[PATCH v6 0/8] Replace xe_hmm with gpusvm

[PATCH v6 0/8] Replace xe_hmm with gpusvm

[Matthew Auld]

As a first step to moving userptr handling over to drm, replace the hmm
usage in xe over to gpusvm, which already offers similar functionality. As
some prep steps we also align on some of the missing pieces that were
already handled in xe_hmm.

v2:
  - Rework the gpusvm API based on feedback.
  - Unify SVM and userptr vm struct so we use the same notifier lock.
  - Drop the mark pages as dirty patch.
  - Various other improvements.
v3:
  - Further unify common handling of userptr and svm in xe.
v4:
  - Allow building xe without userptr, like when building on UML.
v5:
  - Rebase on lastest drm-tip. Lots of conflicts.
v6:
  - Various small fixes around kernel-doc etc.
  - Also convert over the newly added xe_vm_advise userptr usage.
-- 
2.51.0

[PATCH v6 1/8] drm/gpusvm: fix hmm_pfn_to_map_order() usage

[Matthew Auld]

Handle the case where the hmm range partially covers a huge page (like
2M), otherwise we can potentially end up doing something nasty like
mapping memory which is outside the range, and maybe not even mapped by
the mm. Fix is based on the xe userptr code, which in a future patch
will directly use gpusvm, so needs alignment here.

v2:
  - Add kernel-doc (Matt B)
  - s/fls/ilog2/ (Thomas)

Reported-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
---
 drivers/gpu/drm/drm_gpusvm.c | 33 +++++++++++++++++++++++++++++++--
 1 file changed, 31 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/drm_gpusvm.c b/drivers/gpu/drm/drm_gpusvm.c
index e2a9a6ae1d54..49f64d26625d 100644
--- a/drivers/gpu/drm/drm_gpusvm.c
+++ b/drivers/gpu/drm/drm_gpusvm.c
@@ -635,6 +635,35 @@ drm_gpusvm_range_alloc(struct drm_gpusvm *gpusvm,
 	return range;
 }
 
+/**
+ * drm_gpusvm_hmm_pfn_to_order() - Get the largest CPU mapping order.
+ * @hmm_pfn: The current hmm_pfn.
+ * @hmm_pfn_index: Index of the @hmm_pfn within the pfn array.
+ * @npages: Number of pages within the pfn array i.e the hmm range size.
+ *
+ * To allow skipping PFNs with the same flags (like when they belong to
+ * the same huge PTE) when looping over the pfn array, take a given a hmm_pfn,
+ * and return the largest order that will fit inside the CPU PTE, but also
+ * crucially accounting for the original hmm range boundaries.
+ *
+ * Return: The largest order that will safely fit within the size of the hmm_pfn
+ * CPU PTE.
+ */
+static unsigned int drm_gpusvm_hmm_pfn_to_order(unsigned long hmm_pfn,
+						unsigned long hmm_pfn_index,
+						unsigned long npages)
+{
+	unsigned long size;
+
+	size = 1UL << hmm_pfn_to_map_order(hmm_pfn);
+	size -= (hmm_pfn & ~HMM_PFN_FLAGS) & (size - 1);
+	hmm_pfn_index += size;
+	if (hmm_pfn_index > npages)
+		size -= (hmm_pfn_index - npages);
+
+	return ilog2(size);
+}
+
 /**
  * drm_gpusvm_check_pages() - Check pages
  * @gpusvm: Pointer to the GPU SVM structure
@@ -693,7 +722,7 @@ static bool drm_gpusvm_check_pages(struct drm_gpusvm *gpusvm,
 			err = -EFAULT;
 			goto err_free;
 		}
-		i += 0x1 << hmm_pfn_to_map_order(pfns[i]);
+		i += 0x1 << drm_gpusvm_hmm_pfn_to_order(pfns[i], i, npages);
 	}
 
 err_free:
@@ -1270,7 +1299,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 	for (i = 0, j = 0; i < npages; ++j) {
 		struct page *page = hmm_pfn_to_page(pfns[i]);
 
-		order = hmm_pfn_to_map_order(pfns[i]);
+		order = drm_gpusvm_hmm_pfn_to_order(pfns[i], i, npages);
 		if (is_device_private_page(page) ||
 		    is_device_coherent_page(page)) {
 			if (zdd != page->zone_device_data && i > 0) {
-- 
2.51.0

[PATCH v6 2/8] drm/gpusvm: use more selective dma dir in get_pages()

[Matthew Auld]

If we are only reading the memory then from the device pov the direction
can be DMA_TO_DEVICE. This aligns with the xe-userptr code. Using the
most restrictive data direction to represent the access is normally a
good idea.

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
---
 drivers/gpu/drm/drm_gpusvm.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/drm_gpusvm.c b/drivers/gpu/drm/drm_gpusvm.c
index 49f64d26625d..b2920b8551f5 100644
--- a/drivers/gpu/drm/drm_gpusvm.c
+++ b/drivers/gpu/drm/drm_gpusvm.c
@@ -1224,6 +1224,8 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 	struct dev_pagemap *pagemap;
 	struct drm_pagemap *dpagemap;
 	struct drm_gpusvm_range_flags flags;
+	enum dma_data_direction dma_dir = ctx->read_only ? DMA_TO_DEVICE :
+							   DMA_BIDIRECTIONAL;
 
 retry:
 	hmm_range.notifier_seq = mmu_interval_read_begin(notifier);
@@ -1329,7 +1331,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 				dpagemap->ops->device_map(dpagemap,
 							  gpusvm->drm->dev,
 							  page, order,
-							  DMA_BIDIRECTIONAL);
+							  dma_dir);
 			if (dma_mapping_error(gpusvm->drm->dev,
 					      range->dma_addr[j].addr)) {
 				err = -EFAULT;
@@ -1351,7 +1353,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 			addr = dma_map_page(gpusvm->drm->dev,
 					    page, 0,
 					    PAGE_SIZE << order,
-					    DMA_BIDIRECTIONAL);
+					    dma_dir);
 			if (dma_mapping_error(gpusvm->drm->dev, addr)) {
 				err = -EFAULT;
 				goto err_unmap;
@@ -1359,7 +1361,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 
 			range->dma_addr[j] = drm_pagemap_addr_encode
 				(addr, DRM_INTERCONNECT_SYSTEM, order,
-				 DMA_BIDIRECTIONAL);
+				 dma_dir);
 		}
 		i += 1 << order;
 		num_dma_mapped = i;
-- 
2.51.0

[PATCH v6 3/8] drm/gpusvm: pull out drm_gpusvm_pages substructure

[Matthew Auld]

Pull the pages stuff from the svm range into its own substructure, with
the idea of having the main pages related routines, like get_pages(),
unmap_pages() and free_pages() all operating on some lower level
structures, which can then be re-used for stuff like userptr.

v2:
  - Move seq into pages struct (Matt B)
v3:
  - Small kernel-doc fixes

Suggested-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
---
 drivers/gpu/drm/drm_gpusvm.c | 67 +++++++++++++++++++-----------------
 drivers/gpu/drm/xe/xe_pt.c   |  2 +-
 drivers/gpu/drm/xe/xe_svm.c  | 16 ++++-----
 drivers/gpu/drm/xe/xe_svm.h  |  6 ++--
 include/drm/drm_gpusvm.h     | 48 +++++++++++++++++---------
 5 files changed, 80 insertions(+), 59 deletions(-)

diff --git a/drivers/gpu/drm/drm_gpusvm.c b/drivers/gpu/drm/drm_gpusvm.c
index b2920b8551f5..fa193dd265b0 100644
--- a/drivers/gpu/drm/drm_gpusvm.c
+++ b/drivers/gpu/drm/drm_gpusvm.c
@@ -629,8 +629,8 @@ drm_gpusvm_range_alloc(struct drm_gpusvm *gpusvm,
 	range->itree.start = ALIGN_DOWN(fault_addr, chunk_size);
 	range->itree.last = ALIGN(fault_addr + 1, chunk_size) - 1;
 	INIT_LIST_HEAD(&range->entry);
-	range->notifier_seq = LONG_MAX;
-	range->flags.migrate_devmem = migrate_devmem ? 1 : 0;
+	range->pages.notifier_seq = LONG_MAX;
+	range->pages.flags.migrate_devmem = migrate_devmem ? 1 : 0;
 
 	return range;
 }
@@ -992,19 +992,20 @@ static void __drm_gpusvm_range_unmap_pages(struct drm_gpusvm *gpusvm,
 					   struct drm_gpusvm_range *range,
 					   unsigned long npages)
 {
-	unsigned long i, j;
-	struct drm_pagemap *dpagemap = range->dpagemap;
+	struct drm_gpusvm_pages *svm_pages = &range->pages;
+	struct drm_pagemap *dpagemap = svm_pages->dpagemap;
 	struct device *dev = gpusvm->drm->dev;
+	unsigned long i, j;
 
 	lockdep_assert_held(&gpusvm->notifier_lock);
 
