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

* 'drm-nouveau-next' of git://anongit.freedesktop.org/git/nouveau/linux-2.6: (57 commits)
  drm/nouveau: map first page of mmio early and determine chipset earlier
  drm/nvd0/disp: disconnect encoders before reprogramming them
  drm/nvd0/disp: move syncs/magic setup to or mode_set
  drm/nouveau/dp: account for channel coding overhead in link training
  drm/nvd0/disp: fix dcb sor link matching in supervisor handler
  drm/nvd0/disp: initial implementation of displayport
  drm/nouveau/dp: make dp dpms function common, call from sor code instead
  drm/nv50/hwsq: some nv92 fixes
  drm/nouveau/dp: move all nv50/sor-specific code out of nouveau_dp.c
  drm/nouveau/dp: make functions for executing various bios tables
  drm/nouveau/pm: fix oops if chipset has no pm support at all
  drm/nouveau/bios: rework vbios shadowing
  drm/nouveau/bios: attempt acpi rom fetch before pcirom
  drm/nvd0/disp: attempt to handle more than 2 crtcs if possible
  drm/nvc0/vram: get part count from PUNITS
  drm/nv40/pm: fix fanspeed regression
  drm/nouveau/pm: several fixes for nvc0 memory timings
  drm/nvc0/pm: restrict pll mode to clocks that can actually use it
  drm/nouveau/dp: fix bad comparison in dp_link_train_commit()
  drm/nouveau/mxm: call mxmi to determine revision before calling mxms
  ...

32 files changed, 2685 insertions, 1340 deletions

diff --git a/drivers/gpu/drm/nouveau/Makefile b/drivers/gpu/drm/nouveau/Makefile
index 9f27e3d..1a2ad7e 100644
--- a/drivers/gpu/drm/nouveau/Makefile
+++ b/drivers/gpu/drm/nouveau/Makefile
@@ -14,7 +14,8 @@ nouveau-y := nouveau_drv.o nouveau_state.o nouveau_channel.o nouveau_mem.o \
 	     nouveau_mm.o nouveau_vm.o nouveau_mxm.o nouveau_gpio.o \
              nv04_timer.o \
              nv04_mc.o nv40_mc.o nv50_mc.o \
-             nv04_fb.o nv10_fb.o nv30_fb.o nv40_fb.o nv50_fb.o nvc0_fb.o \
+             nv04_fb.o nv10_fb.o nv20_fb.o nv30_fb.o nv40_fb.o \
+             nv50_fb.o nvc0_fb.o \
              nv04_fifo.o nv10_fifo.o nv40_fifo.o nv50_fifo.o nvc0_fifo.o \
              nv04_graph.o nv10_graph.o nv20_graph.o \
              nv40_graph.o nv50_graph.o nvc0_graph.o \
diff --git a/drivers/gpu/drm/nouveau/nouveau_bios.c b/drivers/gpu/drm/nouveau/nouveau_bios.c
index e5cbead..8dbeeea 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bios.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bios.c
@@ -65,195 +65,232 @@ static bool nv_cksum(const uint8_t *data, unsigned int length)
 }
 
 static int
-score_vbios(struct drm_device *dev, const uint8_t *data, const bool writeable)
+score_vbios(struct nvbios *bios, const bool writeable)
 {
-	if (!(data[0] == 0x55 && data[1] == 0xAA)) {
-		NV_TRACEWARN(dev, "... BIOS signature not found\n");
+	if (!bios->data || bios->data[0] != 0x55 || bios->data[1] != 0xAA) {
+		NV_TRACEWARN(bios->dev, "... BIOS signature not found\n");
 		return 0;
 	}
 
-	if (nv_cksum(data, data[2] * 512)) {
-		NV_TRACEWARN(dev, "... BIOS checksum invalid\n");
+	if (nv_cksum(bios->data, bios->data[2] * 512)) {
+		NV_TRACEWARN(bios->dev, "... BIOS checksum invalid\n");
 		/* if a ro image is somewhat bad, it's probably all rubbish */
 		return writeable ? 2 : 1;
-	} else
-		NV_TRACE(dev, "... appears to be valid\n");
+	}
 
+	NV_TRACE(bios->dev, "... appears to be valid\n");
 	return 3;
 }
 
-static void load_vbios_prom(struct drm_device *dev, uint8_t *data)
+static void
+bios_shadow_prom(struct nvbios *bios)
 {
+	struct drm_device *dev = bios->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t pci_nv_20, save_pci_nv_20;
-	int pcir_ptr;
+	u32 pcireg, access;
+	u16 pcir;
 	int i;
 
+	/* enable access to rom */
 	if (dev_priv->card_type >= NV_50)
-		pci_nv_20 = 0x88050;
+		pcireg = 0x088050;
 	else
-		pci_nv_20 = NV_PBUS_PCI_NV_20;
+		pcireg = NV_PBUS_PCI_NV_20;
+	access = nv_mask(dev, pcireg, 0x00000001, 0x00000000);
 
-	/* enable ROM access */
-	save_pci_nv_20 = nvReadMC(dev, pci_nv_20);
-	nvWriteMC(dev, pci_nv_20,
-		  save_pci_nv_20 & ~NV_PBUS_PCI_NV_20_ROM_SHADOW_ENABLED);
+	/* bail if no rom signature, with a workaround for a PROM reading
+	 * issue on some chipsets.  the first read after a period of
+	 * inactivity returns the wrong result, so retry the first header
+	 * byte a few times before giving up as a workaround
+	 */
+	i = 16;
+	do {
+		if (nv_rd08(dev, NV_PROM_OFFSET + 0) == 0x55)
+			break;
+	} while (i--);
 
-	/* bail if no rom signature */
-	if (nv_rd08(dev, NV_PROM_OFFSET) != 0x55 ||
-	    nv_rd08(dev, NV_PROM_OFFSET + 1) != 0xaa)
+	if (!i || nv_rd08(dev, NV_PROM_OFFSET + 1) != 0xaa)
 		goto out;
 
 	/* additional check (see note below) - read PCI record header */
-	pcir_ptr = nv_rd08(dev, NV_PROM_OFFSET + 0x18) |
-		   nv_rd08(dev, NV_PROM_OFFSET + 0x19) << 8;
-	if (nv_rd08(dev, NV_PROM_OFFSET + pcir_ptr) != 'P' ||
-	    nv_rd08(dev, NV_PROM_OFFSET + pcir_ptr + 1) != 'C' ||
-	    nv_rd08(dev, NV_PROM_OFFSET + pcir_ptr + 2) != 'I' ||
-	    nv_rd08(dev, NV_PROM_OFFSET + pcir_ptr + 3) != 'R')
+	pcir = nv_rd08(dev, NV_PROM_OFFSET + 0x18) |
+	       nv_rd08(dev, NV_PROM_OFFSET + 0x19) << 8;
+	if (nv_rd08(dev, NV_PROM_OFFSET + pcir + 0) != 'P' ||
+	    nv_rd08(dev, NV_PROM_OFFSET + pcir + 1) != 'C' ||
+	    nv_rd08(dev, NV_PROM_OFFSET + pcir + 2) != 'I' ||
+	    nv_rd08(dev, NV_PROM_OFFSET + pcir + 3) != 'R')
 		goto out;
 
-	/* on some 6600GT/6800LE prom reads are messed up.  nvclock alleges a
-	 * a good read may be obtained by waiting or re-reading (cargocult: 5x)
-	 * each byte.  we'll hope pramin has something usable instead
-	 */
-	for (i = 0; i < NV_PROM_SIZE; i++)
-		data[i] = nv_rd08(dev, NV_PROM_OFFSET + i);
+	/* read entire bios image to system memory */
+	bios->length = nv_rd08(dev, NV_PROM_OFFSET + 2) * 512;
+	bios->data = kmalloc(bios->length, GFP_KERNEL);
+	if (bios->data) {
+		for (i = 0; i < bios->length; i++)
+			bios->data[i] = nv_rd08(dev, NV_PROM_OFFSET + i);
+	}
 
 out:
-	/* disable ROM access */
-	nvWriteMC(dev, pci_nv_20,
-		  save_pci_nv_20 | NV_PBUS_PCI_NV_20_ROM_SHADOW_ENABLED);
+	/* disable access to rom */
+	nv_wr32(dev, pcireg, access);
 }
 
-static void load_vbios_pramin(struct drm_device *dev, uint8_t *data)
+static void
+bios_shadow_pramin(struct nvbios *bios)
 {
+	struct drm_device *dev = bios->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t old_bar0_pramin = 0;
+	u32 bar0 = 0;
 	int i;
 
 	if (dev_priv->card_type >= NV_50) {
 		u64 addr = (u64)(nv_rd32(dev, 0x619f04) & 0xffffff00) << 8;
 		if (!addr) {
-			addr  = (u64)nv_rd32(dev, 0x1700) << 16;
+			addr  = (u64)nv_rd32(dev, 0x001700) << 16;
 			addr += 0xf0000;
 		}
 
-		old_bar0_pramin = nv_rd32(dev, 0x1700);
-		nv_wr32(dev, 0x1700, addr >> 16);
+		bar0 = nv_mask(dev, 0x001700, 0xffffffff, addr >> 16);
 	}
 
 	/* bail if no rom signature */
-	if (nv_rd08(dev, NV_PRAMIN_OFFSET) != 0x55 ||
+	if (nv_rd08(dev, NV_PRAMIN_OFFSET + 0) != 0x55 ||
 	    nv_rd08(dev, NV_PRAMIN_OFFSET + 1) != 0xaa)
 		goto out;
 
-	for (i = 0; i < NV_PROM_SIZE; i++)
-		data[i] = nv_rd08(dev, NV_PRAMIN_OFFSET + i);
+	bios->length = nv_rd08(dev, NV_PRAMIN_OFFSET + 2) * 512;
+	bios->data = kmalloc(bios->length, GFP_KERNEL);
+	if (bios->data) {
+		for (i = 0; i < bios->length; i++)
+			bios->data[i] = nv_rd08(dev, NV_PRAMIN_OFFSET + i);
+	}
 
 out:
 	if (dev_priv->card_type >= NV_50)
-		nv_wr32(dev, 0x1700, old_bar0_pramin);
+		nv_wr32(dev, 0x001700, bar0);
 }
 
-static void load_vbios_pci(struct drm_device *dev, uint8_t *data)
+static void
+bios_shadow_pci(struct nvbios *bios)
+{
+	struct pci_dev *pdev = bios->dev->pdev;
+	size_t length;
+
+	if (!pci_enable_rom(pdev)) {
+		void __iomem *rom = pci_map_rom(pdev, &length);
+		if (rom) {
+			bios->data = kmalloc(length, GFP_KERNEL);
+			if (bios->data) {
+				memcpy_fromio(bios->data, rom, length);
+				bios->length = length;
+			}
+			pci_unmap_rom(pdev, rom);
+		}
+
+		pci_disable_rom(pdev);
+	}
+}
+
+static void
+bios_shadow_acpi(struct nvbios *bios)
 {
-	void __iomem *rom = NULL;
-	size_t rom_len;
-	int ret;
+	struct pci_dev *pdev = bios->dev->pdev;
+	int ptr, len, ret;
+	u8 data[3];
 
-	ret = pci_enable_rom(dev->pdev);
-	if (ret)
+	if (!nouveau_acpi_rom_supported(pdev))
 		return;
 
-	rom = pci_map_rom(dev->pdev, &rom_len);
-	if (!rom)
-		goto out;
-	memcpy_fromio(data, rom, rom_len);
-	pci_unmap_rom(dev->pdev, rom);
+	ret = nouveau_acpi_get_bios_chunk(data, 0, sizeof(data));
+	if (ret != sizeof(data))
+		return;
 
-out:
-	pci_disable_rom(dev->pdev);
-}
+	bios->length = min(data[2] * 512, 65536);
+	bios->data = kmalloc(bios->length, GFP_KERNEL);
+	if (!bios->data)
+		return;
 
-static void load_vbios_acpi(struct drm_device *dev, uint8_t *data)
-{
-	int i;
-	int ret;
-	int size = 64 * 1024;
+	len = bios->length;
+	ptr = 0;
+	while (len) {
+		int size = (len > ROM_BIOS_PAGE) ? ROM_BIOS_PAGE : len;
 
-	if (!nouveau_acpi_rom_supported(dev->pdev))
-		return;
+		ret = nouveau_acpi_get_bios_chunk(bios->data, ptr, size);
+		if (ret != size) {
+			kfree(bios->data);
+			bios->data = NULL;
+			return;
+		}
 
-	for (i = 0; i < (size / ROM_BIOS_PAGE); i++) {
-		ret = nouveau_acpi_get_bios_chunk(data,
-						  (i * ROM_BIOS_PAGE),
-						  ROM_BIOS_PAGE);
-		if (ret <= 0)
-			break;
+		len -= size;
+		ptr += size;
 	}
-	return;
 }
 
 struct methods {
 	const char desc[8];
-	void (*loadbios)(struct drm_device *, uint8_t *);
+	void (*shadow)(struct nvbios *);
 	const bool rw;
+	int score;
+	u32 size;
+	u8 *data;
 };
 
-static struct methods shadow_methods[] = {
-	{ "PRAMIN", load_vbios_pramin, true },
-	{ "PROM", load_vbios_prom, false },
-	{ "PCIROM", load_vbios_pci, true },
-	{ "ACPI", load_vbios_acpi, true },
-};
-#define NUM_SHADOW_METHODS ARRAY_SIZE(shadow_methods)
-
-static bool NVShadowVBIOS(struct drm_device *dev, uint8_t *data)
-{
-	struct methods *methods = shadow_methods;
-	int testscore = 3;
-	int scores[NUM_SHADOW_METHODS], i;
+static bool
+bios_shadow(struct drm_device *dev)
+{
+	struct methods shadow_methods[] = {
+		{ "PRAMIN", bios_shadow_pramin, true, 0, 0, NULL },
+		{ "PROM", bios_shadow_prom, false, 0, 0, NULL },
+		{ "ACPI", bios_shadow_acpi, true, 0, 0, NULL },
+		{ "PCIROM", bios_shadow_pci, true, 0, 0, NULL },
+		{}
+	};
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nvbios *bios = &dev_priv->vbios;
+	struct methods *mthd, *best;
 
 	if (nouveau_vbios) {
-		for (i = 0; i < NUM_SHADOW_METHODS; i++)
-			if (!strcasecmp(nouveau_vbios, methods[i].desc))
-				break;
-
-		if (i < NUM_SHADOW_METHODS) {
-			NV_INFO(dev, "Attempting to use BIOS image from %s\n",
-				methods[i].desc);
+		mthd = shadow_methods;
+		do {
+			if (strcasecmp(nouveau_vbios, mthd->desc))
+				continue;
+			NV_INFO(dev, "VBIOS source: %s\n", mthd->desc);
 
-			methods[i].loadbios(dev, data);
-			if (score_vbios(dev, data, methods[i].rw))
+			mthd->shadow(bios);
+			mthd->score = score_vbios(bios, mthd->rw);
+			if (mthd->score)
 				return true;
-		}
+		} while ((++mthd)->shadow);
 
 		NV_ERROR(dev, "VBIOS source \'%s\' invalid\n", nouveau_vbios);
 	}
 
-	for (i = 0; i < NUM_SHADOW_METHODS; i++) {
-		NV_TRACE(dev, "Attempting to load BIOS image from %s\n",
-			 methods[i].desc);
-		data[0] = data[1] = 0;	/* avoid reuse of previous image */
-		methods[i].loadbios(dev, data);
-		scores[i] = score_vbios(dev, data, methods[i].rw);
-		if (scores[i] == testscore)
-			return true;
-	}
-
-	while (--testscore > 0) {
-		for (i = 0; i < NUM_SHADOW_METHODS; i++) {
-			if (scores[i] == testscore) {
-				NV_TRACE(dev, "Using BIOS image from %s\n",
-					 methods[i].desc);
-				methods[i].loadbios(dev, data);
-				return true;
-			}
+	mthd = shadow_methods;
+	do {
+		NV_TRACE(dev, "Checking %s for VBIOS\n", mthd->desc);
+		mthd->shadow(bios);
+		mthd->score = score_vbios(bios, mthd->rw);
+		mthd->size = bios->length;
+		mthd->data = bios->data;
+	} while (mthd->score != 3 && (++mthd)->shadow);
+
+	mthd = shadow_methods;
+	best = mthd;
+	do {
+		if (mthd->score > best->score) {
+			kfree(best->data);
+			best = mthd;
 		}
+	} while ((++mthd)->shadow);
+
+	if (best->score) {
+		NV_TRACE(dev, "Using VBIOS from %s\n", best->desc);
+		bios->length = best->size;
+		bios->data = best->data;
+		return true;
 	}
 
-	NV_ERROR(dev, "No valid BIOS image found\n");
+	NV_ERROR(dev, "No valid VBIOS image found\n");
 	return false;
 }
 
@@ -6334,11 +6371,7 @@ static bool NVInitVBIOS(struct drm_device *dev)
 	spin_lock_init(&bios->lock);
 	bios->dev = dev;
 
-	if (!NVShadowVBIOS(dev, bios->data))
-		return false;
-
-	bios->length = NV_PROM_SIZE;
-	return true;
+	return bios_shadow(dev);
 }
 
 static int nouveau_parse_vbios_struct(struct drm_device *dev)
@@ -6498,6 +6531,10 @@ nouveau_bios_init(struct drm_device *dev)
 void
 nouveau_bios_takedown(struct drm_device *dev)
 {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
 	nouveau_mxm_fini(dev);
 	nouveau_i2c_fini(dev);
+
+	kfree(dev_priv->vbios.data);
 }
diff --git a/drivers/gpu/drm/nouveau/nouveau_bios.h b/drivers/gpu/drm/nouveau/nouveau_bios.h
index a37c31e..1f3233d 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bios.h
+++ b/drivers/gpu/drm/nouveau/nouveau_bios.h
@@ -75,6 +75,8 @@ enum dcb_connector_type {
 	DCB_CONNECTOR_eDP = 0x47,
 	DCB_CONNECTOR_HDMI_0 = 0x60,
 	DCB_CONNECTOR_HDMI_1 = 0x61,
+	DCB_CONNECTOR_DMS59_DP0 = 0x64,
+	DCB_CONNECTOR_DMS59_DP1 = 0x65,
 	DCB_CONNECTOR_NONE = 0xff
 };
 
@@ -209,6 +211,8 @@ struct nvbios {
 		NVBIOS_BIT
 	} type;
 	uint16_t offset;
+	uint32_t length;
+	uint8_t *data;
 
 	uint8_t chip_version;
 
@@ -219,8 +223,6 @@ struct nvbios {
 
 	spinlock_t lock;
 
-	uint8_t data[NV_PROM_SIZE];
-	unsigned int length;
 	bool execute;
 
 	uint8_t major_version;
diff --git a/drivers/gpu/drm/nouveau/nouveau_connector.c b/drivers/gpu/drm/nouveau/nouveau_connector.c
index f3ce34b..9f9d50d 100644
--- a/drivers/gpu/drm/nouveau/nouveau_connector.c
+++ b/drivers/gpu/drm/nouveau/nouveau_connector.c
@@ -519,6 +519,19 @@ nouveau_connector_set_property(struct drm_connector *connector,
 		return nv_crtc->set_dither(nv_crtc, true);
 	}
 
+	if (nv_crtc && nv_crtc->set_color_vibrance) {
+		/* Hue */
+		if (property == disp->vibrant_hue_property) {
+			nv_crtc->vibrant_hue = value - 90;
+			return nv_crtc->set_color_vibrance(nv_crtc, true);
+		}
+		/* Saturation */
+		if (property == disp->color_vibrance_property) {
+			nv_crtc->color_vibrance = value - 100;
+			return nv_crtc->set_color_vibrance(nv_crtc, true);
+		}
+	}
+
 	if (nv_encoder && nv_encoder->dcb->type == OUTPUT_TV)
 		return get_slave_funcs(encoder)->set_property(
 			encoder, connector, property, value);
@@ -858,6 +871,8 @@ drm_conntype_from_dcb(enum dcb_connector_type dcb)
 	case DCB_CONNECTOR_DVI_D    : return DRM_MODE_CONNECTOR_DVID;
 	case DCB_CONNECTOR_LVDS     :
 	case DCB_CONNECTOR_LVDS_SPWG: return DRM_MODE_CONNECTOR_LVDS;
+	case DCB_CONNECTOR_DMS59_DP0:
+	case DCB_CONNECTOR_DMS59_DP1:
 	case DCB_CONNECTOR_DP       : return DRM_MODE_CONNECTOR_DisplayPort;
 	case DCB_CONNECTOR_eDP      : return DRM_MODE_CONNECTOR_eDP;
 	case DCB_CONNECTOR_HDMI_0   :
@@ -1002,7 +1017,9 @@ nouveau_connector_create(struct drm_device *dev, int index)
 	     nv_connector->type == DCB_CONNECTOR_DVI_I ||
 	     nv_connector->type == DCB_CONNECTOR_HDMI_0 ||
 	     nv_connector->type == DCB_CONNECTOR_HDMI_1 ||
-	     nv_connector->type == DCB_CONNECTOR_DP)) {
+	     nv_connector->type == DCB_CONNECTOR_DP ||
+	     nv_connector->type == DCB_CONNECTOR_DMS59_DP0 ||
+	     nv_connector->type == DCB_CONNECTOR_DMS59_DP1)) {
 		drm_connector_attach_property(connector,
 					      disp->underscan_property,
 					      UNDERSCAN_OFF);
@@ -1014,6 +1031,16 @@ nouveau_connector_create(struct drm_device *dev, int index)
 					      0);
 	}
 
+	/* Add hue and saturation options */
+	if (disp->vibrant_hue_property)
+		drm_connector_attach_property(connector,
+					      disp->vibrant_hue_property,
+					      90);
+	if (disp->color_vibrance_property)
+		drm_connector_attach_property(connector,
+					      disp->color_vibrance_property,
+					      150);
+
 	switch (nv_connector->type) {
 	case DCB_CONNECTOR_VGA:
 		if (dev_priv->card_type >= NV_50) {
diff --git a/drivers/gpu/drm/nouveau/nouveau_crtc.h b/drivers/gpu/drm/nouveau/nouveau_crtc.h
index 686f6b4..e6d0d1e 100644
--- a/drivers/gpu/drm/nouveau/nouveau_crtc.h
+++ b/drivers/gpu/drm/nouveau/nouveau_crtc.h
@@ -35,6 +35,8 @@ struct nouveau_crtc {
 	uint32_t dpms_saved_fp_control;
 	uint32_t fp_users;
 	int saturation;
+	int color_vibrance;
+	int vibrant_hue;
 	int sharpness;
 	int last_dpms;
 
@@ -67,6 +69,7 @@ struct nouveau_crtc {
 
 	int (*set_dither)(struct nouveau_crtc *crtc, bool update);
 	int (*set_scale)(struct nouveau_crtc *crtc, bool update);
+	int (*set_color_vibrance)(struct nouveau_crtc *crtc, bool update);
 };
 
 static inline struct nouveau_crtc *nouveau_crtc(struct drm_crtc *crtc)
diff --git a/drivers/gpu/drm/nouveau/nouveau_display.c b/drivers/gpu/drm/nouveau/nouveau_display.c
index 5565e50..35acc92 100644
--- a/drivers/gpu/drm/nouveau/nouveau_display.c
+++ b/drivers/gpu/drm/nouveau/nouveau_display.c
@@ -286,6 +286,20 @@ nouveau_display_create(struct drm_device *dev)
 	disp->underscan_vborder_property =
 		drm_property_create_range(dev, 0, "underscan vborder", 0, 128);
 
+	if (gen == 1) {
+		disp->vibrant_hue_property =
+			drm_property_create(dev, DRM_MODE_PROP_RANGE,
+					    "vibrant hue", 2);
+		disp->vibrant_hue_property->values[0] = 0;
+		disp->vibrant_hue_property->values[1] = 180; /* -90..+90 */
+
+		disp->color_vibrance_property =
+			drm_property_create(dev, DRM_MODE_PROP_RANGE,
+					    "color vibrance", 2);
+		disp->color_vibrance_property->values[0] = 0;
+		disp->color_vibrance_property->values[1] = 200; /* -100..+100 */
+	}
+
 	dev->mode_config.funcs = (void *)&nouveau_mode_config_funcs;
 	dev->mode_config.fb_base = pci_resource_start(dev->pdev, 1);
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_dp.c b/drivers/gpu/drm/nouveau/nouveau_dp.c
index 9b93b70..302b2f7 100644
--- a/drivers/gpu/drm/nouveau/nouveau_dp.c
+++ b/drivers/gpu/drm/nouveau/nouveau_dp.c
@@ -161,116 +161,6 @@ out:
 	return ret;
 }
 
-static u32
-dp_link_bw_get(struct drm_device *dev, int or, int link)
-{
-	u32 ctrl = nv_rd32(dev, 0x614300 + (or * 0x800));
-	if (!(ctrl & 0x000c0000))
-		return 162000;
-	return 270000;
-}
-
-static int
-dp_lane_count_get(struct drm_device *dev, int or, int link)
-{
-	u32 ctrl = nv_rd32(dev, NV50_SOR_DP_CTRL(or, link));
-	switch (ctrl & 0x000f0000) {
-	case 0x00010000: return 1;
-	case 0x00030000: return 2;
-	default:
-		return 4;
-	}
-}
-
-void
-nouveau_dp_tu_update(struct drm_device *dev, int or, int link, u32 clk, u32 bpp)
-{
-	const u32 symbol = 100000;
-	int bestTU = 0, bestVTUi = 0, bestVTUf = 0, bestVTUa = 0;
-	int TU, VTUi, VTUf, VTUa;
-	u64 link_data_rate, link_ratio, unk;
-	u32 best_diff = 64 * symbol;
-	u32 link_nr, link_bw, r;
-
-	/* calculate packed data rate for each lane */
-	link_nr = dp_lane_count_get(dev, or, link);
-	link_data_rate = (clk * bpp / 8) / link_nr;
-
-	/* calculate ratio of packed data rate to link symbol rate */
-	link_bw = dp_link_bw_get(dev, or, link);
-	link_ratio = link_data_rate * symbol;
-	r = do_div(link_ratio, link_bw);
-
-	for (TU = 64; TU >= 32; TU--) {
-		/* calculate average number of valid symbols in each TU */
-		u32 tu_valid = link_ratio * TU;
-		u32 calc, diff;
-
-		/* find a hw representation for the fraction.. */
-		VTUi = tu_valid / symbol;
-		calc = VTUi * symbol;
-		diff = tu_valid - calc;
-		if (diff) {
-			if (diff >= (symbol / 2)) {
-				VTUf = symbol / (symbol - diff);
-				if (symbol - (VTUf * diff))
-					VTUf++;
-
-				if (VTUf <= 15) {
-					VTUa  = 1;
-					calc += symbol - (symbol / VTUf);
-				} else {
-					VTUa  = 0;
-					VTUf  = 1;
-					calc += symbol;
-				}
-			} else {
-				VTUa  = 0;
-				VTUf  = min((int)(symbol / diff), 15);
-				calc += symbol / VTUf;
-			}
-
-			diff = calc - tu_valid;
-		} else {
-			/* no remainder, but the hw doesn't like the fractional
-			 * part to be zero.  decrement the integer part and
-			 * have the fraction add a whole symbol back
-			 */
-			VTUa = 0;
-			VTUf = 1;
-			VTUi--;
-		}
-
-		if (diff < best_diff) {
-			best_diff = diff;
-			bestTU = TU;
-			bestVTUa = VTUa;
-			bestVTUf = VTUf;
-			bestVTUi = VTUi;
-			if (diff == 0)
-				break;
-		}
-	}
-
-	if (!bestTU) {
-		NV_ERROR(dev, "DP: unable to find suitable config\n");
-		return;
-	}
-
-	/* XXX close to vbios numbers, but not right */
-	unk  = (symbol - link_ratio) * bestTU;
-	unk *= link_ratio;
-	r = do_div(unk, symbol);
-	r = do_div(unk, symbol);
-	unk += 6;
-
-	nv_mask(dev, NV50_SOR_DP_CTRL(or, link), 0x000001fc, bestTU << 2);
-	nv_mask(dev, NV50_SOR_DP_SCFG(or, link), 0x010f7f3f, bestVTUa << 24 |
-							     bestVTUf << 16 |
-							     bestVTUi << 8 |
-							     unk);
-}
-
 u8 *
 nouveau_dp_bios_data(struct drm_device *dev, struct dcb_entry *dcb, u8 **entry)
 {
@@ -318,13 +208,10 @@ nouveau_dp_bios_data(struct drm_device *dev, struct dcb_entry *dcb, u8 **entry)
  * link training
  *****************************************************************************/
 struct dp_state {
+	struct dp_train_func *func;
 	struct dcb_entry *dcb;
-	u8 *table;
-	u8 *entry;
 	int auxch;
 	int crtc;
-	int or;
-	int link;
 	u8 *dpcd;
 	int link_nr;
 	u32 link_bw;
@@ -335,142 +222,58 @@ struct dp_state {
 static void
 dp_set_link_config(struct drm_device *dev, struct dp_state *dp)
 {
-	int or = dp->or, link = dp->link;
-	u8 *entry, sink[2];
-	u32 dp_ctrl;
-	u16 script;
+	u8 sink[2];
 
 	NV_DEBUG_KMS(dev, "%d lanes at %d KB/s\n", dp->link_nr, dp->link_bw);
 
