drm/nouveau/gpuobj: separate allocation from nvkm_object

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

20 files changed, 373 insertions, 379 deletions

diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/gpuobj.h b/drivers/gpu/drm/nouveau/include/nvkm/core/gpuobj.h
index 7613107c05fa6..d171535b8eef1 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/gpuobj.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/gpuobj.h
@@ -7,30 +7,33 @@ struct nvkm_vma;
 struct nvkm_vm;
 
 #define NVOBJ_FLAG_ZERO_ALLOC 0x00000001
-#define NVOBJ_FLAG_ZERO_FREE  0x00000002
 #define NVOBJ_FLAG_HEAP       0x00000004
 
 struct nvkm_gpuobj {
 	struct nvkm_object object;
-	struct nvkm_memory *memory;
+	const struct nvkm_gpuobj_func *func;
 	struct nvkm_gpuobj *parent;
+	struct nvkm_memory *memory;
 	struct nvkm_mm_node *node;
-	struct nvkm_mm heap;
 
-	u32 flags;
 	u64 addr;
 	u32 size;
+	struct nvkm_mm heap;
 
-	const struct nvkm_gpuobj_func *func;
+	void __iomem *map;
 };
 
 struct nvkm_gpuobj_func {
-	void (*acquire)(struct nvkm_gpuobj *);
+	void *(*acquire)(struct nvkm_gpuobj *);
 	void (*release)(struct nvkm_gpuobj *);
 	u32 (*rd32)(struct nvkm_gpuobj *, u32 offset);
 	void (*wr32)(struct nvkm_gpuobj *, u32 offset, u32 data);
 };
 
+int  nvkm_gpuobj_new(struct nvkm_device *, u32 size, int align, bool zero,
+		     struct nvkm_gpuobj *parent, struct nvkm_gpuobj **);
+void nvkm_gpuobj_del(struct nvkm_gpuobj **);
+
 static inline struct nvkm_gpuobj *
 nv_gpuobj(void *obj)
 {
@@ -51,12 +54,9 @@ int  nvkm_gpuobj_create_(struct nvkm_object *, struct nvkm_object *,
 			    u32 flags, int length, void **);
 void nvkm_gpuobj_destroy(struct nvkm_gpuobj *);
 
-int  nvkm_gpuobj_new(struct nvkm_object *, struct nvkm_object *, u32 size,
-		     u32 align, u32 flags, struct nvkm_gpuobj **);
-int  nvkm_gpuobj_dup(struct nvkm_object *, struct nvkm_memory *,
-		     struct nvkm_gpuobj **);
-int  nvkm_gpuobj_map_vm(struct nvkm_gpuobj *, struct nvkm_vm *, u32 access,
-			struct nvkm_vma *);
+int nvkm_gpuobj_wrap(struct nvkm_memory *, struct nvkm_gpuobj **);
+int nvkm_gpuobj_map(struct nvkm_gpuobj *, struct nvkm_vm *, u32 access,
+		    struct nvkm_vma *);
 void nvkm_gpuobj_unmap(struct nvkm_vma *);
 
 static inline void
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/engine/dmaobj.h b/drivers/gpu/drm/nouveau/include/nvkm/engine/dmaobj.h
index 2c3cc61adb74b..343d2d73cd943 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/engine/dmaobj.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/engine/dmaobj.h
@@ -15,7 +15,7 @@ struct nvkm_dmaeng {
 	struct nvkm_engine engine;
 
 	/* creates a "physical" dma object from a struct nvkm_dmaobj */
-	int (*bind)(struct nvkm_dmaobj *dmaobj, struct nvkm_object *parent,
+	int (*bind)(struct nvkm_dmaobj *dmaobj, struct nvkm_gpuobj *parent,
 		    struct nvkm_gpuobj **);
 };
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.c b/drivers/gpu/drm/nouveau/nouveau_drm.c
index 650e911dd7042..ebcbe7bbdddf6 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drm.c
@@ -141,7 +141,7 @@ nouveau_accel_fini(struct nouveau_drm *drm)
 {
 	nouveau_channel_del(&drm->channel);
 	nvif_object_fini(&drm->ntfy);
-	nvkm_gpuobj_ref(NULL, &drm->notify);
+	nvkm_gpuobj_del(&drm->notify);
 	nvif_object_fini(&drm->nvsw);
 	nouveau_channel_del(&drm->cechan);
 	nvif_object_fini(&drm->ttm.copy);
@@ -264,8 +264,8 @@ nouveau_accel_init(struct nouveau_drm *drm)
 	}
 
 	if (device->info.family < NV_DEVICE_INFO_V0_FERMI) {
-		ret = nvkm_gpuobj_new(nvxx_object(&drm->device.object), NULL, 32,
-				      0, 0, &drm->notify);
+		ret = nvkm_gpuobj_new(nvxx_device(&drm->device), 32, 0, false,
+				      NULL, &drm->notify);
 		if (ret) {
 			NV_ERROR(drm, "failed to allocate notifier, %d\n", ret);
 			nouveau_accel_fini(drm);
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/gpuobj.c b/drivers/gpu/drm/nouveau/nvkm/core/gpuobj.c
index c14469c3a861e..54b46037f4ba3 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/gpuobj.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/gpuobj.c
@@ -28,74 +28,209 @@
 #include <subdev/bar.h>
 #include <subdev/mmu.h>
 
-static void
-nvkm_gpuobj_release(struct nvkm_gpuobj *gpuobj)
+/* fast-path, where backend is able to provide direct pointer to memory */
+static u32
+nvkm_gpuobj_rd32_fast(struct nvkm_gpuobj *gpuobj, u32 offset)
 {
-	if (gpuobj->node) {
-		nvkm_done(gpuobj->parent);
-		return;
-	}
-	nvkm_done(gpuobj->memory);
+	return ioread32_native(gpuobj->map + offset);
 }
 
 static void
-nvkm_gpuobj_acquire(struct nvkm_gpuobj *gpuobj)
+nvkm_gpuobj_wr32_fast(struct nvkm_gpuobj *gpuobj, u32 offset, u32 data)
 {
-	if (gpuobj->node) {
-		nvkm_kmap(gpuobj->parent);
-		return;
-	}
-	nvkm_kmap(gpuobj->memory);
+	iowrite32_native(data, gpuobj->map + offset);
 }
 
+/* accessor functions for gpuobjs allocated directly from instmem */
 static u32
-nvkm_gpuobj_rd32(struct nvkm_gpuobj *gpuobj, u32 offset)
+nvkm_gpuobj_heap_rd32(struct nvkm_gpuobj *gpuobj, u32 offset)
 {
-	if (gpuobj->node)
-		return nvkm_ro32(gpuobj->parent, gpuobj->node->offset + offset);
 	return nvkm_ro32(gpuobj->memory, offset);
 }
 
 static void
-nvkm_gpuobj_wr32(struct nvkm_gpuobj *gpuobj, u32 offset, u32 data)
+nvkm_gpuobj_heap_wr32(struct nvkm_gpuobj *gpuobj, u32 offset, u32 data)
 {
-	if (gpuobj->node) {
-		nvkm_wo32(gpuobj->parent, gpuobj->node->offset + offset, data);
-		return;
-	}
 	nvkm_wo32(gpuobj->memory, offset, data);
 }
 
