drm/nouveau: separate constant-va tracking from nvkm vma structure

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

21 files changed, 245 insertions, 151 deletions

diff --git a/drivers/gpu/drm/nouveau/Kbuild b/drivers/gpu/drm/nouveau/Kbuild
index 3476b61..9c0c650 100644
--- a/drivers/gpu/drm/nouveau/Kbuild
+++ b/drivers/gpu/drm/nouveau/Kbuild
@@ -34,6 +34,7 @@ nouveau-y += nouveau_mem.o
 nouveau-y += nouveau_prime.o
 nouveau-y += nouveau_sgdma.o
 nouveau-y += nouveau_ttm.o
+nouveau-y += nouveau_vmm.o
 
 # DRM - modesetting
 nouveau-$(CONFIG_DRM_NOUVEAU_BACKLIGHT) += nouveau_backlight.o
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/mmu.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/mmu.h
index 439938c..549a878 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/mmu.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/mmu.h
@@ -16,8 +16,6 @@ struct nvkm_vm_pgd {
 };
 
 struct nvkm_vma {
-	struct list_head head;
-	int refcount;
 	struct nvkm_vm *vm;
 	struct nvkm_mm_node *node;
 	union {
diff --git a/drivers/gpu/drm/nouveau/nouveau_abi16.c b/drivers/gpu/drm/nouveau/nouveau_abi16.c
index f98f800..9b0b210 100644
--- a/drivers/gpu/drm/nouveau/nouveau_abi16.c
+++ b/drivers/gpu/drm/nouveau/nouveau_abi16.c
@@ -34,6 +34,7 @@
 #include "nouveau_gem.h"
 #include "nouveau_chan.h"
 #include "nouveau_abi16.h"
+#include "nouveau_vmm.h"
 
 static struct nouveau_abi16 *
 nouveau_abi16(struct drm_file *file_priv)
@@ -134,7 +135,7 @@ nouveau_abi16_chan_fini(struct nouveau_abi16 *abi16,
 	}
 
 	if (chan->ntfy) {
-		nouveau_bo_vma_del(chan->ntfy, &chan->ntfy_vma);
+		nouveau_vma_del(&chan->ntfy_vma);
 		nouveau_bo_unpin(chan->ntfy);
 		drm_gem_object_unreference_unlocked(&chan->ntfy->gem);
 	}
@@ -329,8 +330,7 @@ nouveau_abi16_ioctl_channel_alloc(ABI16_IOCTL_ARGS)
 		goto done;
 
 	if (device->info.family >= NV_DEVICE_INFO_V0_TESLA) {
-		ret = nouveau_bo_vma_add(chan->ntfy, cli->vm,
-					&chan->ntfy_vma);
+		ret = nouveau_vma_new(chan->ntfy, &cli->vmm, &chan->ntfy_vma);
 		if (ret)
 			goto done;
 	}
@@ -548,8 +548,8 @@ nouveau_abi16_ioctl_notifierobj_alloc(ABI16_IOCTL_ARGS)
 	if (device->info.family >= NV_DEVICE_INFO_V0_TESLA) {
 		args.target = NV_DMA_V0_TARGET_VM;
 		args.access = NV_DMA_V0_ACCESS_VM;
-		args.start += chan->ntfy_vma.offset;
-		args.limit += chan->ntfy_vma.offset;
+		args.start += chan->ntfy_vma->addr;
+		args.limit += chan->ntfy_vma->addr;
 	} else
 	if (drm->agp.bridge) {
 		args.target = NV_DMA_V0_TARGET_AGP;
diff --git a/drivers/gpu/drm/nouveau/nouveau_abi16.h b/drivers/gpu/drm/nouveau/nouveau_abi16.h
index 841cc55..3277476 100644
--- a/drivers/gpu/drm/nouveau/nouveau_abi16.h
+++ b/drivers/gpu/drm/nouveau/nouveau_abi16.h
@@ -23,7 +23,7 @@ struct nouveau_abi16_chan {
 	struct nouveau_channel *chan;
 	struct list_head notifiers;
 	struct nouveau_bo *ntfy;
-	struct nvkm_vma ntfy_vma;
+	struct nouveau_vma *ntfy_vma;
 	struct nvkm_mm  heap;
 };
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 415c0e3..d89f3e7 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -38,6 +38,7 @@
 #include "nouveau_ttm.h"
 #include "nouveau_gem.h"
 #include "nouveau_mem.h"
+#include "nouveau_vmm.h"
 
 /*
  * NV10-NV40 tiling helpers
@@ -1223,7 +1224,7 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
 {
 	struct nouveau_mem *mem = new_reg ? nouveau_mem(new_reg) : NULL;
 	struct nouveau_bo *nvbo = nouveau_bo(bo);
-	struct nvkm_vma *vma;
+	struct nouveau_vma *vma;
 
