drm/nouveau/imem: improve management of instance memory

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

34 files changed, 968 insertions, 656 deletions

diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/gpuobj.h b/drivers/gpu/drm/nouveau/include/nvkm/core/gpuobj.h
index 2260aef3ec3e8..4e4a9964d235b 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/gpuobj.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/gpuobj.h
@@ -1,6 +1,7 @@
 #ifndef __NVKM_GPUOBJ_H__
 #define __NVKM_GPUOBJ_H__
 #include <core/object.h>
+#include <core/memory.h>
 #include <core/mm.h>
 struct nvkm_vma;
 struct nvkm_vm;
@@ -11,13 +12,23 @@ struct nvkm_vm;
 
 struct nvkm_gpuobj {
 	struct nvkm_object object;
-	struct nvkm_object *parent;
+	struct nvkm_memory *memory;
+	struct nvkm_gpuobj *parent;
 	struct nvkm_mm_node *node;
 	struct nvkm_mm heap;
 
 	u32 flags;
 	u64 addr;
 	u32 size;
+
+	const struct nvkm_gpuobj_func *func;
+};
+
+struct nvkm_gpuobj_func {
+	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);
 };
 
 static inline struct nvkm_gpuobj *
@@ -60,24 +71,4 @@ int  _nvkm_gpuobj_init(struct nvkm_object *);
 int  _nvkm_gpuobj_fini(struct nvkm_object *, bool);
 u32  _nvkm_gpuobj_rd32(struct nvkm_object *, u64);
 void _nvkm_gpuobj_wr32(struct nvkm_object *, u64, u32);
-
-/* accessor macros - kmap()/done() must bracket use of the other accessor
- * macros to guarantee correct behaviour across all chipsets
- */
-#define nvkm_kmap(o) do {                                                      \
-	struct nvkm_gpuobj *_gpuobj = (o);                                     \
-	(void)_gpuobj;                                                         \
-} while(0)
-#define nvkm_ro32(o,a) ({                                                      \
-	u32 _data;                                                             \
-	nvkm_object_rd32(&(o)->object, (a), &_data);                           \
-	_data;                                                                 \
-})
-#define nvkm_wo32(o,a,d) nvkm_object_wr32(&(o)->object, (a), (d))
-#define nvkm_mo32(o,a,m,d) ({                                                  \
-	u32 _addr = (a), _data = nvkm_ro32((o), _addr);                        \
-	nvkm_wo32((o), _addr, (_data & ~(m)) | (d));                           \
-	_data;                                                                 \
-})
-#define nvkm_done(o) nvkm_kmap(o)
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/memory.h b/drivers/gpu/drm/nouveau/include/nvkm/core/memory.h
new file mode 100644
index 0000000000000..9363b839a9da0
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/memory.h
@@ -0,0 +1,53 @@
+#ifndef __NVKM_MEMORY_H__
+#define __NVKM_MEMORY_H__
+#include <core/os.h>
+struct nvkm_device;
+struct nvkm_vma;
+struct nvkm_vm;
+
+enum nvkm_memory_target {
+	NVKM_MEM_TARGET_INST,
+	NVKM_MEM_TARGET_VRAM,
+	NVKM_MEM_TARGET_HOST,
+};
+
+struct nvkm_memory {
+	const struct nvkm_memory_func *func;
+};
+
+struct nvkm_memory_func {
+	void *(*dtor)(struct nvkm_memory *);
+	enum nvkm_memory_target (*target)(struct nvkm_memory *);
+	u64 (*addr)(struct nvkm_memory *);
+	u64 (*size)(struct nvkm_memory *);
+	void (*boot)(struct nvkm_memory *, struct nvkm_vm *);
+	void __iomem *(*acquire)(struct nvkm_memory *);
+	void (*release)(struct nvkm_memory *);
+	u32 (*rd32)(struct nvkm_memory *, u64 offset);
+	void (*wr32)(struct nvkm_memory *, u64 offset, u32 data);
+	void (*map)(struct nvkm_memory *, struct nvkm_vma *, u64 offset);
+};
+
+void nvkm_memory_ctor(const struct nvkm_memory_func *, struct nvkm_memory *);
+int nvkm_memory_new(struct nvkm_device *, enum nvkm_memory_target,
+		    u64 size, u32 align, bool zero, struct nvkm_memory **);
+void nvkm_memory_del(struct nvkm_memory **);
+#define nvkm_memory_target(p) (p)->func->target(p)
+#define nvkm_memory_addr(p) (p)->func->addr(p)
+#define nvkm_memory_size(p) (p)->func->size(p)
+#define nvkm_memory_boot(p,v) (p)->func->boot((p),(v))
+#define nvkm_memory_map(p,v,o) (p)->func->map((p),(v),(o))
+
+/* accessor macros - kmap()/done() must bracket use of the other accessor
+ * macros to guarantee correct behaviour across all chipsets
+ */
+#define nvkm_kmap(o)     (o)->func->acquire(o)
+#define nvkm_ro32(o,a)   (o)->func->rd32((o), (a))
+#define nvkm_wo32(o,a,d) (o)->func->wr32((o), (a), (d))
+#define nvkm_mo32(o,a,m,d) ({                                                  \
+	u32 _addr = (a), _data = nvkm_ro32((o), _addr);                        \
+	nvkm_wo32((o), _addr, (_data & ~(m)) | (d));                           \
+	_data;                                                                 \
+})
+#define nvkm_done(o)     (o)->func->release(o)
+#endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/bar.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/bar.h
index 753b7e9530351..1eaf7de79d507 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/bar.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/bar.h
@@ -7,13 +7,8 @@ struct nvkm_vma;
 struct nvkm_bar {
 	struct nvkm_subdev subdev;
 
-	int  (*alloc)(struct nvkm_bar *, struct nvkm_object *,
-		      struct nvkm_mem *, struct nvkm_object **);
-
-	int  (*kmap)(struct nvkm_bar *, struct nvkm_mem *, u32 flags,
-		     struct nvkm_vma *);
-	int  (*umap)(struct nvkm_bar *, struct nvkm_mem *, u32 flags,
-		     struct nvkm_vma *);
+	struct nvkm_vm *(*kmap)(struct nvkm_bar *);
+	int  (*umap)(struct nvkm_bar *, u64 size, int type, struct nvkm_vma *);
 	void (*unmap)(struct nvkm_bar *, struct nvkm_vma *);
 	void (*flush)(struct nvkm_bar *);
 
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/instmem.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/instmem.h
index 7ba3ff27e89ec..2e9e6f058dc08 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/instmem.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/instmem.h
@@ -1,33 +1,22 @@
 #ifndef __NVKM_INSTMEM_H__
 #define __NVKM_INSTMEM_H__
 #include <core/subdev.h>
-
-struct nvkm_instobj {
-	struct nvkm_object object;
-	struct list_head head;
-	u32 *suspend;
-	u64 addr;
-	u32 size;
-};
-
-static inline struct nvkm_instobj *
-nv_memobj(void *obj)
-{
-#if CONFIG_NOUVEAU_DEBUG >= NV_DBG_PARANOIA
-	BUG_ON(!nv_iclass(obj, NV_MEMOBJ_CLASS));
-#endif
-	return obj;
-}
+struct nvkm_memory;
 
 struct nvkm_instmem {
 	struct nvkm_subdev subdev;
 	struct list_head list;
 
 	u32 reserved;
-	int (*alloc)(struct nvkm_instmem *, struct nvkm_object *,
-		     u32 size, u32 align, struct nvkm_object **);
+	int (*alloc)(struct nvkm_instmem *, u32 size, u32 align, bool zero,
+		     struct nvkm_memory **);
 
 	const struct nvkm_instmem_func *func;
+
+	struct nvkm_gpuobj *vbios;
+	struct nvkm_ramht  *ramht;
+	struct nvkm_gpuobj *ramro;
+	struct nvkm_gpuobj *ramfc;
 };
 
 struct nvkm_instmem_func {
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/mmu.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/mmu.h
index 0991c9011dc19..677f6c0cbb9b7 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/mmu.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/mmu.h
@@ -97,6 +97,7 @@ int  nvkm_vm_create(struct nvkm_mmu *, u64 offset, u64 length, u64 mm_offset,
 int  nvkm_vm_new(struct nvkm_device *, u64 offset, u64 length, u64 mm_offset,
 		 struct lock_class_key *, struct nvkm_vm **);
 int  nvkm_vm_ref(struct nvkm_vm *, struct nvkm_vm **, struct nvkm_gpuobj *pgd);
+int  nvkm_vm_boot(struct nvkm_vm *, u64 size);
 int  nvkm_vm_get(struct nvkm_vm *, u64 size, u32 page_shift, u32 access,
 		 struct nvkm_vma *);
 void nvkm_vm_put(struct nvkm_vma *);
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 496c00d585cd2..982c0ed163ebd 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -1388,12 +1388,16 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 		mem->bus.is_iomem = true;
 		if (drm->device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
 			struct nvkm_bar *bar = nvxx_bar(&drm->device);
+			int page_shift = 12;
+			if (drm->device.info.family >= NV_DEVICE_INFO_V0_FERMI)
+				page_shift = node->page_shift;
 
-			ret = bar->umap(bar, node, NV_MEM_ACCESS_RW,
+			ret = bar->umap(bar, node->size << 12, page_shift,
 					&node->bar_vma);
 			if (ret)
 				return ret;
 
+			nvkm_vm_map(&node->bar_vma, node);
 			mem->bus.offset = node->bar_vma.offset;
 		}
 		break;
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/Kbuild b/drivers/gpu/drm/nouveau/nvkm/core/Kbuild
index c4198164bc591..98cf39f732c14 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/core/Kbuild
@@ -6,6 +6,7 @@ nvkm-y += nvkm/core/event.o
 nvkm-y += nvkm/core/gpuobj.o
 nvkm-y += nvkm/core/handle.o
 nvkm-y += nvkm/core/ioctl.o
+nvkm-y += nvkm/core/memory.o
 nvkm-y += nvkm/core/mm.o
 nvkm-y += nvkm/core/namedb.o
 nvkm-y += nvkm/core/notify.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/gpuobj.c b/drivers/gpu/drm/nouveau/nvkm/core/gpuobj.c
index bc4b3c2d075e6..0370360698395 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/gpuobj.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/gpuobj.c
@@ -28,6 +28,44 @@
 #include <subdev/bar.h>
 #include <subdev/mmu.h>
 
+static void
+nvkm_gpuobj_release(struct nvkm_gpuobj *gpuobj)
+{
+	if (gpuobj->node) {
+		nvkm_done(gpuobj->parent);
+		return;
+	}
+	nvkm_done(gpuobj->memory);
+}
+
+static void
+nvkm_gpuobj_acquire(struct nvkm_gpuobj *gpuobj)
+{
+	if (gpuobj->node) {
+		nvkm_kmap(gpuobj->parent);
+		return;
+	}
+	nvkm_kmap(gpuobj->memory);
+}
+
+static u32
+nvkm_gpuobj_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)
+{
+	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)
 {
@@ -46,17 +84,27 @@ nvkm_gpuobj_destroy(struct nvkm_gpuobj *gpuobj)
 	if (gpuobj->heap.block_size)
 		nvkm_mm_fini(&gpuobj->heap);
 
+	nvkm_memory_del(&gpuobj->memory);
 	nvkm_object_destroy(&gpuobj->object);
 }
 
+static const struct nvkm_gpuobj_func
+nvkm_gpuobj_func = {
+	.acquire = nvkm_gpuobj_acquire,
+	.release = nvkm_gpuobj_release,
+	.rd32 = nvkm_gpuobj_rd32,
+	.wr32 = nvkm_gpuobj_wr32,
+};
+
 int
 nvkm_gpuobj_create_(struct nvkm_object *parent, struct nvkm_object *engine,
 		    struct nvkm_oclass *oclass, u32 pclass,
-		    struct nvkm_object *pargpu, u32 size, u32 align, u32 flags,
+		    struct nvkm_object *objgpu, u32 size, u32 align, u32 flags,
 		    int length, void **pobject)
 {
-	struct nvkm_instmem *imem = nvkm_instmem(parent);
-	struct nvkm_bar *bar = nvkm_bar(parent);
+	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;
@@ -64,46 +112,39 @@ nvkm_gpuobj_create_(struct nvkm_object *parent, struct nvkm_object *engine,
 
 	*pobject = NULL;
 
-	if (pargpu) {
-		while ((pargpu = nv_pclass(pargpu, NV_GPUOBJ_CLASS))) {
-			if (nv_gpuobj(pargpu)->heap.block_size)
+	if (objgpu) {
+		while ((objgpu = nv_pclass(objgpu, NV_GPUOBJ_CLASS))) {
+			if (nv_gpuobj(objgpu)->heap.block_size)
 				break;
-			pargpu = pargpu->parent;
+			objgpu = objgpu->parent;
 		}
 
-		if (WARN_ON(pargpu == NULL))
+		if (WARN_ON(objgpu == NULL))
 			return -EINVAL;
+		pargpu = nv_gpuobj(objgpu);
 
-		addr =  nv_gpuobj(pargpu)->addr;
-		heap = &nv_gpuobj(pargpu)->heap;
-		atomic_inc(&parent->refcount);
+		addr =  pargpu->addr;
+		heap = &pargpu->heap;
 	} else {
-		ret = imem->alloc(imem, parent, size, align, &parent);
-		pargpu = parent;
+		ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST,
+				      size, align, false, &memory);
 		if (ret)
 			return ret;
 
