Merge remote branch 'nouveau/drm-nouveau-next' of /ssd/git/drm-nouveau-next into drm-core-next

* 'nouveau/drm-nouveau-next' of /ssd/git/drm-nouveau-next:
  drm/nvc0: accelerate ttm buffer moves
  drm/nvc0: initial support for tiled buffer objects
  drm/nvc0: implement fbcon acceleration
  drm/nvc0: implement pgraph engine hooks
  drm/nvc0: implement pfifo engine hooks
  drm/nvc0: implement fencing
  drm/nvc0: fix channel dma init paths
  drm/nvc0: skip dma object creation for drm channel
  drm/nvc0: implement channel structure initialisation
  drm/nvc0: gpuobj_new need only check validity and init the relevant engine
  drm/nvc0: reject the notifier_alloc ioctl
  drm/nvc0: create shared channel vm
  drm/nvc0: initial vm implementation, use for bar1/bar3 management
  drm/nvc0: import initial vm backend
  drm/nouveau: modify vm to accomodate dual page tables for nvc0
  drm/nv50: add missing license header to nv50_fbcon.c
  drm/nv50: fix smatch warning in nv50_vram.c
  drm/nouveau: sizeof() vs ARRAY_SIZE()

30 files changed, 5044 insertions, 341 deletions

diff --git a/drivers/gpu/drm/nouveau/Makefile b/drivers/gpu/drm/nouveau/Makefile
index b1d8941e04d..e12c97fd8db 100644
--- a/drivers/gpu/drm/nouveau/Makefile
+++ b/drivers/gpu/drm/nouveau/Makefile
@@ -18,17 +18,19 @@ nouveau-y := nouveau_drv.o nouveau_state.o nouveau_channel.o nouveau_mem.o \
              nv04_fifo.o nv10_fifo.o nv40_fifo.o nv50_fifo.o nvc0_fifo.o \
              nv04_graph.o nv10_graph.o nv20_graph.o \
              nv40_graph.o nv50_graph.o nvc0_graph.o \
-             nv40_grctx.o nv50_grctx.o \
+             nv40_grctx.o nv50_grctx.o nvc0_grctx.o \
              nv84_crypt.o \
              nv04_instmem.o nv50_instmem.o nvc0_instmem.o \
              nv50_evo.o nv50_crtc.o nv50_dac.o nv50_sor.o \
-             nv50_cursor.o nv50_display.o nv50_fbcon.o \
+             nv50_cursor.o nv50_display.o \
              nv04_dac.o nv04_dfp.o nv04_tv.o nv17_tv.o nv17_tv_modes.o \
-             nv04_crtc.o nv04_display.o nv04_cursor.o nv04_fbcon.o \
+             nv04_crtc.o nv04_display.o nv04_cursor.o \
+             nv04_fbcon.o nv50_fbcon.o nvc0_fbcon.o \
              nv10_gpio.o nv50_gpio.o \
 	     nv50_calc.o \
 	     nv04_pm.o nv50_pm.o nva3_pm.o \
-	     nv50_vram.o nv50_vm.o
+	     nv50_vram.o nvc0_vram.o \
+	     nv50_vm.o nvc0_vm.o
 
 nouveau-$(CONFIG_DRM_NOUVEAU_DEBUG) += nouveau_debugfs.o
 nouveau-$(CONFIG_COMPAT) += nouveau_ioc32.o
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 42d1ad62b38..a7fae26f465 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -413,7 +413,7 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 		man->default_caching = TTM_PL_FLAG_CACHED;
 		break;
 	case TTM_PL_VRAM:
-		if (dev_priv->card_type == NV_50) {
+		if (dev_priv->card_type >= NV_50) {
 			man->func = &nouveau_vram_manager;
 			man->io_reserve_fastpath = false;
 			man->use_io_reserve_lru = true;
@@ -515,6 +515,58 @@ nouveau_bo_mem_ctxdma(struct ttm_buffer_object *bo,
 }
 
