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: (55 commits)
  drm/nouveau: make cursor_set implementation consistent with other drivers
  drm/nva3/clk: better pll calculation when no fractional fb div available
  drm/nouveau/pm: translate ramcfg strap through ram restrict table
  drm/nva3/pm: allow use of divisor 16
  drm/nvc0/pm: parse clock for pll 0x0a (0x137020) from perf table
  drm/nvc0/pm: correct core/mem/shader perflvl parsing
  drm/nouveau/pm: remove memtiming support check when assigning to perflvl
  drm/nva3: support for memory timing map table
  drm/nouveau: Associate memtimings with performance levels on cards <= nv98
  drm/nva3/pm: initial pass at set_clock() hook
  drm/nvc0/gr: calculate some more of our magic numbers
  drm/nv50: respect LVDS link count from EDID on SPWG panels
  drm/nouveau: recognise DCB connector type 0x41 as LVDS
  drm/nouveau: fix uninitialised variable warning
  drm/nouveau: Fix a crash at card takedown for NV40 and older cards
  drm/nouveau: Free nv04 instmem ramin heap at card takedown
  drm/nva3: somewhat improve clock reporting
  drm/nouveau: pull refclk from vbios on limits 0x40 boards
  drm/nv40/gr: oops, fix random bits getting set in engine obj
  drm/nv50: improve nv50_pm_get_clock()
  ...

44 files changed, 5071 insertions, 1837 deletions

diff --git a/drivers/gpu/drm/nouveau/Makefile b/drivers/gpu/drm/nouveau/Makefile
index e12c97fd8db8..0583677e4581 100644
--- a/drivers/gpu/drm/nouveau/Makefile
+++ b/drivers/gpu/drm/nouveau/Makefile
@@ -20,6 +20,8 @@ nouveau-y := nouveau_drv.o nouveau_state.o nouveau_channel.o nouveau_mem.o \
              nv40_graph.o nv50_graph.o nvc0_graph.o \
              nv40_grctx.o nv50_grctx.o nvc0_grctx.o \
              nv84_crypt.o \
+             nva3_copy.o nvc0_copy.o \
+             nv40_mpeg.o nv50_mpeg.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 \
diff --git a/drivers/gpu/drm/nouveau/nouveau_bios.c b/drivers/gpu/drm/nouveau/nouveau_bios.c
index 90aef64b76f2..729d5fd7c88d 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bios.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bios.c
@@ -5049,11 +5049,7 @@ int get_pll_limits(struct drm_device *dev, uint32_t limit_match, struct pll_lims
 		pll_lim->vco1.max_n = record[11];
 		pll_lim->min_p = record[12];
 		pll_lim->max_p = record[13];
-		/* where did this go to?? */
-		if ((entry[0] & 0xf0) == 0x80)
-			pll_lim->refclk = 27000;
-		else
-			pll_lim->refclk = 100000;
+		pll_lim->refclk = ROM16(entry[9]) * 1000;
 	}
 
 	/*
@@ -6035,6 +6031,7 @@ parse_dcb_connector_table(struct nvbios *bios)
 		case DCB_CONNECTOR_DVI_I:
 		case DCB_CONNECTOR_DVI_D:
 		case DCB_CONNECTOR_LVDS:
+		case DCB_CONNECTOR_LVDS_SPWG:
 		case DCB_CONNECTOR_DP:
 		case DCB_CONNECTOR_eDP:
 		case DCB_CONNECTOR_HDMI_0:
diff --git a/drivers/gpu/drm/nouveau/nouveau_bios.h b/drivers/gpu/drm/nouveau/nouveau_bios.h
index 8a54fa7edf5c..050c314119df 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bios.h
+++ b/drivers/gpu/drm/nouveau/nouveau_bios.h
@@ -82,6 +82,7 @@ enum dcb_connector_type {
 	DCB_CONNECTOR_DVI_I = 0x30,
 	DCB_CONNECTOR_DVI_D = 0x31,
 	DCB_CONNECTOR_LVDS = 0x40,
+	DCB_CONNECTOR_LVDS_SPWG = 0x41,
 	DCB_CONNECTOR_DP = 0x46,
 	DCB_CONNECTOR_eDP = 0x47,
 	DCB_CONNECTOR_HDMI_0 = 0x60,
diff --git a/drivers/gpu/drm/nouveau/nouveau_channel.c b/drivers/gpu/drm/nouveau/nouveau_channel.c
index 4cea35c57d15..a7583a8ddb01 100644
--- a/drivers/gpu/drm/nouveau/nouveau_channel.c
+++ b/drivers/gpu/drm/nouveau/nouveau_channel.c
@@ -268,9 +268,8 @@ nouveau_channel_put_unlocked(struct nouveau_channel **pchan)
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct nouveau_crypt_engine *pcrypt = &dev_priv->engine.crypt;
 	unsigned long flags;
+	int i;
 
 	/* decrement the refcount, and we're done if there's still refs */
 	if (likely(!atomic_dec_and_test(&chan->users))) {
@@ -294,19 +293,12 @@ nouveau_channel_put_unlocked(struct nouveau_channel **pchan)
 	/* boot it off the hardware */
 	pfifo->reassign(dev, false);
 
-	/* We want to give pgraph a chance to idle and get rid of all
-	 * potential errors. We need to do this without the context
-	 * switch lock held, otherwise the irq handler is unable to
-	 * process them.
-	 */
-	if (pgraph->channel(dev) == chan)
-		nouveau_wait_for_idle(dev);
-
 	/* destroy the engine specific contexts */
 	pfifo->destroy_context(chan);
-	pgraph->destroy_context(chan);
-	if (pcrypt->destroy_context)
-		pcrypt->destroy_context(chan);
+	for (i = 0; i < NVOBJ_ENGINE_NR; i++) {
+		if (chan->engctx[i])
+			dev_priv->eng[i]->context_del(chan, i);
+	}
 
 	pfifo->reassign(dev, true);
 
@@ -414,7 +406,7 @@ nouveau_ioctl_fifo_alloc(struct drm_device *dev, void *data,
 	struct nouveau_channel *chan;
 	int ret;
 
-	if (dev_priv->engine.graph.accel_blocked)
+	if (!dev_priv->eng[NVOBJ_ENGINE_GR])
 		return -ENODEV;
 
 	if (init->fb_ctxdma_handle == ~0 || init->tt_ctxdma_handle == ~0)
diff --git a/drivers/gpu/drm/nouveau/nouveau_connector.c b/drivers/gpu/drm/nouveau/nouveau_connector.c
index 7ae151109a66..1595d0b6e815 100644
--- a/drivers/gpu/drm/nouveau/nouveau_connector.c
+++ b/drivers/gpu/drm/nouveau/nouveau_connector.c
@@ -442,7 +442,7 @@ nouveau_connector_set_property(struct drm_connector *connector,
 		}
 
 		/* LVDS always needs gpu scaling */
-		if (nv_connector->dcb->type == DCB_CONNECTOR_LVDS &&
+		if (connector->connector_type == DRM_MODE_CONNECTOR_LVDS &&
 		    value == DRM_MODE_SCALE_NONE)
 			return -EINVAL;
 
@@ -650,6 +650,7 @@ nouveau_connector_get_modes(struct drm_connector *connector)
 		ret = get_slave_funcs(encoder)->get_modes(encoder, connector);
 
 	if (nv_connector->dcb->type == DCB_CONNECTOR_LVDS ||
+	    nv_connector->dcb->type == DCB_CONNECTOR_LVDS_SPWG ||
 	    nv_connector->dcb->type == DCB_CONNECTOR_eDP)
 		ret += nouveau_connector_scaler_modes_add(connector);
 
@@ -810,6 +811,7 @@ nouveau_connector_create(struct drm_device *dev, int index)
 		type = DRM_MODE_CONNECTOR_HDMIA;
 		break;
 	case DCB_CONNECTOR_LVDS:
+	case DCB_CONNECTOR_LVDS_SPWG:
 		type = DRM_MODE_CONNECTOR_LVDS;
 		funcs = &nouveau_connector_funcs_lvds;
 		break;
@@ -838,7 +840,7 @@ nouveau_connector_create(struct drm_device *dev, int index)
 	drm_connector_helper_add(connector, &nouveau_connector_helper_funcs);
 
 	/* Check if we need dithering enabled */
-	if (dcb->type == DCB_CONNECTOR_LVDS) {
+	if (connector->connector_type == DRM_MODE_CONNECTOR_LVDS) {
 		bool dummy, is_24bit = false;
 
 		ret = nouveau_bios_parse_lvds_table(dev, 0, &dummy, &is_24bit);
@@ -883,7 +885,7 @@ nouveau_connector_create(struct drm_device *dev, int index)
 				nv_connector->use_dithering ?
 				DRM_MODE_DITHERING_ON : DRM_MODE_DITHERING_OFF);
 
-		if (dcb->type != DCB_CONNECTOR_LVDS) {
+		if (connector->connector_type != DRM_MODE_CONNECTOR_LVDS) {
 			if (dev_priv->card_type >= NV_50)
 				connector->polled = DRM_CONNECTOR_POLL_HPD;
 			else
diff --git a/drivers/gpu/drm/nouveau/nouveau_display.c b/drivers/gpu/drm/nouveau/nouveau_display.c
index 764c15d537ba..eb514ea29377 100644
--- a/drivers/gpu/drm/nouveau/nouveau_display.c
+++ b/drivers/gpu/drm/nouveau/nouveau_display.c
@@ -276,7 +276,7 @@ nouveau_crtc_page_flip(struct drm_crtc *crtc, struct drm_framebuffer *fb,
 	struct nouveau_fence *fence;
 	int ret;
 
-	if (dev_priv->engine.graph.accel_blocked)
+	if (!dev_priv->channel)
 		return -ENODEV;
 
 	s = kzalloc(sizeof(*s), GFP_KERNEL);
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.c b/drivers/gpu/drm/nouveau/nouveau_drv.c
index 155ebdcbf06f..02c6f37d8bd7 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.c
@@ -162,11 +162,10 @@ nouveau_pci_suspend(struct pci_dev *pdev, pm_message_t pm_state)
 	struct drm_device *dev = pci_get_drvdata(pdev);
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_instmem_engine *pinstmem = &dev_priv->engine.instmem;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
 	struct nouveau_channel *chan;
 	struct drm_crtc *crtc;
-	int ret, i;
+	int ret, i, e;
 
 	if (pm_state.event == PM_EVENT_PRETHAW)
 		return 0;
@@ -206,12 +205,17 @@ nouveau_pci_suspend(struct pci_dev *pdev, pm_message_t pm_state)
 			nouveau_channel_idle(chan);
 	}
 
-	pgraph->fifo_access(dev, false);
-	nouveau_wait_for_idle(dev);
 	pfifo->reassign(dev, false);
 	pfifo->disable(dev);
 	pfifo->unload_context(dev);
-	pgraph->unload_context(dev);
+
+	for (e = NVOBJ_ENGINE_NR - 1; e >= 0; e--) {
+		if (dev_priv->eng[e]) {
+			ret = dev_priv->eng[e]->fini(dev, e);
+			if (ret)
+				goto out_abort;
+		}
+	}
 
 	ret = pinstmem->suspend(dev);
 	if (ret) {
@@ -242,9 +246,12 @@ nouveau_pci_suspend(struct pci_dev *pdev, pm_message_t pm_state)
 
 out_abort:
 	NV_INFO(dev, "Re-enabling acceleration..\n");
+	for (e = e + 1; e < NVOBJ_ENGINE_NR; e++) {
+		if (dev_priv->eng[e])
+			dev_priv->eng[e]->init(dev, e);
+	}
 	pfifo->enable(dev);
 	pfifo->reassign(dev, true);
-	pgraph->fifo_access(dev, true);
 	return ret;
 }
 
@@ -299,8 +306,10 @@ nouveau_pci_resume(struct pci_dev *pdev)
 	engine->mc.init(dev);
 	engine->timer.init(dev);
 	engine->fb.init(dev);
-	engine->graph.init(dev);
-	engine->crypt.init(dev);
+	for (i = 0; i < NVOBJ_ENGINE_NR; i++) {
+		if (dev_priv->eng[i])
+			dev_priv->eng[i]->init(dev, i);
+	}
 	engine->fifo.init(dev);
 
 	nouveau_irq_postinstall(dev);
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index a76514a209b3..9c56331941e2 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -150,13 +150,12 @@ enum nouveau_flags {
 
 #define NVOBJ_ENGINE_SW		0
 #define NVOBJ_ENGINE_GR		1
-#define NVOBJ_ENGINE_PPP	2
-#define NVOBJ_ENGINE_COPY	3
-#define NVOBJ_ENGINE_VP		4
-#define NVOBJ_ENGINE_CRYPT      5
-#define NVOBJ_ENGINE_BSP	6
-#define NVOBJ_ENGINE_DISPLAY	0xcafe0001
-#define NVOBJ_ENGINE_INT	0xdeadbeef
+#define NVOBJ_ENGINE_CRYPT	2
+#define NVOBJ_ENGINE_COPY0	3
+#define NVOBJ_ENGINE_COPY1	4
+#define NVOBJ_ENGINE_MPEG	5
+#define NVOBJ_ENGINE_DISPLAY	15
+#define NVOBJ_ENGINE_NR		16
 
 #define NVOBJ_FLAG_DONT_MAP             (1 << 0)
 #define NVOBJ_FLAG_ZERO_ALLOC		(1 << 1)
@@ -245,11 +244,8 @@ struct nouveau_channel {
 	struct nouveau_gpuobj *cache;
 	void *fifo_priv;
 
-	/* PGRAPH context */
-	/* XXX may be merge 2 pointers as private data ??? */
-	struct nouveau_gpuobj *ramin_grctx;
-	struct nouveau_gpuobj *crypt_ctx;
-	void *pgraph_ctx;
+	/* Execution engine contexts */
+	void *engctx[NVOBJ_ENGINE_NR];
 
 	/* NV50 VM */
 	struct nouveau_vm     *vm;
@@ -298,6 +294,18 @@ struct nouveau_channel {
 	} debugfs;
 };
 
+struct nouveau_exec_engine {
+	void (*destroy)(struct drm_device *, int engine);
+	int  (*init)(struct drm_device *, int engine);
+	int  (*fini)(struct drm_device *, int engine);
+	int  (*context_new)(struct nouveau_channel *, int engine);
+	void (*context_del)(struct nouveau_channel *, int engine);
+	int  (*object_new)(struct nouveau_channel *, int engine,
+			   u32 handle, u16 class);
+	void (*set_tile_region)(struct drm_device *dev, int i);
+	void (*tlb_flush)(struct drm_device *, int engine);
+};
+
 struct nouveau_instmem_engine {
 	void	*priv;
 
@@ -364,30 +372,6 @@ struct nouveau_fifo_engine {
 	void (*tlb_flush)(struct drm_device *dev);
 };
 
-struct nouveau_pgraph_engine {
-	bool accel_blocked;
-	bool registered;
-	int grctx_size;
-	void *priv;
-
-	/* NV2x/NV3x context table (0x400780) */
-	struct nouveau_gpuobj *ctx_table;
-
-	int  (*init)(struct drm_device *);
-	void (*takedown)(struct drm_device *);
-
-	void (*fifo_access)(struct drm_device *, bool);
-
-	struct nouveau_channel *(*channel)(struct drm_device *);
-	int  (*create_context)(struct nouveau_channel *);
-	void (*destroy_context)(struct nouveau_channel *);
-	int  (*load_context)(struct nouveau_channel *);
-	int  (*unload_context)(struct drm_device *);
-	void (*tlb_flush)(struct drm_device *dev);
-
-	void (*set_tile_region)(struct drm_device *dev, int i);
-};
-
 struct nouveau_display_engine {
 	void *priv;
 	int (*early_init)(struct drm_device *);
@@ -426,6 +410,19 @@ struct nouveau_pm_voltage {
 	int nr_level;
 };
 
+struct nouveau_pm_memtiming {
+	int id;
+	u32 reg_100220;
+	u32 reg_100224;
+	u32 reg_100228;
+	u32 reg_10022c;
+	u32 reg_100230;
+	u32 reg_100234;
+	u32 reg_100238;
+	u32 reg_10023c;
+	u32 reg_100240;
+};
+
 #define NOUVEAU_PM_MAX_LEVEL 8
 struct nouveau_pm_level {
 	struct device_attribute dev_attr;
@@ -436,11 +433,13 @@ struct nouveau_pm_level {
 	u32 memory;
 	u32 shader;
 	u32 unk05;
+	u32 unk0a;
 
 	u8 voltage;
 	u8 fanspeed;
 
 	u16 memscript;
+	struct nouveau_pm_memtiming *timing;
 };
 
 struct nouveau_pm_temp_sensor_constants {
@@ -457,17 +456,6 @@ struct nouveau_pm_threshold_temp {
 	s16 fan_boost;
 };
 
-struct nouveau_pm_memtiming {
-	u32 reg_100220;
-	u32 reg_100224;
-	u32 reg_100228;
-	u32 reg_10022c;
-	u32 reg_100230;
-	u32 reg_100234;
-	u32 reg_100238;
-	u32 reg_10023c;
-};
-
 struct nouveau_pm_memtimings {
 	bool supported;
 	struct nouveau_pm_memtiming *timing;
@@ -499,16 +487,6 @@ struct nouveau_pm_engine {
 	int (*temp_get)(struct drm_device *);
 };
 
-struct nouveau_crypt_engine {
-	bool registered;
-
-	int  (*init)(struct drm_device *);
-	void (*takedown)(struct drm_device *);
-	int  (*create_context)(struct nouveau_channel *);
-	void (*destroy_context)(struct nouveau_channel *);
-	void (*tlb_flush)(struct drm_device *dev);
-};
-
 struct nouveau_vram_engine {
 	int  (*init)(struct drm_device *);
 	int  (*get)(struct drm_device *, u64, u32 align, u32 size_nc,
@@ -523,12 +501,10 @@ struct nouveau_engine {
 	struct nouveau_mc_engine      mc;
 	struct nouveau_timer_engine   timer;
 	struct nouveau_fb_engine      fb;
-	struct nouveau_pgraph_engine  graph;
 	struct nouveau_fifo_engine    fifo;
 	struct nouveau_display_engine display;
 	struct nouveau_gpio_engine    gpio;
 	struct nouveau_pm_engine      pm;
-	struct nouveau_crypt_engine   crypt;
 	struct nouveau_vram_engine    vram;
 };
 
@@ -637,6 +613,7 @@ struct drm_nouveau_private {
 	enum nouveau_card_type card_type;
 	/* exact chipset, derived from NV_PMC_BOOT_0 */
 	int chipset;
+	int stepping;
 	int flags;
 
 	void __iomem *mmio;
@@ -647,6 +624,7 @@ struct drm_nouveau_private {
 	u32 ramin_base;
 	bool ramin_available;
 	struct drm_mm ramin_heap;
+	struct nouveau_exec_engine *eng[NVOBJ_ENGINE_NR];
 	struct list_head gpuobj_list;
 	struct list_head classes;
 
@@ -745,10 +723,6 @@ struct drm_nouveau_private {
 	uint32_t crtc_owner;
 	uint32_t dac_users[4];
 
-	struct nouveau_suspend_resume {
-		uint32_t *ramin_copy;
-	} susres;
-
 	struct backlight_device *backlight;
 
 	struct {
@@ -757,8 +731,6 @@ struct drm_nouveau_private {
 
 	struct nouveau_fbdev *nfbdev;
 	struct apertures_struct *apertures;
-
-	bool powered_down;
 };
 
 static inline struct drm_nouveau_private *
@@ -883,17 +855,27 @@ extern void nouveau_channel_ref(struct nouveau_channel *chan,
 extern void nouveau_channel_idle(struct nouveau_channel *chan);
 
 /* nouveau_object.c */
-#define NVOBJ_CLASS(d,c,e) do {                                                \
+#define NVOBJ_ENGINE_ADD(d, e, p) do {                                         \
+	struct drm_nouveau_private *dev_priv = (d)->dev_private;               \
+	dev_priv->eng[NVOBJ_ENGINE_##e] = (p);                                 \
+} while (0)
+
+#define NVOBJ_ENGINE_DEL(d, e) do {                                            \
+	struct drm_nouveau_private *dev_priv = (d)->dev_private;               \
+	dev_priv->eng[NVOBJ_ENGINE_##e] = NULL;                                \
+} while (0)
+
+#define NVOBJ_CLASS(d, c, e) do {                                              \
 	int ret = nouveau_gpuobj_class_new((d), (c), NVOBJ_ENGINE_##e);        \
 	if (ret)                                                               \
 		return ret;                                                    \
-} while(0)
+} while (0)
 
-#define NVOBJ_MTHD(d,c,m,e) do {                                               \
+#define NVOBJ_MTHD(d, c, m, e) do {                                            \
 	int ret = nouveau_gpuobj_mthd_new((d), (c), (m), (e));                 \
 	if (ret)                                                               \
 		return ret;                                                    \
-} while(0)
+} while (0)
 
 extern int  nouveau_gpuobj_early_init(struct drm_device *);
 extern int  nouveau_gpuobj_init(struct drm_device *);
@@ -903,7 +885,7 @@ extern void nouveau_gpuobj_resume(struct drm_device *dev);
 extern int  nouveau_gpuobj_class_new(struct drm_device *, u32 class, u32 eng);
 extern int  nouveau_gpuobj_mthd_new(struct drm_device *, u32 class, u32 mthd,
 				    int (*exec)(struct nouveau_channel *,
-					        u32 class, u32 mthd, u32 data));
+						u32 class, u32 mthd, u32 data));
 extern int  nouveau_gpuobj_mthd_call(struct nouveau_channel *, u32, u32, u32);
 extern int  nouveau_gpuobj_mthd_call2(struct drm_device *, int, u32, u32, u32);
 extern int nouveau_gpuobj_channel_init(struct nouveau_channel *,
@@ -1137,81 +1119,50 @@ extern int  nvc0_fifo_load_context(struct nouveau_channel *);
 extern int  nvc0_fifo_unload_context(struct drm_device *);
 
 /* nv04_graph.c */
-extern int  nv04_graph_init(struct drm_device *);
-extern void nv04_graph_takedown(struct drm_device *);
+extern int  nv04_graph_create(struct drm_device *);
 extern void nv04_graph_fifo_access(struct drm_device *, bool);
-extern struct nouveau_channel *nv04_graph_channel(struct drm_device *);
-extern int  nv04_graph_create_context(struct nouveau_channel *);
-extern void nv04_graph_destroy_context(struct nouveau_channel *);
-extern int  nv04_graph_load_context(struct nouveau_channel *);
-extern int  nv04_graph_unload_context(struct drm_device *);
+extern int  nv04_graph_object_new(struct nouveau_channel *, int, u32, u16);
 extern int  nv04_graph_mthd_page_flip(struct nouveau_channel *chan,
 				      u32 class, u32 mthd, u32 data);
 extern struct nouveau_bitfield nv04_graph_nsource[];
 
 /* nv10_graph.c */
-extern int  nv10_graph_init(struct drm_device *);
-extern void nv10_graph_takedown(struct drm_device *);
+extern int  nv10_graph_create(struct drm_device *);
 extern struct nouveau_channel *nv10_graph_channel(struct drm_device *);
-extern int  nv10_graph_create_context(struct nouveau_channel *);
-extern void nv10_graph_destroy_context(struct nouveau_channel *);
-extern int  nv10_graph_load_context(struct nouveau_channel *);
-extern int  nv10_graph_unload_context(struct drm_device *);
-extern void nv10_graph_set_tile_region(struct drm_device *dev, int i);
 extern struct nouveau_bitfield nv10_graph_intr[];
 extern struct nouveau_bitfield nv10_graph_nstatus[];
 
 /* nv20_graph.c */
-extern int  nv20_graph_create_context(struct nouveau_channel *);
-extern void nv20_graph_destroy_context(struct nouveau_channel *);
-extern int  nv20_graph_load_context(struct nouveau_channel *);
-extern int  nv20_graph_unload_context(struct drm_device *);
-extern int  nv20_graph_init(struct drm_device *);
-extern void nv20_graph_takedown(struct drm_device *);
-extern int  nv30_graph_init(struct drm_device *);
-extern void nv20_graph_set_tile_region(struct drm_device *dev, int i);
+extern int  nv20_graph_create(struct drm_device *);
 
 /* nv40_graph.c */
-extern int  nv40_graph_init(struct drm_device *);
-extern void nv40_graph_takedown(struct drm_device *);
-extern struct nouveau_channel *nv40_graph_channel(struct drm_device *);
-extern int  nv40_graph_create_context(struct nouveau_channel *);
-extern void nv40_graph_destroy_context(struct nouveau_channel *);
-extern int  nv40_graph_load_context(struct nouveau_channel *);
-extern int  nv40_graph_unload_context(struct drm_device *);
+extern int  nv40_graph_create(struct drm_device *);
 extern void nv40_grctx_init(struct nouveau_grctx *);
-extern void nv40_graph_set_tile_region(struct drm_device *dev, int i);
 
 /* nv50_graph.c */
-extern int  nv50_graph_init(struct drm_device *);
-extern void nv50_graph_takedown(struct drm_device *);
-extern void nv50_graph_fifo_access(struct drm_device *, bool);
-extern struct nouveau_channel *nv50_graph_channel(struct drm_device *);
-extern int  nv50_graph_create_context(struct nouveau_channel *);
-extern void nv50_graph_destroy_context(struct nouveau_channel *);
-extern int  nv50_graph_load_context(struct nouveau_channel *);
-extern int  nv50_graph_unload_context(struct drm_device *);
+extern int  nv50_graph_create(struct drm_device *);
 extern int  nv50_grctx_init(struct nouveau_grctx *);
-extern void nv50_graph_tlb_flush(struct drm_device *dev);
-extern void nv84_graph_tlb_flush(struct drm_device *dev);
 extern struct nouveau_enum nv50_data_error_names[];
+extern int  nv50_graph_isr_chid(struct drm_device *dev, u64 inst);
 
 /* nvc0_graph.c */
-extern int  nvc0_graph_init(struct drm_device *);
-extern void nvc0_graph_takedown(struct drm_device *);
-extern void nvc0_graph_fifo_access(struct drm_device *, bool);
-extern struct nouveau_channel *nvc0_graph_channel(struct drm_device *);
-extern int  nvc0_graph_create_context(struct nouveau_channel *);
-extern void nvc0_graph_destroy_context(struct nouveau_channel *);
-extern int  nvc0_graph_load_context(struct nouveau_channel *);
-extern int  nvc0_graph_unload_context(struct drm_device *);
+extern int  nvc0_graph_create(struct drm_device *);
+extern int  nvc0_graph_isr_chid(struct drm_device *dev, u64 inst);
 
 /* nv84_crypt.c */
-extern int  nv84_crypt_init(struct drm_device *dev);
-extern void nv84_crypt_fini(struct drm_device *dev);
-extern int  nv84_crypt_create_context(struct nouveau_channel *);
-extern void nv84_crypt_destroy_context(struct nouveau_channel *);
-extern void nv84_crypt_tlb_flush(struct drm_device *dev);
+extern int  nv84_crypt_create(struct drm_device *);
+
+/* nva3_copy.c */
+extern int  nva3_copy_create(struct drm_device *dev);
+
+/* nvc0_copy.c */
+extern int  nvc0_copy_create(struct drm_device *dev, int engine);
+
+/* nv40_mpeg.c */
+extern int  nv40_mpeg_create(struct drm_device *dev);
+
+/* nv50_mpeg.c */
+extern int  nv50_mpeg_create(struct drm_device *dev);
 
 /* nv04_instmem.c */
 extern int  nv04_instmem_init(struct drm_device *);
@@ -1402,8 +1353,8 @@ bool nv50_gpio_irq_enable(struct drm_device *, enum dcb_gpio_tag, bool on);
 /* nv50_calc. */
 int nv50_calc_pll(struct drm_device *, struct pll_lims *, int clk,
 		  int *N1, int *M1, int *N2, int *M2, int *P);
-int nv50_calc_pll2(struct drm_device *, struct pll_lims *,
-		   int clk, int *N, int *fN, int *M, int *P);
+int nva3_calc_pll(struct drm_device *, struct pll_lims *,
+		  int clk, int *N, int *fN, int *M, int *P);
 
 #ifndef ioread32_native
 #ifdef __BIG_ENDIAN
@@ -1579,6 +1530,13 @@ nv_match_device(struct drm_device *dev, unsigned device,
 		dev->pdev->subsystem_device == sub_device;
 }
 
+static inline void *
+nv_engine(struct drm_device *dev, int engine)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	return (void *)dev_priv->eng[engine];
+}
+
 /* returns 1 if device is one of the nv4x using the 0x4497 object class,
  * helpful to determine a number of other hardware features
  */
diff --git a/drivers/gpu/drm/nouveau/nouveau_grctx.h b/drivers/gpu/drm/nouveau/nouveau_grctx.h
index 4a8ad1307fa4..86c2e374e938 100644
--- a/drivers/gpu/drm/nouveau/nouveau_grctx.h
+++ b/drivers/gpu/drm/nouveau/nouveau_grctx.h
@@ -87,10 +87,10 @@ _cp_bra(struct nouveau_grctx *ctx, u32 mod, int flag, int state, int name)
 	cp_out(ctx, CP_BRA | (mod << 18) | ip | flag |
 		    (state ? 0 : CP_BRA_IF_CLEAR));
 }
-#define cp_bra(c,f,s,n) _cp_bra((c), 0, CP_FLAG_##f, CP_FLAG_##f##_##s, n)
+#define cp_bra(c, f, s, n) _cp_bra((c), 0, CP_FLAG_##f, CP_FLAG_##f##_##s, n)
 #ifdef CP_BRA_MOD
-#define cp_cal(c,f,s,n) _cp_bra((c), 1, CP_FLAG_##f, CP_FLAG_##f##_##s, n)
-#define cp_ret(c,f,s) _cp_bra((c), 2, CP_FLAG_##f, CP_FLAG_##f##_##s, 0)
+#define cp_cal(c, f, s, n) _cp_bra((c), 1, CP_FLAG_##f, CP_FLAG_##f##_##s, n)
+#define cp_ret(c, f, s) _cp_bra((c), 2, CP_FLAG_##f, CP_FLAG_##f##_##s, 0)
 #endif
 
 static inline void
@@ -98,14 +98,14 @@ _cp_wait(struct nouveau_grctx *ctx, int flag, int state)
 {
 	cp_out(ctx, CP_WAIT | flag | (state ? CP_WAIT_SET : 0));
 }
-#define cp_wait(c,f,s) _cp_wait((c), CP_FLAG_##f, CP_FLAG_##f##_##s)
+#define cp_wait(c, f, s) _cp_wait((c), CP_FLAG_##f, CP_FLAG_##f##_##s)
 
 static inline void
 _cp_set(struct nouveau_grctx *ctx, int flag, int state)
 {
 	cp_out(ctx, CP_SET | flag | (state ? CP_SET_1 : 0));
 }
-#define cp_set(c,f,s) _cp_set((c), CP_FLAG_##f, CP_FLAG_##f##_##s)
+#define cp_set(c, f, s) _cp_set((c), CP_FLAG_##f, CP_FLAG_##f##_##s)
 
 static inline void
 cp_pos(struct nouveau_grctx *ctx, int offset)
diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c b/drivers/gpu/drm/nouveau/nouveau_mem.c
index 5045f8b921d6..2960f583dc38 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_mem.c
@@ -51,8 +51,7 @@ nv10_mem_update_tile_region(struct drm_device *dev,
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
 	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	int i = tile - dev_priv->tile.reg;
+	int i = tile - dev_priv->tile.reg, j;
 	unsigned long save;
 
 	nouveau_fence_unref(&tile->fence);
@@ -70,7 +69,10 @@ nv10_mem_update_tile_region(struct drm_device *dev,
 	nouveau_wait_for_idle(dev);
 
 	pfb->set_tile_region(dev, i);
-	pgraph->set_tile_region(dev, i);
+	for (j = 0; j < NVOBJ_ENGINE_NR; j++) {
+		if (dev_priv->eng[j] && dev_priv->eng[j]->set_tile_region)
+			dev_priv->eng[j]->set_tile_region(dev, i);
+	}
 
 	pfifo->cache_pull(dev, true);
 	pfifo->reassign(dev, true);
@@ -152,8 +154,6 @@ nouveau_mem_vram_fini(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 
-	nouveau_bo_ref(NULL, &dev_priv->vga_ram);
-
 	ttm_bo_device_release(&dev_priv->ttm.bdev);
 
 	nouveau_ttm_global_release(dev_priv);
@@ -597,10 +597,10 @@ nouveau_mem_timing_init(struct drm_device *dev)
 	if (!memtimings->timing)
 		return;
 
-	/* Get "some number" from the timing reg for NV_40
+	/* Get "some number" from the timing reg for NV_40 and NV_50
 	 * Used in calculations later */
-	if(dev_priv->card_type == NV_40) {
-		magic_number = (nv_rd32(dev,0x100228) & 0x0f000000) >> 24;
+	if (dev_priv->card_type >= NV_40 && dev_priv->chipset < 0x98) {
+		magic_number = (nv_rd32(dev, 0x100228) & 0x0f000000) >> 24;
 	}
 
 	entry = mem + mem[1];
@@ -643,51 +643,68 @@ nouveau_mem_timing_init(struct drm_device *dev)
 		/* XXX: I don't trust the -1's and +1's... they must come
 		 *      from somewhere! */
 		timing->reg_100224 = (tUNK_0 + tUNK_19 + 1 + magic_number) << 24 |
-				      tUNK_18 << 16 |
+				      max(tUNK_18, (u8) 1) << 16 |
 				      (tUNK_1 + tUNK_19 + 1 + magic_number) << 8;
-		if(dev_priv->chipset == 0xa8) {
+		if (dev_priv->chipset == 0xa8) {
 			timing->reg_100224 |= (tUNK_2 - 1);
 		} else {
 			timing->reg_100224 |= (tUNK_2 + 2 - magic_number);
 		}
 
 		timing->reg_100228 = (tUNK_12 << 16 | tUNK_11 << 8 | tUNK_10);
-		if(dev_priv->chipset >= 0xa3 && dev_priv->chipset < 0xaa) {
+		if (dev_priv->chipset >= 0xa3 && dev_priv->chipset < 0xaa)
 			timing->reg_100228 |= (tUNK_19 - 1) << 24;
-		}
+		else
+			timing->reg_100228 |= magic_number << 24;
 
-		if(dev_priv->card_type == NV_40) {
+		if (dev_priv->card_type == NV_40) {
 			/* NV40: don't know what the rest of the regs are..
 			 * And don't need to know either */
-			timing->reg_100228 |= 0x20200000 | magic_number << 24;
-		} else if(dev_priv->card_type >= NV_50) {
-			/* XXX: reg_10022c */
-			timing->reg_10022c = tUNK_2 - 1;
+			timing->reg_100228 |= 0x20200000;
+		} else if (dev_priv->card_type >= NV_50) {
+			if (dev_priv->chipset < 0x98 ||
+			    (dev_priv->chipset == 0x98 &&
+			     dev_priv->stepping <= 0xa1)) {
+				timing->reg_10022c = (0x14 + tUNK_2) << 24 |
+						     0x16 << 16 |
+						     (tUNK_2 - 1) << 8 |
+						     (tUNK_2 - 1);
+			} else {
+				/* XXX: reg_10022c for recentish cards */
+				timing->reg_10022c = tUNK_2 - 1;
+			}
 
 			timing->reg_100230 = (tUNK_20 << 24 | tUNK_21 << 16 |
 						  tUNK_13 << 8  | tUNK_13);
 
 			timing->reg_100234 = (tRAS << 24 | tRC);
-			timing->reg_100234 += max(tUNK_10,tUNK_11) << 16;
+			timing->reg_100234 += max(tUNK_10, tUNK_11) << 16;
 
-			if(dev_priv->chipset < 0xa3) {
+			if (dev_priv->chipset < 0x98 ||
+			    (dev_priv->chipset == 0x98 &&
+			     dev_priv->stepping <= 0xa1)) {
 				timing->reg_100234 |= (tUNK_2 + 2) << 8;
 			} else {
 				/* XXX: +6? */
 				timing->reg_100234 |= (tUNK_19 + 6) << 8;
 			}
 
-			/* XXX; reg_100238, reg_10023c
-			 * reg_100238: 0x00??????
-			 * reg_10023c: 0x!!??0202 for NV50+ cards (empirical evidence) */
+			/* XXX; reg_100238
+			 * reg_100238: 0x00?????? */
 			timing->reg_10023c = 0x202;
-			if(dev_priv->chipset < 0xa3) {
+			if (dev_priv->chipset < 0x98 ||
+			    (dev_priv->chipset == 0x98 &&
+			     dev_priv->stepping <= 0xa1)) {
 				timing->reg_10023c |= 0x4000000 | (tUNK_2 - 1) << 16;
 			} else {
-				/* currently unknown
+				/* XXX: reg_10023c
+				 * currently unknown
 				 * 10023c seen as 06xxxxxx, 0bxxxxxx or 0fxxxxxx */
 			}
+
+			/* XXX: reg_100240? */
 		}
+		timing->id = i;
 
 		NV_DEBUG(dev, "Entry %d: 220: %08x %08x %08x %08x\n", i,
 			 timing->reg_100220, timing->reg_100224,
@@ -695,10 +712,11 @@ nouveau_mem_timing_init(struct drm_device *dev)
 		NV_DEBUG(dev, "         230: %08x %08x %08x %08x\n",
 			 timing->reg_100230, timing->reg_100234,
 			 timing->reg_100238, timing->reg_10023c);
+		NV_DEBUG(dev, "         240: %08x\n", timing->reg_100240);
 	}
 
 	memtimings->nr_timing = entries;
-	memtimings->supported = true;
+	memtimings->supported = (dev_priv->chipset <= 0x98);
 }
 
 void
diff --git a/drivers/gpu/drm/nouveau/nouveau_object.c b/drivers/gpu/drm/nouveau/nouveau_object.c
index 67a16e01ffa6..8f97016f5b26 100644
--- a/drivers/gpu/drm/nouveau/nouveau_object.c
+++ b/drivers/gpu/drm/nouveau/nouveau_object.c
@@ -361,20 +361,6 @@ nouveau_gpuobj_new_fake(struct drm_device *dev, u32 pinst, u64 vinst,
 	return 0;
 }
 
-
-static uint32_t
-nouveau_gpuobj_class_instmem_size(struct drm_device *dev, int class)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	/*XXX: dodgy hack for now */
-	if (dev_priv->card_type >= NV_50)
-		return 24;
-	if (dev_priv->card_type >= NV_40)
-		return 32;
-	return 16;
-}
-
 /*
    DMA objects are used to reference a piece of memory in the
    framebuffer, PCI or AGP address space. Each object is 16 bytes big
@@ -606,11 +592,11 @@ nouveau_gpuobj_dma_new(struct nouveau_channel *chan, int class, u64 base,
    set to 0?
 */
 static int
-nouveau_gpuobj_sw_new(struct nouveau_channel *chan, int class,
-		      struct nouveau_gpuobj **gpuobj_ret)
+nouveau_gpuobj_sw_new(struct nouveau_channel *chan, u32 handle, u16 class)
 {
 	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
 	struct nouveau_gpuobj *gpuobj;
+	int ret;
 
 	gpuobj = kzalloc(sizeof(*gpuobj), GFP_KERNEL);
 	if (!gpuobj)
@@ -624,8 +610,10 @@ nouveau_gpuobj_sw_new(struct nouveau_channel *chan, int class,
 	spin_lock(&dev_priv->ramin_lock);
 	list_add_tail(&gpuobj->list, &dev_priv->gpuobj_list);
 	spin_unlock(&dev_priv->ramin_lock);
-	*gpuobj_ret = gpuobj;
-	return 0;
+
+	ret = nouveau_ramht_insert(chan, handle, gpuobj);
+	nouveau_gpuobj_ref(NULL, &gpuobj);
+	return ret;
 }
 
 int
@@ -634,101 +622,30 @@ nouveau_gpuobj_gr_new(struct nouveau_channel *chan, u32 handle, int class)
 	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
 	struct drm_device *dev = chan->dev;
 	struct nouveau_gpuobj_class *oc;
-	struct nouveau_gpuobj *gpuobj;
 	int ret;
 
 	NV_DEBUG(dev, "ch%d class=0x%04x\n", chan->id, class);
 
 	list_for_each_entry(oc, &dev_priv->classes, head) {
-		if (oc->id == class)
-			goto found;
-	}
-
-	NV_ERROR(dev, "illegal object class: 0x%x\n", class);
-	return -EINVAL;
+		struct nouveau_exec_engine *eng = dev_priv->eng[oc->engine];
 
-found:
-	switch (oc->engine) {
-	case NVOBJ_ENGINE_SW:
-		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_20 && !chan->ramin_grctx) ||
-		    (dev_priv->card_type  < NV_20 && !chan->pgraph_ctx)) {
-			struct nouveau_pgraph_engine *pgraph =
-				&dev_priv->engine.graph;
+		if (oc->id != class)
+			continue;
 
-			ret = pgraph->create_context(chan);
-			if (ret)
-				return ret;
-		}
-		break;
-	case NVOBJ_ENGINE_CRYPT:
-		if (!chan->crypt_ctx) {
-			struct nouveau_crypt_engine *pcrypt =
-				&dev_priv->engine.crypt;
+		if (oc->engine == NVOBJ_ENGINE_SW)
+			return nouveau_gpuobj_sw_new(chan, handle, class);
 
-			ret = pcrypt->create_context(chan);
+		if (!chan->engctx[oc->engine]) {
+			ret = eng->context_new(chan, oc->engine);
 			if (ret)
 				return ret;
 		}
-		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,
-				 NVOBJ_FLAG_ZERO_ALLOC | NVOBJ_FLAG_ZERO_FREE,
-				 &gpuobj);
-	if (ret) {
-		NV_ERROR(dev, "error creating gpuobj: %d\n", ret);
-		return ret;
-	}
 
-	if (dev_priv->card_type >= NV_50) {
-		nv_wo32(gpuobj,  0, class);
-		nv_wo32(gpuobj, 20, 0x00010000);
-	} else {
-		switch (class) {
-		case NV_CLASS_NULL:
-			nv_wo32(gpuobj, 0, 0x00001030);
-			nv_wo32(gpuobj, 4, 0xFFFFFFFF);
-			break;
-		default:
-			if (dev_priv->card_type >= NV_40) {
-				nv_wo32(gpuobj, 0, class);
-#ifdef __BIG_ENDIAN
-				nv_wo32(gpuobj, 8, 0x01000000);
-#endif
-			} else {
-#ifdef __BIG_ENDIAN
-				nv_wo32(gpuobj, 0, class | 0x00080000);
-#else
-				nv_wo32(gpuobj, 0, class);
-#endif
-			}
-		}
+		return eng->object_new(chan, oc->engine, handle, class);
 	}
-	dev_priv->engine.instmem.flush(dev);
-
-	gpuobj->engine = oc->engine;
-	gpuobj->class  = oc->id;
 