-		addr = nv_memobj(pargpu)->addr;
-		size = nv_memobj(pargpu)->size;
-
-		if (bar && bar->alloc) {
-			struct nvkm_instobj *iobj = (void *)parent;
-			struct nvkm_mem **mem = (void *)(iobj + 1);
-			struct nvkm_mem *node = *mem;
-			if (!bar->alloc(bar, parent, node, &pargpu)) {
-				nvkm_object_ref(NULL, &parent);
-				parent = pargpu;
-			}
-		}
+		addr = nvkm_memory_addr(memory);
+		size = nvkm_memory_size(memory);
 	}
 
 	ret = nvkm_object_create_(parent, engine, oclass, pclass |
 				  NV_GPUOBJ_CLASS, length, pobject);
-	nvkm_object_ref(NULL, &parent);
 	gpuobj = *pobject;
-	if (ret)
+	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;
@@ -182,20 +223,14 @@ u32
 _nvkm_gpuobj_rd32(struct nvkm_object *object, u64 addr)
 {
 	struct nvkm_gpuobj *gpuobj = nv_gpuobj(object);
-	u32 data;
-	if (gpuobj->node)
-		addr += gpuobj->node->offset;
-	nvkm_object_rd32(gpuobj->parent, addr, &data);
-	return data;
+	return nvkm_ro32(gpuobj, addr);
 }
 
 void
 _nvkm_gpuobj_wr32(struct nvkm_object *object, u64 addr, u32 data)
 {
 	struct nvkm_gpuobj *gpuobj = nv_gpuobj(object);
-	if (gpuobj->node)
-		addr += gpuobj->node->offset;
-	nvkm_object_wr32(gpuobj->parent, addr, data);
+	nvkm_wo32(gpuobj, addr, data);
 }
 
 static struct nvkm_oclass
@@ -231,14 +266,14 @@ nvkm_gpuobj_new(struct nvkm_object *parent, struct nvkm_object *pargpu,
 int
 nvkm_gpuobj_map(struct nvkm_gpuobj *gpuobj, u32 access, struct nvkm_vma *vma)
 {
+	struct nvkm_memory *memory = gpuobj->memory;
 	struct nvkm_bar *bar = nvkm_bar(gpuobj);
 	int ret = -EINVAL;
 
 	if (bar && bar->umap) {
-		struct nvkm_instobj *iobj = (void *)
-			nv_pclass(nv_object(gpuobj), NV_MEMOBJ_CLASS);
-		struct nvkm_mem **mem = (void *)(iobj + 1);
-		ret = bar->umap(bar, *mem, access, vma);
+		ret = bar->umap(bar, gpuobj->size, 12, vma);
+		if (ret == 0)
+			nvkm_memory_map(memory, vma, 0);
 	}
 
 	return ret;
@@ -248,17 +283,11 @@ int
 nvkm_gpuobj_map_vm(struct nvkm_gpuobj *gpuobj, struct nvkm_vm *vm,
 		   u32 access, struct nvkm_vma *vma)
 {
-	struct nvkm_instobj *iobj = (void *)
-		nv_pclass(nv_object(gpuobj), NV_MEMOBJ_CLASS);
-	struct nvkm_mem **mem = (void *)(iobj + 1);
-	int ret;
-
-	ret = nvkm_vm_get(vm, gpuobj->size, 12, access, vma);
-	if (ret)
-		return ret;
-
-	nvkm_vm_map(vma, *mem);
-	return 0;
+	struct nvkm_memory *memory = gpuobj->memory;
+	int ret = nvkm_vm_get(vm, gpuobj->size, 12, access, vma);
+	if (ret == 0)
+		nvkm_memory_map(memory, vma, 0);
+	return ret;
 }
 
 void
@@ -279,7 +308,7 @@ static void
 nvkm_gpudup_dtor(struct nvkm_object *object)
 {
 	struct nvkm_gpuobj *gpuobj = (void *)object;
-	nvkm_object_ref(NULL, &gpuobj->parent);
+	nvkm_object_ref(NULL, (struct nvkm_object **)&gpuobj->parent);
 	nvkm_object_destroy(&gpuobj->object);
 }
 
@@ -306,7 +335,7 @@ nvkm_gpuobj_dup(struct nvkm_object *parent, struct nvkm_gpuobj *base,
 	if (ret)
 		return ret;
 
-	nvkm_object_ref(nv_object(base), &gpuobj->parent);
+	nvkm_object_ref(nv_object(base), (struct nvkm_object **)&gpuobj->parent);
 	gpuobj->addr = base->addr;
 	gpuobj->size = base->size;
 	return 0;
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/memory.c b/drivers/gpu/drm/nouveau/nvkm/core/memory.c
new file mode 100644
index 0000000000000..0b88faa845f37
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/core/memory.c
@@ -0,0 +1,64 @@
+/*
+ * Copyright 2015 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include <core/memory.h>
+#include <subdev/instmem.h>
+
+void
+nvkm_memory_ctor(const struct nvkm_memory_func *func,
+		 struct nvkm_memory *memory)
+{
+	memory->func = func;
+}
+
+void
+nvkm_memory_del(struct nvkm_memory **pmemory)
+{
+	struct nvkm_memory *memory = *pmemory;
+	if (memory && !WARN_ON(!memory->func)) {
+		if (memory->func->dtor)
+			*pmemory = memory->func->dtor(memory);
+		kfree(*pmemory);
+		*pmemory = NULL;
+	}
+}
+
+int
+nvkm_memory_new(struct nvkm_device *device, enum nvkm_memory_target target,
+		u64 size, u32 align, bool zero,
+		struct nvkm_memory **pmemory)
+{
+	struct nvkm_instmem *imem = device->imem;
+	struct nvkm_memory *memory;
+	int ret = -ENOSYS;
+
+	if (unlikely(target != NVKM_MEM_TARGET_INST || !imem))
+		return -ENOSYS;
+
+	ret = imem->alloc(imem, size, align, zero, &memory);
+	if (ret)
+		return ret;
+
+	*pmemory = memory;
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/ramht.c b/drivers/gpu/drm/nouveau/nvkm/core/ramht.c
index 4717af0800e91..061adedc6e2cf 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/ramht.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/ramht.c
@@ -22,8 +22,6 @@
 #include <core/ramht.h>
 #include <core/engine.h>
 
-#include <subdev/bar.h>
-
 static u32
 nvkm_ramht_hash(struct nvkm_ramht *ramht, int chid, u32 handle)
 {
@@ -43,7 +41,6 @@ int
 nvkm_ramht_insert(struct nvkm_ramht *ramht, int chid, u32 handle, u32 context)
 {
 	struct nvkm_gpuobj *gpuobj = &ramht->gpuobj;
-	struct nvkm_bar *bar = nvkm_bar(ramht);
 	int ret = -ENOSPC;
 	u32 co, ho;
 
@@ -53,8 +50,6 @@ nvkm_ramht_insert(struct nvkm_ramht *ramht, int chid, u32 handle, u32 context)
 		if (!nvkm_ro32(gpuobj, co + 4)) {
 			nvkm_wo32(gpuobj, co + 0, handle);
 			nvkm_wo32(gpuobj, co + 4, context);
-			if (bar)
-				bar->flush(bar);
 			ret = co;
 			break;
 		}
@@ -72,12 +67,9 @@ void
 nvkm_ramht_remove(struct nvkm_ramht *ramht, int cookie)
 {
 	struct nvkm_gpuobj *gpuobj = &ramht->gpuobj;
-	struct nvkm_bar *bar = nvkm_bar(ramht);
 	nvkm_kmap(gpuobj);
 	nvkm_wo32(gpuobj, cookie + 0, 0x00000000);
 	nvkm_wo32(gpuobj, cookie + 4, 0x00000000);
-	if (bar)
-		bar->flush(bar);
 	nvkm_done(gpuobj);
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c
index 64523d7eea589..c6150eaf16f08 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c
@@ -27,7 +27,6 @@
 #include <core/client.h>
 #include <core/engctx.h>
 #include <core/ramht.h>
-#include <subdev/bar.h>
 #include <subdev/mmu.h>
 #include <subdev/timer.h>
 
@@ -41,7 +40,6 @@
 static int
 g84_fifo_context_attach(struct nvkm_object *parent, struct nvkm_object *object)
 {
-	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct nv50_fifo_base *base = (void *)parent->parent;
 	struct nvkm_gpuobj *ectx = (void *)object;
 	u64 limit = ectx->addr + ectx->size - 1;
@@ -73,7 +71,6 @@ g84_fifo_context_attach(struct nvkm_object *parent, struct nvkm_object *object)
 					  upper_32_bits(start));
 	nvkm_wo32(base->eng, addr + 0x10, 0x00000000);
 	nvkm_wo32(base->eng, addr + 0x14, 0x00000000);
-	bar->flush(bar);
 	nvkm_done(base->eng);
 	return 0;
 }
@@ -87,7 +84,6 @@ g84_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 	struct nv50_fifo_chan *chan = (void *)parent;
 	struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
 	struct nvkm_device *device = subdev->device;
-	struct nvkm_bar *bar = device->bar;
 	u32 addr, save, engn;
 	bool done;
 
@@ -128,7 +124,6 @@ g84_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 	nvkm_wo32(base->eng, addr + 0x0c, 0x00000000);
 	nvkm_wo32(base->eng, addr + 0x10, 0x00000000);
 	nvkm_wo32(base->eng, addr + 0x14, 0x00000000);
-	bar->flush(bar);
 	nvkm_done(base->eng);
 	return 0;
 }
@@ -175,7 +170,6 @@ g84_fifo_chan_ctor_dma(struct nvkm_object *parent, struct nvkm_object *engine,
 	union {
 		struct nv50_channel_dma_v0 v0;
 	} *args = data;
-	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct nv50_fifo_base *base = (void *)parent;
 	struct nv50_fifo_chan *chan;
 	int ret;
@@ -239,7 +233,6 @@ g84_fifo_chan_ctor_dma(struct nvkm_object *parent, struct nvkm_object *engine,
 				     (chan->ramht->gpuobj.node->offset >> 4));
 	nvkm_wo32(base->ramfc, 0x88, base->cache->addr >> 10);
 	nvkm_wo32(base->ramfc, 0x98, nv_gpuobj(base)->addr >> 12);
-	bar->flush(bar);
 	nvkm_done(base->ramfc);
 	return 0;
 }
@@ -252,7 +245,6 @@ g84_fifo_chan_ctor_ind(struct nvkm_object *parent, struct nvkm_object *engine,
 	union {
 		struct nv50_channel_gpfifo_v0 v0;
 	} *args = data;
-	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct nv50_fifo_base *base = (void *)parent;
 	struct nv50_fifo_chan *chan;
 	u64 ioffset, ilength;
@@ -318,7 +310,6 @@ g84_fifo_chan_ctor_ind(struct nvkm_object *parent, struct nvkm_object *engine,
 				     (chan->ramht->gpuobj.node->offset >> 4));
 	nvkm_wo32(base->ramfc, 0x88, base->cache->addr >> 10);
 	nvkm_wo32(base->ramfc, 0x98, nv_gpuobj(base)->addr >> 12);
-	bar->flush(bar);
 	nvkm_done(base->ramfc);
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
index ba6b390a1fef4..763a2db7603a5 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
@@ -27,7 +27,6 @@
 #include <core/engctx.h>
 #include <core/enum.h>
 #include <core/handle.h>
-#include <subdev/bar.h>
 #include <subdev/fb.h>
 #include <subdev/mmu.h>
 #include <subdev/timer.h>
@@ -79,7 +78,6 @@ gf100_fifo_runlist_update(struct gf100_fifo *fifo)
 {
 	struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
 	struct nvkm_device *device = subdev->device;
-	struct nvkm_bar *bar = device->bar;
 	struct nvkm_gpuobj *cur;
 	int i, p;
 
@@ -96,7 +94,6 @@ gf100_fifo_runlist_update(struct gf100_fifo *fifo)
 			p += 8;
 		}
 	}
-	bar->flush(bar);
 	nvkm_done(cur);
 
 	nvkm_wr32(device, 0x002270, cur->addr >> 12);
@@ -113,7 +110,6 @@ static int
 gf100_fifo_context_attach(struct nvkm_object *parent,
 			  struct nvkm_object *object)
 {
-	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct gf100_fifo_base *base = (void *)parent->parent;
 	struct nvkm_gpuobj *engn = &base->base.gpuobj;
 	struct nvkm_engctx *ectx = (void *)object;
@@ -144,7 +140,6 @@ gf100_fifo_context_attach(struct nvkm_object *parent,
 	nvkm_kmap(engn);
 	nvkm_wo32(engn, addr + 0x00, lower_32_bits(ectx->vma.offset) | 4);
 	nvkm_wo32(engn, addr + 0x04, upper_32_bits(ectx->vma.offset));
-	bar->flush(bar);
 	nvkm_done(engn);
 	return 0;
 }
@@ -159,7 +154,6 @@ gf100_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 	struct nvkm_gpuobj *engn = &base->base.gpuobj;
 	struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
 	struct nvkm_device *device = subdev->device;
-	struct nvkm_bar *bar = device->bar;
 	u32 addr;
 
 	switch (nv_engidx(object->engine)) {
@@ -188,7 +182,6 @@ gf100_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 	nvkm_kmap(engn);
 	nvkm_wo32(engn, addr + 0x00, 0x00000000);
 	nvkm_wo32(engn, addr + 0x04, 0x00000000);
-	bar->flush(bar);
 	nvkm_done(engn);
 	return 0;
 }
@@ -201,7 +194,6 @@ gf100_fifo_chan_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	union {
 		struct fermi_channel_gpfifo_v0 v0;
 	} *args = data;
-	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct gf100_fifo *fifo = (void *)engine;
 	struct gf100_fifo_base *base = (void *)parent;
 	struct gf100_fifo_chan *chan;
@@ -264,7 +256,6 @@ gf100_fifo_chan_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	nvkm_wo32(ramfc, 0xb8, 0xf8000000);
 	nvkm_wo32(ramfc, 0xf8, 0x10003080); /* 0x002310 */
 	nvkm_wo32(ramfc, 0xfc, 0x10000010); /* 0x002350 */
-	bar->flush(bar);
 	nvkm_done(ramfc);
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
index 62b3de4e9353d..d1238aa7bec55 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
@@ -27,7 +27,6 @@
 #include <core/engctx.h>
 #include <core/enum.h>
 #include <core/handle.h>
-#include <subdev/bar.h>
 #include <subdev/fb.h>
 #include <subdev/mmu.h>
 #include <subdev/timer.h>
@@ -99,7 +98,6 @@ gk104_fifo_runlist_update(struct gk104_fifo *fifo, u32 engine)
 	struct gk104_fifo_engn *engn = &fifo->engine[engine];
 	struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
 	struct nvkm_device *device = subdev->device;
-	struct nvkm_bar *bar = device->bar;
 	struct nvkm_gpuobj *cur;
 	int i, p;
 