-void
-nvkm_gpuobj_destroy(struct nvkm_gpuobj *gpuobj)
+static const struct nvkm_gpuobj_func nvkm_gpuobj_heap;
+static void
+nvkm_gpuobj_heap_release(struct nvkm_gpuobj *gpuobj)
 {
-	int i;
+	gpuobj->func = &nvkm_gpuobj_heap;
+	nvkm_done(gpuobj->memory);
+}
 
-	if (gpuobj->flags & NVOBJ_FLAG_ZERO_FREE) {
-		nvkm_kmap(gpuobj);
-		for (i = 0; i < gpuobj->size; i += 4)
-			nvkm_wo32(gpuobj, i, 0x00000000);
-		nvkm_done(gpuobj);
-	}
+static const struct nvkm_gpuobj_func
+nvkm_gpuobj_heap_fast = {
+	.release = nvkm_gpuobj_heap_release,
+	.rd32 = nvkm_gpuobj_rd32_fast,
+	.wr32 = nvkm_gpuobj_wr32_fast,
+};
 
-	if (gpuobj->node)
-		nvkm_mm_free(&nv_gpuobj(gpuobj->parent)->heap, &gpuobj->node);
+static const struct nvkm_gpuobj_func
+nvkm_gpuobj_heap_slow = {
+	.release = nvkm_gpuobj_heap_release,
+	.rd32 = nvkm_gpuobj_heap_rd32,
+	.wr32 = nvkm_gpuobj_heap_wr32,
+};
 
-	if (gpuobj->heap.block_size)
-		nvkm_mm_fini(&gpuobj->heap);
+static void *
+nvkm_gpuobj_heap_acquire(struct nvkm_gpuobj *gpuobj)
+{
+	gpuobj->map = nvkm_kmap(gpuobj->memory);
+	if (likely(gpuobj->map))
+		gpuobj->func = &nvkm_gpuobj_heap_fast;
+	else
+		gpuobj->func = &nvkm_gpuobj_heap_slow;
+	return gpuobj->map;
+}
 
-	nvkm_memory_del(&gpuobj->memory);
-	nvkm_object_destroy(&gpuobj->object);
+static const struct nvkm_gpuobj_func
+nvkm_gpuobj_heap = {
+	.acquire = nvkm_gpuobj_heap_acquire,
+};
+
+/* accessor functions for gpuobjs sub-allocated from a parent gpuobj */
+static u32
+nvkm_gpuobj_rd32(struct nvkm_gpuobj *gpuobj, u32 offset)
+{
+	return nvkm_ro32(gpuobj->parent, gpuobj->node->offset + offset);
+}
+
+static void
+nvkm_gpuobj_wr32(struct nvkm_gpuobj *gpuobj, u32 offset, u32 data)
+{
+	nvkm_wo32(gpuobj->parent, gpuobj->node->offset + offset, data);
+}
+
+static const struct nvkm_gpuobj_func nvkm_gpuobj_func;
+static void
+nvkm_gpuobj_release(struct nvkm_gpuobj *gpuobj)
+{
+	gpuobj->func = &nvkm_gpuobj_func;
+	nvkm_done(gpuobj->parent);
 }
 
 static const struct nvkm_gpuobj_func
-nvkm_gpuobj_func = {
-	.acquire = nvkm_gpuobj_acquire,
+nvkm_gpuobj_fast = {
+	.release = nvkm_gpuobj_release,
+	.rd32 = nvkm_gpuobj_rd32_fast,
+	.wr32 = nvkm_gpuobj_wr32_fast,
+};
+
+static const struct nvkm_gpuobj_func
+nvkm_gpuobj_slow = {
 	.release = nvkm_gpuobj_release,
 	.rd32 = nvkm_gpuobj_rd32,
 	.wr32 = nvkm_gpuobj_wr32,
 };
 
+static void *
+nvkm_gpuobj_acquire(struct nvkm_gpuobj *gpuobj)
+{
+	gpuobj->map = nvkm_kmap(gpuobj->parent);
+	if (likely(gpuobj->map)) {
+		gpuobj->map  = (u8 *)gpuobj->map + gpuobj->node->offset;
+		gpuobj->func = &nvkm_gpuobj_fast;
+	} else {
+		gpuobj->func = &nvkm_gpuobj_slow;
+	}
+	return gpuobj->map;
+}
+
+static const struct nvkm_gpuobj_func
+nvkm_gpuobj_func = {
+	.acquire = nvkm_gpuobj_acquire,
+};
+
+static int
+nvkm_gpuobj_ctor(struct nvkm_device *device, u32 size, int align, bool zero,
+		 struct nvkm_gpuobj *parent, struct nvkm_gpuobj *gpuobj)
+{
+	u32 offset;
+	int ret;
+
+	if (parent) {
+		if (align >= 0) {
+			ret = nvkm_mm_head(&parent->heap, 0, 1, size, size,
+					   max(align, 1), &gpuobj->node);
+		} else {
+			ret = nvkm_mm_tail(&parent->heap, 0, 1, size, size,
+					   -align, &gpuobj->node);
+		}
+		if (ret)
+			return ret;
+
+		gpuobj->parent = parent;
+		gpuobj->func = &nvkm_gpuobj_func;
+		gpuobj->addr = parent->addr + gpuobj->node->offset;
+		gpuobj->size = gpuobj->node->length;
+
+		if (zero) {
+			nvkm_kmap(gpuobj);
+			for (offset = 0; offset < gpuobj->size; offset += 4)
+				nvkm_wo32(gpuobj, offset, 0x00000000);
+			nvkm_done(gpuobj);
+		}
+	} else {
+		ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, size,
+				      abs(align), zero, &gpuobj->memory);
+		if (ret)
+			return ret;
+
+		gpuobj->func = &nvkm_gpuobj_heap;
+		gpuobj->addr = nvkm_memory_addr(gpuobj->memory);
+		gpuobj->size = nvkm_memory_size(gpuobj->memory);
+	}
+
+	return nvkm_mm_init(&gpuobj->heap, 0, gpuobj->size, 1);
+}
+
+void
+nvkm_gpuobj_del(struct nvkm_gpuobj **pgpuobj)
+{
+	struct nvkm_gpuobj *gpuobj = *pgpuobj;
+	if (gpuobj) {
+		if (gpuobj->parent)
+			nvkm_mm_free(&gpuobj->parent->heap, &gpuobj->node);
+		nvkm_mm_fini(&gpuobj->heap);
+		nvkm_memory_del(&gpuobj->memory);
+		kfree(*pgpuobj);
+		*pgpuobj = NULL;
+	}
+}
+
+int
+nvkm_gpuobj_new(struct nvkm_device *device, u32 size, int align, bool zero,
+		struct nvkm_gpuobj *parent, struct nvkm_gpuobj **pgpuobj)
+{
+	struct nvkm_gpuobj *gpuobj;
+	int ret;
+
+	if (!(gpuobj = *pgpuobj = kzalloc(sizeof(*gpuobj), GFP_KERNEL)))
+		return -ENOMEM;
+
+	ret = nvkm_gpuobj_ctor(device, size, align, zero, parent, gpuobj);
+	if (ret)
+		nvkm_gpuobj_del(pgpuobj);
+	return ret;
+}
+
+void
+nvkm_gpuobj_destroy(struct nvkm_gpuobj *gpuobj)
+{
+	if (gpuobj->node)
+		nvkm_mm_free(&gpuobj->parent->heap, &gpuobj->node);
+
+	gpuobj->heap.block_size = 1;
+	nvkm_mm_fini(&gpuobj->heap);
+
+	nvkm_memory_del(&gpuobj->memory);
+	nvkm_object_destroy(&gpuobj->object);
+}
+
 int
 nvkm_gpuobj_create_(struct nvkm_object *parent, struct nvkm_object *engine,
 		    struct nvkm_oclass *oclass, u32 pclass,
@@ -103,12 +238,10 @@ nvkm_gpuobj_create_(struct nvkm_object *parent, struct nvkm_object *engine,
 		    int length, void **pobject)
 {
 	struct nvkm_device *device = nv_device(parent);
-	struct nvkm_memory *memory = NULL;
 	struct nvkm_gpuobj *pargpu = NULL;
 	struct nvkm_gpuobj *gpuobj;
-	struct nvkm_mm *heap = NULL;
-	int ret, i;
-	u64 addr;
+	const bool zero = (flags & NVOBJ_FLAG_ZERO_ALLOC);
+	int ret;
 