 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 drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev);
+	struct nouveau_bo *nvbo = nouveau_bo(bo);
+	u64 src_offset = old_mem->start << PAGE_SHIFT;
+	u64 dst_offset = new_mem->start << PAGE_SHIFT;
+	u32 page_count = new_mem->num_pages;
+	int ret;
+
+	if (!nvbo->no_vm) {
+		if (old_mem->mem_type == TTM_PL_VRAM)
+			src_offset  = nvbo->vma.offset;
+		else
+			src_offset += dev_priv->gart_info.aper_base;
+
+		if (new_mem->mem_type == TTM_PL_VRAM)
+			dst_offset  = nvbo->vma.offset;
+		else
+			dst_offset += dev_priv->gart_info.aper_base;
+	}
+
+	page_count = new_mem->num_pages;
+	while (page_count) {
+		int line_count = (page_count > 2047) ? 2047 : page_count;
+
+		ret = RING_SPACE(chan, 12);
+		if (ret)
+			return ret;
+
+		BEGIN_NVC0(chan, 2, NvSubM2MF, 0x0238, 2);
+		OUT_RING  (chan, upper_32_bits(dst_offset));
+		OUT_RING  (chan, lower_32_bits(dst_offset));
+		BEGIN_NVC0(chan, 2, NvSubM2MF, 0x030c, 6);
+		OUT_RING  (chan, upper_32_bits(src_offset));
+		OUT_RING  (chan, lower_32_bits(src_offset));
+		OUT_RING  (chan, PAGE_SIZE); /* src_pitch */
+		OUT_RING  (chan, PAGE_SIZE); /* dst_pitch */
+		OUT_RING  (chan, PAGE_SIZE); /* line_length */
+		OUT_RING  (chan, line_count);
+		BEGIN_NVC0(chan, 2, NvSubM2MF, 0x0300, 1);
+		OUT_RING  (chan, 0x00100110);
+
+		page_count -= line_count;
+		src_offset += (PAGE_SIZE * line_count);
+		dst_offset += (PAGE_SIZE * line_count);
+	}
+
+	return 0;
+}
+
+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)
 {
@@ -690,7 +742,10 @@ nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr,
 	if (dev_priv->card_type < NV_50)
 		ret = nv04_bo_move_m2mf(chan, bo, &bo->mem, new_mem);
 	else
+	if (dev_priv->card_type < NV_C0)
 		ret = nv50_bo_move_m2mf(chan, bo, &bo->mem, new_mem);
+	else
+		ret = nvc0_bo_move_m2mf(chan, bo, &bo->mem, new_mem);
 	if (ret == 0) {
 		ret = nouveau_bo_move_accel_cleanup(chan, nvbo, evict,
 						    no_wait_reserve,
@@ -901,6 +956,7 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 	case TTM_PL_VRAM:
 	{
 		struct nouveau_vram *vram = mem->mm_node;
+		u8 page_shift;
 
 		if (!dev_priv->bar1_vm) {
 			mem->bus.offset = mem->start << PAGE_SHIFT;
@@ -909,8 +965,14 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 			break;
 		}
 
-		ret = nouveau_vm_get(dev_priv->bar1_vm, mem->bus.size, 12,
-				     NV_MEM_ACCESS_RW, &vram->bar_vma);
+		if (dev_priv->card_type == NV_C0)
+			page_shift = vram->page_shift;
+		else
+			page_shift = 12;
+
+		ret = nouveau_vm_get(dev_priv->bar1_vm, mem->bus.size,
+				     page_shift, NV_MEM_ACCESS_RW,
+				     &vram->bar_vma);
 		if (ret)
 			return ret;
 
@@ -920,8 +982,9 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 			return ret;
 		}
 
-		mem->bus.offset  = vram->bar_vma.offset;
-		mem->bus.offset -= 0x0020000000ULL;
+		mem->bus.offset = vram->bar_vma.offset;
+		if (dev_priv->card_type == NV_50) /*XXX*/
+			mem->bus.offset -= 0x0020000000ULL;
 		mem->bus.base = pci_resource_start(dev->pdev, 1);
 		mem->bus.is_iomem = true;
 	}
diff --git a/drivers/gpu/drm/nouveau/nouveau_channel.c b/drivers/gpu/drm/nouveau/nouveau_channel.c
index 6f37995aee2..e37977d0246 100644
--- a/drivers/gpu/drm/nouveau/nouveau_channel.c
+++ b/drivers/gpu/drm/nouveau/nouveau_channel.c
@@ -38,9 +38,14 @@ nouveau_channel_pushbuf_ctxdma_init(struct nouveau_channel *chan)
 	int ret;
 