-	if (range->flags.has_dma_mapping) {
-		struct drm_gpusvm_range_flags flags = {
-			.__flags = range->flags.__flags,
+	if (svm_pages->flags.has_dma_mapping) {
+		struct drm_gpusvm_pages_flags flags = {
+			.__flags = svm_pages->flags.__flags,
 		};
 
 		for (i = 0, j = 0; i < npages; j++) {
-			struct drm_pagemap_addr *addr = &range->dma_addr[j];
+			struct drm_pagemap_addr *addr = &svm_pages->dma_addr[j];
 
 			if (addr->proto == DRM_INTERCONNECT_SYSTEM)
 				dma_unmap_page(dev,
@@ -1020,9 +1021,9 @@ static void __drm_gpusvm_range_unmap_pages(struct drm_gpusvm *gpusvm,
 		/* WRITE_ONCE pairs with READ_ONCE for opportunistic checks */
 		flags.has_devmem_pages = false;
 		flags.has_dma_mapping = false;
-		WRITE_ONCE(range->flags.__flags, flags.__flags);
+		WRITE_ONCE(svm_pages->flags.__flags, flags.__flags);
 
-		range->dpagemap = NULL;
+		svm_pages->dpagemap = NULL;
 	}
 }
 
@@ -1036,11 +1037,13 @@ static void __drm_gpusvm_range_unmap_pages(struct drm_gpusvm *gpusvm,
 static void drm_gpusvm_range_free_pages(struct drm_gpusvm *gpusvm,
 					struct drm_gpusvm_range *range)
 {
+	struct drm_gpusvm_pages *svm_pages = &range->pages;
+
 	lockdep_assert_held(&gpusvm->notifier_lock);
 
-	if (range->dma_addr) {
-		kvfree(range->dma_addr);
-		range->dma_addr = NULL;
+	if (svm_pages->dma_addr) {
+		kvfree(svm_pages->dma_addr);
+		svm_pages->dma_addr = NULL;
 	}
 }
 
@@ -1152,9 +1155,11 @@ EXPORT_SYMBOL_GPL(drm_gpusvm_range_put);
 bool drm_gpusvm_range_pages_valid(struct drm_gpusvm *gpusvm,
 				  struct drm_gpusvm_range *range)
 {
+	struct drm_gpusvm_pages *svm_pages = &range->pages;
+
 	lockdep_assert_held(&gpusvm->notifier_lock);
 
-	return range->flags.has_devmem_pages || range->flags.has_dma_mapping;
+	return svm_pages->flags.has_devmem_pages || svm_pages->flags.has_dma_mapping;
 }
 EXPORT_SYMBOL_GPL(drm_gpusvm_range_pages_valid);
 
@@ -1172,9 +1177,10 @@ static bool
 drm_gpusvm_range_pages_valid_unlocked(struct drm_gpusvm *gpusvm,
 				      struct drm_gpusvm_range *range)
 {
+	struct drm_gpusvm_pages *svm_pages = &range->pages;
 	bool pages_valid;
 
-	if (!range->dma_addr)
+	if (!svm_pages->dma_addr)
 		return false;
 
 	drm_gpusvm_notifier_lock(gpusvm);
@@ -1201,6 +1207,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 			       struct drm_gpusvm_range *range,
 			       const struct drm_gpusvm_ctx *ctx)
 {
+	struct drm_gpusvm_pages *svm_pages = &range->pages;
 	struct mmu_interval_notifier *notifier = &range->notifier->notifier;
 	struct hmm_range hmm_range = {
 		.default_flags = HMM_PFN_REQ_FAULT | (ctx->read_only ? 0 :
@@ -1223,7 +1230,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 	int err = 0;
 	struct dev_pagemap *pagemap;
 	struct drm_pagemap *dpagemap;
-	struct drm_gpusvm_range_flags flags;
+	struct drm_gpusvm_pages_flags flags;
 	enum dma_data_direction dma_dir = ctx->read_only ? DMA_TO_DEVICE :
 							   DMA_BIDIRECTIONAL;
 
@@ -1269,7 +1276,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 	 */
 	drm_gpusvm_notifier_lock(gpusvm);
 
-	flags.__flags = range->flags.__flags;
+	flags.__flags = svm_pages->flags.__flags;
 	if (flags.unmapped) {
 		drm_gpusvm_notifier_unlock(gpusvm);
 		err = -EFAULT;
@@ -1282,13 +1289,12 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 		goto retry;
 	}
 
-	if (!range->dma_addr) {
+	if (!svm_pages->dma_addr) {
 		/* Unlock and restart mapping to allocate memory. */
 		drm_gpusvm_notifier_unlock(gpusvm);
-		range->dma_addr = kvmalloc_array(npages,
-						 sizeof(*range->dma_addr),
-						 GFP_KERNEL);
-		if (!range->dma_addr) {
+		svm_pages->dma_addr =
+			kvmalloc_array(npages, sizeof(*svm_pages->dma_addr), GFP_KERNEL);
+		if (!svm_pages->dma_addr) {
 			err = -ENOMEM;
 			goto err_free;
 		}
@@ -1327,13 +1333,13 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 					goto err_unmap;
 				}
 			}
-			range->dma_addr[j] =
+			svm_pages->dma_addr[j] =
 				dpagemap->ops->device_map(dpagemap,
 							  gpusvm->drm->dev,
 							  page, order,
 							  dma_dir);
 			if (dma_mapping_error(gpusvm->drm->dev,
-					      range->dma_addr[j].addr)) {
+					      svm_pages->dma_addr[j].addr)) {
 				err = -EFAULT;
 				goto err_unmap;
 			}
@@ -1359,7 +1365,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 				goto err_unmap;
 			}
 
-			range->dma_addr[j] = drm_pagemap_addr_encode
+			svm_pages->dma_addr[j] = drm_pagemap_addr_encode
 				(addr, DRM_INTERCONNECT_SYSTEM, order,
 				 dma_dir);
 		}
@@ -1370,16 +1376,16 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 
 	if (pagemap) {
 		flags.has_devmem_pages = true;
-		range->dpagemap = dpagemap;
+		svm_pages->dpagemap = dpagemap;
 	}
 
 	/* WRITE_ONCE pairs with READ_ONCE for opportunistic checks */
-	WRITE_ONCE(range->flags.__flags, flags.__flags);
+	WRITE_ONCE(svm_pages->flags.__flags, flags.__flags);
 
 	drm_gpusvm_notifier_unlock(gpusvm);
 	kvfree(pfns);
 set_seqno:
-	range->notifier_seq = hmm_range.notifier_seq;
+	svm_pages->notifier_seq = hmm_range.notifier_seq;
 
 	return 0;
 
@@ -1396,7 +1402,6 @@ EXPORT_SYMBOL_GPL(drm_gpusvm_range_get_pages);
 
 /**
  * drm_gpusvm_range_unmap_pages() - Unmap pages associated with a GPU SVM range
- * drm_gpusvm_range_evict() - Evict GPU SVM range
  * @gpusvm: Pointer to the GPU SVM structure
  * @range: Pointer to the GPU SVM range structure
  * @ctx: GPU SVM context
@@ -1520,10 +1525,10 @@ void drm_gpusvm_range_set_unmapped(struct drm_gpusvm_range *range,
 {
 	lockdep_assert_held_write(&range->gpusvm->notifier_lock);
 