 	*pobject = NULL;
 
@@ -122,83 +255,18 @@ nvkm_gpuobj_create_(struct nvkm_object *parent, struct nvkm_object *engine,
 		if (WARN_ON(objgpu == NULL))
 			return -EINVAL;
 		pargpu = nv_gpuobj(objgpu);
-
-		addr =  pargpu->addr;
-		heap = &pargpu->heap;
-	} else {
-		ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST,
-				      size, align, false, &memory);
-		if (ret)
-			return ret;
-
-		addr = nvkm_memory_addr(memory);
-		size = nvkm_memory_size(memory);
 	}
 
 	ret = nvkm_object_create_(parent, engine, oclass, pclass |
 				  NV_GPUOBJ_CLASS, length, pobject);
 	gpuobj = *pobject;
-	if (ret) {
-		nvkm_memory_del(&memory);
-		return ret;
-	}
-
-	gpuobj->func = &nvkm_gpuobj_func;
-	gpuobj->memory = memory;
-	gpuobj->parent = pargpu;
-	gpuobj->flags = flags;
-	gpuobj->addr = addr;
-	gpuobj->size = size;
-
-	if (heap) {
-		ret = nvkm_mm_head(heap, 0, 1, size, size, max(align, (u32)1),
-				   &gpuobj->node);
-		if (ret)
-			return ret;
-
-		gpuobj->addr += gpuobj->node->offset;
-	}
-
-	if (gpuobj->flags & NVOBJ_FLAG_HEAP) {
-		ret = nvkm_mm_init(&gpuobj->heap, 0, gpuobj->size, 1);
-		if (ret)
-			return ret;
-	}
-
-	if (flags & NVOBJ_FLAG_ZERO_ALLOC) {
-		nvkm_kmap(gpuobj);
-		for (i = 0; i < gpuobj->size; i += 4)
-			nvkm_wo32(gpuobj, i, 0x00000000);
-		nvkm_done(gpuobj);
-	}
-
-	return ret;
-}
-
-struct nvkm_gpuobj_class {
-	struct nvkm_object *pargpu;
-	u64 size;
-	u32 align;
-	u32 flags;
-};
-
-static int
-_nvkm_gpuobj_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
-		  struct nvkm_oclass *oclass, void *data, u32 size,
-		  struct nvkm_object **pobject)
-{
-	struct nvkm_gpuobj_class *args = data;
-	struct nvkm_gpuobj *object;
-	int ret;
-
-	ret = nvkm_gpuobj_create(parent, engine, oclass, 0, args->pargpu,
-				 args->size, args->align, args->flags,
-				 &object);
-	*pobject = nv_object(object);
 	if (ret)
 		return ret;
 
-	return 0;
+	ret = nvkm_gpuobj_ctor(device, size, align, zero, pargpu, gpuobj);
+	if (!(flags & NVOBJ_FLAG_HEAP))
+		gpuobj->heap.block_size = 0;
+	return ret;
 }
 
 void
@@ -233,39 +301,9 @@ _nvkm_gpuobj_wr32(struct nvkm_object *object, u64 addr, u32 data)
 	nvkm_wo32(gpuobj, addr, data);
 }
 
-static struct nvkm_oclass
-_nvkm_gpuobj_oclass = {
-	.handle = 0x00000000,
-	.ofuncs = &(struct nvkm_ofuncs) {
-		.ctor = _nvkm_gpuobj_ctor,
-		.dtor = _nvkm_gpuobj_dtor,
-		.init = _nvkm_gpuobj_init,
-		.fini = _nvkm_gpuobj_fini,
-		.rd32 = _nvkm_gpuobj_rd32,
-		.wr32 = _nvkm_gpuobj_wr32,
-	},
-};
-
-int
-nvkm_gpuobj_new(struct nvkm_object *parent, struct nvkm_object *pargpu,
-		u32 size, u32 align, u32 flags,
-		struct nvkm_gpuobj **pgpuobj)
-{
-	struct nvkm_gpuobj_class args = {
-		.pargpu = pargpu,
-		.size = size,
-		.align = align,
-		.flags = flags,
-	};
-
-	return nvkm_object_old(parent, &parent->engine->subdev.object,
-			       &_nvkm_gpuobj_oclass, &args, sizeof(args),
-			       (struct nvkm_object **)pgpuobj);
-}
-
 int
-nvkm_gpuobj_map_vm(struct nvkm_gpuobj *gpuobj, struct nvkm_vm *vm,
-		   u32 access, struct nvkm_vma *vma)
+nvkm_gpuobj_map(struct nvkm_gpuobj *gpuobj, struct nvkm_vm *vm,
+		u32 access, struct nvkm_vma *vma)
 {
 	struct nvkm_memory *memory = gpuobj->memory;
 	int ret = nvkm_vm_get(vm, gpuobj->size, 12, access, vma);
@@ -288,37 +326,13 @@ nvkm_gpuobj_unmap(struct nvkm_vma *vma)
  * anywhere else.
  */
 
