drm/nouveau/mmu: convert to new-style nvkm_subdev

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

8 files changed, 299 insertions, 278 deletions

diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
index 6fa1bdb02dfd..e04a2296ecd0 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
@@ -21,10 +21,10 @@
  *
  * Authors: Ben Skeggs
  */
-#include <subdev/mmu.h>
-#include <subdev/fb.h>
+#include "priv.h"
 
 #include <core/gpuobj.h>
+#include <subdev/fb.h>
 
 void
 nvkm_vm_map_at(struct nvkm_vma *vma, u64 delta, struct nvkm_mem *node)
@@ -32,12 +32,12 @@ nvkm_vm_map_at(struct nvkm_vma *vma, u64 delta, struct nvkm_mem *node)
 	struct nvkm_vm *vm = vma->vm;
 	struct nvkm_mmu *mmu = vm->mmu;
 	struct nvkm_mm_node *r;
-	int big = vma->node->type != mmu->spg_shift;
+	int big = vma->node->type != mmu->func->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 pde  = (offset >> mmu->func->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (mmu->func->pgt_bits - bits);
 	u32 end, len;
 
 	delta = 0;
@@ -53,7 +53,7 @@ nvkm_vm_map_at(struct nvkm_vma *vma, u64 delta, struct nvkm_mem *node)
 				end = max;
 			len = end - pte;
 
-			mmu->map(vma, pgt, node, pte, len, phys, delta);
+			mmu->func->map(vma, pgt, node, pte, len, phys, delta);
 
 			num -= len;
 			pte += len;
@@ -67,7 +67,7 @@ nvkm_vm_map_at(struct nvkm_vma *vma, u64 delta, struct nvkm_mem *node)
 		}
 	}
 
-	mmu->flush(vm);
+	mmu->func->flush(vm);
 }
 
 static void
@@ -76,13 +76,13 @@ nvkm_vm_map_sg_table(struct nvkm_vma *vma, u64 delta, u64 length,
 {
 	struct nvkm_vm *vm = vma->vm;
 	struct nvkm_mmu *mmu = vm->mmu;
-	int big = vma->node->type != mmu->spg_shift;
+	int big = vma->node->type != mmu->func->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 pde  = (offset >> mmu->func->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (mmu->func->pgt_bits - bits);
 	unsigned m, sglen;
 	u32 end, len;
 	int i;
@@ -100,7 +100,7 @@ nvkm_vm_map_sg_table(struct nvkm_vma *vma, u64 delta, u64 length,
 		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);
+			mmu->func->map_sg(vma, pgt, mem, pte, 1, &addr);
 			num--;
 			pte++;
 
@@ -115,7 +115,7 @@ nvkm_vm_map_sg_table(struct nvkm_vma *vma, u64 delta, u64 length,
 			for (; m < sglen; m++) {
 				dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);
 
-				mmu->map_sg(vma, pgt, mem, pte, 1, &addr);
+				mmu->func->map_sg(vma, pgt, mem, pte, 1, &addr);
 				num--;
 				pte++;
 				if (num == 0)
@@ -125,7 +125,7 @@ nvkm_vm_map_sg_table(struct nvkm_vma *vma, u64 delta, u64 length,
 
 	}
 finish:
-	mmu->flush(vm);
+	mmu->func->flush(vm);
 }
 
 static void
@@ -135,13 +135,13 @@ nvkm_vm_map_sg(struct nvkm_vma *vma, u64 delta, u64 length,
 	struct nvkm_vm *vm = vma->vm;
 	struct nvkm_mmu *mmu = vm->mmu;
 	dma_addr_t *list = mem->pages;
-	int big = vma->node->type != mmu->spg_shift;
+	int big = vma->node->type != mmu->func->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 pde  = (offset >> mmu->func->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (mmu->func->pgt_bits - bits);
 	u32 end, len;
 
 	while (num) {
@@ -152,7 +152,7 @@ nvkm_vm_map_sg(struct nvkm_vma *vma, u64 delta, u64 length,
 			end = max;
 		len = end - pte;
 
-		mmu->map_sg(vma, pgt, mem, pte, len, list);
+		mmu->func->map_sg(vma, pgt, mem, pte, len, list);
 
 		num  -= len;
 		pte  += len;
@@ -163,7 +163,7 @@ nvkm_vm_map_sg(struct nvkm_vma *vma, u64 delta, u64 length,
 		}
 	}
 
-	mmu->flush(vm);
+	mmu->func->flush(vm);
 }
 
 void
@@ -183,13 +183,13 @@ nvkm_vm_unmap_at(struct nvkm_vma *vma, u64 delta, u64 length)
 {
 	struct nvkm_vm *vm = vma->vm;
 	struct nvkm_mmu *mmu = vm->mmu;
-	int big = vma->node->type != mmu->spg_shift;
+	int big = vma->node->type != mmu->func->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 pde  = (offset >> mmu->func->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (mmu->func->pgt_bits - bits);
 	u32 end, len;
 
 	while (num) {
@@ -200,7 +200,7 @@ nvkm_vm_unmap_at(struct nvkm_vma *vma, u64 delta, u64 length)
 			end = max;
 		len = end - pte;
 