-	range->flags.unmapped = true;
+	range->pages.flags.unmapped = true;
 	if (drm_gpusvm_range_start(range) < mmu_range->start ||
 	    drm_gpusvm_range_end(range) > mmu_range->end)
-		range->flags.partial_unmap = true;
+		range->pages.flags.partial_unmap = true;
 }
 EXPORT_SYMBOL_GPL(drm_gpusvm_range_set_unmapped);
 
diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
index c129048a9a09..029480c9cebd 100644
--- a/drivers/gpu/drm/xe/xe_pt.c
+++ b/drivers/gpu/drm/xe/xe_pt.c
@@ -729,7 +729,7 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
 			return -EAGAIN;
 		}
 		if (xe_svm_range_has_dma_mapping(range)) {
-			xe_res_first_dma(range->base.dma_addr, 0,
+			xe_res_first_dma(range->base.pages.dma_addr, 0,
 					 range->base.itree.last + 1 - range->base.itree.start,
 					 &curs);
 			xe_svm_range_debug(range, "BIND PREPARE - MIXED");
diff --git a/drivers/gpu/drm/xe/xe_svm.c b/drivers/gpu/drm/xe/xe_svm.c
index 76c6d74c1208..d6ef0cd8a02c 100644
--- a/drivers/gpu/drm/xe/xe_svm.c
+++ b/drivers/gpu/drm/xe/xe_svm.c
@@ -25,9 +25,9 @@ static bool xe_svm_range_in_vram(struct xe_svm_range *range)
 	 * memory.
 	 */
 
-	struct drm_gpusvm_range_flags flags = {
+	struct drm_gpusvm_pages_flags flags = {
 		/* Pairs with WRITE_ONCE in drm_gpusvm.c */
-		.__flags = READ_ONCE(range->base.flags.__flags),
+		.__flags = READ_ONCE(range->base.pages.flags.__flags),
 	};
 
 	return flags.has_devmem_pages;
@@ -57,7 +57,7 @@ static struct xe_vm *range_to_vm(struct drm_gpusvm_range *r)
 	       (r__)->base.gpusvm,					\
 	       xe_svm_range_in_vram((r__)) ? 1 : 0,			\
 	       xe_svm_range_has_vram_binding((r__)) ? 1 : 0,		\
-	       (r__)->base.notifier_seq,				\
+	       (r__)->base.pages.notifier_seq,				\
 	       xe_svm_range_start((r__)), xe_svm_range_end((r__)),	\
 	       xe_svm_range_size((r__)))
 
@@ -128,7 +128,7 @@ xe_svm_range_notifier_event_begin(struct xe_vm *vm, struct drm_gpusvm_range *r,
 	range_debug(range, "NOTIFIER");
 
 	/* Skip if already unmapped or if no binding exist */
-	if (range->base.flags.unmapped || !range->tile_present)
+	if (range->base.pages.flags.unmapped || !range->tile_present)
 		return 0;
 
 	range_debug(range, "NOTIFIER - EXECUTE");
@@ -716,7 +716,7 @@ bool xe_svm_range_validate(struct xe_vm *vm,
 	xe_svm_notifier_lock(vm);
 
 	ret = (range->tile_present & ~range->tile_invalidated & tile_mask) == tile_mask &&
-	       (devmem_preferred == range->base.flags.has_devmem_pages);
+	       (devmem_preferred == range->base.pages.flags.has_devmem_pages);
 
 	xe_svm_notifier_unlock(vm);
 
@@ -827,7 +827,7 @@ bool xe_svm_range_needs_migrate_to_vram(struct xe_svm_range *range, struct xe_vm
 	struct xe_vm *vm = range_to_vm(&range->base);
 	u64 range_size = xe_svm_range_size(range);
 
-	if (!range->base.flags.migrate_devmem || !preferred_region_is_vram)
+	if (!range->base.pages.flags.migrate_devmem || !preferred_region_is_vram)
 		return false;
 
 	xe_assert(vm->xe, IS_DGFX(vm->xe));
@@ -884,7 +884,7 @@ static int __xe_svm_handle_pagefault(struct xe_vm *vm, struct xe_vma *vma,
 	if (IS_ERR(range))
 		return PTR_ERR(range);
 
-	if (ctx.devmem_only && !range->base.flags.migrate_devmem)
+	if (ctx.devmem_only && !range->base.pages.flags.migrate_devmem)
 		return -EACCES;
 
 	if (xe_svm_range_is_valid(range, tile, ctx.devmem_only))
@@ -1221,7 +1221,7 @@ int xe_svm_alloc_vram(struct xe_tile *tile, struct xe_svm_range *range,
 {
 	struct drm_pagemap *dpagemap;
 
-	xe_assert(tile_to_xe(tile), range->base.flags.migrate_devmem);
+	xe_assert(tile_to_xe(tile), range->base.pages.flags.migrate_devmem);
 	range_debug(range, "ALLOCATE VRAM");
 
 	dpagemap = tile_local_pagemap(tile);
diff --git a/drivers/gpu/drm/xe/xe_svm.h b/drivers/gpu/drm/xe/xe_svm.h
index 9d6a8840a8b7..f0e0e009c7fc 100644
--- a/drivers/gpu/drm/xe/xe_svm.h
+++ b/drivers/gpu/drm/xe/xe_svm.h
@@ -105,7 +105,7 @@ struct drm_pagemap *xe_vma_resolve_pagemap(struct xe_vma *vma, struct xe_tile *t
 static inline bool xe_svm_range_has_dma_mapping(struct xe_svm_range *range)
 {
 	lockdep_assert_held(&range->base.gpusvm->notifier_lock);
-	return range->base.flags.has_dma_mapping;
+	return range->base.pages.flags.has_dma_mapping;
 }
 
 /**
@@ -184,7 +184,9 @@ struct xe_vram_region;
 struct xe_svm_range {
 	struct {
 		struct interval_tree_node itree;
-		const struct drm_pagemap_addr *dma_addr;
+		struct {
+			const struct drm_pagemap_addr *dma_addr;
+		} pages;
 	} base;
 	u32 tile_present;
 	u32 tile_invalidated;
diff --git a/include/drm/drm_gpusvm.h b/include/drm/drm_gpusvm.h
index 0e336148309d..1ee4188c3067 100644
--- a/include/drm/drm_gpusvm.h
+++ b/include/drm/drm_gpusvm.h
@@ -106,16 +106,16 @@ struct drm_gpusvm_notifier {
 };
 
 /**
- * struct drm_gpusvm_range_flags - Structure representing a GPU SVM range flags
+ * struct drm_gpusvm_pages_flags - Structure representing a GPU SVM pages flags
  *
- * @migrate_devmem: Flag indicating whether the range can be migrated to device memory
- * @unmapped: Flag indicating if the range has been unmapped
- * @partial_unmap: Flag indicating if the range has been partially unmapped
- * @has_devmem_pages: Flag indicating if the range has devmem pages
- * @has_dma_mapping: Flag indicating if the range has a DMA mapping
- * @__flags: Flags for range in u16 form (used for READ_ONCE)
+ * @migrate_devmem: Flag indicating whether the pages can be migrated to device memory
+ * @unmapped: Flag indicating if the pages has been unmapped
+ * @partial_unmap: Flag indicating if the pages has been partially unmapped
+ * @has_devmem_pages: Flag indicating if the pages has devmem pages
+ * @has_dma_mapping: Flag indicating if the pages has a DMA mapping
+ * @__flags: Flags for pages in u16 form (used for READ_ONCE)
  */
-struct drm_gpusvm_range_flags {
+struct drm_gpusvm_pages_flags {
 	union {
 		struct {
 			/* All flags below must be set upon creation */
@@ -130,6 +130,27 @@ struct drm_gpusvm_range_flags {
 	};
 };
 
+/**
+ * struct drm_gpusvm_pages - Structure representing a GPU SVM mapped pages
+ *
+ * @dma_addr: Device address array
+ * @dpagemap: The struct drm_pagemap of the device pages we're dma-mapping.
+ *            Note this is assuming only one drm_pagemap per range is allowed.
+ * @notifier_seq: Notifier sequence number of the range's pages
+ * @flags: Flags for range
+ * @flags.migrate_devmem: Flag indicating whether the range can be migrated to device memory
+ * @flags.unmapped: Flag indicating if the range has been unmapped
+ * @flags.partial_unmap: Flag indicating if the range has been partially unmapped
+ * @flags.has_devmem_pages: Flag indicating if the range has devmem pages
+ * @flags.has_dma_mapping: Flag indicating if the range has a DMA mapping
+ */
+struct drm_gpusvm_pages {
+	struct drm_pagemap_addr *dma_addr;
+	struct drm_pagemap *dpagemap;
+	unsigned long notifier_seq;
+	struct drm_gpusvm_pages_flags flags;
+};
+
 /**
  * struct drm_gpusvm_range - Structure representing a GPU SVM range
  *
@@ -138,11 +159,7 @@ struct drm_gpusvm_range_flags {
  * @refcount: Reference count for the range
  * @itree: Interval tree node for the range (inserted in GPU SVM notifier)
  * @entry: List entry to fast interval tree traversal
- * @notifier_seq: Notifier sequence number of the range's pages
- * @dma_addr: Device address array
- * @dpagemap: The struct drm_pagemap of the device pages we're dma-mapping.
- *            Note this is assuming only one drm_pagemap per range is allowed.
- * @flags: Flags for range
+ * @pages: The pages for this range.
  *
  * This structure represents a GPU SVM range used for tracking memory ranges
  * mapped in a DRM device.
@@ -153,10 +170,7 @@ struct drm_gpusvm_range {
 	struct kref refcount;
 	struct interval_tree_node itree;
 	struct list_head entry;
-	unsigned long notifier_seq;
-	struct drm_pagemap_addr *dma_addr;
-	struct drm_pagemap *dpagemap;
-	struct drm_gpusvm_range_flags flags;
+	struct drm_gpusvm_pages pages;
 };
 
 /**
-- 
2.51.0

[PATCH v6 4/8] drm/gpusvm: refactor core API to use pages struct

[Matthew Auld]

Refactor the core API of get/unmap/free pages to all operate on
drm_gpusvm_pages. In the next patch we want to export a simplified core
API without needing fully blown svm range etc.