-insert:
-	ret = nouveau_ramht_insert(chan, handle, gpuobj);
-	if (ret)
-		NV_ERROR(dev, "error adding gpuobj to RAMHT: %d\n", ret);
-	nouveau_gpuobj_ref(NULL, &gpuobj);
-	return ret;
+	NV_ERROR(dev, "illegal object class: 0x%x\n", class);
+	return -EINVAL;
 }
 
 static int
@@ -746,9 +663,6 @@ nouveau_gpuobj_channel_init_pramin(struct nouveau_channel *chan)
 	size = 0x2000;
 	base = 0;
 
-	/* PGRAPH context */
-	size += dev_priv->engine.graph.grctx_size;
-
 	if (dev_priv->card_type == NV_50) {
 		/* Various fixed table thingos */
 		size += 0x1400; /* mostly unknown stuff */
diff --git a/drivers/gpu/drm/nouveau/nouveau_perf.c b/drivers/gpu/drm/nouveau/nouveau_perf.c
index 670e3cb697ec..922fb6b664ed 100644
--- a/drivers/gpu/drm/nouveau/nouveau_perf.c
+++ b/drivers/gpu/drm/nouveau/nouveau_perf.c
@@ -72,6 +72,68 @@ legacy_perf_init(struct drm_device *dev)
 	pm->nr_perflvl = 1;
 }
 
+static struct nouveau_pm_memtiming *
+nouveau_perf_timing(struct drm_device *dev, struct bit_entry *P,
+		    u16 memclk, u8 *entry, u8 recordlen, u8 entries)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_pm_engine *pm = &dev_priv->engine.pm;
+	struct nvbios *bios = &dev_priv->vbios;
+	u8 ramcfg;
+	int i;
+
+	/* perf v2 has a separate "timing map" table, we have to match
+	 * the target memory clock to a specific entry, *then* use
+	 * ramcfg to select the correct subentry
+	 */
+	if (P->version == 2) {
+		u8 *tmap = ROMPTR(bios, P->data[4]);
+		if (!tmap) {
+			NV_DEBUG(dev, "no timing map pointer\n");
+			return NULL;
+		}
+
+		if (tmap[0] != 0x10) {
+			NV_WARN(dev, "timing map 0x%02x unknown\n", tmap[0]);
+			return NULL;
+		}
+
+		entry = tmap + tmap[1];
+		recordlen = tmap[2] + (tmap[4] * tmap[3]);
+		for (i = 0; i < tmap[5]; i++, entry += recordlen) {
+			if (memclk >= ROM16(entry[0]) &&
+			    memclk <= ROM16(entry[2]))
+				break;
+		}
+
+		if (i == tmap[5]) {
+			NV_WARN(dev, "no match in timing map table\n");
+			return NULL;
+		}
+
+		entry += tmap[2];
+		recordlen = tmap[3];
+		entries   = tmap[4];
+	}
+
+	ramcfg = (nv_rd32(dev, NV_PEXTDEV_BOOT_0) & 0x0000003c) >> 2;
+	if (bios->ram_restrict_tbl_ptr)
+		ramcfg = bios->data[bios->ram_restrict_tbl_ptr + ramcfg];
+
+	if (ramcfg >= entries) {
+		NV_WARN(dev, "ramcfg strap out of bounds!\n");
+		return NULL;
+	}
+
+	entry += ramcfg * recordlen;
+	if (entry[1] >= pm->memtimings.nr_timing) {
+		NV_WARN(dev, "timingset %d does not exist\n", entry[1]);
+		return NULL;
+	}
+
+	return &pm->memtimings.timing[entry[1]];
+}
+
 void
 nouveau_perf_init(struct drm_device *dev)
 {
@@ -124,6 +186,8 @@ nouveau_perf_init(struct drm_device *dev)
 	for (i = 0; i < entries; i++) {
 		struct nouveau_pm_level *perflvl = &pm->perflvl[pm->nr_perflvl];
 
+		perflvl->timing = NULL;
+
 		if (entry[0] == 0xff) {
 			entry += recordlen;
 			continue;
@@ -174,9 +238,21 @@ nouveau_perf_init(struct drm_device *dev)
 #define subent(n) entry[perf[2] + ((n) * perf[3])]
 			perflvl->fanspeed = 0; /*XXX*/
 			perflvl->voltage = entry[2];
-			perflvl->core = (ROM16(subent(0)) & 0xfff) * 1000;
-			perflvl->shader = (ROM16(subent(1)) & 0xfff) * 1000;
-			perflvl->memory = (ROM16(subent(2)) & 0xfff) * 1000;
+			if (dev_priv->card_type == NV_50) {
+				perflvl->core = ROM16(subent(0)) & 0xfff;
+				perflvl->shader = ROM16(subent(1)) & 0xfff;
+				perflvl->memory = ROM16(subent(2)) & 0xfff;
+			} else {
+				perflvl->shader = ROM16(subent(3)) & 0xfff;
+				perflvl->core   = perflvl->shader / 2;
+				perflvl->unk0a  = ROM16(subent(4)) & 0xfff;
+				perflvl->memory = ROM16(subent(5)) & 0xfff;
+			}
+
+			perflvl->core *= 1000;
+			perflvl->shader *= 1000;
+			perflvl->memory *= 1000;
+			perflvl->unk0a *= 1000;
 			break;
 		}
 
@@ -190,6 +266,16 @@ nouveau_perf_init(struct drm_device *dev)
 			}
 		}
 
+		/* get the corresponding memory timings */
+		if (version > 0x15) {
+			/* last 3 args are for < 0x40, ignored for >= 0x40 */
+			perflvl->timing =
+				nouveau_perf_timing(dev, &P,
+						    perflvl->memory / 1000,
+						    entry + perf[3],
+						    perf[5], perf[4]);
+		}
+
 		snprintf(perflvl->name, sizeof(perflvl->name),
 			 "performance_level_%d", i);
 		perflvl->id = i;
diff --git a/drivers/gpu/drm/nouveau/nouveau_pm.c b/drivers/gpu/drm/nouveau/nouveau_pm.c
index 4399e2f34db4..da8d994d5e8a 100644
--- a/drivers/gpu/drm/nouveau/nouveau_pm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_pm.c
@@ -156,7 +156,7 @@ nouveau_pm_perflvl_get(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 static void
 nouveau_pm_perflvl_info(struct nouveau_pm_level *perflvl, char *ptr, int len)
 {
-	char c[16], s[16], v[16], f[16];
+	char c[16], s[16], v[16], f[16], t[16];
 
 	c[0] = '\0';
 	if (perflvl->core)
@@ -174,8 +174,12 @@ nouveau_pm_perflvl_info(struct nouveau_pm_level *perflvl, char *ptr, int len)
 	if (perflvl->fanspeed)
 		snprintf(f, sizeof(f), " fanspeed %d%%", perflvl->fanspeed);
 
-	snprintf(ptr, len, "memory %dMHz%s%s%s%s\n", perflvl->memory / 1000,
-		 c, s, v, f);
+	t[0] = '\0';
+	if (perflvl->timing)
+		snprintf(t, sizeof(t), " timing %d", perflvl->timing->id);
+
+	snprintf(ptr, len, "memory %dMHz%s%s%s%s%s\n", perflvl->memory / 1000,
+		 c, s, v, f, t);
 }
 
 static ssize_t
@@ -449,7 +453,7 @@ nouveau_hwmon_fini(struct drm_device *dev)
 #endif
 }
 
-#ifdef CONFIG_ACPI
+#if defined(CONFIG_ACPI) && defined(CONFIG_POWER_SUPPLY)
 static int
 nouveau_pm_acpi_event(struct notifier_block *nb, unsigned long val, void *data)
 {
@@ -476,10 +480,10 @@ nouveau_pm_init(struct drm_device *dev)
 	char info[256];
 	int ret, i;
 
+	nouveau_mem_timing_init(dev);
 	nouveau_volt_init(dev);
 	nouveau_perf_init(dev);
 	nouveau_temp_init(dev);
-	nouveau_mem_timing_init(dev);
 
 	NV_INFO(dev, "%d available performance level(s)\n", pm->nr_perflvl);
 	for (i = 0; i < pm->nr_perflvl; i++) {
@@ -490,6 +494,7 @@ nouveau_pm_init(struct drm_device *dev)
 	/* determine current ("boot") performance level */
 	ret = nouveau_pm_perflvl_get(dev, &pm->boot);
 	if (ret == 0) {
+		strncpy(pm->boot.name, "boot", 4);
 		pm->cur = &pm->boot;
 
 		nouveau_pm_perflvl_info(&pm->boot, info, sizeof(info));
@@ -507,7 +512,7 @@ nouveau_pm_init(struct drm_device *dev)
 
 	nouveau_sysfs_init(dev);
 	nouveau_hwmon_init(dev);
-#ifdef CONFIG_ACPI
+#if defined(CONFIG_ACPI) && defined(CONFIG_POWER_SUPPLY)
 	pm->acpi_nb.notifier_call = nouveau_pm_acpi_event;
 	register_acpi_notifier(&pm->acpi_nb);
 #endif
@@ -524,12 +529,12 @@ nouveau_pm_fini(struct drm_device *dev)
 	if (pm->cur != &pm->boot)
 		nouveau_pm_perflvl_set(dev, &pm->boot);
 
-	nouveau_mem_timing_fini(dev);
 	nouveau_temp_fini(dev);
 	nouveau_perf_fini(dev);
 	nouveau_volt_fini(dev);
+	nouveau_mem_timing_fini(dev);
 
-#ifdef CONFIG_ACPI
+#if defined(CONFIG_ACPI) && defined(CONFIG_POWER_SUPPLY)
 	unregister_acpi_notifier(&pm->acpi_nb);
 #endif
 	nouveau_hwmon_fini(dev);
diff --git a/drivers/gpu/drm/nouveau/nouveau_reg.h b/drivers/gpu/drm/nouveau/nouveau_reg.h
index 04e8fb795269..f18cdfc3400f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_reg.h
+++ b/drivers/gpu/drm/nouveau/nouveau_reg.h
@@ -639,9 +639,9 @@
 #    define NV50_PCONNECTOR_I2C_PORT_4                      0x0000e240
 #    define NV50_PCONNECTOR_I2C_PORT_5                      0x0000e258
 
-#define NV50_AUXCH_DATA_OUT(i,n)             ((n) * 4 + (i) * 0x50 + 0x0000e4c0)
+#define NV50_AUXCH_DATA_OUT(i, n)            ((n) * 4 + (i) * 0x50 + 0x0000e4c0)
 #define NV50_AUXCH_DATA_OUT__SIZE                                             4
-#define NV50_AUXCH_DATA_IN(i,n)              ((n) * 4 + (i) * 0x50 + 0x0000e4d0)
+#define NV50_AUXCH_DATA_IN(i, n)             ((n) * 4 + (i) * 0x50 + 0x0000e4d0)
 #define NV50_AUXCH_DATA_IN__SIZE                                              4
 #define NV50_AUXCH_ADDR(i)                             ((i) * 0x50 + 0x0000e4e0)
 #define NV50_AUXCH_CTRL(i)                             ((i) * 0x50 + 0x0000e4e4)
@@ -829,7 +829,7 @@
 #define NV50_PDISPLAY_SOR_BACKLIGHT                                  0x0061c084
 #define NV50_PDISPLAY_SOR_BACKLIGHT_ENABLE                           0x80000000
 #define NV50_PDISPLAY_SOR_BACKLIGHT_LEVEL                            0x00000fff
-#define NV50_SOR_DP_CTRL(i,l)            (0x0061c10c + (i) * 0x800 + (l) * 0x80)
+#define NV50_SOR_DP_CTRL(i, l)           (0x0061c10c + (i) * 0x800 + (l) * 0x80)
 #define NV50_SOR_DP_CTRL_ENABLED                                     0x00000001
 #define NV50_SOR_DP_CTRL_ENHANCED_FRAME_ENABLED                      0x00004000
 #define NV50_SOR_DP_CTRL_LANE_MASK                                   0x001f0000
@@ -841,10 +841,10 @@
 #define NV50_SOR_DP_CTRL_TRAINING_PATTERN_DISABLED                   0x00000000
 #define NV50_SOR_DP_CTRL_TRAINING_PATTERN_1                          0x01000000
 #define NV50_SOR_DP_CTRL_TRAINING_PATTERN_2                          0x02000000
-#define NV50_SOR_DP_UNK118(i,l)          (0x0061c118 + (i) * 0x800 + (l) * 0x80)
-#define NV50_SOR_DP_UNK120(i,l)          (0x0061c120 + (i) * 0x800 + (l) * 0x80)
-#define NV50_SOR_DP_UNK128(i,l)          (0x0061c128 + (i) * 0x800 + (l) * 0x80)
-#define NV50_SOR_DP_UNK130(i,l)          (0x0061c130 + (i) * 0x800 + (l) * 0x80)
+#define NV50_SOR_DP_UNK118(i, l)         (0x0061c118 + (i) * 0x800 + (l) * 0x80)
+#define NV50_SOR_DP_UNK120(i, l)         (0x0061c120 + (i) * 0x800 + (l) * 0x80)
+#define NV50_SOR_DP_UNK128(i, l)         (0x0061c128 + (i) * 0x800 + (l) * 0x80)
+#define NV50_SOR_DP_UNK130(i, l)         (0x0061c130 + (i) * 0x800 + (l) * 0x80)
 
 #define NV50_PDISPLAY_USER(i)                        ((i) * 0x1000 + 0x00640000)
 #define NV50_PDISPLAY_USER_PUT(i)                    ((i) * 0x1000 + 0x00640000)
diff --git a/drivers/gpu/drm/nouveau/nouveau_state.c b/drivers/gpu/drm/nouveau/nouveau_state.c
index a30adec5beaa..38ea662568c1 100644
--- a/drivers/gpu/drm/nouveau/nouveau_state.c
+++ b/drivers/gpu/drm/nouveau/nouveau_state.c
@@ -65,14 +65,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->timer.takedown		= nv04_timer_takedown;
 		engine->fb.init			= nv04_fb_init;
 		engine->fb.takedown		= nv04_fb_takedown;
-		engine->graph.init		= nv04_graph_init;
-		engine->graph.takedown		= nv04_graph_takedown;
-		engine->graph.fifo_access	= nv04_graph_fifo_access;
-		engine->graph.channel		= nv04_graph_channel;
-		engine->graph.create_context	= nv04_graph_create_context;
-		engine->graph.destroy_context	= nv04_graph_destroy_context;
-		engine->graph.load_context	= nv04_graph_load_context;
-		engine->graph.unload_context	= nv04_graph_unload_context;
 		engine->fifo.channels		= 16;
 		engine->fifo.init		= nv04_fifo_init;
 		engine->fifo.takedown		= nv04_fifo_fini;
@@ -98,8 +90,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clock_get		= nv04_pm_clock_get;
 		engine->pm.clock_pre		= nv04_pm_clock_pre;
 		engine->pm.clock_set		= nv04_pm_clock_set;
-		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;
 		break;
@@ -123,15 +113,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->fb.init_tile_region	= nv10_fb_init_tile_region;
 		engine->fb.set_tile_region	= nv10_fb_set_tile_region;
 		engine->fb.free_tile_region	= nv10_fb_free_tile_region;
-		engine->graph.init		= nv10_graph_init;
-		engine->graph.takedown		= nv10_graph_takedown;
-		engine->graph.channel		= nv10_graph_channel;
-		engine->graph.create_context	= nv10_graph_create_context;
-		engine->graph.destroy_context	= nv10_graph_destroy_context;
-		engine->graph.fifo_access	= nv04_graph_fifo_access;
-		engine->graph.load_context	= nv10_graph_load_context;
-		engine->graph.unload_context	= nv10_graph_unload_context;
-		engine->graph.set_tile_region	= nv10_graph_set_tile_region;
 		engine->fifo.channels		= 32;
 		engine->fifo.init		= nv10_fifo_init;
 		engine->fifo.takedown		= nv04_fifo_fini;
@@ -157,8 +138,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clock_get		= nv04_pm_clock_get;
 		engine->pm.clock_pre		= nv04_pm_clock_pre;
 		engine->pm.clock_set		= nv04_pm_clock_set;
-		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;
 		break;
@@ -182,15 +161,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->fb.init_tile_region	= nv10_fb_init_tile_region;
 		engine->fb.set_tile_region	= nv10_fb_set_tile_region;
 		engine->fb.free_tile_region	= nv10_fb_free_tile_region;
-		engine->graph.init		= nv20_graph_init;
-		engine->graph.takedown		= nv20_graph_takedown;
-		engine->graph.channel		= nv10_graph_channel;
-		engine->graph.create_context	= nv20_graph_create_context;
-		engine->graph.destroy_context	= nv20_graph_destroy_context;
-		engine->graph.fifo_access	= nv04_graph_fifo_access;
-		engine->graph.load_context	= nv20_graph_load_context;
-		engine->graph.unload_context	= nv20_graph_unload_context;
-		engine->graph.set_tile_region	= nv20_graph_set_tile_region;
 		engine->fifo.channels		= 32;
 		engine->fifo.init		= nv10_fifo_init;
 		engine->fifo.takedown		= nv04_fifo_fini;
@@ -216,8 +186,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clock_get		= nv04_pm_clock_get;
 		engine->pm.clock_pre		= nv04_pm_clock_pre;
 		engine->pm.clock_set		= nv04_pm_clock_set;
-		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;
 		break;
@@ -241,15 +209,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->fb.init_tile_region	= nv30_fb_init_tile_region;
 		engine->fb.set_tile_region	= nv10_fb_set_tile_region;
 		engine->fb.free_tile_region	= nv30_fb_free_tile_region;
-		engine->graph.init		= nv30_graph_init;
-		engine->graph.takedown		= nv20_graph_takedown;
-		engine->graph.fifo_access	= nv04_graph_fifo_access;
-		engine->graph.channel		= nv10_graph_channel;
-		engine->graph.create_context	= nv20_graph_create_context;
-		engine->graph.destroy_context	= nv20_graph_destroy_context;
-		engine->graph.load_context	= nv20_graph_load_context;
-		engine->graph.unload_context	= nv20_graph_unload_context;
-		engine->graph.set_tile_region	= nv20_graph_set_tile_region;
 		engine->fifo.channels		= 32;
 		engine->fifo.init		= nv10_fifo_init;
 		engine->fifo.takedown		= nv04_fifo_fini;
@@ -277,8 +236,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clock_set		= nv04_pm_clock_set;
 		engine->pm.voltage_get		= nouveau_voltage_gpio_get;
 		engine->pm.voltage_set		= nouveau_voltage_gpio_set;
-		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;
 		break;
@@ -303,15 +260,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->fb.init_tile_region	= nv30_fb_init_tile_region;
 		engine->fb.set_tile_region	= nv40_fb_set_tile_region;
 		engine->fb.free_tile_region	= nv30_fb_free_tile_region;
-		engine->graph.init		= nv40_graph_init;
-		engine->graph.takedown		= nv40_graph_takedown;
-		engine->graph.fifo_access	= nv04_graph_fifo_access;
-		engine->graph.channel		= nv40_graph_channel;
-		engine->graph.create_context	= nv40_graph_create_context;
-		engine->graph.destroy_context	= nv40_graph_destroy_context;
-		engine->graph.load_context	= nv40_graph_load_context;
-		engine->graph.unload_context	= nv40_graph_unload_context;
-		engine->graph.set_tile_region	= nv40_graph_set_tile_region;
 		engine->fifo.channels		= 32;
 		engine->fifo.init		= nv40_fifo_init;
 		engine->fifo.takedown		= nv04_fifo_fini;
@@ -340,8 +288,6 @@ 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->pm.temp_get		= nv40_temp_get;
-		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;
 		break;
@@ -368,19 +314,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->timer.takedown		= nv04_timer_takedown;
 		engine->fb.init			= nv50_fb_init;
 		engine->fb.takedown		= nv50_fb_takedown;
-		engine->graph.init		= nv50_graph_init;
-		engine->graph.takedown		= nv50_graph_takedown;
-		engine->graph.fifo_access	= nv50_graph_fifo_access;
-		engine->graph.channel		= nv50_graph_channel;
-		engine->graph.create_context	= nv50_graph_create_context;
-		engine->graph.destroy_context	= nv50_graph_destroy_context;
-		engine->graph.load_context	= nv50_graph_load_context;
-		engine->graph.unload_context	= nv50_graph_unload_context;
-		if (dev_priv->chipset == 0x50 ||
-		    dev_priv->chipset == 0xac)
-			engine->graph.tlb_flush	= nv50_graph_tlb_flush;
-		else
-			engine->graph.tlb_flush	= nv84_graph_tlb_flush;
 		engine->fifo.channels		= 128;
 		engine->fifo.init		= nv50_fifo_init;
 		engine->fifo.takedown		= nv50_fifo_takedown;
@@ -432,24 +365,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 			engine->pm.temp_get	= nv84_temp_get;
 		else
 			engine->pm.temp_get	= nv40_temp_get;
-		switch (dev_priv->chipset) {
-		case 0x84:
-		case 0x86:
-		case 0x92:
-		case 0x94:
-		case 0x96:
-		case 0xa0:
-			engine->crypt.init	= nv84_crypt_init;
-			engine->crypt.takedown	= nv84_crypt_fini;
-			engine->crypt.create_context = nv84_crypt_create_context;
-			engine->crypt.destroy_context = nv84_crypt_destroy_context;
-			engine->crypt.tlb_flush	= nv84_crypt_tlb_flush;
-			break;
-		default:
-			engine->crypt.init	= nouveau_stub_init;
-			engine->crypt.takedown	= nouveau_stub_takedown;
-			break;
-		}
 		engine->vram.init		= nv50_vram_init;
 		engine->vram.get		= nv50_vram_new;
 		engine->vram.put		= nv50_vram_del;
@@ -472,14 +387,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->timer.takedown		= nv04_timer_takedown;
 		engine->fb.init			= nvc0_fb_init;
 		engine->fb.takedown		= nvc0_fb_takedown;
-		engine->graph.init		= nvc0_graph_init;
-		engine->graph.takedown		= nvc0_graph_takedown;
-		engine->graph.fifo_access	= nvc0_graph_fifo_access;
-		engine->graph.channel		= nvc0_graph_channel;
-		engine->graph.create_context	= nvc0_graph_create_context;
-		engine->graph.destroy_context	= nvc0_graph_destroy_context;
-		engine->graph.load_context	= nvc0_graph_load_context;
-		engine->graph.unload_context	= nvc0_graph_unload_context;
 		engine->fifo.channels		= 128;
 		engine->fifo.init		= nvc0_fifo_init;
 		engine->fifo.takedown		= nvc0_fifo_takedown;
@@ -503,8 +410,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->gpio.irq_register	= nv50_gpio_irq_register;
 		engine->gpio.irq_unregister	= nv50_gpio_irq_unregister;
 		engine->gpio.irq_enable		= nv50_gpio_irq_enable;
-		engine->crypt.init		= nouveau_stub_init;
-		engine->crypt.takedown		= nouveau_stub_takedown;
 		engine->vram.init		= nvc0_vram_init;
 		engine->vram.get		= nvc0_vram_new;
 		engine->vram.put		= nv50_vram_del;
@@ -593,7 +498,7 @@ nouveau_card_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_engine *engine;
-	int ret;
+	int ret, e = 0;
 
 	vga_client_register(dev->pdev, dev, NULL, nouveau_vga_set_decode);
 	vga_switcheroo_register_client(dev->pdev, nouveau_switcheroo_set_state,
@@ -658,23 +563,80 @@ nouveau_card_init(struct drm_device *dev)
 	if (ret)
 		goto out_timer;
 
-	if (nouveau_noaccel)
-		engine->graph.accel_blocked = true;
-	else {
-		/* PGRAPH */
-		ret = engine->graph.init(dev);
-		if (ret)
-			goto out_fb;
+	switch (dev_priv->card_type) {
+	case NV_04:
+		nv04_graph_create(dev);
+		break;
+	case NV_10:
+		nv10_graph_create(dev);
+		break;
+	case NV_20:
+	case NV_30:
+		nv20_graph_create(dev);
+		break;
+	case NV_40:
+		nv40_graph_create(dev);
+		break;
+	case NV_50:
+		nv50_graph_create(dev);
+		break;
+	case NV_C0:
+		nvc0_graph_create(dev);
+		break;
+	default:
+		break;
+	}
 
-		/* PCRYPT */
-		ret = engine->crypt.init(dev);
-		if (ret)
-			goto out_graph;
+	switch (dev_priv->chipset) {
+	case 0x84:
+	case 0x86:
+	case 0x92:
+	case 0x94:
+	case 0x96:
+	case 0xa0:
+		nv84_crypt_create(dev);
+		break;
+	}
+
+	switch (dev_priv->card_type) {
+	case NV_50:
+		switch (dev_priv->chipset) {
+		case 0xa3:
+		case 0xa5:
+		case 0xa8:
+		case 0xaf:
+			nva3_copy_create(dev);
+			break;
+		}
+		break;
+	case NV_C0:
+		nvc0_copy_create(dev, 0);
+		nvc0_copy_create(dev, 1);
+		break;
+	default:
+		break;
+	}
+
+	if (dev_priv->card_type == NV_40)
+		nv40_mpeg_create(dev);
+	else
+	if (dev_priv->card_type == NV_50 &&
+	    (dev_priv->chipset < 0x98 || dev_priv->chipset == 0xa0))
+		nv50_mpeg_create(dev);
+
+	if (!nouveau_noaccel) {
+		for (e = 0; e < NVOBJ_ENGINE_NR; e++) {
+			if (dev_priv->eng[e]) {
+				ret = dev_priv->eng[e]->init(dev, e);
+				if (ret)
+					goto out_engine;
+			}
+		}
 
 		/* PFIFO */
 		ret = engine->fifo.init(dev);
 		if (ret)
-			goto out_crypt;
+			goto out_engine;
 	}
 
 	ret = engine->display.create(dev);
@@ -691,7 +653,7 @@ nouveau_card_init(struct drm_device *dev)
 
 	/* what about PVIDEO/PCRTC/PRAMDAC etc? */
 
-	if (!engine->graph.accel_blocked) {
+	if (dev_priv->eng[NVOBJ_ENGINE_GR]) {
 		ret = nouveau_fence_init(dev);
 		if (ret)
 			goto out_irq;
@@ -715,13 +677,16 @@ out_vblank:
 out_fifo:
 	if (!nouveau_noaccel)
 		engine->fifo.takedown(dev);
-out_crypt:
-	if (!nouveau_noaccel)
-		engine->crypt.takedown(dev);
-out_graph:
-	if (!nouveau_noaccel)
-		engine->graph.takedown(dev);
-out_fb:
+out_engine:
+	if (!nouveau_noaccel) {
+		for (e = e - 1; e >= 0; e--) {
+			if (!dev_priv->eng[e])
+				continue;
+			dev_priv->eng[e]->fini(dev, e);
+			dev_priv->eng[e]->destroy(dev,e );
+		}
+	}
+
 	engine->fb.takedown(dev);
 out_timer:
 	engine->timer.takedown(dev);
@@ -751,16 +716,21 @@ static void nouveau_card_takedown(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_engine *engine = &dev_priv->engine;
+	int e;
 
-	if (!engine->graph.accel_blocked) {
+	if (dev_priv->channel) {
 		nouveau_fence_fini(dev);
 		nouveau_channel_put_unlocked(&dev_priv->channel);
 	}
 
 	if (!nouveau_noaccel) {
 		engine->fifo.takedown(dev);
-		engine->crypt.takedown(dev);
-		engine->graph.takedown(dev);
+		for (e = NVOBJ_ENGINE_NR - 1; e >= 0; e--) {
+			if (dev_priv->eng[e]) {
+				dev_priv->eng[e]->fini(dev, e);
+				dev_priv->eng[e]->destroy(dev,e );
+			}
+		}
 	}
 	engine->fb.takedown(dev);
 	engine->timer.takedown(dev);
@@ -768,6 +738,11 @@ static void nouveau_card_takedown(struct drm_device *dev)
 	engine->mc.takedown(dev);
 	engine->display.late_takedown(dev);
 
+	if (dev_priv->vga_ram) {
+		nouveau_bo_unpin(dev_priv->vga_ram);
+		nouveau_bo_ref(NULL, &dev_priv->vga_ram);
+	}
+
 	mutex_lock(&dev->struct_mutex);
 	ttm_bo_clean_mm(&dev_priv->ttm.bdev, TTM_PL_VRAM);
 	ttm_bo_clean_mm(&dev_priv->ttm.bdev, TTM_PL_TT);
@@ -861,7 +836,7 @@ static int nouveau_remove_conflicting_drivers(struct drm_device *dev)
 #ifdef CONFIG_X86
 	primary = dev->pdev->resource[PCI_ROM_RESOURCE].flags & IORESOURCE_ROM_SHADOW;
 #endif
-	
+
 	remove_conflicting_framebuffers(dev_priv->apertures, "nouveaufb", primary);
 	return 0;
 }
@@ -913,11 +888,13 @@ int nouveau_load(struct drm_device *dev, unsigned long flags)
 
 	/* Time to determine the card architecture */
 	reg0 = nv_rd32(dev, NV03_PMC_BOOT_0);
+	dev_priv->stepping = 0; /* XXX: add stepping for pre-NV10? */
 
 	/* We're dealing with >=NV10 */
 	if ((reg0 & 0x0f000000) > 0) {
 		/* Bit 27-20 contain the architecture in hex */
 		dev_priv->chipset = (reg0 & 0xff00000) >> 20;
+		dev_priv->stepping = (reg0 & 0xff);
 	/* NV04 or NV05 */
 	} else if ((reg0 & 0xff00fff0) == 0x20004000) {
 		if (reg0 & 0x00f00000)
diff --git a/drivers/gpu/drm/nouveau/nouveau_vm.h b/drivers/gpu/drm/nouveau/nouveau_vm.h
index 2e06b55cfdc1..c48a9fc2b47b 100644
--- a/drivers/gpu/drm/nouveau/nouveau_vm.h
+++ b/drivers/gpu/drm/nouveau/nouveau_vm.h
@@ -53,8 +53,7 @@ struct nouveau_vm {
 	int refcount;
 
 	struct list_head pgd_list;
-	atomic_t pgraph_refs;
-	atomic_t pcrypt_refs;
+	atomic_t engref[16];
 
 	struct nouveau_vm_pgt *pgt;
 	u32 fpde;
diff --git a/drivers/gpu/drm/nouveau/nouveau_volt.c b/drivers/gpu/drm/nouveau/nouveau_volt.c
index 04fdc00a67d5..75e872741d92 100644
--- a/drivers/gpu/drm/nouveau/nouveau_volt.c
+++ b/drivers/gpu/drm/nouveau/nouveau_volt.c
@@ -159,8 +159,16 @@ nouveau_volt_init(struct drm_device *dev)
 		headerlen = volt[1];
 		recordlen = volt[2];
 		entries   = volt[3];
-		vidshift  = hweight8(volt[5]);
 		vidmask   = volt[4];
+		/* no longer certain what volt[5] is, if it's related to
+		 * the vid shift then it's definitely not a function of
+		 * how many bits are set.
+		 *
+		 * after looking at a number of nva3+ vbios images, they
+		 * all seem likely to have a static shift of 2.. lets
+		 * go with that for now until proven otherwise.
+		 */
+		vidshift  = 2;
 		break;
 	default:
 		NV_WARN(dev, "voltage table 0x%02x unknown\n", volt[0]);
diff --git a/drivers/gpu/drm/nouveau/nv04_crtc.c b/drivers/gpu/drm/nouveau/nv04_crtc.c
index 748b9d9c2949..3c78bc81357e 100644
--- a/drivers/gpu/drm/nouveau/nv04_crtc.c
+++ b/drivers/gpu/drm/nouveau/nv04_crtc.c
@@ -790,8 +790,7 @@ nv04_crtc_do_mode_set_base(struct drm_crtc *crtc,
 	if (atomic) {
 		drm_fb = passed_fb;
 		fb = nouveau_framebuffer(passed_fb);
-	}
-	else {
+	} else {
 		/* If not atomic, we can go ahead and pin, and unpin the
 		 * old fb we were passed.
 		 */
@@ -944,14 +943,14 @@ nv04_crtc_cursor_set(struct drm_crtc *crtc, struct drm_file *file_priv,
 	struct drm_gem_object *gem;
 	int ret = 0;
 
-	if (width != 64 || height != 64)
-		return -EINVAL;
-
 	if (!buffer_handle) {
 		nv_crtc->cursor.hide(nv_crtc, true);
 		return 0;
 	}
 
+	if (width != 64 || height != 64)
+		return -EINVAL;
+
 	gem = drm_gem_object_lookup(dev, file_priv, buffer_handle);
 	if (!gem)
 		return -ENOENT;
diff --git a/drivers/gpu/drm/nouveau/nv04_graph.c b/drivers/gpu/drm/nouveau/nv04_graph.c
index af75015068d6..3626ee7db3ba 100644
--- a/drivers/gpu/drm/nouveau/nv04_graph.c
+++ b/drivers/gpu/drm/nouveau/nv04_graph.c
@@ -28,9 +28,11 @@
 #include "nouveau_drv.h"
 #include "nouveau_hw.h"
 #include "nouveau_util.h"
+#include "nouveau_ramht.h"
 
-static int  nv04_graph_register(struct drm_device *dev);
-static void nv04_graph_isr(struct drm_device *dev);
+struct nv04_graph_engine {
+	struct nouveau_exec_engine base;
+};
 
 static uint32_t nv04_graph_ctx_regs[] = {
 	0x0040053c,
@@ -350,7 +352,7 @@ struct graph_state {
 	uint32_t nv04[ARRAY_SIZE(nv04_graph_ctx_regs)];
 };
 
-struct nouveau_channel *
+static struct nouveau_channel *
 nv04_graph_channel(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
@@ -365,26 +367,6 @@ nv04_graph_channel(struct drm_device *dev)
 	return dev_priv->channels.ptr[chid];
 }
 
-static void
-nv04_graph_context_switch(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct nouveau_channel *chan = NULL;
-	int chid;
-
-	nouveau_wait_for_idle(dev);
-
-	/* If previous context is valid, we need to save it */
-	pgraph->unload_context(dev);
-
-	/* Load context for next channel */
-	chid = dev_priv->engine.fifo.channel_id(dev);
-	chan = dev_priv->channels.ptr[chid];
-	if (chan)
-		nv04_graph_load_context(chan);
-}
-
 static uint32_t *ctx_reg(struct graph_state *ctx, uint32_t reg)
 {
 	int i;
@@ -397,48 +379,11 @@ static uint32_t *ctx_reg(struct graph_state *ctx, uint32_t reg)
 	return NULL;
 }
 
-int nv04_graph_create_context(struct nouveau_channel *chan)
-{
-	struct graph_state *pgraph_ctx;
-	NV_DEBUG(chan->dev, "nv04_graph_context_create %d\n", chan->id);
-
-	chan->pgraph_ctx = pgraph_ctx = kzalloc(sizeof(*pgraph_ctx),
-						GFP_KERNEL);
-	if (pgraph_ctx == NULL)
-		return -ENOMEM;
-
-	*ctx_reg(pgraph_ctx, NV04_PGRAPH_DEBUG_3) = 0xfad4ff31;
-
-	return 0;
-}
-
-void nv04_graph_destroy_context(struct nouveau_channel *chan)
-{
-	struct drm_device *dev = chan->dev;
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct graph_state *pgraph_ctx = chan->pgraph_ctx;
-	unsigned long flags;
-
-	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
-	pgraph->fifo_access(dev, false);
-
-	/* Unload the context if it's the currently active one */
-	if (pgraph->channel(dev) == chan)
-		pgraph->unload_context(dev);
-
-	/* Free the context resources */
-	kfree(pgraph_ctx);
-	chan->pgraph_ctx = NULL;
-
-	pgraph->fifo_access(dev, true);
-	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
-}
-
-int nv04_graph_load_context(struct nouveau_channel *chan)
+static int
+nv04_graph_load_context(struct nouveau_channel *chan)
 {
+	struct graph_state *pgraph_ctx = chan->engctx[NVOBJ_ENGINE_GR];
 	struct drm_device *dev = chan->dev;
-	struct graph_state *pgraph_ctx = chan->pgraph_ctx;
 	uint32_t tmp;
 	int i;
 
@@ -456,20 +401,19 @@ int nv04_graph_load_context(struct nouveau_channel *chan)
 	return 0;
 }
 
-int
+static int
 nv04_graph_unload_context(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	struct nouveau_channel *chan = NULL;
 	struct graph_state *ctx;
 	uint32_t tmp;
 	int i;
 
-	chan = pgraph->channel(dev);
+	chan = nv04_graph_channel(dev);
 	if (!chan)
 		return 0;
-	ctx = chan->pgraph_ctx;
+	ctx = chan->engctx[NVOBJ_ENGINE_GR];
 
 	for (i = 0; i < ARRAY_SIZE(nv04_graph_ctx_regs); i++)
 		ctx->nv04[i] = nv_rd32(dev, nv04_graph_ctx_regs[i]);
@@ -481,23 +425,85 @@ nv04_graph_unload_context(struct drm_device *dev)
 	return 0;
 }
 
-int nv04_graph_init(struct drm_device *dev)
+static int
+nv04_graph_context_new(struct nouveau_channel *chan, int engine)
 {
+	struct graph_state *pgraph_ctx;
+	NV_DEBUG(chan->dev, "nv04_graph_context_create %d\n", chan->id);
+
+	pgraph_ctx = kzalloc(sizeof(*pgraph_ctx), GFP_KERNEL);
+	if (pgraph_ctx == NULL)
+		return -ENOMEM;
+
+	*ctx_reg(pgraph_ctx, NV04_PGRAPH_DEBUG_3) = 0xfad4ff31;
+
+	chan->engctx[engine] = pgraph_ctx;
+	return 0;
+}
+
+static void
+nv04_graph_context_del(struct nouveau_channel *chan, int engine)
+{
+	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t tmp;
+	struct graph_state *pgraph_ctx = chan->engctx[engine];
+	unsigned long flags;
+
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	nv04_graph_fifo_access(dev, false);
+
+	/* Unload the context if it's the currently active one */
+	if (nv04_graph_channel(dev) == chan)
+		nv04_graph_unload_context(dev);
+
+	nv04_graph_fifo_access(dev, true);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
+
+	/* Free the context resources */
+	kfree(pgraph_ctx);
+	chan->engctx[engine] = NULL;
+}
+
+int
+nv04_graph_object_new(struct nouveau_channel *chan, int engine,
+		      u32 handle, u16 class)
+{
+	struct drm_device *dev = chan->dev;
+	struct nouveau_gpuobj *obj = NULL;
 	int ret;
 
+	ret = nouveau_gpuobj_new(dev, chan, 16, 16, NVOBJ_FLAG_ZERO_FREE, &obj);
+	if (ret)
+		return ret;
+	obj->engine = 1;
+	obj->class  = class;
+
+#ifdef __BIG_ENDIAN
+	nv_wo32(obj, 0x00, 0x00080000 | class);
+#else
+	nv_wo32(obj, 0x00, class);
+#endif
+	nv_wo32(obj, 0x04, 0x00000000);
+	nv_wo32(obj, 0x08, 0x00000000);
+	nv_wo32(obj, 0x0c, 0x00000000);
+
+	ret = nouveau_ramht_insert(chan, handle, obj);
+	nouveau_gpuobj_ref(NULL, &obj);
+	return ret;
+}
+
+static int
+nv04_graph_init(struct drm_device *dev, int engine)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	uint32_t tmp;
+
 	nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) &
 			~NV_PMC_ENABLE_PGRAPH);
 	nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) |
 			 NV_PMC_ENABLE_PGRAPH);
 
-	ret = nv04_graph_register(dev);
-	if (ret)
-		return ret;
-
 	/* Enable PGRAPH interrupts */
-	nouveau_irq_register(dev, 12, nv04_graph_isr);
 	nv_wr32(dev, NV03_PGRAPH_INTR, 0xFFFFFFFF);
 	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
 
@@ -507,7 +513,7 @@ int nv04_graph_init(struct drm_device *dev)
 	nv_wr32(dev, NV04_PGRAPH_DEBUG_0, 0x001FFFFF);*/
 	nv_wr32(dev, NV04_PGRAPH_DEBUG_0, 0x1231c000);
 	/*1231C000 blob, 001 haiku*/
-	//*V_WRITE(NV04_PGRAPH_DEBUG_1, 0xf2d91100);*/
+	/*V_WRITE(NV04_PGRAPH_DEBUG_1, 0xf2d91100);*/
 	nv_wr32(dev, NV04_PGRAPH_DEBUG_1, 0x72111100);
 	/*0x72111100 blob , 01 haiku*/
 	/*nv_wr32(dev, NV04_PGRAPH_DEBUG_2, 0x11d5f870);*/
@@ -531,10 +537,12 @@ int nv04_graph_init(struct drm_device *dev)
 	return 0;
 }
 
-void nv04_graph_takedown(struct drm_device *dev)
+static int
+nv04_graph_fini(struct drm_device *dev, int engine)
 {
+	nv04_graph_unload_context(dev);
 	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0x00000000);
-	nouveau_irq_unregister(dev, 12);
+	return 0;
 }
 
 void
@@ -969,13 +977,138 @@ nv04_graph_mthd_bind_chroma(struct nouveau_channel *chan,
 	return 1;
 }
 