-	/* set selected link rate on source */
-	switch (dp->link_bw) {
-	case 270000:
-		nv_mask(dev, 0x614300 + (or * 0x800), 0x000c0000, 0x00040000);
-		sink[0] = DP_LINK_BW_2_7;
-		break;
-	default:
-		nv_mask(dev, 0x614300 + (or * 0x800), 0x000c0000, 0x00000000);
-		sink[0] = DP_LINK_BW_1_62;
-		break;
-	}
-
-	/* offset +0x0a of each dp encoder table entry is a pointer to another
-	 * table, that has (among other things) pointers to more scripts that
-	 * need to be executed, this time depending on link speed.
-	 */
-	entry = ROMPTR(dev, dp->entry[10]);
-	if (entry) {
-		if (dp->table[0] < 0x30) {
-			while (dp->link_bw < (ROM16(entry[0]) * 10))
-				entry += 4;
-			script = ROM16(entry[2]);
-		} else {
-			while (dp->link_bw < (entry[0] * 27000))
-				entry += 3;
-			script = ROM16(entry[1]);
-		}
-
-		nouveau_bios_run_init_table(dev, script, dp->dcb, dp->crtc);
-	}
+	/* set desired link configuration on the source */
+	dp->func->link_set(dev, dp->dcb, dp->crtc, dp->link_nr, dp->link_bw,
+			   dp->dpcd[2] & DP_ENHANCED_FRAME_CAP);
 
-	/* configure lane count on the source */
-	dp_ctrl = ((1 << dp->link_nr) - 1) << 16;
+	/* inform the sink of the new configuration */
+	sink[0] = dp->link_bw / 27000;
 	sink[1] = dp->link_nr;
-	if (dp->dpcd[2] & DP_ENHANCED_FRAME_CAP) {
-		dp_ctrl |= 0x00004000;
+	if (dp->dpcd[2] & DP_ENHANCED_FRAME_CAP)
 		sink[1] |= DP_LANE_COUNT_ENHANCED_FRAME_EN;
-	}
-
-	nv_mask(dev, NV50_SOR_DP_CTRL(or, link), 0x001f4000, dp_ctrl);
 
-	/* inform the sink of the new configuration */
 	auxch_tx(dev, dp->auxch, 8, DP_LINK_BW_SET, sink, 2);
 }
 
 static void
-dp_set_training_pattern(struct drm_device *dev, struct dp_state *dp, u8 tp)
+dp_set_training_pattern(struct drm_device *dev, struct dp_state *dp, u8 pattern)
 {
 	u8 sink_tp;
 
-	NV_DEBUG_KMS(dev, "training pattern %d\n", tp);
+	NV_DEBUG_KMS(dev, "training pattern %d\n", pattern);
 
-	nv_mask(dev, NV50_SOR_DP_CTRL(dp->or, dp->link), 0x0f000000, tp << 24);
+	dp->func->train_set(dev, dp->dcb, pattern);
 
 	auxch_tx(dev, dp->auxch, 9, DP_TRAINING_PATTERN_SET, &sink_tp, 1);
 	sink_tp &= ~DP_TRAINING_PATTERN_MASK;
-	sink_tp |= tp;
+	sink_tp |= pattern;
 	auxch_tx(dev, dp->auxch, 8, DP_TRAINING_PATTERN_SET, &sink_tp, 1);
 }
 
-static const u8 nv50_lane_map[] = { 16, 8, 0, 24 };
-static const u8 nvaf_lane_map[] = { 24, 16, 8, 0 };
-
 static int
 dp_link_train_commit(struct drm_device *dev, struct dp_state *dp)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	u32 mask = 0, drv = 0, pre = 0, unk = 0;
-	const u8 *shifts;
-	int link = dp->link;
-	int or = dp->or;
 	int i;
 
-	if (dev_priv->chipset != 0xaf)
-		shifts = nv50_lane_map;
-	else
-		shifts = nvaf_lane_map;
-
 	for (i = 0; i < dp->link_nr; i++) {
-		u8 *conf = dp->entry + dp->table[4];
 		u8 lane = (dp->stat[4 + (i >> 1)] >> ((i & 1) * 4)) & 0xf;
 		u8 lpre = (lane & 0x0c) >> 2;
 		u8 lvsw = (lane & 0x03) >> 0;
 
-		mask |= 0xff << shifts[i];
-		unk |= 1 << (shifts[i] >> 3);
-
 		dp->conf[i] = (lpre << 3) | lvsw;
 		if (lvsw == DP_TRAIN_VOLTAGE_SWING_1200)
 			dp->conf[i] |= DP_TRAIN_MAX_SWING_REACHED;
-		if (lpre == DP_TRAIN_PRE_EMPHASIS_9_5)
+		if ((lpre << 3) == DP_TRAIN_PRE_EMPHASIS_9_5)
 			dp->conf[i] |= DP_TRAIN_MAX_PRE_EMPHASIS_REACHED;
 
 		NV_DEBUG_KMS(dev, "config lane %d %02x\n", i, dp->conf[i]);
-
-		if (dp->table[0] < 0x30) {
-			u8 *last = conf + (dp->entry[4] * dp->table[5]);
-			while (lvsw != conf[0] || lpre != conf[1]) {
-				conf += dp->table[5];
-				if (conf >= last)
-					return -EINVAL;
-			}
-
-			conf += 2;
-		} else {
-			/* no lookup table anymore, set entries for each
-			 * combination of voltage swing and pre-emphasis
-			 * level allowed by the DP spec.
-			 */
-			switch (lvsw) {
-			case 0: lpre += 0; break;
-			case 1: lpre += 4; break;
-			case 2: lpre += 7; break;
-			case 3: lpre += 9; break;
-			}
-
-			conf = conf + (lpre * dp->table[5]);
-			conf++;
-		}
-
-		drv |= conf[0] << shifts[i];
-		pre |= conf[1] << shifts[i];
-		unk  = (unk & ~0x0000ff00) | (conf[2] << 8);
+		dp->func->train_adj(dev, dp->dcb, i, lvsw, lpre);
 	}
 
-	nv_mask(dev, NV50_SOR_DP_UNK118(or, link), mask, drv);
-	nv_mask(dev, NV50_SOR_DP_UNK120(or, link), mask, pre);
-	nv_mask(dev, NV50_SOR_DP_UNK130(or, link), 0x0000ff0f, unk);
-
 	return auxch_tx(dev, dp->auxch, 8, DP_TRAINING_LANE0_SET, dp->conf, 4);
 }
 
@@ -554,8 +357,50 @@ dp_link_train_eq(struct drm_device *dev, struct dp_state *dp)
 	return eq_done ? 0 : -1;
 }
 
+static void
+dp_set_downspread(struct drm_device *dev, struct dp_state *dp, bool enable)
+{
+	u16 script = 0x0000;
+	u8 *entry, *table = nouveau_dp_bios_data(dev, dp->dcb, &entry);
+	if (table) {
+		if (table[0] >= 0x20 && table[0] <= 0x30) {
+			if (enable) script = ROM16(entry[12]);
+			else        script = ROM16(entry[14]);
+		}
+	}
+
+	nouveau_bios_run_init_table(dev, script, dp->dcb, dp->crtc);
+}
+
+static void
+dp_link_train_init(struct drm_device *dev, struct dp_state *dp)
+{
+	u16 script = 0x0000;
+	u8 *entry, *table = nouveau_dp_bios_data(dev, dp->dcb, &entry);
+	if (table) {
+		if (table[0] >= 0x20 && table[0] <= 0x30)
+			script = ROM16(entry[6]);
+	}
+
+	nouveau_bios_run_init_table(dev, script, dp->dcb, dp->crtc);
+}
+
+static void
+dp_link_train_fini(struct drm_device *dev, struct dp_state *dp)
+{
+	u16 script = 0x0000;
+	u8 *entry, *table = nouveau_dp_bios_data(dev, dp->dcb, &entry);
+	if (table) {
+		if (table[0] >= 0x20 && table[0] <= 0x30)
+			script = ROM16(entry[8]);
+	}
+
+	nouveau_bios_run_init_table(dev, script, dp->dcb, dp->crtc);
+}
+
 bool
-nouveau_dp_link_train(struct drm_encoder *encoder, u32 datarate)
+nouveau_dp_link_train(struct drm_encoder *encoder, u32 datarate,
+		      struct dp_train_func *func)
 {
 	struct nouveau_encoder *nv_encoder = nouveau_encoder(encoder);
 	struct nouveau_crtc *nv_crtc = nouveau_crtc(encoder->crtc);
@@ -571,17 +416,15 @@ nouveau_dp_link_train(struct drm_encoder *encoder, u32 datarate)
 	if (!auxch)
 		return false;
 
-	dp.table = nouveau_dp_bios_data(dev, nv_encoder->dcb, &dp.entry);
-	if (!dp.table)
-		return -EINVAL;
-
+	dp.func = func;
 	dp.dcb = nv_encoder->dcb;
 	dp.crtc = nv_crtc->index;
 	dp.auxch = auxch->drive;
-	dp.or = nv_encoder->or;
-	dp.link = !(nv_encoder->dcb->sorconf.link & 1);
 	dp.dpcd = nv_encoder->dp.dpcd;
 
+	/* adjust required bandwidth for 8B/10B coding overhead */
+	datarate = (datarate / 8) * 10;
+
 	/* some sinks toggle hotplug in response to some of the actions
 	 * we take during link training (DP_SET_POWER is one), we need
 	 * to ignore them for the moment to avoid races.
@@ -589,16 +432,10 @@ nouveau_dp_link_train(struct drm_encoder *encoder, u32 datarate)
 	nouveau_gpio_irq(dev, 0, nv_connector->hpd, 0xff, false);
 
 	/* enable down-spreading, if possible */
-	if (dp.table[1] >= 16) {
-		u16 script = ROM16(dp.entry[14]);
-		if (nv_encoder->dp.dpcd[3] & 1)
-			script = ROM16(dp.entry[12]);
-
-		nouveau_bios_run_init_table(dev, script, dp.dcb, dp.crtc);
-	}
+	dp_set_downspread(dev, &dp, nv_encoder->dp.dpcd[3] & 1);
 
 	/* execute pre-train script from vbios */
-	nouveau_bios_run_init_table(dev, ROM16(dp.entry[6]), dp.dcb, dp.crtc);
+	dp_link_train_init(dev, &dp);
 
 	/* start off at highest link rate supported by encoder and display */
 	while (*link_bw > nv_encoder->dp.link_bw)
@@ -632,13 +469,36 @@ nouveau_dp_link_train(struct drm_encoder *encoder, u32 datarate)
 	dp_set_training_pattern(dev, &dp, DP_TRAINING_PATTERN_DISABLE);
 
 	/* execute post-train script from vbios */
-	nouveau_bios_run_init_table(dev, ROM16(dp.entry[8]), dp.dcb, dp.crtc);
+	dp_link_train_fini(dev, &dp);
 
 	/* re-enable hotplug detect */
 	nouveau_gpio_irq(dev, 0, nv_connector->hpd, 0xff, true);
 	return true;
 }
 
+void
+nouveau_dp_dpms(struct drm_encoder *encoder, int mode, u32 datarate,
+		struct dp_train_func *func)
+{
+	struct nouveau_encoder *nv_encoder = nouveau_encoder(encoder);
+	struct nouveau_i2c_chan *auxch;
+	u8 status;
+
+	auxch = nouveau_i2c_find(encoder->dev, nv_encoder->dcb->i2c_index);
+	if (!auxch)
+		return;
+
+	if (mode == DRM_MODE_DPMS_ON)
+		status = DP_SET_POWER_D0;
+	else
+		status = DP_SET_POWER_D3;
+
+	nouveau_dp_auxch(auxch, 8, DP_SET_POWER, &status, 1);
+
+	if (mode == DRM_MODE_DPMS_ON)
+		nouveau_dp_link_train(encoder, datarate, func);
+}
+
 bool
 nouveau_dp_detect(struct drm_encoder *encoder)
 {
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.c b/drivers/gpu/drm/nouveau/nouveau_drv.c
index 81d7962..4f2030b 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.c
@@ -57,6 +57,10 @@ MODULE_PARM_DESC(vram_notify, "Force DMA notifiers to be in VRAM");
 int nouveau_vram_notify = 0;
 module_param_named(vram_notify, nouveau_vram_notify, int, 0400);
 
+MODULE_PARM_DESC(vram_type, "Override detected VRAM type");
+char *nouveau_vram_type;
+module_param_named(vram_type, nouveau_vram_type, charp, 0400);
+
 MODULE_PARM_DESC(duallink, "Allow dual-link TMDS (>=GeForce 8)");
 int nouveau_duallink = 1;
 module_param_named(duallink, nouveau_duallink, int, 0400);
@@ -89,7 +93,7 @@ MODULE_PARM_DESC(override_conntype, "Ignore DCB connector type");
 int nouveau_override_conntype = 0;
 module_param_named(override_conntype, nouveau_override_conntype, int, 0400);
 
-MODULE_PARM_DESC(tv_disable, "Disable TV-out detection\n");
+MODULE_PARM_DESC(tv_disable, "Disable TV-out detection");
 int nouveau_tv_disable = 0;
 module_param_named(tv_disable, nouveau_tv_disable, int, 0400);
 
@@ -104,27 +108,27 @@ module_param_named(tv_norm, nouveau_tv_norm, charp, 0400);
 MODULE_PARM_DESC(reg_debug, "Register access debug bitmask:\n"
 		"\t\t0x1 mc, 0x2 video, 0x4 fb, 0x8 extdev,\n"
 		"\t\t0x10 crtc, 0x20 ramdac, 0x40 vgacrtc, 0x80 rmvio,\n"
-		"\t\t0x100 vgaattr, 0x200 EVO (G80+). ");
+		"\t\t0x100 vgaattr, 0x200 EVO (G80+)");
 int nouveau_reg_debug;
 module_param_named(reg_debug, nouveau_reg_debug, int, 0600);
 
-MODULE_PARM_DESC(perflvl, "Performance level (default: boot)\n");
+MODULE_PARM_DESC(perflvl, "Performance level (default: boot)");
 char *nouveau_perflvl;
 module_param_named(perflvl, nouveau_perflvl, charp, 0400);
 
-MODULE_PARM_DESC(perflvl_wr, "Allow perflvl changes (warning: dangerous!)\n");
+MODULE_PARM_DESC(perflvl_wr, "Allow perflvl changes (warning: dangerous!)");
 int nouveau_perflvl_wr;
 module_param_named(perflvl_wr, nouveau_perflvl_wr, int, 0400);
 
-MODULE_PARM_DESC(msi, "Enable MSI (default: off)\n");
+MODULE_PARM_DESC(msi, "Enable MSI (default: off)");
 int nouveau_msi;
 module_param_named(msi, nouveau_msi, int, 0400);
 
-MODULE_PARM_DESC(ctxfw, "Use external HUB/GPC ucode (fermi)\n");
+MODULE_PARM_DESC(ctxfw, "Use external HUB/GPC ucode (fermi)");
 int nouveau_ctxfw;
 module_param_named(ctxfw, nouveau_ctxfw, int, 0400);
 
-MODULE_PARM_DESC(mxmdcb, "Santise DCB table according to MXM-SIS\n");
+MODULE_PARM_DESC(mxmdcb, "Santise DCB table according to MXM-SIS");
 int nouveau_mxmdcb = 1;
 module_param_named(mxmdcb, nouveau_mxmdcb, int, 0400);
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index b827098..a184ba3 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -406,6 +406,9 @@ struct nouveau_display_engine {
 	struct drm_property *underscan_property;
 	struct drm_property *underscan_hborder_property;
 	struct drm_property *underscan_vborder_property;
+	/* not really hue and saturation: */
+	struct drm_property *vibrant_hue_property;
+	struct drm_property *color_vibrance_property;
 };
 
 struct nouveau_gpio_engine {
@@ -432,58 +435,85 @@ struct nouveau_pm_voltage {
 	int nr_level;
 };
 
+/* Exclusive upper limits */
+#define NV_MEM_CL_DDR2_MAX 8
+#define NV_MEM_WR_DDR2_MAX 9
+#define NV_MEM_CL_DDR3_MAX 17
+#define NV_MEM_WR_DDR3_MAX 17
+#define NV_MEM_CL_GDDR3_MAX 16
+#define NV_MEM_WR_GDDR3_MAX 18
+#define NV_MEM_CL_GDDR5_MAX 21
+#define NV_MEM_WR_GDDR5_MAX 20
+
 struct nouveau_pm_memtiming {
 	int id;
-	u32 reg_0; /* 0x10f290 on Fermi, 0x100220 for older */
-	u32 reg_1;
-	u32 reg_2;
-	u32 reg_3;
-	u32 reg_4;
-	u32 reg_5;
-	u32 reg_6;
-	u32 reg_7;
-	u32 reg_8;
-	/* To be written to 0x1002c0 */
-	u8 CL;
-	u8 WR;
+
+	u32 reg[9];
+	u32 mr[4];
+
+	u8 tCWL;
+
+	u8 odt;
+	u8 drive_strength;
 };
 
-struct nouveau_pm_tbl_header{
+struct nouveau_pm_tbl_header {
 	u8 version;
 	u8 header_len;
 	u8 entry_cnt;
 	u8 entry_len;
 };
 
-struct nouveau_pm_tbl_entry{
+struct nouveau_pm_tbl_entry {
 	u8 tWR;
-	u8 tUNK_1;
+	u8 tWTR;
 	u8 tCL;
-	u8 tRP;		/* Byte 3 */
+	u8 tRC;
 	u8 empty_4;
-	u8 tRAS;	/* Byte 5 */
+	u8 tRFC;	/* Byte 5 */
 	u8 empty_6;
-	u8 tRFC;	/* Byte 7 */
+	u8 tRAS;	/* Byte 7 */
 	u8 empty_8;
-	u8 tRC;		/* Byte 9 */
-	u8 tUNK_10, tUNK_11, tUNK_12, tUNK_13, tUNK_14;
-	u8 empty_15,empty_16,empty_17;
-	u8 tUNK_18, tUNK_19, tUNK_20, tUNK_21;
+	u8 tRP;		/* Byte 9 */
+	u8 tRCDRD;
+	u8 tRCDWR;
+	u8 tRRD;
+	u8 tUNK_13;
+	u8 RAM_FT1;		/* 14, a bitmask of random RAM features */
+	u8 empty_15;
+	u8 tUNK_16;
+	u8 empty_17;
+	u8 tUNK_18;
+	u8 tCWL;
+	u8 tUNK_20, tUNK_21;
 };
 
-/* nouveau_mem.c */
-void nv30_mem_timing_entry(struct drm_device *dev, struct nouveau_pm_tbl_header *hdr,
-							struct nouveau_pm_tbl_entry *e, uint8_t magic_number,
-							struct nouveau_pm_memtiming *timing);
+struct nouveau_pm_profile;
+struct nouveau_pm_profile_func {
+	void (*destroy)(struct nouveau_pm_profile *);
+	void (*init)(struct nouveau_pm_profile *);
+	void (*fini)(struct nouveau_pm_profile *);
+	struct nouveau_pm_level *(*select)(struct nouveau_pm_profile *);
+};
+
+struct nouveau_pm_profile {
+	const struct nouveau_pm_profile_func *func;
+	struct list_head head;
+	char name[8];
+};
 
 #define NOUVEAU_PM_MAX_LEVEL 8
 struct nouveau_pm_level {
+	struct nouveau_pm_profile profile;
 	struct device_attribute dev_attr;
 	char name[32];
 	int id;
 
-	u32 core;
+	struct nouveau_pm_memtiming timing;
 	u32 memory;
+	u16 memscript;
+
+	u32 core;
 	u32 shader;
 	u32 rop;
 	u32 copy;
@@ -498,9 +528,6 @@ struct nouveau_pm_level {
 	u32 volt_min; /* microvolts */
 	u32 volt_max;
 	u8  fanspeed;
-
-	u16 memscript;
-	struct nouveau_pm_memtiming *timing;
 };
 
 struct nouveau_pm_temp_sensor_constants {
@@ -517,27 +544,26 @@ struct nouveau_pm_threshold_temp {
 	s16 fan_boost;
 };
 
-struct nouveau_pm_memtimings {
-	bool supported;
-	struct nouveau_pm_memtiming *timing;
-	int nr_timing;
-};
-
 struct nouveau_pm_fan {
+	u32 percent;
 	u32 min_duty;
 	u32 max_duty;
 	u32 pwm_freq;
+	u32 pwm_divisor;
 };
 
 struct nouveau_pm_engine {
 	struct nouveau_pm_voltage voltage;
 	struct nouveau_pm_level perflvl[NOUVEAU_PM_MAX_LEVEL];
 	int nr_perflvl;
-	struct nouveau_pm_memtimings memtimings;
 	struct nouveau_pm_temp_sensor_constants sensor_constants;
 	struct nouveau_pm_threshold_temp threshold_temp;
 	struct nouveau_pm_fan fan;
-	u32 pwm_divisor;
+
+	struct nouveau_pm_profile *profile_ac;
+	struct nouveau_pm_profile *profile_dc;
+	struct nouveau_pm_profile *profile;
+	struct list_head profiles;
 
 	struct nouveau_pm_level boot;
 	struct nouveau_pm_level *cur;
@@ -669,14 +695,14 @@ struct nv04_mode_state {
 };
 
 enum nouveau_card_type {
-	NV_04      = 0x00,
+	NV_04      = 0x04,
 	NV_10      = 0x10,
 	NV_20      = 0x20,
 	NV_30      = 0x30,
 	NV_40      = 0x40,
 	NV_50      = 0x50,
 	NV_C0      = 0xc0,
-	NV_D0      = 0xd0
+	NV_D0      = 0xd0,
 };
 
 struct drm_nouveau_private {
@@ -772,8 +798,22 @@ struct drm_nouveau_private {
 	} tile;
 
 	/* VRAM/fb configuration */
+	enum {
+		NV_MEM_TYPE_UNKNOWN = 0,
+		NV_MEM_TYPE_STOLEN,
+		NV_MEM_TYPE_SGRAM,
+		NV_MEM_TYPE_SDRAM,
+		NV_MEM_TYPE_DDR1,
+		NV_MEM_TYPE_DDR2,
+		NV_MEM_TYPE_DDR3,
+		NV_MEM_TYPE_GDDR2,
+		NV_MEM_TYPE_GDDR3,
+		NV_MEM_TYPE_GDDR4,
+		NV_MEM_TYPE_GDDR5
+	} vram_type;
 	uint64_t vram_size;
 	uint64_t vram_sys_base;
+	bool vram_rank_B;
 
 	uint64_t fb_available_size;
 	uint64_t fb_mappable_pages;
@@ -846,6 +886,7 @@ extern int nouveau_uscript_lvds;
 extern int nouveau_uscript_tmds;
 extern int nouveau_vram_pushbuf;
 extern int nouveau_vram_notify;
+extern char *nouveau_vram_type;
 extern int nouveau_fbpercrtc;
 extern int nouveau_tv_disable;
 extern char *nouveau_tv_norm;
@@ -894,8 +935,12 @@ extern void nouveau_mem_gart_fini(struct drm_device *);
 extern int  nouveau_mem_init_agp(struct drm_device *);
 extern int  nouveau_mem_reset_agp(struct drm_device *);
 extern void nouveau_mem_close(struct drm_device *);
-extern int  nouveau_mem_detect(struct drm_device *);
 extern bool nouveau_mem_flags_valid(struct drm_device *, u32 tile_flags);
+extern int  nouveau_mem_timing_calc(struct drm_device *, u32 freq,
+				    struct nouveau_pm_memtiming *);
+extern void nouveau_mem_timing_read(struct drm_device *,
+				    struct nouveau_pm_memtiming *);
+extern int nouveau_mem_vbios_type(struct drm_device *);
 extern struct nouveau_tile_reg *nv10_mem_set_tiling(
 	struct drm_device *dev, uint32_t addr, uint32_t size,
 	uint32_t pitch, uint32_t flags);
@@ -1117,19 +1162,14 @@ int nouveau_ttm_mmap(struct file *, struct vm_area_struct *);
 /* nouveau_hdmi.c */
 void nouveau_hdmi_mode_set(struct drm_encoder *, struct drm_display_mode *);
 
-/* nouveau_dp.c */
-int nouveau_dp_auxch(struct nouveau_i2c_chan *auxch, int cmd, int addr,
-		     uint8_t *data, int data_nr);
-bool nouveau_dp_detect(struct drm_encoder *);
-bool nouveau_dp_link_train(struct drm_encoder *, u32 datarate);
-void nouveau_dp_tu_update(struct drm_device *, int, int, u32, u32);
-u8 *nouveau_dp_bios_data(struct drm_device *, struct dcb_entry *, u8 **);
-
 /* nv04_fb.c */
+extern int  nv04_fb_vram_init(struct drm_device *);
 extern int  nv04_fb_init(struct drm_device *);
 extern void nv04_fb_takedown(struct drm_device *);
 
 /* nv10_fb.c */
+extern int  nv10_fb_vram_init(struct drm_device *dev);
+extern int  nv1a_fb_vram_init(struct drm_device *dev);
 extern int  nv10_fb_init(struct drm_device *);
 extern void nv10_fb_takedown(struct drm_device *);
 extern void nv10_fb_init_tile_region(struct drm_device *dev, int i,
@@ -1138,6 +1178,16 @@ extern void nv10_fb_init_tile_region(struct drm_device *dev, int i,
 extern void nv10_fb_set_tile_region(struct drm_device *dev, int i);
 extern void nv10_fb_free_tile_region(struct drm_device *dev, int i);
 
+/* nv20_fb.c */
+extern int  nv20_fb_vram_init(struct drm_device *dev);
+extern int  nv20_fb_init(struct drm_device *);
+extern void nv20_fb_takedown(struct drm_device *);
+extern void nv20_fb_init_tile_region(struct drm_device *dev, int i,
+				     uint32_t addr, uint32_t size,
+				     uint32_t pitch, uint32_t flags);
+extern void nv20_fb_set_tile_region(struct drm_device *dev, int i);
+extern void nv20_fb_free_tile_region(struct drm_device *dev, int i);
+
 /* nv30_fb.c */
 extern int  nv30_fb_init(struct drm_device *);
 extern void nv30_fb_takedown(struct drm_device *);
@@ -1147,6 +1197,7 @@ extern void nv30_fb_init_tile_region(struct drm_device *dev, int i,
 extern void nv30_fb_free_tile_region(struct drm_device *dev, int i);
 
 /* nv40_fb.c */
+extern int  nv40_fb_vram_init(struct drm_device *dev);
 extern int  nv40_fb_init(struct drm_device *);
 extern void nv40_fb_takedown(struct drm_device *);
 extern void nv40_fb_set_tile_region(struct drm_device *dev, int i);
@@ -1703,6 +1754,7 @@ nv44_graph_class(struct drm_device *dev)
 #define NV_MEM_ACCESS_RW (NV_MEM_ACCESS_RO | NV_MEM_ACCESS_WO)
 #define NV_MEM_ACCESS_SYS 4
 #define NV_MEM_ACCESS_VM  8
+#define NV_MEM_ACCESS_NOSNOOP 16
 
 #define NV_MEM_TARGET_VRAM        0
 #define NV_MEM_TARGET_PCI         1
diff --git a/drivers/gpu/drm/nouveau/nouveau_encoder.h b/drivers/gpu/drm/nouveau/nouveau_encoder.h
index e5d6e3f..3dc14a3 100644
--- a/drivers/gpu/drm/nouveau/nouveau_encoder.h
+++ b/drivers/gpu/drm/nouveau/nouveau_encoder.h
@@ -32,6 +32,14 @@
 
 #define NV_DPMS_CLEARED 0x80
 
+struct dp_train_func {
+	void (*link_set)(struct drm_device *, struct dcb_entry *, int crtc,
+			 int nr, u32 bw, bool enhframe);
+	void (*train_set)(struct drm_device *, struct dcb_entry *, u8 pattern);
+	void (*train_adj)(struct drm_device *, struct dcb_entry *,
+			  u8 lane, u8 swing, u8 preem);
+};
+
 struct nouveau_encoder {
 	struct drm_encoder_slave base;
 
@@ -78,9 +86,19 @@ get_slave_funcs(struct drm_encoder *enc)
 	return to_encoder_slave(enc)->slave_funcs;
 }
 
+/* nouveau_dp.c */
+int nouveau_dp_auxch(struct nouveau_i2c_chan *auxch, int cmd, int addr,
+		     uint8_t *data, int data_nr);
+bool nouveau_dp_detect(struct drm_encoder *);
+void nouveau_dp_dpms(struct drm_encoder *, int mode, u32 datarate,
+		     struct dp_train_func *);
+u8 *nouveau_dp_bios_data(struct drm_device *, struct dcb_entry *, u8 **);
+
 struct nouveau_connector *
 nouveau_encoder_connector_get(struct nouveau_encoder *encoder);
 int nv50_sor_create(struct drm_connector *, struct dcb_entry *);
+void nv50_sor_dp_calc_tu(struct drm_device *, int, int, u32, u32);
 int nv50_dac_create(struct drm_connector *, struct dcb_entry *);
 
+
 #endif /* __NOUVEAU_ENCODER_H__ */
diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c b/drivers/gpu/drm/nouveau/nouveau_mem.c
index c3a5745..b08065f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_mem.c
@@ -26,7 +26,8 @@
  * DEALINGS IN THE SOFTWARE.
  *
  * Authors:
- *    Keith Whitwell <keith@tungstengraphics.com>
+ *    Ben Skeggs <bskeggs@redhat.com>
+ *    Roy Spliet <r.spliet@student.tudelft.nl>
  */
 
