Merge branch 'master' of git+ssh://git.freedesktop.org/git/mesa/drm into modesetting-101

Conflicts: linux-core/drm_drv.c linux-core/drm_fops.c linux-core/drm_objects.h linux-core/drm_stub.c shared-core/i915_dma.c
author: Alan Hourihane <alanh@fairlite.demon.co.uk> 2007-06-29 20:09:44 +0100
committer: Alan Hourihane <alanh@fairlite.demon.co.uk> 2007-06-29 20:09:44 +0100
commit: 8a78dead291ffdb5a8774419cdca369a1e27cad9 (patch)
tree: 399b8378e32036bdc7dadc10c639465a8fb21681 /shared-core
parent: e79e2a58161d44754fd55507e155b7e12a09c4d2 (diff)
parent: a27af4c4a665864df09123f177ca7269e48f6171 (diff)
36 files changed, 2446 insertions, 827 deletions
diff --git a/shared-core/drm_pciids.txt b/shared-core/drm_pciids.txt
index ad67dda2..126974d0 100644
--- a/shared-core/drm_pciids.txt
+++ b/shared-core/drm_pciids.txt
@@ -98,13 +98,20 @@
 0x1002 0x5653 CHIP_RV410|RADEON_IS_MOBILITY|RADEON_NEW_MEMMAP "ATI Radeon Mobility X700 M26"
 0x1002 0x5834 CHIP_RS300|RADEON_IS_IGP "ATI Radeon RS300 9100 IGP"
 0x1002 0x5835 CHIP_RS300|RADEON_IS_IGP|RADEON_IS_MOBILITY "ATI Radeon RS300 Mobility IGP"
+0x1002 0x5954 CHIP_RS400|RADEON_IS_IGP|RADEON_IS_MOBILITY|RADEON_IS_IGPGART "ATI RS480 XPRESS 200G"
 0x1002 0x5955 CHIP_RS400|RADEON_IS_IGP|RADEON_IS_MOBILITY|RADEON_IS_IGPGART "ATI Radeon XPRESS 200M 5955"
+0x1002 0x5974 CHIP_RS400|RADEON_IS_IGP|RADEON_IS_MOBILITY|RADEON_IS_IGPGART "ATI Radeon RS482 XPRESS 200"
+0x1002 0x5975 CHIP_RS400|RADEON_IS_IGP|RADEON_IS_MOBILITY|RADEON_IS_IGPGART "ATI Radeon RS485 XPRESS 1100 IGP"
 0x1002 0x5960 CHIP_RV280 "ATI Radeon RV280 9250"
 0x1002 0x5961 CHIP_RV280 "ATI Radeon RV280 9200"
 0x1002 0x5962 CHIP_RV280 "ATI Radeon RV280 9200"
 0x1002 0x5964 CHIP_RV280 "ATI Radeon RV280 9200 SE"
 0x1002 0x5965 CHIP_RV280 "ATI FireMV 2200 PCI"
 0x1002 0x5969 CHIP_RV100 "ATI ES1000 RN50"
+0x1002 0x5a41 CHIP_RS400|RADEON_IS_IGP|RADEON_IS_MOBILITY|RADEON_IS_IGPGART "ATI Radeon RS400 XPRESS 200"
+0x1002 0x5a42 CHIP_RS400|RADEON_IS_IGP|RADEON_IS_MOBILITY|RADEON_IS_IGPGART "ATI Radeon RS400 XPRESS 200M"
+0x1002 0x5a61 CHIP_RS400|RADEON_IS_IGP|RADEON_IS_MOBILITY|RADEON_IS_IGPGART "ATI Radeon RC410 XPRESS 200"
+0x1002 0x5a62 CHIP_RS400|RADEON_IS_IGP|RADEON_IS_MOBILITY|RADEON_IS_IGPGART "ATI Radeon RC410 XPRESS 200M"
 0x1002 0x5b60 CHIP_RV380|RADEON_NEW_MEMMAP "ATI Radeon RV370 X300 SE"
 0x1002 0x5b62 CHIP_RV380|RADEON_NEW_MEMMAP "ATI Radeon RV370 X600 Pro"
 0x1002 0x5b63 CHIP_RV380|RADEON_NEW_MEMMAP "ATI Radeon RV370 X550"
@@ -204,6 +211,8 @@
 0x1039 0x6300 0 "SiS 630"
 0x1039 0x6330 SIS_CHIP_315 "SiS 661"
 0x1039 0x7300 0 "SiS 730"
+0x18CA 0x0040 SIS_CHIP_315 "Volari V3XT/V5/V8"
+0x18CA 0x0042 SIS_CHIP_315 "Volari Unknown"
 
 [tdfx]
 0x121a 0x0003 0 "3dfx Voodoo Banshee"
@@ -275,11 +284,16 @@
 0x8086 0x2592 CHIP_I9XX|CHIP_I915 "Intel i915GM"
 0x8086 0x2772 CHIP_I9XX|CHIP_I915 "Intel i945G"
 0x8086 0x27A2 CHIP_I9XX|CHIP_I915 "Intel i945GM"
+0x8086 0x27AE CHIP_I9XX|CHIP_I915 "Intel i945GME"
 0x8086 0x2972 CHIP_I9XX|CHIP_I965 "Intel i946GZ"
 0x8086 0x2982 CHIP_I9XX|CHIP_I965 "Intel i965G"
 0x8086 0x2992 CHIP_I9XX|CHIP_I965 "Intel i965Q"
 0x8086 0x29A2 CHIP_I9XX|CHIP_I965 "Intel i965G"
 0x8086 0x2A02 CHIP_I9XX|CHIP_I965 "Intel i965GM"
+0x8086 0x2A12 CHIP_I9XX|CHIP_I965 "Intel i965GME/GLE"
+0x8086 0x29C2 CHIP_I9XX|CHIP_I915 "Intel G33"
+0x8086 0x29B2 CHIP_I9XX|CHIP_I915 "Intel Q35"
+0x8086 0x29D2 CHIP_I9XX|CHIP_I915 "Intel Q33"
 
 [imagine]
 0x105d 0x2309 IMAGINE_128 "Imagine 128"
@@ -582,6 +596,9 @@
 0x10de 0x018d NV_17 "GeForce4 448 Go"
 0x10de 0x0191 NV_50 "GeForce 8800 GTX"
 0x10de 0x0193 NV_50 "GeForce 8800 GTS"
+0x10de 0x0194 NV_50 "GeForce 8800 Ultra"
+0x10de 0x019d NV_50 "Quadro FX 5600"
+0x10de 0x019e NV_50 "Quadro FX 4600"
 0x10de 0x01a0 NV_11|NV_NFORCE "GeForce2 MX Integrated Graphics"
 0x10de 0x01d1 NV_44 "GeForce 7300 LE"
 0x10de 0x01d6 NV_44 "GeForce Go 7200"
@@ -688,6 +705,11 @@
 0x10de 0x03d1 NV_44 "GeForce 6100 nForce 405"
 0x10de 0x03d2 NV_44 "GeForce 6100 nForce 400"
 0x10de 0x03d5 NV_44 "GeForce 6100 nForce 420"
+0x10de 0x0400 NV_50 "GeForce 8600 GTS"
+0x10de 0x0402 NV_50 "GeForce 8600 GT"
+0x10de 0x0421 NV_50 "GeForce 8500 GT"
+0x10de 0x0422 NV_50 "GeForce 8400 GS"
+0x10de 0x0423 NV_50 "GeForce 8300 GS"
 0x12d2 0x0008 NV_03 "NV1"
 0x12d2 0x0009 NV_03 "DAC64"
 0x12d2 0x0018 NV_03 "Riva128"
diff --git a/shared-core/i915_dma.c b/shared-core/i915_dma.c
index a5e51de4..c95f1857 100644
--- a/shared-core/i915_dma.c
+++ b/shared-core/i915_dma.c
@@ -78,19 +78,15 @@ void i915_kernel_lost_context(drm_device_t * dev)
 		dev_priv->sarea_priv->perf_boxes |= I915_BOX_RING_EMPTY;
 }
 
-int i915_dma_cleanup(drm_device_t * dev)
+static int i915_dma_cleanup(drm_device_t * dev)
 {
-	drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private;
 	/* Make sure interrupts are disabled here because the uninstall ioctl
 	 * may not have been called from userspace and after dev_private
 	 * is freed, it's too late.
 	 */
-	I915_WRITE(LP_RING + RING_LEN, 0);
-
 	if (dev->irq)
 		drm_irq_uninstall(dev);
 
-
 	return 0;
 }
 
@@ -98,6 +94,8 @@ static int i915_initialize(drm_device_t * dev,
 			   drm_i915_private_t * dev_priv,
 			   drm_i915_init_t * init)
 {
+	memset(dev_priv, 0, sizeof(drm_i915_private_t));
+
 	dev_priv->sarea = drm_getsarea(dev);
 	if (!dev_priv->sarea) {
 		DRM_ERROR("can not find sarea!\n");
@@ -106,11 +104,13 @@ static int i915_initialize(drm_device_t * dev,
 		return DRM_ERR(EINVAL);
 	}
 
-	memset((void *)(dev_priv->ring.virtual_start), 0, dev_priv->ring.Size);
-
-	I915_WRITE(LP_RING + RING_START, dev_priv->ring.Start);
-	I915_WRITE(LP_RING + RING_LEN, ((dev_priv->ring.Size - 4096) & RING_NR_PAGES) | (RING_NO_REPORT | RING_VALID));
-
+	dev_priv->mmio_map = drm_core_findmap(dev, init->mmio_offset);
+	if (!dev_priv->mmio_map) {
+		dev->dev_private = (void *)dev_priv;
+		i915_dma_cleanup(dev);
+		DRM_ERROR("can not find mmio map!\n");
+		return DRM_ERR(EINVAL);
+	}
 
 	dev_priv->cpp = init->cpp;
 	dev_priv->sarea_priv->pf_current_page = 0;
@@ -128,24 +128,6 @@ static int i915_initialize(drm_device_t * dev,
 	 */
     	dev_priv->vblank_pipe = DRM_I915_VBLANK_PIPE_A;
 
-	/* Program Hardware Status Page */
-	dev_priv->status_page_dmah = drm_pci_alloc(dev, PAGE_SIZE, PAGE_SIZE, 
-	    0xffffffff);
-
-	if (!dev_priv->status_page_dmah) {
-		dev->dev_private = (void *)dev_priv;
-		i915_dma_cleanup(dev);
-		DRM_ERROR("Can not allocate hardware status page\n");
-		return DRM_ERR(ENOMEM);
-	}
-	dev_priv->hw_status_page = dev_priv->status_page_dmah->vaddr;
-	dev_priv->dma_status_page = dev_priv->status_page_dmah->busaddr;
-	
-	memset(dev_priv->hw_status_page, 0, PAGE_SIZE);
-	DRM_DEBUG("hw status page @ %p\n", dev_priv->hw_status_page);
-
-	I915_WRITE(0x02080, dev_priv->dma_status_page);
-	DRM_DEBUG("Enabled hardware status page\n");
 	dev->dev_private = (void *)dev_priv;
 	return 0;
 }
@@ -156,8 +138,15 @@ static int i915_dma_resume(drm_device_t * dev)
 
 	DRM_DEBUG("%s\n", __FUNCTION__);
 
-	I915_WRITE(I915REG_HWS_PGA, dev_priv->dma_status_page);
-	DRM_DEBUG("Enabled hardware status page\n");
+	if (!dev_priv->sarea) {
+		DRM_ERROR("can not find sarea!\n");
+		return DRM_ERR(EINVAL);
+	}
+
+	if (!dev_priv->mmio_map) {
+		DRM_ERROR("can not find mmio map!\n");
+		return DRM_ERR(EINVAL);
+	}
 
 	return 0;
 }
@@ -353,10 +342,12 @@ void i915_emit_breadcrumb(drm_device_t *dev)
 	drm_i915_private_t *dev_priv = dev->dev_private;
 	RING_LOCALS;
 
-	dev_priv->sarea_priv->last_enqueue = ++dev_priv->counter;
+	if (++dev_priv->counter > BREADCRUMB_MASK) {
+		 dev_priv->counter = 1;
+		 DRM_DEBUG("Breadcrumb counter wrapped around\n");
+	}
 
-	if (dev_priv->counter > 0x7FFFFFFFUL)
-		 dev_priv->sarea_priv->last_enqueue = dev_priv->counter = 1;
+	dev_priv->sarea_priv->last_enqueue = dev_priv->counter;
 
 	BEGIN_LP_RING(4);
 	OUT_RING(CMD_STORE_DWORD_IDX);
@@ -391,7 +382,9 @@ int i915_emit_mi_flush(drm_device_t *dev, uint32_t flush)
 static int i915_dispatch_cmdbuffer(drm_device_t * dev,
 				   drm_i915_cmdbuffer_t * cmd)
 {
+#ifdef I915_HAVE_FENCE
 	drm_i915_private_t *dev_priv = dev->dev_private;
+#endif
 	int nbox = cmd->num_cliprects;
 	int i = 0, count, ret;
 
@@ -776,7 +769,7 @@ static int i915_mmio(DRM_IOCTL_ARGS)
 		return DRM_ERR(EINVAL);
 
 	e = &mmio_table[mmio.reg];
-	base = dev_priv->mmio_map->handle + e->offset;
+	base = (u8 *) dev_priv->mmio_map->handle + e->offset;
 
         switch (mmio.read_write) {
 		case I915_MMIO_READ:
@@ -802,6 +795,47 @@ static int i915_mmio(DRM_IOCTL_ARGS)
 	return 0;
 }
 
+static int i915_set_status_page(DRM_IOCTL_ARGS)
+{
+	DRM_DEVICE;
+	drm_i915_private_t *dev_priv = dev->dev_private;
+	drm_i915_hws_addr_t hws;
+
+	if (!dev_priv) {
+		DRM_ERROR("%s called with no initialization\n", __FUNCTION__);
+		return DRM_ERR(EINVAL);
+	}
+	DRM_COPY_FROM_USER_IOCTL(hws, (drm_i915_hws_addr_t __user *) data,
+			sizeof(hws));
+	printk(KERN_DEBUG "set status page addr 0x%08x\n", (u32)hws.addr);
+
+	dev_priv->status_gfx_addr = hws.addr & (0x1ffff<<12);
+
+	dev_priv->hws_map.offset = dev->agp->agp_info.aper_base + hws.addr;
+	dev_priv->hws_map.size = 4*1024;
+	dev_priv->hws_map.type = 0;
+	dev_priv->hws_map.flags = 0;
+	dev_priv->hws_map.mtrr = 0;
+
+	drm_core_ioremap(&dev_priv->hws_map, dev);
+	if (dev_priv->hws_map.handle == NULL) {
+		dev->dev_private = (void *)dev_priv;
+		i915_dma_cleanup(dev);
+		dev_priv->status_gfx_addr = 0;
+		DRM_ERROR("can not ioremap virtual address for"
+				" G33 hw status page\n");
+		return DRM_ERR(ENOMEM);
+	}
+	dev_priv->hw_status_page = dev_priv->hws_map.handle;
+
+	memset(dev_priv->hw_status_page, 0, PAGE_SIZE);
+	I915_WRITE(I915REG_HWS_PGA, dev_priv->status_gfx_addr);
+	DRM_DEBUG("load hws 0x2080 with gfx mem 0x%x\n",
+			dev_priv->status_gfx_addr);
+	DRM_DEBUG("load hws at %p\n", dev_priv->hw_status_page);
+	return 0;
+}
+
 drm_ioctl_desc_t i915_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_I915_INIT)] = {i915_dma_init, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY},
 	[DRM_IOCTL_NR(DRM_I915_FLUSH)] = {i915_flush_ioctl, DRM_AUTH},
@@ -820,6 +854,7 @@ drm_ioctl_desc_t i915_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_I915_GET_VBLANK_PIPE)] = { i915_vblank_pipe_get, DRM_AUTH },
 	[DRM_IOCTL_NR(DRM_I915_VBLANK_SWAP)] = {i915_vblank_swap, DRM_AUTH},
 	[DRM_IOCTL_NR(DRM_I915_MMIO)] = {i915_mmio, DRM_AUTH},
+	[DRM_IOCTL_NR(DRM_I915_HWS_ADDR)] = {i915_set_status_page, DRM_AUTH},
 };
 
 int i915_max_ioctl = DRM_ARRAY_SIZE(i915_ioctls);
diff --git a/shared-core/i915_drm.h b/shared-core/i915_drm.h
index 1f32313e..1c6ff4d3 100644
--- a/shared-core/i915_drm.h
+++ b/shared-core/i915_drm.h
@@ -159,6 +159,7 @@ typedef struct _drm_i915_sarea {
 #define DRM_I915_GET_VBLANK_PIPE	0x0e
 #define DRM_I915_VBLANK_SWAP	0x0f
 #define DRM_I915_MMIO		0x10
+#define DRM_I915_HWS_ADDR	0x11
 
 #define DRM_IOCTL_I915_INIT		DRM_IOW( DRM_COMMAND_BASE + DRM_I915_INIT, drm_i915_init_t)
 #define DRM_IOCTL_I915_FLUSH		DRM_IO ( DRM_COMMAND_BASE + DRM_I915_FLUSH)
@@ -314,4 +315,8 @@ typedef struct drm_i915_mmio {
 	void __user *data;	
 } drm_i915_mmio_t;
 
+typedef struct drm_i915_hws_addr {
+	uint64_t addr;
+} drm_i915_hws_addr_t;
+
 #endif				/* _I915_DRM_H_ */
diff --git a/shared-core/i915_drv.h b/shared-core/i915_drv.h
index b0aa5df1..35451feb 100644
--- a/shared-core/i915_drv.h
+++ b/shared-core/i915_drv.h
@@ -103,6 +103,8 @@ typedef struct drm_i915_private {
 	void *hw_status_page;
 	dma_addr_t dma_status_page;
 	uint32_t counter;
+	unsigned int status_gfx_addr;
+	drm_local_map_t hws_map;
 
 	unsigned int cpp;
 	int use_mi_batchbuffer_start;
@@ -526,6 +528,9 @@ extern int i915_wait_ring(drm_device_t * dev, int n, const char *caller);
 
 #define CMD_OP_DESTBUFFER_INFO	 ((0x3<<29)|(0x1d<<24)|(0x8e<<16)|1)
 
+#define BREADCRUMB_BITS 31
+#define BREADCRUMB_MASK ((1U << BREADCRUMB_BITS) - 1)
+
 #define READ_BREADCRUMB(dev_priv)  (((volatile u32*)(dev_priv->hw_status_page))[5])
 #define READ_HWSP(dev_priv, reg)  (((volatile u32*)(dev_priv->hw_status_page))[reg])
 
@@ -1012,12 +1017,18 @@ extern int i915_wait_ring(drm_device_t * dev, int n, const char *caller);
 #define IS_I965G(dev) ((dev)->pci_device == 0x2972 || \
 		       (dev)->pci_device == 0x2982 || \
 		       (dev)->pci_device == 0x2992 || \
-		       (dev)->pci_device == 0x29A2)
+		       (dev)->pci_device == 0x29A2 || \
+		       (dev)->pci_device == 0x2A02 || \
+		       (dev)->pci_device == 0x2A12)
 
 
 #define IS_I9XX(dev) (IS_I915G(dev) || IS_I915GM(dev) || IS_I945G(dev) || \
 		      IS_I945GM(dev) || IS_I965G(dev))
 
+#define IS_G33(dev)    ((dev)->pci_device == 0x29C2 || \
+		   	(dev)->pci_device == 0x29B2 || \
+			(dev)->pci_device == 0x29D2) 
+
 #define IS_MOBILE(dev) (IS_I830(dev) || IS_I85X(dev) || IS_I915GM(dev) || \
 			IS_I945GM(dev))
 
diff --git a/shared-core/i915_init.c b/shared-core/i915_init.c
index 7516b2ce..83219e49 100644
--- a/shared-core/i915_init.c
+++ b/shared-core/i915_init.c
@@ -228,22 +228,23 @@ int i915_driver_load(drm_device_t *dev, unsigned long flags)
 	dev_priv->allow_batchbuffer = 1;
 
 	/* Program Hardware Status Page */
-	dev_priv->status_page_dmah = drm_pci_alloc(dev, PAGE_SIZE, PAGE_SIZE, 
-	    0xffffffff);
-
-	if (!dev_priv->status_page_dmah) {
-		dev->dev_private = (void *)dev_priv;
-		i915_dma_cleanup(dev);
-		DRM_ERROR("Can not allocate hardware status page\n");
-		return DRM_ERR(ENOMEM);
+	if (!IS_G33(dev)) {
+		dev_priv->status_page_dmah = 
+			drm_pci_alloc(dev, PAGE_SIZE, PAGE_SIZE, 0xffffffff);
+
+		if (!dev_priv->status_page_dmah) {
+			dev->dev_private = (void *)dev_priv;
+			i915_dma_cleanup(dev);
+			DRM_ERROR("Can not allocate hardware status page\n");
+			return DRM_ERR(ENOMEM);
+		}
+		dev_priv->hw_status_page = dev_priv->status_page_dmah->vaddr;
+		dev_priv->dma_status_page = dev_priv->status_page_dmah->busaddr;
+
+		memset(dev_priv->hw_status_page, 0, PAGE_SIZE);
+
+		I915_WRITE(I915REG_HWS_PGA, dev_priv->dma_status_page);
 	}
-	dev_priv->hw_status_page = dev_priv->status_page_dmah->vaddr;
-	dev_priv->dma_status_page = dev_priv->status_page_dmah->busaddr;
-	
-	memset(dev_priv->hw_status_page, 0, PAGE_SIZE);
-	DRM_DEBUG("hw status page @ %p\n", dev_priv->hw_status_page);
-
-	I915_WRITE(I915REG_HWS_PGA, dev_priv->dma_status_page);
 	DRM_DEBUG("Enabled hardware status page\n");
 
 	intel_modeset_init(dev);
diff --git a/shared-core/i915_irq.c b/shared-core/i915_irq.c
index 2d4df76e..98f51b2b 100644
--- a/shared-core/i915_irq.c
+++ b/shared-core/i915_irq.c
@@ -381,13 +381,6 @@ void i915_user_irq_off(drm_i915_private_t *dev_priv)
 	spin_unlock(&dev_priv->user_irq_lock);
 }
 		
-static int wait_compare(struct drm_device *dev, void *priv)
-{
-	drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private;
-	int irq_nr = (u64)priv;
-
-	return (READ_BREADCRUMB(dev_priv) >= irq_nr);
-}
 
 static int i915_wait_irq(drm_device_t * dev, int irq_nr)
 {
@@ -403,8 +396,8 @@ static int i915_wait_irq(drm_device_t * dev, int irq_nr)
 	dev_priv->sarea_priv->perf_boxes |= I915_BOX_WAIT;
 	
 	i915_user_irq_on(dev_priv);
-	ret = drm_wait_on(dev, &dev_priv->irq_queue, 3 * DRM_HZ, wait_compare,
-			  (void *)(u64)irq_nr);
+	DRM_WAIT_ON(ret, dev_priv->irq_queue, 3 * DRM_HZ,
+		    READ_BREADCRUMB(dev_priv) >= irq_nr);
 	i915_user_irq_off(dev_priv);
 
 	if (ret == DRM_ERR(EBUSY)) {
@@ -735,11 +728,7 @@ void i915_driver_irq_postinstall(drm_device_t * dev)
 {
 	drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private;
 
-	dev_priv->swaps_lock = SPIN_LOCK_UNLOCKED;
-	INIT_LIST_HEAD(&dev_priv->vbl_swaps.head);
-	dev_priv->swaps_pending = 0;
-
-	dev_priv->swaps_lock = SPIN_LOCK_UNLOCKED;
+	spin_lock_init(&dev_priv->swaps_lock);
 	INIT_LIST_HEAD(&dev_priv->vbl_swaps.head);
 	dev_priv->swaps_pending = 0;
 
diff --git a/shared-core/mga_dma.c b/shared-core/mga_dma.c
index 78448282..d48313c7 100644
--- a/shared-core/mga_dma.c
+++ b/shared-core/mga_dma.c
@@ -551,7 +551,7 @@ static int mga_do_agp_dma_bootstrap(drm_device_t * dev,
 		drm_map_list_t *_entry;
 		unsigned long agp_token = 0;
 
-		list_for_each_entry(_entry, &dev->maplist->head, head) {
+		list_for_each_entry(_entry, &dev->maplist, head) {
 			if (_entry->map == dev->agp_buffer_map)
 				agp_token = _entry->user_token;
 		}
diff --git a/shared-core/nouveau_drm.h b/shared-core/nouveau_drm.h
index 1e7322e0..0758991a 100644
--- a/shared-core/nouveau_drm.h
+++ b/shared-core/nouveau_drm.h
@@ -25,9 +25,12 @@
 #ifndef __NOUVEAU_DRM_H__
 #define __NOUVEAU_DRM_H__
 
-#define NOUVEAU_DRM_HEADER_PATCHLEVEL 6
+#define NOUVEAU_DRM_HEADER_PATCHLEVEL 7
 
 typedef struct drm_nouveau_fifo_alloc {
+	uint32_t     fb_ctxdma_handle;
+	uint32_t     tt_ctxdma_handle;
+
 	int          channel;
 	uint32_t     put_base;
 	/* FIFO control regs */
@@ -36,29 +39,30 @@ typedef struct drm_nouveau_fifo_alloc {
 	/* DMA command buffer */
 	drm_handle_t cmdbuf;
 	int          cmdbuf_size;
+	/* Notifier memory */
+	drm_handle_t notifier;
+	int          notifier_size;
 }
 drm_nouveau_fifo_alloc_t;
 
-typedef struct drm_nouveau_object_init {
+typedef struct drm_nouveau_grobj_alloc {
 	int      channel;
 	uint32_t handle;
 	int      class;
 }
-drm_nouveau_object_init_t;
+drm_nouveau_grobj_alloc_t;
 
 #define NOUVEAU_MEM_ACCESS_RO	1
 #define NOUVEAU_MEM_ACCESS_WO	2
 #define NOUVEAU_MEM_ACCESS_RW	3
-typedef struct drm_nouveau_dma_object_init {
+typedef struct drm_nouveau_notifier_alloc {
 	int      channel;
 	uint32_t handle;
-	int      class;
-	int      access;
-	int      target;
+	int      count;
+
 	uint32_t offset;
-	int      size;
 }
-drm_nouveau_dma_object_init_t;
+drm_nouveau_notifier_alloc_t;
 
 #define NOUVEAU_MEM_FB			0x00000001
 #define NOUVEAU_MEM_AGP			0x00000002
@@ -68,7 +72,7 @@ drm_nouveau_dma_object_init_t;
 #define NOUVEAU_MEM_USER_BACKED		0x00000020
 #define NOUVEAU_MEM_MAPPED		0x00000040
 #define NOUVEAU_MEM_INSTANCE		0x00000080 /* internal */
-
+#define NOUVEAU_MEM_NOTIFIER            0x00000100 /* internal */
 typedef struct drm_nouveau_mem_alloc {
 	int flags;
 	int alignment;
@@ -141,8 +145,8 @@ typedef struct drm_nouveau_sarea {
 drm_nouveau_sarea_t;
 
