Merge tag 'for-linus-hmm' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma

Pull hmm updates from Jason Gunthorpe:
 "Ralph has been working on nouveau's use of hmm_range_fault() and
  migrate_vma() which resulted in this small series. It adds reporting
  of the page table order from hmm_range_fault() and some optimization
  of migrate_vma():

   - Report the size of the page table mapping out of hmm_range_fault().

     This makes it easier to establish a large/huge/etc mapping in the
     device's page table.

   - Allow devices to ignore the invalidations during migration in cases
     where the migration is not going to change pages.

     For instance migrating pages to a device does not require the
     device to invalidate pages already in the device.

   - Update nouveau and hmm_tests to use the above"

* tag 'for-linus-hmm' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
  mm/hmm/test: use the new migration invalidation
  nouveau/svm: use the new migration invalidation
  mm/notifier: add migration invalidation type
  mm/migrate: add a flags parameter to migrate_vma
  nouveau: fix storing invalid ptes
  nouveau/hmm: support mapping large sysmem pages
  nouveau: fix mapping 2MB sysmem pages
  nouveau/hmm: fault one page at a time
  mm/hmm: add tests for hmm_pfn_to_map_order()
  mm/hmm: provide the page mapping order in hmm_range_fault()

5 files changed, 225 insertions, 160 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_dmem.c b/drivers/gpu/drm/nouveau/nouveau_dmem.c
index cc9993837508..951bd51f4834 100644
--- a/drivers/gpu/drm/nouveau/nouveau_dmem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_dmem.c
@@ -140,6 +140,7 @@ static vm_fault_t nouveau_dmem_fault_copy_one(struct nouveau_drm *drm,
 {
 	struct device *dev = drm->dev->dev;
 	struct page *dpage, *spage;
+	struct nouveau_svmm *svmm;
 
 	spage = migrate_pfn_to_page(args->src[0]);
 	if (!spage || !(args->src[0] & MIGRATE_PFN_MIGRATE))
@@ -154,14 +155,19 @@ static vm_fault_t nouveau_dmem_fault_copy_one(struct nouveau_drm *drm,
 	if (dma_mapping_error(dev, *dma_addr))
 		goto error_free_page;
 
+	svmm = spage->zone_device_data;
+	mutex_lock(&svmm->mutex);
+	nouveau_svmm_invalidate(svmm, args->start, args->end);
 	if (drm->dmem->migrate.copy_func(drm, 1, NOUVEAU_APER_HOST, *dma_addr,
 			NOUVEAU_APER_VRAM, nouveau_dmem_page_addr(spage)))
 		goto error_dma_unmap;
+	mutex_unlock(&svmm->mutex);
 
 	args->dst[0] = migrate_pfn(page_to_pfn(dpage)) | MIGRATE_PFN_LOCKED;
 	return 0;
 
 error_dma_unmap:
+	mutex_unlock(&svmm->mutex);
 	dma_unmap_page(dev, *dma_addr, PAGE_SIZE, DMA_BIDIRECTIONAL);
 error_free_page:
 	__free_page(dpage);
@@ -182,7 +188,8 @@ static vm_fault_t nouveau_dmem_migrate_to_ram(struct vm_fault *vmf)
 		.end		= vmf->address + PAGE_SIZE,
 		.src		= &src,
 		.dst		= &dst,
-		.src_owner	= drm->dev,
+		.pgmap_owner	= drm->dev,
+		.flags		= MIGRATE_VMA_SELECT_DEVICE_PRIVATE,
 	};
 
 	/*
@@ -530,7 +537,8 @@ nouveau_dmem_init(struct nouveau_drm *drm)
 }
 
 static unsigned long nouveau_dmem_migrate_copy_one(struct nouveau_drm *drm,
-		unsigned long src, dma_addr_t *dma_addr, u64 *pfn)
+		struct nouveau_svmm *svmm, unsigned long src,
+		dma_addr_t *dma_addr, u64 *pfn)
 {
 	struct device *dev = drm->dev->dev;
 	struct page *dpage, *spage;
@@ -560,6 +568,7 @@ static unsigned long nouveau_dmem_migrate_copy_one(struct nouveau_drm *drm,
 			goto out_free_page;
 	}
 