 
@@ -192,75 +193,6 @@ nouveau_mem_gart_fini(struct drm_device *dev)
 	}
 }
 
-static uint32_t
-nouveau_mem_detect_nv04(struct drm_device *dev)
-{
-	uint32_t boot0 = nv_rd32(dev, NV04_PFB_BOOT_0);
-
-	if (boot0 & 0x00000100)
-		return (((boot0 >> 12) & 0xf) * 2 + 2) * 1024 * 1024;
-
-	switch (boot0 & NV04_PFB_BOOT_0_RAM_AMOUNT) {
-	case NV04_PFB_BOOT_0_RAM_AMOUNT_32MB:
-		return 32 * 1024 * 1024;
-	case NV04_PFB_BOOT_0_RAM_AMOUNT_16MB:
-		return 16 * 1024 * 1024;
-	case NV04_PFB_BOOT_0_RAM_AMOUNT_8MB:
-		return 8 * 1024 * 1024;
-	case NV04_PFB_BOOT_0_RAM_AMOUNT_4MB:
-		return 4 * 1024 * 1024;
-	}
-
-	return 0;
-}
-
-static uint32_t
-nouveau_mem_detect_nforce(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct pci_dev *bridge;
-	uint32_t mem;
-
-	bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 1));
-	if (!bridge) {
-		NV_ERROR(dev, "no bridge device\n");
-		return 0;
-	}
-
-	if (dev_priv->flags & NV_NFORCE) {
-		pci_read_config_dword(bridge, 0x7C, &mem);
-		return (uint64_t)(((mem >> 6) & 31) + 1)*1024*1024;
-	} else
-	if (dev_priv->flags & NV_NFORCE2) {
-		pci_read_config_dword(bridge, 0x84, &mem);
-		return (uint64_t)(((mem >> 4) & 127) + 1)*1024*1024;
-	}
-
-	NV_ERROR(dev, "impossible!\n");
-	return 0;
-}
-
-int
-nouveau_mem_detect(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	if (dev_priv->card_type == NV_04) {
-		dev_priv->vram_size = nouveau_mem_detect_nv04(dev);
-	} else
-	if (dev_priv->flags & (NV_NFORCE | NV_NFORCE2)) {
-		dev_priv->vram_size = nouveau_mem_detect_nforce(dev);
-	} else
-	if (dev_priv->card_type < NV_50) {
-		dev_priv->vram_size  = nv_rd32(dev, NV04_PFB_FIFO_DATA);
-		dev_priv->vram_size &= NV10_PFB_FIFO_DATA_RAM_AMOUNT_MB_MASK;
-	}
-
-	if (dev_priv->vram_size)
-		return 0;
-	return -ENOMEM;
-}
-
 bool
 nouveau_mem_flags_valid(struct drm_device *dev, u32 tile_flags)
 {
@@ -385,11 +317,29 @@ nouveau_mem_init_agp(struct drm_device *dev)
 	return 0;
 }
 
+static const struct vram_types {
+	int value;
+	const char *name;
+} vram_type_map[] = {
+	{ NV_MEM_TYPE_STOLEN , "stolen system memory" },
+	{ NV_MEM_TYPE_SGRAM  , "SGRAM" },
+	{ NV_MEM_TYPE_SDRAM  , "SDRAM" },
+	{ NV_MEM_TYPE_DDR1   , "DDR1" },
+	{ NV_MEM_TYPE_DDR2   , "DDR2" },
+	{ NV_MEM_TYPE_DDR3   , "DDR3" },
+	{ NV_MEM_TYPE_GDDR2  , "GDDR2" },
+	{ NV_MEM_TYPE_GDDR3  , "GDDR3" },
+	{ NV_MEM_TYPE_GDDR4  , "GDDR4" },
+	{ NV_MEM_TYPE_GDDR5  , "GDDR5" },
+	{ NV_MEM_TYPE_UNKNOWN, "unknown type" }
+};
+
 int
 nouveau_mem_vram_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct ttm_bo_device *bdev = &dev_priv->ttm.bdev;
+	const struct vram_types *vram_type;
 	int ret, dma_bits;
 
 	dma_bits = 32;
@@ -427,7 +377,21 @@ nouveau_mem_vram_init(struct drm_device *dev)
 		return ret;
 	}
 
-	NV_INFO(dev, "Detected %dMiB VRAM\n", (int)(dev_priv->vram_size >> 20));
+	vram_type = vram_type_map;
+	while (vram_type->value != NV_MEM_TYPE_UNKNOWN) {
+		if (nouveau_vram_type) {
+			if (!strcasecmp(nouveau_vram_type, vram_type->name))
+				break;
+			dev_priv->vram_type = vram_type->value;
+		} else {
+			if (vram_type->value == dev_priv->vram_type)
+				break;
+		}
+		vram_type++;
+	}
+
+	NV_INFO(dev, "Detected %dMiB VRAM (%s)\n",
+		(int)(dev_priv->vram_size >> 20), vram_type->name);
 	if (dev_priv->vram_sys_base) {
 		NV_INFO(dev, "Stolen system memory at: 0x%010llx\n",
 			dev_priv->vram_sys_base);
@@ -508,216 +472,617 @@ nouveau_mem_gart_init(struct drm_device *dev)
 	return 0;
 }
 
-/* XXX: For now a dummy. More samples required, possibly even a card
- * Called from nouveau_perf.c */
-void nv30_mem_timing_entry(struct drm_device *dev, struct nouveau_pm_tbl_header *hdr,
-							struct nouveau_pm_tbl_entry *e, uint8_t magic_number,
-							struct nouveau_pm_memtiming *timing) {
-
-	NV_DEBUG(dev,"Timing entry format unknown, please contact nouveau developers");
-}
-
-void nv40_mem_timing_entry(struct drm_device *dev, struct nouveau_pm_tbl_header *hdr,
-							struct nouveau_pm_tbl_entry *e, uint8_t magic_number,
-							struct nouveau_pm_memtiming *timing) {
-
-	timing->reg_0 = (e->tRC << 24 | e->tRFC << 16 | e->tRAS << 8 | e->tRP);
+static int
+nv40_mem_timing_calc(struct drm_device *dev, u32 freq,
+		     struct nouveau_pm_tbl_entry *e, u8 len,
+		     struct nouveau_pm_memtiming *boot,
+		     struct nouveau_pm_memtiming *t)
+{
+	t->reg[0] = (e->tRP << 24 | e->tRAS << 16 | e->tRFC << 8 | e->tRC);
 
 	/* XXX: I don't trust the -1's and +1's... they must come
 	 *      from somewhere! */
-	timing->reg_1 = (e->tWR + 2 + magic_number) << 24 |
-				  1 << 16 |
-				  (e->tUNK_1 + 2 + magic_number) << 8 |
-				  (e->tCL + 2 - magic_number);
-	timing->reg_2 = (magic_number << 24 | e->tUNK_12 << 16 | e->tUNK_11 << 8 | e->tUNK_10);
-	timing->reg_2 |= 0x20200000;
-
-	NV_DEBUG(dev, "Entry %d: 220: %08x %08x %08x\n", timing->id,
-		 timing->reg_0, timing->reg_1,timing->reg_2);
+	t->reg[1] = (e->tWR + 2 + (t->tCWL - 1)) << 24 |
+		    1 << 16 |
+		    (e->tWTR + 2 + (t->tCWL - 1)) << 8 |
+		    (e->tCL + 2 - (t->tCWL - 1));
+
+	t->reg[2] = 0x20200000 |
+		    ((t->tCWL - 1) << 24 |
+		     e->tRRD << 16 |
+		     e->tRCDWR << 8 |
+		     e->tRCDRD);
+
+	NV_DEBUG(dev, "Entry %d: 220: %08x %08x %08x\n", t->id,
+		 t->reg[0], t->reg[1], t->reg[2]);
+	return 0;
 }
 
-void nv50_mem_timing_entry(struct drm_device *dev, struct bit_entry *P, struct nouveau_pm_tbl_header *hdr,
-							struct nouveau_pm_tbl_entry *e, uint8_t magic_number,struct nouveau_pm_memtiming *timing) {
+static int
+nv50_mem_timing_calc(struct drm_device *dev, u32 freq,
+		     struct nouveau_pm_tbl_entry *e, u8 len,
+		     struct nouveau_pm_memtiming *boot,
+		     struct nouveau_pm_memtiming *t)
+{
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct bit_entry P;
+	uint8_t unk18 = 1, unk20 = 0, unk21 = 0, tmp7_3;
 
-	uint8_t unk18 = 1,
-		unk19 = 1,
-		unk20 = 0,
-		unk21 = 0;
+	if (bit_table(dev, 'P', &P))
+		return -EINVAL;
 
-	switch (min(hdr->entry_len, (u8) 22)) {
+	switch (min(len, (u8) 22)) {
 	case 22:
 		unk21 = e->tUNK_21;
 	case 21:
 		unk20 = e->tUNK_20;
 	case 20:
-		unk19 = e->tUNK_19;
+		if (e->tCWL > 0)
+			t->tCWL = e->tCWL;
 	case 19:
 		unk18 = e->tUNK_18;
 		break;
 	}
 
-	timing->reg_0 = (e->tRC << 24 | e->tRFC << 16 | e->tRAS << 8 | e->tRP);
+	t->reg[0] = (e->tRP << 24 | e->tRAS << 16 | e->tRFC << 8 | e->tRC);
 
-	/* XXX: I don't trust the -1's and +1's... they must come
-	 *      from somewhere! */
-	timing->reg_1 = (e->tWR + unk19 + 1 + magic_number) << 24 |
-				  max(unk18, (u8) 1) << 16 |
-				  (e->tUNK_1 + unk19 + 1 + magic_number) << 8;
-	if (dev_priv->chipset == 0xa8) {
-		timing->reg_1 |= (e->tCL - 1);
-	} else {
-		timing->reg_1 |= (e->tCL + 2 - magic_number);
-	}
-	timing->reg_2 = (e->tUNK_12 << 16 | e->tUNK_11 << 8 | e->tUNK_10);
-
-	timing->reg_5 = (e->tRAS << 24 | e->tRC);
-	timing->reg_5 += max(e->tUNK_10, e->tUNK_11) << 16;
-
-	if (P->version == 1) {
-		timing->reg_2 |= magic_number << 24;
-		timing->reg_3 = (0x14 + e->tCL) << 24 |
-						0x16 << 16 |
-						(e->tCL - 1) << 8 |
-						(e->tCL - 1);
-		timing->reg_4 = (nv_rd32(dev,0x10022c) & 0xffff0000) | e->tUNK_13 << 8  | e->tUNK_13;
-		timing->reg_5 |= (e->tCL + 2) << 8;
-		timing->reg_7 = 0x4000202 | (e->tCL - 1) << 16;
+	t->reg[1] = (e->tWR + 2 + (t->tCWL - 1)) << 24 |
+				max(unk18, (u8) 1) << 16 |
+				(e->tWTR + 2 + (t->tCWL - 1)) << 8;
+
+	t->reg[2] = ((t->tCWL - 1) << 24 |
+		    e->tRRD << 16 |
+		    e->tRCDWR << 8 |
+		    e->tRCDRD);
+
+	t->reg[4] = e->tUNK_13 << 8  | e->tUNK_13;
+
+	t->reg[5] = (e->tRFC << 24 | max(e->tRCDRD, e->tRCDWR) << 16 | e->tRP);
+
+	t->reg[8] = boot->reg[8] & 0xffffff00;
+
+	if (P.version == 1) {
+		t->reg[1] |= (e->tCL + 2 - (t->tCWL - 1));
+
+		t->reg[3] = (0x14 + e->tCL) << 24 |
+			    0x16 << 16 |
+			    (e->tCL - 1) << 8 |
+			    (e->tCL - 1);
+
+		t->reg[4] |= boot->reg[4] & 0xffff0000;
+
+		t->reg[6] = (0x33 - t->tCWL) << 16 |
+			    t->tCWL << 8 |
+			    (0x2e + e->tCL - t->tCWL);
+
+		t->reg[7] = 0x4000202 | (e->tCL - 1) << 16;
+
+		/* XXX: P.version == 1 only has DDR2 and GDDR3? */
+		if (dev_priv->vram_type == NV_MEM_TYPE_DDR2) {
+			t->reg[5] |= (e->tCL + 3) << 8;
+			t->reg[6] |= (t->tCWL - 2) << 8;
+			t->reg[8] |= (e->tCL - 4);
+		} else {
+			t->reg[5] |= (e->tCL + 2) << 8;
+			t->reg[6] |= t->tCWL << 8;
+			t->reg[8] |= (e->tCL - 2);
+		}
 	} else {
-		timing->reg_2 |= (unk19 - 1) << 24;
-		/* XXX: reg_10022c for recentish cards pretty much unknown*/
-		timing->reg_3 = e->tCL - 1;
-		timing->reg_4 = (unk20 << 24 | unk21 << 16 |
-							e->tUNK_13 << 8  | e->tUNK_13);
+		t->reg[1] |= (5 + e->tCL - (t->tCWL));
+
+		/* XXX: 0xb? 0x30? */
+		t->reg[3] = (0x30 + e->tCL) << 24 |
+			    (boot->reg[3] & 0x00ff0000)|
+			    (0xb + e->tCL) << 8 |
+			    (e->tCL - 1);
+
+		t->reg[4] |= (unk20 << 24 | unk21 << 16);
+
 		/* XXX: +6? */
-		timing->reg_5 |= (unk19 + 6) << 8;
+		t->reg[5] |= (t->tCWL + 6) << 8;
 
-		/* XXX: reg_10023c currently unknown
-		 * 10023c seen as 06xxxxxx, 0bxxxxxx or 0fxxxxxx */
-		timing->reg_7 = 0x202;
+		t->reg[6] = (0x5a + e->tCL) << 16 |
+			    (6 - e->tCL + t->tCWL) << 8 |
+			    (0x50 + e->tCL - t->tCWL);
+
+		tmp7_3 = (boot->reg[7] & 0xff000000) >> 24;
+		t->reg[7] = (tmp7_3 << 24) |
+			    ((tmp7_3 - 6 + e->tCL) << 16) |
+			    0x202;
 	}
 
-	NV_DEBUG(dev, "Entry %d: 220: %08x %08x %08x %08x\n", timing->id,
-		 timing->reg_0, timing->reg_1,
-		 timing->reg_2, timing->reg_3);
+	NV_DEBUG(dev, "Entry %d: 220: %08x %08x %08x %08x\n", t->id,
+		 t->reg[0], t->reg[1], t->reg[2], t->reg[3]);
 	NV_DEBUG(dev, "         230: %08x %08x %08x %08x\n",
-		 timing->reg_4, timing->reg_5,
-		 timing->reg_6, timing->reg_7);
-	NV_DEBUG(dev, "         240: %08x\n", timing->reg_8);
-}
-
-void nvc0_mem_timing_entry(struct drm_device *dev, struct nouveau_pm_tbl_header *hdr,
-							struct nouveau_pm_tbl_entry *e, struct nouveau_pm_memtiming *timing) {
-	timing->reg_0 = (e->tRC << 24 | (e->tRFC & 0x7f) << 17 | e->tRAS << 8 | e->tRP);
-	timing->reg_1 = (nv_rd32(dev,0x10f294) & 0xff000000) | (e->tUNK_11&0x0f) << 20 | (e->tUNK_19 << 7) | (e->tCL & 0x0f);
-	timing->reg_2 = (nv_rd32(dev,0x10f298) & 0xff0000ff) | e->tWR << 16 | e->tUNK_1 << 8;
-	timing->reg_3 = e->tUNK_20 << 9 | e->tUNK_13;
-	timing->reg_4 = (nv_rd32(dev,0x10f2a0) & 0xfff000ff) | e->tUNK_12 << 15;
-	NV_DEBUG(dev, "Entry %d: 290: %08x %08x %08x %08x\n", timing->id,
-		 timing->reg_0, timing->reg_1,
-		 timing->reg_2, timing->reg_3);
-	NV_DEBUG(dev, "         2a0: %08x %08x %08x %08x\n",
-		 timing->reg_4, timing->reg_5,
-		 timing->reg_6, timing->reg_7);
+		 t->reg[4], t->reg[5], t->reg[6], t->reg[7]);
+	NV_DEBUG(dev, "         240: %08x\n", t->reg[8]);
+	return 0;
+}
+
+static int
+nvc0_mem_timing_calc(struct drm_device *dev, u32 freq,
+		     struct nouveau_pm_tbl_entry *e, u8 len,
+		     struct nouveau_pm_memtiming *boot,
+		     struct nouveau_pm_memtiming *t)
+{
+	if (e->tCWL > 0)
+		t->tCWL = e->tCWL;
+
+	t->reg[0] = (e->tRP << 24 | (e->tRAS & 0x7f) << 17 |
+		     e->tRFC << 8 | e->tRC);
+
+	t->reg[1] = (boot->reg[1] & 0xff000000) |
+		    (e->tRCDWR & 0x0f) << 20 |
+		    (e->tRCDRD & 0x0f) << 14 |
+		    (t->tCWL << 7) |
+		    (e->tCL & 0x0f);
+
+	t->reg[2] = (boot->reg[2] & 0xff0000ff) |
+		    e->tWR << 16 | e->tWTR << 8;
+
+	t->reg[3] = (e->tUNK_20 & 0x1f) << 9 |
+		    (e->tUNK_21 & 0xf) << 5 |
+		    (e->tUNK_13 & 0x1f);
+
+	t->reg[4] = (boot->reg[4] & 0xfff00fff) |
+		    (e->tRRD&0x1f) << 15;
+
+	NV_DEBUG(dev, "Entry %d: 290: %08x %08x %08x %08x\n", t->id,
+		 t->reg[0], t->reg[1], t->reg[2], t->reg[3]);
+	NV_DEBUG(dev, "         2a0: %08x\n", t->reg[4]);
+	return 0;
 }
 
 /**
- * Processes the Memory Timing BIOS table, stores generated
- * register values
- * @pre init scripts were run, memtiming regs are initialized
+ * MR generation methods
  */
-void
-nouveau_mem_timing_init(struct drm_device *dev)
+
+static int
+nouveau_mem_ddr2_mr(struct drm_device *dev, u32 freq,
+		    struct nouveau_pm_tbl_entry *e, u8 len,
+		    struct nouveau_pm_memtiming *boot,
+		    struct nouveau_pm_memtiming *t)
+{
+	t->drive_strength = 0;
+	if (len < 15) {
+		t->odt = boot->odt;
+	} else {
+		t->odt = e->RAM_FT1 & 0x07;
+	}
+
+	if (e->tCL >= NV_MEM_CL_DDR2_MAX) {
+		NV_WARN(dev, "(%u) Invalid tCL: %u", t->id, e->tCL);
+		return -ERANGE;
+	}
+
+	if (e->tWR >= NV_MEM_WR_DDR2_MAX) {
+		NV_WARN(dev, "(%u) Invalid tWR: %u", t->id, e->tWR);
+		return -ERANGE;
+	}
+
+	if (t->odt > 3) {
+		NV_WARN(dev, "(%u) Invalid odt value, assuming disabled: %x",
+			t->id, t->odt);
+		t->odt = 0;
+	}
+
+	t->mr[0] = (boot->mr[0] & 0x100f) |
+		   (e->tCL) << 4 |
+		   (e->tWR - 1) << 9;
+	t->mr[1] = (boot->mr[1] & 0x101fbb) |
+		   (t->odt & 0x1) << 2 |
+		   (t->odt & 0x2) << 5;
+
+	NV_DEBUG(dev, "(%u) MR: %08x", t->id, t->mr[0]);
+	return 0;
+}
+
+uint8_t nv_mem_wr_lut_ddr3[NV_MEM_WR_DDR3_MAX] = {
+	0, 0, 0, 0, 0, 1, 2, 3, 4, 5, 5, 6, 6, 7, 7, 0, 0};
+
+static int
+nouveau_mem_ddr3_mr(struct drm_device *dev, u32 freq,
+		    struct nouveau_pm_tbl_entry *e, u8 len,
+		    struct nouveau_pm_memtiming *boot,
+		    struct nouveau_pm_memtiming *t)
+{
+	u8 cl = e->tCL - 4;
+
+	t->drive_strength = 0;
+	if (len < 15) {
+		t->odt = boot->odt;
+	} else {
+		t->odt = e->RAM_FT1 & 0x07;
+	}
+
+	if (e->tCL >= NV_MEM_CL_DDR3_MAX || e->tCL < 4) {
+		NV_WARN(dev, "(%u) Invalid tCL: %u", t->id, e->tCL);
+		return -ERANGE;
+	}
+
+	if (e->tWR >= NV_MEM_WR_DDR3_MAX || e->tWR < 4) {
+		NV_WARN(dev, "(%u) Invalid tWR: %u", t->id, e->tWR);
+		return -ERANGE;
+	}
+
+	if (e->tCWL < 5) {
+		NV_WARN(dev, "(%u) Invalid tCWL: %u", t->id, e->tCWL);
+		return -ERANGE;
+	}
+
+	t->mr[0] = (boot->mr[0] & 0x180b) |
+		   /* CAS */
+		   (cl & 0x7) << 4 |
+		   (cl & 0x8) >> 1 |
+		   (nv_mem_wr_lut_ddr3[e->tWR]) << 9;
+	t->mr[1] = (boot->mr[1] & 0x101dbb) |
+		   (t->odt & 0x1) << 2 |
+		   (t->odt & 0x2) << 5 |
+		   (t->odt & 0x4) << 7;
+	t->mr[2] = (boot->mr[2] & 0x20ffb7) | (e->tCWL - 5) << 3;
+
+	NV_DEBUG(dev, "(%u) MR: %08x %08x", t->id, t->mr[0], t->mr[2]);
+	return 0;
+}
+
+uint8_t nv_mem_cl_lut_gddr3[NV_MEM_CL_GDDR3_MAX] = {
+	0, 0, 0, 0, 4, 5, 6, 7, 0, 1, 2, 3, 8, 9, 10, 11};
+uint8_t nv_mem_wr_lut_gddr3[NV_MEM_WR_GDDR3_MAX] = {
+	0, 0, 0, 0, 0, 2, 3, 8, 9, 10, 11, 0, 0, 1, 1, 0, 3};
+
+static int
+nouveau_mem_gddr3_mr(struct drm_device *dev, u32 freq,
+		     struct nouveau_pm_tbl_entry *e, u8 len,
+		     struct nouveau_pm_memtiming *boot,
+		     struct nouveau_pm_memtiming *t)
+{
+	if (len < 15) {
+		t->drive_strength = boot->drive_strength;
+		t->odt = boot->odt;
+	} else {
+		t->drive_strength = (e->RAM_FT1 & 0x30) >> 4;
+		t->odt = e->RAM_FT1 & 0x07;
+	}
+
+	if (e->tCL >= NV_MEM_CL_GDDR3_MAX) {
+		NV_WARN(dev, "(%u) Invalid tCL: %u", t->id, e->tCL);
+		return -ERANGE;
+	}
+
+	if (e->tWR >= NV_MEM_WR_GDDR3_MAX) {
+		NV_WARN(dev, "(%u) Invalid tWR: %u", t->id, e->tWR);
+		return -ERANGE;
+	}
+
+	if (t->odt > 3) {
+		NV_WARN(dev, "(%u) Invalid odt value, assuming autocal: %x",
+			t->id, t->odt);
+		t->odt = 0;
+	}
+
+	t->mr[0] = (boot->mr[0] & 0xe0b) |
+		   /* CAS */
+		   ((nv_mem_cl_lut_gddr3[e->tCL] & 0x7) << 4) |
+		   ((nv_mem_cl_lut_gddr3[e->tCL] & 0x8) >> 2);
+	t->mr[1] = (boot->mr[1] & 0x100f40) | t->drive_strength |
+		   (t->odt << 2) |
+		   (nv_mem_wr_lut_gddr3[e->tWR] & 0xf) << 4;
+	t->mr[2] = boot->mr[2];
+
+	NV_DEBUG(dev, "(%u) MR: %08x %08x %08x", t->id,
+		      t->mr[0], t->mr[1], t->mr[2]);
+	return 0;
+}
+
+static int
+nouveau_mem_gddr5_mr(struct drm_device *dev, u32 freq,
+		     struct nouveau_pm_tbl_entry *e, u8 len,
+		     struct nouveau_pm_memtiming *boot,
+		     struct nouveau_pm_memtiming *t)
+{
+	if (len < 15) {
+		t->drive_strength = boot->drive_strength;
+		t->odt = boot->odt;
+	} else {
+		t->drive_strength = (e->RAM_FT1 & 0x30) >> 4;
+		t->odt = e->RAM_FT1 & 0x03;
+	}
+
+	if (e->tCL >= NV_MEM_CL_GDDR5_MAX) {
+		NV_WARN(dev, "(%u) Invalid tCL: %u", t->id, e->tCL);
+		return -ERANGE;
+	}
+
+	if (e->tWR >= NV_MEM_WR_GDDR5_MAX) {
+		NV_WARN(dev, "(%u) Invalid tWR: %u", t->id, e->tWR);
+		return -ERANGE;
+	}
+
+	if (t->odt > 3) {
+		NV_WARN(dev, "(%u) Invalid odt value, assuming autocal: %x",
+			t->id, t->odt);
+		t->odt = 0;
+	}
+
+	t->mr[0] = (boot->mr[0] & 0x007) |
+		   ((e->tCL - 5) << 3) |
+		   ((e->tWR - 4) << 8);
+	t->mr[1] = (boot->mr[1] & 0x1007f0) |
+		   t->drive_strength |
+		   (t->odt << 2);
+
+	NV_DEBUG(dev, "(%u) MR: %08x %08x", t->id, t->mr[0], t->mr[1]);
+	return 0;
+}
+
+int
+nouveau_mem_timing_calc(struct drm_device *dev, u32 freq,
+			struct nouveau_pm_memtiming *t)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_pm_engine *pm = &dev_priv->engine.pm;
-	struct nouveau_pm_memtimings *memtimings = &pm->memtimings;
-	struct nvbios *bios = &dev_priv->vbios;
-	struct bit_entry P;
-	struct nouveau_pm_tbl_header *hdr = NULL;
-	uint8_t magic_number;
-	u8 *entry;
-	int i;
+	struct nouveau_pm_memtiming *boot = &pm->boot.timing;
+	struct nouveau_pm_tbl_entry *e;
+	u8 ver, len, *ptr, *ramcfg;
+	int ret;
+
+	ptr = nouveau_perf_timing(dev, freq, &ver, &len);
+	if (!ptr || ptr[0] == 0x00) {
+		*t = *boot;
+		return 0;
+	}
+	e = (struct nouveau_pm_tbl_entry *)ptr;
+
+	t->tCWL = boot->tCWL;
+
+	switch (dev_priv->card_type) {
+	case NV_40:
+		ret = nv40_mem_timing_calc(dev, freq, e, len, boot, t);
+		break;
+	case NV_50:
+		ret = nv50_mem_timing_calc(dev, freq, e, len, boot, t);
+		break;
+	case NV_C0:
+		ret = nvc0_mem_timing_calc(dev, freq, e, len, boot, t);
+		break;
+	default:
+		ret = -ENODEV;
+		break;
+	}
 
-	if (bios->type == NVBIOS_BIT) {
-		if (bit_table(dev, 'P', &P))
-			return;
+	switch (dev_priv->vram_type * !ret) {
+	case NV_MEM_TYPE_GDDR3:
+		ret = nouveau_mem_gddr3_mr(dev, freq, e, len, boot, t);
+		break;
+	case NV_MEM_TYPE_GDDR5:
+		ret = nouveau_mem_gddr5_mr(dev, freq, e, len, boot, t);
+		break;
+	case NV_MEM_TYPE_DDR2:
+		ret = nouveau_mem_ddr2_mr(dev, freq, e, len, boot, t);
+		break;
+	case NV_MEM_TYPE_DDR3:
+		ret = nouveau_mem_ddr3_mr(dev, freq, e, len, boot, t);
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+
+	ramcfg = nouveau_perf_ramcfg(dev, freq, &ver, &len);
+	if (ramcfg) {
+		int dll_off;
 
-		if (P.version == 1)
-			hdr = (struct nouveau_pm_tbl_header *) ROMPTR(dev, P.data[4]);
+		if (ver == 0x00)
+			dll_off = !!(ramcfg[3] & 0x04);
 		else
-		if (P.version == 2)
-			hdr = (struct nouveau_pm_tbl_header *) ROMPTR(dev, P.data[8]);
-		else {
-			NV_WARN(dev, "unknown mem for BIT P %d\n", P.version);
+			dll_off = !!(ramcfg[2] & 0x40);
+
+		switch (dev_priv->vram_type) {
+		case NV_MEM_TYPE_GDDR3:
+			t->mr[1] &= ~0x00000040;
+			t->mr[1] |=  0x00000040 * dll_off;
+			break;
+		default:
+			t->mr[1] &= ~0x00000001;
+			t->mr[1] |=  0x00000001 * dll_off;
+			break;
 		}
+	}
+
+	return ret;
+}
+
+void
+nouveau_mem_timing_read(struct drm_device *dev, struct nouveau_pm_memtiming *t)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	u32 timing_base, timing_regs, mr_base;
+	int i;
+
+	if (dev_priv->card_type >= 0xC0) {
+		timing_base = 0x10f290;
+		mr_base = 0x10f300;
 	} else {
-		NV_DEBUG(dev, "BMP version too old for memory\n");
-		return;
+		timing_base = 0x100220;
+		mr_base = 0x1002c0;
 	}
 