 #define DRM_NOUVEAU_FIFO_ALLOC      0x00
-#define DRM_NOUVEAU_OBJECT_INIT     0x01
-#define DRM_NOUVEAU_DMA_OBJECT_INIT 0x02
+#define DRM_NOUVEAU_GROBJ_ALLOC     0x01
+#define DRM_NOUVEAU_NOTIFIER_ALLOC  0x02
 #define DRM_NOUVEAU_MEM_ALLOC       0x03
 #define DRM_NOUVEAU_MEM_FREE        0x04
 #define DRM_NOUVEAU_GETPARAM        0x05
diff --git a/shared-core/nouveau_drv.h b/shared-core/nouveau_drv.h
index debee8e4..3cca07fc 100644
--- a/shared-core/nouveau_drv.h
+++ b/shared-core/nouveau_drv.h
@@ -34,7 +34,7 @@
 
 #define DRIVER_MAJOR		0
 #define DRIVER_MINOR		0
-#define DRIVER_PATCHLEVEL	6
+#define DRIVER_PATCHLEVEL	7
 
 #define NOUVEAU_FAMILY   0x0000FFFF
 #define NOUVEAU_FLAGS    0xFFFF0000
@@ -64,7 +64,6 @@ struct nouveau_object
 	int channel;
 
 	struct mem_block *instance;
-	uint32_t          ht_loc;
 
 	uint32_t handle;
 	int      class;
@@ -83,6 +82,11 @@ struct nouveau_fifo
 	/* dma object for the command buffer itself */
 	struct mem_block      *cmdbuf_mem;
 	struct nouveau_object *cmdbuf_obj;
+	uint32_t pushbuf_base;
+	/* notifier memory */
+	struct mem_block *notifier_block;
+	struct mem_block *notifier_heap;
+	drm_local_map_t  *notifier_map;
 	/* PGRAPH context, for cards that keep it in RAMIN */
 	struct mem_block *ramin_grctx;
 	/* objects belonging to this fifo */
@@ -99,32 +103,42 @@ struct nouveau_config {
 	} cmdbuf;
 };
 
-struct nouveau_engine_func {
+typedef struct nouveau_engine_func {
 	struct {
-		int	(*Init)(drm_device_t *dev);
-		void	(*Takedown)(drm_device_t *dev);
-	} Mc;
+		int	(*init)(drm_device_t *dev);
+		void	(*takedown)(drm_device_t *dev);
+	} mc;
 
 	struct {
-		int	(*Init)(drm_device_t *dev);
-		void	(*Takedown)(drm_device_t *dev);
-	} Timer;
+		int	(*init)(drm_device_t *dev);
+		void	(*takedown)(drm_device_t *dev);
+	} timer;
 
 	struct {
-		int	(*Init)(drm_device_t *dev);
-		void	(*Takedown)(drm_device_t *dev);
-	} Fb;
+		int	(*init)(drm_device_t *dev);
+		void	(*takedown)(drm_device_t *dev);
+	} fb;
 
 	struct {
-		int	(*Init)(drm_device_t *dev);
-		void	(*Takedown)(drm_device_t *dev);
-	} Graph;
+		int	(*init)(drm_device_t *);
+		void	(*takedown)(drm_device_t *);
+
+		int	(*create_context)(drm_device_t *, int channel);
+		void	(*destroy_context)(drm_device_t *, int channel);
+		int	(*load_context)(drm_device_t *, int channel);
+		int	(*save_context)(drm_device_t *, int channel);
+	} graph;
 
 	struct {
-		int	(*Init)(drm_device_t *dev);
-		void	(*Takedown)(drm_device_t *dev);
-	} Fifo;
-};
+		int	(*init)(drm_device_t *);
+		void	(*takedown)(drm_device_t *);
+
+		int	(*create_context)(drm_device_t *, int channel);
+		void	(*destroy_context)(drm_device_t *, int channel);
+		int	(*load_context)(drm_device_t *, int channel);
+		int	(*save_context)(drm_device_t *, int channel);
+	} fifo;
+} nouveau_engine_func_t;
 
 typedef struct drm_nouveau_private {
 	/* the card type, takes NV_* as values */
@@ -186,6 +200,12 @@ extern void nouveau_wait_for_idle(struct drm_device *dev);
 extern int nouveau_ioctl_card_init(DRM_IOCTL_ARGS);
 
 /* nouveau_mem.c */
+extern int               nouveau_mem_init_heap(struct mem_block **,
+					       uint64_t start, uint64_t size);
+extern struct mem_block *nouveau_mem_alloc_block(struct mem_block *,
+						 uint64_t size, int align2,
+						 DRMFILE);
+extern void              nouveau_mem_free_block(struct mem_block *);
 extern uint64_t          nouveau_mem_fb_amount(struct drm_device *dev);
 extern void              nouveau_mem_release(DRMFILE filp, struct mem_block *heap);
 extern int               nouveau_ioctl_mem_alloc(DRM_IOCTL_ARGS);
@@ -199,11 +219,13 @@ extern struct mem_block* nouveau_instmem_alloc(struct drm_device *dev,
 					       uint32_t size, uint32_t align);
 extern void              nouveau_instmem_free(struct drm_device *dev,
 					      struct mem_block *block);
-extern uint32_t          nouveau_instmem_r32(drm_nouveau_private_t *dev_priv,
-					     struct mem_block *mem, int index);
-extern void              nouveau_instmem_w32(drm_nouveau_private_t *dev_priv,
-					     struct mem_block *mem, int index,
-					     uint32_t val);
+
+/* nouveau_notifier.c */
+extern int  nouveau_notifier_init_channel(drm_device_t *, int channel, DRMFILE);
+extern void nouveau_notifier_takedown_channel(drm_device_t *, int channel);
+extern int  nouveau_notifier_alloc(drm_device_t *, int channel,
+				   uint32_t handle, int cout, uint32_t *offset);
+extern int  nouveau_ioctl_notifier_alloc(DRM_IOCTL_ARGS);
 
 /* nouveau_fifo.c */
 extern int  nouveau_fifo_init(drm_device_t *dev);
@@ -214,7 +236,13 @@ extern int  nouveau_fifo_owner(drm_device_t *dev, DRMFILE filp, int channel);
 extern void nouveau_fifo_free(drm_device_t *dev, int channel);
 
 /* nouveau_object.c */
+extern int  nouveau_object_init_channel(drm_device_t *, int channel,
+					uint32_t vram_handle,
+					uint32_t tt_handle);
+extern void nouveau_object_takedown_channel(drm_device_t *dev, int channel);
 extern void nouveau_object_cleanup(drm_device_t *dev, int channel);
+extern int  nouveau_ramht_insert(drm_device_t *, int channel,
+				 uint32_t handle, struct nouveau_object *);
 extern struct nouveau_object *
 nouveau_object_gr_create(drm_device_t *dev, int channel, int class);
 extern struct nouveau_object *
@@ -222,8 +250,7 @@ nouveau_object_dma_create(drm_device_t *dev, int channel, int class,
 			  uint32_t offset, uint32_t size,
 			  int access, int target);
 extern void nouveau_object_free(drm_device_t *dev, struct nouveau_object *obj);
-extern int  nouveau_ioctl_object_init(DRM_IOCTL_ARGS);
-extern int  nouveau_ioctl_dma_object_init(DRM_IOCTL_ARGS);
+extern int  nouveau_ioctl_grobj_alloc(DRM_IOCTL_ARGS);
 extern uint32_t nouveau_chip_instance_get(drm_device_t *dev, struct mem_block *mem);
 
 /* nouveau_irq.c */
@@ -244,35 +271,82 @@ extern void nv10_fb_takedown(drm_device_t *dev);
 extern int  nv40_fb_init(drm_device_t *dev);
 extern void nv40_fb_takedown(drm_device_t *dev);
 
+/* nv04_fifo.c */
+extern int  nv04_fifo_create_context(drm_device_t *dev, int channel);
+extern void nv04_fifo_destroy_context(drm_device_t *dev, int channel);
+extern int  nv04_fifo_load_context(drm_device_t *dev, int channel);
+extern int  nv04_fifo_save_context(drm_device_t *dev, int channel);
+
+/* nv10_fifo.c */
+extern int  nv10_fifo_create_context(drm_device_t *dev, int channel);
+extern void nv10_fifo_destroy_context(drm_device_t *dev, int channel);
+extern int  nv10_fifo_load_context(drm_device_t *dev, int channel);
+extern int  nv10_fifo_save_context(drm_device_t *dev, int channel);
+
+/* nv40_fifo.c */
+extern int  nv40_fifo_create_context(drm_device_t *, int channel);
+extern void nv40_fifo_destroy_context(drm_device_t *, int channel);
+extern int  nv40_fifo_load_context(drm_device_t *, int channel);
+extern int  nv40_fifo_save_context(drm_device_t *, int channel);
+
+/* nv50_fifo.c */
+extern int  nv50_fifo_init(drm_device_t *);
+extern void nv50_fifo_takedown(drm_device_t *);
+extern int  nv50_fifo_create_context(drm_device_t *, int channel);
+extern void nv50_fifo_destroy_context(drm_device_t *, int channel);
+extern int  nv50_fifo_load_context(drm_device_t *, int channel);
+extern int  nv50_fifo_save_context(drm_device_t *, int channel);
+
 /* nv04_graph.c */
 extern void nouveau_nv04_context_switch(drm_device_t *dev);
-extern int nv04_graph_init(drm_device_t *dev);
+extern int  nv04_graph_init(drm_device_t *dev);
 extern void nv04_graph_takedown(drm_device_t *dev);
-extern int nv04_graph_context_create(drm_device_t *dev, int channel);
+extern int  nv04_graph_create_context(drm_device_t *dev, int channel);
+extern void nv04_graph_destroy_context(drm_device_t *dev, int channel);
+extern int  nv04_graph_load_context(drm_device_t *dev, int channel);
+extern int  nv04_graph_save_context(drm_device_t *dev, int channel);
 
 /* nv10_graph.c */
 extern void nouveau_nv10_context_switch(drm_device_t *dev);
-extern int nv10_graph_init(drm_device_t *dev);
+extern int  nv10_graph_init(drm_device_t *dev);
 extern void nv10_graph_takedown(drm_device_t *dev);
-extern int nv10_graph_context_create(drm_device_t *dev, int channel);
+extern int  nv10_graph_create_context(drm_device_t *dev, int channel);
+extern void nv10_graph_destroy_context(drm_device_t *dev, int channel);
+extern int  nv10_graph_load_context(drm_device_t *dev, int channel);
+extern int  nv10_graph_save_context(drm_device_t *dev, int channel);
 
 /* nv20_graph.c */
 extern void nouveau_nv20_context_switch(drm_device_t *dev);
-extern int nv20_graph_init(drm_device_t *dev);
+extern int  nv20_graph_init(drm_device_t *dev);
 extern void nv20_graph_takedown(drm_device_t *dev);
-extern int nv20_graph_context_create(drm_device_t *dev, int channel);
+extern int  nv20_graph_create_context(drm_device_t *dev, int channel);
+extern void nv20_graph_destroy_context(drm_device_t *dev, int channel);
+extern int  nv20_graph_load_context(drm_device_t *dev, int channel);
+extern int  nv20_graph_save_context(drm_device_t *dev, int channel);
 
 /* nv30_graph.c */
-extern int nv30_graph_init(drm_device_t *dev);
+extern int  nv30_graph_init(drm_device_t *dev);
 extern void nv30_graph_takedown(drm_device_t *dev);
-extern int nv30_graph_context_create(drm_device_t *dev, int channel);
+extern int  nv30_graph_create_context(drm_device_t *, int channel);
+extern void nv30_graph_destroy_context(drm_device_t *, int channel);
+extern int  nv30_graph_load_context(drm_device_t *, int channel);
+extern int  nv30_graph_save_context(drm_device_t *, int channel);
 
 /* nv40_graph.c */
-extern int  nv40_graph_init(drm_device_t *dev);
-extern void nv40_graph_takedown(drm_device_t *dev);
-extern int  nv40_graph_context_create(drm_device_t *dev, int channel);
-extern void nv40_graph_context_save_current(drm_device_t *dev);
-extern void nv40_graph_context_restore(drm_device_t *dev, int channel);
+extern int  nv40_graph_init(drm_device_t *);
+extern void nv40_graph_takedown(drm_device_t *);
+extern int  nv40_graph_create_context(drm_device_t *, int channel);
+extern void nv40_graph_destroy_context(drm_device_t *, int channel);
+extern int  nv40_graph_load_context(drm_device_t *, int channel);
+extern int  nv40_graph_save_context(drm_device_t *, int channel);
+
+/* nv50_graph.c */
+extern int  nv50_graph_init(drm_device_t *);
+extern void nv50_graph_takedown(drm_device_t *);
+extern int  nv50_graph_create_context(drm_device_t *, int channel);
+extern void nv50_graph_destroy_context(drm_device_t *, int channel);
+extern int  nv50_graph_load_context(drm_device_t *, int channel);
+extern int  nv50_graph_save_context(drm_device_t *, int channel);
 
 /* nv04_mc.c */
 extern int  nv04_mc_init(drm_device_t *dev);
@@ -282,6 +356,10 @@ extern void nv04_mc_takedown(drm_device_t *dev);
 extern int  nv40_mc_init(drm_device_t *dev);
 extern void nv40_mc_takedown(drm_device_t *dev);
 
+/* nv50_mc.c */
+extern int  nv50_mc_init(drm_device_t *dev);
+extern void nv50_mc_takedown(drm_device_t *dev);
+
 /* nv04_timer.c */
 extern int  nv04_timer_init(drm_device_t *dev);
 extern void nv04_timer_takedown(drm_device_t *dev);
@@ -297,8 +375,17 @@ extern long nouveau_compat_ioctl(struct file *filp, unsigned int cmd,
 #define NV_WRITE(reg,val)   DRM_WRITE32( dev_priv->mmio, (reg), (val) )
 #endif
 
-#define INSTANCE_WR(mem,ofs,val) nouveau_instmem_w32(dev_priv,(mem),(ofs),(val))
-#define INSTANCE_RD(mem,ofs)     nouveau_instmem_r32(dev_priv,(mem),(ofs))
+/* PRAMIN access */
+#if defined(__powerpc__)
+#define NV_RI32(o) in_be32((void __iomem *)(dev_priv->ramin)->handle+(o))
+#define NV_WI32(o,v) out_be32((void __iomem*)(dev_priv->ramin)->handle+(o), (v))
+#else
+#define NV_RI32(o) DRM_READ32(dev_priv->ramin, (o))
+#define NV_WI32(o,v) DRM_WRITE32(dev_priv->ramin, (o), (v))
+#endif
+
+#define INSTANCE_RD(o,i) NV_RI32((o)->start + ((i)<<2))
+#define INSTANCE_WR(o,i,v) NV_WI32((o)->start + ((i)<<2), (v))
 
 #endif /* __NOUVEAU_DRV_H__ */
 
diff --git a/shared-core/nouveau_fifo.c b/shared-core/nouveau_fifo.c
index 8e66ca2e..81dbfcda 100644
--- a/shared-core/nouveau_fifo.c
+++ b/shared-core/nouveau_fifo.c
@@ -83,6 +83,8 @@ static int nouveau_fifo_instmem_configure(drm_device_t *dev)
 		case NV_50:
 		case NV_40:
 			NV_WRITE(NV40_PFIFO_RAMFC, 0x30002);
+			if((dev_priv->chipset == 0x49) || (dev_priv->chipset == 0x4b))
+				NV_WRITE(0x2230,0x00000001);
 			break;
 		case NV_44:
 			NV_WRITE(NV40_PFIFO_RAMFC, ((nouveau_mem_fb_amount(dev)-512*1024+dev_priv->ramfc_offset)>>16) |
@@ -232,235 +234,20 @@ nouveau_fifo_cmdbuf_alloc(struct drm_device *dev, int channel)
 		return DRM_ERR(ENOMEM);
 	}
 
+	dev_priv->fifos[channel].pushbuf_base = 0;
 	dev_priv->fifos[channel].cmdbuf_mem = cb;
 	dev_priv->fifos[channel].cmdbuf_obj = cb_dma;
 	return 0;
 }
 
-#define RAMFC_WR(offset, val) NV_WRITE(fifoctx + NV04_RAMFC_##offset, (val))
-static void nouveau_nv04_context_init(drm_device_t *dev, int channel)
-{
-        drm_nouveau_private_t *dev_priv = dev->dev_private;
-	struct nouveau_object *cb_obj;
-	uint32_t fifoctx, ctx_size = 32;
-	int i;
-
-	cb_obj = dev_priv->fifos[channel].cmdbuf_obj;
-
-	fifoctx=NV_RAMIN+dev_priv->ramfc_offset+channel*ctx_size;
-	
-        // clear the fifo context
-        for(i=0;i<ctx_size/4;i++)
-                NV_WRITE(fifoctx+4*i,0x0);
-
-	RAMFC_WR(DMA_INSTANCE	, nouveau_chip_instance_get(dev, cb_obj->instance));
-
-        RAMFC_WR(DMA_FETCH,	NV_PFIFO_CACHE1_DMA_FETCH_TRIG_112_BYTES |
-                                NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
-                                NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_4     |
-#ifdef __BIG_ENDIAN
-                                NV_PFIFO_CACHE1_BIG_ENDIAN |
-#endif
-				0x00000000);
-}
-#undef RAMFC_WR
-
-#define RAMFC_WR(offset, val) NV_WRITE(fifoctx + NV10_RAMFC_##offset, (val))
-static void nouveau_nv10_context_init(drm_device_t *dev, int channel)
-{
-        drm_nouveau_private_t *dev_priv = dev->dev_private;
-        struct nouveau_object *cb_obj;
-        uint32_t fifoctx;
-        int ctx_size = nouveau_fifo_ctx_size(dev);
-        int i;
-        cb_obj  = dev_priv->fifos[channel].cmdbuf_obj;
-        fifoctx = NV_RAMIN + dev_priv->ramfc_offset + channel*ctx_size;
-
-        for (i=0;i<ctx_size;i+=4)
-                NV_WRITE(fifoctx + i, 0);
-
-        /* Fill entries that are seen filled in dumps of nvidia driver just
-         * after channel's is put into DMA mode
-         */
-
-        RAMFC_WR(DMA_INSTANCE  , nouveau_chip_instance_get(dev,
-                                cb_obj->instance));
-
-        RAMFC_WR(DMA_FETCH, NV_PFIFO_CACHE1_DMA_FETCH_TRIG_112_BYTES | 
-                        NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
-                        NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_4     |
-#ifdef __BIG_ENDIAN
-                        NV_PFIFO_CACHE1_BIG_ENDIAN |
-#endif
-			0x00000000);
-}
-
-static void nouveau_nv30_context_init(drm_device_t *dev, int channel)
-{
-        drm_nouveau_private_t *dev_priv = dev->dev_private;
-        struct nouveau_fifo *chan = &dev_priv->fifos[channel];
-	struct nouveau_object *cb_obj;
-	uint32_t fifoctx, grctx_inst, cb_inst, ctx_size = 64;
-	int i;
-
-	cb_obj = dev_priv->fifos[channel].cmdbuf_obj;
-        cb_inst = nouveau_chip_instance_get(dev, chan->cmdbuf_obj->instance);
-        grctx_inst = nouveau_chip_instance_get(dev, chan->ramin_grctx);
-        fifoctx = NV_RAMIN + dev_priv->ramfc_offset + channel * ctx_size;
-        
-        for (i = 0; i < ctx_size; i += 4)
-                NV_WRITE(fifoctx + i, 0);
-
-        RAMFC_WR(REF_CNT,       NV_READ(NV10_PFIFO_CACHE1_REF_CNT));
-        RAMFC_WR(DMA_INSTANCE,  cb_inst);
-        RAMFC_WR(DMA_STATE,     NV_READ(NV04_PFIFO_CACHE1_DMA_STATE));
-        RAMFC_WR(DMA_FETCH,     NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES | 
-                                NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
-                                NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8 |
-#ifdef __BIG_ENDIAN
-                                NV_PFIFO_CACHE1_BIG_ENDIAN |
-#endif
-                                0x00000000);
-        
-        RAMFC_WR(ENGINE,                NV_READ(NV04_PFIFO_CACHE1_ENGINE));
-        RAMFC_WR(PULL1_ENGINE,          NV_READ(NV04_PFIFO_CACHE1_PULL1)); 
-        RAMFC_WR(ACQUIRE_VALUE,         NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_VALUE));
-        RAMFC_WR(ACQUIRE_TIMESTAMP,     NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_TIMESTAMP));
-        RAMFC_WR(ACQUIRE_TIMEOUT,       NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_TIMEOUT));
-        RAMFC_WR(SEMAPHORE,             NV_READ(NV10_PFIFO_CACHE1_SEMAPHORE));
-}
-
-#if 0
-static void nouveau_nv10_context_save(drm_device_t *dev)
-{
-	drm_nouveau_private_t *dev_priv = dev->dev_private;
-	uint32_t fifoctx;
-	int channel;
-
-	channel = NV_READ(NV03_PFIFO_CACHE1_PUSH1) & (nouveau_fifo_number(dev)-1);
-	fifoctx = NV_RAMIN + dev_priv->ramfc_offset + channel*64;
-
-	RAMFC_WR(DMA_PUT          , NV_READ(NV04_PFIFO_CACHE1_DMA_PUT));
-	RAMFC_WR(DMA_GET          , NV_READ(NV04_PFIFO_CACHE1_DMA_GET));
-	RAMFC_WR(REF_CNT          , NV_READ(NV10_PFIFO_CACHE1_REF_CNT));
-	RAMFC_WR(DMA_INSTANCE     , NV_READ(NV04_PFIFO_CACHE1_DMA_INSTANCE));
-	RAMFC_WR(DMA_STATE        , NV_READ(NV04_PFIFO_CACHE1_DMA_STATE));
-	RAMFC_WR(DMA_FETCH        , NV_READ(NV04_PFIFO_CACHE1_DMA_FETCH));
-	RAMFC_WR(ENGINE           , NV_READ(NV04_PFIFO_CACHE1_ENGINE));
-	RAMFC_WR(PULL1_ENGINE     , NV_READ(NV04_PFIFO_CACHE1_PULL1));
-	RAMFC_WR(ACQUIRE_VALUE    , NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_VALUE));
-	RAMFC_WR(ACQUIRE_TIMESTAMP, NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_TIMESTAMP));
-	RAMFC_WR(ACQUIRE_TIMEOUT  , NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_TIMEOUT));
-	RAMFC_WR(SEMAPHORE        , NV_READ(NV10_PFIFO_CACHE1_SEMAPHORE));
-	RAMFC_WR(DMA_SUBROUTINE   , NV_READ(NV10_PFIFO_CACHE1_DMA_SUBROUTINE));
-}
-#endif
-#undef RAMFC_WR
-
-#define RAMFC_WR(offset, val) NV_WRITE(fifoctx + NV40_RAMFC_##offset, (val))
-static void nouveau_nv40_context_init(drm_device_t *dev, int channel)
-{
-	drm_nouveau_private_t *dev_priv = dev->dev_private;
-	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
-	uint32_t fifoctx, cb_inst, grctx_inst;
-	int i;
-
-	cb_inst = nouveau_chip_instance_get(dev, chan->cmdbuf_obj->instance);
-	grctx_inst = nouveau_chip_instance_get(dev, chan->ramin_grctx);
-	fifoctx = NV_RAMIN + dev_priv->ramfc_offset + channel*128;
-	for (i=0;i<128;i+=4)
-		NV_WRITE(fifoctx + i, 0);
-
-	/* Fill entries that are seen filled in dumps of nvidia driver just
-	 * after channel's is put into DMA mode
-	 */
-	RAMFC_WR(DMA_INSTANCE  , cb_inst);
-	RAMFC_WR(DMA_FETCH     , NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
-				 NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
-				 NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8 |
-#ifdef __BIG_ENDIAN
-				 NV_PFIFO_CACHE1_BIG_ENDIAN |
-#endif
-				 0x30000000 /* no idea.. */);
-	RAMFC_WR(GRCTX_INSTANCE, grctx_inst);
-	RAMFC_WR(DMA_TIMESLICE , 0x0001FFFF);
-}
-
-static void nouveau_nv40_context_save(drm_device_t *dev)
-{
-	drm_nouveau_private_t *dev_priv = dev->dev_private;
-	uint32_t fifoctx;
-	int channel;
-
-	channel = NV_READ(NV03_PFIFO_CACHE1_PUSH1) & (nouveau_fifo_number(dev)-1);
-	fifoctx = NV_RAMIN + dev_priv->ramfc_offset + channel*128;
-
-	RAMFC_WR(DMA_PUT          , NV_READ(NV04_PFIFO_CACHE1_DMA_PUT));
-	RAMFC_WR(DMA_GET          , NV_READ(NV04_PFIFO_CACHE1_DMA_GET));
-	RAMFC_WR(REF_CNT          , NV_READ(NV10_PFIFO_CACHE1_REF_CNT));
-	RAMFC_WR(DMA_INSTANCE     , NV_READ(NV04_PFIFO_CACHE1_DMA_INSTANCE));
-	RAMFC_WR(DMA_DCOUNT       , NV_READ(NV10_PFIFO_CACHE1_DMA_DCOUNT));
-	RAMFC_WR(DMA_STATE        , NV_READ(NV04_PFIFO_CACHE1_DMA_STATE));
-	RAMFC_WR(DMA_FETCH	  , NV_READ(NV04_PFIFO_CACHE1_DMA_FETCH));
-	RAMFC_WR(ENGINE           , NV_READ(NV04_PFIFO_CACHE1_ENGINE));
-	RAMFC_WR(PULL1_ENGINE     , NV_READ(NV04_PFIFO_CACHE1_PULL1));
-	RAMFC_WR(ACQUIRE_VALUE    , NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_VALUE));
-	RAMFC_WR(ACQUIRE_TIMESTAMP, NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_TIMESTAMP));
-	RAMFC_WR(ACQUIRE_TIMEOUT  , NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_TIMEOUT));
-	RAMFC_WR(SEMAPHORE        , NV_READ(NV10_PFIFO_CACHE1_SEMAPHORE));
-	RAMFC_WR(DMA_SUBROUTINE   , NV_READ(NV04_PFIFO_CACHE1_DMA_GET));
-	RAMFC_WR(GRCTX_INSTANCE   , NV_READ(NV40_PFIFO_GRCTX_INSTANCE));
-	RAMFC_WR(DMA_TIMESLICE    , NV_READ(NV04_PFIFO_DMA_TIMESLICE) & 0x1FFFF);
-	RAMFC_WR(UNK_40           , NV_READ(NV40_PFIFO_UNK32E4));
-}
-#undef RAMFC_WR
-
-/* This function should load values from RAMFC into PFIFO, but for now
- * it just clobbers PFIFO with what nouveau_fifo_alloc used to setup
- * unconditionally.
- */
-static void
-nouveau_fifo_context_restore(drm_device_t *dev, int channel)
-{
-	drm_nouveau_private_t *dev_priv = dev->dev_private;
-	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
-	uint32_t cb_inst;
-
-	cb_inst = nouveau_chip_instance_get(dev, chan->cmdbuf_obj->instance);
-
-	// FIXME check if we need to refill the time quota with something like NV_WRITE(0x204C, 0x0003FFFF);
-
-	if (dev_priv->card_type >= NV_40)
-		NV_WRITE(NV03_PFIFO_CACHE1_PUSH1, 0x00010000|channel);
-	else
-		NV_WRITE(NV03_PFIFO_CACHE1_PUSH1, 0x00000100|channel);
-
-	NV_WRITE(NV04_PFIFO_CACHE1_DMA_PUT, 0 /*RAMFC_DMA_PUT*/);
-	NV_WRITE(NV04_PFIFO_CACHE1_DMA_GET, 0 /*RAMFC_DMA_GET*/);
-	NV_WRITE(NV04_PFIFO_CACHE1_DMA_INSTANCE, cb_inst);
-	NV_WRITE(NV04_PFIFO_SIZE , 0x0000FFFF);
-	NV_WRITE(NV04_PFIFO_CACHE1_HASH, 0x0000FFFF);
-
-	NV_WRITE(NV04_PFIFO_CACHE0_PULL1, 0x00000001);
-	NV_WRITE(NV04_PFIFO_CACHE1_DMA_CTL, 0x00000000);
-	NV_WRITE(NV04_PFIFO_CACHE1_DMA_STATE, 0x00000000);
-	NV_WRITE(NV04_PFIFO_CACHE1_ENGINE, 0x00000000);
-
-	NV_WRITE(NV04_PFIFO_CACHE1_DMA_FETCH,	NV_PFIFO_CACHE1_DMA_FETCH_TRIG_112_BYTES |
-					NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
-					NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_4 |
-#ifdef __BIG_ENDIAN
-					NV_PFIFO_CACHE1_BIG_ENDIAN |
-#endif
-					0x00000000);
-}
-
 /* allocates and initializes a fifo for user space consumption */
-static int nouveau_fifo_alloc(drm_device_t* dev, int *chan_ret, DRMFILE filp)
+int nouveau_fifo_alloc(drm_device_t* dev, int *chan_ret, DRMFILE filp,
+		       uint32_t vram_handle, uint32_t tt_handle)
 {
 	int ret;
 	drm_nouveau_private_t *dev_priv = dev->dev_private;
-	struct nouveau_object *cb_obj;
+	nouveau_engine_func_t *engine = &dev_priv->Engine;
+	struct nouveau_fifo *chan;
 	int channel;
 