+	dpage->zone_device_data = svmm;
 	*pfn = NVIF_VMM_PFNMAP_V0_V | NVIF_VMM_PFNMAP_V0_VRAM |
 		((paddr >> PAGE_SHIFT) << NVIF_VMM_PFNMAP_V0_ADDR_SHIFT);
 	if (src & MIGRATE_PFN_WRITE)
@@ -583,8 +592,8 @@ static void nouveau_dmem_migrate_chunk(struct nouveau_drm *drm,
 	unsigned long addr = args->start, nr_dma = 0, i;
 
 	for (i = 0; addr < args->end; i++) {
-		args->dst[i] = nouveau_dmem_migrate_copy_one(drm, args->src[i],
-				dma_addrs + nr_dma, pfns + i);
+		args->dst[i] = nouveau_dmem_migrate_copy_one(drm, svmm,
+				args->src[i], dma_addrs + nr_dma, pfns + i);
 		if (!dma_mapping_error(drm->dev->dev, dma_addrs[nr_dma]))
 			nr_dma++;
 		addr += PAGE_SIZE;
@@ -615,6 +624,8 @@ nouveau_dmem_migrate_vma(struct nouveau_drm *drm,
 	struct migrate_vma args = {
 		.vma		= vma,
 		.start		= start,
+		.pgmap_owner	= drm->dev,
+		.flags		= MIGRATE_VMA_SELECT_SYSTEM,
 	};
 	unsigned long i;
 	u64 *pfns;
diff --git a/drivers/gpu/drm/nouveau/nouveau_svm.c b/drivers/gpu/drm/nouveau/nouveau_svm.c
index 6586d9d39874..03c515eaf10f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_svm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_svm.c
@@ -93,17 +93,6 @@ nouveau_ivmm_find(struct nouveau_svm *svm, u64 inst)
 	return NULL;
 }
 
-struct nouveau_svmm {
-	struct mmu_notifier notifier;
-	struct nouveau_vmm *vmm;
-	struct {
-		unsigned long start;
-		unsigned long limit;
-	} unmanaged;
-
-	struct mutex mutex;
-};
-
 #define SVMM_DBG(s,f,a...)                                                     \
 	NV_DEBUG((s)->vmm->cli->drm, "svm-%p: "f"\n", (s), ##a)
 #define SVMM_ERR(s,f,a...)                                                     \
@@ -246,7 +235,7 @@ nouveau_svmm_join(struct nouveau_svmm *svmm, u64 inst)
 }
 
 /* Invalidate SVMM address-range on GPU. */
-static void
+void
 nouveau_svmm_invalidate(struct nouveau_svmm *svmm, u64 start, u64 limit)
 {
 	if (limit > start) {
@@ -279,6 +268,14 @@ nouveau_svmm_invalidate_range_start(struct mmu_notifier *mn,
 	if (unlikely(!svmm->vmm))
 		goto out;
 
+	/*
+	 * Ignore invalidation callbacks for device private pages since
+	 * the invalidation is handled as part of the migration process.
+	 */
+	if (update->event == MMU_NOTIFY_MIGRATE &&
+	    update->migrate_pgmap_owner == svmm->vmm->cli->drm->dev)
+		goto out;
+
 	if (limit > svmm->unmanaged.start && start < svmm->unmanaged.limit) {
 		if (start < svmm->unmanaged.start) {
 			nouveau_svmm_invalidate(svmm, start,
@@ -514,53 +511,68 @@ static const struct mmu_interval_notifier_ops nouveau_svm_mni_ops = {
 };
 
 static void nouveau_hmm_convert_pfn(struct nouveau_drm *drm,
-				    struct hmm_range *range, u64 *ioctl_addr)
+				    struct hmm_range *range,
+				    struct nouveau_pfnmap_args *args)
 {
-	unsigned long i, npages;
+	struct page *page;
 
 	/*
-	 * The ioctl_addr prepared here is passed through nvif_object_ioctl()
+	 * The address prepared here is passed through nvif_object_ioctl()
 	 * to an eventual DMA map in something like gp100_vmm_pgt_pfn()
 	 *
 	 * This is all just encoding the internal hmm representation into a
 	 * different nouveau internal representation.
 	 */
-	npages = (range->end - range->start) >> PAGE_SHIFT;
-	for (i = 0; i < npages; ++i) {
-		struct page *page;
-
-		if (!(range->hmm_pfns[i] & HMM_PFN_VALID)) {
-			ioctl_addr[i] = 0;
-			continue;
-		}
+	if (!(range->hmm_pfns[0] & HMM_PFN_VALID)) {
+		args->p.phys[0] = 0;
+		return;
+	}
 
