drivers/gpu/drm/nouveau/nouveau_bios.c    |   25 +--
 drivers/gpu/drm/nouveau/nouveau_bo.c      |    6 +-
 drivers/gpu/drm/nouveau/nouveau_channel.c |    2 +-
 drivers/gpu/drm/nouveau/nouveau_dp.c      |    2 -
 drivers/gpu/drm/nouveau/nouveau_drv.h     |   13 +-
 drivers/gpu/drm/nouveau/nouveau_fence.c   |  190 ++++++++++------
 drivers/gpu/drm/nouveau/nouveau_mem.c     |   16 +-
 drivers/gpu/drm/nouveau/nouveau_object.c  |   22 ++-
 drivers/gpu/drm/nouveau/nouveau_sgdma.c   |  342 +++++++++++++++++++++++++----
 drivers/gpu/drm/nouveau/nouveau_state.c   |   10 +-
 drivers/gpu/drm/nouveau/nouveau_temp.c    |    4 +-
 drivers/gpu/drm/nouveau/nv04_fifo.c       |   17 ++-
 drivers/gpu/drm/nouveau/nv40_fb.c         |   59 +++++-
 drivers/gpu/drm/nouveau/nv50_display.c    |    7 +-
 drivers/gpu/drm/nouveau/nv50_fifo.c       |    3 +-
 drivers/gpu/drm/nouveau/nv50_gpio.c       |   13 +-
 drivers/gpu/drm/nouveau/nv50_graph.c      |    8 +-
 drivers/gpu/drm/nouveau/nv50_vm.c         |    1 -
 drivers/gpu/drm/nouveau/nvc0_fifo.c       |   15 +-
 drivers/gpu/drm/nouveau/nvc0_graph.c      |    2 -
 20 files changed, 567 insertions(+), 190 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/nouveau_bios.c b/drivers/gpu/drm/nouveau/nouveau_bios.c
index 6bdab89..b8ff1e7 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bios.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bios.c
@@ -5950,6 +5950,11 @@ apply_dcb_connector_quirks(struct nvbios *bios, int idx)
 	}
 }
 
+static const u8 hpd_gpio[16] = {
+	0xff, 0x07, 0x08, 0xff, 0xff, 0x51, 0x52, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0x5e, 0x5f, 0x60,
+};
+
 static void
 parse_dcb_connector_table(struct nvbios *bios)
 {
@@ -5986,23 +5991,9 @@ parse_dcb_connector_table(struct nvbios *bios)
 
 		cte->type  = (cte->entry & 0x000000ff) >> 0;
 		cte->index2 = (cte->entry & 0x00000f00) >> 8;
-		switch (cte->entry & 0x00033000) {
-		case 0x00001000:
-			cte->gpio_tag = 0x07;
-			break;
-		case 0x00002000:
-			cte->gpio_tag = 0x08;
-			break;
-		case 0x00010000:
-			cte->gpio_tag = 0x51;
-			break;
-		case 0x00020000:
-			cte->gpio_tag = 0x52;
-			break;
-		default:
-			cte->gpio_tag = 0xff;
-			break;
-		}
+
+		cte->gpio_tag = ffs((cte->entry & 0x07033000) >> 12);
+		cte->gpio_tag = hpd_gpio[cte->gpio_tag];
 
 		if (cte->type == 0xff)
 			continue;
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index d38a4d9..bf260af 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -382,7 +382,8 @@ nouveau_bo_create_ttm_backend_entry(struct ttm_bo_device *bdev)
 	case NOUVEAU_GART_AGP:
 		return ttm_agp_backend_init(bdev, dev->agp->bridge);
 #endif
-	case NOUVEAU_GART_SGDMA:
+	case NOUVEAU_GART_PDMA:
+	case NOUVEAU_GART_HW:
 		return nouveau_sgdma_init_ttm(dev);
 	default:
 		NV_ERROR(dev, "Unknown GART type %d\n",
@@ -436,7 +437,8 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 				TTM_PL_FLAG_WC;
 			man->default_caching = TTM_PL_FLAG_WC;
 			break;
-		case NOUVEAU_GART_SGDMA:
+		case NOUVEAU_GART_PDMA:
+		case NOUVEAU_GART_HW:
 			man->flags = TTM_MEMTYPE_FLAG_MAPPABLE |
 				     TTM_MEMTYPE_FLAG_CMA;
 			man->available_caching = TTM_PL_MASK_CACHING;
diff --git a/drivers/gpu/drm/nouveau/nouveau_channel.c b/drivers/gpu/drm/nouveau/nouveau_channel.c
index 3960d66..3d7b316 100644
--- a/drivers/gpu/drm/nouveau/nouveau_channel.c
+++ b/drivers/gpu/drm/nouveau/nouveau_channel.c
@@ -35,7 +35,7 @@ nouveau_channel_pushbuf_ctxdma_init(struct nouveau_channel *chan)
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_bo *pb = chan->pushbuf_bo;
 	struct nouveau_gpuobj *pushbuf = NULL;
-	int ret;
+	int ret = 0;
 
 	if (dev_priv->card_type >= NV_50) {
 		if (dev_priv->card_type < NV_C0) {
diff --git a/drivers/gpu/drm/nouveau/nouveau_dp.c b/drivers/gpu/drm/nouveau/nouveau_dp.c
index 38d5995..7beb82a 100644
--- a/drivers/gpu/drm/nouveau/nouveau_dp.c
+++ b/drivers/gpu/drm/nouveau/nouveau_dp.c
@@ -175,7 +175,6 @@ nouveau_dp_link_train_adjust(struct drm_encoder *encoder, uint8_t *config)
 {
 	struct nouveau_encoder *nv_encoder = nouveau_encoder(encoder);
 	struct drm_device *dev = encoder->dev;
-	struct bit_displayport_encoder_table_entry *dpse;
 	struct bit_displayport_encoder_table *dpe;
 	int ret, i, dpe_headerlen, vs = 0, pre = 0;
 	uint8_t request[2];
@@ -183,7 +182,6 @@ nouveau_dp_link_train_adjust(struct drm_encoder *encoder, uint8_t *config)
 	dpe = nouveau_bios_dp_table(dev, nv_encoder->dcb, &dpe_headerlen);
 	if (!dpe)
 		return false;
-	dpse = (void *)((char *)dpe + dpe_headerlen);
 
 	ret = auxch_rd(encoder, DP_ADJUST_REQUEST_LANE0_1, request, 2);
 	if (ret)
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index 9821fca..2e3d7fb 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -652,7 +652,6 @@ struct drm_nouveau_private {
 	/* interrupt handling */
 	void (*irq_handler[32])(struct drm_device *);
 	bool msi_enabled;
-	struct workqueue_struct *wq;
 	struct work_struct irq_work;
 
 	struct list_head vbl_waiting;
@@ -691,13 +690,21 @@ struct drm_nouveau_private {
 	struct {
 		enum {
 			NOUVEAU_GART_NONE = 0,
-			NOUVEAU_GART_AGP,
-			NOUVEAU_GART_SGDMA
+			NOUVEAU_GART_AGP,	/* AGP */
+			NOUVEAU_GART_PDMA,	/* paged dma object */
+			NOUVEAU_GART_HW		/* on-chip gart/vm */
 		} type;
 		uint64_t aper_base;
 		uint64_t aper_size;
 		uint64_t aper_free;
 