 	/*
@@ -478,14 +265,16 @@ static int nouveau_fifo_alloc(drm_device_t* dev, int *chan_ret, DRMFILE filp)
 	if (channel==nouveau_fifo_number(dev))
 		return DRM_ERR(EINVAL);
 	(*chan_ret) = channel;
+	chan = &dev_priv->fifos[channel];
 
 	DRM_INFO("Allocating FIFO number %d\n", channel);
 
 	/* that fifo is used */
-	dev_priv->fifos[channel].used = 1;
-	dev_priv->fifos[channel].filp = filp;
+	chan->used = 1;
+	chan->filp = filp;
+
 	/* FIFO has no objects yet */
-	dev_priv->fifos[channel].objs = NULL;
+	chan->objs = NULL;
 
 	/* allocate a command buffer, and create a dma object for the gpu */
 	ret = nouveau_fifo_cmdbuf_alloc(dev, channel);
@@ -493,7 +282,20 @@ static int nouveau_fifo_alloc(drm_device_t* dev, int *chan_ret, DRMFILE filp)
 		nouveau_fifo_free(dev, channel);
 		return ret;
 	}
-	cb_obj = dev_priv->fifos[channel].cmdbuf_obj;
+
+	/* Setup channel's default objects */
+	ret = nouveau_object_init_channel(dev, channel, vram_handle, tt_handle);
+	if (ret) {
+		nouveau_fifo_free(dev, channel);
+		return ret;
+	}
+
+	/* Allocate space for per-channel fixed notifier memory */
+	ret = nouveau_notifier_init_channel(dev, channel, filp);
+	if (ret) {
+		nouveau_fifo_free(dev, channel);
+		return ret;
+	}
 
 	nouveau_wait_for_idle(dev);
 
@@ -503,73 +305,54 @@ static int nouveau_fifo_alloc(drm_device_t* dev, int *chan_ret, DRMFILE filp)
 	NV_WRITE(NV03_PFIFO_CACHE1_PUSH0, 0x00000000);
 	NV_WRITE(NV04_PFIFO_CACHE1_PULL0, 0x00000000);
 
+	/* Create a graphics context for new channel */
+	ret = engine->graph.create_context(dev, channel);
+	if (ret) {
+		nouveau_fifo_free(dev, channel);
+		return ret;
+	}
+
 	/* Construct inital RAMFC for new channel */
-	switch(dev_priv->card_type)
-	{
-		case NV_04:
-		case NV_05:
-			nv04_graph_context_create(dev, channel);
-			nouveau_nv04_context_init(dev, channel);
-			break;
-		case NV_10:
-		case NV_17:
-			nv10_graph_context_create(dev, channel);
-			nouveau_nv10_context_init(dev, channel);
-			break;
-		case NV_20:
-			ret = nv20_graph_context_create(dev, channel);
-			if (ret) {
-				nouveau_fifo_free(dev, channel);
-				return ret;
-			}
-			nouveau_nv10_context_init(dev, channel);
-			break;
-		case NV_30:
-			ret = nv30_graph_context_create(dev, channel);
-			if (ret) {
-				nouveau_fifo_free(dev, channel);
-				return ret;
-			}
-			nouveau_nv30_context_init(dev, channel);
-			break;
-		case NV_40:
-		case NV_44:
-		case NV_50:
-			ret = nv40_graph_context_create(dev, channel);
-			if (ret) {
-				nouveau_fifo_free(dev, channel);
-				return ret;
-			}
-			nouveau_nv40_context_init(dev, channel);
-			break;
+	ret = engine->fifo.create_context(dev, channel);
+	if (ret) {
+		nouveau_fifo_free(dev, channel);
+		return ret;
 	}
 
 	/* enable the fifo dma operation */
 	NV_WRITE(NV04_PFIFO_MODE,NV_READ(NV04_PFIFO_MODE)|(1<<channel));
 
 	/* setup channel's default get/put values */
-	NV_WRITE(NV03_FIFO_REGS_DMAPUT(channel), 0);
-	NV_WRITE(NV03_FIFO_REGS_DMAGET(channel), 0);
+	NV_WRITE(NV03_FIFO_REGS_DMAPUT(channel), chan->pushbuf_base);
+	NV_WRITE(NV03_FIFO_REGS_DMAGET(channel), chan->pushbuf_base);
 
 	/* If this is the first channel, setup PFIFO ourselves.  For any
 	 * other case, the GPU will handle this when it switches contexts.
 	 */
 	if (dev_priv->fifo_alloc_count == 0) {
-		nouveau_fifo_context_restore(dev, channel);
-		if (dev_priv->card_type >= NV_30) {
-			struct nouveau_fifo *chan;
-			uint32_t inst;
-
-			chan = &dev_priv->fifos[channel];
-			inst = nouveau_chip_instance_get(dev,
-							 chan->ramin_grctx);
-
-			/* see comments in nv40_graph_context_restore() */
-			NV_WRITE(NV10_PGRAPH_CHANNEL_CTX_SIZE, inst);
-                        if (dev_priv->card_type >= NV_40) {
-                                NV_WRITE(0x40032C, inst | 0x01000000);
-                                NV_WRITE(NV40_PFIFO_GRCTX_INSTANCE, inst);
-                        }
+		ret = engine->fifo.load_context(dev, channel);
+		if (ret) {
+			nouveau_fifo_free(dev, channel);
+			return ret;
+		}
+
+		ret = engine->graph.load_context(dev, channel);
+		if (ret) {
+			nouveau_fifo_free(dev, channel);
+			return ret;
+		}
+
+		/* Temporary hack, to avoid breaking Xv on cards where the
+		 * initial context value for 0x400710 doesn't have these bits
+		 * set.  Proper fix would be to find which object+method is
+		 * responsible for modifying this state.
+		 */
+		if (dev_priv->chipset >= 0x10) {
+			uint32_t tmp;
+			tmp = NV_READ(NV10_PGRAPH_SURFACE) & 0x0007ff00;
+			NV_WRITE(NV10_PGRAPH_SURFACE, tmp);
+			tmp = NV_READ(NV10_PGRAPH_SURFACE) | 0x00020100;
+			NV_WRITE(NV10_PGRAPH_SURFACE, tmp);
 		}
 	}
 
@@ -591,9 +374,8 @@ static int nouveau_fifo_alloc(drm_device_t* dev, int *chan_ret, DRMFILE filp)
 void nouveau_fifo_free(drm_device_t* dev, int channel)
 {
 	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	nouveau_engine_func_t *engine = &dev_priv->Engine;
 	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
-	int i;
-	int ctx_size = nouveau_fifo_ctx_size(dev);
 
 	chan->used = 0;
 	DRM_INFO("%s: freeing fifo %d\n", __func__, channel);
@@ -604,25 +386,10 @@ void nouveau_fifo_free(drm_device_t* dev, int channel)
 	NV_WRITE(NV04_PFIFO_MODE, NV_READ(NV04_PFIFO_MODE)&~(1<<channel));
 	// FIXME XXX needs more code
 	
-	/* Clean RAMFC */
-	for (i=0;i<ctx_size;i+=4) {
-		DRM_DEBUG("RAMFC +%02x: 0x%08x\n", i, NV_READ(NV_RAMIN +
-					dev_priv->ramfc_offset + 
-					channel*ctx_size + i));
-		NV_WRITE(NV_RAMIN + dev_priv->ramfc_offset +
-				channel*ctx_size + i, 0);
-	}
+	engine->fifo.destroy_context(dev, channel);
 
 	/* Cleanup PGRAPH state */
-	if (dev_priv->card_type >= NV_40)
-		nouveau_instmem_free(dev, chan->ramin_grctx);
-	else if (dev_priv->card_type >= NV_30) {
-	}
-	else if (dev_priv->card_type >= NV_20) {
-		/* clear ctx table */
-		INSTANCE_WR(dev_priv->ctx_table, channel, 0);
-		nouveau_instmem_free(dev, chan->ramin_grctx);
-	}
+	engine->graph.destroy_context(dev, channel);
 
 	/* reenable the fifo caches */
 	NV_WRITE(NV03_PFIFO_CACHES, 0x00000001);
@@ -631,6 +398,8 @@ void nouveau_fifo_free(drm_device_t* dev, int channel)
 	if (chan->cmdbuf_mem)
 		nouveau_mem_free(dev, chan->cmdbuf_mem);
 
+	nouveau_notifier_takedown_channel(dev, channel);
+
 	/* Destroy objects belonging to the channel */
 	nouveau_object_cleanup(dev, channel);
 
@@ -669,31 +438,42 @@ static int nouveau_ioctl_fifo_alloc(DRM_IOCTL_ARGS)
 {
 	DRM_DEVICE;
 	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	struct nouveau_fifo *chan;
 	drm_nouveau_fifo_alloc_t init;
 	int res;
 
 	DRM_COPY_FROM_USER_IOCTL(init, (drm_nouveau_fifo_alloc_t __user *) data,
 				 sizeof(init));
 
-	res = nouveau_fifo_alloc(dev, &init.channel, filp);
+	res = nouveau_fifo_alloc(dev, &init.channel, filp,
+				 init.fb_ctxdma_handle,
+				 init.tt_ctxdma_handle);
 	if (res)
 		return res;
+	chan = &dev_priv->fifos[init.channel];
 
-	/* this should probably disappear in the next abi break? */
-	init.put_base = 0;
+	init.put_base = chan->pushbuf_base;
 
 	/* make the fifo available to user space */
 	/* first, the fifo control regs */
 	init.ctrl      = dev_priv->mmio->offset + NV03_FIFO_REGS(init.channel);
 	init.ctrl_size = NV03_FIFO_REGS_SIZE;
 	res = drm_addmap(dev, init.ctrl, init.ctrl_size, _DRM_REGISTERS,
-			 0, &dev_priv->fifos[init.channel].regs);
+			 0, &chan->regs);
 	if (res != 0)
 		return res;
 
 	/* pass back FIFO map info to the caller */
-	init.cmdbuf      = dev_priv->fifos[init.channel].cmdbuf_mem->start;
-	init.cmdbuf_size = dev_priv->fifos[init.channel].cmdbuf_mem->size;
+	init.cmdbuf      = chan->cmdbuf_mem->start;
+	init.cmdbuf_size = chan->cmdbuf_mem->size;
+
+	/* and the notifier block */
+	init.notifier      = chan->notifier_block->start;
+	init.notifier_size = chan->notifier_block->size;
+	res = drm_addmap(dev, init.notifier, init.notifier_size, _DRM_REGISTERS,
+			 0, &chan->notifier_map);
+	if (res != 0)
+		return res;
 
 	DRM_COPY_TO_USER_IOCTL((drm_nouveau_fifo_alloc_t __user *)data,
 			       init, sizeof(init));
@@ -706,8 +486,8 @@ static int nouveau_ioctl_fifo_alloc(DRM_IOCTL_ARGS)
 
 drm_ioctl_desc_t nouveau_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_NOUVEAU_FIFO_ALLOC)] = {nouveau_ioctl_fifo_alloc, DRM_AUTH},	
-	[DRM_IOCTL_NR(DRM_NOUVEAU_OBJECT_INIT)] = {nouveau_ioctl_object_init, DRM_AUTH},
-	[DRM_IOCTL_NR(DRM_NOUVEAU_DMA_OBJECT_INIT)] = {nouveau_ioctl_dma_object_init, DRM_AUTH},
+	[DRM_IOCTL_NR(DRM_NOUVEAU_GROBJ_ALLOC)] = {nouveau_ioctl_grobj_alloc, DRM_AUTH},
+	[DRM_IOCTL_NR(DRM_NOUVEAU_NOTIFIER_ALLOC)] = {nouveau_ioctl_notifier_alloc, DRM_AUTH},
 	[DRM_IOCTL_NR(DRM_NOUVEAU_MEM_ALLOC)] = {nouveau_ioctl_mem_alloc, DRM_AUTH},
 	[DRM_IOCTL_NR(DRM_NOUVEAU_MEM_FREE)] = {nouveau_ioctl_mem_free, DRM_AUTH},
 	[DRM_IOCTL_NR(DRM_NOUVEAU_GETPARAM)] = {nouveau_ioctl_getparam, DRM_AUTH},
diff --git a/shared-core/nouveau_mem.c b/shared-core/nouveau_mem.c
index a5343b99..d8ae52b7 100644
--- a/shared-core/nouveau_mem.c
+++ b/shared-core/nouveau_mem.c
@@ -77,8 +77,8 @@ out:
 	return p;
 }
 
-static struct mem_block *alloc_block(struct mem_block *heap, uint64_t size,
-		int align2, DRMFILE filp)
+struct mem_block *nouveau_mem_alloc_block(struct mem_block *heap, uint64_t size,
+					  int align2, DRMFILE filp)
 {
 	struct mem_block *p;
 	uint64_t mask = (1 << align2) - 1;
@@ -106,7 +106,7 @@ static struct mem_block *find_block(struct mem_block *heap, uint64_t start)
 	return NULL;
 }
 
-static void free_block(struct mem_block *p)
+void nouveau_mem_free_block(struct mem_block *p)
 {
 	p->filp = NULL;
 
@@ -132,7 +132,8 @@ static void free_block(struct mem_block *p)
 
 /* Initialize.  How to check for an uninitialized heap?
  */
-static int init_heap(struct mem_block **heap, uint64_t start, uint64_t size)
+int nouveau_mem_init_heap(struct mem_block **heap, uint64_t start,
+			  uint64_t size)
 {
 	struct mem_block *blocks = drm_alloc(sizeof(*blocks), DRM_MEM_BUFS);
 
@@ -331,7 +332,9 @@ int nouveau_mem_init(struct drm_device *dev)
 			goto no_agp;
 		}
 
-		if (init_heap(&dev_priv->agp_heap, info.aperture_base, info.aperture_size))
+		if (nouveau_mem_init_heap(&dev_priv->agp_heap,
+					  info.aperture_base,
+					  info.aperture_size))
 			goto no_agp;
 
 		dev_priv->agp_phys		= info.aperture_base;
@@ -357,12 +360,19 @@ no_agp:
 	if (fb_size>256*1024*1024) {
 		/* On cards with > 256Mb, you can't map everything. 
 		 * So we create a second FB heap for that type of memory */
-		if (init_heap(&dev_priv->fb_heap, drm_get_resource_start(dev,1), 256*1024*1024))
+		if (nouveau_mem_init_heap(&dev_priv->fb_heap,
+					  drm_get_resource_start(dev,1),
+					  256*1024*1024))
 			return DRM_ERR(ENOMEM);
-		if (init_heap(&dev_priv->fb_nomap_heap, drm_get_resource_start(dev,1)+256*1024*1024, fb_size-256*1024*1024))
+		if (nouveau_mem_init_heap(&dev_priv->fb_nomap_heap,
+					  drm_get_resource_start(dev,1) + 
+					  256*1024*1024,
+					  fb_size-256*1024*1024))
 			return DRM_ERR(ENOMEM);
 	} else {
-		if (init_heap(&dev_priv->fb_heap, drm_get_resource_start(dev,1), fb_size))
+		if (nouveau_mem_init_heap(&dev_priv->fb_heap,
+					  drm_get_resource_start(dev,1),
+					  fb_size))
 			return DRM_ERR(ENOMEM);
 		dev_priv->fb_nomap_heap=NULL;
 	}
@@ -397,21 +407,25 @@ struct mem_block* nouveau_mem_alloc(struct drm_device *dev, int alignment, uint6
 
 	if (flags&NOUVEAU_MEM_AGP) {
 		type=NOUVEAU_MEM_AGP;
-		block = alloc_block(dev_priv->agp_heap, size, alignment, filp);
+		block = nouveau_mem_alloc_block(dev_priv->agp_heap, size,
+						alignment, filp);
 		if (block) goto alloc_ok;
 	}
 	if (flags&(NOUVEAU_MEM_FB|NOUVEAU_MEM_FB_ACCEPTABLE)) {
 		type=NOUVEAU_MEM_FB;
 		if (!(flags&NOUVEAU_MEM_MAPPED)) {
-			block = alloc_block(dev_priv->fb_nomap_heap, size, alignment, filp);
+			block = nouveau_mem_alloc_block(dev_priv->fb_nomap_heap,
+							size, alignment, filp);
 			if (block) goto alloc_ok;
 		}
-		block = alloc_block(dev_priv->fb_heap, size, alignment, filp);
+		block = nouveau_mem_alloc_block(dev_priv->fb_heap, size,
+						alignment, filp);
 		if (block) goto alloc_ok;	
 	}
 	if (flags&NOUVEAU_MEM_AGP_ACCEPTABLE) {
 		type=NOUVEAU_MEM_AGP;
-		block = alloc_block(dev_priv->agp_heap, size, alignment, filp);
+		block = nouveau_mem_alloc_block(dev_priv->agp_heap, size,
+						alignment, filp);
 		if (block) goto alloc_ok;
 	}
 
@@ -432,7 +446,7 @@ alloc_ok:
 			ret = drm_addmap(dev, block->start, block->size,
 					_DRM_FRAME_BUFFER, 0, &block->map);
 		if (ret) { 
-			free_block(block);
+			nouveau_mem_free_block(block);
 			return NULL;
 		}
 	}
@@ -446,7 +460,7 @@ void nouveau_mem_free(struct drm_device* dev, struct mem_block* block)
 	DRM_INFO("freeing 0x%llx\n", block->start);
 	if (block->flags&NOUVEAU_MEM_MAPPED)
 		drm_rmmap(dev, block->map);
-	free_block(block);
+	nouveau_mem_free_block(block);
 }
 
 static void
@@ -475,13 +489,8 @@ nouveau_instmem_determine_amount(struct drm_device *dev)
 	DRM_DEBUG("RAMIN size: %dKiB\n", dev_priv->ramin_size>>10);
 
 	/* Clear all of it, except the BIOS image that's in the first 64KiB */
-	if (dev_priv->ramin) {
-		for (i=(64*1024); i<dev_priv->ramin_size; i+=4)
-			DRM_WRITE32(dev_priv->ramin, i, 0x00000000);
-	} else {
-		for (i=(64*1024); i<dev_priv->ramin_size; i+=4)
-			DRM_WRITE32(dev_priv->mmio, NV_RAMIN + i, 0x00000000);
-	}
+	for (i=(64*1024); i<dev_priv->ramin_size; i+=4)
+		NV_WI32(i, 0x00000000);
 }
 
 static void
@@ -549,8 +558,8 @@ int nouveau_instmem_init(struct drm_device *dev)
 	 * the space that was reserved for RAMHT/FC/RO.
 	 */
 	offset = dev_priv->ramfc_offset + dev_priv->ramfc_size;
-	ret = init_heap(&dev_priv->ramin_heap,
-			 offset, dev_priv->ramin_size - offset);
+	ret = nouveau_mem_init_heap(&dev_priv->ramin_heap,
+				    offset, dev_priv->ramin_size - offset);
 	if (ret) {
 		dev_priv->ramin_heap = NULL;
 		DRM_ERROR("Failed to init RAMIN heap\n");
@@ -570,7 +579,8 @@ struct mem_block *nouveau_instmem_alloc(struct drm_device *dev,
 		return NULL;
 	}
 
-	block = alloc_block(dev_priv->ramin_heap, size, align, (DRMFILE)-2);
+	block = nouveau_mem_alloc_block(dev_priv->ramin_heap, size, align,
+					(DRMFILE)-2);
 	if (block) {
 		block->flags = NOUVEAU_MEM_INSTANCE;
 		DRM_DEBUG("instance(size=%d, align=%d) alloc'd at 0x%08x\n",
@@ -583,39 +593,7 @@ struct mem_block *nouveau_instmem_alloc(struct drm_device *dev,
 void nouveau_instmem_free(struct drm_device *dev, struct mem_block *block)
 {
 	if (dev && block) {
-		free_block(block);
-	}
-}
-
-uint32_t nouveau_instmem_r32(drm_nouveau_private_t *dev_priv,
-			     struct mem_block *mem, int index)
-{
-	uint32_t ofs = (uint32_t)mem->start + (index<<2);
-
-	if (dev_priv->ramin) {
-#if defined(__powerpc__)
-		return in_be32((void __iomem *)(dev_priv->ramin)->handle + ofs);
-#else
-		return DRM_READ32(dev_priv->ramin, ofs);
-#endif
-	} else {
-		return NV_READ(NV_RAMIN+ofs);
-	}
-}
-
-void nouveau_instmem_w32(drm_nouveau_private_t *dev_priv,
-			 struct mem_block *mem, int index, uint32_t val)
-{
-	uint32_t ofs = (uint32_t)mem->start + (index<<2);
-
-	if (dev_priv->ramin) {
-#if defined(__powerpc__)
-		out_be32((void __iomem *)(dev_priv->ramin)->handle + ofs, val);
-#else
-		DRM_WRITE32(dev_priv->ramin, ofs, val);
-#endif
-	} else {
-		NV_WRITE(NV_RAMIN+ofs, val);
+		nouveau_mem_free_block(block);
 	}
 }
 
diff --git a/shared-core/nouveau_notifier.c b/shared-core/nouveau_notifier.c
new file mode 100644
index 00000000..0cfe733e
--- /dev/null
+++ b/shared-core/nouveau_notifier.c
@@ -0,0 +1,154 @@
+/*
+ * Copyright (C) 2007 Ben Skeggs.
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "drmP.h"
+#include "drm.h"
+#include "nouveau_drv.h"
+
+int
+nouveau_notifier_init_channel(drm_device_t *dev, int channel, DRMFILE filp)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
+	int flags, ret;
+
+	/*TODO: PCI notifier blocks */
+	if (dev_priv->agp_heap)
+		flags = NOUVEAU_MEM_AGP | NOUVEAU_MEM_FB_ACCEPTABLE;
+	else
+		flags = NOUVEAU_MEM_FB;
+
+	chan->notifier_block = nouveau_mem_alloc(dev, 0, PAGE_SIZE, flags,filp);
+	if (!chan->notifier_block)
+		return DRM_ERR(ENOMEM);
+
+	ret = nouveau_mem_init_heap(&chan->notifier_heap,
+				    0, chan->notifier_block->size);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+void
+nouveau_notifier_takedown_channel(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
+
+	if (chan->notifier_block) {
+		nouveau_mem_free(dev, chan->notifier_block);
+		chan->notifier_block = NULL;
+	}
+
+	/*XXX: heap destroy */
+}
+
+int
+nouveau_notifier_alloc(drm_device_t *dev, int channel, uint32_t handle,
+		       int count, uint32_t *b_offset)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
+	struct nouveau_object *obj;
+	struct mem_block *mem;
+	uint32_t offset;
+	int target;
+
+	if (!chan->notifier_heap) {
+		DRM_ERROR("Channel %d doesn't have a notifier heap!\n",
+			  channel);
+		return DRM_ERR(EINVAL);
+	}
+
+	mem = nouveau_mem_alloc_block(chan->notifier_heap, 32, 0, chan->filp);
+	if (!mem) {
+		DRM_ERROR("Channel %d notifier block full\n", channel);
+		return DRM_ERR(ENOMEM);
+	}
+	mem->flags = NOUVEAU_MEM_NOTIFIER;
+
+	offset = chan->notifier_block->start + mem->start;
+	if (chan->notifier_block->flags & NOUVEAU_MEM_FB) {
+		offset -= drm_get_resource_start(dev, 1);
+		target = NV_DMA_TARGET_VIDMEM;
+	} else if (chan->notifier_block->flags & NOUVEAU_MEM_AGP) {
+		offset -= dev_priv->agp_phys;
+		target = NV_DMA_TARGET_AGP;
+	} else {
+		DRM_ERROR("Bad DMA target, flags 0x%08x!\n",
+			  chan->notifier_block->flags);
+		return DRM_ERR(EINVAL);
+	}
+
+	obj = nouveau_object_dma_create(dev, channel, NV_CLASS_DMA_IN_MEMORY,
+					offset, mem->size, NV_DMA_ACCESS_RW,
+					target);
+	if (!obj) {
+		nouveau_mem_free_block(mem);
+		DRM_ERROR("Error creating notifier ctxdma\n");
+		return DRM_ERR(ENOMEM);
+	}
+
+	obj->handle = handle;
+	if (nouveau_ramht_insert(dev, channel, handle, obj)) {
+		nouveau_object_free(dev, obj);
+		nouveau_mem_free_block(mem);
+		DRM_ERROR("Error inserting notifier ctxdma into RAMHT\n");
+		return DRM_ERR(ENOMEM);
+	}
+
+	*b_offset = mem->start;
+	return 0;
+}
+
+int
+nouveau_ioctl_notifier_alloc(DRM_IOCTL_ARGS)
+{
+	DRM_DEVICE;
+	drm_nouveau_notifier_alloc_t na;
+	int ret;
+
+	DRM_COPY_FROM_USER_IOCTL(na, (drm_nouveau_notifier_alloc_t __user*)data,
+				 sizeof(na));
+
+	if (!nouveau_fifo_owner(dev, filp, na.channel)) {
+		DRM_ERROR("pid %d doesn't own channel %d\n",
+			  DRM_CURRENTPID, na.channel);
+		return DRM_ERR(EPERM);
+	}
+
+	ret = nouveau_notifier_alloc(dev, na.channel, na.handle,
+				     na.count, &na.offset);
+	if (ret)
+		return ret;
+
+	DRM_COPY_TO_USER_IOCTL((drm_nouveau_notifier_alloc_t __user*)data,
+			       na, sizeof(na));
+	return 0;
+}
+
diff --git a/shared-core/nouveau_object.c b/shared-core/nouveau_object.c
index ace7c2aa..dac08df4 100644
--- a/shared-core/nouveau_object.c
+++ b/shared-core/nouveau_object.c
@@ -139,7 +139,7 @@ nouveau_object_handle_find(drm_device_t *dev, int channel, uint32_t handle)
    is given as:
 */
 static uint32_t
-nouveau_ht_handle_hash(drm_device_t *dev, int channel, uint32_t handle)
+nouveau_ramht_hash_handle(drm_device_t *dev, int channel, uint32_t handle)
 {
 	drm_nouveau_private_t *dev_priv=dev->dev_private;
 	uint32_t hash = 0;
@@ -154,62 +154,89 @@ nouveau_ht_handle_hash(drm_device_t *dev, int channel, uint32_t handle)
 }
 
