drm/nouveau/mmu: rename from vmmgr (no binary change)

Switch to NVIDIA's name for the device.

The namespace of NVKM is being changed to nvkm_ instead of nouveau_,
which will be used for the DRM part of the driver.  This is being
done in order to make it very clear as to what part of the driver a
given symbol belongs to, and as a minor step towards splitting the
DRM driver out to be able to stand on its own (for virt).

Because there's already a large amount of churn here anyway, this is
as good a time as any to also switch to NVIDIA's device and chipset
naming to ease collaboration with them.

A comparison of objdump disassemblies proves no code changes.

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

8 files changed, 1548 insertions, 0 deletions

diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/Kbuild b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/Kbuild
new file mode 100644
index 000000000000..729e27f2cbd4
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/Kbuild
@@ -0,0 +1,6 @@
+nvkm-y += nvkm/subdev/mmu/base.o
+nvkm-y += nvkm/subdev/mmu/nv04.o
+nvkm-y += nvkm/subdev/mmu/nv41.o
+nvkm-y += nvkm/subdev/mmu/nv44.o
+nvkm-y += nvkm/subdev/mmu/nv50.o
+nvkm-y += nvkm/subdev/mmu/nvc0.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
new file mode 100644
index 000000000000..e3cb186c440b
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
@@ -0,0 +1,483 @@
+/*
+ * Copyright 2010 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
+ */
+
+#include <core/gpuobj.h>
+#include <core/mm.h>
+
+#include <subdev/fb.h>
+#include <subdev/mmu.h>
+
+void
+nouveau_vm_map_at(struct nouveau_vma *vma, u64 delta, struct nouveau_mem *node)
+{
+	struct nouveau_vm *vm = vma->vm;
+	struct nouveau_mmu *mmu = vm->mmu;
+	struct nouveau_mm_node *r;
+	int big = vma->node->type != mmu->spg_shift;
+	u32 offset = vma->node->offset + (delta >> 12);
+	u32 bits = vma->node->type - 12;
+	u32 pde  = (offset >> mmu->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << mmu->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (mmu->pgt_bits - bits);
+	u32 end, len;
+
+	delta = 0;
+	list_for_each_entry(r, &node->regions, rl_entry) {
+		u64 phys = (u64)r->offset << 12;
+		u32 num  = r->length >> bits;
+
+		while (num) {
+			struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
+
+			end = (pte + num);
+			if (unlikely(end >= max))
+				end = max;
+			len = end - pte;
+
+			mmu->map(vma, pgt, node, pte, len, phys, delta);
+
+			num -= len;
+			pte += len;
+			if (unlikely(end >= max)) {
+				phys += len << (bits + 12);
+				pde++;
+				pte = 0;
+			}
+
+			delta += (u64)len << vma->node->type;
+		}
+	}
+
+	mmu->flush(vm);
+}
+
+static void
+nouveau_vm_map_sg_table(struct nouveau_vma *vma, u64 delta, u64 length,
+			struct nouveau_mem *mem)
+{
+	struct nouveau_vm *vm = vma->vm;
+	struct nouveau_mmu *mmu = vm->mmu;
+	int big = vma->node->type != mmu->spg_shift;
+	u32 offset = vma->node->offset + (delta >> 12);
+	u32 bits = vma->node->type - 12;
+	u32 num  = length >> vma->node->type;
+	u32 pde  = (offset >> mmu->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << mmu->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (mmu->pgt_bits - bits);
+	unsigned m, sglen;
+	u32 end, len;
+	int i;
+	struct scatterlist *sg;
+
+	for_each_sg(mem->sg->sgl, sg, mem->sg->nents, i) {
+		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
+		sglen = sg_dma_len(sg) >> PAGE_SHIFT;
+
+		end = pte + sglen;
+		if (unlikely(end >= max))
+			end = max;
+		len = end - pte;
+
+		for (m = 0; m < len; m++) {
+			dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);
+
+			mmu->map_sg(vma, pgt, mem, pte, 1, &addr);
+			num--;
+			pte++;
+
+			if (num == 0)
+				goto finish;
+		}
+		if (unlikely(end >= max)) {
+			pde++;
+			pte = 0;
+		}
+		if (m < sglen) {
+			for (; m < sglen; m++) {
+				dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);
+
+				mmu->map_sg(vma, pgt, mem, pte, 1, &addr);
+				num--;
+				pte++;
+				if (num == 0)
+					goto finish;
+			}
+		}
+
+	}
+finish:
+	mmu->flush(vm);
+}
+
+static void
+nouveau_vm_map_sg(struct nouveau_vma *vma, u64 delta, u64 length,
+		  struct nouveau_mem *mem)
+{
+	struct nouveau_vm *vm = vma->vm;
+	struct nouveau_mmu *mmu = vm->mmu;
+	dma_addr_t *list = mem->pages;
+	int big = vma->node->type != mmu->spg_shift;
+	u32 offset = vma->node->offset + (delta >> 12);
+	u32 bits = vma->node->type - 12;
+	u32 num  = length >> vma->node->type;
+	u32 pde  = (offset >> mmu->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << mmu->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (mmu->pgt_bits - bits);