-static void
-nvkm_gpudup_dtor(struct nvkm_object *object)
-{
-	struct nvkm_gpuobj *gpuobj = (void *)object;
-	nvkm_object_destroy(&gpuobj->object);
-}
-
-static struct nvkm_oclass
-nvkm_gpudup_oclass = {
-	.handle = NV_GPUOBJ_CLASS,
-	.ofuncs = &(struct nvkm_ofuncs) {
-		.dtor = nvkm_gpudup_dtor,
-		.init = _nvkm_object_init,
-		.fini = _nvkm_object_fini,
-	},
-};
-
 int
-nvkm_gpuobj_dup(struct nvkm_object *parent, struct nvkm_memory *base,
-		struct nvkm_gpuobj **pgpuobj)
+nvkm_gpuobj_wrap(struct nvkm_memory *memory, struct nvkm_gpuobj **pgpuobj)
 {
-	struct nvkm_gpuobj *gpuobj;
-	int ret;
-
-	ret = nvkm_object_create(parent, &parent->engine->subdev.object,
-				 &nvkm_gpudup_oclass, 0, &gpuobj);
-	*pgpuobj = gpuobj;
-	if (ret)
-		return ret;
+	if (!(*pgpuobj = kzalloc(sizeof(**pgpuobj), GFP_KERNEL)))
+		return -ENOMEM;
 
-	gpuobj->addr = nvkm_memory_addr(base);
-	gpuobj->size = nvkm_memory_size(base);
+	(*pgpuobj)->addr = nvkm_memory_addr(memory);
+	(*pgpuobj)->size = nvkm_memory_size(memory);
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/base.c
index f9f38ee49058d..7f9870789a0f3 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/base.c
@@ -24,32 +24,74 @@
 #include "priv.h"
 
 #include <core/client.h>
+#include <core/gpuobj.h>
 #include <subdev/fb.h>
 #include <subdev/instmem.h>
 
 #include <nvif/class.h>
 #include <nvif/unpack.h>
 
+struct hack {
+	struct nvkm_gpuobj object;
+	struct nvkm_gpuobj *parent;
+};
+
+static void
+dtor(struct nvkm_object *object)
+{
+	struct hack *hack = (void *)object;
+	nvkm_gpuobj_del(&hack->parent);
+	nvkm_object_destroy(&hack->object.object);
+}
+
+static struct nvkm_oclass
+hack = {
+	.handle = NV_GPUOBJ_CLASS,
+	.ofuncs = &(struct nvkm_ofuncs) {
+		.dtor = dtor,
+		.init = _nvkm_object_init,
+		.fini = _nvkm_object_fini,
+	},
+};
+
 static int
-nvkm_dmaobj_bind(struct nvkm_dmaobj *dmaobj, struct nvkm_object *parent,
+nvkm_dmaobj_bind(struct nvkm_dmaobj *dmaobj, struct nvkm_gpuobj *pargpu,
 		 struct nvkm_gpuobj **pgpuobj)
 {
 	const struct nvkm_dmaeng_impl *impl = (void *)
 		nv_oclass(nv_object(dmaobj)->engine);
 	int ret = 0;
 
-	if (nv_object(dmaobj) == parent) { /* ctor bind */
+	if (&dmaobj->base == &pargpu->object) { /* ctor bind */
+		struct nvkm_object *parent = (void *)pargpu;
+		struct hack *object;
+
 		if (nv_mclass(parent->parent) == NV_DEVICE) {
 			/* delayed, or no, binding */
 			return 0;
 		}
-		ret = impl->bind(dmaobj, parent, pgpuobj);
-		if (ret == 0)
+
+		pargpu = (void *)nv_pclass((void *)pargpu, NV_GPUOBJ_CLASS);
+
+		ret = nvkm_object_create(parent, NULL, &hack, NV_GPUOBJ_CLASS, &object);
+		if (ret == 0) {
 			nvkm_object_ref(NULL, &parent);
+			*pgpuobj = &object->object;
+
+			ret = impl->bind(dmaobj, pargpu, &object->parent);
+			if (ret)
+				return ret;
+
+			object->object.node = object->parent->node;
+			object->object.addr = object->parent->addr;
+			object->object.size = object->parent->size;
+			return 0;
+		}
+
 		return ret;
 	}
 
-	return impl->bind(dmaobj, parent, pgpuobj);
+	return impl->bind(dmaobj, pargpu, pgpuobj);
 }
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/gf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/gf100.c
index 499a7c7e024a5..96a604de4b632 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/gf100.c
@@ -37,25 +37,14 @@ struct gf100_dmaobj {
 };
 
 static int
-gf100_dmaobj_bind(struct nvkm_dmaobj *obj, struct nvkm_object *parent,
+gf100_dmaobj_bind(struct nvkm_dmaobj *obj, struct nvkm_gpuobj *parent,
 		  struct nvkm_gpuobj **pgpuobj)
 {
 	struct gf100_dmaobj *dmaobj = container_of(obj, typeof(*dmaobj), base);
+	struct nvkm_device *device = dmaobj->base.base.engine->subdev.device;
 	int ret;
 
-	if (!nv_iclass(parent, NV_ENGCTX_CLASS)) {
-		switch (nv_mclass(parent->parent)) {
-		case GT214_DISP_CORE_CHANNEL_DMA:
-		case GT214_DISP_BASE_CHANNEL_DMA:
-		case GT214_DISP_OVERLAY_CHANNEL_DMA:
-			break;
-		default:
-			return -EINVAL;
-		}
-	} else
-		return 0;
-
-	ret = nvkm_gpuobj_new(parent, parent, 24, 32, 0, pgpuobj);
+	ret = nvkm_gpuobj_new(device, 24, 32, false, parent, pgpuobj);
 	if (ret == 0) {
 		nvkm_kmap(*pgpuobj);
 		nvkm_wo32(*pgpuobj, 0x00, dmaobj->flags0 | nv_mclass(dmaobj));
@@ -146,7 +135,7 @@ gf100_dmaobj_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 		break;
 	}
 
-	return dmaeng->bind(&dmaobj->base, nv_object(dmaobj), (void *)pobject);
+	return dmaeng->bind(&dmaobj->base, (void *)dmaobj, (void *)pobject);
 }
 
 static struct nvkm_ofuncs
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/gf110.c b/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/gf110.c
index a28cf56454e46..017c7a2affe51 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/gf110.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/gf110.c
@@ -36,32 +36,14 @@ struct gf110_dmaobj {
 };
 
 static int
-gf110_dmaobj_bind(struct nvkm_dmaobj *obj, struct nvkm_object *parent,
+gf110_dmaobj_bind(struct nvkm_dmaobj *obj, struct nvkm_gpuobj *parent,
 		  struct nvkm_gpuobj **pgpuobj)
 {
 	struct gf110_dmaobj *dmaobj = container_of(obj, typeof(*dmaobj), base);
+	struct nvkm_device *device = dmaobj->base.base.engine->subdev.device;
 	int ret;
 