+		struct ttm_backend_func *func;
+
+		struct {
+			struct page *page;
+			dma_addr_t   addr;
+		} dummy;
+
 		struct nouveau_gpuobj *sg_ctxdma;
 		struct nouveau_vma vma;
 	} gart_info;
diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.c b/drivers/gpu/drm/nouveau/nouveau_fence.c
index 221b846..8b46392 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fence.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fence.c
@@ -32,8 +32,7 @@
 #include "nouveau_dma.h"
 
 #define USE_REFCNT(dev) (nouveau_private(dev)->chipset >= 0x10)
-#define USE_SEMA(dev) (nouveau_private(dev)->chipset >= 0x17 && \
-		       nouveau_private(dev)->card_type < NV_C0)
+#define USE_SEMA(dev) (nouveau_private(dev)->chipset >= 0x17)
 
 struct nouveau_fence {
 	struct nouveau_channel *channel;
@@ -259,11 +258,12 @@ __nouveau_fence_wait(void *sync_obj, void *sync_arg, bool lazy, bool intr)
 }
 
 static struct nouveau_semaphore *
-alloc_semaphore(struct drm_device *dev)
+semaphore_alloc(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_semaphore *sema;
-	int ret;
+	int size = (dev_priv->chipset < 0x84) ? 4 : 16;
+	int ret, i;
 
 	if (!USE_SEMA(dev))
 		return NULL;
@@ -277,9 +277,9 @@ alloc_semaphore(struct drm_device *dev)
 		goto fail;
 
 	spin_lock(&dev_priv->fence.lock);
-	sema->mem = drm_mm_search_free(&dev_priv->fence.heap, 4, 0, 0);
+	sema->mem = drm_mm_search_free(&dev_priv->fence.heap, size, 0, 0);
 	if (sema->mem)
-		sema->mem = drm_mm_get_block_atomic(sema->mem, 4, 0);
+		sema->mem = drm_mm_get_block_atomic(sema->mem, size, 0);
 	spin_unlock(&dev_priv->fence.lock);
 
 	if (!sema->mem)
@@ -287,7 +287,8 @@ alloc_semaphore(struct drm_device *dev)
 
 	kref_init(&sema->ref);
 	sema->dev = dev;
-	nouveau_bo_wr32(dev_priv->fence.bo, sema->mem->start / 4, 0);
+	for (i = sema->mem->start; i < sema->mem->start + size; i += 4)
+		nouveau_bo_wr32(dev_priv->fence.bo, i / 4, 0);
 
 	return sema;
 fail:
@@ -296,7 +297,7 @@ fail:
 }
 
 static void
-free_semaphore(struct kref *ref)
+semaphore_free(struct kref *ref)
 {
 	struct nouveau_semaphore *sema =
 		container_of(ref, struct nouveau_semaphore, ref);
@@ -318,61 +319,107 @@ semaphore_work(void *priv, bool signalled)
 	if (unlikely(!signalled))
 		nouveau_bo_wr32(dev_priv->fence.bo, sema->mem->start / 4, 1);
 
-	kref_put(&sema->ref, free_semaphore);
+	kref_put(&sema->ref, semaphore_free);
 }
 
 static int
-emit_semaphore(struct nouveau_channel *chan, int method,
-	       struct nouveau_semaphore *sema)
+semaphore_acquire(struct nouveau_channel *chan, struct nouveau_semaphore *sema)
 {
-	struct drm_nouveau_private *dev_priv = sema->dev->dev_private;
-	struct nouveau_fence *fence;
-	bool smart = (dev_priv->card_type >= NV_50);
+	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
+	struct nouveau_fence *fence = NULL;
 	int ret;
 
-	ret = RING_SPACE(chan, smart ? 8 : 4);
+	if (dev_priv->chipset < 0x84) {
+		ret = RING_SPACE(chan, 3);
+		if (ret)
+			return ret;
+
+		BEGIN_RING(chan, NvSubSw, NV_SW_SEMAPHORE_OFFSET, 2);
+		OUT_RING  (chan, sema->mem->start);
+		OUT_RING  (chan, 1);
+	} else
+	if (dev_priv->chipset < 0xc0) {
+		struct nouveau_vma *vma = &dev_priv->fence.bo->vma;
+		u64 offset = vma->offset + sema->mem->start;
+
+		ret = RING_SPACE(chan, 5);
+		if (ret)
+			return ret;
+
+		BEGIN_RING(chan, NvSubSw, 0x0010, 4);
+		OUT_RING  (chan, upper_32_bits(offset));
+		OUT_RING  (chan, lower_32_bits(offset));
+		OUT_RING  (chan, 1);
+		OUT_RING  (chan, 1); /* ACQUIRE_EQ */
+	} else {
+		struct nouveau_vma *vma = &dev_priv->fence.bo->vma;
+		u64 offset = vma->offset + sema->mem->start;
+
+		ret = RING_SPACE(chan, 5);
+		if (ret)
+			return ret;
+
+		BEGIN_NVC0(chan, 2, NvSubM2MF, 0x0010, 4);
+		OUT_RING  (chan, upper_32_bits(offset));
+		OUT_RING  (chan, lower_32_bits(offset));
+		OUT_RING  (chan, 1);
+		OUT_RING  (chan, 0x1001); /* ACQUIRE_EQ */
+	}
+
+	/* Delay semaphore destruction until its work is done */
+	ret = nouveau_fence_new(chan, &fence, true);
 	if (ret)
 		return ret;
 
-	if (smart) {
-		BEGIN_RING(chan, NvSubSw, NV_SW_DMA_SEMAPHORE, 1);
-		OUT_RING(chan, NvSema);
-	}
-	BEGIN_RING(chan, NvSubSw, NV_SW_SEMAPHORE_OFFSET, 1);
-	OUT_RING(chan, sema->mem->start);
-
-	if (smart && method == NV_SW_SEMAPHORE_ACQUIRE) {
-		/*
-		 * NV50 tries to be too smart and context-switch
-		 * between semaphores instead of doing a "first come,
-		 * first served" strategy like previous cards
-		 * do.
-		 *
-		 * That's bad because the ACQUIRE latency can get as
-		 * large as the PFIFO context time slice in the
-		 * typical DRI2 case where you have several
-		 * outstanding semaphores at the same moment.
-		 *
-		 * If we're going to ACQUIRE, force the card to
-		 * context switch before, just in case the matching
-		 * RELEASE is already scheduled to be executed in
-		 * another channel.
-		 */
-		BEGIN_RING(chan, NvSubSw, NV_SW_YIELD, 1);
-		OUT_RING(chan, 0);
-	}
+	kref_get(&sema->ref);
+	nouveau_fence_work(fence, semaphore_work, sema);
+	nouveau_fence_unref(&fence);
+	return 0;
+}
+
+static int
+semaphore_release(struct nouveau_channel *chan, struct nouveau_semaphore *sema)
+{
+	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
+	struct nouveau_fence *fence = NULL;
+	int ret;
+
+	if (dev_priv->chipset < 0x84) {
+		ret = RING_SPACE(chan, 4);
+		if (ret)
+			return ret;
+
+		BEGIN_RING(chan, NvSubSw, NV_SW_SEMAPHORE_OFFSET, 1);
+		OUT_RING  (chan, sema->mem->start);
+		BEGIN_RING(chan, NvSubSw, NV_SW_SEMAPHORE_RELEASE, 1);
+		OUT_RING  (chan, 1);
+	} else
+	if (dev_priv->chipset < 0xc0) {
+		struct nouveau_vma *vma = &dev_priv->fence.bo->vma;
+		u64 offset = vma->offset + sema->mem->start;
 