 	/* ttm can now (stupidly) pass the driver bos it didn't create... */
 	if (bo->destroy != nouveau_bo_del_ttm)
@@ -1232,12 +1233,12 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
 	if (mem && new_reg->mem_type != TTM_PL_SYSTEM &&
 	    mem->mem.page == nvbo->page) {
 		list_for_each_entry(vma, &nvbo->vma_list, head) {
-			nvkm_vm_map(vma, mem->_mem);
+			nouveau_vma_map(vma, mem);
 		}
 	} else {
 		list_for_each_entry(vma, &nvbo->vma_list, head) {
 			WARN_ON(ttm_bo_wait(bo, false, false));
-			nvkm_vm_unmap(vma);
+			nouveau_vma_unmap(vma);
 		}
 	}
 }
@@ -1599,47 +1600,3 @@ struct ttm_bo_driver nouveau_bo_driver = {
 	.io_mem_free = &nouveau_ttm_io_mem_free,
 	.io_mem_pfn = ttm_bo_default_io_mem_pfn,
 };
-
-struct nvkm_vma *
-nouveau_bo_vma_find(struct nouveau_bo *nvbo, struct nvkm_vm *vm)
-{
-	struct nvkm_vma *vma;
-	list_for_each_entry(vma, &nvbo->vma_list, head) {
-		if (vma->vm == vm)
-			return vma;
-	}
-
-	return NULL;
-}
-
-int
-nouveau_bo_vma_add(struct nouveau_bo *nvbo, struct nvkm_vm *vm,
-		   struct nvkm_vma *vma)
-{
-	const u32 size = nvbo->bo.mem.num_pages << PAGE_SHIFT;
-	struct nouveau_mem *mem = nouveau_mem(&nvbo->bo.mem);
-	int ret;
-
-	ret = nvkm_vm_get(vm, size, nvbo->page, NV_MEM_ACCESS_RW, vma);
-	if (ret)
-		return ret;
-
-	if (nvbo->bo.mem.mem_type != TTM_PL_SYSTEM &&
-	    mem->mem.page == nvbo->page)
-		nvkm_vm_map(vma, mem->_mem);
-
-	list_add_tail(&vma->head, &nvbo->vma_list);
-	vma->refcount = 1;
-	return 0;
-}
-
-void
-nouveau_bo_vma_del(struct nouveau_bo *nvbo, struct nvkm_vma *vma)
-{
-	if (vma->node) {
-		if (nvbo->bo.mem.mem_type != TTM_PL_SYSTEM)
-			nvkm_vm_unmap(vma);
-		nvkm_vm_put(vma);
-		list_del(&vma->head);
-	}
-}
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h
index a179bbf..23002bd 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.h
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.h
@@ -93,13 +93,6 @@ int  nouveau_bo_validate(struct nouveau_bo *, bool interruptible,
 void nouveau_bo_sync_for_device(struct nouveau_bo *nvbo);
 void nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo);
 
-struct nvkm_vma *
-nouveau_bo_vma_find(struct nouveau_bo *, struct nvkm_vm *);
-
-int  nouveau_bo_vma_add(struct nouveau_bo *, struct nvkm_vm *,
-			struct nvkm_vma *);
-void nouveau_bo_vma_del(struct nouveau_bo *, struct nvkm_vma *);
-
 /* TODO: submit equivalent to TTM generic API upstream? */
 static inline void __iomem *
 nvbo_kmap_obj_iovirtual(struct nouveau_bo *nvbo)
diff --git a/drivers/gpu/drm/nouveau/nouveau_chan.c b/drivers/gpu/drm/nouveau/nouveau_chan.c
index 4af09e9..9fdca7b 100644
--- a/drivers/gpu/drm/nouveau/nouveau_chan.c
+++ b/drivers/gpu/drm/nouveau/nouveau_chan.c
@@ -40,6 +40,7 @@
 #include "nouveau_chan.h"
 #include "nouveau_fence.h"
 #include "nouveau_abi16.h"
+#include "nouveau_vmm.h"
 
 MODULE_PARM_DESC(vram_pushbuf, "Create DMA push buffers in VRAM");
 int nouveau_vram_pushbuf;
@@ -91,7 +92,7 @@ nouveau_channel_del(struct nouveau_channel **pchan)
 		nvif_notify_fini(&chan->kill);
 		nvif_object_fini(&chan->user);
 		nvif_object_fini(&chan->push.ctxdma);
-		nouveau_bo_vma_del(chan->push.buffer, &chan->push.vma);
+		nouveau_vma_del(&chan->push.vma);
 		nouveau_bo_unmap(chan->push.buffer);
 		if (chan->push.buffer && chan->push.buffer->pin_refcnt)
 			nouveau_bo_unpin(chan->push.buffer);
@@ -142,11 +143,11 @@ nouveau_channel_prep(struct nouveau_drm *drm, struct nvif_device *device,
 	 * pushbuf lives in, this is because the GEM code requires that
 	 * we be able to call out to other (indirect) push buffers
 	 */
-	chan->push.vma.offset = chan->push.buffer->bo.offset;
+	chan->push.addr = chan->push.buffer->bo.offset;
 
