTR
+	and $r11 $r10 0x00000004
+	bra e ih_no_fifo
+		// queue incoming fifo command for later processing
+		mov $r11 0x1900
+		mov $r13 cmd_queue
+		iord $r14 I[$r11 + 0x100]	// FIFO_CMD
+		iord $r15 I[$r11 + 0x000]	// FIFO_DATA
+		call queue_put
+		add b32 $r11 0x400
+		mov $r14 1
+		iowr I[$r11 + 0x000] $r14	// FIFO_ACK
+
+	// context switch request?
+	ih_no_fifo:
+	and $r11 $r10 0x00000100
+	bra e ih_no_ctxsw
+		// enqueue a context switch for later processing
+		mov $r13 cmd_queue
+		mov $r14 0x4001
+		call queue_put
+
+	// anything we didn't handle, bring it to the host's attention
+	ih_no_ctxsw:
+	mov $r11 0x104
+	not b32 $r11
+	and $r11 $r10 $r11
+	bra e ih_no_other
+		mov $r10 0xc1c
+		shl b32 $r10 6
+		iowr I[$r10] $r11	// INTR_UP_SET
+
+	// ack, and wake up main()
+	ih_no_other:
+	iowr I[$r0 + 0x100] $r10	// INTR_ACK
+
+	pop $r15
+	pop $r14
+	pop $r13
+	pop $r11
+	pop $r10
+	pop $r9
+	pop $r8
+	mov $flags $r8
+	pop $r8
+	bclr $flags $p0
+	iret
+
+// Not real sure, but, MEM_CMD 7 will hang forever if this isn't done
+ctx_4160s:
+	mov $r14 0x4160
+	sethi $r14 0x400000
+	mov $r15 1
+	call nv_wr32
+	ctx_4160s_wait:
+		call nv_rd32
+		xbit $r15 $r15 4
+		bra e ctx_4160s_wait
+	ret
+
+// Without clearing again at end of xfer, some things cause PGRAPH
+// to hang with STATUS=0x00000007 until it's cleared.. fbcon can
+// still function with it set however...
+ctx_4160c:
+	mov $r14 0x4160
+	sethi $r14 0x400000
+	clear b32 $r15
+	call nv_wr32
+	ret
+
+// Again, not real sure
+//
+// In: $r15 value to set 0x404170 to
+//
+ctx_4170s:
+	mov $r14 0x4170
+	sethi $r14 0x400000
+	or $r15 0x10
+	call nv_wr32
+	ret
+
+// Waits for a ctx_4170s() call to complete
+//
+ctx_4170w:
+	mov $r14 0x4170
+	sethi $r14 0x400000
+	call nv_rd32
+	and $r15 0x10
+	bra ne ctx_4170w
+	ret
+
+// Disables various things, waits a bit, and re-enables them..
+//
+// Not sure how exactly this helps, perhaps "ENABLE" is not such a
+// good description for the bits we turn off?  Anyways, without this,
+// funny things happen.
+//
+ctx_redswitch:
+	mov $r14 0x614
+	shl b32 $r14 6
+	mov $r15 0x270
+	iowr I[$r14] $r15	// HUB_RED_SWITCH = ENABLE_GPC, POWER_ALL
+	mov $r15 8
+	ctx_redswitch_delay:
+		sub b32 $r15 1
+		bra ne ctx_redswitch_delay
+	mov $r15 0x770
+	iowr I[$r14] $r15	// HUB_RED_SWITCH = ENABLE_ALL, POWER_ALL
+	ret
+
+// Not a clue what this is for, except that unless the value is 0x10, the
+// strand context is saved (and presumably restored) incorrectly..
+//
+// In: $r15 value to set to (0x00/0x10 are used)
+//
+ctx_86c:
+	mov $r14 0x86c
+	shl b32 $r14 6
+	iowr I[$r14] $r15	// HUB(0x86c) = val
+	mov $r14 -0x75ec
+	sethi $r14 0x400000
+	call nv_wr32		// ROP(0xa14) = val
+	mov $r14 -0x5794
+	sethi $r14 0x410000
+	call nv_wr32		// GPC(0x86c) = val
+	ret
+
+// ctx_load - load's a channel's ctxctl data, and selects its vm
+//
+// In: $r2 channel address
+//
+ctx_load:
+	trace_set(T_CHAN)
+
+	// switch to channel, somewhat magic in parts..