-static int
-nv04_graph_register(struct drm_device *dev)
+static struct nouveau_bitfield nv04_graph_intr[] = {
+	{ NV_PGRAPH_INTR_NOTIFY, "NOTIFY" },
+	{}
+};
+
+static struct nouveau_bitfield nv04_graph_nstatus[] = {
+	{ NV04_PGRAPH_NSTATUS_STATE_IN_USE,       "STATE_IN_USE" },
+	{ NV04_PGRAPH_NSTATUS_INVALID_STATE,      "INVALID_STATE" },
+	{ NV04_PGRAPH_NSTATUS_BAD_ARGUMENT,       "BAD_ARGUMENT" },
+	{ NV04_PGRAPH_NSTATUS_PROTECTION_FAULT,   "PROTECTION_FAULT" },
+	{}
+};
+
+struct nouveau_bitfield nv04_graph_nsource[] = {
+	{ NV03_PGRAPH_NSOURCE_NOTIFICATION,       "NOTIFICATION" },
+	{ NV03_PGRAPH_NSOURCE_DATA_ERROR,         "DATA_ERROR" },
+	{ NV03_PGRAPH_NSOURCE_PROTECTION_ERROR,   "PROTECTION_ERROR" },
+	{ NV03_PGRAPH_NSOURCE_RANGE_EXCEPTION,    "RANGE_EXCEPTION" },
+	{ NV03_PGRAPH_NSOURCE_LIMIT_COLOR,        "LIMIT_COLOR" },
+	{ NV03_PGRAPH_NSOURCE_LIMIT_ZETA,         "LIMIT_ZETA" },
+	{ NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD,       "ILLEGAL_MTHD" },
+	{ NV03_PGRAPH_NSOURCE_DMA_R_PROTECTION,   "DMA_R_PROTECTION" },
+	{ NV03_PGRAPH_NSOURCE_DMA_W_PROTECTION,   "DMA_W_PROTECTION" },
+	{ NV03_PGRAPH_NSOURCE_FORMAT_EXCEPTION,   "FORMAT_EXCEPTION" },
+	{ NV03_PGRAPH_NSOURCE_PATCH_EXCEPTION,    "PATCH_EXCEPTION" },
+	{ NV03_PGRAPH_NSOURCE_STATE_INVALID,      "STATE_INVALID" },
+	{ NV03_PGRAPH_NSOURCE_DOUBLE_NOTIFY,      "DOUBLE_NOTIFY" },
+	{ NV03_PGRAPH_NSOURCE_NOTIFY_IN_USE,      "NOTIFY_IN_USE" },
+	{ NV03_PGRAPH_NSOURCE_METHOD_CNT,         "METHOD_CNT" },
+	{ NV03_PGRAPH_NSOURCE_BFR_NOTIFICATION,   "BFR_NOTIFICATION" },
+	{ NV03_PGRAPH_NSOURCE_DMA_VTX_PROTECTION, "DMA_VTX_PROTECTION" },
+	{ NV03_PGRAPH_NSOURCE_DMA_WIDTH_A,        "DMA_WIDTH_A" },
+	{ NV03_PGRAPH_NSOURCE_DMA_WIDTH_B,        "DMA_WIDTH_B" },
+	{}
+};
+
+static void
+nv04_graph_context_switch(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_channel *chan = NULL;
+	int chid;
 
-	if (dev_priv->engine.graph.registered)
-		return 0;
+	nouveau_wait_for_idle(dev);
+
+	/* If previous context is valid, we need to save it */
+	nv04_graph_unload_context(dev);
+
+	/* Load context for next channel */
+	chid = dev_priv->engine.fifo.channel_id(dev);
+	chan = dev_priv->channels.ptr[chid];
+	if (chan)
+		nv04_graph_load_context(chan);
+}
+
+static void
+nv04_graph_isr(struct drm_device *dev)
+{
+	u32 stat;
+
+	while ((stat = nv_rd32(dev, NV03_PGRAPH_INTR))) {
+		u32 nsource = nv_rd32(dev, NV03_PGRAPH_NSOURCE);
+		u32 nstatus = nv_rd32(dev, NV03_PGRAPH_NSTATUS);
+		u32 addr = nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR);
+		u32 chid = (addr & 0x0f000000) >> 24;
+		u32 subc = (addr & 0x0000e000) >> 13;
+		u32 mthd = (addr & 0x00001ffc);
+		u32 data = nv_rd32(dev, NV04_PGRAPH_TRAPPED_DATA);
+		u32 class = nv_rd32(dev, 0x400180 + subc * 4) & 0xff;
+		u32 show = stat;
+
+		if (stat & NV_PGRAPH_INTR_NOTIFY) {
+			if (nsource & NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD) {
+				if (!nouveau_gpuobj_mthd_call2(dev, chid, class, mthd, data))
+					show &= ~NV_PGRAPH_INTR_NOTIFY;
+			}
+		}
+
+		if (stat & NV_PGRAPH_INTR_CONTEXT_SWITCH) {
+			nv_wr32(dev, NV03_PGRAPH_INTR, NV_PGRAPH_INTR_CONTEXT_SWITCH);
+			stat &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
+			show &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
+			nv04_graph_context_switch(dev);
+		}
+
+		nv_wr32(dev, NV03_PGRAPH_INTR, stat);
+		nv_wr32(dev, NV04_PGRAPH_FIFO, 0x00000001);
+
+		if (show && nouveau_ratelimit()) {
+			NV_INFO(dev, "PGRAPH -");
+			nouveau_bitfield_print(nv04_graph_intr, show);
+			printk(" nsource:");
+			nouveau_bitfield_print(nv04_graph_nsource, nsource);
+			printk(" nstatus:");
+			nouveau_bitfield_print(nv04_graph_nstatus, nstatus);
+			printk("\n");
+			NV_INFO(dev, "PGRAPH - ch %d/%d class 0x%04x "
+				     "mthd 0x%04x data 0x%08x\n",
+				chid, subc, class, mthd, data);
+		}
+	}
+}
+
+static void
+nv04_graph_destroy(struct drm_device *dev, int engine)
+{
+	struct nv04_graph_engine *pgraph = nv_engine(dev, engine);
+
+	nouveau_irq_unregister(dev, 12);
+
+	NVOBJ_ENGINE_DEL(dev, GR);
+	kfree(pgraph);
+}
+
+int
+nv04_graph_create(struct drm_device *dev)
+{
+	struct nv04_graph_engine *pgraph;
+
+	pgraph = kzalloc(sizeof(*pgraph), GFP_KERNEL);
+	if (!pgraph)
+		return -ENOMEM;
+
+	pgraph->base.destroy = nv04_graph_destroy;
+	pgraph->base.init = nv04_graph_init;
+	pgraph->base.fini = nv04_graph_fini;
+	pgraph->base.context_new = nv04_graph_context_new;
+	pgraph->base.context_del = nv04_graph_context_del;
+	pgraph->base.object_new = nv04_graph_object_new;
+
+	NVOBJ_ENGINE_ADD(dev, GR, &pgraph->base);
+	nouveau_irq_register(dev, 12, nv04_graph_isr);
 
 	/* dvd subpicture */
 	NVOBJ_CLASS(dev, 0x0038, GR);
@@ -1222,93 +1355,5 @@ nv04_graph_register(struct drm_device *dev)
 	NVOBJ_CLASS(dev, 0x506e, SW);
 	NVOBJ_MTHD (dev, 0x506e, 0x0150, nv04_graph_mthd_set_ref);
 	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
-
-	dev_priv->engine.graph.registered = true;
 	return 0;
-};
-
-static struct nouveau_bitfield nv04_graph_intr[] = {
-	{ NV_PGRAPH_INTR_NOTIFY, "NOTIFY" },
-	{}
-};
-
-static struct nouveau_bitfield nv04_graph_nstatus[] =
-{
-	{ NV04_PGRAPH_NSTATUS_STATE_IN_USE,       "STATE_IN_USE" },
-	{ NV04_PGRAPH_NSTATUS_INVALID_STATE,      "INVALID_STATE" },
-	{ NV04_PGRAPH_NSTATUS_BAD_ARGUMENT,       "BAD_ARGUMENT" },
-	{ NV04_PGRAPH_NSTATUS_PROTECTION_FAULT,   "PROTECTION_FAULT" },
-	{}
-};
-
-struct nouveau_bitfield nv04_graph_nsource[] =
-{
-	{ NV03_PGRAPH_NSOURCE_NOTIFICATION,       "NOTIFICATION" },
-	{ NV03_PGRAPH_NSOURCE_DATA_ERROR,         "DATA_ERROR" },
-	{ NV03_PGRAPH_NSOURCE_PROTECTION_ERROR,   "PROTECTION_ERROR" },
-	{ NV03_PGRAPH_NSOURCE_RANGE_EXCEPTION,    "RANGE_EXCEPTION" },
-	{ NV03_PGRAPH_NSOURCE_LIMIT_COLOR,        "LIMIT_COLOR" },
-	{ NV03_PGRAPH_NSOURCE_LIMIT_ZETA,         "LIMIT_ZETA" },
-	{ NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD,       "ILLEGAL_MTHD" },
-	{ NV03_PGRAPH_NSOURCE_DMA_R_PROTECTION,   "DMA_R_PROTECTION" },
-	{ NV03_PGRAPH_NSOURCE_DMA_W_PROTECTION,   "DMA_W_PROTECTION" },
-	{ NV03_PGRAPH_NSOURCE_FORMAT_EXCEPTION,   "FORMAT_EXCEPTION" },
-	{ NV03_PGRAPH_NSOURCE_PATCH_EXCEPTION,    "PATCH_EXCEPTION" },
-	{ NV03_PGRAPH_NSOURCE_STATE_INVALID,      "STATE_INVALID" },
-	{ NV03_PGRAPH_NSOURCE_DOUBLE_NOTIFY,      "DOUBLE_NOTIFY" },
-	{ NV03_PGRAPH_NSOURCE_NOTIFY_IN_USE,      "NOTIFY_IN_USE" },
-	{ NV03_PGRAPH_NSOURCE_METHOD_CNT,         "METHOD_CNT" },
-	{ NV03_PGRAPH_NSOURCE_BFR_NOTIFICATION,   "BFR_NOTIFICATION" },
-	{ NV03_PGRAPH_NSOURCE_DMA_VTX_PROTECTION, "DMA_VTX_PROTECTION" },
-	{ NV03_PGRAPH_NSOURCE_DMA_WIDTH_A,        "DMA_WIDTH_A" },
-	{ NV03_PGRAPH_NSOURCE_DMA_WIDTH_B,        "DMA_WIDTH_B" },
-	{}
-};
-
-static void
-nv04_graph_isr(struct drm_device *dev)
-{
-	u32 stat;
-
-	while ((stat = nv_rd32(dev, NV03_PGRAPH_INTR))) {
-		u32 nsource = nv_rd32(dev, NV03_PGRAPH_NSOURCE);
-		u32 nstatus = nv_rd32(dev, NV03_PGRAPH_NSTATUS);
-		u32 addr = nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR);
-		u32 chid = (addr & 0x0f000000) >> 24;
-		u32 subc = (addr & 0x0000e000) >> 13;
-		u32 mthd = (addr & 0x00001ffc);
-		u32 data = nv_rd32(dev, NV04_PGRAPH_TRAPPED_DATA);
-		u32 class = nv_rd32(dev, 0x400180 + subc * 4) & 0xff;
-		u32 show = stat;
-
-		if (stat & NV_PGRAPH_INTR_NOTIFY) {
-			if (nsource & NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD) {
-				if (!nouveau_gpuobj_mthd_call2(dev, chid, class, mthd, data))
-					show &= ~NV_PGRAPH_INTR_NOTIFY;
-			}
-		}
-
-		if (stat & NV_PGRAPH_INTR_CONTEXT_SWITCH) {
-			nv_wr32(dev, NV03_PGRAPH_INTR, NV_PGRAPH_INTR_CONTEXT_SWITCH);
-			stat &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
-			show &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
-			nv04_graph_context_switch(dev);
-		}
-
-		nv_wr32(dev, NV03_PGRAPH_INTR, stat);
-		nv_wr32(dev, NV04_PGRAPH_FIFO, 0x00000001);
-
-		if (show && nouveau_ratelimit()) {
-			NV_INFO(dev, "PGRAPH -");
-			nouveau_bitfield_print(nv04_graph_intr, show);
-			printk(" nsource:");
-			nouveau_bitfield_print(nv04_graph_nsource, nsource);
-			printk(" nstatus:");
-			nouveau_bitfield_print(nv04_graph_nstatus, nstatus);
-			printk("\n");
-			NV_INFO(dev, "PGRAPH - ch %d/%d class 0x%04x "
-				     "mthd 0x%04x data 0x%08x\n",
-				chid, subc, class, mthd, data);
-		}
-	}
 }
diff --git a/drivers/gpu/drm/nouveau/nv04_instmem.c b/drivers/gpu/drm/nouveau/nv04_instmem.c
index b8e3edb5c063..b8611b955313 100644
--- a/drivers/gpu/drm/nouveau/nv04_instmem.c
+++ b/drivers/gpu/drm/nouveau/nv04_instmem.c
@@ -95,6 +95,9 @@ nv04_instmem_takedown(struct drm_device *dev)
 	nouveau_ramht_ref(NULL, &dev_priv->ramht, NULL);
 	nouveau_gpuobj_ref(NULL, &dev_priv->ramro);
 	nouveau_gpuobj_ref(NULL, &dev_priv->ramfc);
+
+	if (drm_mm_initialized(&dev_priv->ramin_heap))
+		drm_mm_takedown(&dev_priv->ramin_heap);
 }
 
 int
diff --git a/drivers/gpu/drm/nouveau/nv10_graph.c b/drivers/gpu/drm/nouveau/nv10_graph.c
index 8c92edb7bbcd..0930c6cb88e0 100644
--- a/drivers/gpu/drm/nouveau/nv10_graph.c
+++ b/drivers/gpu/drm/nouveau/nv10_graph.c
@@ -28,10 +28,9 @@
 #include "nouveau_drv.h"
 #include "nouveau_util.h"
 
-static int  nv10_graph_register(struct drm_device *);
-static void nv10_graph_isr(struct drm_device *);
-
-#define NV10_FIFO_NUMBER 32
+struct nv10_graph_engine {
+	struct nouveau_exec_engine base;
+};
 
 struct pipe_state {
 	uint32_t pipe_0x0000[0x040/4];
@@ -414,9 +413,9 @@ struct graph_state {
 
 static void nv10_graph_save_pipe(struct nouveau_channel *chan)
 {
-	struct drm_device *dev = chan->dev;
-	struct graph_state *pgraph_ctx = chan->pgraph_ctx;
+	struct graph_state *pgraph_ctx = chan->engctx[NVOBJ_ENGINE_GR];
 	struct pipe_state *pipe = &pgraph_ctx->pipe_state;
+	struct drm_device *dev = chan->dev;
 
 	PIPE_SAVE(dev, pipe->pipe_0x4400, 0x4400);
 	PIPE_SAVE(dev, pipe->pipe_0x0200, 0x0200);
@@ -432,9 +431,9 @@ static void nv10_graph_save_pipe(struct nouveau_channel *chan)
 
 static void nv10_graph_load_pipe(struct nouveau_channel *chan)
 {
-	struct drm_device *dev = chan->dev;
-	struct graph_state *pgraph_ctx = chan->pgraph_ctx;
+	struct graph_state *pgraph_ctx = chan->engctx[NVOBJ_ENGINE_GR];
 	struct pipe_state *pipe = &pgraph_ctx->pipe_state;
+	struct drm_device *dev = chan->dev;
 	uint32_t xfmode0, xfmode1;
 	int i;
 
@@ -482,9 +481,9 @@ static void nv10_graph_load_pipe(struct nouveau_channel *chan)
 
 static void nv10_graph_create_pipe(struct nouveau_channel *chan)
 {
-	struct drm_device *dev = chan->dev;
-	struct graph_state *pgraph_ctx = chan->pgraph_ctx;
+	struct graph_state *pgraph_ctx = chan->engctx[NVOBJ_ENGINE_GR];
 	struct pipe_state *fifo_pipe_state = &pgraph_ctx->pipe_state;
+	struct drm_device *dev = chan->dev;
 	uint32_t *fifo_pipe_state_addr;
 	int i;
 #define PIPE_INIT(addr) \
@@ -661,8 +660,6 @@ static void nv10_graph_load_dma_vtxbuf(struct nouveau_channel *chan,
 				       uint32_t inst)
 {
 	struct drm_device *dev = chan->dev;
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	uint32_t st2, st2_dl, st2_dh, fifo_ptr, fifo[0x60/4];
 	uint32_t ctx_user, ctx_switch[5];
 	int i, subchan = -1;
@@ -711,8 +708,8 @@ static void nv10_graph_load_dma_vtxbuf(struct nouveau_channel *chan,
 		0x2c000000 | chan->id << 20 | subchan << 16 | 0x18c);
 	nv_wr32(dev, NV10_PGRAPH_FFINTFC_ST2_DL, inst);
 	nv_mask(dev, NV10_PGRAPH_CTX_CONTROL, 0, 0x10000);
-	pgraph->fifo_access(dev, true);
-	pgraph->fifo_access(dev, false);
+	nv04_graph_fifo_access(dev, true);
+	nv04_graph_fifo_access(dev, false);
 
 	/* Restore the FIFO state */
 	for (i = 0; i < ARRAY_SIZE(fifo); i++)
@@ -729,11 +726,12 @@ static void nv10_graph_load_dma_vtxbuf(struct nouveau_channel *chan,
 	nv_wr32(dev, NV10_PGRAPH_CTX_USER, ctx_user);
 }
 
-int nv10_graph_load_context(struct nouveau_channel *chan)
+static int
+nv10_graph_load_context(struct nouveau_channel *chan)
 {
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct graph_state *pgraph_ctx = chan->pgraph_ctx;
+	struct graph_state *pgraph_ctx = chan->engctx[NVOBJ_ENGINE_GR];
 	uint32_t tmp;
 	int i;
 
@@ -757,21 +755,20 @@ int nv10_graph_load_context(struct nouveau_channel *chan)
 	return 0;
 }
 
-int
+static int
 nv10_graph_unload_context(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
 	struct nouveau_channel *chan;
 	struct graph_state *ctx;
 	uint32_t tmp;
 	int i;
 
-	chan = pgraph->channel(dev);
+	chan = nv10_graph_channel(dev);
 	if (!chan)
 		return 0;
-	ctx = chan->pgraph_ctx;
+	ctx = chan->engctx[NVOBJ_ENGINE_GR];
 
 	for (i = 0; i < ARRAY_SIZE(nv10_graph_ctx_regs); i++)
 		ctx->nv10[i] = nv_rd32(dev, nv10_graph_ctx_regs[i]);
@@ -805,7 +802,7 @@ nv10_graph_context_switch(struct drm_device *dev)
 	/* Load context for next channel */
 	chid = (nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR) >> 20) & 0x1f;
 	chan = dev_priv->channels.ptr[chid];
-	if (chan && chan->pgraph_ctx)
+	if (chan && chan->engctx[NVOBJ_ENGINE_GR])
 		nv10_graph_load_context(chan);
 }
 
@@ -836,7 +833,8 @@ nv10_graph_channel(struct drm_device *dev)
 	return dev_priv->channels.ptr[chid];
 }
 
-int nv10_graph_create_context(struct nouveau_channel *chan)
+static int
+nv10_graph_context_new(struct nouveau_channel *chan, int engine)
 {
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
@@ -844,11 +842,10 @@ int nv10_graph_create_context(struct nouveau_channel *chan)
 
 	NV_DEBUG(dev, "nv10_graph_context_create %d\n", chan->id);
 
-	chan->pgraph_ctx = pgraph_ctx = kzalloc(sizeof(*pgraph_ctx),
-						GFP_KERNEL);
+	pgraph_ctx = kzalloc(sizeof(*pgraph_ctx), GFP_KERNEL);
 	if (pgraph_ctx == NULL)
 		return -ENOMEM;
-
+	chan->engctx[engine] = pgraph_ctx;
 
 	NV_WRITE_CTX(0x00400e88, 0x08000000);
 	NV_WRITE_CTX(0x00400e9c, 0x4b7fffff);
@@ -873,30 +870,30 @@ int nv10_graph_create_context(struct nouveau_channel *chan)
 	return 0;
 }
 
-void nv10_graph_destroy_context(struct nouveau_channel *chan)
+static void
+nv10_graph_context_del(struct nouveau_channel *chan, int engine)
 {
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct graph_state *pgraph_ctx = chan->pgraph_ctx;
+	struct graph_state *pgraph_ctx = chan->engctx[engine];
 	unsigned long flags;
 
 	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
-	pgraph->fifo_access(dev, false);
+	nv04_graph_fifo_access(dev, false);
 
 	/* Unload the context if it's the currently active one */
-	if (pgraph->channel(dev) == chan)
-		pgraph->unload_context(dev);
+	if (nv10_graph_channel(dev) == chan)
+		nv10_graph_unload_context(dev);
+
+	nv04_graph_fifo_access(dev, true);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
 
 	/* Free the context resources */
+	chan->engctx[engine] = NULL;
 	kfree(pgraph_ctx);
-	chan->pgraph_ctx = NULL;
-
-	pgraph->fifo_access(dev, true);
-	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
 }
 
-void
+static void
 nv10_graph_set_tile_region(struct drm_device *dev, int i)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
@@ -907,22 +904,18 @@ nv10_graph_set_tile_region(struct drm_device *dev, int i)
 	nv_wr32(dev, NV10_PGRAPH_TILE(i), tile->addr);
 }
 
-int nv10_graph_init(struct drm_device *dev)
+static int
+nv10_graph_init(struct drm_device *dev, int engine)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t tmp;
-	int ret, i;
+	u32 tmp;
+	int i;
 
 	nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) &
 			~NV_PMC_ENABLE_PGRAPH);
 	nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) |
 			 NV_PMC_ENABLE_PGRAPH);
 
-	ret = nv10_graph_register(dev);
-	if (ret)
-		return ret;
-
-	nouveau_irq_register(dev, 12, nv10_graph_isr);
 	nv_wr32(dev, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
 	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
 
@@ -963,18 +956,20 @@ int nv10_graph_init(struct drm_device *dev)
 	return 0;
 }
 
-void nv10_graph_takedown(struct drm_device *dev)
+static int
+nv10_graph_fini(struct drm_device *dev, int engine)
 {
+	nv10_graph_unload_context(dev);
 	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0x00000000);
-	nouveau_irq_unregister(dev, 12);
+	return 0;
 }
 
 static int
 nv17_graph_mthd_lma_window(struct nouveau_channel *chan,
 			   u32 class, u32 mthd, u32 data)
 {
+	struct graph_state *ctx = chan->engctx[NVOBJ_ENGINE_GR];
 	struct drm_device *dev = chan->dev;
-	struct graph_state *ctx = chan->pgraph_ctx;
 	struct pipe_state *pipe = &ctx->pipe_state;
 	uint32_t pipe_0x0040[1], pipe_0x64c0[8], pipe_0x6a80[3], pipe_0x6ab0[3];
 	uint32_t xfmode0, xfmode1;
@@ -1061,64 +1056,13 @@ nv17_graph_mthd_lma_enable(struct nouveau_channel *chan,
 	return 0;
 }
 
-static int
-nv10_graph_register(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	if (dev_priv->engine.graph.registered)
-		return 0;
-
-	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
-	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
-	NVOBJ_CLASS(dev, 0x0039, GR); /* m2mf */
-	NVOBJ_CLASS(dev, 0x004a, GR); /* gdirect */
-	NVOBJ_CLASS(dev, 0x005f, GR); /* imageblit */
-	NVOBJ_CLASS(dev, 0x009f, GR); /* imageblit (nv12) */
-	NVOBJ_CLASS(dev, 0x008a, GR); /* ifc */
-	NVOBJ_CLASS(dev, 0x0089, GR); /* sifm */
-	NVOBJ_CLASS(dev, 0x0062, GR); /* surf2d */
-	NVOBJ_CLASS(dev, 0x0043, GR); /* rop */
-	NVOBJ_CLASS(dev, 0x0012, GR); /* beta1 */
-	NVOBJ_CLASS(dev, 0x0072, GR); /* beta4 */
-	NVOBJ_CLASS(dev, 0x0019, GR); /* cliprect */
-	NVOBJ_CLASS(dev, 0x0044, GR); /* pattern */
-	NVOBJ_CLASS(dev, 0x0052, GR); /* swzsurf */
-	NVOBJ_CLASS(dev, 0x0093, GR); /* surf3d */
-	NVOBJ_CLASS(dev, 0x0094, GR); /* tex_tri */
-	NVOBJ_CLASS(dev, 0x0095, GR); /* multitex_tri */
-
-	/* celcius */
-	if (dev_priv->chipset <= 0x10) {
-		NVOBJ_CLASS(dev, 0x0056, GR);
-	} else
-	if (dev_priv->chipset < 0x17 || dev_priv->chipset == 0x1a) {
-		NVOBJ_CLASS(dev, 0x0096, GR);
-	} else {
-		NVOBJ_CLASS(dev, 0x0099, GR);
-		NVOBJ_MTHD (dev, 0x0099, 0x1638, nv17_graph_mthd_lma_window);
-		NVOBJ_MTHD (dev, 0x0099, 0x163c, nv17_graph_mthd_lma_window);
-		NVOBJ_MTHD (dev, 0x0099, 0x1640, nv17_graph_mthd_lma_window);
-		NVOBJ_MTHD (dev, 0x0099, 0x1644, nv17_graph_mthd_lma_window);
-		NVOBJ_MTHD (dev, 0x0099, 0x1658, nv17_graph_mthd_lma_enable);
-	}
-
-	/* nvsw */
-	NVOBJ_CLASS(dev, 0x506e, SW);
-	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
-
-	dev_priv->engine.graph.registered = true;
-	return 0;
-}
-
 struct nouveau_bitfield nv10_graph_intr[] = {
 	{ NV_PGRAPH_INTR_NOTIFY, "NOTIFY" },
 	{ NV_PGRAPH_INTR_ERROR,  "ERROR"  },
 	{}
 };
 
-struct nouveau_bitfield nv10_graph_nstatus[] =
-{
+struct nouveau_bitfield nv10_graph_nstatus[] = {
 	{ NV10_PGRAPH_NSTATUS_STATE_IN_USE,       "STATE_IN_USE" },
 	{ NV10_PGRAPH_NSTATUS_INVALID_STATE,      "INVALID_STATE" },
 	{ NV10_PGRAPH_NSTATUS_BAD_ARGUMENT,       "BAD_ARGUMENT" },
@@ -1173,3 +1117,73 @@ nv10_graph_isr(struct drm_device *dev)
 		}
 	}
 }
+
+static void
+nv10_graph_destroy(struct drm_device *dev, int engine)
+{
+	struct nv10_graph_engine *pgraph = nv_engine(dev, engine);
+
+	nouveau_irq_unregister(dev, 12);
+	kfree(pgraph);
+}
+
+int
+nv10_graph_create(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nv10_graph_engine *pgraph;
+
+	pgraph = kzalloc(sizeof(*pgraph), GFP_KERNEL);
+	if (!pgraph)
+		return -ENOMEM;
+
+	pgraph->base.destroy = nv10_graph_destroy;
+	pgraph->base.init = nv10_graph_init;
+	pgraph->base.fini = nv10_graph_fini;
+	pgraph->base.context_new = nv10_graph_context_new;
+	pgraph->base.context_del = nv10_graph_context_del;
+	pgraph->base.object_new = nv04_graph_object_new;
+	pgraph->base.set_tile_region = nv10_graph_set_tile_region;
+
+	NVOBJ_ENGINE_ADD(dev, GR, &pgraph->base);
+	nouveau_irq_register(dev, 12, nv10_graph_isr);
+
+	/* nvsw */
+	NVOBJ_CLASS(dev, 0x506e, SW);
+	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
+
+	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
+	NVOBJ_CLASS(dev, 0x0039, GR); /* m2mf */
+	NVOBJ_CLASS(dev, 0x004a, GR); /* gdirect */
+	NVOBJ_CLASS(dev, 0x005f, GR); /* imageblit */
+	NVOBJ_CLASS(dev, 0x009f, GR); /* imageblit (nv12) */
+	NVOBJ_CLASS(dev, 0x008a, GR); /* ifc */
+	NVOBJ_CLASS(dev, 0x0089, GR); /* sifm */
+	NVOBJ_CLASS(dev, 0x0062, GR); /* surf2d */
+	NVOBJ_CLASS(dev, 0x0043, GR); /* rop */
+	NVOBJ_CLASS(dev, 0x0012, GR); /* beta1 */
+	NVOBJ_CLASS(dev, 0x0072, GR); /* beta4 */
+	NVOBJ_CLASS(dev, 0x0019, GR); /* cliprect */
+	NVOBJ_CLASS(dev, 0x0044, GR); /* pattern */
+	NVOBJ_CLASS(dev, 0x0052, GR); /* swzsurf */
+	NVOBJ_CLASS(dev, 0x0093, GR); /* surf3d */
+	NVOBJ_CLASS(dev, 0x0094, GR); /* tex_tri */
+	NVOBJ_CLASS(dev, 0x0095, GR); /* multitex_tri */
+
+	/* celcius */
+	if (dev_priv->chipset <= 0x10) {
+		NVOBJ_CLASS(dev, 0x0056, GR);
+	} else
+	if (dev_priv->chipset < 0x17 || dev_priv->chipset == 0x1a) {
+		NVOBJ_CLASS(dev, 0x0096, GR);
+	} else {
+		NVOBJ_CLASS(dev, 0x0099, GR);
+		NVOBJ_MTHD (dev, 0x0099, 0x1638, nv17_graph_mthd_lma_window);
+		NVOBJ_MTHD (dev, 0x0099, 0x163c, nv17_graph_mthd_lma_window);
+		NVOBJ_MTHD (dev, 0x0099, 0x1640, nv17_graph_mthd_lma_window);
+		NVOBJ_MTHD (dev, 0x0099, 0x1644, nv17_graph_mthd_lma_window);
+		NVOBJ_MTHD (dev, 0x0099, 0x1658, nv17_graph_mthd_lma_enable);
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nv20_graph.c b/drivers/gpu/drm/nouveau/nv20_graph.c
index 8464b76798d5..affc7d7dd029 100644
--- a/drivers/gpu/drm/nouveau/nv20_graph.c
+++ b/drivers/gpu/drm/nouveau/nv20_graph.c
@@ -24,6 +24,14 @@
  *
  */
 
+struct nv20_graph_engine {
+	struct nouveau_exec_engine base;
+	struct nouveau_gpuobj *ctxtab;
+	void (*grctx_init)(struct nouveau_gpuobj *);
+	u32 grctx_size;
+	u32 grctx_user;
+};
+
 #define NV20_GRCTX_SIZE (3580*4)
 #define NV25_GRCTX_SIZE (3529*4)
 #define NV2A_GRCTX_SIZE (3500*4)
@@ -32,12 +40,54 @@
 #define NV34_GRCTX_SIZE    (18140)
 #define NV35_36_GRCTX_SIZE (22396)
 
-static int nv20_graph_register(struct drm_device *);
-static int nv30_graph_register(struct drm_device *);
-static void nv20_graph_isr(struct drm_device *);
+int
+nv20_graph_unload_context(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
+	struct nouveau_channel *chan;
+	struct nouveau_gpuobj *grctx;
+	u32 tmp;
+
+	chan = nv10_graph_channel(dev);
+	if (!chan)
+		return 0;
+	grctx = chan->engctx[NVOBJ_ENGINE_GR];
+
+	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_POINTER, grctx->pinst >> 4);
+	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_XFER,
+		     NV20_PGRAPH_CHANNEL_CTX_XFER_SAVE);
+
+	nouveau_wait_for_idle(dev);
+
+	nv_wr32(dev, NV10_PGRAPH_CTX_CONTROL, 0x10000000);
+	tmp  = nv_rd32(dev, NV10_PGRAPH_CTX_USER) & 0x00ffffff;
+	tmp |= (pfifo->channels - 1) << 24;
+	nv_wr32(dev, NV10_PGRAPH_CTX_USER, tmp);
+	return 0;
+}
+
+static void
+nv20_graph_rdi(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	int i, writecount = 32;
+	uint32_t rdi_index = 0x2c80000;
+
+	if (dev_priv->chipset == 0x20) {
+		rdi_index = 0x3d0000;
+		writecount = 15;
+	}
+
+	nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, rdi_index);
+	for (i = 0; i < writecount; i++)
+		nv_wr32(dev, NV10_PGRAPH_RDI_DATA, 0);
+
+	nouveau_wait_for_idle(dev);
+}
 
 static void
-nv20_graph_context_init(struct drm_device *dev, struct nouveau_gpuobj *ctx)
+nv20_graph_context_init(struct nouveau_gpuobj *ctx)
 {
 	int i;
 
@@ -87,7 +137,7 @@ nv20_graph_context_init(struct drm_device *dev, struct nouveau_gpuobj *ctx)
 }
 
 static void
-nv25_graph_context_init(struct drm_device *dev, struct nouveau_gpuobj *ctx)
+nv25_graph_context_init(struct nouveau_gpuobj *ctx)
 {
 	int i;
 
@@ -146,7 +196,7 @@ nv25_graph_context_init(struct drm_device *dev, struct nouveau_gpuobj *ctx)
 }
 
 static void
-nv2a_graph_context_init(struct drm_device *dev, struct nouveau_gpuobj *ctx)
+nv2a_graph_context_init(struct nouveau_gpuobj *ctx)
 {
 	int i;
 
@@ -196,7 +246,7 @@ nv2a_graph_context_init(struct drm_device *dev, struct nouveau_gpuobj *ctx)
 }
 
 static void
-nv30_31_graph_context_init(struct drm_device *dev, struct nouveau_gpuobj *ctx)
+nv30_31_graph_context_init(struct nouveau_gpuobj *ctx)
 {
 	int i;
 
@@ -254,7 +304,7 @@ nv30_31_graph_context_init(struct drm_device *dev, struct nouveau_gpuobj *ctx)
 }
 
 static void
-nv34_graph_context_init(struct drm_device *dev, struct nouveau_gpuobj *ctx)
+nv34_graph_context_init(struct nouveau_gpuobj *ctx)
 {
 	int i;
 
@@ -312,7 +362,7 @@ nv34_graph_context_init(struct drm_device *dev, struct nouveau_gpuobj *ctx)
 }
 
 static void
-nv35_36_graph_context_init(struct drm_device *dev, struct nouveau_gpuobj *ctx)
+nv35_36_graph_context_init(struct nouveau_gpuobj *ctx)
 {
 	int i;
 
@@ -370,148 +420,57 @@ nv35_36_graph_context_init(struct drm_device *dev, struct nouveau_gpuobj *ctx)
 }
 
 int
-nv20_graph_create_context(struct nouveau_channel *chan)
+nv20_graph_context_new(struct nouveau_channel *chan, int engine)
 {
+	struct nv20_graph_engine *pgraph = nv_engine(chan->dev, engine);
+	struct nouveau_gpuobj *grctx = NULL;
 	struct drm_device *dev = chan->dev;
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	void (*ctx_init)(struct drm_device *, struct nouveau_gpuobj *);
-	unsigned int idoffs = 0x28;
 	int ret;
 
-	switch (dev_priv->chipset) {
-	case 0x20:
-		ctx_init = nv20_graph_context_init;
-		idoffs = 0;
-		break;
-	case 0x25:
-	case 0x28:
-		ctx_init = nv25_graph_context_init;
-		break;
-	case 0x2a:
-		ctx_init = nv2a_graph_context_init;
-		idoffs = 0;
-		break;
-	case 0x30:
-	case 0x31:
-		ctx_init = nv30_31_graph_context_init;
-		break;
-	case 0x34:
-		ctx_init = nv34_graph_context_init;
-		break;
-	case 0x35:
-	case 0x36:
-		ctx_init = nv35_36_graph_context_init;
-		break;
-	default:
-		BUG_ON(1);
-	}
-
-	ret = nouveau_gpuobj_new(dev, chan, pgraph->grctx_size, 16,
-				 NVOBJ_FLAG_ZERO_ALLOC, &chan->ramin_grctx);
+	ret = nouveau_gpuobj_new(dev, NULL, pgraph->grctx_size, 16,
+				 NVOBJ_FLAG_ZERO_ALLOC, &grctx);
 	if (ret)
 		return ret;
 
 	/* Initialise default context values */
-	ctx_init(dev, chan->ramin_grctx);
+	pgraph->grctx_init(grctx);
 
 	/* nv20: nv_wo32(dev, chan->ramin_grctx->gpuobj, 10, chan->id<<24); */
-	nv_wo32(chan->ramin_grctx, idoffs,
-		(chan->id << 24) | 0x1); /* CTX_USER */
+	/* CTX_USER */
+	nv_wo32(grctx, pgraph->grctx_user, (chan->id << 24) | 0x1);
 
-	nv_wo32(pgraph->ctx_table, chan->id * 4, chan->ramin_grctx->pinst >> 4);
+	nv_wo32(pgraph->ctxtab, chan->id * 4, grctx->pinst >> 4);
+	chan->engctx[engine] = grctx;
 	return 0;
 }
 
 void
-nv20_graph_destroy_context(struct nouveau_channel *chan)
+nv20_graph_context_del(struct nouveau_channel *chan, int engine)
 {
+	struct nv20_graph_engine *pgraph = nv_engine(chan->dev, engine);
+	struct nouveau_gpuobj *grctx = chan->engctx[engine];
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	unsigned long flags;
 
 	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
-	pgraph->fifo_access(dev, false);
+	nv04_graph_fifo_access(dev, false);
 
 	/* Unload the context if it's the currently active one */
-	if (pgraph->channel(dev) == chan)
-		pgraph->unload_context(dev);
+	if (nv10_graph_channel(dev) == chan)
+		nv20_graph_unload_context(dev);
 
-	pgraph->fifo_access(dev, true);
+	nv04_graph_fifo_access(dev, true);
 	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
 
 	/* Free the context resources */
-	nv_wo32(pgraph->ctx_table, chan->id * 4, 0);
-	nouveau_gpuobj_ref(NULL, &chan->ramin_grctx);
-}
-
-int
-nv20_graph_load_context(struct nouveau_channel *chan)
-{
-	struct drm_device *dev = chan->dev;
-	uint32_t inst;
+	nv_wo32(pgraph->ctxtab, chan->id * 4, 0);
 
-	if (!chan->ramin_grctx)
-		return -EINVAL;
-	inst = chan->ramin_grctx->pinst >> 4;
-
-	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_POINTER, inst);
-	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_XFER,
-		     NV20_PGRAPH_CHANNEL_CTX_XFER_LOAD);
-	nv_wr32(dev, NV10_PGRAPH_CTX_CONTROL, 0x10010100);
-
-	nouveau_wait_for_idle(dev);
-	return 0;
-}
-
-int
-nv20_graph_unload_context(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
-	struct nouveau_channel *chan;
-	uint32_t inst, tmp;
-
-	chan = pgraph->channel(dev);
-	if (!chan)
-		return 0;
-	inst = chan->ramin_grctx->pinst >> 4;
-
-	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_POINTER, inst);
-	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_XFER,
-		     NV20_PGRAPH_CHANNEL_CTX_XFER_SAVE);
-
-	nouveau_wait_for_idle(dev);
-
-	nv_wr32(dev, NV10_PGRAPH_CTX_CONTROL, 0x10000000);
-	tmp  = nv_rd32(dev, NV10_PGRAPH_CTX_USER) & 0x00ffffff;
-	tmp |= (pfifo->channels - 1) << 24;
-	nv_wr32(dev, NV10_PGRAPH_CTX_USER, tmp);
-	return 0;
+	nouveau_gpuobj_ref(NULL, &grctx);
+	chan->engctx[engine] = NULL;
 }
 
 static void
-nv20_graph_rdi(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	int i, writecount = 32;
-	uint32_t rdi_index = 0x2c80000;
-
-	if (dev_priv->chipset == 0x20) {
-		rdi_index = 0x3d0000;
-		writecount = 15;
-	}
-
-	nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, rdi_index);
-	for (i = 0; i < writecount; i++)
-		nv_wr32(dev, NV10_PGRAPH_RDI_DATA, 0);
-
-	nouveau_wait_for_idle(dev);
-}
-
-void
 nv20_graph_set_tile_region(struct drm_device *dev, int i)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
@@ -536,56 +495,22 @@ nv20_graph_set_tile_region(struct drm_device *dev, int i)
 }
 
 int
-nv20_graph_init(struct drm_device *dev)
+nv20_graph_init(struct drm_device *dev, int engine)
 {
+	struct nv20_graph_engine *pgraph = nv_engine(dev, engine);
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	uint32_t tmp, vramsz;
-	int ret, i;
-
-	switch (dev_priv->chipset) {
-	case 0x20:
-		pgraph->grctx_size = NV20_GRCTX_SIZE;
-		break;
-	case 0x25:
-	case 0x28:
-		pgraph->grctx_size = NV25_GRCTX_SIZE;
-		break;
-	case 0x2a:
-		pgraph->grctx_size = NV2A_GRCTX_SIZE;
-		break;
-	default:
-		NV_ERROR(dev, "unknown chipset, disabling acceleration\n");
-		pgraph->accel_blocked = true;
-		return 0;
-	}
+	int i;
 
 	nv_wr32(dev, NV03_PMC_ENABLE,
 		nv_rd32(dev, NV03_PMC_ENABLE) & ~NV_PMC_ENABLE_PGRAPH);
 	nv_wr32(dev, NV03_PMC_ENABLE,
 		nv_rd32(dev, NV03_PMC_ENABLE) |  NV_PMC_ENABLE_PGRAPH);
 
-	if (!pgraph->ctx_table) {
-		/* Create Context Pointer Table */
-		ret = nouveau_gpuobj_new(dev, NULL, 32 * 4, 16,
-					 NVOBJ_FLAG_ZERO_ALLOC,
-					 &pgraph->ctx_table);
-		if (ret)
-			return ret;
-	}
-
-	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_TABLE,
-		     pgraph->ctx_table->pinst >> 4);
+	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_TABLE, pgraph->ctxtab->pinst >> 4);
 
 	nv20_graph_rdi(dev);
 
-	ret = nv20_graph_register(dev);
-	if (ret) {
-		nouveau_gpuobj_ref(NULL, &pgraph->ctx_table);
-		return ret;
-	}
-
-	nouveau_irq_register(dev, 12, nv20_graph_isr);
 	nv_wr32(dev, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
 	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
 
@@ -657,67 +582,20 @@ nv20_graph_init(struct drm_device *dev)
 	return 0;
 }
 
-void
-nv20_graph_takedown(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-
-	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0x00000000);
-	nouveau_irq_unregister(dev, 12);
-
-	nouveau_gpuobj_ref(NULL, &pgraph->ctx_table);
-}
-
 int
-nv30_graph_init(struct drm_device *dev)
+nv30_graph_init(struct drm_device *dev, int engine)
 {
+	struct nv20_graph_engine *pgraph = nv_engine(dev, engine);
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	int ret, i;
-
-	switch (dev_priv->chipset) {
-	case 0x30:
-	case 0x31:
-		pgraph->grctx_size = NV30_31_GRCTX_SIZE;
-		break;
-	case 0x34:
-		pgraph->grctx_size = NV34_GRCTX_SIZE;
-		break;
-	case 0x35:
-	case 0x36:
-		pgraph->grctx_size = NV35_36_GRCTX_SIZE;
-		break;
-	default:
-		NV_ERROR(dev, "unknown chipset, disabling acceleration\n");
-		pgraph->accel_blocked = true;
-		return 0;
-	}
+	int i;
 
 	nv_wr32(dev, NV03_PMC_ENABLE,
 		nv_rd32(dev, NV03_PMC_ENABLE) & ~NV_PMC_ENABLE_PGRAPH);
 	nv_wr32(dev, NV03_PMC_ENABLE,
 		nv_rd32(dev, NV03_PMC_ENABLE) |  NV_PMC_ENABLE_PGRAPH);
 
-	if (!pgraph->ctx_table) {
-		/* Create Context Pointer Table */
-		ret = nouveau_gpuobj_new(dev, NULL, 32 * 4, 16,
-					 NVOBJ_FLAG_ZERO_ALLOC,
-					 &pgraph->ctx_table);
-		if (ret)
-			return ret;
-	}
-
-	ret = nv30_graph_register(dev);
-	if (ret) {
-		nouveau_gpuobj_ref(NULL, &pgraph->ctx_table);
-		return ret;
-	}
+	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_TABLE, pgraph->ctxtab->pinst >> 4);
 