-	if (!nv_iclass(parent, NV_ENGCTX_CLASS)) {
-		switch (nv_mclass(parent->parent)) {
-		case GF110_DISP_CORE_CHANNEL_DMA:
-		case GK104_DISP_CORE_CHANNEL_DMA:
-		case GK110_DISP_CORE_CHANNEL_DMA:
-		case GM107_DISP_CORE_CHANNEL_DMA:
-		case GM204_DISP_CORE_CHANNEL_DMA:
-		case GF110_DISP_BASE_CHANNEL_DMA:
-		case GK104_DISP_BASE_CHANNEL_DMA:
-		case GK110_DISP_BASE_CHANNEL_DMA:
-		case GF110_DISP_OVERLAY_CONTROL_DMA:
-		case GK104_DISP_OVERLAY_CONTROL_DMA:
-			break;
-		default:
-			return -EINVAL;
-		}
-	} else
-		return 0;
-
-	ret = nvkm_gpuobj_new(parent, parent, 24, 32, 0, pgpuobj);
+	ret = nvkm_gpuobj_new(device, 24, 32, false, parent, pgpuobj);
 	if (ret == 0) {
 		nvkm_kmap(*pgpuobj);
 		nvkm_wo32(*pgpuobj, 0x00, dmaobj->flags0);
@@ -135,7 +117,7 @@ gf110_dmaobj_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 		return -EINVAL;
 	}
 
-	return dmaeng->bind(&dmaobj->base, nv_object(dmaobj), (void *)pobject);
+	return dmaeng->bind(&dmaobj->base, (void *)dmaobj, (void *)pobject);
 }
 
 static struct nvkm_ofuncs
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/nv04.c b/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/nv04.c
index 9c68eb0e4a70f..6be2dfc70cb71 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/nv04.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/nv04.c
@@ -37,41 +37,28 @@ struct nv04_dmaobj {
 };
 
 static int
-nv04_dmaobj_bind(struct nvkm_dmaobj *obj, struct nvkm_object *parent,
+nv04_dmaobj_bind(struct nvkm_dmaobj *obj, struct nvkm_gpuobj *parent,
 		 struct nvkm_gpuobj **pgpuobj)
 {
 	struct nv04_dmaobj *dmaobj = container_of(obj, typeof(*dmaobj), base);
-	struct nvkm_gpuobj *gpuobj;
+	struct nvkm_device *device = dmaobj->base.base.engine->subdev.device;
 	u64 offset = dmaobj->base.start & 0xfffff000;
 	u64 adjust = dmaobj->base.start & 0x00000fff;
 	u32 length = dmaobj->base.limit - dmaobj->base.start;
 	int ret;
 
-	if (!nv_iclass(parent, NV_ENGCTX_CLASS)) {
-		switch (nv_mclass(parent->parent)) {
-		case NV03_CHANNEL_DMA:
-		case NV10_CHANNEL_DMA:
-		case NV17_CHANNEL_DMA:
-		case NV40_CHANNEL_DMA:
-			break;
-		default:
-			return -EINVAL;
-		}
-	}
-
 	if (dmaobj->clone) {
 		struct nv04_mmu *mmu = nv04_mmu(dmaobj);
 		struct nvkm_memory *pgt = mmu->vm->pgt[0].mem[0];
 		if (!dmaobj->base.start)
-			return nvkm_gpuobj_dup(parent, pgt, pgpuobj);
+			return nvkm_gpuobj_wrap(pgt, pgpuobj);
 		nvkm_kmap(pgt);
 		offset  = nvkm_ro32(pgt, 8 + (offset >> 10));
 		offset &= 0xfffff000;
 		nvkm_done(pgt);
 	}
 
-	ret = nvkm_gpuobj_new(parent, parent, 16, 16, 0, &gpuobj);
-	*pgpuobj = gpuobj;
+	ret = nvkm_gpuobj_new(device, 16, 16, false, parent, pgpuobj);
 	if (ret == 0) {
 		nvkm_kmap(*pgpuobj);
 		nvkm_wo32(*pgpuobj, 0x00, dmaobj->flags0 | (adjust << 20));
@@ -134,7 +121,7 @@ nv04_dmaobj_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 		return -EINVAL;
 	}
 
-	return dmaeng->bind(&dmaobj->base, nv_object(dmaobj), (void *)pobject);
+	return dmaeng->bind(&dmaobj->base, (void *)dmaobj, (void *)pobject);
 }
 
 static struct nvkm_ofuncs
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/nv50.c b/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/nv50.c
index 3566fa9b3ba97..f87134ec5b341 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/nv50.c
@@ -37,37 +37,14 @@ struct nv50_dmaobj {
 };
 
 static int
-nv50_dmaobj_bind(struct nvkm_dmaobj *obj, struct nvkm_object *parent,
+nv50_dmaobj_bind(struct nvkm_dmaobj *obj, struct nvkm_gpuobj *parent,
 		 struct nvkm_gpuobj **pgpuobj)
 {
 	struct nv50_dmaobj *dmaobj = container_of(obj, typeof(*dmaobj), base);
+	struct nvkm_device *device = dmaobj->base.base.engine->subdev.device;
 	int ret;
 
-	if (!nv_iclass(parent, NV_ENGCTX_CLASS)) {
-		switch (nv_mclass(parent->parent)) {
-		case NV40_CHANNEL_DMA:
-		case NV50_CHANNEL_GPFIFO:
-		case G82_CHANNEL_GPFIFO:
-		case NV50_DISP_CORE_CHANNEL_DMA:
-		case G82_DISP_CORE_CHANNEL_DMA:
-		case GT206_DISP_CORE_CHANNEL_DMA:
-		case GT200_DISP_CORE_CHANNEL_DMA:
-		case GT214_DISP_CORE_CHANNEL_DMA:
-		case NV50_DISP_BASE_CHANNEL_DMA:
-		case G82_DISP_BASE_CHANNEL_DMA:
-		case GT200_DISP_BASE_CHANNEL_DMA:
-		case GT214_DISP_BASE_CHANNEL_DMA:
-		case NV50_DISP_OVERLAY_CHANNEL_DMA:
-		case G82_DISP_OVERLAY_CHANNEL_DMA:
-		case GT200_DISP_OVERLAY_CHANNEL_DMA:
-		case GT214_DISP_OVERLAY_CHANNEL_DMA:
-			break;
-		default:
-			return -EINVAL;
-		}
-	}
-
-	ret = nvkm_gpuobj_new(parent, parent, 24, 32, 0, pgpuobj);
+	ret = nvkm_gpuobj_new(device, 24, 32, false, parent, pgpuobj);
 	if (ret == 0) {
 		nvkm_kmap(*pgpuobj);
 		nvkm_wo32(*pgpuobj, 0x00, dmaobj->flags0 | nv_mclass(dmaobj));
@@ -164,7 +141,7 @@ nv50_dmaobj_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 		return -EINVAL;
 	}
 