-	if (!hdr) {
-		NV_DEBUG(dev, "memory timing table pointer invalid\n");
+	t->id = -1;
+
+	switch (dev_priv->card_type) {
+	case NV_50:
+		timing_regs = 9;
+		break;
+	case NV_C0:
+	case NV_D0:
+		timing_regs = 5;
+		break;
+	case NV_30:
+	case NV_40:
+		timing_regs = 3;
+		break;
+	default:
+		timing_regs = 0;
 		return;
 	}
+	for(i = 0; i < timing_regs; i++)
+		t->reg[i] = nv_rd32(dev, timing_base + (0x04 * i));
+
+	t->tCWL = 0;
+	if (dev_priv->card_type < NV_C0) {
+		t->tCWL = ((nv_rd32(dev, 0x100228) & 0x0f000000) >> 24) + 1;
+	} else if (dev_priv->card_type <= NV_D0) {
+		t->tCWL = ((nv_rd32(dev, 0x10f294) & 0x00000f80) >> 7);
+	}
 
-	if (hdr->version != 0x10) {
-		NV_WARN(dev, "memory timing table 0x%02x unknown\n", hdr->version);
-		return;
+	t->mr[0] = nv_rd32(dev, mr_base);
+	t->mr[1] = nv_rd32(dev, mr_base + 0x04);
+	t->mr[2] = nv_rd32(dev, mr_base + 0x20);
+	t->mr[3] = nv_rd32(dev, mr_base + 0x24);
+
+	t->odt = 0;
+	t->drive_strength = 0;
+
+	switch (dev_priv->vram_type) {
+	case NV_MEM_TYPE_DDR3:
+		t->odt |= (t->mr[1] & 0x200) >> 7;
+	case NV_MEM_TYPE_DDR2:
+		t->odt |= (t->mr[1] & 0x04) >> 2 |
+			  (t->mr[1] & 0x40) >> 5;
+		break;
+	case NV_MEM_TYPE_GDDR3:
+	case NV_MEM_TYPE_GDDR5:
+		t->drive_strength = t->mr[1] & 0x03;
+		t->odt = (t->mr[1] & 0x0c) >> 2;
+		break;
+	default:
+		break;
 	}
+}
 
-	/* validate record length */
-	if (hdr->entry_len < 15) {
-		NV_ERROR(dev, "mem timing table length unknown: %d\n", hdr->entry_len);
-		return;
+int
+nouveau_mem_exec(struct nouveau_mem_exec_func *exec,
+		 struct nouveau_pm_level *perflvl)
+{
+	struct drm_nouveau_private *dev_priv = exec->dev->dev_private;
+	struct nouveau_pm_memtiming *info = &perflvl->timing;
+	u32 tMRD = 1000, tCKSRE = 0, tCKSRX = 0, tXS = 0, tDLLK = 0;
+	u32 mr[3] = { info->mr[0], info->mr[1], info->mr[2] };
+	u32 mr1_dlloff;
+
+	switch (dev_priv->vram_type) {
+	case NV_MEM_TYPE_DDR2:
+		tDLLK = 2000;
+		mr1_dlloff = 0x00000001;
+		break;
+	case NV_MEM_TYPE_DDR3:
+		tDLLK = 12000;
+		mr1_dlloff = 0x00000001;
+		break;
+	case NV_MEM_TYPE_GDDR3:
+		tDLLK = 40000;
+		mr1_dlloff = 0x00000040;
+		break;
+	default:
+		NV_ERROR(exec->dev, "cannot reclock unsupported memtype\n");
+		return -ENODEV;
 	}
 
-	/* parse vbios entries into common format */
-	memtimings->timing =
-		kcalloc(hdr->entry_cnt, sizeof(*memtimings->timing), GFP_KERNEL);
-	if (!memtimings->timing)
-		return;
+	/* fetch current MRs */
+	switch (dev_priv->vram_type) {
+	case NV_MEM_TYPE_GDDR3:
+	case NV_MEM_TYPE_DDR3:
+		mr[2] = exec->mrg(exec, 2);
+	default:
+		mr[1] = exec->mrg(exec, 1);
+		mr[0] = exec->mrg(exec, 0);
+		break;
+	}
 
-	/* Get "some number" from the timing reg for NV_40 and NV_50
-	 * Used in calculations later... source unknown */
-	magic_number = 0;
-	if (P.version == 1) {
-		magic_number = (nv_rd32(dev, 0x100228) & 0x0f000000) >> 24;
+	/* DLL 'on' -> DLL 'off' mode, disable before entering self-refresh  */
+	if (!(mr[1] & mr1_dlloff) && (info->mr[1] & mr1_dlloff)) {
+		exec->precharge(exec);
+		exec->mrs (exec, 1, mr[1] | mr1_dlloff);
+		exec->wait(exec, tMRD);
 	}
 
-	entry = (u8*) hdr + hdr->header_len;
-	for (i = 0; i < hdr->entry_cnt; i++, entry += hdr->entry_len) {
-		struct nouveau_pm_memtiming *timing = &pm->memtimings.timing[i];
-		if (entry[0] == 0)
-			continue;
+	/* enter self-refresh mode */
+	exec->precharge(exec);
+	exec->refresh(exec);
+	exec->refresh(exec);
+	exec->refresh_auto(exec, false);
+	exec->refresh_self(exec, true);
+	exec->wait(exec, tCKSRE);
+
+	/* modify input clock frequency */
+	exec->clock_set(exec);
+
+	/* exit self-refresh mode */
+	exec->wait(exec, tCKSRX);
+	exec->precharge(exec);
+	exec->refresh_self(exec, false);
+	exec->refresh_auto(exec, true);
+	exec->wait(exec, tXS);
+
+	/* update MRs */
+	if (mr[2] != info->mr[2]) {
+		exec->mrs (exec, 2, info->mr[2]);
+		exec->wait(exec, tMRD);
+	}
+
+	if (mr[1] != info->mr[1]) {
+		/* need to keep DLL off until later, at least on GDDR3 */
+		exec->mrs (exec, 1, info->mr[1] | (mr[1] & mr1_dlloff));
+		exec->wait(exec, tMRD);
+	}
+
+	if (mr[0] != info->mr[0]) {
+		exec->mrs (exec, 0, info->mr[0]);
+		exec->wait(exec, tMRD);
+	}
 
-		timing->id = i;
-		timing->WR = entry[0];
-		timing->CL = entry[2];
+	/* update PFB timing registers */
+	exec->timing_set(exec);
 
-		if(dev_priv->card_type <= NV_40) {
-			nv40_mem_timing_entry(dev,hdr,(struct nouveau_pm_tbl_entry*) entry,magic_number,&pm->memtimings.timing[i]);
-		} else if(dev_priv->card_type == NV_50){
-			nv50_mem_timing_entry(dev,&P,hdr,(struct nouveau_pm_tbl_entry*) entry,magic_number,&pm->memtimings.timing[i]);
-		} else if(dev_priv->card_type == NV_C0) {
-			nvc0_mem_timing_entry(dev,hdr,(struct nouveau_pm_tbl_entry*) entry,&pm->memtimings.timing[i]);
+	/* DLL (enable + ) reset */
+	if (!(info->mr[1] & mr1_dlloff)) {
+		if (mr[1] & mr1_dlloff) {
+			exec->mrs (exec, 1, info->mr[1]);
+			exec->wait(exec, tMRD);
 		}
+		exec->mrs (exec, 0, info->mr[0] | 0x00000100);
+		exec->wait(exec, tMRD);
+		exec->mrs (exec, 0, info->mr[0] | 0x00000000);
+		exec->wait(exec, tMRD);
+		exec->wait(exec, tDLLK);
+		if (dev_priv->vram_type == NV_MEM_TYPE_GDDR3)
+			exec->precharge(exec);
 	}
 
-	memtimings->nr_timing = hdr->entry_cnt;
-	memtimings->supported = P.version == 1;
+	return 0;
 }
 
-void
-nouveau_mem_timing_fini(struct drm_device *dev)
+int
+nouveau_mem_vbios_type(struct drm_device *dev)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pm_memtimings *mem = &dev_priv->engine.pm.memtimings;
+	struct bit_entry M;
+	u8 ramcfg = (nv_rd32(dev, 0x101000) & 0x0000003c) >> 2;
+	if (!bit_table(dev, 'M', &M) || M.version != 2 || M.length < 5) {
+		u8 *table = ROMPTR(dev, M.data[3]);
+		if (table && table[0] == 0x10 && ramcfg < table[3]) {
+			u8 *entry = table + table[1] + (ramcfg * table[2]);
+			switch (entry[0] & 0x0f) {
+			case 0: return NV_MEM_TYPE_DDR2;
+			case 1: return NV_MEM_TYPE_DDR3;
+			case 2: return NV_MEM_TYPE_GDDR3;
+			case 3: return NV_MEM_TYPE_GDDR5;
+			default:
+				break;
+			}
 
-	if(mem->timing) {
-		kfree(mem->timing);
-		mem->timing = NULL;
+		}
 	}
+	return NV_MEM_TYPE_UNKNOWN;
 }
 
 static int
diff --git a/drivers/gpu/drm/nouveau/nouveau_mxm.c b/drivers/gpu/drm/nouveau/nouveau_mxm.c
index e5a64f0..07d0d1e 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mxm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_mxm.c
@@ -582,6 +582,35 @@ mxm_shadow_dsm(struct drm_device *dev, u8 version)
 
 #define WMI_WMMX_GUID "F6CB5C3C-9CAE-4EBD-B577-931EA32A2CC0"
 
+static u8
+wmi_wmmx_mxmi(struct drm_device *dev, u8 version)
+{
+	u32 mxmi_args[] = { 0x494D584D /* MXMI */, version, 0 };
+	struct acpi_buffer args = { sizeof(mxmi_args), mxmi_args };
+	struct acpi_buffer retn = { ACPI_ALLOCATE_BUFFER, NULL };
+	union acpi_object *obj;
+	acpi_status status;
+
+	status = wmi_evaluate_method(WMI_WMMX_GUID, 0, 0, &args, &retn);
+	if (ACPI_FAILURE(status)) {
+		MXM_DBG(dev, "WMMX MXMI returned %d\n", status);
+		return 0x00;
+	}
+
+	obj = retn.pointer;
+	if (obj->type == ACPI_TYPE_INTEGER) {
+		version = obj->integer.value;
+		MXM_DBG(dev, "WMMX MXMI version %d.%d\n",
+			     (version >> 4), version & 0x0f);
+	} else {
+		version = 0;
+		MXM_DBG(dev, "WMMX MXMI returned non-integer\n");
+	}
+
+	kfree(obj);
+	return version;
+}
+
 static bool
 mxm_shadow_wmi(struct drm_device *dev, u8 version)
 {
@@ -592,7 +621,15 @@ mxm_shadow_wmi(struct drm_device *dev, u8 version)
 	union acpi_object *obj;
 	acpi_status status;
 
-	if (!wmi_has_guid(WMI_WMMX_GUID))
+	if (!wmi_has_guid(WMI_WMMX_GUID)) {
+		MXM_DBG(dev, "WMMX GUID not found\n");
+		return false;
+	}
+
+	mxms_args[1] = wmi_wmmx_mxmi(dev, 0x00);
+	if (!mxms_args[1])
+		mxms_args[1] = wmi_wmmx_mxmi(dev, version);
+	if (!mxms_args[1])
 		return false;
 
 	status = wmi_evaluate_method(WMI_WMMX_GUID, 0, 0, &args, &retn);
diff --git a/drivers/gpu/drm/nouveau/nouveau_perf.c b/drivers/gpu/drm/nouveau/nouveau_perf.c
index 58f4973..69a528d 100644
--- a/drivers/gpu/drm/nouveau/nouveau_perf.c
+++ b/drivers/gpu/drm/nouveau/nouveau_perf.c
@@ -27,6 +27,178 @@
 #include "nouveau_drv.h"
 #include "nouveau_pm.h"
 
+static u8 *
+nouveau_perf_table(struct drm_device *dev, u8 *ver)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nvbios *bios = &dev_priv->vbios;
+	struct bit_entry P;
+
+	if (!bit_table(dev, 'P', &P) && P.version && P.version <= 2) {
+		u8 *perf = ROMPTR(dev, P.data[0]);
+		if (perf) {
+			*ver = perf[0];
+			return perf;
+		}
+	}
+
+	if (bios->type == NVBIOS_BMP) {
+		if (bios->data[bios->offset + 6] >= 0x25) {
+			u8 *perf = ROMPTR(dev, bios->data[bios->offset + 0x94]);
+			if (perf) {
+				*ver = perf[1];
+				return perf;
+			}
+		}
+	}
+
+	return NULL;
+}
+
+static u8 *
+nouveau_perf_entry(struct drm_device *dev, int idx,
+		   u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
+{
+	u8 *perf = nouveau_perf_table(dev, ver);
+	if (perf) {
+		if (*ver >= 0x12 && *ver < 0x20 && idx < perf[2]) {
+			*hdr = perf[3];
+			*cnt = 0;
+			*len = 0;
+			return perf + perf[0] + idx * perf[3];
+		} else
+		if (*ver >= 0x20 && *ver < 0x40 && idx < perf[2]) {
+			*hdr = perf[3];
+			*cnt = perf[4];
+			*len = perf[5];
+			return perf + perf[1] + idx * (*hdr + (*cnt * *len));
+		} else
+		if (*ver >= 0x40 && *ver < 0x41 && idx < perf[5]) {
+			*hdr = perf[2];
+			*cnt = perf[4];
+			*len = perf[3];
+			return perf + perf[1] + idx * (*hdr + (*cnt * *len));
+		}
+	}
+	return NULL;
+}
+
+static u8 *
+nouveau_perf_rammap(struct drm_device *dev, u32 freq,
+		    u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct bit_entry P;
+	u8 *perf, i = 0;
+
+	if (!bit_table(dev, 'P', &P) && P.version == 2) {
+		u8 *rammap = ROMPTR(dev, P.data[4]);
+		if (rammap) {
+			u8 *ramcfg = rammap + rammap[1];
+
+			*ver = rammap[0];
+			*hdr = rammap[2];
+			*cnt = rammap[4];
+			*len = rammap[3];
+
+			freq /= 1000;
+			for (i = 0; i < rammap[5]; i++) {
+				if (freq >= ROM16(ramcfg[0]) &&
+				    freq <= ROM16(ramcfg[2]))
+					return ramcfg;
+
+				ramcfg += *hdr + (*cnt * *len);
+			}
+		}
+
+		return NULL;
+	}
+
+	if (dev_priv->chipset == 0x49 ||
+	    dev_priv->chipset == 0x4b)
+		freq /= 2;
+
+	while ((perf = nouveau_perf_entry(dev, i++, ver, hdr, cnt, len))) {
+		if (*ver >= 0x20 && *ver < 0x25) {
+			if (perf[0] != 0xff && freq <= ROM16(perf[11]) * 1000)
+				break;
+		} else
+		if (*ver >= 0x25 && *ver < 0x40) {
+			if (perf[0] != 0xff && freq <= ROM16(perf[12]) * 1000)
+				break;
+		}
+	}
+
+	if (perf) {
+		u8 *ramcfg = perf + *hdr;
+		*ver = 0x00;
+		*hdr = 0;
+		return ramcfg;
+	}
+
+	return NULL;
+}
+
+u8 *
+nouveau_perf_ramcfg(struct drm_device *dev, u32 freq, u8 *ver, u8 *len)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nvbios *bios = &dev_priv->vbios;
+	u8 strap, hdr, cnt;
+	u8 *rammap;
+
+	strap = (nv_rd32(dev, 0x101000) & 0x0000003c) >> 2;
+	if (bios->ram_restrict_tbl_ptr)
+		strap = bios->data[bios->ram_restrict_tbl_ptr + strap];
+
+	rammap = nouveau_perf_rammap(dev, freq, ver, &hdr, &cnt, len);
+	if (rammap && strap < cnt)
+		return rammap + hdr + (strap * *len);
+
+	return NULL;
+}
+
+u8 *
+nouveau_perf_timing(struct drm_device *dev, u32 freq, u8 *ver, u8 *len)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nvbios *bios = &dev_priv->vbios;
+	struct bit_entry P;
+	u8 *perf, *timing = NULL;
+	u8 i = 0, hdr, cnt;
+
+	if (bios->type == NVBIOS_BMP) {
+		while ((perf = nouveau_perf_entry(dev, i++, ver, &hdr, &cnt,
+						  len)) && *ver == 0x15) {
+			if (freq <= ROM32(perf[5]) * 20) {
+				*ver = 0x00;
+				*len = 14;
+				return perf + 41;
+			}
+		}
+		return NULL;
+	}
+
+	if (!bit_table(dev, 'P', &P)) {
+		if (P.version == 1)
+			timing = ROMPTR(dev, P.data[4]);
+		else
+		if (P.version == 2)
+			timing = ROMPTR(dev, P.data[8]);
+	}
+
+	if (timing && timing[0] == 0x10) {
+		u8 *ramcfg = nouveau_perf_ramcfg(dev, freq, ver, len);
+		if (ramcfg && ramcfg[1] < timing[2]) {
+			*ver = timing[0];
+			*len = timing[3];
+			return timing + timing[1] + (ramcfg[1] * timing[3]);
+		}
+	}
+
+	return NULL;
+}
+
 static void
 legacy_perf_init(struct drm_device *dev)
 {
@@ -72,74 +244,11 @@ 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(dev, 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) {
-		if (entry[1] != 0xff)
-			NV_WARN(dev, "timingset %d does not exist\n", entry[1]);
-		return NULL;
-	}
-
-	return &pm->memtimings.timing[entry[1]];
-}
-
 static void
-nouveau_perf_voltage(struct drm_device *dev, struct bit_entry *P,
-		     struct nouveau_pm_level *perflvl)
+nouveau_perf_voltage(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct bit_entry P;
 	u8 *vmap;
 	int id;
 
@@ -158,13 +267,13 @@ nouveau_perf_voltage(struct drm_device *dev, struct bit_entry *P,
 	/* on newer ones, the perflvl stores an index into yet another
 	 * vbios table containing a min/max voltage value for the perflvl
 	 */
-	if (P->version != 2 || P->length < 34) {
+	if (bit_table(dev, 'P', &P) || P.version != 2 || P.length < 34) {
 		NV_DEBUG(dev, "where's our volt map table ptr? %d %d\n",
-			 P->version, P->length);
+			 P.version, P.length);
 		return;
 	}
 
-	vmap = ROMPTR(dev, P->data[32]);
+	vmap = ROMPTR(dev, P.data[32]);
 	if (!vmap) {
 		NV_DEBUG(dev, "volt map table pointer invalid\n");
 		return;
@@ -183,130 +292,70 @@ nouveau_perf_init(struct drm_device *dev)
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_pm_engine *pm = &dev_priv->engine.pm;
 	struct nvbios *bios = &dev_priv->vbios;
-	struct bit_entry P;
-	struct nouveau_pm_memtimings *memtimings = &pm->memtimings;
-	struct nouveau_pm_tbl_header mt_hdr;
-	u8 version, headerlen, recordlen, entries;
-	u8 *perf, *entry;
-	int vid, i;
-
-	if (bios->type == NVBIOS_BIT) {
-		if (bit_table(dev, 'P', &P))
-			return;
-
-		if (P.version != 1 && P.version != 2) {
-			NV_WARN(dev, "unknown perf for BIT P %d\n", P.version);
-			return;
-		}
-
-		perf = ROMPTR(dev, P.data[0]);
-		version   = perf[0];
-		headerlen = perf[1];
-		if (version < 0x40) {
-			recordlen = perf[3] + (perf[4] * perf[5]);
-			entries   = perf[2];
-
-			pm->pwm_divisor = ROM16(perf[6]);
-		} else {
-			recordlen = perf[2] + (perf[3] * perf[4]);
-			entries   = perf[5];
-		}
-	} else {
-		if (bios->data[bios->offset + 6] < 0x25) {
-			legacy_perf_init(dev);
-			return;
-		}
+	u8 *perf, ver, hdr, cnt, len;
+	int ret, vid, i = -1;
 
-		perf = ROMPTR(dev, bios->data[bios->offset + 0x94]);
-		if (!perf) {
-			NV_DEBUG(dev, "perf table pointer invalid\n");
-			return;
-		}
-
-		version   = perf[1];
-		headerlen = perf[0];
-		recordlen = perf[3];
-		entries   = perf[2];
-	}
-
-	if (entries > NOUVEAU_PM_MAX_LEVEL) {
-		NV_DEBUG(dev, "perf table has too many entries - buggy vbios?\n");
-		entries = NOUVEAU_PM_MAX_LEVEL;
+	if (bios->type == NVBIOS_BMP && bios->data[bios->offset + 6] < 0x25) {
+		legacy_perf_init(dev);
+		return;
 	}
 
-	entry = perf + headerlen;
-
-	/* For version 0x15, initialize memtiming table */
-	if(version == 0x15) {
-		memtimings->timing =
-				kcalloc(entries, sizeof(*memtimings->timing), GFP_KERNEL);
-		if (!memtimings->timing) {
-			NV_WARN(dev,"Could not allocate memtiming table\n");
-			return;
-		}
-
-		mt_hdr.entry_cnt = entries;
-		mt_hdr.entry_len = 14;
-		mt_hdr.version = version;
-		mt_hdr.header_len = 4;
-	}
+	perf = nouveau_perf_table(dev, &ver);
+	if (ver >= 0x20 && ver < 0x40)
+		pm->fan.pwm_divisor = ROM16(perf[6]);
 
-	for (i = 0; i < entries; i++) {
+	while ((perf = nouveau_perf_entry(dev, ++i, &ver, &hdr, &cnt, &len))) {
 		struct nouveau_pm_level *perflvl = &pm->perflvl[pm->nr_perflvl];
 
-		perflvl->timing = NULL;
-
-		if (entry[0] == 0xff) {
-			entry += recordlen;
+		if (perf[0] == 0xff)
 			continue;
-		}
 
-		switch (version) {
+		switch (ver) {
 		case 0x12:
 		case 0x13:
 		case 0x15:
-			perflvl->fanspeed = entry[55];
-			if (recordlen > 56)
-				perflvl->volt_min = entry[56];
-			perflvl->core = ROM32(entry[1]) * 10;
-			perflvl->memory = ROM32(entry[5]) * 20;
+			perflvl->fanspeed = perf[55];
+			if (hdr > 56)
+				perflvl->volt_min = perf[56];
+			perflvl->core = ROM32(perf[1]) * 10;
+			perflvl->memory = ROM32(perf[5]) * 20;
 			break;
 		case 0x21:
 		case 0x23:
 		case 0x24:
-			perflvl->fanspeed = entry[4];
-			perflvl->volt_min = entry[5];
-			perflvl->shader = ROM16(entry[6]) * 1000;
+			perflvl->fanspeed = perf[4];
+			perflvl->volt_min = perf[5];
+			perflvl->shader = ROM16(perf[6]) * 1000;
 			perflvl->core = perflvl->shader;
-			perflvl->core += (signed char)entry[8] * 1000;
+			perflvl->core += (signed char)perf[8] * 1000;
 			if (dev_priv->chipset == 0x49 ||
 			    dev_priv->chipset == 0x4b)
-				perflvl->memory = ROM16(entry[11]) * 1000;
+				perflvl->memory = ROM16(perf[11]) * 1000;
 			else
-				perflvl->memory = ROM16(entry[11]) * 2000;
+				perflvl->memory = ROM16(perf[11]) * 2000;
 			break;
 		case 0x25:
-			perflvl->fanspeed = entry[4];
-			perflvl->volt_min = entry[5];
-			perflvl->core = ROM16(entry[6]) * 1000;
-			perflvl->shader = ROM16(entry[10]) * 1000;
-			perflvl->memory = ROM16(entry[12]) * 1000;
+			perflvl->fanspeed = perf[4];
+			perflvl->volt_min = perf[5];
+			perflvl->core = ROM16(perf[6]) * 1000;
+			perflvl->shader = ROM16(perf[10]) * 1000;
+			perflvl->memory = ROM16(perf[12]) * 1000;
 			break;
 		case 0x30:
-			perflvl->memscript = ROM16(entry[2]);
+			perflvl->memscript = ROM16(perf[2]);
 		case 0x35:
-			perflvl->fanspeed = entry[6];
-			perflvl->volt_min = entry[7];
-			perflvl->core = ROM16(entry[8]) * 1000;
-			perflvl->shader = ROM16(entry[10]) * 1000;
-			perflvl->memory = ROM16(entry[12]) * 1000;
-			perflvl->vdec = ROM16(entry[16]) * 1000;
-			perflvl->dom6 = ROM16(entry[20]) * 1000;
+			perflvl->fanspeed = perf[6];
+			perflvl->volt_min = perf[7];
+			perflvl->core = ROM16(perf[8]) * 1000;
+			perflvl->shader = ROM16(perf[10]) * 1000;
+			perflvl->memory = ROM16(perf[12]) * 1000;
+			perflvl->vdec = ROM16(perf[16]) * 1000;
+			perflvl->dom6 = ROM16(perf[20]) * 1000;
 			break;
 		case 0x40:
-#define subent(n) (ROM16(entry[perf[2] + ((n) * perf[3])]) & 0xfff) * 1000
+#define subent(n) ((ROM16(perf[hdr + (n) * len]) & 0xfff) * 1000)
 			perflvl->fanspeed = 0; /*XXX*/
-			perflvl->volt_min = entry[2];
+			perflvl->volt_min = perf[2];
 			if (dev_priv->card_type == NV_50) {
 				perflvl->core   = subent(0);
 				perflvl->shader = subent(1);
@@ -329,36 +378,34 @@ nouveau_perf_init(struct drm_device *dev)
 		}
 
 		/* make sure vid is valid */
-		nouveau_perf_voltage(dev, &P, perflvl);
+		nouveau_perf_voltage(dev, perflvl);
 		if (pm->voltage.supported && perflvl->volt_min) {
 			vid = nouveau_volt_vid_lookup(dev, perflvl->volt_min);
 			if (vid < 0) {
-				NV_DEBUG(dev, "drop perflvl %d, bad vid\n", i);
-				entry += recordlen;
+				NV_DEBUG(dev, "perflvl %d, bad vid\n", i);
 				continue;
 			}
 		}
 
 		/* get the corresponding memory timings */
-		if (version == 0x15) {
-			memtimings->timing[i].id = i;
-			nv30_mem_timing_entry(dev,&mt_hdr,(struct nouveau_pm_tbl_entry*) &entry[41],0,&memtimings->timing[i]);
-			perflvl->timing = &memtimings->timing[i];
-		} else 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]);
+		ret = nouveau_mem_timing_calc(dev, perflvl->memory,
+					          &perflvl->timing);
+		if (ret) {
+			NV_DEBUG(dev, "perflvl %d, bad timing: %d\n", i, ret);
+			continue;
 		}
 
 		snprintf(perflvl->name, sizeof(perflvl->name),
 			 "performance_level_%d", i);
 		perflvl->id = i;
-		pm->nr_perflvl++;
 
-		entry += recordlen;
+		snprintf(perflvl->profile.name, sizeof(perflvl->profile.name),
+			 "%d", perflvl->id);
+		perflvl->profile.func = &nouveau_pm_static_profile_func;
+		list_add_tail(&perflvl->profile.head, &pm->profiles);
+
+
+		pm->nr_perflvl++;
 	}
 }
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_pm.c b/drivers/gpu/drm/nouveau/nouveau_pm.c
index 9064d7f..34d591b 100644
--- a/drivers/gpu/drm/nouveau/nouveau_pm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_pm.c
@@ -50,7 +50,7 @@ nouveau_pwmfan_get(struct drm_device *dev)
 	ret = nouveau_gpio_find(dev, 0, DCB_GPIO_PWM_FAN, 0xff, &gpio);
 	if (ret == 0) {
 		ret = pm->pwm_get(dev, gpio.line, &divs, &duty);
-		if (ret == 0) {
+		if (ret == 0 && divs) {
 			divs = max(divs, duty);
 			if (dev_priv->card_type <= NV_40 || (gpio.log[0] & 1))
 				duty = divs - duty;
@@ -77,7 +77,7 @@ nouveau_pwmfan_set(struct drm_device *dev, int percent)
 
 	ret = nouveau_gpio_find(dev, 0, DCB_GPIO_PWM_FAN, 0xff, &gpio);
 	if (ret == 0) {
-		divs = pm->pwm_divisor;
+		divs = pm->fan.pwm_divisor;
 		if (pm->fan.pwm_freq) {
 			/*XXX: PNVIO clock more than likely... */
 			divs = 135000 / pm->fan.pwm_freq;
@@ -89,7 +89,10 @@ nouveau_pwmfan_set(struct drm_device *dev, int percent)
 		if (dev_priv->card_type <= NV_40 || (gpio.log[0] & 1))
 			duty = divs - duty;
 