 static int
-nouveau_ht_object_insert(drm_device_t* dev, int channel, uint32_t handle,
-			 struct nouveau_object *obj)
+nouveau_ramht_entry_valid(drm_device_t *dev, uint32_t ramht, uint32_t offset)
 {
 	drm_nouveau_private_t *dev_priv=dev->dev_private;
-	int ht_base = NV_RAMIN + dev_priv->ramht_offset;
-	int ht_end  = ht_base + dev_priv->ramht_size;
-	int o_ofs, ofs;
-
-	obj->handle = handle;
-	o_ofs = ofs = nouveau_ht_handle_hash(dev, channel, obj->handle);
-
-	while (NV_READ(ht_base + ofs) || NV_READ(ht_base + ofs + 4)) {
-		ofs += 8;
-		if (ofs == ht_end) ofs = ht_base;
-		if (ofs == o_ofs) {
-			DRM_ERROR("no free hash table entries\n");
-			return 1;
-		}
+	uint32_t ctx = NV_RI32(ramht + offset + 4);
+
+	if (dev_priv->card_type < NV_40)
+		return ((ctx & NV_RAMHT_CONTEXT_VALID) != 0);
+	return (ctx != 0);
+}
+
+int
+nouveau_ramht_insert(drm_device_t* dev, int channel, uint32_t handle,
+		     struct nouveau_object *obj)
+{
+	drm_nouveau_private_t *dev_priv=dev->dev_private;
+	uint32_t ramht = dev_priv->ramht_offset;
+	uint32_t ctx, co, ho;
+	uint32_t inst;
+
+	inst = nouveau_chip_instance_get(dev, obj->instance);
+	if (dev_priv->card_type < NV_40) {
+		ctx = NV_RAMHT_CONTEXT_VALID | inst |
+		      (channel     << NV_RAMHT_CONTEXT_CHANNEL_SHIFT) |
+		      (obj->engine << NV_RAMHT_CONTEXT_ENGINE_SHIFT);
+	} else
+	if (dev_priv->card_type < NV_50) {
+		ctx = inst |
+		      (channel     << NV40_RAMHT_CONTEXT_CHANNEL_SHIFT) |
+		      (obj->engine << NV40_RAMHT_CONTEXT_ENGINE_SHIFT);
+	} else {
+		ctx = inst |
+		      (obj->engine << NV40_RAMHT_CONTEXT_ENGINE_SHIFT);
 	}
-	ofs += ht_base;
-
-	DRM_DEBUG("Channel %d - Handle 0x%08x at 0x%08x\n",
-		channel, obj->handle, ofs);
-
-	NV_WRITE(NV_RAMHT_HANDLE_OFFSET + ofs, obj->handle);
-	if (dev_priv->card_type >= NV_40)
-		NV_WRITE(NV_RAMHT_CONTEXT_OFFSET + ofs,
-			(channel     << NV40_RAMHT_CONTEXT_CHANNEL_SHIFT) |
-			(obj->engine << NV40_RAMHT_CONTEXT_ENGINE_SHIFT) |
-			nouveau_chip_instance_get(dev, obj->instance)
-			);	    
-	else
-		NV_WRITE(NV_RAMHT_CONTEXT_OFFSET + ofs,
-			NV_RAMHT_CONTEXT_VALID |
-			(channel     << NV_RAMHT_CONTEXT_CHANNEL_SHIFT) |
-			(obj->engine << NV_RAMHT_CONTEXT_ENGINE_SHIFT) |
-			nouveau_chip_instance_get(dev, obj->instance)
-			);
-
-	obj->ht_loc = ofs;
-	return 0;
+
+	co = ho = nouveau_ramht_hash_handle(dev, channel, handle);
+	do {
+		if (!nouveau_ramht_entry_valid(dev, ramht, co)) {
+			DRM_DEBUG("insert ch%d 0x%08x: h=0x%08x, c=0x%08x\n",
+				  channel, co, handle, ctx);
+			NV_WI32(ramht + co + 0, handle);
+			NV_WI32(ramht + co + 4, ctx);
+			obj->handle = handle;
+			return 0;
+		}
+		DRM_DEBUG("collision ch%d 0x%08x: h=0x%08x\n",
+			  channel, co, NV_RI32(ramht + co));
+
+		co += 8;
+		if (co == dev_priv->ramht_size)
+			co = 0;
+	} while (co != ho);
+
+	DRM_ERROR("RAMHT space exhausted. ch=%d\n", channel);
+	return DRM_ERR(ENOMEM);
 }
 
-static void nouveau_hash_table_remove(drm_device_t* dev,
-				      struct nouveau_object *obj)
+static void
+nouveau_ramht_remove(drm_device_t* dev, struct nouveau_object *obj)
 {
 	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	uint32_t ramht = dev_priv->ramht_offset;
+	uint32_t co, ho;
+
+	co = ho = nouveau_ramht_hash_handle(dev, obj->channel, obj->handle);
+	do {
+		if (nouveau_ramht_entry_valid(dev, ramht, co) &&
+		    (obj->handle == NV_RI32(ramht + co))) {
+			DRM_DEBUG("remove ch%d 0x%08x: h=0x%08x, c=0x%08x\n",
+				  obj->channel, co, obj->handle,
+				  NV_RI32(ramht + co + 4));
+			NV_WI32(ramht + co + 0, 0x00000000);
+			NV_WI32(ramht + co + 4, 0x00000000);
+			obj->handle = ~0;
+			return;
+		}
 
-	DRM_DEBUG("Remove handle 0x%08x at 0x%08x from HT\n",
-			obj->handle, obj->ht_loc);
-	if (obj->ht_loc) {
-		DRM_DEBUG("... HT entry was: 0x%08x/0x%08x\n",
-				NV_READ(obj->ht_loc), NV_READ(obj->ht_loc+4));
-		NV_WRITE(obj->ht_loc  , 0x00000000);
-		NV_WRITE(obj->ht_loc+4, 0x00000000);
-	}
+		co += 8;
+		if (co == dev_priv->ramht_size)
+			co = 0;
+	} while (co != ho);
+
+	DRM_ERROR("RAMHT entry not found. ch=%d, handle=0x%08x\n",
+		  obj->channel, obj->handle);
 }
 
 static struct nouveau_object *
@@ -457,119 +484,74 @@ nouveau_object_free(drm_device_t *dev, struct nouveau_object *obj)
 {
 	nouveau_object_instance_free(dev, obj);
 	if (obj->handle != ~0)
-		nouveau_hash_table_remove(dev, obj);
+		nouveau_ramht_remove(dev, obj);
 	drm_free(obj, sizeof(struct nouveau_object), DRM_MEM_DRIVER);
 }
 
-void nouveau_object_cleanup(drm_device_t *dev, int channel)
-{
-	drm_nouveau_private_t *dev_priv=dev->dev_private;
-
-	while (dev_priv->fifos[channel].objs) {
-		nouveau_object_free(dev, dev_priv->fifos[channel].objs);
-	}
-}
-
-int nouveau_ioctl_object_init(DRM_IOCTL_ARGS)
+int
+nouveau_object_init_channel(drm_device_t *dev, int channel,
+			    uint32_t vram_handle,
+			    uint32_t tt_handle)
 {
-	DRM_DEVICE;
-	drm_nouveau_object_init_t init;
-	struct nouveau_object *obj;
-
-	DRM_COPY_FROM_USER_IOCTL(init, (drm_nouveau_object_init_t __user *)
-		data, sizeof(init));
-
-	if (!nouveau_fifo_owner(dev, filp, init.channel)) {
-		DRM_ERROR("pid %d doesn't own channel %d\n",
-				DRM_CURRENTPID, init.channel);
-		return DRM_ERR(EINVAL);
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	struct nouveau_object *gpuobj;
+	int ret;
+
+	/* VRAM ctxdma */
+	gpuobj = nouveau_object_dma_create(dev, channel, NV_CLASS_DMA_IN_MEMORY,
+					   0, dev_priv->fb_available_size,
+					   NV_DMA_ACCESS_RW,
+					   NV_DMA_TARGET_VIDMEM);
+	if (!gpuobj) {
+		DRM_ERROR("Error creating VRAM ctxdma: %d\n", DRM_ERR(ENOMEM));
+		return DRM_ERR(ENOMEM);
 	}
 
-	//FIXME: check args, only allow trusted objects to be created
-
-	if (nouveau_object_handle_find(dev, init.channel, init.handle)) {
-		DRM_ERROR("Channel %d: handle 0x%08x already exists\n",
-			init.channel, init.handle);
-		return DRM_ERR(EINVAL);
+	ret = nouveau_ramht_insert(dev, channel, vram_handle, gpuobj);
+	if (ret) {
+		DRM_ERROR("Error referencing VRAM ctxdma: %d\n", ret);
+		return ret;
 	}
 
-	obj = nouveau_object_gr_create(dev, init.channel, init.class);
-	if (!obj)
+	/* non-AGP unimplemented */
+	if (dev_priv->agp_heap == NULL)
+		return 0;
+
+	/* GART ctxdma */
+	gpuobj = nouveau_object_dma_create(dev, channel, NV_CLASS_DMA_IN_MEMORY,
+					   0, dev_priv->agp_available_size,
+					   NV_DMA_ACCESS_RW,
+					   NV_DMA_TARGET_AGP);
+	if (!gpuobj) {
+		DRM_ERROR("Error creating TT ctxdma: %d\n", DRM_ERR(ENOMEM));
 		return DRM_ERR(ENOMEM);
+	}
 
-	if (nouveau_ht_object_insert(dev, init.channel, init.handle, obj)) {
-		nouveau_object_free(dev, obj);
-		return DRM_ERR(ENOMEM);
+	ret = nouveau_ramht_insert(dev, channel, tt_handle, gpuobj);
+	if (ret) {
+		DRM_ERROR("Error referencing TT ctxdma: %d\n", ret);
+		return ret;
 	}
 
 	return 0;
 }
 
-static int
-nouveau_dma_object_check_access(drm_device_t *dev,
-				drm_nouveau_dma_object_init_t *init)
+void nouveau_object_cleanup(drm_device_t *dev, int channel)
 {
-	drm_nouveau_private_t *dev_priv = dev->dev_private;
-	uint64_t limit;
-
-	/* Check for known DMA object classes */
-	switch (init->class) {
-	case NV_CLASS_DMA_IN_MEMORY:
-	case NV_CLASS_DMA_FROM_MEMORY:
-	case NV_CLASS_DMA_TO_MEMORY:
-		break;
-	default:
-		DRM_ERROR("invalid class = 0x%x\n", init->class);
-		return DRM_ERR(EPERM);
-	}
-
-	/* Check access mode, and translate to NV_DMA_ACCESS_* */
-	switch (init->access) {
-	case NOUVEAU_MEM_ACCESS_RO:
-		init->access = NV_DMA_ACCESS_RO;
-		break;
-	case NOUVEAU_MEM_ACCESS_WO:
-		init->access = NV_DMA_ACCESS_WO;
-		break;
-	case NOUVEAU_MEM_ACCESS_RW:
-		init->access = NV_DMA_ACCESS_RW;
-		break;
-	default:
-		DRM_ERROR("invalid access mode = %d\n", init->access);
-		return DRM_ERR(EPERM);
-	}
-
-	/* Check that request is within the allowed limits of "target" */
-	switch (init->target) {
-	case NOUVEAU_MEM_FB:
-		limit = dev_priv->fb_available_size;
-		init->target = NV_DMA_TARGET_VIDMEM;
-		break;
-	case NOUVEAU_MEM_AGP:
-		limit = dev_priv->agp_available_size;
-		init->target = NV_DMA_TARGET_AGP;
-		break;
-	default:
-		DRM_ERROR("invalid target = 0x%x\n", init->target);
-		return DRM_ERR(EPERM);
-	}
+	drm_nouveau_private_t *dev_priv=dev->dev_private;
 
-	if ((init->offset > limit) || (init->offset + init->size) > limit) {
-		DRM_ERROR("access out of allowed range (%d,0x%08x,0x%08x)\n",
-				init->target, init->offset, init->size);
-		return DRM_ERR(EPERM);
+	while (dev_priv->fifos[channel].objs) {
+		nouveau_object_free(dev, dev_priv->fifos[channel].objs);
 	}
-
-	return 0;
 }
 
-int nouveau_ioctl_dma_object_init(DRM_IOCTL_ARGS)
+int nouveau_ioctl_grobj_alloc(DRM_IOCTL_ARGS)
 {
 	DRM_DEVICE;
-	drm_nouveau_dma_object_init_t init;
+	drm_nouveau_grobj_alloc_t init;
 	struct nouveau_object *obj;
 
-	DRM_COPY_FROM_USER_IOCTL(init, (drm_nouveau_dma_object_init_t __user *)
+	DRM_COPY_FROM_USER_IOCTL(init, (drm_nouveau_grobj_alloc_t __user *)
 		data, sizeof(init));
 
 	if (!nouveau_fifo_owner(dev, filp, init.channel)) {
@@ -578,8 +560,7 @@ int nouveau_ioctl_dma_object_init(DRM_IOCTL_ARGS)
 		return DRM_ERR(EINVAL);
 	}
 
-	if (nouveau_dma_object_check_access(dev, &init))
-		return DRM_ERR(EPERM);
+	//FIXME: check args, only allow trusted objects to be created
 
 	if (nouveau_object_handle_find(dev, init.channel, init.handle)) {
 		DRM_ERROR("Channel %d: handle 0x%08x already exists\n",
@@ -587,14 +568,11 @@ int nouveau_ioctl_dma_object_init(DRM_IOCTL_ARGS)
 		return DRM_ERR(EINVAL);
 	}
 
-	obj = nouveau_object_dma_create(dev, init.channel, init.class,
-					     init.offset, init.size,
-					     init.access, init.target);
+	obj = nouveau_object_gr_create(dev, init.channel, init.class);
 	if (!obj)
 		return DRM_ERR(ENOMEM);
 
-	obj->handle = init.handle;
-	if (nouveau_ht_object_insert(dev, init.channel, init.handle, obj)) {
+	if (nouveau_ramht_insert(dev, init.channel, init.handle, obj)) {
 		nouveau_object_free(dev, obj);
 		return DRM_ERR(ENOMEM);
 	}
diff --git a/shared-core/nouveau_reg.h b/shared-core/nouveau_reg.h
index ea4a2f6b..4c013c53 100644
--- a/shared-core/nouveau_reg.h
+++ b/shared-core/nouveau_reg.h
@@ -135,6 +135,17 @@
 #define NV10_PGRAPH_CTX_CACHE4                             0x004001C0
 #define NV04_PGRAPH_CTX_CACHE4                             0x004001E0
 #define NV10_PGRAPH_CTX_CACHE5                             0x004001E0
+#define NV40_PGRAPH_CTXCTL_0304                            0x00400304
+#define NV40_PGRAPH_CTXCTL_0304_XFER_CTX                   0x00000001
+#define NV40_PGRAPH_CTXCTL_0310                            0x00400310
+#define NV40_PGRAPH_CTXCTL_0310_XFER_SAVE                  0x00000020
+#define NV40_PGRAPH_CTXCTL_0310_XFER_LOAD                  0x00000040
+#define NV40_PGRAPH_CTXCTL_030C                            0x0040030c
+#define NV40_PGRAPH_CTXCTL_UCODE_INDEX                     0x00400324
+#define NV40_PGRAPH_CTXCTL_UCODE_DATA                      0x00400328
+#define NV40_PGRAPH_CTXCTL_CUR                             0x0040032c
+#define NV40_PGRAPH_CTXCTL_CUR_LOADED                      0x01000000
+#define NV40_PGRAPH_CTXCTL_CUR_INST_MASK                   0x000FFFFF
 #define NV03_PGRAPH_ABS_X_RAM                              0x00400400
 #define NV03_PGRAPH_ABS_Y_RAM                              0x00400480
 #define NV03_PGRAPH_X_MISC                                 0x00400500
@@ -230,7 +241,11 @@
 #define NV10_PGRAPH_SCALED_FORMAT                          0x00400778
 #define NV10_PGRAPH_CHANNEL_CTX_TABLE                      0x00400780
 #define NV10_PGRAPH_CHANNEL_CTX_SIZE                       0x00400784
+#define NV20_PGRAPH_CHANNEL_CTX_POINTER                    0x00400784
 #define NV10_PGRAPH_CHANNEL_CTX_POINTER                    0x00400788
+#define NV20_PGRAPH_CHANNEL_CTX_XFER                       0x00400788
+#define NV20_PGRAPH_CHANNEL_CTX_XFER_LOAD                  0x00000001
+#define NV20_PGRAPH_CHANNEL_CTX_XFER_SAVE                  0x00000002
 #define NV04_PGRAPH_PATT_COLOR0                            0x00400800
 #define NV04_PGRAPH_PATT_COLOR1                            0x00400804
 #define NV04_PGRAPH_PATTERN                                0x00400808
@@ -404,7 +419,7 @@
 #define NV10_PFIFO_CACHE1_SEMAPHORE                        0x0000326C
 #define NV03_PFIFO_CACHE1_GET                              0x00003270
 #define NV04_PFIFO_CACHE1_ENGINE                           0x00003280
-#define NV10_PFIFO_CACHE1_DMA_DCOUNT                       0x000032A0
+#define NV04_PFIFO_CACHE1_DMA_DCOUNT                       0x000032A0
 #define NV40_PFIFO_GRCTX_INSTANCE                          0x000032E0
 #define NV40_PFIFO_UNK32E4                                 0x000032E4
 #define NV04_PFIFO_CACHE1_METHOD(i)                (0x00003800+(i*8))
@@ -427,7 +442,10 @@
 #define NV04_RAMFC_DMA_PUT                                       0x00
 #define NV04_RAMFC_DMA_GET                                       0x04
 #define NV04_RAMFC_DMA_INSTANCE                                  0x08
+#define NV04_RAMFC_DMA_STATE                                     0x0C
 #define NV04_RAMFC_DMA_FETCH                                     0x10
+#define NV04_RAMFC_ENGINE                                        0x14
+#define NV04_RAMFC_PULL1_ENGINE                                  0x18
 
 #define NV10_RAMFC_DMA_PUT                                       0x00
 #define NV10_RAMFC_DMA_GET                                       0x04
@@ -462,6 +480,6 @@
 #define NV40_RAMFC_UNK_40                                        0x40
 #define NV40_RAMFC_UNK_44                                        0x44
 #define NV40_RAMFC_UNK_48                                        0x48
-#define NV40_RAMFC_2088                                          0x4C
-#define NV40_RAMFC_3300                                          0x50
+#define NV40_RAMFC_UNK_4C                                        0x4C
+#define NV40_RAMFC_UNK_50                                        0x50
 
diff --git a/shared-core/nouveau_state.c b/shared-core/nouveau_state.c
index e7930b9e..fa773d28 100644
--- a/shared-core/nouveau_state.c
+++ b/shared-core/nouveau_state.c
@@ -51,6 +51,7 @@ static int nouveau_init_card_mappings(drm_device_t *dev)
 	DRM_DEBUG("regs mapped ok at 0x%lx\n", dev_priv->mmio->offset);
 
 	/* map larger RAMIN aperture on NV40 cards */
+	dev_priv->ramin = NULL;
 	if (dev_priv->card_type >= NV_40) {
 		int ramin_resource = 2;
 		if (drm_get_resource_len(dev, ramin_resource) == 0)
@@ -66,12 +67,26 @@ static int nouveau_init_card_mappings(drm_device_t *dev)
 				  "limited instance memory available\n");
 			dev_priv->ramin = NULL;
 		}
-	} else
-		dev_priv->ramin = NULL;
+	}
+
+	/* On older cards (or if the above failed), create a map covering
+	 * the BAR0 PRAMIN aperture */
+	if (!dev_priv->ramin) {
+		ret = drm_addmap(dev,
+				 drm_get_resource_start(dev, 0) + NV_RAMIN,
+				 (1*1024*1024),
+				 _DRM_REGISTERS, _DRM_READ_ONLY,
+				 &dev_priv->ramin);
+		if (ret) {
+			DRM_ERROR("Failed to map BAR0 PRAMIN: %d\n", ret);
+			return ret;
+		}
+	}
 
 	return 0;
 }
 
+static int nouveau_stub_init(drm_device_t *dev) { return 0; }
 static void nouveau_stub_takedown(drm_device_t *dev) {}
 static int nouveau_init_engine_ptrs(drm_device_t *dev)
 {
@@ -80,66 +95,126 @@ static int nouveau_init_engine_ptrs(drm_device_t *dev)
 
 	switch (dev_priv->chipset & 0xf0) {
 	case 0x00:
-		engine->Mc.Init		= nv04_mc_init;
-		engine->Mc.Takedown	= nv04_mc_takedown;
-		engine->Timer.Init	= nv04_timer_init;
-		engine->Timer.Takedown	= nv04_timer_takedown;
-		engine->Fb.Init		= nv04_fb_init;
-		engine->Fb.Takedown	= nv04_fb_takedown;
-		engine->Graph.Init	= nv04_graph_init;
-		engine->Graph.Takedown	= nv04_graph_takedown;
-		engine->Fifo.Init	= nouveau_fifo_init;
-		engine->Fifo.Takedown	= nouveau_stub_takedown;
+		engine->mc.init		= nv04_mc_init;
+		engine->mc.takedown	= nv04_mc_takedown;
+		engine->timer.init	= nv04_timer_init;
+		engine->timer.takedown	= nv04_timer_takedown;
+		engine->fb.init		= nv04_fb_init;
+		engine->fb.takedown	= nv04_fb_takedown;
+		engine->graph.init	= nv04_graph_init;
+		engine->graph.takedown	= nv04_graph_takedown;
+		engine->graph.create_context	= nv04_graph_create_context;
+		engine->graph.destroy_context	= nv04_graph_destroy_context;
+		engine->graph.load_context	= nv04_graph_load_context;
+		engine->graph.save_context	= nv04_graph_save_context;
+		engine->fifo.init	= nouveau_fifo_init;
+		engine->fifo.takedown	= nouveau_stub_takedown;
+		engine->fifo.create_context	= nv04_fifo_create_context;
+		engine->fifo.destroy_context	= nv04_fifo_destroy_context;
+		engine->fifo.load_context	= nv04_fifo_load_context;
+		engine->fifo.save_context	= nv04_fifo_save_context;
 		break;
 	case 0x10:
-		engine->Mc.Init		= nv04_mc_init;
-		engine->Mc.Takedown	= nv04_mc_takedown;
-		engine->Timer.Init	= nv04_timer_init;
-		engine->Timer.Takedown	= nv04_timer_takedown;
-		engine->Fb.Init		= nv10_fb_init;
-		engine->Fb.Takedown	= nv10_fb_takedown;
-		engine->Graph.Init	= nv10_graph_init;
-		engine->Graph.Takedown	= nv10_graph_takedown;
-		engine->Fifo.Init	= nouveau_fifo_init;
-		engine->Fifo.Takedown	= nouveau_stub_takedown;
+		engine->mc.init		= nv04_mc_init;
+		engine->mc.takedown	= nv04_mc_takedown;
+		engine->timer.init	= nv04_timer_init;
+		engine->timer.takedown	= nv04_timer_takedown;
+		engine->fb.init		= nv10_fb_init;
+		engine->fb.takedown	= nv10_fb_takedown;
+		engine->graph.init	= nv10_graph_init;
+		engine->graph.takedown	= nv10_graph_takedown;
+		engine->graph.create_context	= nv10_graph_create_context;
+		engine->graph.destroy_context	= nv10_graph_destroy_context;
+		engine->graph.load_context	= nv10_graph_load_context;
+		engine->graph.save_context	= nv10_graph_save_context;
+		engine->fifo.init	= nouveau_fifo_init;
+		engine->fifo.takedown	= nouveau_stub_takedown;
+		engine->fifo.create_context	= nv10_fifo_create_context;
+		engine->fifo.destroy_context	= nv10_fifo_destroy_context;
+		engine->fifo.load_context	= nv10_fifo_load_context;
+		engine->fifo.save_context	= nv10_fifo_save_context;
 		break;
 	case 0x20:
-		engine->Mc.Init		= nv04_mc_init;
-		engine->Mc.Takedown	= nv04_mc_takedown;
-		engine->Timer.Init	= nv04_timer_init;
-		engine->Timer.Takedown	= nv04_timer_takedown;
-		engine->Fb.Init		= nv10_fb_init;
-		engine->Fb.Takedown	= nv10_fb_takedown;
-		engine->Graph.Init	= nv20_graph_init;
-		engine->Graph.Takedown	= nv20_graph_takedown;
-		engine->Fifo.Init	= nouveau_fifo_init;
-		engine->Fifo.Takedown	= nouveau_stub_takedown;
+		engine->mc.init		= nv04_mc_init;
+		engine->mc.takedown	= nv04_mc_takedown;
+		engine->timer.init	= nv04_timer_init;
+		engine->timer.takedown	= nv04_timer_takedown;
+		engine->fb.init		= nv10_fb_init;
+		engine->fb.takedown	= nv10_fb_takedown;
+		engine->graph.init	= nv20_graph_init;
+		engine->graph.takedown	= nv20_graph_takedown;
+		engine->graph.create_context	= nv20_graph_create_context;
+		engine->graph.destroy_context	= nv20_graph_destroy_context;
+		engine->graph.load_context	= nv20_graph_load_context;
+		engine->graph.save_context	= nv20_graph_save_context;
+		engine->fifo.init	= nouveau_fifo_init;
+		engine->fifo.takedown	= nouveau_stub_takedown;
+		engine->fifo.create_context	= nv10_fifo_create_context;
+		engine->fifo.destroy_context	= nv10_fifo_destroy_context;
+		engine->fifo.load_context	= nv10_fifo_load_context;
+		engine->fifo.save_context	= nv10_fifo_save_context;
 		break;
 	case 0x30:
-		engine->Mc.Init		= nv04_mc_init;
-		engine->Mc.Takedown	= nv04_mc_takedown;
-		engine->Timer.Init	= nv04_timer_init;
-		engine->Timer.Takedown	= nv04_timer_takedown;
-		engine->Fb.Init		= nv10_fb_init;
-		engine->Fb.Takedown	= nv10_fb_takedown;
-		engine->Graph.Init	= nv30_graph_init;
-		engine->Graph.Takedown	= nv30_graph_takedown;
-		engine->Fifo.Init	= nouveau_fifo_init;
-		engine->Fifo.Takedown	= nouveau_stub_takedown;
+		engine->mc.init		= nv04_mc_init;
+		engine->mc.takedown	= nv04_mc_takedown;
+		engine->timer.init	= nv04_timer_init;
+		engine->timer.takedown	= nv04_timer_takedown;
+		engine->fb.init		= nv10_fb_init;
+		engine->fb.takedown	= nv10_fb_takedown;
+		engine->graph.init	= nv30_graph_init;
+		engine->graph.takedown	= nv30_graph_takedown;
+		engine->graph.create_context	= nv30_graph_create_context;
+		engine->graph.destroy_context	= nv30_graph_destroy_context;
+		engine->graph.load_context	= nv30_graph_load_context;
+		engine->graph.save_context	= nv30_graph_save_context;
+		engine->fifo.init	= nouveau_fifo_init;
+		engine->fifo.takedown	= nouveau_stub_takedown;
+		engine->fifo.create_context	= nv10_fifo_create_context;
+		engine->fifo.destroy_context	= nv10_fifo_destroy_context;
+		engine->fifo.load_context	= nv10_fifo_load_context;
+		engine->fifo.save_context	= nv10_fifo_save_context;
 		break;
 	case 0x40:
-		engine->Mc.Init		= nv40_mc_init;
-		engine->Mc.Takedown	= nv40_mc_takedown;
-		engine->Timer.Init	= nv04_timer_init;
-		engine->Timer.Takedown	= nv04_timer_takedown;
-		engine->Fb.Init		= nv40_fb_init;
-		engine->Fb.Takedown	= nv40_fb_takedown;
-		engine->Graph.Init	= nv40_graph_init;
-		engine->Graph.Takedown	= nv40_graph_takedown;
-		engine->Fifo.Init	= nouveau_fifo_init;
-		engine->Fifo.Takedown	= nouveau_stub_takedown;
+		engine->mc.init		= nv40_mc_init;
+		engine->mc.takedown	= nv40_mc_takedown;
+		engine->timer.init	= nv04_timer_init;
+		engine->timer.takedown	= nv04_timer_takedown;
+		engine->fb.init		= nv40_fb_init;
+		engine->fb.takedown	= nv40_fb_takedown;
+		engine->graph.init	= nv40_graph_init;
+		engine->graph.takedown	= nv40_graph_takedown;
+		engine->graph.create_context	= nv40_graph_create_context;
+		engine->graph.destroy_context	= nv40_graph_destroy_context;
+		engine->graph.load_context	= nv40_graph_load_context;
+		engine->graph.save_context	= nv40_graph_save_context;
+		engine->fifo.init	= nouveau_fifo_init;
+		engine->fifo.takedown	= nouveau_stub_takedown;
+		engine->fifo.create_context	= nv40_fifo_create_context;
+		engine->fifo.destroy_context	= nv40_fifo_destroy_context;
+		engine->fifo.load_context	= nv40_fifo_load_context;
+		engine->fifo.save_context	= nv40_fifo_save_context;
 		break;
 	case 0x50:
+	case 0x80: /* gotta love NVIDIA's consistency.. */
+		engine->mc.init		= nv50_mc_init;
+		engine->mc.takedown	= nv50_mc_takedown;
+		engine->timer.init	= nouveau_stub_init;
+		engine->timer.takedown	= nouveau_stub_takedown;
+		engine->fb.init		= nouveau_stub_init;
+		engine->fb.takedown	= nouveau_stub_takedown;
+		engine->graph.init	= nv50_graph_init;
+		engine->graph.takedown	= nv50_graph_takedown;
+		engine->graph.create_context	= nv50_graph_create_context;
+		engine->graph.destroy_context	= nv50_graph_destroy_context;
+		engine->graph.load_context	= nv50_graph_load_context;
+		engine->graph.save_context	= nv50_graph_save_context;
+		engine->fifo.init	= nv50_fifo_init;
+		engine->fifo.takedown	= nv50_fifo_takedown;
+		engine->fifo.create_context	= nv50_fifo_create_context;
+		engine->fifo.destroy_context	= nv50_fifo_destroy_context;
+		engine->fifo.load_context	= nv50_fifo_load_context;
+		engine->fifo.save_context	= nv50_fifo_save_context;
+		break;
 	default:
 		DRM_ERROR("NV%02x unsupported\n", dev_priv->chipset);
 		return 1;
@@ -184,23 +259,23 @@ static int nouveau_card_init(drm_device_t *dev)
 	/* Parse BIOS tables / Run init tables? */
 