-		page = hmm_pfn_to_page(range->hmm_pfns[i]);
-		if (is_device_private_page(page))
-			ioctl_addr[i] = nouveau_dmem_page_addr(page) |
-					NVIF_VMM_PFNMAP_V0_V |
-					NVIF_VMM_PFNMAP_V0_VRAM;
-		else
-			ioctl_addr[i] = page_to_phys(page) |
-					NVIF_VMM_PFNMAP_V0_V |
-					NVIF_VMM_PFNMAP_V0_HOST;
-		if (range->hmm_pfns[i] & HMM_PFN_WRITE)
-			ioctl_addr[i] |= NVIF_VMM_PFNMAP_V0_W;
+	page = hmm_pfn_to_page(range->hmm_pfns[0]);
+	/*
+	 * Only map compound pages to the GPU if the CPU is also mapping the
+	 * page as a compound page. Otherwise, the PTE protections might not be
+	 * consistent (e.g., CPU only maps part of a compound page).
+	 * Note that the underlying page might still be larger than the
+	 * CPU mapping (e.g., a PUD sized compound page partially mapped with
+	 * a PMD sized page table entry).
+	 */
+	if (hmm_pfn_to_map_order(range->hmm_pfns[0])) {
+		unsigned long addr = args->p.addr;
+
+		args->p.page = hmm_pfn_to_map_order(range->hmm_pfns[0]) +
+				PAGE_SHIFT;
+		args->p.size = 1UL << args->p.page;
+		args->p.addr &= ~(args->p.size - 1);
+		page -= (addr - args->p.addr) >> PAGE_SHIFT;
 	}
+	if (is_device_private_page(page))
+		args->p.phys[0] = nouveau_dmem_page_addr(page) |
+				NVIF_VMM_PFNMAP_V0_V |
+				NVIF_VMM_PFNMAP_V0_VRAM;
+	else
+		args->p.phys[0] = page_to_phys(page) |
+				NVIF_VMM_PFNMAP_V0_V |
+				NVIF_VMM_PFNMAP_V0_HOST;
+	if (range->hmm_pfns[0] & HMM_PFN_WRITE)
+		args->p.phys[0] |= NVIF_VMM_PFNMAP_V0_W;
 }
 
 static int nouveau_range_fault(struct nouveau_svmm *svmm,
-			       struct nouveau_drm *drm, void *data, u32 size,
-			       unsigned long hmm_pfns[], u64 *ioctl_addr,
+			       struct nouveau_drm *drm,
+			       struct nouveau_pfnmap_args *args, u32 size,
+			       unsigned long hmm_flags,
 			       struct svm_notifier *notifier)
 {
 	unsigned long timeout =
 		jiffies + msecs_to_jiffies(HMM_RANGE_DEFAULT_TIMEOUT);
 	/* Have HMM fault pages within the fault window to the GPU. */
+	unsigned long hmm_pfns[1];
 	struct hmm_range range = {
 		.notifier = &notifier->notifier,
 		.start = notifier->notifier.interval_tree.start,
 		.end = notifier->notifier.interval_tree.last + 1,
-		.pfn_flags_mask = HMM_PFN_REQ_FAULT | HMM_PFN_REQ_WRITE,
+		.default_flags = hmm_flags,
 		.hmm_pfns = hmm_pfns,
 		.dev_private_owner = drm->dev,
 	};
@@ -576,11 +588,6 @@ static int nouveau_range_fault(struct nouveau_svmm *svmm,
 		ret = hmm_range_fault(&range);
 		mmap_read_unlock(mm);
 		if (ret) {
-			/*
-			 * FIXME: the input PFN_REQ flags are destroyed on
-			 * -EBUSY, we need to regenerate them, also for the
-			 * other continue below
-			 */
 			if (ret == -EBUSY)
 				continue;
 			return ret;
@@ -595,10 +602,10 @@ static int nouveau_range_fault(struct nouveau_svmm *svmm,
 		break;
 	}
 