-		return pm->pwm_set(dev, gpio.line, divs, duty);
+		ret = pm->pwm_set(dev, gpio.line, divs, duty);
+		if (!ret)
+			pm->fan.percent = percent;
+		return ret;
 	}
 
 	return -ENODEV;
@@ -144,9 +147,13 @@ nouveau_pm_perflvl_set(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 		return ret;
 
 	state = pm->clocks_pre(dev, perflvl);
-	if (IS_ERR(state))
-		return PTR_ERR(state);
-	pm->clocks_set(dev, state);
+	if (IS_ERR(state)) {
+		ret = PTR_ERR(state);
+		goto error;
+	}
+	ret = pm->clocks_set(dev, state);
+	if (ret)
+		goto error;
 
 	ret = nouveau_pm_perflvl_aux(dev, perflvl, perflvl, pm->cur);
 	if (ret)
@@ -154,6 +161,65 @@ nouveau_pm_perflvl_set(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 
 	pm->cur = perflvl;
 	return 0;
+
+error:
+	/* restore the fan speed and voltage before leaving */
+	nouveau_pm_perflvl_aux(dev, perflvl, perflvl, pm->cur);
+	return ret;
+}
+
+void
+nouveau_pm_trigger(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_pm_engine *pm = &dev_priv->engine.pm;
+	struct nouveau_pm_profile *profile = NULL;
+	struct nouveau_pm_level *perflvl = NULL;
+	int ret;
+
+	/* select power profile based on current power source */
+	if (power_supply_is_system_supplied())
+		profile = pm->profile_ac;
+	else
+		profile = pm->profile_dc;
+
+	if (profile != pm->profile) {
+		pm->profile->func->fini(pm->profile);
+		pm->profile = profile;
+		pm->profile->func->init(pm->profile);
+	}
+
+	/* select performance level based on profile */
+	perflvl = profile->func->select(profile);
+
+	/* change perflvl, if necessary */
+	if (perflvl != pm->cur) {
+		struct nouveau_timer_engine *ptimer = &dev_priv->engine.timer;
+		u64 time0 = ptimer->read(dev);
+
+		NV_INFO(dev, "setting performance level: %d", perflvl->id);
+		ret = nouveau_pm_perflvl_set(dev, perflvl);
+		if (ret)
+			NV_INFO(dev, "> reclocking failed: %d\n\n", ret);
+
+		NV_INFO(dev, "> reclocking took %lluns\n\n",
+			     ptimer->read(dev) - time0);
+	}
+}
+
+static struct nouveau_pm_profile *
+profile_find(struct drm_device *dev, const char *string)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_pm_engine *pm = &dev_priv->engine.pm;
+	struct nouveau_pm_profile *profile;
+
+	list_for_each_entry(profile, &pm->profiles, head) {
+		if (!strncmp(profile->name, string, sizeof(profile->name)))
+			return profile;
+	}
+
+	return NULL;
 }
 
 static int
@@ -161,33 +227,54 @@ nouveau_pm_profile_set(struct drm_device *dev, const char *profile)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_pm_engine *pm = &dev_priv->engine.pm;
-	struct nouveau_pm_level *perflvl = NULL;
+	struct nouveau_pm_profile *ac = NULL, *dc = NULL;
+	char string[16], *cur = string, *ptr;
 
 	/* safety precaution, for now */
 	if (nouveau_perflvl_wr != 7777)
 		return -EPERM;
 
-	if (!strncmp(profile, "boot", 4))
-		perflvl = &pm->boot;
-	else {
-		int pl = simple_strtol(profile, NULL, 10);
-		int i;
+	strncpy(string, profile, sizeof(string));
+	if ((ptr = strchr(string, '\n')))
+		*ptr = '\0';
 
-		for (i = 0; i < pm->nr_perflvl; i++) {
-			if (pm->perflvl[i].id == pl) {
-				perflvl = &pm->perflvl[i];
-				break;
-			}
-		}
+	ptr = strsep(&cur, ",");
+	if (ptr)
+		ac = profile_find(dev, ptr);
 
-		if (!perflvl)
-			return -EINVAL;
-	}
+	ptr = strsep(&cur, ",");
+	if (ptr)
+		dc = profile_find(dev, ptr);
+	else
+		dc = ac;
+
+	if (ac == NULL || dc == NULL)
+		return -EINVAL;
+
+	pm->profile_ac = ac;
+	pm->profile_dc = dc;
+	nouveau_pm_trigger(dev);
+	return 0;
+}
+
+static void
+nouveau_pm_static_dummy(struct nouveau_pm_profile *profile)
+{
+}
 
-	NV_INFO(dev, "setting performance level: %s\n", profile);
-	return nouveau_pm_perflvl_set(dev, perflvl);
+static struct nouveau_pm_level *
+nouveau_pm_static_select(struct nouveau_pm_profile *profile)
+{
+	return container_of(profile, struct nouveau_pm_level, profile);
 }
 
+const struct nouveau_pm_profile_func nouveau_pm_static_profile_func = {
+	.destroy = nouveau_pm_static_dummy,
+	.init = nouveau_pm_static_dummy,
+	.fini = nouveau_pm_static_dummy,
+	.select = nouveau_pm_static_select,
+};
+
 static int
 nouveau_pm_perflvl_get(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 {
@@ -197,9 +284,11 @@ nouveau_pm_perflvl_get(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 
 	memset(perflvl, 0, sizeof(*perflvl));
 
-	ret = pm->clocks_get(dev, perflvl);
-	if (ret)
-		return ret;
+	if (pm->clocks_get) {
+		ret = pm->clocks_get(dev, perflvl);
+		if (ret)
+			return ret;
+	}
 
 	if (pm->voltage.supported && pm->voltage_get) {
 		ret = pm->voltage_get(dev);
@@ -213,13 +302,14 @@ nouveau_pm_perflvl_get(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 	if (ret > 0)
 		perflvl->fanspeed = ret;
 
+	nouveau_mem_timing_read(dev, &perflvl->timing);
 	return 0;
 }
 
 static void
 nouveau_pm_perflvl_info(struct nouveau_pm_level *perflvl, char *ptr, int len)
 {
-	char c[16], s[16], v[32], f[16], t[16], m[16];
+	char c[16], s[16], v[32], f[16], m[16];
 
 	c[0] = '\0';
 	if (perflvl->core)
@@ -247,18 +337,15 @@ nouveau_pm_perflvl_info(struct nouveau_pm_level *perflvl, char *ptr, int len)
 	if (perflvl->fanspeed)
 		snprintf(f, sizeof(f), " fanspeed %d%%", perflvl->fanspeed);
 
-	t[0] = '\0';
-	if (perflvl->timing)
-		snprintf(t, sizeof(t), " timing %d", perflvl->timing->id);
-
-	snprintf(ptr, len, "%s%s%s%s%s%s\n", c, s, m, t, v, f);
+	snprintf(ptr, len, "%s%s%s%s%s\n", c, s, m, v, f);
 }
 
 static ssize_t
 nouveau_pm_get_perflvl_info(struct device *d,
 			    struct device_attribute *a, char *buf)
 {
-	struct nouveau_pm_level *perflvl = (struct nouveau_pm_level *)a;
+	struct nouveau_pm_level *perflvl =
+		container_of(a, struct nouveau_pm_level, dev_attr);
 	char *ptr = buf;
 	int len = PAGE_SIZE;
 
@@ -280,12 +367,8 @@ nouveau_pm_get_perflvl(struct device *d, struct device_attribute *a, char *buf)
 	int len = PAGE_SIZE, ret;
 	char *ptr = buf;
 
-	if (!pm->cur)
-		snprintf(ptr, len, "setting: boot\n");
-	else if (pm->cur == &pm->boot)
-		snprintf(ptr, len, "setting: boot\nc:");
-	else
-		snprintf(ptr, len, "setting: static %d\nc:", pm->cur->id);
+	snprintf(ptr, len, "profile: %s, %s\nc:",
+		 pm->profile_ac->name, pm->profile_dc->name);
 	ptr += strlen(buf);
 	len -= strlen(buf);
 
@@ -397,7 +480,7 @@ nouveau_hwmon_set_max_temp(struct device *d, struct device_attribute *a,
 	struct nouveau_pm_threshold_temp *temp = &pm->threshold_temp;
 	long value;
 
-	if (strict_strtol(buf, 10, &value) == -EINVAL)
+	if (kstrtol(buf, 10, &value) == -EINVAL)
 		return count;
 
 	temp->down_clock = value/1000;
@@ -432,7 +515,7 @@ nouveau_hwmon_set_critical_temp(struct device *d, struct device_attribute *a,
 	struct nouveau_pm_threshold_temp *temp = &pm->threshold_temp;
 	long value;
 
-	if (strict_strtol(buf, 10, &value) == -EINVAL)
+	if (kstrtol(buf, 10, &value) == -EINVAL)
 		return count;
 
 	temp->critical = value/1000;
@@ -529,7 +612,7 @@ nouveau_hwmon_set_pwm0(struct device *d, struct device_attribute *a,
 	if (nouveau_perflvl_wr != 7777)
 		return -EPERM;
 
-	if (strict_strtol(buf, 10, &value) == -EINVAL)
+	if (kstrtol(buf, 10, &value) == -EINVAL)
 		return -EINVAL;
 
 	if (value < pm->fan.min_duty)
@@ -568,7 +651,7 @@ nouveau_hwmon_set_pwm0_min(struct device *d, struct device_attribute *a,
 	struct nouveau_pm_engine *pm = &dev_priv->engine.pm;
 	long value;
 
-	if (strict_strtol(buf, 10, &value) == -EINVAL)
+	if (kstrtol(buf, 10, &value) == -EINVAL)
 		return -EINVAL;
 
 	if (value < 0)
@@ -609,7 +692,7 @@ nouveau_hwmon_set_pwm0_max(struct device *d, struct device_attribute *a,
 	struct nouveau_pm_engine *pm = &dev_priv->engine.pm;
 	long value;
 
-	if (strict_strtol(buf, 10, &value) == -EINVAL)
+	if (kstrtol(buf, 10, &value) == -EINVAL)
 		return -EINVAL;
 
 	if (value < 0)
@@ -731,8 +814,10 @@ nouveau_hwmon_fini(struct drm_device *dev)
 
 	if (pm->hwmon) {
 		sysfs_remove_group(&dev->pdev->dev.kobj, &hwmon_attrgroup);
-		sysfs_remove_group(&dev->pdev->dev.kobj, &hwmon_pwm_fan_attrgroup);
-		sysfs_remove_group(&dev->pdev->dev.kobj, &hwmon_fan_rpm_attrgroup);
+		sysfs_remove_group(&dev->pdev->dev.kobj,
+				   &hwmon_pwm_fan_attrgroup);
+		sysfs_remove_group(&dev->pdev->dev.kobj,
+				   &hwmon_fan_rpm_attrgroup);
 
 		hwmon_device_unregister(pm->hwmon);
 	}
@@ -752,6 +837,7 @@ nouveau_pm_acpi_event(struct notifier_block *nb, unsigned long val, void *data)
 		bool ac = power_supply_is_system_supplied();
 
 		NV_DEBUG(dev, "power supply changed: %s\n", ac ? "AC" : "DC");
+		nouveau_pm_trigger(dev);
 	}
 
 	return NOTIFY_OK;
@@ -766,35 +852,48 @@ nouveau_pm_init(struct drm_device *dev)
 	char info[256];
 	int ret, i;
 
-	nouveau_mem_timing_init(dev);
+	/* parse aux tables from vbios */
 	nouveau_volt_init(dev);
-	nouveau_perf_init(dev);
 	nouveau_temp_init(dev);
 
+	/* determine current ("boot") performance level */
+	ret = nouveau_pm_perflvl_get(dev, &pm->boot);
+	if (ret) {
+		NV_ERROR(dev, "failed to determine boot perflvl\n");
+		return ret;
+	}
+
+	strncpy(pm->boot.name, "boot", 4);
+	strncpy(pm->boot.profile.name, "boot", 4);
+	pm->boot.profile.func = &nouveau_pm_static_profile_func;
+
+	INIT_LIST_HEAD(&pm->profiles);
+	list_add(&pm->boot.profile.head, &pm->profiles);
+
+	pm->profile_ac = &pm->boot.profile;
+	pm->profile_dc = &pm->boot.profile;
+	pm->profile = &pm->boot.profile;
+	pm->cur = &pm->boot;
+
+	/* add performance levels from vbios */
+	nouveau_perf_init(dev);
+
+	/* display available performance levels */
 	NV_INFO(dev, "%d available performance level(s)\n", pm->nr_perflvl);
 	for (i = 0; i < pm->nr_perflvl; i++) {
 		nouveau_pm_perflvl_info(&pm->perflvl[i], info, sizeof(info));
 		NV_INFO(dev, "%d:%s", pm->perflvl[i].id, info);
 	}
 
-	/* 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));
-		NV_INFO(dev, "c:%s", info);
-	}
+	nouveau_pm_perflvl_info(&pm->boot, info, sizeof(info));
+	NV_INFO(dev, "c:%s", info);
 
 	/* switch performance levels now if requested */
-	if (nouveau_perflvl != NULL) {
-		ret = nouveau_pm_profile_set(dev, nouveau_perflvl);
-		if (ret) {
-			NV_ERROR(dev, "error setting perflvl \"%s\": %d\n",
-				 nouveau_perflvl, ret);
-		}
-	}
+	if (nouveau_perflvl != NULL)
+		nouveau_pm_profile_set(dev, nouveau_perflvl);
+
+	/* determine the current fan speed */
+	pm->fan.percent = nouveau_pwmfan_get(dev);
 
 	nouveau_sysfs_init(dev);
 	nouveau_hwmon_init(dev);
@@ -811,6 +910,12 @@ nouveau_pm_fini(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_pm_engine *pm = &dev_priv->engine.pm;
+	struct nouveau_pm_profile *profile, *tmp;
+
+	list_for_each_entry_safe(profile, tmp, &pm->profiles, head) {
+		list_del(&profile->head);
+		profile->func->destroy(profile);
+	}
 
 	if (pm->cur != &pm->boot)
 		nouveau_pm_perflvl_set(dev, &pm->boot);
@@ -818,7 +923,6 @@ nouveau_pm_fini(struct drm_device *dev)
 	nouveau_temp_fini(dev);
 	nouveau_perf_fini(dev);
 	nouveau_volt_fini(dev);
-	nouveau_mem_timing_fini(dev);
 
 #if defined(CONFIG_ACPI) && defined(CONFIG_POWER_SUPPLY)
 	unregister_acpi_notifier(&pm->acpi_nb);
@@ -840,4 +944,5 @@ nouveau_pm_resume(struct drm_device *dev)
 	perflvl = pm->cur;
 	pm->cur = &pm->boot;
 	nouveau_pm_perflvl_set(dev, perflvl);
+	nouveau_pwmfan_set(dev, pm->fan.percent);
 }
diff --git a/drivers/gpu/drm/nouveau/nouveau_pm.h b/drivers/gpu/drm/nouveau/nouveau_pm.h
index 2f8e14f..3f82dfe 100644
--- a/drivers/gpu/drm/nouveau/nouveau_pm.h
+++ b/drivers/gpu/drm/nouveau/nouveau_pm.h
@@ -25,10 +25,30 @@
 #ifndef __NOUVEAU_PM_H__
 #define __NOUVEAU_PM_H__
 
+struct nouveau_mem_exec_func {
+	struct drm_device *dev;
+	void (*precharge)(struct nouveau_mem_exec_func *);
+	void (*refresh)(struct nouveau_mem_exec_func *);
+	void (*refresh_auto)(struct nouveau_mem_exec_func *, bool);
+	void (*refresh_self)(struct nouveau_mem_exec_func *, bool);
+	void (*wait)(struct nouveau_mem_exec_func *, u32 nsec);
+	u32  (*mrg)(struct nouveau_mem_exec_func *, int mr);
+	void (*mrs)(struct nouveau_mem_exec_func *, int mr, u32 data);
+	void (*clock_set)(struct nouveau_mem_exec_func *);
+	void (*timing_set)(struct nouveau_mem_exec_func *);
+	void *priv;
+};
+
+/* nouveau_mem.c */
+int  nouveau_mem_exec(struct nouveau_mem_exec_func *,
+		      struct nouveau_pm_level *);
+
 /* nouveau_pm.c */
 int  nouveau_pm_init(struct drm_device *dev);
 void nouveau_pm_fini(struct drm_device *dev);
 void nouveau_pm_resume(struct drm_device *dev);
+extern const struct nouveau_pm_profile_func nouveau_pm_static_profile_func;
+void nouveau_pm_trigger(struct drm_device *dev);
 
 /* nouveau_volt.c */
 void nouveau_volt_init(struct drm_device *);
@@ -41,6 +61,8 @@ int  nouveau_voltage_gpio_set(struct drm_device *, int voltage);
 /* nouveau_perf.c */
 void nouveau_perf_init(struct drm_device *);
 void nouveau_perf_fini(struct drm_device *);
+u8 *nouveau_perf_timing(struct drm_device *, u32 freq, u8 *ver, u8 *len);
+u8 *nouveau_perf_ramcfg(struct drm_device *, u32 freq, u8 *ver, u8 *len);
 
 /* nouveau_mem.c */
 void nouveau_mem_timing_init(struct drm_device *);
diff --git a/drivers/gpu/drm/nouveau/nouveau_state.c b/drivers/gpu/drm/nouveau/nouveau_state.c
index 912839c..9c144fb 100644
--- a/drivers/gpu/drm/nouveau/nouveau_state.c
+++ b/drivers/gpu/drm/nouveau/nouveau_state.c
@@ -87,7 +87,7 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clocks_get		= nv04_pm_clocks_get;
 		engine->pm.clocks_pre		= nv04_pm_clocks_pre;
 		engine->pm.clocks_set		= nv04_pm_clocks_set;
-		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.init		= nv04_fb_vram_init;
 		engine->vram.takedown		= nouveau_stub_takedown;
 		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
@@ -134,7 +134,11 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clocks_get		= nv04_pm_clocks_get;
 		engine->pm.clocks_pre		= nv04_pm_clocks_pre;
 		engine->pm.clocks_set		= nv04_pm_clocks_set;
-		engine->vram.init		= nouveau_mem_detect;
+		if (dev_priv->chipset == 0x1a ||
+		    dev_priv->chipset == 0x1f)
+			engine->vram.init	= nv1a_fb_vram_init;
+		else
+			engine->vram.init	= nv10_fb_vram_init;
 		engine->vram.takedown		= nouveau_stub_takedown;
 		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
@@ -153,11 +157,11 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->timer.init		= nv04_timer_init;
 		engine->timer.read		= nv04_timer_read;
 		engine->timer.takedown		= nv04_timer_takedown;
-		engine->fb.init			= nv10_fb_init;
-		engine->fb.takedown		= nv10_fb_takedown;
-		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->fb.init			= nv20_fb_init;
+		engine->fb.takedown		= nv20_fb_takedown;
+		engine->fb.init_tile_region	= nv20_fb_init_tile_region;
+		engine->fb.set_tile_region	= nv20_fb_set_tile_region;
+		engine->fb.free_tile_region	= nv20_fb_free_tile_region;
 		engine->fifo.channels		= 32;
 		engine->fifo.init		= nv10_fifo_init;
 		engine->fifo.takedown		= nv04_fifo_fini;
@@ -181,7 +185,7 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clocks_get		= nv04_pm_clocks_get;
 		engine->pm.clocks_pre		= nv04_pm_clocks_pre;
 		engine->pm.clocks_set		= nv04_pm_clocks_set;
-		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.init		= nv20_fb_vram_init;
 		engine->vram.takedown		= nouveau_stub_takedown;
 		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
@@ -230,7 +234,7 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clocks_set		= nv04_pm_clocks_set;
 		engine->pm.voltage_get		= nouveau_voltage_gpio_get;
 		engine->pm.voltage_set		= nouveau_voltage_gpio_set;
-		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.init		= nv20_fb_vram_init;
 		engine->vram.takedown		= nouveau_stub_takedown;
 		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
@@ -286,7 +290,7 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.temp_get		= nv40_temp_get;
 		engine->pm.pwm_get		= nv40_pm_pwm_get;
 		engine->pm.pwm_set		= nv40_pm_pwm_set;
-		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.init		= nv40_fb_vram_init;
 		engine->vram.takedown		= nouveau_stub_takedown;
 		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
@@ -588,47 +592,45 @@ nouveau_card_init(struct drm_device *dev)
 		nv_mask(dev, 0x00088080, 0x00000800, 0x00000000);
 	}
 
-	nouveau_pm_init(dev);
-
-	ret = engine->vram.init(dev);
+	/* PMC */
+	ret = engine->mc.init(dev);
 	if (ret)
 		goto out_bios;
 
-	ret = nouveau_gpuobj_init(dev);
+	/* PTIMER */
+	ret = engine->timer.init(dev);
 	if (ret)
-		goto out_vram;
+		goto out_mc;
 
-	ret = engine->instmem.init(dev);
+	/* PFB */
+	ret = engine->fb.init(dev);
 	if (ret)
-		goto out_gpuobj;
+		goto out_timer;
 
-	ret = nouveau_mem_vram_init(dev);
+	ret = engine->vram.init(dev);
 	if (ret)
-		goto out_instmem;
+		goto out_fb;
 
-	ret = nouveau_mem_gart_init(dev);
+	/* PGPIO */
+	ret = nouveau_gpio_create(dev);
 	if (ret)
-		goto out_ttmvram;
+		goto out_vram;
 
-	/* PMC */
-	ret = engine->mc.init(dev);
+	ret = nouveau_gpuobj_init(dev);
 	if (ret)
-		goto out_gart;
+		goto out_gpio;
 
-	/* PGPIO */
-	ret = nouveau_gpio_create(dev);
+	ret = engine->instmem.init(dev);
 	if (ret)
-		goto out_mc;
+		goto out_gpuobj;
 
-	/* PTIMER */
-	ret = engine->timer.init(dev);
+	ret = nouveau_mem_vram_init(dev);
 	if (ret)
-		goto out_gpio;
+		goto out_instmem;
 
-	/* PFB */
-	ret = engine->fb.init(dev);
+	ret = nouveau_mem_gart_init(dev);
 	if (ret)
-		goto out_timer;
+		goto out_ttmvram;
 
 	if (!dev_priv->noaccel) {
 		switch (dev_priv->card_type) {
@@ -734,11 +736,12 @@ nouveau_card_init(struct drm_device *dev)
 		goto out_irq;
 
 	nouveau_backlight_init(dev);
+	nouveau_pm_init(dev);
 
 	if (dev_priv->eng[NVOBJ_ENGINE_GR]) {
 		ret = nouveau_fence_init(dev);
 		if (ret)
-			goto out_disp;
+			goto out_pm;
 
 		ret = nouveau_channel_alloc(dev, &dev_priv->channel, NULL,
 					    NvDmaFB, NvDmaTT);
@@ -762,7 +765,8 @@ out_chan:
 	nouveau_channel_put_unlocked(&dev_priv->channel);
 out_fence:
 	nouveau_fence_fini(dev);
-out_disp:
+out_pm:
+	nouveau_pm_fini(dev);
 	nouveau_backlight_exit(dev);
 	nouveau_display_destroy(dev);
 out_irq:
@@ -779,15 +783,6 @@ out_engine:
 			dev_priv->eng[e]->destroy(dev,e );
 		}
 	}
-
-	engine->fb.takedown(dev);
-out_timer:
-	engine->timer.takedown(dev);
-out_gpio:
-	nouveau_gpio_destroy(dev);
-out_mc:
-	engine->mc.takedown(dev);
-out_gart:
 	nouveau_mem_gart_fini(dev);
 out_ttmvram:
 	nouveau_mem_vram_fini(dev);
@@ -795,10 +790,17 @@ out_instmem:
 	engine->instmem.takedown(dev);
 out_gpuobj:
 	nouveau_gpuobj_takedown(dev);
+out_gpio:
+	nouveau_gpio_destroy(dev);
 out_vram:
 	engine->vram.takedown(dev);
+out_fb:
+	engine->fb.takedown(dev);
+out_timer:
+	engine->timer.takedown(dev);
+out_mc:
+	engine->mc.takedown(dev);
 out_bios:
-	nouveau_pm_fini(dev);
 	nouveau_bios_takedown(dev);
 out_display_early:
 	engine->display.late_takedown(dev);
@@ -823,6 +825,7 @@ static void nouveau_card_takedown(struct drm_device *dev)
 		nouveau_fence_fini(dev);
 	}
 
+	nouveau_pm_fini(dev);
 	nouveau_backlight_exit(dev);
 	nouveau_display_destroy(dev);
 
@@ -835,11 +838,6 @@ static void nouveau_card_takedown(struct drm_device *dev)
 			}
 		}
 	}
-	engine->fb.takedown(dev);
-	engine->timer.takedown(dev);
-	nouveau_gpio_destroy(dev);
-	engine->mc.takedown(dev);
-	engine->display.late_takedown(dev);
 
 	if (dev_priv->vga_ram) {
 		nouveau_bo_unpin(dev_priv->vga_ram);
@@ -855,12 +853,17 @@ static void nouveau_card_takedown(struct drm_device *dev)
 
 	engine->instmem.takedown(dev);
 	nouveau_gpuobj_takedown(dev);
-	engine->vram.takedown(dev);
 
-	nouveau_irq_fini(dev);
+	nouveau_gpio_destroy(dev);
+	engine->vram.takedown(dev);
+	engine->fb.takedown(dev);
+	engine->timer.takedown(dev);
+	engine->mc.takedown(dev);
 
-	nouveau_pm_fini(dev);
 	nouveau_bios_takedown(dev);
+	engine->display.late_takedown(dev);
+
+	nouveau_irq_fini(dev);
 
 	vga_client_register(dev->pdev, NULL, NULL, NULL);
 }
@@ -990,7 +993,7 @@ static int nouveau_remove_conflicting_drivers(struct drm_device *dev)
 int nouveau_load(struct drm_device *dev, unsigned long flags)
 {
 	struct drm_nouveau_private *dev_priv;
-	uint32_t reg0, strap;
+	uint32_t reg0 = ~0, strap;
 	resource_size_t mmio_start_offs;
 	int ret;
 
@@ -1009,10 +1012,65 @@ int nouveau_load(struct drm_device *dev, unsigned long flags)
 	NV_DEBUG(dev, "vendor: 0x%X device: 0x%X class: 0x%X\n",
 		 dev->pci_vendor, dev->pci_device, dev->pdev->class);
 
-	/* resource 0 is mmio regs */
-	/* resource 1 is linear FB */
-	/* resource 2 is RAMIN (mmio regs + 0x1000000) */
-	/* resource 6 is bios */
+	/* first up, map the start of mmio and determine the chipset */
+	dev_priv->mmio = ioremap(pci_resource_start(dev->pdev, 0), PAGE_SIZE);
+	if (dev_priv->mmio) {
+#ifdef __BIG_ENDIAN
+		/* put the card into big-endian mode if it's not */
+		if (nv_rd32(dev, NV03_PMC_BOOT_1) != 0x01000001)
+			nv_wr32(dev, NV03_PMC_BOOT_1, 0x01000001);
+		DRM_MEMORYBARRIER();
+#endif
+
+		/* determine chipset and derive architecture from it */
+		reg0 = nv_rd32(dev, NV03_PMC_BOOT_0);
+		if ((reg0 & 0x0f000000) > 0) {
+			dev_priv->chipset = (reg0 & 0xff00000) >> 20;
+			switch (dev_priv->chipset & 0xf0) {
+			case 0x10:
+			case 0x20:
+			case 0x30:
+				dev_priv->card_type = dev_priv->chipset & 0xf0;
+				break;
+			case 0x40:
+			case 0x60:
+				dev_priv->card_type = NV_40;
+				break;
+			case 0x50:
+			case 0x80:
+			case 0x90:
+			case 0xa0:
+				dev_priv->card_type = NV_50;
+				break;
+			case 0xc0:
+				dev_priv->card_type = NV_C0;
+				break;
+			case 0xd0:
+				dev_priv->card_type = NV_D0;
+				break;
+			default:
+				break;
+			}
+		} else
+		if ((reg0 & 0xff00fff0) == 0x20004000) {
+			if (reg0 & 0x00f00000)
+				dev_priv->chipset = 0x05;
+			else
+				dev_priv->chipset = 0x04;
+			dev_priv->card_type = NV_04;
+		}
+
+		iounmap(dev_priv->mmio);
+	}
+
+	if (!dev_priv->card_type) {
+		NV_ERROR(dev, "unsupported chipset 0x%08x\n", reg0);
+		ret = -EINVAL;
+		goto err_priv;
+	}
+
+	NV_INFO(dev, "Detected an NV%2x generation card (0x%08x)\n",
+		     dev_priv->card_type, reg0);
 
 	/* map the mmio regs */
 	mmio_start_offs = pci_resource_start(dev->pdev, 0);
@@ -1026,62 +1084,6 @@ int nouveau_load(struct drm_device *dev, unsigned long flags)
 	NV_DEBUG(dev, "regs mapped ok at 0x%llx\n",
 					(unsigned long long)mmio_start_offs);
 