-		mmu->unmap(vma, pgt, pte, len);
+		mmu->func->unmap(vma, pgt, pte, len);
 
 		num -= len;
 		pte += len;
@@ -210,7 +210,7 @@ nvkm_vm_unmap_at(struct nvkm_vma *vma, u64 delta, u64 length)
 		}
 	}
 
-	mmu->flush(vm);
+	mmu->func->flush(vm);
 }
 
 void
@@ -237,7 +237,7 @@ nvkm_vm_unmap_pgt(struct nvkm_vm *vm, int big, u32 fpde, u32 lpde)
 		vpgt->mem[big] = NULL;
 
 		list_for_each_entry(vpgd, &vm->pgd_list, head) {
-			mmu->map_pgt(vpgd->obj, pde, vpgt->mem);
+			mmu->func->map_pgt(vpgd->obj, pde, vpgt->mem);
 		}
 
 		nvkm_memory_del(&pgt);
@@ -250,11 +250,11 @@ nvkm_vm_map_pgt(struct nvkm_vm *vm, u32 pde, u32 type)
 	struct nvkm_mmu *mmu = vm->mmu;
 	struct nvkm_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
 	struct nvkm_vm_pgd *vpgd;
-	int big = (type != mmu->spg_shift);
+	int big = (type != mmu->func->spg_shift);
 	u32 pgt_size;
 	int ret;
 
-	pgt_size  = (1 << (mmu->pgt_bits + 12)) >> type;
+	pgt_size  = (1 << (mmu->func->pgt_bits + 12)) >> type;
 	pgt_size *= 8;
 
 	ret = nvkm_memory_new(mmu->subdev.device, NVKM_MEM_TARGET_INST,
@@ -263,7 +263,7 @@ nvkm_vm_map_pgt(struct nvkm_vm *vm, u32 pde, u32 type)
 		return ret;
 
 	list_for_each_entry(vpgd, &vm->pgd_list, head) {
-		mmu->map_pgt(vpgd->obj, pde, vpgt->mem);
+		mmu->func->map_pgt(vpgd->obj, pde, vpgt->mem);
 	}
 
 	vpgt->refcount[big]++;
@@ -288,12 +288,12 @@ nvkm_vm_get(struct nvkm_vm *vm, u64 size, u32 page_shift, u32 access,
 		return ret;
 	}
 
-	fpde = (vma->node->offset >> mmu->pgt_bits);
-	lpde = (vma->node->offset + vma->node->length - 1) >> mmu->pgt_bits;
+	fpde = (vma->node->offset >> mmu->func->pgt_bits);
+	lpde = (vma->node->offset + vma->node->length - 1) >> mmu->func->pgt_bits;
 
 	for (pde = fpde; pde <= lpde; pde++) {
 		struct nvkm_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
-		int big = (vma->node->type != mmu->spg_shift);
+		int big = (vma->node->type != mmu->func->spg_shift);
 
 		if (likely(vpgt->refcount[big])) {
 			vpgt->refcount[big]++;
@@ -330,11 +330,11 @@ nvkm_vm_put(struct nvkm_vma *vma)
 	vm = vma->vm;
 	mmu = vm->mmu;
 
-	fpde = (vma->node->offset >> mmu->pgt_bits);
-	lpde = (vma->node->offset + vma->node->length - 1) >> mmu->pgt_bits;
+	fpde = (vma->node->offset >> mmu->func->pgt_bits);
+	lpde = (vma->node->offset + vma->node->length - 1) >> mmu->func->pgt_bits;
 
 	mutex_lock(&vm->mutex);
-	nvkm_vm_unmap_pgt(vm, vma->node->type != mmu->spg_shift, fpde, lpde);
+	nvkm_vm_unmap_pgt(vm, vma->node->type != mmu->func->spg_shift, fpde, lpde);
 	nvkm_mm_free(&vm->mm, &vma->node);
 	mutex_unlock(&vm->mutex);
 