-	nouveau_hmm_convert_pfn(drm, &range, ioctl_addr);
+	nouveau_hmm_convert_pfn(drm, &range, args);
 
 	svmm->vmm->vmm.object.client->super = true;
-	ret = nvif_object_ioctl(&svmm->vmm->vmm.object, data, size, NULL);
+	ret = nvif_object_ioctl(&svmm->vmm->vmm.object, args, size, NULL);
 	svmm->vmm->vmm.object.client->super = false;
 	mutex_unlock(&svmm->mutex);
 
@@ -615,17 +622,12 @@ nouveau_svm_fault(struct nvif_notify *notify)
 	struct nvif_object *device = &svm->drm->client.device.object;
 	struct nouveau_svmm *svmm;
 	struct {
-		struct {
-			struct nvif_ioctl_v0 i;
-			struct nvif_ioctl_mthd_v0 m;
-			struct nvif_vmm_pfnmap_v0 p;
-		} i;
-		u64 phys[16];
+		struct nouveau_pfnmap_args i;
+		u64 phys[1];
 	} args;
-	unsigned long hmm_pfns[ARRAY_SIZE(args.phys)];
-	struct vm_area_struct *vma;
+	unsigned long hmm_flags;
 	u64 inst, start, limit;
-	int fi, fn, pi, fill;
+	int fi, fn;
 	int replay = 0, ret;
 
 	/* Parse available fault buffer entries into a cache, and update
@@ -692,128 +694,83 @@ nouveau_svm_fault(struct nvif_notify *notify)
 		 * window into a single update.
 		 */
 		start = buffer->fault[fi]->addr;
-		limit = start + (ARRAY_SIZE(args.phys) << PAGE_SHIFT);
+		limit = start + PAGE_SIZE;
 		if (start < svmm->unmanaged.limit)
 			limit = min_t(u64, limit, svmm->unmanaged.start);
-		SVMM_DBG(svmm, "wndw %016llx-%016llx", start, limit);
 
-		mm = svmm->notifier.mm;
-		if (!mmget_not_zero(mm)) {
-			nouveau_svm_fault_cancel_fault(svm, buffer->fault[fi]);
-			continue;
-		}
-
-		/* Intersect fault window with the CPU VMA, cancelling
-		 * the fault if the address is invalid.
+		/*
+		 * Prepare the GPU-side update of all pages within the
+		 * fault window, determining required pages and access
+		 * permissions based on pending faults.
 		 */
-		mmap_read_lock(mm);
-		vma = find_vma_intersection(mm, start, limit);
-		if (!vma) {
-			SVMM_ERR(svmm, "wndw %016llx-%016llx", start, limit);
-			mmap_read_unlock(mm);
-			mmput(mm);
-			nouveau_svm_fault_cancel_fault(svm, buffer->fault[fi]);
-			continue;
+		args.i.p.addr = start;
+		args.i.p.page = PAGE_SHIFT;
+		args.i.p.size = PAGE_SIZE;
+		/*
+		 * Determine required permissions based on GPU fault
+		 * access flags.
+		 * XXX: atomic?
+		 */
+		switch (buffer->fault[fi]->access) {
+		case 0: /* READ. */
+			hmm_flags = HMM_PFN_REQ_FAULT;
+			break;
+		case 3: /* PREFETCH. */
+			hmm_flags = 0;
+			break;
+		default:
+			hmm_flags = HMM_PFN_REQ_FAULT | HMM_PFN_REQ_WRITE;
+			break;
 		}
-		start = max_t(u64, start, vma->vm_start);
-		limit = min_t(u64, limit, vma->vm_end);
-		mmap_read_unlock(mm);
-		SVMM_DBG(svmm, "wndw %016llx-%016llx", start, limit);
 
-		if (buffer->fault[fi]->addr != start) {
-			SVMM_ERR(svmm, "addr %016llx", buffer->fault[fi]->addr);
-			mmput(mm);
+		mm = svmm->notifier.mm;
+		if (!mmget_not_zero(mm)) {
 			nouveau_svm_fault_cancel_fault(svm, buffer->fault[fi]);
 			continue;
 		}
 