-#ifdef __BIG_ENDIAN
-	/* Put the card in BE mode if it's not */
-	if (nv_rd32(dev, NV03_PMC_BOOT_1) != 0x01000001)
-		nv_wr32(dev, NV03_PMC_BOOT_1, 0x01000001);
-
-	DRM_MEMORYBARRIER();
-#endif
-
-	/* Time to determine the card architecture */
-	reg0 = nv_rd32(dev, NV03_PMC_BOOT_0);
-
-	/* We're dealing with >=NV10 */
-	if ((reg0 & 0x0f000000) > 0) {
-		/* Bit 27-20 contain the architecture in hex */
-		dev_priv->chipset = (reg0 & 0xff00000) >> 20;
-	/* NV04 or NV05 */
-	} else if ((reg0 & 0xff00fff0) == 0x20004000) {
-		if (reg0 & 0x00f00000)
-			dev_priv->chipset = 0x05;
-		else
-			dev_priv->chipset = 0x04;
-	} else
-		dev_priv->chipset = 0xff;
-
-	switch (dev_priv->chipset & 0xf0) {
-	case 0x00:
-	case 0x10:
-	case 0x20:
-	case 0x30:
-		dev_priv->card_type = dev_priv->chipset & 0xf0;
-		break;
-	case 0x40:
-	case 0x60:
-		dev_priv->card_type = NV_40;
-		break;
-	case 0x50:
-	case 0x80:
-	case 0x90:
-	case 0xa0:
-		dev_priv->card_type = NV_50;
-		break;
-	case 0xc0:
-		dev_priv->card_type = NV_C0;
-		break;
-	case 0xd0:
-		dev_priv->card_type = NV_D0;
-		break;
-	default:
-		NV_INFO(dev, "Unsupported chipset 0x%08x\n", reg0);
-		ret = -EINVAL;
-		goto err_mmio;
-	}
-
-	NV_INFO(dev, "Detected an NV%2x generation card (0x%08x)\n",
-		dev_priv->card_type, reg0);
-
 	/* determine frequency of timing crystal */
 	strap = nv_rd32(dev, 0x101000);
 	if ( dev_priv->chipset < 0x17 ||
diff --git a/drivers/gpu/drm/nouveau/nv04_fb.c b/drivers/gpu/drm/nouveau/nv04_fb.c
index 638cf60..d5eedd6 100644
--- a/drivers/gpu/drm/nouveau/nv04_fb.c
+++ b/drivers/gpu/drm/nouveau/nv04_fb.c
@@ -4,6 +4,40 @@
 #include "nouveau_drm.h"
 
 int
+nv04_fb_vram_init(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	u32 boot0 = nv_rd32(dev, NV04_PFB_BOOT_0);
+
+	if (boot0 & 0x00000100) {
+		dev_priv->vram_size  = ((boot0 >> 12) & 0xf) * 2 + 2;
+		dev_priv->vram_size *= 1024 * 1024;
+	} else {
+		switch (boot0 & NV04_PFB_BOOT_0_RAM_AMOUNT) {
+		case NV04_PFB_BOOT_0_RAM_AMOUNT_32MB:
+			dev_priv->vram_size = 32 * 1024 * 1024;
+			break;
+		case NV04_PFB_BOOT_0_RAM_AMOUNT_16MB:
+			dev_priv->vram_size = 16 * 1024 * 1024;
+			break;
+		case NV04_PFB_BOOT_0_RAM_AMOUNT_8MB:
+			dev_priv->vram_size = 8 * 1024 * 1024;
+			break;
+		case NV04_PFB_BOOT_0_RAM_AMOUNT_4MB:
+			dev_priv->vram_size = 4 * 1024 * 1024;
+			break;
+		}
+	}
+
+	if ((boot0 & 0x00000038) <= 0x10)
+		dev_priv->vram_type = NV_MEM_TYPE_SGRAM;
+	else
+		dev_priv->vram_type = NV_MEM_TYPE_SDRAM;
+
+	return 0;
+}
+
+int
 nv04_fb_init(struct drm_device *dev)
 {
 	/* This is what the DDX did for NV_ARCH_04, but a mmio-trace shows
diff --git a/drivers/gpu/drm/nouveau/nv10_fb.c b/drivers/gpu/drm/nouveau/nv10_fb.c
index f78181a..420b1608 100644
--- a/drivers/gpu/drm/nouveau/nv10_fb.c
+++ b/drivers/gpu/drm/nouveau/nv10_fb.c
@@ -3,81 +3,16 @@
 #include "nouveau_drv.h"
 #include "nouveau_drm.h"
 
-static struct drm_mm_node *
-nv20_fb_alloc_tag(struct drm_device *dev, uint32_t size)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
-	struct drm_mm_node *mem;
-	int ret;
-
-	ret = drm_mm_pre_get(&pfb->tag_heap);
-	if (ret)
-		return NULL;
-
-	spin_lock(&dev_priv->tile.lock);
-	mem = drm_mm_search_free(&pfb->tag_heap, size, 0, 0);
-	if (mem)
-		mem = drm_mm_get_block_atomic(mem, size, 0);
-	spin_unlock(&dev_priv->tile.lock);
-
-	return mem;
-}
-
-static void
-nv20_fb_free_tag(struct drm_device *dev, struct drm_mm_node *mem)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	spin_lock(&dev_priv->tile.lock);
-	drm_mm_put_block(mem);
-	spin_unlock(&dev_priv->tile.lock);
-}
-
 void
 nv10_fb_init_tile_region(struct drm_device *dev, int i, uint32_t addr,
 			 uint32_t size, uint32_t pitch, uint32_t flags)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i];
-	int bpp = (flags & NOUVEAU_GEM_TILE_32BPP ? 32 : 16);
 
-	tile->addr = addr;
+	tile->addr  = 0x80000000 | addr;
 	tile->limit = max(1u, addr + size) - 1;
 	tile->pitch = pitch;
-
-	if (dev_priv->card_type == NV_20) {
-		if (flags & NOUVEAU_GEM_TILE_ZETA) {
-			/*
-			 * Allocate some of the on-die tag memory,
-			 * used to store Z compression meta-data (most
-			 * likely just a bitmap determining if a given
-			 * tile is compressed or not).
-			 */
-			tile->tag_mem = nv20_fb_alloc_tag(dev, size / 256);
-
-			if (tile->tag_mem) {
-				/* Enable Z compression */
-				if (dev_priv->chipset >= 0x25)
-					tile->zcomp = tile->tag_mem->start |
-						(bpp == 16 ?
-						 NV25_PFB_ZCOMP_MODE_16 :
-						 NV25_PFB_ZCOMP_MODE_32);
-				else
-					tile->zcomp = tile->tag_mem->start |
-						NV20_PFB_ZCOMP_EN |
-						(bpp == 16 ? 0 :
-						 NV20_PFB_ZCOMP_MODE_32);
-			}
-
-			tile->addr |= 3;
-		} else {
-			tile->addr |= 1;
-		}
-
-	} else {
-		tile->addr |= 1 << 31;
-	}
 }
 
 void
@@ -86,11 +21,6 @@ nv10_fb_free_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];
 
-	if (tile->tag_mem) {
-		nv20_fb_free_tag(dev, tile->tag_mem);
-		tile->tag_mem = NULL;
-	}
-
 	tile->addr = tile->limit = tile->pitch = tile->zcomp = 0;
 }
 
@@ -103,9 +33,48 @@ nv10_fb_set_tile_region(struct drm_device *dev, int i)
 	nv_wr32(dev, NV10_PFB_TLIMIT(i), tile->limit);
 	nv_wr32(dev, NV10_PFB_TSIZE(i), tile->pitch);
 	nv_wr32(dev, NV10_PFB_TILE(i), tile->addr);
+}
+
+int
+nv1a_fb_vram_init(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct pci_dev *bridge;
+	uint32_t mem, mib;
+
+	bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 1));
+	if (!bridge) {
+		NV_ERROR(dev, "no bridge device\n");
+		return 0;
+	}
+
+	if (dev_priv->chipset == 0x1a) {
+		pci_read_config_dword(bridge, 0x7c, &mem);
+		mib = ((mem >> 6) & 31) + 1;
+	} else {
+		pci_read_config_dword(bridge, 0x84, &mem);
+		mib = ((mem >> 4) & 127) + 1;
+	}
+
+	dev_priv->vram_size = mib * 1024 * 1024;
+	return 0;
+}
+
+int
+nv10_fb_vram_init(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	u32 fifo_data = nv_rd32(dev, NV04_PFB_FIFO_DATA);
+	u32 cfg0 = nv_rd32(dev, 0x100200);
 
-	if (dev_priv->card_type == NV_20)
-		nv_wr32(dev, NV20_PFB_ZCOMP(i), tile->zcomp);
+	dev_priv->vram_size = fifo_data & NV10_PFB_FIFO_DATA_RAM_AMOUNT_MB_MASK;
+
+	if (cfg0 & 0x00000001)
+		dev_priv->vram_type = NV_MEM_TYPE_DDR1;
+	else
+		dev_priv->vram_type = NV_MEM_TYPE_SDRAM;
+
+	return 0;
 }
 
 int
@@ -115,14 +84,8 @@ nv10_fb_init(struct drm_device *dev)
 	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
 	int i;
 
-	pfb->num_tiles = NV10_PFB_TILE__SIZE;
-
-	if (dev_priv->card_type == NV_20)
-		drm_mm_init(&pfb->tag_heap, 0,
-			    (dev_priv->chipset >= 0x25 ?
-			     64 * 1024 : 32 * 1024));
-
 	/* Turn all the tiling regions off. */
+	pfb->num_tiles = NV10_PFB_TILE__SIZE;
 	for (i = 0; i < pfb->num_tiles; i++)
 		pfb->set_tile_region(dev, i);
 
@@ -138,7 +101,4 @@ nv10_fb_takedown(struct drm_device *dev)
 
 	for (i = 0; i < pfb->num_tiles; i++)
 		pfb->free_tile_region(dev, i);
-
-	if (dev_priv->card_type == NV_20)
-		drm_mm_takedown(&pfb->tag_heap);
 }
diff --git a/drivers/gpu/drm/nouveau/nv20_fb.c b/drivers/gpu/drm/nouveau/nv20_fb.c
new file mode 100644
index 0000000..19bd640
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nv20_fb.c
@@ -0,0 +1,148 @@
+#include "drmP.h"
+#include "drm.h"
+#include "nouveau_drv.h"
+#include "nouveau_drm.h"
+
+static struct drm_mm_node *
+nv20_fb_alloc_tag(struct drm_device *dev, uint32_t size)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
+	struct drm_mm_node *mem;
+	int ret;
+
+	ret = drm_mm_pre_get(&pfb->tag_heap);
+	if (ret)
+		return NULL;
+
+	spin_lock(&dev_priv->tile.lock);
+	mem = drm_mm_search_free(&pfb->tag_heap, size, 0, 0);
+	if (mem)
+		mem = drm_mm_get_block_atomic(mem, size, 0);
+	spin_unlock(&dev_priv->tile.lock);
+
+	return mem;
+}
+
+static void
+nv20_fb_free_tag(struct drm_device *dev, struct drm_mm_node **pmem)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct drm_mm_node *mem = *pmem;
+	if (mem) {
+		spin_lock(&dev_priv->tile.lock);
+		drm_mm_put_block(mem);
+		spin_unlock(&dev_priv->tile.lock);
+		*pmem = NULL;
+	}
+}
+
+void
+nv20_fb_init_tile_region(struct drm_device *dev, int i, uint32_t addr,
+			 uint32_t size, uint32_t pitch, uint32_t flags)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i];
+	int bpp = (flags & NOUVEAU_GEM_TILE_32BPP ? 32 : 16);
+
+	tile->addr  = 0x00000001 | addr;
+	tile->limit = max(1u, addr + size) - 1;
+	tile->pitch = pitch;
+
+	/* Allocate some of the on-die tag memory, used to store Z
+	 * compression meta-data (most likely just a bitmap determining
+	 * if a given tile is compressed or not).
+	 */
+	if (flags & NOUVEAU_GEM_TILE_ZETA) {
+		tile->tag_mem = nv20_fb_alloc_tag(dev, size / 256);
+		if (tile->tag_mem) {
+			/* Enable Z compression */
+			tile->zcomp = tile->tag_mem->start;
+			if (dev_priv->chipset >= 0x25) {
+				if (bpp == 16)
+					tile->zcomp |= NV25_PFB_ZCOMP_MODE_16;
+				else
+					tile->zcomp |= NV25_PFB_ZCOMP_MODE_32;
+			} else {
+				tile->zcomp |= NV20_PFB_ZCOMP_EN;
+				if (bpp != 16)
+					tile->zcomp |= NV20_PFB_ZCOMP_MODE_32;
+			}
+		}
+
+		tile->addr |= 2;
+	}
+}
+
+void
+nv20_fb_free_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];
+
+	tile->addr = tile->limit = tile->pitch = tile->zcomp = 0;
+	nv20_fb_free_tag(dev, &tile->tag_mem);
+}
+
+void
+nv20_fb_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, NV10_PFB_TLIMIT(i), tile->limit);
+	nv_wr32(dev, NV10_PFB_TSIZE(i), tile->pitch);
+	nv_wr32(dev, NV10_PFB_TILE(i), tile->addr);
+	nv_wr32(dev, NV20_PFB_ZCOMP(i), tile->zcomp);
+}
+
+int
+nv20_fb_vram_init(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	u32 mem_size = nv_rd32(dev, 0x10020c);
+	u32 pbus1218 = nv_rd32(dev, 0x001218);
+
+	dev_priv->vram_size = mem_size & 0xff000000;
+	switch (pbus1218 & 0x00000300) {
+	case 0x00000000: dev_priv->vram_type = NV_MEM_TYPE_SDRAM; break;
+	case 0x00000100: dev_priv->vram_type = NV_MEM_TYPE_DDR1; break;
+	case 0x00000200: dev_priv->vram_type = NV_MEM_TYPE_GDDR3; break;
+	case 0x00000300: dev_priv->vram_type = NV_MEM_TYPE_GDDR2; break;
+	}
+
+	return 0;
+}
+
+int
+nv20_fb_init(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
+	int i;
+
+	if (dev_priv->chipset >= 0x25)
+		drm_mm_init(&pfb->tag_heap, 0, 64 * 1024);
+	else
+		drm_mm_init(&pfb->tag_heap, 0, 32 * 1024);
+
+	/* Turn all the tiling regions off. */
+	pfb->num_tiles = NV10_PFB_TILE__SIZE;
+	for (i = 0; i < pfb->num_tiles; i++)
+		pfb->set_tile_region(dev, i);
+
+	return 0;
+}
+
+void
+nv20_fb_takedown(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
+	int i;
+
+	for (i = 0; i < pfb->num_tiles; i++)
+		pfb->free_tile_region(dev, i);
+
+	drm_mm_takedown(&pfb->tag_heap);
+}
diff --git a/drivers/gpu/drm/nouveau/nv40_fb.c b/drivers/gpu/drm/nouveau/nv40_fb.c
index f0ac2a7..7fbcb33 100644
--- a/drivers/gpu/drm/nouveau/nv40_fb.c
+++ b/drivers/gpu/drm/nouveau/nv40_fb.c
@@ -72,6 +72,51 @@ nv44_fb_init_gart(struct drm_device *dev)
 }
 
 int
+nv40_fb_vram_init(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	/* 0x001218 is actually present on a few other NV4X I looked at,
+	 * and even contains sane values matching 0x100474.  From looking
+	 * at various vbios images however, this isn't the case everywhere.
+	 * So, I chose to use the same regs I've seen NVIDIA reading around
+	 * the memory detection, hopefully that'll get us the right numbers
+	 */
+	if (dev_priv->chipset == 0x40) {
+		u32 pbus1218 = nv_rd32(dev, 0x001218);
+		switch (pbus1218 & 0x00000300) {
+		case 0x00000000: dev_priv->vram_type = NV_MEM_TYPE_SDRAM; break;
+		case 0x00000100: dev_priv->vram_type = NV_MEM_TYPE_DDR1; break;
+		case 0x00000200: dev_priv->vram_type = NV_MEM_TYPE_GDDR3; break;
+		case 0x00000300: dev_priv->vram_type = NV_MEM_TYPE_DDR2; break;
+		}
+	} else
+	if (dev_priv->chipset == 0x49 || dev_priv->chipset == 0x4b) {
+		u32 pfb914 = nv_rd32(dev, 0x100914);
+		switch (pfb914 & 0x00000003) {
+		case 0x00000000: dev_priv->vram_type = NV_MEM_TYPE_DDR1; break;
+		case 0x00000001: dev_priv->vram_type = NV_MEM_TYPE_DDR2; break;
+		case 0x00000002: dev_priv->vram_type = NV_MEM_TYPE_GDDR3; break;
+		case 0x00000003: break;
+		}
+	} else
+	if (dev_priv->chipset != 0x4e) {
+		u32 pfb474 = nv_rd32(dev, 0x100474);
+		if (pfb474 & 0x00000004)
+			dev_priv->vram_type = NV_MEM_TYPE_GDDR3;
+		if (pfb474 & 0x00000002)
+			dev_priv->vram_type = NV_MEM_TYPE_DDR2;
+		if (pfb474 & 0x00000001)
+			dev_priv->vram_type = NV_MEM_TYPE_DDR1;
+	} else {
+		dev_priv->vram_type = NV_MEM_TYPE_STOLEN;
+	}
+
+	dev_priv->vram_size = nv_rd32(dev, 0x10020c) & 0xff000000;
+	return 0;
+}
+
+int
 nv40_fb_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
diff --git a/drivers/gpu/drm/nouveau/nv50_crtc.c b/drivers/gpu/drm/nouveau/nv50_crtc.c
index 8f6c2ac..701b927 100644
--- a/drivers/gpu/drm/nouveau/nv50_crtc.c
+++ b/drivers/gpu/drm/nouveau/nv50_crtc.c
@@ -170,6 +170,41 @@ nv50_crtc_set_dither(struct nouveau_crtc *nv_crtc, bool update)
 	return ret;
 }
 
+static int
+nv50_crtc_set_color_vibrance(struct nouveau_crtc *nv_crtc, bool update)
+{
+	struct drm_device *dev = nv_crtc->base.dev;
+	struct nouveau_channel *evo = nv50_display(dev)->master;
+	int ret;
+	int adj;
+	u32 hue, vib;
+
+	NV_DEBUG_KMS(dev, "vibrance = %i, hue = %i\n",
+		     nv_crtc->color_vibrance, nv_crtc->vibrant_hue);
+
+	ret = RING_SPACE(evo, 2 + (update ? 2 : 0));
+	if (ret) {
+		NV_ERROR(dev, "no space while setting color vibrance\n");
+		return ret;
+	}
+
+	adj = (nv_crtc->color_vibrance > 0) ? 50 : 0;
+	vib = ((nv_crtc->color_vibrance * 2047 + adj) / 100) & 0xfff;
+
+	hue = ((nv_crtc->vibrant_hue * 2047) / 100) & 0xfff;
+
+	BEGIN_RING(evo, 0, NV50_EVO_CRTC(nv_crtc->index, COLOR_CTRL), 1);
+	OUT_RING  (evo, (hue << 20) | (vib << 8));
+
+	if (update) {
+		BEGIN_RING(evo, 0, NV50_EVO_UPDATE, 1);
+		OUT_RING  (evo, 0);
+		FIRE_RING (evo);
+	}
+
+	return 0;
+}
+
 struct nouveau_connector *
 nouveau_crtc_connector_get(struct nouveau_crtc *nv_crtc)
 {
@@ -577,8 +612,6 @@ nv50_crtc_do_mode_set_base(struct drm_crtc *crtc,
 	OUT_RING  (evo, fb->base.depth == 8 ?
 		   NV50_EVO_CRTC_CLUT_MODE_OFF : NV50_EVO_CRTC_CLUT_MODE_ON);
 
-	BEGIN_RING(evo, 0, NV50_EVO_CRTC(nv_crtc->index, COLOR_CTRL), 1);
-	OUT_RING  (evo, NV50_EVO_CRTC_COLOR_CTRL_COLOR);
 	BEGIN_RING(evo, 0, NV50_EVO_CRTC(nv_crtc->index, FB_POS), 1);
 	OUT_RING  (evo, (y << 16) | x);
 
@@ -661,6 +694,7 @@ nv50_crtc_mode_set(struct drm_crtc *crtc, struct drm_display_mode *umode,
 
 	nv_crtc->set_dither(nv_crtc, false);
 	nv_crtc->set_scale(nv_crtc, false);
+	nv_crtc->set_color_vibrance(nv_crtc, false);
 
 	return nv50_crtc_do_mode_set_base(crtc, old_fb, x, y, false);
 }
@@ -721,6 +755,9 @@ nv50_crtc_create(struct drm_device *dev, int index)
 	if (!nv_crtc)
 		return -ENOMEM;
 
+	nv_crtc->color_vibrance = 50;
+	nv_crtc->vibrant_hue = 0;
+
 	/* Default CLUT parameters, will be activated on the hw upon
 	 * first mode set.
 	 */
@@ -751,6 +788,7 @@ nv50_crtc_create(struct drm_device *dev, int index)
 	/* set function pointers */
 	nv_crtc->set_dither = nv50_crtc_set_dither;
 	nv_crtc->set_scale = nv50_crtc_set_scale;
+	nv_crtc->set_color_vibrance = nv50_crtc_set_color_vibrance;
 
 	drm_crtc_init(dev, &nv_crtc->base, &nv50_crtc_funcs);
 	drm_crtc_helper_add(&nv_crtc->base, &nv50_crtc_helper_funcs);
diff --git a/drivers/gpu/drm/nouveau/nv50_display.c b/drivers/gpu/drm/nouveau/nv50_display.c
index 7ba28e0..0e47a89 100644
--- a/drivers/gpu/drm/nouveau/nv50_display.c
+++ b/drivers/gpu/drm/nouveau/nv50_display.c
@@ -50,6 +50,29 @@ nv50_sor_nr(struct drm_device *dev)
 	return 4;
 }
 
+u32
+nv50_display_active_crtcs(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	u32 mask = 0;
+	int i;
+
+	if (dev_priv->chipset  < 0x90 ||
+	    dev_priv->chipset == 0x92 ||
+	    dev_priv->chipset == 0xa0) {
+		for (i = 0; i < 2; i++)
+			mask |= nv_rd32(dev, NV50_PDISPLAY_SOR_MODE_CTRL_C(i));
+	} else {
+		for (i = 0; i < 4; i++)
+			mask |= nv_rd32(dev, NV90_PDISPLAY_SOR_MODE_CTRL_C(i));
+	}
+
+	for (i = 0; i < 3; i++)
+		mask |= nv_rd32(dev, NV50_PDISPLAY_DAC_MODE_CTRL_C(i));
+
+	return mask & 3;
+}
+
 static int
 evo_icmd(struct drm_device *dev, int ch, u32 mthd, u32 data)
 {
@@ -840,9 +863,9 @@ nv50_display_unk20_handler(struct drm_device *dev)
 	if (type == OUTPUT_DP) {
 		int link = !(dcb->dpconf.sor.link & 1);
 		if ((mc & 0x000f0000) == 0x00020000)
-			nouveau_dp_tu_update(dev, or, link, pclk, 18);
+			nv50_sor_dp_calc_tu(dev, or, link, pclk, 18);
 		else
-			nouveau_dp_tu_update(dev, or, link, pclk, 24);
+			nv50_sor_dp_calc_tu(dev, or, link, pclk, 24);
 	}
 
 	if (dcb->type != OUTPUT_ANALOG) {
diff --git a/drivers/gpu/drm/nouveau/nv50_display.h b/drivers/gpu/drm/nouveau/nv50_display.h
index 95874f7..5d3dd14 100644
--- a/drivers/gpu/drm/nouveau/nv50_display.h
+++ b/drivers/gpu/drm/nouveau/nv50_display.h
@@ -74,6 +74,8 @@ void nv50_display_destroy(struct drm_device *dev);
 int nv50_crtc_blank(struct nouveau_crtc *, bool blank);
 int nv50_crtc_set_clock(struct drm_device *, int head, int pclk);
 
+u32  nv50_display_active_crtcs(struct drm_device *);
+
 int  nv50_display_sync(struct drm_device *);
 int  nv50_display_flip_next(struct drm_crtc *, struct drm_framebuffer *,
 			    struct nouveau_channel *chan);
diff --git a/drivers/gpu/drm/nouveau/nv50_evo.h b/drivers/gpu/drm/nouveau/nv50_evo.h
index 3860ca6..771d879 100644
--- a/drivers/gpu/drm/nouveau/nv50_evo.h
+++ b/drivers/gpu/drm/nouveau/nv50_evo.h
@@ -104,7 +104,8 @@
 #define NV50_EVO_CRTC_SCALE_CTRL_INACTIVE                            0x00000000
 #define NV50_EVO_CRTC_SCALE_CTRL_ACTIVE                              0x00000009
 #define NV50_EVO_CRTC_COLOR_CTRL                                     0x000008a8
-#define NV50_EVO_CRTC_COLOR_CTRL_COLOR                               0x00040000
+#define NV50_EVO_CRTC_COLOR_CTRL_VIBRANCE                            0x000fff00
+#define NV50_EVO_CRTC_COLOR_CTRL_HUE                                 0xfff00000
 #define NV50_EVO_CRTC_FB_POS                                         0x000008c0
 #define NV50_EVO_CRTC_REAL_RES                                       0x000008c8
 #define NV50_EVO_CRTC_SCALE_CENTER_OFFSET                            0x000008d4
diff --git a/drivers/gpu/drm/nouveau/nv50_pm.c b/drivers/gpu/drm/nouveau/nv50_pm.c
index ec5481d..d020ed4 100644
--- a/drivers/gpu/drm/nouveau/nv50_pm.c
+++ b/drivers/gpu/drm/nouveau/nv50_pm.c
@@ -28,6 +28,7 @@
 #include "nouveau_hw.h"
 #include "nouveau_pm.h"
 #include "nouveau_hwsq.h"
+#include "nv50_display.h"
 
 enum clk_src {
 	clk_src_crystal,
@@ -352,17 +353,13 @@ nv50_pm_clocks_get(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 }
 
 struct nv50_pm_state {
+	struct nouveau_pm_level *perflvl;
+	struct hwsq_ucode eclk_hwsq;
 	struct hwsq_ucode mclk_hwsq;
 	u32 mscript;
-
-	u32 emast;
-	u32 nctrl;
-	u32 ncoef;
-	u32 sctrl;
-	u32 scoef;
-
-	u32 amast;
-	u32 pdivs;
+	u32 mmast;
+	u32 mctrl;
+	u32 mcoef;
 };
 
 static u32
@@ -415,40 +412,153 @@ clk_same(u32 a, u32 b)
 	return ((a / 1000) == (b / 1000));
 }
 
+static void
+mclk_precharge(struct nouveau_mem_exec_func *exec)
+{
+	struct nv50_pm_state *info = exec->priv;
+	struct hwsq_ucode *hwsq = &info->mclk_hwsq;
+
+	hwsq_wr32(hwsq, 0x1002d4, 0x00000001);
+}
+
+static void
+mclk_refresh(struct nouveau_mem_exec_func *exec)
+{
+	struct nv50_pm_state *info = exec->priv;
+	struct hwsq_ucode *hwsq = &info->mclk_hwsq;
+
+	hwsq_wr32(hwsq, 0x1002d0, 0x00000001);
+}
+
+static void
+mclk_refresh_auto(struct nouveau_mem_exec_func *exec, bool enable)
+{
+	struct nv50_pm_state *info = exec->priv;
+	struct hwsq_ucode *hwsq = &info->mclk_hwsq;
+
+	hwsq_wr32(hwsq, 0x100210, enable ? 0x80000000 : 0x00000000);
+}
+
+static void
+mclk_refresh_self(struct nouveau_mem_exec_func *exec, bool enable)
+{
+	struct nv50_pm_state *info = exec->priv;
+	struct hwsq_ucode *hwsq = &info->mclk_hwsq;
+
+	hwsq_wr32(hwsq, 0x1002dc, enable ? 0x00000001 : 0x00000000);
+}
+
+static void
+mclk_wait(struct nouveau_mem_exec_func *exec, u32 nsec)
+{
+	struct nv50_pm_state *info = exec->priv;
+	struct hwsq_ucode *hwsq = &info->mclk_hwsq;
+
+	if (nsec > 1000)
+		hwsq_usec(hwsq, (nsec + 500) / 1000);
+}
+
+static u32
+mclk_mrg(struct nouveau_mem_exec_func *exec, int mr)
+{
+	if (mr <= 1)
+		return nv_rd32(exec->dev, 0x1002c0 + ((mr - 0) * 4));
+	if (mr <= 3)
+		return nv_rd32(exec->dev, 0x1002e0 + ((mr - 2) * 4));
+	return 0;
+}
+
+static void
+mclk_mrs(struct nouveau_mem_exec_func *exec, int mr, u32 data)
+{
+	struct drm_nouveau_private *dev_priv = exec->dev->dev_private;
+	struct nv50_pm_state *info = exec->priv;
+	struct hwsq_ucode *hwsq = &info->mclk_hwsq;
+
+	if (mr <= 1) {
+		if (dev_priv->vram_rank_B)
+			hwsq_wr32(hwsq, 0x1002c8 + ((mr - 0) * 4), data);
+		hwsq_wr32(hwsq, 0x1002c0 + ((mr - 0) * 4), data);
+	} else
+	if (mr <= 3) {
+		if (dev_priv->vram_rank_B)
+			hwsq_wr32(hwsq, 0x1002e8 + ((mr - 2) * 4), data);
+		hwsq_wr32(hwsq, 0x1002e0 + ((mr - 2) * 4), data);
+	}
+}
+
+static void
+mclk_clock_set(struct nouveau_mem_exec_func *exec)
+{
+	struct nv50_pm_state *info = exec->priv;
+	struct hwsq_ucode *hwsq = &info->mclk_hwsq;
+	u32 ctrl = nv_rd32(exec->dev, 0x004008);
+
+	info->mmast = nv_rd32(exec->dev, 0x00c040);
+	info->mmast &= ~0xc0000000; /* get MCLK_2 from HREF */
+	info->mmast |=  0x0000c000; /* use MCLK_2 as MPLL_BYPASS clock */
+
+	hwsq_wr32(hwsq, 0xc040, info->mmast);
+	hwsq_wr32(hwsq, 0x4008, ctrl | 0x00000200); /* bypass MPLL */
+	if (info->mctrl & 0x80000000)
+		hwsq_wr32(hwsq, 0x400c, info->mcoef);
+	hwsq_wr32(hwsq, 0x4008, info->mctrl);
+}
+
+static void
+mclk_timing_set(struct nouveau_mem_exec_func *exec)
+{
+	struct drm_device *dev = exec->dev;
+	struct nv50_pm_state *info = exec->priv;
+	struct nouveau_pm_level *perflvl = info->perflvl;
+	struct hwsq_ucode *hwsq = &info->mclk_hwsq;
+	int i;
+
+	for (i = 0; i < 9; i++) {
+		u32 reg = 0x100220 + (i * 4);
+		u32 val = nv_rd32(dev, reg);
+		if (val != perflvl->timing.reg[i])
+			hwsq_wr32(hwsq, reg, perflvl->timing.reg[i]);
+	}
+}
+
 static int
-calc_mclk(struct drm_device *dev, u32 freq, struct hwsq_ucode *hwsq)
+calc_mclk(struct drm_device *dev, struct nouveau_pm_level *perflvl,
+	  struct nv50_pm_state *info)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	u32 crtc_mask = nv50_display_active_crtcs(dev);
+	struct nouveau_mem_exec_func exec = {
+		.dev = dev,
+		.precharge = mclk_precharge,
+		.refresh = mclk_refresh,
+		.refresh_auto = mclk_refresh_auto,
+		.refresh_self = mclk_refresh_self,
+		.wait = mclk_wait,
+		.mrg = mclk_mrg,
+		.mrs = mclk_mrs,
+		.clock_set = mclk_clock_set,
+		.timing_set = mclk_timing_set,
+		.priv = info
+	};
+	struct hwsq_ucode *hwsq = &info->mclk_hwsq;
 	struct pll_lims pll;
-	u32 mast = nv_rd32(dev, 0x00c040);
-	u32 ctrl = nv_rd32(dev, 0x004008);
-	u32 coef = nv_rd32(dev, 0x00400c);
-	u32 orig = ctrl;
-	u32 crtc_mask = 0;
 	int N, M, P;
-	int ret, i;
+	int ret;
 