-	return dmaeng->bind(&dmaobj->base, nv_object(dmaobj), (void *)pobject);
+	return dmaeng->bind(&dmaobj->base, (void *)dmaobj, (void *)pobject);
 }
 
 static struct nvkm_ofuncs
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/priv.h b/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/priv.h
index c4c210bdbf7e9..614d9e697bd42 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/dmaobj/priv.h
@@ -22,7 +22,7 @@ int _nvkm_dmaeng_ctor(struct nvkm_object *, struct nvkm_object *,
 struct nvkm_dmaeng_impl {
 	struct nvkm_oclass base;
 	struct nvkm_oclass *sclass;
-	int (*bind)(struct nvkm_dmaobj *, struct nvkm_object *,
+	int (*bind)(struct nvkm_dmaobj *, struct nvkm_gpuobj *,
 		    struct nvkm_gpuobj **);
 };
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
index f937d705d6687..1934dfb5a434a 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
@@ -61,6 +61,7 @@ nvkm_fifo_channel_create_(struct nvkm_object *parent,
 	struct nvkm_handle *handle;
 	struct nvkm_dmaobj *dmaobj;
 	struct nvkm_fifo *fifo = (void *)engine;
+	struct nvkm_fifo_base *base = (void *)parent;
 	struct nvkm_fifo_chan *chan;
 	struct nvkm_dmaeng *dmaeng;
 	struct nvkm_subdev *subdev = &fifo->engine.subdev;
@@ -91,7 +92,7 @@ nvkm_fifo_channel_create_(struct nvkm_object *parent,
 			return -EINVAL;
 		}
 
-		ret = dmaeng->bind(dmaobj, parent, &chan->pushgpu);
+		ret = dmaeng->bind(dmaobj, &base->gpuobj, &chan->pushgpu);
 		if (ret)
 			return ret;
 	}
@@ -131,7 +132,7 @@ nvkm_fifo_channel_destroy(struct nvkm_fifo_chan *chan)
 	fifo->channel[chan->chid] = NULL;
 	spin_unlock_irqrestore(&fifo->lock, flags);
 
-	nvkm_gpuobj_ref(NULL, &chan->pushgpu);
+	nvkm_gpuobj_del(&chan->pushgpu);
 	nvkm_namedb_destroy(&chan->namedb);
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c
index 96c6835e9f149..058296b2f2852 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c
@@ -140,23 +140,25 @@ g84_fifo_object_attach(struct nvkm_object *parent,
 	else
 		context = 0x00000004; /* just non-zero */
 
-	switch (nv_engidx(object->engine)) {
-	case NVDEV_ENGINE_DMAOBJ:
-	case NVDEV_ENGINE_SW    : context |= 0x00000000; break;
-	case NVDEV_ENGINE_GR    : context |= 0x00100000; break;
-	case NVDEV_ENGINE_MPEG  :
-	case NVDEV_ENGINE_MSPPP : context |= 0x00200000; break;
-	case NVDEV_ENGINE_ME    :
-	case NVDEV_ENGINE_CE0   : context |= 0x00300000; break;
-	case NVDEV_ENGINE_VP    :
-	case NVDEV_ENGINE_MSPDEC: context |= 0x00400000; break;
-	case NVDEV_ENGINE_CIPHER:
-	case NVDEV_ENGINE_SEC   :
-	case NVDEV_ENGINE_VIC   : context |= 0x00500000; break;
-	case NVDEV_ENGINE_BSP   :
-	case NVDEV_ENGINE_MSVLD : context |= 0x00600000; break;
-	default:
-		return -EINVAL;
+	if (object->engine) {
+		switch (nv_engidx(object->engine)) {
+		case NVDEV_ENGINE_DMAOBJ:
+		case NVDEV_ENGINE_SW    : context |= 0x00000000; break;
+		case NVDEV_ENGINE_GR    : context |= 0x00100000; break;
+		case NVDEV_ENGINE_MPEG  :
+		case NVDEV_ENGINE_MSPPP : context |= 0x00200000; break;
+		case NVDEV_ENGINE_ME    :
+		case NVDEV_ENGINE_CE0   : context |= 0x00300000; break;
+		case NVDEV_ENGINE_VP    :
+		case NVDEV_ENGINE_MSPDEC: context |= 0x00400000; break;
+		case NVDEV_ENGINE_CIPHER:
+		case NVDEV_ENGINE_SEC   :
+		case NVDEV_ENGINE_VIC   : context |= 0x00500000; break;
+		case NVDEV_ENGINE_BSP   :
+		case NVDEV_ENGINE_MSVLD : context |= 0x00600000; break;
+		default:
+			return -EINVAL;
+		}
 	}
 
 	return nvkm_ramht_insert(chan->ramht, 0, handle, context);
@@ -374,6 +376,7 @@ g84_fifo_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 		      struct nvkm_oclass *oclass, void *data, u32 size,
 		      struct nvkm_object **pobject)
 {
+	struct nvkm_device *device = nv_engine(engine)->subdev.device;
 	struct nv50_fifo_base *base;
 	int ret;
 
@@ -383,13 +386,13 @@ g84_fifo_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(base), nv_object(base), 0x0200, 0,
-			      NVOBJ_FLAG_ZERO_ALLOC, &base->eng);
+	ret = nvkm_gpuobj_new(device, 0x0200, 0, true, &base->base.gpuobj,
+			      &base->eng);
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(base), nv_object(base), 0x4000, 0,
-			      0, &base->pgd);
+	ret = nvkm_gpuobj_new(device, 0x4000, 0, false, &base->base.gpuobj,
+			      &base->pgd);
 	if (ret)
 		return ret;
 
@@ -397,13 +400,13 @@ g84_fifo_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(base), nv_object(base), 0x1000,
-			      0x400, NVOBJ_FLAG_ZERO_ALLOC, &base->cache);
+	ret = nvkm_gpuobj_new(device, 0x1000, 0x400, true, &base->base.gpuobj,
+			      &base->cache);
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(base), nv_object(base), 0x0100,
-			      0x100, NVOBJ_FLAG_ZERO_ALLOC, &base->ramfc);
+	ret = nvkm_gpuobj_new(device, 0x100, 0x100, true, &base->base.gpuobj,
+			      &base->ramfc);
 	if (ret)
 		return ret;
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
index cbb8d249e7fbd..77b8df1f57fba 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
@@ -130,8 +130,8 @@ gf100_fifo_context_attach(struct nvkm_object *parent,
 	}
 
 	if (!ectx->vma.node) {
-		ret = nvkm_gpuobj_map_vm(nv_gpuobj(ectx), base->vm,
-					 NV_MEM_ACCESS_RW, &ectx->vma);
+		ret = nvkm_gpuobj_map(nv_gpuobj(ectx), base->vm,
+				      NV_MEM_ACCESS_RW, &ectx->vma);
 		if (ret)
 			return ret;
 
@@ -334,6 +334,7 @@ gf100_fifo_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 			struct nvkm_oclass *oclass, void *data, u32 size,
 			struct nvkm_object **pobject)
 {
+	struct nvkm_device *device = nv_engine(engine)->subdev.device;
 	struct gf100_fifo_base *base;
 	int ret;
 