@@ -116,7 +114,6 @@ gk104_fifo_runlist_update(struct gk104_fifo *fifo, u32 engine)
 			p += 8;
 		}
 	}
-	bar->flush(bar);
 	nvkm_done(cur);
 
 	nvkm_wr32(device, 0x002270, cur->addr >> 12);
@@ -133,7 +130,6 @@ static int
 gk104_fifo_context_attach(struct nvkm_object *parent,
 			  struct nvkm_object *object)
 {
-	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct gk104_fifo_base *base = (void *)parent->parent;
 	struct nvkm_gpuobj *engn = &base->base.gpuobj;
 	struct nvkm_engctx *ectx = (void *)object;
@@ -168,7 +164,6 @@ gk104_fifo_context_attach(struct nvkm_object *parent,
 	nvkm_kmap(engn);
 	nvkm_wo32(engn, addr + 0x00, lower_32_bits(ectx->vma.offset) | 4);
 	nvkm_wo32(engn, addr + 0x04, upper_32_bits(ectx->vma.offset));
-	bar->flush(bar);
 	nvkm_done(engn);
 	return 0;
 }
@@ -198,7 +193,6 @@ static int
 gk104_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 			  struct nvkm_object *object)
 {
-	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct gk104_fifo_base *base = (void *)parent->parent;
 	struct gk104_fifo_chan *chan = (void *)parent;
 	struct nvkm_gpuobj *engn = &base->base.gpuobj;
@@ -226,7 +220,6 @@ gk104_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 		nvkm_kmap(engn);
 		nvkm_wo32(engn, addr + 0x00, 0x00000000);
 		nvkm_wo32(engn, addr + 0x04, 0x00000000);
-		bar->flush(bar);
 		nvkm_done(engn);
 	}
 
@@ -241,7 +234,6 @@ gk104_fifo_chan_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	union {
 		struct kepler_channel_gpfifo_a_v0 v0;
 	} *args = data;
-	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct gk104_fifo *fifo = (void *)engine;
 	struct gk104_fifo_base *base = (void *)parent;
 	struct gk104_fifo_chan *chan;
@@ -320,7 +312,6 @@ gk104_fifo_chan_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	nvkm_wo32(ramfc, 0xb8, 0xf8000000);
 	nvkm_wo32(ramfc, 0xf8, 0x10003080); /* 0x002310 */
 	nvkm_wo32(ramfc, 0xfc, 0x10000010); /* 0x002350 */
-	bar->flush(bar);
 	nvkm_done(ramfc);
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c
index 51b8fce9e0fc1..b01490f71f098 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c
@@ -27,7 +27,7 @@
 #include <core/engctx.h>
 #include <core/handle.h>
 #include <core/ramht.h>
-#include <subdev/instmem/nv04.h>
+#include <subdev/instmem.h>
 #include <subdev/timer.h>
 
 #include <nvif/class.h>
@@ -574,7 +574,8 @@ nv04_fifo_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	       struct nvkm_oclass *oclass, void *data, u32 size,
 	       struct nvkm_object **pobject)
 {
-	struct nv04_instmem *imem = nv04_instmem(parent);
+	struct nvkm_device *device = (void *)parent;
+	struct nvkm_instmem *imem = device->imem;
 	struct nv04_fifo *fifo;
 	int ret;
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c
index 046acc2813817..7491b10a76b7d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c
@@ -26,7 +26,7 @@
 #include <core/client.h>
 #include <core/engctx.h>
 #include <core/ramht.h>
-#include <subdev/instmem/nv04.h>
+#include <subdev/instmem.h>
 
 #include <nvif/class.h>
 #include <nvif/unpack.h>
@@ -145,7 +145,8 @@ nv10_fifo_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	       struct nvkm_oclass *oclass, void *data, u32 size,
 	       struct nvkm_object **pobject)
 {
-	struct nv04_instmem *imem = nv04_instmem(parent);
+	struct nvkm_device *device = (void *)parent;
+	struct nvkm_instmem *imem = device->imem;
 	struct nv04_fifo *fifo;
 	int ret;
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c
index 7be6fea900100..e652941d04f3b 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c
@@ -26,7 +26,7 @@
 #include <core/client.h>
 #include <core/engctx.h>
 #include <core/ramht.h>
-#include <subdev/instmem/nv04.h>
+#include <subdev/instmem.h>
 
 #include <nvif/class.h>
 #include <nvif/unpack.h>
@@ -152,7 +152,8 @@ nv17_fifo_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	       struct nvkm_oclass *oclass, void *data, u32 size,
 	       struct nvkm_object **pobject)
 {
-	struct nv04_instmem *imem = nv04_instmem(parent);
+	struct nvkm_device *device = (void *)parent;
+	struct nvkm_instmem *imem = device->imem;
 	struct nv04_fifo *fifo;
 	int ret;
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c
index cad4b4be19387..f35ae76472392 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c
@@ -27,7 +27,7 @@
 #include <core/engctx.h>
 #include <core/ramht.h>
 #include <subdev/fb.h>
-#include <subdev/instmem/nv04.h>
+#include <subdev/instmem.h>
 
 #include <nvif/class.h>
 #include <nvif/unpack.h>
@@ -276,7 +276,8 @@ nv40_fifo_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	       struct nvkm_oclass *oclass, void *data, u32 size,
 	       struct nvkm_object **pobject)
 {
-	struct nv04_instmem *imem = nv04_instmem(parent);
+	struct nvkm_device *device = (void *)parent;
+	struct nvkm_instmem *imem = device->imem;
 	struct nv04_fifo *fifo;
 	int ret;
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c
index 402639cb4ec9e..ffa495c1692bb 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c
@@ -27,7 +27,6 @@
 #include <core/client.h>
 #include <core/engctx.h>
 #include <core/ramht.h>
-#include <subdev/bar.h>
 #include <subdev/mmu.h>
 #include <subdev/timer.h>
 
@@ -42,7 +41,6 @@ static void
 nv50_fifo_playlist_update_locked(struct nv50_fifo *fifo)
 {
 	struct nvkm_device *device = fifo->base.engine.subdev.device;
-	struct nvkm_bar *bar = device->bar;
 	struct nvkm_gpuobj *cur;
 	int i, p;
 
@@ -54,7 +52,6 @@ nv50_fifo_playlist_update_locked(struct nv50_fifo *fifo)
 		if (nvkm_rd32(device, 0x002600 + (i * 4)) & 0x80000000)
 			nvkm_wo32(cur, p++ * 4, i);
 	}
-	bar->flush(bar);
 	nvkm_done(cur);
 
 	nvkm_wr32(device, 0x0032f4, cur->addr >> 12);
@@ -73,7 +70,6 @@ nv50_fifo_playlist_update(struct nv50_fifo *fifo)
 static int
 nv50_fifo_context_attach(struct nvkm_object *parent, struct nvkm_object *object)
 {
-	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct nv50_fifo_base *base = (void *)parent->parent;
 	struct nvkm_gpuobj *ectx = (void *)object;
 	u64 limit = ectx->addr + ectx->size - 1;
@@ -98,7 +94,6 @@ nv50_fifo_context_attach(struct nvkm_object *parent, struct nvkm_object *object)
 					  upper_32_bits(start));
 	nvkm_wo32(base->eng, addr + 0x10, 0x00000000);
 	nvkm_wo32(base->eng, addr + 0x14, 0x00000000);
-	bar->flush(bar);
 	nvkm_done(base->eng);
 	return 0;
 }
@@ -112,7 +107,6 @@ nv50_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 	struct nv50_fifo_chan *chan = (void *)parent;
 	struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
 	struct nvkm_device *device = subdev->device;
-	struct nvkm_bar *bar = device->bar;
 	u32 addr, me;
 	int ret = 0;
 
@@ -159,7 +153,6 @@ nv50_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 		nvkm_wo32(base->eng, addr + 0x0c, 0x00000000);
 		nvkm_wo32(base->eng, addr + 0x10, 0x00000000);
 		nvkm_wo32(base->eng, addr + 0x14, 0x00000000);
-		bar->flush(bar);
 		nvkm_done(base->eng);
 	}
 
@@ -205,7 +198,6 @@ nv50_fifo_chan_ctor_dma(struct nvkm_object *parent, struct nvkm_object *engine,
 	union {
 		struct nv50_channel_dma_v0 v0;
 	} *args = data;
-	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct nv50_fifo_base *base = (void *)parent;
 	struct nv50_fifo_chan *chan;
 	int ret;
@@ -257,7 +249,6 @@ nv50_fifo_chan_ctor_dma(struct nvkm_object *parent, struct nvkm_object *engine,
 	nvkm_wo32(base->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
 				     (4 << 24) /* SEARCH_FULL */ |
 				     (chan->ramht->gpuobj.node->offset >> 4));
-	bar->flush(bar);
 	nvkm_done(base->ramfc);
 	return 0;
 }
@@ -270,7 +261,6 @@ nv50_fifo_chan_ctor_ind(struct nvkm_object *parent, struct nvkm_object *engine,
 	union {
 		struct nv50_channel_gpfifo_v0 v0;
 	} *args = data;
-	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct nv50_fifo_base *base = (void *)parent;
 	struct nv50_fifo_chan *chan;
 	u64 ioffset, ilength;
@@ -324,7 +314,6 @@ nv50_fifo_chan_ctor_ind(struct nvkm_object *parent, struct nvkm_object *engine,
 	nvkm_wo32(base->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
 				     (4 << 24) /* SEARCH_FULL */ |
 				     (chan->ramht->gpuobj.node->offset >> 4));
-	bar->flush(bar);
 	nvkm_done(base->ramfc);
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf100.c
index f36e0896ae9c6..a2e60b1ddb27c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf100.c
@@ -23,7 +23,6 @@
  */
 #include "ctxgf100.h"
 
-#include <subdev/bar.h>
 #include <subdev/fb.h>
 #include <subdev/mc.h>
 #include <subdev/timer.h>
@@ -1273,7 +1272,6 @@ gf100_grctx_generate(struct gf100_gr *gr)
 	struct gf100_grctx_oclass *oclass = (void *)nv_engine(gr)->cclass;
 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
 	struct nvkm_device *device = subdev->device;
-	struct nvkm_bar *bar = device->bar;
 	struct nvkm_gpuobj *chan;
 	struct gf100_grctx info;
 	int ret, i;
@@ -1309,7 +1307,6 @@ gf100_grctx_generate(struct gf100_gr *gr)
 	/* context pointer (virt) */
 	nvkm_wo32(chan, 0x0210, 0x00080004);
 	nvkm_wo32(chan, 0x0214, 0x00000000);
-	bar->flush(bar);
 	nvkm_done(chan);
 
 	nvkm_wr32(device, 0x100cb8, (chan->addr + 0x1000) >> 8);
@@ -1341,7 +1338,6 @@ gf100_grctx_generate(struct gf100_gr *gr)
 		nvkm_wo32(chan, 0x80020, 0);
 		nvkm_wo32(chan, 0x80028, 0);
 		nvkm_wo32(chan, 0x8002c, 0);
-		bar->flush(bar);
 		nvkm_done(chan);
 	} else {
 		nvkm_wr32(device, 0x409840, 0x80000000);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/mpeg/nv50.c b/drivers/gpu/drm/nouveau/nvkm/engine/mpeg/nv50.c
index 6af707b5be189..c906c43e9065c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/mpeg/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/mpeg/nv50.c
@@ -23,7 +23,6 @@
  */
 #include <engine/mpeg.h>
 
-#include <subdev/bar.h>
 #include <subdev/timer.h>
 
 struct nv50_mpeg_chan {
@@ -84,7 +83,6 @@ nv50_mpeg_context_ctor(struct nvkm_object *parent,
 		       struct nvkm_oclass *oclass, void *data, u32 size,
 		       struct nvkm_object **pobject)
 {
-	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct nv50_mpeg_chan *chan;
 	struct nvkm_gpuobj *image;
 	int ret;
@@ -100,7 +98,6 @@ nv50_mpeg_context_ctor(struct nvkm_object *parent,
 	nvkm_kmap(image);
 	nvkm_wo32(image, 0x0070, 0x00801ec1);
 	nvkm_wo32(image, 0x007c, 0x0000037c);
-	bar->flush(bar);
 	nvkm_done(image);
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bar/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bar/base.c
index 6fe71b2276cca..8daaa65fc8cf5 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bar/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bar/base.c
@@ -23,96 +23,6 @@
  */
 #include "priv.h"
 