 	/* PMC */
-	ret = engine->Mc.Init(dev);
+	ret = engine->mc.init(dev);
 	if (ret) return ret;
 
 	/* PTIMER */
-	ret = engine->Timer.Init(dev);
+	ret = engine->timer.init(dev);
 	if (ret) return ret;
 
 	/* PFB */
-	ret = engine->Fb.Init(dev);
+	ret = engine->fb.init(dev);
 	if (ret) return ret;
 
 	/* PGRAPH */
-	ret = engine->Graph.Init(dev);
+	ret = engine->graph.init(dev);
 	if (ret) return ret;
 
 	/* PFIFO */
-	ret = engine->Fifo.Init(dev);
+	ret = engine->fifo.init(dev);
 	if (ret) return ret;
 
 	/* what about PVIDEO/PCRTC/PRAMDAC etc? */
@@ -213,9 +288,9 @@ void nouveau_preclose(drm_device_t * dev, DRMFILE filp)
 {
 	drm_nouveau_private_t *dev_priv = dev->dev_private;
 
+	nouveau_fifo_cleanup(dev, filp);
 	nouveau_mem_release(filp,dev_priv->fb_heap);
 	nouveau_mem_release(filp,dev_priv->agp_heap);
-	nouveau_fifo_cleanup(dev, filp);
 }
 
 /* first module load, setup the mmio/fb mapping */
@@ -235,7 +310,6 @@ int nouveau_firstopen(struct drm_device *dev)
 int nouveau_load(struct drm_device *dev, unsigned long flags)
 {
 	drm_nouveau_private_t *dev_priv;
-	int ret;
 
 	if (flags==NV_UNKNOWN)
 		return DRM_ERR(EINVAL);
diff --git a/shared-core/nv04_fifo.c b/shared-core/nv04_fifo.c
new file mode 100644
index 00000000..bfae432e
--- /dev/null
+++ b/shared-core/nv04_fifo.c
@@ -0,0 +1,126 @@
+/*
+ * Copyright (C) 2007 Ben Skeggs.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "drmP.h"
+#include "drm.h"
+#include "nouveau_drv.h"
+
+#define RAMFC_WR(offset, val) NV_WI32(fifoctx + NV04_RAMFC_##offset, (val))
+#define RAMFC_RD(offset)      NV_RI32(fifoctx + NV04_RAMFC_##offset)
+#define NV04_RAMFC(c) (dev_priv->ramfc_offset + ((c) * NV04_RAMFC__SIZE))
+#define NV04_RAMFC__SIZE 32
+
+int
+nv04_fifo_create_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
+	struct nouveau_object *pb = chan->cmdbuf_obj;
+	uint32_t fifoctx = NV04_RAMFC(channel);
+	int i;
+
+	if (!pb || !pb->instance)
+		return DRM_ERR(EINVAL);
+
+	/* Clear RAMFC */
+	for (i=0; i<NV04_RAMFC__SIZE; i+=4)
+		NV_WI32(fifoctx + i, 0);
+	
+	/* Setup initial state */
+	RAMFC_WR(DMA_PUT, chan->pushbuf_base);
+	RAMFC_WR(DMA_GET, chan->pushbuf_base);
+	RAMFC_WR(DMA_INSTANCE, nouveau_chip_instance_get(dev, pb->instance));
+	RAMFC_WR(DMA_FETCH, (NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
+			     NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
+			     NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8 |
+#ifdef __BIG_ENDIAN
+			     NV_PFIFO_CACHE1_BIG_ENDIAN |
+#endif
+			     0));
+	return 0;
+}
+
+void
+nv04_fifo_destroy_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	uint32_t fifoctx = NV04_RAMFC(channel);
+	int i;
+
+	for (i=0; i<NV04_RAMFC__SIZE; i+=4)
+		NV_WI32(fifoctx + i, 0);
+}
+
+int
+nv04_fifo_load_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	uint32_t fifoctx = NV04_RAMFC(channel);
+	uint32_t tmp;
+
+	NV_WRITE(NV03_PFIFO_CACHE1_PUSH1, (1<<8) | channel);
+
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_GET, RAMFC_RD(DMA_GET));
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_PUT, RAMFC_RD(DMA_PUT));
+	
+	tmp = RAMFC_RD(DMA_INSTANCE);
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_INSTANCE, tmp & 0xFFFF);
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_DCOUNT, tmp >> 16);
+	
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_STATE, RAMFC_RD(DMA_STATE));
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_FETCH, RAMFC_RD(DMA_FETCH));
+	NV_WRITE(NV04_PFIFO_CACHE1_ENGINE, RAMFC_RD(ENGINE));
+	NV_WRITE(NV04_PFIFO_CACHE1_PULL1, RAMFC_RD(PULL1_ENGINE));
+
+	/* Reset NV04_PFIFO_CACHE1_DMA_CTL_AT_INFO to INVALID */
+	tmp = NV_READ(NV04_PFIFO_CACHE1_DMA_CTL) & ~(1<<31);
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_CTL, tmp);
+
+	return 0;
+}
+
+int
+nv04_fifo_save_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	uint32_t fifoctx = NV04_RAMFC(channel);
+	uint32_t tmp;
+
+	RAMFC_WR(DMA_PUT, NV04_PFIFO_CACHE1_DMA_PUT);
+	RAMFC_WR(DMA_GET, NV04_PFIFO_CACHE1_DMA_GET);
+
+	tmp  = NV_READ(NV04_PFIFO_CACHE1_DMA_DCOUNT) << 16;
+	tmp |= NV_READ(NV04_PFIFO_CACHE1_DMA_INSTANCE);
+	RAMFC_WR(DMA_INSTANCE, tmp);
+
+	RAMFC_WR(DMA_STATE, NV_READ(NV04_PFIFO_CACHE1_DMA_STATE));
+	RAMFC_WR(DMA_FETCH, NV_READ(NV04_PFIFO_CACHE1_DMA_FETCH));
+	RAMFC_WR(ENGINE, NV_READ(NV04_PFIFO_CACHE1_ENGINE));
+	RAMFC_WR(PULL1_ENGINE, NV_READ(NV04_PFIFO_CACHE1_PULL1));
+	
+	return 0;
+}
+
diff --git a/shared-core/nv04_graph.c b/shared-core/nv04_graph.c
index 0cd4d3b8..1aaae33c 100644
--- a/shared-core/nv04_graph.c
+++ b/shared-core/nv04_graph.c
@@ -336,7 +336,7 @@ void nouveau_nv04_context_switch(drm_device_t *dev)
 	NV_WRITE(NV04_PGRAPH_FIFO,0x1);
 }
 
-int nv04_graph_context_create(drm_device_t *dev, int channel) {
+int nv04_graph_create_context(drm_device_t *dev, int channel) {
 	drm_nouveau_private_t *dev_priv = dev->dev_private;
 	DRM_DEBUG("nv04_graph_context_create %d\n", channel);
 
@@ -351,6 +351,21 @@ int nv04_graph_context_create(drm_device_t *dev, int channel) {
 	return 0;
 }
 
+void nv04_graph_destroy_context(drm_device_t *dev, int channel)
+{
+}
+
+int nv04_graph_load_context(drm_device_t *dev, int channel)
+{
+	DRM_ERROR("stub!\n");
+	return 0;
+}
+
+int nv04_graph_save_context(drm_device_t *dev, int channel)
+{
+	DRM_ERROR("stub!\n");
+	return 0;
+}
 
 int nv04_graph_init(drm_device_t *dev) {
 	drm_nouveau_private_t *dev_priv = dev->dev_private;
diff --git a/shared-core/nv10_fifo.c b/shared-core/nv10_fifo.c
new file mode 100644
index 00000000..b84971de
--- /dev/null
+++ b/shared-core/nv10_fifo.c
@@ -0,0 +1,154 @@
+/*
+ * Copyright (C) 2007 Ben Skeggs.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "drmP.h"
+#include "drm.h"
+#include "nouveau_drv.h"
+
+#define RAMFC_WR(offset, val)	NV_WI32(fifoctx + NV10_RAMFC_##offset, (val))
+#define RAMFC_RD(offset)	NV_RI32(fifoctx + NV10_RAMFC_##offset)
+#define NV10_RAMFC(c) (dev_priv->ramfc_offset + NV10_RAMFC__SIZE)
+#define NV10_RAMFC__SIZE ((dev_priv->chipset) >= 0x17 ? 64 : 32)
+
+int
+nv10_fifo_create_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
+	uint32_t fifoctx = NV10_RAMFC(channel), pushbuf;
+	int i;
+
+	pushbuf = nouveau_chip_instance_get(dev, chan->cmdbuf_obj->instance);
+
+	for (i=0; i<NV10_RAMFC__SIZE; i+=4)
+		NV_WI32(fifoctx + i, 0);
+
+	/* Fill entries that are seen filled in dumps of nvidia driver just
+	 * after channel's is put into DMA mode
+	 */
+	RAMFC_WR(DMA_PUT       , chan->pushbuf_base);
+	RAMFC_WR(DMA_GET       , chan->pushbuf_base);
+	RAMFC_WR(DMA_INSTANCE  , pushbuf);
+	RAMFC_WR(DMA_FETCH     , NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
+				 NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
+				 NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8 |
+#ifdef __BIG_ENDIAN
+				 NV_PFIFO_CACHE1_BIG_ENDIAN |
+#endif
+				 0);
+
+	return 0;
+}
+
+void
+nv10_fifo_destroy_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	uint32_t fifoctx = NV10_RAMFC(channel);
+	int i;
+
+	for (i=0; i<NV10_RAMFC__SIZE; i+=4)
+		NV_WI32(fifoctx + i, 0);
+}
+
+int
+nv10_fifo_load_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	uint32_t fifoctx = NV10_RAMFC(channel);
+	uint32_t tmp;
+
+	NV_WRITE(NV03_PFIFO_CACHE1_PUSH1            , 0x00000100 | channel);
+
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_GET          , RAMFC_RD(DMA_GET));
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_PUT          , RAMFC_RD(DMA_PUT));
+	NV_WRITE(NV10_PFIFO_CACHE1_REF_CNT          , RAMFC_RD(REF_CNT));
+
+	tmp = RAMFC_RD(DMA_INSTANCE);
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_INSTANCE     , tmp & 0xFFFF);
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_DCOUNT       , tmp >> 16);
+
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_STATE        , RAMFC_RD(DMA_STATE));
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_FETCH        , RAMFC_RD(DMA_FETCH));
+	NV_WRITE(NV04_PFIFO_CACHE1_ENGINE           , RAMFC_RD(ENGINE));
+	NV_WRITE(NV04_PFIFO_CACHE1_PULL1            , RAMFC_RD(PULL1_ENGINE));
+
+	if (dev_priv->chipset >= 0x17) {
+		NV_WRITE(NV10_PFIFO_CACHE1_ACQUIRE_VALUE,
+			 RAMFC_RD(ACQUIRE_VALUE));
+		NV_WRITE(NV10_PFIFO_CACHE1_ACQUIRE_TIMESTAMP,
+			 RAMFC_RD(ACQUIRE_TIMESTAMP));
+		NV_WRITE(NV10_PFIFO_CACHE1_ACQUIRE_TIMEOUT,
+			 RAMFC_RD(ACQUIRE_TIMEOUT));
+		NV_WRITE(NV10_PFIFO_CACHE1_SEMAPHORE,
+			 RAMFC_RD(SEMAPHORE));
+		NV_WRITE(NV10_PFIFO_CACHE1_DMA_SUBROUTINE,
+			 RAMFC_RD(DMA_SUBROUTINE));
+	}
+
+	/* Reset NV04_PFIFO_CACHE1_DMA_CTL_AT_INFO to INVALID */
+	tmp = NV_READ(NV04_PFIFO_CACHE1_DMA_CTL) & ~(1<<31);
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_CTL, tmp);
+
+	return 0;
+}
+
+int
+nv10_fifo_save_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	uint32_t fifoctx = NV10_RAMFC(channel);
+	uint32_t tmp;
+
+	RAMFC_WR(DMA_PUT          , NV_READ(NV04_PFIFO_CACHE1_DMA_PUT));
+	RAMFC_WR(DMA_GET          , NV_READ(NV04_PFIFO_CACHE1_DMA_GET));
+	RAMFC_WR(REF_CNT          , NV_READ(NV10_PFIFO_CACHE1_REF_CNT));
+
+	tmp  = NV_READ(NV04_PFIFO_CACHE1_DMA_INSTANCE) & 0xFFFF;
+	tmp |= (NV_READ(NV04_PFIFO_CACHE1_DMA_DCOUNT) << 16);
+	RAMFC_WR(DMA_INSTANCE     , tmp);
+
+	RAMFC_WR(DMA_STATE        , NV_READ(NV04_PFIFO_CACHE1_DMA_STATE));
+	RAMFC_WR(DMA_FETCH	  , NV_READ(NV04_PFIFO_CACHE1_DMA_FETCH));
+	RAMFC_WR(ENGINE           , NV_READ(NV04_PFIFO_CACHE1_ENGINE));
+	RAMFC_WR(PULL1_ENGINE     , NV_READ(NV04_PFIFO_CACHE1_PULL1));
+
+	if (dev_priv->chipset >= 0x17) {
+		RAMFC_WR(ACQUIRE_VALUE,
+			 NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_VALUE));
+		RAMFC_WR(ACQUIRE_TIMESTAMP,
+			 NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_TIMESTAMP));
+		RAMFC_WR(ACQUIRE_TIMEOUT,
+			 NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_TIMEOUT));
+		RAMFC_WR(SEMAPHORE,
+			 NV_READ(NV10_PFIFO_CACHE1_SEMAPHORE));
+		RAMFC_WR(DMA_SUBROUTINE,
+			 NV_READ(NV04_PFIFO_CACHE1_DMA_GET));
+	}
+
+	return 0;
+}
+
diff --git a/shared-core/nv10_graph.c b/shared-core/nv10_graph.c
index fb189709..d1fe0a54 100644
--- a/shared-core/nv10_graph.c
+++ b/shared-core/nv10_graph.c
@@ -611,7 +611,7 @@ void nouveau_nv10_context_switch(drm_device_t *dev)
 	if (offset > 0) \
 		fifo->pgraph_ctx[offset] = val; \
 	} while (0)
-int nv10_graph_context_create(drm_device_t *dev, int channel) {
+int nv10_graph_create_context(drm_device_t *dev, int channel) {
 	drm_nouveau_private_t *dev_priv = dev->dev_private;
 	struct nouveau_fifo *fifo = &dev_priv->fifos[channel];
 	uint32_t tmp, vramsz;
@@ -663,6 +663,21 @@ int nv10_graph_context_create(drm_device_t *dev, int channel) {
 	return 0;
 }
 
+void nv10_graph_destroy_context(drm_device_t *dev, int channel)
+{
+}
+
+int nv10_graph_load_context(drm_device_t *dev, int channel)
+{
+	DRM_ERROR("stub!\n");
+	return 0;
+}
+
+int nv10_graph_save_context(drm_device_t *dev, int channel)
+{
+	DRM_ERROR("stub!\n");
+	return 0;
+}
 
 int nv10_graph_init(drm_device_t *dev) {
 	drm_nouveau_private_t *dev_priv = dev->dev_private;
diff --git a/shared-core/nv20_graph.c b/shared-core/nv20_graph.c
index 7190fc84..1b8a6727 100644
--- a/shared-core/nv20_graph.c
+++ b/shared-core/nv20_graph.c
@@ -29,7 +29,7 @@
 
 #define NV20_GRCTX_SIZE (3529*4)
 
-int nv20_graph_context_create(drm_device_t *dev, int channel) {
+int nv20_graph_create_context(drm_device_t *dev, int channel) {
 	drm_nouveau_private_t *dev_priv =
 		(drm_nouveau_private_t *)dev->dev_private;
 	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
@@ -47,10 +47,21 @@ int nv20_graph_context_create(drm_device_t *dev, int channel) {
 	INSTANCE_WR(chan->ramin_grctx, 10, channel << 24); /* CTX_USER */
 
 	INSTANCE_WR(dev_priv->ctx_table, channel, nouveau_chip_instance_get(dev, chan->ramin_grctx));
-
 	return 0;
 }
 
+void nv20_graph_destroy_context(drm_device_t *dev, int channel) {
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
+
+	if (chan->ramin_grctx) {
+		nouveau_instmem_free(dev, chan->ramin_grctx);
+		chan->ramin_grctx = NULL;
+	}
+
+	INSTANCE_WR(dev_priv->ctx_table, channel, 0);
+}
+
 static void nv20_graph_rdi(drm_device_t *dev) {
 	drm_nouveau_private_t *dev_priv =
 		(drm_nouveau_private_t *)dev->dev_private;
@@ -65,40 +76,42 @@ static void nv20_graph_rdi(drm_device_t *dev) {
 
 /* Save current context (from PGRAPH) into the channel's context
  */
-static void nv20_graph_context_save_current(drm_device_t *dev, int channel) {
+int nv20_graph_save_context(drm_device_t *dev, int channel) {
 	drm_nouveau_private_t *dev_priv =
 		(drm_nouveau_private_t *)dev->dev_private;
 	uint32_t instance;
 
 	instance = INSTANCE_RD(dev_priv->ctx_table, channel);
 	if (!instance) {
-		return;
+		return DRM_ERR(EINVAL);
 	}
 	if (instance != nouveau_chip_instance_get(dev, dev_priv->fifos[channel].ramin_grctx))
-		DRM_ERROR("nv20_graph_context_save_current : bad instance\n");
+		DRM_ERROR("nv20_graph_save_context : bad instance\n");
 
 	NV_WRITE(NV10_PGRAPH_CHANNEL_CTX_SIZE, instance);
 	NV_WRITE(NV10_PGRAPH_CHANNEL_CTX_POINTER, 2 /* save ctx */);
+	return 0;
 }
 
 
 /* Restore the context for a specific channel into PGRAPH
  */
-static void nv20_graph_context_restore(drm_device_t *dev, int channel) {
+int nv20_graph_load_context(drm_device_t *dev, int channel) {
 	drm_nouveau_private_t *dev_priv =
 		(drm_nouveau_private_t *)dev->dev_private;
 	uint32_t instance;
 
 	instance = INSTANCE_RD(dev_priv->ctx_table, channel);
 	if (!instance) {
-		return;
+		return DRM_ERR(EINVAL);
 	}
 	if (instance != nouveau_chip_instance_get(dev, dev_priv->fifos[channel].ramin_grctx))
-		DRM_ERROR("nv20_graph_context_restore_current : bad instance\n");
+		DRM_ERROR("nv20_graph_load_context_current : bad instance\n");
 
 	NV_WRITE(NV10_PGRAPH_CTX_USER, channel << 24);
 	NV_WRITE(NV10_PGRAPH_CHANNEL_CTX_SIZE, instance);
 	NV_WRITE(NV10_PGRAPH_CHANNEL_CTX_POINTER, 1 /* restore ctx */);
+	return 0;
 }
 
 void nouveau_nv20_context_switch(drm_device_t *dev)
@@ -113,13 +126,13 @@ void nouveau_nv20_context_switch(drm_device_t *dev)
 
 	NV_WRITE(NV04_PGRAPH_FIFO,0x0);
 
-	nv20_graph_context_save_current(dev, channel_old);
+	nv20_graph_save_context(dev, channel_old);
 	
 	nouveau_wait_for_idle(dev);
 
 	NV_WRITE(NV10_PGRAPH_CTX_CONTROL, 0x10000000);
 
-	nv20_graph_context_restore(dev, channel);
+	nv20_graph_load_context(dev, channel);
 
 	nouveau_wait_for_idle(dev);
 	
diff --git a/shared-core/nv30_graph.c b/shared-core/nv30_graph.c
index f4faadd8..7a87990a 100644
--- a/shared-core/nv30_graph.c
+++ b/shared-core/nv30_graph.c
@@ -100,7 +100,7 @@ static void nv30_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
 }
 
 
-int nv30_graph_context_create(drm_device_t *dev, int channel)
+int nv30_graph_create_context(drm_device_t *dev, int channel)
 {
 	drm_nouveau_private_t *dev_priv =
 		(drm_nouveau_private_t *)dev->dev_private;
@@ -132,6 +132,72 @@ int nv30_graph_context_create(drm_device_t *dev, int channel)
 	return 0;
 }
 