 	if (device->info.family >= NV_DEVICE_INFO_V0_TESLA) {
-		ret = nouveau_bo_vma_add(chan->push.buffer, cli->vm,
-					&chan->push.vma);
+		ret = nouveau_vma_new(chan->push.buffer, &cli->vmm,
+				      &chan->push.vma);
 		if (ret) {
 			nouveau_channel_del(pchan);
 			return ret;
@@ -156,6 +157,8 @@ nouveau_channel_prep(struct nouveau_drm *drm, struct nvif_device *device,
 		args.access = NV_DMA_V0_ACCESS_VM;
 		args.start = 0;
 		args.limit = cli->vm->mmu->limit - 1;
+
+		chan->push.addr = chan->push.vma->addr;
 	} else
 	if (chan->push.buffer->bo.mem.mem_type == TTM_PL_VRAM) {
 		if (device->info.family == NV_DEVICE_INFO_V0_TNT) {
@@ -233,20 +236,20 @@ nouveau_channel_ind(struct nouveau_drm *drm, struct nvif_device *device,
 			args.kepler.version = 0;
 			args.kepler.engines = engine;
 			args.kepler.ilength = 0x02000;
-			args.kepler.ioffset = 0x10000 + chan->push.vma.offset;
+			args.kepler.ioffset = 0x10000 + chan->push.addr;
 			args.kepler.vm = 0;
 			size = sizeof(args.kepler);
 		} else
 		if (oclass[0] >= FERMI_CHANNEL_GPFIFO) {
 			args.fermi.version = 0;
 			args.fermi.ilength = 0x02000;
-			args.fermi.ioffset = 0x10000 + chan->push.vma.offset;
+			args.fermi.ioffset = 0x10000 + chan->push.addr;
 			args.fermi.vm = 0;
 			size = sizeof(args.fermi);
 		} else {
 			args.nv50.version = 0;
 			args.nv50.ilength = 0x02000;
-			args.nv50.ioffset = 0x10000 + chan->push.vma.offset;
+			args.nv50.ioffset = 0x10000 + chan->push.addr;
 			args.nv50.pushbuf = nvif_handle(&chan->push.ctxdma);
 			args.nv50.vm = 0;
 			size = sizeof(args.nv50);
@@ -293,7 +296,7 @@ nouveau_channel_dma(struct nouveau_drm *drm, struct nvif_device *device,
 	/* create channel object */
 	args.version = 0;
 	args.pushbuf = nvif_handle(&chan->push.ctxdma);
-	args.offset = chan->push.vma.offset;
+	args.offset = chan->push.addr;
 
 	do {
 		ret = nvif_object_init(&device->object, 0, *oclass++,
diff --git a/drivers/gpu/drm/nouveau/nouveau_chan.h b/drivers/gpu/drm/nouveau/nouveau_chan.h
index 46b947b..f29d3a7 100644
--- a/drivers/gpu/drm/nouveau/nouveau_chan.h
+++ b/drivers/gpu/drm/nouveau/nouveau_chan.h
@@ -16,8 +16,9 @@ struct nouveau_channel {
 
 	struct {
 		struct nouveau_bo *buffer;
-		struct nvkm_vma vma;
+		struct nouveau_vma *vma;
 		struct nvif_object ctxdma;
+		u64 addr;
 	} push;
 
 	/* TODO: this will be reworked in the near future */
diff --git a/drivers/gpu/drm/nouveau/nouveau_display.h b/drivers/gpu/drm/nouveau/nouveau_display.h
index 201aec2..1411bf0 100644
--- a/drivers/gpu/drm/nouveau/nouveau_display.h
+++ b/drivers/gpu/drm/nouveau/nouveau_display.h
@@ -1,14 +1,11 @@
 #ifndef __NOUVEAU_DISPLAY_H__
 #define __NOUVEAU_DISPLAY_H__
-
-#include <subdev/mmu.h>
-
 #include "nouveau_drv.h"
 
 struct nouveau_framebuffer {
 	struct drm_framebuffer base;
 	struct nouveau_bo *nvbo;
-	struct nvkm_vma vma;
+	struct nouveau_vma *vma;
 	u32 r_handle;
 	u32 r_format;
 	u32 r_pitch;
diff --git a/drivers/gpu/drm/nouveau/nouveau_dma.c b/drivers/gpu/drm/nouveau/nouveau_dma.c
index 2634a1a..10e84f6 100644
--- a/drivers/gpu/drm/nouveau/nouveau_dma.c
+++ b/drivers/gpu/drm/nouveau/nouveau_dma.c
@@ -26,6 +26,7 @@
 
 #include "nouveau_drv.h"
 #include "nouveau_dma.h"
+#include "nouveau_vmm.h"
 
 void
 OUT_RINGp(struct nouveau_channel *chan, const void *data, unsigned nr_dwords)
@@ -71,11 +72,11 @@ READ_GET(struct nouveau_channel *chan, uint64_t *prev_get, int *timeout)
 			return -EBUSY;
 	}
 
-	if (val < chan->push.vma.offset ||
-	    val > chan->push.vma.offset + (chan->dma.max << 2))
+	if (val < chan->push.addr ||
+	    val > chan->push.addr + (chan->dma.max << 2))
 		return -EINVAL;
 