@@ -344,8 +345,7 @@ gf100_fifo_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(base), NULL, 0x10000, 0x1000, 0,
-			      &base->pgd);
+	ret = nvkm_gpuobj_new(device, 0x10000, 0x1000, false, NULL, &base->pgd);
 	if (ret)
 		return ret;
 
@@ -368,7 +368,7 @@ gf100_fifo_context_dtor(struct nvkm_object *object)
 {
 	struct gf100_fifo_base *base = (void *)object;
 	nvkm_vm_ref(NULL, &base->vm, base->pgd);
-	nvkm_gpuobj_ref(NULL, &base->pgd);
+	nvkm_gpuobj_del(&base->pgd);
 	nvkm_fifo_context_destroy(&base->base);
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
index 0e8356d5fa02e..39dae1a28dd1e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
@@ -154,8 +154,8 @@ gk104_fifo_context_attach(struct nvkm_object *parent,
 	}
 
 	if (!ectx->vma.node) {
-		ret = nvkm_gpuobj_map_vm(nv_gpuobj(ectx), base->vm,
-					 NV_MEM_ACCESS_RW, &ectx->vma);
+		ret = nvkm_gpuobj_map(nv_gpuobj(ectx), base->vm,
+				      NV_MEM_ACCESS_RW, &ectx->vma);
 		if (ret)
 			return ret;
 
@@ -388,6 +388,7 @@ gk104_fifo_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 			struct nvkm_oclass *oclass, void *data, u32 size,
 			struct nvkm_object **pobject)
 {
+	struct nvkm_device *device = nv_engine(engine)->subdev.device;
 	struct gk104_fifo_base *base;
 	int ret;
 
@@ -397,8 +398,7 @@ gk104_fifo_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(base), NULL, 0x10000, 0x1000, 0,
-			      &base->pgd);
+	ret = nvkm_gpuobj_new(device, 0x10000, 0x1000, false, NULL, &base->pgd);
 	if (ret)
 		return ret;
 
@@ -421,7 +421,7 @@ gk104_fifo_context_dtor(struct nvkm_object *object)
 {
 	struct gk104_fifo_base *base = (void *)object;
 	nvkm_vm_ref(NULL, &base->vm, base->pgd);
-	nvkm_gpuobj_ref(NULL, &base->pgd);
+	nvkm_gpuobj_del(&base->pgd);
 	nvkm_fifo_context_destroy(&base->base);
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c
index f0fd1ce8f671b..4bec707305586 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c
@@ -65,19 +65,21 @@ nv04_fifo_object_attach(struct nvkm_object *parent,
 	else
 		context = 0x00000004; /* just non-zero */
 
-	switch (nv_engidx(object->engine)) {
-	case NVDEV_ENGINE_DMAOBJ:
-	case NVDEV_ENGINE_SW:
-		context |= 0x00000000;
-		break;
-	case NVDEV_ENGINE_GR:
-		context |= 0x00010000;
-		break;
-	case NVDEV_ENGINE_MPEG:
-		context |= 0x00020000;
-		break;
-	default:
-		return -EINVAL;
+	if (object->engine) {
+		switch (nv_engidx(object->engine)) {
+		case NVDEV_ENGINE_DMAOBJ:
+		case NVDEV_ENGINE_SW:
+			context |= 0x00000000;
+			break;
+		case NVDEV_ENGINE_GR:
+			context |= 0x00010000;
+			break;
+		case NVDEV_ENGINE_MPEG:
+			context |= 0x00020000;
+			break;
+		default:
+			return -EINVAL;
+		}
 	}
 
 	context |= 0x80000000; /* valid */
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c
index eb6588ec0b6d4..a2d8da8a9341e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c
@@ -78,19 +78,21 @@ nv40_fifo_object_attach(struct nvkm_object *parent,
 	else
 		context = 0x00000004; /* just non-zero */
 
-	switch (nv_engidx(object->engine)) {
-	case NVDEV_ENGINE_DMAOBJ:
-	case NVDEV_ENGINE_SW:
-		context |= 0x00000000;
-		break;
-	case NVDEV_ENGINE_GR:
-		context |= 0x00100000;
-		break;
-	case NVDEV_ENGINE_MPEG:
-		context |= 0x00200000;
-		break;
-	default:
-		return -EINVAL;
+	if (object->engine) {
+		switch (nv_engidx(object->engine)) {
+		case NVDEV_ENGINE_DMAOBJ:
+		case NVDEV_ENGINE_SW:
+			context |= 0x00000000;
+			break;
+		case NVDEV_ENGINE_GR:
+			context |= 0x00100000;
+			break;
+		case NVDEV_ENGINE_MPEG:
+			context |= 0x00200000;
+			break;
+		default:
+			return -EINVAL;
+		}
 	}
 
 	context |= chid << 23;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c
index 96e5b61daf10c..620c0cfb2453b 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c
@@ -171,13 +171,15 @@ nv50_fifo_object_attach(struct nvkm_object *parent,
 	else
 		context = 0x00000004; /* just non-zero */
 
-	switch (nv_engidx(object->engine)) {
-	case NVDEV_ENGINE_DMAOBJ:
-	case NVDEV_ENGINE_SW    : context |= 0x00000000; break;
-	case NVDEV_ENGINE_GR    : context |= 0x00100000; break;
-	case NVDEV_ENGINE_MPEG  : context |= 0x00200000; break;
-	default:
-		return -EINVAL;
+	if (object->engine) {
+		switch (nv_engidx(object->engine)) {
+		case NVDEV_ENGINE_DMAOBJ:
+		case NVDEV_ENGINE_SW    : context |= 0x00000000; break;
+		case NVDEV_ENGINE_GR    : context |= 0x00100000; break;
+		case NVDEV_ENGINE_MPEG  : context |= 0x00200000; break;
+		default:
+			return -EINVAL;
+		}
 	}
 
 	return nvkm_ramht_insert(chan->ramht, 0, handle, context);
@@ -402,6 +404,7 @@ nv50_fifo_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 		       struct nvkm_oclass *oclass, void *data, u32 size,
 		       struct nvkm_object **pobject)
 {
+	struct nvkm_device *device = nv_engine(engine)->subdev.device;
 	struct nv50_fifo_base *base;
 	int ret;
 