+	u32 end, len;
+
+	while (num) {
+		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
+
+		end = (pte + num);
+		if (unlikely(end >= max))
+			end = max;
+		len = end - pte;
+
+		mmu->map_sg(vma, pgt, mem, pte, len, list);
+
+		num  -= len;
+		pte  += len;
+		list += len;
+		if (unlikely(end >= max)) {
+			pde++;
+			pte = 0;
+		}
+	}
+
+	mmu->flush(vm);
+}
+
+void
+nouveau_vm_map(struct nouveau_vma *vma, struct nouveau_mem *node)
+{
+	if (node->sg)
+		nouveau_vm_map_sg_table(vma, 0, node->size << 12, node);
+	else
+	if (node->pages)
+		nouveau_vm_map_sg(vma, 0, node->size << 12, node);
+	else
+		nouveau_vm_map_at(vma, 0, node);
+}
+
+void
+nouveau_vm_unmap_at(struct nouveau_vma *vma, u64 delta, u64 length)
+{
+	struct nouveau_vm *vm = vma->vm;
+	struct nouveau_mmu *mmu = vm->mmu;
+	int big = vma->node->type != mmu->spg_shift;
+	u32 offset = vma->node->offset + (delta >> 12);
+	u32 bits = vma->node->type - 12;
+	u32 num  = length >> vma->node->type;
+	u32 pde  = (offset >> mmu->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << mmu->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (mmu->pgt_bits - bits);
+	u32 end, len;
+
+	while (num) {
+		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
+
+		end = (pte + num);
+		if (unlikely(end >= max))
+			end = max;
+		len = end - pte;
+
+		mmu->unmap(pgt, pte, len);
+
+		num -= len;
+		pte += len;
+		if (unlikely(end >= max)) {
+			pde++;
+			pte = 0;
+		}
+	}
+
+	mmu->flush(vm);
+}
+
+void
+nouveau_vm_unmap(struct nouveau_vma *vma)
+{
+	nouveau_vm_unmap_at(vma, 0, (u64)vma->node->length << 12);
+}
+
+static void
+nouveau_vm_unmap_pgt(struct nouveau_vm *vm, int big, u32 fpde, u32 lpde)
+{
+	struct nouveau_mmu *mmu = vm->mmu;
+	struct nouveau_vm_pgd *vpgd;
+	struct nouveau_vm_pgt *vpgt;
+	struct nouveau_gpuobj *pgt;
+	u32 pde;
+
+	for (pde = fpde; pde <= lpde; pde++) {
+		vpgt = &vm->pgt[pde - vm->fpde];
+		if (--vpgt->refcount[big])
+			continue;
+
+		pgt = vpgt->obj[big];
+		vpgt->obj[big] = NULL;
+
+		list_for_each_entry(vpgd, &vm->pgd_list, head) {
+			mmu->map_pgt(vpgd->obj, pde, vpgt->obj);
+		}
+
+		mutex_unlock(&nv_subdev(mmu)->mutex);
+		nouveau_gpuobj_ref(NULL, &pgt);
+		mutex_lock(&nv_subdev(mmu)->mutex);
+	}
+}
+
+static int
+nouveau_vm_map_pgt(struct nouveau_vm *vm, u32 pde, u32 type)
+{
+	struct nouveau_mmu *mmu = vm->mmu;
+	struct nouveau_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
+	struct nouveau_vm_pgd *vpgd;
+	struct nouveau_gpuobj *pgt;
+	int big = (type != mmu->spg_shift);
+	u32 pgt_size;
+	int ret;
+
+	pgt_size  = (1 << (mmu->pgt_bits + 12)) >> type;
+	pgt_size *= 8;
+
+	mutex_unlock(&nv_subdev(mmu)->mutex);
+	ret = nouveau_gpuobj_new(nv_object(vm->mmu), NULL, pgt_size, 0x1000,
+				 NVOBJ_FLAG_ZERO_ALLOC, &pgt);
+	mutex_lock(&nv_subdev(mmu)->mutex);
+	if (unlikely(ret))
+		return ret;
+
+	/* someone beat us to filling the PDE while we didn't have the lock */
+	if (unlikely(vpgt->refcount[big]++)) {
+		mutex_unlock(&nv_subdev(mmu)->mutex);
+		nouveau_gpuobj_ref(NULL, &pgt);
+		mutex_lock(&nv_subdev(mmu)->mutex);
+		return 0;
+	}
+
+	vpgt->obj[big] = pgt;
+	list_for_each_entry(vpgd, &vm->pgd_list, head) {
+		mmu->map_pgt(vpgd->obj, pde, vpgt->obj);
+	}
+
+	return 0;
+}
+
+int
+nouveau_vm_get(struct nouveau_vm *vm, u64 size, u32 page_shift,
+	       u32 access, struct nouveau_vma *vma)
+{
+	struct nouveau_mmu *mmu = vm->mmu;
+	u32 align = (1 << page_shift) >> 12;
+	u32 msize = size >> 12;
+	u32 fpde, lpde, pde;
+	int ret;
+
+	mutex_lock(&nv_subdev(mmu)->mutex);
+	ret = nouveau_mm_head(&vm->mm, 0, page_shift, msize, msize, align,
+			     &vma->node);
+	if (unlikely(ret != 0)) {
+		mutex_unlock(&nv_subdev(mmu)->mutex);
+		return ret;
+	}
+
+	fpde = (vma->node->offset >> mmu->pgt_bits);
+	lpde = (vma->node->offset + vma->node->length - 1) >> mmu->pgt_bits;
+
+	for (pde = fpde; pde <= lpde; pde++) {
+		struct nouveau_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
+		int big = (vma->node->type != mmu->spg_shift);
+
+		if (likely(vpgt->refcount[big])) {
+			vpgt->refcount[big]++;
+			continue;
+		}
+
+		ret = nouveau_vm_map_pgt(vm, pde, vma->node->type);
+		if (ret) {
+			if (pde != fpde)
+				nouveau_vm_unmap_pgt(vm, big, fpde, pde - 1);
+			nouveau_mm_free(&vm->mm, &vma->node);
+			mutex_unlock(&nv_subdev(mmu)->mutex);
+			return ret;
+		}
+	}
+	mutex_unlock(&nv_subdev(mmu)->mutex);
+
+	vma->vm = NULL;
+	nouveau_vm_ref(vm, &vma->vm, NULL);
+	vma->offset = (u64)vma->node->offset << 12;
+	vma->access = access;
+	return 0;
+}
+
+void
+nouveau_vm_put(struct nouveau_vma *vma)
+{
+	struct nouveau_vm *vm = vma->vm;
+	struct nouveau_mmu *mmu = vm->mmu;