+void nv30_graph_destroy_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv =
+		(drm_nouveau_private_t *)dev->dev_private;
+	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
+
+	if (chan->ramin_grctx) {
+		nouveau_instmem_free(dev, chan->ramin_grctx);
+		chan->ramin_grctx = NULL;
+	}
+
+	INSTANCE_WR(dev_priv->ctx_table, channel, 0);
+}
+
+static int
+nouveau_graph_wait_idle(drm_device_t *dev)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	int tv = 1000;
+
+	while (tv--) {
+		if (NV_READ(0x400700) == 0)
+			break;
+	}
+
+	if (NV_READ(0x400700)) {
+		DRM_ERROR("timeout!\n");
+		return DRM_ERR(EBUSY);
+	}
+	return 0;
+}
+
+int nv30_graph_load_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
+	uint32_t inst;
+
+	if (!chan->ramin_grctx)
+		return DRM_ERR(EINVAL);
+	inst = nouveau_chip_instance_get(dev, chan->ramin_grctx);
+
+	NV_WRITE(NV20_PGRAPH_CHANNEL_CTX_POINTER, inst);
+	NV_WRITE(NV20_PGRAPH_CHANNEL_CTX_XFER,
+		 NV20_PGRAPH_CHANNEL_CTX_XFER_LOAD);
+
+	return nouveau_graph_wait_idle(dev);
+}
+
+int nv30_graph_save_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
+	uint32_t inst;
+
+	if (!chan->ramin_grctx)
+		return DRM_ERR(EINVAL);
+	inst = nouveau_chip_instance_get(dev, chan->ramin_grctx);
+
+	NV_WRITE(NV20_PGRAPH_CHANNEL_CTX_POINTER, inst);
+	NV_WRITE(NV20_PGRAPH_CHANNEL_CTX_XFER,
+		 NV20_PGRAPH_CHANNEL_CTX_XFER_SAVE);
+
+	return nouveau_graph_wait_idle(dev);
+}
+
 int nv30_graph_init(drm_device_t *dev)
 {
 	drm_nouveau_private_t *dev_priv =
diff --git a/shared-core/nv40_fifo.c b/shared-core/nv40_fifo.c
new file mode 100644
index 00000000..6f25349c
--- /dev/null
+++ b/shared-core/nv40_fifo.c
@@ -0,0 +1,189 @@
+/*
+ * Copyright (C) 2007 Ben Skeggs.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_drm.h"
+
+#define RAMFC_WR(offset, val)	NV_WI32(fifoctx + NV40_RAMFC_##offset, (val))
+#define RAMFC_RD(offset)	NV_RI32(fifoctx + NV40_RAMFC_##offset)
+#define NV40_RAMFC(c) (dev_priv->ramfc_offset + ((c)*NV40_RAMFC__SIZE))
+#define NV40_RAMFC__SIZE 128
+
+int
+nv40_fifo_create_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
+	uint32_t fifoctx = NV40_RAMFC(channel), grctx, pushbuf;
+	int i;
+
+	for (i = 0; i < NV40_RAMFC__SIZE; i+=4)
+		NV_WI32(fifoctx + i, 0);
+
+	grctx   = nouveau_chip_instance_get(dev, chan->ramin_grctx);
+	pushbuf = nouveau_chip_instance_get(dev, chan->cmdbuf_obj->instance);
+
+	/* Fill entries that are seen filled in dumps of nvidia driver just
+	 * after channel's is put into DMA mode
+	 */
+	RAMFC_WR(DMA_PUT       , chan->pushbuf_base);
+	RAMFC_WR(DMA_GET       , chan->pushbuf_base);
+	RAMFC_WR(DMA_INSTANCE  , pushbuf);
+	RAMFC_WR(DMA_FETCH     , NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
+				 NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
+				 NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8 |
+#ifdef __BIG_ENDIAN
+				 NV_PFIFO_CACHE1_BIG_ENDIAN |
+#endif
+				 0x30000000 /* no idea.. */);
+	RAMFC_WR(DMA_SUBROUTINE, 0);
+	RAMFC_WR(GRCTX_INSTANCE, grctx);
+	RAMFC_WR(DMA_TIMESLICE , 0x0001FFFF);
+
+	return 0;
+}
+
+void
+nv40_fifo_destroy_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	uint32_t fifoctx = NV40_RAMFC(channel);
+	int i;
+
+	for (i = 0; i < NV40_RAMFC__SIZE; i+=4)
+		NV_WI32(fifoctx + i, 0);
+}
+
+int
+nv40_fifo_load_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	uint32_t fifoctx = NV40_RAMFC(channel);
+	uint32_t tmp, tmp2;
+
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_GET          , RAMFC_RD(DMA_GET));
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_PUT          , RAMFC_RD(DMA_PUT));
+	NV_WRITE(NV10_PFIFO_CACHE1_REF_CNT          , RAMFC_RD(REF_CNT));
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_INSTANCE     , RAMFC_RD(DMA_INSTANCE));
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_DCOUNT       , RAMFC_RD(DMA_DCOUNT));
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_STATE        , RAMFC_RD(DMA_STATE));
+
+	/* No idea what 0x2058 is.. */
+	tmp   = RAMFC_RD(DMA_FETCH);
+	tmp2  = NV_READ(0x2058) & 0xFFF;
+	tmp2 |= (tmp & 0x30000000);
+	NV_WRITE(0x2058, tmp2);
+	tmp  &= ~0x30000000;
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_FETCH        , tmp);
+
+	NV_WRITE(NV04_PFIFO_CACHE1_ENGINE           , RAMFC_RD(ENGINE));
+	NV_WRITE(NV04_PFIFO_CACHE1_PULL1            , RAMFC_RD(PULL1_ENGINE));
+	NV_WRITE(NV10_PFIFO_CACHE1_ACQUIRE_VALUE    , RAMFC_RD(ACQUIRE_VALUE));
+	NV_WRITE(NV10_PFIFO_CACHE1_ACQUIRE_TIMESTAMP, RAMFC_RD(ACQUIRE_TIMESTAMP));
+	NV_WRITE(NV10_PFIFO_CACHE1_ACQUIRE_TIMEOUT  , RAMFC_RD(ACQUIRE_TIMEOUT));
+	NV_WRITE(NV10_PFIFO_CACHE1_SEMAPHORE        , RAMFC_RD(SEMAPHORE));
+	NV_WRITE(NV10_PFIFO_CACHE1_DMA_SUBROUTINE   , RAMFC_RD(DMA_SUBROUTINE));
+	NV_WRITE(NV40_PFIFO_GRCTX_INSTANCE          , RAMFC_RD(GRCTX_INSTANCE));
+	NV_WRITE(0x32e4, RAMFC_RD(UNK_40));
+	/* NVIDIA does this next line twice... */
+	NV_WRITE(0x32e8, RAMFC_RD(UNK_44));
+	NV_WRITE(0x2088, RAMFC_RD(UNK_4C));
+	NV_WRITE(0x3300, RAMFC_RD(UNK_50));
+
+	/* not sure what part is PUT, and which is GET.. never seen a non-zero
+	 * value appear in a mmio-trace yet..
+	 */
+#if 0
+	tmp = NV_READ(UNK_84);
+	NV_WRITE(NV_PFIFO_CACHE1_GET, tmp ???);
+	NV_WRITE(NV_PFIFO_CACHE1_PUT, tmp ???);
+#endif
+
+	/* Don't clobber the TIMEOUT_ENABLED flag when restoring from RAMFC */
+	tmp  = NV_READ(NV04_PFIFO_DMA_TIMESLICE) & ~0x1FFFF;
+	tmp |= RAMFC_RD(DMA_TIMESLICE) & 0x1FFFF;
+	NV_WRITE(NV04_PFIFO_DMA_TIMESLICE, tmp);
+
+	/* Set channel active, and in DMA mode */
+	NV_WRITE(NV03_PFIFO_CACHE1_PUSH1  , 0x00010000 | channel);
+	/* Reset DMA_CTL_AT_INFO to INVALID */
+	tmp = NV_READ(NV04_PFIFO_CACHE1_DMA_CTL) & ~(1<<31);
+	NV_WRITE(NV04_PFIFO_CACHE1_DMA_CTL, tmp);
+
+	return 0;
+}
+
+int
+nv40_fifo_save_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	uint32_t fifoctx = NV40_RAMFC(channel);
+	uint32_t tmp;
+
+	RAMFC_WR(DMA_PUT          , NV_READ(NV04_PFIFO_CACHE1_DMA_PUT));
+	RAMFC_WR(DMA_GET          , NV_READ(NV04_PFIFO_CACHE1_DMA_GET));
+	RAMFC_WR(REF_CNT          , NV_READ(NV10_PFIFO_CACHE1_REF_CNT));
+	RAMFC_WR(DMA_INSTANCE     , NV_READ(NV04_PFIFO_CACHE1_DMA_INSTANCE));
+	RAMFC_WR(DMA_DCOUNT       , NV_READ(NV04_PFIFO_CACHE1_DMA_DCOUNT));
+	RAMFC_WR(DMA_STATE        , NV_READ(NV04_PFIFO_CACHE1_DMA_STATE));
+
+	tmp  = NV_READ(NV04_PFIFO_CACHE1_DMA_FETCH);
+	tmp |= NV_READ(0x2058) & 0x30000000;
+	RAMFC_WR(DMA_FETCH	  , tmp);
+
+	RAMFC_WR(ENGINE           , NV_READ(NV04_PFIFO_CACHE1_ENGINE));
+	RAMFC_WR(PULL1_ENGINE     , NV_READ(NV04_PFIFO_CACHE1_PULL1));
+	RAMFC_WR(ACQUIRE_VALUE    , NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_VALUE));
+	tmp = NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_TIMESTAMP);
+	RAMFC_WR(ACQUIRE_TIMESTAMP, tmp);
+	RAMFC_WR(ACQUIRE_TIMEOUT  , NV_READ(NV10_PFIFO_CACHE1_ACQUIRE_TIMEOUT));
+	RAMFC_WR(SEMAPHORE        , NV_READ(NV10_PFIFO_CACHE1_SEMAPHORE));
+
+	/* NVIDIA read 0x3228 first, then write DMA_GET here.. maybe something
+	 * more involved depending on the value of 0x3228?
+	 */
+	RAMFC_WR(DMA_SUBROUTINE   , NV_READ(NV04_PFIFO_CACHE1_DMA_GET));
+
+	RAMFC_WR(GRCTX_INSTANCE   , NV_READ(NV40_PFIFO_GRCTX_INSTANCE));
+
+	/* No idea what the below is for exactly, ripped from a mmio-trace */
+	RAMFC_WR(UNK_40           , NV_READ(NV40_PFIFO_UNK32E4));
+
+	/* NVIDIA do this next line twice.. bug? */
+	RAMFC_WR(UNK_44           , NV_READ(0x32e8));
+	RAMFC_WR(UNK_4C           , NV_READ(0x2088));
+	RAMFC_WR(UNK_50           , NV_READ(0x3300));
+
+#if 0 /* no real idea which is PUT/GET in UNK_48.. */
+	tmp  = NV_READ(NV04_PFIFO_CACHE1_GET);
+	tmp |= (NV_READ(NV04_PFIFO_CACHE1_PUT) << 16);
+	RAMFC_WR(UNK_48           , tmp);
+#endif
+
+	return 0;
+}
+
diff --git a/shared-core/nv40_graph.c b/shared-core/nv40_graph.c
index 792734ed..6fb575db 100644
--- a/shared-core/nv40_graph.c
+++ b/shared-core/nv40_graph.c
@@ -1,7 +1,32 @@
+/*
+ * Copyright (C) 2007 Ben Skeggs.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
 #include "drmP.h"
 #include "drm.h"
 #include "nouveau_drv.h"
-#include "nouveau_drm.h"
 
 /* The sizes are taken from the difference between the start of two
  * grctx addresses while running the nvidia driver.  Probably slightly
@@ -11,7 +36,9 @@
 #define NV40_GRCTX_SIZE (175*1024)
 #define NV43_GRCTX_SIZE (70*1024)
 #define NV46_GRCTX_SIZE (70*1024) /* probably ~64KiB */
+#define NV49_GRCTX_SIZE (164640)
 #define NV4A_GRCTX_SIZE (64*1024)
+#define NV4B_GRCTX_SIZE (164640)
 #define NV4C_GRCTX_SIZE (25*1024)
 #define NV4E_GRCTX_SIZE (25*1024)
 
@@ -19,7 +46,8 @@
  *      contexts are taken from dumps just after the 3D object is
  *      created.
  */
-static void nv40_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
+static void
+nv40_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
 {
 	drm_nouveau_private_t *dev_priv = dev->dev_private;
 	int i;
@@ -275,7 +303,8 @@ nv43_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
 		INSTANCE_WR(ctx, i/4, 0x3f800000);
 };
 
-static void nv46_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
+static void
+nv46_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
 {
 	drm_nouveau_private_t *dev_priv = dev->dev_private;
 	int i;
@@ -425,7 +454,231 @@ static void nv46_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
 		INSTANCE_WR(ctx, i/4, 0x3f800000);
 }
 
-static void nv4a_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
+static void
+nv49_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	int i;
+
+	INSTANCE_WR(ctx, 0x00000/4, nouveau_chip_instance_get(dev, ctx));
+	INSTANCE_WR(ctx, 0x00004/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x00008/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x0000c/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x00010/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x00014/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x00018/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x0001c/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x00020/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x000c4/4, 0x0000ffff);
+	INSTANCE_WR(ctx, 0x000c8/4, 0x0000ffff);
+	INSTANCE_WR(ctx, 0x000d0/4, 0x00000001);
+	INSTANCE_WR(ctx, 0x001bc/4, 0x20010001);
+	INSTANCE_WR(ctx, 0x001c0/4, 0x0f73ef00);
+	INSTANCE_WR(ctx, 0x001c8/4, 0x02008821);
+	INSTANCE_WR(ctx, 0x00218/4, 0x00000040);
+	INSTANCE_WR(ctx, 0x0021c/4, 0x00000040);
+	INSTANCE_WR(ctx, 0x00220/4, 0x00000040);
+	INSTANCE_WR(ctx, 0x00228/4, 0x00000040);
+	INSTANCE_WR(ctx, 0x00234/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x00238/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x0023c/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x00240/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x00244/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x00248/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x0024c/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x00250/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x00270/4, 0x0b0b0b0c);
+	INSTANCE_WR(ctx, 0x003e0/4, 0x00040000);
+	INSTANCE_WR(ctx, 0x003f0/4, 0x55555555);
+	INSTANCE_WR(ctx, 0x003f4/4, 0x55555555);
+	INSTANCE_WR(ctx, 0x003f8/4, 0x55555555);
+	INSTANCE_WR(ctx, 0x003fc/4, 0x55555555);
+	INSTANCE_WR(ctx, 0x00428/4, 0x00000008);
+	INSTANCE_WR(ctx, 0x0043c/4, 0x00001010);
+	INSTANCE_WR(ctx, 0x00460/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00464/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00468/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x0046c/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00470/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00474/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00478/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x0047c/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00480/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00484/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00488/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x0048c/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00490/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00494/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00498/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x0049c/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x004f4/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x004f8/4, 0x00080060);
+	INSTANCE_WR(ctx, 0x00514/4, 0x00000080);
+	INSTANCE_WR(ctx, 0x00518/4, 0xffff0000);
+	INSTANCE_WR(ctx, 0x0051c/4, 0x00000001);
+	INSTANCE_WR(ctx, 0x00530/4, 0x46400000);
+	INSTANCE_WR(ctx, 0x00540/4, 0xffff0000);
+	INSTANCE_WR(ctx, 0x00544/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00548/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x0054c/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00550/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00554/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00558/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x0055c/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00560/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00564/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00568/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x0056c/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00570/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00574/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00578/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x0057c/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00580/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00594/4, 0x0fff0000);
+	INSTANCE_WR(ctx, 0x00598/4, 0x0fff0000);
+	INSTANCE_WR(ctx, 0x005a0/4, 0x00011100);
+	INSTANCE_WR(ctx, 0x005bc/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005c0/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005c4/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005c8/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005cc/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005d0/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005d4/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005d8/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005dc/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005e0/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005e4/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005e8/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005ec/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005f0/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005f4/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005f8/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x00604/4, 0x4b7fffff);
+	INSTANCE_WR(ctx, 0x0062c/4, 0x30201000);
+	INSTANCE_WR(ctx, 0x00630/4, 0x70605040);
+	INSTANCE_WR(ctx, 0x00634/4, 0xb8a89888);
+	INSTANCE_WR(ctx, 0x00638/4, 0xf8e8d8c8);
+	INSTANCE_WR(ctx, 0x0064c/4, 0x40100000);
+	INSTANCE_WR(ctx, 0x00668/4, 0x0000ffff);
+	INSTANCE_WR(ctx, 0x0069c/4, 0x435185d6);
+	INSTANCE_WR(ctx, 0x006a0/4, 0x2155b699);
+	INSTANCE_WR(ctx, 0x006a4/4, 0xfedcba98);
+	INSTANCE_WR(ctx, 0x006a8/4, 0x00000098);
+	INSTANCE_WR(ctx, 0x006b8/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x006bc/4, 0x00ff7000);
+	INSTANCE_WR(ctx, 0x006c0/4, 0x0000ffff);
+	INSTANCE_WR(ctx, 0x006d0/4, 0x00ff0000);
+	INSTANCE_WR(ctx, 0x0070c/4, 0x00ffff00);
+	for (i=0x00750; i<=0x0078c; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x00018488);
+	for (i=0x00790; i<=0x007cc; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x00028202);
+	for (i=0x00810; i<=0x0084c; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x0000aae4);
+	for (i=0x00850; i<=0x0088c; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x01012000);
+	for (i=0x00890; i<=0x008cc; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x00080008);
+	for (i=0x00910; i<=0x0094c; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x00100008);
+	for (i=0x009a0; i<=0x009ac; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x0001bc80);
+	for (i=0x009b0; i<=0x009bc; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x00000202);
+	for (i=0x009d0; i<=0x009dc; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x00000008);
+	for (i=0x009f0; i<=0x009fc; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x00080008);
+	INSTANCE_WR(ctx, 0x00a10/4, 0x00000002);
+	INSTANCE_WR(ctx, 0x00a44/4, 0x00000421);
+	INSTANCE_WR(ctx, 0x00a48/4, 0x030c30c3);
+	INSTANCE_WR(ctx, 0x00a54/4, 0x3e020200);
+	INSTANCE_WR(ctx, 0x00a58/4, 0x00ffffff);
+	INSTANCE_WR(ctx, 0x00a5c/4, 0x20103f00);
+	INSTANCE_WR(ctx, 0x00a68/4, 0x00040000);
+	INSTANCE_WR(ctx, 0x00aa0/4, 0x00008100);
+	INSTANCE_WR(ctx, 0x00b2c/4, 0x00000001);
+	INSTANCE_WR(ctx, 0x00b70/4, 0x00001001);
+	INSTANCE_WR(ctx, 0x00b7c/4, 0x00000003);
+	INSTANCE_WR(ctx, 0x00b80/4, 0x00888001);
+	INSTANCE_WR(ctx, 0x00bb0/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bb4/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bb8/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bbc/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bc0/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bc4/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bc8/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bcc/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bd0/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bd4/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bd8/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bdc/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00be0/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00be4/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00be8/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bec/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bf0/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bf4/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bf8/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bfc/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c00/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c04/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c08/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c0c/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c10/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c14/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c18/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c1c/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c20/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c24/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c28/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c2c/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c54/4, 0x00000005);
+	INSTANCE_WR(ctx, 0x00c60/4, 0x0000ffff);
+	INSTANCE_WR(ctx, 0x00c7c/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c80/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c84/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c88/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c8c/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c90/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c94/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c98/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c9c/4, 0x00000001);
+	INSTANCE_WR(ctx, 0x00cd4/4, 0x00000001);
+	INSTANCE_WR(ctx, 0x00cd8/4, 0x08e00001);
+	INSTANCE_WR(ctx, 0x00cdc/4, 0x000e3000);
+	for(i=0x030a0; i<=0x03118; i+=8)
+		INSTANCE_WR(ctx, i/4, 0x3f800000);
+	for(i=0x098a0; i<=0x0ba90; i+=24)
+		INSTANCE_WR(ctx, i/4, 0x00000001);
+	for(i=0x0baa0; i<=0x0be90; i+=16)
+		INSTANCE_WR(ctx, i/4, 0x3f800000);
+	for(i=0x0e2e0; i<=0x0fff0; i+=24)
+		INSTANCE_WR(ctx, i/4, 0x00000001);
+	for(i=0x10008; i<=0x104d0; i+=24)
+		INSTANCE_WR(ctx, i/4, 0x00000001);
+	for(i=0x104e0; i<=0x108d0; i+=16)
+		INSTANCE_WR(ctx, i/4, 0x3f800000);
+	for(i=0x12d20; i<=0x14f10; i+=24)
+		INSTANCE_WR(ctx, i/4, 0x00000001);
+	for(i=0x14f20; i<=0x15310; i+=16)
+		INSTANCE_WR(ctx, i/4, 0x3f800000);
+	for(i=0x17760; i<=0x19950; i+=24)
+		INSTANCE_WR(ctx, i/4, 0x00000001);
+	for(i=0x19960; i<=0x19d50; i+=16)
+		INSTANCE_WR(ctx, i/4, 0x3f800000);
+	for(i=0x1c1a0; i<=0x1e390; i+=24)
+		INSTANCE_WR(ctx, i/4, 0x00000001);
+	for(i=0x1e3a0; i<=0x1e790; i+=16)
+		INSTANCE_WR(ctx, i/4, 0x3f800000);
+	for(i=0x20be0; i<=0x22dd0; i+=24)
+		INSTANCE_WR(ctx, i/4, 0x00000001);
+	for(i=0x22de0; i<=0x231d0; i+=16)
+		INSTANCE_WR(ctx, i/4, 0x3f800000);
+}
+
+static void
+nv4a_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
 {
 	drm_nouveau_private_t *dev_priv = dev->dev_private;
 	int i;
@@ -541,8 +794,223 @@ static void nv4a_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
 		INSTANCE_WR(ctx, i/4, 0x3f800000);
 }
 
+static void
+nv4b_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	int i;
+
+	INSTANCE_WR(ctx, 0x00000/4, nouveau_chip_instance_get(dev, ctx));
+	INSTANCE_WR(ctx, 0x00004/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x00008/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x0000c/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x00010/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x00014/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x00018/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x0001c/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x00020/4, 0x0000c040);
+	INSTANCE_WR(ctx, 0x000c4/4, 0x0000ffff);
+	INSTANCE_WR(ctx, 0x000c8/4, 0x0000ffff);
+	INSTANCE_WR(ctx, 0x000d0/4, 0x00000001);
+	INSTANCE_WR(ctx, 0x001bc/4, 0x20010001);
+	INSTANCE_WR(ctx, 0x001c0/4, 0x0f73ef00);
+	INSTANCE_WR(ctx, 0x001c8/4, 0x02008821);
+	INSTANCE_WR(ctx, 0x00218/4, 0x00000040);
+	INSTANCE_WR(ctx, 0x0021c/4, 0x00000040);
+	INSTANCE_WR(ctx, 0x00220/4, 0x00000040);
+	INSTANCE_WR(ctx, 0x00228/4, 0x00000040);
+	INSTANCE_WR(ctx, 0x00234/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x00238/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x0023c/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x00240/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x00244/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x00248/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x0024c/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x00250/4, 0x80000000);
+	INSTANCE_WR(ctx, 0x00270/4, 0x0b0b0b0c);
+	INSTANCE_WR(ctx, 0x003e0/4, 0x00040000);
+	INSTANCE_WR(ctx, 0x003f0/4, 0x55555555);
+	INSTANCE_WR(ctx, 0x003f4/4, 0x55555555);
+	INSTANCE_WR(ctx, 0x003f8/4, 0x55555555);
+	INSTANCE_WR(ctx, 0x003fc/4, 0x55555555);
+	INSTANCE_WR(ctx, 0x00428/4, 0x00000008);
+	INSTANCE_WR(ctx, 0x0043c/4, 0x00001010);
+	INSTANCE_WR(ctx, 0x00460/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00464/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00468/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x0046c/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00470/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00474/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00478/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x0047c/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00480/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00484/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00488/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x0048c/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00490/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00494/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x00498/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x0049c/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x004f4/4, 0x00000111);
+	INSTANCE_WR(ctx, 0x004f8/4, 0x00080060);
+	INSTANCE_WR(ctx, 0x00514/4, 0x00000080);
+	INSTANCE_WR(ctx, 0x00518/4, 0xffff0000);
+	INSTANCE_WR(ctx, 0x0051c/4, 0x00000001);
+	INSTANCE_WR(ctx, 0x00530/4, 0x46400000);
+	INSTANCE_WR(ctx, 0x00540/4, 0xffff0000);
+	INSTANCE_WR(ctx, 0x00544/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00548/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x0054c/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00550/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00554/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00558/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x0055c/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00560/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00564/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00568/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x0056c/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00570/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00574/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00578/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x0057c/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00580/4, 0x88888888);
+	INSTANCE_WR(ctx, 0x00594/4, 0x0fff0000);
+	INSTANCE_WR(ctx, 0x00598/4, 0x0fff0000);
+	INSTANCE_WR(ctx, 0x005a0/4, 0x00011100);
+	INSTANCE_WR(ctx, 0x005bc/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005c0/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005c4/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005c8/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005cc/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005d0/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005d4/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005d8/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005dc/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005e0/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005e4/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005e8/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005ec/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005f0/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005f4/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x005f8/4, 0x07ff0000);
+	INSTANCE_WR(ctx, 0x00604/4, 0x4b7fffff);
+	INSTANCE_WR(ctx, 0x0062c/4, 0x30201000);
+	INSTANCE_WR(ctx, 0x00630/4, 0x70605040);
+	INSTANCE_WR(ctx, 0x00634/4, 0xb8a89888);
+	INSTANCE_WR(ctx, 0x00638/4, 0xf8e8d8c8);
+	INSTANCE_WR(ctx, 0x0064c/4, 0x40100000);
+	INSTANCE_WR(ctx, 0x00668/4, 0x0000ffff);
+	INSTANCE_WR(ctx, 0x0069c/4, 0x435185d6);
+	INSTANCE_WR(ctx, 0x006a0/4, 0x2155b699);
+	INSTANCE_WR(ctx, 0x006a4/4, 0xfedcba98);
+	INSTANCE_WR(ctx, 0x006a8/4, 0x00000098);
+	INSTANCE_WR(ctx, 0x006b8/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x006bc/4, 0x00ff7000);
+	INSTANCE_WR(ctx, 0x006c0/4, 0x0000ffff);
+	INSTANCE_WR(ctx, 0x006d0/4, 0x00ff0000);
+	INSTANCE_WR(ctx, 0x0070c/4, 0x00ffff00);
+	for (i=0x00750; i<=0x0078c; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x00018488);
+	for (i=0x00790; i<=0x007cc; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x00028202);
+	for (i=0x00810; i<=0x0084c; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x0000aae4);
+	for (i=0x00850; i<=0x0088c; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x01012000);
+	for (i=0x00890; i<=0x008cc; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x00080008);
+	for (i=0x00910; i<=0x0094c; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x00100008);
+	for (i=0x009a0; i<=0x009ac; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x0001bc80);
+	for (i=0x009b0; i<=0x009bc; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x00000202);
+	for (i=0x009d0; i<=0x009dc; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x00000008);
+	for (i=0x009f0; i<=0x009fc; i+=4)
+		INSTANCE_WR(ctx, i/4, 0x00080008);
+	INSTANCE_WR(ctx, 0x00a10/4, 0x00000002);
+	INSTANCE_WR(ctx, 0x00a44/4, 0x00000421);
+	INSTANCE_WR(ctx, 0x00a48/4, 0x030c30c3);
+	INSTANCE_WR(ctx, 0x00a54/4, 0x3e020200);
+	INSTANCE_WR(ctx, 0x00a58/4, 0x00ffffff);
+	INSTANCE_WR(ctx, 0x00a5c/4, 0x20103f00);
+	INSTANCE_WR(ctx, 0x00a68/4, 0x00040000);
+	INSTANCE_WR(ctx, 0x00aa0/4, 0x00008100);
+	INSTANCE_WR(ctx, 0x00b2c/4, 0x00000001);
+	INSTANCE_WR(ctx, 0x00b70/4, 0x00001001);
+	INSTANCE_WR(ctx, 0x00b7c/4, 0x00000003);
+	INSTANCE_WR(ctx, 0x00b80/4, 0x00888001);
+	INSTANCE_WR(ctx, 0x00bb0/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bb4/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bb8/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bbc/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bc0/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bc4/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bc8/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bcc/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bd0/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bd4/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bd8/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bdc/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00be0/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00be4/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00be8/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bec/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bf0/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bf4/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bf8/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00bfc/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c00/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c04/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c08/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c0c/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c10/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c14/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c18/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c1c/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c20/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c24/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c28/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c2c/4, 0xffffffff);
+	INSTANCE_WR(ctx, 0x00c54/4, 0x00000005);
+	INSTANCE_WR(ctx, 0x00c60/4, 0x0000ffff);
+	INSTANCE_WR(ctx, 0x00c7c/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c80/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c84/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c88/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c8c/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c90/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c94/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c98/4, 0x00005555);
+	INSTANCE_WR(ctx, 0x00c9c/4, 0x00000001);
+	INSTANCE_WR(ctx, 0x00cd4/4, 0x00000001);
+	INSTANCE_WR(ctx, 0x00cd8/4, 0x08e00001);
+	INSTANCE_WR(ctx, 0x00cdc/4, 0x000e3000);
+	for(i=0x030a0; i<=0x03118; i+=8)
+		INSTANCE_WR(ctx, i/4, 0x3f800000);
+	for(i=0x098a0; i<=0x0ba90; i+=24)
+		INSTANCE_WR(ctx, i/4, 0x00000001);
+	for(i=0x0baa0; i<=0x0be90; i+=16)
+		INSTANCE_WR(ctx, i/4, 0x3f800000);
+	for(i=0x0e2e0; i<=0x0fff0; i+=24)
+		INSTANCE_WR(ctx, i/4, 0x00000001);
+	for(i=0x10008; i<=0x104d0; i+=24)
+		INSTANCE_WR(ctx, i/4, 0x00000001);
+	for(i=0x104e0; i<=0x108d0; i+=16)
+		INSTANCE_WR(ctx, i/4, 0x3f800000);
+	for(i=0x12d20; i<=0x14f10; i+=24)
+		INSTANCE_WR(ctx, i/4, 0x00000001);
+	for(i=0x14f20; i<=0x15310; i+=16)
+		INSTANCE_WR(ctx, i/4, 0x3f800000);
+	for(i=0x17760; i<=0x19950; i+=24)
+		INSTANCE_WR(ctx, i/4, 0x00000001);
+	for(i=0x19960; i<=0x19d50; i+=16)
+		INSTANCE_WR(ctx, i/4, 0x3f800000);
+}
 