-	BEGIN_RING(chan, NvSubSw, method, 1);
-	OUT_RING(chan, 1);
-
-	if (smart && method == NV_SW_SEMAPHORE_RELEASE) {
-		/*
-		 * Force the card to context switch, there may be
-		 * another channel waiting for the semaphore we just
-		 * released.
-		 */
-		BEGIN_RING(chan, NvSubSw, NV_SW_YIELD, 1);
-		OUT_RING(chan, 0);
+		ret = RING_SPACE(chan, 5);
+		if (ret)
+			return ret;
+
+		BEGIN_RING(chan, NvSubSw, 0x0010, 4);
+		OUT_RING  (chan, upper_32_bits(offset));
+		OUT_RING  (chan, lower_32_bits(offset));
+		OUT_RING  (chan, 1);
+		OUT_RING  (chan, 2); /* RELEASE */
+	} else {
+		struct nouveau_vma *vma = &dev_priv->fence.bo->vma;
+		u64 offset = vma->offset + sema->mem->start;
+
+		ret = RING_SPACE(chan, 5);
+		if (ret)
+			return ret;
+
+		BEGIN_NVC0(chan, 2, NvSubM2MF, 0x0010, 4);
+		OUT_RING  (chan, upper_32_bits(offset));
+		OUT_RING  (chan, lower_32_bits(offset));
+		OUT_RING  (chan, 1);
+		OUT_RING  (chan, 0x1002); /* RELEASE */
 	}
 
 	/* Delay semaphore destruction until its work is done */
@@ -383,7 +430,6 @@ emit_semaphore(struct nouveau_channel *chan, int method,
 	kref_get(&sema->ref);
 	nouveau_fence_work(fence, semaphore_work, sema);
 	nouveau_fence_unref(&fence);
-
 	return 0;
 }
 
@@ -400,7 +446,7 @@ nouveau_fence_sync(struct nouveau_fence *fence,
 		   nouveau_fence_signalled(fence)))
 		goto out;
 
-	sema = alloc_semaphore(dev);
+	sema = semaphore_alloc(dev);
 	if (!sema) {
 		/* Early card or broken userspace, fall back to
 		 * software sync. */
@@ -418,17 +464,17 @@ nouveau_fence_sync(struct nouveau_fence *fence,
 	}
 
 	/* Make wchan wait until it gets signalled */
-	ret = emit_semaphore(wchan, NV_SW_SEMAPHORE_ACQUIRE, sema);
+	ret = semaphore_acquire(wchan, sema);
 	if (ret)
 		goto out_unlock;
 
 	/* Signal the semaphore from chan */
-	ret = emit_semaphore(chan, NV_SW_SEMAPHORE_RELEASE, sema);
+	ret = semaphore_release(chan, sema);
 
 out_unlock:
 	mutex_unlock(&chan->mutex);
 out_unref:
-	kref_put(&sema->ref, free_semaphore);
+	kref_put(&sema->ref, semaphore_free);
 out:
 	if (chan)
 		nouveau_channel_put_unlocked(&chan);
@@ -449,22 +495,23 @@ nouveau_fence_channel_init(struct nouveau_channel *chan)
 	struct nouveau_gpuobj *obj = NULL;
 	int ret;
 
+	if (dev_priv->card_type >= NV_C0)
+		goto out_initialised;
+
 	/* Create an NV_SW object for various sync purposes */
 	ret = nouveau_gpuobj_gr_new(chan, NvSw, NV_SW);
 	if (ret)
 		return ret;
 
 	/* we leave subchannel empty for nvc0 */
-	if (dev_priv->card_type < NV_C0) {
-		ret = RING_SPACE(chan, 2);
-		if (ret)
-			return ret;
-		BEGIN_RING(chan, NvSubSw, 0, 1);
-		OUT_RING(chan, NvSw);
-	}
+	ret = RING_SPACE(chan, 2);
+	if (ret)
+		return ret;
+	BEGIN_RING(chan, NvSubSw, 0, 1);
+	OUT_RING(chan, NvSw);
 
 	/* Create a DMA object for the shared cross-channel sync area. */
-	if (USE_SEMA(dev)) {
+	if (USE_SEMA(dev) && dev_priv->chipset < 0x84) {
 		struct ttm_mem_reg *mem = &dev_priv->fence.bo->bo.mem;
 
 		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
@@ -484,14 +531,20 @@ nouveau_fence_channel_init(struct nouveau_channel *chan)
 			return ret;
 		BEGIN_RING(chan, NvSubSw, NV_SW_DMA_SEMAPHORE, 1);
 		OUT_RING(chan, NvSema);
+	} else {
+		ret = RING_SPACE(chan, 2);
+		if (ret)
+			return ret;
+		BEGIN_RING(chan, NvSubSw, NV_SW_DMA_SEMAPHORE, 1);
+		OUT_RING  (chan, chan->vram_handle); /* whole VM */
 	}
 
 	FIRE_RING(chan);
 
+out_initialised:
 	INIT_LIST_HEAD(&chan->fence.pending);
 	spin_lock_init(&chan->fence.lock);
 	atomic_set(&chan->fence.last_sequence_irq, 0);
-
 	return 0;
 }
 
@@ -519,11 +572,12 @@ int
 nouveau_fence_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	int size = (dev_priv->chipset < 0x84) ? 4096 : 16384;
 	int ret;
 
 	/* Create a shared VRAM heap for cross-channel sync. */
 	if (USE_SEMA(dev)) {
-		ret = nouveau_bo_new(dev, NULL, 4096, 0, TTM_PL_FLAG_VRAM,
+		ret = nouveau_bo_new(dev, NULL, size, 0, TTM_PL_FLAG_VRAM,
 				     0, 0, false, true, &dev_priv->fence.bo);
 		if (ret)
 			return ret;
diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c b/drivers/gpu/drm/nouveau/nouveau_mem.c
index 26347b7..30bd230 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_mem.c
@@ -393,11 +393,17 @@ nouveau_mem_vram_init(struct drm_device *dev)
 	struct ttm_bo_device *bdev = &dev_priv->ttm.bdev;
 	int ret, dma_bits;
 