+	u32 fpde, lpde;
+
+	if (unlikely(vma->node == NULL))
+		return;
+	fpde = (vma->node->offset >> mmu->pgt_bits);
+	lpde = (vma->node->offset + vma->node->length - 1) >> mmu->pgt_bits;
+
+	mutex_lock(&nv_subdev(mmu)->mutex);
+	nouveau_vm_unmap_pgt(vm, vma->node->type != mmu->spg_shift, fpde, lpde);
+	nouveau_mm_free(&vm->mm, &vma->node);
+	mutex_unlock(&nv_subdev(mmu)->mutex);
+
+	nouveau_vm_ref(NULL, &vma->vm, NULL);
+}
+
+int
+nouveau_vm_create(struct nouveau_mmu *mmu, u64 offset, u64 length,
+		  u64 mm_offset, u32 block, struct nouveau_vm **pvm)
+{
+	struct nouveau_vm *vm;
+	u64 mm_length = (offset + length) - mm_offset;
+	int ret;
+
+	vm = kzalloc(sizeof(*vm), GFP_KERNEL);
+	if (!vm)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&vm->pgd_list);
+	vm->mmu = mmu;
+	kref_init(&vm->refcount);
+	vm->fpde = offset >> (mmu->pgt_bits + 12);
+	vm->lpde = (offset + length - 1) >> (mmu->pgt_bits + 12);
+
+	vm->pgt  = vzalloc((vm->lpde - vm->fpde + 1) * sizeof(*vm->pgt));
+	if (!vm->pgt) {
+		kfree(vm);
+		return -ENOMEM;
+	}
+
+	ret = nouveau_mm_init(&vm->mm, mm_offset >> 12, mm_length >> 12,
+			      block >> 12);
+	if (ret) {
+		vfree(vm->pgt);
+		kfree(vm);
+		return ret;
+	}
+
+	*pvm = vm;
+
+	return 0;
+}
+
+int
+nouveau_vm_new(struct nouveau_device *device, u64 offset, u64 length,
+	       u64 mm_offset, struct nouveau_vm **pvm)
+{
+	struct nouveau_mmu *mmu = nouveau_mmu(device);
+	return mmu->create(mmu, offset, length, mm_offset, pvm);
+}
+
+static int
+nouveau_vm_link(struct nouveau_vm *vm, struct nouveau_gpuobj *pgd)
+{
+	struct nouveau_mmu *mmu = vm->mmu;
+	struct nouveau_vm_pgd *vpgd;
+	int i;
+
+	if (!pgd)
+		return 0;
+
+	vpgd = kzalloc(sizeof(*vpgd), GFP_KERNEL);
+	if (!vpgd)
+		return -ENOMEM;
+
+	nouveau_gpuobj_ref(pgd, &vpgd->obj);
+
+	mutex_lock(&nv_subdev(mmu)->mutex);
+	for (i = vm->fpde; i <= vm->lpde; i++)
+		mmu->map_pgt(pgd, i, vm->pgt[i - vm->fpde].obj);
+	list_add(&vpgd->head, &vm->pgd_list);
+	mutex_unlock(&nv_subdev(mmu)->mutex);
+	return 0;
+}
+
+static void
+nouveau_vm_unlink(struct nouveau_vm *vm, struct nouveau_gpuobj *mpgd)
+{
+	struct nouveau_mmu *mmu = vm->mmu;
+	struct nouveau_vm_pgd *vpgd, *tmp;
+	struct nouveau_gpuobj *pgd = NULL;
+
+	if (!mpgd)
+		return;
+
+	mutex_lock(&nv_subdev(mmu)->mutex);
+	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
+		if (vpgd->obj == mpgd) {
+			pgd = vpgd->obj;
+			list_del(&vpgd->head);
+			kfree(vpgd);
+			break;
+		}
+	}
+	mutex_unlock(&nv_subdev(mmu)->mutex);
+
+	nouveau_gpuobj_ref(NULL, &pgd);
+}
+
+static void
+nouveau_vm_del(struct kref *kref)
+{
+	struct nouveau_vm *vm = container_of(kref, typeof(*vm), refcount);
+	struct nouveau_vm_pgd *vpgd, *tmp;
+
+	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
+		nouveau_vm_unlink(vm, vpgd->obj);
+	}
+
+	nouveau_mm_fini(&vm->mm);
+	vfree(vm->pgt);
+	kfree(vm);
+}
+
+int
+nouveau_vm_ref(struct nouveau_vm *ref, struct nouveau_vm **ptr,
+	       struct nouveau_gpuobj *pgd)
+{
+	if (ref) {
+		int ret = nouveau_vm_link(ref, pgd);
+		if (ret)
+			return ret;
+
+		kref_get(&ref->refcount);
+	}
+
+	if (*ptr) {
+		nouveau_vm_unlink(*ptr, pgd);
+		kref_put(&(*ptr)->refcount, nouveau_vm_del);
+	}
+
+	*ptr = ref;
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.c
new file mode 100644
index 000000000000..a317d8f13570
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.c
@@ -0,0 +1,151 @@
+/*
+ * Copyright 2012 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
+ */
+
+#include <core/gpuobj.h>
+
+#include "nv04.h"
+
+#define NV04_PDMA_SIZE (128 * 1024 * 1024)
+#define NV04_PDMA_PAGE (  4 * 1024)
+
+/*******************************************************************************
+ * VM map/unmap callbacks
+ ******************************************************************************/
+
+static void
+nv04_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	       struct nouveau_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
+{
+	pte = 0x00008 + (pte * 4);
+	while (cnt) {
+		u32 page = PAGE_SIZE / NV04_PDMA_PAGE;
+		u32 phys = (u32)*list++;
+		while (cnt && page--) {
+			nv_wo32(pgt, pte, phys | 3);
+			phys += NV04_PDMA_PAGE;
+			pte += 4;
+			cnt -= 1;
+		}
+	}
+}
+
+static void
+nv04_vm_unmap(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt)
+{
+	pte = 0x00008 + (pte * 4);
+	while (cnt--) {
+		nv_wo32(pgt, pte, 0x00000000);
+		pte += 4;
+	}
+}
+
+static void
+nv04_vm_flush(struct nouveau_vm *vm)
+{
+}
+
+/*******************************************************************************
+ * VM object
+ ******************************************************************************/
+
+int
+nv04_vm_create(struct nouveau_mmu *mmu, u64 offset, u64 length, u64 mmstart,
+	       struct nouveau_vm **pvm)
+{
+	return -EINVAL;
+}
+
+/*******************************************************************************
+ * MMU subdev
+ ******************************************************************************/
+
+static int
+nv04_mmu_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nv04_mmu_priv *priv;
+	struct nouveau_gpuobj *dma;
+	int ret;
+
+	ret = nouveau_mmu_create(parent, engine, oclass, "PCIGART",
+				   "pcigart", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.create = nv04_vm_create;
+	priv->base.limit = NV04_PDMA_SIZE;
+	priv->base.dma_bits = 32;
+	priv->base.pgt_bits = 32 - 12;
+	priv->base.spg_shift = 12;
+	priv->base.lpg_shift = 12;
+	priv->base.map_sg = nv04_vm_map_sg;
+	priv->base.unmap = nv04_vm_unmap;
+	priv->base.flush = nv04_vm_flush;
+
+	ret = nouveau_vm_create(&priv->base, 0, NV04_PDMA_SIZE, 0, 4096,
+				&priv->vm);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(nv_object(priv), NULL,
+				 (NV04_PDMA_SIZE / NV04_PDMA_PAGE) * 4 +
+				 8, 16, NVOBJ_FLAG_ZERO_ALLOC,
+				 &priv->vm->pgt[0].obj[0]);
+	dma = priv->vm->pgt[0].obj[0];
+	priv->vm->pgt[0].refcount[0] = 1;
+	if (ret)
+		return ret;
+
+	nv_wo32(dma, 0x00000, 0x0002103d); /* PCI, RW, PT, !LN */
+	nv_wo32(dma, 0x00004, NV04_PDMA_SIZE - 1);
+	return 0;
+}
+
+void
+nv04_mmu_dtor(struct nouveau_object *object)
+{
+	struct nv04_mmu_priv *priv = (void *)object;
+	if (priv->vm) {
+		nouveau_gpuobj_ref(NULL, &priv->vm->pgt[0].obj[0]);
+		nouveau_vm_ref(NULL, &priv->vm, NULL);
+	}
+	if (priv->nullp) {
+		pci_free_consistent(nv_device(priv)->pdev, 16 * 1024,
+				    priv->nullp, priv->null);
+	}
+	nouveau_mmu_destroy(&priv->base);
+}
+
+struct nouveau_oclass
+nv04_mmu_oclass = {
+	.handle = NV_SUBDEV(MMU, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_mmu_ctor,
+		.dtor = nv04_mmu_dtor,
+		.init = _nouveau_mmu_init,
+		.fini = _nouveau_mmu_fini,
+	},
+};
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.h b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.h
new file mode 100644
index 000000000000..919b254ef6a1
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.h
@@ -0,0 +1,19 @@
+#ifndef __NV04_MMU_PRIV__
+#define __NV04_MMU_PRIV__
+
+#include <subdev/mmu.h>
+
+struct nv04_mmu_priv {
+	struct nouveau_mmu base;
+	struct nouveau_vm *vm;
+	dma_addr_t null;
+	void *nullp;
+};
+
+static inline struct nv04_mmu_priv *
+nv04_mmu(void *obj)
+{
+	return (void *)nouveau_mmu(obj);
+}
+
+#endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c
new file mode 100644
index 000000000000..61af036f1252
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c
@@ -0,0 +1,159 @@
+/*
+ * Copyright 2012 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
+ */
+
+#include <core/gpuobj.h>
+#include <core/option.h>
+
+#include <subdev/timer.h>
+#include <subdev/mmu.h>
+
+#include "nv04.h"
+
+#define NV41_GART_SIZE (512 * 1024 * 1024)
+#define NV41_GART_PAGE (  4 * 1024)
+
+/*******************************************************************************
+ * VM map/unmap callbacks
+ ******************************************************************************/
+
+static void
+nv41_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	       struct nouveau_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
+{
+	pte = pte * 4;
+	while (cnt) {
+		u32 page = PAGE_SIZE / NV41_GART_PAGE;
+		u64 phys = (u64)*list++;
+		while (cnt && page--) {
+			nv_wo32(pgt, pte, (phys >> 7) | 1);
+			phys += NV41_GART_PAGE;
+			pte += 4;
+			cnt -= 1;
+		}
+	}
+}
+
+static void
+nv41_vm_unmap(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt)
+{
+	pte = pte * 4;
+	while (cnt--) {
+		nv_wo32(pgt, pte, 0x00000000);
+		pte += 4;
+	}
+}
+
+static void
+nv41_vm_flush(struct nouveau_vm *vm)
+{
+	struct nv04_mmu_priv *priv = (void *)vm->mmu;
+
+	mutex_lock(&nv_subdev(priv)->mutex);
+	nv_wr32(priv, 0x100810, 0x00000022);
+	if (!nv_wait(priv, 0x100810, 0x00000020, 0x00000020)) {
+		nv_warn(priv, "flush timeout, 0x%08x\n",
+			nv_rd32(priv, 0x100810));
+	}
+	nv_wr32(priv, 0x100810, 0x00000000);
+	mutex_unlock(&nv_subdev(priv)->mutex);
+}
+
+/*******************************************************************************
+ * MMU subdev
+ ******************************************************************************/
+
+static int
+nv41_mmu_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nv04_mmu_priv *priv;
+	int ret;
+
+	if (pci_find_capability(device->pdev, PCI_CAP_ID_AGP) ||
+	    !nouveau_boolopt(device->cfgopt, "NvPCIE", true)) {
+		return nouveau_object_ctor(parent, engine, &nv04_mmu_oclass,
+					   data, size, pobject);
+	}
+
+	ret = nouveau_mmu_create(parent, engine, oclass, "PCIEGART",
+				   "pciegart", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.create = nv04_vm_create;
+	priv->base.limit = NV41_GART_SIZE;
+	priv->base.dma_bits = 39;
+	priv->base.pgt_bits = 32 - 12;
+	priv->base.spg_shift = 12;
+	priv->base.lpg_shift = 12;
+	priv->base.map_sg = nv41_vm_map_sg;
+	priv->base.unmap = nv41_vm_unmap;
+	priv->base.flush = nv41_vm_flush;
+
+	ret = nouveau_vm_create(&priv->base, 0, NV41_GART_SIZE, 0, 4096,
+				&priv->vm);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(nv_object(priv), NULL,
+				(NV41_GART_SIZE / NV41_GART_PAGE) * 4,
+				 16, NVOBJ_FLAG_ZERO_ALLOC,
+				 &priv->vm->pgt[0].obj[0]);
+	priv->vm->pgt[0].refcount[0] = 1;
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static int
+nv41_mmu_init(struct nouveau_object *object)
+{
+	struct nv04_mmu_priv *priv = (void *)object;
+	struct nouveau_gpuobj *dma = priv->vm->pgt[0].obj[0];
+	int ret;
+
+	ret = nouveau_mmu_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x100800, dma->addr | 0x00000002);
+	nv_mask(priv, 0x10008c, 0x00000100, 0x00000100);
+	nv_wr32(priv, 0x100820, 0x00000000);
+	return 0;
+}
+
+struct nouveau_oclass
+nv41_mmu_oclass = {
+	.handle = NV_SUBDEV(MMU, 0x41),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv41_mmu_ctor,
+		.dtor = nv04_mmu_dtor,
+		.init = nv41_mmu_init,
+		.fini = _nouveau_mmu_fini,
+	},
+};
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv44.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv44.c
new file mode 100644
index 000000000000..f5319e3e7fe5
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv44.c
@@ -0,0 +1,249 @@
+/*
+ * Copyright 2012 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
+ */
+
+#include <core/gpuobj.h>
+#include <core/option.h>
+
+#include <subdev/timer.h>
+#include <subdev/mmu.h>
+
+#include "nv04.h"
+
+#define NV44_GART_SIZE (512 * 1024 * 1024)
+#define NV44_GART_PAGE (  4 * 1024)
+
+/*******************************************************************************
+ * VM map/unmap callbacks
+ ******************************************************************************/
+
+static void
+nv44_vm_fill(struct nouveau_gpuobj *pgt, dma_addr_t null,
+	     dma_addr_t *list, u32 pte, u32 cnt)
+{
+	u32 base = (pte << 2) & ~0x0000000f;
+	u32 tmp[4];
+
+	tmp[0] = nv_ro32(pgt, base + 0x0);
+	tmp[1] = nv_ro32(pgt, base + 0x4);
+	tmp[2] = nv_ro32(pgt, base + 0x8);
+	tmp[3] = nv_ro32(pgt, base + 0xc);
+
+	while (cnt--) {
+		u32 addr = list ? (*list++ >> 12) : (null >> 12);
+		switch (pte++ & 0x3) {
+		case 0:
+			tmp[0] &= ~0x07ffffff;
+			tmp[0] |= addr;
+			break;
+		case 1:
+			tmp[0] &= ~0xf8000000;
+			tmp[0] |= addr << 27;
+			tmp[1] &= ~0x003fffff;
+			tmp[1] |= addr >> 5;
+			break;
+		case 2:
+			tmp[1] &= ~0xffc00000;
+			tmp[1] |= addr << 22;
+			tmp[2] &= ~0x0001ffff;
+			tmp[2] |= addr >> 10;
+			break;
+		case 3:
+			tmp[2] &= ~0xfffe0000;
+			tmp[2] |= addr << 17;
+			tmp[3] &= ~0x00000fff;
+			tmp[3] |= addr >> 15;
+			break;
+		}
+	}
+
+	nv_wo32(pgt, base + 0x0, tmp[0]);
+	nv_wo32(pgt, base + 0x4, tmp[1]);
+	nv_wo32(pgt, base + 0x8, tmp[2]);
+	nv_wo32(pgt, base + 0xc, tmp[3] | 0x40000000);
+}
+
+static void
+nv44_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	       struct nouveau_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
+{
+	struct nv04_mmu_priv *priv = (void *)vma->vm->mmu;
+	u32 tmp[4];
+	int i;
+
+	if (pte & 3) {
+		u32  max = 4 - (pte & 3);
+		u32 part = (cnt > max) ? max : cnt;
+		nv44_vm_fill(pgt, priv->null, list, pte, part);
+		pte  += part;
+		list += part;
+		cnt  -= part;
+	}
+
+	while (cnt >= 4) {
+		for (i = 0; i < 4; i++)
+			tmp[i] = *list++ >> 12;
+		nv_wo32(pgt, pte++ * 4, tmp[0] >>  0 | tmp[1] << 27);
+		nv_wo32(pgt, pte++ * 4, tmp[1] >>  5 | tmp[2] << 22);
+		nv_wo32(pgt, pte++ * 4, tmp[2] >> 10 | tmp[3] << 17);
+		nv_wo32(pgt, pte++ * 4, tmp[3] >> 15 | 0x40000000);
+		cnt -= 4;
+	}
+
+	if (cnt)
+		nv44_vm_fill(pgt, priv->null, list, pte, cnt);
+}
+
+static void
+nv44_vm_unmap(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt)
+{
+	struct nv04_mmu_priv *priv = (void *)nouveau_mmu(pgt);
+
+	if (pte & 3) {
+		u32  max = 4 - (pte & 3);
+		u32 part = (cnt > max) ? max : cnt;
+		nv44_vm_fill(pgt, priv->null, NULL, pte, part);
+		pte  += part;
+		cnt  -= part;
+	}
+
+	while (cnt >= 4) {
+		nv_wo32(pgt, pte++ * 4, 0x00000000);
+		nv_wo32(pgt, pte++ * 4, 0x00000000);
+		nv_wo32(pgt, pte++ * 4, 0x00000000);
+		nv_wo32(pgt, pte++ * 4, 0x00000000);
+		cnt -= 4;
+	}
+
+	if (cnt)
+		nv44_vm_fill(pgt, priv->null, NULL, pte, cnt);
+}
+
+static void
+nv44_vm_flush(struct nouveau_vm *vm)
+{
+	struct nv04_mmu_priv *priv = (void *)vm->mmu;
+	nv_wr32(priv, 0x100814, priv->base.limit - NV44_GART_PAGE);
+	nv_wr32(priv, 0x100808, 0x00000020);
+	if (!nv_wait(priv, 0x100808, 0x00000001, 0x00000001))
+		nv_error(priv, "timeout: 0x%08x\n", nv_rd32(priv, 0x100808));
+	nv_wr32(priv, 0x100808, 0x00000000);
+}
+
+/*******************************************************************************
+ * MMU subdev
+ ******************************************************************************/
+
+static int
+nv44_mmu_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nv04_mmu_priv *priv;
+	int ret;
+
+	if (pci_find_capability(device->pdev, PCI_CAP_ID_AGP) ||
+	    !nouveau_boolopt(device->cfgopt, "NvPCIE", true)) {
+		return nouveau_object_ctor(parent, engine, &nv04_mmu_oclass,
+					   data, size, pobject);
+	}
+
+	ret = nouveau_mmu_create(parent, engine, oclass, "PCIEGART",
+				   "pciegart", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.create = nv04_vm_create;
+	priv->base.limit = NV44_GART_SIZE;
+	priv->base.dma_bits = 39;
+	priv->base.pgt_bits = 32 - 12;
+	priv->base.spg_shift = 12;
+	priv->base.lpg_shift = 12;
+	priv->base.map_sg = nv44_vm_map_sg;
+	priv->base.unmap = nv44_vm_unmap;
+	priv->base.flush = nv44_vm_flush;
+
+	priv->nullp = pci_alloc_consistent(device->pdev, 16 * 1024, &priv->null);
+	if (!priv->nullp) {
+		nv_error(priv, "unable to allocate dummy pages\n");
+		return -ENOMEM;
+	}
+
+	ret = nouveau_vm_create(&priv->base, 0, NV44_GART_SIZE, 0, 4096,
+				&priv->vm);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(nv_object(priv), NULL,
+				(NV44_GART_SIZE / NV44_GART_PAGE) * 4,
+				 512 * 1024, NVOBJ_FLAG_ZERO_ALLOC,
+				 &priv->vm->pgt[0].obj[0]);
+	priv->vm->pgt[0].refcount[0] = 1;
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static int
+nv44_mmu_init(struct nouveau_object *object)
+{
+	struct nv04_mmu_priv *priv = (void *)object;
+	struct nouveau_gpuobj *gart = priv->vm->pgt[0].obj[0];
+	u32 addr;
+	int ret;
+
+	ret = nouveau_mmu_init(&priv->base);
+	if (ret)
+		return ret;
+
+	/* calculate vram address of this PRAMIN block, object must be
+	 * allocated on 512KiB alignment, and not exceed a total size
+	 * of 512KiB for this to work correctly
+	 */
+	addr  = nv_rd32(priv, 0x10020c);
+	addr -= ((gart->addr >> 19) + 1) << 19;
+
+	nv_wr32(priv, 0x100850, 0x80000000);
+	nv_wr32(priv, 0x100818, priv->null);
+	nv_wr32(priv, 0x100804, NV44_GART_SIZE);
+	nv_wr32(priv, 0x100850, 0x00008000);
+	nv_mask(priv, 0x10008c, 0x00000200, 0x00000200);
+	nv_wr32(priv, 0x100820, 0x00000000);
+	nv_wr32(priv, 0x10082c, 0x00000001);
+	nv_wr32(priv, 0x100800, addr | 0x00000010);
+	return 0;
+}
+
+struct nouveau_oclass
+nv44_mmu_oclass = {
+	.handle = NV_SUBDEV(MMU, 0x44),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv44_mmu_ctor,
+		.dtor = nv04_mmu_dtor,
+		.init = nv44_mmu_init,
+		.fini = _nouveau_mmu_fini,
+	},
+};
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c
new file mode 100644
index 000000000000..6ddc65dc684d
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c
@@ -0,0 +1,240 @@
+/*
+ * Copyright 2010 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
+ */
+
+#include <core/device.h>
+#include <core/gpuobj.h>
+
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+#include <subdev/bar.h>
+#include <subdev/mmu.h>
+
+struct nv50_mmu_priv {
+	struct nouveau_mmu base;
+};
+
+static void
+nv50_vm_map_pgt(struct nouveau_gpuobj *pgd, u32 pde,
+		struct nouveau_gpuobj *pgt[2])
+{
+	u64 phys = 0xdeadcafe00000000ULL;
+	u32 coverage = 0;
+
+	if (pgt[0]) {
+		phys = 0x00000003 | pgt[0]->addr; /* present, 4KiB pages */
+		coverage = (pgt[0]->size >> 3) << 12;
+	} else
+	if (pgt[1]) {
+		phys = 0x00000001 | pgt[1]->addr; /* present */
+		coverage = (pgt[1]->size >> 3) << 16;
+	}
+
+	if (phys & 1) {
+		if (coverage <= 32 * 1024 * 1024)
+			phys |= 0x60;
+		else if (coverage <= 64 * 1024 * 1024)
+			phys |= 0x40;
+		else if (coverage <= 128 * 1024 * 1024)
+			phys |= 0x20;
+	}
+
+	nv_wo32(pgd, (pde * 8) + 0, lower_32_bits(phys));
+	nv_wo32(pgd, (pde * 8) + 4, upper_32_bits(phys));
+}
+
+static inline u64
+vm_addr(struct nouveau_vma *vma, u64 phys, u32 memtype, u32 target)
+{
+	phys |= 1; /* present */
+	phys |= (u64)memtype << 40;
+	phys |= target << 4;
+	if (vma->access & NV_MEM_ACCESS_SYS)
+		phys |= (1 << 6);
+	if (!(vma->access & NV_MEM_ACCESS_WO))
+		phys |= (1 << 3);
+	return phys;
+}
+
+static void
+nv50_vm_map(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	    struct nouveau_mem *mem, u32 pte, u32 cnt, u64 phys, u64 delta)
+{
+	u32 comp = (mem->memtype & 0x180) >> 7;
+	u32 block, target;
+	int i;
+
+	/* IGPs don't have real VRAM, re-target to stolen system memory */
+	target = 0;
+	if (nouveau_fb(vma->vm->mmu)->ram->stolen) {
+		phys += nouveau_fb(vma->vm->mmu)->ram->stolen;
+		target = 3;
+	}
+
+	phys  = vm_addr(vma, phys, mem->memtype, target);
+	pte <<= 3;
+	cnt <<= 3;
+
+	while (cnt) {
+		u32 offset_h = upper_32_bits(phys);
+		u32 offset_l = lower_32_bits(phys);
+
+		for (i = 7; i >= 0; i--) {
+			block = 1 << (i + 3);
+			if (cnt >= block && !(pte & (block - 1)))
+				break;
+		}
+		offset_l |= (i << 7);
+
+		phys += block << (vma->node->type - 3);
+		cnt  -= block;
+		if (comp) {
+			u32 tag = mem->tag->offset + ((delta >> 16) * comp);
+			offset_h |= (tag << 17);
+			delta    += block << (vma->node->type - 3);
+		}
+
+		while (block) {
+			nv_wo32(pgt, pte + 0, offset_l);
+			nv_wo32(pgt, pte + 4, offset_h);
+			pte += 8;
+			block -= 8;
+		}
+	}
+}
+
+static void
+nv50_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	       struct nouveau_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
+{
+	u32 target = (vma->access & NV_MEM_ACCESS_NOSNOOP) ? 3 : 2;
+	pte <<= 3;
+	while (cnt--) {
+		u64 phys = vm_addr(vma, (u64)*list++, mem->memtype, target);
+		nv_wo32(pgt, pte + 0, lower_32_bits(phys));
+		nv_wo32(pgt, pte + 4, upper_32_bits(phys));
+		pte += 8;
+	}
+}
+
+static void
+nv50_vm_unmap(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt)
+{
+	pte <<= 3;
+	while (cnt--) {
+		nv_wo32(pgt, pte + 0, 0x00000000);
+		nv_wo32(pgt, pte + 4, 0x00000000);
+		pte += 8;
+	}
+}
+
+static void
+nv50_vm_flush(struct nouveau_vm *vm)
+{
+	struct nv50_mmu_priv *priv = (void *)vm->mmu;
+	struct nouveau_bar *bar = nouveau_bar(priv);
+	struct nouveau_engine *engine;
+	int i, vme;
+
+	bar->flush(bar);
+
+	mutex_lock(&nv_subdev(priv)->mutex);
+	for (i = 0; i < NVDEV_SUBDEV_NR; i++) {
+		if (!atomic_read(&vm->engref[i]))
+			continue;
+
+		/* unfortunate hw bug workaround... */
+		engine = nouveau_engine(priv, i);
+		if (engine && engine->tlb_flush) {
+			engine->tlb_flush(engine);
+			continue;
+		}
+
+		switch (i) {
+		case NVDEV_ENGINE_GR   : vme = 0x00; break;
+		case NVDEV_ENGINE_VP   : vme = 0x01; break;
+		case NVDEV_SUBDEV_BAR  : vme = 0x06; break;
+		case NVDEV_ENGINE_PPP  :
+		case NVDEV_ENGINE_MPEG : vme = 0x08; break;
+		case NVDEV_ENGINE_BSP  : vme = 0x09; break;
+		case NVDEV_ENGINE_CRYPT: vme = 0x0a; break;
+		case NVDEV_ENGINE_COPY0: vme = 0x0d; break;
+		default:
+			continue;
+		}
+
+		nv_wr32(priv, 0x100c80, (vme << 16) | 1);
+		if (!nv_wait(priv, 0x100c80, 0x00000001, 0x00000000))
+			nv_error(priv, "vm flush timeout: engine %d\n", vme);
+	}
+	mutex_unlock(&nv_subdev(priv)->mutex);
+}
+
+static int
+nv50_vm_create(struct nouveau_mmu *mmu, u64 offset, u64 length,
+	       u64 mm_offset, struct nouveau_vm **pvm)
+{
+	u32 block = (1 << (mmu->pgt_bits + 12));
+	if (block > length)
+		block = length;
+
+	return nouveau_vm_create(mmu, offset, length, mm_offset, block, pvm);
+}
+
+static int
+nv50_mmu_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nv50_mmu_priv *priv;
+	int ret;
+
+	ret = nouveau_mmu_create(parent, engine, oclass, "VM", "vm", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.limit = 1ULL << 40;
+	priv->base.dma_bits = 40;
+	priv->base.pgt_bits  = 29 - 12;
+	priv->base.spg_shift = 12;
+	priv->base.lpg_shift = 16;
+	priv->base.create = nv50_vm_create;
+	priv->base.map_pgt = nv50_vm_map_pgt;
+	priv->base.map = nv50_vm_map;
+	priv->base.map_sg = nv50_vm_map_sg;
+	priv->base.unmap = nv50_vm_unmap;
+	priv->base.flush = nv50_vm_flush;
+	return 0;
+}
+
+struct nouveau_oclass
+nv50_mmu_oclass = {
+	.handle = NV_SUBDEV(MMU, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_mmu_ctor,
+		.dtor = _nouveau_mmu_dtor,
+		.init = _nouveau_mmu_init,
+		.fini = _nouveau_mmu_fini,
+	},
+};
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nvc0.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nvc0.c
new file mode 100644
index 000000000000..bd695c59aac7
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nvc0.c
@@ -0,0 +1,241 @@
+/*
+ * Copyright 2010 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
+ */
+
+#include <core/device.h>
+#include <core/gpuobj.h>
+
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+#include <subdev/mmu.h>
+#include <subdev/ltc.h>
+#include <subdev/bar.h>
+
+struct nvc0_mmu_priv {
+	struct nouveau_mmu base;
+};
+
+
+/* Map from compressed to corresponding uncompressed storage type.
+ * The value 0xff represents an invalid storage type.
+ */
+const u8 nvc0_pte_storage_type_map[256] =
+{
+	0x00, 0x01, 0x01, 0x01, 0x01, 0x01, 0xff, 0x01, /* 0x00 */
+	0x01, 0x01, 0x01, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0x11, 0xff, 0xff, 0xff, 0xff, 0xff, 0x11, /* 0x10 */
+	0x11, 0x11, 0x11, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0x26, 0x27, /* 0x20 */
+	0x28, 0x29, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, /* 0x30 */
+	0xff, 0xff, 0x26, 0x27, 0x28, 0x29, 0x26, 0x27,
+	0x28, 0x29, 0xff, 0xff, 0xff, 0xff, 0x46, 0xff, /* 0x40 */
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0x46, 0x46, 0x46, 0x46, 0xff, 0xff, 0xff, /* 0x50 */
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, /* 0x60 */
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, /* 0x70 */
+	0xff, 0xff, 0xff, 0x7b, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0x7b, 0x7b, /* 0x80 */
+	0x7b, 0x7b, 0xff, 0x8b, 0x8c, 0x8d, 0x8e, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, /* 0x90 */
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0xff, 0xff, 0x8b, 0x8c, 0x8d, 0x8e, 0xa7, /* 0xa0 */
+	0xa8, 0xa9, 0xaa, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, /* 0xb0 */
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xa7,
+	0xa8, 0xa9, 0xaa, 0xc3, 0xff, 0xff, 0xff, 0xff, /* 0xc0 */
+	0xff, 0xff, 0xff, 0xff, 0xfe, 0xfe, 0xc3, 0xc3,
+	0xc3, 0xc3, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, /* 0xd0 */
+	0xfe, 0xff, 0xff, 0xfe, 0xff, 0xfe, 0xff, 0xfe,
+	0xfe, 0xff, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xff, /* 0xe0 */
+	0xff, 0xfe, 0xff, 0xfe, 0xff, 0xfe, 0xfe, 0xff,
+	0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, /* 0xf0 */
+	0xfe, 0xfe, 0xfe, 0xfe, 0xff, 0xfd, 0xfe, 0xff
+};
+
+
+static void
+nvc0_vm_map_pgt(struct nouveau_gpuobj *pgd, u32 index,
+		struct nouveau_gpuobj *pgt[2])
+{
+	u32 pde[2] = { 0, 0 };
+
+	if (pgt[0])
+		pde[1] = 0x00000001 | (pgt[0]->addr >> 8);
+	if (pgt[1])
+		pde[0] = 0x00000001 | (pgt[1]->addr >> 8);
+
+	nv_wo32(pgd, (index * 8) + 0, pde[0]);
+	nv_wo32(pgd, (index * 8) + 4, pde[1]);
+}
+
+static inline u64
+nvc0_vm_addr(struct nouveau_vma *vma, u64 phys, u32 memtype, u32 target)
+{
+	phys >>= 8;
+
+	phys |= 0x00000001; /* present */
+	if (vma->access & NV_MEM_ACCESS_SYS)
+		phys |= 0x00000002;
+
+	phys |= ((u64)target  << 32);
+	phys |= ((u64)memtype << 36);
+
+	return phys;
+}
+
+static void
+nvc0_vm_map(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	    struct nouveau_mem *mem, u32 pte, u32 cnt, u64 phys, u64 delta)
+{
+	u64 next = 1 << (vma->node->type - 8);
+
+	phys  = nvc0_vm_addr(vma, phys, mem->memtype, 0);
+	pte <<= 3;
+
+	if (mem->tag) {
+		struct nouveau_ltc *ltc = nouveau_ltc(vma->vm->mmu);
+		u32 tag = mem->tag->offset + (delta >> 17);
+		phys |= (u64)tag << (32 + 12);
+		next |= (u64)1   << (32 + 12);
+		ltc->tags_clear(ltc, tag, cnt);
+	}
+
+	while (cnt--) {
+		nv_wo32(pgt, pte + 0, lower_32_bits(phys));
+		nv_wo32(pgt, pte + 4, upper_32_bits(phys));
+		phys += next;
+		pte  += 8;
+	}
+}
+
+static void
+nvc0_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	       struct nouveau_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
+{
+	u32 target = (vma->access & NV_MEM_ACCESS_NOSNOOP) ? 7 : 5;
+	/* compressed storage types are invalid for system memory */
+	u32 memtype = nvc0_pte_storage_type_map[mem->memtype & 0xff];
+
+	pte <<= 3;
+	while (cnt--) {
+		u64 phys = nvc0_vm_addr(vma, *list++, memtype, target);
+		nv_wo32(pgt, pte + 0, lower_32_bits(phys));
+		nv_wo32(pgt, pte + 4, upper_32_bits(phys));
+		pte += 8;
+	}
+}
+
+static void
+nvc0_vm_unmap(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt)
+{
+	pte <<= 3;
+	while (cnt--) {
+		nv_wo32(pgt, pte + 0, 0x00000000);
+		nv_wo32(pgt, pte + 4, 0x00000000);
+		pte += 8;
+	}
+}
+
+static void
+nvc0_vm_flush(struct nouveau_vm *vm)
+{
+	struct nvc0_mmu_priv *priv = (void *)vm->mmu;
+	struct nouveau_bar *bar = nouveau_bar(priv);
+	struct nouveau_vm_pgd *vpgd;
+	u32 type;
+
+	bar->flush(bar);
+
+	type = 0x00000001; /* PAGE_ALL */
+	if (atomic_read(&vm->engref[NVDEV_SUBDEV_BAR]))
+		type |= 0x00000004; /* HUB_ONLY */
+
+	mutex_lock(&nv_subdev(priv)->mutex);
+	list_for_each_entry(vpgd, &vm->pgd_list, head) {
+		/* looks like maybe a "free flush slots" counter, the
+		 * faster you write to 0x100cbc to more it decreases
+		 */
+		if (!nv_wait_ne(priv, 0x100c80, 0x00ff0000, 0x00000000)) {
+			nv_error(priv, "vm timeout 0: 0x%08x %d\n",
+				 nv_rd32(priv, 0x100c80), type);
+		}
+
+		nv_wr32(priv, 0x100cb8, vpgd->obj->addr >> 8);
+		nv_wr32(priv, 0x100cbc, 0x80000000 | type);
+
+		/* wait for flush to be queued? */
+		if (!nv_wait(priv, 0x100c80, 0x00008000, 0x00008000)) {
+			nv_error(priv, "vm timeout 1: 0x%08x %d\n",
+				 nv_rd32(priv, 0x100c80), type);
+		}
+	}
+	mutex_unlock(&nv_subdev(priv)->mutex);
+}
+
+static int
+nvc0_vm_create(struct nouveau_mmu *mmu, u64 offset, u64 length,
+	       u64 mm_offset, struct nouveau_vm **pvm)
+{
+	return nouveau_vm_create(mmu, offset, length, mm_offset, 4096, pvm);
+}
+
+static int
+nvc0_mmu_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nvc0_mmu_priv *priv;
+	int ret;
+
+	ret = nouveau_mmu_create(parent, engine, oclass, "VM", "vm", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.limit = 1ULL << 40;
+	priv->base.dma_bits = 40;
+	priv->base.pgt_bits  = 27 - 12;
+	priv->base.spg_shift = 12;
+	priv->base.lpg_shift = 17;
+	priv->base.create = nvc0_vm_create;
+	priv->base.map_pgt = nvc0_vm_map_pgt;
+	priv->base.map = nvc0_vm_map;
+	priv->base.map_sg = nvc0_vm_map_sg;
+	priv->base.unmap = nvc0_vm_unmap;
+	priv->base.flush = nvc0_vm_flush;
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_mmu_oclass = {
+	.handle = NV_SUBDEV(MMU, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_mmu_ctor,
+		.dtor = _nouveau_mmu_dtor,
+		.init = _nouveau_mmu_init,
+		.fini = _nouveau_mmu_fini,
+	},
+};