-		/* Prepare the GPU-side update of all pages within the
-		 * fault window, determining required pages and access
-		 * permissions based on pending faults.
-		 */
-		args.i.p.page = PAGE_SHIFT;
-		args.i.p.addr = start;
-		for (fn = fi, pi = 0;;) {
-			/* Determine required permissions based on GPU fault
-			 * access flags.
-			 *XXX: atomic?
-			 */
-			switch (buffer->fault[fn]->access) {
-			case 0: /* READ. */
-				hmm_pfns[pi++] = HMM_PFN_REQ_FAULT;
-				break;
-			case 3: /* PREFETCH. */
-				hmm_pfns[pi++] = 0;
-				break;
-			default:
-				hmm_pfns[pi++] = HMM_PFN_REQ_FAULT |
-						 HMM_PFN_REQ_WRITE;
-				break;
-			}
-			args.i.p.size = pi << PAGE_SHIFT;
+		notifier.svmm = svmm;
+		ret = mmu_interval_notifier_insert(&notifier.notifier, mm,
+						   args.i.p.addr, args.i.p.size,
+						   &nouveau_svm_mni_ops);
+		if (!ret) {
+			ret = nouveau_range_fault(svmm, svm->drm, &args.i,
+				sizeof(args), hmm_flags, &notifier);
+			mmu_interval_notifier_remove(&notifier.notifier);
+		}
+		mmput(mm);
 
+		limit = args.i.p.addr + args.i.p.size;
+		for (fn = fi; ++fn < buffer->fault_nr; ) {
 			/* It's okay to skip over duplicate addresses from the
 			 * same SVMM as faults are ordered by access type such
 			 * that only the first one needs to be handled.
 			 *
 			 * ie. WRITE faults appear first, thus any handling of
 			 * pending READ faults will already be satisfied.
+			 * But if a large page is mapped, make sure subsequent
+			 * fault addresses have sufficient access permission.
 			 */
-			while (++fn < buffer->fault_nr &&
-			       buffer->fault[fn]->svmm == svmm &&
-			       buffer->fault[fn    ]->addr ==
-			       buffer->fault[fn - 1]->addr);
-
-			/* If the next fault is outside the window, or all GPU
-			 * faults have been dealt with, we're done here.
-			 */
-			if (fn >= buffer->fault_nr ||
-			    buffer->fault[fn]->svmm != svmm ||
-			    buffer->fault[fn]->addr >= limit)
+			if (buffer->fault[fn]->svmm != svmm ||
+			    buffer->fault[fn]->addr >= limit ||
+			    (buffer->fault[fi]->access == 0 /* READ. */ &&
+			     !(args.phys[0] & NVIF_VMM_PFNMAP_V0_V)) ||
+			    (buffer->fault[fi]->access != 0 /* READ. */ &&
+			     buffer->fault[fi]->access != 3 /* PREFETCH. */ &&
+			     !(args.phys[0] & NVIF_VMM_PFNMAP_V0_W)))
 				break;
-
-			/* Fill in the gap between this fault and the next. */
-			fill = (buffer->fault[fn    ]->addr -
-				buffer->fault[fn - 1]->addr) >> PAGE_SHIFT;
-			while (--fill)
-				hmm_pfns[pi++] = 0;
 		}
 
-		SVMM_DBG(svmm, "wndw %016llx-%016llx covering %d fault(s)",
-			 args.i.p.addr,
-			 args.i.p.addr + args.i.p.size, fn - fi);
-
-		notifier.svmm = svmm;
-		ret = mmu_interval_notifier_insert(&notifier.notifier,
-						   svmm->notifier.mm,
-						   args.i.p.addr, args.i.p.size,
-						   &nouveau_svm_mni_ops);
-		if (!ret) {
-			ret = nouveau_range_fault(
-				svmm, svm->drm, &args,
-				sizeof(args.i) + pi * sizeof(args.phys[0]),
-				hmm_pfns, args.phys, &notifier);
-			mmu_interval_notifier_remove(&notifier.notifier);
-		}
-		mmput(mm);
+		/* If handling failed completely, cancel all faults. */
+		if (ret) {
+			while (fi < fn) {
+				struct nouveau_svm_fault *fault =
+					buffer->fault[fi++];
 
-		/* Cancel any faults in the window whose pages didn't manage
-		 * to keep their valid bit, or stay writeable when required.
-		 *
-		 * If handling failed completely, cancel all faults.
-		 */
-		while (fi < fn) {
-			struct nouveau_svm_fault *fault = buffer->fault[fi++];
-			pi = (fault->addr - args.i.p.addr) >> PAGE_SHIFT;
-			if (ret ||
-			     !(args.phys[pi] & NVIF_VMM_PFNMAP_V0_V) ||
-			    (!(args.phys[pi] & NVIF_VMM_PFNMAP_V0_W) &&
-			     fault->access != 0 && fault->access != 3)) {
 				nouveau_svm_fault_cancel_fault(svm, fault);
-				continue;
 			}
+		} else
 			replay++;
-		}
 	}
 