-#include <subdev/fb.h>
-#include <subdev/mmu.h>
-
-struct nvkm_barobj {
-	struct nvkm_object base;
-	struct nvkm_vma vma;
-	void __iomem *iomem;
-};
-
-static int
-nvkm_barobj_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_device(parent);
-	struct nvkm_bar *bar = nvkm_bar(device);
-	struct nvkm_mem *mem = data;
-	struct nvkm_barobj *barobj;
-	int ret;
-
-	ret = nvkm_object_create(parent, engine, oclass, 0, &barobj);
-	*pobject = nv_object(barobj);
-	if (ret)
-		return ret;
-
-	ret = bar->kmap(bar, mem, NV_MEM_ACCESS_RW, &barobj->vma);
-	if (ret)
-		return ret;
-
-	barobj->iomem = ioremap(nv_device_resource_start(device, 3) +
-				(u32)barobj->vma.offset, mem->size << 12);
-	if (!barobj->iomem) {
-		nvkm_warn(&bar->subdev, "PRAMIN ioremap failed\n");
-		return -ENOMEM;
-	}
-
-	return 0;
-}
-
-static void
-nvkm_barobj_dtor(struct nvkm_object *object)
-{
-	struct nvkm_bar *bar = nvkm_bar(object);
-	struct nvkm_barobj *barobj = (void *)object;
-	if (barobj->vma.node) {
-		if (barobj->iomem)
-			iounmap(barobj->iomem);
-		bar->unmap(bar, &barobj->vma);
-	}
-	nvkm_object_destroy(&barobj->base);
-}
-
-static u32
-nvkm_barobj_rd32(struct nvkm_object *object, u64 addr)
-{
-	struct nvkm_barobj *barobj = (void *)object;
-	return ioread32_native(barobj->iomem + addr);
-}
-
-static void
-nvkm_barobj_wr32(struct nvkm_object *object, u64 addr, u32 data)
-{
-	struct nvkm_barobj *barobj = (void *)object;
-	iowrite32_native(data, barobj->iomem + addr);
-}
-
-static struct nvkm_oclass
-nvkm_barobj_oclass = {
-	.ofuncs = &(struct nvkm_ofuncs) {
-		.ctor = nvkm_barobj_ctor,
-		.dtor = nvkm_barobj_dtor,
-		.init = _nvkm_object_init,
-		.fini = _nvkm_object_fini,
-		.rd32 = nvkm_barobj_rd32,
-		.wr32 = nvkm_barobj_wr32,
-	},
-};
-
-int
-nvkm_bar_alloc(struct nvkm_bar *bar, struct nvkm_object *parent,
-	       struct nvkm_mem *mem, struct nvkm_object **pobject)
-{
-	struct nvkm_object *gpuobj;
-	int ret = nvkm_object_old(parent, &parent->engine->subdev.object,
-				   &nvkm_barobj_oclass, mem, 0, &gpuobj);
-	if (ret == 0)
-		*pobject = gpuobj;
-	return ret;
-}
-
 int
 nvkm_bar_create_(struct nvkm_object *parent, struct nvkm_object *engine,
 		 struct nvkm_oclass *oclass, int length, void **pobject)
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bar/gf100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bar/gf100.c
index 01e26213fd88c..1ea6b3909c3a5 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bar/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bar/gf100.c
@@ -39,35 +39,18 @@ struct gf100_bar {
 	struct gf100_bar_vm bar[2];
 };
 
-static int
-gf100_bar_kmap(struct nvkm_bar *obj, struct nvkm_mem *mem, u32 flags,
-	       struct nvkm_vma *vma)
+static struct nvkm_vm *
+gf100_bar_kmap(struct nvkm_bar *obj)
 {
 	struct gf100_bar *bar = container_of(obj, typeof(*bar), base);
-	int ret;
-
-	ret = nvkm_vm_get(bar->bar[0].vm, mem->size << 12, 12, flags, vma);
-	if (ret)
-		return ret;
-
-	nvkm_vm_map(vma, mem);
-	return 0;
+	return bar->bar[0].vm;
 }
 
 static int
-gf100_bar_umap(struct nvkm_bar *obj, struct nvkm_mem *mem, u32 flags,
-	       struct nvkm_vma *vma)
+gf100_bar_umap(struct nvkm_bar *obj, u64 size, int type, struct nvkm_vma *vma)
 {
 	struct gf100_bar *bar = container_of(obj, typeof(*bar), base);
-	int ret;
-
-	ret = nvkm_vm_get(bar->bar[1].vm, mem->size << 12,
-			  mem->page_shift, flags, vma);
-	if (ret)
-		return ret;
-
-	nvkm_vm_map(vma, mem);
-	return 0;
+	return nvkm_vm_get(bar->bar[1].vm, size, type, NV_MEM_ACCESS_RW, vma);
 }
 
 static void
@@ -109,11 +92,7 @@ gf100_bar_ctor_vm(struct gf100_bar *bar, struct gf100_bar_vm *bar_vm,
 	 * Bootstrap page table lookup.
 	 */
 	if (bar_nr == 3) {
-		ret = nvkm_gpuobj_new(nv_object(bar), NULL,
-				      (bar_len >> 12) * 8, 0x1000,
-				      NVOBJ_FLAG_ZERO_ALLOC,
-				      &vm->pgt[0].obj[0]);
-		vm->pgt[0].refcount[0] = 1;
+		ret = nvkm_vm_boot(vm, bar_len);
 		if (ret)
 			return ret;
 	}
@@ -149,6 +128,10 @@ gf100_bar_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	if (ret)
 		return ret;
 
+	device->bar = &bar->base;
+	bar->base.flush = g84_bar_flush;
+	spin_lock_init(&bar->lock);
+
 	/* BAR3 */
 	if (has_bar3) {
 		ret = gf100_bar_ctor_vm(bar, &bar->bar[0], &bar3_lock, 3);
@@ -161,14 +144,10 @@ gf100_bar_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	if (ret)
 		return ret;
 
-	if (has_bar3) {
-		bar->base.alloc = nvkm_bar_alloc;
+	if (has_bar3)
 		bar->base.kmap = gf100_bar_kmap;
-	}
 	bar->base.umap = gf100_bar_umap;
 	bar->base.unmap = gf100_bar_unmap;
-	bar->base.flush = g84_bar_flush;
-	spin_lock_init(&bar->lock);
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bar/nv50.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bar/nv50.c
index cb58cc5b2b579..ec864afc8862c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bar/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bar/nv50.c
@@ -40,34 +40,18 @@ struct nv50_bar {
 	struct nvkm_gpuobj *bar3;
 };
 
-static int
-nv50_bar_kmap(struct nvkm_bar *obj, struct nvkm_mem *mem, u32 flags,
-	      struct nvkm_vma *vma)
+static struct nvkm_vm *
+nv50_bar_kmap(struct nvkm_bar *obj)
 {
 	struct nv50_bar *bar = container_of(obj, typeof(*bar), base);
-	int ret;
-
-	ret = nvkm_vm_get(bar->bar3_vm, mem->size << 12, 12, flags, vma);
-	if (ret)
-		return ret;
-
-	nvkm_vm_map(vma, mem);
-	return 0;
+	return bar->bar3_vm;
 }
 
 static int
-nv50_bar_umap(struct nvkm_bar *obj, struct nvkm_mem *mem, u32 flags,
-	      struct nvkm_vma *vma)
+nv50_bar_umap(struct nvkm_bar *obj, u64 size, int type, struct nvkm_vma *vma)
 {
 	struct nv50_bar *bar = container_of(obj, typeof(*bar), base);
-	int ret;
-
-	ret = nvkm_vm_get(bar->bar1_vm, mem->size << 12, 12, flags, vma);
-	if (ret)
-		return ret;
-
-	nvkm_vm_map(vma, mem);
-	return 0;
+	return nvkm_vm_get(bar->bar1_vm, size, type, NV_MEM_ACCESS_RW, vma);
 }
 
 static void
@@ -152,10 +136,7 @@ nv50_bar_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 
 	atomic_inc(&vm->engref[NVDEV_SUBDEV_BAR]);
 
-	ret = nvkm_gpuobj_new(nv_object(bar), heap,
-			      ((limit-- - start) >> 12) * 8, 0x1000,
-			      NVOBJ_FLAG_ZERO_ALLOC, &vm->pgt[0].obj[0]);
-	vm->pgt[0].refcount[0] = 1;
+	ret = nvkm_vm_boot(vm, limit-- - start);
 	if (ret)
 		return ret;
 
@@ -207,7 +188,6 @@ nv50_bar_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	nvkm_wo32(bar->bar1, 0x14, 0x00000000);
 	nvkm_done(bar->bar1);
 
-	bar->base.alloc = nvkm_bar_alloc;
 	bar->base.kmap = nv50_bar_kmap;
 	bar->base.umap = nv50_bar_umap;
 	bar->base.unmap = nv50_bar_unmap;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bar/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/bar/priv.h
index e9d2f7b6b22f8..3a8fbaea582d9 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bar/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bar/priv.h
@@ -17,9 +17,6 @@ void _nvkm_bar_dtor(struct nvkm_object *);
 #define _nvkm_bar_init _nvkm_subdev_init
 #define _nvkm_bar_fini _nvkm_subdev_fini
 
-int  nvkm_bar_alloc(struct nvkm_bar *, struct nvkm_object *,
-		    struct nvkm_mem *, struct nvkm_object **);
-
 void g84_bar_flush(struct nvkm_bar *);
 
 int gf100_bar_ctor(struct nvkm_object *, struct nvkm_object *,
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c
index 8ac8e4f4fa4b5..2a1dab304087c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c
@@ -23,83 +23,221 @@
  */
 #include "priv.h"
 
-#include <core/engine.h>
+#include <core/memory.h>
+#include <subdev/bar.h>
 
 /******************************************************************************
  * instmem object base implementation
  *****************************************************************************/
+#define nvkm_instobj(p) container_of((p), struct nvkm_instobj, memory)
 
-void
-_nvkm_instobj_dtor(struct nvkm_object *object)
+struct nvkm_instobj {
+	struct nvkm_memory memory;
+	struct nvkm_memory *parent;
+	struct nvkm_instmem *imem;
+	struct list_head head;
+	u32 *suspend;
+	void __iomem *map;
+};
+
+static enum nvkm_memory_target
+nvkm_instobj_target(struct nvkm_memory *memory)
+{
+	memory = nvkm_instobj(memory)->parent;
+	return nvkm_memory_target(memory);
+}
+
+static u64
+nvkm_instobj_addr(struct nvkm_memory *memory)
+{
+	memory = nvkm_instobj(memory)->parent;
+	return nvkm_memory_addr(memory);
+}
+
+static u64
+nvkm_instobj_size(struct nvkm_memory *memory)
+{
+	memory = nvkm_instobj(memory)->parent;
+	return nvkm_memory_size(memory);
+}
+
+static void
+nvkm_instobj_release(struct nvkm_memory *memory)
+{
+	struct nvkm_instobj *iobj = nvkm_instobj(memory);
+	struct nvkm_bar *bar = iobj->imem->subdev.device->bar;
+	if (bar && bar->flush)
+		bar->flush(bar);
+}
+
+static void __iomem *
+nvkm_instobj_acquire(struct nvkm_memory *memory)
+{
+	return nvkm_instobj(memory)->map;
+}
+
+static u32
+nvkm_instobj_rd32(struct nvkm_memory *memory, u64 offset)
+{
+	return ioread32_native(nvkm_instobj(memory)->map + offset);
+}
+
+static void
+nvkm_instobj_wr32(struct nvkm_memory *memory, u64 offset, u32 data)
 {
-	struct nvkm_instmem *imem = nvkm_instmem(object);
-	struct nvkm_instobj *iobj = (void *)object;
+	iowrite32_native(data, nvkm_instobj(memory)->map + offset);
+}
+
+static void
+nvkm_instobj_map(struct nvkm_memory *memory, struct nvkm_vma *vma, u64 offset)
+{
+	memory = nvkm_instobj(memory)->parent;
+	nvkm_memory_map(memory, vma, offset);
+}
 
-	mutex_lock(&nv_subdev(imem)->mutex);
+static void *
+nvkm_instobj_dtor(struct nvkm_memory *memory)
+{
+	struct nvkm_instobj *iobj = nvkm_instobj(memory);
 	list_del(&iobj->head);
-	mutex_unlock(&nv_subdev(imem)->mutex);
+	nvkm_memory_del(&iobj->parent);
+	return iobj;
+}
 
-	return nvkm_object_destroy(&iobj->object);
+const struct nvkm_memory_func
+nvkm_instobj_func = {
+	.dtor = nvkm_instobj_dtor,
+	.target = nvkm_instobj_target,
+	.addr = nvkm_instobj_addr,
+	.size = nvkm_instobj_size,
+	.acquire = nvkm_instobj_acquire,
+	.release = nvkm_instobj_release,
+	.rd32 = nvkm_instobj_rd32,
+	.wr32 = nvkm_instobj_wr32,
+	.map = nvkm_instobj_map,
+};
+
+static void
+nvkm_instobj_boot(struct nvkm_memory *memory, struct nvkm_vm *vm)
+{
+	memory = nvkm_instobj(memory)->parent;
+	nvkm_memory_boot(memory, vm);
 }
 
-int
-nvkm_instobj_create_(struct nvkm_object *parent, struct nvkm_object *engine,
-		     struct nvkm_oclass *oclass, int length, void **pobject)
+static void
+nvkm_instobj_release_slow(struct nvkm_memory *memory)
+{
+	struct nvkm_instobj *iobj = nvkm_instobj(memory);
+	nvkm_instobj_release(memory);
+	nvkm_done(iobj->parent);
+}
+
+static void __iomem *
+nvkm_instobj_acquire_slow(struct nvkm_memory *memory)
+{
+	struct nvkm_instobj *iobj = nvkm_instobj(memory);
+	iobj->map = nvkm_kmap(iobj->parent);
+	if (iobj->map)
+		memory->func = &nvkm_instobj_func;
+	return iobj->map;
+}
+
+static u32
+nvkm_instobj_rd32_slow(struct nvkm_memory *memory, u64 offset)
+{
+	struct nvkm_instobj *iobj = nvkm_instobj(memory);
+	return nvkm_ro32(iobj->parent, offset);
+}
+
+static void
+nvkm_instobj_wr32_slow(struct nvkm_memory *memory, u64 offset, u32 data)
+{
+	struct nvkm_instobj *iobj = nvkm_instobj(memory);
+	return nvkm_wo32(iobj->parent, offset, data);
+}
+
+const struct nvkm_memory_func
+nvkm_instobj_func_slow = {
+	.dtor = nvkm_instobj_dtor,
+	.target = nvkm_instobj_target,
+	.addr = nvkm_instobj_addr,
+	.size = nvkm_instobj_size,
+	.boot = nvkm_instobj_boot,
+	.acquire = nvkm_instobj_acquire_slow,
+	.release = nvkm_instobj_release_slow,
+	.rd32 = nvkm_instobj_rd32_slow,
+	.wr32 = nvkm_instobj_wr32_slow,
+	.map = nvkm_instobj_map,
+};
+
+static int
+nvkm_instobj_new(struct nvkm_instmem *imem, u32 size, u32 align, bool zero,
+		 struct nvkm_memory **pmemory)
 {
-	struct nvkm_instmem *imem = nvkm_instmem(parent);
+	struct nvkm_instmem_impl *impl = (void *)imem->subdev.object.oclass;
+	struct nvkm_memory *memory;
 	struct nvkm_instobj *iobj;
+	u32 offset;
 	int ret;
 