-static void nv4c_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
+static void
+nv4c_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
 {
 	drm_nouveau_private_t *dev_priv = dev->dev_private;
 	int i;
@@ -648,7 +1116,8 @@ static void nv4c_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
 		INSTANCE_WR(ctx, i/4, 0x3f800000);
 }
 
-static void nv4e_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
+static void
+nv4e_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
 {
 	drm_nouveau_private_t *dev_priv = dev->dev_private;
 	int i;
@@ -755,7 +1224,7 @@ static void nv4e_graph_context_init(drm_device_t *dev, struct mem_block *ctx)
 }
 
 int
-nv40_graph_context_create(drm_device_t *dev, int channel)
+nv40_graph_create_context(drm_device_t *dev, int channel)
 {
 	drm_nouveau_private_t *dev_priv =
 		(drm_nouveau_private_t *)dev->dev_private;
@@ -777,10 +1246,18 @@ nv40_graph_context_create(drm_device_t *dev, int channel)
 		ctx_size = NV46_GRCTX_SIZE;
 		ctx_init = nv46_graph_context_init;
 		break;
+	case 0x49:
+		ctx_size = NV49_GRCTX_SIZE;
+		ctx_init = nv49_graph_context_init;
+		break;
 	case 0x4a:
 		ctx_size = NV4A_GRCTX_SIZE;
 		ctx_init = nv4a_graph_context_init;
 		break;
+	case 0x4b:
+		ctx_size = NV4B_GRCTX_SIZE;
+		ctx_init = nv4b_graph_context_init;
+		break;
 	case 0x4c:
 		ctx_size = NV4C_GRCTX_SIZE;
 		ctx_init = nv4c_graph_context_init;
@@ -808,89 +1285,99 @@ nv40_graph_context_create(drm_device_t *dev, int channel)
 	return 0;
 }
 
-/* Save current context (from PGRAPH) into the channel's context
- *XXX: fails sometimes, not sure why..
- */
 void
-nv40_graph_context_save_current(drm_device_t *dev)
+nv40_graph_destroy_context(drm_device_t *dev, int channel)
 {
-	drm_nouveau_private_t *dev_priv =
-		(drm_nouveau_private_t *)dev->dev_private;
-	uint32_t instance;
-	int i;
-
-	NV_WRITE(NV04_PGRAPH_FIFO, 0);
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
 
-	instance = NV_READ(0x40032C) & 0xFFFFF;
-	if (!instance) {
-		NV_WRITE(NV04_PGRAPH_FIFO, 1);
-		return;
+	if (chan->ramin_grctx) {
+		nouveau_instmem_free(dev, chan->ramin_grctx);
+		chan->ramin_grctx = NULL;
 	}
+}
+
+static int
+nv40_graph_transfer_context(drm_device_t *dev, uint32_t inst, int save)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	uint32_t old_cp, tv = 1000;
+	int i;
 
-	NV_WRITE(0x400784, instance);
-	NV_WRITE(0x400310, NV_READ(0x400310) | 0x20);
-	NV_WRITE(0x400304, 1);
-	/* just in case, we don't want to spin in-kernel forever */
-	for (i=0; i<1000; i++) {
-		if (NV_READ(0x40030C) == 0)
+	old_cp = NV_READ(NV20_PGRAPH_CHANNEL_CTX_POINTER);
+	NV_WRITE(NV20_PGRAPH_CHANNEL_CTX_POINTER, inst);
+	NV_WRITE(NV40_PGRAPH_CTXCTL_0310,
+		 save ? NV40_PGRAPH_CTXCTL_0310_XFER_SAVE :
+		 	NV40_PGRAPH_CTXCTL_0310_XFER_LOAD);
+	NV_WRITE(NV40_PGRAPH_CTXCTL_0304, NV40_PGRAPH_CTXCTL_0304_XFER_CTX);
+
+	for (i = 0; i < tv; i++) {
+		if (NV_READ(NV40_PGRAPH_CTXCTL_030C) == 0)
 			break;
 	}
-	if (i==1000) {
-		DRM_ERROR("failed to save current grctx to ramin\n");
-		DRM_ERROR("instance = 0x%08x\n", NV_READ(0x40032C));
-		DRM_ERROR("0x40030C = 0x%08x\n", NV_READ(0x40030C));
-		NV_WRITE(NV04_PGRAPH_FIFO, 1);
-		return;
+	NV_WRITE(NV20_PGRAPH_CHANNEL_CTX_POINTER, old_cp);
+
+	if (i == tv) {
+		DRM_ERROR("failed: inst=0x%08x save=%d\n", inst, save);
+		DRM_ERROR("0x40030C = 0x%08x\n",
+			  NV_READ(NV40_PGRAPH_CTXCTL_030C));
+		return DRM_ERR(EBUSY);
 	}
 
-	NV_WRITE(NV04_PGRAPH_FIFO, 1);
+	return 0;
+}
+
+/* Save current context (from PGRAPH) into the channel's context
+ *XXX: fails sometimes, not sure why..
+ */
+int
+nv40_graph_save_context(drm_device_t *dev, int channel)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
+	uint32_t inst;
+
+	if (!chan->ramin_grctx)
+		return DRM_ERR(EINVAL);
+	inst = nouveau_chip_instance_get(dev, chan->ramin_grctx);
+
+	return nv40_graph_transfer_context(dev, inst, 1);
 }
 
 /* Restore the context for a specific channel into PGRAPH
  * XXX: fails sometimes.. not sure why
  */
-void
-nv40_graph_context_restore(drm_device_t *dev, int channel)
+int
+nv40_graph_load_context(drm_device_t *dev, int channel)
 {
-	drm_nouveau_private_t *dev_priv =
-		(drm_nouveau_private_t *)dev->dev_private;
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
 	struct nouveau_fifo *chan = &dev_priv->fifos[channel];
-	uint32_t instance;
-	int i;
+	uint32_t inst;
+	int ret;
 
-	instance = nouveau_chip_instance_get(dev, chan->ramin_grctx);
-
-	NV_WRITE(NV04_PGRAPH_FIFO, 0);
-	NV_WRITE(0x400784, instance);
-	NV_WRITE(0x400310, NV_READ(0x400310) | 0x40);
-	NV_WRITE(0x400304, 1);
-	/* just in case, we don't want to spin in-kernel forever */
-	for (i=0; i<1000; i++) {
-		if (NV_READ(0x40030C) == 0)
-			break;
-	}
-	if (i==1000) {
-		DRM_ERROR("failed to restore grctx for ch%d to PGRAPH\n",
-				channel);
-		DRM_ERROR("instance = 0x%08x\n", instance);
-		DRM_ERROR("0x40030C = 0x%08x\n", NV_READ(0x40030C));
-		NV_WRITE(NV04_PGRAPH_FIFO, 1);
-		return;
-	}
+	if (!chan->ramin_grctx)
+		return DRM_ERR(EINVAL);
+	inst = nouveau_chip_instance_get(dev, chan->ramin_grctx);
 
+	ret = nv40_graph_transfer_context(dev, inst, 0);
+	if (ret)
+		return ret;
 
 	/* 0x40032C, no idea of it's exact function.  Could simply be a
 	 * record of the currently active PGRAPH context.  It's currently
 	 * unknown as to what bit 24 does.  The nv ddx has it set, so we will
 	 * set it here too.
 	 */
-	NV_WRITE(0x40032C, instance | 0x01000000);
+	NV_WRITE(NV20_PGRAPH_CHANNEL_CTX_POINTER, inst);
+	NV_WRITE(NV40_PGRAPH_CTXCTL_CUR,
+		 (inst & NV40_PGRAPH_CTXCTL_CUR_INST_MASK) |
+		  NV40_PGRAPH_CTXCTL_CUR_LOADED);
 	/* 0x32E0 records the instance address of the active FIFO's PGRAPH
 	 * context.  If at any time this doesn't match 0x40032C, you will
 	 * recieve PGRAPH_INTR_CONTEXT_SWITCH
 	 */
-	NV_WRITE(NV40_PFIFO_GRCTX_INSTANCE, instance);
-	NV_WRITE(NV04_PGRAPH_FIFO, 1);
+	NV_WRITE(NV40_PFIFO_GRCTX_INSTANCE, inst);
+	return 0;
 }
 
 /* Some voodoo that makes context switching work without the binary driver
@@ -1007,6 +1494,39 @@ static uint32_t nv46_ctx_voodoo[] = {
 	0x00600009, 0x00700005, 0x00700006, 0x0060000e, ~0
 };
 
+//this is used for nv49 and nv4b
+static uint32_t nv49_4b_ctx_voodoo[] ={
+	0x00400564, 0x00400505, 0x00408165, 0x00408206, 0x00409e68, 0x00200020,
+	0x0060000a, 0x00700080, 0x00104042, 0x00200020, 0x0060000a, 0x00700000,
+	0x001040c5, 0x00400f26, 0x00401068, 0x0060000d, 0x0070008f, 0x0070000e,
+	0x00408d68, 0x004015e6, 0x007000a0, 0x00700080, 0x0040180f, 0x00700000,
+	0x00200029, 0x0060000a, 0x0011814d, 0x00110158, 0x00105401, 0x0020003a,
+	0x00100051, 0x001040c5, 0x0010c1c4, 0x001041c9, 0x0010c1dc, 0x00150210,
+	0x0012c225, 0x00108238, 0x0010823e, 0x001242c0, 0x00200040, 0x00100280,
+	0x00128100, 0x00128120, 0x00128143, 0x0011415f, 0x0010815c, 0x0010c140,
+	0x00104029, 0x00110400, 0x00104d12, 0x00500060, 0x004071e6, 0x00200118,
+	0x0060000a, 0x00200020, 0x00100620, 0x00154650, 0x00104668, 0x0017466d,
+	0x0011068b, 0x00168691, 0x001046ae, 0x001046b0, 0x001206b4, 0x001046c4,
+	0x001146c6, 0x00200022, 0x001006cc, 0x001246f0, 0x002000c0, 0x00100700,
+	0x0010c3d7, 0x001043e1, 0x00500060, 0x00200290, 0x0060000a, 0x00104800,
+	0x00108901, 0x00124920, 0x0020001f, 0x00100940, 0x00140965, 0x00144a00,
+	0x00104a19, 0x0010ca1c, 0x00110b00, 0x00200028, 0x00100b08, 0x00134c2e,
+	0x0010cd00, 0x0010cd04, 0x00120d08, 0x00104d80, 0x00104e00, 0x0012d600,
+	0x00105c00, 0x00104f06, 0x00105406, 0x00105709, 0x00200340, 0x0060000a,
+	0x00300000, 0x00200680, 0x00406a0f, 0x00200684, 0x00800001, 0x00200b88,
+	0x0060000a, 0x00209540, 0x0040708a, 0x00201350, 0x00800041, 0x00407c0f,
+	0x00600006, 0x00407ce6, 0x00700080, 0x002000a2, 0x0060000a, 0x00104280,
+	0x00200340, 0x0060000a, 0x00200004, 0x00800001, 0x0070008e, 0x00408d68,
+	0x0040020f, 0x00600006, 0x00409e68, 0x00600007, 0x0070000f, 0x0070000e,
+	0x00408d68, 0x0091a880, 0x00901ffe, 0x10940000, 0x00200020, 0x0060000b,
+	0x00500069, 0x0060000c, 0x00401568, 0x00700000, 0x00200001, 0x0040910e,
+	0x00200021, 0x0060000a, 0x00409b0d, 0x00104a40, 0x00104a50, 0x00104a60,
+	0x00104a70, 0x00104a80, 0x00104a90, 0x00104aa0, 0x00104ab0, 0x00407e0e,
+	0x0040130f, 0x00408568, 0x0040a006, 0x0040a105, 0x00600009, 0x00700005,
+	0x00700006, 0x0060000e, ~0
+};
+
+
 static uint32_t nv4a_ctx_voodoo[] = {
 	0x00400889, 0x00200000, 0x0060000a, 0x00200000, 0x00300000, 0x00800001, 
 	0x00700009, 0x0060000e, 0x00400d64, 0x00400d05, 0x00409965, 0x00409e06, 
@@ -1100,7 +1620,9 @@ nv40_graph_init(drm_device_t *dev)
 	case 0x40: ctx_voodoo = nv40_ctx_voodoo; break;
 	case 0x43: ctx_voodoo = nv43_ctx_voodoo; break;
 	case 0x46: ctx_voodoo = nv46_ctx_voodoo; break;
+	case 0x49: ctx_voodoo = nv49_4b_ctx_voodoo; break;
 	case 0x4a: ctx_voodoo = nv4a_ctx_voodoo; break;
+	case 0x4b: ctx_voodoo = nv49_4b_ctx_voodoo; break;
 	case 0x4e: ctx_voodoo = nv4e_ctx_voodoo; break;
 	default:
 		DRM_ERROR("Unknown ctx_voodoo for chipset 0x%02x\n",
@@ -1114,15 +1636,15 @@ nv40_graph_init(drm_device_t *dev)
 		DRM_DEBUG("Loading context-switch voodoo\n");
 		i = 0;
 
-		NV_WRITE(0x400324, 0);
+		NV_WRITE(NV40_PGRAPH_CTXCTL_UCODE_INDEX, 0);
 		while (ctx_voodoo[i] != ~0) {
-			NV_WRITE(0x400328, ctx_voodoo[i]);
+			NV_WRITE(NV40_PGRAPH_CTXCTL_UCODE_DATA, ctx_voodoo[i]);
 			i++;
 		}
 	}	
 
 	/* No context present currently */
-	NV_WRITE(0x40032C, 0x00000000);
+	NV_WRITE(NV40_PGRAPH_CTXCTL_CUR, 0x00000000);
 
 	NV_WRITE(NV03_PGRAPH_INTR_EN, 0x00000000);
 	NV_WRITE(NV03_PGRAPH_INTR   , 0xFFFFFFFF);
diff --git a/shared-core/nv50_fifo.c b/shared-core/nv50_fifo.c
new file mode 100644
index 00000000..e5d37949
--- /dev/null
+++ b/shared-core/nv50_fifo.c
@@ -0,0 +1,83 @@
+/*
+ * Copyright (C) 2007 Ben Skeggs.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "drmP.h"
+#include "drm.h"
+#include "nouveau_drv.h"
+
+static void
+nv50_fifo_init_reset(drm_device_t *dev)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	uint32_t pmc_e;
+
+	pmc_e = NV_READ(NV03_PMC_ENABLE);
+	NV_WRITE(NV03_PMC_ENABLE, pmc_e & ~NV_PMC_ENABLE_PFIFO);
+	pmc_e = NV_READ(NV03_PMC_ENABLE);
+	NV_WRITE(NV03_PMC_ENABLE, pmc_e |  NV_PMC_ENABLE_PFIFO);
+}
+
+int
+nv50_fifo_init(drm_device_t *dev)
+{
+	nv50_fifo_init_reset(dev);
+
+	DRM_ERROR("stub!\n");
+	return 0;
+}
+
+void
+nv50_fifo_takedown(drm_device_t *dev)
+{
+	DRM_ERROR("stub!\n");
+}
+
+int
+nv50_fifo_create_context(drm_device_t *dev, int channel)
+{
+	DRM_ERROR("stub!\n");
+	return 0;
+}
+
+void
+nv50_fifo_destroy_context(drm_device_t *dev, int channel)
+{
+}
+
+int
+nv50_fifo_load_context(drm_device_t *dev, int channel)
+{
+	DRM_ERROR("stub!\n");
+	return 0;
+}
+
+int
+nv50_fifo_save_context(drm_device_t *dev, int channel)
+{
+	DRM_ERROR("stub!\n");
+	return 0;
+}
+
diff --git a/shared-core/nv50_graph.c b/shared-core/nv50_graph.c
new file mode 100644
index 00000000..8c3e2b9b
--- /dev/null
+++ b/shared-core/nv50_graph.c
@@ -0,0 +1,84 @@
+/*
+ * Copyright (C) 2007 Ben Skeggs.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "drmP.h"
+#include "drm.h"
+#include "nouveau_drv.h"
+
+static void
+nv50_graph_init_reset(drm_device_t *dev)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+	uint32_t pmc_e;
+
+	pmc_e = NV_READ(NV03_PMC_ENABLE);
+	NV_WRITE(NV03_PMC_ENABLE, pmc_e & ~NV_PMC_ENABLE_PGRAPH);
+	pmc_e = NV_READ(NV03_PMC_ENABLE);
+	NV_WRITE(NV03_PMC_ENABLE, pmc_e |  NV_PMC_ENABLE_PGRAPH);
+}
+
+int 
+nv50_graph_init(drm_device_t *dev)
+{
+	nv50_graph_init_reset(dev);
+
+	DRM_ERROR("stub!\n");
+	return 0;
+}
+
+void
+nv50_graph_takedown(drm_device_t *dev)
+{
+	DRM_ERROR("stub!\n");
+}
+
+int
+nv50_graph_create_context(drm_device_t *dev, int channel)
+{
+	DRM_ERROR("stub!\n");
+	return 0;
+}
+
+void
+nv50_graph_destroy_context(drm_device_t *dev, int channel)
+{
+	DRM_ERROR("stub!\n");
+}
+
+int
+nv50_graph_load_context(drm_device_t *dev, int channel)
+{
+	DRM_ERROR("stub!\n");
+	return 0;
+}
+
+int
+nv50_graph_save_context(drm_device_t *dev, int channel)
+{
+	DRM_ERROR("stub!\n");
+	return 0;
+}
+
diff --git a/shared-core/nv50_mc.c b/shared-core/nv50_mc.c
new file mode 100644
index 00000000..7f7537f0
--- /dev/null
+++ b/shared-core/nv50_mc.c
@@ -0,0 +1,42 @@
+/*
+ * Copyright (C) 2007 Ben Skeggs.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "drmP.h"
+#include "drm.h"
+#include "nouveau_drv.h"
+
+int
+nv50_mc_init(drm_device_t *dev)
+{
+	drm_nouveau_private_t *dev_priv = dev->dev_private;
+
+	NV_WRITE(NV03_PMC_ENABLE, 0xFFFFFFFF);
+	return 0;
+}
+
+void nv50_mc_takedown(drm_device_t *dev)
+{
+}
diff --git a/shared-core/r300_cmdbuf.c b/shared-core/r300_cmdbuf.c
index d3c52d43..0cd5d7e2 100644
--- a/shared-core/r300_cmdbuf.c
+++ b/shared-core/r300_cmdbuf.c
@@ -155,7 +155,8 @@ void r300_init_reg_flags(void)
 	ADD_RANGE(R300_VAP_INPUT_CNTL_0, 2);
 	ADD_RANGE(0x21DC, 1);
 	ADD_RANGE(R300_VAP_UNKNOWN_221C, 1);
-	ADD_RANGE(0x2220, 4);
+	ADD_RANGE(R300_VAP_CLIP_X_0, 4);
+	ADD_RANGE(R300_VAP_PVS_WAITIDLE, 1);
 	ADD_RANGE(R300_VAP_UNKNOWN_2288, 1);
 	ADD_RANGE(R300_VAP_OUTPUT_VTX_FMT_0, 2);
 	ADD_RANGE(R300_VAP_PVS_CNTL_1, 3);
diff --git a/shared-core/r300_reg.h b/shared-core/r300_reg.h
index 0a31f0b9..e59919be 100644
--- a/shared-core/r300_reg.h
+++ b/shared-core/r300_reg.h
@@ -116,6 +116,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #       define R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT (1<<16) /* GUESS */
 
 #define R300_VAP_OUTPUT_VTX_FMT_1           0x2094
+	/* each of the following is 3 bits wide, specifies number
+	   of components */
 #       define R300_VAP_OUTPUT_VTX_FMT_1__TEX_0_COMP_CNT_SHIFT 0
 #       define R300_VAP_OUTPUT_VTX_FMT_1__TEX_1_COMP_CNT_SHIFT 3
 #       define R300_VAP_OUTPUT_VTX_FMT_1__TEX_2_COMP_CNT_SHIFT 6
@@ -299,6 +301,18 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #       define R300_221C_NORMAL                  0x00000000
 #       define R300_221C_CLEAR                   0x0001C000
 
+/* These seem to be per-pixel and per-vertex X and Y clipping planes. The first
+ * plane is per-pixel and the second plane is per-vertex.
+ *
+ * This was determined by experimentation alone but I believe it is correct.
+ *
+ * These registers are called X_QUAD0_1_FL to X_QUAD0_4_FL by glxtest.
+ */
+#define R300_VAP_CLIP_X_0                   0x2220
+#define R300_VAP_CLIP_X_1                   0x2224
+#define R300_VAP_CLIP_Y_0                   0x2228
+#define R300_VAP_CLIP_Y_1                   0x2230
+
 /* gap */
 
 /* Sometimes, END_OF_PKT and 0x2284=0 are the only commands sent between
@@ -322,13 +336,15 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
  * The meaning of the two UNKNOWN fields is obviously not known. However,
  * experiments so far have shown that both *must* point to an instruction
  * inside the vertex program, otherwise the GPU locks up.
+ *
  * fglrx usually sets CNTL_3_UNKNOWN to the end of the program and
- * CNTL_1_UNKNOWN points to instruction where last write to position takes
- * place.
+ * R300_PVS_CNTL_1_POS_END_SHIFT points to instruction where last write to
+ * position takes place.
+ *
  * Most likely this is used to ignore rest of the program in cases
  * where group of verts arent visible. For some reason this "section"
  * is sometimes accepted other instruction that have no relationship with
- *position calculations.
+ * position calculations.
  */
 #define R300_VAP_PVS_CNTL_1                 0x22D0
 #       define R300_PVS_CNTL_1_PROGRAM_START_SHIFT   0
@@ -967,7 +983,6 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
  * first node is stored in NODE_2, the second node is stored in NODE_3.
  *
  * Offsets are relative to the master offset from PFS_CNTL_2.
- * LAST_NODE is set for the last node, and only for the last node.
  */
 #define R300_PFS_NODE_0                     0x4610
 #define R300_PFS_NODE_1                     0x4614
@@ -981,7 +996,6 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #       define R300_PFS_NODE_TEX_OFFSET_MASK     (31 << 12)
 #       define R300_PFS_NODE_TEX_END_SHIFT       17
 #       define R300_PFS_NODE_TEX_END_MASK        (31 << 17)
-/*#       define R300_PFS_NODE_LAST_NODE           (1 << 22) */
 #		define R300_PFS_NODE_OUTPUT_COLOR        (1 << 22)
 #		define R300_PFS_NODE_OUTPUT_DEPTH        (1 << 23)
 
@@ -1591,6 +1605,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #    define R300_EB_UNK1_SHIFT                      24
 #    define R300_EB_UNK1                    (0x80<<24)
 #    define R300_EB_UNK2                        0x0810
+#define R300_PACKET3_3D_DRAW_VBUF_2         0x00003400
 #define R300_PACKET3_3D_DRAW_INDX_2         0x00003600
 
 /* END: Packet 3 commands */
diff --git a/shared-core/radeon_cp.c b/shared-core/radeon_cp.c
index bde98b30..ec2e688b 100644
--- a/shared-core/radeon_cp.c
+++ b/shared-core/radeon_cp.c
@@ -1190,9 +1190,15 @@ static void radeon_cp_init_ring_buffer(drm_device_t * dev,
 	/* Set ring buffer size */
 #ifdef __BIG_ENDIAN
 	RADEON_WRITE(RADEON_CP_RB_CNTL,
-		     dev_priv->ring.size_l2qw | RADEON_BUF_SWAP_32BIT);
+		     RADEON_BUF_SWAP_32BIT |
+		     (dev_priv->ring.fetch_size_l2ow << 18) |
+		     (dev_priv->ring.rptr_update_l2qw << 8) |
+		     dev_priv->ring.size_l2qw);
 #else
-	RADEON_WRITE(RADEON_CP_RB_CNTL, dev_priv->ring.size_l2qw);
+	RADEON_WRITE(RADEON_CP_RB_CNTL,
+		     (dev_priv->ring.fetch_size_l2ow << 18) |
+		     (dev_priv->ring.rptr_update_l2qw << 8) |
+		     dev_priv->ring.size_l2qw);
 #endif
 
 	/* Start with assuming that writeback doesn't work */
@@ -1391,8 +1397,7 @@ static int radeon_do_init_cp(drm_device_t * dev, drm_radeon_init_t * init)
 	DRM_DEBUG("\n");
 
 	/* if we require new memory map but we don't have it fail */
-	if ((dev_priv->flags & RADEON_NEW_MEMMAP) && !dev_priv->new_memmap)
-	{
+	if ((dev_priv->flags & RADEON_NEW_MEMMAP) && !dev_priv->new_memmap) {
 		DRM_ERROR("Cannot initialise DRM on this card\nThis card requires a new X.org DDX for 3D\n");
 		radeon_do_cleanup_cp(dev);
 		return DRM_ERR(EINVAL);
@@ -1424,6 +1429,10 @@ static int radeon_do_init_cp(drm_device_t * dev, drm_radeon_init_t * init)
 		return DRM_ERR(EINVAL);
 	}
 
+	/* Enable vblank on CRTC1 for older X servers
+	 */
+	dev_priv->vblank_crtc = DRM_RADEON_VBLANK_CRTC1;
+
 	switch(init->func) {
 	case RADEON_INIT_R200_CP:
 		dev_priv->microcode_version = UCODE_R200;
@@ -1663,6 +1672,12 @@ static int radeon_do_init_cp(drm_device_t * dev, drm_radeon_init_t * init)
 	dev_priv->ring.size = init->ring_size;
 	dev_priv->ring.size_l2qw = drm_order(init->ring_size / 8);
 
+	dev_priv->ring.rptr_update = /* init->rptr_update */ 4096;
+	dev_priv->ring.rptr_update_l2qw = drm_order( /* init->rptr_update */ 4096 / 8);
+
+	dev_priv->ring.fetch_size = /* init->fetch_size */ 32;
+	dev_priv->ring.fetch_size_l2ow = drm_order( /* init->fetch_size */ 32 / 16);
+
 	dev_priv->ring.tail_mask = (dev_priv->ring.size / sizeof(u32)) - 1;
 
 	dev_priv->ring.high_mark = RADEON_RING_HIGH_MARK;
diff --git a/shared-core/radeon_drm.h b/shared-core/radeon_drm.h
index e96e7851..6a57b804 100644
--- a/shared-core/radeon_drm.h
+++ b/shared-core/radeon_drm.h
@@ -655,6 +655,7 @@ typedef struct drm_radeon_indirect {
 #define RADEON_PARAM_GART_TEX_HANDLE       10
 #define RADEON_PARAM_SCRATCH_OFFSET        11
 #define RADEON_PARAM_CARD_TYPE             12
+#define RADEON_PARAM_VBLANK_CRTC           13   /* VBLANK CRTC */
 
 typedef struct drm_radeon_getparam {
 	int param;
@@ -709,7 +710,7 @@ typedef struct drm_radeon_setparam {
 
 #define RADEON_SETPARAM_NEW_MEMMAP 4		/* Use new memory map */
 #define RADEON_SETPARAM_PCIGART_TABLE_SIZE 5    /* PCI GART Table Size */
-
+#define RADEON_SETPARAM_VBLANK_CRTC 6           /* VBLANK CRTC */
 /* 1.14: Clients can allocate/free a surface
  */
 typedef struct drm_radeon_surface_alloc {
@@ -722,5 +723,7 @@ typedef struct drm_radeon_surface_free {
 	unsigned int address;
 } drm_radeon_surface_free_t;
 