 	/* Issue fault replay to the GPU. */
diff --git a/drivers/gpu/drm/nouveau/nouveau_svm.h b/drivers/gpu/drm/nouveau/nouveau_svm.h
index f0fcd1b72e8b..e7d63d7f0c2d 100644
--- a/drivers/gpu/drm/nouveau/nouveau_svm.h
+++ b/drivers/gpu/drm/nouveau/nouveau_svm.h
@@ -1,11 +1,21 @@
 #ifndef __NOUVEAU_SVM_H__
 #define __NOUVEAU_SVM_H__
 #include <nvif/os.h>
+#include <linux/mmu_notifier.h>
 struct drm_device;
 struct drm_file;
 struct nouveau_drm;
 
-struct nouveau_svmm;
+struct nouveau_svmm {
+	struct mmu_notifier notifier;
+	struct nouveau_vmm *vmm;
+	struct {
+		unsigned long start;
+		unsigned long limit;
+	} unmanaged;
+
+	struct mutex mutex;
+};
 
 #if IS_ENABLED(CONFIG_DRM_NOUVEAU_SVM)
 void nouveau_svm_init(struct nouveau_drm *);
@@ -19,6 +29,7 @@ int nouveau_svmm_join(struct nouveau_svmm *, u64 inst);
 void nouveau_svmm_part(struct nouveau_svmm *, u64 inst);
 int nouveau_svmm_bind(struct drm_device *, void *, struct drm_file *);
 
+void nouveau_svmm_invalidate(struct nouveau_svmm *svmm, u64 start, u64 limit);
 u64 *nouveau_pfns_alloc(unsigned long npages);
 void nouveau_pfns_free(u64 *pfns);
 void nouveau_pfns_map(struct nouveau_svmm *svmm, struct mm_struct *mm,
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c
index 199f94e15c5f..19a6804e3989 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c
@@ -1204,7 +1204,6 @@ nvkm_vmm_pfn_unmap(struct nvkm_vmm *vmm, u64 addr, u64 size)
 /*TODO:
  * - Avoid PT readback (for dma_unmap etc), this might end up being dealt
  *   with inside HMM, which would be a lot nicer for us to deal with.
- * - Multiple page sizes (particularly for huge page support).
  * - Support for systems without a 4KiB page size.
  */
 int
@@ -1220,8 +1219,8 @@ nvkm_vmm_pfn_map(struct nvkm_vmm *vmm, u8 shift, u64 addr, u64 size, u64 *pfn)
 	/* Only support mapping where the page size of the incoming page
 	 * array matches a page size available for direct mapping.
 	 */
-	while (page->shift && page->shift != shift &&
-	       page->desc->func->pfn == NULL)
+	while (page->shift && (page->shift != shift ||
+	       page->desc->func->pfn == NULL))
 		page++;
 
 	if (!page->shift || !IS_ALIGNED(addr, 1ULL << shift) ||
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmgp100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmgp100.c
index d86287565542..7eabe9fe0d2b 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmgp100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmgp100.c
@@ -79,8 +79,12 @@ gp100_vmm_pgt_pfn(struct nvkm_vmm *vmm, struct nvkm_mmu_pt *pt,
 	dma_addr_t addr;
 
 	nvkm_kmap(pt->memory);
-	while (ptes--) {
+	for (; ptes; ptes--, map->pfn++) {
 		u64 data = 0;
+
+		if (!(*map->pfn & NVKM_VMM_PFN_V))
+			continue;
+
 		if (!(*map->pfn & NVKM_VMM_PFN_W))
 			data |= BIT_ULL(6); /* RO. */
 
@@ -100,7 +104,6 @@ gp100_vmm_pgt_pfn(struct nvkm_vmm *vmm, struct nvkm_mmu_pt *pt,
 		}
 