Suggested-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
---
 drivers/gpu/drm/drm_gpusvm.c | 161 ++++++++++++++++++++++++-----------
 1 file changed, 110 insertions(+), 51 deletions(-)

diff --git a/drivers/gpu/drm/drm_gpusvm.c b/drivers/gpu/drm/drm_gpusvm.c
index fa193dd265b0..eb50ed0d0f55 100644
--- a/drivers/gpu/drm/drm_gpusvm.c
+++ b/drivers/gpu/drm/drm_gpusvm.c
@@ -980,19 +980,18 @@ drm_gpusvm_range_find_or_insert(struct drm_gpusvm *gpusvm,
 EXPORT_SYMBOL_GPL(drm_gpusvm_range_find_or_insert);
 
 /**
- * __drm_gpusvm_range_unmap_pages() - Unmap pages associated with a GPU SVM range (internal)
+ * __drm_gpusvm_unmap_pages() - Unmap pages associated with GPU SVM pages (internal)
  * @gpusvm: Pointer to the GPU SVM structure
- * @range: Pointer to the GPU SVM range structure
+ * @svm_pages: Pointer to the GPU SVM pages structure
  * @npages: Number of pages to unmap
  *
- * This function unmap pages associated with a GPU SVM range. Assumes and
+ * This function unmap pages associated with a GPU SVM pages struct. Assumes and
  * asserts correct locking is in place when called.
  */
-static void __drm_gpusvm_range_unmap_pages(struct drm_gpusvm *gpusvm,
-					   struct drm_gpusvm_range *range,
-					   unsigned long npages)
+static void __drm_gpusvm_unmap_pages(struct drm_gpusvm *gpusvm,
+				     struct drm_gpusvm_pages *svm_pages,
+				     unsigned long npages)
 {
-	struct drm_gpusvm_pages *svm_pages = &range->pages;
 	struct drm_pagemap *dpagemap = svm_pages->dpagemap;
 	struct device *dev = gpusvm->drm->dev;
 	unsigned long i, j;
@@ -1028,17 +1027,15 @@ static void __drm_gpusvm_range_unmap_pages(struct drm_gpusvm *gpusvm,
 }
 
 /**
- * drm_gpusvm_range_free_pages() - Free pages associated with a GPU SVM range
+ * __drm_gpusvm_free_pages() - Free dma array associated with GPU SVM pages
  * @gpusvm: Pointer to the GPU SVM structure
- * @range: Pointer to the GPU SVM range structure
+ * @svm_pages: Pointer to the GPU SVM pages structure
  *
  * This function frees the dma address array associated with a GPU SVM range.
  */
-static void drm_gpusvm_range_free_pages(struct drm_gpusvm *gpusvm,
-					struct drm_gpusvm_range *range)
+static void __drm_gpusvm_free_pages(struct drm_gpusvm *gpusvm,
+				    struct drm_gpusvm_pages *svm_pages)
 {
-	struct drm_gpusvm_pages *svm_pages = &range->pages;
-
 	lockdep_assert_held(&gpusvm->notifier_lock);
 
 	if (svm_pages->dma_addr) {
@@ -1072,8 +1069,8 @@ void drm_gpusvm_range_remove(struct drm_gpusvm *gpusvm,
 		return;
 
 	drm_gpusvm_notifier_lock(gpusvm);
-	__drm_gpusvm_range_unmap_pages(gpusvm, range, npages);
-	drm_gpusvm_range_free_pages(gpusvm, range);
+	__drm_gpusvm_unmap_pages(gpusvm, &range->pages, npages);
+	__drm_gpusvm_free_pages(gpusvm, &range->pages);
 	__drm_gpusvm_range_remove(notifier, range);
 	drm_gpusvm_notifier_unlock(gpusvm);
 
@@ -1138,6 +1135,28 @@ void drm_gpusvm_range_put(struct drm_gpusvm_range *range)
 }
 EXPORT_SYMBOL_GPL(drm_gpusvm_range_put);
 
+/**
+ * drm_gpusvm_pages_valid() - GPU SVM range pages valid
+ * @gpusvm: Pointer to the GPU SVM structure
+ * @svm_pages: Pointer to the GPU SVM pages structure
+ *
+ * This function determines if a GPU SVM range pages are valid. Expected be
+ * called holding gpusvm->notifier_lock and as the last step before committing a
+ * GPU binding. This is akin to a notifier seqno check in the HMM documentation
+ * but due to wider notifiers (i.e., notifiers which span multiple ranges) this
+ * function is required for finer grained checking (i.e., per range) if pages
+ * are valid.
+ *
+ * Return: True if GPU SVM range has valid pages, False otherwise
+ */
+static bool drm_gpusvm_pages_valid(struct drm_gpusvm *gpusvm,
+				   struct drm_gpusvm_pages *svm_pages)
+{
+	lockdep_assert_held(&gpusvm->notifier_lock);
+
+	return svm_pages->flags.has_devmem_pages || svm_pages->flags.has_dma_mapping;
+}
+
 /**
  * drm_gpusvm_range_pages_valid() - GPU SVM range pages valid
  * @gpusvm: Pointer to the GPU SVM structure
@@ -1155,11 +1174,7 @@ EXPORT_SYMBOL_GPL(drm_gpusvm_range_put);
 bool drm_gpusvm_range_pages_valid(struct drm_gpusvm *gpusvm,
 				  struct drm_gpusvm_range *range)
 {
-	struct drm_gpusvm_pages *svm_pages = &range->pages;
-
-	lockdep_assert_held(&gpusvm->notifier_lock);
-
-	return svm_pages->flags.has_devmem_pages || svm_pages->flags.has_dma_mapping;
+	return drm_gpusvm_pages_valid(gpusvm, &range->pages);
 }
 EXPORT_SYMBOL_GPL(drm_gpusvm_range_pages_valid);
 
@@ -1173,57 +1188,59 @@ EXPORT_SYMBOL_GPL(drm_gpusvm_range_pages_valid);
  *
  * Return: True if GPU SVM range has valid pages, False otherwise
  */
-static bool
-drm_gpusvm_range_pages_valid_unlocked(struct drm_gpusvm *gpusvm,
-				      struct drm_gpusvm_range *range)
+static bool drm_gpusvm_pages_valid_unlocked(struct drm_gpusvm *gpusvm,
+					    struct drm_gpusvm_pages *svm_pages)
 {
-	struct drm_gpusvm_pages *svm_pages = &range->pages;
 	bool pages_valid;
 
 	if (!svm_pages->dma_addr)
 		return false;
 
 	drm_gpusvm_notifier_lock(gpusvm);
-	pages_valid = drm_gpusvm_range_pages_valid(gpusvm, range);
+	pages_valid = drm_gpusvm_pages_valid(gpusvm, svm_pages);
 	if (!pages_valid)
-		drm_gpusvm_range_free_pages(gpusvm, range);
+		__drm_gpusvm_free_pages(gpusvm, svm_pages);
 	drm_gpusvm_notifier_unlock(gpusvm);
 