-	ret = nvkm_object_create_(parent, engine, oclass, NV_MEMOBJ_CLASS,
-				  length, pobject);
-	iobj = *pobject;
+	ret = impl->memory_new(imem, size, align, zero, &memory);
 	if (ret)
-		return ret;
+		goto done;
 
-	mutex_lock(&imem->subdev.mutex);
-	list_add(&iobj->head, &imem->list);
-	mutex_unlock(&imem->subdev.mutex);
-	return 0;
+	if (!impl->persistent) {
+		if (!(iobj = kzalloc(sizeof(*iobj), GFP_KERNEL))) {
+			ret = -ENOMEM;
+			goto done;
+		}
+
+		nvkm_memory_ctor(&nvkm_instobj_func_slow, &iobj->memory);
+		iobj->parent = memory;
+		iobj->imem = imem;
+		list_add_tail(&iobj->head, &imem->list);
+		memory = &iobj->memory;
+	}
+
+	if (!impl->zero && zero) {
+		void __iomem *map = nvkm_kmap(memory);
+		if (unlikely(!map)) {
+			for (offset = 0; offset < size; offset += 4)
+				nvkm_wo32(memory, offset, 0x00000000);
+		} else {
+			memset_io(map, 0x00, size);
+		}
+		nvkm_done(memory);
+	}
+
+done:
+	if (ret)
+		nvkm_memory_del(&memory);
+	*pmemory = memory;
+	return ret;
 }
 
 /******************************************************************************
  * instmem subdev base implementation
  *****************************************************************************/
 
-static int
-nvkm_instmem_alloc(struct nvkm_instmem *imem, struct nvkm_object *parent,
-		   u32 size, u32 align, struct nvkm_object **pobject)
-{
-	struct nvkm_instmem_impl *impl = (void *)imem->subdev.object.oclass;
-	struct nvkm_instobj_args args = { .size = size, .align = align };
-	return nvkm_object_old(parent, &parent->engine->subdev.object,
-				impl->instobj, &args, sizeof(args), pobject);
-}
-
 int
 _nvkm_instmem_fini(struct nvkm_object *object, bool suspend)
 {
 	struct nvkm_instmem *imem = (void *)object;
 	struct nvkm_instobj *iobj;
-	int i, ret = 0;
+	int i;
 
 	if (suspend) {
-		mutex_lock(&imem->subdev.mutex);
 		list_for_each_entry(iobj, &imem->list, head) {
-			iobj->suspend = vmalloc(iobj->size);
-			if (!iobj->suspend) {
-				ret = -ENOMEM;
-				break;
-			}
-
-			for (i = 0; i < iobj->size; i += 4) {
-				nvkm_object_rd32(&iobj->object, i, (u32 *)
-						 &iobj->suspend[i/4]);
-			}
+			struct nvkm_memory *memory = iobj->parent;
+			u64 size = nvkm_memory_size(memory);
+
+			iobj->suspend = vmalloc(size);
+			if (!iobj->suspend)
+				return -ENOMEM;
+
+			for (i = 0; i < size; i += 4)
+				iobj->suspend[i / 4] = nvkm_ro32(memory, i);
 		}
-		mutex_unlock(&imem->subdev.mutex);
-		if (ret)
-			return ret;
 	}
 
 	return nvkm_subdev_fini_old(&imem->subdev, suspend);
@@ -116,18 +254,17 @@ _nvkm_instmem_init(struct nvkm_object *object)
 	if (ret)
 		return ret;
 
-	mutex_lock(&imem->subdev.mutex);
 	list_for_each_entry(iobj, &imem->list, head) {
 		if (iobj->suspend) {
-			for (i = 0; i < iobj->size; i += 4) {
-				nvkm_object_wr32(&iobj->object, i, *(u32 *)
-						 &iobj->suspend[i/4]);
-			}
+			struct nvkm_memory *memory = iobj->parent;
+			u64 size = nvkm_memory_size(memory);
+			for (i = 0; i < size; i += 4)
+				nvkm_wo32(memory, i, iobj->suspend[i / 4]);
 			vfree(iobj->suspend);
 			iobj->suspend = NULL;
 		}
 	}
-	mutex_unlock(&imem->subdev.mutex);
+
 	return 0;
 }
 
@@ -135,6 +272,7 @@ int
 nvkm_instmem_create_(struct nvkm_object *parent, struct nvkm_object *engine,
 		     struct nvkm_oclass *oclass, int length, void **pobject)
 {
+	struct nvkm_device *device = (void *)parent;
 	struct nvkm_instmem *imem;
 	int ret;
 
@@ -144,7 +282,9 @@ nvkm_instmem_create_(struct nvkm_object *parent, struct nvkm_object *engine,
 	if (ret)
 		return ret;
 
+	device->imem = imem;
+
 	INIT_LIST_HEAD(&imem->list);
-	imem->alloc = nvkm_instmem_alloc;
+	imem->alloc = nvkm_instobj_new;
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c
index e6370382109b4..a64c3f9bfc3dc 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c
@@ -37,9 +37,12 @@
  * to use more "relaxed" allocation parameters when using the DMA API, since we
  * never need a kernel mapping.
  */
+#define gk20a_instmem(p) container_of((p), struct gk20a_instmem, base)
+#include "priv.h"
 
-#include <subdev/fb.h>
+#include <core/memory.h>
 #include <core/mm.h>
+#include <subdev/fb.h>
 
 #ifdef __KERNEL__
 #include <linux/dma-attrs.h>
@@ -47,14 +50,12 @@
 #include <nouveau_platform.h>
 #endif
 
-#include "priv.h"
+#define gk20a_instobj(p) container_of((p), struct gk20a_instobj, memory)
 
 struct gk20a_instobj {
-	struct nvkm_instobj base;
-	/* Must be second member here - see nouveau_gpuobj_map_vm() */
-	struct nvkm_mem *mem;
-	/* Pointed by mem */
-	struct nvkm_mem _mem;
+	struct nvkm_memory memory;
+	struct gk20a_instmem *imem;
+	struct nvkm_mem mem;
 };
 
 /*
@@ -80,6 +81,7 @@ struct gk20a_instobj_iommu {
 
 struct gk20a_instmem {
 	struct nvkm_instmem base;
+	unsigned long lock_flags;
 	spinlock_t lock;
 	u64 addr;
 
@@ -93,6 +95,42 @@ struct gk20a_instmem {
 	struct dma_attrs attrs;
 };
 
+static enum nvkm_memory_target
+gk20a_instobj_target(struct nvkm_memory *memory)
+{
+	return NVKM_MEM_TARGET_HOST;
+}
+
+static u64
+gk20a_instobj_addr(struct nvkm_memory *memory)
+{
+	return gk20a_instobj(memory)->mem.offset;
+
+}
+
+static u64
+gk20a_instobj_size(struct nvkm_memory *memory)
+{
+	return (u64)gk20a_instobj(memory)->mem.size << 12;
+}
+
+static void __iomem *
+gk20a_instobj_acquire(struct nvkm_memory *memory)
+{
+	struct gk20a_instmem *imem = gk20a_instobj(memory)->imem;
+	unsigned long flags;
+	spin_lock_irqsave(&imem->lock, flags);
+	imem->lock_flags = flags;
+	return NULL;
+}
+
+static void
+gk20a_instobj_release(struct nvkm_memory *memory)
+{
+	struct gk20a_instmem *imem = gk20a_instobj(memory)->imem;
+	spin_unlock_irqrestore(&imem->lock, imem->lock_flags);
+}
+
 /*
  * Use PRAMIN to read/write data and avoid coherency issues.
  * PRAMIN uses the GPU path and ensures data will always be coherent.
@@ -103,56 +141,57 @@ struct gk20a_instmem {
  */
 
 static u32
-gk20a_instobj_rd32(struct nvkm_object *object, u64 offset)
+gk20a_instobj_rd32(struct nvkm_memory *memory, u64 offset)
 {
-	struct gk20a_instmem *imem = (void *)nvkm_instmem(object);
-	struct gk20a_instobj *node = (void *)object;
+	struct gk20a_instobj *node = gk20a_instobj(memory);
+	struct gk20a_instmem *imem = node->imem;
 	struct nvkm_device *device = imem->base.subdev.device;
-	unsigned long flags;
-	u64 base = (node->mem->offset + offset) & 0xffffff00000ULL;
-	u64 addr = (node->mem->offset + offset) & 0x000000fffffULL;
+	u64 base = (node->mem.offset + offset) & 0xffffff00000ULL;
+	u64 addr = (node->mem.offset + offset) & 0x000000fffffULL;
 	u32 data;
 
-	spin_lock_irqsave(&imem->lock, flags);
 	if (unlikely(imem->addr != base)) {
 		nvkm_wr32(device, 0x001700, base >> 16);
 		imem->addr = base;
 	}
 	data = nvkm_rd32(device, 0x700000 + addr);
-	spin_unlock_irqrestore(&imem->lock, flags);
 	return data;
 }
 
 static void
-gk20a_instobj_wr32(struct nvkm_object *object, u64 offset, u32 data)
+gk20a_instobj_wr32(struct nvkm_memory *memory, u64 offset, u32 data)
 {
-	struct gk20a_instmem *imem = (void *)nvkm_instmem(object);
-	struct gk20a_instobj *node = (void *)object;
+	struct gk20a_instobj *node = gk20a_instobj(memory);
+	struct gk20a_instmem *imem = node->imem;
 	struct nvkm_device *device = imem->base.subdev.device;
-	unsigned long flags;
-	u64 base = (node->mem->offset + offset) & 0xffffff00000ULL;
-	u64 addr = (node->mem->offset + offset) & 0x000000fffffULL;
+	u64 base = (node->mem.offset + offset) & 0xffffff00000ULL;
+	u64 addr = (node->mem.offset + offset) & 0x000000fffffULL;
 
-	spin_lock_irqsave(&imem->lock, flags);
 	if (unlikely(imem->addr != base)) {
 		nvkm_wr32(device, 0x001700, base >> 16);
 		imem->addr = base;
 	}
 	nvkm_wr32(device, 0x700000 + addr, data);
-	spin_unlock_irqrestore(&imem->lock, flags);
+}
+
+static void
+gk20a_instobj_map(struct nvkm_memory *memory, struct nvkm_vma *vma, u64 offset)
+{
+	struct gk20a_instobj *node = gk20a_instobj(memory);
+	nvkm_vm_map_at(vma, offset, &node->mem);
 }
 
 static void
 gk20a_instobj_dtor_dma(struct gk20a_instobj *_node)
 {
 	struct gk20a_instobj_dma *node = (void *)_node;
-	struct gk20a_instmem *imem = (void *)nvkm_instmem(node);
+	struct gk20a_instmem *imem = _node->imem;
 	struct device *dev = nv_device_base(nv_device(imem));
 
 	if (unlikely(!node->cpuaddr))
 		return;
 
-	dma_free_attrs(dev, _node->mem->size << PAGE_SHIFT, node->cpuaddr,
+	dma_free_attrs(dev, _node->mem.size << PAGE_SHIFT, node->cpuaddr,
 		       node->handle, &imem->attrs);
 }
 
@@ -160,21 +199,21 @@ static void
 gk20a_instobj_dtor_iommu(struct gk20a_instobj *_node)
 {
 	struct gk20a_instobj_iommu *node = (void *)_node;
-	struct gk20a_instmem *imem = (void *)nvkm_instmem(node);
+	struct gk20a_instmem *imem = _node->imem;
 	struct nvkm_mm_node *r;
 	int i;
 
-	if (unlikely(list_empty(&_node->mem->regions)))
+	if (unlikely(list_empty(&_node->mem.regions)))
 		return;
 
-	r = list_first_entry(&_node->mem->regions, struct nvkm_mm_node,
+	r = list_first_entry(&_node->mem.regions, struct nvkm_mm_node,
 			     rl_entry);
 
 	/* clear bit 34 to unmap pages */
 	r->offset &= ~BIT(34 - imem->iommu_pgshift);
 
 	/* Unmap pages from GPU address space and free them */
-	for (i = 0; i < _node->mem->size; i++) {
+	for (i = 0; i < _node->mem.size; i++) {
 		iommu_unmap(imem->domain,
 			    (r->offset + i) << imem->iommu_pgshift, PAGE_SIZE);
 		__free_page(node->pages[i]);
@@ -186,36 +225,44 @@ gk20a_instobj_dtor_iommu(struct gk20a_instobj *_node)
 	mutex_unlock(imem->mm_mutex);
 }
 
-static void
-gk20a_instobj_dtor(struct nvkm_object *object)
+static void *
+gk20a_instobj_dtor(struct nvkm_memory *memory)
 {
-	struct gk20a_instobj *node = (void *)object;
-	struct gk20a_instmem *imem = (void *)nvkm_instmem(node);
+	struct gk20a_instobj *node = gk20a_instobj(memory);
+	struct gk20a_instmem *imem = node->imem;
 
 	if (imem->domain)
 		gk20a_instobj_dtor_iommu(node);
 	else
 		gk20a_instobj_dtor_dma(node);
 