@@ -349,7 +349,7 @@ nvkm_vm_boot(struct nvkm_vm *vm, u64 size)
 	int ret;
 
 	ret = nvkm_memory_new(mmu->subdev.device, NVKM_MEM_TARGET_INST,
-			      (size >> mmu->spg_shift) * 8, 0x1000, true, &pgt);
+			      (size >> mmu->func->spg_shift) * 8, 0x1000, true, &pgt);
 	if (ret == 0) {
 		vm->pgt[0].refcount[0] = 1;
 		vm->pgt[0].mem[0] = pgt;
@@ -376,8 +376,8 @@ nvkm_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset,
 	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->fpde = offset >> (mmu->func->pgt_bits + 12);
+	vm->lpde = (offset + length - 1) >> (mmu->func->pgt_bits + 12);
 
 	vm->pgt  = vzalloc((vm->lpde - vm->fpde + 1) * sizeof(*vm->pgt));
 	if (!vm->pgt) {
@@ -402,8 +402,10 @@ int
 nvkm_vm_new(struct nvkm_device *device, u64 offset, u64 length, u64 mm_offset,
 	    struct lock_class_key *key, struct nvkm_vm **pvm)
 {
-	struct nvkm_mmu *mmu = nvkm_mmu(device);
-	return mmu->create(mmu, offset, length, mm_offset, key, pvm);
+	struct nvkm_mmu *mmu = device->mmu;
+	if (!mmu->func->create)
+		return -EINVAL;
+	return mmu->func->create(mmu, offset, length, mm_offset, key, pvm);
 }
 
 static int
@@ -424,7 +426,7 @@ nvkm_vm_link(struct nvkm_vm *vm, struct nvkm_gpuobj *pgd)
 
 	mutex_lock(&vm->mutex);
 	for (i = vm->fpde; i <= vm->lpde; i++)
-		mmu->map_pgt(pgd, i, vm->pgt[i - vm->fpde].mem);
+		mmu->func->map_pgt(pgd, i, vm->pgt[i - vm->fpde].mem);
 	list_add(&vpgd->head, &vm->pgd_list);
 	mutex_unlock(&vm->mutex);
 	return 0;
@@ -483,3 +485,58 @@ nvkm_vm_ref(struct nvkm_vm *ref, struct nvkm_vm **ptr, struct nvkm_gpuobj *pgd)
 	*ptr = ref;
 	return 0;
 }
+
+static int
+nvkm_mmu_oneinit(struct nvkm_subdev *subdev)
+{
+	struct nvkm_mmu *mmu = nvkm_mmu(subdev);
+	if (mmu->func->oneinit)
+		return mmu->func->oneinit(mmu);
+	return 0;
+}
+
+static int
+nvkm_mmu_init(struct nvkm_subdev *subdev)
+{
+	struct nvkm_mmu *mmu = nvkm_mmu(subdev);
+	if (mmu->func->init)
+		mmu->func->init(mmu);
+	return 0;
+}
+
+static void *
+nvkm_mmu_dtor(struct nvkm_subdev *subdev)
+{
+	struct nvkm_mmu *mmu = nvkm_mmu(subdev);
+	if (mmu->func->dtor)
+		return mmu->func->dtor(mmu);
+	return mmu;
+}
+
+static const struct nvkm_subdev_func
+nvkm_mmu = {
+	.dtor = nvkm_mmu_dtor,
+	.oneinit = nvkm_mmu_oneinit,
+	.init = nvkm_mmu_init,
+};
+
+void
+nvkm_mmu_ctor(const struct nvkm_mmu_func *func, struct nvkm_device *device,
+	      int index, struct nvkm_mmu *mmu)
+{
+	nvkm_subdev_ctor(&nvkm_mmu, device, index, 0, &mmu->subdev);
+	mmu->func = func;
+	mmu->limit = func->limit;
+	mmu->dma_bits = func->dma_bits;
+	mmu->lpg_shift = func->lpg_shift;
+}
+
+int
+nvkm_mmu_new_(const struct nvkm_mmu_func *func, struct nvkm_device *device,
+	      int index, struct nvkm_mmu **pmmu)
+{
+	if (!(*pmmu = kzalloc(sizeof(**pmmu), GFP_KERNEL)))
+		return -ENOMEM;
+	nvkm_mmu_ctor(func, device, index, *pmmu);
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/gf100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/gf100.c
index 6a2a2d575f16..5f6c5df15a95 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/gf100.c
@@ -21,7 +21,8 @@
  *
  * Authors: Ben Skeggs
  */
-#include <subdev/mmu.h>
+#include "priv.h"
+
 #include <subdev/fb.h>
 #include <subdev/ltc.h>
 #include <subdev/timer.h>
@@ -160,7 +161,7 @@ gf100_vm_unmap(struct nvkm_vma *vma, struct nvkm_memory *pgt, u32 pte, u32 cnt)
 static void
 gf100_vm_flush(struct nvkm_vm *vm)
 {
-	struct nvkm_mmu *mmu = (void *)vm->mmu;
+	struct nvkm_mmu *mmu = vm->mmu;
 	struct nvkm_device *device = mmu->subdev.device;
 	struct nvkm_vm_pgd *vpgd;
 	u32 type;
@@ -169,7 +170,7 @@ gf100_vm_flush(struct nvkm_vm *vm)
 	if (atomic_read(&vm->engref[NVDEV_SUBDEV_BAR]))
 		type |= 0x00000004; /* HUB_ONLY */
 