@@ -411,17 +414,17 @@ nv50_fifo_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(base), nv_object(base), 0x0200,
-			      0x1000, NVOBJ_FLAG_ZERO_ALLOC, &base->ramfc);
+	ret = nvkm_gpuobj_new(device, 0x0200, 0x1000, true, &base->base.gpuobj,
+			      &base->ramfc);
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(base), nv_object(base), 0x1200, 0,
-			      NVOBJ_FLAG_ZERO_ALLOC, &base->eng);
+	ret = nvkm_gpuobj_new(device, 0x1200, 0, true, &base->base.gpuobj,
+			      &base->eng);
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(base), nv_object(base), 0x4000, 0, 0,
+	ret = nvkm_gpuobj_new(device, 0x4000, 0, false, &base->base.gpuobj,
 			      &base->pgd);
 	if (ret)
 		return ret;
@@ -438,10 +441,10 @@ nv50_fifo_context_dtor(struct nvkm_object *object)
 {
 	struct nv50_fifo_base *base = (void *)object;
 	nvkm_vm_ref(NULL, &base->vm, base->pgd);
-	nvkm_gpuobj_ref(NULL, &base->pgd);
-	nvkm_gpuobj_ref(NULL, &base->eng);
-	nvkm_gpuobj_ref(NULL, &base->ramfc);
-	nvkm_gpuobj_ref(NULL, &base->cache);
+	nvkm_gpuobj_del(&base->pgd);
+	nvkm_gpuobj_del(&base->eng);
+	nvkm_gpuobj_del(&base->ramfc);
+	nvkm_gpuobj_del(&base->cache);
 	nvkm_fifo_context_destroy(&base->base);
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bar/gf100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bar/gf100.c
index 8e3574631068f..7d304ccc6a64a 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bar/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bar/gf100.c
@@ -75,8 +75,7 @@ gf100_bar_ctor_vm(struct gf100_bar *bar, struct gf100_bar_vm *bar_vm,
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(bar), NULL, 0x8000, 0, 0,
-			      &bar_vm->pgd);
+	ret = nvkm_gpuobj_new(device, 0x8000, 0, false, NULL, &bar_vm->pgd);
 	if (ret)
 		return ret;
 
@@ -157,14 +156,14 @@ gf100_bar_dtor(struct nvkm_object *object)
 	struct gf100_bar *bar = (void *)object;
 
 	nvkm_vm_ref(NULL, &bar->bar[1].vm, bar->bar[1].pgd);
-	nvkm_gpuobj_ref(NULL, &bar->bar[1].pgd);
+	nvkm_gpuobj_del(&bar->bar[1].pgd);
 	nvkm_memory_del(&bar->bar[1].mem);
 
 	if (bar->bar[0].vm) {
 		nvkm_memory_del(&bar->bar[0].vm->pgt[0].mem[0]);
 		nvkm_vm_ref(NULL, &bar->bar[0].vm, bar->bar[0].pgd);
 	}
-	nvkm_gpuobj_ref(NULL, &bar->bar[0].pgd);
+	nvkm_gpuobj_del(&bar->bar[0].pgd);
 	nvkm_memory_del(&bar->bar[0].mem);
 
 	nvkm_bar_destroy(&bar->base);
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bar/nv50.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bar/nv50.c
index 8e17b62fa1dae..add132b2ed7aa 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bar/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bar/nv50.c
@@ -99,7 +99,6 @@ nv50_bar_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	static struct lock_class_key bar1_lock;
 	static struct lock_class_key bar3_lock;
 	struct nvkm_device *device = nv_device(parent);
-	struct nvkm_object *heap;
 	struct nvkm_vm *vm;
 	struct nv50_bar *bar;
 	u64 start, limit;
@@ -110,19 +109,17 @@ nv50_bar_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(bar), NULL, 0x20000, 0,
-			      NVOBJ_FLAG_HEAP, &bar->mem);
-	heap = nv_object(bar->mem);
+	ret = nvkm_gpuobj_new(device, 0x20000, 0, false, NULL, &bar->mem);
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(bar), heap,
-			      (device->chipset == 0x50) ? 0x1400 : 0x0200,
-			      0, 0, &bar->pad);
+	ret = nvkm_gpuobj_new(device, (device->chipset == 0x50) ?
+			      0x1400 : 0x200, 0, false, bar->mem,
+			      &bar->pad);
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(bar), heap, 0x4000, 0, 0, &bar->pgd);
+	ret = nvkm_gpuobj_new(device, 0x4000, 0, false, bar->mem, &bar->pgd);
 	if (ret)
 		return ret;
 
@@ -145,7 +142,7 @@ nv50_bar_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(bar), heap, 24, 16, 0, &bar->bar3);
+	ret = nvkm_gpuobj_new(device, 24, 16, false, bar->mem, &bar->bar3);
 	if (ret)
 		return ret;
 
@@ -174,7 +171,7 @@ nv50_bar_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	if (ret)
 		return ret;
 
-	ret = nvkm_gpuobj_new(nv_object(bar), heap, 24, 16, 0, &bar->bar1);
+	ret = nvkm_gpuobj_new(device, 24, 16, false, bar->mem, &bar->bar1);
 	if (ret)
 		return ret;
 
@@ -203,16 +200,16 @@ static void
 nv50_bar_dtor(struct nvkm_object *object)
 {
 	struct nv50_bar *bar = (void *)object;
-	nvkm_gpuobj_ref(NULL, &bar->bar1);
+	nvkm_gpuobj_del(&bar->bar1);
 	nvkm_vm_ref(NULL, &bar->bar1_vm, bar->pgd);
-	nvkm_gpuobj_ref(NULL, &bar->bar3);
+	nvkm_gpuobj_del(&bar->bar3);
 	if (bar->bar3_vm) {
 		nvkm_memory_del(&bar->bar3_vm->pgt[0].mem[0]);
 		nvkm_vm_ref(NULL, &bar->bar3_vm, bar->pgd);
 	}
-	nvkm_gpuobj_ref(NULL, &bar->pgd);
-	nvkm_gpuobj_ref(NULL, &bar->pad);
-	nvkm_gpuobj_ref(NULL, &bar->mem);
+	nvkm_gpuobj_del(&bar->pgd);
+	nvkm_gpuobj_del(&bar->pad);
+	nvkm_gpuobj_del(&bar->mem);
 	nvkm_bar_destroy(&bar->base);
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
index 9c712818528b9..6fa1bdb02dfdb 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
@@ -420,7 +420,7 @@ nvkm_vm_link(struct nvkm_vm *vm, struct nvkm_gpuobj *pgd)
 	if (!vpgd)
 		return -ENOMEM;
 
-	nvkm_gpuobj_ref(pgd, &vpgd->obj);
+	vpgd->obj = pgd;
 
 	mutex_lock(&vm->mutex);
 	for (i = vm->fpde; i <= vm->lpde; i++)
@@ -434,7 +434,6 @@ static void
 nvkm_vm_unlink(struct nvkm_vm *vm, struct nvkm_gpuobj *mpgd)
 {
 	struct nvkm_vm_pgd *vpgd, *tmp;
-	struct nvkm_gpuobj *pgd = NULL;
 
 	if (!mpgd)
 		return;
@@ -442,15 +441,12 @@ nvkm_vm_unlink(struct nvkm_vm *vm, struct nvkm_gpuobj *mpgd)
 	mutex_lock(&vm->mutex);
 	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
 		if (vpgd->obj == mpgd) {
-			pgd = vpgd->obj;
 			list_del(&vpgd->head);
 			kfree(vpgd);
 			break;
 		}
 	}
 	mutex_unlock(&vm->mutex);
-
-	nvkm_gpuobj_ref(NULL, &pgd);
 }
 
 static void