-	nvkm_instobj_destroy(&node->base);
+	return node;
 }
 
+static const struct nvkm_memory_func
+gk20a_instobj_func = {
+	.dtor = gk20a_instobj_dtor,
+	.target = gk20a_instobj_target,
+	.addr = gk20a_instobj_addr,
+	.size = gk20a_instobj_size,
+	.acquire = gk20a_instobj_acquire,
+	.release = gk20a_instobj_release,
+	.rd32 = gk20a_instobj_rd32,
+	.wr32 = gk20a_instobj_wr32,
+	.map = gk20a_instobj_map,
+};
+
 static int
-gk20a_instobj_ctor_dma(struct nvkm_object *parent, struct nvkm_object *engine,
-		       struct nvkm_oclass *oclass, u32 npages, u32 align,
+gk20a_instobj_ctor_dma(struct gk20a_instmem *imem, u32 npages, u32 align,
 		       struct gk20a_instobj **_node)
 {
 	struct gk20a_instobj_dma *node;
-	struct gk20a_instmem *imem = (void *)nvkm_instmem(parent);
 	struct nvkm_subdev *subdev = &imem->base.subdev;
-	struct device *dev = nv_device_base(nv_device(parent));
-	int ret;
+	struct device *dev = subdev->device->dev;
 
-	ret = nvkm_instobj_create_(parent, engine, oclass, sizeof(*node),
-				   (void **)&node);
+	if (!(node = kzalloc(sizeof(*node), GFP_KERNEL)))
+		return -ENOMEM;
 	*_node = &node->base;
-	if (ret)
-		return ret;
 
 	node->cpuaddr = dma_alloc_attrs(dev, npages << PAGE_SHIFT,
 					&node->handle, GFP_KERNEL,
@@ -236,32 +283,28 @@ gk20a_instobj_ctor_dma(struct nvkm_object *parent, struct nvkm_object *engine,
 	node->r.offset = node->handle >> 12;
 	node->r.length = (npages << PAGE_SHIFT) >> 12;
 
-	node->base._mem.offset = node->handle;
+	node->base.mem.offset = node->handle;
 
-	INIT_LIST_HEAD(&node->base._mem.regions);
-	list_add_tail(&node->r.rl_entry, &node->base._mem.regions);
+	INIT_LIST_HEAD(&node->base.mem.regions);
+	list_add_tail(&node->r.rl_entry, &node->base.mem.regions);
 
 	return 0;
 }
 
 static int
-gk20a_instobj_ctor_iommu(struct nvkm_object *parent, struct nvkm_object *engine,
-			 struct nvkm_oclass *oclass, u32 npages, u32 align,
+gk20a_instobj_ctor_iommu(struct gk20a_instmem *imem, u32 npages, u32 align,
 			 struct gk20a_instobj **_node)
 {
 	struct gk20a_instobj_iommu *node;
-	struct gk20a_instmem *imem = (void *)nvkm_instmem(parent);
 	struct nvkm_subdev *subdev = &imem->base.subdev;
 	struct nvkm_mm_node *r;
 	int ret;
 	int i;
 
-	ret = nvkm_instobj_create_(parent, engine, oclass,
-				sizeof(*node) + sizeof(node->pages[0]) * npages,
-				(void **)&node);
+	if (!(node = kzalloc(sizeof(*node) +
+			     sizeof( node->pages[0]) * npages, GFP_KERNEL)))
+		return -ENOMEM;
 	*_node = &node->base;
-	if (ret)
-		return ret;
 
 	/* Allocate backing memory */
 	for (i = 0; i < npages; i++) {
@@ -305,10 +348,10 @@ gk20a_instobj_ctor_iommu(struct nvkm_object *parent, struct nvkm_object *engine,
 	/* Bit 34 tells that an address is to be resolved through the IOMMU */
 	r->offset |= BIT(34 - imem->iommu_pgshift);
 
-	node->base._mem.offset = ((u64)r->offset) << imem->iommu_pgshift;
+	node->base.mem.offset = ((u64)r->offset) << imem->iommu_pgshift;
 
-	INIT_LIST_HEAD(&node->base._mem.regions);
-	list_add_tail(&r->rl_entry, &node->base._mem.regions);
+	INIT_LIST_HEAD(&node->base.mem.regions);
+	list_add_tail(&r->rl_entry, &node->base.mem.regions);
 
 	return 0;
 
@@ -325,64 +368,45 @@ free_pages:
 }
 
 static int
-gk20a_instobj_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
-		   struct nvkm_oclass *oclass, void *data, u32 _size,
-		   struct nvkm_object **pobject)
+gk20a_instobj_new(struct nvkm_instmem *base, u32 size, u32 align, bool zero,
+		  struct nvkm_memory **pmemory)
 {
-	struct nvkm_instobj_args *args = data;
-	struct gk20a_instmem *imem = (void *)nvkm_instmem(parent);
+	struct gk20a_instmem *imem = gk20a_instmem(base);
 	struct gk20a_instobj *node;
 	struct nvkm_subdev *subdev = &imem->base.subdev;
-	u32 size, align;
 	int ret;
 
 	nvkm_debug(subdev, "%s (%s): size: %x align: %x\n", __func__,
-		   imem->domain ? "IOMMU" : "DMA", args->size, args->align);
+		   imem->domain ? "IOMMU" : "DMA", size, align);
 
 	/* Round size and align to page bounds */
-	size = max(roundup(args->size, PAGE_SIZE), PAGE_SIZE);
-	align = max(roundup(args->align, PAGE_SIZE), PAGE_SIZE);
+	size = max(roundup(size, PAGE_SIZE), PAGE_SIZE);
+	align = max(roundup(align, PAGE_SIZE), PAGE_SIZE);
 
 	if (imem->domain)
-		ret = gk20a_instobj_ctor_iommu(parent, engine, oclass,
-					      size >> PAGE_SHIFT, align, &node);
+		ret = gk20a_instobj_ctor_iommu(imem, size >> PAGE_SHIFT,
+					       align, &node);
 	else
-		ret = gk20a_instobj_ctor_dma(parent, engine, oclass,
-					     size >> PAGE_SHIFT, align, &node);
-	*pobject = nv_object(node);
+		ret = gk20a_instobj_ctor_dma(imem, size >> PAGE_SHIFT,
+					     align, &node);
 	if (ret)
 		return ret;
+	*pmemory = &node->memory;
 
-	node->mem = &node->_mem;
+	nvkm_memory_ctor(&gk20a_instobj_func, &node->memory);
+	node->imem = imem;
 
 	/* present memory for being mapped using small pages */
-	node->mem->size = size >> 12;
-	node->mem->memtype = 0;
-	node->mem->page_shift = 12;
-
-	node->base.addr = node->mem->offset;
-	node->base.size = size;
+	node->mem.size = size >> 12;
+	node->mem.memtype = 0;
+	node->mem.page_shift = 12;
 
 	nvkm_debug(subdev, "alloc size: 0x%x, align: 0x%x, gaddr: 0x%llx\n",
-		   size, align, node->mem->offset);
+		   size, align, node->mem.offset);
 
 	return 0;
 }
 
-static struct nvkm_instobj_impl
-gk20a_instobj_oclass = {
-	.base.ofuncs = &(struct nvkm_ofuncs) {
-		.ctor = gk20a_instobj_ctor,
-		.dtor = gk20a_instobj_dtor,
-		.init = _nvkm_instobj_init,
-		.fini = _nvkm_instobj_fini,
-		.rd32 = gk20a_instobj_rd32,
-		.wr32 = gk20a_instobj_wr32,
-	},
-};
-
-
-
 static int
 gk20a_instmem_fini(struct nvkm_object *object, bool suspend)
 {
@@ -440,5 +464,7 @@ gk20a_instmem_oclass = &(struct nvkm_instmem_impl) {
 		.init = _nvkm_instmem_init,
 		.fini = gk20a_instmem_fini,
 	},
-	.instobj = &gk20a_instobj_oclass.base,
+	.memory_new = gk20a_instobj_new,
+	.persistent = true,
+	.zero = false,
 }.base;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.c
index 8ba95f366e2fe..c499e485373b8 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.c
@@ -21,83 +21,119 @@
  *
  * Authors: Ben Skeggs
  */
-#include "nv04.h"
+#define nv04_instmem(p) container_of((p), struct nv04_instmem, base)
+#include "priv.h"
 
+#include <core/memory.h>
 #include <core/ramht.h>
 
+struct nv04_instmem {
+	struct nvkm_instmem base;
+	struct nvkm_mm heap;
+};
+
 /******************************************************************************
  * instmem object implementation
  *****************************************************************************/
+#define nv04_instobj(p) container_of((p), struct nv04_instobj, memory)
 
-static u32
-nv04_instobj_rd32(struct nvkm_object *object, u64 addr)
+struct nv04_instobj {
+	struct nvkm_memory memory;
+	struct nv04_instmem *imem;
+	struct nvkm_mm_node *node;
+};
+
+static enum nvkm_memory_target
+nv04_instobj_target(struct nvkm_memory *memory)
+{
+	return NVKM_MEM_TARGET_INST;
+}
+
+static u64
+nv04_instobj_addr(struct nvkm_memory *memory)
+{
+	return nv04_instobj(memory)->node->offset;
+}
+
+static u64
+nv04_instobj_size(struct nvkm_memory *memory)
+{
+	return nv04_instobj(memory)->node->length;
+}
+
+static void __iomem *
+nv04_instobj_acquire(struct nvkm_memory *memory)
 {
-	struct nvkm_instmem *imem = nvkm_instmem(object);
-	struct nv04_instobj *node = (void *)object;
-	return imem->func->rd32(imem, node->mem->offset + addr);
+	struct nv04_instobj *iobj = nv04_instobj(memory);
+	struct nvkm_device *device = iobj->imem->base.subdev.device;
+	return device->pri + 0x700000 + iobj->node->offset;
 }
 
 static void
-nv04_instobj_wr32(struct nvkm_object *object, u64 addr, u32 data)
+nv04_instobj_release(struct nvkm_memory *memory)
 {
-	struct nvkm_instmem *imem = nvkm_instmem(object);
-	struct nv04_instobj *node = (void *)object;
-	imem->func->wr32(imem, node->mem->offset + addr, data);
+}
+
+static u32
+nv04_instobj_rd32(struct nvkm_memory *memory, u64 offset)
+{
+	struct nv04_instobj *iobj = nv04_instobj(memory);
+	struct nvkm_device *device = iobj->imem->base.subdev.device;
+	return nvkm_rd32(device, 0x700000 + iobj->node->offset + offset);
 }
 
 static void
-nv04_instobj_dtor(struct nvkm_object *object)
+nv04_instobj_wr32(struct nvkm_memory *memory, u64 offset, u32 data)
 {
-	struct nv04_instmem *imem = (void *)nvkm_instmem(object);
-	struct nv04_instobj *node = (void *)object;
-	mutex_lock(&imem->base.subdev.mutex);
-	nvkm_mm_free(&imem->heap, &node->mem);
-	mutex_unlock(&imem->base.subdev.mutex);
-	nvkm_instobj_destroy(&node->base);
+	struct nv04_instobj *iobj = nv04_instobj(memory);
+	struct nvkm_device *device = iobj->imem->base.subdev.device;
+	nvkm_wr32(device, 0x700000 + iobj->node->offset + offset, data);
 }
 
+static void *
+nv04_instobj_dtor(struct nvkm_memory *memory)
+{
+	struct nv04_instobj *iobj = nv04_instobj(memory);
+	mutex_lock(&iobj->imem->base.subdev.mutex);
+	nvkm_mm_free(&iobj->imem->heap, &iobj->node);
+	mutex_unlock(&iobj->imem->base.subdev.mutex);
+	return iobj;
+}
+
+static const struct nvkm_memory_func
+nv04_instobj_func = {
+	.dtor = nv04_instobj_dtor,
+	.target = nv04_instobj_target,
+	.size = nv04_instobj_size,
+	.addr = nv04_instobj_addr,
+	.acquire = nv04_instobj_acquire,
+	.release = nv04_instobj_release,
+	.rd32 = nv04_instobj_rd32,
+	.wr32 = nv04_instobj_wr32,
+};
+
 static int
-nv04_instobj_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
-		  struct nvkm_oclass *oclass, void *data, u32 size,
-		  struct nvkm_object **pobject)
+nv04_instobj_new(struct nvkm_instmem *base, u32 size, u32 align, bool zero,
+		 struct nvkm_memory **pmemory)
 {
-	struct nv04_instmem *imem = (void *)nvkm_instmem(parent);
-	struct nv04_instobj *node;
-	struct nvkm_instobj_args *args = data;
+	struct nv04_instmem *imem = nv04_instmem(base);
+	struct nv04_instobj *iobj;
 	int ret;
 
-	if (!args->align)
-		args->align = 1;
+	if (!(iobj = kzalloc(sizeof(*iobj), GFP_KERNEL)))
+		return -ENOMEM;
+	*pmemory = &iobj->memory;
 
-	ret = nvkm_instobj_create(parent, engine, oclass, &node);
-	*pobject = nv_object(node);
-	if (ret)
-		return ret;
+	nvkm_memory_ctor(&nv04_instobj_func, &iobj->memory);
+	iobj->imem = imem;
 
 	mutex_lock(&imem->base.subdev.mutex);
-	ret = nvkm_mm_head(&imem->heap, 0, 1, args->size, args->size,
-			   args->align, &node->mem);
+	ret = nvkm_mm_head(&imem->heap, 0, 1, size, size,
+			   align ? align : 1, &iobj->node);
 	mutex_unlock(&imem->base.subdev.mutex);
-	if (ret)
-		return ret;
-
-	node->base.addr = node->mem->offset;
-	node->base.size = node->mem->length;
-	return 0;
+	return ret;
 }
 
-struct nvkm_instobj_impl
-nv04_instobj_oclass = {
-	.base.ofuncs = &(struct nvkm_ofuncs) {
-		.ctor = nv04_instobj_ctor,
-		.dtor = nv04_instobj_dtor,
-		.init = _nvkm_instobj_init,
-		.fini = _nvkm_instobj_fini,
-		.rd32 = nv04_instobj_rd32,
-		.wr32 = nv04_instobj_wr32,
-	},
-};
-
 /******************************************************************************
  * instmem subdev implementation
  *****************************************************************************/
@@ -114,17 +150,15 @@ nv04_instmem_wr32(struct nvkm_instmem *imem, u32 addr, u32 data)
 	nvkm_wr32(imem->subdev.device, 0x700000 + addr, data);
 }
 