-	mutex_lock(&nv_subdev(mmu)->mutex);
+	mutex_lock(&mmu->subdev.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
@@ -188,7 +189,7 @@ gf100_vm_flush(struct nvkm_vm *vm)
 				break;
 		);
 	}
-	mutex_unlock(&nv_subdev(mmu)->mutex);
+	mutex_unlock(&mmu->subdev.mutex);
 }
 
 static int
@@ -198,40 +199,23 @@ gf100_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset,
 	return nvkm_vm_create(mmu, offset, length, mm_offset, 4096, key, pvm);
 }
 
-static int
-gf100_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
-	       struct nvkm_oclass *oclass, void *data, u32 size,
-	       struct nvkm_object **pobject)
+static const struct nvkm_mmu_func
+gf100_mmu = {
+	.limit = (1ULL << 40),
+	.dma_bits = 40,
+	.pgt_bits  = 27 - 12,
+	.spg_shift = 12,
+	.lpg_shift = 17,
+	.create = gf100_vm_create,
+	.map_pgt = gf100_vm_map_pgt,
+	.map = gf100_vm_map,
+	.map_sg = gf100_vm_map_sg,
+	.unmap = gf100_vm_unmap,
+	.flush = gf100_vm_flush,
+};
+
+int
+gf100_mmu_new(struct nvkm_device *device, int index, struct nvkm_mmu **pmmu)
 {
-	struct nvkm_mmu *mmu;
-	int ret;
-
-	ret = nvkm_mmu_create(parent, engine, oclass, "VM", "mmu", &mmu);
-	*pobject = nv_object(mmu);
-	if (ret)
-		return ret;
-
-	mmu->limit = 1ULL << 40;
-	mmu->dma_bits = 40;
-	mmu->pgt_bits  = 27 - 12;
-	mmu->spg_shift = 12;
-	mmu->lpg_shift = 17;
-	mmu->create = gf100_vm_create;
-	mmu->map_pgt = gf100_vm_map_pgt;
-	mmu->map = gf100_vm_map;
-	mmu->map_sg = gf100_vm_map_sg;
-	mmu->unmap = gf100_vm_unmap;
-	mmu->flush = gf100_vm_flush;
-	return 0;
+	return nvkm_mmu_new_(&gf100_mmu, device, index, pmmu);
 }
-
-struct nvkm_oclass
-gf100_mmu_oclass = {
-	.handle = NV_SUBDEV(MMU, 0xc0),
-	.ofuncs = &(struct nvkm_ofuncs) {
-		.ctor = gf100_mmu_ctor,
-		.dtor = _nvkm_mmu_dtor,
-		.init = _nvkm_mmu_init,
-		.fini = _nvkm_mmu_fini,
-	},
-};
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.c
index 6995845ac6a1..5def412f0467 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.c
@@ -69,46 +69,17 @@ nv04_vm_flush(struct nvkm_vm *vm)
 }
 
 /*******************************************************************************
- * VM object
- ******************************************************************************/
-
-int
-nv04_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mmstart,
-	       struct lock_class_key *key, struct nvkm_vm **pvm)
-{
-	return -EINVAL;
-}
-
-/*******************************************************************************
  * MMU subdev
  ******************************************************************************/
 
 static int
-nv04_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
-	      struct nvkm_oclass *oclass, void *data, u32 size,
-	      struct nvkm_object **pobject)
+nv04_mmu_oneinit(struct nvkm_mmu *base)
 {
-	struct nvkm_device *device = (void *)parent;
-	struct nv04_mmu *mmu;
+	struct nv04_mmu *mmu = nv04_mmu(base);
+	struct nvkm_device *device = mmu->base.subdev.device;
 	struct nvkm_memory *dma;
 	int ret;
 
-	ret = nvkm_mmu_create(parent, engine, oclass, "PCIGART",
-			      "mmu", &mmu);
-	*pobject = nv_object(mmu);
-	if (ret)
-		return ret;
-
-	mmu->base.create = nv04_vm_create;
-	mmu->base.limit = NV04_PDMA_SIZE;
-	mmu->base.dma_bits = 32;
-	mmu->base.pgt_bits = 32 - 12;
-	mmu->base.spg_shift = 12;
-	mmu->base.lpg_shift = 12;
-	mmu->base.map_sg = nv04_vm_map_sg;
-	mmu->base.unmap = nv04_vm_unmap;
-	mmu->base.flush = nv04_vm_flush;
-
 	ret = nvkm_vm_create(&mmu->base, 0, NV04_PDMA_SIZE, 0, 4096, NULL,
 			     &mmu->vm);
 	if (ret)
@@ -129,28 +100,50 @@ nv04_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	return 0;
 }
 