 	if (dev_priv->card_type >= NV_50) {
-		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY, 0,
-					     (1ULL << 40), NV_MEM_ACCESS_RO,
-					     NV_MEM_TARGET_VM, &pushbuf);
+		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_base = pb->bo.offset;
 	} else
 	if (pb->bo.mem.mem_type == TTM_PL_TT) {
@@ -71,7 +76,7 @@ nouveau_channel_pushbuf_ctxdma_init(struct nouveau_channel *chan)
 
 	nouveau_gpuobj_ref(pushbuf, &chan->pushbuf);
 	nouveau_gpuobj_ref(NULL, &pushbuf);
-	return 0;
+	return ret;
 }
 
 static struct nouveau_bo *
@@ -99,6 +104,13 @@ nouveau_channel_user_pushbuf_alloc(struct drm_device *dev)
 		return NULL;
 	}
 
+	ret = nouveau_bo_map(pushbuf);
+	if (ret) {
+		nouveau_bo_unpin(pushbuf);
+		nouveau_bo_ref(NULL, &pushbuf);
+		return NULL;
+	}
+
 	return pushbuf;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_dma.c b/drivers/gpu/drm/nouveau/nouveau_dma.c
index 6ff77cedc00..65699bfaaae 100644
--- a/drivers/gpu/drm/nouveau/nouveau_dma.c
+++ b/drivers/gpu/drm/nouveau/nouveau_dma.c
@@ -36,7 +36,7 @@ nouveau_dma_pre_init(struct nouveau_channel *chan)
 	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
 	struct nouveau_bo *pushbuf = chan->pushbuf_bo;
 
-	if (dev_priv->card_type == NV_50) {
+	if (dev_priv->card_type >= NV_50) {
 		const int ib_size = pushbuf->bo.mem.size / 2;
 
 		chan->dma.ib_base = (pushbuf->bo.mem.size - ib_size) >> 2;
@@ -61,6 +61,21 @@ nouveau_dma_init(struct nouveau_channel *chan)
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	int ret, i;
 
+	if (dev_priv->card_type >= NV_C0) {
+		ret = nouveau_gpuobj_gr_new(chan, 0x9039, 0x9039);
+		if (ret)
+			return ret;
+
+		ret = RING_SPACE(chan, 2);
+		if (ret)
+			return ret;
+
+		BEGIN_NVC0(chan, 2, NvSubM2MF, 0x0000, 1);
+		OUT_RING  (chan, 0x00009039);
+		FIRE_RING (chan);
+		return 0;
+	}
+
 	/* Create NV_MEMORY_TO_MEMORY_FORMAT for buffer moves */
 	ret = nouveau_gpuobj_gr_new(chan, NvM2MF, dev_priv->card_type < NV_50 ?
 				    0x0039 : 0x5039);
@@ -72,11 +87,6 @@ nouveau_dma_init(struct nouveau_channel *chan)
 	if (ret)
 		return ret;
 
-	/* Map push buffer */
-	ret = nouveau_bo_map(chan->pushbuf_bo);
-	if (ret)
-		return ret;
-
 	/* Insert NOPS for NOUVEAU_DMA_SKIPS */
 	ret = RING_SPACE(chan, NOUVEAU_DMA_SKIPS);
 	if (ret)
diff --git a/drivers/gpu/drm/nouveau/nouveau_dma.h b/drivers/gpu/drm/nouveau/nouveau_dma.h
index d578c21d3c8..c36f1763fea 100644
--- a/drivers/gpu/drm/nouveau/nouveau_dma.h
+++ b/drivers/gpu/drm/nouveau/nouveau_dma.h
@@ -77,7 +77,8 @@ enum {
 	/* G80+ display objects */
 	NvEvoVRAM	= 0x01000000,
 	NvEvoFB16	= 0x01000001,
-	NvEvoFB32	= 0x01000002
+	NvEvoFB32	= 0x01000002,
+	NvEvoVRAM_LP	= 0x01000003
 };
 