 	/* use pcie refclock if possible, otherwise use mpll */
-	ctrl &= ~0x81ff0200;
-	if (clk_same(freq, read_clk(dev, clk_src_href))) {
-		ctrl |= 0x00000200 | (pll.log2p_bias << 19);
+	info->mctrl  = nv_rd32(dev, 0x004008);
+	info->mctrl &= ~0x81ff0200;
+	if (clk_same(perflvl->memory, read_clk(dev, clk_src_href))) {
+		info->mctrl |= 0x00000200 | (pll.log2p_bias << 19);
 	} else {
-		ret = calc_pll(dev, 0x4008, &pll, freq, &N, &M, &P);
+		ret = calc_pll(dev, 0x4008, &pll, perflvl->memory, &N, &M, &P);
 		if (ret == 0)
 			return -EINVAL;
 
-		ctrl |= 0x80000000 | (P << 22) | (P << 16);
-		ctrl |= pll.log2p_bias << 19;
-		coef  = (N << 8) | M;
-	}
-
-	mast &= ~0xc0000000; /* get MCLK_2 from HREF */
-	mast |=  0x0000c000; /* use MCLK_2 as MPLL_BYPASS clock */
-
-	/* determine active crtcs */
-	for (i = 0; i < 2; i++) {
-		if (nv_rd32(dev, NV50_PDISPLAY_CRTC_C(i, CLOCK)))
-			crtc_mask |= (1 << i);
+		info->mctrl |= 0x80000000 | (P << 22) | (P << 16);
+		info->mctrl |= pll.log2p_bias << 19;
+		info->mcoef  = (N << 8) | M;
 	}
 
 	/* build the ucode which will reclock the memory for us */
@@ -462,25 +572,10 @@ calc_mclk(struct drm_device *dev, u32 freq, struct hwsq_ucode *hwsq)
 	hwsq_setf(hwsq, 0x10, 0); /* disable bus access */
 	hwsq_op5f(hwsq, 0x00, 0x01); /* no idea :s */
 
-	/* prepare memory controller */
-	hwsq_wr32(hwsq, 0x1002d4, 0x00000001); /* precharge banks and idle */
-	hwsq_wr32(hwsq, 0x1002d0, 0x00000001); /* force refresh */
-	hwsq_wr32(hwsq, 0x100210, 0x00000000); /* stop the automatic refresh */
-	hwsq_wr32(hwsq, 0x1002dc, 0x00000001); /* start self refresh mode */
-
-	/* reclock memory */
-	hwsq_wr32(hwsq, 0xc040, mast);
-	hwsq_wr32(hwsq, 0x4008, orig | 0x00000200); /* bypass MPLL */
-	hwsq_wr32(hwsq, 0x400c, coef);
-	hwsq_wr32(hwsq, 0x4008, ctrl);
-
-	/* restart memory controller */
-	hwsq_wr32(hwsq, 0x1002d4, 0x00000001); /* precharge banks and idle */
-	hwsq_wr32(hwsq, 0x1002dc, 0x00000000); /* stop self refresh mode */
-	hwsq_wr32(hwsq, 0x100210, 0x80000000); /* restart automatic refresh */
-	hwsq_usec(hwsq, 12); /* wait for the PLL to stabilize */
-
-	hwsq_usec(hwsq, 48); /* may be unnecessary: causes flickering */
+	ret = nouveau_mem_exec(&exec, perflvl);
+	if (ret)
+		return ret;
+
 	hwsq_setf(hwsq, 0x10, 1); /* enable bus access */
 	hwsq_op5f(hwsq, 0x00, 0x00); /* no idea, reverse of 0x00, 0x01? */
 	if (dev_priv->chipset >= 0x92)
@@ -494,10 +589,11 @@ nv50_pm_clocks_pre(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nv50_pm_state *info;
+	struct hwsq_ucode *hwsq;
 	struct pll_lims pll;
+	u32 out, mast, divs, ctrl;
 	int clk, ret = -EINVAL;
 	int N, M, P1, P2;
-	u32 out;
 
 	if (dev_priv->chipset == 0xaa ||
 	    dev_priv->chipset == 0xac)
@@ -506,54 +602,44 @@ nv50_pm_clocks_pre(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 	info = kmalloc(sizeof(*info), GFP_KERNEL);
 	if (!info)
 		return ERR_PTR(-ENOMEM);
+	info->perflvl = perflvl;
 
-	/* core: for the moment at least, always use nvpll */
-	clk = calc_pll(dev, 0x4028, &pll, perflvl->core, &N, &M, &P1);
-	if (clk == 0)
-		goto error;
+	/* memory: build hwsq ucode which we'll use to reclock memory.
+	 *         use pcie refclock if possible, otherwise use mpll */
+	info->mclk_hwsq.len = 0;
+	if (perflvl->memory) {
+		ret = calc_mclk(dev, perflvl, info);
+		if (ret)
+			goto error;
+		info->mscript = perflvl->memscript;
+	}
 
-	info->emast = 0x00000003;
-	info->nctrl = 0x80000000 | (P1 << 19) | (P1 << 16);
-	info->ncoef = (N << 8) | M;
+	divs = read_div(dev);
+	mast = info->mmast;
 
-	/* shader: tie to nvclk if possible, otherwise use spll.  have to be
-	 * very careful that the shader clock is at least twice the core, or
-	 * some chipsets will be very unhappy.  i expect most or all of these
-	 * cases will be handled by tying to nvclk, but it's possible there's
-	 * corners
-	 */
-	if (P1-- && perflvl->shader == (perflvl->core << 1)) {
-		info->emast |= 0x00000020;
-		info->sctrl  = 0x00000000 | (P1 << 19) | (P1 << 16);
-		info->scoef  = nv_rd32(dev, 0x004024);
-	} else {
-		clk = calc_pll(dev, 0x4020, &pll, perflvl->shader, &N, &M, &P1);
-		if (clk == 0)
-			goto error;
+	/* start building HWSQ script for engine reclocking */
+	hwsq = &info->eclk_hwsq;
+	hwsq_init(hwsq);
+	hwsq_setf(hwsq, 0x10, 0); /* disable bus access */
+	hwsq_op5f(hwsq, 0x00, 0x01); /* wait for access disabled? */
 
-		info->emast |= 0x00000030;
-		info->sctrl  = 0x80000000 | (P1 << 19) | (P1 << 16);
-		info->scoef  = (N << 8) | M;
+	/* vdec/dom6: switch to "safe" clocks temporarily */
+	if (perflvl->vdec) {
+		mast &= ~0x00000c00;
+		divs &= ~0x00000700;
 	}
 
-	/* memory: build hwsq ucode which we'll use to reclock memory */
-	info->mclk_hwsq.len = 0;
-	if (perflvl->memory) {
-		clk = calc_mclk(dev, perflvl->memory, &info->mclk_hwsq);
-		if (clk < 0) {
-			ret = clk;
-			goto error;
-		}
-
-		info->mscript = perflvl->memscript;
+	if (perflvl->dom6) {
+		mast &= ~0x0c000000;
+		divs &= ~0x00000007;
 	}
 
+	hwsq_wr32(hwsq, 0x00c040, mast);
+
 	/* vdec: avoid modifying xpll until we know exactly how the other
 	 * clock domains work, i suspect at least some of them can also be
 	 * tied to xpll...
 	 */
-	info->amast = nv_rd32(dev, 0x00c040);
-	info->pdivs = read_div(dev);
 	if (perflvl->vdec) {
 		/* see how close we can get using nvclk as a source */
 		clk = calc_div(perflvl->core, perflvl->vdec, &P1);
@@ -566,16 +652,14 @@ nv50_pm_clocks_pre(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 		out = calc_div(out, perflvl->vdec, &P2);
 
 		/* select whichever gets us closest */
-		info->amast &= ~0x00000c00;
-		info->pdivs &= ~0x00000700;
 		if (abs((int)perflvl->vdec - clk) <=
 		    abs((int)perflvl->vdec - out)) {
 			if (dev_priv->chipset != 0x98)
-				info->amast |= 0x00000c00;
-			info->pdivs |= P1 << 8;
+				mast |= 0x00000c00;
+			divs |= P1 << 8;
 		} else {
-			info->amast |= 0x00000800;
-			info->pdivs |= P2 << 8;
+			mast |= 0x00000800;
+			divs |= P2 << 8;
 		}
 	}
 
@@ -583,21 +667,82 @@ nv50_pm_clocks_pre(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 	 * of the host clock frequency
 	 */
 	if (perflvl->dom6) {
-		info->amast &= ~0x0c000000;
 		if (clk_same(perflvl->dom6, read_clk(dev, clk_src_href))) {
-			info->amast |= 0x00000000;
+			mast |= 0x00000000;
 		} else
 		if (clk_same(perflvl->dom6, read_clk(dev, clk_src_hclk))) {
-			info->amast |= 0x08000000;
+			mast |= 0x08000000;
 		} else {
 			clk = read_clk(dev, clk_src_hclk) * 3;
 			clk = calc_div(clk, perflvl->dom6, &P1);
 
-			info->amast |= 0x0c000000;
-			info->pdivs  = (info->pdivs & ~0x00000007) | P1;
+			mast |= 0x0c000000;
+			divs |= P1;
 		}
 	}
 
+	/* vdec/dom6: complete switch to new clocks */
+	switch (dev_priv->chipset) {
+	case 0x92:
+	case 0x94:
+	case 0x96:
+		hwsq_wr32(hwsq, 0x004800, divs);
+		break;
+	default:
+		hwsq_wr32(hwsq, 0x004700, divs);
+		break;
+	}
+
+	hwsq_wr32(hwsq, 0x00c040, mast);
+
+	/* core/shader: make sure sclk/nvclk are disconnected from their
+	 * PLLs (nvclk to dom6, sclk to hclk)
+	 */
+	if (dev_priv->chipset < 0x92)
+		mast = (mast & ~0x001000b0) | 0x00100080;
+	else
+		mast = (mast & ~0x000000b3) | 0x00000081;
+
+	hwsq_wr32(hwsq, 0x00c040, mast);
+
+	/* core: for the moment at least, always use nvpll */
+	clk = calc_pll(dev, 0x4028, &pll, perflvl->core, &N, &M, &P1);
+	if (clk == 0)
+		goto error;
+
+	ctrl  = nv_rd32(dev, 0x004028) & ~0xc03f0100;
+	mast &= ~0x00100000;
+	mast |= 3;
+
+	hwsq_wr32(hwsq, 0x004028, 0x80000000 | (P1 << 19) | (P1 << 16) | ctrl);
+	hwsq_wr32(hwsq, 0x00402c, (N << 8) | M);
+
+	/* shader: tie to nvclk if possible, otherwise use spll.  have to be
+	 * very careful that the shader clock is at least twice the core, or
+	 * some chipsets will be very unhappy.  i expect most or all of these
+	 * cases will be handled by tying to nvclk, but it's possible there's
+	 * corners
+	 */
+	ctrl = nv_rd32(dev, 0x004020) & ~0xc03f0100;
+
+	if (P1-- && perflvl->shader == (perflvl->core << 1)) {
+		hwsq_wr32(hwsq, 0x004020, (P1 << 19) | (P1 << 16) | ctrl);
+		hwsq_wr32(hwsq, 0x00c040, 0x00000020 | mast);
+	} else {
+		clk = calc_pll(dev, 0x4020, &pll, perflvl->shader, &N, &M, &P1);
+		if (clk == 0)
+			goto error;
+		ctrl |= 0x80000000;
+
+		hwsq_wr32(hwsq, 0x004020, (P1 << 19) | (P1 << 16) | ctrl);
+		hwsq_wr32(hwsq, 0x004024, (N << 8) | M);
+		hwsq_wr32(hwsq, 0x00c040, 0x00000030 | mast);
+	}
+
+	hwsq_setf(hwsq, 0x10, 1); /* enable bus access */
+	hwsq_op5f(hwsq, 0x00, 0x00); /* wait for access enabled? */
+	hwsq_fini(hwsq);
+
 	return info;
 error:
 	kfree(info);
@@ -605,23 +750,24 @@ error:
 }
 
 static int
-prog_mclk(struct drm_device *dev, struct hwsq_ucode *hwsq)
+prog_hwsq(struct drm_device *dev, struct hwsq_ucode *hwsq)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	u32 hwsq_data, hwsq_kick;
 	int i;
 
-	if (dev_priv->chipset < 0x90) {
+	if (dev_priv->chipset < 0x94) {
 		hwsq_data = 0x001400;
 		hwsq_kick = 0x00000003;
 	} else {
 		hwsq_data = 0x080000;
 		hwsq_kick = 0x00000001;
 	}
-
 	/* upload hwsq ucode */
 	nv_mask(dev, 0x001098, 0x00000008, 0x00000000);
 	nv_wr32(dev, 0x001304, 0x00000000);
+	if (dev_priv->chipset >= 0x92)
+		nv_wr32(dev, 0x001318, 0x00000000);
 	for (i = 0; i < hwsq->len / 4; i++)
 		nv_wr32(dev, hwsq_data + (i * 4), hwsq->ptr.u32[i]);
 	nv_mask(dev, 0x001098, 0x00000018, 0x00000018);
@@ -645,20 +791,19 @@ prog_mclk(struct drm_device *dev, struct hwsq_ucode *hwsq)
 int
 nv50_pm_clocks_set(struct drm_device *dev, void *data)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nv50_pm_state *info = data;
 	struct bit_entry M;
-	int ret = 0;
+	int ret = -EBUSY;
 
 	/* halt and idle execution engines */
 	nv_mask(dev, 0x002504, 0x00000001, 0x00000001);
 	if (!nv_wait(dev, 0x002504, 0x00000010, 0x00000010))
-		goto error;
+		goto resume;
+	if (!nv_wait(dev, 0x00251c, 0x0000003f, 0x0000003f))
+		goto resume;
 
-	/* memory: it is *very* important we change this first, the ucode
-	 * we build in pre() now has hardcoded 0xc040 values, which can't
-	 * change before we execute it or the engine clocks may end up
-	 * messed up.
+	/* program memory clock, if necessary - must come before engine clock
+	 * reprogramming due to how we construct the hwsq scripts in pre()
 	 */
 	if (info->mclk_hwsq.len) {
 		/* execute some scripts that do ??? from the vbios.. */
@@ -672,42 +817,14 @@ nv50_pm_clocks_set(struct drm_device *dev, void *data)
 			nouveau_bios_init_exec(dev, info->mscript);
 		}
 
-		ret = prog_mclk(dev, &info->mclk_hwsq);
+		ret = prog_hwsq(dev, &info->mclk_hwsq);
 		if (ret)
 			goto resume;
 	}
 
-	/* reclock vdec/dom6 */
-	nv_mask(dev, 0x00c040, 0x00000c00, 0x00000000);
-	switch (dev_priv->chipset) {
-	case 0x92:
-	case 0x94:
-	case 0x96:
-		nv_mask(dev, 0x004800, 0x00000707, info->pdivs);
-		break;
-	default:
-		nv_mask(dev, 0x004700, 0x00000707, info->pdivs);
-		break;
-	}
-	nv_mask(dev, 0x00c040, 0x0c000c00, info->amast);
+	/* program engine clocks */
+	ret = prog_hwsq(dev, &info->eclk_hwsq);
 
-	/* core/shader: make sure sclk/nvclk are disconnected from their
-	 * plls (nvclk to dom6, sclk to hclk), modify the plls, and
-	 * reconnect sclk/nvclk to their new clock source
-	 */
-	if (dev_priv->chipset < 0x92)
-		nv_mask(dev, 0x00c040, 0x001000b0, 0x00100080); /* grrr! */
-	else
-		nv_mask(dev, 0x00c040, 0x000000b3, 0x00000081);
-	nv_mask(dev, 0x004020, 0xc03f0100, info->sctrl);
-	nv_wr32(dev, 0x004024, info->scoef);
-	nv_mask(dev, 0x004028, 0xc03f0100, info->nctrl);
-	nv_wr32(dev, 0x00402c, info->ncoef);
-	nv_mask(dev, 0x00c040, 0x00100033, info->emast);
-
-	goto resume;
-error:
-	ret = -EBUSY;
 resume:
 	nv_mask(dev, 0x002504, 0x00000001, 0x00000000);
 	kfree(info);
diff --git a/drivers/gpu/drm/nouveau/nv50_sor.c b/drivers/gpu/drm/nouveau/nv50_sor.c
index c4423ba..9a5c2c9 100644
--- a/drivers/gpu/drm/nouveau/nv50_sor.c
+++ b/drivers/gpu/drm/nouveau/nv50_sor.c
@@ -36,6 +36,193 @@
 #include "nouveau_crtc.h"
 #include "nv50_display.h"
 
+static u32
+nv50_sor_dp_lane_map(struct drm_device *dev, struct dcb_entry *dcb, u8 lane)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	static const u8 nvaf[] = { 24, 16, 8, 0 }; /* thanks, apple.. */
+	static const u8 nv50[] = { 16, 8, 0, 24 };
+	if (dev_priv->card_type == 0xaf)
+		return nvaf[lane];
+	return nv50[lane];
+}
+
+static void
+nv50_sor_dp_train_set(struct drm_device *dev, struct dcb_entry *dcb, u8 pattern)
+{
+	u32 or = ffs(dcb->or) - 1, link = !(dcb->sorconf.link & 1);
+	nv_mask(dev, NV50_SOR_DP_CTRL(or, link), 0x0f000000, pattern << 24);
+}
+
+static void
+nv50_sor_dp_train_adj(struct drm_device *dev, struct dcb_entry *dcb,
+		      u8 lane, u8 swing, u8 preem)
+{
+	u32 or = ffs(dcb->or) - 1, link = !(dcb->sorconf.link & 1);
+	u32 shift = nv50_sor_dp_lane_map(dev, dcb, lane);
+	u32 mask = 0x000000ff << shift;
+	u8 *table, *entry, *config;
+
+	table = nouveau_dp_bios_data(dev, dcb, &entry);
+	if (!table || (table[0] != 0x20 && table[0] != 0x21)) {
+		NV_ERROR(dev, "PDISP: unsupported DP table for chipset\n");
+		return;
+	}
+
+	config = entry + table[4];
+	while (config[0] != swing || config[1] != preem) {
+		config += table[5];
+		if (config >= entry + table[4] + entry[4] * table[5])
+			return;
+	}
+
+	nv_mask(dev, NV50_SOR_DP_UNK118(or, link), mask, config[2] << shift);
+	nv_mask(dev, NV50_SOR_DP_UNK120(or, link), mask, config[3] << shift);
+	nv_mask(dev, NV50_SOR_DP_UNK130(or, link), 0x0000ff00, config[4] << 8);
+}
+
+static void
+nv50_sor_dp_link_set(struct drm_device *dev, struct dcb_entry *dcb, int crtc,
+		     int link_nr, u32 link_bw, bool enhframe)
+{
+	u32 or = ffs(dcb->or) - 1, link = !(dcb->sorconf.link & 1);
+	u32 dpctrl = nv_rd32(dev, NV50_SOR_DP_CTRL(or, link)) & ~0x001f4000;
+	u32 clksor = nv_rd32(dev, 0x614300 + (or * 0x800)) & ~0x000c0000;
+	u8 *table, *entry, mask;
+	int i;
+
+	table = nouveau_dp_bios_data(dev, dcb, &entry);
+	if (!table || (table[0] != 0x20 && table[0] != 0x21)) {
+		NV_ERROR(dev, "PDISP: unsupported DP table for chipset\n");
+		return;
+	}
+
+	entry = ROMPTR(dev, entry[10]);
+	if (entry) {
+		while (link_bw < ROM16(entry[0]) * 10)
+			entry += 4;
+
+		nouveau_bios_run_init_table(dev, ROM16(entry[2]), dcb, crtc);
+	}
+
+	dpctrl |= ((1 << link_nr) - 1) << 16;
+	if (enhframe)
+		dpctrl |= 0x00004000;
+
+	if (link_bw > 162000)
+		clksor |= 0x00040000;
+
+	nv_wr32(dev, 0x614300 + (or * 0x800), clksor);
+	nv_wr32(dev, NV50_SOR_DP_CTRL(or, link), dpctrl);
+
+	mask = 0;
+	for (i = 0; i < link_nr; i++)
+		mask |= 1 << (nv50_sor_dp_lane_map(dev, dcb, i) >> 3);
+	nv_mask(dev, NV50_SOR_DP_UNK130(or, link), 0x0000000f, mask);
+}
+
+static void
+nv50_sor_dp_link_get(struct drm_device *dev, u32 or, u32 link, u32 *nr, u32 *bw)
+{
+	u32 dpctrl = nv_rd32(dev, NV50_SOR_DP_CTRL(or, link)) & 0x000f0000;
+	u32 clksor = nv_rd32(dev, 0x614300 + (or * 0x800));
+	if (clksor & 0x000c0000)
+		*bw = 270000;
+	else
+		*bw = 162000;
+
+	if      (dpctrl > 0x00030000) *nr = 4;
+	else if (dpctrl > 0x00010000) *nr = 2;
+	else			      *nr = 1;
+}
+
+void
+nv50_sor_dp_calc_tu(struct drm_device *dev, int or, int link, u32 clk, u32 bpp)
+{
+	const u32 symbol = 100000;
+	int bestTU = 0, bestVTUi = 0, bestVTUf = 0, bestVTUa = 0;
+	int TU, VTUi, VTUf, VTUa;
+	u64 link_data_rate, link_ratio, unk;
+	u32 best_diff = 64 * symbol;
+	u32 link_nr, link_bw, r;
+
+	/* calculate packed data rate for each lane */
+	nv50_sor_dp_link_get(dev, or, link, &link_nr, &link_bw);
+	link_data_rate = (clk * bpp / 8) / link_nr;
+
+	/* calculate ratio of packed data rate to link symbol rate */
+	link_ratio = link_data_rate * symbol;
+	r = do_div(link_ratio, link_bw);
+
+	for (TU = 64; TU >= 32; TU--) {
+		/* calculate average number of valid symbols in each TU */
+		u32 tu_valid = link_ratio * TU;
+		u32 calc, diff;
+
+		/* find a hw representation for the fraction.. */
+		VTUi = tu_valid / symbol;
+		calc = VTUi * symbol;
+		diff = tu_valid - calc;
+		if (diff) {
+			if (diff >= (symbol / 2)) {
+				VTUf = symbol / (symbol - diff);
+				if (symbol - (VTUf * diff))
+					VTUf++;
+
+				if (VTUf <= 15) {
+					VTUa  = 1;
+					calc += symbol - (symbol / VTUf);
+				} else {
+					VTUa  = 0;
+					VTUf  = 1;
+					calc += symbol;
+				}
+			} else {
+				VTUa  = 0;
+				VTUf  = min((int)(symbol / diff), 15);
+				calc += symbol / VTUf;
+			}
+
+			diff = calc - tu_valid;
+		} else {
+			/* no remainder, but the hw doesn't like the fractional
+			 * part to be zero.  decrement the integer part and
+			 * have the fraction add a whole symbol back
+			 */
+			VTUa = 0;
+			VTUf = 1;
+			VTUi--;
+		}
+
+		if (diff < best_diff) {
+			best_diff = diff;
+			bestTU = TU;
+			bestVTUa = VTUa;
+			bestVTUf = VTUf;
+			bestVTUi = VTUi;
+			if (diff == 0)
+				break;
+		}
+	}
+
+	if (!bestTU) {
+		NV_ERROR(dev, "DP: unable to find suitable config\n");
+		return;
+	}
+
+	/* XXX close to vbios numbers, but not right */
+	unk  = (symbol - link_ratio) * bestTU;
+	unk *= link_ratio;
+	r = do_div(unk, symbol);
+	r = do_div(unk, symbol);
+	unk += 6;
+
+	nv_mask(dev, NV50_SOR_DP_CTRL(or, link), 0x000001fc, bestTU << 2);
+	nv_mask(dev, NV50_SOR_DP_SCFG(or, link), 0x010f7f3f, bestVTUa << 24 |
+							     bestVTUf << 16 |
+							     bestVTUi << 8 |
+							     unk);
+}
 static void
 nv50_sor_disconnect(struct drm_encoder *encoder)
 {
@@ -117,20 +304,13 @@ nv50_sor_dpms(struct drm_encoder *encoder, int mode)
 	}
 
 	if (nv_encoder->dcb->type == OUTPUT_DP) {
-		struct nouveau_i2c_chan *auxch;
-
-		auxch = nouveau_i2c_find(dev, nv_encoder->dcb->i2c_index);
-		if (!auxch)
-			return;
+		struct dp_train_func func = {
+			.link_set = nv50_sor_dp_link_set,
+			.train_set = nv50_sor_dp_train_set,
+			.train_adj = nv50_sor_dp_train_adj
+		};
 
-		if (mode == DRM_MODE_DPMS_ON) {
-			u8 status = DP_SET_POWER_D0;
-			nouveau_dp_auxch(auxch, 8, DP_SET_POWER, &status, 1);
-			nouveau_dp_link_train(encoder, nv_encoder->dp.datarate);
-		} else {
-			u8 status = DP_SET_POWER_D3;
-			nouveau_dp_auxch(auxch, 8, DP_SET_POWER, &status, 1);
-		}
+		nouveau_dp_dpms(encoder, mode, nv_encoder->dp.datarate, &func);
 	}
 }
 
diff --git a/drivers/gpu/drm/nouveau/nv50_vm.c b/drivers/gpu/drm/nouveau/nv50_vm.c
index 6f38cea..44fbac9 100644
--- a/drivers/gpu/drm/nouveau/nv50_vm.c
+++ b/drivers/gpu/drm/nouveau/nv50_vm.c
@@ -57,27 +57,15 @@ nv50_vm_map_pgt(struct nouveau_gpuobj *pgd, u32 pde,
 }
 
 static inline u64
-nv50_vm_addr(struct nouveau_vma *vma, u64 phys, u32 memtype, u32 target)
+vm_addr(struct nouveau_vma *vma, u64 phys, u32 memtype, u32 target)
 {
-	struct drm_nouveau_private *dev_priv = vma->vm->dev->dev_private;
-
 	phys |= 1; /* present */
 	phys |= (u64)memtype << 40;
-
-	/* IGPs don't have real VRAM, re-target to stolen system memory */
-	if (target == 0 && dev_priv->vram_sys_base) {
-		phys  += dev_priv->vram_sys_base;
-		target = 3;
-	}
-
 	phys |= target << 4;
-
 	if (vma->access & NV_MEM_ACCESS_SYS)
 		phys |= (1 << 6);
-
 	if (!(vma->access & NV_MEM_ACCESS_WO))
 		phys |= (1 << 3);
-
 	return phys;
 }
 
@@ -85,11 +73,19 @@ void
 nv50_vm_map(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
 	    struct nouveau_mem *mem, u32 pte, u32 cnt, u64 phys, u64 delta)
 {
+	struct drm_nouveau_private *dev_priv = vma->vm->dev->dev_private;
 	u32 comp = (mem->memtype & 0x180) >> 7;
-	u32 block;
+	u32 block, target;
 	int i;
 
-	phys  = nv50_vm_addr(vma, phys, mem->memtype, 0);
+	/* IGPs don't have real VRAM, re-target to stolen system memory */
+	target = 0;
+	if (dev_priv->vram_sys_base) {
+		phys += dev_priv->vram_sys_base;
+		target = 3;
+	}
+
+	phys  = vm_addr(vma, phys, mem->memtype, target);
 	pte <<= 3;
 	cnt <<= 3;
 