-	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_TABLE,
-		     pgraph->ctx_table->pinst >> 4);
-
-	nouveau_irq_register(dev, 12, nv20_graph_isr);
 	nv_wr32(dev, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
 	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
 
@@ -775,85 +653,11 @@ nv30_graph_init(struct drm_device *dev)
 	return 0;
 }
 
-static int
-nv20_graph_register(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	if (dev_priv->engine.graph.registered)
-		return 0;
-
-	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
-	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
-	NVOBJ_CLASS(dev, 0x0039, GR); /* m2mf */
-	NVOBJ_CLASS(dev, 0x004a, GR); /* gdirect */
-	NVOBJ_CLASS(dev, 0x009f, GR); /* imageblit (nv12) */
-	NVOBJ_CLASS(dev, 0x008a, GR); /* ifc */
-	NVOBJ_CLASS(dev, 0x0089, GR); /* sifm */
-	NVOBJ_CLASS(dev, 0x0062, GR); /* surf2d */
-	NVOBJ_CLASS(dev, 0x0043, GR); /* rop */
-	NVOBJ_CLASS(dev, 0x0012, GR); /* beta1 */
-	NVOBJ_CLASS(dev, 0x0072, GR); /* beta4 */
-	NVOBJ_CLASS(dev, 0x0019, GR); /* cliprect */
-	NVOBJ_CLASS(dev, 0x0044, GR); /* pattern */
-	NVOBJ_CLASS(dev, 0x009e, GR); /* swzsurf */
-	NVOBJ_CLASS(dev, 0x0096, GR); /* celcius */
-
-	/* kelvin */
-	if (dev_priv->chipset < 0x25)
-		NVOBJ_CLASS(dev, 0x0097, GR);
-	else
-		NVOBJ_CLASS(dev, 0x0597, GR);
-
-	/* nvsw */
-	NVOBJ_CLASS(dev, 0x506e, SW);
-	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
-
-	dev_priv->engine.graph.registered = true;
-	return 0;
-}
-
-static int
-nv30_graph_register(struct drm_device *dev)
+int
+nv20_graph_fini(struct drm_device *dev, int engine)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	if (dev_priv->engine.graph.registered)
-		return 0;
-
-	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
-	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
-	NVOBJ_CLASS(dev, 0x0039, GR); /* m2mf */
-	NVOBJ_CLASS(dev, 0x004a, GR); /* gdirect */
-	NVOBJ_CLASS(dev, 0x009f, GR); /* imageblit (nv12) */
-	NVOBJ_CLASS(dev, 0x008a, GR); /* ifc */
-	NVOBJ_CLASS(dev, 0x038a, GR); /* ifc (nv30) */
-	NVOBJ_CLASS(dev, 0x0089, GR); /* sifm */
-	NVOBJ_CLASS(dev, 0x0389, GR); /* sifm (nv30) */
-	NVOBJ_CLASS(dev, 0x0062, GR); /* surf2d */
-	NVOBJ_CLASS(dev, 0x0362, GR); /* surf2d (nv30) */
-	NVOBJ_CLASS(dev, 0x0043, GR); /* rop */
-	NVOBJ_CLASS(dev, 0x0012, GR); /* beta1 */
-	NVOBJ_CLASS(dev, 0x0072, GR); /* beta4 */
-	NVOBJ_CLASS(dev, 0x0019, GR); /* cliprect */
-	NVOBJ_CLASS(dev, 0x0044, GR); /* pattern */
-	NVOBJ_CLASS(dev, 0x039e, GR); /* swzsurf */
-
-	/* rankine */
-	if (0x00000003 & (1 << (dev_priv->chipset & 0x0f)))
-		NVOBJ_CLASS(dev, 0x0397, GR);
-	else
-	if (0x00000010 & (1 << (dev_priv->chipset & 0x0f)))
-		NVOBJ_CLASS(dev, 0x0697, GR);
-	else
-	if (0x000001e0 & (1 << (dev_priv->chipset & 0x0f)))
-		NVOBJ_CLASS(dev, 0x0497, GR);
-
-	/* nvsw */
-	NVOBJ_CLASS(dev, 0x506e, SW);
-	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
-
-	dev_priv->engine.graph.registered = true;
+	nv20_graph_unload_context(dev);
+	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0x00000000);
 	return 0;
 }
 
@@ -897,3 +701,135 @@ nv20_graph_isr(struct drm_device *dev)
 		}
 	}
 }
+
+static void
+nv20_graph_destroy(struct drm_device *dev, int engine)
+{
+	struct nv20_graph_engine *pgraph = nv_engine(dev, engine);
+
+	nouveau_irq_unregister(dev, 12);
+	nouveau_gpuobj_ref(NULL, &pgraph->ctxtab);
+
+	NVOBJ_ENGINE_DEL(dev, GR);
+	kfree(pgraph);
+}
+
+int
+nv20_graph_create(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nv20_graph_engine *pgraph;
+	int ret;
+
+	pgraph = kzalloc(sizeof(*pgraph), GFP_KERNEL);
+	if (!pgraph)
+		return -ENOMEM;
+
+	pgraph->base.destroy = nv20_graph_destroy;
+	pgraph->base.fini = nv20_graph_fini;
+	pgraph->base.context_new = nv20_graph_context_new;
+	pgraph->base.context_del = nv20_graph_context_del;
+	pgraph->base.object_new = nv04_graph_object_new;
+	pgraph->base.set_tile_region = nv20_graph_set_tile_region;
+
+	pgraph->grctx_user = 0x0028;
+	if (dev_priv->card_type == NV_20) {
+		pgraph->base.init = nv20_graph_init;
+		switch (dev_priv->chipset) {
+		case 0x20:
+			pgraph->grctx_init = nv20_graph_context_init;
+			pgraph->grctx_size = NV20_GRCTX_SIZE;
+			pgraph->grctx_user = 0x0000;
+			break;
+		case 0x25:
+		case 0x28:
+			pgraph->grctx_init = nv25_graph_context_init;
+			pgraph->grctx_size = NV25_GRCTX_SIZE;
+			break;
+		case 0x2a:
+			pgraph->grctx_init = nv2a_graph_context_init;
+			pgraph->grctx_size = NV2A_GRCTX_SIZE;
+			pgraph->grctx_user = 0x0000;
+			break;
+		default:
+			NV_ERROR(dev, "PGRAPH: unknown chipset\n");
+			return 0;
+		}
+	} else {
+		pgraph->base.init = nv30_graph_init;
+		switch (dev_priv->chipset) {
+		case 0x30:
+		case 0x31:
+			pgraph->grctx_init = nv30_31_graph_context_init;
+			pgraph->grctx_size = NV30_31_GRCTX_SIZE;
+			break;
+		case 0x34:
+			pgraph->grctx_init = nv34_graph_context_init;
+			pgraph->grctx_size = NV34_GRCTX_SIZE;
+			break;
+		case 0x35:
+		case 0x36:
+			pgraph->grctx_init = nv35_36_graph_context_init;
+			pgraph->grctx_size = NV35_36_GRCTX_SIZE;
+			break;
+		default:
+			NV_ERROR(dev, "PGRAPH: unknown chipset\n");
+			return 0;
+		}
+	}
+
+	/* Create Context Pointer Table */
+	ret = nouveau_gpuobj_new(dev, NULL, 32 * 4, 16, NVOBJ_FLAG_ZERO_ALLOC,
+				 &pgraph->ctxtab);
+	if (ret) {
+		kfree(pgraph);
+		return ret;
+	}
+
+	NVOBJ_ENGINE_ADD(dev, GR, &pgraph->base);
+	nouveau_irq_register(dev, 12, nv20_graph_isr);
+
+	/* nvsw */
+	NVOBJ_CLASS(dev, 0x506e, SW);
+	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
+
+	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
+	NVOBJ_CLASS(dev, 0x0039, GR); /* m2mf */
+	NVOBJ_CLASS(dev, 0x004a, GR); /* gdirect */
+	NVOBJ_CLASS(dev, 0x009f, GR); /* imageblit (nv12) */
+	NVOBJ_CLASS(dev, 0x008a, GR); /* ifc */
+	NVOBJ_CLASS(dev, 0x0089, GR); /* sifm */
+	NVOBJ_CLASS(dev, 0x0062, GR); /* surf2d */
+	NVOBJ_CLASS(dev, 0x0043, GR); /* rop */
+	NVOBJ_CLASS(dev, 0x0012, GR); /* beta1 */
+	NVOBJ_CLASS(dev, 0x0072, GR); /* beta4 */
+	NVOBJ_CLASS(dev, 0x0019, GR); /* cliprect */
+	NVOBJ_CLASS(dev, 0x0044, GR); /* pattern */
+	if (dev_priv->card_type == NV_20) {
+		NVOBJ_CLASS(dev, 0x009e, GR); /* swzsurf */
+		NVOBJ_CLASS(dev, 0x0096, GR); /* celcius */
+
+		/* kelvin */
+		if (dev_priv->chipset < 0x25)
+			NVOBJ_CLASS(dev, 0x0097, GR);
+		else
+			NVOBJ_CLASS(dev, 0x0597, GR);
+	} else {
+		NVOBJ_CLASS(dev, 0x038a, GR); /* ifc (nv30) */
+		NVOBJ_CLASS(dev, 0x0389, GR); /* sifm (nv30) */
+		NVOBJ_CLASS(dev, 0x0362, GR); /* surf2d (nv30) */
+		NVOBJ_CLASS(dev, 0x039e, GR); /* swzsurf */
+
+		/* rankine */
+		if (0x00000003 & (1 << (dev_priv->chipset & 0x0f)))
+			NVOBJ_CLASS(dev, 0x0397, GR);
+		else
+		if (0x00000010 & (1 << (dev_priv->chipset & 0x0f)))
+			NVOBJ_CLASS(dev, 0x0697, GR);
+		else
+		if (0x000001e0 & (1 << (dev_priv->chipset & 0x0f)))
+			NVOBJ_CLASS(dev, 0x0497, GR);
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nv40_fifo.c b/drivers/gpu/drm/nouveau/nv40_fifo.c
index 49b9a35a9cd6..68cb2d991c88 100644
--- a/drivers/gpu/drm/nouveau/nv40_fifo.c
+++ b/drivers/gpu/drm/nouveau/nv40_fifo.c
@@ -115,6 +115,7 @@ nv40_fifo_do_load_context(struct drm_device *dev, int chid)
 	nv_wr32(dev, 0x32e8, nv_ri32(dev, fc + 68));
 	nv_wr32(dev, 0x2088, nv_ri32(dev, fc + 76));
 	nv_wr32(dev, 0x3300, nv_ri32(dev, fc + 80));
+	nv_wr32(dev, 0x330c, nv_ri32(dev, fc + 84));
 
 	nv_wr32(dev, NV03_PFIFO_CACHE1_GET, 0);
 	nv_wr32(dev, NV03_PFIFO_CACHE1_PUT, 0);
@@ -186,6 +187,7 @@ nv40_fifo_unload_context(struct drm_device *dev)
 	tmp |= (nv_rd32(dev, NV04_PFIFO_CACHE1_PUT) << 16);
 	nv_wi32(dev, fc + 72, tmp);
 #endif
+	nv_wi32(dev, fc + 84, nv_rd32(dev, 0x330c));
 
 	nv40_fifo_do_load_context(dev, pfifo->channels - 1);
 	nv_wr32(dev, NV03_PFIFO_CACHE1_PUSH1,
diff --git a/drivers/gpu/drm/nouveau/nv40_graph.c b/drivers/gpu/drm/nouveau/nv40_graph.c
index fceb44c0ec74..5beb01b8ace1 100644
--- a/drivers/gpu/drm/nouveau/nv40_graph.c
+++ b/drivers/gpu/drm/nouveau/nv40_graph.c
@@ -28,14 +28,18 @@
 #include "drm.h"
 #include "nouveau_drv.h"
 #include "nouveau_grctx.h"
+#include "nouveau_ramht.h"
 
-static int nv40_graph_register(struct drm_device *);
-static void nv40_graph_isr(struct drm_device *);
+struct nv40_graph_engine {
+	struct nouveau_exec_engine base;
+	u32 grctx_size;
+};
 
-struct nouveau_channel *
+static struct nouveau_channel *
 nv40_graph_channel(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *grctx;
 	uint32_t inst;
 	int i;
 
@@ -45,74 +49,17 @@ nv40_graph_channel(struct drm_device *dev)
 	inst = (inst & NV40_PGRAPH_CTXCTL_CUR_INSTANCE) << 4;
 
 	for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
-		struct nouveau_channel *chan = dev_priv->channels.ptr[i];
+		if (!dev_priv->channels.ptr[i])
+			continue;
 
-		if (chan && chan->ramin_grctx &&
-		    chan->ramin_grctx->pinst == inst)
-			return chan;
+		grctx = dev_priv->channels.ptr[i]->engctx[NVOBJ_ENGINE_GR];
+		if (grctx && grctx->pinst == inst)
+			return dev_priv->channels.ptr[i];
 	}
 
 	return NULL;
 }
 
-int
-nv40_graph_create_context(struct nouveau_channel *chan)
-{
-	struct drm_device *dev = chan->dev;
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct nouveau_grctx ctx = {};
-	unsigned long flags;
-	int ret;
-
-	ret = nouveau_gpuobj_new(dev, chan, pgraph->grctx_size, 16,
-				 NVOBJ_FLAG_ZERO_ALLOC, &chan->ramin_grctx);
-	if (ret)
-		return ret;
-
-	/* Initialise default context values */
-	ctx.dev = chan->dev;
-	ctx.mode = NOUVEAU_GRCTX_VALS;
-	ctx.data = chan->ramin_grctx;
-	nv40_grctx_init(&ctx);
-
-	nv_wo32(chan->ramin_grctx, 0, chan->ramin_grctx->pinst);
-
-	/* init grctx pointer in ramfc, and on PFIFO if channel is
-	 * already active there
-	 */
-	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
-	nv_wo32(chan->ramfc, 0x38, chan->ramin_grctx->pinst >> 4);
-	nv_mask(dev, 0x002500, 0x00000001, 0x00000000);
-	if ((nv_rd32(dev, 0x003204) & 0x0000001f) == chan->id)
-		nv_wr32(dev, 0x0032e0, chan->ramin_grctx->pinst >> 4);
-	nv_mask(dev, 0x002500, 0x00000001, 0x00000001);
-	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
-	return 0;
-}
-
-void
-nv40_graph_destroy_context(struct nouveau_channel *chan)
-{
-	struct drm_device *dev = chan->dev;
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	unsigned long flags;
-
-	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
-	pgraph->fifo_access(dev, false);
-
-	/* Unload the context if it's the currently active one */
-	if (pgraph->channel(dev) == chan)
-		pgraph->unload_context(dev);
-
-	pgraph->fifo_access(dev, true);
-	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
-
-	/* Free the context resources */
-	nouveau_gpuobj_ref(NULL, &chan->ramin_grctx);
-}
-
 static int
 nv40_graph_transfer_context(struct drm_device *dev, uint32_t inst, int save)
 {
@@ -154,57 +101,115 @@ nv40_graph_transfer_context(struct drm_device *dev, uint32_t inst, int save)
 	return 0;
 }
 
-/* Restore the context for a specific channel into PGRAPH */
-int
-nv40_graph_load_context(struct nouveau_channel *chan)
+static int
+nv40_graph_unload_context(struct drm_device *dev)
 {
-	struct drm_device *dev = chan->dev;
 	uint32_t inst;
 	int ret;
 
-	if (!chan->ramin_grctx)
-		return -EINVAL;
-	inst = chan->ramin_grctx->pinst >> 4;
+	inst = nv_rd32(dev, NV40_PGRAPH_CTXCTL_CUR);
+	if (!(inst & NV40_PGRAPH_CTXCTL_CUR_LOADED))
+		return 0;
+	inst &= NV40_PGRAPH_CTXCTL_CUR_INSTANCE;
+
+	ret = nv40_graph_transfer_context(dev, inst, 1);
+
+	nv_wr32(dev, NV40_PGRAPH_CTXCTL_CUR, inst);
+	return ret;
+}
+
+static int
+nv40_graph_context_new(struct nouveau_channel *chan, int engine)
+{
+	struct nv40_graph_engine *pgraph = nv_engine(chan->dev, engine);
+	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *grctx = NULL;
+	struct nouveau_grctx ctx = {};
+	unsigned long flags;
+	int ret;
 
-	ret = nv40_graph_transfer_context(dev, inst, 0);
+	ret = nouveau_gpuobj_new(dev, NULL, pgraph->grctx_size, 16,
+				 NVOBJ_FLAG_ZERO_ALLOC, &grctx);
 	if (ret)
 		return ret;
 
-	/* 0x40032C, no idea of it's exact function.  Could simply be a
-	 * record of the currently active PGRAPH context.  It's currently
-	 * unknown as to what bit 24 does.  The nv ddx has it set, so we will
-	 * set it here too.
-	 */
-	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_POINTER, inst);
-	nv_wr32(dev, NV40_PGRAPH_CTXCTL_CUR,
-		 (inst & NV40_PGRAPH_CTXCTL_CUR_INSTANCE) |
-		  NV40_PGRAPH_CTXCTL_CUR_LOADED);
-	/* 0x32E0 records the instance address of the active FIFO's PGRAPH
-	 * context.  If at any time this doesn't match 0x40032C, you will
-	 * receive PGRAPH_INTR_CONTEXT_SWITCH
+	/* Initialise default context values */
+	ctx.dev = chan->dev;
+	ctx.mode = NOUVEAU_GRCTX_VALS;
+	ctx.data = grctx;
+	nv40_grctx_init(&ctx);
+
+	nv_wo32(grctx, 0, grctx->vinst);
+
+	/* init grctx pointer in ramfc, and on PFIFO if channel is
+	 * already active there
 	 */
-	nv_wr32(dev, NV40_PFIFO_GRCTX_INSTANCE, inst);
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	nv_wo32(chan->ramfc, 0x38, grctx->vinst >> 4);
+	nv_mask(dev, 0x002500, 0x00000001, 0x00000000);
+	if ((nv_rd32(dev, 0x003204) & 0x0000001f) == chan->id)
+		nv_wr32(dev, 0x0032e0, grctx->vinst >> 4);
+	nv_mask(dev, 0x002500, 0x00000001, 0x00000001);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
+
+	chan->engctx[engine] = grctx;
 	return 0;
 }
 
-int
-nv40_graph_unload_context(struct drm_device *dev)
+static void
+nv40_graph_context_del(struct nouveau_channel *chan, int engine)
 {
-	uint32_t inst;
-	int ret;
+	struct nouveau_gpuobj *grctx = chan->engctx[engine];
+	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	unsigned long flags;
 
-	inst = nv_rd32(dev, NV40_PGRAPH_CTXCTL_CUR);
-	if (!(inst & NV40_PGRAPH_CTXCTL_CUR_LOADED))
-		return 0;
-	inst &= NV40_PGRAPH_CTXCTL_CUR_INSTANCE;
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	nv04_graph_fifo_access(dev, false);
 
-	ret = nv40_graph_transfer_context(dev, inst, 1);
+	/* Unload the context if it's the currently active one */
+	if (nv40_graph_channel(dev) == chan)
+		nv40_graph_unload_context(dev);
 
-	nv_wr32(dev, NV40_PGRAPH_CTXCTL_CUR, inst);
+	nv04_graph_fifo_access(dev, true);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
+
+	/* Free the context resources */
+	nouveau_gpuobj_ref(NULL, &grctx);
+	chan->engctx[engine] = NULL;
+}
+
+int
+nv40_graph_object_new(struct nouveau_channel *chan, int engine,
+		      u32 handle, u16 class)
+{
+	struct drm_device *dev = chan->dev;
+	struct nouveau_gpuobj *obj = NULL;
+	int ret;
+
+	ret = nouveau_gpuobj_new(dev, chan, 20, 16, NVOBJ_FLAG_ZERO_FREE, &obj);
+	if (ret)
+		return ret;
+	obj->engine = 1;
+	obj->class  = class;
+
+	nv_wo32(obj, 0x00, class);
+	nv_wo32(obj, 0x04, 0x00000000);
+#ifndef __BIG_ENDIAN
+	nv_wo32(obj, 0x08, 0x00000000);
+#else
+	nv_wo32(obj, 0x08, 0x01000000);
+#endif
+	nv_wo32(obj, 0x0c, 0x00000000);
+	nv_wo32(obj, 0x10, 0x00000000);
+
+	ret = nouveau_ramht_insert(chan, handle, obj);
+	nouveau_gpuobj_ref(NULL, &obj);
 	return ret;
 }
 
-void
+static void
 nv40_graph_set_tile_region(struct drm_device *dev, int i)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
@@ -257,14 +262,14 @@ nv40_graph_set_tile_region(struct drm_device *dev, int i)
  * C51		0x4e
  */
 int
-nv40_graph_init(struct drm_device *dev)
+nv40_graph_init(struct drm_device *dev, int engine)
 {
-	struct drm_nouveau_private *dev_priv =
-		(struct drm_nouveau_private *)dev->dev_private;
+	struct nv40_graph_engine *pgraph = nv_engine(dev, engine);
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
 	struct nouveau_grctx ctx = {};
 	uint32_t vramsz, *cp;
-	int ret, i, j;
+	int i, j;
 
 	nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) &
 			~NV_PMC_ENABLE_PGRAPH);
@@ -280,7 +285,7 @@ nv40_graph_init(struct drm_device *dev)
 	ctx.data = cp;
 	ctx.ctxprog_max = 256;
 	nv40_grctx_init(&ctx);
-	dev_priv->engine.graph.grctx_size = ctx.ctxvals_pos * 4;
+	pgraph->grctx_size = ctx.ctxvals_pos * 4;
 
 	nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_INDEX, 0);
 	for (i = 0; i < ctx.ctxprog_len; i++)
@@ -288,14 +293,9 @@ nv40_graph_init(struct drm_device *dev)
 
 	kfree(cp);
 
-	ret = nv40_graph_register(dev);
-	if (ret)
-		return ret;
-
 	/* No context present currently */
 	nv_wr32(dev, NV40_PGRAPH_CTXCTL_CUR, 0x00000000);
 
-	nouveau_irq_register(dev, 12, nv40_graph_isr);
 	nv_wr32(dev, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
 	nv_wr32(dev, NV40_PGRAPH_INTR_EN, 0xFFFFFFFF);
 
@@ -428,47 +428,10 @@ nv40_graph_init(struct drm_device *dev)
 	return 0;
 }
 
-void nv40_graph_takedown(struct drm_device *dev)
-{
-	nouveau_irq_unregister(dev, 12);
-}
-
 static int
-nv40_graph_register(struct drm_device *dev)
+nv40_graph_fini(struct drm_device *dev, int engine)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	if (dev_priv->engine.graph.registered)
-		return 0;
-
-	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
-	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
-	NVOBJ_CLASS(dev, 0x0039, GR); /* m2mf */
-	NVOBJ_CLASS(dev, 0x004a, GR); /* gdirect */
-	NVOBJ_CLASS(dev, 0x009f, GR); /* imageblit (nv12) */
-	NVOBJ_CLASS(dev, 0x008a, GR); /* ifc */
-	NVOBJ_CLASS(dev, 0x0089, GR); /* sifm */
-	NVOBJ_CLASS(dev, 0x3089, GR); /* sifm (nv40) */
-	NVOBJ_CLASS(dev, 0x0062, GR); /* surf2d */
-	NVOBJ_CLASS(dev, 0x3062, GR); /* surf2d (nv40) */
-	NVOBJ_CLASS(dev, 0x0043, GR); /* rop */
-	NVOBJ_CLASS(dev, 0x0012, GR); /* beta1 */
-	NVOBJ_CLASS(dev, 0x0072, GR); /* beta4 */
-	NVOBJ_CLASS(dev, 0x0019, GR); /* cliprect */
-	NVOBJ_CLASS(dev, 0x0044, GR); /* pattern */
-	NVOBJ_CLASS(dev, 0x309e, GR); /* swzsurf */
-
-	/* curie */
-	if (nv44_graph_class(dev))
-		NVOBJ_CLASS(dev, 0x4497, GR);
-	else
-		NVOBJ_CLASS(dev, 0x4097, GR);
-
-	/* nvsw */
-	NVOBJ_CLASS(dev, 0x506e, SW);
-	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
-
-	dev_priv->engine.graph.registered = true;
+	nv40_graph_unload_context(dev);
 	return 0;
 }
 
@@ -476,17 +439,17 @@ static int
 nv40_graph_isr_chid(struct drm_device *dev, u32 inst)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_channel *chan;
+	struct nouveau_gpuobj *grctx;
 	unsigned long flags;
 	int i;
 
 	spin_lock_irqsave(&dev_priv->channels.lock, flags);
 	for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
-		chan = dev_priv->channels.ptr[i];
-		if (!chan || !chan->ramin_grctx)
+		if (!dev_priv->channels.ptr[i])
 			continue;
+		grctx = dev_priv->channels.ptr[i]->engctx[NVOBJ_ENGINE_GR];
 
-		if (inst == chan->ramin_grctx->pinst)
+		if (grctx && grctx->pinst == inst)
 			break;
 	}
 	spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
@@ -537,3 +500,63 @@ nv40_graph_isr(struct drm_device *dev)
 		}
 	}
 }
+
+static void
+nv40_graph_destroy(struct drm_device *dev, int engine)
+{
+	struct nv40_graph_engine *pgraph = nv_engine(dev, engine);
+
+	nouveau_irq_unregister(dev, 12);
+
+	NVOBJ_ENGINE_DEL(dev, GR);
+	kfree(pgraph);
+}
+
+int
+nv40_graph_create(struct drm_device *dev)
+{
+	struct nv40_graph_engine *pgraph;
+
+	pgraph = kzalloc(sizeof(*pgraph), GFP_KERNEL);
+	if (!pgraph)
+		return -ENOMEM;
+
+	pgraph->base.destroy = nv40_graph_destroy;
+	pgraph->base.init = nv40_graph_init;
+	pgraph->base.fini = nv40_graph_fini;
+	pgraph->base.context_new = nv40_graph_context_new;
+	pgraph->base.context_del = nv40_graph_context_del;
+	pgraph->base.object_new = nv40_graph_object_new;
+	pgraph->base.set_tile_region = nv40_graph_set_tile_region;
+
+	NVOBJ_ENGINE_ADD(dev, GR, &pgraph->base);
+	nouveau_irq_register(dev, 12, nv40_graph_isr);
+
+	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
+	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
+	NVOBJ_CLASS(dev, 0x0039, GR); /* m2mf */
+	NVOBJ_CLASS(dev, 0x004a, GR); /* gdirect */
+	NVOBJ_CLASS(dev, 0x009f, GR); /* imageblit (nv12) */
+	NVOBJ_CLASS(dev, 0x008a, GR); /* ifc */
+	NVOBJ_CLASS(dev, 0x0089, GR); /* sifm */
+	NVOBJ_CLASS(dev, 0x3089, GR); /* sifm (nv40) */
+	NVOBJ_CLASS(dev, 0x0062, GR); /* surf2d */
+	NVOBJ_CLASS(dev, 0x3062, GR); /* surf2d (nv40) */
+	NVOBJ_CLASS(dev, 0x0043, GR); /* rop */
+	NVOBJ_CLASS(dev, 0x0012, GR); /* beta1 */
+	NVOBJ_CLASS(dev, 0x0072, GR); /* beta4 */
+	NVOBJ_CLASS(dev, 0x0019, GR); /* cliprect */
+	NVOBJ_CLASS(dev, 0x0044, GR); /* pattern */
+	NVOBJ_CLASS(dev, 0x309e, GR); /* swzsurf */
+
+	/* curie */
+	if (nv44_graph_class(dev))
+		NVOBJ_CLASS(dev, 0x4497, GR);
+	else
+		NVOBJ_CLASS(dev, 0x4097, GR);
+
+	/* nvsw */
+	NVOBJ_CLASS(dev, 0x506e, SW);
+	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nv40_mpeg.c b/drivers/gpu/drm/nouveau/nv40_mpeg.c
new file mode 100644
index 000000000000..6d2af292a2e3
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nv40_mpeg.c
@@ -0,0 +1,311 @@
+/*
+ * 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
+ */
+
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_ramht.h"
+
+struct nv40_mpeg_engine {
+	struct nouveau_exec_engine base;
+};
+
+static int
+nv40_mpeg_context_new(struct nouveau_channel *chan, int engine)
+{
+	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *ctx = NULL;
+	unsigned long flags;
+	int ret;
+
+	NV_DEBUG(dev, "ch%d\n", chan->id);
+
+	ret = nouveau_gpuobj_new(dev, NULL, 264 * 4, 16, NVOBJ_FLAG_ZERO_ALLOC |
+				 NVOBJ_FLAG_ZERO_FREE, &ctx);
+	if (ret)
+		return ret;
+
+	nv_wo32(ctx, 0x78, 0x02001ec1);
+
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	nv_mask(dev, 0x002500, 0x00000001, 0x00000000);
+	if ((nv_rd32(dev, 0x003204) & 0x1f) == chan->id)
+		nv_wr32(dev, 0x00330c, ctx->pinst >> 4);
+	nv_wo32(chan->ramfc, 0x54, ctx->pinst >> 4);
+	nv_mask(dev, 0x002500, 0x00000001, 0x00000001);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
+
+	chan->engctx[engine] = ctx;
+	return 0;
+}
+
+static void
+nv40_mpeg_context_del(struct nouveau_channel *chan, int engine)
+{
+	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
+	struct nouveau_gpuobj *ctx = chan->engctx[engine];
+	struct drm_device *dev = chan->dev;
+	unsigned long flags;
+	u32 inst = 0x80000000 | (ctx->pinst >> 4);
+
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	nv_mask(dev, 0x00b32c, 0x00000001, 0x00000000);
+	if (nv_rd32(dev, 0x00b318) == inst)
+		nv_mask(dev, 0x00b318, 0x80000000, 0x00000000);
+	nv_mask(dev, 0x00b32c, 0x00000001, 0x00000001);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
+
+	nouveau_gpuobj_ref(NULL, &ctx);
+	chan->engctx[engine] = NULL;
+}
+
+static int
+nv40_mpeg_object_new(struct nouveau_channel *chan, int engine,
+		      u32 handle, u16 class)
+{
+	struct drm_device *dev = chan->dev;
+	struct nouveau_gpuobj *obj = NULL;
+	int ret;
+
+	ret = nouveau_gpuobj_new(dev, chan, 20, 16, NVOBJ_FLAG_ZERO_ALLOC |
+				 NVOBJ_FLAG_ZERO_FREE, &obj);
+	if (ret)
+		return ret;
+	obj->engine = 2;
+	obj->class  = class;
+
+	nv_wo32(obj, 0x00, class);
+
+	ret = nouveau_ramht_insert(chan, handle, obj);
+	nouveau_gpuobj_ref(NULL, &obj);
+	return ret;
+}
+
+static int
+nv40_mpeg_init(struct drm_device *dev, int engine)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nv40_mpeg_engine *pmpeg = nv_engine(dev, engine);
+	int i;
+
+	/* VPE init */
+	nv_mask(dev, 0x000200, 0x00000002, 0x00000000);
+	nv_mask(dev, 0x000200, 0x00000002, 0x00000002);
+	nv_wr32(dev, 0x00b0e0, 0x00000020); /* nvidia: rd 0x01, wr 0x20 */
+	nv_wr32(dev, 0x00b0e8, 0x00000020); /* nvidia: rd 0x01, wr 0x20 */
+
+	for (i = 0; i < dev_priv->engine.fb.num_tiles; i++)
+		pmpeg->base.set_tile_region(dev, i);
+
+	/* PMPEG init */
+	nv_wr32(dev, 0x00b32c, 0x00000000);
+	nv_wr32(dev, 0x00b314, 0x00000100);
+	nv_wr32(dev, 0x00b220, 0x00000044);
+	nv_wr32(dev, 0x00b300, 0x02001ec1);
+	nv_mask(dev, 0x00b32c, 0x00000001, 0x00000001);
+
+	nv_wr32(dev, 0x00b100, 0xffffffff);
+	nv_wr32(dev, 0x00b140, 0xffffffff);
+
+	if (!nv_wait(dev, 0x00b200, 0x00000001, 0x00000000)) {
+		NV_ERROR(dev, "PMPEG init: 0x%08x\n", nv_rd32(dev, 0x00b200));
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+static int
+nv40_mpeg_fini(struct drm_device *dev, int engine)
+{
+	/*XXX: context save? */
+	nv_mask(dev, 0x00b32c, 0x00000001, 0x00000000);
+	nv_wr32(dev, 0x00b140, 0x00000000);
+	return 0;
+}
+
+static int
+nv40_mpeg_mthd_dma(struct nouveau_channel *chan, u32 class, u32 mthd, u32 data)
+{
+	struct drm_device *dev = chan->dev;
+	u32 inst = data << 4;
+	u32 dma0 = nv_ri32(dev, inst + 0);
+	u32 dma1 = nv_ri32(dev, inst + 4);
+	u32 dma2 = nv_ri32(dev, inst + 8);
+	u32 base = (dma2 & 0xfffff000) | (dma0 >> 20);
+	u32 size = dma1 + 1;
+
+	/* only allow linear DMA objects */
+	if (!(dma0 & 0x00002000))
+		return -EINVAL;
+
+	if (mthd == 0x0190) {
+		/* DMA_CMD */
+		nv_mask(dev, 0x00b300, 0x00030000, (dma0 & 0x00030000));
+		nv_wr32(dev, 0x00b334, base);
+		nv_wr32(dev, 0x00b324, size);
+	} else
+	if (mthd == 0x01a0) {
+		/* DMA_DATA */
+		nv_mask(dev, 0x00b300, 0x000c0000, (dma0 & 0x00030000) << 2);
+		nv_wr32(dev, 0x00b360, base);
+		nv_wr32(dev, 0x00b364, size);
+	} else {
+		/* DMA_IMAGE, VRAM only */
+		if (dma0 & 0x000c0000)
+			return -EINVAL;
+
+		nv_wr32(dev, 0x00b370, base);
+		nv_wr32(dev, 0x00b374, size);
+	}
+
+	return 0;
+}
+
+static int
+nv40_mpeg_isr_chid(struct drm_device *dev, u32 inst)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *ctx;
+	unsigned long flags;
+	int i;
+
+	spin_lock_irqsave(&dev_priv->channels.lock, flags);
+	for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
+		if (!dev_priv->channels.ptr[i])
+			continue;
+
+		ctx = dev_priv->channels.ptr[i]->engctx[NVOBJ_ENGINE_MPEG];
+		if (ctx && ctx->pinst == inst)
+			break;
+	}
+	spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
+	return i;
+}
+
+static void
+nv40_vpe_set_tile_region(struct drm_device *dev, int i)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i];
+
+	nv_wr32(dev, 0x00b008 + (i * 0x10), tile->pitch);
+	nv_wr32(dev, 0x00b004 + (i * 0x10), tile->limit);
+	nv_wr32(dev, 0x00b000 + (i * 0x10), tile->addr);
+}
+
+static void
+nv40_mpeg_isr(struct drm_device *dev)
+{
+	u32 inst = (nv_rd32(dev, 0x00b318) & 0x000fffff) << 4;
+	u32 chid = nv40_mpeg_isr_chid(dev, inst);
+	u32 stat = nv_rd32(dev, 0x00b100);
+	u32 type = nv_rd32(dev, 0x00b230);
+	u32 mthd = nv_rd32(dev, 0x00b234);
+	u32 data = nv_rd32(dev, 0x00b238);
+	u32 show = stat;
+
+	if (stat & 0x01000000) {
+		/* happens on initial binding of the object */
+		if (type == 0x00000020 && mthd == 0x0000) {
+			nv_mask(dev, 0x00b308, 0x00000000, 0x00000000);
+			show &= ~0x01000000;
+		}
+
+		if (type == 0x00000010) {
+			if (!nouveau_gpuobj_mthd_call2(dev, chid, 0x3174, mthd, data))
+				show &= ~0x01000000;
+		}
+	}
+
+	nv_wr32(dev, 0x00b100, stat);
+	nv_wr32(dev, 0x00b230, 0x00000001);
+
+	if (show && nouveau_ratelimit()) {
+		NV_INFO(dev, "PMPEG: Ch %d [0x%08x] 0x%08x 0x%08x 0x%08x 0x%08x\n",
+			chid, inst, stat, type, mthd, data);
+	}
+}
+
+static void
+nv40_vpe_isr(struct drm_device *dev)
+{
+	if (nv_rd32(dev, 0x00b100))
+		nv40_mpeg_isr(dev);
+
+	if (nv_rd32(dev, 0x00b800)) {
+		u32 stat = nv_rd32(dev, 0x00b800);
+		NV_INFO(dev, "PMSRCH: 0x%08x\n", stat);
+		nv_wr32(dev, 0xb800, stat);
+	}
+}
+
+static void
+nv40_mpeg_destroy(struct drm_device *dev, int engine)
+{
+	struct nv40_mpeg_engine *pmpeg = nv_engine(dev, engine);
+
+	nouveau_irq_unregister(dev, 0);
+
+	NVOBJ_ENGINE_DEL(dev, MPEG);
+	kfree(pmpeg);
+}
+
+int
+nv40_mpeg_create(struct drm_device *dev)
+{
+	struct nv40_mpeg_engine *pmpeg;
+
+	pmpeg = kzalloc(sizeof(*pmpeg), GFP_KERNEL);
+	if (!pmpeg)
+		return -ENOMEM;
+
+	pmpeg->base.destroy = nv40_mpeg_destroy;
+	pmpeg->base.init = nv40_mpeg_init;
+	pmpeg->base.fini = nv40_mpeg_fini;
+	pmpeg->base.context_new = nv40_mpeg_context_new;
+	pmpeg->base.context_del = nv40_mpeg_context_del;
+	pmpeg->base.object_new = nv40_mpeg_object_new;
+
+	/* ISR vector, PMC_ENABLE bit,  and TILE regs are shared between
+	 * all VPE engines, for this driver's purposes the PMPEG engine
+	 * will be treated as the "master" and handle the global VPE
+	 * bits too
+	 */
+	pmpeg->base.set_tile_region = nv40_vpe_set_tile_region;
+	nouveau_irq_register(dev, 0, nv40_vpe_isr);
+
+	NVOBJ_ENGINE_ADD(dev, MPEG, &pmpeg->base);
+	NVOBJ_CLASS(dev, 0x3174, MPEG);
+	NVOBJ_MTHD (dev, 0x3174, 0x0190, nv40_mpeg_mthd_dma);
+	NVOBJ_MTHD (dev, 0x3174, 0x01a0, nv40_mpeg_mthd_dma);
+	NVOBJ_MTHD (dev, 0x3174, 0x01b0, nv40_mpeg_mthd_dma);
+
+#if 0
+	NVOBJ_ENGINE_ADD(dev, ME, &pme->base);
+	NVOBJ_CLASS(dev, 0x4075, ME);
+#endif
+	return 0;
+
+}
diff --git a/drivers/gpu/drm/nouveau/nv50_calc.c b/drivers/gpu/drm/nouveau/nv50_calc.c
index de81151648f8..8cf63a8b30cd 100644
--- a/drivers/gpu/drm/nouveau/nv50_calc.c
+++ b/drivers/gpu/drm/nouveau/nv50_calc.c
@@ -23,7 +23,6 @@
  */
 
 #include "drmP.h"
-#include "drm_fixed.h"
 #include "nouveau_drv.h"
 #include "nouveau_hw.h"
 
@@ -47,45 +46,52 @@ nv50_calc_pll(struct drm_device *dev, struct pll_lims *pll, int clk,
 }
 
 int
-nv50_calc_pll2(struct drm_device *dev, struct pll_lims *pll, int clk,
-	       int *N, int *fN, int *M, int *P)
+nva3_calc_pll(struct drm_device *dev, struct pll_lims *pll, int clk,
+	      int *pN, int *pfN, int *pM, int *P)
 {
-	fixed20_12 fb_div, a, b;
-	u32 refclk = pll->refclk / 10;
-	u32 max_vco_freq = pll->vco1.maxfreq / 10;
-	u32 max_vco_inputfreq = pll->vco1.max_inputfreq / 10;
-	clk /= 10;
+	u32 best_err = ~0, err;
+	int M, lM, hM, N, fN;
 
-	*P = max_vco_freq / clk;
+	*P = pll->vco1.maxfreq / clk;
 	if (*P > pll->max_p)
 		*P = pll->max_p;
 	if (*P < pll->min_p)
 		*P = pll->min_p;
 
-	/* *M = floor((refclk + max_vco_inputfreq) / max_vco_inputfreq); */
-	a.full = dfixed_const(refclk + max_vco_inputfreq);
-	b.full = dfixed_const(max_vco_inputfreq);
-	a.full = dfixed_div(a, b);
-	a.full = dfixed_floor(a);
-	*M = dfixed_trunc(a);
+	lM = (pll->refclk + pll->vco1.max_inputfreq) / pll->vco1.max_inputfreq;
+	lM = max(lM, (int)pll->vco1.min_m);
+	hM = (pll->refclk + pll->vco1.min_inputfreq) / pll->vco1.min_inputfreq;
+	hM = min(hM, (int)pll->vco1.max_m);
 
-	/* fb_div = (vco * *M) / refclk; */
-	fb_div.full = dfixed_const(clk * *P);
-	fb_div.full = dfixed_mul(fb_div, a);
-	a.full = dfixed_const(refclk);
-	fb_div.full = dfixed_div(fb_div, a);
+	for (M = lM; M <= hM; M++) {
+		u32 tmp = clk * *P * M;
+		N  = tmp / pll->refclk;
+		fN = tmp % pll->refclk;
+		if (!pfN && fN >= pll->refclk / 2)
+			N++;
 
-	/* *N = floor(fb_div); */
-	a.full = dfixed_floor(fb_div);
-	*N = dfixed_trunc(fb_div);
+		if (N < pll->vco1.min_n)
+			continue;
+		if (N > pll->vco1.max_n)
+			break;
 
-	/* *fN = (fmod(fb_div, 1.0) * 8192) - 4096; */
-	b.full = dfixed_const(8192);
-	a.full = dfixed_mul(a, b);
-	fb_div.full = dfixed_mul(fb_div, b);
-	fb_div.full = fb_div.full - a.full;
-	*fN = dfixed_trunc(fb_div) - 4096;
-	*fN &= 0xffff;
+		err = abs(clk - (pll->refclk * N / M / *P));
+		if (err < best_err) {
+			best_err = err;
+			*pN = N;
+			*pM = M;
+		}
 