-	if (dev_priv->card_type >= NV_50 &&
-	    pci_dma_supported(dev->pdev, DMA_BIT_MASK(40)))
-		dma_bits = 40;
-	else
-		dma_bits = 32;
+	dma_bits = 32;
+	if (dev_priv->card_type >= NV_50) {
+		if (pci_dma_supported(dev->pdev, DMA_BIT_MASK(40)))
+			dma_bits = 40;
+	} else
+	if (drm_device_is_pcie(dev) &&
+	    dev_priv->chipset != 0x40 &&
+	    dev_priv->chipset != 0x45) {
+		if (pci_dma_supported(dev->pdev, DMA_BIT_MASK(39)))
+			dma_bits = 39;
+	}
 
 	ret = pci_set_dma_mask(dev->pdev, DMA_BIT_MASK(dma_bits));
 	if (ret)
diff --git a/drivers/gpu/drm/nouveau/nouveau_object.c b/drivers/gpu/drm/nouveau/nouveau_object.c
index 30b6544..3c12461 100644
--- a/drivers/gpu/drm/nouveau/nouveau_object.c
+++ b/drivers/gpu/drm/nouveau/nouveau_object.c
@@ -490,16 +490,22 @@ nouveau_gpuobj_dma_new(struct nouveau_channel *chan, int class, u64 base,
 	}
 
 	if (target == NV_MEM_TARGET_GART) {
-		if (dev_priv->gart_info.type == NOUVEAU_GART_AGP) {
-			target = NV_MEM_TARGET_PCI_NOSNOOP;
-			base  += dev_priv->gart_info.aper_base;
-		} else
-		if (base != 0) {
-			base = nouveau_sgdma_get_physical(dev, base);
+		struct nouveau_gpuobj *gart = dev_priv->gart_info.sg_ctxdma;
+
+		if (dev_priv->gart_info.type == NOUVEAU_GART_PDMA) {
+			if (base == 0) {
+				nouveau_gpuobj_ref(gart, pobj);
+				return 0;
+			}
+
+			base   = nouveau_sgdma_get_physical(dev, base);
 			target = NV_MEM_TARGET_PCI;
 		} else {
-			nouveau_gpuobj_ref(dev_priv->gart_info.sg_ctxdma, pobj);
-			return 0;
+			base += dev_priv->gart_info.aper_base;
+			if (dev_priv->gart_info.type == NOUVEAU_GART_AGP)
+				target = NV_MEM_TARGET_PCI_NOSNOOP;
+			else
+				target = NV_MEM_TARGET_PCI;
 		}
 	}
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_sgdma.c b/drivers/gpu/drm/nouveau/nouveau_sgdma.c
index 9a250eb..bdbaa54 100644
--- a/drivers/gpu/drm/nouveau/nouveau_sgdma.c
+++ b/drivers/gpu/drm/nouveau/nouveau_sgdma.c
@@ -74,8 +74,24 @@ nouveau_sgdma_clear(struct ttm_backend *be)
 	}
 }
 
+static void
+nouveau_sgdma_destroy(struct ttm_backend *be)
+{
+	struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
+
+	if (be) {
+		NV_DEBUG(nvbe->dev, "\n");
+
+		if (nvbe) {
+			if (nvbe->pages)
+				be->func->clear(be);
+			kfree(nvbe);
+		}
+	}
+}
+
 static int
-nouveau_sgdma_bind(struct ttm_backend *be, struct ttm_mem_reg *mem)
+nv04_sgdma_bind(struct ttm_backend *be, struct ttm_mem_reg *mem)
 {
 	struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
 	struct drm_device *dev = nvbe->dev;
@@ -102,7 +118,7 @@ nouveau_sgdma_bind(struct ttm_backend *be, struct ttm_mem_reg *mem)
 }
 
 static int
-nouveau_sgdma_unbind(struct ttm_backend *be)
+nv04_sgdma_unbind(struct ttm_backend *be)
 {
 	struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
 	struct drm_device *dev = nvbe->dev;
@@ -125,23 +141,222 @@ nouveau_sgdma_unbind(struct ttm_backend *be)
 	return 0;
 }
 
+static struct ttm_backend_func nv04_sgdma_backend = {
+	.populate		= nouveau_sgdma_populate,
+	.clear			= nouveau_sgdma_clear,
+	.bind			= nv04_sgdma_bind,
+	.unbind			= nv04_sgdma_unbind,
+	.destroy		= nouveau_sgdma_destroy
+};
+
 static void
-nouveau_sgdma_destroy(struct ttm_backend *be)
+nv41_sgdma_flush(struct nouveau_sgdma_be *nvbe)
+{
+	struct drm_device *dev = nvbe->dev;
+
+	nv_wr32(dev, 0x100810, 0x00000022);
+	if (!nv_wait(dev, 0x100810, 0x00000100, 0x00000100))
+		NV_ERROR(dev, "vm flush timeout: 0x%08x\n",
+			 nv_rd32(dev, 0x100810));
+	nv_wr32(dev, 0x100810, 0x00000000);
+}
+
+static int
+nv41_sgdma_bind(struct ttm_backend *be, struct ttm_mem_reg *mem)
 {
 	struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
+	struct drm_nouveau_private *dev_priv = nvbe->dev->dev_private;
+	struct nouveau_gpuobj *pgt = dev_priv->gart_info.sg_ctxdma;
+	dma_addr_t *list = nvbe->pages;
+	u32 pte = mem->start << 2;
+	u32 cnt = nvbe->nr_pages;
 
-	if (be) {
-		NV_DEBUG(nvbe->dev, "\n");
+	nvbe->offset = mem->start << PAGE_SHIFT;
 
-		if (nvbe) {
-			if (nvbe->pages)
-				be->func->clear(be);
-			kfree(nvbe);
+	while (cnt--) {
+		nv_wo32(pgt, pte, (*list++ >> 7) | 1);
+		pte += 4;
+	}
+
+	nv41_sgdma_flush(nvbe);
+	nvbe->bound = true;
+	return 0;
+}
+
+static int
+nv41_sgdma_unbind(struct ttm_backend *be)
+{
+	struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
+	struct drm_nouveau_private *dev_priv = nvbe->dev->dev_private;
+	struct nouveau_gpuobj *pgt = dev_priv->gart_info.sg_ctxdma;
+	u32 pte = (nvbe->offset >> 12) << 2;
+	u32 cnt = nvbe->nr_pages;
+
+	while (cnt--) {
+		nv_wo32(pgt, pte, 0x00000000);
+		pte += 4;
+	}
+
+	nv41_sgdma_flush(nvbe);
+	nvbe->bound = false;
+	return 0;
+}
+
+static struct ttm_backend_func nv41_sgdma_backend = {
+	.populate		= nouveau_sgdma_populate,
+	.clear			= nouveau_sgdma_clear,
+	.bind			= nv41_sgdma_bind,
+	.unbind			= nv41_sgdma_unbind,
+	.destroy		= nouveau_sgdma_destroy
+};
+
+static void
+nv44_sgdma_flush(struct nouveau_sgdma_be *nvbe)
+{
+	struct drm_device *dev = nvbe->dev;
+
+	nv_wr32(dev, 0x100814, (nvbe->nr_pages - 1) << 12);
+	nv_wr32(dev, 0x100808, nvbe->offset | 0x20);
+	if (!nv_wait(dev, 0x100808, 0x00000001, 0x00000001))
+		NV_ERROR(dev, "gart flush timeout: 0x%08x\n",
+			 nv_rd32(dev, 0x100808));
+	nv_wr32(dev, 0x100808, 0x00000000);
+}
+
+static void
+nv44_sgdma_fill(struct nouveau_gpuobj *pgt, dma_addr_t *list, u32 base, u32 cnt)
+{
+	struct drm_nouveau_private *dev_priv = pgt->dev->dev_private;
+	dma_addr_t dummy = dev_priv->gart_info.dummy.addr;
+	u32 pte, tmp[4];
+
+	pte   = base >> 2;
+	base &= ~0x0000000f;
+
+	tmp[0] = nv_ro32(pgt, base + 0x0);
+	tmp[1] = nv_ro32(pgt, base + 0x4);
+	tmp[2] = nv_ro32(pgt, base + 0x8);
+	tmp[3] = nv_ro32(pgt, base + 0xc);
+	while (cnt--) {
+		u32 addr = list ? (*list++ >> 12) : (dummy >> 12);
+		switch (pte++ & 0x3) {
+		case 0:
+			tmp[0] &= ~0x07ffffff;
+			tmp[0] |= addr;
+			break;
+		case 1:
+			tmp[0] &= ~0xf8000000;
+			tmp[0] |= addr << 27;
+			tmp[1] &= ~0x003fffff;
+			tmp[1] |= addr >> 5;
+			break;
+		case 2:
+			tmp[1] &= ~0xffc00000;
+			tmp[1] |= addr << 22;
+			tmp[2] &= ~0x0001ffff;
+			tmp[2] |= addr >> 10;
+			break;
+		case 3:
+			tmp[2] &= ~0xfffe0000;
+			tmp[2] |= addr << 17;
+			tmp[3] &= ~0x00000fff;
+			tmp[3] |= addr >> 15;
+			break;
 		}
 	}
+
+	tmp[3] |= 0x40000000;
+
+	nv_wo32(pgt, base + 0x0, tmp[0]);
+	nv_wo32(pgt, base + 0x4, tmp[1]);
+	nv_wo32(pgt, base + 0x8, tmp[2]);
+	nv_wo32(pgt, base + 0xc, tmp[3]);
 }
 
 static int
+nv44_sgdma_bind(struct ttm_backend *be, struct ttm_mem_reg *mem)
+{
+	struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
+	struct drm_nouveau_private *dev_priv = nvbe->dev->dev_private;
+	struct nouveau_gpuobj *pgt = dev_priv->gart_info.sg_ctxdma;
+	dma_addr_t *list = nvbe->pages;
+	u32 pte = mem->start << 2, tmp[4];
+	u32 cnt = nvbe->nr_pages;
+	int i;
+
+	nvbe->offset = mem->start << PAGE_SHIFT;
+
+	if (pte & 0x0000000c) {
+		u32  max = 4 - ((pte >> 2) & 0x3);
+		u32 part = (cnt > max) ? max : cnt;
+		nv44_sgdma_fill(pgt, list, pte, part);
+		pte  += (part << 2);
+		list += part;
+		cnt  -= part;
+	}
+
+	while (cnt >= 4) {
+		for (i = 0; i < 4; i++)
+			tmp[i] = *list++ >> 12;
+		nv_wo32(pgt, pte + 0x0, tmp[0] >>  0 | tmp[1] << 27);
+		nv_wo32(pgt, pte + 0x4, tmp[1] >>  5 | tmp[2] << 22);
+		nv_wo32(pgt, pte + 0x8, tmp[2] >> 10 | tmp[3] << 17);
+		nv_wo32(pgt, pte + 0xc, tmp[3] >> 15 | 0x40000000);
+		pte  += 0x10;
+		cnt  -= 4;
+	}
+
+	if (cnt)
+		nv44_sgdma_fill(pgt, list, pte, cnt);
+
+	nv44_sgdma_flush(nvbe);
+	nvbe->bound = true;
+	return 0;
+}
+
+static int
+nv44_sgdma_unbind(struct ttm_backend *be)
+{
+	struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
+	struct drm_nouveau_private *dev_priv = nvbe->dev->dev_private;
+	struct nouveau_gpuobj *pgt = dev_priv->gart_info.sg_ctxdma;
+	u32 pte = (nvbe->offset >> 12) << 2;
+	u32 cnt = nvbe->nr_pages;
+
+	if (pte & 0x0000000c) {
+		u32  max = 4 - ((pte >> 2) & 0x3);
+		u32 part = (cnt > max) ? max : cnt;
+		nv44_sgdma_fill(pgt, NULL, pte, part);
+		pte  += (part << 2);
+		cnt  -= part;
+	}
+
+	while (cnt >= 4) {
+		nv_wo32(pgt, pte + 0x0, 0x00000000);
+		nv_wo32(pgt, pte + 0x4, 0x00000000);
+		nv_wo32(pgt, pte + 0x8, 0x00000000);
+		nv_wo32(pgt, pte + 0xc, 0x00000000);
+		pte  += 0x10;
+		cnt  -= 4;
+	}
+
+	if (cnt)
+		nv44_sgdma_fill(pgt, NULL, pte, cnt);
+
+	nv44_sgdma_flush(nvbe);
+	nvbe->bound = false;
+	return 0;
+}
+
+static struct ttm_backend_func nv44_sgdma_backend = {
+	.populate		= nouveau_sgdma_populate,
+	.clear			= nouveau_sgdma_clear,
+	.bind			= nv44_sgdma_bind,
+	.unbind			= nv44_sgdma_unbind,
+	.destroy		= nouveau_sgdma_destroy
+};
+
+static int
 nv50_sgdma_bind(struct ttm_backend *be, struct ttm_mem_reg *mem)
 {
 	struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
@@ -170,14 +385,6 @@ nv50_sgdma_unbind(struct ttm_backend *be)
 	return 0;
 }
 
-static struct ttm_backend_func nouveau_sgdma_backend = {
-	.populate		= nouveau_sgdma_populate,
-	.clear			= nouveau_sgdma_clear,
-	.bind			= nouveau_sgdma_bind,
-	.unbind			= nouveau_sgdma_unbind,
-	.destroy		= nouveau_sgdma_destroy
-};
-
 static struct ttm_backend_func nv50_sgdma_backend = {
 	.populate		= nouveau_sgdma_populate,
 	.clear			= nouveau_sgdma_clear,
@@ -198,10 +405,7 @@ nouveau_sgdma_init_ttm(struct drm_device *dev)
 
 	nvbe->dev = dev;
 
-	if (dev_priv->card_type < NV_50)
-		nvbe->backend.func = &nouveau_sgdma_backend;
-	else
-		nvbe->backend.func = &nv50_sgdma_backend;
+	nvbe->backend.func = dev_priv->gart_info.func;
 	return &nvbe->backend;
 }
 
@@ -210,21 +414,71 @@ nouveau_sgdma_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_gpuobj *gpuobj = NULL;
-	uint32_t aper_size, obj_size;
-	int i, ret;
+	u32 aper_size, align;
+	int ret;
+
+	if (dev_priv->card_type >= NV_50 ||
+	    dev_priv->ramin_rsvd_vram >= 2 * 1024 * 1024)
+		aper_size = 512 * 1024 * 1024;
+	else
+		aper_size = 64 * 1024 * 1024;
+
+	/* Dear NVIDIA, NV44+ would like proper present bits in PTEs for
+	 * christmas.  The cards before it have them, the cards after
+	 * it have them, why is NV44 so unloved?
+	 */
+	dev_priv->gart_info.dummy.page = alloc_page(GFP_DMA32 | GFP_KERNEL);
+	if (!dev_priv->gart_info.dummy.page)
+		return -ENOMEM;
+
+	dev_priv->gart_info.dummy.addr =
+		pci_map_page(dev->pdev, dev_priv->gart_info.dummy.page,
+			     0, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
+	if (pci_dma_mapping_error(dev->pdev, dev_priv->gart_info.dummy.addr)) {
+		NV_ERROR(dev, "error mapping dummy page\n");
+		__free_page(dev_priv->gart_info.dummy.page);
+		dev_priv->gart_info.dummy.page = NULL;
+		return -ENOMEM;
+	}
 
-	if (dev_priv->card_type < NV_50) {
-		if(dev_priv->ramin_rsvd_vram < 2 * 1024 * 1024)
-			aper_size = 64 * 1024 * 1024;
-		else
-			aper_size = 512 * 1024 * 1024;
+	if (dev_priv->card_type >= NV_50) {
+		ret = nouveau_vm_get(dev_priv->chan_vm, aper_size,
+				     12, NV_MEM_ACCESS_RW,
+				     &dev_priv->gart_info.vma);
+		if (ret)
+			return ret;
+
+		dev_priv->gart_info.aper_base = dev_priv->gart_info.vma.offset;
+		dev_priv->gart_info.aper_size = aper_size;
+		dev_priv->gart_info.type = NOUVEAU_GART_HW;
+		dev_priv->gart_info.func = &nv50_sgdma_backend;
+	} else
+	if (drm_device_is_pcie(dev) &&
+	    dev_priv->chipset != 0x40 && dev_priv->chipset != 0x45) {
+		if (nv44_graph_class(dev)) {
+			dev_priv->gart_info.func = &nv44_sgdma_backend;
+			align = 512 * 1024;
+		} else {
+			dev_priv->gart_info.func = &nv41_sgdma_backend;
+			align = 16;
+		}
 
-		obj_size  = (aper_size >> NV_CTXDMA_PAGE_SHIFT) * 4;
-		obj_size += 8; /* ctxdma header */
+		ret = nouveau_gpuobj_new(dev, NULL, aper_size / 1024, align,
+					 NVOBJ_FLAG_ZERO_ALLOC |
+					 NVOBJ_FLAG_ZERO_FREE, &gpuobj);
+		if (ret) {
+			NV_ERROR(dev, "Error creating sgdma object: %d\n", ret);
+			return ret;
+		}
 
-		ret = nouveau_gpuobj_new(dev, NULL, obj_size, 16,
-					      NVOBJ_FLAG_ZERO_ALLOC |
-					      NVOBJ_FLAG_ZERO_FREE, &gpuobj);
+		dev_priv->gart_info.sg_ctxdma = gpuobj;
+		dev_priv->gart_info.aper_base = 0;
+		dev_priv->gart_info.aper_size = aper_size;
+		dev_priv->gart_info.type = NOUVEAU_GART_HW;
+	} else {
+		ret = nouveau_gpuobj_new(dev, NULL, (aper_size / 1024) + 8, 16,
+					 NVOBJ_FLAG_ZERO_ALLOC |
+					 NVOBJ_FLAG_ZERO_FREE, &gpuobj);
 		if (ret) {
 			NV_ERROR(dev, "Error creating sgdma object: %d\n", ret);
 			return ret;
@@ -236,25 +490,14 @@ nouveau_sgdma_init(struct drm_device *dev)
 				   (0 << 14) /* RW */ |
 				   (2 << 16) /* PCI */);
 		nv_wo32(gpuobj, 4, aper_size - 1);
-		for (i = 2; i < 2 + (aper_size >> 12); i++)
-			nv_wo32(gpuobj, i * 4, 0x00000000);
 
 		dev_priv->gart_info.sg_ctxdma = gpuobj;
 		dev_priv->gart_info.aper_base = 0;
 		dev_priv->gart_info.aper_size = aper_size;
-	} else
-	if (dev_priv->chan_vm) {
-		ret = nouveau_vm_get(dev_priv->chan_vm, 512 * 1024 * 1024,
-				     12, NV_MEM_ACCESS_RW,
-				     &dev_priv->gart_info.vma);
-		if (ret)
-			return ret;
-
-		dev_priv->gart_info.aper_base = dev_priv->gart_info.vma.offset;
-		dev_priv->gart_info.aper_size = 512 * 1024 * 1024;
+		dev_priv->gart_info.type = NOUVEAU_GART_PDMA;
+		dev_priv->gart_info.func = &nv04_sgdma_backend;
 	}
 
-	dev_priv->gart_info.type      = NOUVEAU_GART_SGDMA;
 	return 0;
 }
 
@@ -265,6 +508,13 @@ nouveau_sgdma_takedown(struct drm_device *dev)
 
 	nouveau_gpuobj_ref(NULL, &dev_priv->gart_info.sg_ctxdma);
 	nouveau_vm_put(&dev_priv->gart_info.vma);
+
+	if (dev_priv->gart_info.dummy.page) {
+		pci_unmap_page(dev->pdev, dev_priv->gart_info.dummy.addr,
+			       PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
+		__free_page(dev_priv->gart_info.dummy.page);
+		dev_priv->gart_info.dummy.page = NULL;
+	}
 }
 
 uint32_t
diff --git a/drivers/gpu/drm/nouveau/nouveau_state.c b/drivers/gpu/drm/nouveau/nouveau_state.c
index a54fc43..916505d 100644
--- a/drivers/gpu/drm/nouveau/nouveau_state.c
+++ b/drivers/gpu/drm/nouveau/nouveau_state.c
@@ -929,12 +929,6 @@ int nouveau_load(struct drm_device *dev, unsigned long flags)
 	NV_DEBUG(dev, "vendor: 0x%X device: 0x%X class: 0x%X\n",
 		 dev->pci_vendor, dev->pci_device, dev->pdev->class);
 
-	dev_priv->wq = create_workqueue("nouveau");
-	if (!dev_priv->wq) {
-		ret = -EINVAL;
-		goto err_priv;
-	}
-
 	/* resource 0 is mmio regs */
 	/* resource 1 is linear FB */
 	/* resource 2 is RAMIN (mmio regs + 0x1000000) */
@@ -947,7 +941,7 @@ int nouveau_load(struct drm_device *dev, unsigned long flags)
 		NV_ERROR(dev, "Unable to initialize the mmio mapping. "
 			 "Please report your setup to " DRIVER_EMAIL "\n");
 		ret = -EINVAL;
-		goto err_wq;
+		goto err_priv;
 	}
 	NV_DEBUG(dev, "regs mapped ok at 0x%llx\n",
 					(unsigned long long)mmio_start_offs);
@@ -1054,8 +1048,6 @@ err_ramin:
 	iounmap(dev_priv->ramin);
 err_mmio:
 	iounmap(dev_priv->mmio);
-err_wq:
-	destroy_workqueue(dev_priv->wq);
 err_priv:
 	kfree(dev_priv);
 	dev->dev_private = NULL;
diff --git a/drivers/gpu/drm/nouveau/nouveau_temp.c b/drivers/gpu/drm/nouveau/nouveau_temp.c
index 8d9968e..649b041 100644
--- a/drivers/gpu/drm/nouveau/nouveau_temp.c
+++ b/drivers/gpu/drm/nouveau/nouveau_temp.c
@@ -239,11 +239,9 @@ static bool
 probe_monitoring_device(struct nouveau_i2c_chan *i2c,
 			struct i2c_board_info *info)
 {
-	char modalias[16] = "i2c:";
 	struct i2c_client *client;
 
-	strlcat(modalias, info->type, sizeof(modalias));
-	request_module(modalias);
+	request_module("%s%s", I2C_MODULE_PREFIX, info->type);
 
 	client = i2c_new_device(&i2c->adapter, info);
 	if (!client)
diff --git a/drivers/gpu/drm/nouveau/nv04_fifo.c b/drivers/gpu/drm/nouveau/nv04_fifo.c
index f89d104..dfa600c 100644
--- a/drivers/gpu/drm/nouveau/nv04_fifo.c
+++ b/drivers/gpu/drm/nouveau/nv04_fifo.c
@@ -379,6 +379,15 @@ out:
 	return handled;
 }
 
+static const char *nv_dma_state_err(u32 state)
+{
+	static const char * const desc[] = {
+		"NONE", "CALL_SUBR_ACTIVE", "INVALID_MTHD", "RET_SUBR_INACTIVE",
+		"INVALID_CMD", "IB_EMPTY"/* NV50+ */, "MEM_FAULT", "UNK"
+	};
+	return desc[(state >> 29) & 0x7];
+}
+
 void
 nv04_fifo_isr(struct drm_device *dev)
 {
@@ -460,9 +469,10 @@ nv04_fifo_isr(struct drm_device *dev)
 				if (nouveau_ratelimit())
 					NV_INFO(dev, "PFIFO_DMA_PUSHER - Ch %d Get 0x%02x%08x "
 					     "Put 0x%02x%08x IbGet 0x%08x IbPut 0x%08x "
-					     "State 0x%08x Push 0x%08x\n",
+					     "State 0x%08x (err: %s) Push 0x%08x\n",
 						chid, ho_get, dma_get, ho_put,
 						dma_put, ib_get, ib_put, state,
+						nv_dma_state_err(state),
 						push);
 
 				/* METHOD_COUNT, in DMA_STATE on earlier chipsets */
@@ -476,8 +486,9 @@ nv04_fifo_isr(struct drm_device *dev)
 				}
 			} else {
 				NV_INFO(dev, "PFIFO_DMA_PUSHER - Ch %d Get 0x%08x "
-					     "Put 0x%08x State 0x%08x Push 0x%08x\n",
-					chid, dma_get, dma_put, state, push);
+					     "Put 0x%08x State 0x%08x (err: %s) Push 0x%08x\n",
+					chid, dma_get, dma_put, state,
+					nv_dma_state_err(state), push);
 
 				if (dma_get != dma_put)
 					nv_wr32(dev, 0x003244, dma_put);
diff --git a/drivers/gpu/drm/nouveau/nv40_fb.c b/drivers/gpu/drm/nouveau/nv40_fb.c
index f3d9c05..f0ac2a7 100644
--- a/drivers/gpu/drm/nouveau/nv40_fb.c
+++ b/drivers/gpu/drm/nouveau/nv40_fb.c
@@ -24,6 +24,53 @@ nv40_fb_set_tile_region(struct drm_device *dev, int i)
 	}
 }
 
+static void
+nv40_fb_init_gart(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *gart = dev_priv->gart_info.sg_ctxdma;
+
+	if (dev_priv->gart_info.type != NOUVEAU_GART_HW) {
+		nv_wr32(dev, 0x100800, 0x00000001);
+		return;
+	}
+
+	nv_wr32(dev, 0x100800, gart->pinst | 0x00000002);
+	nv_mask(dev, 0x10008c, 0x00000100, 0x00000100);
+	nv_wr32(dev, 0x100820, 0x00000000);
+}
+
+static void
+nv44_fb_init_gart(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *gart = dev_priv->gart_info.sg_ctxdma;
+	u32 vinst;
+
+	if (dev_priv->gart_info.type != NOUVEAU_GART_HW) {
+		nv_wr32(dev, 0x100850, 0x80000000);
+		nv_wr32(dev, 0x100800, 0x00000001);
+		return;
+	}
+
+	/* calculate vram address of this PRAMIN block, object
+	 * must be allocated on 512KiB alignment, and not exceed
+	 * a total size of 512KiB for this to work correctly
+	 */
+	vinst  = nv_rd32(dev, 0x10020c);
+	vinst -= ((gart->pinst >> 19) + 1) << 19;
+
+	nv_wr32(dev, 0x100850, 0x80000000);
+	nv_wr32(dev, 0x100818, dev_priv->gart_info.dummy.addr);
+
+	nv_wr32(dev, 0x100804, dev_priv->gart_info.aper_size);
+	nv_wr32(dev, 0x100850, 0x00008000);
+	nv_mask(dev, 0x10008c, 0x00000200, 0x00000200);
+	nv_wr32(dev, 0x100820, 0x00000000);
+	nv_wr32(dev, 0x10082c, 0x00000001);
+	nv_wr32(dev, 0x100800, vinst | 0x00000010);
+}
+
 int
 nv40_fb_init(struct drm_device *dev)
 {
@@ -32,12 +79,12 @@ nv40_fb_init(struct drm_device *dev)
 	uint32_t tmp;
 	int i;
 
-	/* This is strictly a NV4x register (don't know about NV5x). */
-	/* The blob sets these to all kinds of values, and they mess up our setup. */
-	/* I got value 0x52802 instead. For some cards the blob even sets it back to 0x1. */
-	/* Note: the blob doesn't read this value, so i'm pretty sure this is safe for all cards. */
-	/* Any idea what this is? */
-	nv_wr32(dev, NV40_PFB_UNK_800, 0x1);
+	if (dev_priv->chipset != 0x40 && dev_priv->chipset != 0x45) {
+		if (nv44_graph_class(dev))
+			nv44_fb_init_gart(dev);
+		else
+			nv40_fb_init_gart(dev);
+	}
 
 	switch (dev_priv->chipset) {
 	case 0x40:
diff --git a/drivers/gpu/drm/nouveau/nv50_display.c b/drivers/gpu/drm/nouveau/nv50_display.c
index 7cc94ed..a804a35 100644
--- a/drivers/gpu/drm/nouveau/nv50_display.c
+++ b/drivers/gpu/drm/nouveau/nv50_display.c
@@ -345,12 +345,15 @@ int nv50_display_create(struct drm_device *dev)
 void
 nv50_display_destroy(struct drm_device *dev)
 {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
 	NV_DEBUG_KMS(dev, "\n");
 
 	drm_mode_config_cleanup(dev);
 
 	nv50_display_disable(dev);
 	nouveau_irq_unregister(dev, 26);
+	flush_work_sync(&dev_priv->irq_work);
 }
 
 static u16
@@ -587,7 +590,7 @@ static void
 nv50_display_unk20_handler(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	u32 unk30 = nv_rd32(dev, 0x610030), tmp, pclk, script, mc;
+	u32 unk30 = nv_rd32(dev, 0x610030), tmp, pclk, script, mc = 0;
 	struct dcb_entry *dcb;
 	int i, crtc, or, type = OUTPUT_ANY;
 
@@ -836,7 +839,7 @@ nv50_display_isr(struct drm_device *dev)
 		if (clock) {
 			nv_wr32(dev, NV03_PMC_INTR_EN_0, 0);
 			if (!work_pending(&dev_priv->irq_work))
-				queue_work(dev_priv->wq, &dev_priv->irq_work);
+				schedule_work(&dev_priv->irq_work);
 			delayed |= clock;
 			intr1 &= ~clock;
 		}
diff --git a/drivers/gpu/drm/nouveau/nv50_fifo.c b/drivers/gpu/drm/nouveau/nv50_fifo.c
index 8dd04c5..c34a074 100644
--- a/drivers/gpu/drm/nouveau/nv50_fifo.c
+++ b/drivers/gpu/drm/nouveau/nv50_fifo.c
@@ -149,6 +149,7 @@ nv50_fifo_init_regs(struct drm_device *dev)
 	nv_wr32(dev, 0x3204, 0);
 	nv_wr32(dev, 0x3210, 0);
 	nv_wr32(dev, 0x3270, 0);
+	nv_wr32(dev, 0x2044, 0x01003fff);
 
 	/* Enable dummy channels setup by nv50_instmem.c */
 	nv50_fifo_channel_enable(dev, 0);
@@ -273,7 +274,7 @@ nv50_fifo_create_context(struct nouveau_channel *chan)
 	nv_wo32(ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
 			     (4 << 24) /* SEARCH_FULL */ |
 			     (chan->ramht->gpuobj->cinst >> 4));
-	nv_wo32(ramfc, 0x44, 0x2101ffff);
+	nv_wo32(ramfc, 0x44, 0x01003fff);
 	nv_wo32(ramfc, 0x60, 0x7fffffff);
 	nv_wo32(ramfc, 0x40, 0x00000000);
 	nv_wo32(ramfc, 0x7c, 0x30000001);
diff --git a/drivers/gpu/drm/nouveau/nv50_gpio.c b/drivers/gpu/drm/nouveau/nv50_gpio.c
index 6b149c0..d4f4206 100644
--- a/drivers/gpu/drm/nouveau/nv50_gpio.c
+++ b/drivers/gpu/drm/nouveau/nv50_gpio.c
@@ -137,6 +137,7 @@ nv50_gpio_irq_unregister(struct drm_device *dev, enum dcb_gpio_tag tag,
 	struct nv50_gpio_priv *priv = pgpio->priv;
 	struct nv50_gpio_handler *gpioh, *tmp;
 	struct dcb_gpio_entry *gpio;
+	LIST_HEAD(tofree);
 	unsigned long flags;
 
 	gpio = nouveau_bios_gpio_entry(dev, tag);
@@ -149,10 +150,14 @@ nv50_gpio_irq_unregister(struct drm_device *dev, enum dcb_gpio_tag tag,
 		    gpioh->handler != handler ||
 		    gpioh->data != data)
 			continue;
-		list_del(&gpioh->head);
-		kfree(gpioh);
+		list_move(&gpioh->head, &tofree);
 	}
 	spin_unlock_irqrestore(&priv->lock, flags);
+
+	list_for_each_entry_safe(gpioh, tmp, &tofree, head) {
+		flush_work_sync(&gpioh->work);
+		kfree(gpioh);
+	}
 }
 
 bool
@@ -205,7 +210,6 @@ nv50_gpio_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_gpio_engine *pgpio = &dev_priv->engine.gpio;
-	struct nv50_gpio_priv *priv;
 	int ret;
 
 	if (!pgpio->priv) {
@@ -213,7 +217,6 @@ nv50_gpio_init(struct drm_device *dev)
 		if (ret)
 			return ret;
 	}
-	priv = pgpio->priv;
 
 	/* disable, and ack any pending gpio interrupts */
 	nv_wr32(dev, 0xe050, 0x00000000);
@@ -293,7 +296,7 @@ nv50_gpio_isr(struct drm_device *dev)
 			continue;
 		gpioh->inhibit = true;
 
-		queue_work(dev_priv->wq, &gpioh->work);
+		schedule_work(&gpioh->work);
 	}
 	spin_unlock(&priv->lock);
 }
diff --git a/drivers/gpu/drm/nouveau/nv50_graph.c b/drivers/gpu/drm/nouveau/nv50_graph.c
index 37e21d2..c75cff1 100644
--- a/drivers/gpu/drm/nouveau/nv50_graph.c
+++ b/drivers/gpu/drm/nouveau/nv50_graph.c
@@ -912,10 +912,10 @@ nv50_pgraph_trap_handler(struct drm_device *dev, u32 display, u64 inst, u32 chid
 			printk("\n");
 			NV_INFO(dev, "PGRAPH - TRAP_CCACHE %08x %08x %08x %08x"
 				     " %08x %08x %08x\n",
-				nv_rd32(dev, 0x405800), nv_rd32(dev, 0x405804),
-				nv_rd32(dev, 0x405808), nv_rd32(dev, 0x40580c),
-				nv_rd32(dev, 0x405810), nv_rd32(dev, 0x405814),
-				nv_rd32(dev, 0x40581c));
+				nv_rd32(dev, 0x405000), nv_rd32(dev, 0x405004),
+				nv_rd32(dev, 0x405008), nv_rd32(dev, 0x40500c),
+				nv_rd32(dev, 0x405010), nv_rd32(dev, 0x405014),
+				nv_rd32(dev, 0x40501c));
 
 		}
 
diff --git a/drivers/gpu/drm/nouveau/nv50_vm.c b/drivers/gpu/drm/nouveau/nv50_vm.c
index 459ff08..03c1a63 100644
--- a/drivers/gpu/drm/nouveau/nv50_vm.c
+++ b/drivers/gpu/drm/nouveau/nv50_vm.c
@@ -31,7 +31,6 @@ void
 nv50_vm_map_pgt(struct nouveau_gpuobj *pgd, u32 pde,
 		struct nouveau_gpuobj *pgt[2])
 {
-	struct drm_nouveau_private *dev_priv = pgd->dev->dev_private;
 	u64 phys = 0xdeadcafe00000000ULL;
 	u32 coverage = 0;
 
diff --git a/drivers/gpu/drm/nouveau/nvc0_fifo.c b/drivers/gpu/drm/nouveau/nvc0_fifo.c
index e6f92c5..e9f8643 100644
--- a/drivers/gpu/drm/nouveau/nvc0_fifo.c
+++ b/drivers/gpu/drm/nouveau/nvc0_fifo.c
@@ -418,6 +418,12 @@ nvc0_fifo_isr(struct drm_device *dev)
 {
 	u32 stat = nv_rd32(dev, 0x002100);
 
+	if (stat & 0x00000100) {
+		NV_INFO(dev, "PFIFO: unknown status 0x00000100\n");
+		nv_wr32(dev, 0x002100, 0x00000100);
+		stat &= ~0x00000100;
+	}
+
 	if (stat & 0x10000000) {
 		u32 units = nv_rd32(dev, 0x00259c);
 		u32 u = units;
@@ -446,10 +452,15 @@ nvc0_fifo_isr(struct drm_device *dev)
 		stat &= ~0x20000000;
 	}
 
+	if (stat & 0x40000000) {
+		NV_INFO(dev, "PFIFO: unknown status 0x40000000\n");
+		nv_mask(dev, 0x002a00, 0x00000000, 0x00000000);
+		stat &= ~0x40000000;
+	}
+
 	if (stat) {
 		NV_INFO(dev, "PFIFO: unhandled status 0x%08x\n", stat);
 		nv_wr32(dev, 0x002100, stat);
+		nv_wr32(dev, 0x002140, 0);
 	}
-
-	nv_wr32(dev, 0x2140, 0);
 }
diff --git a/drivers/gpu/drm/nouveau/nvc0_graph.c b/drivers/gpu/drm/nouveau/nvc0_graph.c
index eb18a7e..afa7afe 100644
--- a/drivers/gpu/drm/nouveau/nvc0_graph.c
+++ b/drivers/gpu/drm/nouveau/nvc0_graph.c
@@ -640,7 +640,6 @@ nvc0_graph_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct nvc0_graph_priv *priv;
 	int ret;
 
 	dev_priv->engine.graph.accel_blocked = true;
@@ -665,7 +664,6 @@ nvc0_graph_init(struct drm_device *dev)
 		if (ret)
 			return ret;
 	}
-	priv = pgraph->priv;
 
 	nvc0_graph_init_obj418880(dev);
 	nvc0_graph_init_regs(dev);