-	return (val - chan->push.vma.offset) >> 2;
+	return (val - chan->push.addr) >> 2;
 }
 
 void
@@ -84,13 +85,13 @@ nv50_dma_push(struct nouveau_channel *chan, struct nouveau_bo *bo,
 {
 	struct nouveau_cli *cli = (void *)chan->user.client;
 	struct nouveau_bo *pb = chan->push.buffer;
-	struct nvkm_vma *vma;
+	struct nouveau_vma *vma;
 	int ip = (chan->dma.ib_put * 2) + chan->dma.ib_base;
 	u64 offset;
 
-	vma = nouveau_bo_vma_find(bo, cli->vm);
+	vma = nouveau_vma_find(bo, &cli->vmm);
 	BUG_ON(!vma);
-	offset = vma->offset + delta;
+	offset = vma->addr + delta;
 
 	BUG_ON(chan->dma.ib_free < 1);
 
@@ -224,7 +225,7 @@ nouveau_dma_wait(struct nouveau_channel *chan, int slots, int size)
 			 * instruct the GPU to jump back to the start right
 			 * after processing the currently pending commands.
 			 */
-			OUT_RING(chan, chan->push.vma.offset | 0x20000000);
+			OUT_RING(chan, chan->push.addr | 0x20000000);
 
 			/* wait for GET to depart from the skips area.
 			 * prevents writing GET==PUT and causing a race
diff --git a/drivers/gpu/drm/nouveau/nouveau_dma.h b/drivers/gpu/drm/nouveau/nouveau_dma.h
index aff3a9d..74e10b1 100644
--- a/drivers/gpu/drm/nouveau/nouveau_dma.h
+++ b/drivers/gpu/drm/nouveau/nouveau_dma.h
@@ -140,7 +140,7 @@ BEGIN_IMC0(struct nouveau_channel *chan, int subc, int mthd, u16 data)
 #define WRITE_PUT(val) do {                                                    \
 	mb();                                                   \
 	nouveau_bo_rd32(chan->push.buffer, 0);                                 \
-	nvif_wr32(&chan->user, chan->user_put, ((val) << 2) + chan->push.vma.offset); \
+	nvif_wr32(&chan->user, chan->user_put, ((val) << 2) + chan->push.addr);\
 } while (0)
 
 static inline void
diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.c b/drivers/gpu/drm/nouveau/nouveau_drm.c
index a705e56..25a47e8 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drm.c
@@ -114,8 +114,8 @@ nouveau_name(struct drm_device *dev)
 static void
 nouveau_cli_fini(struct nouveau_cli *cli)
 {
-	nvkm_vm_ref(NULL, &nvxx_client(&cli->base)->vm, NULL);
 	usif_client_fini(cli);
+	nouveau_vmm_fini(&cli->vmm);
 	nvif_device_fini(&cli->device);
 	mutex_lock(&cli->drm->master.lock);
 	nvif_client_fini(&cli->base);
@@ -472,12 +472,11 @@ nouveau_drm_load(struct drm_device *dev, unsigned long flags)
 			goto fail_device;
 		}
 
-		ret = nvkm_vm_new(nvxx_device(&drm->client.device),
-				  0, (1ULL << 40), 0x1000, NULL,
-				  &drm->client.vm);
+		ret = nouveau_vmm_init(&drm->client, 0, &drm->client.vmm);
 		if (ret)
 			goto fail_device;
 
+		drm->client.vm = drm->client.vmm.vm;
 		nvxx_client(&drm->client.base)->vm = drm->client.vm;
 	}
 
@@ -863,11 +862,11 @@ nouveau_drm_open(struct drm_device *dev, struct drm_file *fpriv)
 	cli->base.super = false;
 
 	if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
-		ret = nvkm_vm_new(nvxx_device(&drm->client.device), 0,
-				  (1ULL << 40), 0x1000, NULL, &cli->vm);
+		ret = nouveau_vmm_init(cli, 0, &cli->vmm);
 		if (ret)
 			goto done;
 
+		cli->vm = cli->vmm.vm;
 		nvxx_client(&cli->base)->vm = cli->vm;
 	}
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index 6102459..ba8a9ab 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -61,6 +61,7 @@ struct platform_device;
 
 #include "nouveau_fence.h"
 #include "nouveau_bios.h"
+#include "nouveau_vmm.h"
 
 struct nouveau_drm_tile {
 	struct nouveau_fence *fence;
@@ -90,8 +91,9 @@ struct nouveau_cli {
 	struct mutex mutex;
 
 	struct nvif_device device;
+	struct nouveau_vmm vmm;
 
-	struct nvkm_vm *vm; /*XXX*/
+	struct nvkm_vm *vm;
 	struct list_head head;
 	void *abi16;
 	struct list_head objects;
diff --git a/drivers/gpu/drm/nouveau/nouveau_fbcon.c b/drivers/gpu/drm/nouveau/nouveau_fbcon.c
index 2b12d82..c533d8e 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fbcon.c
@@ -48,6 +48,7 @@
 #include "nouveau_bo.h"
 #include "nouveau_fbcon.h"
 #include "nouveau_chan.h"
+#include "nouveau_vmm.h"
 
 #include "nouveau_crtc.h"
 