 	return pages_valid;
 }
 
 /**
- * drm_gpusvm_range_get_pages() - Get pages for a GPU SVM range
+ * drm_gpusvm_get_pages() - Get pages and populate GPU SVM pages struct
  * @gpusvm: Pointer to the GPU SVM structure
- * @range: Pointer to the GPU SVM range structure
+ * @svm_pages: The SVM pages to populate. This will contain the dma-addresses
+ * @mm: The mm corresponding to the CPU range
+ * @notifier: The corresponding notifier for the given CPU range
+ * @pages_start: Start CPU address for the pages
+ * @pages_end: End CPU address for the pages (exclusive)
  * @ctx: GPU SVM context
  *
- * This function gets pages for a GPU SVM range and ensures they are mapped for
- * DMA access.
+ * This function gets and maps pages for CPU range and ensures they are
+ * mapped for DMA access.
  *
  * Return: 0 on success, negative error code on failure.
  */
-int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
-			       struct drm_gpusvm_range *range,
-			       const struct drm_gpusvm_ctx *ctx)
+static int drm_gpusvm_get_pages(struct drm_gpusvm *gpusvm,
+				struct drm_gpusvm_pages *svm_pages,
+				struct mm_struct *mm,
+				struct mmu_interval_notifier *notifier,
+				unsigned long pages_start,
+				unsigned long pages_end,
+				const struct drm_gpusvm_ctx *ctx)
 {
-	struct drm_gpusvm_pages *svm_pages = &range->pages;
-	struct mmu_interval_notifier *notifier = &range->notifier->notifier;
 	struct hmm_range hmm_range = {
 		.default_flags = HMM_PFN_REQ_FAULT | (ctx->read_only ? 0 :
 			HMM_PFN_REQ_WRITE),
 		.notifier = notifier,
-		.start = drm_gpusvm_range_start(range),
-		.end = drm_gpusvm_range_end(range),
+		.start = pages_start,
+		.end = pages_end,
 		.dev_private_owner = gpusvm->device_private_page_owner,
 	};
-	struct mm_struct *mm = gpusvm->mm;
 	void *zdd;
 	unsigned long timeout =
 		jiffies + msecs_to_jiffies(HMM_RANGE_DEFAULT_TIMEOUT);
 	unsigned long i, j;
-	unsigned long npages = npages_in_range(drm_gpusvm_range_start(range),
-					       drm_gpusvm_range_end(range));
+	unsigned long npages = npages_in_range(pages_start, pages_end);
 	unsigned long num_dma_mapped;
 	unsigned int order = 0;
 	unsigned long *pfns;
@@ -1236,7 +1253,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 
 retry:
 	hmm_range.notifier_seq = mmu_interval_read_begin(notifier);
-	if (drm_gpusvm_range_pages_valid_unlocked(gpusvm, range))
+	if (drm_gpusvm_pages_valid_unlocked(gpusvm, svm_pages))
 		goto set_seqno;
 
 	pfns = kvmalloc_array(npages, sizeof(*pfns), GFP_KERNEL);
@@ -1390,7 +1407,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 	return 0;
 
 err_unmap:
-	__drm_gpusvm_range_unmap_pages(gpusvm, range, num_dma_mapped);
+	__drm_gpusvm_unmap_pages(gpusvm, svm_pages, num_dma_mapped);
 	drm_gpusvm_notifier_unlock(gpusvm);
 err_free:
 	kvfree(pfns);
@@ -1398,8 +1415,58 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
 		goto retry;
 	return err;
 }
+
+/**
+ * drm_gpusvm_range_get_pages() - Get pages for a GPU SVM range
+ * @gpusvm: Pointer to the GPU SVM structure
+ * @range: Pointer to the GPU SVM range structure
+ * @ctx: GPU SVM context
+ *
+ * This function gets pages for a GPU SVM range and ensures they are mapped for
+ * DMA access.
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,
+			       struct drm_gpusvm_range *range,
+			       const struct drm_gpusvm_ctx *ctx)
+{
+	return drm_gpusvm_get_pages(gpusvm, &range->pages, gpusvm->mm,
+				    &range->notifier->notifier,
+				    drm_gpusvm_range_start(range),
+				    drm_gpusvm_range_end(range), ctx);
+}
 EXPORT_SYMBOL_GPL(drm_gpusvm_range_get_pages);
 
+/**
+ * drm_gpusvm_unmap_pages() - Unmap GPU svm pages
+ * @gpusvm: Pointer to the GPU SVM structure
+ * @svm_pages: Pointer to the GPU SVM pages structure
+ * @npages: Number of pages in @svm_pages.
+ * @ctx: GPU SVM context
+ *
+ * This function unmaps pages associated with a GPU SVM pages struct. If
+ * @in_notifier is set, it is assumed that gpusvm->notifier_lock is held in
+ * write mode; if it is clear, it acquires gpusvm->notifier_lock in read mode.
+ * Must be called in the invalidate() callback of the corresponding notifier for
+ * IOMMU security model.
+ */
+static void drm_gpusvm_unmap_pages(struct drm_gpusvm *gpusvm,
+				   struct drm_gpusvm_pages *svm_pages,
+				   unsigned long npages,
+				   const struct drm_gpusvm_ctx *ctx)
+{
+	if (ctx->in_notifier)
+		lockdep_assert_held_write(&gpusvm->notifier_lock);
+	else
+		drm_gpusvm_notifier_lock(gpusvm);
+
+	__drm_gpusvm_unmap_pages(gpusvm, svm_pages, npages);
+
+	if (!ctx->in_notifier)
+		drm_gpusvm_notifier_unlock(gpusvm);
+}
+
 /**
  * drm_gpusvm_range_unmap_pages() - Unmap pages associated with a GPU SVM range
  * @gpusvm: Pointer to the GPU SVM structure
@@ -1419,15 +1486,7 @@ void drm_gpusvm_range_unmap_pages(struct drm_gpusvm *gpusvm,
 	unsigned long npages = npages_in_range(drm_gpusvm_range_start(range),
 					       drm_gpusvm_range_end(range));
 
-	if (ctx->in_notifier)
-		lockdep_assert_held_write(&gpusvm->notifier_lock);
-	else
-		drm_gpusvm_notifier_lock(gpusvm);
-
-	__drm_gpusvm_range_unmap_pages(gpusvm, range, npages);
-
-	if (!ctx->in_notifier)
-		drm_gpusvm_notifier_unlock(gpusvm);
+	return drm_gpusvm_unmap_pages(gpusvm, &range->pages, npages, ctx);
 }
 EXPORT_SYMBOL_GPL(drm_gpusvm_range_unmap_pages);
 
-- 
2.51.0

[PATCH v6 5/8] drm/gpusvm: export drm_gpusvm_pages API

[Matthew Auld]

Export get/unmap/free pages API. We also need to tweak the SVM init to
allow skipping much of the unneeded parts.

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
---
 drivers/gpu/drm/drm_gpusvm.c | 66 ++++++++++++++++++++++++++++--------
 include/drm/drm_gpusvm.h     | 16 +++++++++
 2 files changed, 67 insertions(+), 15 deletions(-)

diff --git a/drivers/gpu/drm/drm_gpusvm.c b/drivers/gpu/drm/drm_gpusvm.c
index eb50ed0d0f55..eeeeb99cfdf6 100644
--- a/drivers/gpu/drm/drm_gpusvm.c
+++ b/drivers/gpu/drm/drm_gpusvm.c
@@ -373,6 +373,12 @@ static const struct mmu_interval_notifier_ops drm_gpusvm_notifier_ops = {
  *
  * This function initializes the GPU SVM.
  *
+ * Note: If only using the simple drm_gpusvm_pages API (get/unmap/free),
+ * then only @gpusvm, @name, and @drm are expected. However, the same base
+ * @gpusvm can also be used with both modes together in which case the full
+ * setup is needed, where the core drm_gpusvm_pages API will simply never use
+ * the other fields.
+ *
  * Return: 0 on success, a negative error code on failure.
  */
 int drm_gpusvm_init(struct drm_gpusvm *gpusvm,
@@ -383,8 +389,16 @@ int drm_gpusvm_init(struct drm_gpusvm *gpusvm,
 		    const struct drm_gpusvm_ops *ops,
 		    const unsigned long *chunk_sizes, int num_chunks)
 {
-	if (!ops->invalidate || !num_chunks)
-		return -EINVAL;
+	if (mm) {
+		if (!ops->invalidate || !num_chunks)
+			return -EINVAL;
+		mmgrab(mm);
+	} else {
+		/* No full SVM mode, only core drm_gpusvm_pages API. */
+		if (ops || num_chunks || mm_range || notifier_size ||
+		    device_private_page_owner)
+			return -EINVAL;
+	}
 
 	gpusvm->name = name;
 	gpusvm->drm = drm;
@@ -397,7 +411,6 @@ int drm_gpusvm_init(struct drm_gpusvm *gpusvm,
 	gpusvm->chunk_sizes = chunk_sizes;
 	gpusvm->num_chunks = num_chunks;
 