+	mov $r10 12		// DONE_UNK12
+	call wait_donez
+	mov $r1 0xa24
+	shl b32 $r1 6
+	iowr I[$r1 + 0x000] $r0	// 0x409a24
+	mov $r3 0xb00
+	shl b32 $r3 6
+	iowr I[$r3 + 0x100] $r2	// CHAN_NEXT
+	mov $r1 0xa0c
+	shl b32 $r1 6
+	mov $r4 7
+	iowr I[$r1 + 0x000] $r2 // MEM_CHAN
+	iowr I[$r1 + 0x100] $r4	// MEM_CMD
+	ctx_chan_wait_0:
+		iord $r4 I[$r1 + 0x100]
+		and $r4 0x1f
+		bra ne ctx_chan_wait_0
+	iowr I[$r3 + 0x000] $r2	// CHAN_CUR
+
+	// load channel header, fetch PGRAPH context pointer
+	mov $xtargets $r0
+	bclr $r2 31
+	shl b32 $r2 4
+	add b32 $r2 2
+
+	trace_set(T_LCHAN)
+	mov $r1 0xa04
+	shl b32 $r1 6
+	iowr I[$r1 + 0x000] $r2		// MEM_BASE
+	mov $r1 0xa20
+	shl b32 $r1 6
+	mov $r2 0x0002
+	sethi $r2 0x80000000
+	iowr I[$r1 + 0x000] $r2		// MEM_TARGET = vram
+	mov $r1 0x10			// chan + 0x0210
+	mov $r2 xfer_data
+	sethi $r2 0x00020000		// 16 bytes
+	xdld $r1 $r2
+	xdwait
+	trace_clr(T_LCHAN)
+
+	// update current context
+	ld b32 $r1 D[$r0 + xfer_data + 4]
+	shl b32 $r1 24
+	ld b32 $r2 D[$r0 + xfer_data + 0]
+	shr b32 $r2 8
+	or $r1 $r2
+	st b32 D[$r0 + ctx_current] $r1
+
+	// set transfer base to start of context, and fetch context header
+	trace_set(T_LCTXH)
+	mov $r2 0xa04
+	shl b32 $r2 6
+	iowr I[$r2 + 0x000] $r1		// MEM_BASE
+	mov $r2 1
+	mov $r1 0xa20
+	shl b32 $r1 6
+	iowr I[$r1 + 0x000] $r2		// MEM_TARGET = vm
+	mov $r1 chan_data
+	sethi $r1 0x00060000		// 256 bytes
+	xdld $r0 $r1
+	xdwait
+	trace_clr(T_LCTXH)
+
+	trace_clr(T_CHAN)
+	ret
+
+// ctx_chan - handler for HUB_SET_CHAN command, will set a channel as
+//            the active channel for ctxctl, but not actually transfer
+//            any context data.  intended for use only during initial
+//            context construction.
+//
+// In: $r2 channel address
+//
+ctx_chan:
+	call ctx_4160s
+	call ctx_load
+	mov $r10 12			// DONE_UNK12
+	call wait_donez
+	mov $r1 0xa10
+	shl b32 $r1 6
+	mov $r2 5
+	iowr I[$r1 + 0x000] $r2		// MEM_CMD = 5 (???)
+	ctx_chan_wait:
+		iord $r2 I[$r1 + 0x000]
+		or $r2 $r2
+		bra ne ctx_chan_wait
+	call ctx_4160c
+	ret
+
+// Execute per-context state overrides list
+//
+// Only executed on the first load of a channel.  Might want to look into
+// removing this and having the host directly modify the channel's context
+// to change this state...  The nouveau DRM already builds this list as
+// it's definitely needed for NVIDIA's, so we may as well use it for now
+//
+// Input: $r1 mmio list length
+//
+ctx_mmio_exec:
+	// set transfer base to be the mmio list
+	ld b32 $r3 D[$r0 + chan_mmio_address]
+	mov $r2 0xa04
+	shl b32 $r2 6
+	iowr I[$r2 + 0x000] $r3		// MEM_BASE
+
+	clear b32 $r3
+	ctx_mmio_loop:
+		// fetch next 256 bytes of mmio list if necessary
+		and $r4 $r3 0xff
+		bra ne ctx_mmio_pull
+			mov $r5 xfer_data
+			sethi $r5 0x00060000	// 256 bytes
+			xdld $r3 $r5
+			xdwait
+
+		// execute a single list entry
+		ctx_mmio_pull:
+		ld b32 $r14 D[$r4 + xfer_data + 0x00]
+		ld b32 $r15 D[$r4 + xfer_data + 0x04]
+		call nv_wr32
+
+		// next!
+		add b32 $r3 8
+		sub b32 $r1 1
+		bra ne ctx_mmio_loop
+
+	// set transfer base back to the current context
+	ctx_mmio_done:
+	ld b32 $r3 D[$r0 + ctx_current]
+	iowr I[$r2 + 0x000] $r3		// MEM_BASE
+
+	// disable the mmio list now, we don't need/want to execute it again
+	st b32 D[$r0 + chan_mmio_count] $r0
+	mov $r1 chan_data
+	sethi $r1 0x00060000		// 256 bytes
+	xdst $r0 $r1
+	xdwait
+	ret
+
+// Transfer HUB context data between GPU and storage area
+//
+// In: $r2 channel address
+//     $p1 clear on save, set on load
+//     $p2 set if opposite direction done/will be done, so:
+//		on save it means: "a load will follow this save"
+//		on load it means: "a save preceeded this load"
+//
+ctx_xfer:
+	bra not $p1 ctx_xfer_pre
+	bra $p2 ctx_xfer_pre_load
+	ctx_xfer_pre:
+		mov $r15 0x10
+		call ctx_86c
+		call ctx_4160s
+		bra not $p1 ctx_xfer_exec
+
+	ctx_xfer_pre_load:
+		mov $r15 2
+		call ctx_4170s
+		call ctx_4170w
+		call ctx_redswitch
+		clear b32 $r15
+		call ctx_4170s
+		call ctx_load
+
+	// fetch context pointer, and initiate xfer on all GPCs
+	ctx_xfer_exec:
+	ld b32 $r1 D[$r0 + ctx_current]
+	mov $r2 0x414
+	shl b32 $r2 6
+	iowr I[$r2 + 0x000] $r0	// BAR_STATUS = reset
+	mov $r14 -0x5b00
+	sethi $r14 0x410000
+	mov b32 $r15 $r1
+	call nv_wr32		// GPC_BCAST_WRCMD_DATA = ctx pointer
+	add b32 $r14 4
+	xbit $r15 $flags $p1
+	xbit $r2 $flags $p2
+	shl b32 $r2 1
+	or $r15 $r2
+	call nv_wr32		// GPC_BCAST_WRCMD_CMD = GPC_XFER(type)
+
+	// strands
+	mov $r1 0x4afc
+	sethi $r1 0x20000
+	mov $r2 0xc
+	iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x0c
+	call strand_wait
+	mov $r2 0x47fc
+	sethi $r2 0x20000
+	iowr I[$r2] $r0		// STRAND_FIRST_GENE(0x3f) = 0x00
+	xbit $r2 $flags $p1
+	add b32 $r2 3
+	iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x03/0x04 (SAVE/LOAD)
+
+	// mmio context
+	xbit $r10 $flags $p1	// direction
+	or $r10 6		// first, last
+	mov $r11 0		// base = 0
+	ld b32 $r12 D[$r0 + hub_mmio_list_head]
+	ld b32 $r13 D[$r0 + hub_mmio_list_tail]
+	mov $r14 0		// not multi
+	call mmctx_xfer
+
+	// wait for GPCs to all complete
+	mov $r10 8		// DONE_BAR
+	call wait_doneo
+
+	// wait for strand xfer to complete
+	call strand_wait
+
+	// post-op
+	bra $p1 ctx_xfer_post
+		mov $r10 12		// DONE_UNK12
+		call wait_donez
+		mov $r1 0xa10
+		shl b32 $r1 6
+		mov $r2 5
+		iowr I[$r1] $r2		// MEM_CMD
+		ctx_xfer_post_save_wait:
+			iord $r2 I[$r1]
+			or $r2 $r2
+			bra ne ctx_xfer_post_save_wait
+
+	bra $p2 ctx_xfer_done
+	ctx_xfer_post:
+		mov $r15 2
+		call ctx_4170s
+		clear b32 $r15
+		call ctx_86c
+		call strand_post
+		call ctx_4170w
+		clear b32 $r15
+		call ctx_4170s
+
+		bra not $p1 ctx_xfer_no_post_mmio
+		ld b32 $r1 D[$r0 + chan_mmio_count]
+		or $r1 $r1
+		bra e ctx_xfer_no_post_mmio
+			call ctx_mmio_exec
+
+		ctx_xfer_no_post_mmio:
+		call ctx_4160c
+
+	ctx_xfer_done:
+	ret
+
+.align 256