-	return clk;
+		if (pfN) {
+			*pfN = (((fN << 13) / pll->refclk) - 4096) & 0xffff;
+			return clk;
+		}
+	}
+
+	if (unlikely(best_err == ~0)) {
+		NV_ERROR(dev, "unable to find matching pll values\n");
+		return -EINVAL;
+	}
+
+	return pll->refclk * *pN / *pM / *P;
 }
diff --git a/drivers/gpu/drm/nouveau/nv50_crtc.c b/drivers/gpu/drm/nouveau/nv50_crtc.c
index a19ccaa025b3..ebabacf38da9 100644
--- a/drivers/gpu/drm/nouveau/nv50_crtc.c
+++ b/drivers/gpu/drm/nouveau/nv50_crtc.c
@@ -286,7 +286,7 @@ nv50_crtc_set_clock(struct drm_device *dev, int head, int pclk)
 		nv_wr32(dev, pll.reg + 8, reg2 | (P << 28) | (M2 << 16) | N2);
 	} else
 	if (dev_priv->chipset < NV_C0) {
-		ret = nv50_calc_pll2(dev, &pll, pclk, &N1, &N2, &M1, &P);
+		ret = nva3_calc_pll(dev, &pll, pclk, &N1, &N2, &M1, &P);
 		if (ret <= 0)
 			return 0;
 
@@ -298,7 +298,7 @@ nv50_crtc_set_clock(struct drm_device *dev, int head, int pclk)
 		nv_wr32(dev, pll.reg + 4, reg1 | (P << 16) | (M1 << 8) | N1);
 		nv_wr32(dev, pll.reg + 8, N2);
 	} else {
-		ret = nv50_calc_pll2(dev, &pll, pclk, &N1, &N2, &M1, &P);
+		ret = nva3_calc_pll(dev, &pll, pclk, &N1, &N2, &M1, &P);
 		if (ret <= 0)
 			return 0;
 
@@ -349,14 +349,14 @@ nv50_crtc_cursor_set(struct drm_crtc *crtc, struct drm_file *file_priv,
 	struct drm_gem_object *gem;
 	int ret = 0, i;
 
-	if (width != 64 || height != 64)
-		return -EINVAL;
-
 	if (!buffer_handle) {
 		nv_crtc->cursor.hide(nv_crtc, true);
 		return 0;
 	}
 
+	if (width != 64 || height != 64)
+		return -EINVAL;
+
 	gem = drm_gem_object_lookup(dev, file_priv, buffer_handle);
 	if (!gem)
 		return -ENOENT;
@@ -532,8 +532,7 @@ nv50_crtc_do_mode_set_base(struct drm_crtc *crtc,
 	if (atomic) {
 		drm_fb = passed_fb;
 		fb = nouveau_framebuffer(passed_fb);
-	}
-	else {
+	} else {
 		/* If not atomic, we can go ahead and pin, and unpin the
 		 * old fb we were passed.
 		 */
diff --git a/drivers/gpu/drm/nouveau/nv50_display.c b/drivers/gpu/drm/nouveau/nv50_display.c
index 75a376cc342a..74a3f6872701 100644
--- a/drivers/gpu/drm/nouveau/nv50_display.c
+++ b/drivers/gpu/drm/nouveau/nv50_display.c
@@ -517,13 +517,25 @@ nv50_display_script_select(struct drm_device *dev, struct dcb_entry *dcb,
 			if (bios->fp.if_is_24bit)
 				script |= 0x0200;
 		} else {
+			/* determine number of lvds links */
+			if (nv_connector && nv_connector->edid &&
+			    nv_connector->dcb->type == DCB_CONNECTOR_LVDS_SPWG) {
+				/* http://www.spwg.org */
+				if (((u8 *)nv_connector->edid)[121] == 2)
+					script |= 0x0100;
+			} else
 			if (pxclk >= bios->fp.duallink_transition_clk) {
 				script |= 0x0100;
+			}
+
+			/* determine panel depth */
+			if (script & 0x0100) {
 				if (bios->fp.strapless_is_24bit & 2)
 					script |= 0x0200;
-			} else
-			if (bios->fp.strapless_is_24bit & 1)
-				script |= 0x0200;
+			} else {
+				if (bios->fp.strapless_is_24bit & 1)
+					script |= 0x0200;
+			}
 
 			if (nv_connector && nv_connector->edid &&
 			    (nv_connector->edid->revision >= 4) &&
diff --git a/drivers/gpu/drm/nouveau/nv50_graph.c b/drivers/gpu/drm/nouveau/nv50_graph.c
index b02a5b1e7d37..e25cbb46789a 100644
--- a/drivers/gpu/drm/nouveau/nv50_graph.c
+++ b/drivers/gpu/drm/nouveau/nv50_graph.c
@@ -31,10 +31,95 @@
 #include "nouveau_grctx.h"
 #include "nouveau_dma.h"
 #include "nouveau_vm.h"
+#include "nouveau_ramht.h"
 #include "nv50_evo.h"
 
-static int  nv50_graph_register(struct drm_device *);
-static void nv50_graph_isr(struct drm_device *);
+struct nv50_graph_engine {
+	struct nouveau_exec_engine base;
+	u32 ctxprog[512];
+	u32 ctxprog_size;
+	u32 grctx_size;
+};
+
+static void
+nv50_graph_fifo_access(struct drm_device *dev, bool enabled)
+{
+	const uint32_t mask = 0x00010001;
+
+	if (enabled)
+		nv_wr32(dev, 0x400500, nv_rd32(dev, 0x400500) | mask);
+	else
+		nv_wr32(dev, 0x400500, nv_rd32(dev, 0x400500) & ~mask);
+}
+
+static struct nouveau_channel *
+nv50_graph_channel(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	uint32_t inst;
+	int i;
+
+	/* Be sure we're not in the middle of a context switch or bad things
+	 * will happen, such as unloading the wrong pgraph context.
+	 */
+	if (!nv_wait(dev, 0x400300, 0x00000001, 0x00000000))
+		NV_ERROR(dev, "Ctxprog is still running\n");
+
+	inst = nv_rd32(dev, NV50_PGRAPH_CTXCTL_CUR);
+	if (!(inst & NV50_PGRAPH_CTXCTL_CUR_LOADED))
+		return NULL;
+	inst = (inst & NV50_PGRAPH_CTXCTL_CUR_INSTANCE) << 12;
+
+	for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
+		struct nouveau_channel *chan = dev_priv->channels.ptr[i];
+
+		if (chan && chan->ramin && chan->ramin->vinst == inst)
+			return chan;
+	}
+
+	return NULL;
+}
+
+static int
+nv50_graph_do_load_context(struct drm_device *dev, uint32_t inst)
+{
+	uint32_t fifo = nv_rd32(dev, 0x400500);
+
+	nv_wr32(dev, 0x400500, fifo & ~1);
+	nv_wr32(dev, 0x400784, inst);
+	nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) | 0x40);
+	nv_wr32(dev, 0x400320, nv_rd32(dev, 0x400320) | 0x11);
+	nv_wr32(dev, 0x400040, 0xffffffff);
+	(void)nv_rd32(dev, 0x400040);
+	nv_wr32(dev, 0x400040, 0x00000000);
+	nv_wr32(dev, 0x400304, nv_rd32(dev, 0x400304) | 1);
+
+	if (nouveau_wait_for_idle(dev))
+		nv_wr32(dev, 0x40032c, inst | (1<<31));
+	nv_wr32(dev, 0x400500, fifo);
+
+	return 0;
+}
+
+static int
+nv50_graph_unload_context(struct drm_device *dev)
+{
+	uint32_t inst;
+
+	inst  = nv_rd32(dev, NV50_PGRAPH_CTXCTL_CUR);
+	if (!(inst & NV50_PGRAPH_CTXCTL_CUR_LOADED))
+		return 0;
+	inst &= NV50_PGRAPH_CTXCTL_CUR_INSTANCE;
+
+	nouveau_wait_for_idle(dev);
+	nv_wr32(dev, 0x400784, inst);
+	nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) | 0x20);
+	nv_wr32(dev, 0x400304, nv_rd32(dev, 0x400304) | 0x01);
+	nouveau_wait_for_idle(dev);
+
+	nv_wr32(dev, NV50_PGRAPH_CTXCTL_CUR, inst);
+	return 0;
+}
 
 static void
 nv50_graph_init_reset(struct drm_device *dev)
@@ -52,7 +137,6 @@ nv50_graph_init_intr(struct drm_device *dev)
 {
 	NV_DEBUG(dev, "\n");
 
-	nouveau_irq_register(dev, 12, nv50_graph_isr);
 	nv_wr32(dev, NV03_PGRAPH_INTR, 0xffffffff);
 	nv_wr32(dev, 0x400138, 0xffffffff);
 	nv_wr32(dev, NV40_PGRAPH_INTR_EN, 0xffffffff);
@@ -135,34 +219,14 @@ nv50_graph_init_zcull(struct drm_device *dev)
 static int
 nv50_graph_init_ctxctl(struct drm_device *dev)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_grctx ctx = {};
-	uint32_t *cp;
+	struct nv50_graph_engine *pgraph = nv_engine(dev, NVOBJ_ENGINE_GR);
 	int i;
 
 	NV_DEBUG(dev, "\n");
 
-	cp = kmalloc(512 * 4, GFP_KERNEL);
-	if (!cp) {
-		NV_ERROR(dev, "failed to allocate ctxprog\n");
-		dev_priv->engine.graph.accel_blocked = true;
-		return 0;
-	}
-
-	ctx.dev = dev;
-	ctx.mode = NOUVEAU_GRCTX_PROG;
-	ctx.data = cp;
-	ctx.ctxprog_max = 512;
-	if (!nv50_grctx_init(&ctx)) {
-		dev_priv->engine.graph.grctx_size = ctx.ctxvals_pos * 4;
-
-		nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_INDEX, 0);
-		for (i = 0; i < ctx.ctxprog_len; i++)
-			nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_DATA, cp[i]);
-	} else {
-		dev_priv->engine.graph.accel_blocked = true;
-	}
-	kfree(cp);
+	nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_INDEX, 0);
+	for (i = 0; i < pgraph->ctxprog_size; i++)
+		nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_DATA, pgraph->ctxprog[i]);
 
 	nv_wr32(dev, 0x40008c, 0x00000004); /* HW_CTX_SWITCH_ENABLED */
 	nv_wr32(dev, 0x400320, 4);
@@ -171,8 +235,8 @@ nv50_graph_init_ctxctl(struct drm_device *dev)
 	return 0;
 }
 
-int
-nv50_graph_init(struct drm_device *dev)
+static int
+nv50_graph_init(struct drm_device *dev, int engine)
 {
 	int ret;
 
@@ -186,105 +250,66 @@ nv50_graph_init(struct drm_device *dev)
 	if (ret)
 		return ret;
 
-	ret = nv50_graph_register(dev);
-	if (ret)
-		return ret;
 	nv50_graph_init_intr(dev);
 	return 0;
 }
 
-void
-nv50_graph_takedown(struct drm_device *dev)
+static int
+nv50_graph_fini(struct drm_device *dev, int engine)
 {
 	NV_DEBUG(dev, "\n");
+	nv50_graph_unload_context(dev);
 	nv_wr32(dev, 0x40013c, 0x00000000);
-	nouveau_irq_unregister(dev, 12);
-}
-
-void
-nv50_graph_fifo_access(struct drm_device *dev, bool enabled)
-{
-	const uint32_t mask = 0x00010001;
-
-	if (enabled)
-		nv_wr32(dev, 0x400500, nv_rd32(dev, 0x400500) | mask);
-	else
-		nv_wr32(dev, 0x400500, nv_rd32(dev, 0x400500) & ~mask);
-}
-
-struct nouveau_channel *
-nv50_graph_channel(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t inst;
-	int i;
-
-	/* Be sure we're not in the middle of a context switch or bad things
-	 * will happen, such as unloading the wrong pgraph context.
-	 */
-	if (!nv_wait(dev, 0x400300, 0x00000001, 0x00000000))
-		NV_ERROR(dev, "Ctxprog is still running\n");
-
-	inst = nv_rd32(dev, NV50_PGRAPH_CTXCTL_CUR);
-	if (!(inst & NV50_PGRAPH_CTXCTL_CUR_LOADED))
-		return NULL;
-	inst = (inst & NV50_PGRAPH_CTXCTL_CUR_INSTANCE) << 12;
-
-	for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
-		struct nouveau_channel *chan = dev_priv->channels.ptr[i];
-
-		if (chan && chan->ramin && chan->ramin->vinst == inst)
-			return chan;
-	}
-
-	return NULL;
+	return 0;
 }
 
-int
-nv50_graph_create_context(struct nouveau_channel *chan)
+static int
+nv50_graph_context_new(struct nouveau_channel *chan, int engine)
 {
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_gpuobj *ramin = chan->ramin;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
+	struct nouveau_gpuobj *grctx = NULL;
+	struct nv50_graph_engine *pgraph = nv_engine(dev, engine);
 	struct nouveau_grctx ctx = {};
 	int hdr, ret;
 
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
-	ret = nouveau_gpuobj_new(dev, chan, pgraph->grctx_size, 0,
+	ret = nouveau_gpuobj_new(dev, NULL, pgraph->grctx_size, 0,
 				 NVOBJ_FLAG_ZERO_ALLOC |
-				 NVOBJ_FLAG_ZERO_FREE, &chan->ramin_grctx);
+				 NVOBJ_FLAG_ZERO_FREE, &grctx);
 	if (ret)
 		return ret;
 
 	hdr = (dev_priv->chipset == 0x50) ? 0x200 : 0x20;
 	nv_wo32(ramin, hdr + 0x00, 0x00190002);
-	nv_wo32(ramin, hdr + 0x04, chan->ramin_grctx->vinst +
-				   pgraph->grctx_size - 1);
-	nv_wo32(ramin, hdr + 0x08, chan->ramin_grctx->vinst);
+	nv_wo32(ramin, hdr + 0x04, grctx->vinst + grctx->size - 1);
+	nv_wo32(ramin, hdr + 0x08, grctx->vinst);
 	nv_wo32(ramin, hdr + 0x0c, 0);
 	nv_wo32(ramin, hdr + 0x10, 0);
 	nv_wo32(ramin, hdr + 0x14, 0x00010000);
 
 	ctx.dev = chan->dev;
 	ctx.mode = NOUVEAU_GRCTX_VALS;
-	ctx.data = chan->ramin_grctx;
+	ctx.data = grctx;
 	nv50_grctx_init(&ctx);
 
-	nv_wo32(chan->ramin_grctx, 0x00000, chan->ramin->vinst >> 12);
+	nv_wo32(grctx, 0x00000, chan->ramin->vinst >> 12);
 
 	dev_priv->engine.instmem.flush(dev);
-	atomic_inc(&chan->vm->pgraph_refs);
+
+	atomic_inc(&chan->vm->engref[NVOBJ_ENGINE_GR]);
+	chan->engctx[NVOBJ_ENGINE_GR] = grctx;
 	return 0;
 }
 
-void
-nv50_graph_destroy_context(struct nouveau_channel *chan)
+static void
+nv50_graph_context_del(struct nouveau_channel *chan, int engine)
 {
+	struct nouveau_gpuobj *grctx = chan->engctx[engine];
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
 	int i, hdr = (dev_priv->chipset == 0x50) ? 0x200 : 0x20;
 	unsigned long flags;
@@ -296,72 +321,49 @@ nv50_graph_destroy_context(struct nouveau_channel *chan)
 
 	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
 	pfifo->reassign(dev, false);
-	pgraph->fifo_access(dev, false);
+	nv50_graph_fifo_access(dev, false);
 
-	if (pgraph->channel(dev) == chan)
-		pgraph->unload_context(dev);
+	if (nv50_graph_channel(dev) == chan)
+		nv50_graph_unload_context(dev);
 
 	for (i = hdr; i < hdr + 24; i += 4)
 		nv_wo32(chan->ramin, i, 0);
 	dev_priv->engine.instmem.flush(dev);
 
-	pgraph->fifo_access(dev, true);
+	nv50_graph_fifo_access(dev, true);
 	pfifo->reassign(dev, true);
 	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
 
-	nouveau_gpuobj_ref(NULL, &chan->ramin_grctx);
+	nouveau_gpuobj_ref(NULL, &grctx);
 
-	atomic_dec(&chan->vm->pgraph_refs);
+	atomic_dec(&chan->vm->engref[engine]);
+	chan->engctx[engine] = NULL;
 }
 
 static int
-nv50_graph_do_load_context(struct drm_device *dev, uint32_t inst)
-{
-	uint32_t fifo = nv_rd32(dev, 0x400500);
-
-	nv_wr32(dev, 0x400500, fifo & ~1);
-	nv_wr32(dev, 0x400784, inst);
-	nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) | 0x40);
-	nv_wr32(dev, 0x400320, nv_rd32(dev, 0x400320) | 0x11);
-	nv_wr32(dev, 0x400040, 0xffffffff);
-	(void)nv_rd32(dev, 0x400040);
-	nv_wr32(dev, 0x400040, 0x00000000);
-	nv_wr32(dev, 0x400304, nv_rd32(dev, 0x400304) | 1);
-
-	if (nouveau_wait_for_idle(dev))
-		nv_wr32(dev, 0x40032c, inst | (1<<31));
-	nv_wr32(dev, 0x400500, fifo);
-
-	return 0;
-}
-
-int
-nv50_graph_load_context(struct nouveau_channel *chan)
-{
-	uint32_t inst = chan->ramin->vinst >> 12;
-
-	NV_DEBUG(chan->dev, "ch%d\n", chan->id);
-	return nv50_graph_do_load_context(chan->dev, inst);
-}
-
-int
-nv50_graph_unload_context(struct drm_device *dev)
+nv50_graph_object_new(struct nouveau_channel *chan, int engine,
+		      u32 handle, u16 class)
 {
-	uint32_t inst;
+	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *obj = NULL;
+	int ret;
 
-	inst  = nv_rd32(dev, NV50_PGRAPH_CTXCTL_CUR);
-	if (!(inst & NV50_PGRAPH_CTXCTL_CUR_LOADED))
-		return 0;
-	inst &= NV50_PGRAPH_CTXCTL_CUR_INSTANCE;
+	ret = nouveau_gpuobj_new(dev, chan, 16, 16, NVOBJ_FLAG_ZERO_FREE, &obj);
+	if (ret)
+		return ret;
+	obj->engine = 1;
+	obj->class  = class;
 
-	nouveau_wait_for_idle(dev);
-	nv_wr32(dev, 0x400784, inst);
-	nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) | 0x20);
-	nv_wr32(dev, 0x400304, nv_rd32(dev, 0x400304) | 0x01);
-	nouveau_wait_for_idle(dev);
+	nv_wo32(obj, 0x00, class);
+	nv_wo32(obj, 0x04, 0x00000000);
+	nv_wo32(obj, 0x08, 0x00000000);
+	nv_wo32(obj, 0x0c, 0x00000000);
+	dev_priv->engine.instmem.flush(dev);
 
-	nv_wr32(dev, NV50_PGRAPH_CTXCTL_CUR, inst);
-	return 0;
+	ret = nouveau_ramht_insert(chan, handle, obj);
+	nouveau_gpuobj_ref(NULL, &obj);
+	return ret;
 }
 
 static void
@@ -442,68 +444,15 @@ nv50_graph_nvsw_mthd_page_flip(struct nouveau_channel *chan,
 	return 0;
 }
 
-static int
-nv50_graph_register(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	if (dev_priv->engine.graph.registered)
-		return 0;
-
-	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
-	NVOBJ_MTHD (dev, 0x506e, 0x018c, nv50_graph_nvsw_dma_vblsem);
-	NVOBJ_MTHD (dev, 0x506e, 0x0400, nv50_graph_nvsw_vblsem_offset);
-	NVOBJ_MTHD (dev, 0x506e, 0x0404, nv50_graph_nvsw_vblsem_release_val);
-	NVOBJ_MTHD (dev, 0x506e, 0x0408, nv50_graph_nvsw_vblsem_release);
-	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv50_graph_nvsw_mthd_page_flip);
-
-	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
-	NVOBJ_CLASS(dev, 0x5039, GR); /* m2mf */
-	NVOBJ_CLASS(dev, 0x502d, GR); /* 2d */
-
-	/* tesla */
-	if (dev_priv->chipset == 0x50)
-		NVOBJ_CLASS(dev, 0x5097, GR); /* tesla (nv50) */
-	else
-	if (dev_priv->chipset < 0xa0)
-		NVOBJ_CLASS(dev, 0x8297, GR); /* tesla (nv8x/nv9x) */
-	else {
-		switch (dev_priv->chipset) {
-		case 0xa0:
-		case 0xaa:
-		case 0xac:
-			NVOBJ_CLASS(dev, 0x8397, GR);
-			break;
-		case 0xa3:
-		case 0xa5:
-		case 0xa8:
-			NVOBJ_CLASS(dev, 0x8597, GR);
-			break;
-		case 0xaf:
-			NVOBJ_CLASS(dev, 0x8697, GR);
-			break;
-		}
-	}
-
-	/* compute */
-	NVOBJ_CLASS(dev, 0x50c0, GR);
-	if (dev_priv->chipset  > 0xa0 &&
-	    dev_priv->chipset != 0xaa &&
-	    dev_priv->chipset != 0xac)
-		NVOBJ_CLASS(dev, 0x85c0, GR);
-
-	dev_priv->engine.graph.registered = true;
-	return 0;
-}
 
-void
-nv50_graph_tlb_flush(struct drm_device *dev)
+static void
+nv50_graph_tlb_flush(struct drm_device *dev, int engine)
 {
 	nv50_vm_flush_engine(dev, 0);
 }
 
-void
-nv84_graph_tlb_flush(struct drm_device *dev)
+static void
+nv84_graph_tlb_flush(struct drm_device *dev, int engine)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_timer_engine *ptimer = &dev_priv->engine.timer;
@@ -548,8 +497,7 @@ nv84_graph_tlb_flush(struct drm_device *dev)
 	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
 }
 
-static struct nouveau_enum nv50_mp_exec_error_names[] =
-{
+static struct nouveau_enum nv50_mp_exec_error_names[] = {
 	{ 3, "STACK_UNDERFLOW", NULL },
 	{ 4, "QUADON_ACTIVE", NULL },
 	{ 8, "TIMEOUT", NULL },
@@ -663,7 +611,7 @@ nv50_pgraph_mp_trap(struct drm_device *dev, int tpid, int display)
 			nv_rd32(dev, addr + 0x20);
 			pc = nv_rd32(dev, addr + 0x24);
 			oplow = nv_rd32(dev, addr + 0x70);
-			ophigh= nv_rd32(dev, addr + 0x74);
+			ophigh = nv_rd32(dev, addr + 0x74);
 			NV_INFO(dev, "PGRAPH_TRAP_MP_EXEC - "
 					"TP %d MP %d: ", tpid, i);
 			nouveau_enum_print(nv50_mp_exec_error_names, status);
@@ -991,7 +939,7 @@ nv50_pgraph_trap_handler(struct drm_device *dev, u32 display, u64 inst, u32 chid
 	return 1;
 }
 
-static int
+int
 nv50_graph_isr_chid(struct drm_device *dev, u64 inst)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
@@ -1073,3 +1021,101 @@ nv50_graph_isr(struct drm_device *dev)
 	if (nv_rd32(dev, 0x400824) & (1 << 31))
 		nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) & ~(1 << 31));
 }
+
+static void
+nv50_graph_destroy(struct drm_device *dev, int engine)
+{
+	struct nv50_graph_engine *pgraph = nv_engine(dev, engine);
+
+	NVOBJ_ENGINE_DEL(dev, GR);
+
+	nouveau_irq_unregister(dev, 12);
+	kfree(pgraph);
+}
+
+int
+nv50_graph_create(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nv50_graph_engine *pgraph;
+	struct nouveau_grctx ctx = {};
+	int ret;
+
+	pgraph = kzalloc(sizeof(*pgraph),GFP_KERNEL);
+	if (!pgraph)
+		return -ENOMEM;
+
+	ctx.dev = dev;
+	ctx.mode = NOUVEAU_GRCTX_PROG;
+	ctx.data = pgraph->ctxprog;
+	ctx.ctxprog_max = ARRAY_SIZE(pgraph->ctxprog);
+
+	ret = nv50_grctx_init(&ctx);
+	if (ret) {
+		NV_ERROR(dev, "PGRAPH: ctxprog build failed\n");
+		kfree(pgraph);
+		return 0;
+	}
+
+	pgraph->grctx_size = ctx.ctxvals_pos * 4;
+	pgraph->ctxprog_size = ctx.ctxprog_len;
+
+	pgraph->base.destroy = nv50_graph_destroy;
+	pgraph->base.init = nv50_graph_init;
+	pgraph->base.fini = nv50_graph_fini;
+	pgraph->base.context_new = nv50_graph_context_new;
+	pgraph->base.context_del = nv50_graph_context_del;
+	pgraph->base.object_new = nv50_graph_object_new;
+	if (dev_priv->chipset == 0x50 || dev_priv->chipset == 0xac)
+		pgraph->base.tlb_flush = nv50_graph_tlb_flush;
+	else
+		pgraph->base.tlb_flush = nv84_graph_tlb_flush;
+
+	nouveau_irq_register(dev, 12, nv50_graph_isr);
+
+	/* NVSW really doesn't live here... */
+	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
+	NVOBJ_MTHD (dev, 0x506e, 0x018c, nv50_graph_nvsw_dma_vblsem);
+	NVOBJ_MTHD (dev, 0x506e, 0x0400, nv50_graph_nvsw_vblsem_offset);
+	NVOBJ_MTHD (dev, 0x506e, 0x0404, nv50_graph_nvsw_vblsem_release_val);
+	NVOBJ_MTHD (dev, 0x506e, 0x0408, nv50_graph_nvsw_vblsem_release);
+	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv50_graph_nvsw_mthd_page_flip);
+
+	NVOBJ_ENGINE_ADD(dev, GR, &pgraph->base);
+	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
+	NVOBJ_CLASS(dev, 0x5039, GR); /* m2mf */
+	NVOBJ_CLASS(dev, 0x502d, GR); /* 2d */
+
+	/* tesla */
+	if (dev_priv->chipset == 0x50)
+		NVOBJ_CLASS(dev, 0x5097, GR); /* tesla (nv50) */
+	else
+	if (dev_priv->chipset < 0xa0)
+		NVOBJ_CLASS(dev, 0x8297, GR); /* tesla (nv8x/nv9x) */
+	else {
+		switch (dev_priv->chipset) {
+		case 0xa0:
+		case 0xaa:
+		case 0xac:
+			NVOBJ_CLASS(dev, 0x8397, GR);
+			break;
+		case 0xa3:
+		case 0xa5:
+		case 0xa8:
+			NVOBJ_CLASS(dev, 0x8597, GR);
+			break;
+		case 0xaf:
+			NVOBJ_CLASS(dev, 0x8697, GR);
+			break;
+		}
+	}
+
+	/* compute */
+	NVOBJ_CLASS(dev, 0x50c0, GR);
+	if (dev_priv->chipset  > 0xa0 &&
+	    dev_priv->chipset != 0xaa &&
+	    dev_priv->chipset != 0xac)
+		NVOBJ_CLASS(dev, 0x85c0, GR);
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nv50_grctx.c b/drivers/gpu/drm/nouveau/nv50_grctx.c
index 336aab2a24a6..de9abff12b90 100644
--- a/drivers/gpu/drm/nouveau/nv50_grctx.c
+++ b/drivers/gpu/drm/nouveau/nv50_grctx.c
@@ -747,7 +747,7 @@ nv50_graph_construct_mmio(struct nouveau_grctx *ctx)
 				gr_def(ctx, offset + 0x64, 0x0000001f);
 				gr_def(ctx, offset + 0x68, 0x0000000f);
 				gr_def(ctx, offset + 0x6c, 0x0000000f);
-			} else if(dev_priv->chipset < 0xa0) {
+			} else if (dev_priv->chipset < 0xa0) {
 				cp_ctx(ctx, offset + 0x50, 1);
 				cp_ctx(ctx, offset + 0x70, 1);
 			} else {
@@ -924,7 +924,7 @@ nv50_graph_construct_mmio_ddata(struct nouveau_grctx *ctx)
 		dd_emit(ctx, 1, 0);	/* 0000007f MULTISAMPLE_SAMPLES_LOG2 */
 	} else {
 		dd_emit(ctx, 1, 0);	/* 0000000f MULTISAMPLE_SAMPLES_LOG2 */
-	} 
+	}
 	dd_emit(ctx, 1, 0xc);		/* 000000ff SEMANTIC_COLOR.BFC0_ID */
 	if (dev_priv->chipset != 0x50)
 		dd_emit(ctx, 1, 0);	/* 00000001 SEMANTIC_COLOR.CLMP_EN */
@@ -1803,9 +1803,7 @@ nv50_graph_construct_gene_unk24xx(struct nouveau_grctx *ctx)
 		xf_emit(ctx, 1, 0);	/* 1ff */
 		xf_emit(ctx, 8, 0);	/* 0? */
 		xf_emit(ctx, 9, 0);	/* ffffffff, 7ff */
-	}
-	else
-	{
+	} else {
 		xf_emit(ctx, 0xc, 0);	/* RO */
 		/* SEEK */
 		xf_emit(ctx, 0xe10, 0); /* 190 * 9: 8*ffffffff, 7ff */
@@ -2836,7 +2834,7 @@ nv50_graph_construct_xfer_tprop(struct nouveau_grctx *ctx)
 	xf_emit(ctx, 1, 1);		/* 00000001 DST_LINEAR */
 	if (IS_NVA3F(dev_priv->chipset))
 		xf_emit(ctx, 1, 1);	/* 0000001f tesla UNK169C */
-	if(dev_priv->chipset == 0x50)
+	if (dev_priv->chipset == 0x50)
 		xf_emit(ctx, 1, 0);	/* ff */
 	else
 		xf_emit(ctx, 3, 0);	/* 1, 7, 3ff */
diff --git a/drivers/gpu/drm/nouveau/nv50_mpeg.c b/drivers/gpu/drm/nouveau/nv50_mpeg.c
new file mode 100644
index 000000000000..1dc5913f78c5
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nv50_mpeg.c
@@ -0,0 +1,256 @@
+/*
+ * 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
+ */
+
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_ramht.h"
+
+struct nv50_mpeg_engine {
+	struct nouveau_exec_engine base;
+};
+
+static inline u32
+CTX_PTR(struct drm_device *dev, u32 offset)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	if (dev_priv->chipset == 0x50)
+		offset += 0x0260;
+	else
+		offset += 0x0060;
+
+	return offset;
+}
+
+static int
+nv50_mpeg_context_new(struct nouveau_channel *chan, int engine)
+{
+	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *ramin = chan->ramin;
+	struct nouveau_gpuobj *ctx = NULL;
+	int ret;
+
+	NV_DEBUG(dev, "ch%d\n", chan->id);
+
+	ret = nouveau_gpuobj_new(dev, chan, 128 * 4, 0, NVOBJ_FLAG_ZERO_ALLOC |
+				 NVOBJ_FLAG_ZERO_FREE, &ctx);
+	if (ret)
+		return ret;
+
+	nv_wo32(ramin, CTX_PTR(dev, 0x00), 0x80190002);
+	nv_wo32(ramin, CTX_PTR(dev, 0x04), ctx->vinst + ctx->size - 1);
+	nv_wo32(ramin, CTX_PTR(dev, 0x08), ctx->vinst);
+	nv_wo32(ramin, CTX_PTR(dev, 0x0c), 0);
+	nv_wo32(ramin, CTX_PTR(dev, 0x10), 0);
+	nv_wo32(ramin, CTX_PTR(dev, 0x14), 0x00010000);
+
+	nv_wo32(ctx, 0x70, 0x00801ec1);
+	nv_wo32(ctx, 0x7c, 0x0000037c);
+	dev_priv->engine.instmem.flush(dev);
+
+	chan->engctx[engine] = ctx;
+	return 0;
+}
+
+static void
+nv50_mpeg_context_del(struct nouveau_channel *chan, int engine)
+{
+	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
+	struct nouveau_gpuobj *ctx = chan->engctx[engine];
+	struct drm_device *dev = chan->dev;
+	unsigned long flags;
+	u32 inst, i;
+
+	if (!chan->ramin)
+		return;
+
+	inst  = chan->ramin->vinst >> 12;
+	inst |= 0x80000000;
+
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	nv_mask(dev, 0x00b32c, 0x00000001, 0x00000000);
+	if (nv_rd32(dev, 0x00b318) == inst)
+		nv_mask(dev, 0x00b318, 0x80000000, 0x00000000);
+	nv_mask(dev, 0x00b32c, 0x00000001, 0x00000001);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
+
+	for (i = 0x00; i <= 0x14; i += 4)
+		nv_wo32(chan->ramin, CTX_PTR(dev, i), 0x00000000);
+	nouveau_gpuobj_ref(NULL, &ctx);
+	chan->engctx[engine] = NULL;
+}
+
+static int
+nv50_mpeg_object_new(struct nouveau_channel *chan, int engine,
+		     u32 handle, u16 class)
+{
+	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *obj = NULL;
+	int ret;
+
+	ret = nouveau_gpuobj_new(dev, chan, 16, 16, NVOBJ_FLAG_ZERO_FREE, &obj);
+	if (ret)
+		return ret;
+	obj->engine = 2;
+	obj->class  = class;
+
+	nv_wo32(obj, 0x00, class);
+	nv_wo32(obj, 0x04, 0x00000000);
+	nv_wo32(obj, 0x08, 0x00000000);
+	nv_wo32(obj, 0x0c, 0x00000000);
+	dev_priv->engine.instmem.flush(dev);
+
+	ret = nouveau_ramht_insert(chan, handle, obj);
+	nouveau_gpuobj_ref(NULL, &obj);
+	return ret;
+}
+
+static void
+nv50_mpeg_tlb_flush(struct drm_device *dev, int engine)
+{
+	nv50_vm_flush_engine(dev, 0x08);
+}
+
+static int
+nv50_mpeg_init(struct drm_device *dev, int engine)
+{
+	nv_wr32(dev, 0x00b32c, 0x00000000);
+	nv_wr32(dev, 0x00b314, 0x00000100);
+	nv_wr32(dev, 0x00b0e0, 0x0000001a);
+
+	nv_wr32(dev, 0x00b220, 0x00000044);
+	nv_wr32(dev, 0x00b300, 0x00801ec1);
+	nv_wr32(dev, 0x00b390, 0x00000000);
+	nv_wr32(dev, 0x00b394, 0x00000000);
+	nv_wr32(dev, 0x00b398, 0x00000000);
+	nv_mask(dev, 0x00b32c, 0x00000001, 0x00000001);
+
+	nv_wr32(dev, 0x00b100, 0xffffffff);
+	nv_wr32(dev, 0x00b140, 0xffffffff);
+
+	if (!nv_wait(dev, 0x00b200, 0x00000001, 0x00000000)) {
+		NV_ERROR(dev, "PMPEG init: 0x%08x\n", nv_rd32(dev, 0x00b200));
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+static int
+nv50_mpeg_fini(struct drm_device *dev, int engine)
+{
+	/*XXX: context save for s/r */
+	nv_mask(dev, 0x00b32c, 0x00000001, 0x00000000);
+	nv_wr32(dev, 0x00b140, 0x00000000);
+	return 0;
+}
+
+static void
+nv50_mpeg_isr(struct drm_device *dev)
+{
+	u32 stat = nv_rd32(dev, 0x00b100);
+	u32 type = nv_rd32(dev, 0x00b230);
+	u32 mthd = nv_rd32(dev, 0x00b234);
+	u32 data = nv_rd32(dev, 0x00b238);
+	u32 show = stat;
+
+	if (stat & 0x01000000) {
+		/* happens on initial binding of the object */
+		if (type == 0x00000020 && mthd == 0x0000) {
+			nv_wr32(dev, 0x00b308, 0x00000100);
+			show &= ~0x01000000;
+		}
+	}
+
+	if (show && nouveau_ratelimit()) {
+		NV_INFO(dev, "PMPEG - 0x%08x 0x%08x 0x%08x 0x%08x\n",
+			stat, type, mthd, data);
+	}
+
+	nv_wr32(dev, 0x00b100, stat);
+	nv_wr32(dev, 0x00b230, 0x00000001);
+	nv50_fb_vm_trap(dev, 1);
+}
+
+static void
+nv50_vpe_isr(struct drm_device *dev)
+{
+	if (nv_rd32(dev, 0x00b100))
+		nv50_mpeg_isr(dev);
+
+	if (nv_rd32(dev, 0x00b800)) {
+		u32 stat = nv_rd32(dev, 0x00b800);
+		NV_INFO(dev, "PMSRCH: 0x%08x\n", stat);
+		nv_wr32(dev, 0xb800, stat);
+	}
+}
+
+static void
+nv50_mpeg_destroy(struct drm_device *dev, int engine)
+{
+	struct nv50_mpeg_engine *pmpeg = nv_engine(dev, engine);
+
+	nouveau_irq_unregister(dev, 0);
+
+	NVOBJ_ENGINE_DEL(dev, MPEG);
+	kfree(pmpeg);
+}
+
+int
+nv50_mpeg_create(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nv50_mpeg_engine *pmpeg;
+
+	pmpeg = kzalloc(sizeof(*pmpeg), GFP_KERNEL);
+	if (!pmpeg)
+		return -ENOMEM;
+
+	pmpeg->base.destroy = nv50_mpeg_destroy;
+	pmpeg->base.init = nv50_mpeg_init;
+	pmpeg->base.fini = nv50_mpeg_fini;
+	pmpeg->base.context_new = nv50_mpeg_context_new;
+	pmpeg->base.context_del = nv50_mpeg_context_del;
+	pmpeg->base.object_new = nv50_mpeg_object_new;
+	pmpeg->base.tlb_flush = nv50_mpeg_tlb_flush;
+
+	if (dev_priv->chipset == 0x50) {
+		nouveau_irq_register(dev, 0, nv50_vpe_isr);
+		NVOBJ_ENGINE_ADD(dev, MPEG, &pmpeg->base);
+		NVOBJ_CLASS(dev, 0x3174, MPEG);
+#if 0
+		NVOBJ_ENGINE_ADD(dev, ME, &pme->base);
+		NVOBJ_CLASS(dev, 0x4075, ME);
+#endif
+	} else {
+		nouveau_irq_register(dev, 0, nv50_mpeg_isr);
+		NVOBJ_ENGINE_ADD(dev, MPEG, &pmpeg->base);
+		NVOBJ_CLASS(dev, 0x8274, MPEG);
+	}
+
+	return 0;
+
+}
diff --git a/drivers/gpu/drm/nouveau/nv50_pm.c b/drivers/gpu/drm/nouveau/nv50_pm.c
index 7dbb305d7e63..8a2810011bda 100644
--- a/drivers/gpu/drm/nouveau/nv50_pm.c
+++ b/drivers/gpu/drm/nouveau/nv50_pm.c
@@ -47,6 +47,21 @@ nv50_pm_clock_get(struct drm_device *dev, u32 id)
 
 	reg0 = nv_rd32(dev, pll.reg + 0);
 	reg1 = nv_rd32(dev, pll.reg + 4);
+
+	if ((reg0 & 0x80000000) == 0) {
+		if (id == PLL_SHADER) {
+			NV_DEBUG(dev, "Shader PLL is disabled. "
+				"Shader clock is twice the core\n");
+			ret = nv50_pm_clock_get(dev, PLL_CORE);
+			if (ret > 0)
+				return ret << 1;
+		} else if (id == PLL_MEMORY) {
+			NV_DEBUG(dev, "Memory PLL is disabled. "
+				"Memory clock is equal to the ref_clk\n");
+			return pll.refclk;
+		}
+	}
+
 	P = (reg0 & 0x00070000) >> 16;
 	N = (reg1 & 0x0000ff00) >> 8;
 	M = (reg1 & 0x000000ff);
diff --git a/drivers/gpu/drm/nouveau/nv50_vm.c b/drivers/gpu/drm/nouveau/nv50_vm.c
index 6c2694490741..1a0dd491a0e4 100644
--- a/drivers/gpu/drm/nouveau/nv50_vm.c
+++ b/drivers/gpu/drm/nouveau/nv50_vm.c
@@ -151,8 +151,7 @@ nv50_vm_flush(struct nouveau_vm *vm)
 	struct drm_nouveau_private *dev_priv = vm->dev->dev_private;
 	struct nouveau_instmem_engine *pinstmem = &dev_priv->engine.instmem;
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct nouveau_crypt_engine *pcrypt = &dev_priv->engine.crypt;
+	int i;
 
 	pinstmem->flush(vm->dev);
 
@@ -163,11 +162,10 @@ nv50_vm_flush(struct nouveau_vm *vm)
 	}
 
 	pfifo->tlb_flush(vm->dev);
-
-	if (atomic_read(&vm->pgraph_refs))
-		pgraph->tlb_flush(vm->dev);
-	if (atomic_read(&vm->pcrypt_refs))
-		pcrypt->tlb_flush(vm->dev);
+	for (i = 0; i < NVOBJ_ENGINE_NR; i++) {
+		if (atomic_read(&vm->engref[i]))
+			dev_priv->eng[i]->tlb_flush(vm->dev, i);
+	}
 }
 
 void
diff --git a/drivers/gpu/drm/nouveau/nv84_crypt.c b/drivers/gpu/drm/nouveau/nv84_crypt.c
index fabc7fd30b1d..75b809a51748 100644
--- a/drivers/gpu/drm/nouveau/nv84_crypt.c
+++ b/drivers/gpu/drm/nouveau/nv84_crypt.c
@@ -26,46 +26,48 @@
 #include "nouveau_drv.h"
 #include "nouveau_util.h"
 #include "nouveau_vm.h"
+#include "nouveau_ramht.h"
 
-static void nv84_crypt_isr(struct drm_device *);
+struct nv84_crypt_engine {
+	struct nouveau_exec_engine base;
+};
 
-int
-nv84_crypt_create_context(struct nouveau_channel *chan)
+static int
+nv84_crypt_context_new(struct nouveau_channel *chan, int engine)
 {
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_gpuobj *ramin = chan->ramin;
+	struct nouveau_gpuobj *ctx;
 	int ret;
 
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
-	ret = nouveau_gpuobj_new(dev, chan, 256, 0,
-				 NVOBJ_FLAG_ZERO_ALLOC | NVOBJ_FLAG_ZERO_FREE,
-				 &chan->crypt_ctx);
+	ret = nouveau_gpuobj_new(dev, chan, 256, 0, NVOBJ_FLAG_ZERO_ALLOC |
+				 NVOBJ_FLAG_ZERO_FREE, &ctx);
 	if (ret)
 		return ret;
 
 	nv_wo32(ramin, 0xa0, 0x00190000);
-	nv_wo32(ramin, 0xa4, chan->crypt_ctx->vinst + 0xff);
-	nv_wo32(ramin, 0xa8, chan->crypt_ctx->vinst);
+	nv_wo32(ramin, 0xa4, ctx->vinst + ctx->size - 1);
+	nv_wo32(ramin, 0xa8, ctx->vinst);
 	nv_wo32(ramin, 0xac, 0);
 	nv_wo32(ramin, 0xb0, 0);
 	nv_wo32(ramin, 0xb4, 0);
-
 	dev_priv->engine.instmem.flush(dev);
-	atomic_inc(&chan->vm->pcrypt_refs);
+
+	atomic_inc(&chan->vm->engref[engine]);
+	chan->engctx[engine] = ctx;
 	return 0;
 }
 