-	mmgrab(mm);
 	gpusvm->root = RB_ROOT_CACHED;
 	INIT_LIST_HEAD(&gpusvm->notifier_list);
 
@@ -489,7 +502,8 @@ void drm_gpusvm_fini(struct drm_gpusvm *gpusvm)
 			drm_gpusvm_range_remove(gpusvm, range);
 	}
 
-	mmdrop(gpusvm->mm);
+	if (gpusvm->mm)
+		mmdrop(gpusvm->mm);
 	WARN_ON(!RB_EMPTY_ROOT(&gpusvm->root.rb_root));
 }
 EXPORT_SYMBOL_GPL(drm_gpusvm_fini);
@@ -1044,6 +1058,27 @@ static void __drm_gpusvm_free_pages(struct drm_gpusvm *gpusvm,
 	}
 }
 
+/**
+ * drm_gpusvm_free_pages() - Free dma-mapping associated with GPU SVM pages
+ * struct
+ * @gpusvm: Pointer to the GPU SVM structure
+ * @svm_pages: Pointer to the GPU SVM pages structure
+ * @npages: Number of mapped pages
+ *
+ * This function unmaps and frees the dma address array associated with a GPU
+ * SVM pages struct.
+ */
+void drm_gpusvm_free_pages(struct drm_gpusvm *gpusvm,
+			   struct drm_gpusvm_pages *svm_pages,
+			   unsigned long npages)
+{
+	drm_gpusvm_notifier_lock(gpusvm);
+	__drm_gpusvm_unmap_pages(gpusvm, svm_pages, npages);
+	__drm_gpusvm_free_pages(gpusvm, svm_pages);
+	drm_gpusvm_notifier_unlock(gpusvm);
+}
+EXPORT_SYMBOL_GPL(drm_gpusvm_free_pages);
+
 /**
  * drm_gpusvm_range_remove() - Remove GPU SVM range
  * @gpusvm: Pointer to the GPU SVM structure
@@ -1220,13 +1255,12 @@ static bool drm_gpusvm_pages_valid_unlocked(struct drm_gpusvm *gpusvm,
  *
  * Return: 0 on success, negative error code on failure.
  */
-static int drm_gpusvm_get_pages(struct drm_gpusvm *gpusvm,
-				struct drm_gpusvm_pages *svm_pages,
-				struct mm_struct *mm,
-				struct mmu_interval_notifier *notifier,
-				unsigned long pages_start,
-				unsigned long pages_end,
-				const struct drm_gpusvm_ctx *ctx)
+int drm_gpusvm_get_pages(struct drm_gpusvm *gpusvm,
+			 struct drm_gpusvm_pages *svm_pages,
+			 struct mm_struct *mm,
+			 struct mmu_interval_notifier *notifier,
+			 unsigned long pages_start, unsigned long pages_end,
+			 const struct drm_gpusvm_ctx *ctx)
 {
 	struct hmm_range hmm_range = {
 		.default_flags = HMM_PFN_REQ_FAULT | (ctx->read_only ? 0 :
@@ -1415,6 +1449,7 @@ static int drm_gpusvm_get_pages(struct drm_gpusvm *gpusvm,
 		goto retry;
 	return err;
 }
+EXPORT_SYMBOL_GPL(drm_gpusvm_get_pages);
 
 /**
  * drm_gpusvm_range_get_pages() - Get pages for a GPU SVM range
@@ -1451,10 +1486,10 @@ EXPORT_SYMBOL_GPL(drm_gpusvm_range_get_pages);
  * Must be called in the invalidate() callback of the corresponding notifier for
  * IOMMU security model.
  */
-static void drm_gpusvm_unmap_pages(struct drm_gpusvm *gpusvm,
-				   struct drm_gpusvm_pages *svm_pages,
-				   unsigned long npages,
-				   const struct drm_gpusvm_ctx *ctx)
+void drm_gpusvm_unmap_pages(struct drm_gpusvm *gpusvm,
+			    struct drm_gpusvm_pages *svm_pages,
+			    unsigned long npages,
+			    const struct drm_gpusvm_ctx *ctx)
 {
 	if (ctx->in_notifier)
 		lockdep_assert_held_write(&gpusvm->notifier_lock);
@@ -1466,6 +1501,7 @@ static void drm_gpusvm_unmap_pages(struct drm_gpusvm *gpusvm,
 	if (!ctx->in_notifier)
 		drm_gpusvm_notifier_unlock(gpusvm);
 }
+EXPORT_SYMBOL_GPL(drm_gpusvm_unmap_pages);
 
 /**
  * drm_gpusvm_range_unmap_pages() - Unmap pages associated with a GPU SVM range
diff --git a/include/drm/drm_gpusvm.h b/include/drm/drm_gpusvm.h
index 1ee4188c3067..5434048a2ca4 100644
--- a/include/drm/drm_gpusvm.h
+++ b/include/drm/drm_gpusvm.h
@@ -307,6 +307,22 @@ drm_gpusvm_range_find(struct drm_gpusvm_notifier *notifier, unsigned long start,
 void drm_gpusvm_range_set_unmapped(struct drm_gpusvm_range *range,
 				   const struct mmu_notifier_range *mmu_range);
 
+int drm_gpusvm_get_pages(struct drm_gpusvm *gpusvm,
+			 struct drm_gpusvm_pages *svm_pages,
+			 struct mm_struct *mm,
+			 struct mmu_interval_notifier *notifier,
+			 unsigned long pages_start, unsigned long pages_end,
+			 const struct drm_gpusvm_ctx *ctx);
+
+void drm_gpusvm_unmap_pages(struct drm_gpusvm *gpusvm,
+			    struct drm_gpusvm_pages *svm_pages,
+			    unsigned long npages,
+			    const struct drm_gpusvm_ctx *ctx);
+
+void drm_gpusvm_free_pages(struct drm_gpusvm *gpusvm,
+			   struct drm_gpusvm_pages *svm_pages,
+			   unsigned long npages);
+
 #ifdef CONFIG_LOCKDEP
 /**
  * drm_gpusvm_driver_set_lock() - Set the lock protecting accesses to GPU SVM
-- 
2.51.0

[PATCH v6 6/8] drm/xe/vm: split userptr bits into separate file

[Matthew Auld]

This will simplify compiling out the bits that depend on DRM_GPUSVM in a
later patch. Without this we end up littering the code with ifdef
checks, plus it becomes hard to be sure that something won't blow at
runtime due to something not being initialised, even though it passed
the build. Should be no functional change here.

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
---
 drivers/gpu/drm/xe/Makefile      |   1 +
 drivers/gpu/drm/xe/xe_pt.c       |   1 +
 drivers/gpu/drm/xe/xe_userptr.c  | 305 +++++++++++++++++++++++++++++++
 drivers/gpu/drm/xe/xe_userptr.h  |  97 ++++++++++
 drivers/gpu/drm/xe/xe_vm.c       | 278 +---------------------------
 drivers/gpu/drm/xe/xe_vm.h       |  17 --
 drivers/gpu/drm/xe/xe_vm_types.h |  60 +-----
 7 files changed, 411 insertions(+), 348 deletions(-)
 create mode 100644 drivers/gpu/drm/xe/xe_userptr.c
 create mode 100644 drivers/gpu/drm/xe/xe_userptr.h

diff --git a/drivers/gpu/drm/xe/Makefile b/drivers/gpu/drm/xe/Makefile
index 987e4fe10538..85ead33f69ab 100644
--- a/drivers/gpu/drm/xe/Makefile
+++ b/drivers/gpu/drm/xe/Makefile
@@ -130,6 +130,7 @@ xe-y += xe_bb.o \
 	xe_tuning.o \
 	xe_uc.o \
 	xe_uc_fw.o \
+	xe_userptr.o \
 	xe_vm.o \
 	xe_vm_madvise.o \
 	xe_vram.o \
diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
index 029480c9cebd..9eac29cbdc48 100644
--- a/drivers/gpu/drm/xe/xe_pt.c
+++ b/drivers/gpu/drm/xe/xe_pt.c
@@ -24,6 +24,7 @@
 #include "xe_tlb_inval_job.h"
 #include "xe_trace.h"
 #include "xe_ttm_stolen_mgr.h"
+#include "xe_userptr.h"
 #include "xe_vm.h"
 