@@ -125,9 +121,10 @@ void
 nv50_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
 	       struct nouveau_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
 {
+	u32 target = (vma->access & NV_MEM_ACCESS_NOSNOOP) ? 3 : 2;
 	pte <<= 3;
 	while (cnt--) {
-		u64 phys = nv50_vm_addr(vma, (u64)*list++, mem->memtype, 2);
+		u64 phys = vm_addr(vma, (u64)*list++, mem->memtype, target);
 		nv_wo32(pgt, pte + 0, lower_32_bits(phys));
 		nv_wo32(pgt, pte + 4, upper_32_bits(phys));
 		pte += 8;
diff --git a/drivers/gpu/drm/nouveau/nv50_vram.c b/drivers/gpu/drm/nouveau/nv50_vram.c
index 2e45e57..9ed9ae39 100644
--- a/drivers/gpu/drm/nouveau/nv50_vram.c
+++ b/drivers/gpu/drm/nouveau/nv50_vram.c
@@ -189,8 +189,25 @@ nv50_vram_init(struct drm_device *dev)
 	struct nouveau_vram_engine *vram = &dev_priv->engine.vram;
 	const u32 rsvd_head = ( 256 * 1024) >> 12; /* vga memory */
 	const u32 rsvd_tail = (1024 * 1024) >> 12; /* vbios etc */
+	u32 pfb714 = nv_rd32(dev, 0x100714);
 	u32 rblock, length;
 
+	switch (pfb714 & 0x00000007) {
+	case 0: dev_priv->vram_type = NV_MEM_TYPE_DDR1; break;
+	case 1:
+		if (nouveau_mem_vbios_type(dev) == NV_MEM_TYPE_DDR3)
+			dev_priv->vram_type = NV_MEM_TYPE_DDR3;
+		else
+			dev_priv->vram_type = NV_MEM_TYPE_DDR2;
+		break;
+	case 2: dev_priv->vram_type = NV_MEM_TYPE_GDDR3; break;
+	case 3: dev_priv->vram_type = NV_MEM_TYPE_GDDR4; break;
+	case 4: dev_priv->vram_type = NV_MEM_TYPE_GDDR5; break;
+	default:
+		break;
+	}
+
+	dev_priv->vram_rank_B = !!(nv_rd32(dev, 0x100200) & 0x4);
 	dev_priv->vram_size  = nv_rd32(dev, 0x10020c);
 	dev_priv->vram_size |= (dev_priv->vram_size & 0xff) << 32;
 	dev_priv->vram_size &= 0xffffffff00ULL;
diff --git a/drivers/gpu/drm/nouveau/nvc0_pm.c b/drivers/gpu/drm/nouveau/nvc0_pm.c
index e9992f6..ce65f81 100644
--- a/drivers/gpu/drm/nouveau/nvc0_pm.c
+++ b/drivers/gpu/drm/nouveau/nvc0_pm.c
@@ -269,7 +269,7 @@ calc_clk(struct drm_device *dev, int clk, struct nvc0_pm_clock *info, u32 freq)
 	clk0 = calc_div(dev, clk, clk0, freq, &div1D);
 
 	/* see if we can get any closer using PLLs */
-	if (clk0 != freq) {
+	if (clk0 != freq && (0x00004387 & (1 << clk))) {
 		if (clk < 7)
 			clk1 = calc_pll(dev, clk, freq, &info->coef);
 		else
diff --git a/drivers/gpu/drm/nouveau/nvc0_vm.c b/drivers/gpu/drm/nouveau/nvc0_vm.c
index 9e35294..30d2bd5 100644
--- a/drivers/gpu/drm/nouveau/nvc0_vm.c
+++ b/drivers/gpu/drm/nouveau/nvc0_vm.c
@@ -77,9 +77,11 @@ void
 nvc0_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
 	       struct nouveau_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
 {
+	u32 target = (vma->access & NV_MEM_ACCESS_NOSNOOP) ? 7 : 5;
+
 	pte <<= 3;
 	while (cnt--) {
-		u64 phys = nvc0_vm_addr(vma, *list++, mem->memtype, 5);
+		u64 phys = nvc0_vm_addr(vma, *list++, mem->memtype, target);
 		nv_wo32(pgt, pte + 0, lower_32_bits(phys));
 		nv_wo32(pgt, pte + 4, upper_32_bits(phys));
 		pte += 8;
diff --git a/drivers/gpu/drm/nouveau/nvc0_vram.c b/drivers/gpu/drm/nouveau/nvc0_vram.c
index ce984d5..a7eef89 100644
--- a/drivers/gpu/drm/nouveau/nvc0_vram.c
+++ b/drivers/gpu/drm/nouveau/nvc0_vram.c
@@ -106,31 +106,32 @@ nvc0_vram_init(struct drm_device *dev)
 	struct nouveau_vram_engine *vram = &dev_priv->engine.vram;
 	const u32 rsvd_head = ( 256 * 1024) >> 12; /* vga memory */
 	const u32 rsvd_tail = (1024 * 1024) >> 12; /* vbios etc */
-	u32 parts = nv_rd32(dev, 0x121c74);
+	u32 parts = nv_rd32(dev, 0x022438);
+	u32 pmask = nv_rd32(dev, 0x022554);
 	u32 bsize = nv_rd32(dev, 0x10f20c);
 	u32 offset, length;
 	bool uniform = true;
 	int ret, part;
 
 	NV_DEBUG(dev, "0x100800: 0x%08x\n", nv_rd32(dev, 0x100800));
-	NV_DEBUG(dev, "parts 0x%08x bcast_mem_amount 0x%08x\n", parts, bsize);
+	NV_DEBUG(dev, "parts 0x%08x mask 0x%08x\n", parts, pmask);
+
+	dev_priv->vram_type = nouveau_mem_vbios_type(dev);
+	dev_priv->vram_rank_B = !!(nv_rd32(dev, 0x10f200) & 0x00000004);
 
 	/* read amount of vram attached to each memory controller */
-	part = 0;
-	while (parts) {
-		u32 psize = nv_rd32(dev, 0x11020c + (part++ * 0x1000));
-		if (psize == 0)
-			continue;
-		parts--;
-
-		if (psize != bsize) {
-			if (psize < bsize)
-				bsize = psize;
-			uniform = false;
+	for (part = 0; part < parts; part++) {
+		if (!(pmask & (1 << part))) {
+			u32 psize = nv_rd32(dev, 0x11020c + (part * 0x1000));
+			if (psize != bsize) {
+				if (psize < bsize)
+					bsize = psize;
+				uniform = false;
+			}
+
+			NV_DEBUG(dev, "%d: mem_amount 0x%08x\n", part, psize);
+			dev_priv->vram_size += (u64)psize << 20;
 		}
-
-		NV_DEBUG(dev, "%d: mem_amount 0x%08x\n", part, psize);
-		dev_priv->vram_size += (u64)psize << 20;
 	}
 
 	/* if all controllers have the same amount attached, there's no holes */
diff --git a/drivers/gpu/drm/nouveau/nvd0_display.c b/drivers/gpu/drm/nouveau/nvd0_display.c
index d2ba2f0..dfb8a95 100644
--- a/drivers/gpu/drm/nouveau/nvd0_display.c
+++ b/drivers/gpu/drm/nouveau/nvd0_display.c
@@ -284,6 +284,8 @@ nvd0_display_flip_next(struct drm_crtc *crtc, struct drm_framebuffer *fb,
 	u32 *push;
 	int ret;
 
+	evo_sync(crtc->dev, EVO_MASTER);
+
 	swap_interval <<= 4;
 	if (swap_interval == 0)
 		swap_interval |= 0x100;
@@ -593,7 +595,7 @@ nvd0_crtc_commit(struct drm_crtc *crtc)
 		evo_kick(push, crtc->dev, EVO_MASTER);
 	}
 
-	nvd0_crtc_cursor_show(nv_crtc, nv_crtc->cursor.visible, false);
+	nvd0_crtc_cursor_show(nv_crtc, nv_crtc->cursor.visible, true);
 	nvd0_display_flip_next(crtc, crtc->fb, NULL, 1);
 }
 
@@ -634,8 +636,7 @@ nvd0_crtc_mode_set(struct drm_crtc *crtc, struct drm_display_mode *umode,
 	u32 hactive, hsynce, hbackp, hfrontp, hblanke, hblanks;
 	u32 vactive, vsynce, vbackp, vfrontp, vblanke, vblanks;
 	u32 vblan2e = 0, vblan2s = 1;
-	u32 magic = 0x31ec6000;
-	u32 syncs, *push;
+	u32 *push;
 	int ret;
 
 	hactive = mode->htotal;
@@ -655,15 +656,8 @@ nvd0_crtc_mode_set(struct drm_crtc *crtc, struct drm_display_mode *umode,
 		vblan2e = vactive + vsynce + vbackp;
 		vblan2s = vblan2e + (mode->vdisplay * vscan / ilace);
 		vactive = (vactive * 2) + 1;
-		magic  |= 0x00000001;
 	}
 
-	syncs = 0x00000001;
-	if (mode->flags & DRM_MODE_FLAG_NHSYNC)
-		syncs |= 0x00000008;
-	if (mode->flags & DRM_MODE_FLAG_NVSYNC)
-		syncs |= 0x00000010;
-
 	ret = nvd0_crtc_swap_fbs(crtc, old_fb);
 	if (ret)
 		return ret;
@@ -683,9 +677,6 @@ nvd0_crtc_mode_set(struct drm_crtc *crtc, struct drm_display_mode *umode,
 		evo_data(push, mode->clock * 1000);
 		evo_data(push, 0x00200000); /* ??? */
 		evo_data(push, mode->clock * 1000);
-		evo_mthd(push, 0x0404 + (nv_crtc->index * 0x300), 2);
-		evo_data(push, syncs);
-		evo_data(push, magic);
 		evo_mthd(push, 0x04d0 + (nv_crtc->index * 0x300), 2);
 		evo_data(push, 0x00000311);
 		evo_data(push, 0x00000100);
@@ -959,11 +950,6 @@ nvd0_dac_mode_fixup(struct drm_encoder *encoder, struct drm_display_mode *mode,
 }
 
 static void
-nvd0_dac_prepare(struct drm_encoder *encoder)
-{
-}
-
-static void
 nvd0_dac_commit(struct drm_encoder *encoder)
 {
 }
@@ -974,13 +960,26 @@ nvd0_dac_mode_set(struct drm_encoder *encoder, struct drm_display_mode *mode,
 {
 	struct nouveau_encoder *nv_encoder = nouveau_encoder(encoder);
 	struct nouveau_crtc *nv_crtc = nouveau_crtc(encoder->crtc);
-	u32 *push;
+	u32 syncs, magic, *push;
+
+	syncs = 0x00000001;
+	if (mode->flags & DRM_MODE_FLAG_NHSYNC)
+		syncs |= 0x00000008;
+	if (mode->flags & DRM_MODE_FLAG_NVSYNC)
+		syncs |= 0x00000010;
+
+	magic = 0x31ec6000 | (nv_crtc->index << 25);
+	if (mode->flags & DRM_MODE_FLAG_INTERLACE)
+		magic |= 0x00000001;
 
 	nvd0_dac_dpms(encoder, DRM_MODE_DPMS_ON);
 
-	push = evo_wait(encoder->dev, EVO_MASTER, 4);
+	push = evo_wait(encoder->dev, EVO_MASTER, 8);
 	if (push) {
-		evo_mthd(push, 0x0180 + (nv_encoder->or * 0x20), 2);
+		evo_mthd(push, 0x0404 + (nv_crtc->index * 0x300), 2);
+		evo_data(push, syncs);
+		evo_data(push, magic);
+		evo_mthd(push, 0x0180 + (nv_encoder->or * 0x020), 2);
 		evo_data(push, 1 << nv_crtc->index);
 		evo_data(push, 0x00ff);
 		evo_kick(push, encoder->dev, EVO_MASTER);
@@ -1043,7 +1042,7 @@ nvd0_dac_destroy(struct drm_encoder *encoder)
 static const struct drm_encoder_helper_funcs nvd0_dac_hfunc = {
 	.dpms = nvd0_dac_dpms,
 	.mode_fixup = nvd0_dac_mode_fixup,
-	.prepare = nvd0_dac_prepare,
+	.prepare = nvd0_dac_disconnect,
 	.commit = nvd0_dac_commit,
 	.mode_set = nvd0_dac_mode_set,
 	.disable = nvd0_dac_disconnect,
@@ -1183,6 +1182,143 @@ nvd0_hdmi_disconnect(struct drm_encoder *encoder)
 /******************************************************************************
  * SOR
  *****************************************************************************/
+static inline u32
+nvd0_sor_dp_lane_map(struct drm_device *dev, struct dcb_entry *dcb, u8 lane)
+{
+	static const u8 nvd0[] = { 16, 8, 0, 24 };
+	return nvd0[lane];
+}
+
+static void
+nvd0_sor_dp_train_set(struct drm_device *dev, struct dcb_entry *dcb, u8 pattern)
+{
+	const u32 or = ffs(dcb->or) - 1, link = !(dcb->sorconf.link & 1);
+	const u32 loff = (or * 0x800) + (link * 0x80);
+	nv_mask(dev, 0x61c110 + loff, 0x0f0f0f0f, 0x01010101 * pattern);
+}
+
+static void
+nvd0_sor_dp_train_adj(struct drm_device *dev, struct dcb_entry *dcb,
+		      u8 lane, u8 swing, u8 preem)
+{
+	const u32 or = ffs(dcb->or) - 1, link = !(dcb->sorconf.link & 1);
+	const u32 loff = (or * 0x800) + (link * 0x80);
+	u32 shift = nvd0_sor_dp_lane_map(dev, dcb, lane);
+	u32 mask = 0x000000ff << shift;
+	u8 *table, *entry, *config = NULL;
+
+	switch (swing) {
+	case 0: preem += 0; break;
+	case 1: preem += 4; break;
+	case 2: preem += 7; break;
+	case 3: preem += 9; break;
+	}
+
+	table = nouveau_dp_bios_data(dev, dcb, &entry);
+	if (table) {
+		if (table[0] == 0x30) {
+			config  = entry + table[4];
+			config += table[5] * preem;
+		}
+	}
+
+	if (!config) {
+		NV_ERROR(dev, "PDISP: unsupported DP table for chipset\n");
+		return;
+	}
+
+	nv_mask(dev, 0x61c118 + loff, mask, config[1] << shift);
+	nv_mask(dev, 0x61c120 + loff, mask, config[2] << shift);
+	nv_mask(dev, 0x61c130 + loff, 0x0000ff00, config[3] << 8);
+	nv_mask(dev, 0x61c13c + loff, 0x00000000, 0x00000000);
+}
+
+static void
+nvd0_sor_dp_link_set(struct drm_device *dev, struct dcb_entry *dcb, int crtc,
+		     int link_nr, u32 link_bw, bool enhframe)
+{
+	const u32 or = ffs(dcb->or) - 1, link = !(dcb->sorconf.link & 1);
+	const u32 loff = (or * 0x800) + (link * 0x80);
+	const u32 soff = (or * 0x800);
+	u32 dpctrl = nv_rd32(dev, 0x61c10c + loff) & ~0x001f4000;
+	u32 clksor = nv_rd32(dev, 0x612300 + soff) & ~0x007c0000;
+	u32 script = 0x0000, lane_mask = 0;
+	u8 *table, *entry;
+	int i;
+
+	link_bw /= 27000;
+
+	table = nouveau_dp_bios_data(dev, dcb, &entry);
+	if (table) {
+		if      (table[0] == 0x30) entry = ROMPTR(dev, entry[10]);
+		else                       entry = NULL;
+
+		while (entry) {
+			if (entry[0] >= link_bw)
+				break;
+			entry += 3;
+		}
+
+		nouveau_bios_run_init_table(dev, script, dcb, crtc);
+	}
+
+	clksor |= link_bw << 18;
+	dpctrl |= ((1 << link_nr) - 1) << 16;
+	if (enhframe)
+		dpctrl |= 0x00004000;
+
+	for (i = 0; i < link_nr; i++)
+		lane_mask |= 1 << (nvd0_sor_dp_lane_map(dev, dcb, i) >> 3);
+
+	nv_wr32(dev, 0x612300 + soff, clksor);
+	nv_wr32(dev, 0x61c10c + loff, dpctrl);
+	nv_mask(dev, 0x61c130 + loff, 0x0000000f, lane_mask);
+}
+
+static void
+nvd0_sor_dp_link_get(struct drm_device *dev, struct dcb_entry *dcb,
+		     u32 *link_nr, u32 *link_bw)
+{
+	const u32 or = ffs(dcb->or) - 1, link = !(dcb->sorconf.link & 1);
+	const u32 loff = (or * 0x800) + (link * 0x80);
+	const u32 soff = (or * 0x800);
+	u32 dpctrl = nv_rd32(dev, 0x61c10c + loff) & 0x000f0000;
+	u32 clksor = nv_rd32(dev, 0x612300 + soff);
+
+	if      (dpctrl > 0x00030000) *link_nr = 4;
+	else if (dpctrl > 0x00010000) *link_nr = 2;
+	else			      *link_nr = 1;
+
+	*link_bw  = (clksor & 0x007c0000) >> 18;
+	*link_bw *= 27000;
+}
+
+static void
+nvd0_sor_dp_calc_tu(struct drm_device *dev, struct dcb_entry *dcb,
+		    u32 crtc, u32 datarate)
+{
+	const u32 symbol = 100000;
+	const u32 TU = 64;
+	u32 link_nr, link_bw;
+	u64 ratio, value;
+
+	nvd0_sor_dp_link_get(dev, dcb, &link_nr, &link_bw);
+
+	ratio  = datarate;
+	ratio *= symbol;
+	do_div(ratio, link_nr * link_bw);
+
+	value  = (symbol - ratio) * TU;
+	value *= ratio;
+	do_div(value, symbol);
+	do_div(value, symbol);
+
+	value += 5;
+	value |= 0x08000000;
+
+	nv_wr32(dev, 0x616610 + (crtc * 0x800), value);
+}
+
 static void
 nvd0_sor_dpms(struct drm_encoder *encoder, int mode)
 {
@@ -1215,6 +1351,16 @@ nvd0_sor_dpms(struct drm_encoder *encoder, int mode)
 	nv_mask(dev, 0x61c004 + (or * 0x0800), 0x80000001, dpms_ctrl);
 	nv_wait(dev, 0x61c004 + (or * 0x0800), 0x80000000, 0x00000000);
 	nv_wait(dev, 0x61c030 + (or * 0x0800), 0x10000000, 0x00000000);
+
+	if (nv_encoder->dcb->type == OUTPUT_DP) {
+		struct dp_train_func func = {
+			.link_set = nvd0_sor_dp_link_set,
+			.train_set = nvd0_sor_dp_train_set,
+			.train_adj = nvd0_sor_dp_train_adj
+		};
+
+		nouveau_dp_dpms(encoder, mode, nv_encoder->dp.datarate, &func);
+	}
 }
 
 static bool
@@ -1237,8 +1383,37 @@ nvd0_sor_mode_fixup(struct drm_encoder *encoder, struct drm_display_mode *mode,
 }
 
 static void
+nvd0_sor_disconnect(struct drm_encoder *encoder)
+{
+	struct nouveau_encoder *nv_encoder = nouveau_encoder(encoder);
+	struct drm_device *dev = encoder->dev;
+	u32 *push;
+
+	if (nv_encoder->crtc) {
+		nvd0_crtc_prepare(nv_encoder->crtc);
+
+		push = evo_wait(dev, EVO_MASTER, 4);
+		if (push) {
+			evo_mthd(push, 0x0200 + (nv_encoder->or * 0x20), 1);
+			evo_data(push, 0x00000000);
+			evo_mthd(push, 0x0080, 1);
+			evo_data(push, 0x00000000);
+			evo_kick(push, dev, EVO_MASTER);
+		}
+
+		nvd0_hdmi_disconnect(encoder);
+
+		nv_encoder->crtc = NULL;
+		nv_encoder->last_dpms = DRM_MODE_DPMS_OFF;
+	}
+}
+
+static void
 nvd0_sor_prepare(struct drm_encoder *encoder)
 {
+	nvd0_sor_disconnect(encoder);
+	if (nouveau_encoder(encoder)->dcb->type == OUTPUT_DP)
+		evo_sync(encoder->dev, EVO_MASTER);
 }
 
 static void
@@ -1257,7 +1432,18 @@ nvd0_sor_mode_set(struct drm_encoder *encoder, struct drm_display_mode *umode,
 	struct nouveau_connector *nv_connector;
 	struct nvbios *bios = &dev_priv->vbios;
 	u32 mode_ctrl = (1 << nv_crtc->index);
-	u32 *push, or_config;
+	u32 syncs, magic, *push;
+	u32 or_config;
+
+	syncs = 0x00000001;
+	if (mode->flags & DRM_MODE_FLAG_NHSYNC)
+		syncs |= 0x00000008;
+	if (mode->flags & DRM_MODE_FLAG_NVSYNC)
+		syncs |= 0x00000010;
+
+	magic = 0x31ec6000 | (nv_crtc->index << 25);
+	if (mode->flags & DRM_MODE_FLAG_INTERLACE)
+		magic |= 0x00000001;
 
 	nv_connector = nouveau_encoder_connector_get(nv_encoder);
 	switch (nv_encoder->dcb->type) {
@@ -1306,6 +1492,22 @@ nvd0_sor_mode_set(struct drm_encoder *encoder, struct drm_display_mode *umode,
 
 		}
 		break;
+	case OUTPUT_DP:
+		if (nv_connector->base.display_info.bpc == 6) {
+			nv_encoder->dp.datarate = mode->clock * 18 / 8;
+			syncs |= 0x00000140;
+		} else {
+			nv_encoder->dp.datarate = mode->clock * 24 / 8;
+			syncs |= 0x00000180;
+		}
+
+		if (nv_encoder->dcb->sorconf.link & 1)
+			mode_ctrl |= 0x00000800;
+		else
+			mode_ctrl |= 0x00000900;
+
+		or_config = (mode_ctrl & 0x00000f00) >> 8;
+		break;
 	default:
 		BUG_ON(1);
 		break;
@@ -1313,9 +1515,17 @@ nvd0_sor_mode_set(struct drm_encoder *encoder, struct drm_display_mode *umode,
 
 	nvd0_sor_dpms(encoder, DRM_MODE_DPMS_ON);
 
-	push = evo_wait(dev, EVO_MASTER, 4);
+	if (nv_encoder->dcb->type == OUTPUT_DP) {
+		nvd0_sor_dp_calc_tu(dev, nv_encoder->dcb, nv_crtc->index,
+					 nv_encoder->dp.datarate);
+	}
+
+	push = evo_wait(dev, EVO_MASTER, 8);
 	if (push) {
-		evo_mthd(push, 0x0200 + (nv_encoder->or * 0x20), 2);
+		evo_mthd(push, 0x0404 + (nv_crtc->index * 0x300), 2);
+		evo_data(push, syncs);
+		evo_data(push, magic);
+		evo_mthd(push, 0x0200 + (nv_encoder->or * 0x020), 2);
 		evo_data(push, mode_ctrl);
 		evo_data(push, or_config);
 		evo_kick(push, dev, EVO_MASTER);
@@ -1325,32 +1535,6 @@ nvd0_sor_mode_set(struct drm_encoder *encoder, struct drm_display_mode *umode,
 }
 
 static void
-nvd0_sor_disconnect(struct drm_encoder *encoder)
-{
-	struct nouveau_encoder *nv_encoder = nouveau_encoder(encoder);
-	struct drm_device *dev = encoder->dev;
-	u32 *push;
-
-	if (nv_encoder->crtc) {
-		nvd0_crtc_prepare(nv_encoder->crtc);
-
-		push = evo_wait(dev, EVO_MASTER, 4);
-		if (push) {
-			evo_mthd(push, 0x0200 + (nv_encoder->or * 0x20), 1);
-			evo_data(push, 0x00000000);
-			evo_mthd(push, 0x0080, 1);
-			evo_data(push, 0x00000000);
-			evo_kick(push, dev, EVO_MASTER);
-		}
-
-		nvd0_hdmi_disconnect(encoder);
-
-		nv_encoder->crtc = NULL;
-		nv_encoder->last_dpms = DRM_MODE_DPMS_OFF;
-	}
-}
-
-static void
 nvd0_sor_destroy(struct drm_encoder *encoder)
 {
 	drm_encoder_cleanup(encoder);
@@ -1402,17 +1586,19 @@ static struct dcb_entry *
 lookup_dcb(struct drm_device *dev, int id, u32 mc)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	int type, or, i;
+	int type, or, i, link = -1;
 
 	if (id < 4) {
 		type = OUTPUT_ANALOG;
 		or   = id;
 	} else {
 		switch (mc & 0x00000f00) {
-		case 0x00000000: type = OUTPUT_LVDS; break;
-		case 0x00000100: type = OUTPUT_TMDS; break;
-		case 0x00000200: type = OUTPUT_TMDS; break;
-		case 0x00000500: type = OUTPUT_TMDS; break;
+		case 0x00000000: link = 0; type = OUTPUT_LVDS; break;
+		case 0x00000100: link = 0; type = OUTPUT_TMDS; break;
+		case 0x00000200: link = 1; type = OUTPUT_TMDS; break;
+		case 0x00000500: link = 0; type = OUTPUT_TMDS; break;
+		case 0x00000800: link = 0; type = OUTPUT_DP; break;
+		case 0x00000900: link = 1; type = OUTPUT_DP; break;
 		default:
 			NV_ERROR(dev, "PDISP: unknown SOR mc 0x%08x\n", mc);
 			return NULL;
@@ -1423,7 +1609,8 @@ lookup_dcb(struct drm_device *dev, int id, u32 mc)
 
 	for (i = 0; i < dev_priv->vbios.dcb.entries; i++) {
 		struct dcb_entry *dcb = &dev_priv->vbios.dcb.entry[i];
-		if (dcb->type == type && (dcb->or & (1 << or)))
+		if (dcb->type == type && (dcb->or & (1 << or)) &&
+		    (link < 0 || link == !(dcb->sorconf.link & 1)))
 			return dcb;
 	}
 
@@ -1498,6 +1685,7 @@ nvd0_display_unk2_handler(struct drm_device *dev, u32 crtc, u32 mask)
 			break;
 		case OUTPUT_TMDS:
 		case OUTPUT_LVDS:
+		case OUTPUT_DP:
 			if (cfg & 0x00000100)
 				tmp = 0x00000101;
 			else
@@ -1548,7 +1736,7 @@ nvd0_display_bh(unsigned long data)
 {
 	struct drm_device *dev = (struct drm_device *)data;
 	struct nvd0_display *disp = nvd0_display(dev);
-	u32 mask, crtc;
+	u32 mask = 0, crtc = ~0;
 	int i;
 
 	if (drm_debug & (DRM_UT_DRIVER | DRM_UT_KMS)) {
@@ -1564,12 +1752,8 @@ nvd0_display_bh(unsigned long data)
 		}
 	}
 
-	mask = nv_rd32(dev, 0x6101d4);
-	crtc = 0;
-	if (!mask) {
-		mask = nv_rd32(dev, 0x6109d4);
-		crtc = 1;
-	}
+	while (!mask && ++crtc < dev->mode_config.num_crtc)
+		mask = nv_rd32(dev, 0x6101d4 + (crtc * 0x800));
 
 	if (disp->modeset & 0x00000001)
 		nvd0_display_unk1_handler(dev, crtc, mask);
@@ -1584,6 +1768,7 @@ nvd0_display_intr(struct drm_device *dev)
 {
 	struct nvd0_display *disp = nvd0_display(dev);
 	u32 intr = nv_rd32(dev, 0x610088);
+	int i;
 
 	if (intr & 0x00000001) {
 		u32 stat = nv_rd32(dev, 0x61008c);
@@ -1628,16 +1813,13 @@ nvd0_display_intr(struct drm_device *dev)
 		intr &= ~0x00100000;
 	}
 
-	if (intr & 0x01000000) {
-		u32 stat = nv_rd32(dev, 0x6100bc);
-		nv_wr32(dev, 0x6100bc, stat);
-		intr &= ~0x01000000;
-	}
-
-	if (intr & 0x02000000) {
-		u32 stat = nv_rd32(dev, 0x6108bc);
-		nv_wr32(dev, 0x6108bc, stat);
-		intr &= ~0x02000000;
+	for (i = 0; i < dev->mode_config.num_crtc; i++) {
+		u32 mask = 0x01000000 << i;
+		if (intr & mask) {
+			u32 stat = nv_rd32(dev, 0x6100bc + (i * 0x800));
+			nv_wr32(dev, 0x6100bc + (i * 0x800), stat);
+			intr &= ~mask;
+		}
 	}
 
 	if (intr)
@@ -1774,7 +1956,7 @@ nvd0_display_create(struct drm_device *dev)
 	struct pci_dev *pdev = dev->pdev;
 	struct nvd0_display *disp;
 	struct dcb_entry *dcbe;
-	int ret, i;
+	int crtcs, ret, i;
 
 	disp = kzalloc(sizeof(*disp), GFP_KERNEL);
 	if (!disp)
@@ -1782,7 +1964,8 @@ nvd0_display_create(struct drm_device *dev)
 	dev_priv->engine.display.priv = disp;
 
 	/* create crtc objects to represent the hw heads */
-	for (i = 0; i < 2; i++) {
+	crtcs = nv_rd32(dev, 0x022448);
+	for (i = 0; i < crtcs; i++) {
 		ret = nvd0_crtc_create(dev, i);
 		if (ret)
 			goto out;
@@ -1803,6 +1986,7 @@ nvd0_display_create(struct drm_device *dev)
 		switch (dcbe->type) {
 		case OUTPUT_TMDS:
 		case OUTPUT_LVDS:
+		case OUTPUT_DP:
 			nvd0_sor_create(connector, dcbe);
 			break;
 		case OUTPUT_ANALOG: