drm/nouveau: remove nouveau_gpuobj_ref completely, replace with sanity

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

21 files changed, 459 insertions, 583 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_channel.c b/drivers/gpu/drm/nouveau/nouveau_channel.c
index 3144ddea593e..e01396747f6f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_channel.c
+++ b/drivers/gpu/drm/nouveau/nouveau_channel.c
@@ -70,14 +70,8 @@ nouveau_channel_pushbuf_ctxdma_init(struct nouveau_channel *chan)
 		chan->pushbuf_base = pb->bo.mem.mm_node->start << PAGE_SHIFT;
 	}
 
-	ret = nouveau_gpuobj_ref_add(dev, chan, 0, pushbuf, &chan->pushbuf);
-	if (ret) {
-		NV_ERROR(dev, "Error referencing pushbuf ctxdma: %d\n", ret);
-		if (pushbuf != dev_priv->gart_info.sg_ctxdma)
-			nouveau_gpuobj_del(dev, &pushbuf);
-		return ret;
-	}
-
+	nouveau_gpuobj_ref(pushbuf, &chan->pushbuf);
+	nouveau_gpuobj_ref(NULL, &pushbuf);
 	return 0;
 }
 
@@ -308,7 +302,7 @@ nouveau_channel_free(struct nouveau_channel *chan)
 	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
 
 	/* Release the channel's resources */
-	nouveau_gpuobj_ref_del(dev, &chan->pushbuf);
+	nouveau_gpuobj_ref(NULL, &chan->pushbuf);
 	if (chan->pushbuf_bo) {
 		nouveau_bo_unmap(chan->pushbuf_bo);
 		nouveau_bo_unpin(chan->pushbuf_bo);
diff --git a/drivers/gpu/drm/nouveau/nouveau_dma.c b/drivers/gpu/drm/nouveau/nouveau_dma.c
index 2d006993378a..9d27acda87e2 100644
--- a/drivers/gpu/drm/nouveau/nouveau_dma.c
+++ b/drivers/gpu/drm/nouveau/nouveau_dma.c
@@ -28,6 +28,7 @@
 #include "drm.h"
 #include "nouveau_drv.h"
 #include "nouveau_dma.h"
+#include "nouveau_ramht.h"
 
 void
 nouveau_dma_pre_init(struct nouveau_channel *chan)
@@ -58,26 +59,27 @@ nouveau_dma_init(struct nouveau_channel *chan)
 {
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj *m2mf = NULL;
-	struct nouveau_gpuobj *nvsw = NULL;
+	struct nouveau_gpuobj *obj = NULL;
 	int ret, i;
 
 	/* Create NV_MEMORY_TO_MEMORY_FORMAT for buffer moves */
 	ret = nouveau_gpuobj_gr_new(chan, dev_priv->card_type < NV_50 ?
-				    0x0039 : 0x5039, &m2mf);
+				    0x0039 : 0x5039, &obj);
 	if (ret)
 		return ret;
 
-	ret = nouveau_gpuobj_ref_add(dev, chan, NvM2MF, m2mf, NULL);
+	ret = nouveau_ramht_insert(chan, NvM2MF, obj);
+	nouveau_gpuobj_ref(NULL, &obj);
 	if (ret)
 		return ret;
 
 	/* Create an NV_SW object for various sync purposes */
-	ret = nouveau_gpuobj_sw_new(chan, NV_SW, &nvsw);
+	ret = nouveau_gpuobj_sw_new(chan, NV_SW, &obj);
 	if (ret)
 		return ret;
 
-	ret = nouveau_gpuobj_ref_add(dev, chan, NvSw, nvsw, NULL);
+	ret = nouveau_ramht_insert(chan, NvSw, obj);
+	nouveau_gpuobj_ref(NULL, &obj);
 	if (ret)
 		return ret;
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index b8511c122f5f..bf89d0297e21 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -133,7 +133,6 @@ enum nouveau_flags {
 #define NVOBJ_ENGINE_DISPLAY	2
 #define NVOBJ_ENGINE_INT	0xdeadbeef
 
-#define NVOBJ_FLAG_ALLOW_NO_REFS	(1 << 0)
 #define NVOBJ_FLAG_ZERO_ALLOC		(1 << 1)
 #define NVOBJ_FLAG_ZERO_FREE		(1 << 2)
 #define NVOBJ_FLAG_FAKE			(1 << 3)
@@ -141,7 +140,6 @@ struct nouveau_gpuobj {
 	struct drm_device *dev;
 	struct list_head list;
 
-	struct nouveau_channel *im_channel;
 	struct drm_mm_node *im_pramin;
 	struct nouveau_bo *im_backing;
 	uint32_t im_backing_start;
@@ -162,16 +160,6 @@ struct nouveau_gpuobj {
 	void *priv;
 };
 
-struct nouveau_gpuobj_ref {
-	struct list_head list;
-
-	struct nouveau_gpuobj *gpuobj;
-	uint32_t instance;
-
-	struct nouveau_channel *channel;
-	int handle;
-};
-
 struct nouveau_channel {
 	struct drm_device *dev;
 	int id;
@@ -197,33 +185,32 @@ struct nouveau_channel {
 	} fence;
 
 	/* DMA push buffer */
-	struct nouveau_gpuobj_ref *pushbuf;
-	struct nouveau_bo         *pushbuf_bo;
-	uint32_t                   pushbuf_base;
+	struct nouveau_gpuobj *pushbuf;
+	struct nouveau_bo     *pushbuf_bo;
+	uint32_t               pushbuf_base;
 
 	/* Notifier memory */
 	struct nouveau_bo *notifier_bo;
 	struct drm_mm notifier_heap;
 
 	/* PFIFO context */
-	struct nouveau_gpuobj_ref *ramfc;
-	struct nouveau_gpuobj_ref *cache;
+	struct nouveau_gpuobj *ramfc;
+	struct nouveau_gpuobj *cache;
 
 	/* PGRAPH context */
 	/* XXX may be merge 2 pointers as private data ??? */
-	struct nouveau_gpuobj_ref *ramin_grctx;
+	struct nouveau_gpuobj *ramin_grctx;
 	void *pgraph_ctx;
 
 	/* NV50 VM */
-	struct nouveau_gpuobj     *vm_pd;
-	struct nouveau_gpuobj_ref *vm_gart_pt;
-	struct nouveau_gpuobj_ref *vm_vram_pt[NV50_VM_VRAM_NR];
+	struct nouveau_gpuobj *vm_pd;
+	struct nouveau_gpuobj *vm_gart_pt;
+	struct nouveau_gpuobj *vm_vram_pt[NV50_VM_VRAM_NR];
 
 	/* Objects */
-	struct nouveau_gpuobj_ref *ramin; /* Private instmem */
-	struct drm_mm              ramin_heap; /* Private PRAMIN heap */
-	struct nouveau_gpuobj_ref *ramht; /* Hash table */
-	struct list_head           ramht_refs; /* Objects referenced by RAMHT */
+	struct nouveau_gpuobj *ramin; /* Private instmem */
+	struct drm_mm          ramin_heap; /* Private PRAMIN heap */
+	struct nouveau_ramht  *ramht; /* Hash table */
 
 	/* GPU object info for stuff used in-kernel (mm_enabled) */
 	uint32_t m2mf_ntfy;
@@ -301,7 +288,7 @@ struct nouveau_fb_engine {
 struct nouveau_fifo_engine {
 	int  channels;
 
-	struct nouveau_gpuobj_ref *playlist[2];
+	struct nouveau_gpuobj *playlist[2];
 	int cur_playlist;
 
 	int  (*init)(struct drm_device *);
@@ -339,7 +326,7 @@ struct nouveau_pgraph_engine {
 	int grctx_size;
 
 	/* NV2x/NV3x context table (0x400780) */
-	struct nouveau_gpuobj_ref *ctx_table;
+	struct nouveau_gpuobj *ctx_table;
 
 	int  (*init)(struct drm_device *);
 	void (*takedown)(struct drm_device *);
@@ -555,7 +542,7 @@ struct drm_nouveau_private {
 	spinlock_t context_switch_lock;
 
 	/* RAMIN configuration, RAMFC, RAMHT and RAMRO offsets */
-	struct nouveau_gpuobj *ramht;
+	struct nouveau_ramht *ramht;
 	uint32_t ramin_rsvd_vram;
 	uint32_t ramht_offset;
 	uint32_t ramht_size;
@@ -764,24 +751,12 @@ extern void nouveau_gpuobj_channel_takedown(struct nouveau_channel *);
 extern int nouveau_gpuobj_new(struct drm_device *, struct nouveau_channel *,
 			      uint32_t size, int align, uint32_t flags,
 			      struct nouveau_gpuobj **);
-extern int nouveau_gpuobj_del(struct drm_device *, struct nouveau_gpuobj **);
-extern int nouveau_gpuobj_ref_add(struct drm_device *, struct nouveau_channel *,
-				  uint32_t handle, struct nouveau_gpuobj *,
-				  struct nouveau_gpuobj_ref **);
-extern int nouveau_gpuobj_ref_del(struct drm_device *,
-				  struct nouveau_gpuobj_ref **);
-extern int nouveau_gpuobj_ref_find(struct nouveau_channel *, uint32_t handle,
-				   struct nouveau_gpuobj_ref **ref_ret);
-extern int nouveau_gpuobj_new_ref(struct drm_device *,
-				  struct nouveau_channel *alloc_chan,
-				  struct nouveau_channel *ref_chan,
-				  uint32_t handle, uint32_t size, int align,
-				  uint32_t flags, struct nouveau_gpuobj_ref **);
+extern void nouveau_gpuobj_ref(struct nouveau_gpuobj *,
+			       struct nouveau_gpuobj **);
 extern int nouveau_gpuobj_new_fake(struct drm_device *,
 				   uint32_t p_offset, uint32_t b_offset,
 				   uint32_t size, uint32_t flags,
-				   struct nouveau_gpuobj **,
-				   struct nouveau_gpuobj_ref**);
+				   struct nouveau_gpuobj **);
 extern int nouveau_gpuobj_dma_new(struct nouveau_channel *, int class,
 				  uint64_t offset, uint64_t size, int access,
 				  int target, struct nouveau_gpuobj **);
diff --git a/drivers/gpu/drm/nouveau/nouveau_irq.c b/drivers/gpu/drm/nouveau/nouveau_irq.c
index 794b0ee30cf6..9cc3259a54b9 100644
--- a/drivers/gpu/drm/nouveau/nouveau_irq.c
+++ b/drivers/gpu/drm/nouveau/nouveau_irq.c
@@ -35,6 +35,7 @@
 #include "nouveau_drm.h"
 #include "nouveau_drv.h"
 #include "nouveau_reg.h"
+#include "nouveau_ramht.h"
 #include <linux/ratelimit.h>
 
 /* needed for hotplug irq */
@@ -106,15 +107,16 @@ nouveau_fifo_swmthd(struct nouveau_channel *chan, uint32_t addr, uint32_t data)
 	const int mthd = addr & 0x1ffc;
 
 	if (mthd == 0x0000) {
-		struct nouveau_gpuobj_ref *ref = NULL;
+		struct nouveau_gpuobj *gpuobj;
 
-		if (nouveau_gpuobj_ref_find(chan, data, &ref))
+		gpuobj = nouveau_ramht_find(chan, data);
+		if (!gpuobj)
 			return false;
 
-		if (ref->gpuobj->engine != NVOBJ_ENGINE_SW)
+		if (gpuobj->engine != NVOBJ_ENGINE_SW)
 			return false;
 
-		chan->sw_subchannel[subc] = ref->gpuobj->class;
+		chan->sw_subchannel[subc] = gpuobj->class;
 		nv_wr32(dev, NV04_PFIFO_CACHE1_ENGINE, nv_rd32(dev,
 			NV04_PFIFO_CACHE1_ENGINE) & ~(0xf << subc * 4));
 		return true;
@@ -357,7 +359,7 @@ nouveau_graph_chid_from_grctx(struct drm_device *dev)
 			if (!chan || !chan->ramin_grctx)
 				continue;
 
-			if (inst == chan->ramin_grctx->instance)
+			if (inst == chan->ramin_grctx->pinst)
 				break;
 		}
 	} else {
@@ -369,7 +371,7 @@ nouveau_graph_chid_from_grctx(struct drm_device *dev)
 			if (!chan || !chan->ramin)
 				continue;
 
-			if (inst == chan->ramin->instance)
+			if (inst == chan->ramin->vinst)
 				break;
 		}
 	}
@@ -625,7 +627,7 @@ nv50_pfb_vm_trap(struct drm_device *dev, int display, const char *name)
 				if (!chan || !chan->ramin)
 					continue;
 
-				if (trap[1] == chan->ramin->instance >> 12)
+				if (trap[1] == chan->ramin->vinst >> 12)
 					break;
 			}
 			NV_INFO(dev, "%s - VM: Trapped %s at %02x%04x%04x status %08x %08x channel %d\n",
diff --git a/drivers/gpu/drm/nouveau/nouveau_notifier.c b/drivers/gpu/drm/nouveau/nouveau_notifier.c
index 3ec181ff50ce..22b86189b7bb 100644
--- a/drivers/gpu/drm/nouveau/nouveau_notifier.c
+++ b/drivers/gpu/drm/nouveau/nouveau_notifier.c
@@ -28,6 +28,7 @@
 #include "drmP.h"
 #include "drm.h"
 #include "nouveau_drv.h"
+#include "nouveau_ramht.h"
 
 int
 nouveau_notifier_init_channel(struct nouveau_channel *chan)
@@ -146,11 +147,11 @@ nouveau_notifier_alloc(struct nouveau_channel *chan, uint32_t handle,
 	nobj->dtor = nouveau_notifier_gpuobj_dtor;
 	nobj->priv = mem;
 
-	ret = nouveau_gpuobj_ref_add(dev, chan, handle, nobj, NULL);
+	ret = nouveau_ramht_insert(chan, handle, nobj);
+	nouveau_gpuobj_ref(NULL, &nobj);
 	if (ret) {
-		nouveau_gpuobj_del(dev, &nobj);
 		drm_mm_put_block(mem);
-		NV_ERROR(dev, "Error referencing notifier ctxdma: %d\n", ret);
+		NV_ERROR(dev, "Error adding notifier to ramht: %d\n", ret);
 		return ret;
 	}
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_object.c b/drivers/gpu/drm/nouveau/nouveau_object.c
index 552f5131650f..d55c50f1a2d3 100644
--- a/drivers/gpu/drm/nouveau/nouveau_object.c
+++ b/drivers/gpu/drm/nouveau/nouveau_object.c
@@ -90,7 +90,7 @@ nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan,
 	NV_DEBUG(dev, "gpuobj %p\n", gpuobj);
 	gpuobj->dev = dev;
 	gpuobj->flags = flags;
-	gpuobj->im_channel = chan;
+	gpuobj->refcount = 1;
 
 	list_add_tail(&gpuobj->list, &dev_priv->gpuobj_list);
 
@@ -108,7 +108,7 @@ nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan,
 
 		ret = engine->instmem.populate(dev, gpuobj, &size);
 		if (ret) {
-			nouveau_gpuobj_del(dev, &gpuobj);
+			nouveau_gpuobj_ref(NULL, &gpuobj);
 			return ret;
 		}
 	}
@@ -119,14 +119,14 @@ nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan,
 		gpuobj->im_pramin = drm_mm_get_block(gpuobj->im_pramin, size, align);
 
 	if (!gpuobj->im_pramin) {
-		nouveau_gpuobj_del(dev, &gpuobj);
+		nouveau_gpuobj_ref(NULL, &gpuobj);
 		return -ENOMEM;
 	}
 
 	if (!chan) {
 		ret = engine->instmem.bind(dev, gpuobj);
 		if (ret) {
-			nouveau_gpuobj_del(dev, &gpuobj);
+			nouveau_gpuobj_ref(NULL, &gpuobj);
 			return ret;
 		}
 	}
@@ -134,13 +134,13 @@ nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan,
 	/* calculate the various different addresses for the object */
 	if (chan) {
 		gpuobj->pinst = gpuobj->im_pramin->start +
-				chan->ramin->gpuobj->im_pramin->start;
+				chan->ramin->im_pramin->start;
 		if (dev_priv->card_type < NV_50) {
 			gpuobj->cinst = gpuobj->pinst;
 		} else {
 			gpuobj->cinst = gpuobj->im_pramin->start;
 			gpuobj->vinst = gpuobj->im_pramin->start +
-					chan->ramin->gpuobj->im_backing_start;
+					chan->ramin->im_backing_start;
 		}
 	} else {
 		gpuobj->pinst = gpuobj->im_pramin->start;
@@ -156,6 +156,7 @@ nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan,
 		engine->instmem.flush(dev);
 	}
 
+
 	*gpuobj_ret = gpuobj;
 	return 0;
 }
@@ -176,20 +177,23 @@ int
 nouveau_gpuobj_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *ramht = NULL;
 	int ret;
 
 	NV_DEBUG(dev, "\n");
 
-	if (dev_priv->card_type < NV_50) {
-		ret = nouveau_gpuobj_new_fake(dev,
-			dev_priv->ramht_offset, ~0, dev_priv->ramht_size,
-			NVOBJ_FLAG_ZERO_ALLOC | NVOBJ_FLAG_ALLOW_NO_REFS,
-						&dev_priv->ramht, NULL);
-		if (ret)
-			return ret;
-	}
+	if (dev_priv->card_type >= NV_50)
+		return 0;
 
-	return 0;
+	ret = nouveau_gpuobj_new_fake(dev, dev_priv->ramht_offset, ~0,
+				      dev_priv->ramht_size,
+				      NVOBJ_FLAG_ZERO_ALLOC, &ramht);
+	if (ret)
+		return ret;
+
+	ret = nouveau_ramht_new(dev, ramht, &dev_priv->ramht);
+	nouveau_gpuobj_ref(NULL, &ramht);
+	return ret;
 }
 
 void
@@ -199,7 +203,7 @@ nouveau_gpuobj_takedown(struct drm_device *dev)
 
 	NV_DEBUG(dev, "\n");
 
-	nouveau_gpuobj_del(dev, &dev_priv->ramht);
+	nouveau_ramht_ref(NULL, &dev_priv->ramht, NULL);
 }
 
 void
@@ -216,29 +220,21 @@ nouveau_gpuobj_late_takedown(struct drm_device *dev)
 
 		NV_ERROR(dev, "gpuobj %p still exists at takedown, refs=%d\n",
 			 gpuobj, gpuobj->refcount);
-		gpuobj->refcount = 0;
-		nouveau_gpuobj_del(dev, &gpuobj);
+
+		gpuobj->refcount = 1;
+		nouveau_gpuobj_ref(NULL, &gpuobj);
 	}
 }
 
-int
-nouveau_gpuobj_del(struct drm_device *dev, struct nouveau_gpuobj **pgpuobj)
+static int
+nouveau_gpuobj_del(struct nouveau_gpuobj *gpuobj)
 {
+	struct drm_device *dev = gpuobj->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_engine *engine = &dev_priv->engine;
-	struct nouveau_gpuobj *gpuobj;
 	int i;
 
-	NV_DEBUG(dev, "gpuobj %p\n", pgpuobj ? *pgpuobj : NULL);
-
-	if (!dev_priv || !pgpuobj || !(*pgpuobj))
-		return -EINVAL;
-	gpuobj = *pgpuobj;
-
-	if (gpuobj->refcount != 0) {
-		NV_ERROR(dev, "gpuobj refcount is %d\n", gpuobj->refcount);
-		return -EINVAL;
-	}
+	NV_DEBUG(dev, "gpuobj %p\n", gpuobj);
 
 	if (gpuobj->im_pramin && (gpuobj->flags & NVOBJ_FLAG_ZERO_FREE)) {
 		for (i = 0; i < gpuobj->im_pramin->size; i += 4)
@@ -261,181 +257,26 @@ nouveau_gpuobj_del(struct drm_device *dev, struct nouveau_gpuobj **pgpuobj)
 
 	list_del(&gpuobj->list);
 
-	*pgpuobj = NULL;
 	kfree(gpuobj);
 	return 0;
 }
 
-static int
-nouveau_gpuobj_instance_get(struct drm_device *dev,
-			    struct nouveau_channel *chan,
-			    struct nouveau_gpuobj *gpuobj, uint32_t *inst)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj *cpramin;
-
-	/* <NV50 use PRAMIN address everywhere */
-	if (dev_priv->card_type < NV_50) {
-		*inst = gpuobj->im_pramin->start;
-		if (gpuobj->im_channel) {
-			cpramin = gpuobj->im_channel->ramin->gpuobj;
-			*inst += cpramin->im_pramin->start;
-		}
-		return 0;
-	}
-
-	/* NV50 channel-local instance */
-	if (chan) {
-		*inst = gpuobj->im_pramin->start;
-		return 0;
-	}
-
-	/* NV50 global (VRAM) instance */
-	if (!gpuobj->im_channel) {
-		/* ...from global heap */
-		if (!gpuobj->im_backing) {
-			NV_ERROR(dev, "AII, no VRAM backing gpuobj\n");
-			return -EINVAL;
-		}
-		*inst = gpuobj->im_backing_start;
-		return 0;
-	} else {
-		/* ...from local heap */
-		cpramin = gpuobj->im_channel->ramin->gpuobj;
-		*inst = cpramin->im_backing_start + gpuobj->im_pramin->start;
-		return 0;
-	}
-
-	return -EINVAL;
-}
-
-int
-nouveau_gpuobj_ref_add(struct drm_device *dev, struct nouveau_channel *chan,
-		       uint32_t handle, struct nouveau_gpuobj *gpuobj,
-		       struct nouveau_gpuobj_ref **ref_ret)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj_ref *ref;
-	uint32_t instance;
-	int ret;
-
-	NV_DEBUG(dev, "ch%d h=0x%08x gpuobj=%p\n",
-		 chan ? chan->id : -1, handle, gpuobj);
-
-	if (!dev_priv || !gpuobj || (ref_ret && *ref_ret != NULL))
-		return -EINVAL;
-
-	if (!chan && !ref_ret)
-		return -EINVAL;
-
-	if (gpuobj->engine == NVOBJ_ENGINE_SW && !gpuobj->im_pramin) {
-		/* sw object */
-		instance = 0x40;
-	} else {
-		ret = nouveau_gpuobj_instance_get(dev, chan, gpuobj, &instance);
-		if (ret)
-			return ret;
-	}
-
-	ref = kzalloc(sizeof(*ref), GFP_KERNEL);
-	if (!ref)
-		return -ENOMEM;
-	INIT_LIST_HEAD(&ref->list);
-	ref->gpuobj   = gpuobj;
-	ref->channel  = chan;
-	ref->instance = instance;
-
-	if (!ref_ret) {
-		ref->handle = handle;
-
-		ret = nouveau_ramht_insert(dev, ref);
-		if (ret) {
-			kfree(ref);
-			return ret;
-		}
-	} else {
-		ref->handle = ~0;
-		*ref_ret = ref;
-	}
-
-	ref->gpuobj->refcount++;
-	return 0;
-}
-
-int nouveau_gpuobj_ref_del(struct drm_device *dev, struct nouveau_gpuobj_ref **pref)
-{
-	struct nouveau_gpuobj_ref *ref;
-
-	NV_DEBUG(dev, "ref %p\n", pref ? *pref : NULL);
-
-	if (!dev || !pref || *pref == NULL)
-		return -EINVAL;
-	ref = *pref;
-
-	if (ref->handle != ~0)
-		nouveau_ramht_remove(dev, ref);
-
-	if (ref->gpuobj) {
-		ref->gpuobj->refcount--;
-
-		if (ref->gpuobj->refcount == 0) {
-			if (!(ref->gpuobj->flags & NVOBJ_FLAG_ALLOW_NO_REFS))
-				nouveau_gpuobj_del(dev, &ref->gpuobj);
-		}
-	}
-
-	*pref = NULL;
-	kfree(ref);
-	return 0;
-}
-
-int
-nouveau_gpuobj_new_ref(struct drm_device *dev,
-		       struct nouveau_channel *oc, struct nouveau_channel *rc,
-		       uint32_t handle, uint32_t size, int align,
-		       uint32_t flags, struct nouveau_gpuobj_ref **ref)
+void
+nouveau_gpuobj_ref(struct nouveau_gpuobj *ref, struct nouveau_gpuobj **ptr)
 {
-	struct nouveau_gpuobj *gpuobj = NULL;
-	int ret;
+	if (ref)
+		ref->refcount++;
 
-	ret = nouveau_gpuobj_new(dev, oc, size, align, flags, &gpuobj);
-	if (ret)
-		return ret;
+	if (*ptr && --(*ptr)->refcount == 0)
+		nouveau_gpuobj_del(*ptr);
 
-	ret = nouveau_gpuobj_ref_add(dev, rc, handle, gpuobj, ref);
-	if (ret) {
-		nouveau_gpuobj_del(dev, &gpuobj);
-		return ret;
-	}
-
-	return 0;
-}
-
-int
-nouveau_gpuobj_ref_find(struct nouveau_channel *chan, uint32_t handle,
-			struct nouveau_gpuobj_ref **ref_ret)
-{
-	struct nouveau_gpuobj_ref *ref;
-	struct list_head *entry, *tmp;
-
-	list_for_each_safe(entry, tmp, &chan->ramht_refs) {
-		ref = list_entry(entry, struct nouveau_gpuobj_ref, list);
-
-		if (ref->handle == handle) {
-			if (ref_ret)
-				*ref_ret = ref;
-			return 0;
-		}
-	}
-
-	return -EINVAL;
+	*ptr = ref;
 }
 
 int
 nouveau_gpuobj_new_fake(struct drm_device *dev, uint32_t p_offset,
 			uint32_t b_offset, uint32_t size,
-			uint32_t flags, struct nouveau_gpuobj **pgpuobj,
-			struct nouveau_gpuobj_ref **pref)
+			uint32_t flags, struct nouveau_gpuobj **pgpuobj)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_gpuobj *gpuobj = NULL;
@@ -450,8 +291,8 @@ nouveau_gpuobj_new_fake(struct drm_device *dev, uint32_t p_offset,
 		return -ENOMEM;
 	NV_DEBUG(dev, "gpuobj %p\n", gpuobj);
 	gpuobj->dev = dev;
-	gpuobj->im_channel = NULL;
 	gpuobj->flags      = flags | NVOBJ_FLAG_FAKE;
+	gpuobj->refcount = 1;
 
 	list_add_tail(&gpuobj->list, &dev_priv->gpuobj_list);
 
@@ -459,7 +300,7 @@ nouveau_gpuobj_new_fake(struct drm_device *dev, uint32_t p_offset,
 		gpuobj->im_pramin = kzalloc(sizeof(struct drm_mm_node),
 					    GFP_KERNEL);
 		if (!gpuobj->im_pramin) {
-			nouveau_gpuobj_del(dev, &gpuobj);
+			nouveau_gpuobj_ref(NULL, &gpuobj);
 			return -ENOMEM;
 		}
 		gpuobj->im_pramin->start = p_offset;
@@ -481,14 +322,6 @@ nouveau_gpuobj_new_fake(struct drm_device *dev, uint32_t p_offset,
 		dev_priv->engine.instmem.flush(dev);
 	}
 
-	if (pref) {
-		i = nouveau_gpuobj_ref_add(dev, NULL, 0, gpuobj, pref);
-		if (i) {
-			nouveau_gpuobj_del(dev, &gpuobj);
-			return i;
-		}
-	}
-
 	if (pgpuobj)
 		*pgpuobj = gpuobj;
 	return 0;
@@ -628,7 +461,7 @@ nouveau_gpuobj_gart_dma_new(struct nouveau_channel *chan,
 			*o_ret = 0;
 	} else
 	if (dev_priv->gart_info.type == NOUVEAU_GART_SGDMA) {
-		*gpuobj = dev_priv->gart_info.sg_ctxdma;
+		nouveau_gpuobj_ref(dev_priv->gart_info.sg_ctxdma, gpuobj);
 		if (offset & ~0xffffffffULL) {
 			NV_ERROR(dev, "obj offset exceeds 32-bits\n");
 			return -EINVAL;
@@ -760,8 +593,11 @@ nouveau_gpuobj_sw_new(struct nouveau_channel *chan, int class,
 	gpuobj = kzalloc(sizeof(*gpuobj), GFP_KERNEL);
 	if (!gpuobj)
 		return -ENOMEM;
+	gpuobj->dev = chan->dev;
 	gpuobj->engine = NVOBJ_ENGINE_SW;
 	gpuobj->class = class;
+	gpuobj->refcount = 1;
+	gpuobj->cinst = 0x40;
 
 	list_add_tail(&gpuobj->list, &dev_priv->gpuobj_list);
 	*gpuobj_ret = gpuobj;
@@ -773,7 +609,6 @@ nouveau_gpuobj_channel_init_pramin(struct nouveau_channel *chan)
 {
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj *pramin = NULL;
 	uint32_t size;
 	uint32_t base;
 	int ret;
@@ -798,18 +633,16 @@ nouveau_gpuobj_channel_init_pramin(struct nouveau_channel *chan)
 		size += 0x1000;
 	}
 
-	ret = nouveau_gpuobj_new_ref(dev, NULL, NULL, 0, size, 0x1000, 0,
-				     &chan->ramin);
+	ret = nouveau_gpuobj_new(dev, NULL, size, 0x1000, 0, &chan->ramin);
 	if (ret) {
 		NV_ERROR(dev, "Error allocating channel PRAMIN: %d\n", ret);
 		return ret;
 	}
-	pramin = chan->ramin->gpuobj;
 
 	ret = drm_mm_init(&chan->ramin_heap, base, size);
 	if (ret) {
 		NV_ERROR(dev, "Error creating PRAMIN heap: %d\n", ret);
-		nouveau_gpuobj_ref_del(dev, &chan->ramin);
+		nouveau_gpuobj_ref(NULL, &chan->ramin);
 		return ret;
 	}
 
@@ -826,8 +659,6 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 	struct nouveau_gpuobj *vram = NULL, *tt = NULL;
 	int ret, i;
 
-	INIT_LIST_HEAD(&chan->ramht_refs);
-
 	NV_DEBUG(dev, "ch%d vram=0x%08x tt=0x%08x\n", chan->id, vram_h, tt_h);
 
 	/* Allocate a chunk of memory for per-channel object storage */
@@ -846,10 +677,10 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 		uint32_t vm_offset, pde;
 
 		vm_offset = (dev_priv->chipset & 0xf0) == 0x50 ? 0x1400 : 0x200;
-		vm_offset += chan->ramin->gpuobj->im_pramin->start;
+		vm_offset += chan->ramin->im_pramin->start;
 
 		ret = nouveau_gpuobj_new_fake(dev, vm_offset, ~0, 0x4000,
-							0, &chan->vm_pd, NULL);
+					      0, &chan->vm_pd);
 		if (ret)
 			return ret;
 		for (i = 0; i < 0x4000; i += 8) {
@@ -857,25 +688,19 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 			nv_wo32(chan->vm_pd, i + 4, 0xdeadcafe);
 		}
 
+		nouveau_gpuobj_ref(dev_priv->gart_info.sg_ctxdma,
+				   &chan->vm_gart_pt);
 		pde = (dev_priv->vm_gart_base / (512*1024*1024)) * 8;
-		ret = nouveau_gpuobj_ref_add(dev, NULL, 0,
-					     dev_priv->gart_info.sg_ctxdma,
-					     &chan->vm_gart_pt);
-		if (ret)
-			return ret;
-		nv_wo32(chan->vm_pd, pde + 0, chan->vm_gart_pt->instance | 3);
+		nv_wo32(chan->vm_pd, pde + 0, chan->vm_gart_pt->vinst | 3);
 		nv_wo32(chan->vm_pd, pde + 4, 0x00000000);
 
 		pde = (dev_priv->vm_vram_base / (512*1024*1024)) * 8;
 		for (i = 0; i < dev_priv->vm_vram_pt_nr; i++) {
-			ret = nouveau_gpuobj_ref_add(dev, NULL, 0,
-						     dev_priv->vm_vram_pt[i],
-						     &chan->vm_vram_pt[i]);
-			if (ret)
-				return ret;
+			nouveau_gpuobj_ref(dev_priv->vm_vram_pt[i],
+					   &chan->vm_vram_pt[i]);
 
 			nv_wo32(chan->vm_pd, pde + 0,
-				chan->vm_vram_pt[i]->instance | 0x61);
+				chan->vm_vram_pt[i]->vinst | 0x61);
 			nv_wo32(chan->vm_pd, pde + 4, 0x00000000);
 			pde += 8;
 		}
@@ -885,15 +710,17 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 
 	/* RAMHT */
 	if (dev_priv->card_type < NV_50) {
-		ret = nouveau_gpuobj_ref_add(dev, NULL, 0, dev_priv->ramht,
-					     &chan->ramht);
+		nouveau_ramht_ref(dev_priv->ramht, &chan->ramht, NULL);
+	} else {
+		struct nouveau_gpuobj *ramht = NULL;
+
+		ret = nouveau_gpuobj_new(dev, chan, 0x8000, 16,
+					 NVOBJ_FLAG_ZERO_ALLOC, &ramht);
 		if (ret)
 			return ret;
-	} else {
-		ret = nouveau_gpuobj_new_ref(dev, chan, chan, 0,
-					     0x8000, 16,
-					     NVOBJ_FLAG_ZERO_ALLOC,
-					     &chan->ramht);
+
+		ret = nouveau_ramht_new(dev, ramht, &chan->ramht);
+		nouveau_gpuobj_ref(NULL, &ramht);
 		if (ret)
 			return ret;
 	}
@@ -910,24 +737,32 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 		}
 	} else {
 		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
-						0, dev_priv->fb_available_size,
-						NV_DMA_ACCESS_RW,
-						NV_DMA_TARGET_VIDMEM, &vram);
+					     0, dev_priv->fb_available_size,
+					     NV_DMA_ACCESS_RW,
+					     NV_DMA_TARGET_VIDMEM, &vram);
 		if (ret) {
 			NV_ERROR(dev, "Error creating VRAM ctxdma: %d\n", ret);
 			return ret;
 		}
 	}
 
-	ret = nouveau_gpuobj_ref_add(dev, chan, vram_h, vram, NULL);
+	ret = nouveau_ramht_insert(chan, vram_h, vram);
+	nouveau_gpuobj_ref(NULL, &vram);
 	if (ret) {
-		NV_ERROR(dev, "Error referencing VRAM ctxdma: %d\n", ret);
+		NV_ERROR(dev, "Error adding VRAM ctxdma to RAMHT: %d\n", ret);
 		return ret;
 	}
 
 	/* TT memory ctxdma */
 	if (dev_priv->card_type >= NV_50) {
-		tt = vram;
+		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
+					     0, dev_priv->vm_end,
+					     NV_DMA_ACCESS_RW,
+					     NV_DMA_TARGET_AGP, &tt);
+		if (ret) {
+			NV_ERROR(dev, "Error creating VRAM ctxdma: %d\n", ret);
+			return ret;
+		}
 	} else
 	if (dev_priv->gart_info.type != NOUVEAU_GART_NONE) {
 		ret = nouveau_gpuobj_gart_dma_new(chan, 0,
@@ -943,9 +778,10 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 		return ret;
 	}
 
-	ret = nouveau_gpuobj_ref_add(dev, chan, tt_h, tt, NULL);
+	ret = nouveau_ramht_insert(chan, tt_h, tt);
+	nouveau_gpuobj_ref(NULL, &tt);
 	if (ret) {
-		NV_ERROR(dev, "Error referencing TT ctxdma: %d\n", ret);
+		NV_ERROR(dev, "Error adding TT ctxdma to RAMHT: %d\n", ret);
 		return ret;
 	}
 
@@ -957,33 +793,23 @@ nouveau_gpuobj_channel_takedown(struct nouveau_channel *chan)
 {
 	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
 	struct drm_device *dev = chan->dev;
-	struct list_head *entry, *tmp;
-	struct nouveau_gpuobj_ref *ref;
 	int i;
 
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
-	if (!chan->ramht_refs.next)
+	if (!chan->ramht)
 		return;
 
-	list_for_each_safe(entry, tmp, &chan->ramht_refs) {
-		ref = list_entry(entry, struct nouveau_gpuobj_ref, list);
-
-		nouveau_gpuobj_ref_del(dev, &ref);
-	}
-
-	nouveau_gpuobj_ref_del(dev, &chan->ramht);
+	nouveau_ramht_ref(NULL, &chan->ramht, chan);
 
-	nouveau_gpuobj_del(dev, &chan->vm_pd);
-	nouveau_gpuobj_ref_del(dev, &chan->vm_gart_pt);
+	nouveau_gpuobj_ref(NULL, &chan->vm_pd);
+	nouveau_gpuobj_ref(NULL, &chan->vm_gart_pt);
 	for (i = 0; i < dev_priv->vm_vram_pt_nr; i++)
-		nouveau_gpuobj_ref_del(dev, &chan->vm_vram_pt[i]);
+		nouveau_gpuobj_ref(NULL, &chan->vm_vram_pt[i]);
 
 	if (chan->ramin_heap.free_stack.next)
 		drm_mm_takedown(&chan->ramin_heap);
-	if (chan->ramin)
-		nouveau_gpuobj_ref_del(dev, &chan->ramin);
-
+	nouveau_gpuobj_ref(NULL, &chan->ramin);
 }
 
 int
@@ -1095,25 +921,24 @@ int nouveau_ioctl_grobj_alloc(struct drm_device *dev, void *data,
 		return -EPERM;
 	}
 
-	if (nouveau_gpuobj_ref_find(chan, init->handle, NULL) == 0)
+	if (nouveau_ramht_find(chan, init->handle))
 		return -EEXIST;
 
 	if (!grc->software)
 		ret = nouveau_gpuobj_gr_new(chan, grc->id, &gr);
 	else
 		ret = nouveau_gpuobj_sw_new(chan, grc->id, &gr);
-
 	if (ret) {
 		NV_ERROR(dev, "Error creating object: %d (%d/0x%08x)\n",
 			 ret, init->channel, init->handle);
 		return ret;
 	}
 
-	ret = nouveau_gpuobj_ref_add(dev, chan, init->handle, gr, NULL);
+	ret = nouveau_ramht_insert(chan, init->handle, gr);
+	nouveau_gpuobj_ref(NULL, &gr);
 	if (ret) {
 		NV_ERROR(dev, "Error referencing object: %d (%d/0x%08x)\n",
 			 ret, init->channel, init->handle);
-		nouveau_gpuobj_del(dev, &gr);
 		return ret;
 	}
 
@@ -1124,17 +949,16 @@ int nouveau_ioctl_gpuobj_free(struct drm_device *dev, void *data,
 			      struct drm_file *file_priv)
 {
 	struct drm_nouveau_gpuobj_free *objfree = data;
-	struct nouveau_gpuobj_ref *ref;
+	struct nouveau_gpuobj *gpuobj;
 	struct nouveau_channel *chan;
-	int ret;
 
 	NOUVEAU_GET_USER_CHANNEL_WITH_RETURN(objfree->channel, file_priv, chan);
 
-	ret = nouveau_gpuobj_ref_find(chan, objfree->handle, &ref);
-	if (ret)
-		return ret;
-	nouveau_gpuobj_ref_del(dev, &ref);
+	gpuobj = nouveau_ramht_find(chan, objfree->handle);
+	if (!gpuobj)
+		return -ENOENT;
 
+	nouveau_ramht_remove(chan, objfree->handle);
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_ramht.c b/drivers/gpu/drm/nouveau/nouveau_ramht.c
index e5cc93c55d80..5f9d52f06305 100644
--- a/drivers/gpu/drm/nouveau/nouveau_ramht.c
+++ b/drivers/gpu/drm/nouveau/nouveau_ramht.c
@@ -62,48 +62,56 @@ nouveau_ramht_entry_valid(struct drm_device *dev, struct nouveau_gpuobj *ramht,
 }
 
 int
-nouveau_ramht_insert(struct drm_device *dev, struct nouveau_gpuobj_ref *ref)
+nouveau_ramht_insert(struct nouveau_channel *chan, u32 handle,
+		     struct nouveau_gpuobj *gpuobj)
 {
+	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_instmem_engine *instmem = &dev_priv->engine.instmem;
-	struct nouveau_channel *chan = ref->channel;
-	struct nouveau_gpuobj *ramht = chan->ramht ? chan->ramht->gpuobj : NULL;
+	struct nouveau_ramht_entry *entry;
+	struct nouveau_gpuobj *ramht = chan->ramht->gpuobj;
 	uint32_t ctx, co, ho;
 
-	if (!ramht) {
-		NV_ERROR(dev, "No hash table!\n");
-		return -EINVAL;
-	}
+	if (nouveau_ramht_find(chan, handle))
+		return -EEXIST;
+
+	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
+	if (!entry)
+		return -ENOMEM;
+	entry->channel = chan;
+	entry->gpuobj = NULL;
+	entry->handle = handle;
+	list_add(&entry->head, &chan->ramht->entries);
+	nouveau_gpuobj_ref(gpuobj, &entry->gpuobj);
 
 	if (dev_priv->card_type < NV_40) {
-		ctx = NV_RAMHT_CONTEXT_VALID | (ref->instance >> 4) |
+		ctx = NV_RAMHT_CONTEXT_VALID | (gpuobj->cinst >> 4) |
 		      (chan->id << NV_RAMHT_CONTEXT_CHANNEL_SHIFT) |
-		      (ref->gpuobj->engine << NV_RAMHT_CONTEXT_ENGINE_SHIFT);
+		      (gpuobj->engine << NV_RAMHT_CONTEXT_ENGINE_SHIFT);
 	} else
 	if (dev_priv->card_type < NV_50) {
-		ctx = (ref->instance >> 4) |
+		ctx = (gpuobj->cinst >> 4) |
 		      (chan->id << NV40_RAMHT_CONTEXT_CHANNEL_SHIFT) |
-		      (ref->gpuobj->engine << NV40_RAMHT_CONTEXT_ENGINE_SHIFT);
+		      (gpuobj->engine << NV40_RAMHT_CONTEXT_ENGINE_SHIFT);
 	} else {
-		if (ref->gpuobj->engine == NVOBJ_ENGINE_DISPLAY) {
-			ctx = (ref->instance << 10) | 2;
+		if (gpuobj->engine == NVOBJ_ENGINE_DISPLAY) {
+			ctx = (gpuobj->cinst << 10) | 2;
 		} else {
-			ctx = (ref->instance >> 4) |
-			      ((ref->gpuobj->engine <<
+			ctx = (gpuobj->cinst >> 4) |
+			      ((gpuobj->engine <<
 				NV40_RAMHT_CONTEXT_ENGINE_SHIFT));
 		}
 	}
 
-	co = ho = nouveau_ramht_hash_handle(dev, chan->id, ref->handle);
+	co = ho = nouveau_ramht_hash_handle(dev, chan->id, handle);
 	do {
 		if (!nouveau_ramht_entry_valid(dev, ramht, co)) {
 			NV_DEBUG(dev,
 				 "insert ch%d 0x%08x: h=0x%08x, c=0x%08x\n",
-				 chan->id, co, ref->handle, ctx);
-			nv_wo32(ramht, co + 0, ref->handle);
+				 chan->id, co, handle, ctx);
+			nv_wo32(ramht, co + 0, handle);
 			nv_wo32(ramht, co + 4, ctx);
 
-			list_add_tail(&ref->list, &chan->ramht_refs);
 			instmem->flush(dev);
 			return 0;
 		}
@@ -116,35 +124,40 @@ nouveau_ramht_insert(struct drm_device *dev, struct nouveau_gpuobj_ref *ref)
 	} while (co != ho);
 
 	NV_ERROR(dev, "RAMHT space exhausted. ch=%d\n", chan->id);
+	list_del(&entry->head);
+	kfree(entry);
 	return -ENOMEM;
 }
 
 void
-nouveau_ramht_remove(struct drm_device *dev, struct nouveau_gpuobj_ref *ref)
+nouveau_ramht_remove(struct nouveau_channel *chan, u32 handle)
 {
+	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_instmem_engine *instmem = &dev_priv->engine.instmem;
-	struct nouveau_channel *chan = ref->channel;
-	struct nouveau_gpuobj *ramht = chan->ramht ? chan->ramht->gpuobj : NULL;
-	uint32_t co, ho;
-
-	if (!ramht) {
-		NV_ERROR(dev, "No hash table!\n");
-		return;
+	struct nouveau_gpuobj *ramht = chan->ramht->gpuobj;
+	struct nouveau_ramht_entry *entry, *tmp;
+	u32 co, ho;
+
+	list_for_each_entry_safe(entry, tmp, &chan->ramht->entries, head) {
+		if (entry->channel != chan || entry->handle != handle)
+			continue;
+
+		nouveau_gpuobj_ref(NULL, &entry->gpuobj);
+		list_del(&entry->head);
+		kfree(entry);
+		break;
 	}
 
-	co = ho = nouveau_ramht_hash_handle(dev, chan->id, ref->handle);
+	co = ho = nouveau_ramht_hash_handle(dev, chan->id, handle);
 	do {
 		if (nouveau_ramht_entry_valid(dev, ramht, co) &&
-		    (ref->handle == nv_ro32(ramht, co))) {
+		    (handle == nv_ro32(ramht, co))) {
 			NV_DEBUG(dev,
 				 "remove ch%d 0x%08x: h=0x%08x, c=0x%08x\n",
-				 chan->id, co, ref->handle,
-				 nv_ro32(ramht, co + 4));
+				 chan->id, co, handle, nv_ro32(ramht, co + 4));
 			nv_wo32(ramht, co + 0, 0x00000000);
 			nv_wo32(ramht, co + 4, 0x00000000);
-
-			list_del(&ref->list);
 			instmem->flush(dev);
 			return;
 		}
@@ -153,8 +166,64 @@ nouveau_ramht_remove(struct drm_device *dev, struct nouveau_gpuobj_ref *ref)
 		if (co >= dev_priv->ramht_size)
 			co = 0;
 	} while (co != ho);
-	list_del(&ref->list);
 
 	NV_ERROR(dev, "RAMHT entry not found. ch=%d, handle=0x%08x\n",
-		 chan->id, ref->handle);
+		 chan->id, handle);
+}
+
+struct nouveau_gpuobj *
+nouveau_ramht_find(struct nouveau_channel *chan, u32 handle)
+{
+	struct nouveau_ramht_entry *entry;
+
+	list_for_each_entry(entry, &chan->ramht->entries, head) {
+		if (entry->channel == chan && entry->handle == handle)
+			return entry->gpuobj;
+	}
+
+	return NULL;
+}
+
+int
+nouveau_ramht_new(struct drm_device *dev, struct nouveau_gpuobj *gpuobj,
+		  struct nouveau_ramht **pramht)
+{
+	struct nouveau_ramht *ramht;
+
+	ramht = kzalloc(sizeof(*ramht), GFP_KERNEL);
+	if (!ramht)
+		return -ENOMEM;
+
+	ramht->dev = dev;
+	ramht->refcount = 1;
+	INIT_LIST_HEAD(&ramht->entries);
+	nouveau_gpuobj_ref(gpuobj, &ramht->gpuobj);
+
+	*pramht = ramht;
+	return 0;
+}
+
+void
+nouveau_ramht_ref(struct nouveau_ramht *ref, struct nouveau_ramht **ptr,
+		  struct nouveau_channel *chan)
+{
+	struct nouveau_ramht_entry *entry, *tmp;
+	struct nouveau_ramht *ramht;
+
+	if (ref)
+		ref->refcount++;
+
+	ramht = *ptr;
+	if (ramht) {
+		list_for_each_entry_safe(entry, tmp, &ramht->entries, head) {
+			if (entry->channel == chan)
+				nouveau_ramht_remove(chan, entry->handle);
+		}
+
+		if (--ramht->refcount == 0) {
+			nouveau_gpuobj_ref(NULL, &ramht->gpuobj);
+			kfree(ramht);
+		}
+	}
+	*ptr = ref;
 }
diff --git a/drivers/gpu/drm/nouveau/nouveau_ramht.h b/drivers/gpu/drm/nouveau/nouveau_ramht.h
index e10455c6e7ff..7076ae4c07a5 100644
--- a/drivers/gpu/drm/nouveau/nouveau_ramht.h
+++ b/drivers/gpu/drm/nouveau/nouveau_ramht.h
@@ -25,7 +25,29 @@
 #ifndef __NOUVEAU_RAMHT_H__
 #define __NOUVEAU_RAMHT_H__
 
-extern int nouveau_ramht_insert(struct drm_device *, struct nouveau_gpuobj_ref *);
-extern void nouveau_ramht_remove(struct drm_device *, struct nouveau_gpuobj_ref *);
+struct nouveau_ramht_entry {
+	struct list_head head;
+	struct nouveau_channel *channel;
+	struct nouveau_gpuobj *gpuobj;
+	u32 handle;
+};
+
+struct nouveau_ramht {
+	struct drm_device *dev;
+	int refcount;
+	struct nouveau_gpuobj *gpuobj;
+	struct list_head entries;
+};
+
+extern int  nouveau_ramht_new(struct drm_device *, struct nouveau_gpuobj *,
+			      struct nouveau_ramht **);
+extern void nouveau_ramht_ref(struct nouveau_ramht *, struct nouveau_ramht **,
+			      struct nouveau_channel *unref_channel);
+
+extern int  nouveau_ramht_insert(struct nouveau_channel *, u32 handle,
+				 struct nouveau_gpuobj *);
+extern void nouveau_ramht_remove(struct nouveau_channel *, u32 handle);
+extern struct nouveau_gpuobj *
+nouveau_ramht_find(struct nouveau_channel *chan, u32 handle);
 
 #endif
diff --git a/drivers/gpu/drm/nouveau/nouveau_sgdma.c b/drivers/gpu/drm/nouveau/nouveau_sgdma.c
index 630988af801c..5a66a7ae6e29 100644
--- a/drivers/gpu/drm/nouveau/nouveau_sgdma.c
+++ b/drivers/gpu/drm/nouveau/nouveau_sgdma.c
@@ -234,7 +234,6 @@ nouveau_sgdma_init(struct drm_device *dev)
 	}
 
 	ret = nouveau_gpuobj_new(dev, NULL, obj_size, 16,
-				      NVOBJ_FLAG_ALLOW_NO_REFS |
 				      NVOBJ_FLAG_ZERO_ALLOC |
 				      NVOBJ_FLAG_ZERO_FREE, &gpuobj);
 	if (ret) {
@@ -245,7 +244,7 @@ nouveau_sgdma_init(struct drm_device *dev)
 	dev_priv->gart_info.sg_dummy_page =
 		alloc_page(GFP_KERNEL|__GFP_DMA32);
 	if (!dev_priv->gart_info.sg_dummy_page) {
-		nouveau_gpuobj_del(dev, &gpuobj);
+		nouveau_gpuobj_ref(NULL, &gpuobj);
 		return -ENOMEM;
 	}
 
@@ -254,11 +253,17 @@ nouveau_sgdma_init(struct drm_device *dev)
 		pci_map_page(pdev, dev_priv->gart_info.sg_dummy_page, 0,
 			     PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
 	if (pci_dma_mapping_error(pdev, dev_priv->gart_info.sg_dummy_bus)) {
-		nouveau_gpuobj_del(dev, &gpuobj);
+		nouveau_gpuobj_ref(NULL, &gpuobj);
 		return -EFAULT;
 	}
 
 	if (dev_priv->card_type < NV_50) {
+		/* special case, allocated from global instmem heap so
+		 * cinst is invalid, we use it on all channels though so
+		 * cinst needs to be valid, set it the same as pinst
+		 */
+		gpuobj->cinst = gpuobj->pinst;
+
 		/* Maybe use NV_DMA_TARGET_AGP for PCIE? NVIDIA do this, and
 		 * confirmed to work on c51.  Perhaps means NV_DMA_TARGET_PCIE
 		 * on those cards? */
@@ -302,7 +307,7 @@ nouveau_sgdma_takedown(struct drm_device *dev)
 		dev_priv->gart_info.sg_dummy_bus = 0;
 	}
 
-	nouveau_gpuobj_del(dev, &dev_priv->gart_info.sg_ctxdma);
+	nouveau_gpuobj_ref(NULL, &dev_priv->gart_info.sg_ctxdma);
 }
 
 int
diff --git a/drivers/gpu/drm/nouveau/nouveau_state.c b/drivers/gpu/drm/nouveau/nouveau_state.c
index 989322be3728..fec29522298d 100644
--- a/drivers/gpu/drm/nouveau/nouveau_state.c
+++ b/drivers/gpu/drm/nouveau/nouveau_state.c
@@ -35,6 +35,7 @@
 #include "nouveau_drv.h"
 #include "nouveau_drm.h"
 #include "nouveau_fbcon.h"
+#include "nouveau_ramht.h"
 #include "nv50_display.h"
 
 static void nouveau_stub_takedown(struct drm_device *dev) {}
@@ -437,16 +438,14 @@ static int
 nouveau_card_init_channel(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj *gpuobj;
+	struct nouveau_gpuobj *gpuobj = NULL;
 	int ret;
 
 	ret = nouveau_channel_alloc(dev, &dev_priv->channel,
-				    (struct drm_file *)-2,
-				    NvDmaFB, NvDmaTT);
+				    (struct drm_file *)-2, NvDmaFB, NvDmaTT);
 	if (ret)
 		return ret;
 
-	gpuobj = NULL;
 	ret = nouveau_gpuobj_dma_new(dev_priv->channel, NV_CLASS_DMA_IN_MEMORY,
 				     0, dev_priv->vram_size,
 				     NV_DMA_ACCESS_RW, NV_DMA_TARGET_VIDMEM,
@@ -454,26 +453,25 @@ nouveau_card_init_channel(struct drm_device *dev)
 	if (ret)
 		goto out_err;
 
-	ret = nouveau_gpuobj_ref_add(dev, dev_priv->channel, NvDmaVRAM,
-				     gpuobj, NULL);
+	ret = nouveau_ramht_insert(dev_priv->channel, NvDmaVRAM, gpuobj);
+	nouveau_gpuobj_ref(NULL, &gpuobj);
 	if (ret)
 		goto out_err;
 
-	gpuobj = NULL;
 	ret = nouveau_gpuobj_gart_dma_new(dev_priv->channel, 0,
 					  dev_priv->gart_info.aper_size,
 					  NV_DMA_ACCESS_RW, &gpuobj, NULL);
 	if (ret)
 		goto out_err;
 
-	ret = nouveau_gpuobj_ref_add(dev, dev_priv->channel, NvDmaGART,
-				     gpuobj, NULL);
+	ret = nouveau_ramht_insert(dev_priv->channel, NvDmaGART, gpuobj);
+	nouveau_gpuobj_ref(NULL, &gpuobj);
 	if (ret)
 		goto out_err;
 
 	return 0;
+
 out_err:
-	nouveau_gpuobj_del(dev, &gpuobj);
 	nouveau_channel_free(dev_priv->channel);
 	dev_priv->channel = NULL;
 	return ret;
diff --git a/drivers/gpu/drm/nouveau/nv04_fbcon.c b/drivers/gpu/drm/nouveau/nv04_fbcon.c
index 1eeac4fae73d..33e4c9388bc1 100644
--- a/drivers/gpu/drm/nouveau/nv04_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nv04_fbcon.c
@@ -25,6 +25,7 @@
 #include "drmP.h"
 #include "nouveau_drv.h"
 #include "nouveau_dma.h"
+#include "nouveau_ramht.h"
 #include "nouveau_fbcon.h"
 
 void
@@ -169,11 +170,9 @@ nv04_fbcon_grobj_new(struct drm_device *dev, int class, uint32_t handle)
 	if (ret)
 		return ret;
 
-	ret = nouveau_gpuobj_ref_add(dev, dev_priv->channel, handle, obj, NULL);
-	if (ret)
-		return ret;
-
-	return 0;
+	ret = nouveau_ramht_insert(dev_priv->channel, handle, obj);
+	nouveau_gpuobj_ref(NULL, &obj);
+	return ret;
 }
 
 int
diff --git a/drivers/gpu/drm/nouveau/nv04_fifo.c b/drivers/gpu/drm/nouveau/nv04_fifo.c
index bbb87ef262c0..b7ecafb78d77 100644
--- a/drivers/gpu/drm/nouveau/nv04_fifo.c
+++ b/drivers/gpu/drm/nouveau/nv04_fifo.c
@@ -38,10 +38,8 @@
 #define NV04_RAMFC_ENGINE                                        0x14
 #define NV04_RAMFC_PULL1_ENGINE                                  0x18
 
-#define RAMFC_WR(offset, val) nv_wo32(chan->ramfc->gpuobj, \
-				      NV04_RAMFC_##offset, (val))
-#define RAMFC_RD(offset)      nv_ro32(chan->ramfc->gpuobj, \
-				      NV04_RAMFC_##offset)
+#define RAMFC_WR(offset, val) nv_wo32(chan->ramfc, NV04_RAMFC_##offset, (val))
+#define RAMFC_RD(offset)      nv_ro32(chan->ramfc, NV04_RAMFC_##offset)
 
 void
 nv04_fifo_disable(struct drm_device *dev)
@@ -130,7 +128,7 @@ nv04_fifo_create_context(struct nouveau_channel *chan)
 						NV04_RAMFC__SIZE,
 						NVOBJ_FLAG_ZERO_ALLOC |
 						NVOBJ_FLAG_ZERO_FREE,
-						NULL, &chan->ramfc);
+						&chan->ramfc);
 	if (ret)
 		return ret;
 
@@ -139,7 +137,7 @@ nv04_fifo_create_context(struct nouveau_channel *chan)
 	/* Setup initial state */
 	RAMFC_WR(DMA_PUT, chan->pushbuf_base);
 	RAMFC_WR(DMA_GET, chan->pushbuf_base);
-	RAMFC_WR(DMA_INSTANCE, chan->pushbuf->instance >> 4);
+	RAMFC_WR(DMA_INSTANCE, chan->pushbuf->pinst >> 4);
 	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 |
@@ -161,7 +159,7 @@ nv04_fifo_destroy_context(struct nouveau_channel *chan)
 	nv_wr32(dev, NV04_PFIFO_MODE,
 		nv_rd32(dev, NV04_PFIFO_MODE) & ~(1 << chan->id));
 
-	nouveau_gpuobj_ref_del(dev, &chan->ramfc);
+	nouveau_gpuobj_ref(NULL, &chan->ramfc);
 }
 
 static void
diff --git a/drivers/gpu/drm/nouveau/nv10_fifo.c b/drivers/gpu/drm/nouveau/nv10_fifo.c
index 7a4069cf5d0b..ccb07fb701ca 100644
--- a/drivers/gpu/drm/nouveau/nv10_fifo.c
+++ b/drivers/gpu/drm/nouveau/nv10_fifo.c
@@ -48,7 +48,7 @@ nv10_fifo_create_context(struct nouveau_channel *chan)
 
 	ret = nouveau_gpuobj_new_fake(dev, NV10_RAMFC(chan->id), ~0,
 				      NV10_RAMFC__SIZE, NVOBJ_FLAG_ZERO_ALLOC |
-				      NVOBJ_FLAG_ZERO_FREE, NULL, &chan->ramfc);
+				      NVOBJ_FLAG_ZERO_FREE, &chan->ramfc);
 	if (ret)
 		return ret;
 
@@ -57,7 +57,7 @@ nv10_fifo_create_context(struct nouveau_channel *chan)
 	 */
 	nv_wi32(dev, fc +  0, chan->pushbuf_base);
 	nv_wi32(dev, fc +  4, chan->pushbuf_base);
-	nv_wi32(dev, fc + 12, chan->pushbuf->instance >> 4);
+	nv_wi32(dev, fc + 12, chan->pushbuf->pinst >> 4);
 	nv_wi32(dev, fc + 20, NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
 			      NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
 			      NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8 |
@@ -80,7 +80,7 @@ nv10_fifo_destroy_context(struct nouveau_channel *chan)
 	nv_wr32(dev, NV04_PFIFO_MODE,
 			nv_rd32(dev, NV04_PFIFO_MODE) & ~(1 << chan->id));
 
-	nouveau_gpuobj_ref_del(dev, &chan->ramfc);
+	nouveau_gpuobj_ref(NULL, &chan->ramfc);
 }
 
 static void
diff --git a/drivers/gpu/drm/nouveau/nv20_graph.c b/drivers/gpu/drm/nouveau/nv20_graph.c
index d8693d32bd0e..12ab9cd56eca 100644
--- a/drivers/gpu/drm/nouveau/nv20_graph.c
+++ b/drivers/gpu/drm/nouveau/nv20_graph.c
@@ -403,21 +403,19 @@ nv20_graph_create_context(struct nouveau_channel *chan)
 		BUG_ON(1);
 	}
 
-	ret = nouveau_gpuobj_new_ref(dev, chan, NULL, 0, pgraph->grctx_size,
-				     16, NVOBJ_FLAG_ZERO_ALLOC,
-				     &chan->ramin_grctx);
+	ret = nouveau_gpuobj_new(dev, chan, pgraph->grctx_size, 16,
+				 NVOBJ_FLAG_ZERO_ALLOC, &chan->ramin_grctx);
 	if (ret)
 		return ret;
 
 	/* Initialise default context values */
-	ctx_init(dev, chan->ramin_grctx->gpuobj);
+	ctx_init(dev, chan->ramin_grctx);
 
 	/* nv20: nv_wo32(dev, chan->ramin_grctx->gpuobj, 10, chan->id<<24); */
-	nv_wo32(chan->ramin_grctx->gpuobj, idoffs,
+	nv_wo32(chan->ramin_grctx, idoffs,
 		(chan->id << 24) | 0x1); /* CTX_USER */
 
-	nv_wo32(pgraph->ctx_table->gpuobj, chan->id * 4,
-		chan->ramin_grctx->instance >> 4);
+	nv_wo32(pgraph->ctx_table, chan->id * 4, chan->ramin_grctx->pinst >> 4);
 	return 0;
 }
 
@@ -428,10 +426,8 @@ nv20_graph_destroy_context(struct nouveau_channel *chan)
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 
-	if (chan->ramin_grctx)
-		nouveau_gpuobj_ref_del(dev, &chan->ramin_grctx);
-
-	nv_wo32(pgraph->ctx_table->gpuobj, chan->id * 4, 0);
+	nouveau_gpuobj_ref(NULL, &chan->ramin_grctx);
+	nv_wo32(pgraph->ctx_table, chan->id * 4, 0);
 }
 
 int
@@ -442,7 +438,7 @@ nv20_graph_load_context(struct nouveau_channel *chan)
 
 	if (!chan->ramin_grctx)
 		return -EINVAL;
-	inst = chan->ramin_grctx->instance >> 4;
+	inst = chan->ramin_grctx->pinst >> 4;
 
 	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_POINTER, inst);
 	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_XFER,
@@ -465,7 +461,7 @@ nv20_graph_unload_context(struct drm_device *dev)
 	chan = pgraph->channel(dev);
 	if (!chan)
 		return 0;
-	inst = chan->ramin_grctx->instance >> 4;
+	inst = chan->ramin_grctx->pinst >> 4;
 
 	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_POINTER, inst);
 	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_XFER,
@@ -552,15 +548,15 @@ nv20_graph_init(struct drm_device *dev)
 
 	if (!pgraph->ctx_table) {
 		/* Create Context Pointer Table */
-		ret = nouveau_gpuobj_new_ref(dev, NULL, NULL, 0, 32 * 4, 16,
-						  NVOBJ_FLAG_ZERO_ALLOC,
-						  &pgraph->ctx_table);
+		ret = nouveau_gpuobj_new(dev, NULL, 32 * 4, 16,
+					 NVOBJ_FLAG_ZERO_ALLOC,
+					 &pgraph->ctx_table);
 		if (ret)
 			return ret;
 	}
 
 	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_TABLE,
-		     pgraph->ctx_table->instance >> 4);
+		     pgraph->ctx_table->pinst >> 4);
 
 	nv20_graph_rdi(dev);
 
@@ -646,7 +642,7 @@ nv20_graph_takedown(struct drm_device *dev)
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 
-	nouveau_gpuobj_ref_del(dev, &pgraph->ctx_table);
+	nouveau_gpuobj_ref(NULL, &pgraph->ctx_table);
 }
 
 int
@@ -681,15 +677,15 @@ nv30_graph_init(struct drm_device *dev)
 
 	if (!pgraph->ctx_table) {
 		/* Create Context Pointer Table */
-		ret = nouveau_gpuobj_new_ref(dev, NULL, NULL, 0, 32 * 4, 16,
-						  NVOBJ_FLAG_ZERO_ALLOC,
-						  &pgraph->ctx_table);
+		ret = nouveau_gpuobj_new(dev, NULL, 32 * 4, 16,
+					 NVOBJ_FLAG_ZERO_ALLOC,
+					 &pgraph->ctx_table);
 		if (ret)
 			return ret;
 	}
 
 	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_TABLE,
-		     pgraph->ctx_table->instance >> 4);
+		     pgraph->ctx_table->pinst >> 4);
 
 	nv_wr32(dev, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
 	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
diff --git a/drivers/gpu/drm/nouveau/nv40_fifo.c b/drivers/gpu/drm/nouveau/nv40_fifo.c
index 2b67f1835c39..03f4dc13725b 100644
--- a/drivers/gpu/drm/nouveau/nv40_fifo.c
+++ b/drivers/gpu/drm/nouveau/nv40_fifo.c
@@ -42,7 +42,7 @@ nv40_fifo_create_context(struct nouveau_channel *chan)
 
 	ret = nouveau_gpuobj_new_fake(dev, NV40_RAMFC(chan->id), ~0,
 				      NV40_RAMFC__SIZE, NVOBJ_FLAG_ZERO_ALLOC |
-				      NVOBJ_FLAG_ZERO_FREE, NULL, &chan->ramfc);
+				      NVOBJ_FLAG_ZERO_FREE, &chan->ramfc);
 	if (ret)
 		return ret;
 
@@ -50,7 +50,7 @@ nv40_fifo_create_context(struct nouveau_channel *chan)
 
 	nv_wi32(dev, fc +  0, chan->pushbuf_base);
 	nv_wi32(dev, fc +  4, chan->pushbuf_base);
-	nv_wi32(dev, fc + 12, chan->pushbuf->instance >> 4);
+	nv_wi32(dev, fc + 12, chan->pushbuf->pinst >> 4);
 	nv_wi32(dev, fc + 24, NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
 			      NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
 			      NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8 |
@@ -58,7 +58,7 @@ nv40_fifo_create_context(struct nouveau_channel *chan)
 			      NV_PFIFO_CACHE1_BIG_ENDIAN |
 #endif
 			      0x30000000 /* no idea.. */);
-	nv_wi32(dev, fc + 56, chan->ramin_grctx->instance >> 4);
+	nv_wi32(dev, fc + 56, chan->ramin_grctx->pinst >> 4);
 	nv_wi32(dev, fc + 60, 0x0001FFFF);
 
 	/* enable the fifo dma operation */
@@ -77,8 +77,7 @@ nv40_fifo_destroy_context(struct nouveau_channel *chan)
 	nv_wr32(dev, NV04_PFIFO_MODE,
 		nv_rd32(dev, NV04_PFIFO_MODE) & ~(1 << chan->id));
 
-	if (chan->ramfc)
-		nouveau_gpuobj_ref_del(dev, &chan->ramfc);
+	nouveau_gpuobj_ref(NULL, &chan->ramfc);
 }
 
 static void
diff --git a/drivers/gpu/drm/nouveau/nv40_graph.c b/drivers/gpu/drm/nouveau/nv40_graph.c
index 6215dfcf1ea9..912940e2457d 100644
--- a/drivers/gpu/drm/nouveau/nv40_graph.c
+++ b/drivers/gpu/drm/nouveau/nv40_graph.c
@@ -45,7 +45,7 @@ nv40_graph_channel(struct drm_device *dev)
 		struct nouveau_channel *chan = dev_priv->fifos[i];
 
 		if (chan && chan->ramin_grctx &&
-		    chan->ramin_grctx->instance == inst)
+		    chan->ramin_grctx->pinst == inst)
 			return chan;
 	}
 
@@ -61,27 +61,25 @@ nv40_graph_create_context(struct nouveau_channel *chan)
 	struct nouveau_grctx ctx = {};
 	int ret;
 
-	ret = nouveau_gpuobj_new_ref(dev, chan, NULL, 0, pgraph->grctx_size,
-				     16, NVOBJ_FLAG_ZERO_ALLOC,
-				     &chan->ramin_grctx);
+	ret = nouveau_gpuobj_new(dev, chan, pgraph->grctx_size, 16,
+				 NVOBJ_FLAG_ZERO_ALLOC, &chan->ramin_grctx);
 	if (ret)
 		return ret;
 
 	/* Initialise default context values */
 	ctx.dev = chan->dev;
 	ctx.mode = NOUVEAU_GRCTX_VALS;
-	ctx.data = chan->ramin_grctx->gpuobj;
+	ctx.data = chan->ramin_grctx;
 	nv40_grctx_init(&ctx);
 
-	nv_wo32(chan->ramin_grctx->gpuobj, 0,
-		chan->ramin_grctx->gpuobj->im_pramin->start);
+	nv_wo32(chan->ramin_grctx, 0, chan->ramin_grctx->im_pramin->start);
 	return 0;
 }
 
 void
 nv40_graph_destroy_context(struct nouveau_channel *chan)
 {
-	nouveau_gpuobj_ref_del(chan->dev, &chan->ramin_grctx);
+	nouveau_gpuobj_ref(NULL, &chan->ramin_grctx);
 }
 
 static int
@@ -135,7 +133,7 @@ nv40_graph_load_context(struct nouveau_channel *chan)
 
 	if (!chan->ramin_grctx)
 		return -EINVAL;
-	inst = chan->ramin_grctx->instance >> 4;
+	inst = chan->ramin_grctx->pinst >> 4;
 
 	ret = nv40_graph_transfer_context(dev, inst, 0);
 	if (ret)
diff --git a/drivers/gpu/drm/nouveau/nv50_display.c b/drivers/gpu/drm/nouveau/nv50_display.c
index 6f89674ebb96..b65d2ddd415d 100644
--- a/drivers/gpu/drm/nouveau/nv50_display.c
+++ b/drivers/gpu/drm/nouveau/nv50_display.c
@@ -30,6 +30,7 @@
 #include "nouveau_connector.h"
 #include "nouveau_fb.h"
 #include "nouveau_fbcon.h"
+#include "nouveau_ramht.h"
 #include "drm_crtc_helper.h"
 
 static void
@@ -66,12 +67,6 @@ nv50_evo_dmaobj_new(struct nouveau_channel *evo, uint32_t class, uint32_t name,
 		return ret;
 	obj->engine = NVOBJ_ENGINE_DISPLAY;
 
-	ret = nouveau_gpuobj_ref_add(dev, evo, name, obj, NULL);
-	if (ret) {
-		nouveau_gpuobj_del(dev, &obj);
-		return ret;
-	}
-
 	nv_wo32(obj,  0, (tile_flags << 22) | (magic_flags << 16) | class);
 	nv_wo32(obj,  4, limit);
 	nv_wo32(obj,  8, offset);
@@ -83,6 +78,12 @@ nv50_evo_dmaobj_new(struct nouveau_channel *evo, uint32_t class, uint32_t name,
 		nv_wo32(obj, 20, 0x00020000);
 	dev_priv->engine.instmem.flush(dev);
 
+	ret = nouveau_ramht_insert(evo, name, obj);
+	nouveau_gpuobj_ref(NULL, &obj);
+	if (ret) {
+		return ret;
+	}
+
 	return 0;
 }
 
@@ -90,6 +91,7 @@ static int
 nv50_evo_channel_new(struct drm_device *dev, struct nouveau_channel **pchan)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *ramht = NULL;
 	struct nouveau_channel *chan;
 	int ret;
 
@@ -103,10 +105,8 @@ nv50_evo_channel_new(struct drm_device *dev, struct nouveau_channel **pchan)
 	chan->user_get = 4;
 	chan->user_put = 0;
 
-	INIT_LIST_HEAD(&chan->ramht_refs);
-
-	ret = nouveau_gpuobj_new_ref(dev, NULL, NULL, 0, 32768, 0x1000,
-				     NVOBJ_FLAG_ZERO_ALLOC, &chan->ramin);
+	ret = nouveau_gpuobj_new(dev, NULL, 32768, 0x1000,
+				 NVOBJ_FLAG_ZERO_ALLOC, &chan->ramin);
 	if (ret) {
 		NV_ERROR(dev, "Error allocating EVO channel memory: %d\n", ret);
 		nv50_evo_channel_del(pchan);
@@ -120,14 +120,20 @@ nv50_evo_channel_new(struct drm_device *dev, struct nouveau_channel **pchan)
 		return ret;
 	}
 
-	ret = nouveau_gpuobj_new_ref(dev, chan, chan, 0, 4096, 16,
-				     0, &chan->ramht);
+	ret = nouveau_gpuobj_new(dev, chan, 4096, 16, 0, &ramht);
 	if (ret) {
 		NV_ERROR(dev, "Unable to allocate EVO RAMHT: %d\n", ret);
 		nv50_evo_channel_del(pchan);
 		return ret;
 	}
 
+	ret = nouveau_ramht_new(dev, ramht, &chan->ramht);
+	nouveau_gpuobj_ref(NULL, &ramht);
+	if (ret) {
+		nv50_evo_channel_del(pchan);
+		return ret;
+	}
+
 	if (dev_priv->chipset != 0x50) {
 		ret = nv50_evo_dmaobj_new(chan, 0x3d, NvEvoFB16, 0x70, 0x19,
 					  0, 0xffffffff);
@@ -321,7 +327,7 @@ nv50_display_init(struct drm_device *dev)
 		}
 	}
 
-	nv_wr32(dev, NV50_PDISPLAY_OBJECTS, (evo->ramin->instance >> 8) | 9);
+	nv_wr32(dev, NV50_PDISPLAY_OBJECTS, (evo->ramin->vinst >> 8) | 9);
 
 	/* initialise fifo */
 	nv_wr32(dev, NV50_PDISPLAY_CHANNEL_DMA_CB(0),
diff --git a/drivers/gpu/drm/nouveau/nv50_fbcon.c b/drivers/gpu/drm/nouveau/nv50_fbcon.c
index 6bf025c6fc6f..6dcf048eddbc 100644
--- a/drivers/gpu/drm/nouveau/nv50_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nv50_fbcon.c
@@ -1,6 +1,7 @@
 #include "drmP.h"
 #include "nouveau_drv.h"
 #include "nouveau_dma.h"
+#include "nouveau_ramht.h"
 #include "nouveau_fbcon.h"
 
 void
@@ -193,7 +194,8 @@ nv50_fbcon_accel_init(struct fb_info *info)
 	if (ret)
 		return ret;
 
-	ret = nouveau_gpuobj_ref_add(dev, dev_priv->channel, Nv2D, eng2d, NULL);
+	ret = nouveau_ramht_insert(dev_priv->channel, Nv2D, eng2d);
+	nouveau_gpuobj_ref(NULL, &eng2d);
 	if (ret)
 		return ret;
 
diff --git a/drivers/gpu/drm/nouveau/nv50_fifo.c b/drivers/gpu/drm/nouveau/nv50_fifo.c
index 9201f35d8277..4fc8b59cc41e 100644
--- a/drivers/gpu/drm/nouveau/nv50_fifo.c
+++ b/drivers/gpu/drm/nouveau/nv50_fifo.c
@@ -27,13 +27,14 @@
 #include "drmP.h"
 #include "drm.h"
 #include "nouveau_drv.h"
+#include "nouveau_ramht.h"
 
 static void
 nv50_fifo_playlist_update(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
-	struct nouveau_gpuobj_ref *cur;
+	struct nouveau_gpuobj *cur;
 	int i, nr;
 
 	NV_DEBUG(dev, "\n");
@@ -44,13 +45,13 @@ nv50_fifo_playlist_update(struct drm_device *dev)
 	/* We never schedule channel 0 or 127 */
 	for (i = 1, nr = 0; i < 127; i++) {
 		if (dev_priv->fifos[i] && dev_priv->fifos[i]->ramfc) {
-			nv_wo32(cur->gpuobj, (nr * 4), i);
+			nv_wo32(cur, (nr * 4), i);
 			nr++;
 		}
 	}
 	dev_priv->engine.instmem.flush(dev);
 
-	nv_wr32(dev, 0x32f4, cur->instance >> 12);
+	nv_wr32(dev, 0x32f4, cur->vinst >> 12);
 	nv_wr32(dev, 0x32ec, nr);
 	nv_wr32(dev, 0x2500, 0x101);
 }
@@ -65,9 +66,9 @@ nv50_fifo_channel_enable(struct drm_device *dev, int channel)
 	NV_DEBUG(dev, "ch%d\n", channel);
 
 	if (dev_priv->chipset == 0x50)
-		inst = chan->ramfc->instance >> 12;
+		inst = chan->ramfc->vinst >> 12;
 	else
-		inst = chan->ramfc->instance >> 8;
+		inst = chan->ramfc->vinst >> 8;
 
 	nv_wr32(dev, NV50_PFIFO_CTX_TABLE(channel), inst |
 		     NV50_PFIFO_CTX_TABLE_CHANNEL_ENABLED);
@@ -165,19 +166,19 @@ nv50_fifo_init(struct drm_device *dev)
 		goto just_reset;
 	}
 
-	ret = nouveau_gpuobj_new_ref(dev, NULL, NULL, 0, 128*4, 0x1000,
-				     NVOBJ_FLAG_ZERO_ALLOC,
-				     &pfifo->playlist[0]);
+	ret = nouveau_gpuobj_new(dev, NULL, 128*4, 0x1000,
+				 NVOBJ_FLAG_ZERO_ALLOC,
+				 &pfifo->playlist[0]);
 	if (ret) {
 		NV_ERROR(dev, "error creating playlist 0: %d\n", ret);
 		return ret;
 	}
 
-	ret = nouveau_gpuobj_new_ref(dev, NULL, NULL, 0, 128*4, 0x1000,
-				     NVOBJ_FLAG_ZERO_ALLOC,
-				     &pfifo->playlist[1]);
+	ret = nouveau_gpuobj_new(dev, NULL, 128*4, 0x1000,
+				 NVOBJ_FLAG_ZERO_ALLOC,
+				 &pfifo->playlist[1]);
 	if (ret) {
-		nouveau_gpuobj_ref_del(dev, &pfifo->playlist[0]);
+		nouveau_gpuobj_ref(NULL, &pfifo->playlist[0]);
 		NV_ERROR(dev, "error creating playlist 1: %d\n", ret);
 		return ret;
 	}
@@ -205,8 +206,8 @@ nv50_fifo_takedown(struct drm_device *dev)
 	if (!pfifo->playlist[0])
 		return;
 
-	nouveau_gpuobj_ref_del(dev, &pfifo->playlist[0]);
-	nouveau_gpuobj_ref_del(dev, &pfifo->playlist[1]);
+	nouveau_gpuobj_ref(NULL, &pfifo->playlist[0]);
+	nouveau_gpuobj_ref(NULL, &pfifo->playlist[1]);
 }
 
 int
@@ -228,42 +229,39 @@ nv50_fifo_create_context(struct nouveau_channel *chan)
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
 	if (dev_priv->chipset == 0x50) {
-		ret = nouveau_gpuobj_new_fake(dev, chan->ramin->gpuobj->pinst,
-					      chan->ramin->gpuobj->vinst, 0x100,
+		ret = nouveau_gpuobj_new_fake(dev, chan->ramin->pinst,
+					      chan->ramin->vinst, 0x100,
 					      NVOBJ_FLAG_ZERO_ALLOC |
-					      NVOBJ_FLAG_ZERO_FREE, &ramfc,
+					      NVOBJ_FLAG_ZERO_FREE,
 					      &chan->ramfc);
 		if (ret)
 			return ret;
 
-		ret = nouveau_gpuobj_new_fake(dev, chan->ramin->gpuobj->pinst +
-					      0x0400,
-					      chan->ramin->gpuobj->vinst +
-					      0x0400, 4096, 0, NULL,
-					      &chan->cache);
+		ret = nouveau_gpuobj_new_fake(dev, chan->ramin->pinst + 0x0400,
+					      chan->ramin->vinst + 0x0400,
+					      4096, 0, &chan->cache);
 		if (ret)
 			return ret;
 	} else {
-		ret = nouveau_gpuobj_new_ref(dev, chan, NULL, 0, 0x100, 256,
-					     NVOBJ_FLAG_ZERO_ALLOC |
-					     NVOBJ_FLAG_ZERO_FREE,
-					     &chan->ramfc);
+		ret = nouveau_gpuobj_new(dev, chan, 0x100, 256,
+					 NVOBJ_FLAG_ZERO_ALLOC |
+					 NVOBJ_FLAG_ZERO_FREE, &chan->ramfc);
 		if (ret)
 			return ret;
-		ramfc = chan->ramfc->gpuobj;
 
-		ret = nouveau_gpuobj_new_ref(dev, chan, NULL, 0, 4096, 1024,
-					     0, &chan->cache);
+		ret = nouveau_gpuobj_new(dev, chan, 4096, 1024,
+					 0, &chan->cache);
 		if (ret)
 			return ret;
 	}
+	ramfc = chan->ramfc;
 
 	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
 
-	nv_wo32(ramfc, 0x48, chan->pushbuf->instance >> 4);
+	nv_wo32(ramfc, 0x48, chan->pushbuf->cinst >> 4);
 	nv_wo32(ramfc, 0x80, (0 << 27) /* 4KiB */ |
 			     (4 << 24) /* SEARCH_FULL */ |
-			     (chan->ramht->instance >> 4));
+			     (chan->ramht->gpuobj->cinst >> 4));
 	nv_wo32(ramfc, 0x44, 0x2101ffff);
 	nv_wo32(ramfc, 0x60, 0x7fffffff);
 	nv_wo32(ramfc, 0x40, 0x00000000);
@@ -274,11 +272,11 @@ nv50_fifo_create_context(struct nouveau_channel *chan)
 	nv_wo32(ramfc, 0x54, drm_order(chan->dma.ib_max + 1) << 16);
 
 	if (dev_priv->chipset != 0x50) {
-		nv_wo32(chan->ramin->gpuobj, 0, chan->id);
-		nv_wo32(chan->ramin->gpuobj, 4, chan->ramfc->instance >> 8);
+		nv_wo32(chan->ramin, 0, chan->id);
+		nv_wo32(chan->ramin, 4, chan->ramfc->vinst >> 8);
 
-		nv_wo32(ramfc, 0x88, chan->cache->instance >> 10);
-		nv_wo32(ramfc, 0x98, chan->ramin->instance >> 12);
+		nv_wo32(ramfc, 0x88, chan->cache->vinst >> 10);
+		nv_wo32(ramfc, 0x98, chan->ramin->vinst >> 12);
 	}
 
 	dev_priv->engine.instmem.flush(dev);
@@ -293,12 +291,13 @@ void
 nv50_fifo_destroy_context(struct nouveau_channel *chan)
 {
 	struct drm_device *dev = chan->dev;
-	struct nouveau_gpuobj_ref *ramfc = chan->ramfc;
+	struct nouveau_gpuobj *ramfc = NULL;
 
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
 	/* This will ensure the channel is seen as disabled. */
-	chan->ramfc = NULL;
+	nouveau_gpuobj_ref(chan->ramfc, &ramfc);
+	nouveau_gpuobj_ref(NULL, &chan->ramfc);
 	nv50_fifo_channel_disable(dev, chan->id);
 
 	/* Dummy channel, also used on ch 127 */
@@ -306,8 +305,8 @@ nv50_fifo_destroy_context(struct nouveau_channel *chan)
 		nv50_fifo_channel_disable(dev, 127);
 	nv50_fifo_playlist_update(dev);
 
-	nouveau_gpuobj_ref_del(dev, &ramfc);
-	nouveau_gpuobj_ref_del(dev, &chan->cache);
+	nouveau_gpuobj_ref(NULL, &ramfc);
+	nouveau_gpuobj_ref(NULL, &chan->cache);
 }
 
 int
@@ -315,8 +314,8 @@ nv50_fifo_load_context(struct nouveau_channel *chan)
 {
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj *ramfc = chan->ramfc->gpuobj;
-	struct nouveau_gpuobj *cache = chan->cache->gpuobj;
+	struct nouveau_gpuobj *ramfc = chan->ramfc;
+	struct nouveau_gpuobj *cache = chan->cache;
 	int ptr, cnt;
 
 	NV_DEBUG(dev, "ch%d\n", chan->id);
@@ -399,8 +398,8 @@ nv50_fifo_unload_context(struct drm_device *dev)
 		return -EINVAL;
 	}
 	NV_DEBUG(dev, "ch%d\n", chan->id);
-	ramfc = chan->ramfc->gpuobj;
-	cache = chan->cache->gpuobj;
+	ramfc = chan->ramfc;
+	cache = chan->cache;
 
 	nv_wo32(ramfc, 0x00, nv_rd32(dev, 0x3330));
 	nv_wo32(ramfc, 0x04, nv_rd32(dev, 0x3334));
diff --git a/drivers/gpu/drm/nouveau/nv50_graph.c b/drivers/gpu/drm/nouveau/nv50_graph.c
index 17a8d788a494..7db0d627c1b9 100644
--- a/drivers/gpu/drm/nouveau/nv50_graph.c
+++ b/drivers/gpu/drm/nouveau/nv50_graph.c
@@ -27,7 +27,7 @@
 #include "drmP.h"
 #include "drm.h"
 #include "nouveau_drv.h"
-
+#include "nouveau_ramht.h"
 #include "nouveau_grctx.h"
 
 static void
@@ -192,7 +192,7 @@ nv50_graph_channel(struct drm_device *dev)
 	for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
 		struct nouveau_channel *chan = dev_priv->fifos[i];
 
-		if (chan && chan->ramin && chan->ramin->instance == inst)
+		if (chan && chan->ramin && chan->ramin->vinst == inst)
 			return chan;
 	}
 
@@ -204,36 +204,34 @@ nv50_graph_create_context(struct nouveau_channel *chan)
 {
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj *ramin = chan->ramin->gpuobj;
-	struct nouveau_gpuobj *obj;
+	struct nouveau_gpuobj *ramin = chan->ramin;
 	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	struct nouveau_grctx ctx = {};
 	int hdr, ret;
 
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
-	ret = nouveau_gpuobj_new_ref(dev, chan, NULL, 0, pgraph->grctx_size,
-				     0x1000, NVOBJ_FLAG_ZERO_ALLOC |
-				     NVOBJ_FLAG_ZERO_FREE, &chan->ramin_grctx);
+	ret = nouveau_gpuobj_new(dev, chan, pgraph->grctx_size, 0x1000,
+				 NVOBJ_FLAG_ZERO_ALLOC |
+				 NVOBJ_FLAG_ZERO_FREE, &chan->ramin_grctx);
 	if (ret)
 		return ret;
-	obj = chan->ramin_grctx->gpuobj;
 
 	hdr = (dev_priv->chipset == 0x50) ? 0x200 : 0x20;
 	nv_wo32(ramin, hdr + 0x00, 0x00190002);
-	nv_wo32(ramin, hdr + 0x04, chan->ramin_grctx->instance +
+	nv_wo32(ramin, hdr + 0x04, chan->ramin_grctx->vinst +
 				   pgraph->grctx_size - 1);
-	nv_wo32(ramin, hdr + 0x08, chan->ramin_grctx->instance);
+	nv_wo32(ramin, hdr + 0x08, chan->ramin_grctx->vinst);
 	nv_wo32(ramin, hdr + 0x0c, 0);
 	nv_wo32(ramin, hdr + 0x10, 0);
 	nv_wo32(ramin, hdr + 0x14, 0x00010000);
 
 	ctx.dev = chan->dev;
 	ctx.mode = NOUVEAU_GRCTX_VALS;
-	ctx.data = obj;
+	ctx.data = chan->ramin_grctx;
 	nv50_grctx_init(&ctx);
 
-	nv_wo32(obj, 0x00000, chan->ramin->instance >> 12);
+	nv_wo32(chan->ramin_grctx, 0x00000, chan->ramin->vinst >> 12);
 
 	dev_priv->engine.instmem.flush(dev);
 	return 0;
@@ -248,14 +246,14 @@ nv50_graph_destroy_context(struct nouveau_channel *chan)
 
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
-	if (!chan->ramin || !chan->ramin->gpuobj)
+	if (!chan->ramin)
 		return;
 
 	for (i = hdr; i < hdr + 24; i += 4)
-		nv_wo32(chan->ramin->gpuobj, i, 0);
+		nv_wo32(chan->ramin, i, 0);
 	dev_priv->engine.instmem.flush(dev);
 
-	nouveau_gpuobj_ref_del(dev, &chan->ramin_grctx);
+	nouveau_gpuobj_ref(NULL, &chan->ramin_grctx);
 }
 
 static int
@@ -282,7 +280,7 @@ nv50_graph_do_load_context(struct drm_device *dev, uint32_t inst)
 int
 nv50_graph_load_context(struct nouveau_channel *chan)
 {
-	uint32_t inst = chan->ramin->instance >> 12;
+	uint32_t inst = chan->ramin->vinst >> 12;
 
 	NV_DEBUG(chan->dev, "ch%d\n", chan->id);
 	return nv50_graph_do_load_context(chan->dev, inst);
@@ -327,15 +325,16 @@ static int
 nv50_graph_nvsw_dma_vblsem(struct nouveau_channel *chan, int grclass,
 			   int mthd, uint32_t data)
 {
-	struct nouveau_gpuobj_ref *ref = NULL;
+	struct nouveau_gpuobj *gpuobj;
 
-	if (nouveau_gpuobj_ref_find(chan, data, &ref))
+	gpuobj = nouveau_ramht_find(chan, data);
+	if (!gpuobj)
 		return -ENOENT;
 
-	if (nouveau_notifier_offset(ref->gpuobj, NULL))
+	if (nouveau_notifier_offset(gpuobj, NULL))
 		return -EINVAL;
 
-	chan->nvsw.vblsem = ref->gpuobj;
+	chan->nvsw.vblsem = gpuobj;
 	chan->nvsw.vblsem_offset = ~0;
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nv50_instmem.c b/drivers/gpu/drm/nouveau/nv50_instmem.c
index 821806c835ba..0af0baf4f1a9 100644
--- a/drivers/gpu/drm/nouveau/nv50_instmem.c
+++ b/drivers/gpu/drm/nouveau/nv50_instmem.c
@@ -32,9 +32,9 @@
 struct nv50_instmem_priv {
 	uint32_t save1700[5]; /* 0x1700->0x1710 */
 
-	struct nouveau_gpuobj_ref *pramin_pt;
-	struct nouveau_gpuobj_ref *pramin_bar;
-	struct nouveau_gpuobj_ref *fb_bar;
+	struct nouveau_gpuobj *pramin_pt;
+	struct nouveau_gpuobj *pramin_bar;
+	struct nouveau_gpuobj *fb_bar;
 };
 
 #define NV50_INSTMEM_PAGE_SHIFT 12
@@ -44,15 +44,8 @@ struct nv50_instmem_priv {
 /*NOTE: - Assumes 0x1700 already covers the correct MiB of PRAMIN
  */
 #define BAR0_WI32(g, o, v) do {                                   \
-	uint32_t offset;                                          \
-	if ((g)->im_backing) {                                    \
-		offset = (g)->im_backing_start;                   \
-	} else {                                                  \
-		offset  = chan->ramin->gpuobj->im_backing_start;  \
-		offset += (g)->im_pramin->start;                  \
-	}                                                         \
-	offset += (o);                                            \
-	nv_wr32(dev, NV_RAMIN + (offset & 0xfffff), (v));              \
+	u32 offset = (g)->vinst + (o);                            \
+	nv_wr32(dev, NV_RAMIN + (offset & 0xfffff), (v));         \
 } while (0)
 
 int
@@ -142,8 +135,7 @@ nv50_instmem_init(struct drm_device *dev)
 	INIT_LIST_HEAD(&chan->ramht_refs);
 
 	/* Channel's PRAMIN object + heap */
-	ret = nouveau_gpuobj_new_fake(dev, 0, c_offset, c_size, 0,
-							NULL, &chan->ramin);
+	ret = nouveau_gpuobj_new_fake(dev, 0, c_offset, c_size, 0, &chan->ramin);
 	if (ret)
 		return ret;
 
@@ -152,16 +144,16 @@ nv50_instmem_init(struct drm_device *dev)
 
 	/* RAMFC + zero channel's PRAMIN up to start of VM pagedir */
 	ret = nouveau_gpuobj_new_fake(dev, c_ramfc, c_offset + c_ramfc,
-						0x4000, 0, NULL, &chan->ramfc);
+				      0x4000, 0, &chan->ramfc);
 	if (ret)
 		return ret;
 
 	for (i = 0; i < c_vmpd; i += 4)
-		BAR0_WI32(chan->ramin->gpuobj, i, 0);
+		BAR0_WI32(chan->ramin, i, 0);
 
 	/* VM page directory */
 	ret = nouveau_gpuobj_new_fake(dev, c_vmpd, c_offset + c_vmpd,
-					   0x4000, 0, &chan->vm_pd, NULL);
+				      0x4000, 0, &chan->vm_pd);
 	if (ret)
 		return ret;
 	for (i = 0; i < 0x4000; i += 8) {
@@ -172,8 +164,8 @@ nv50_instmem_init(struct drm_device *dev)
 	/* PRAMIN page table, cheat and map into VM at 0x0000000000.
 	 * We map the entire fake channel into the start of the PRAMIN BAR
 	 */
-	ret = nouveau_gpuobj_new_ref(dev, chan, NULL, 0, pt_size, 0x1000,
-				     0, &priv->pramin_pt);
+	ret = nouveau_gpuobj_new(dev, chan, pt_size, 0x1000, 0,
+				 &priv->pramin_pt);
 	if (ret)
 		return ret;
 
@@ -185,76 +177,74 @@ nv50_instmem_init(struct drm_device *dev)
 
 	i = 0;
 	while (v < dev_priv->vram_sys_base + c_offset + c_size) {
-		BAR0_WI32(priv->pramin_pt->gpuobj, i + 0, lower_32_bits(v));
-		BAR0_WI32(priv->pramin_pt->gpuobj, i + 4, upper_32_bits(v));
+		BAR0_WI32(priv->pramin_pt, i + 0, lower_32_bits(v));
+		BAR0_WI32(priv->pramin_pt, i + 4, upper_32_bits(v));
 		v += 0x1000;
 		i += 8;
 	}
 
 	while (i < pt_size) {
-		BAR0_WI32(priv->pramin_pt->gpuobj, i + 0, 0x00000000);
-		BAR0_WI32(priv->pramin_pt->gpuobj, i + 4, 0x00000000);
+		BAR0_WI32(priv->pramin_pt, i + 0, 0x00000000);
+		BAR0_WI32(priv->pramin_pt, i + 4, 0x00000000);
 		i += 8;
 	}
 
-	BAR0_WI32(chan->vm_pd, 0x00, priv->pramin_pt->instance | 0x63);
+	BAR0_WI32(chan->vm_pd, 0x00, priv->pramin_pt->vinst | 0x63);
 	BAR0_WI32(chan->vm_pd, 0x04, 0x00000000);
 
 	/* VRAM page table(s), mapped into VM at +1GiB  */
 	for (i = 0; i < dev_priv->vm_vram_pt_nr; i++) {
-		ret = nouveau_gpuobj_new_ref(dev, chan, NULL, 0,
-					     NV50_VM_BLOCK/65536*8, 0, 0,
-					     &chan->vm_vram_pt[i]);
+		ret = nouveau_gpuobj_new(dev, chan, NV50_VM_BLOCK / 0x10000 * 8,
+					 0, 0, &chan->vm_vram_pt[i]);
 		if (ret) {
 			NV_ERROR(dev, "Error creating VRAM page tables: %d\n",
 									ret);
 			dev_priv->vm_vram_pt_nr = i;
 			return ret;
 		}
-		dev_priv->vm_vram_pt[i] = chan->vm_vram_pt[i]->gpuobj;
+		/*XXX: double-check this is ok */
+		dev_priv->vm_vram_pt[i] = chan->vm_vram_pt[i];
 
 		for (v = 0; v < dev_priv->vm_vram_pt[i]->im_pramin->size;
 								v += 4)
 			BAR0_WI32(dev_priv->vm_vram_pt[i], v, 0);
 
 		BAR0_WI32(chan->vm_pd, 0x10 + (i*8),
-			  chan->vm_vram_pt[i]->instance | 0x61);
+			  chan->vm_vram_pt[i]->vinst | 0x61);
 		BAR0_WI32(chan->vm_pd, 0x14 + (i*8), 0);
 	}
 
 	/* DMA object for PRAMIN BAR */
-	ret = nouveau_gpuobj_new_ref(dev, chan, chan, 0, 6*4, 16, 0,
-							&priv->pramin_bar);
+	ret = nouveau_gpuobj_new(dev, chan, 6*4, 16, 0, &priv->pramin_bar);
 	if (ret)
 		return ret;
-	BAR0_WI32(priv->pramin_bar->gpuobj, 0x00, 0x7fc00000);
-	BAR0_WI32(priv->pramin_bar->gpuobj, 0x04, dev_priv->ramin_size - 1);
-	BAR0_WI32(priv->pramin_bar->gpuobj, 0x08, 0x00000000);
-	BAR0_WI32(priv->pramin_bar->gpuobj, 0x0c, 0x00000000);
-	BAR0_WI32(priv->pramin_bar->gpuobj, 0x10, 0x00000000);
-	BAR0_WI32(priv->pramin_bar->gpuobj, 0x14, 0x00000000);
+	BAR0_WI32(priv->pramin_bar, 0x00, 0x7fc00000);
+	BAR0_WI32(priv->pramin_bar, 0x04, dev_priv->ramin_size - 1);
+	BAR0_WI32(priv->pramin_bar, 0x08, 0x00000000);
+	BAR0_WI32(priv->pramin_bar, 0x0c, 0x00000000);
+	BAR0_WI32(priv->pramin_bar, 0x10, 0x00000000);
+	BAR0_WI32(priv->pramin_bar, 0x14, 0x00000000);
 
 	/* DMA object for FB BAR */
-	ret = nouveau_gpuobj_new_ref(dev, chan, chan, 0, 6*4, 16, 0,
-							&priv->fb_bar);
+	ret = nouveau_gpuobj_new(dev, chan, 6*4, 16, 0, &priv->fb_bar);
 	if (ret)
 		return ret;
-	BAR0_WI32(priv->fb_bar->gpuobj, 0x00, 0x7fc00000);
-	BAR0_WI32(priv->fb_bar->gpuobj, 0x04, 0x40000000 +
-					      pci_resource_len(dev->pdev, 1) - 1);
-	BAR0_WI32(priv->fb_bar->gpuobj, 0x08, 0x40000000);
-	BAR0_WI32(priv->fb_bar->gpuobj, 0x0c, 0x00000000);
-	BAR0_WI32(priv->fb_bar->gpuobj, 0x10, 0x00000000);
-	BAR0_WI32(priv->fb_bar->gpuobj, 0x14, 0x00000000);
+	BAR0_WI32(priv->fb_bar, 0x00, 0x7fc00000);
+	BAR0_WI32(priv->fb_bar, 0x04, 0x40000000 +
+				      pci_resource_len(dev->pdev, 1) - 1);
+	BAR0_WI32(priv->fb_bar, 0x08, 0x40000000);
+	BAR0_WI32(priv->fb_bar, 0x0c, 0x00000000);
+	BAR0_WI32(priv->fb_bar, 0x10, 0x00000000);
+	BAR0_WI32(priv->fb_bar, 0x14, 0x00000000);
 
 	/* Poke the relevant regs, and pray it works :) */
-	nv_wr32(dev, NV50_PUNK_BAR_CFG_BASE, (chan->ramin->instance >> 12));
+	nv_wr32(dev, NV50_PUNK_BAR_CFG_BASE, (chan->ramin->vinst >> 12));
 	nv_wr32(dev, NV50_PUNK_UNK1710, 0);
-	nv_wr32(dev, NV50_PUNK_BAR_CFG_BASE, (chan->ramin->instance >> 12) |
+	nv_wr32(dev, NV50_PUNK_BAR_CFG_BASE, (chan->ramin->vinst >> 12) |
 					 NV50_PUNK_BAR_CFG_BASE_VALID);
-	nv_wr32(dev, NV50_PUNK_BAR1_CTXDMA, (priv->fb_bar->instance >> 4) |
+	nv_wr32(dev, NV50_PUNK_BAR1_CTXDMA, (priv->fb_bar->cinst >> 4) |
 					NV50_PUNK_BAR1_CTXDMA_VALID);
-	nv_wr32(dev, NV50_PUNK_BAR3_CTXDMA, (priv->pramin_bar->instance >> 4) |
+	nv_wr32(dev, NV50_PUNK_BAR3_CTXDMA, (priv->pramin_bar->cinst >> 4) |
 					NV50_PUNK_BAR3_CTXDMA_VALID);
 
 	for (i = 0; i < 8; i++)
@@ -301,21 +291,19 @@ nv50_instmem_takedown(struct drm_device *dev)
 	for (i = 0x1700; i <= 0x1710; i += 4)
 		nv_wr32(dev, i, priv->save1700[(i - 0x1700) / 4]);
 
-	nouveau_gpuobj_ref_del(dev, &priv->fb_bar);
-	nouveau_gpuobj_ref_del(dev, &priv->pramin_bar);
-	nouveau_gpuobj_ref_del(dev, &priv->pramin_pt);
+	nouveau_gpuobj_ref(NULL, &priv->fb_bar);
+	nouveau_gpuobj_ref(NULL, &priv->pramin_bar);
+	nouveau_gpuobj_ref(NULL, &priv->pramin_pt);
 
 	/* Destroy dummy channel */
 	if (chan) {
-		for (i = 0; i < dev_priv->vm_vram_pt_nr; i++) {
-			nouveau_gpuobj_ref_del(dev, &chan->vm_vram_pt[i]);
-			dev_priv->vm_vram_pt[i] = NULL;
-		}
+		for (i = 0; i < dev_priv->vm_vram_pt_nr; i++)
+			nouveau_gpuobj_ref(NULL, &chan->vm_vram_pt[i]);
 		dev_priv->vm_vram_pt_nr = 0;
 
-		nouveau_gpuobj_del(dev, &chan->vm_pd);
-		nouveau_gpuobj_ref_del(dev, &chan->ramfc);
-		nouveau_gpuobj_ref_del(dev, &chan->ramin);
+		nouveau_gpuobj_ref(NULL, &chan->vm_pd);
+		nouveau_gpuobj_ref(NULL, &chan->ramfc);
+		nouveau_gpuobj_ref(NULL, &chan->ramin);
 		drm_mm_takedown(&chan->ramin_heap);
 
 		dev_priv->fifos[0] = dev_priv->fifos[127] = NULL;
@@ -331,7 +319,7 @@ nv50_instmem_suspend(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_channel *chan = dev_priv->fifos[0];
-	struct nouveau_gpuobj *ramin = chan->ramin->gpuobj;
+	struct nouveau_gpuobj *ramin = chan->ramin;
 	int i;
 
 	ramin->im_backing_suspend = vmalloc(ramin->im_pramin->size);
@@ -349,7 +337,7 @@ nv50_instmem_resume(struct drm_device *dev)
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nv50_instmem_priv *priv = dev_priv->engine.instmem.priv;
 	struct nouveau_channel *chan = dev_priv->fifos[0];
-	struct nouveau_gpuobj *ramin = chan->ramin->gpuobj;
+	struct nouveau_gpuobj *ramin = chan->ramin;
 	int i;
 
 	nv_wr32(dev, NV50_PUNK_BAR0_PRAMIN, (ramin->im_backing_start >> 16));
@@ -359,13 +347,13 @@ nv50_instmem_resume(struct drm_device *dev)
 	ramin->im_backing_suspend = NULL;
 
 	/* Poke the relevant regs, and pray it works :) */
-	nv_wr32(dev, NV50_PUNK_BAR_CFG_BASE, (chan->ramin->instance >> 12));
+	nv_wr32(dev, NV50_PUNK_BAR_CFG_BASE, (chan->ramin->vinst >> 12));
 	nv_wr32(dev, NV50_PUNK_UNK1710, 0);
-	nv_wr32(dev, NV50_PUNK_BAR_CFG_BASE, (chan->ramin->instance >> 12) |
+	nv_wr32(dev, NV50_PUNK_BAR_CFG_BASE, (chan->ramin->vinst >> 12) |
 					 NV50_PUNK_BAR_CFG_BASE_VALID);
-	nv_wr32(dev, NV50_PUNK_BAR1_CTXDMA, (priv->fb_bar->instance >> 4) |
+	nv_wr32(dev, NV50_PUNK_BAR1_CTXDMA, (priv->fb_bar->cinst >> 4) |
 					NV50_PUNK_BAR1_CTXDMA_VALID);
-	nv_wr32(dev, NV50_PUNK_BAR3_CTXDMA, (priv->pramin_bar->instance >> 4) |
+	nv_wr32(dev, NV50_PUNK_BAR3_CTXDMA, (priv->pramin_bar->cinst >> 4) |
 					NV50_PUNK_BAR3_CTXDMA_VALID);
 
 	for (i = 0; i < 8; i++)
@@ -424,7 +412,7 @@ nv50_instmem_bind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nv50_instmem_priv *priv = dev_priv->engine.instmem.priv;
-	struct nouveau_gpuobj *pramin_pt = priv->pramin_pt->gpuobj;
+	struct nouveau_gpuobj *pramin_pt = priv->pramin_pt;
 	uint32_t pte, pte_end;
 	uint64_t vram;
 
@@ -477,8 +465,8 @@ nv50_instmem_unbind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
 	pte_end = ((gpuobj->im_pramin->size >> 12) << 1) + pte;
 
 	while (pte < pte_end) {
-		nv_wo32(priv->pramin_pt->gpuobj, (pte * 4) + 0, 0x00000000);
-		nv_wo32(priv->pramin_pt->gpuobj, (pte * 4) + 4, 0x00000000);
+		nv_wo32(priv->pramin_pt, (pte * 4) + 0, 0x00000000);
+		nv_wo32(priv->pramin_pt, (pte * 4) + 4, 0x00000000);
 		pte += 2;
 	}
 	dev_priv->engine.instmem.flush(dev);