 struct xe_pt_dir {
diff --git a/drivers/gpu/drm/xe/xe_userptr.c b/drivers/gpu/drm/xe/xe_userptr.c
new file mode 100644
index 000000000000..17ecff17c993
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_userptr.c
@@ -0,0 +1,305 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2025 Intel Corporation
+ */
+
+#include "xe_userptr.h"
+
+#include <linux/mm.h>
+
+#include "xe_hmm.h"
+#include "xe_trace_bo.h"
+
+/**
+ * xe_vma_userptr_check_repin() - Advisory check for repin needed
+ * @uvma: The userptr vma
+ *
+ * Check if the userptr vma has been invalidated since last successful
+ * repin. The check is advisory only and can the function can be called
+ * without the vm->svm.gpusvm.notifier_lock held. There is no guarantee that the
+ * vma userptr will remain valid after a lockless check, so typically
+ * the call needs to be followed by a proper check under the notifier_lock.
+ *
+ * Return: 0 if userptr vma is valid, -EAGAIN otherwise; repin recommended.
+ */
+int xe_vma_userptr_check_repin(struct xe_userptr_vma *uvma)
+{
+	return mmu_interval_check_retry(&uvma->userptr.notifier,
+					uvma->userptr.notifier_seq) ?
+		-EAGAIN : 0;
+}
+
+/**
+ * __xe_vm_userptr_needs_repin() - Check whether the VM does have userptrs
+ * that need repinning.
+ * @vm: The VM.
+ *
+ * This function checks for whether the VM has userptrs that need repinning,
+ * and provides a release-type barrier on the userptr.notifier_lock after
+ * checking.
+ *
+ * Return: 0 if there are no userptrs needing repinning, -EAGAIN if there are.
+ */
+int __xe_vm_userptr_needs_repin(struct xe_vm *vm)
+{
+	lockdep_assert_held_read(&vm->userptr.notifier_lock);
+
+	return (list_empty(&vm->userptr.repin_list) &&
+		list_empty(&vm->userptr.invalidated)) ? 0 : -EAGAIN;
+}
+
+int xe_vma_userptr_pin_pages(struct xe_userptr_vma *uvma)
+{
+	struct xe_vma *vma = &uvma->vma;
+	struct xe_vm *vm = xe_vma_vm(vma);
+	struct xe_device *xe = vm->xe;
+
+	lockdep_assert_held(&vm->lock);
+	xe_assert(xe, xe_vma_is_userptr(vma));
+
+	return xe_hmm_userptr_populate_range(uvma, false);
+}
+
+static void __vma_userptr_invalidate(struct xe_vm *vm, struct xe_userptr_vma *uvma)
+{
+	struct xe_userptr *userptr = &uvma->userptr;
+	struct xe_vma *vma = &uvma->vma;
+	struct dma_resv_iter cursor;
+	struct dma_fence *fence;
+	long err;
+
+	/*
+	 * Tell exec and rebind worker they need to repin and rebind this
+	 * userptr.
+	 */
+	if (!xe_vm_in_fault_mode(vm) &&
+	    !(vma->gpuva.flags & XE_VMA_DESTROYED)) {
+		spin_lock(&vm->userptr.invalidated_lock);
+		list_move_tail(&userptr->invalidate_link,
+			       &vm->userptr.invalidated);
+		spin_unlock(&vm->userptr.invalidated_lock);
+	}
+
+	/*
+	 * Preempt fences turn into schedule disables, pipeline these.
+	 * Note that even in fault mode, we need to wait for binds and
+	 * unbinds to complete, and those are attached as BOOKMARK fences
+	 * to the vm.
+	 */
+	dma_resv_iter_begin(&cursor, xe_vm_resv(vm),
+			    DMA_RESV_USAGE_BOOKKEEP);
+	dma_resv_for_each_fence_unlocked(&cursor, fence)
+		dma_fence_enable_sw_signaling(fence);
+	dma_resv_iter_end(&cursor);
+
+	err = dma_resv_wait_timeout(xe_vm_resv(vm),
+				    DMA_RESV_USAGE_BOOKKEEP,
+				    false, MAX_SCHEDULE_TIMEOUT);
+	XE_WARN_ON(err <= 0);
+
+	if (xe_vm_in_fault_mode(vm) && userptr->initial_bind) {
+		err = xe_vm_invalidate_vma(vma);
+		XE_WARN_ON(err);
+	}
+
+	xe_hmm_userptr_unmap(uvma);
+}
+
+static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni,
+				   const struct mmu_notifier_range *range,
+				   unsigned long cur_seq)
+{
+	struct xe_userptr_vma *uvma = container_of(mni, typeof(*uvma), userptr.notifier);
+	struct xe_vma *vma = &uvma->vma;
+	struct xe_vm *vm = xe_vma_vm(vma);
+
+	xe_assert(vm->xe, xe_vma_is_userptr(vma));
+	trace_xe_vma_userptr_invalidate(vma);
+
+	if (!mmu_notifier_range_blockable(range))
+		return false;
+
+	vm_dbg(&xe_vma_vm(vma)->xe->drm,
+	       "NOTIFIER: addr=0x%016llx, range=0x%016llx",
+		xe_vma_start(vma), xe_vma_size(vma));
+
+	down_write(&vm->userptr.notifier_lock);
+	mmu_interval_set_seq(mni, cur_seq);
+
+	__vma_userptr_invalidate(vm, uvma);
+	up_write(&vm->userptr.notifier_lock);
+	trace_xe_vma_userptr_invalidate_complete(vma);
+
+	return true;
+}
+
+static const struct mmu_interval_notifier_ops vma_userptr_notifier_ops = {
+	.invalidate = vma_userptr_invalidate,
+};
+
+#if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
+/**
+ * xe_vma_userptr_force_invalidate() - force invalidate a userptr
+ * @uvma: The userptr vma to invalidate
+ *
+ * Perform a forced userptr invalidation for testing purposes.
+ */
+void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma)
+{
+	struct xe_vm *vm = xe_vma_vm(&uvma->vma);
+
+	/* Protect against concurrent userptr pinning */
+	lockdep_assert_held(&vm->lock);
+	/* Protect against concurrent notifiers */
+	lockdep_assert_held(&vm->userptr.notifier_lock);
+	/*
+	 * Protect against concurrent instances of this function and
+	 * the critical exec sections
+	 */
+	xe_vm_assert_held(vm);
+
+	if (!mmu_interval_read_retry(&uvma->userptr.notifier,
+				     uvma->userptr.notifier_seq))
+		uvma->userptr.notifier_seq -= 2;
+	__vma_userptr_invalidate(vm, uvma);
+}
+#endif
+
+int xe_vm_userptr_pin(struct xe_vm *vm)
+{
+	struct xe_userptr_vma *uvma, *next;
+	int err = 0;
+
+	xe_assert(vm->xe, !xe_vm_in_fault_mode(vm));
+	lockdep_assert_held_write(&vm->lock);
+
+	/* Collect invalidated userptrs */
+	spin_lock(&vm->userptr.invalidated_lock);
+	xe_assert(vm->xe, list_empty(&vm->userptr.repin_list));
+	list_for_each_entry_safe(uvma, next, &vm->userptr.invalidated,
+				 userptr.invalidate_link) {
+		list_del_init(&uvma->userptr.invalidate_link);
+		list_add_tail(&uvma->userptr.repin_link,
+			      &vm->userptr.repin_list);
+	}
+	spin_unlock(&vm->userptr.invalidated_lock);
+
+	/* Pin and move to bind list */
+	list_for_each_entry_safe(uvma, next, &vm->userptr.repin_list,
+				 userptr.repin_link) {
+		err = xe_vma_userptr_pin_pages(uvma);
+		if (err == -EFAULT) {
+			list_del_init(&uvma->userptr.repin_link);
+			/*
+			 * We might have already done the pin once already, but
+			 * then had to retry before the re-bind happened, due
+			 * some other condition in the caller, but in the
+			 * meantime the userptr got dinged by the notifier such
+			 * that we need to revalidate here, but this time we hit
+			 * the EFAULT. In such a case make sure we remove
+			 * ourselves from the rebind list to avoid going down in
+			 * flames.
+			 */
+			if (!list_empty(&uvma->vma.combined_links.rebind))
+				list_del_init(&uvma->vma.combined_links.rebind);
+
+			/* Wait for pending binds */
+			xe_vm_lock(vm, false);
+			dma_resv_wait_timeout(xe_vm_resv(vm),
+					      DMA_RESV_USAGE_BOOKKEEP,
+					      false, MAX_SCHEDULE_TIMEOUT);
+
+			down_read(&vm->userptr.notifier_lock);
+			err = xe_vm_invalidate_vma(&uvma->vma);
+			up_read(&vm->userptr.notifier_lock);