+#define	DRM_RADEON_VBLANK_CRTC1 	1
+#define	DRM_RADEON_VBLANK_CRTC2 	2
 
 #endif
diff --git a/shared-core/radeon_drv.h b/shared-core/radeon_drv.h
index 4a36ea70..92a9b65e 100644
--- a/shared-core/radeon_drv.h
+++ b/shared-core/radeon_drv.h
@@ -97,10 +97,11 @@
  *       new packet type)
  * 1.26- Add support for variable size PCI(E) gart aperture
  * 1.27- Add support for IGP GART
+ * 1.28- Add support for VBL on CRTC2
  */
 
 #define DRIVER_MAJOR		1
-#define DRIVER_MINOR		27
+#define DRIVER_MINOR		28
 #define DRIVER_PATCHLEVEL	0
 
 /*
@@ -163,8 +164,14 @@ typedef struct drm_radeon_freelist {
 typedef struct drm_radeon_ring_buffer {
 	u32 *start;
 	u32 *end;
-	int size;
-	int size_l2qw;
+	int size; /* Double Words */
+	int size_l2qw; /* log2 Quad Words */
+
+	int rptr_update; /* Double Words */
+	int rptr_update_l2qw; /* log2 Quad Words */
+
+	int fetch_size; /* Double Words */
+	int fetch_size_l2ow; /* log2 Oct Words */
 
 	u32 tail;
 	u32 tail_mask;
@@ -279,6 +286,9 @@ typedef struct drm_radeon_private {
 	/* SW interrupt */
 	wait_queue_head_t swi_queue;
 	atomic_t swi_emitted;
+	int vblank_crtc;
+	uint32_t irq_enable_reg;
+	int irq_enabled;
 
 	struct radeon_surface surfaces[RADEON_MAX_SURFACES];
 	struct radeon_virt_surface virt_surfaces[2*RADEON_MAX_SURFACES];
@@ -355,10 +365,14 @@ extern int radeon_irq_wait(DRM_IOCTL_ARGS);
 extern void radeon_do_release(drm_device_t * dev);
 extern int radeon_driver_vblank_wait(drm_device_t * dev,
 				     unsigned int *sequence);
+extern int radeon_driver_vblank_wait2(drm_device_t * dev,
+				      unsigned int *sequence);
 extern irqreturn_t radeon_driver_irq_handler(DRM_IRQ_ARGS);
 extern void radeon_driver_irq_preinstall(drm_device_t * dev);
 extern void radeon_driver_irq_postinstall(drm_device_t * dev);
 extern void radeon_driver_irq_uninstall(drm_device_t * dev);
+extern int radeon_vblank_crtc_get(drm_device_t *dev);
+extern int radeon_vblank_crtc_set(drm_device_t *dev, int64_t value);
 
 extern int radeon_driver_load(struct drm_device *dev, unsigned long flags);
 extern int radeon_driver_unload(struct drm_device *dev);
@@ -495,12 +509,15 @@ extern int r300_do_cp_cmdbuf(drm_device_t *dev, DRMFILE filp,
 
 #define RADEON_GEN_INT_CNTL		0x0040
 #	define RADEON_CRTC_VBLANK_MASK		(1 << 0)
+#	define RADEON_CRTC2_VBLANK_MASK		(1 << 9)
 #	define RADEON_GUI_IDLE_INT_ENABLE	(1 << 19)
 #	define RADEON_SW_INT_ENABLE		(1 << 25)
 
 #define RADEON_GEN_INT_STATUS		0x0044
 #	define RADEON_CRTC_VBLANK_STAT		(1 << 0)
 #	define RADEON_CRTC_VBLANK_STAT_ACK   	(1 << 0)
+#	define RADEON_CRTC2_VBLANK_STAT		(1 << 9)
+#	define RADEON_CRTC2_VBLANK_STAT_ACK   	(1 << 9)
 #	define RADEON_GUI_IDLE_INT_TEST_ACK     (1 << 19)
 #	define RADEON_SW_INT_TEST		(1 << 25)
 #	define RADEON_SW_INT_TEST_ACK   	(1 << 25)
@@ -601,9 +618,51 @@ extern int r300_do_cp_cmdbuf(drm_device_t *dev, DRMFILE filp,
 #	define RADEON_SOFT_RESET_E2		(1 <<  5)
 #	define RADEON_SOFT_RESET_RB		(1 <<  6)
 #	define RADEON_SOFT_RESET_HDP		(1 <<  7)
+/*
+ *   6:0  Available slots in the FIFO
+ *   8    Host Interface active
+ *   9    CP request active
+ *   10   FIFO request active
+ *   11   Host Interface retry active
+ *   12   CP retry active
+ *   13   FIFO retry active
+ *   14   FIFO pipeline busy
+ *   15   Event engine busy
+ *   16   CP command stream busy
+ *   17   2D engine busy
+ *   18   2D portion of render backend busy
+ *   20   3D setup engine busy
+ *   26   GA engine busy
+ *   27   CBA 2D engine busy
+ *   31   2D engine busy or 3D engine busy or FIFO not empty or CP busy or
+ *           command stream queue not empty or Ring Buffer not empty
+ */
 #define RADEON_RBBM_STATUS		0x0e40
+/* Same as the previous RADEON_RBBM_STATUS; this is a mirror of that register.  */
+/* #define RADEON_RBBM_STATUS 		0x1740 */
+/* bits 6:0 are dword slots available in the cmd fifo */
 #	define RADEON_RBBM_FIFOCNT_MASK		0x007f
-#	define RADEON_RBBM_ACTIVE		(1 << 31)
+#	define RADEON_HIRQ_ON_RBB 	(1 <<  8)
+#	define RADEON_CPRQ_ON_RBB 	(1 <<  9)
+#	define RADEON_CFRQ_ON_RBB 	(1 << 10)
+#	define RADEON_HIRQ_IN_RTBUF 	(1 << 11)
+#	define RADEON_CPRQ_IN_RTBUF 	(1 << 12)
+#	define RADEON_CFRQ_IN_RTBUF 	(1 << 13)
+#	define RADEON_PIPE_BUSY 	(1 << 14)
+#	define RADEON_ENG_EV_BUSY 	(1 << 15)
+#	define RADEON_CP_CMDSTRM_BUSY 	(1 << 16)
+#	define RADEON_E2_BUSY 		(1 << 17)
+#	define RADEON_RB2D_BUSY 	(1 << 18)
+#	define RADEON_RB3D_BUSY 	(1 << 19) /* not used on r300 */
+#	define RADEON_VAP_BUSY 		(1 << 20)
+#	define RADEON_RE_BUSY 		(1 << 21) /* not used on r300 */
+#	define RADEON_TAM_BUSY 		(1 << 22) /* not used on r300 */
+#	define RADEON_TDM_BUSY 		(1 << 23) /* not used on r300 */
+#	define RADEON_PB_BUSY 		(1 << 24) /* not used on r300 */
+#	define RADEON_TIM_BUSY 		(1 << 25) /* not used on r300 */
+#	define RADEON_GA_BUSY 		(1 << 26)
+#	define RADEON_CBA2D_BUSY 	(1 << 27)
+#	define RADEON_RBBM_ACTIVE 	(1 << 31)
 #define RADEON_RE_LINE_PATTERN		0x1cd0
 #define RADEON_RE_MISC			0x26c4
 #define RADEON_RE_TOP_LEFT		0x26c0
diff --git a/shared-core/radeon_irq.c b/shared-core/radeon_irq.c
index 3ff0baa2..a4be86e3 100644
--- a/shared-core/radeon_irq.c
+++ b/shared-core/radeon_irq.c
@@ -73,18 +73,35 @@ irqreturn_t radeon_driver_irq_handler(DRM_IRQ_ARGS)
 	 * outside the DRM
 	 */
 	stat = radeon_acknowledge_irqs(dev_priv, (RADEON_SW_INT_TEST_ACK |
-						  RADEON_CRTC_VBLANK_STAT));
+						  RADEON_CRTC_VBLANK_STAT |
+						  RADEON_CRTC2_VBLANK_STAT));
 	if (!stat)
 		return IRQ_NONE;
 
+	stat &= dev_priv->irq_enable_reg;
+
 	/* SW interrupt */
 	if (stat & RADEON_SW_INT_TEST) {
 		DRM_WAKEUP(&dev_priv->swi_queue);
 	}
 
 	/* VBLANK interrupt */
-	if (stat & RADEON_CRTC_VBLANK_STAT) {
-		atomic_inc(&dev->vbl_received);
+	if (stat & (RADEON_CRTC_VBLANK_STAT|RADEON_CRTC2_VBLANK_STAT)) {
+		int vblank_crtc = dev_priv->vblank_crtc;
+
+		if ((vblank_crtc &
+		     (DRM_RADEON_VBLANK_CRTC1 | DRM_RADEON_VBLANK_CRTC2)) ==
+		    (DRM_RADEON_VBLANK_CRTC1 | DRM_RADEON_VBLANK_CRTC2)) {
+			if (stat & RADEON_CRTC_VBLANK_STAT)
+				atomic_inc(&dev->vbl_received);
+			if (stat & RADEON_CRTC2_VBLANK_STAT)
+				atomic_inc(&dev->vbl_received2);
+		} else if (((stat & RADEON_CRTC_VBLANK_STAT) &&
+			   (vblank_crtc & DRM_RADEON_VBLANK_CRTC1)) ||
+			   ((stat & RADEON_CRTC2_VBLANK_STAT) &&
+			    (vblank_crtc & DRM_RADEON_VBLANK_CRTC2)))
+			atomic_inc(&dev->vbl_received);
+
 		DRM_WAKEUP(&dev->vbl_queue);
 		drm_vbl_send_signals(dev);
 	}
@@ -127,19 +144,30 @@ static int radeon_wait_irq(drm_device_t * dev, int swi_nr)
 	return ret;
 }
 
-int radeon_driver_vblank_wait(drm_device_t * dev, unsigned int *sequence)
+int radeon_driver_vblank_do_wait(drm_device_t * dev, unsigned int *sequence,
+				 int crtc)
 {
 	drm_radeon_private_t *dev_priv =
 	    (drm_radeon_private_t *) dev->dev_private;
 	unsigned int cur_vblank;
 	int ret = 0;
-
+	int ack = 0;
+	atomic_t *counter;
 	if (!dev_priv) {
 		DRM_ERROR("%s called with no initialization\n", __FUNCTION__);
 		return DRM_ERR(EINVAL);
 	}
 
-	radeon_acknowledge_irqs(dev_priv, RADEON_CRTC_VBLANK_STAT);
+	if (crtc == DRM_RADEON_VBLANK_CRTC1) {
+		counter = &dev->vbl_received;
+		ack |= RADEON_CRTC_VBLANK_STAT;
+	} else if (crtc == DRM_RADEON_VBLANK_CRTC2) {
+		counter = &dev->vbl_received2;
+		ack |= RADEON_CRTC2_VBLANK_STAT;
+	} else
+		return DRM_ERR(EINVAL);
+
+	radeon_acknowledge_irqs(dev_priv, ack);
 
 	dev_priv->stats.boxes |= RADEON_BOX_WAIT_IDLE;
 
@@ -148,7 +176,7 @@ int radeon_driver_vblank_wait(drm_device_t * dev, unsigned int *sequence)
 	 * using vertical blanks...
 	 */
 	DRM_WAIT_ON(ret, dev->vbl_queue, 3 * DRM_HZ,
-		    (((cur_vblank = atomic_read(&dev->vbl_received))
+		    (((cur_vblank = atomic_read(counter))
 		      - *sequence) <= (1 << 23)));
 
 	*sequence = cur_vblank;
@@ -156,6 +184,16 @@ int radeon_driver_vblank_wait(drm_device_t * dev, unsigned int *sequence)
 	return ret;
 }
 
+int radeon_driver_vblank_wait(drm_device_t *dev, unsigned int *sequence)
+{
+	return radeon_driver_vblank_do_wait(dev, sequence, DRM_RADEON_VBLANK_CRTC1);
+}
+
+int radeon_driver_vblank_wait2(drm_device_t *dev, unsigned int *sequence)
+{
+	return radeon_driver_vblank_do_wait(dev, sequence, DRM_RADEON_VBLANK_CRTC2);
+}
+
 /* Needs the lock as it touches the ring.
  */
 int radeon_irq_emit(DRM_IOCTL_ARGS)
@@ -204,6 +242,21 @@ int radeon_irq_wait(DRM_IOCTL_ARGS)
 	return radeon_wait_irq(dev, irqwait.irq_seq);
 }
 
+static void radeon_enable_interrupt(drm_device_t *dev)
+{
+	drm_radeon_private_t *dev_priv = (drm_radeon_private_t *) dev->dev_private;
+
+	dev_priv->irq_enable_reg = RADEON_SW_INT_ENABLE;
+	if (dev_priv->vblank_crtc & DRM_RADEON_VBLANK_CRTC1)
+		dev_priv->irq_enable_reg |= RADEON_CRTC_VBLANK_MASK;
+
+	if (dev_priv->vblank_crtc & DRM_RADEON_VBLANK_CRTC2)
+		dev_priv->irq_enable_reg |= RADEON_CRTC2_VBLANK_MASK;
+
+	RADEON_WRITE(RADEON_GEN_INT_CNTL, dev_priv->irq_enable_reg);
+	dev_priv->irq_enabled = 1;
+}
+
 /* drm_dma.h hooks
 */
 void radeon_driver_irq_preinstall(drm_device_t * dev)
@@ -216,7 +269,8 @@ void radeon_driver_irq_preinstall(drm_device_t * dev)
 
 	/* Clear bits if they're already high */
 	radeon_acknowledge_irqs(dev_priv, (RADEON_SW_INT_TEST_ACK |
-					   RADEON_CRTC_VBLANK_STAT));
+					   RADEON_CRTC_VBLANK_STAT |
+					   RADEON_CRTC2_VBLANK_STAT));
 }
 
 void radeon_driver_irq_postinstall(drm_device_t * dev)
@@ -227,9 +281,7 @@ void radeon_driver_irq_postinstall(drm_device_t * dev)
 	atomic_set(&dev_priv->swi_emitted, 0);
 	DRM_INIT_WAITQUEUE(&dev_priv->swi_queue);
 
-	/* Turn on SW and VBL ints */
-	RADEON_WRITE(RADEON_GEN_INT_CNTL,
-		     RADEON_CRTC_VBLANK_MASK | RADEON_SW_INT_ENABLE);
+	radeon_enable_interrupt(dev);
 }
 
 void radeon_driver_irq_uninstall(drm_device_t * dev)
@@ -239,6 +291,38 @@ void radeon_driver_irq_uninstall(drm_device_t * dev)
 	if (!dev_priv)
 		return;
 
+	dev_priv->irq_enabled = 0;
+
 	/* Disable *all* interrupts */
 	RADEON_WRITE(RADEON_GEN_INT_CNTL, 0);
 }
+
+
+int radeon_vblank_crtc_get(drm_device_t *dev)
+{
+	drm_radeon_private_t *dev_priv = (drm_radeon_private_t *) dev->dev_private;
+	u32 flag;
+	u32 value;
+
+	flag = RADEON_READ(RADEON_GEN_INT_CNTL);
+	value = 0;
+
+	if (flag & RADEON_CRTC_VBLANK_MASK)
+		value |= DRM_RADEON_VBLANK_CRTC1;
+
+	if (flag & RADEON_CRTC2_VBLANK_MASK)
+		value |= DRM_RADEON_VBLANK_CRTC2;
+	return value;
+}
+
+int radeon_vblank_crtc_set(drm_device_t *dev, int64_t value)
+{
+	drm_radeon_private_t *dev_priv = (drm_radeon_private_t *) dev->dev_private;
+	if (value & ~(DRM_RADEON_VBLANK_CRTC1 | DRM_RADEON_VBLANK_CRTC2)) {
+		DRM_ERROR("called with invalid crtc 0x%x\n", (unsigned int)value);
+		return DRM_ERR(EINVAL);
+	}
+	dev_priv->vblank_crtc = (unsigned int)value;
+	radeon_enable_interrupt(dev);
+	return 0;
+}
diff --git a/shared-core/radeon_state.c b/shared-core/radeon_state.c
index bad054d7..8ccd0981 100644
--- a/shared-core/radeon_state.c
+++ b/shared-core/radeon_state.c
@@ -3131,6 +3131,9 @@ static int radeon_cp_getparam(DRM_IOCTL_ARGS)
 		else
 			value = RADEON_CARD_PCI;
 		break;
+	case RADEON_PARAM_VBLANK_CRTC:
+		value = radeon_vblank_crtc_get(dev);
+		break;
 	default:
 		DRM_DEBUG( "Invalid parameter %d\n", param.param );
 		return DRM_ERR(EINVAL);
@@ -3192,6 +3195,9 @@ static int radeon_cp_setparam(DRM_IOCTL_ARGS)
 		if (dev_priv->gart_info.table_size < RADEON_PCIGART_TABLE_SIZE)
 			dev_priv->gart_info.table_size = RADEON_PCIGART_TABLE_SIZE;
 		break;
+	case RADEON_SETPARAM_VBLANK_CRTC:
+		return radeon_vblank_crtc_set(dev, sp.value);
+		break;
 	default:
 		DRM_DEBUG("Invalid parameter %d\n", sp.param);
 		return DRM_ERR(EINVAL);
diff --git a/shared-core/sis_drv.h b/shared-core/sis_drv.h
index 006d148c..ec572ad4 100644
--- a/shared-core/sis_drv.h
+++ b/shared-core/sis_drv.h
@@ -33,11 +33,11 @@
 
 #define DRIVER_AUTHOR		"SIS, Tungsten Graphics"
 #define DRIVER_NAME		"sis"
-#define DRIVER_DESC		"SIS 300/630/540"
-#define DRIVER_DATE		"20060619"
+#define DRIVER_DESC		"SIS 300/630/540 and XGI V3XE/V5/V8"
+#define DRIVER_DATE		"20070626"
 #define DRIVER_MAJOR		1
-#define DRIVER_MINOR		2
-#define DRIVER_PATCHLEVEL	1
+#define DRIVER_MINOR		3
+#define DRIVER_PATCHLEVEL	0
 
 enum sis_family {
 	SIS_OTHER = 0,
diff --git a/shared-core/via_dma.c b/shared-core/via_dma.c
index c73634e4..333c4bcf 100644
--- a/shared-core/via_dma.c
+++ b/shared-core/via_dma.c
@@ -84,9 +84,9 @@ static uint32_t via_cmdbuf_lag(drm_via_private_t *dev_priv)
 {
 	uint32_t agp_base = dev_priv->dma_offset + (uint32_t) dev_priv->agpAddr;
 	uint32_t hw_addr = *(dev_priv->hw_addr_ptr) - agp_base;
-	
-	return ((hw_addr <= dev_priv->dma_low) ? 
-		(dev_priv->dma_low - hw_addr) : 
+
+	return ((hw_addr <= dev_priv->dma_low) ?
+		(dev_priv->dma_low - hw_addr) :
 		(dev_priv->dma_wrap + dev_priv->dma_low - hw_addr));
 }
 
@@ -103,7 +103,7 @@ via_cmdbuf_wait(drm_via_private_t * dev_priv, unsigned int size)
 	uint32_t count;
 	hw_addr_ptr = dev_priv->hw_addr_ptr;
 	cur_addr = dev_priv->dma_low;
-	next_addr = cur_addr + size + 512*1024;
+	next_addr = cur_addr + size + 512 * 1024;
 	count = 1000000;
 	do {
 		hw_addr = *hw_addr_ptr - agp_base;
@@ -207,8 +207,8 @@ static int via_initialize(drm_device_t * dev,
 	dev_priv->dma_offset = init->offset;
 	dev_priv->last_pause_ptr = NULL;
 	dev_priv->hw_addr_ptr =
-	    (volatile uint32_t *)((char *)dev_priv->mmio->handle +
-	    init->reg_pause_addr);
+		(volatile uint32_t *)((char *)dev_priv->mmio->handle +
+		init->reg_pause_addr);
 
 	via_cmdbuf_start(dev_priv);
 
@@ -239,8 +239,8 @@ static int via_dma_init(DRM_IOCTL_ARGS)
 			retcode = via_dma_cleanup(dev);
 		break;
 	case VIA_DMA_INITIALIZED:
-		retcode = (dev_priv->ring.virtual_start != NULL) ? 
-			0: DRM_ERR( EFAULT );
+		retcode = (dev_priv->ring.virtual_start != NULL) ?
+			0 : DRM_ERR(EFAULT);
 		break;
 	default:
 		retcode = DRM_ERR(EINVAL);
@@ -268,8 +268,7 @@ static int via_dispatch_cmdbuffer(drm_device_t * dev, drm_via_cmdbuffer_t * cmd)
 
 	if (cmd->size > VIA_PCI_BUF_SIZE) {
 		return DRM_ERR(ENOMEM);
-	} 
-
+	}
 
 	if (DRM_COPY_FROM_USER(dev_priv->pci_buf, cmd->buf, cmd->size))
 		return DRM_ERR(EFAULT);
@@ -292,7 +291,7 @@ static int via_dispatch_cmdbuffer(drm_device_t * dev, drm_via_cmdbuffer_t * cmd)
 	}
 
 	memcpy(vb, dev_priv->pci_buf, cmd->size);
-	
+
 	dev_priv->dma_low += cmd->size;
 
 	/*
@@ -301,7 +300,7 @@ static int via_dispatch_cmdbuffer(drm_device_t * dev, drm_via_cmdbuffer_t * cmd)
 	 */
 
 	if (cmd->size < 0x100)
-	  via_pad_cache(dev_priv,(0x100 - cmd->size) >> 3);
+		via_pad_cache(dev_priv, (0x100 - cmd->size) >> 3);
 	via_cmdbuf_pause(dev_priv);
 
 	return 0;
@@ -321,7 +320,7 @@ static int via_flush_ioctl(DRM_IOCTL_ARGS)
 {
 	DRM_DEVICE;
 
-	LOCK_TEST_WITH_RETURN( dev, filp );
+	LOCK_TEST_WITH_RETURN(dev, filp);
 
 	return via_driver_dma_quiescent(dev);
 }
@@ -332,7 +331,7 @@ static int via_cmdbuffer(DRM_IOCTL_ARGS)
 	drm_via_cmdbuffer_t cmdbuf;
 	int ret;
 
-	LOCK_TEST_WITH_RETURN( dev, filp );
+	LOCK_TEST_WITH_RETURN(dev, filp);
 
 	DRM_COPY_FROM_USER_IOCTL(cmdbuf, (drm_via_cmdbuffer_t __user *) data,
 				 sizeof(cmdbuf));
@@ -355,16 +354,16 @@ static int via_dispatch_pci_cmdbuffer(drm_device_t * dev,
 
 	if (cmd->size > VIA_PCI_BUF_SIZE) {
 		return DRM_ERR(ENOMEM);
-	} 
+	}
 	if (DRM_COPY_FROM_USER(dev_priv->pci_buf, cmd->buf, cmd->size))
 		return DRM_ERR(EFAULT);
-	
-	if ((ret = 
-	     via_verify_command_stream((uint32_t *)dev_priv->pci_buf,
+
+	if ((ret =
+	     via_verify_command_stream((uint32_t *) dev_priv->pci_buf,
 				       cmd->size, dev, 0))) {
 		return ret;
 	}
-	
+
 	ret =
 	    via_parse_command_stream(dev, (const uint32_t *)dev_priv->pci_buf,
 				     cmd->size);
@@ -377,7 +376,7 @@ static int via_pci_cmdbuffer(DRM_IOCTL_ARGS)
 	drm_via_cmdbuffer_t cmdbuf;
 	int ret;
 
-	LOCK_TEST_WITH_RETURN( dev, filp );
+	LOCK_TEST_WITH_RETURN(dev, filp);
 
 	DRM_COPY_FROM_USER_IOCTL(cmdbuf, (drm_via_cmdbuffer_t __user *) data,
 				 sizeof(cmdbuf));
@@ -393,7 +392,6 @@ static int via_pci_cmdbuffer(DRM_IOCTL_ARGS)
 	return 0;
 }
 
-
 static inline uint32_t *via_align_buffer(drm_via_private_t * dev_priv,
 					 uint32_t * vb, int qw_count)
 {
@@ -403,7 +401,6 @@ static inline uint32_t *via_align_buffer(drm_via_private_t * dev_priv,
 	return vb;
 }
 
-
 /*
  * This function is used internally by ring buffer mangement code.
  *
@@ -419,8 +416,7 @@ static inline uint32_t *via_get_dma(drm_via_private_t * dev_priv)
  * modifying the pause address stored in the buffer itself. If
  * the regulator has already paused, restart it.
  */
-
-static int via_hook_segment(drm_via_private_t *dev_priv,
+static int via_hook_segment(drm_via_private_t * dev_priv,
 			    uint32_t pause_addr_hi, uint32_t pause_addr_lo,
 			    int no_pci_fire)
 {
@@ -479,7 +475,7 @@ static int via_wait_idle(drm_via_private_t * dev_priv)
 }
 
 static uint32_t *via_align_cmd(drm_via_private_t * dev_priv, uint32_t cmd_type,
-			       uint32_t addr, uint32_t *cmd_addr_hi, 
+			       uint32_t addr, uint32_t *cmd_addr_hi,
 			       uint32_t *cmd_addr_lo, int skip_wait)
 {
 	uint32_t agp_base;
@@ -508,9 +504,6 @@ static uint32_t *via_align_cmd(drm_via_private_t * dev_priv, uint32_t cmd_type,
 	return vb;
 }
 
-
-
-
 static void via_cmdbuf_start(drm_via_private_t * dev_priv)
 {
 	uint32_t pause_addr_lo, pause_addr_hi;
diff --git a/shared-core/via_verifier.c b/shared-core/via_verifier.c
index b5a1d33a..4b844af0 100644
--- a/shared-core/via_verifier.c
+++ b/shared-core/via_verifier.c
@@ -255,7 +255,6 @@ static __inline__ drm_local_map_t *via_drm_lookup_agp_map(drm_via_state_t * seq,
 							  drm_device_t * dev)
 {
 #ifdef __linux__
-	struct list_head *list;
 	drm_map_list_t *r_list;
 #endif
 	drm_local_map_t *map = seq->map_cache;
@@ -265,8 +264,7 @@ static __inline__ drm_local_map_t *via_drm_lookup_agp_map(drm_via_state_t * seq,
 		return map;
 	}
 #ifdef __linux__
-	list_for_each(list, &dev->maplist->head) {
-		r_list = (drm_map_list_t *) list;
+	list_for_each_entry(r_list, &dev->maplist, head) {
 		map = r_list->map;
 		if (!map)
 			continue;
author	Alan Hourihane <alanh@fairlite.demon.co.uk>	2007-06-29 20:09:44 +0100
committer	Alan Hourihane <alanh@fairlite.demon.co.uk>	2007-06-29 20:09:44 +0100
commit	8a78dead291ffdb5a8774419cdca369a1e27cad9 (patch)
tree	399b8378e32036bdc7dadc10c639465a8fb21681 /shared-core
parent	e79e2a58161d44754fd55507e155b7e12a09c4d2 (diff)
parent	a27af4c4a665864df09123f177ca7269e48f6171 (diff)