drivers/gpu/drm/nouveau/nvc0_grhub.fuc.h

+uint32_t nvc0_grhub_data[] = {
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x000000c0,
+	0x012c0090,
+	0x000000c1,
+	0x01300090,
+	0x000000c3,
+	0x012c0090,
+	0x000000c4,
+	0x012c0090,
+	0x000000c8,
+	0x012c0090,
+	0x000000ce,
+	0x012c0090,
+	0x00000000,
+	0x0417e91c,
+	0x04400204,
+	0x28404004,
+	0x00404044,
+	0x34404094,
+	0x184040d0,
+	0x004040f8,
+	0x08404130,
+	0x08404150,
+	0x04404164,
+	0x08404174,
+	0x1c404200,
+	0x34404404,
+	0x0c404460,
+	0x00404480,
+	0x00404498,
+	0x0c404604,
+	0x7c404618,
+	0x50404698,
+	0x044046f0,
+	0x54404700,
+	0x00405800,
+	0x08405830,
+	0x00405854,
+	0x0c405870,
+	0x04405a00,
+	0x00405a18,
+	0x00406020,
+	0x0c406028,
+	0x044064a8,
+	0x044064b4,
+	0x00407804,
+	0x1440780c,
+	0x004078bc,
+	0x18408000,
+	0x00408064,
+	0x08408800,
+	0x0c408900,
+	0x00408980,
+	0x044064c0,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+};
+
+uint32_t nvc0_grhub_code[] = {
+	0x03090ef5,
+	0x9800d898,
+	0x86f001d9,
+	0x0489b808,
+	0xf00c1bf4,
+	0x21f502f7,
+	0x00f802ec,
+	0xb60798c4,
+	0x8dbb0384,
+	0x0880b600,
+	0x80008e80,
+	0x90b6018f,
+	0x0f94f001,
+	0xf801d980,
+	0x0131f400,
+	0x9800d898,
+	0x89b801d9,
+	0x210bf404,
+	0xb60789c4,
+	0x9dbb0394,
+	0x0890b600,
+	0x98009e98,
+	0x80b6019f,
+	0x0f84f001,
+	0xf400d880,
+	0x00f80132,
+	0x0728b7f1,
+	0xb906b4b6,
+	0xc9f002ec,
+	0x00bcd01f,
+	0xc800bccf,
+	0x1bf41fcc,
+	0x06a7f0fa,
+	0x010321f5,
+	0xf840bfcf,
+	0x28b7f100,
+	0x06b4b607,
+	0xb980bfd0,
+	0xc9f002ec,
+	0x1ec9f01f,
+	0xcf00bcd0,
+	0xccc800bc,
+	0xfa1bf41f,
+	0x87f100f8,
+	0x84b60430,
+	0x1ff9f006,
+	0xf8008fd0,
+	0x3087f100,
+	0x0684b604,
+	0xf80080d0,
+	0x3c87f100,
+	0x0684b608,
+	0x99f094bd,
+	0x0089d000,
+	0x081887f1,
+	0xd00684b6,
+	0x87f1008a,
+	0x84b60400,
+	0x0088cf06,
+	0xf4888aff,
+	0x87f1f31b,
+	0x84b6085c,
+	0xf094bd06,
+	0x89d00099,
+	0xf100f800,
+	0xb6083c87,
+	0x94bd0684,
+	0xd00099f0,
+	0x87f10089,
+	0x84b60818,
+	0x008ad006,
+	0x040087f1,
+	0xcf0684b6,
+	0x8aff0088,
+	0xf30bf488,
+	0x085c87f1,
+	0xbd0684b6,
+	0x0099f094,
+	0xf80089d0,
+	0x9894bd00,
+	0x85b600e8,
+	0x0180b61a,
+	0xbb0284b6,
+	0xe0b60098,
+	0x04efb804,
+	0xb9eb1bf4,
+	0x00f8029f,
+	0x083c87f1,
+	0xbd0684b6,
+	0x0199f094,
+	0xf10089d0,
+	0xb6071087,
+	0x94bd0684,
+	0xf405bbfd,
+	0x8bd0090b,
+	0x0099f000,