 #define NV_MEMORY_TO_MEMORY_FORMAT                                    0x00000039
@@ -125,6 +126,12 @@ extern void
 OUT_RINGp(struct nouveau_channel *chan, const void *data, unsigned nr_dwords);
 
 static inline void
+BEGIN_NVC0(struct nouveau_channel *chan, int op, int subc, int mthd, int size)
+{
+	OUT_RING(chan, (op << 28) | (size << 16) | (subc << 13) | (mthd >> 2));
+}
+
+static inline void
 BEGIN_RING(struct nouveau_channel *chan, int subc, int mthd, int size)
 {
 	OUT_RING(chan, (subc << 13) | (size << 18) | mthd);
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index 8f13906185b..e8157568735 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -69,6 +69,7 @@ struct nouveau_vram {
 	struct drm_device *dev;
 
 	struct nouveau_vma bar_vma;
+	u8  page_shift;
 
 	struct list_head regions;
 	u32 memtype;
@@ -239,6 +240,7 @@ struct nouveau_channel {
 	/* PFIFO context */
 	struct nouveau_gpuobj *ramfc;
 	struct nouveau_gpuobj *cache;
+	void *fifo_priv;
 
 	/* PGRAPH context */
 	/* XXX may be merge 2 pointers as private data ??? */
@@ -336,6 +338,7 @@ struct nouveau_fb_engine {
 };
 
 struct nouveau_fifo_engine {
+	void *priv;
 	int  channels;
 
 	struct nouveau_gpuobj *playlist[2];
@@ -362,6 +365,7 @@ struct nouveau_pgraph_engine {
 	bool accel_blocked;
 	bool registered;
 	int grctx_size;
+	void *priv;
 
 	/* NV2x/NV3x context table (0x400780) */
 	struct nouveau_gpuobj *ctx_table;
@@ -841,6 +845,9 @@ extern void nv10_mem_put_tile_region(struct drm_device *dev,
 				     struct nouveau_fence *fence);
 extern const struct ttm_mem_type_manager_func nouveau_vram_manager;
 
+/* nvc0_vram.c */
+extern const struct ttm_mem_type_manager_func nvc0_vram_manager;
+
 /* nouveau_notifier.c */
 extern int  nouveau_notifier_init_channel(struct nouveau_channel *);
 extern void nouveau_notifier_takedown_channel(struct nouveau_channel *);
@@ -1228,11 +1235,6 @@ extern int  nvc0_instmem_init(struct drm_device *);
 extern void nvc0_instmem_takedown(struct drm_device *);
 extern int  nvc0_instmem_suspend(struct drm_device *);
 extern void nvc0_instmem_resume(struct drm_device *);
-extern int  nvc0_instmem_get(struct nouveau_gpuobj *, u32 size, u32 align);
-extern void nvc0_instmem_put(struct nouveau_gpuobj *);
-extern int  nvc0_instmem_map(struct nouveau_gpuobj *);
-extern void nvc0_instmem_unmap(struct nouveau_gpuobj *);
-extern void nvc0_instmem_flush(struct drm_device *);
 
 /* nv04_mc.c */
 extern int  nv04_mc_init(struct drm_device *);
diff --git a/drivers/gpu/drm/nouveau/nouveau_fbcon.c b/drivers/gpu/drm/nouveau/nouveau_fbcon.c
index 9dbe1eff2ae..6d56a54b6e2 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fbcon.c
@@ -68,6 +68,8 @@ nouveau_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect)
 		else
 		if (dev_priv->card_type < NV_C0)
 			ret = nv50_fbcon_fillrect(info, rect);
+		else
+			ret = nvc0_fbcon_fillrect(info, rect);
 		mutex_unlock(&dev_priv->channel->mutex);
 	}
 
@@ -98,6 +100,8 @@ nouveau_fbcon_copyarea(struct fb_info *info, const struct fb_copyarea *image)
 		else
 		if (dev_priv->card_type < NV_C0)
 			ret = nv50_fbcon_copyarea(info, image);
+		else
+			ret = nvc0_fbcon_copyarea(info, image);
 		mutex_unlock(&dev_priv->channel->mutex);
 	}
 
@@ -128,6 +132,8 @@ nouveau_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 		else
 		if (dev_priv->card_type < NV_C0)
 			ret = nv50_fbcon_imageblit(info, image);
+		else
+			ret = nvc0_fbcon_imageblit(info, image);
 		mutex_unlock(&dev_priv->channel->mutex);
 	}
 
@@ -163,10 +169,18 @@ nouveau_fbcon_sync(struct fb_info *info)
 		return 0;
 	}
 
-	BEGIN_RING(chan, 0, 0x0104, 1);
-	OUT_RING(chan, 0);
-	BEGIN_RING(chan, 0, 0x0100, 1);
-	OUT_RING(chan, 0);
+	if (dev_priv->card_type >= NV_C0) {
+		BEGIN_NVC0(chan, 2, NvSub2D, 0x010c, 1);
+		OUT_RING  (chan, 0);
+		BEGIN_NVC0(chan, 2, NvSub2D, 0x0100, 1);
+		OUT_RING  (chan, 0);
+	} else {
+		BEGIN_RING(chan, 0, 0x0104, 1);
+		OUT_RING  (chan, 0);
+		BEGIN_RING(chan, 0, 0x0100, 1);
+		OUT_RING  (chan, 0);
+	}
+
 	nouveau_bo_wr32(chan->notifier_bo, chan->m2mf_ntfy + 3, 0xffffffff);
 	FIRE_RING(chan);
 	mutex_unlock(&chan->mutex);
@@ -370,6 +384,8 @@ nouveau_fbcon_create(struct nouveau_fbdev *nfbdev,
 		else
 		if (dev_priv->card_type < NV_C0)
 			ret = nv50_fbcon_accel_init(info);
+		else
+			ret = nvc0_fbcon_accel_init(info);
 
 		if (ret == 0)
 			info->fbops = &nouveau_fbcon_ops;
diff --git a/drivers/gpu/drm/nouveau/nouveau_fbcon.h b/drivers/gpu/drm/nouveau/nouveau_fbcon.h
index 6b933f2c3a5..b73c29f87fc 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fbcon.h
+++ b/drivers/gpu/drm/nouveau/nouveau_fbcon.h
@@ -44,11 +44,17 @@ int nv04_fbcon_copyarea(struct fb_info *info, const struct fb_copyarea *region);
 int nv04_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect);
 int nv04_fbcon_imageblit(struct fb_info *info, const struct fb_image *image);
 int nv04_fbcon_accel_init(struct fb_info *info);
+
 int nv50_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect);
 int nv50_fbcon_copyarea(struct fb_info *info, const struct fb_copyarea *region);
 int nv50_fbcon_imageblit(struct fb_info *info, const struct fb_image *image);
 int nv50_fbcon_accel_init(struct fb_info *info);
 
+int nvc0_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect);
+int nvc0_fbcon_copyarea(struct fb_info *info, const struct fb_copyarea *region);
+int nvc0_fbcon_imageblit(struct fb_info *info, const struct fb_image *image);
+int nvc0_fbcon_accel_init(struct fb_info *info);
+
 void nouveau_fbcon_gpu_lockup(struct fb_info *info);
 
 int nouveau_fbcon_init(struct drm_device *dev);
diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.c b/drivers/gpu/drm/nouveau/nouveau_fence.c
index 374a9793b85..88b2f29ca3e 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fence.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fence.c
@@ -32,7 +32,8 @@
 #include "nouveau_dma.h"
 
 #define USE_REFCNT(dev) (nouveau_private(dev)->chipset >= 0x10)
-#define USE_SEMA(dev) (nouveau_private(dev)->chipset >= 0x17)
+#define USE_SEMA(dev) (nouveau_private(dev)->chipset >= 0x17 && \
+		       nouveau_private(dev)->card_type < NV_C0)
 
 struct nouveau_fence {
 	struct nouveau_channel *channel;
@@ -139,6 +140,7 @@ nouveau_fence_emit(struct nouveau_fence *fence)
 {
 	struct nouveau_channel *chan = fence->channel;
 	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	int ret;
 
 	ret = RING_SPACE(chan, 2);
@@ -159,8 +161,15 @@ nouveau_fence_emit(struct nouveau_fence *fence)
 	list_add_tail(&fence->entry, &chan->fence.pending);
 	spin_unlock(&chan->fence.lock);
 
-	BEGIN_RING(chan, NvSubSw, USE_REFCNT(dev) ? 0x0050 : 0x0150, 1);
-	OUT_RING(chan, fence->sequence);
+	if (USE_REFCNT(dev)) {
+		if (dev_priv->card_type < NV_C0)
+			BEGIN_RING(chan, NvSubSw, 0x0050, 1);
+		else
+			BEGIN_NVC0(chan, 2, NvSubSw, 0x0050, 1);
+	} else {
+		BEGIN_RING(chan, NvSubSw, 0x0150, 1);
+	}
+	OUT_RING (chan, fence->sequence);
 	FIRE_RING(chan);
 
 	return 0;
@@ -445,11 +454,14 @@ nouveau_fence_channel_init(struct nouveau_channel *chan)
 	if (ret)
 		return ret;
 
-	ret = RING_SPACE(chan, 2);
-	if (ret)
-		return ret;
-	BEGIN_RING(chan, NvSubSw, 0, 1);
-	OUT_RING(chan, NvSw);
+	/* we leave subchannel empty for nvc0 */
+	if (dev_priv->card_type < NV_C0) {
+		ret = RING_SPACE(chan, 2);
+		if (ret)
+			return ret;
+		BEGIN_RING(chan, NvSubSw, 0, 1);
+		OUT_RING(chan, NvSw);
+	}
 
 	/* Create a DMA object for the shared cross-channel sync area. */
 	if (USE_SEMA(dev)) {
diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c b/drivers/gpu/drm/nouveau/nouveau_mem.c
index 224181193a1..69044eb104b 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_mem.c
@@ -255,9 +255,6 @@ nouveau_mem_detect(struct drm_device *dev)
 	if (dev_priv->card_type < NV_50) {
 		dev_priv->vram_size  = nv_rd32(dev, NV04_PFB_FIFO_DATA);
 		dev_priv->vram_size &= NV10_PFB_FIFO_DATA_RAM_AMOUNT_MB_MASK;
-	} else {
-		dev_priv->vram_size  = nv_rd32(dev, 0x10f20c) << 20;
-		dev_priv->vram_size *= nv_rd32(dev, 0x121c74);
 	}
 
 	if (dev_priv->vram_size)
@@ -731,6 +728,10 @@ nouveau_vram_manager_new(struct ttm_mem_type_manager *man,
 	if (ret)
 		return ret;
 
+	node->page_shift = 12;
+	if (nvbo->vma.node)
+		node->page_shift = nvbo->vma.node->type;
+
 	mem->mm_node = node;
 	mem->start   = node->offset >> PAGE_SHIFT;
 	return 0;
diff --git a/drivers/gpu/drm/nouveau/nouveau_mm.h b/drivers/gpu/drm/nouveau/nouveau_mm.h
index 250e642de0a..af384493303 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mm.h
+++ b/drivers/gpu/drm/nouveau/nouveau_mm.h
@@ -59,4 +59,9 @@ int  nv50_vram_new(struct drm_device *, u64 size, u32 align, u32 size_nc,
 void nv50_vram_del(struct drm_device *, struct nouveau_vram **);
 bool nv50_vram_flags_valid(struct drm_device *, u32 tile_flags);
 
+int  nvc0_vram_init(struct drm_device *);
+int  nvc0_vram_new(struct drm_device *, u64 size, u32 align, u32 ncmin,
+		    u32 memtype, struct nouveau_vram **);
+bool nvc0_vram_flags_valid(struct drm_device *, u32 tile_flags);
+
 #endif
diff --git a/drivers/gpu/drm/nouveau/nouveau_notifier.c b/drivers/gpu/drm/nouveau/nouveau_notifier.c
index a050b7b6978..fe29d604b82 100644
--- a/drivers/gpu/drm/nouveau/nouveau_notifier.c
+++ b/drivers/gpu/drm/nouveau/nouveau_notifier.c
@@ -164,10 +164,15 @@ int
 nouveau_ioctl_notifier_alloc(struct drm_device *dev, void *data,
 			     struct drm_file *file_priv)
 {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct drm_nouveau_notifierobj_alloc *na = data;
 	struct nouveau_channel *chan;
 	int ret;
 
+	/* completely unnecessary for these chipsets... */
+	if (unlikely(dev_priv->card_type >= NV_C0))
+		return -EINVAL;
+
 	chan = nouveau_channel_get(dev, file_priv, na->channel);
 	if (IS_ERR(chan))
 		return PTR_ERR(chan);
diff --git a/drivers/gpu/drm/nouveau/nouveau_object.c b/drivers/gpu/drm/nouveau/nouveau_object.c
index 55c9fdcfa67..d77b1fcd19d 100644
--- a/drivers/gpu/drm/nouveau/nouveau_object.c
+++ b/drivers/gpu/drm/nouveau/nouveau_object.c
@@ -643,10 +643,13 @@ nouveau_gpuobj_gr_new(struct nouveau_channel *chan, u32 handle, int class)
 found:
 	switch (oc->engine) {
 	case NVOBJ_ENGINE_SW:
-		ret = nouveau_gpuobj_sw_new(chan, class, &gpuobj);
-		if (ret)
-			return ret;
-		goto insert;
+		if (dev_priv->card_type < NV_C0) {
+			ret = nouveau_gpuobj_sw_new(chan, class, &gpuobj);
+			if (ret)
+				return ret;
+			goto insert;
+		}
+		break;
 	case NVOBJ_ENGINE_GR:
 		if (dev_priv->card_type >= NV_50 && !chan->ramin_grctx) {
 			struct nouveau_pgraph_engine *pgraph =
@@ -669,6 +672,10 @@ found:
 		break;
 	}
 
+	/* we're done if this is fermi */
+	if (dev_priv->card_type >= NV_C0)
+		return 0;
+
 	ret = nouveau_gpuobj_new(dev, chan,
 				 nouveau_gpuobj_class_instmem_size(dev, class),
 				 16,
@@ -772,6 +779,25 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 
 	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_vm_pgd *vpgd;
+
+		ret = nouveau_gpuobj_new(dev, NULL, 4096, 0x1000, 0,
+					 &chan->ramin);
+		if (ret)
+			return ret;
+
+		nouveau_vm_ref(vm, &chan->vm, NULL);
+
+		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;
+	}
+
 	/* Allocate a chunk of memory for per-channel object storage */
 	ret = nouveau_gpuobj_channel_init_pramin(chan);
 	if (ret) {
@@ -779,7 +805,7 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 		return ret;
 	}
 
-	/* NV50/NVC0 VM
+	/* NV50 VM
 	 *  - Allocate per-channel page-directory
 	 *  - Link with shared channel VM
 	 */
@@ -877,9 +903,6 @@ nouveau_gpuobj_channel_takedown(struct nouveau_channel *chan)
 
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
-	if (!chan->ramht)
-		return;
-
 	nouveau_ramht_ref(NULL, &chan->ramht, chan);
 
 	nouveau_vm_ref(NULL, &chan->vm, chan->vm_pd);
diff --git a/drivers/gpu/drm/nouveau/nouveau_state.c b/drivers/gpu/drm/nouveau/nouveau_state.c
index 8eac943e8fd..d5b17b6ccd3 100644
--- a/drivers/gpu/drm/nouveau/nouveau_state.c
+++ b/drivers/gpu/drm/nouveau/nouveau_state.c
@@ -464,11 +464,11 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->instmem.takedown	= nvc0_instmem_takedown;
 		engine->instmem.suspend		= nvc0_instmem_suspend;
 		engine->instmem.resume		= nvc0_instmem_resume;
-		engine->instmem.get		= nvc0_instmem_get;
-		engine->instmem.put		= nvc0_instmem_put;
-		engine->instmem.map		= nvc0_instmem_map;
-		engine->instmem.unmap		= nvc0_instmem_unmap;
-		engine->instmem.flush		= nvc0_instmem_flush;
+		engine->instmem.get		= nv50_instmem_get;
+		engine->instmem.put		= nv50_instmem_put;
+		engine->instmem.map		= nv50_instmem_map;
+		engine->instmem.unmap		= nv50_instmem_unmap;
+		engine->instmem.flush		= nv84_instmem_flush;
 		engine->mc.init			= nv50_mc_init;
 		engine->mc.takedown		= nv50_mc_takedown;
 		engine->timer.init		= nv04_timer_init;
@@ -509,8 +509,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->gpio.irq_enable		= nv50_gpio_irq_enable;
 		engine->crypt.init		= nouveau_stub_init;
 		engine->crypt.takedown		= nouveau_stub_takedown;
-		engine->vram.init		= nouveau_mem_detect;
-		engine->vram.flags_valid	= nouveau_mem_flags_valid;
+		engine->vram.init		= nvc0_vram_init;
+		engine->vram.get		= nvc0_vram_new;
+		engine->vram.put		= nv50_vram_del;
+		engine->vram.flags_valid	= nvc0_vram_flags_valid;
 		break;
 	default:
 		NV_ERROR(dev, "NV%02x unsupported\n", dev_priv->chipset);
@@ -550,6 +552,10 @@ nouveau_card_init_channel(struct drm_device *dev)
 	if (ret)
 		return ret;
 