@@ -348,7 +349,7 @@ nouveau_fbcon_create(struct drm_fb_helper *helper,
 
 	chan = nouveau_nofbaccel ? NULL : drm->channel;
 	if (chan && device->info.family >= NV_DEVICE_INFO_V0_TESLA) {
-		ret = nouveau_bo_vma_add(nvbo, drm->client.vm, &fb->vma);
+		ret = nouveau_vma_new(nvbo, &drm->client.vmm, &fb->vma);
 		if (ret) {
 			NV_ERROR(drm, "failed to map fb into chan: %d\n", ret);
 			chan = NULL;
@@ -402,7 +403,7 @@ nouveau_fbcon_create(struct drm_fb_helper *helper,
 
 out_unlock:
 	if (chan)
-		nouveau_bo_vma_del(fb->nvbo, &fb->vma);
+		nouveau_vma_del(&fb->vma);
 	nouveau_bo_unmap(fb->nvbo);
 out_unpin:
 	nouveau_bo_unpin(fb->nvbo);
@@ -429,7 +430,7 @@ nouveau_fbcon_destroy(struct drm_device *dev, struct nouveau_fbdev *fbcon)
 	drm_fb_helper_fini(&fbcon->helper);
 
 	if (nouveau_fb->nvbo) {
-		nouveau_bo_vma_del(nouveau_fb->nvbo, &nouveau_fb->vma);
+		nouveau_vma_del(&nouveau_fb->vma);
 		nouveau_bo_unmap(nouveau_fb->nvbo);
 		nouveau_bo_unpin(nouveau_fb->nvbo);
 		drm_framebuffer_unreference(&nouveau_fb->base);
diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.h b/drivers/gpu/drm/nouveau/nouveau_fence.h
index 28a2135..d32e58b 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fence.h
+++ b/drivers/gpu/drm/nouveau/nouveau_fence.h
@@ -88,7 +88,7 @@ int nouveau_flip_complete(struct nvif_notify *);
 
 struct nv84_fence_chan {
 	struct nouveau_fence_chan base;
-	struct nvkm_vma vma;
+	struct nouveau_vma *vma;
 };
 
 struct nv84_fence_priv {
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
index 2d92473..9f82b13 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -31,6 +31,7 @@
 
 #include "nouveau_ttm.h"
 #include "nouveau_gem.h"
+#include "nouveau_vmm.h"
 
 void
 nouveau_gem_object_del(struct drm_gem_object *gem)
@@ -64,8 +65,8 @@ nouveau_gem_object_open(struct drm_gem_object *gem, struct drm_file *file_priv)
 	struct nouveau_cli *cli = nouveau_cli(file_priv);
 	struct nouveau_bo *nvbo = nouveau_gem_object(gem);
 	struct nouveau_drm *drm = nouveau_bdev(nvbo->bo.bdev);
-	struct nvkm_vma *vma;
 	struct device *dev = drm->dev->dev;
+	struct nouveau_vma *vma;
 	int ret;
 
 	if (!cli->vm)
@@ -75,30 +76,13 @@ nouveau_gem_object_open(struct drm_gem_object *gem, struct drm_file *file_priv)
 	if (ret)
 		return ret;
 
-	vma = nouveau_bo_vma_find(nvbo, cli->vm);
-	if (!vma) {
-		vma = kzalloc(sizeof(*vma), GFP_KERNEL);
-		if (!vma) {
-			ret = -ENOMEM;
-			goto out;
-		}
-
-		ret = pm_runtime_get_sync(dev);
-		if (ret < 0 && ret != -EACCES) {
-			kfree(vma);
-			goto out;
-		}
-
-		ret = nouveau_bo_vma_add(nvbo, cli->vm, vma);
-		if (ret)
-			kfree(vma);
-
-		pm_runtime_mark_last_busy(dev);
-		pm_runtime_put_autosuspend(dev);
-	} else {
-		vma->refcount++;
-	}
+	ret = pm_runtime_get_sync(dev);
+	if (ret < 0 && ret != -EACCES)
+		goto out;
 
+	ret = nouveau_vma_new(nvbo, &cli->vmm, &vma);
+	pm_runtime_mark_last_busy(dev);
+	pm_runtime_put_autosuspend(dev);
 out:
 	ttm_bo_unreserve(&nvbo->bo);
 	return ret;
@@ -107,14 +91,12 @@ out:
 static void
 nouveau_gem_object_delete(void *data)
 {
-	struct nvkm_vma *vma = data;
-	nvkm_vm_unmap(vma);
-	nvkm_vm_put(vma);
-	kfree(vma);
+	struct nouveau_vma *vma = data;
+	nouveau_vma_del(&vma);
 }
 
 static void
-nouveau_gem_object_unmap(struct nouveau_bo *nvbo, struct nvkm_vma *vma)
+nouveau_gem_object_unmap(struct nouveau_bo *nvbo, struct nouveau_vma *vma)
 {
 	const bool mapped = nvbo->bo.mem.mem_type != TTM_PL_SYSTEM;
 	struct reservation_object *resv = nvbo->bo.resv;
@@ -123,7 +105,7 @@ nouveau_gem_object_unmap(struct nouveau_bo *nvbo, struct nvkm_vma *vma)
 
 	fobj = reservation_object_get_list(resv);
 