+	0xf405eefd,
+	0x8ed00c0b,
+	0xc08fd080,
+	0xb70199f0,
+	0xc8010080,
+	0xb4b600ab,
+	0x0cb9f010,
+	0xb601aec8,
+	0xbefd11e4,
+	0x008bd005,
+	0xf0008ecf,
+	0x0bf41fe4,
+	0x00ce98fa,
+	0xd005e9fd,
+	0xc0b6c08e,
+	0x04cdb804,
+	0xc8e81bf4,
+	0x1bf402ab,
+	0x008bcf18,
+	0xb01fb4f0,
+	0x1bf410b4,
+	0x02a7f0f7,
+	0xf4c921f4,
+	0xabc81b0e,
+	0x10b4b600,
+	0xf00cb9f0,
+	0x8bd012b9,
+	0x008bcf00,
+	0xf412bbc8,
+	0x87f1fa1b,
+	0x84b6085c,
+	0xf094bd06,
+	0x89d00199,
+	0xf900f800,
+	0x02a7f0a0,
+	0xfcc921f4,
+	0xf100f8a0,
+	0xf04afc87,
+	0x97f00283,
+	0x0089d00c,
+	0x020721f5,
+	0x87f100f8,
+	0x83f04afc,
+	0x0d97f002,
+	0xf50089d0,
+	0xf8020721,
+	0xfca7f100,
+	0x02a3f04f,
+	0x0500aba2,
+	0xd00fc7f0,
+	0xc7f000ac,
+	0x00bcd00b,
+	0x020721f5,
+	0xf000aed0,
+	0xbcd00ac7,
+	0x0721f500,
+	0xf100f802,
+	0xb6083c87,
+	0x94bd0684,
+	0xd00399f0,
+	0x21f50089,
+	0xe7f00213,
+	0x3921f503,
+	0xfca7f102,
+	0x02a3f046,
+	0x0400aba0,
+	0xf040a0d0,
+	0xbcd001c7,
+	0x0721f500,
+	0x010c9202,
+	0xf000acd0,
+	0xbcd002c7,
+	0x0721f500,
+	0x2621f502,
+	0x8087f102,
+	0x0684b608,
+	0xb70089cf,
+	0x95220080,
+	0x8ed008fe,
+	0x408ed000,
+	0xb6808acf,
+	0xa0b606a5,
+	0x00eabb01,
+	0xb60480b6,
+	0x1bf40192,
+	0x08e4b6e8,
+	0xf1f2efbc,
+	0xb6085c87,
+	0x94bd0684,
+	0xd00399f0,
+	0x00f80089,
+	0xe7f1e0f9,
+	0xe4b60814,
+	0x00efd006,
+	0x0c1ce7f1,
+	0xf006e4b6,
+	0xefd001f7,
+	0xf8e0fc00,
+	0xfe04bd00,
+	0x07fe0004,
+	0x0017f100,
+	0x0227f012,
+	0xf10012d0,
+	0xfe05b917,
+	0x17f10010,
+	0x10d00400,
+	0x0437f1c0,
+	0x0634b604,
+	0x200327f1,
+	0xf10032d0,
+	0xd0200427,
+	0x27f10132,
+	0x32d0200b,
+	0x0c27f102,
+	0x0732d020,
+	0x0c2427f1,
+	0xb90624b6,
+	0x23d00003,
+	0x0427f100,
+	0x0023f087,
+	0xb70012d0,
+	0xf0010012,
+	0x12d00427,
+	0x1031f400,
+	0x9604e7f1,
+	0xf440e3f0,
+	0xf1c76821,
+	0x01018090,
+	0x801ff4f0,
+	0x17f0000f,
+	0x041fbb01,
+	0xf10112b6,
+	0xb6040c27,
+	0x21d00624,
+	0x4021d000,
+	0x080027f1,
+	0xcf0624b6,
+	0xf7f00022,
+	0x08f0b654,
+	0xb800f398,
+	0x0bf40432,
+	0x0034b00b,
+	0xf8f11bf4,
+	0x0017f100,
+	0x02fe5801,
+	0xf003ff58,
+	0x0e8000e3,
+	0x150f8014,
+	0x013d21f5,
+	0x070037f1,
+	0x950634b6,
+	0x34d00814,
+	0x4034d000,
+	0x130030b7,
+	0xb6001fbb,
+	0x3fd002f5,
+	0x0815b600,
+	0xb60110b6,
+	0x1fb90814,
+	0x6321f502,
+	0x001fbb02,
+	0xf1000398,
+	0xf0200047,