-void
+static void
 nv04_instmem_dtor(struct nvkm_object *object)
 {
 	struct nv04_instmem *imem = (void *)object;
-	nvkm_gpuobj_ref(NULL, &imem->ramfc);
-	nvkm_gpuobj_ref(NULL, &imem->ramro);
-	nvkm_ramht_ref(NULL, &imem->ramht);
-	nvkm_gpuobj_ref(NULL, &imem->vbios);
+	nvkm_gpuobj_ref(NULL, &imem->base.ramfc);
+	nvkm_gpuobj_ref(NULL, &imem->base.ramro);
+	nvkm_ramht_ref(NULL, &imem->base.ramht);
+	nvkm_gpuobj_ref(NULL, &imem->base.vbios);
 	nvkm_mm_fini(&imem->heap);
-	if (imem->iomem)
-		iounmap(imem->iomem);
 	nvkm_instmem_destroy(&imem->base);
 }
 
@@ -158,24 +192,26 @@ nv04_instmem_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 
 	/* 0x00000-0x10000: reserve for probable vbios image */
 	ret = nvkm_gpuobj_new(nv_object(imem), NULL, 0x10000, 0, 0,
-			      &imem->vbios);
+			      &imem->base.vbios);
 	if (ret)
 		return ret;
 
 	/* 0x10000-0x18000: reserve for RAMHT */
-	ret = nvkm_ramht_new(nv_object(imem), NULL, 0x08000, 0, &imem->ramht);
+	ret = nvkm_ramht_new(nv_object(imem), NULL, 0x08000, 0,
+			     &imem->base.ramht);
 	if (ret)
 		return ret;
 
 	/* 0x18000-0x18800: reserve for RAMFC (enough for 32 nv30 channels) */
 	ret = nvkm_gpuobj_new(nv_object(imem), NULL, 0x00800, 0,
-			      NVOBJ_FLAG_ZERO_ALLOC, &imem->ramfc);
+			      NVOBJ_FLAG_ZERO_ALLOC,
+			      &imem->base.ramfc);
 	if (ret)
 		return ret;
 
 	/* 0x18800-0x18a00: reserve for RAMRO */
 	ret = nvkm_gpuobj_new(nv_object(imem), NULL, 0x00200, 0, 0,
-			      &imem->ramro);
+			      &imem->base.ramro);
 	if (ret)
 		return ret;
 
@@ -191,5 +227,7 @@ nv04_instmem_oclass = &(struct nvkm_instmem_impl) {
 		.init = _nvkm_instmem_init,
 		.fini = _nvkm_instmem_fini,
 	},
-	.instobj = &nv04_instobj_oclass.base,
+	.memory_new = nv04_instobj_new,
+	.persistent = false,
+	.zero = false,
 }.base;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.h b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.h
deleted file mode 100644
index 6065e34e2f0d6..0000000000000
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.h
+++ /dev/null
@@ -1,36 +0,0 @@
-#ifndef __NV04_INSTMEM_H__
-#define __NV04_INSTMEM_H__
-#include "priv.h"
-
-#include <core/mm.h>
-
-extern struct nvkm_instobj_impl nv04_instobj_oclass;
-
-struct nv04_instmem {
-	struct nvkm_instmem base;
-
-	void __iomem *iomem;
-	struct nvkm_mm heap;
-
-	struct nvkm_gpuobj *vbios;
-	struct nvkm_ramht  *ramht;
-	struct nvkm_gpuobj *ramro;
-	struct nvkm_gpuobj *ramfc;
-};
-
-static inline struct nv04_instmem *
-nv04_instmem(void *obj)
-{
-	return (void *)nvkm_instmem(obj);
-}
-
-struct nv04_instobj {
-	struct nvkm_instobj base;
-	struct nvkm_mm_node *mem;
-};
-
-void nv04_instmem_dtor(struct nvkm_object *);
-
-int nv04_instmem_alloc(struct nvkm_instmem *, struct nvkm_object *,
-		       u32 size, u32 align, struct nvkm_object **pobject);
-#endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv40.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv40.c
index c645e02615301..3ac55481cc97d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv40.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv40.c
@@ -21,11 +21,118 @@
  *
  * Authors: Ben Skeggs
  */
-#include "nv04.h"
+#define nv40_instmem(p) container_of((p), struct nv40_instmem, base)
+#include "priv.h"
 
+#include <core/memory.h>
 #include <core/ramht.h>
 #include <engine/gr/nv40.h>
 
+struct nv40_instmem {
+	struct nvkm_instmem base;
+	struct nvkm_mm heap;
+	void __iomem *iomem;
+};
+
+/******************************************************************************
+ * instmem object implementation
+ *****************************************************************************/
+#define nv40_instobj(p) container_of((p), struct nv40_instobj, memory)
+
+struct nv40_instobj {
+	struct nvkm_memory memory;
+	struct nv40_instmem *imem;
+	struct nvkm_mm_node *node;
+};
+
+static enum nvkm_memory_target
+nv40_instobj_target(struct nvkm_memory *memory)
+{
+	return NVKM_MEM_TARGET_INST;
+}
+
+static u64
+nv40_instobj_addr(struct nvkm_memory *memory)
+{
+	return nv40_instobj(memory)->node->offset;
+}
+
+static u64
+nv40_instobj_size(struct nvkm_memory *memory)
+{
+	return nv40_instobj(memory)->node->length;
+}
+
+static void __iomem *
+nv40_instobj_acquire(struct nvkm_memory *memory)
+{
+	struct nv40_instobj *iobj = nv40_instobj(memory);
+	return iobj->imem->iomem + iobj->node->offset;
+}
+
+static void
+nv40_instobj_release(struct nvkm_memory *memory)
+{
+}
+
+static u32
+nv40_instobj_rd32(struct nvkm_memory *memory, u64 offset)
+{
+	struct nv40_instobj *iobj = nv40_instobj(memory);
+	return ioread32_native(iobj->imem->iomem + iobj->node->offset + offset);
+}
+
+static void
+nv40_instobj_wr32(struct nvkm_memory *memory, u64 offset, u32 data)
+{
+	struct nv40_instobj *iobj = nv40_instobj(memory);
+	iowrite32_native(data, iobj->imem->iomem + iobj->node->offset + offset);
+}
+
+static void *
+nv40_instobj_dtor(struct nvkm_memory *memory)
+{
+	struct nv40_instobj *iobj = nv40_instobj(memory);
+	mutex_lock(&iobj->imem->base.subdev.mutex);
+	nvkm_mm_free(&iobj->imem->heap, &iobj->node);
+	mutex_unlock(&iobj->imem->base.subdev.mutex);
+	return iobj;
+}
+
+static const struct nvkm_memory_func
+nv40_instobj_func = {
+	.dtor = nv40_instobj_dtor,
+	.target = nv40_instobj_target,
+	.size = nv40_instobj_size,
+	.addr = nv40_instobj_addr,
+	.acquire = nv40_instobj_acquire,
+	.release = nv40_instobj_release,
+	.rd32 = nv40_instobj_rd32,
+	.wr32 = nv40_instobj_wr32,
+};
+
+static int
+nv40_instobj_new(struct nvkm_instmem *base, u32 size, u32 align, bool zero,
+		 struct nvkm_memory **pmemory)
+{
+	struct nv40_instmem *imem = nv40_instmem(base);
+	struct nv40_instobj *iobj;
+	int ret;
+
+	if (!(iobj = kzalloc(sizeof(*iobj), GFP_KERNEL)))
+		return -ENOMEM;
+	*pmemory = &iobj->memory;
+
+	nvkm_memory_ctor(&nv40_instobj_func, &iobj->memory);
+	iobj->imem = imem;
+
+	mutex_lock(&imem->base.subdev.mutex);
+	ret = nvkm_mm_head(&imem->heap, 0, 1, size, size,
+			   align ? align : 1, &iobj->node);
+	mutex_unlock(&imem->base.subdev.mutex);
+	return ret;
+}
+
 /******************************************************************************
  * instmem subdev implementation
  *****************************************************************************/
@@ -33,17 +140,31 @@
 static u32
 nv40_instmem_rd32(struct nvkm_instmem *obj, u32 addr)
 {
-	struct nv04_instmem *imem = container_of(obj, typeof(*imem), base);
+	struct nv40_instmem *imem = container_of(obj, typeof(*imem), base);
 	return ioread32_native(imem->iomem + addr);
 }
 
 static void
 nv40_instmem_wr32(struct nvkm_instmem *obj, u32 addr, u32 data)
 {
-	struct nv04_instmem *imem = container_of(obj, typeof(*imem), base);
+	struct nv40_instmem *imem = container_of(obj, typeof(*imem), base);
 	iowrite32_native(data, imem->iomem + addr);
 }
 
+static void
+nv40_instmem_dtor(struct nvkm_object *object)
+{
+	struct nv40_instmem *imem = (void *)object;
+	nvkm_gpuobj_ref(NULL, &imem->base.ramfc);
+	nvkm_gpuobj_ref(NULL, &imem->base.ramro);
+	nvkm_ramht_ref(NULL, &imem->base.ramht);
+	nvkm_gpuobj_ref(NULL, &imem->base.vbios);
+	nvkm_mm_fini(&imem->heap);
+	if (imem->iomem)
+		iounmap(imem->iomem);
+	nvkm_instmem_destroy(&imem->base);
+}
+
 static const struct nvkm_instmem_func
 nv40_instmem_func = {
 	.rd32 = nv40_instmem_rd32,
@@ -56,7 +177,7 @@ nv40_instmem_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 		  struct nvkm_object **pobject)
 {
 	struct nvkm_device *device = (void *)parent;
-	struct nv04_instmem *imem;
+	struct nv40_instmem *imem;
 	int ret, bar, vs;
 
 	ret = nvkm_instmem_create(parent, engine, oclass, &imem);
@@ -86,7 +207,7 @@ nv40_instmem_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	vs = hweight8((nvkm_rd32(device, 0x001540) & 0x0000ff00) >> 8);
 	if      (device->chipset == 0x40) imem->base.reserved = 0x6aa0 * vs;
 	else if (device->chipset  < 0x43) imem->base.reserved = 0x4f00 * vs;
-	else if (nv44_gr_class(imem))  imem->base.reserved = 0x4980 * vs;
+	else if (nv44_gr_class(imem))     imem->base.reserved = 0x4980 * vs;
 	else				  imem->base.reserved = 0x4a40 * vs;
 	imem->base.reserved += 16 * 1024;
 	imem->base.reserved *= 32;		/* per-channel */
@@ -101,12 +222,13 @@ nv40_instmem_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 
 	/* 0x00000-0x10000: reserve for probable vbios image */
 	ret = nvkm_gpuobj_new(nv_object(imem), NULL, 0x10000, 0, 0,
-			      &imem->vbios);
+			      &imem->base.vbios);
 	if (ret)
 		return ret;
 
 	/* 0x10000-0x18000: reserve for RAMHT */
-	ret = nvkm_ramht_new(nv_object(imem), NULL, 0x08000, 0, &imem->ramht);
+	ret = nvkm_ramht_new(nv_object(imem), NULL, 0x08000, 0,
+			     &imem->base.ramht);
 	if (ret)
 		return ret;
 
@@ -114,7 +236,7 @@ nv40_instmem_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	 * 0x18200-0x20000: padding
 	 */
 	ret = nvkm_gpuobj_new(nv_object(imem), NULL, 0x08000, 0, 0,
-			      &imem->ramro);
+			      &imem->base.ramro);
 	if (ret)
 		return ret;
 
@@ -122,7 +244,7 @@ nv40_instmem_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	 * 0x21000-0x40000: padding and some unknown crap
 	 */
 	ret = nvkm_gpuobj_new(nv_object(imem), NULL, 0x20000, 0,
-			      NVOBJ_FLAG_ZERO_ALLOC, &imem->ramfc);
+			      NVOBJ_FLAG_ZERO_ALLOC, &imem->base.ramfc);
 	if (ret)
 		return ret;
 
@@ -134,9 +256,11 @@ nv40_instmem_oclass = &(struct nvkm_instmem_impl) {
 	.base.handle = NV_SUBDEV(INSTMEM, 0x40),
 	.base.ofuncs = &(struct nvkm_ofuncs) {
 		.ctor = nv40_instmem_ctor,
-		.dtor = nv04_instmem_dtor,
+		.dtor = nv40_instmem_dtor,
 		.init = _nvkm_instmem_init,
 		.fini = _nvkm_instmem_fini,
 	},
-	.instobj = &nv04_instobj_oclass.base,
+	.memory_new = nv40_instobj_new,
+	.persistent = false,
+	.zero = false,
 }.base;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c
index 4f6354df538a4..535a8f9c23ce9 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c
@@ -21,115 +21,201 @@
  *
  * Authors: Ben Skeggs
  */
+#define nv50_instmem(p) container_of((p), struct nv50_instmem, base)
 #include "priv.h"
 
+#include <core/memory.h>
+#include <subdev/bar.h>
 #include <subdev/fb.h>
+#include <subdev/mmu.h>
 
 struct nv50_instmem {
 	struct nvkm_instmem base;
+	unsigned long lock_flags;
 	spinlock_t lock;
 	u64 addr;
 };
 
+/******************************************************************************
+ * instmem object implementation
+ *****************************************************************************/
+#define nv50_instobj(p) container_of((p), struct nv50_instobj, memory)
+
 struct nv50_instobj {
-	struct nvkm_instobj base;
+	struct nvkm_memory memory;
+	struct nv50_instmem *imem;
 	struct nvkm_mem *mem;
+	struct nvkm_vma bar;
+	void *map;
 };
 