 		VMM_WO064(pt, vmm, ptei++ * 8, data);
-		map->pfn++;
 	}
 	nvkm_done(pt->memory);
 }
@@ -258,12 +261,96 @@ gp100_vmm_pd0_unmap(struct nvkm_vmm *vmm,
 	VMM_FO128(pt, vmm, pdei * 0x10, 0ULL, 0ULL, pdes);
 }
 
+static void
+gp100_vmm_pd0_pfn_unmap(struct nvkm_vmm *vmm,
+			struct nvkm_mmu_pt *pt, u32 ptei, u32 ptes)
+{
+	struct device *dev = vmm->mmu->subdev.device->dev;
+	dma_addr_t addr;
+
+	nvkm_kmap(pt->memory);
+	while (ptes--) {
+		u32 datalo = nvkm_ro32(pt->memory, pt->base + ptei * 16 + 0);
+		u32 datahi = nvkm_ro32(pt->memory, pt->base + ptei * 16 + 4);
+		u64 data   = (u64)datahi << 32 | datalo;
+
+		if ((data & (3ULL << 1)) != 0) {
+			addr = (data >> 8) << 12;
+			dma_unmap_page(dev, addr, 1UL << 21, DMA_BIDIRECTIONAL);
+		}
+		ptei++;
+	}
+	nvkm_done(pt->memory);
+}
+
+static bool
+gp100_vmm_pd0_pfn_clear(struct nvkm_vmm *vmm,
+			struct nvkm_mmu_pt *pt, u32 ptei, u32 ptes)
+{
+	bool dma = false;
+
+	nvkm_kmap(pt->memory);
+	while (ptes--) {
+		u32 datalo = nvkm_ro32(pt->memory, pt->base + ptei * 16 + 0);
+		u32 datahi = nvkm_ro32(pt->memory, pt->base + ptei * 16 + 4);
+		u64 data   = (u64)datahi << 32 | datalo;
+
+		if ((data & BIT_ULL(0)) && (data & (3ULL << 1)) != 0) {
+			VMM_WO064(pt, vmm, ptei * 16, data & ~BIT_ULL(0));
+			dma = true;
+		}
+		ptei++;
+	}
+	nvkm_done(pt->memory);
+	return dma;
+}
+
+static void
+gp100_vmm_pd0_pfn(struct nvkm_vmm *vmm, struct nvkm_mmu_pt *pt,
+		  u32 ptei, u32 ptes, struct nvkm_vmm_map *map)
+{
+	struct device *dev = vmm->mmu->subdev.device->dev;
+	dma_addr_t addr;
+
+	nvkm_kmap(pt->memory);
+	for (; ptes; ptes--, map->pfn++) {
+		u64 data = 0;
+
+		if (!(*map->pfn & NVKM_VMM_PFN_V))
+			continue;
+
+		if (!(*map->pfn & NVKM_VMM_PFN_W))
+			data |= BIT_ULL(6); /* RO. */
+
+		if (!(*map->pfn & NVKM_VMM_PFN_VRAM)) {
+			addr = *map->pfn >> NVKM_VMM_PFN_ADDR_SHIFT;
+			addr = dma_map_page(dev, pfn_to_page(addr), 0,
+					    1UL << 21, DMA_BIDIRECTIONAL);
+			if (!WARN_ON(dma_mapping_error(dev, addr))) {
+				data |= addr >> 4;
+				data |= 2ULL << 1; /* SYSTEM_COHERENT_MEMORY. */
+				data |= BIT_ULL(3); /* VOL. */
+				data |= BIT_ULL(0); /* VALID. */
+			}
+		} else {
+			data |= (*map->pfn & NVKM_VMM_PFN_ADDR) >> 4;
+			data |= BIT_ULL(0); /* VALID. */
+		}
+
+		VMM_WO064(pt, vmm, ptei++ * 16, data);
+	}
+	nvkm_done(pt->memory);
+}
+
 static const struct nvkm_vmm_desc_func
 gp100_vmm_desc_pd0 = {
 	.unmap = gp100_vmm_pd0_unmap,
 	.sparse = gp100_vmm_pd0_sparse,
 	.pde = gp100_vmm_pd0_pde,
 	.mem = gp100_vmm_pd0_mem,
+	.pfn = gp100_vmm_pd0_pfn,
+	.pfn_clear = gp100_vmm_pd0_pfn_clear,
+	.pfn_unmap = gp100_vmm_pd0_pfn_unmap,
 };
 
 static void