+	0x4ea05043,
+	0x1fb90804,
+	0x8d21f402,
+	0x08004ea0,
+	0xf4022fb9,
+	0x4ea08d21,
+	0xf4bd010c,
+	0xa08d21f4,
+	0xf401044e,
+	0x4ea08d21,
+	0xf7f00100,
+	0x8d21f402,
+	0x08004ea0,
+	0xc86821f4,
+	0x0bf41fff,
+	0x044ea0fa,
+	0x6821f408,
+	0xb7001fbb,
+	0xb6800040,
+	0x1bf40132,
+	0x0027f1b4,
+	0x0624b608,
+	0xb74021d0,
+	0xbd080020,
+	0x1f19f014,
+	0xf40021d0,
+	0x28f40031,
+	0x08d7f000,
+	0xf43921f4,
+	0xe4b1f401,
+	0x1bf54001,
+	0x87f100d1,
+	0x84b6083c,
+	0xf094bd06,
+	0x89d00499,
+	0x0017f100,
+	0x0614b60b,
+	0xcf4012cf,
+	0x13c80011,
+	0x7e0bf41f,
+	0xf41f23c8,
+	0x20f95a0b,
+	0xf10212b9,
+	0xb6083c87,
+	0x94bd0684,
+	0xd00799f0,
+	0x32f40089,
+	0x0231f401,
+	0x082921f5,
+	0x085c87f1,
+	0xbd0684b6,
+	0x0799f094,
+	0xfc0089d0,
+	0x3c87f120,
+	0x0684b608,
+	0x99f094bd,
+	0x0089d006,
+	0xf50131f4,
+	0xf1082921,
+	0xb6085c87,
+	0x94bd0684,
+	0xd00699f0,
+	0x0ef40089,
+	0xb920f931,
+	0x32f40212,
+	0x0232f401,
+	0x082921f5,
+	0x17f120fc,
+	0x14b60b00,
+	0x0012d006,
+	0xc8130ef4,
+	0x0bf41f23,
+	0x0131f40d,
+	0xf50232f4,
+	0xf1082921,
+	0xb60b0c17,
+	0x27f00614,
+	0x0012d001,
+	0x085c87f1,
+	0xbd0684b6,
+	0x0499f094,
+	0xf50089d0,
+	0xb0ff200e,
+	0x1bf401e4,
+	0x02f2b90d,
+	0x07b521f5,
+	0xb0420ef4,
+	0x1bf402e4,
+	0x3c87f12e,
+	0x0684b608,
+	0x99f094bd,
+	0x0089d007,
+	0xf40132f4,
+	0x21f50232,
+	0x87f10829,
+	0x84b6085c,
+	0xf094bd06,
+	0x89d00799,
+	0x110ef400,
+	0xf010ef94,
+	0x21f501f5,
+	0x0ef502ec,
+	0x17f1fed1,
+	0x14b60820,
+	0xf024bd06,
+	0x12d01f29,
+	0xbe0ef500,
+	0xfe80f9fe,
+	0x80f90188,
+	0xa0f990f9,
+	0xd0f9b0f9,
+	0xf0f9e0f9,
+	0xc4800acf,
+	0x0bf404ab,
+	0x00b7f11d,
+	0x08d7f019,
+	0xcf40becf,
+	0x21f400bf,
+	0x00b0b704,
+	0x01e7f004,
+	0xe400bed0,
+	0xf40100ab,
+	0xd7f00d0b,
+	0x01e7f108,
+	0x0421f440,
+	0x0104b7f1,
+	0xabffb0bd,
+	0x0d0bf4b4,
+	0x0c1ca7f1,
+	0xd006a4b6,
+	0x0ad000ab,
+	0xfcf0fc40,
+	0xfcd0fce0,
+	0xfca0fcb0,
+	0xfe80fc90,
+	0x80fc0088,
+	0xf80032f4,
+	0x60e7f101,
+	0x40e3f041,
+	0xf401f7f0,
+	0x21f48d21,
+	0x04ffc868,
+	0xf8fa0bf4,
+	0x60e7f100,
+	0x40e3f041,
+	0x21f4f4bd,
+	0xf100f88d,
+	0xf04170e7,
+	0xf5f040e3,
+	0x8d21f410,
+	0xe7f100f8,
+	0xe3f04170,
+	0x6821f440,
+	0xf410f4f0,
+	0x00f8f31b,
+	0x0614e7f1,
+	0xf106e4b6,
+	0xd00270f7,
+	0xf7f000ef,
+	0x01f2b608,
+	0xf1fd1bf4,
+	0xd00770f7,
+	0x00f800ef,
+	0x086ce7f1,