-void
-nv04_mmu_dtor(struct nvkm_object *object)
+void *
+nv04_mmu_dtor(struct nvkm_mmu *base)
 {
-	struct nv04_mmu *mmu = (void *)object;
+	struct nv04_mmu *mmu = nv04_mmu(base);
+	struct nvkm_device *device = mmu->base.subdev.device;
 	if (mmu->vm) {
 		nvkm_memory_del(&mmu->vm->pgt[0].mem[0]);
 		nvkm_vm_ref(NULL, &mmu->vm, NULL);
 	}
 	if (mmu->nullp) {
-		pci_free_consistent(nv_device(mmu)->pdev, 16 * 1024,
+		pci_free_consistent(device->pdev, 16 * 1024,
 				    mmu->nullp, mmu->null);
 	}
-	nvkm_mmu_destroy(&mmu->base);
+	return mmu;
 }
 
-struct nvkm_oclass
-nv04_mmu_oclass = {
-	.handle = NV_SUBDEV(MMU, 0x04),
-	.ofuncs = &(struct nvkm_ofuncs) {
-		.ctor = nv04_mmu_ctor,
-		.dtor = nv04_mmu_dtor,
-		.init = _nvkm_mmu_init,
-		.fini = _nvkm_mmu_fini,
-	},
+int
+nv04_mmu_new_(const struct nvkm_mmu_func *func, struct nvkm_device *device,
+	      int index, struct nvkm_mmu **pmmu)
+{
+	struct nv04_mmu *mmu;
+	if (!(mmu = kzalloc(sizeof(*mmu), GFP_KERNEL)))
+		return -ENOMEM;
+	*pmmu = &mmu->base;
+	nvkm_mmu_ctor(func, device, index, &mmu->base);
+	return 0;
+}
+
+const struct nvkm_mmu_func
+nv04_mmu = {
+	.oneinit = nv04_mmu_oneinit,
+	.dtor = nv04_mmu_dtor,
+	.limit = NV04_PDMA_SIZE,
+	.dma_bits = 32,
+	.pgt_bits = 32 - 12,
+	.spg_shift = 12,
+	.lpg_shift = 12,
+	.map_sg = nv04_vm_map_sg,
+	.unmap = nv04_vm_unmap,
+	.flush = nv04_vm_flush,
 };
+
+int
+nv04_mmu_new(struct nvkm_device *device, int index, struct nvkm_mmu **pmmu)
+{
+	return nv04_mmu_new_(&nv04_mmu, device, index, pmmu);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.h b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.h
index 80a404eab441..363e33b296d5 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.h
@@ -1,6 +1,7 @@
 #ifndef __NV04_MMU_PRIV__
 #define __NV04_MMU_PRIV__
-#include <subdev/mmu.h>
+#define nv04_mmu(p) container_of((p), struct nv04_mmu, base)
+#include "priv.h"
 
 struct nv04_mmu {
 	struct nvkm_mmu base;
@@ -9,9 +10,9 @@ struct nv04_mmu {
 	void *nullp;
 };
 
-static inline struct nv04_mmu *
-nv04_mmu(void *obj)
-{
-	return (void *)nvkm_mmu(obj);
-}
+int nv04_mmu_new_(const struct nvkm_mmu_func *, struct nvkm_device *,
+		  int index, struct nvkm_mmu **);
+void *nv04_mmu_dtor(struct nvkm_mmu *);
+
+extern const struct nvkm_mmu_func nv04_mmu;
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c
index 0f91d7aeb53f..f30c3b890626 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c
@@ -71,14 +71,14 @@ nv41_vm_flush(struct nvkm_vm *vm)
 	struct nv04_mmu *mmu = nv04_mmu(vm->mmu);
 	struct nvkm_device *device = mmu->base.subdev.device;
 
-	mutex_lock(&nv_subdev(mmu)->mutex);
+	mutex_lock(&mmu->base.subdev.mutex);
 	nvkm_wr32(device, 0x100810, 0x00000022);
 	nvkm_msec(device, 2000,
 		if (nvkm_rd32(device, 0x100810) & 0x00000020)
 			break;
 	);
 	nvkm_wr32(device, 0x100810, 0x00000000);
-	mutex_unlock(&nv_subdev(mmu)->mutex);
+	mutex_unlock(&mmu->base.subdev.mutex);
 }
 
 /*******************************************************************************
@@ -86,36 +86,12 @@ nv41_vm_flush(struct nvkm_vm *vm)
  ******************************************************************************/
 
 static int
-nv41_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
-	      struct nvkm_oclass *oclass, void *data, u32 size,
-	      struct nvkm_object **pobject)
+nv41_mmu_oneinit(struct nvkm_mmu *base)
 {
-	struct nvkm_device *device = nv_device(parent);
-	struct nv04_mmu *mmu;
+	struct nv04_mmu *mmu = nv04_mmu(base);
+	struct nvkm_device *device = mmu->base.subdev.device;
 	int ret;
 