-void
-nv84_crypt_destroy_context(struct nouveau_channel *chan)
+static void
+nv84_crypt_context_del(struct nouveau_channel *chan, int engine)
 {
+	struct nouveau_gpuobj *ctx = chan->engctx[engine];
 	struct drm_device *dev = chan->dev;
 	u32 inst;
 
-	if (!chan->crypt_ctx)
-		return;
-
 	inst  = (chan->ramin->vinst >> 12);
 	inst |= 0x80000000;
 
@@ -80,43 +82,39 @@ nv84_crypt_destroy_context(struct nouveau_channel *chan)
 		nv_mask(dev, 0x10218c, 0x80000000, 0x00000000);
 	nv_wr32(dev, 0x10200c, 0x00000010);
 
-	nouveau_gpuobj_ref(NULL, &chan->crypt_ctx);
-	atomic_dec(&chan->vm->pcrypt_refs);
-}
+	nouveau_gpuobj_ref(NULL, &ctx);
 
-void
-nv84_crypt_tlb_flush(struct drm_device *dev)
-{
-	nv50_vm_flush_engine(dev, 0x0a);
+	atomic_dec(&chan->vm->engref[engine]);
+	chan->engctx[engine] = NULL;
 }
 
-int
-nv84_crypt_init(struct drm_device *dev)
+static int
+nv84_crypt_object_new(struct nouveau_channel *chan, int engine,
+		      u32 handle, u16 class)
 {
+	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_crypt_engine *pcrypt = &dev_priv->engine.crypt;
-
-	if (!pcrypt->registered) {
-		NVOBJ_CLASS(dev, 0x74c1, CRYPT);
-		pcrypt->registered = true;
-	}
+	struct nouveau_gpuobj *obj = NULL;
+	int ret;
 
-	nv_mask(dev, 0x000200, 0x00004000, 0x00000000);
-	nv_mask(dev, 0x000200, 0x00004000, 0x00004000);
+	ret = nouveau_gpuobj_new(dev, chan, 16, 16, NVOBJ_FLAG_ZERO_FREE, &obj);
+	if (ret)
+		return ret;
+	obj->engine = 5;
+	obj->class  = class;
 
-	nouveau_irq_register(dev, 14, nv84_crypt_isr);
-	nv_wr32(dev, 0x102130, 0xffffffff);
-	nv_wr32(dev, 0x102140, 0xffffffbf);
+	nv_wo32(obj, 0x00, class);
+	dev_priv->engine.instmem.flush(dev);
 
-	nv_wr32(dev, 0x10200c, 0x00000010);
-	return 0;
+	ret = nouveau_ramht_insert(chan, handle, obj);
+	nouveau_gpuobj_ref(NULL, &obj);
+	return ret;
 }
 
-void
-nv84_crypt_fini(struct drm_device *dev)
+static void
+nv84_crypt_tlb_flush(struct drm_device *dev, int engine)
 {
-	nv_wr32(dev, 0x102140, 0x00000000);
-	nouveau_irq_unregister(dev, 14);
+	nv50_vm_flush_engine(dev, 0x0a);
 }
 
 static void
@@ -138,3 +136,58 @@ nv84_crypt_isr(struct drm_device *dev)
 
 	nv50_fb_vm_trap(dev, show);
 }
+
+static int
+nv84_crypt_fini(struct drm_device *dev, int engine)
+{
+	nv_wr32(dev, 0x102140, 0x00000000);
+	return 0;
+}
+
+static int
+nv84_crypt_init(struct drm_device *dev, int engine)
+{
+	nv_mask(dev, 0x000200, 0x00004000, 0x00000000);
+	nv_mask(dev, 0x000200, 0x00004000, 0x00004000);
+
+	nv_wr32(dev, 0x102130, 0xffffffff);
+	nv_wr32(dev, 0x102140, 0xffffffbf);
+
+	nv_wr32(dev, 0x10200c, 0x00000010);
+	return 0;
+}
+
+static void
+nv84_crypt_destroy(struct drm_device *dev, int engine)
+{
+	struct nv84_crypt_engine *pcrypt = nv_engine(dev, engine);
+
+	NVOBJ_ENGINE_DEL(dev, CRYPT);
+
+	nouveau_irq_unregister(dev, 14);
+	kfree(pcrypt);
+}
+
+int
+nv84_crypt_create(struct drm_device *dev)
+{
+	struct nv84_crypt_engine *pcrypt;
+
+	pcrypt = kzalloc(sizeof(*pcrypt), GFP_KERNEL);
+	if (!pcrypt)
+		return -ENOMEM;
+
+	pcrypt->base.destroy = nv84_crypt_destroy;
+	pcrypt->base.init = nv84_crypt_init;
+	pcrypt->base.fini = nv84_crypt_fini;
+	pcrypt->base.context_new = nv84_crypt_context_new;
+	pcrypt->base.context_del = nv84_crypt_context_del;
+	pcrypt->base.object_new = nv84_crypt_object_new;
+	pcrypt->base.tlb_flush = nv84_crypt_tlb_flush;
+
+	nouveau_irq_register(dev, 14, nv84_crypt_isr);
+
+	NVOBJ_ENGINE_ADD(dev, CRYPT, &pcrypt->base);
+	NVOBJ_CLASS (dev, 0x74c1, CRYPT);
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nva3_copy.c b/drivers/gpu/drm/nouveau/nva3_copy.c
new file mode 100644
index 000000000000..b86820a61220
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nva3_copy.c
@@ -0,0 +1,226 @@
+/*
+ * 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
+ */
+
+#include <linux/firmware.h>
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_util.h"
+#include "nouveau_vm.h"
+#include "nouveau_ramht.h"
+#include "nva3_copy.fuc.h"
+
+struct nva3_copy_engine {
+	struct nouveau_exec_engine base;
+};
+
+static int
+nva3_copy_context_new(struct nouveau_channel *chan, int engine)
+{
+	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *ramin = chan->ramin;
+	struct nouveau_gpuobj *ctx = NULL;
+	int ret;
+
+	NV_DEBUG(dev, "ch%d\n", chan->id);
+
+	ret = nouveau_gpuobj_new(dev, chan, 256, 0, NVOBJ_FLAG_ZERO_ALLOC |
+				 NVOBJ_FLAG_ZERO_FREE, &ctx);
+	if (ret)
+		return ret;
+
+	nv_wo32(ramin, 0xc0, 0x00190000);
+	nv_wo32(ramin, 0xc4, ctx->vinst + ctx->size - 1);
+	nv_wo32(ramin, 0xc8, ctx->vinst);
+	nv_wo32(ramin, 0xcc, 0x00000000);
+	nv_wo32(ramin, 0xd0, 0x00000000);
+	nv_wo32(ramin, 0xd4, 0x00000000);
+	dev_priv->engine.instmem.flush(dev);
+
+	atomic_inc(&chan->vm->engref[engine]);
+	chan->engctx[engine] = ctx;
+	return 0;
+}
+
+static int
+nva3_copy_object_new(struct nouveau_channel *chan, int engine,
+		     u32 handle, u16 class)
+{
+	struct nouveau_gpuobj *ctx = chan->engctx[engine];
+
+	/* fuc engine doesn't need an object, our ramht code does.. */
+	ctx->engine = 3;
+	ctx->class  = class;
+	return nouveau_ramht_insert(chan, handle, ctx);
+}
+
+static void
+nva3_copy_context_del(struct nouveau_channel *chan, int engine)
+{
+	struct nouveau_gpuobj *ctx = chan->engctx[engine];
+	struct drm_device *dev = chan->dev;
+	u32 inst;
+
+	inst  = (chan->ramin->vinst >> 12);
+	inst |= 0x40000000;
+
+	/* disable fifo access */
+	nv_wr32(dev, 0x104048, 0x00000000);
+	/* mark channel as unloaded if it's currently active */
+	if (nv_rd32(dev, 0x104050) == inst)
+		nv_mask(dev, 0x104050, 0x40000000, 0x00000000);
+	/* mark next channel as invalid if it's about to be loaded */
+	if (nv_rd32(dev, 0x104054) == inst)
+		nv_mask(dev, 0x104054, 0x40000000, 0x00000000);
+	/* restore fifo access */
+	nv_wr32(dev, 0x104048, 0x00000003);
+
+	for (inst = 0xc0; inst <= 0xd4; inst += 4)
+		nv_wo32(chan->ramin, inst, 0x00000000);
+
+	nouveau_gpuobj_ref(NULL, &ctx);
+
+	atomic_dec(&chan->vm->engref[engine]);
+	chan->engctx[engine] = ctx;
+}
+
+static void
+nva3_copy_tlb_flush(struct drm_device *dev, int engine)
+{
+	nv50_vm_flush_engine(dev, 0x0d);
+}
+
+static int
+nva3_copy_init(struct drm_device *dev, int engine)
+{
+	int i;
+
+	nv_mask(dev, 0x000200, 0x00002000, 0x00000000);
+	nv_mask(dev, 0x000200, 0x00002000, 0x00002000);
+	nv_wr32(dev, 0x104014, 0xffffffff); /* disable all interrupts */
+
+	/* upload ucode */
+	nv_wr32(dev, 0x1041c0, 0x01000000);
+	for (i = 0; i < sizeof(nva3_pcopy_data) / 4; i++)
+		nv_wr32(dev, 0x1041c4, nva3_pcopy_data[i]);
+
+	nv_wr32(dev, 0x104180, 0x01000000);
+	for (i = 0; i < sizeof(nva3_pcopy_code) / 4; i++) {
+		if ((i & 0x3f) == 0)
+			nv_wr32(dev, 0x104188, i >> 6);
+		nv_wr32(dev, 0x104184, nva3_pcopy_code[i]);
+	}
+
+	/* start it running */
+	nv_wr32(dev, 0x10410c, 0x00000000);
+	nv_wr32(dev, 0x104104, 0x00000000); /* ENTRY */
+	nv_wr32(dev, 0x104100, 0x00000002); /* TRIGGER */
+	return 0;
+}
+
+static int
+nva3_copy_fini(struct drm_device *dev, int engine)
+{
+	nv_mask(dev, 0x104048, 0x00000003, 0x00000000);
+
+	/* trigger fuc context unload */
+	nv_wait(dev, 0x104008, 0x0000000c, 0x00000000);
+	nv_mask(dev, 0x104054, 0x40000000, 0x00000000);
+	nv_wr32(dev, 0x104000, 0x00000008);
+	nv_wait(dev, 0x104008, 0x00000008, 0x00000000);
+
+	nv_wr32(dev, 0x104014, 0xffffffff);
+	return 0;
+}
+
+static struct nouveau_enum nva3_copy_isr_error_name[] = {
+	{ 0x0001, "ILLEGAL_MTHD" },
+	{ 0x0002, "INVALID_ENUM" },
+	{ 0x0003, "INVALID_BITFIELD" },
+	{}
+};
+
+static void
+nva3_copy_isr(struct drm_device *dev)
+{
+	u32 dispatch = nv_rd32(dev, 0x10401c);
+	u32 stat = nv_rd32(dev, 0x104008) & dispatch & ~(dispatch >> 16);
+	u32 inst = nv_rd32(dev, 0x104050) & 0x3fffffff;
+	u32 ssta = nv_rd32(dev, 0x104040) & 0x0000ffff;
+	u32 addr = nv_rd32(dev, 0x104040) >> 16;
+	u32 mthd = (addr & 0x07ff) << 2;
+	u32 subc = (addr & 0x3800) >> 11;
+	u32 data = nv_rd32(dev, 0x104044);
+	int chid = nv50_graph_isr_chid(dev, inst);
+
+	if (stat & 0x00000040) {
+		NV_INFO(dev, "PCOPY: DISPATCH_ERROR [");
+		nouveau_enum_print(nva3_copy_isr_error_name, ssta);
+		printk("] ch %d [0x%08x] subc %d mthd 0x%04x data 0x%08x\n",
+			chid, inst, subc, mthd, data);
+		nv_wr32(dev, 0x104004, 0x00000040);
+		stat &= ~0x00000040;
+	}
+
+	if (stat) {
+		NV_INFO(dev, "PCOPY: unhandled intr 0x%08x\n", stat);
+		nv_wr32(dev, 0x104004, stat);
+	}
+	nv50_fb_vm_trap(dev, 1);
+}
+
+static void
+nva3_copy_destroy(struct drm_device *dev, int engine)
+{
+	struct nva3_copy_engine *pcopy = nv_engine(dev, engine);
+
+	nouveau_irq_unregister(dev, 22);
+
+	NVOBJ_ENGINE_DEL(dev, COPY0);
+	kfree(pcopy);
+}
+
+int
+nva3_copy_create(struct drm_device *dev)
+{
+	struct nva3_copy_engine *pcopy;
+
+	pcopy = kzalloc(sizeof(*pcopy), GFP_KERNEL);
+	if (!pcopy)
+		return -ENOMEM;
+
+	pcopy->base.destroy = nva3_copy_destroy;
+	pcopy->base.init = nva3_copy_init;
+	pcopy->base.fini = nva3_copy_fini;
+	pcopy->base.context_new = nva3_copy_context_new;
+	pcopy->base.context_del = nva3_copy_context_del;
+	pcopy->base.object_new = nva3_copy_object_new;
+	pcopy->base.tlb_flush = nva3_copy_tlb_flush;
+
+	nouveau_irq_register(dev, 22, nva3_copy_isr);
+
+	NVOBJ_ENGINE_ADD(dev, COPY0, &pcopy->base);
+	NVOBJ_CLASS(dev, 0x85b5, COPY0);
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nva3_copy.fuc b/drivers/gpu/drm/nouveau/nva3_copy.fuc
new file mode 100644
index 000000000000..eaf35f8321ee
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nva3_copy.fuc
@@ -0,0 +1,870 @@
+/* fuc microcode for copy engine on nva3- chipsets
+ *
+ * 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 for nva3:nvc0
+ *    m4 -DNVA3 nva3_copy.fuc | envyas -a -w -m fuc -V nva3 -o nva3_copy.fuc.h
+ *
+ * To build for nvc0-
+ *    m4 -DNVC0 nva3_copy.fuc | envyas -a -w -m fuc -V nva3 -o nvc0_copy.fuc.h
+ */
+
+ifdef(`NVA3',
+.section nva3_pcopy_data,
+.section nvc0_pcopy_data
+)
+
+ctx_object:                   .b32 0
+ifdef(`NVA3',
+ctx_dma:
+ctx_dma_query:                .b32 0
+ctx_dma_src:                  .b32 0
+ctx_dma_dst:                  .b32 0
+,)
+.equ ctx_dma_count 3
+ctx_query_address_high:       .b32 0
+ctx_query_address_low:        .b32 0
+ctx_query_counter:            .b32 0
+ctx_src_address_high:         .b32 0
+ctx_src_address_low:          .b32 0
+ctx_src_pitch:                .b32 0
+ctx_src_tile_mode:            .b32 0
+ctx_src_xsize:                .b32 0
+ctx_src_ysize:                .b32 0
+ctx_src_zsize:                .b32 0
+ctx_src_zoff:                 .b32 0
+ctx_src_xoff:                 .b32 0
+ctx_src_yoff:                 .b32 0
+ctx_src_cpp:                  .b32 0
+ctx_dst_address_high:         .b32 0
+ctx_dst_address_low:          .b32 0
+ctx_dst_pitch:                .b32 0
+ctx_dst_tile_mode:            .b32 0
+ctx_dst_xsize:                .b32 0
+ctx_dst_ysize:                .b32 0
+ctx_dst_zsize:                .b32 0
+ctx_dst_zoff:                 .b32 0
+ctx_dst_xoff:                 .b32 0
+ctx_dst_yoff:                 .b32 0
+ctx_dst_cpp:                  .b32 0
+ctx_format:                   .b32 0
+ctx_swz_const0:               .b32 0
+ctx_swz_const1:               .b32 0
+ctx_xcnt:                     .b32 0
+ctx_ycnt:                     .b32 0
+.align 256
+
+dispatch_table:
+// mthd 0x0000, NAME
+.b16 0x000 1
+.b32 ctx_object                     ~0xffffffff
+// mthd 0x0100, NOP
+.b16 0x040 1
+.b32 0x00010000 + cmd_nop           ~0xffffffff
+// mthd 0x0140, PM_TRIGGER
+.b16 0x050 1
+.b32 0x00010000 + cmd_pm_trigger    ~0xffffffff
+ifdef(`NVA3', `
+// mthd 0x0180-0x018c, DMA_
+.b16 0x060 ctx_dma_count
+dispatch_dma:
+.b32 0x00010000 + cmd_dma           ~0xffffffff
+.b32 0x00010000 + cmd_dma           ~0xffffffff
+.b32 0x00010000 + cmd_dma           ~0xffffffff
+',)
+// mthd 0x0200-0x0218, SRC_TILE
+.b16 0x80 7
+.b32 ctx_src_tile_mode              ~0x00000fff
+.b32 ctx_src_xsize                  ~0x0007ffff
+.b32 ctx_src_ysize                  ~0x00001fff
+.b32 ctx_src_zsize                  ~0x000007ff
+.b32 ctx_src_zoff                   ~0x00000fff
+.b32 ctx_src_xoff                   ~0x0007ffff
+.b32 ctx_src_yoff                   ~0x00001fff
+// mthd 0x0220-0x0238, DST_TILE
+.b16 0x88 7
+.b32 ctx_dst_tile_mode              ~0x00000fff
+.b32 ctx_dst_xsize                  ~0x0007ffff
+.b32 ctx_dst_ysize                  ~0x00001fff
+.b32 ctx_dst_zsize                  ~0x000007ff
+.b32 ctx_dst_zoff                   ~0x00000fff
+.b32 ctx_dst_xoff                   ~0x0007ffff
+.b32 ctx_dst_yoff                   ~0x00001fff
+// mthd 0x0300-0x0304, EXEC, WRCACHE_FLUSH
+.b16 0xc0 2
+.b32 0x00010000 + cmd_exec          ~0xffffffff
+.b32 0x00010000 + cmd_wrcache_flush ~0xffffffff
+// mthd 0x030c-0x0340, various stuff
+.b16 0xc3 14
+.b32 ctx_src_address_high           ~0x000000ff
+.b32 ctx_src_address_low            ~0xfffffff0
+.b32 ctx_dst_address_high           ~0x000000ff
+.b32 ctx_dst_address_low            ~0xfffffff0
+.b32 ctx_src_pitch                  ~0x0007ffff
+.b32 ctx_dst_pitch                  ~0x0007ffff
+.b32 ctx_xcnt                       ~0x0000ffff
+.b32 ctx_ycnt                       ~0x00001fff
+.b32 ctx_format                     ~0x0333ffff
+.b32 ctx_swz_const0                 ~0xffffffff
+.b32 ctx_swz_const1                 ~0xffffffff
+.b32 ctx_query_address_high         ~0x000000ff
+.b32 ctx_query_address_low          ~0xffffffff
+.b32 ctx_query_counter              ~0xffffffff
+.b16 0x800 0
+
+ifdef(`NVA3',
+.section nva3_pcopy_code,
+.section nvc0_pcopy_code
+)
+
+main:
+   clear b32 $r0
+   mov $sp $r0
+
+   // setup i0 handler and route fifo and ctxswitch to it
+   mov $r1 ih
+   mov $iv0 $r1
+   mov $r1 0x400
+   movw $r2 0xfff3
+   sethi $r2 0
+   iowr I[$r2 + 0x300] $r2
+
+   // enable interrupts
+   or $r2 0xc
+   iowr I[$r1] $r2
+   bset $flags ie0
+
+   // enable fifo access and context switching
+   mov $r1 0x1200
+   mov $r2 3
+   iowr I[$r1] $r2
+
+   // sleep forever, waking for interrupts
+   bset $flags $p0
+   spin:
+      sleep $p0
+      bra spin
+
+// i0 handler
+ih:
+   iord $r1 I[$r0 + 0x200]
+
+   and $r2 $r1 0x00000008
+   bra e ih_no_chsw
+      call chsw
+   ih_no_chsw:
+   and $r2 $r1 0x00000004
+   bra e ih_no_cmd
+      call dispatch
+
+   ih_no_cmd:
+   and $r1 $r1 0x0000000c
+   iowr I[$r0 + 0x100] $r1
+   iret
+
+// $p1 direction (0 = unload, 1 = load)
+// $r3 channel
+swctx:
+   mov $r4 0x7700
+   mov $xtargets $r4
+ifdef(`NVA3', `
+   // target 7 hardcoded to ctx dma object
+   mov $xdbase $r0
+', ` // NVC0
+   // read SCRATCH3 to decide if we are PCOPY0 or PCOPY1
+   mov $r4 0x2100
+   iord $r4 I[$r4 + 0]
+   and $r4 1
+   shl b32 $r4 4
+   add b32 $r4 0x30
+
+   // channel is in vram
+   mov $r15 0x61c
+   shl b32 $r15 6
+   mov $r5 0x114
+   iowrs I[$r15] $r5
+
+   // read 16-byte PCOPYn info, containing context pointer, from channel
+   shl b32 $r5 $r3 4
+   add b32 $r5 2
+   mov $xdbase $r5
+   mov $r5 $sp
+   // get a chunk of stack space, aligned to 256 byte boundary
+   sub b32 $r5 0x100
+   mov $r6 0xff
+   not b32 $r6
+   and $r5 $r6
+   sethi $r5 0x00020000
+   xdld $r4 $r5
+   xdwait
+   sethi $r5 0
+
+   // set context pointer, from within channel VM
+   mov $r14 0
+   iowrs I[$r15] $r14
+   ld b32 $r4 D[$r5 + 0]
+   shr b32 $r4 8
+   ld b32 $r6 D[$r5 + 4]
+   shl b32 $r6 24
+   or $r4 $r6
+   mov $xdbase $r4
+')
+   // 256-byte context, at start of data segment
+   mov b32 $r4 $r0
+   sethi $r4 0x60000
+
+   // swap!
+   bra $p1 swctx_load
+      xdst $r0 $r4
+      bra swctx_done
+   swctx_load:
+      xdld $r0 $r4
+   swctx_done:
+   xdwait
+   ret
+
+chsw:
+   // read current channel
+   mov $r2 0x1400
+   iord $r3 I[$r2]
+
+   // if it's active, unload it and return
+   xbit $r15 $r3 0x1e
+   bra e chsw_no_unload
+      bclr $flags $p1
+      call swctx
+      bclr $r3 0x1e
+      iowr I[$r2] $r3
+      mov $r4 1
+      iowr I[$r2 + 0x200] $r4
+      ret
+
+   // read next channel
+   chsw_no_unload:
+   iord $r3 I[$r2 + 0x100]
+
+   // is there a channel waiting to be loaded?
+   xbit $r13 $r3 0x1e
+   bra e chsw_finish_load
+      bset $flags $p1
+      call swctx
+ifdef(`NVA3',
+      // load dma objects back into TARGET regs
+      mov $r5 ctx_dma
+      mov $r6 ctx_dma_count
+      chsw_load_ctx_dma:
+         ld b32 $r7 D[$r5 + $r6 * 4]
+         add b32 $r8 $r6 0x180
+         shl b32 $r8 8
+         iowr I[$r8] $r7
+         sub b32 $r6 1
+         bra nc chsw_load_ctx_dma
+,)
+
+   chsw_finish_load:
+   mov $r3 2
+   iowr I[$r2 + 0x200] $r3
+   ret
+
+dispatch:
+   // read incoming fifo command
+   mov $r3 0x1900
+   iord $r2 I[$r3 + 0x100]
+   iord $r3 I[$r3 + 0x000]
+   and $r4 $r2 0x7ff
+   // $r2 will be used to store exception data
+   shl b32 $r2 0x10
+
+   // lookup method in the dispatch table, ILLEGAL_MTHD if not found
+   mov $r5 dispatch_table
+   clear b32 $r6
+   clear b32 $r7
+   dispatch_loop:
+      ld b16 $r6 D[$r5 + 0]
+      ld b16 $r7 D[$r5 + 2]
+      add b32 $r5 4
+      cmpu b32 $r4 $r6
+      bra c dispatch_illegal_mthd
+      add b32 $r7 $r6
+      cmpu b32 $r4 $r7
+      bra c dispatch_valid_mthd
+      sub b32 $r7 $r6
+      shl b32 $r7 3
+      add b32 $r5 $r7
+      bra dispatch_loop
+
+   // ensure no bits set in reserved fields, INVALID_BITFIELD
+   dispatch_valid_mthd:
+   sub b32 $r4 $r6
+   shl b32 $r4 3
+   add b32 $r4 $r5
+   ld b32 $r5 D[$r4 + 4]
+   and $r5 $r3
+   cmpu b32 $r5 0
+   bra ne dispatch_invalid_bitfield
+
+   // depending on dispatch flags: execute method, or save data as state
+   ld b16 $r5 D[$r4 + 0]
+   ld b16 $r6 D[$r4 + 2]
+   cmpu b32 $r6 0
+   bra ne dispatch_cmd
+      st b32 D[$r5] $r3
+      bra dispatch_done
+   dispatch_cmd:
+      bclr $flags $p1
+      call $r5
+      bra $p1 dispatch_error
+      bra dispatch_done
+
+   dispatch_invalid_bitfield:
+   or $r2 2
+   dispatch_illegal_mthd:
+   or $r2 1
+
+   // store exception data in SCRATCH0/SCRATCH1, signal hostirq
+   dispatch_error:
+   mov $r4 0x1000
+   iowr I[$r4 + 0x000] $r2
+   iowr I[$r4 + 0x100] $r3
+   mov $r2 0x40
+   iowr I[$r0] $r2
+   hostirq_wait:
+      iord $r2 I[$r0 + 0x200]
+      and $r2 0x40
+      cmpu b32 $r2 0
+      bra ne hostirq_wait
+
+   dispatch_done:
+   mov $r2 0x1d00
+   mov $r3 1
+   iowr I[$r2] $r3
+   ret
+
+// No-operation
+//
+// Inputs:
+//    $r1: irqh state
+//    $r2: hostirq state
+//    $r3: data
+//    $r4: dispatch table entry
+// Outputs:
+//    $r1: irqh state
+//    $p1: set on error
+//       $r2: hostirq state
+//       $r3: data
+cmd_nop:
+   ret
+
+// PM_TRIGGER
+//
+// Inputs:
+//    $r1: irqh state
+//    $r2: hostirq state
+//    $r3: data
+//    $r4: dispatch table entry
+// Outputs:
+//    $r1: irqh state
+//    $p1: set on error
+//       $r2: hostirq state
+//       $r3: data
+cmd_pm_trigger:
+   mov $r2 0x2200
+   clear b32 $r3
+   sethi $r3 0x20000
+   iowr I[$r2] $r3
+   ret
+
+ifdef(`NVA3',
+// SET_DMA_* method handler
+//
+// Inputs:
+//    $r1: irqh state
+//    $r2: hostirq state
+//    $r3: data
+//    $r4: dispatch table entry
+// Outputs:
+//    $r1: irqh state
+//    $p1: set on error
+//       $r2: hostirq state
+//       $r3: data
+cmd_dma:
+   sub b32 $r4 dispatch_dma
+   shr b32 $r4 1
+   bset $r3 0x1e
+   st b32 D[$r4 + ctx_dma] $r3
+   add b32 $r4 0x600
+   shl b32 $r4 6
+   iowr I[$r4] $r3
+   ret
+,)
+
+// Calculates the hw swizzle mask and adjusts the surface's xcnt to match
+//
+cmd_exec_set_format:
+   // zero out a chunk of the stack to store the swizzle into
+   add $sp -0x10
+   st b32 D[$sp + 0x00] $r0
+   st b32 D[$sp + 0x04] $r0
+   st b32 D[$sp + 0x08] $r0
+   st b32 D[$sp + 0x0c] $r0
+
+   // extract cpp, src_ncomp and dst_ncomp from FORMAT
+   ld b32 $r4 D[$r0 + ctx_format]
+   extr $r5 $r4 16:17
+   add b32 $r5 1
+   extr $r6 $r4 20:21
+   add b32 $r6 1
+   extr $r7 $r4 24:25
+   add b32 $r7 1
+
+   // convert FORMAT swizzle mask to hw swizzle mask
+   bclr $flags $p2
+   clear b32 $r8
+   clear b32 $r9
+   ncomp_loop:
+      and $r10 $r4 0xf
+      shr b32 $r4 4
+      clear b32 $r11
+      bpc_loop:
+         cmpu b8 $r10 4
+         bra nc cmp_c0
+            mulu $r12 $r10 $r5
+            add b32 $r12 $r11
+            bset $flags $p2
+            bra bpc_next
+         cmp_c0:
+         bra ne cmp_c1
+            mov $r12 0x10
+            add b32 $r12 $r11
+            bra bpc_next
+         cmp_c1:
+         cmpu b8 $r10 6
+         bra nc cmp_zero
+            mov $r12 0x14
+            add b32 $r12 $r11
+            bra bpc_next
+         cmp_zero:
+            mov $r12 0x80
+         bpc_next:
+         st b8 D[$sp + $r8] $r12
+         add b32 $r8 1
+         add b32 $r11 1
+         cmpu b32 $r11 $r5
+         bra c bpc_loop
+      add b32 $r9 1
+      cmpu b32 $r9 $r7
+      bra c ncomp_loop
+
+   // SRC_XCNT = (xcnt * src_cpp), or 0 if no src ref in swz (hw will hang)
+   mulu $r6 $r5
+   st b32 D[$r0 + ctx_src_cpp] $r6
+   ld b32 $r8 D[$r0 + ctx_xcnt]
+   mulu $r6 $r8
+   bra $p2 dst_xcnt
+   clear b32 $r6
+
+   dst_xcnt:
+   mulu $r7 $r5
+   st b32 D[$r0 + ctx_dst_cpp] $r7
+   mulu $r7 $r8
+
+   mov $r5 0x810
+   shl b32 $r5 6
+   iowr I[$r5 + 0x000] $r6
+   iowr I[$r5 + 0x100] $r7
+   add b32 $r5 0x800
+   ld b32 $r6 D[$r0 + ctx_dst_cpp]
+   sub b32 $r6 1
+   shl b32 $r6 8
+   ld b32 $r7 D[$r0 + ctx_src_cpp]
+   sub b32 $r7 1
+   or $r6 $r7
+   iowr I[$r5 + 0x000] $r6
+   add b32 $r5 0x100
+   ld b32 $r6 D[$sp + 0x00]
+   iowr I[$r5 + 0x000] $r6
+   ld b32 $r6 D[$sp + 0x04]
+   iowr I[$r5 + 0x100] $r6
+   ld b32 $r6 D[$sp + 0x08]
+   iowr I[$r5 + 0x200] $r6
+   ld b32 $r6 D[$sp + 0x0c]
+   iowr I[$r5 + 0x300] $r6
+   add b32 $r5 0x400
+   ld b32 $r6 D[$r0 + ctx_swz_const0]
+   iowr I[$r5 + 0x000] $r6
+   ld b32 $r6 D[$r0 + ctx_swz_const1]
+   iowr I[$r5 + 0x100] $r6
+   add $sp 0x10
+   ret
+
+// Setup to handle a tiled surface
+//
+// Calculates a number of parameters the hardware requires in order
+// to correctly handle tiling.
+//
+// Offset calculation is performed as follows (Tp/Th/Td from TILE_MODE):
+//    nTx = round_up(w * cpp, 1 << Tp) >> Tp
+//    nTy = round_up(h, 1 << Th) >> Th
+//    Txo = (x * cpp) & ((1 << Tp) - 1)
+//     Tx = (x * cpp) >> Tp
+//    Tyo = y & ((1 << Th) - 1)
+//     Ty = y >> Th
+//    Tzo = z & ((1 << Td) - 1)
+//     Tz = z >> Td
+//
+//    off  = (Tzo << Tp << Th) + (Tyo << Tp) + Txo
+//    off += ((Tz * nTy * nTx)) + (Ty * nTx) + Tx) << Td << Th << Tp;
+//
+// Inputs:
+//    $r4: hw command (0x104800)
+//    $r5: ctx offset adjustment for src/dst selection
+//    $p2: set if dst surface
+//
+cmd_exec_set_surface_tiled:
+   // translate TILE_MODE into Tp, Th, Td shift values
+   ld b32 $r7 D[$r5 + ctx_src_tile_mode]
+   extr $r9 $r7 8:11
+   extr $r8 $r7 4:7
+ifdef(`NVA3',
+   add b32 $r8 2
+,
+   add b32 $r8 3
+)
+   extr $r7 $r7 0:3
+   cmp b32 $r7 0xe
+   bra ne xtile64
+   mov $r7 4
+   bra xtileok
+   xtile64:
+   xbit $r7 $flags $p2
+   add b32 $r7 17
+   bset $r4 $r7
+   mov $r7 6
+   xtileok:
+
+   // Op = (x * cpp) & ((1 << Tp) - 1)
+   // Tx = (x * cpp) >> Tp
+   ld b32 $r10 D[$r5 + ctx_src_xoff]
+   ld b32 $r11 D[$r5 + ctx_src_cpp]
+   mulu $r10 $r11
+   mov $r11 1
+   shl b32 $r11 $r7
+   sub b32 $r11 1
+   and $r12 $r10 $r11
+   shr b32 $r10 $r7
+
+   // Tyo = y & ((1 << Th) - 1)
+   // Ty  = y >> Th
+   ld b32 $r13 D[$r5 + ctx_src_yoff]
+   mov $r14 1
+   shl b32 $r14 $r8
+   sub b32 $r14 1
+   and $r11 $r13 $r14
+   shr b32 $r13 $r8
+
+   // YTILE = ((1 << Th) << 12) | ((1 << Th) - Tyo)
+   add b32 $r14 1
+   shl b32 $r15 $r14 12
+   sub b32 $r14 $r11
+   or $r15 $r14
+   xbit $r6 $flags $p2
+   add b32 $r6 0x208
+   shl b32 $r6 8
+   iowr I[$r6 + 0x000] $r15
+
+   // Op += Tyo << Tp
+   shl b32 $r11 $r7
+   add b32 $r12 $r11
+
+   // nTx = ((w * cpp) + ((1 << Tp) - 1) >> Tp)
+   ld b32 $r15 D[$r5 + ctx_src_xsize]
+   ld b32 $r11 D[$r5 + ctx_src_cpp]
+   mulu $r15 $r11
+   mov $r11 1
+   shl b32 $r11 $r7
+   sub b32 $r11 1
+   add b32 $r15 $r11
+   shr b32 $r15 $r7
+   push $r15
+
+   // nTy = (h + ((1 << Th) - 1)) >> Th
+   ld b32 $r15 D[$r5 + ctx_src_ysize]
+   mov $r11 1
+   shl b32 $r11 $r8
+   sub b32 $r11 1
+   add b32 $r15 $r11
+   shr b32 $r15 $r8
+   push $r15
+
+   // Tys = Tp + Th
+   // CFG_YZ_TILE_SIZE = ((1 << Th) >> 2) << Td
+   add b32 $r7 $r8
+   sub b32 $r8 2
+   mov $r11 1
+   shl b32 $r11 $r8
+   shl b32 $r11 $r9
+
+   // Tzo = z & ((1 << Td) - 1)
+   // Tz  = z >> Td
+   // Op += Tzo << Tys
+   // Ts  = Tys + Td
+   ld b32 $r8 D[$r5 + ctx_src_zoff]
+   mov $r14 1
+   shl b32 $r14 $r9
+   sub b32 $r14 1
+   and $r15 $r8 $r14
+   shl b32 $r15 $r7
+   add b32 $r12 $r15
+   add b32 $r7 $r9
+   shr b32 $r8 $r9
+
+   // Ot = ((Tz * nTy * nTx) + (Ty * nTx) + Tx) << Ts
+   pop $r15
+   pop $r9
+   mulu $r13 $r9
+   add b32 $r10 $r13
+   mulu $r8 $r9
+   mulu $r8 $r15
+   add b32 $r10 $r8
+   shl b32 $r10 $r7
+
+   // PITCH = (nTx - 1) << Ts
+   sub b32 $r9 1
+   shl b32 $r9 $r7
+   iowr I[$r6 + 0x200] $r9
+
+   // SRC_ADDRESS_LOW   = (Ot + Op) & 0xffffffff
+   // CFG_ADDRESS_HIGH |= ((Ot + Op) >> 32) << 16
+   ld b32 $r7 D[$r5 + ctx_src_address_low]
+   ld b32 $r8 D[$r5 + ctx_src_address_high]
+   add b32 $r10 $r12
+   add b32 $r7 $r10
+   adc b32 $r8 0
+   shl b32 $r8 16
+   or $r8 $r11
+   sub b32 $r6 0x600
+   iowr I[$r6 + 0x000] $r7
+   add b32 $r6 0x400
+   iowr I[$r6 + 0x000] $r8
+   ret
+
+// Setup to handle a linear surface
+//
+// Nothing to see here.. Sets ADDRESS and PITCH, pretty non-exciting
+//
+cmd_exec_set_surface_linear:
+   xbit $r6 $flags $p2
+   add b32 $r6 0x202
+   shl b32 $r6 8
+   ld b32 $r7 D[$r5 + ctx_src_address_low]
+   iowr I[$r6 + 0x000] $r7
+   add b32 $r6 0x400
+   ld b32 $r7 D[$r5 + ctx_src_address_high]
+   shl b32 $r7 16
+   iowr I[$r6 + 0x000] $r7
+   add b32 $r6 0x400
+   ld b32 $r7 D[$r5 + ctx_src_pitch]
+   iowr I[$r6 + 0x000] $r7
+   ret
+
+// wait for regs to be available for use
+cmd_exec_wait:
+   push $r0
+   push $r1
+   mov $r0 0x800
+   shl b32 $r0 6
+   loop:
+      iord $r1 I[$r0]
+      and $r1 1
+      bra ne loop
+   pop $r1
+   pop $r0
+   ret
+
+cmd_exec_query:
+   // if QUERY_SHORT not set, write out { -, 0, TIME_LO, TIME_HI }
+   xbit $r4 $r3 13
+   bra ne query_counter
+      call cmd_exec_wait
+      mov $r4 0x80c
+      shl b32 $r4 6
+      ld b32 $r5 D[$r0 + ctx_query_address_low]
+      add b32 $r5 4
+      iowr I[$r4 + 0x000] $r5
+      iowr I[$r4 + 0x100] $r0
+      mov $r5 0xc
+      iowr I[$r4 + 0x200] $r5
+      add b32 $r4 0x400
+      ld b32 $r5 D[$r0 + ctx_query_address_high]
+      shl b32 $r5 16
+      iowr I[$r4 + 0x000] $r5
+      add b32 $r4 0x500
+      mov $r5 0x00000b00
+      sethi $r5 0x00010000
+      iowr I[$r4 + 0x000] $r5
+      mov $r5 0x00004040
+      shl b32 $r5 1
+      sethi $r5 0x80800000
+      iowr I[$r4 + 0x100] $r5
+      mov $r5 0x00001110
+      sethi $r5 0x13120000
+      iowr I[$r4 + 0x200] $r5
+      mov $r5 0x00001514
+      sethi $r5 0x17160000
+      iowr I[$r4 + 0x300] $r5
+      mov $r5 0x00002601
+      sethi $r5 0x00010000
+      mov $r4 0x800
+      shl b32 $r4 6
+      iowr I[$r4 + 0x000] $r5
+
+   // write COUNTER
+   query_counter:
+   call cmd_exec_wait
+   mov $r4 0x80c
+   shl b32 $r4 6
+   ld b32 $r5 D[$r0 + ctx_query_address_low]
+   iowr I[$r4 + 0x000] $r5
+   iowr I[$r4 + 0x100] $r0
+   mov $r5 0x4
+   iowr I[$r4 + 0x200] $r5
+   add b32 $r4 0x400
+   ld b32 $r5 D[$r0 + ctx_query_address_high]
+   shl b32 $r5 16
+   iowr I[$r4 + 0x000] $r5
+   add b32 $r4 0x500
+   mov $r5 0x00000300
+   iowr I[$r4 + 0x000] $r5
+   mov $r5 0x00001110
+   sethi $r5 0x13120000
+   iowr I[$r4 + 0x100] $r5
+   ld b32 $r5 D[$r0 + ctx_query_counter]
+   add b32 $r4 0x500
+   iowr I[$r4 + 0x000] $r5
+   mov $r5 0x00002601
+   sethi $r5 0x00010000
+   mov $r4 0x800
+   shl b32 $r4 6
+   iowr I[$r4 + 0x000] $r5
+   ret
+
+// Execute a copy operation
+//
+// Inputs:
+//    $r1: irqh state
+//    $r2: hostirq state
+//    $r3: data
+//       000002000 QUERY_SHORT
+//       000001000 QUERY
+//       000000100 DST_LINEAR
+//       000000010 SRC_LINEAR
+//       000000001 FORMAT
+//    $r4: dispatch table entry
+// Outputs:
+//    $r1: irqh state
+//    $p1: set on error
+//       $r2: hostirq state
+//       $r3: data
+cmd_exec:
+   call cmd_exec_wait
+
+   // if format requested, call function to calculate it, otherwise
+   // fill in cpp/xcnt for both surfaces as if (cpp == 1)
+   xbit $r15 $r3 0
+   bra e cmd_exec_no_format
+      call cmd_exec_set_format
+      mov $r4 0x200
+      bra cmd_exec_init_src_surface
+   cmd_exec_no_format:
+      mov $r6 0x810
+      shl b32 $r6 6
+      mov $r7 1
+      st b32 D[$r0 + ctx_src_cpp] $r7
+      st b32 D[$r0 + ctx_dst_cpp] $r7
+      ld b32 $r7 D[$r0 + ctx_xcnt]
+      iowr I[$r6 + 0x000] $r7
+      iowr I[$r6 + 0x100] $r7
+      clear b32 $r4
+
+   cmd_exec_init_src_surface:
+   bclr $flags $p2
+   clear b32 $r5
+   xbit $r15 $r3 4
+   bra e src_tiled
+      call cmd_exec_set_surface_linear
+      bra cmd_exec_init_dst_surface
+   src_tiled:
+      call cmd_exec_set_surface_tiled
+      bset $r4 7
+
+   cmd_exec_init_dst_surface:
+   bset $flags $p2
+   mov $r5 ctx_dst_address_high - ctx_src_address_high
+   xbit $r15 $r3 8
+   bra e dst_tiled
+      call cmd_exec_set_surface_linear
+      bra cmd_exec_kick
+   dst_tiled:
+      call cmd_exec_set_surface_tiled
+      bset $r4 8
+
+   cmd_exec_kick:
+   mov $r5 0x800
+   shl b32 $r5 6
+   ld b32 $r6 D[$r0 + ctx_ycnt]
+   iowr I[$r5 + 0x100] $r6
+   mov $r6 0x0041
+   // SRC_TARGET = 1, DST_TARGET = 2
+   sethi $r6 0x44000000
+   or $r4 $r6
+   iowr I[$r5] $r4
+
+   // if requested, queue up a QUERY write after the copy has completed
+   xbit $r15 $r3 12
+   bra e cmd_exec_done
+      call cmd_exec_query
+
+   cmd_exec_done:
+   ret
+
+// Flush write cache
+//
+// Inputs:
+//    $r1: irqh state
+//    $r2: hostirq state
+//    $r3: data
+//    $r4: dispatch table entry
+// Outputs:
+//    $r1: irqh state
+//    $p1: set on error
+//       $r2: hostirq state
+//       $r3: data
+cmd_wrcache_flush:
+   mov $r2 0x2200
+   clear b32 $r3
+   sethi $r3 0x10000
+   iowr I[$r2] $r3
+   ret
+
+.align 0x100
diff --git a/drivers/gpu/drm/nouveau/nva3_copy.fuc.h b/drivers/gpu/drm/nouveau/nva3_copy.fuc.h
new file mode 100644
index 000000000000..2731de22ebe9
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nva3_copy.fuc.h
@@ -0,0 +1,534 @@
+uint32_t nva3_pcopy_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,
+	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,
+	0x00010000,
+	0x00000000,
+	0x00000000,
+	0x00010040,
+	0x00010160,
+	0x00000000,
+	0x00010050,
+	0x00010162,
+	0x00000000,
+	0x00030060,
+	0x00010170,
+	0x00000000,
+	0x00010170,
+	0x00000000,
+	0x00010170,
+	0x00000000,
+	0x00070080,
+	0x00000028,
+	0xfffff000,
+	0x0000002c,
+	0xfff80000,
+	0x00000030,
+	0xffffe000,
+	0x00000034,
+	0xfffff800,
+	0x00000038,
+	0xfffff000,
+	0x0000003c,
+	0xfff80000,
+	0x00000040,
+	0xffffe000,
+	0x00070088,
+	0x00000054,
+	0xfffff000,
+	0x00000058,
+	0xfff80000,
+	0x0000005c,
+	0xffffe000,
+	0x00000060,
+	0xfffff800,
+	0x00000064,
+	0xfffff000,
+	0x00000068,
+	0xfff80000,
+	0x0000006c,
+	0xffffe000,
+	0x000200c0,
+	0x00010492,
+	0x00000000,
+	0x0001051b,
+	0x00000000,
+	0x000e00c3,
+	0x0000001c,
+	0xffffff00,
+	0x00000020,
+	0x0000000f,
+	0x00000048,
+	0xffffff00,
+	0x0000004c,
+	0x0000000f,
+	0x00000024,
+	0xfff80000,
+	0x00000050,
+	0xfff80000,
+	0x00000080,
+	0xffff0000,
+	0x00000084,
+	0xffffe000,
+	0x00000074,
+	0xfccc0000,
+	0x00000078,
+	0x00000000,
+	0x0000007c,
+	0x00000000,
+	0x00000010,
+	0xffffff00,
+	0x00000014,
+	0x00000000,
+	0x00000018,
+	0x00000000,
+	0x00000800,
+};
+
+uint32_t nva3_pcopy_code[] = {
+	0x04fe04bd,
+	0x3517f000,
+	0xf10010fe,
+	0xf1040017,
+	0xf0fff327,
+	0x22d00023,
+	0x0c25f0c0,
+	0xf40012d0,
+	0x17f11031,
+	0x27f01200,
+	0x0012d003,
+	0xf40031f4,
+	0x0ef40028,
+	0x8001cffd,
+	0xf40812c4,
+	0x21f4060b,
+	0x0412c472,
+	0xf4060bf4,
+	0x11c4c321,
+	0x4001d00c,
+	0x47f101f8,
+	0x4bfe7700,
+	0x0007fe00,
+	0xf00204b9,
+	0x01f40643,
+	0x0604fa09,
+	0xfa060ef4,
+	0x03f80504,
+	0x27f100f8,
+	0x23cf1400,
+	0x1e3fc800,
+	0xf4170bf4,
+	0x21f40132,
+	0x1e3af052,
+	0xf00023d0,
+	0x24d00147,
+	0xcf00f880,
+	0x3dc84023,
+	0x220bf41e,
+	0xf40131f4,
+	0x57f05221,
+	0x0367f004,
+	0xa07856bc,
+	0xb6018068,
+	0x87d00884,
+	0x0162b600,
+	0xf0f018f4,
+	0x23d00237,
+	0xf100f880,
+	0xcf190037,
+	0x33cf4032,
+	0xff24e400,
+	0x1024b607,
+	0x010057f1,
+	0x74bd64bd,
+	0x58005658,
+	0x50b60157,
+	0x0446b804,
+	0xbb4d08f4,
+	0x47b80076,
+	0x0f08f404,
+	0xb60276bb,
+	0x57bb0374,
+	0xdf0ef400,
+	0xb60246bb,
+	0x45bb0344,
+	0x01459800,
+	0xb00453fd,
+	0x1bf40054,
+	0x00455820,
+	0xb0014658,
+	0x1bf40064,
+	0x00538009,
+	0xf4300ef4,
+	0x55f90132,
+	0xf40c01f4,
+	0x25f0250e,
+	0x0125f002,
+	0x100047f1,
+	0xd00042d0,
+	0x27f04043,
+	0x0002d040,
+	0xf08002cf,
+	0x24b04024,
+	0xf71bf400,
+	0x1d0027f1,
+	0xd00137f0,
+	0x00f80023,
+	0x27f100f8,
+	0x34bd2200,
+	0xd00233f0,
+	0x00f80023,
+	0x012842b7,
+	0xf00145b6,
+	0x43801e39,
+	0x0040b701,
+	0x0644b606,
+	0xf80043d0,
+	0xf030f400,
+	0xb00001b0,
+	0x01b00101,
+	0x0301b002,
+	0xc71d0498,
+	0x50b63045,
+	0x3446c701,
+	0xc70160b6,
+	0x70b63847,
+	0x0232f401,
+	0x94bd84bd,
+	0xb60f4ac4,
+	0xb4bd0445,
+	0xf404a430,
+	0xa5ff0f18,
+	0x00cbbbc0,
+	0xf40231f4,
+	0x1bf4220e,
+	0x10c7f00c,
+	0xf400cbbb,
+	0xa430160e,
+	0x0c18f406,
+	0xbb14c7f0,
+	0x0ef400cb,
+	0x80c7f107,
+	0x01c83800,
+	0xb60180b6,
+	0xb5b801b0,
+	0xc308f404,
+	0xb80190b6,
+	0x08f40497,
+	0x0065fdb2,
+	0x98110680,
+	0x68fd2008,
+	0x0502f400,
+	0x75fd64bd,
+	0x1c078000,
+	0xf10078fd,
+	0xb6081057,
+	0x56d00654,
+	0x4057d000,
+	0x080050b7,
+	0xb61c0698,
+	0x64b60162,
+	0x11079808,
+	0xfd0172b6,
+	0x56d00567,
+	0x0050b700,
+	0x0060b401,
+	0xb40056d0,
+	0x56d00160,
+	0x0260b440,
+	0xb48056d0,
+	0x56d00360,
+	0x0050b7c0,
+	0x1e069804,
+	0x980056d0,
+	0x56d01f06,
+	0x1030f440,
+	0x579800f8,
+	0x6879c70a,
+	0xb66478c7,
+	0x77c70280,
+	0x0e76b060,
+	0xf0091bf4,
+	0x0ef40477,
+	0x027cf00f,
+	0xfd1170b6,
+	0x77f00947,
+	0x0f5a9806,
+	0xfd115b98,
+	0xb7f000ab,
+	0x04b7bb01,
+	0xff01b2b6,
+	0xa7bbc4ab,
+	0x105d9805,
+	0xbb01e7f0,
+	0xe2b604e8,
+	0xb4deff01,
+	0xb605d8bb,
+	0xef9401e0,
+	0x02ebbb0c,
+	0xf005fefd,
+	0x60b7026c,
+	0x64b60208,
+	0x006fd008,
+	0xbb04b7bb,
+	0x5f9800cb,
+	0x115b980b,
+	0xf000fbfd,
+	0xb7bb01b7,
+	0x01b2b604,
+	0xbb00fbbb,
+	0xf0f905f7,
+	0xf00c5f98,
+	0xb8bb01b7,
+	0x01b2b604,
+	0xbb00fbbb,
+	0xf0f905f8,
+	0xb60078bb,
+	0xb7f00282,
+	0x04b8bb01,
+	0x9804b9bb,
+	0xe7f00e58,
+	0x04e9bb01,
+	0xff01e2b6,
+	0xf7bbf48e,
+	0x00cfbb04,
+	0xbb0079bb,
+	0xf0fc0589,
+	0xd9fd90fc,
+	0x00adbb00,
+	0xfd0089fd,
+	0xa8bb008f,
+	0x04a7bb00,
+	0xbb0192b6,
+	0x69d00497,
+	0x08579880,
+	0xbb075898,
+	0x7abb00ac,
+	0x0081b600,
+	0xfd1084b6,
+	0x62b7058b,
+	0x67d00600,
+	0x0060b700,
+	0x0068d004,
+	0x6cf000f8,
+	0x0260b702,
+	0x0864b602,
+	0xd0085798,
+	0x60b70067,
+	0x57980400,
+	0x1074b607,
+	0xb70067d0,
+	0x98040060,
+	0x67d00957,
+	0xf900f800,
+	0xf110f900,
+	0xb6080007,
+	0x01cf0604,
+	0x0114f000,
+	0xfcfa1bf4,
+	0xf800fc10,
+	0x0d34c800,
+	0xf5701bf4,
+	0xf103ab21,
+	0xb6080c47,
+	0x05980644,
+	0x0450b605,
+	0xd00045d0,
+	0x57f04040,
+	0x8045d00c,
+	0x040040b7,
+	0xb6040598,
+	0x45d01054,
+	0x0040b700,
+	0x0057f105,
+	0x0153f00b,
+	0xf10045d0,
+	0xb6404057,
+	0x53f10154,
+	0x45d08080,
+	0x1057f140,
+	0x1253f111,
+	0x8045d013,
+	0x151457f1,
+	0x171653f1,
+	0xf1c045d0,
+	0xf0260157,
+	0x47f10153,
+	0x44b60800,
+	0x0045d006,
+	0x03ab21f5,
+	0x080c47f1,
+	0x980644b6,
+	0x45d00505,
+	0x4040d000,
+	0xd00457f0,
+	0x40b78045,
+	0x05980400,
+	0x1054b604,
+	0xb70045d0,
+	0xf1050040,
+	0xd0030057,
+	0x57f10045,
+	0x53f11110,
+	0x45d01312,
+	0x06059840,
+	0x050040b7,
+	0xf10045d0,
+	0xf0260157,
+	0x47f10153,
+	0x44b60800,
+	0x0045d006,
+	0x21f500f8,
+	0x3fc803ab,
+	0x0e0bf400,
+	0x018921f5,
+	0x020047f1,
+	0xf11e0ef4,
+	0xb6081067,
+	0x77f00664,
+	0x11078001,
+	0x981c0780,
+	0x67d02007,
+	0x4067d000,
+	0x32f444bd,
+	0xc854bd02,
+	0x0bf4043f,
+	0x8221f50a,
+	0x0a0ef403,
+	0x027621f5,
+	0xf40749f0,
+	0x57f00231,
+	0x083fc82c,
+	0xf50a0bf4,
+	0xf4038221,
+	0x21f50a0e,
+	0x49f00276,
+	0x0057f108,
+	0x0654b608,
+	0xd0210698,
+	0x67f04056,
+	0x0063f141,
+	0x0546fd44,
+	0xc80054d0,
+	0x0bf40c3f,
+	0xc521f507,
+	0xf100f803,
+	0xbd220027,
+	0x0133f034,
+	0xf80023d0,
+	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,
+};
diff --git a/drivers/gpu/drm/nouveau/nva3_pm.c b/drivers/gpu/drm/nouveau/nva3_pm.c
index dbbafed36406..e4b2b9e934b2 100644
--- a/drivers/gpu/drm/nouveau/nva3_pm.c
+++ b/drivers/gpu/drm/nouveau/nva3_pm.c
@@ -27,32 +27,74 @@
 #include "nouveau_bios.h"
 #include "nouveau_pm.h"
 