-	list_del(&vma->head);
+	list_del_init(&vma->head);
 
 	if (fobj && fobj->shared_count > 1)
 		ttm_bo_wait(&nvbo->bo, false, false);
@@ -133,14 +115,10 @@ nouveau_gem_object_unmap(struct nouveau_bo *nvbo, struct nvkm_vma *vma)
 	else
 		fence = reservation_object_get_excl(nvbo->bo.resv);
 
-	if (fence && mapped) {
+	if (fence && mapped)
 		nouveau_fence_work(fence, nouveau_gem_object_delete, vma);
-	} else {
-		if (mapped)
-			nvkm_vm_unmap(vma);
-		nvkm_vm_put(vma);
-		kfree(vma);
-	}
+	else
+		nouveau_vma_del(&vma);
 }
 
 void
@@ -150,7 +128,7 @@ nouveau_gem_object_close(struct drm_gem_object *gem, struct drm_file *file_priv)
 	struct nouveau_bo *nvbo = nouveau_gem_object(gem);
 	struct nouveau_drm *drm = nouveau_bdev(nvbo->bo.bdev);
 	struct device *dev = drm->dev->dev;
-	struct nvkm_vma *vma;
+	struct nouveau_vma *vma;
 	int ret;
 
 	if (!cli->vm)
@@ -160,9 +138,9 @@ nouveau_gem_object_close(struct drm_gem_object *gem, struct drm_file *file_priv)
 	if (ret)
 		return;
 
-	vma = nouveau_bo_vma_find(nvbo, cli->vm);
+	vma = nouveau_vma_find(nvbo, &cli->vmm);
 	if (vma) {
-		if (--vma->refcount == 0) {
+		if (--vma->refs == 0) {
 			ret = pm_runtime_get_sync(dev);
 			if (!WARN_ON(ret < 0 && ret != -EACCES)) {
 				nouveau_gem_object_unmap(nvbo, vma);
@@ -227,7 +205,7 @@ nouveau_gem_info(struct drm_file *file_priv, struct drm_gem_object *gem,
 {
 	struct nouveau_cli *cli = nouveau_cli(file_priv);
 	struct nouveau_bo *nvbo = nouveau_gem_object(gem);
-	struct nvkm_vma *vma;
+	struct nouveau_vma *vma;
 
 	if (is_power_of_2(nvbo->valid_domains))
 		rep->domain = nvbo->valid_domains;
@@ -237,11 +215,11 @@ nouveau_gem_info(struct drm_file *file_priv, struct drm_gem_object *gem,
 		rep->domain = NOUVEAU_GEM_DOMAIN_VRAM;
 	rep->offset = nvbo->bo.offset;
 	if (cli->vm) {
-		vma = nouveau_bo_vma_find(nvbo, cli->vm);
+		vma = nouveau_vma_find(nvbo, &cli->vmm);
 		if (!vma)
 			return -EINVAL;
 
-		rep->offset = vma->offset;
+		rep->offset = vma->addr;
 	}
 
 	rep->size = nvbo->bo.mem.num_pages << PAGE_SHIFT;
@@ -798,7 +776,7 @@ nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
 				bo[push[i].bo_index].user_priv;
 			uint32_t cmd;
 