+	/* no dma objects on fermi... */
+	if (dev_priv->card_type >= NV_C0)
+		goto out_done;
+
 	ret = nouveau_gpuobj_dma_new(dev_priv->channel, NV_CLASS_DMA_IN_MEMORY,
 				     0, dev_priv->vram_size,
 				     NV_MEM_ACCESS_RW, NV_MEM_TARGET_VRAM,
@@ -574,6 +580,7 @@ nouveau_card_init_channel(struct drm_device *dev)
 	if (ret)
 		goto out_err;
 
+out_done:
 	mutex_unlock(&dev_priv->channel->mutex);
 	return 0;
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_vm.c b/drivers/gpu/drm/nouveau/nouveau_vm.c
index 07ab1749cf7..97d82aedf86 100644
--- a/drivers/gpu/drm/nouveau/nouveau_vm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_vm.c
@@ -32,6 +32,7 @@ nouveau_vm_map_at(struct nouveau_vma *vma, u64 delta, struct nouveau_vram *vram)
 {
 	struct nouveau_vm *vm = vma->vm;
 	struct nouveau_mm_node *r;
+	int big = vma->node->type != vm->spg_shift;
 	u32 offset = vma->node->offset + (delta >> 12);
 	u32 bits = vma->node->type - 12;
 	u32 pde  = (offset >> vm->pgt_bits) - vm->fpde;
@@ -44,7 +45,7 @@ nouveau_vm_map_at(struct nouveau_vma *vma, u64 delta, struct nouveau_vram *vram)
 		u32 num  = r->length >> bits;
 
 		while (num) {
-			struct nouveau_gpuobj *pgt = vm->pgt[pde].obj;
+			struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
 
 			end = (pte + num);
 			if (unlikely(end >= max))
@@ -76,6 +77,7 @@ nouveau_vm_map_sg(struct nouveau_vma *vma, u64 delta, u64 length,
 		  dma_addr_t *list)
 {
 	struct nouveau_vm *vm = vma->vm;
+	int big = vma->node->type != vm->spg_shift;
 	u32 offset = vma->node->offset + (delta >> 12);
 	u32 bits = vma->node->type - 12;
 	u32 num  = length >> vma->node->type;
@@ -85,7 +87,7 @@ nouveau_vm_map_sg(struct nouveau_vma *vma, u64 delta, u64 length,
 	u32 end, len;
 
 	while (num) {
-		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj;
+		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
 
 		end = (pte + num);
 		if (unlikely(end >= max))
@@ -110,6 +112,7 @@ void
 nouveau_vm_unmap_at(struct nouveau_vma *vma, u64 delta, u64 length)
 {
 	struct nouveau_vm *vm = vma->vm;
+	int big = vma->node->type != vm->spg_shift;
 	u32 offset = vma->node->offset + (delta >> 12);
 	u32 bits = vma->node->type - 12;
 	u32 num  = length >> vma->node->type;
@@ -119,7 +122,7 @@ nouveau_vm_unmap_at(struct nouveau_vma *vma, u64 delta, u64 length)
 	u32 end, len;
 
 	while (num) {
-		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj;
+		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
 
 		end = (pte + num);
 		if (unlikely(end >= max))
@@ -146,7 +149,7 @@ nouveau_vm_unmap(struct nouveau_vma *vma)
 }
 
 static void
-nouveau_vm_unmap_pgt(struct nouveau_vm *vm, u32 fpde, u32 lpde)
+nouveau_vm_unmap_pgt(struct nouveau_vm *vm, int big, u32 fpde, u32 lpde)
 {
 	struct nouveau_vm_pgd *vpgd;
 	struct nouveau_vm_pgt *vpgt;
@@ -155,16 +158,16 @@ nouveau_vm_unmap_pgt(struct nouveau_vm *vm, u32 fpde, u32 lpde)
 
 	for (pde = fpde; pde <= lpde; pde++) {
 		vpgt = &vm->pgt[pde - vm->fpde];
-		if (--vpgt->refcount)
+		if (--vpgt->refcount[big])
 			con