+	0xd006e4b6,
+	0xe7f100ef,
+	0xe3f08a14,
+	0x8d21f440,
+	0xa86ce7f1,
+	0xf441e3f0,
+	0x00f88d21,
+	0x083c87f1,
+	0xbd0684b6,
+	0x0599f094,
+	0xf00089d0,
+	0x21f40ca7,
+	0x2417f1c9,
+	0x0614b60a,
+	0xf10010d0,
+	0xb60b0037,
+	0x32d00634,
+	0x0c17f140,
+	0x0614b60a,
+	0xd00747f0,
+	0x14d00012,
+	0x4014cf40,
+	0xf41f44f0,
+	0x32d0fa1b,
+	0x000bfe00,
+	0xb61f2af0,
+	0x20b60424,
+	0x3c87f102,
+	0x0684b608,
+	0x99f094bd,
+	0x0089d008,
+	0x0a0417f1,
+	0xd00614b6,
+	0x17f10012,
+	0x14b60a20,
+	0x0227f006,
+	0x800023f1,
+	0xf00012d0,
+	0x27f11017,
+	0x23f00300,
+	0x0512fa02,
+	0x87f103f8,
+	0x84b6085c,
+	0xf094bd06,
+	0x89d00899,
+	0xc1019800,
+	0x981814b6,
+	0x25b6c002,
+	0x0512fd08,
+	0xf1160180,
+	0xb6083c87,
+	0x94bd0684,
+	0xd00999f0,
+	0x27f10089,
+	0x24b60a04,
+	0x0021d006,
+	0xf10127f0,
+	0xb60a2017,
+	0x12d00614,
+	0x0017f100,
+	0x0613f002,
+	0xf80501fa,
+	0x5c87f103,
+	0x0684b608,
+	0x99f094bd,
+	0x0089d009,
+	0x085c87f1,
+	0xbd0684b6,
+	0x0599f094,
+	0xf80089d0,
+	0x3121f500,
+	0xb821f506,
+	0x0ca7f006,
+	0xf1c921f4,
+	0xb60a1017,
+	0x27f00614,
+	0x0012d005,
+	0xfd0012cf,
+	0x1bf40522,
+	0x4921f5fa,
+	0x9800f806,
+	0x27f18103,
+	0x24b60a04,
+	0x0023d006,
+	0x34c434bd,
+	0x0f1bf4ff,
+	0x030057f1,
+	0xfa0653f0,
+	0x03f80535,
+	0x98c04e98,
+	0x21f4c14f,
+	0x0830b68d,
+	0xf40112b6,
+	0x0398df1b,
+	0x0023d016,
+	0xf1800080,
+	0xf0020017,
+	0x01fa0613,
+	0xf803f806,
+	0x0611f400,
+	0xf01102f4,
+	0x21f510f7,
+	0x21f50698,
+	0x11f40631,
+	0x02f7f01c,
+	0x065721f5,
+	0x066621f5,
+	0x067821f5,
+	0x21f5f4bd,
+	0x21f50657,
+	0x019806b8,
+	0x1427f116,
+	0x0624b604,
+	0xf10020d0,
+	0xf0a500e7,
+	0x1fb941e3,
+	0x8d21f402,
+	0xf004e0b6,
+	0x2cf001fc,
+	0x0124b602,
+	0xf405f2fd,
+	0x17f18d21,
+	0x13f04afc,
+	0x0c27f002,
+	0xf50012d0,
+	0xf1020721,
+	0xf047fc27,
+	0x20d00223,
+	0x012cf000,
+	0xd00320b6,
+	0xacf00012,
+	0x06a5f001,
+	0x9800b7f0,
+	0x0d98140c,
+	0x00e7f015,
+	0x015c21f5,
+	0xf508a7f0,
+	0xf5010321,
+	0xf4020721,
+	0xa7f02201,
+	0xc921f40c,
+	0x0a1017f1,
+	0xf00614b6,
+	0x12d00527,
+	0x0012cf00,
+	0xf40522fd,
+	0x02f4fa1b,
+	0x02f7f032,
+	0x065721f5,
+	0x21f5f4bd,
+	0x21f50698,
+	0x21f50226,
+	0xf4bd0666,
+	0x065721f5,
+	0x981011f4,
+	0x11fd8001,
+	0x070bf405,
+	0x07df21f5,
+	0x064921f5,
+	0x000000f8,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+};