-			cmd = chan->push.vma.offset + ((chan->dma.cur + 2) << 2);
+			cmd = chan->push.addr + ((chan->dma.cur + 2) << 2);
 			cmd |= 0x20000000;
 			if (unlikely(cmd != req->suffix0)) {
 				if (!nvbo->kmap.virtual) {
@@ -850,7 +828,7 @@ out_next:
 		req->suffix1 = 0x00000000;
 	} else {
 		req->suffix0 = 0x20000000 |
-			      (chan->push.vma.offset + ((chan->dma.cur + 2) << 2));
+			      (chan->push.addr + ((chan->dma.cur + 2) << 2));
 		req->suffix1 = 0x00000000;
 	}
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_vmm.c b/drivers/gpu/drm/nouveau/nouveau_vmm.c
new file mode 100644
index 0000000..855d549
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nouveau_vmm.c
@@ -0,0 +1,128 @@
+/*
+ * Copyright 2017 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.
+ */
+#include "nouveau_vmm.h"
+#include "nouveau_drv.h"
+#include "nouveau_bo.h"
+#include "nouveau_mem.h"
+
+void
+nouveau_vma_unmap(struct nouveau_vma *vma)
+{
+	if (vma->mem) {
+		nvkm_vm_unmap(&vma->_vma);
+		vma->mem = NULL;
+	}
+}
+
+int
+nouveau_vma_map(struct nouveau_vma *vma, struct nouveau_mem *mem)
+{
+	int ret = nouveau_mem_map(mem, vma->vmm->vm, &vma->_vma);
+	if (ret)
+		return ret;
+	vma->mem = mem;
+	return 0;
+}
+
+struct nouveau_vma *
+nouveau_vma_find(struct nouveau_bo *nvbo, struct nouveau_vmm *vmm)
+{
+	struct nouveau_vma *vma;
+
+	list_for_each_entry(vma, &nvbo->vma_list, head) {
+		if (vma->vmm == vmm)
+			return vma;
+	}
+
+	return NULL;
+}
+
+void
+nouveau_vma_del(struct nouveau_vma **pvma)
+{
+	struct nouveau_vma *vma = *pvma;
+	if (vma && --vma->refs <= 0) {
+		if (likely(vma->addr != ~0ULL)) {
+			nouveau_vma_unmap(vma);
+			nvkm_vm_put(&vma->_vma);
+		}
+		list_del(&vma->head);
+		*pvma = NULL;
+		kfree(*pvma);
+	}
+}
+
+int
+nouveau_vma_new(struct nouveau_bo *nvbo, struct nouveau_vmm *vmm,
+		struct nouveau_vma **pvma)
+{
+	struct nouveau_mem *mem = nouveau_mem(&nvbo->bo.mem);
+	struct nouveau_vma *vma;
+	int ret;
+
+	if ((vma = *pvma = nouveau_vma_find(nvbo, vmm))) {
+		vma->refs++;
+		return 0;
+	}
+
+	if (!(vma = *pvma = kmalloc(sizeof(*vma), GFP_KERNEL)))
+		return -ENOMEM;
+	vma->vmm = vmm;
+	vma->refs = 1;
+	vma->addr = ~0ULL;
+	vma->mem = NULL;
+	list_add_tail(&vma->head, &nvbo->vma_list);
+
+	if (nvbo->bo.mem.mem_type != TTM_PL_SYSTEM &&
+	    mem->mem.page == nvbo->page) {
+		ret = nvkm_vm_get(vmm->vm, mem->_mem->size << 12, mem->mem.page,
+				  NV_MEM_ACCESS_RW, &vma->_vma);
+		if (ret)
+			goto done;
+
+		vma->addr = vma->_vma.offset;
+		ret = nouveau_vma_map(vma, mem);
+	} else {
+		ret = nvkm_vm_get(vmm->vm, mem->_mem->size << 12, mem->mem.page,
+				  NV_MEM_ACCESS_RW, &vma->_vma);
+		vma->addr = vma->_vma.offset;
+	}
+
+done:
+	if (ret)
+		nouveau_vma_del(pvma);
+	return ret;
+}
+
+void
+nouveau_vmm_fini(struct nouveau_vmm *vmm)
+{
+	nvkm_vm_ref(NULL, &vmm->vm, NULL);
+}
+
+int
+nouveau_vmm_init(struct nouveau_cli *cli, s32 oclass, struct nouveau_vmm *vmm)
+{
+	vmm->cli = cli;
+	return nvkm_vm_new(nvxx_device(&cli->device), 0, (1ULL << 40),
+			   0x1000, NULL, &vmm->vm);
+}
diff --git a/drivers/gpu/drm/nouveau/nouveau_vmm.h b/drivers/gpu/drm/nouveau/nouveau_vmm.h
new file mode 100644
index 0000000..86ab599
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nouveau_vmm.h
@@ -0,0 +1,32 @@
+#ifndef __NOUVEAU_VMA_H__
+#define __NOUVEAU_VMA_H__
+#include <subdev/mmu.h>
+struct nouveau_bo;
+struct nouveau_mem;
+
+struct nouveau_vma {
+	struct nouveau_vmm *vmm;
+	int refs;
+	struct list_head head;
+	u64 addr;
+
+	struct nouveau_mem *mem;
+
+	struct nvkm_vma _vma;
+};
+
+struct nouveau_vma *nouveau_vma_find(struct nouveau_bo *, struct nouveau_vmm *);
+int nouveau_vma_new(struct nouveau_bo *, struct nouveau_vmm *,
+		    struct nouveau_vma **);
+void nouveau_vma_del(struct nouveau_vma **);
+int nouveau_vma_map(struct nouveau_vma *, struct nouveau_mem *);
+void nouveau_vma_unmap(struct nouveau_vma *);
+
+struct nouveau_vmm {
+	struct nouveau_cli *cli;
+	struct nvkm_vm *vm;
+};
+
+int nouveau_vmm_init(struct nouveau_cli *, s32 oclass, struct nouveau_vmm *);
+void nouveau_vmm_fini(struct nouveau_vmm *);
+#endif
diff --git a/drivers/gpu/drm/nouveau/nv50_fbcon.c b/drivers/gpu/drm/nouveau/nv50_fbcon.c
index 327dcd7..facd185 100644
--- a/drivers/gpu/drm/nouveau/nv50_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nv50_fbcon.c
@@ -25,6 +25,7 @@
 #include "nouveau_drv.h"
 #include "nouveau_dma.h"
 #include "nouveau_fbcon.h"
+#include "nouveau_vmm.h"
 