-/*XXX: boards using limits 0x40 need fixing, the register layout
- *     is correct here, but, there's some other funny magic
- *     that modifies things, so it's not likely we'll set/read
- *     the correct timings yet..  working on it...
+/* This is actually a lot more complex than it appears here, but hopefully
+ * this should be able to deal with what the VBIOS leaves for us..
+ *
+ * If not, well, I'll jump off that bridge when I come to it.
  */
 
 struct nva3_pm_state {
-	struct pll_lims pll;
-	int N, M, P;
+	enum pll_types type;
+	u32 src0;
+	u32 src1;
+	u32 ctrl;
+	u32 coef;
+	u32 old_pnm;
+	u32 new_pnm;
+	u32 new_div;
 };
 
+static int
+nva3_pm_pll_offset(u32 id)
+{
+	static const u32 pll_map[] = {
+		0x00, PLL_CORE,
+		0x01, PLL_SHADER,
+		0x02, PLL_MEMORY,
+		0x00, 0x00
+	};
+	const u32 *map = pll_map;
+
+	while (map[1]) {
+		if (id == map[1])
+			return map[0];
+		map += 2;
+	}
+
+	return -ENOENT;
+}
+
 int
 nva3_pm_clock_get(struct drm_device *dev, u32 id)
 {
+	u32 src0, src1, ctrl, coef;
 	struct pll_lims pll;
-	int P, N, M, ret;
-	u32 reg;
+	int ret, off;
+	int P, N, M;
 
 	ret = get_pll_limits(dev, id, &pll);
 	if (ret)
 		return ret;
 
-	reg = nv_rd32(dev, pll.reg + 4);
-	P = (reg & 0x003f0000) >> 16;
-	N = (reg & 0x0000ff00) >> 8;
-	M = (reg & 0x000000ff);
+	off = nva3_pm_pll_offset(id);
+	if (off < 0)
+		return off;
+
+	src0 = nv_rd32(dev, 0x4120 + (off * 4));
+	src1 = nv_rd32(dev, 0x4160 + (off * 4));
+	ctrl = nv_rd32(dev, pll.reg + 0);
+	coef = nv_rd32(dev, pll.reg + 4);
+	NV_DEBUG(dev, "PLL %02x: 0x%08x 0x%08x 0x%08x 0x%08x\n",
+		      id, src0, src1, ctrl, coef);
+
+	if (ctrl & 0x00000008) {
+		u32 div = ((src1 & 0x003c0000) >> 18) + 1;
+		return (pll.refclk * 2) / div;
+	}
+
+	P = (coef & 0x003f0000) >> 16;
+	N = (coef & 0x0000ff00) >> 8;
+	M = (coef & 0x000000ff);
 	return pll.refclk * N / M / P;
 }
 
@@ -60,36 +102,103 @@ void *
 nva3_pm_clock_pre(struct drm_device *dev, struct nouveau_pm_level *perflvl,
 		  u32 id, int khz)
 {
-	struct nva3_pm_state *state;
-	int dummy, ret;
+	struct nva3_pm_state *pll;
+	struct pll_lims limits;
+	int N, M, P, diff;
+	int ret, off;
+
+	ret = get_pll_limits(dev, id, &limits);
+	if (ret < 0)
+		return (ret == -ENOENT) ? NULL : ERR_PTR(ret);
+
+	off = nva3_pm_pll_offset(id);
+	if (id < 0)
+		return ERR_PTR(-EINVAL);
 
-	state = kzalloc(sizeof(*state), GFP_KERNEL);
-	if (!state)
+
+	pll = kzalloc(sizeof(*pll), GFP_KERNEL);
+	if (!pll)
 		return ERR_PTR(-ENOMEM);
+	pll->type = id;
+	pll->src0 = 0x004120 + (off * 4);
+	pll->src1 = 0x004160 + (off * 4);
+	pll->ctrl = limits.reg + 0;
+	pll->coef = limits.reg + 4;
 
-	ret = get_pll_limits(dev, id, &state->pll);
-	if (ret < 0) {
-		kfree(state);
-		return (ret == -ENOENT) ? NULL : ERR_PTR(ret);
+	/* If target clock is within [-2, 3) MHz of a divisor, we'll
+	 * use that instead of calculating MNP values
+	 */
+	pll->new_div = min((limits.refclk * 2) / (khz - 2999), 16);
+	if (pll->new_div) {
+		diff = khz - ((limits.refclk * 2) / pll->new_div);
+		if (diff < -2000 || diff >= 3000)
+			pll->new_div = 0;
 	}
 
-	ret = nv50_calc_pll2(dev, &state->pll, khz, &state->N, &dummy,
-			     &state->M, &state->P);
-	if (ret < 0) {
-		kfree(state);
-		return ERR_PTR(ret);
+	if (!pll->new_div) {
+		ret = nva3_calc_pll(dev, &limits, khz, &N, NULL, &M, &P);
+		if (ret < 0)
+			return ERR_PTR(ret);
+
+		pll->new_pnm = (P << 16) | (N << 8) | M;
+		pll->new_div = 2 - 1;
+	} else {
+		pll->new_pnm = 0;
+		pll->new_div--;
 	}
 
-	return state;
+	if ((nv_rd32(dev, pll->src1) & 0x00000101) != 0x00000101)
+		pll->old_pnm = nv_rd32(dev, pll->coef);
+	return pll;
 }
 
 void
 nva3_pm_clock_set(struct drm_device *dev, void *pre_state)
 {
-	struct nva3_pm_state *state = pre_state;
-	u32 reg = state->pll.reg;
+	struct nva3_pm_state *pll = pre_state;
+	u32 ctrl = 0;
+
+	/* For the memory clock, NVIDIA will build a "script" describing
+	 * the reclocking process and ask PDAEMON to execute it.
+	 */
+	if (pll->type == PLL_MEMORY) {
+		nv_wr32(dev, 0x100210, 0);
+		nv_wr32(dev, 0x1002dc, 1);
+		nv_wr32(dev, 0x004018, 0x00001000);
+		ctrl = 0x18000100;
+	}
+
+	if (pll->old_pnm || !pll->new_pnm) {
+		nv_mask(dev, pll->src1, 0x003c0101, 0x00000101 |
+						    (pll->new_div << 18));
+		nv_wr32(dev, pll->ctrl, 0x0001001d | ctrl);
+		nv_mask(dev, pll->ctrl, 0x00000001, 0x00000000);
+	}
+
+	if (pll->new_pnm) {
+		nv_mask(dev, pll->src0, 0x00000101, 0x00000101);
+		nv_wr32(dev, pll->coef, pll->new_pnm);
+		nv_wr32(dev, pll->ctrl, 0x0001001d | ctrl);
+		nv_mask(dev, pll->ctrl, 0x00000010, 0x00000000);
+		nv_mask(dev, pll->ctrl, 0x00020010, 0x00020010);
+		nv_wr32(dev, pll->ctrl, 0x00010015 | ctrl);
+		nv_mask(dev, pll->src1, 0x00000100, 0x00000000);
+		nv_mask(dev, pll->src1, 0x00000001, 0x00000000);
+		if (pll->type == PLL_MEMORY)
+			nv_wr32(dev, 0x4018, 0x10005000);
+	} else {
+		nv_mask(dev, pll->ctrl, 0x00000001, 0x00000000);
+		nv_mask(dev, pll->src0, 0x00000100, 0x00000000);
+		nv_mask(dev, pll->src0, 0x00000001, 0x00000000);
+		if (pll->type == PLL_MEMORY)
+			nv_wr32(dev, 0x4018, 0x1000d000);
+	}
+
+	if (pll->type == PLL_MEMORY) {
+		nv_wr32(dev, 0x1002dc, 0);
+		nv_wr32(dev, 0x100210, 0x80000000);
+	}
 
-	nv_wr32(dev, reg + 4, (state->P << 16) | (state->N << 8) | state->M);
-	kfree(state);
+	kfree(pll);
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvc0_copy.c b/drivers/gpu/drm/nouveau/nvc0_copy.c
new file mode 100644
index 000000000000..208fa7ab3f42
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvc0_copy.c
@@ -0,0 +1,243 @@
+/*
+ * 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
+ */
+
+#include <linux/firmware.h>
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_util.h"
+#include "nouveau_vm.h"
+#include "nouveau_ramht.h"
+#include "nvc0_copy.fuc.h"
+
+struct nvc0_copy_engine {
+	struct nouveau_exec_engine base;
+	u32 irq;
+	u32 pmc;
+	u32 fuc;
+	u32 ctx;
+};
+
+static int
+nvc0_copy_context_new(struct nouveau_channel *chan, int engine)
+{
+	struct nvc0_copy_engine *pcopy = nv_engine(chan->dev, engine);
+	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *ramin = chan->ramin;
+	struct nouveau_gpuobj *ctx = NULL;
+	int ret;
+
+	ret = nouveau_gpuobj_new(dev, NULL, 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));
+	dev_priv->engine.instmem.flush(dev);
+
+	chan->engctx[engine] = ctx;
+	return 0;
+}
+
+static int
+nvc0_copy_object_new(struct nouveau_channel *chan, int engine,
+		     u32 handle, u16 class)
+{
+	return 0;
+}
+
+static void
+nvc0_copy_context_del(struct nouveau_channel *chan, int engine)
+{
+	struct nvc0_copy_engine *pcopy = nv_engine(chan->dev, engine);
+	struct nouveau_gpuobj *ctx = chan->engctx[engine];
+	struct drm_device *dev = chan->dev;
+	u32 inst;
+
+	inst  = (chan->ramin->vinst >> 12);
+	inst |= 0x40000000;
+
+	/* disable fifo access */
+	nv_wr32(dev, pcopy->fuc + 0x048, 0x00000000);
+	/* mark channel as unloaded if it's currently active */
+	if (nv_rd32(dev, pcopy->fuc + 0x050) == inst)
+		nv_mask(dev, pcopy->fuc + 0x050, 0x40000000, 0x00000000);
+	/* mark next channel as invalid if it's about to be loaded */
+	if (nv_rd32(dev, pcopy->fuc + 0x054) == inst)
+		nv_mask(dev, pcopy->fuc + 0x054, 0x40000000, 0x00000000);
+	/* restore fifo access */
+	nv_wr32(dev, pcopy->fuc + 0x048, 0x00000003);
+
+	nv_wo32(chan->ramin, pcopy->ctx + 0, 0x00000000);
+	nv_wo32(chan->ramin, pcopy->ctx + 4, 0x00000000);
+	nouveau_gpuobj_ref(NULL, &ctx);
+
+	chan->engctx[engine] = ctx;
+}
+
+static int
+nvc0_copy_init(struct drm_device *dev, int engine)
+{
+	struct nvc0_copy_engine *pcopy = nv_engine(dev, engine);
+	int i;
+
+	nv_mask(dev, 0x000200, pcopy->pmc, 0x00000000);
+	nv_mask(dev, 0x000200, pcopy->pmc, pcopy->pmc);
+	nv_wr32(dev, pcopy->fuc + 0x014, 0xffffffff);
+
+	nv_wr32(dev, pcopy->fuc + 0x1c0, 0x01000000);
+	for (i = 0; i < sizeof(nvc0_pcopy_data) / 4; i++)
+		nv_wr32(dev, pcopy->fuc + 0x1c4, nvc0_pcopy_data[i]);
+
+	nv_wr32(dev, pcopy->fuc + 0x180, 0x01000000);
+	for (i = 0; i < sizeof(nvc0_pcopy_code) / 4; i++) {
+		if ((i & 0x3f) == 0)
+			nv_wr32(dev, pcopy->fuc + 0x188, i >> 6);
+		nv_wr32(dev, pcopy->fuc + 0x184, nvc0_pcopy_code[i]);
+	}
+
+	nv_wr32(dev, pcopy->fuc + 0x084, engine - NVOBJ_ENGINE_COPY0);
+	nv_wr32(dev, pcopy->fuc + 0x10c, 0x00000000);
+	nv_wr32(dev, pcopy->fuc + 0x104, 0x00000000); /* ENTRY */
+	nv_wr32(dev, pcopy->fuc + 0x100, 0x00000002); /* TRIGGER */
+	return 0;
+}
+
+static int
+nvc0_copy_fini(struct drm_device *dev, int engine)
+{
+	struct nvc0_copy_engine *pcopy = nv_engine(dev, engine);
+
+	nv_mask(dev, pcopy->fuc + 0x048, 0x00000003, 0x00000000);
+
+	/* trigger fuc context unload */
+	nv_wait(dev, pcopy->fuc + 0x008, 0x0000000c, 0x00000000);
+	nv_mask(dev, pcopy->fuc + 0x054, 0x40000000, 0x00000000);
+	nv_wr32(dev, pcopy->fuc + 0x000, 0x00000008);
+	nv_wait(dev, pcopy->fuc + 0x008, 0x00000008, 0x00000000);
+
+	nv_wr32(dev, pcopy->fuc + 0x014, 0xffffffff);
+	return 0;
+}
+
+static struct nouveau_enum nvc0_copy_isr_error_name[] = {
+	{ 0x0001, "ILLEGAL_MTHD" },
+	{ 0x0002, "INVALID_ENUM" },
+	{ 0x0003, "INVALID_BITFIELD" },
+	{}
+};
+
+static void
+nvc0_copy_isr(struct drm_device *dev, int engine)
+{
+	struct nvc0_copy_engine *pcopy = nv_engine(dev, engine);
+	u32 disp = nv_rd32(dev, pcopy->fuc + 0x01c);
+	u32 stat = nv_rd32(dev, pcopy->fuc + 0x008) & disp & ~(disp >> 16);
+	u64 inst = (u64)(nv_rd32(dev, pcopy->fuc + 0x050) & 0x0fffffff) << 12;
+	u32 chid = nvc0_graph_isr_chid(dev, inst);
+	u32 ssta = nv_rd32(dev, pcopy->fuc + 0x040) & 0x0000ffff;
+	u32 addr = nv_rd32(dev, pcopy->fuc + 0x040) >> 16;
+	u32 mthd = (addr & 0x07ff) << 2;
+	u32 subc = (addr & 0x3800) >> 11;
+	u32 data = nv_rd32(dev, pcopy->fuc + 0x044);
+
+	if (stat & 0x00000040) {
+		NV_INFO(dev, "PCOPY: DISPATCH_ERROR [");
+		nouveau_enum_print(nvc0_copy_isr_error_name, ssta);
+		printk("] ch %d [0x%010llx] subc %d mthd 0x%04x data 0x%08x\n",
+			chid, inst, subc, mthd, data);
+		nv_wr32(dev, pcopy->fuc + 0x004, 0x00000040);
+		stat &= ~0x00000040;
+	}
+
+	if (stat) {
+		NV_INFO(dev, "PCOPY: unhandled intr 0x%08x\n", stat);
+		nv_wr32(dev, pcopy->fuc + 0x004, stat);
+	}
+}
+
+static void
+nvc0_copy_isr_0(struct drm_device *dev)
+{
+	nvc0_copy_isr(dev, NVOBJ_ENGINE_COPY0);
+}
+
+static void
+nvc0_copy_isr_1(struct drm_device *dev)
+{
+	nvc0_copy_isr(dev, NVOBJ_ENGINE_COPY1);
+}
+
+static void
+nvc0_copy_destroy(struct drm_device *dev, int engine)
+{
+	struct nvc0_copy_engine *pcopy = nv_engine(dev, engine);
+
+	nouveau_irq_unregister(dev, pcopy->irq);
+
+	if (engine == NVOBJ_ENGINE_COPY0)
+		NVOBJ_ENGINE_DEL(dev, COPY0);
+	else
+		NVOBJ_ENGINE_DEL(dev, COPY1);
+	kfree(pcopy);
+}
+
+int
+nvc0_copy_create(struct drm_device *dev, int engine)
+{
+	struct nvc0_copy_engine *pcopy;
+
+	pcopy = kzalloc(sizeof(*pcopy), GFP_KERNEL);
+	if (!pcopy)
+		return -ENOMEM;
+
+	pcopy->base.destroy = nvc0_copy_destroy;
+	pcopy->base.init = nvc0_copy_init;
+	pcopy->base.fini = nvc0_copy_fini;
+	pcopy->base.context_new = nvc0_copy_context_new;
+	pcopy->base.context_del = nvc0_copy_context_del;
+	pcopy->base.object_new = nvc0_copy_object_new;
+
+	if (engine == 0) {
+		pcopy->irq = 5;
+		pcopy->pmc = 0x00000040;
+		pcopy->fuc = 0x104000;
+		pcopy->ctx = 0x0230;
+		nouveau_irq_register(dev, pcopy->irq, nvc0_copy_isr_0);
+		NVOBJ_ENGINE_ADD(dev, COPY0, &pcopy->base);
+		NVOBJ_CLASS(dev, 0x90b5, COPY0);
+	} else {
+		pcopy->irq = 6;
+		pcopy->pmc = 0x00000080;
+		pcopy->fuc = 0x105000;
+		pcopy->ctx = 0x0240;
+		nouveau_irq_register(dev, pcopy->irq, nvc0_copy_isr_1);
+		NVOBJ_ENGINE_ADD(dev, COPY1, &pcopy->base);
+		NVOBJ_CLASS(dev, 0x90b8, COPY1);
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nvc0_copy.fuc.h b/drivers/gpu/drm/nouveau/nvc0_copy.fuc.h
new file mode 100644
index 000000000000..419903880e9d
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvc0_copy.fuc.h
@@ -0,0 +1,527 @@
+uint32_t nvc0_pcopy_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,
+	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,
+	0x00010000,
+	0x00000000,
+	0x00000000,
+	0x00010040,
+	0x0001019f,
+	0x00000000,
+	0x00010050,
+	0x000101a1,
+	0x00000000,
+	0x00070080,
+	0x0000001c,
+	0xfffff000,
+	0x00000020,
+	0xfff80000,
+	0x00000024,
+	0xffffe000,
+	0x00000028,
+	0xfffff800,
+	0x0000002c,
+	0xfffff000,
+	0x00000030,
+	0xfff80000,
+	0x00000034,
+	0xffffe000,
+	0x00070088,
+	0x00000048,
+	0xfffff000,
+	0x0000004c,
+	0xfff80000,
+	0x00000050,
+	0xffffe000,
+	0x00000054,
+	0xfffff800,
+	0x00000058,
+	0xfffff000,
+	0x0000005c,
+	0xfff80000,
+	0x00000060,
+	0xffffe000,
+	0x000200c0,
+	0x000104b8,
+	0x00000000,
+	0x00010541,
+	0x00000000,
+	0x000e00c3,
+	0x00000010,
+	0xffffff00,
+	0x00000014,
+	0x0000000f,
+	0x0000003c,
+	0xffffff00,
+	0x00000040,
+	0x0000000f,
+	0x00000018,
+	0xfff80000,
+	0x00000044,
+	0xfff80000,
+	0x00000074,
+	0xffff0000,
+	0x00000078,
+	0xffffe000,
+	0x00000068,
+	0xfccc0000,
+	0x0000006c,
+	0x00000000,
+	0x00000070,
+	0x00000000,
+	0x00000004,
+	0xffffff00,
+	0x00000008,
+	0x00000000,
+	0x0000000c,
+	0x00000000,
+	0x00000800,
+};
+
+uint32_t nvc0_pcopy_code[] = {
+	0x04fe04bd,
+	0x3517f000,
+	0xf10010fe,
+	0xf1040017,
+	0xf0fff327,
+	0x22d00023,
+	0x0c25f0c0,
+	0xf40012d0,
+	0x17f11031,
+	0x27f01200,
+	0x0012d003,
+	0xf40031f4,
+	0x0ef40028,
+	0x8001cffd,
+	0xf40812c4,
+	0x21f4060b,
+	0x0412c4ca,
+	0xf5070bf4,
+	0xc4010221,
+	0x01d00c11,
+	0xf101f840,
+	0xfe770047,
+	0x47f1004b,
+	0x44cf2100,
+	0x0144f000,
+	0xb60444b6,
+	0xf7f13040,
+	0xf4b6061c,
+	0x1457f106,
+	0x00f5d101,
+	0xb6043594,
+	0x57fe0250,
+	0x0145fe00,
+	0x010052b7,
+	0x00ff67f1,
+	0x56fd60bd,
+	0x0253f004,
+	0xf80545fa,
+	0x0053f003,
+	0xd100e7f0,
+	0x549800fe,
+	0x0845b600,
+	0xb6015698,
+	0x46fd1864,
+	0x0047fe05,
+	0xf00204b9,
+	0x01f40643,
+	0x0604fa09,
+	0xfa060ef4,
+	0x03f80504,
+	0x27f100f8,
+	0x23cf1400,
+	0x1e3fc800,
+	0xf4170bf4,
+	0x21f40132,
+	0x1e3af053,
+	0xf00023d0,
+	0x24d00147,
+	0xcf00f880,
+	0x3dc84023,
+	0x090bf41e,
+	0xf40131f4,
+	0x37f05321,
+	0x8023d002,
+	0x37f100f8,
+	0x32cf1900,
+	0x0033cf40,
+	0x07ff24e4,
+	0xf11024b6,
+	0xbd010057,
+	0x5874bd64,
+	0x57580056,
+	0x0450b601,
+	0xf40446b8,
+	0x76bb4d08,
+	0x0447b800,
+	0xbb0f08f4,
+	0x74b60276,
+	0x0057bb03,
+	0xbbdf0ef4,
+	0x44b60246,
+	0x0045bb03,
+	0xfd014598,
+	0x54b00453,
+	0x201bf400,
+	0x58004558,
+	0x64b00146,
+	0x091bf400,
+	0xf4005380,
+	0x32f4300e,
+	0xf455f901,
+	0x0ef40c01,
+	0x0225f025,
+	0xf10125f0,
+	0xd0100047,
+	0x43d00042,
+	0x4027f040,
+	0xcf0002d0,
+	0x24f08002,
+	0x0024b040,
+	0xf1f71bf4,
+	0xf01d0027,
+	0x23d00137,
+	0xf800f800,
+	0x0027f100,
+	0xf034bd22,
+	0x23d00233,
+	0xf400f800,
+	0x01b0f030,
+	0x0101b000,
+	0xb00201b0,
+	0x04980301,
+	0x3045c71a,
+	0xc70150b6,
+	0x60b63446,
+	0x3847c701,
+	0xf40170b6,
+	0x84bd0232,
+	0x4ac494bd,
+	0x0445b60f,
+	0xa430b4bd,
+	0x0f18f404,
+	0xbbc0a5ff,
+	0x31f400cb,
+	0x220ef402,
+	0xf00c1bf4,
+	0xcbbb10c7,
+	0x160ef400,
+	0xf406a430,
+	0xc7f00c18,
+	0x00cbbb14,
+	0xf1070ef4,
+	0x380080c7,
+	0x80b601c8,
+	0x01b0b601,
+	0xf404b5b8,
+	0x90b6c308,
+	0x0497b801,
+	0xfdb208f4,
+	0x06800065,
+	0x1d08980e,
+	0xf40068fd,
+	0x64bd0502,
+	0x800075fd,
+	0x78fd1907,
+	0x1057f100,
+	0x0654b608,
+	0xd00056d0,
+	0x50b74057,
+	0x06980800,
+	0x0162b619,
+	0x980864b6,
+	0x72b60e07,
+	0x0567fd01,
+	0xb70056d0,
+	0xb4010050,
+	0x56d00060,
+	0x0160b400,
+	0xb44056d0,
+	0x56d00260,
+	0x0360b480,
+	0xb7c056d0,
+	0x98040050,
+	0x56d01b06,
+	0x1c069800,
+	0xf44056d0,
+	0x00f81030,
+	0xc7075798,
+	0x78c76879,
+	0x0380b664,
+	0xb06077c7,
+	0x1bf40e76,
+	0x0477f009,
+	0xf00f0ef4,
+	0x70b6027c,
+	0x0947fd11,
+	0x980677f0,
+	0x5b980c5a,
+	0x00abfd0e,
+	0xbb01b7f0,
+	0xb2b604b7,
+	0xc4abff01,
+	0x9805a7bb,
+	0xe7f00d5d,
+	0x04e8bb01,
+	0xff01e2b6,
+	0xd8bbb4de,
+	0x01e0b605,
+	0xbb0cef94,
+	0xfefd02eb,
+	0x026cf005,
+	0x020860b7,
+	0xd00864b6,
+	0xb7bb006f,
+	0x00cbbb04,
+	0x98085f98,
+	0xfbfd0e5b,
+	0x01b7f000,
+	0xb604b7bb,
+	0xfbbb01b2,
+	0x05f7bb00,
+	0x5f98f0f9,
+	0x01b7f009,
+	0xb604b8bb,
+	0xfbbb01b2,
+	0x05f8bb00,
+	0x78bbf0f9,
+	0x0282b600,
+	0xbb01b7f0,
+	0xb9bb04b8,
+	0x0b589804,
+	0xbb01e7f0,
+	0xe2b604e9,
+	0xf48eff01,
+	0xbb04f7bb,
+	0x79bb00cf,
+	0x0589bb00,
+	0x90fcf0fc,
+	0xbb00d9fd,
+	0x89fd00ad,
+	0x008ffd00,
+	0xbb00a8bb,
+	0x92b604a7,
+	0x0497bb01,
+	0x988069d0,
+	0x58980557,
+	0x00acbb04,
+	0xb6007abb,
+	0x84b60081,
+	0x058bfd10,
+	0x060062b7,
+	0xb70067d0,
+	0xd0040060,
+	0x00f80068,
+	0xb7026cf0,
+	0xb6020260,
+	0x57980864,
+	0x0067d005,
+	0x040060b7,
+	0xb6045798,
+	0x67d01074,
+	0x0060b700,
+	0x06579804,
+	0xf80067d0,
+	0xf900f900,
+	0x0007f110,
+	0x0604b608,
+	0xf00001cf,
+	0x1bf40114,
+	0xfc10fcfa,
+	0xc800f800,
+	0x1bf40d34,
+	0xd121f570,
+	0x0c47f103,
+	0x0644b608,
+	0xb6020598,
+	0x45d00450,
+	0x4040d000,
+	0xd00c57f0,
+	0x40b78045,
+	0x05980400,
+	0x1054b601,
+	0xb70045d0,
+	0xf1050040,
+	0xf00b0057,
+	0x45d00153,
+	0x4057f100,
+	0x0154b640,
+	0x808053f1,
+	0xf14045d0,
+	0xf1111057,
+	0xd0131253,
+	0x57f18045,
+	0x53f11514,
+	0x45d01716,
+	0x0157f1c0,
+	0x0153f026,
+	0x080047f1,
+	0xd00644b6,
+	0x21f50045,
+	0x47f103d1,
+	0x44b6080c,
+	0x02059806,
+	0xd00045d0,
+	0x57f04040,
+	0x8045d004,
+	0x040040b7,
+	0xb6010598,
+	0x45d01054,
+	0x0040b700,
+	0x0057f105,
+	0x0045d003,
+	0x111057f1,
+	0x131253f1,
+	0x984045d0,
+	0x40b70305,
+	0x45d00500,
+	0x0157f100,
+	0x0153f026,
+	0x080047f1,
+	0xd00644b6,
+	0x00f80045,
+	0x03d121f5,
+	0xf4003fc8,
+	0x21f50e0b,
+	0x47f101af,
+	0x0ef40200,
+	0x1067f11e,
+	0x0664b608,
+	0x800177f0,
+	0x07800e07,
+	0x1d079819,
+	0xd00067d0,
+	0x44bd4067,
+	0xbd0232f4,
+	0x043fc854,
+	0xf50a0bf4,
+	0xf403a821,
+	0x21f50a0e,
+	0x49f0029c,
+	0x0231f407,
+	0xc82c57f0,
+	0x0bf4083f,
+	0xa821f50a,
+	0x0a0ef403,
+	0x029c21f5,
+	0xf10849f0,
+	0xb6080057,
+	0x06980654,
+	0x4056d01e,
+	0xf14167f0,
+	0xfd440063,
+	0x54d00546,
+	0x0c3fc800,
+	0xf5070bf4,
+	0xf803eb21,
+	0x0027f100,
+	0xf034bd22,
+	0x23d00133,
+	0x0000f800,
+	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,
+};
diff --git a/drivers/gpu/drm/nouveau/nvc0_fifo.c b/drivers/gpu/drm/nouveau/nvc0_fifo.c
index 2886f2726a9e..fb4f5943e01b 100644
--- a/drivers/gpu/drm/nouveau/nvc0_fifo.c
+++ b/drivers/gpu/drm/nouveau/nvc0_fifo.c
@@ -37,7 +37,7 @@ struct nvc0_fifo_priv {
 };
 
 struct nvc0_fifo_chan {
-	struct nouveau_bo *user;
+	struct nouveau_gpuobj *user;
 	struct nouveau_gpuobj *ramfc;
 };
 
@@ -106,7 +106,7 @@ nvc0_fifo_create_context(struct nouveau_channel *chan)
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
 	struct nvc0_fifo_priv *priv = pfifo->priv;
 	struct nvc0_fifo_chan *fifoch;
-	u64 ib_virt, user_vinst;
+	u64 ib_virt = chan->pushbuf_base + chan->dma.ib_base * 4;
 	int ret;
 
 	chan->fifo_priv = kzalloc(sizeof(*fifoch), GFP_KERNEL);
@@ -115,28 +115,13 @@ nvc0_fifo_create_context(struct nouveau_channel *chan)
 	fifoch = chan->fifo_priv;
 
 	/* allocate vram for control regs, map into polling area */
-	ret = nouveau_bo_new(dev, NULL, 0x1000, 0, TTM_PL_FLAG_VRAM,
-			     0, 0, &fifoch->user);
+	ret = nouveau_gpuobj_new(dev, NULL, 0x1000, 0x1000,
+				 NVOBJ_FLAG_ZERO_ALLOC, &fifoch->user);
 	if (ret)
 		goto error;
 
-	ret = nouveau_bo_pin(fifoch->user, TTM_PL_FLAG_VRAM);
-	if (ret) {
-		nouveau_bo_ref(NULL, &fifoch->user);
-		goto error;
-	}
-
-	user_vinst = fifoch->user->bo.mem.start << PAGE_SHIFT;
-
-	ret = nouveau_bo_map(fifoch->user);
-	if (ret) {
-		nouveau_bo_unpin(fifoch->user);
-		nouveau_bo_ref(NULL, &fifoch->user);
-		goto error;
-	}
-
 	nouveau_vm_map_at(&priv->user_vma, chan->id * 0x1000,
-			  fifoch->user->bo.mem.mm_node);
+			  *(struct nouveau_mem **)fifoch->user->node);
 
 	chan->user = ioremap_wc(pci_resource_start(dev->pdev, 1) +
 				priv->user_vma.offset + (chan->id * 0x1000),
@@ -146,20 +131,6 @@ nvc0_fifo_create_context(struct nouveau_channel *chan)
 		goto error;
 	}
 