-/******************************************************************************
- * instmem object implementation
- *****************************************************************************/
+static enum nvkm_memory_target
+nv50_instobj_target(struct nvkm_memory *memory)
+{
+	return NVKM_MEM_TARGET_VRAM;
+}
+
+static u64
+nv50_instobj_addr(struct nvkm_memory *memory)
+{
+	return nv50_instobj(memory)->mem->offset;
+}
+
+static u64
+nv50_instobj_size(struct nvkm_memory *memory)
+{
+	return (u64)nv50_instobj(memory)->mem->size << NVKM_RAM_MM_SHIFT;
+}
+
+static void
+nv50_instobj_boot(struct nvkm_memory *memory, struct nvkm_vm *vm)
+{
+	struct nv50_instobj *iobj = nv50_instobj(memory);
+	struct nvkm_subdev *subdev = &iobj->imem->base.subdev;
+	struct nvkm_device *device = subdev->device;
+	u64 size = nvkm_memory_size(memory);
+	void __iomem *map;
+	int ret;
+
+	iobj->map = ERR_PTR(-ENOMEM);
+
+	ret = nvkm_vm_get(vm, size, 12, NV_MEM_ACCESS_RW, &iobj->bar);
+	if (ret == 0) {
+		map = ioremap(nv_device_resource_start(device, 3) +
+			      (u32)iobj->bar.offset, size);
+		if (map) {
+			nvkm_memory_map(memory, &iobj->bar, 0);
+			iobj->map = map;
+		} else {
+			nvkm_warn(subdev, "PRAMIN ioremap failed\n");
+			nvkm_vm_put(&iobj->bar);
+		}
+	} else {
+		nvkm_warn(subdev, "PRAMIN exhausted\n");
+	}
+}
+
+static void
+nv50_instobj_release(struct nvkm_memory *memory)
+{
+	struct nv50_instmem *imem = nv50_instobj(memory)->imem;
+	spin_unlock_irqrestore(&imem->lock, imem->lock_flags);
+}
+
+static void __iomem *
+nv50_instobj_acquire(struct nvkm_memory *memory)
+{
+	struct nv50_instobj *iobj = nv50_instobj(memory);
+	struct nv50_instmem *imem = iobj->imem;
+	struct nvkm_bar *bar = imem->base.subdev.device->bar;
+	struct nvkm_vm *vm;
+	unsigned long flags;
+
+	if (!iobj->map && bar && bar->kmap && (vm = bar->kmap(bar)))
+		nvkm_memory_boot(memory, vm);
+	if (!IS_ERR_OR_NULL(iobj->map))
+		return iobj->map;
+
+	spin_lock_irqsave(&imem->lock, flags);
+	imem->lock_flags = flags;
+	return NULL;
+}
 
 static u32
-nv50_instobj_rd32(struct nvkm_object *object, u64 offset)
+nv50_instobj_rd32(struct nvkm_memory *memory, u64 offset)
 {
-	struct nv50_instmem *imem = (void *)nvkm_instmem(object);
-	struct nv50_instobj *node = (void *)object;
+	struct nv50_instobj *iobj = nv50_instobj(memory);
+	struct nv50_instmem *imem = iobj->imem;
 	struct nvkm_device *device = imem->base.subdev.device;
-	unsigned long flags;
-	u64 base = (node->mem->offset + offset) & 0xffffff00000ULL;
-	u64 addr = (node->mem->offset + offset) & 0x000000fffffULL;
+	u64 base = (iobj->mem->offset + offset) & 0xffffff00000ULL;
+	u64 addr = (iobj->mem->offset + offset) & 0x000000fffffULL;
 	u32 data;
 
-	spin_lock_irqsave(&imem->lock, flags);
 	if (unlikely(imem->addr != base)) {
 		nvkm_wr32(device, 0x001700, base >> 16);
 		imem->addr = base;
 	}
 	data = nvkm_rd32(device, 0x700000 + addr);
-	spin_unlock_irqrestore(&imem->lock, flags);
 	return data;
 }
 
 static void
-nv50_instobj_wr32(struct nvkm_object *object, u64 offset, u32 data)
+nv50_instobj_wr32(struct nvkm_memory *memory, u64 offset, u32 data)
 {
-	struct nv50_instmem *imem = (void *)nvkm_instmem(object);
-	struct nv50_instobj *node = (void *)object;
+	struct nv50_instobj *iobj = nv50_instobj(memory);
+	struct nv50_instmem *imem = iobj->imem;
 	struct nvkm_device *device = imem->base.subdev.device;
-	unsigned long flags;
-	u64 base = (node->mem->offset + offset) & 0xffffff00000ULL;
-	u64 addr = (node->mem->offset + offset) & 0x000000fffffULL;
+	u64 base = (iobj->mem->offset + offset) & 0xffffff00000ULL;
+	u64 addr = (iobj->mem->offset + offset) & 0x000000fffffULL;
 
-	spin_lock_irqsave(&imem->lock, flags);
 	if (unlikely(imem->addr != base)) {
 		nvkm_wr32(device, 0x001700, base >> 16);
 		imem->addr = base;
 	}
 	nvkm_wr32(device, 0x700000 + addr, data);
-	spin_unlock_irqrestore(&imem->lock, flags);
 }
 
 static void
-nv50_instobj_dtor(struct nvkm_object *object)
+nv50_instobj_map(struct nvkm_memory *memory, struct nvkm_vma *vma, u64 offset)
 {
-	struct nv50_instobj *node = (void *)object;
-	struct nvkm_ram *ram = nvkm_fb(object)->ram;
-	ram->func->put(ram, &node->mem);
-	nvkm_instobj_destroy(&node->base);
+	struct nv50_instobj *iobj = nv50_instobj(memory);
+	nvkm_vm_map_at(vma, offset, iobj->mem);
 }
 
+static void *
+nv50_instobj_dtor(struct nvkm_memory *memory)
+{
+	struct nv50_instobj *iobj = nv50_instobj(memory);
+	struct nvkm_ram *ram = iobj->imem->base.subdev.device->fb->ram;
+	if (!IS_ERR_OR_NULL(iobj->map)) {
+		nvkm_vm_put(&iobj->bar);
+		iounmap(iobj->map);
+	}
+	ram->func->put(ram, &iobj->mem);
+	return iobj;
+}
+
+static const struct nvkm_memory_func
+nv50_instobj_func = {
+	.dtor = nv50_instobj_dtor,
+	.target = nv50_instobj_target,
+	.size = nv50_instobj_size,
+	.addr = nv50_instobj_addr,
+	.boot = nv50_instobj_boot,
+	.acquire = nv50_instobj_acquire,
+	.release = nv50_instobj_release,
+	.rd32 = nv50_instobj_rd32,
+	.wr32 = nv50_instobj_wr32,
+	.map = nv50_instobj_map,
+};
+
 static int
-nv50_instobj_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
-		  struct nvkm_oclass *oclass, void *data, u32 size,
-		  struct nvkm_object **pobject)
+nv50_instobj_new(struct nvkm_instmem *base, u32 size, u32 align, bool zero,
+		 struct nvkm_memory **pmemory)
 {
-	struct nvkm_ram *ram = nvkm_fb(parent)->ram;
-	struct nvkm_instobj_args *args = data;
-	struct nv50_instobj *node;
+	struct nv50_instmem *imem = nv50_instmem(base);
+	struct nv50_instobj *iobj;
+	struct nvkm_ram *ram = imem->base.subdev.device->fb->ram;
 	int ret;
 
-	args->size  = max((args->size  + 4095) & ~4095, (u32)4096);
-	args->align = max((args->align + 4095) & ~4095, (u32)4096);
+	if (!(iobj = kzalloc(sizeof(*iobj), GFP_KERNEL)))
+		return -ENOMEM;
+	*pmemory = &iobj->memory;
 
-	ret = nvkm_instobj_create(parent, engine, oclass, &node);
-	*pobject = nv_object(node);
-	if (ret)
-		return ret;
+	nvkm_memory_ctor(&nv50_instobj_func, &iobj->memory);
+	iobj->imem = imem;
+
+	size  = max((size  + 4095) & ~4095, (u32)4096);
+	align = max((align + 4095) & ~4095, (u32)4096);
 
-	ret = ram->func->get(ram, args->size, args->align, 0, 0x800,
-			     &node->mem);
+	ret = ram->func->get(ram, size, align, 0, 0x800, &iobj->mem);
 	if (ret)
 		return ret;
 
-	node->base.addr = node->mem->offset;
-	node->base.size = node->mem->size << 12;
-	node->mem->page_shift = 12;
+	iobj->mem->page_shift = 12;
 	return 0;
 }
 
-static struct nvkm_instobj_impl
-nv50_instobj_oclass = {
-	.base.ofuncs = &(struct nvkm_ofuncs) {
-		.ctor = nv50_instobj_ctor,
-		.dtor = nv50_instobj_dtor,
-		.init = _nvkm_instobj_init,
-		.fini = _nvkm_instobj_fini,
-		.rd32 = nv50_instobj_rd32,
-		.wr32 = nv50_instobj_wr32,
-	},
-};
-
 /******************************************************************************
  * instmem subdev implementation
  *****************************************************************************/
@@ -168,5 +254,7 @@ nv50_instmem_oclass = &(struct nvkm_instmem_impl) {
 		.init = _nvkm_instmem_init,
 		.fini = nv50_instmem_fini,
 	},
-	.instobj = &nv50_instobj_oclass.base,
+	.memory_new = nv50_instobj_new,
+	.persistent = false,
+	.zero = false,
 }.base;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/priv.h
index 819f615782cf8..2b6d73005767b 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/priv.h
@@ -2,35 +2,12 @@
 #define __NVKM_INSTMEM_PRIV_H__
 #include <subdev/instmem.h>
 
-struct nvkm_instobj_impl {
-	struct nvkm_oclass base;
-};
-
-struct nvkm_instobj_args {
-	u32 size;
-	u32 align;
-};
-
-#define nvkm_instobj_create(p,e,o,d)                                        \
-	nvkm_instobj_create_((p), (e), (o), sizeof(**d), (void **)d)
-#define nvkm_instobj_destroy(p) ({                                          \
-	struct nvkm_instobj *iobj = (p);                                    \
-	_nvkm_instobj_dtor(nv_object(iobj));                                \
-})
-#define nvkm_instobj_init(p)                                                \
-	_nvkm_object_init(&(p)->base)
-#define nvkm_instobj_fini(p,s)                                              \
-	_nvkm_object_fini(&(p)->base, (s))
-
-int  nvkm_instobj_create_(struct nvkm_object *, struct nvkm_object *,
-			     struct nvkm_oclass *, int, void **);
-void _nvkm_instobj_dtor(struct nvkm_object *);
-#define _nvkm_instobj_init _nvkm_object_init
-#define _nvkm_instobj_fini _nvkm_object_fini
-
 struct nvkm_instmem_impl {
 	struct nvkm_oclass base;
-	struct nvkm_oclass *instobj;
+	int (*memory_new)(struct nvkm_instmem *, u32 size, u32 align,
+			  bool zero, struct nvkm_memory **);
+	bool persistent;
+	bool zero;
 };
 
 #define nvkm_instmem_create(p,e,o,d)                                        \
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
index e81d3170325f6..35b6d33f6669c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
@@ -339,6 +339,25 @@ nvkm_vm_put(struct nvkm_vma *vma)
 }
 
 int
+nvkm_vm_boot(struct nvkm_vm *vm, u64 size)
+{
+	struct nvkm_mmu *mmu = vm->mmu;
+	struct nvkm_gpuobj *pgt;
+	int ret;
+
+	ret = nvkm_gpuobj_new(nv_object(mmu), NULL,
+			      (size >> mmu->spg_shift) * 8, 0x1000,
+			      NVOBJ_FLAG_ZERO_ALLOC, &pgt);
+	if (ret == 0) {
+		vm->pgt[0].refcount[0] = 1;
+		vm->pgt[0].obj[0] = pgt;
+		nvkm_memory_boot(pgt->memory, vm);
+	}
+
+	return ret;
+}
+
+int
 nvkm_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset,
 	       u32 block, struct lock_class_key *key, struct nvkm_vm **pvm)
 {
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/gf100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/gf100.c
index e801e57946ad8..12d6ef461c62c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/gf100.c
@@ -22,7 +22,6 @@
  * Authors: Ben Skeggs
  */
 #include <subdev/mmu.h>
-#include <subdev/bar.h>
 #include <subdev/fb.h>
 #include <subdev/ltc.h>
 #include <subdev/timer.h>
@@ -163,12 +162,9 @@ gf100_vm_flush(struct nvkm_vm *vm)
 {
 	struct nvkm_mmu *mmu = (void *)vm->mmu;
 	struct nvkm_device *device = mmu->subdev.device;
-	struct nvkm_bar *bar = device->bar;
 	struct nvkm_vm_pgd *vpgd;
 	u32 type;
 
-	bar->flush(bar);
-
 	type = 0x00000001; /* PAGE_ALL */
 	if (atomic_read(&vm->engref[NVDEV_SUBDEV_BAR]))
 		type |= 0x00000004; /* HUB_ONLY */
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c
index b87fef9ee1988..cd63b9288507d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c
@@ -22,7 +22,6 @@
  * Authors: Ben Skeggs
  */
 #include <subdev/mmu.h>
-#include <subdev/bar.h>
 #include <subdev/fb.h>
 #include <subdev/timer.h>
 
@@ -156,12 +155,9 @@ nv50_vm_flush(struct nvkm_vm *vm)
 	struct nvkm_mmu *mmu = (void *)vm->mmu;
 	struct nvkm_subdev *subdev = &mmu->subdev;
 	struct nvkm_device *device = subdev->device;
-	struct nvkm_bar *bar = device->bar;
 	struct nvkm_engine *engine;
 	int i, vme;
 
-	bar->flush(bar);
-
 	mutex_lock(&subdev->mutex);
 	for (i = 0; i < NVDEV_SUBDEV_NR; i++) {
 		if (!atomic_read(&vm->engref[i]))