-	if (pci_find_capability(device->pdev, PCI_CAP_ID_AGP) ||
-	    !nvkm_boolopt(device->cfgopt, "NvPCIE", true)) {
-		return nvkm_object_old(parent, engine, &nv04_mmu_oclass,
-					data, size, pobject);
-	}
-
-	ret = nvkm_mmu_create(parent, engine, oclass, "PCIEGART",
-			      "mmu", &mmu);
-	*pobject = nv_object(mmu);
-	if (ret)
-		return ret;
-
-	mmu->base.create = nv04_vm_create;
-	mmu->base.limit = NV41_GART_SIZE;
-	mmu->base.dma_bits = 39;
-	mmu->base.pgt_bits = 32 - 12;
-	mmu->base.spg_shift = 12;
-	mmu->base.lpg_shift = 12;
-	mmu->base.map_sg = nv41_vm_map_sg;
-	mmu->base.unmap = nv41_vm_unmap;
-	mmu->base.flush = nv41_vm_flush;
-
 	ret = nvkm_vm_create(&mmu->base, 0, NV41_GART_SIZE, 0, 4096, NULL,
 			     &mmu->vm);
 	if (ret)
@@ -125,37 +101,41 @@ nv41_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 			      (NV41_GART_SIZE / NV41_GART_PAGE) * 4, 16, true,
 			      &mmu->vm->pgt[0].mem[0]);
 	mmu->vm->pgt[0].refcount[0] = 1;
-	if (ret)
-		return ret;
-
-	return 0;
+	return ret;
 }
 
-static int
-nv41_mmu_init(struct nvkm_object *object)
+static void
+nv41_mmu_init(struct nvkm_mmu *base)
 {
-	struct nv04_mmu *mmu = (void *)object;
+	struct nv04_mmu *mmu = nv04_mmu(base);
 	struct nvkm_device *device = mmu->base.subdev.device;
 	struct nvkm_memory *dma = mmu->vm->pgt[0].mem[0];
-	int ret;
-
-	ret = nvkm_mmu_init(&mmu->base);
-	if (ret)
-		return ret;
-
 	nvkm_wr32(device, 0x100800, 0x00000002 | nvkm_memory_addr(dma));
 	nvkm_mask(device, 0x10008c, 0x00000100, 0x00000100);
 	nvkm_wr32(device, 0x100820, 0x00000000);
-	return 0;
 }
 
-struct nvkm_oclass
-nv41_mmu_oclass = {
-	.handle = NV_SUBDEV(MMU, 0x41),
-	.ofuncs = &(struct nvkm_ofuncs) {
-		.ctor = nv41_mmu_ctor,
-		.dtor = nv04_mmu_dtor,
-		.init = nv41_mmu_init,
-		.fini = _nvkm_mmu_fini,
-	},
+static const struct nvkm_mmu_func
+nv41_mmu = {
+	.dtor = nv04_mmu_dtor,
+	.oneinit = nv41_mmu_oneinit,
+	.init = nv41_mmu_init,
+	.limit = NV41_GART_SIZE,
+	.dma_bits = 39,
+	.pgt_bits = 32 - 12,
+	.spg_shift = 12,
+	.lpg_shift = 12,
+	.map_sg = nv41_vm_map_sg,
+	.unmap = nv41_vm_unmap,
+	.flush = nv41_vm_flush,
 };
+
+int
+nv41_mmu_new(struct nvkm_device *device, int index, struct nvkm_mmu **pmmu)
+{
+	if (pci_find_capability(device->pdev, PCI_CAP_ID_AGP) ||
+	    !nvkm_boolopt(device->cfgopt, "NvPCIE", true))
+		return nv04_mmu_new(device, index, pmmu);
+
+	return nv04_mmu_new_(&nv41_mmu, device, index, pmmu);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv44.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv44.c
index d2b586bc57a2..7c37bd84b862 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv44.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv44.c
@@ -159,36 +159,12 @@ nv44_vm_flush(struct nvkm_vm *vm)
  ******************************************************************************/
 
 static int
-nv44_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
-	      struct nvkm_oclass *oclass, void *data, u32 size,
-	      struct nvkm_object **pobject)
+nv44_mmu_oneinit(struct nvkm_mmu *base)
 {
-	struct nvkm_device *device = nv_device(parent);
-	struct nv04_mmu *mmu;
+	struct nv04_mmu *mmu = nv04_mmu(base);
+	struct nvkm_device *device = mmu->base.subdev.device;
 	int ret;
 