 int
 nv50_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect)
@@ -239,8 +240,8 @@ nv50_fbcon_accel_init(struct fb_info *info)
 	OUT_RING(chan, info->fix.line_length);
 	OUT_RING(chan, info->var.xres_virtual);
 	OUT_RING(chan, info->var.yres_virtual);
-	OUT_RING(chan, upper_32_bits(fb->vma.offset));
-	OUT_RING(chan, lower_32_bits(fb->vma.offset));
+	OUT_RING(chan, upper_32_bits(fb->vma->addr));
+	OUT_RING(chan, lower_32_bits(fb->vma->addr));
 	BEGIN_NV04(chan, NvSub2D, 0x0230, 2);
 	OUT_RING(chan, format);
 	OUT_RING(chan, 1);
@@ -248,8 +249,8 @@ nv50_fbcon_accel_init(struct fb_info *info)
 	OUT_RING(chan, info->fix.line_length);
 	OUT_RING(chan, info->var.xres_virtual);
 	OUT_RING(chan, info->var.yres_virtual);
-	OUT_RING(chan, upper_32_bits(fb->vma.offset));
-	OUT_RING(chan, lower_32_bits(fb->vma.offset));
+	OUT_RING(chan, upper_32_bits(fb->vma->addr));
+	OUT_RING(chan, lower_32_bits(fb->vma->addr));
 	FIRE_RING(chan);
 
 	return 0;
diff --git a/drivers/gpu/drm/nouveau/nv84_fence.c b/drivers/gpu/drm/nouveau/nv84_fence.c
index e5f6249..5f0c0c2 100644
--- a/drivers/gpu/drm/nouveau/nv84_fence.c
+++ b/drivers/gpu/drm/nouveau/nv84_fence.c
@@ -25,6 +25,7 @@
 #include "nouveau_drv.h"
 #include "nouveau_dma.h"
 #include "nouveau_fence.h"
+#include "nouveau_vmm.h"
 
 #include "nv50_display.h"
 
@@ -68,7 +69,7 @@ nv84_fence_emit(struct nouveau_fence *fence)
 {
 	struct nouveau_channel *chan = fence->channel;
 	struct nv84_fence_chan *fctx = chan->fence;
-	u64 addr = fctx->vma.offset + chan->chid * 16;
+	u64 addr = fctx->vma->addr + chan->chid * 16;
 
 	return fctx->base.emit32(chan, addr, fence->base.seqno);
 }
@@ -78,7 +79,7 @@ nv84_fence_sync(struct nouveau_fence *fence,
 		struct nouveau_channel *prev, struct nouveau_channel *chan)
 {
 	struct nv84_fence_chan *fctx = chan->fence;
-	u64 addr = fctx->vma.offset + prev->chid * 16;
+	u64 addr = fctx->vma->addr + prev->chid * 16;
 
 	return fctx->base.sync32(chan, addr, fence->base.seqno);
 }
@@ -98,7 +99,7 @@ nv84_fence_context_del(struct nouveau_channel *chan)
 
 	nouveau_bo_wr32(priv->bo, chan->chid * 16 / 4, fctx->base.sequence);
 	mutex_lock(&priv->mutex);
-	nouveau_bo_vma_del(priv->bo, &fctx->vma);
+	nouveau_vma_del(&fctx->vma);
 	mutex_unlock(&priv->mutex);
 	nouveau_fence_context_del(&fctx->base);
 	chan->fence = NULL;
@@ -126,7 +127,7 @@ nv84_fence_context_new(struct nouveau_channel *chan)
 	fctx->base.sequence = nv84_fence_read(chan);
 
 	mutex_lock(&priv->mutex);
-	ret = nouveau_bo_vma_add(priv->bo, cli->vm, &fctx->vma);
+	ret = nouveau_vma_new(priv->bo, &cli->vmm, &fctx->vma);
 	mutex_unlock(&priv->mutex);
 
 	if (ret)
diff --git a/drivers/gpu/drm/nouveau/nvc0_fbcon.c b/drivers/gpu/drm/nouveau/nvc0_fbcon.c
index 90f27bf..c0deef4 100644
--- a/drivers/gpu/drm/nouveau/nvc0_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nvc0_fbcon.c
@@ -25,6 +25,7 @@
 #include "nouveau_drv.h"
 #include "nouveau_dma.h"
 #include "nouveau_fbcon.h"
+#include "nouveau_vmm.h"
 
 int
 nvc0_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect)
@@ -239,8 +240,8 @@ nvc0_fbcon_accel_init(struct fb_info *info)
 	OUT_RING  (chan, info->fix.line_length);
 	OUT_RING  (chan, info->var.xres_virtual);
 	OUT_RING  (chan, info->var.yres_virtual);
-	OUT_RING  (chan, upper_32_bits(fb->vma.offset));
-	OUT_RING  (chan, lower_32_bits(fb->vma.offset));
+	OUT_RING  (chan, upper_32_bits(fb->vma->addr));
+	OUT_RING  (chan, lower_32_bits(fb->vma->addr));
 	BEGIN_NVC0(chan, NvSub2D, 0x0230, 10);
 	OUT_RING  (chan, format);
 	OUT_RING  (chan, 1);
@@ -250,8 +251,8 @@ nvc0_fbcon_accel_init(struct fb_info *info)
 	OUT_RING  (chan, info->fix.line_length);
 	OUT_RING  (chan, info->var.xres_virtual);
 	OUT_RING  (chan, info->var.yres_virtual);
-	OUT_RING  (chan, upper_32_bits(fb->vma.offset));
-	OUT_RING  (chan, lower_32_bits(fb->vma.offset));
+	OUT_RING  (chan, upper_32_bits(fb->vma->addr));
+	OUT_RING  (chan, lower_32_bits(fb->vma->addr));
 	FIRE_RING (chan);
 
 	return 0;