drivers/gpu/drm/nouveau/nvc0_instmem.c

@@ -32,7 +32,6 @@ struct nvc0_instmem_priv {
 	struct nouveau_channel *bar1;
 	struct nouveau_gpuobj  *bar3_pgd;
 	struct nouveau_channel *bar3;
-	struct nouveau_gpuobj  *chan_pgd;
 };
 
 int
@@ -181,17 +180,11 @@ nvc0_instmem_init(struct drm_device *dev)
 		goto error;
 
 	/* channel vm */
-	ret = nouveau_vm_new(dev, 0, (1ULL << 40), 0x0008000000ULL, &vm);
+	ret = nouveau_vm_new(dev, 0, (1ULL << 40), 0x0008000000ULL,
+			     &dev_priv->chan_vm);
 	if (ret)
 		goto error;
 
-	ret = nouveau_gpuobj_new(dev, NULL, 0x8000, 4096, 0, &priv->chan_pgd);
-	if (ret)
-		goto error;
-
-	nouveau_vm_ref(vm, &dev_priv->chan_vm, priv->chan_pgd);
-	nouveau_vm_ref(NULL, &vm, NULL);
-
 	nvc0_instmem_resume(dev);
 	return 0;
 error:
@@ -211,8 +204,7 @@ nvc0_instmem_takedown(struct drm_device *dev)
 	nv_wr32(dev, 0x1704, 0x00000000);
 	nv_wr32(dev, 0x1714, 0x00000000);
 
-	nouveau_vm_ref(NULL, &dev_priv->chan_vm, priv->chan_pgd);
-	nouveau_gpuobj_ref(NULL, &priv->chan_pgd);
+	nouveau_vm_ref(NULL, &dev_priv->chan_vm, NULL);
 
 	nvc0_channel_del(&priv->bar1);
 	nouveau_vm_ref(NULL, &dev_priv->bar1_vm, priv->bar1_pgd);

drivers/gpu/drm/nouveau/nvc0_vm.c

@@ -105,7 +105,11 @@ nvc0_vm_flush(struct nouveau_vm *vm)
 	struct drm_device *dev = vm->dev;
 	struct nouveau_vm_pgd *vpgd;
 	unsigned long flags;
-	u32 engine = (dev_priv->chan_vm == vm) ? 1 : 5;
+	u32 engine;
+
+	engine = 1;
+	if (vm == dev_priv->bar1_vm || vm == dev_priv->bar3_vm)
+		engine |= 4;
 
 	pinstmem->flush(vm->dev);
 

drivers/gpu/drm/nouveau/nvc0_vram.c

@@ -61,9 +61,7 @@ nvc0_vram_new(struct drm_device *dev, u64 size, u32 align, u32 ncmin,
 	      u32 type, struct nouveau_mem **pmem)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct ttm_bo_device *bdev = &dev_priv->ttm.bdev;
-	struct ttm_mem_type_manager *man = &bdev->man[TTM_PL_VRAM];
-	struct nouveau_mm *mm = man->priv;
+	struct nouveau_mm *mm = dev_priv->engine.vram.mm;
 	struct nouveau_mm_node *r;
 	struct nouveau_mem *mem;
 	int ret;
@@ -105,9 +103,15 @@ int
 nvc0_vram_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_vram_engine *vram = &dev_priv->engine.vram;
+	const u32 rsvd_head = ( 256 * 1024) >> 12; /* vga memory */
+	const u32 rsvd_tail = (1024 * 1024) >> 12; /* vbios etc */
+	u32 length;
 
 	dev_priv->vram_size  = nv_rd32(dev, 0x10f20c) << 20;
 	dev_priv->vram_size *= nv_rd32(dev, 0x121c74);
-	dev_priv->vram_rblock_size = 4096;
-	return 0;
+
+	length = (dev_priv->vram_size >> 12) - rsvd_head - rsvd_tail;
+
+	return nouveau_mm_init(&vram->mm, rsvd_head, length, 1);
 }