-	if (pci_find_capability(device->pdev, PCI_CAP_ID_AGP) ||
-	    !nvkm_boolopt(device->cfgopt, "NvPCIE", true)) {
-		return nvkm_object_old(parent, engine, &nv04_mmu_oclass,
-					data, size, pobject);
-	}
-
-	ret = nvkm_mmu_create(parent, engine, oclass, "PCIEGART",
-			      "mmu", &mmu);
-	*pobject = nv_object(mmu);
-	if (ret)
-		return ret;
-
-	mmu->base.create = nv04_vm_create;
-	mmu->base.limit = NV44_GART_SIZE;
-	mmu->base.dma_bits = 39;
-	mmu->base.pgt_bits = 32 - 12;
-	mmu->base.spg_shift = 12;
-	mmu->base.lpg_shift = 12;
-	mmu->base.map_sg = nv44_vm_map_sg;
-	mmu->base.unmap = nv44_vm_unmap;
-	mmu->base.flush = nv44_vm_flush;
-
 	mmu->nullp = pci_alloc_consistent(device->pdev, 16 * 1024, &mmu->null);
 	if (!mmu->nullp) {
 		nvkm_warn(&mmu->base.subdev, "unable to allocate dummy pages\n");
@@ -205,24 +181,16 @@ nv44_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 			      512 * 1024, true,
 			      &mmu->vm->pgt[0].mem[0]);
 	mmu->vm->pgt[0].refcount[0] = 1;
-	if (ret)
-		return ret;
-
-	return 0;
+	return ret;
 }
 
-static int
-nv44_mmu_init(struct nvkm_object *object)
+static void
+nv44_mmu_init(struct nvkm_mmu *base)
 {
-	struct nv04_mmu *mmu = (void *)object;
+	struct nv04_mmu *mmu = nv04_mmu(base);
 	struct nvkm_device *device = mmu->base.subdev.device;
 	struct nvkm_memory *gart = mmu->vm->pgt[0].mem[0];
 	u32 addr;
-	int ret;
-
-	ret = nvkm_mmu_init(&mmu->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
@@ -239,16 +207,29 @@ nv44_mmu_init(struct nvkm_object *object)
 	nvkm_wr32(device, 0x100820, 0x00000000);
 	nvkm_wr32(device, 0x10082c, 0x00000001);
 	nvkm_wr32(device, 0x100800, addr | 0x00000010);
-	return 0;
 }
 
-struct nvkm_oclass
-nv44_mmu_oclass = {
-	.handle = NV_SUBDEV(MMU, 0x44),
-	.ofuncs = &(struct nvkm_ofuncs) {
-		.ctor = nv44_mmu_ctor,
-		.dtor = nv04_mmu_dtor,
-		.init = nv44_mmu_init,
-		.fini = _nvkm_mmu_fini,
-	},
+static const struct nvkm_mmu_func
+nv44_mmu = {
+	.dtor = nv04_mmu_dtor,
+	.oneinit = nv44_mmu_oneinit,
+	.init = nv44_mmu_init,
+	.limit = NV44_GART_SIZE,
+	.dma_bits = 39,
+	.pgt_bits = 32 - 12,
+	.spg_shift = 12,
+	.lpg_shift = 12,
+	.map_sg = nv44_vm_map_sg,
+	.unmap = nv44_vm_unmap,
+	.flush = nv44_vm_flush,
 };
+
+int
+nv44_mmu_new(struct nvkm_device *device, int index, struct nvkm_mmu **pmmu)
+{
+	if (pci_find_capability(device->pdev, PCI_CAP_ID_AGP) ||
+	    !nvkm_boolopt(device->cfgopt, "NvPCIE", true))
+		return nv04_mmu_new(device, index, pmmu);
+
+	return nv04_mmu_new_(&nv44_mmu, device, index, pmmu);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c
index f1f62a217e91..931f14094eb5 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c
@@ -21,7 +21,8 @@
  *
  * Authors: Ben Skeggs
  */
-#include <subdev/mmu.h>
+#include "priv.h"
+
 #include <subdev/fb.h>
 #include <subdev/timer.h>
 
@@ -155,10 +156,9 @@ nv50_vm_unmap(struct nvkm_vma *vma, struct nvkm_memory *pgt, u32 pte, u32 cnt)
 static void
 nv50_vm_flush(struct nvkm_vm *vm)
 {
-	struct nvkm_mmu *mmu = (void *)vm->mmu;
+	struct nvkm_mmu *mmu = vm->mmu;
 	struct nvkm_subdev *subdev = &mmu->subdev;
 	struct nvkm_device *device = subdev->device;
-	struct nvkm_engine *engine;
 	int i, vme;
 
 	mutex_lock(&subdev->mutex);
@@ -167,10 +167,13 @@ nv50_vm_flush(struct nvkm_vm *vm)
 			continue;
 
 		/* unfortunate hw bug workaround... */
-		engine = nvkm_engine(mmu, i);
-		if (engine && engine->tlb_flush) {
-			engine->tlb_flush(engine);
-			continue;
+		if (i == NVDEV_ENGINE_GR) {
+			struct nvkm_engine *engine =
+				nvkm_device_engine(device, i);
+			if (engine && engine->tlb_flush) {
+				engine->tlb_flush(engine);
+				continue;
+			}
 		}
 
 		switch (i) {
@@ -203,47 +206,30 @@ static int
 nv50_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset,
 	       struct lock_class_key *key, struct nvkm_vm **pvm)
 {
-	u32 block = (1 << (mmu->pgt_bits + 12));
+	u32 block = (1 << (mmu->func->pgt_bits + 12));
 	if (block > length)
 		block = length;
 
 	return nvkm_vm_create(mmu, offset, length, mm_offset, block, key, pvm);
 }
 
-static int
-nv50_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
-	      struct nvkm_oclass *oclass, void *data, u32 size,
-	      struct nvkm_object **pobject)
+static const struct nvkm_mmu_func
+nv50_mmu = {
+	.limit = (1ULL << 40),
+	.dma_bits = 40,
+	.pgt_bits  = 29 - 12,
+	.spg_shift = 12,
+	.lpg_shift = 16,
+	.create = nv50_vm_create,
+	.map_pgt = nv50_vm_map_pgt,
+	.map = nv50_vm_map,
+	.map_sg = nv50_vm_map_sg,
+	.unmap = nv50_vm_unmap,
+	.flush = nv50_vm_flush,
+};
+
+int
+nv50_mmu_new(struct nvkm_device *device, int index, struct nvkm_mmu **pmmu)
 {
-	struct nvkm_mmu *mmu;
-	int ret;
-
-	ret = nvkm_mmu_create(parent, engine, oclass, "VM", "mmu", &mmu);
-	*pobject = nv_object(mmu);
-	if (ret)
-		return ret;
-
-	mmu->limit = 1ULL << 40;
-	mmu->dma_bits = 40;
-	mmu->pgt_bits  = 29 - 12;
-	mmu->spg_shift = 12;
-	mmu->lpg_shift = 16;
-	mmu->create = nv50_vm_create;
-	mmu->map_pgt = nv50_vm_map_pgt;
-	mmu->map = nv50_vm_map;
-	mmu->map_sg = nv50_vm_map_sg;
-	mmu->unmap = nv50_vm_unmap;
-	mmu->flush = nv50_vm_flush;
-	return 0;
+	return nvkm_mmu_new_(&nv50_mmu, device, index, pmmu);
 }
-
-struct nvkm_oclass
-nv50_mmu_oclass = {
-	.handle = NV_SUBDEV(MMU, 0x50),
-	.ofuncs = &(struct nvkm_ofuncs) {
-		.ctor = nv50_mmu_ctor,
-		.dtor = _nvkm_mmu_dtor,
-		.init = _nvkm_mmu_init,
-		.fini = _nvkm_mmu_fini,
-	},
-};
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/priv.h
new file mode 100644
index 000000000000..27cedc60b507
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/priv.h
@@ -0,0 +1,39 @@
+#ifndef __NVKM_MMU_PRIV_H__
+#define __NVKM_MMU_PRIV_H__
+#define nvkm_mmu(p) container_of((p), struct nvkm_mmu, subdev)
+#include <subdev/mmu.h>
+
+void nvkm_mmu_ctor(const struct nvkm_mmu_func *, struct nvkm_device *,
+		   int index, struct nvkm_mmu *);
+int nvkm_mmu_new_(const struct nvkm_mmu_func *, struct nvkm_device *,
+		  int index, struct nvkm_mmu **);
+
+struct nvkm_mmu_func {
+	void *(*dtor)(struct nvkm_mmu *);
+	int (*oneinit)(struct nvkm_mmu *);
+	void (*init)(struct nvkm_mmu *);
+
+	u64 limit;
+	u8  dma_bits;
+	u32 pgt_bits;
+	u8  spg_shift;
+	u8  lpg_shift;
+
+	int  (*create)(struct nvkm_mmu *, u64 offset, u64 length, u64 mm_offset,
+		       struct lock_class_key *, struct nvkm_vm **);
+
+	void (*map_pgt)(struct nvkm_gpuobj *pgd, u32 pde,
+			struct nvkm_memory *pgt[2]);
+	void (*map)(struct nvkm_vma *, struct nvkm_memory *,
+		    struct nvkm_mem *, u32 pte, u32 cnt,
+		    u64 phys, u64 delta);
+	void (*map_sg)(struct nvkm_vma *, struct nvkm_memory *,
+		       struct nvkm_mem *, u32 pte, u32 cnt, dma_addr_t *);
+	void (*unmap)(struct nvkm_vma *, struct nvkm_memory *pgt,
+		      u32 pte, u32 cnt);
+	void (*flush)(struct nvkm_vm *);
+};
+
+int nvkm_vm_create(struct nvkm_mmu *, u64, u64, u64, u32,
+		   struct lock_class_key *, struct nvkm_vm **);
+#endif