-	ib_virt = chan->pushbuf_base + chan->dma.ib_base * 4;
-
-	/* zero channel regs */
-	nouveau_bo_wr32(fifoch->user, 0x0040/4, 0);
-	nouveau_bo_wr32(fifoch->user, 0x0044/4, 0);
-	nouveau_bo_wr32(fifoch->user, 0x0048/4, 0);
-	nouveau_bo_wr32(fifoch->user, 0x004c/4, 0);
-	nouveau_bo_wr32(fifoch->user, 0x0050/4, 0);
-	nouveau_bo_wr32(fifoch->user, 0x0058/4, 0);
-	nouveau_bo_wr32(fifoch->user, 0x005c/4, 0);
-	nouveau_bo_wr32(fifoch->user, 0x0060/4, 0);
-	nouveau_bo_wr32(fifoch->user, 0x0088/4, 0);
-	nouveau_bo_wr32(fifoch->user, 0x008c/4, 0);
-
 	/* ramfc */
 	ret = nouveau_gpuobj_new_fake(dev, chan->ramin->pinst,
 				      chan->ramin->vinst, 0x100,
@@ -167,8 +138,8 @@ nvc0_fifo_create_context(struct nouveau_channel *chan)
 	if (ret)
 		goto error;
 
-	nv_wo32(fifoch->ramfc, 0x08, lower_32_bits(user_vinst));
-	nv_wo32(fifoch->ramfc, 0x0c, upper_32_bits(user_vinst));
+	nv_wo32(fifoch->ramfc, 0x08, lower_32_bits(fifoch->user->vinst));
+	nv_wo32(fifoch->ramfc, 0x0c, upper_32_bits(fifoch->user->vinst));
 	nv_wo32(fifoch->ramfc, 0x10, 0x0000face);
 	nv_wo32(fifoch->ramfc, 0x30, 0xfffff902);
 	nv_wo32(fifoch->ramfc, 0x48, lower_32_bits(ib_virt));
@@ -223,11 +194,7 @@ nvc0_fifo_destroy_context(struct nouveau_channel *chan)
 		return;
 
 	nouveau_gpuobj_ref(NULL, &fifoch->ramfc);
-	if (fifoch->user) {
-		nouveau_bo_unmap(fifoch->user);
-		nouveau_bo_unpin(fifoch->user);
-		nouveau_bo_ref(NULL, &fifoch->user);
-	}
+	nouveau_gpuobj_ref(NULL, &fifoch->user);
 	kfree(fifoch);
 }
 
@@ -240,6 +207,21 @@ nvc0_fifo_load_context(struct nouveau_channel *chan)
 int
 nvc0_fifo_unload_context(struct drm_device *dev)
 {
+	int i;
+
+	for (i = 0; i < 128; i++) {
+		if (!(nv_rd32(dev, 0x003004 + (i * 4)) & 1))
+			continue;
+
+		nv_mask(dev, 0x003004 + (i * 4), 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",
+				i, nv_rd32(dev, 0x002634));
+			return -EBUSY;
+		}
+	}
+
 	return 0;
 }
 
@@ -309,6 +291,7 @@ nvc0_fifo_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
+	struct nouveau_channel *chan;
 	struct nvc0_fifo_priv *priv;
 	int ret, i;
 
@@ -351,23 +334,74 @@ nvc0_fifo_init(struct drm_device *dev)
 	nv_wr32(dev, 0x002a00, 0xffffffff); /* clears PFIFO.INTR bit 30 */
 	nv_wr32(dev, 0x002100, 0xffffffff);
 	nv_wr32(dev, 0x002140, 0xbfffffff);
+
+	/* restore PFIFO context table */
+	for (i = 0; i < 128; i++) {
+		chan = dev_priv->channels.ptr[i];
+		if (!chan || !chan->fifo_priv)
+			continue;
+
+		nv_wr32(dev, 0x003000 + (i * 8), 0xc0000000 |
+						 (chan->ramin->vinst >> 12));
+		nv_wr32(dev, 0x003004 + (i * 8), 0x001f0001);
+	}
+	nvc0_fifo_playlist_update(dev);
+
 	return 0;
 }
 
 struct nouveau_enum nvc0_fifo_fault_unit[] = {
-	{ 0, "PGRAPH" },
-	{ 3, "PEEPHOLE" },
-	{ 4, "BAR1" },
-	{ 5, "BAR3" },
-	{ 7, "PFIFO" },
+	{ 0x00, "PGRAPH" },
+	{ 0x03, "PEEPHOLE" },
+	{ 0x04, "BAR1" },
+	{ 0x05, "BAR3" },
+	{ 0x07, "PFIFO" },
+	{ 0x10, "PBSP" },
+	{ 0x11, "PPPP" },
+	{ 0x13, "PCOUNTER" },
+	{ 0x14, "PVP" },
+	{ 0x15, "PCOPY0" },
+	{ 0x16, "PCOPY1" },
+	{ 0x17, "PDAEMON" },
 	{}
 };
 
 struct nouveau_enum nvc0_fifo_fault_reason[] = {
-	{ 0, "PT_NOT_PRESENT" },
-	{ 1, "PT_TOO_SHORT" },
-	{ 2, "PAGE_NOT_PRESENT" },
-	{ 3, "VM_LIMIT_EXCEEDED" },
+	{ 0x00, "PT_NOT_PRESENT" },
+	{ 0x01, "PT_TOO_SHORT" },
+	{ 0x02, "PAGE_NOT_PRESENT" },
+	{ 0x03, "VM_LIMIT_EXCEEDED" },
+	{ 0x04, "NO_CHANNEL" },
+	{ 0x05, "PAGE_SYSTEM_ONLY" },
+	{ 0x06, "PAGE_READ_ONLY" },
+	{ 0x0a, "COMPRESSED_SYSRAM" },
+	{ 0x0c, "INVALID_STORAGE_TYPE" },
+	{}
+};
+
+struct nouveau_enum nvc0_fifo_fault_hubclient[] = {
+	{ 0x01, "PCOPY0" },
+	{ 0x02, "PCOPY1" },
+	{ 0x04, "DISPATCH" },
+	{ 0x05, "CTXCTL" },
+	{ 0x06, "PFIFO" },
+	{ 0x07, "BAR_READ" },
+	{ 0x08, "BAR_WRITE" },
+	{ 0x0b, "PVP" },
+	{ 0x0c, "PPPP" },
+	{ 0x0d, "PBSP" },
+	{ 0x11, "PCOUNTER" },
+	{ 0x12, "PDAEMON" },
+	{ 0x14, "CCACHE" },
+	{ 0x15, "CCACHE_POST" },
+	{}
+};
+
+struct nouveau_enum nvc0_fifo_fault_gpcclient[] = {
+	{ 0x01, "TEX" },
+	{ 0x0c, "ESETUP" },
+	{ 0x0e, "CTXCTL" },
+	{ 0x0f, "PROP" },
 	{}
 };
 
@@ -385,12 +419,20 @@ nvc0_fifo_isr_vm_fault(struct drm_device *dev, int unit)
 	u32 valo = nv_rd32(dev, 0x2804 + (unit * 0x10));
 	u32 vahi = nv_rd32(dev, 0x2808 + (unit * 0x10));
 	u32 stat = nv_rd32(dev, 0x280c + (unit * 0x10));
+	u32 client = (stat & 0x00001f00) >> 8;
 
 	NV_INFO(dev, "PFIFO: %s fault at 0x%010llx [",
 		(stat & 0x00000080) ? "write" : "read", (u64)vahi << 32 | valo);
 	nouveau_enum_print(nvc0_fifo_fault_reason, stat & 0x0000000f);
 	printk("] from ");
 	nouveau_enum_print(nvc0_fifo_fault_unit, unit);
+	if (stat & 0x00000040) {
+		printk("/");
+		nouveau_enum_print(nvc0_fifo_fault_hubclient, client);
+	} else {
+		printk("/GPC%d/", (stat & 0x1f000000) >> 24);
+		nouveau_enum_print(nvc0_fifo_fault_gpcclient, client);
+	}
 	printk(" on channel 0x%010llx\n", (u64)inst << 12);
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvc0_graph.c b/drivers/gpu/drm/nouveau/nvc0_graph.c
index 3de9b721d8db..ca6db204d644 100644
--- a/drivers/gpu/drm/nouveau/nvc0_graph.c
+++ b/drivers/gpu/drm/nouveau/nvc0_graph.c
@@ -30,27 +30,40 @@
 #include "nouveau_mm.h"
 #include "nvc0_graph.h"
 
-static void nvc0_graph_isr(struct drm_device *);
-static void nvc0_runk140_isr(struct drm_device *);
-static int  nvc0_graph_unload_context_to(struct drm_device *dev, u64 chan);
-
-void
-nvc0_graph_fifo_access(struct drm_device *dev, bool enabled)
+static int
+nvc0_graph_load_context(struct nouveau_channel *chan)
 {
+	struct drm_device *dev = chan->dev;
+
+	nv_wr32(dev, 0x409840, 0x00000030);
+	nv_wr32(dev, 0x409500, 0x80000000 | chan->ramin->vinst >> 12);
+	nv_wr32(dev, 0x409504, 0x00000003);
+	if (!nv_wait(dev, 0x409800, 0x00000010, 0x00000010))
+		NV_ERROR(dev, "PGRAPH: load_ctx timeout\n");
+
+	return 0;
 }
 
-struct nouveau_channel *
-nvc0_graph_channel(struct drm_device *dev)
+static int
+nvc0_graph_unload_context_to(struct drm_device *dev, u64 chan)
 {
-	return NULL;
+	nv_wr32(dev, 0x409840, 0x00000003);
+	nv_wr32(dev, 0x409500, 0x80000000 | chan >> 12);
+	nv_wr32(dev, 0x409504, 0x00000009);
+	if (!nv_wait(dev, 0x409800, 0x00000001, 0x00000000)) {
+		NV_ERROR(dev, "PGRAPH: unload_ctx timeout\n");
+		return -EBUSY;
+	}
+
+	return 0;
 }
 
 static int
 nvc0_graph_construct_context(struct nouveau_channel *chan)
 {
 	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
-	struct nvc0_graph_priv *priv = dev_priv->engine.graph.priv;
-	struct nvc0_graph_chan *grch = chan->pgraph_ctx;
+	struct nvc0_graph_priv *priv = nv_engine(chan->dev, NVOBJ_ENGINE_GR);
+	struct nvc0_graph_chan *grch = chan->engctx[NVOBJ_ENGINE_GR];
 	struct drm_device *dev = chan->dev;
 	int ret, i;
 	u32 *ctx;
@@ -89,9 +102,8 @@ nvc0_graph_construct_context(struct nouveau_channel *chan)
 static int
 nvc0_graph_create_context_mmio_list(struct nouveau_channel *chan)
 {
-	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
-	struct nvc0_graph_priv *priv = dev_priv->engine.graph.priv;
-	struct nvc0_graph_chan *grch = chan->pgraph_ctx;
+	struct nvc0_graph_priv *priv = nv_engine(chan->dev, NVOBJ_ENGINE_GR);
+	struct nvc0_graph_chan *grch = chan->engctx[NVOBJ_ENGINE_GR];
 	struct drm_device *dev = chan->dev;
 	int i = 0, gpc, tp, ret;
 	u32 magic;
@@ -158,29 +170,27 @@ nvc0_graph_create_context_mmio_list(struct nouveau_channel *chan)
 	return 0;
 }
 
-int
-nvc0_graph_create_context(struct nouveau_channel *chan)
+static int
+nvc0_graph_context_new(struct nouveau_channel *chan, int engine)
 {
-	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
+	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_instmem_engine *pinstmem = &dev_priv->engine.instmem;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct nvc0_graph_priv *priv = pgraph->priv;
+	struct nvc0_graph_priv *priv = nv_engine(dev, engine);
 	struct nvc0_graph_chan *grch;
-	struct drm_device *dev = chan->dev;
 	struct nouveau_gpuobj *grctx;
 	int ret, i;
 
-	chan->pgraph_ctx = kzalloc(sizeof(*grch), GFP_KERNEL);
-	if (!chan->pgraph_ctx)
+	grch = kzalloc(sizeof(*grch), GFP_KERNEL);
+	if (!grch)
 		return -ENOMEM;
-	grch = chan->pgraph_ctx;
+	chan->engctx[NVOBJ_ENGINE_GR] = grch;
 
 	ret = nouveau_gpuobj_new(dev, NULL, priv->grctx_size, 256,
 				 NVOBJ_FLAG_VM | NVOBJ_FLAG_ZERO_ALLOC,
 				 &grch->grctx);
 	if (ret)
 		goto error;
-	chan->ramin_grctx = grch->grctx;
 	grctx = grch->grctx;
 
 	ret = nvc0_graph_create_context_mmio_list(chan);
@@ -200,104 +210,49 @@ nvc0_graph_create_context(struct nouveau_channel *chan)
 	for (i = 0; i < priv->grctx_size; i += 4)
 		nv_wo32(grctx, i, priv->grctx_vals[i / 4]);
 
-        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, 0x1c, 1);
-        nv_wo32(grctx, 0x20, 0);
-        nv_wo32(grctx, 0x28, 0);
-        nv_wo32(grctx, 0x2c, 0);
+	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, 0x1c, 1);
+	nv_wo32(grctx, 0x20, 0);
+	nv_wo32(grctx, 0x28, 0);
+	nv_wo32(grctx, 0x2c, 0);
 	pinstmem->flush(dev);
 	return 0;
 
 error:
-	pgraph->destroy_context(chan);
+	priv->base.context_del(chan, engine);
 	return ret;
 }
 
-void
-nvc0_graph_destroy_context(struct nouveau_channel *chan)
+static void
+nvc0_graph_context_del(struct nouveau_channel *chan, int engine)
 {
-	struct nvc0_graph_chan *grch;
-
-	grch = chan->pgraph_ctx;
-	chan->pgraph_ctx = NULL;
-	if (!grch)
-		return;
+	struct nvc0_graph_chan *grch = chan->engctx[engine];
 
 	nouveau_gpuobj_ref(NULL, &grch->mmio);
 	nouveau_gpuobj_ref(NULL, &grch->unk418810);
 	nouveau_gpuobj_ref(NULL, &grch->unk40800c);
 	nouveau_gpuobj_ref(NULL, &grch->unk408004);
 	nouveau_gpuobj_ref(NULL, &grch->grctx);
-	chan->ramin_grctx = NULL;
+	chan->engctx[engine] = NULL;
 }
 
-int
-nvc0_graph_load_context(struct nouveau_channel *chan)
+static int
+nvc0_graph_object_new(struct nouveau_channel *chan, int engine,
+		      u32 handle, u16 class)
 {
-	struct drm_device *dev = chan->dev;
-
-	nv_wr32(dev, 0x409840, 0x00000030);
-	nv_wr32(dev, 0x409500, 0x80000000 | chan->ramin->vinst >> 12);
-	nv_wr32(dev, 0x409504, 0x00000003);
-	if (!nv_wait(dev, 0x409800, 0x00000010, 0x00000010))
-		NV_ERROR(dev, "PGRAPH: load_ctx timeout\n");
-
 	return 0;
 }
 
 static int
-nvc0_graph_unload_context_to(struct drm_device *dev, u64 chan)
+nvc0_graph_fini(struct drm_device *dev, int engine)
 {
-	nv_wr32(dev, 0x409840, 0x00000003);
-	nv_wr32(dev, 0x409500, 0x80000000 | chan >> 12);
-	nv_wr32(dev, 0x409504, 0x00000009);
-	if (!nv_wait(dev, 0x409800, 0x00000001, 0x00000000)) {
-		NV_ERROR(dev, "PGRAPH: unload_ctx timeout\n");
-		return -EBUSY;
-	}
-
 	return 0;
 }
 
-int
-nvc0_graph_unload_context(struct drm_device *dev)
-{
-	u64 inst = (u64)(nv_rd32(dev, 0x409b00) & 0x0fffffff) << 12;
-	return nvc0_graph_unload_context_to(dev, inst);
-}
-
-static void
-nvc0_graph_destroy(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct nvc0_graph_priv *priv;
-
-	priv = pgraph->priv;
-	if (!priv)
-		return;
-
-	nouveau_irq_unregister(dev, 12);
-	nouveau_irq_unregister(dev, 25);
-
-	nouveau_gpuobj_ref(NULL, &priv->unk4188b8);
-	nouveau_gpuobj_ref(NULL, &priv->unk4188b4);
-
-	if (priv->grctx_vals)
-		kfree(priv->grctx_vals);
-	kfree(priv);
-}
-
-void
-nvc0_graph_takedown(struct drm_device *dev)
-{
-	nvc0_graph_destroy(dev);
-}
-
 static int
 nvc0_graph_mthd_page_flip(struct nouveau_channel *chan,
 			  u32 class, u32 mthd, u32 data)
@@ -306,119 +261,10 @@ nvc0_graph_mthd_page_flip(struct nouveau_channel *chan,
 	return 0;
 }
 
-static int
-nvc0_graph_create(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct nvc0_graph_priv *priv;
-	int ret, gpc, i;
-
-	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
-	if (!priv)
-		return -ENOMEM;
-	pgraph->priv = priv;
-
-	ret = nouveau_gpuobj_new(dev, NULL, 0x1000, 256, 0, &priv->unk4188b4);
-	if (ret)
-		goto error;
-
-	ret = nouveau_gpuobj_new(dev, NULL, 0x1000, 256, 0, &priv->unk4188b8);
-	if (ret)
-		goto error;
-
-	for (i = 0; i < 0x1000; i += 4) {
-		nv_wo32(priv->unk4188b4, i, 0x00000010);
-		nv_wo32(priv->unk4188b8, i, 0x00000010);
-	}
-
-	priv->gpc_nr  =  nv_rd32(dev, 0x409604) & 0x0000001f;
-	priv->rop_nr = (nv_rd32(dev, 0x409604) & 0x001f0000) >> 16;
-	for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
-		priv->tp_nr[gpc] = nv_rd32(dev, GPC_UNIT(gpc, 0x2608));
-		priv->tp_total += priv->tp_nr[gpc];
-	}
-
-	/*XXX: these need figuring out... */
-	switch (dev_priv->chipset) {
-	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->magicgpc980[0]   = 0x22111000;
-			priv->magicgpc980[1]   = 0x00000233;
-			priv->magicgpc980[2]   = 0x00000000;
-			priv->magicgpc980[3]   = 0x00000000;
-			priv->magicgpc918      = 0x000ba2e9;
-		} else
-		if (priv->tp_total == 14) { /* 470, 3/3/4/4, 5 */
-			priv->magic_not_rop_nr = 0x05;
-			priv->magicgpc980[0]   = 0x11110000;
-			priv->magicgpc980[1]   = 0x00233222;
-			priv->magicgpc980[2]   = 0x00000000;
-			priv->magicgpc980[3]   = 0x00000000;
-			priv->magicgpc918      = 0x00092493;
-		} else
-		if (priv->tp_total == 15) { /* 480, 3/4/4/4, 6 */
-			priv->magic_not_rop_nr = 0x06;
-			priv->magicgpc980[0]   = 0x11110000;
-			priv->magicgpc980[1]   = 0x03332222;
-			priv->magicgpc980[2]   = 0x00000000;
-			priv->magicgpc980[3]   = 0x00000000;
-			priv->magicgpc918      = 0x00088889;
-		}
-		break;
-	case 0xc3: /* 450, 4/0/0/0, 2 */
-		priv->magic_not_rop_nr = 0x03;
-		priv->magicgpc980[0]   = 0x00003210;
-		priv->magicgpc980[1]   = 0x00000000;
-		priv->magicgpc980[2]   = 0x00000000;
-		priv->magicgpc980[3]   = 0x00000000;
-		priv->magicgpc918      = 0x00200000;
-		break;
-	case 0xc4: /* 460, 3/4/0/0, 4 */
-		priv->magic_not_rop_nr = 0x01;
-		priv->magicgpc980[0]   = 0x02321100;
-		priv->magicgpc980[1]   = 0x00000000;
-		priv->magicgpc980[2]   = 0x00000000;
-		priv->magicgpc980[3]   = 0x00000000;
-		priv->magicgpc918      = 0x00124925;
-		break;
-	}
-
-	if (!priv->magic_not_rop_nr) {
-		NV_ERROR(dev, "PGRAPH: unknown config: %d/%d/%d/%d, %d\n",
-			 priv->tp_nr[0], priv->tp_nr[1], priv->tp_nr[2],
-			 priv->tp_nr[3], priv->rop_nr);
-		/* use 0xc3's values... */
-		priv->magic_not_rop_nr = 0x03;
-		priv->magicgpc980[0]   = 0x00003210;
-		priv->magicgpc980[1]   = 0x00000000;
-		priv->magicgpc980[2]   = 0x00000000;
-		priv->magicgpc980[3]   = 0x00000000;
-		priv->magicgpc918      = 0x00200000;
-	}
-
-	nouveau_irq_register(dev, 12, nvc0_graph_isr);
-	nouveau_irq_register(dev, 25, nvc0_runk140_isr);
-	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 */
-	NVOBJ_CLASS(dev, 0x90c0, GR); /* COMPUTE */
-	return 0;
-
-error:
-	nvc0_graph_destroy(dev);
-	return ret;
-}
-
 static void
 nvc0_graph_init_obj418880(struct drm_device *dev)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct nvc0_graph_priv *priv = pgraph->priv;
+	struct nvc0_graph_priv *priv = nv_engine(dev, NVOBJ_ENGINE_GR);
 	int i;
 
 	nv_wr32(dev, GPC_BCAST(0x0880), 0x00000000);
@@ -449,35 +295,42 @@ nvc0_graph_init_regs(struct drm_device *dev)
 static void
 nvc0_graph_init_gpc_0(struct drm_device *dev)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nvc0_graph_priv *priv = dev_priv->engine.graph.priv;
-	int gpc;
-	
-	//      TP      ROP UNKVAL(magic_not_rop_nr)
-	// 450: 4/0/0/0 2        3
-	// 460: 3/4/0/0 4        1
-	// 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
-
-	/* filled values up to tp_total, remainder 0 */
-	// 450: 00003210 00000000 00000000 00000000
-	// 460: 02321100 00000000 00000000 00000000
-	// 465: 22111000 00000233 00000000 00000000
-	// 470: 11110000 00233222 00000000 00000000
-	// 480: 11110000 03332222 00000000 00000000
-	
-	nv_wr32(dev, GPC_BCAST(0x0980), priv->magicgpc980[0]);
-	nv_wr32(dev, GPC_BCAST(0x0984), priv->magicgpc980[1]);
-	nv_wr32(dev, GPC_BCAST(0x0988), priv->magicgpc980[2]);
-	nv_wr32(dev, GPC_BCAST(0x098c), priv->magicgpc980[3]);
+	struct nvc0_graph_priv *priv = nv_engine(dev, NVOBJ_ENGINE_GR);
+	u32 data[TP_MAX / 8];
+	u8  tpnr[GPC_MAX];
+	int i, gpc, tpc;
+
+	/*
+	 *      TP      ROP UNKVAL(magic_not_rop_nr)
+	 * 450: 4/0/0/0 2        3
+	 * 460: 3/4/0/0 4        1
+	 * 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));
+	memcpy(tpnr, priv->tp_nr, sizeof(priv->tp_nr));
+	for (i = 0, gpc = -1; i < priv->tp_total; i++) {
+		do {
+			gpc = (gpc + 1) % priv->gpc_nr;
+		} while (!tpnr[gpc]);
+		tpc = priv->tp_nr[gpc] - tpnr[gpc]--;
+
+		data[i / 8] |= tpc << ((i % 8) * 4);
+	}
+
+	nv_wr32(dev, GPC_BCAST(0x0980), data[0]);
+	nv_wr32(dev, GPC_BCAST(0x0984), data[1]);
+	nv_wr32(dev, GPC_BCAST(0x0988), data[2]);
+	nv_wr32(dev, GPC_BCAST(0x098c), data[3]);
 
 	for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
 		nv_wr32(dev, GPC_UNIT(gpc, 0x0914), priv->magic_not_rop_nr << 8 |
@@ -509,8 +362,7 @@ nvc0_graph_init_units(struct drm_device *dev)
 static void
 nvc0_graph_init_gpc_1(struct drm_device *dev)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nvc0_graph_priv *priv = dev_priv->engine.graph.priv;
+	struct nvc0_graph_priv *priv = nv_engine(dev, NVOBJ_ENGINE_GR);
 	int gpc, tp;
 
 	for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
@@ -535,8 +387,7 @@ nvc0_graph_init_gpc_1(struct drm_device *dev)
 static void
 nvc0_graph_init_rop(struct drm_device *dev)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nvc0_graph_priv *priv = dev_priv->engine.graph.priv;
+	struct nvc0_graph_priv *priv = nv_engine(dev, NVOBJ_ENGINE_GR);
 	int rop;
 
 	for (rop = 0; rop < priv->rop_nr; rop++) {
@@ -547,62 +398,36 @@ nvc0_graph_init_rop(struct drm_device *dev)
 	}
 }
 
-static int
-nvc0_fuc_load_fw(struct drm_device *dev, u32 fuc_base,
-		 const char *code_fw, const char *data_fw)
+static void
+nvc0_graph_init_fuc(struct drm_device *dev, u32 fuc_base,
+		    struct nvc0_graph_fuc *code, struct nvc0_graph_fuc *data)
 {
-	const struct firmware *fw;
-	char name[32];
-	int ret, i;
-
-	snprintf(name, sizeof(name), "nouveau/%s", data_fw);
-	ret = request_firmware(&fw, name, &dev->pdev->dev);
-	if (ret) {
-		NV_ERROR(dev, "failed to load %s\n", data_fw);
-		return ret;
-	}
+	int i;
 
 	nv_wr32(dev, fuc_base + 0x01c0, 0x01000000);
-	for (i = 0; i < fw->size / 4; i++)
-		nv_wr32(dev, fuc_base + 0x01c4, ((u32 *)fw->data)[i]);
-	release_firmware(fw);
-
-	snprintf(name, sizeof(name), "nouveau/%s", code_fw);
-	ret = request_firmware(&fw, name, &dev->pdev->dev);
-	if (ret) {
-		NV_ERROR(dev, "failed to load %s\n", code_fw);
-		return ret;
-	}
+	for (i = 0; i < data->size / 4; i++)
+		nv_wr32(dev, fuc_base + 0x01c4, data->data[i]);
 
 	nv_wr32(dev, fuc_base + 0x0180, 0x01000000);
-	for (i = 0; i < fw->size / 4; i++) {
+	for (i = 0; i < code->size / 4; i++) {
 		if ((i & 0x3f) == 0)
 			nv_wr32(dev, fuc_base + 0x0188, i >> 6);
-		nv_wr32(dev, fuc_base + 0x0184, ((u32 *)fw->data)[i]);
+		nv_wr32(dev, fuc_base + 0x0184, code->data[i]);
 	}
-	release_firmware(fw);
-
-	return 0;
 }
 
 static int
 nvc0_graph_init_ctxctl(struct drm_device *dev)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nvc0_graph_priv *priv = dev_priv->engine.graph.priv;
+	struct nvc0_graph_priv *priv = nv_engine(dev, NVOBJ_ENGINE_GR);
 	u32 r000260;
-	int ret;
 
 	/* load fuc microcode */
 	r000260 = nv_mask(dev, 0x000260, 0x00000001, 0x00000000);
-	ret = nvc0_fuc_load_fw(dev, 0x409000, "fuc409c", "fuc409d");
-	if (ret == 0)
-		ret = nvc0_fuc_load_fw(dev, 0x41a000, "fuc41ac", "fuc41ad");
+	nvc0_graph_init_fuc(dev, 0x409000, &priv->fuc409c, &priv->fuc409d);
+	nvc0_graph_init_fuc(dev, 0x41a000, &priv->fuc41ac, &priv->fuc41ad);
 	nv_wr32(dev, 0x000260, r000260);
 
-	if (ret)
-		return ret;
-
 	/* start both of them running */
 	nv_wr32(dev, 0x409840, 0xffffffff);
 	nv_wr32(dev, 0x41a10c, 0x00000000);
@@ -644,41 +469,19 @@ nvc0_graph_init_ctxctl(struct drm_device *dev)
 	return 0;
 }
 
-int
-nvc0_graph_init(struct drm_device *dev)
+static int
+nvc0_graph_init(struct drm_device *dev, int engine)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	int ret;
 
-	dev_priv->engine.graph.accel_blocked = true;
-
-	switch (dev_priv->chipset) {
-	case 0xc0:
-	case 0xc3:
-	case 0xc4:
-		break;
-	default:
-		NV_ERROR(dev, "PGRAPH: unsupported chipset, please report!\n");
-		if (nouveau_noaccel != 0)
-			return 0;
-		break;
-	}
-
 	nv_mask(dev, 0x000200, 0x18001000, 0x00000000);
 	nv_mask(dev, 0x000200, 0x18001000, 0x18001000);
 
-	if (!pgraph->priv) {
-		ret = nvc0_graph_create(dev);
-		if (ret)
-			return ret;
-	}
-
 	nvc0_graph_init_obj418880(dev);
 	nvc0_graph_init_regs(dev);
-	//nvc0_graph_init_unitplemented_magics(dev);
+	/*nvc0_graph_init_unitplemented_magics(dev);*/
 	nvc0_graph_init_gpc_0(dev);
-	//nvc0_graph_init_unitplemented_c242(dev);
+	/*nvc0_graph_init_unitplemented_c242(dev);*/
 
 	nv_wr32(dev, 0x400500, 0x00010001);
 	nv_wr32(dev, 0x400100, 0xffffffff);
@@ -697,12 +500,13 @@ nvc0_graph_init(struct drm_device *dev)
 	nv_wr32(dev, 0x400054, 0x34ce3464);
 
 	ret = nvc0_graph_init_ctxctl(dev);
-	if (ret == 0)
-		dev_priv->engine.graph.accel_blocked = false;
+	if (ret)
+		return ret;
+
 	return 0;
 }
 
-static int
+int
 nvc0_graph_isr_chid(struct drm_device *dev, u64 inst)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
@@ -806,3 +610,187 @@ nvc0_runk140_isr(struct drm_device *dev)
 		units &= ~(1 << unit);
 	}
 }
+
+static int
+nvc0_graph_create_fw(struct drm_device *dev, const char *fwname,
+		     struct nvc0_graph_fuc *fuc)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	const struct firmware *fw;
+	char f[32];
+	int ret;
+
+	snprintf(f, sizeof(f), "nouveau/nv%02x_%s", dev_priv->chipset, fwname);
+	ret = request_firmware(&fw, f, &dev->pdev->dev);
+	if (ret) {
+		snprintf(f, sizeof(f), "nouveau/%s", fwname);
+		ret = request_firmware(&fw, f, &dev->pdev->dev);
+		if (ret) {
+			NV_ERROR(dev, "failed to load %s\n", fwname);
+			return ret;
+		}
+	}
+
+	fuc->size = fw->size;
+	fuc->data = kmemdup(fw->data, fuc->size, GFP_KERNEL);
+	release_firmware(fw);
+	return (fuc->data != NULL) ? 0 : -ENOMEM;
+}
+
+static void
+nvc0_graph_destroy_fw(struct nvc0_graph_fuc *fuc)
+{
+	if (fuc->data) {
+		kfree(fuc->data);
+		fuc->data = NULL;
+	}
+}
+
+static void
+nvc0_graph_destroy(struct drm_device *dev, int engine)
+{
+	struct nvc0_graph_priv *priv = nv_engine(dev, engine);
+
+	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);
+
+	nouveau_gpuobj_ref(NULL, &priv->unk4188b8);
+	nouveau_gpuobj_ref(NULL, &priv->unk4188b4);
+
+	if (priv->grctx_vals)
+		kfree(priv->grctx_vals);
+
+	NVOBJ_ENGINE_DEL(dev, GR);
+	kfree(priv);
+}
+
+int
+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;
+
+	switch (dev_priv->chipset) {
+	case 0xc0:
+	case 0xc3:
+	case 0xc4:
+		break;
+	default:
+		NV_ERROR(dev, "PGRAPH: unsupported chipset, please report!\n");
+		return 0;
+	}
+
+	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
+	if (!priv)
+		return -ENOMEM;
+
+	priv->base.destroy = nvc0_graph_destroy;
+	priv->base.init = nvc0_graph_init;
+	priv->base.fini = nvc0_graph_fini;
+	priv->base.context_new = nvc0_graph_context_new;
+	priv->base.context_del = nvc0_graph_context_del;
+	priv->base.object_new = nvc0_graph_object_new;
+
+	NVOBJ_ENGINE_ADD(dev, GR, &priv->base);
+	nouveau_irq_register(dev, 12, nvc0_graph_isr);
+	nouveau_irq_register(dev, 25, nvc0_runk140_isr);
+
+	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) ||
+	    nvc0_graph_create_fw(dev, "fuc41ad", &priv->fuc41ad)) {
+		ret = 0;
+		goto error;
+	}
+
+
+	ret = nouveau_gpuobj_new(dev, NULL, 0x1000, 256, 0, &priv->unk4188b4);
+	if (ret)
+		goto error;
+
+	ret = nouveau_gpuobj_new(dev, NULL, 0x1000, 256, 0, &priv->unk4188b8);
+	if (ret)
+		goto error;
+
+	for (i = 0; i < 0x1000; i += 4) {
+		nv_wo32(priv->unk4188b4, i, 0x00000010);
+		nv_wo32(priv->unk4188b8, i, 0x00000010);
+	}
+
+	priv->gpc_nr  =  nv_rd32(dev, 0x409604) & 0x0000001f;
+	priv->rop_nr = (nv_rd32(dev, 0x409604) & 0x001f0000) >> 16;
+	for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
+		priv->tp_nr[gpc] = nv_rd32(dev, GPC_UNIT(gpc, 0x2608));
+		priv->tp_total += priv->tp_nr[gpc];
+	}
+
+	/*XXX: these need figuring out... */
+	switch (dev_priv->chipset) {
+	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;
+	}
+
+	if (!priv->magic_not_rop_nr) {
+		NV_ERROR(dev, "PGRAPH: unknown config: %d/%d/%d/%d, %d\n",
+			 priv->tp_nr[0], priv->tp_nr[1], priv->tp_nr[2],
+			 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 */
+	NVOBJ_CLASS(dev, 0x90c0, GR); /* COMPUTE */
+	return 0;
+
+error:
+	nvc0_graph_destroy(dev, NVOBJ_ENGINE_GR);
+	return ret;
+}
+
+MODULE_FIRMWARE("nouveau/nvc0_fuc409c");
+MODULE_FIRMWARE("nouveau/nvc0_fuc409d");
+MODULE_FIRMWARE("nouveau/nvc0_fuc41ac");
+MODULE_FIRMWARE("nouveau/nvc0_fuc41ad");
+MODULE_FIRMWARE("nouveau/nvc3_fuc409c");
+MODULE_FIRMWARE("nouveau/nvc3_fuc409d");
+MODULE_FIRMWARE("nouveau/nvc3_fuc41ac");
+MODULE_FIRMWARE("nouveau/nvc3_fuc41ad");
+MODULE_FIRMWARE("nouveau/nvc4_fuc409c");
+MODULE_FIRMWARE("nouveau/nvc4_fuc409d");
+MODULE_FIRMWARE("nouveau/nvc4_fuc41ac");
+MODULE_FIRMWARE("nouveau/nvc4_fuc41ad");
+MODULE_FIRMWARE("nouveau/fuc409c");
+MODULE_FIRMWARE("nouveau/fuc409d");
+MODULE_FIRMWARE("nouveau/fuc41ac");
+MODULE_FIRMWARE("nouveau/fuc41ad");
diff --git a/drivers/gpu/drm/nouveau/nvc0_graph.h b/drivers/gpu/drm/nouveau/nvc0_graph.h
index 40e26f9c56c4..f5d184e0689d 100644
--- a/drivers/gpu/drm/nouveau/nvc0_graph.h
+++ b/drivers/gpu/drm/nouveau/nvc0_graph.h
@@ -28,13 +28,25 @@
 #define GPC_MAX 4
 #define TP_MAX 32
 
-#define ROP_BCAST(r)   (0x408800 + (r))
-#define ROP_UNIT(u,r)  (0x410000 + (u) * 0x400 + (r))
-#define GPC_BCAST(r)   (0x418000 + (r))
-#define GPC_UNIT(t,r)  (0x500000 + (t) * 0x8000 + (r))
-#define TP_UNIT(t,m,r) (0x504000 + (t) * 0x8000 + (m) * 0x800 + (r))
+#define ROP_BCAST(r)     (0x408800 + (r))
+#define ROP_UNIT(u, r)   (0x410000 + (u) * 0x400 + (r))
+#define GPC_BCAST(r)     (0x418000 + (r))
+#define GPC_UNIT(t, r)   (0x500000 + (t) * 0x8000 + (r))
+#define TP_UNIT(t, m, r) (0x504000 + (t) * 0x8000 + (m) * 0x800 + (r))
+
+struct nvc0_graph_fuc {
+	u32 *data;
+	u32  size;
+};
 
 struct nvc0_graph_priv {
+	struct nouveau_exec_engine base;
+
+	struct nvc0_graph_fuc fuc409c;
+	struct nvc0_graph_fuc fuc409d;
+	struct nvc0_graph_fuc fuc41ac;
+	struct nvc0_graph_fuc fuc41ad;
+
 	u8 gpc_nr;
 	u8 rop_nr;
 	u8 tp_nr[GPC_MAX];
@@ -46,15 +58,14 @@ struct nvc0_graph_priv {
 	struct nouveau_gpuobj *unk4188b8;
 
 	u8  magic_not_rop_nr;
-	u32 magicgpc980[4];
 	u32 magicgpc918;
 };
 
 struct nvc0_graph_chan {
 	struct nouveau_gpuobj *grctx;
-	struct nouveau_gpuobj *unk408004; // 0x418810 too
-	struct nouveau_gpuobj *unk40800c; // 0x419004 too
-	struct nouveau_gpuobj *unk418810; // 0x419848 too
+	struct nouveau_gpuobj *unk408004; /* 0x418810 too */
+	struct nouveau_gpuobj *unk40800c; /* 0x419004 too */
+	struct nouveau_gpuobj *unk418810; /* 0x419848 too */
 	struct nouveau_gpuobj *mmio;
 	int mmio_nr;
 };
diff --git a/drivers/gpu/drm/nouveau/nvc0_grctx.c b/drivers/gpu/drm/nouveau/nvc0_grctx.c
index f880ff776db8..6df066114133 100644
--- a/drivers/gpu/drm/nouveau/nvc0_grctx.c
+++ b/drivers/gpu/drm/nouveau/nvc0_grctx.c
@@ -1623,7 +1623,7 @@ nvc0_grctx_generate_rop(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 
-	// ROPC_BROADCAST
+	/* ROPC_BROADCAST */
 	nv_wr32(dev, 0x408800, 0x02802a3c);
 	nv_wr32(dev, 0x408804, 0x00000040);
 	nv_wr32(dev, 0x408808, 0x0003e00d);
@@ -1647,7 +1647,7 @@ nvc0_grctx_generate_gpc(struct drm_device *dev)
 {
 	int i;
 
-	// GPC_BROADCAST
+	/* GPC_BROADCAST */
 	nv_wr32(dev, 0x418380, 0x00000016);
 	nv_wr32(dev, 0x418400, 0x38004e00);
 	nv_wr32(dev, 0x418404, 0x71e0ffff);
@@ -1728,7 +1728,7 @@ nvc0_grctx_generate_tp(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 
-	// GPC_BROADCAST.TP_BROADCAST
+	/* GPC_BROADCAST.TP_BROADCAST */
 	nv_wr32(dev, 0x419848, 0x00000000);
 	nv_wr32(dev, 0x419864, 0x0000012a);
 	nv_wr32(dev, 0x419888, 0x00000000);
@@ -1741,7 +1741,7 @@ nvc0_grctx_generate_tp(struct drm_device *dev)
 	nv_wr32(dev, 0x419a1c, 0x00000000);
 	nv_wr32(dev, 0x419a20, 0x00000800);
 	if (dev_priv->chipset != 0xc0)
-		nv_wr32(dev, 0x00419ac4, 0x0007f440); // 0xc3
+		nv_wr32(dev, 0x00419ac4, 0x0007f440); /* 0xc3 */
 	nv_wr32(dev, 0x419b00, 0x0a418820);
 	nv_wr32(dev, 0x419b04, 0x062080e6);
 	nv_wr32(dev, 0x419b08, 0x020398a4);
@@ -1797,8 +1797,8 @@ int
 nvc0_grctx_generate(struct nouveau_channel *chan)
 {
 	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
-	struct nvc0_graph_priv *priv = dev_priv->engine.graph.priv;
-	struct nvc0_graph_chan *grch = chan->pgraph_ctx;
+	struct nvc0_graph_priv *priv = nv_engine(chan->dev, NVOBJ_ENGINE_GR);
+	struct nvc0_graph_chan *grch = chan->engctx[NVOBJ_ENGINE_GR];
 	struct drm_device *dev = chan->dev;
 	int i, gpc, tp, id;
 	u32 r000260, tmp;
@@ -1912,13 +1912,13 @@ nvc0_grctx_generate(struct nouveau_channel *chan)
 		for (i = 1; i < 7; i++)
 			data2[1] |= ((1 << (i + 5)) % ntpcv) << ((i - 1) * 5);
 
-		// GPC_BROADCAST
+		/* GPC_BROADCAST */
 		nv_wr32(dev, 0x418bb8, (priv->tp_total << 8) |
 					priv->magic_not_rop_nr);
 		for (i = 0; i < 6; i++)
 			nv_wr32(dev, 0x418b08 + (i * 4), data[i]);
 
-		// GPC_BROADCAST.TP_BROADCAST
+		/* GPC_BROADCAST.TP_BROADCAST */
 		nv_wr32(dev, 0x419bd0, (priv->tp_total << 8) |
 				       priv->magic_not_rop_nr |
 				       data2[0]);
@@ -1926,7 +1926,7 @@ nvc0_grctx_generate(struct nouveau_channel *chan)
 		for (i = 0; i < 6; i++)
 			nv_wr32(dev, 0x419b00 + (i * 4), data[i]);
 
-		// UNK78xx
+		/* UNK78xx */
 		nv_wr32(dev, 0x4078bc, (priv->tp_total << 8) |
 					priv->magic_not_rop_nr);
 		for (i = 0; i < 6; i++)
@@ -1944,7 +1944,7 @@ nvc0_grctx_generate(struct nouveau_channel *chan)
 		gpc = -1;
 		for (i = 0, gpc = -1; i < 32; i++) {
 			int ltp = i * (priv->tp_total - 1) / 32;
-			
+
 			do {
 				gpc = (gpc + 1) % priv->gpc_nr;
 			} while (!tpnr[gpc]);