+			xe_vm_unlock(vm);
+			if (err)
+				break;
+		} else {
+			if (err)
+				break;
+
+			list_del_init(&uvma->userptr.repin_link);
+			list_move_tail(&uvma->vma.combined_links.rebind,
+				       &vm->rebind_list);
+		}
+	}
+
+	if (err) {
+		down_write(&vm->userptr.notifier_lock);
+		spin_lock(&vm->userptr.invalidated_lock);
+		list_for_each_entry_safe(uvma, next, &vm->userptr.repin_list,
+					 userptr.repin_link) {
+			list_del_init(&uvma->userptr.repin_link);
+			list_move_tail(&uvma->userptr.invalidate_link,
+				       &vm->userptr.invalidated);
+		}
+		spin_unlock(&vm->userptr.invalidated_lock);
+		up_write(&vm->userptr.notifier_lock);
+	}
+	return err;
+}
+
+/**
+ * xe_vm_userptr_check_repin() - Check whether the VM might have userptrs
+ * that need repinning.
+ * @vm: The VM.
+ *
+ * This function does an advisory check for whether the VM has userptrs that
+ * need repinning.
+ *
+ * Return: 0 if there are no indications of userptrs needing repinning,
+ * -EAGAIN if there are.
+ */
+int xe_vm_userptr_check_repin(struct xe_vm *vm)
+{
+	return (list_empty_careful(&vm->userptr.repin_list) &&
+		list_empty_careful(&vm->userptr.invalidated)) ? 0 : -EAGAIN;
+}
+
+int xe_userptr_setup(struct xe_userptr_vma *uvma, unsigned long start,
+		     unsigned long range)
+{
+	struct xe_userptr *userptr = &uvma->userptr;
+	int err;
+
+	INIT_LIST_HEAD(&userptr->invalidate_link);
+	INIT_LIST_HEAD(&userptr->repin_link);
+	mutex_init(&userptr->unmap_mutex);
+
+	err = mmu_interval_notifier_insert(&userptr->notifier, current->mm,
+					   start, range,
+					   &vma_userptr_notifier_ops);
+	if (err)
+		return err;
+
+	userptr->notifier_seq = LONG_MAX;
+
+	return 0;
+}
+
+void xe_userptr_remove(struct xe_userptr_vma *uvma)
+{
+	struct xe_userptr *userptr = &uvma->userptr;
+
+	if (userptr->sg)
+		xe_hmm_userptr_free_sg(uvma);
+
+	/*
+	 * Since userptr pages are not pinned, we can't remove
+	 * the notifier until we're sure the GPU is not accessing
+	 * them anymore
+	 */
+	mmu_interval_notifier_remove(&userptr->notifier);
+	mutex_destroy(&userptr->unmap_mutex);
+}
+
+void xe_userptr_destroy(struct xe_userptr_vma *uvma)
+{
+	struct xe_vm *vm = xe_vma_vm(&uvma->vma);
+
+	spin_lock(&vm->userptr.invalidated_lock);
+	xe_assert(vm->xe, list_empty(&uvma->userptr.repin_link));
+	list_del(&uvma->userptr.invalidate_link);
+	spin_unlock(&vm->userptr.invalidated_lock);
+}
diff --git a/drivers/gpu/drm/xe/xe_userptr.h b/drivers/gpu/drm/xe/xe_userptr.h
new file mode 100644
index 000000000000..dc49348ac42d
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_userptr.h
@@ -0,0 +1,97 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2025 Intel Corporation
+ */
+
+#ifndef _XE_USERPTR_H_
+#define _XE_USERPTR_H_
+
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/notifier.h>
+#include <linux/scatterlist.h>
+#include <linux/spinlock.h>
+
+struct xe_vm;
+struct xe_vma;
+struct xe_userptr_vma;
+
+/** struct xe_userptr_vm - User pointer VM level state */
+struct xe_userptr_vm {
+	/**
+	 * @userptr.repin_list: list of VMAs which are user pointers,
+	 * and needs repinning. Protected by @lock.
+	 */
+	struct list_head repin_list;
+	/**
+	 * @notifier_lock: protects notifier in write mode and
+	 * submission in read mode.o
+	 */
+	struct rw_semaphore notifier_lock;
+	/**
+	 * @userptr.invalidated_lock: Protects the
+	 * @userptr.invalidated list.
+	 */
+	spinlock_t invalidated_lock;
+	/**
+	 * @userptr.invalidated: List of invalidated userptrs, not yet
+	 * picked
+	 * up for revalidation. Protected from access with the
+	 * @invalidated_lock. Removing items from the list
+	 * additionally requires @lock in write mode, and adding
+	 * items to the list requires either the @userptr.notifier_lock in
+	 * write mode, OR @lock in write mode.
+	 */
+	struct list_head invalidated;
+};
+
+/** struct xe_userptr - User pointer */
+struct xe_userptr {
+	/** @invalidate_link: Link for the vm::userptr.invalidated list */
+	struct list_head invalidate_link;
+	/** @userptr: link into VM repin list if userptr. */
+	struct list_head repin_link;
+	/**
+	 * @notifier: MMU notifier for user pointer (invalidation call back)
+	 */
+	struct mmu_interval_notifier notifier;
+	/** @sgt: storage for a scatter gather table */
+	struct sg_table sgt;
+	/** @sg: allocated scatter gather table */
+	struct sg_table *sg;
+	/** @notifier_seq: notifier sequence number */
+	unsigned long notifier_seq;
+	/** @unmap_mutex: Mutex protecting dma-unmapping */
+	struct mutex unmap_mutex;
+	/**
+	 * @initial_bind: user pointer has been bound at least once.
+	 * write: vm->userptr.notifier_lock in read mode and vm->resv held.
+	 * read: vm->userptr.notifier_lock in write mode or vm->resv held.
+	 */
+	bool initial_bind;
+	/** @mapped: Whether the @sgt sg-table is dma-mapped. Protected by @unmap_mutex. */
+	bool mapped;
+#if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
+	u32 divisor;
+#endif
+};
+
+void xe_userptr_remove(struct xe_userptr_vma *uvma);
+int xe_userptr_setup(struct xe_userptr_vma *uvma, unsigned long start,
+		     unsigned long range);
+void xe_userptr_destroy(struct xe_userptr_vma *uvma);
+
+int xe_vm_userptr_pin(struct xe_vm *vm);
+int __xe_vm_userptr_needs_repin(struct xe_vm *vm);
+int xe_vm_userptr_check_repin(struct xe_vm *vm);
+int xe_vma_userptr_pin_pages(struct xe_userptr_vma *uvma);
+int xe_vma_userptr_check_repin(struct xe_userptr_vma *uvma);
+
+#if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
+void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma);
+#else
+static inline void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma)
+{
+}
+#endif
+#endif
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index 3ff3c67aa79d..57afe9819842 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -40,6 +40,7 @@
 #include "xe_tile.h"
 #include "xe_tlb_inval.h"
 #include "xe_trace_bo.h"
+#include "xe_userptr.h"
 #include "xe_wa.h"
 #include "xe_hmm.h"
 
@@ -48,37 +49,6 @@ static struct drm_gem_object *xe_vm_obj(struct xe_vm *vm)
 	return vm->gpuvm.r_obj;
 }
 
-/**
- * xe_vma_userptr_check_repin() - Advisory check for repin needed
- * @uvma: The userptr vma
- *
- * Check if the userptr vma has been invalidated since last successful
- * repin. The check is advisory only and can the function can be called
- * without the vm->userptr.notifier_lock held. There is no guarantee that the
- * vma userptr will remain valid after a lockless check, so typically
- * the call needs to be followed by a proper check under the notifier_lock.
- *
- * Return: 0 if userptr vma is valid, -EAGAIN otherwise; repin recommended.
- */
-int xe_vma_userptr_check_repin(struct xe_userptr_vma *uvma)
-{
-	return mmu_interval_check_retry(&uvma->userptr.notifier,
-					uvma->userptr.notifier_seq) ?
-		-EAGAIN : 0;
-}
-
-int xe_vma_userptr_pin_pages(struct xe_userptr_vma *uvma)
-{
-	struct xe_vma *vma = &uvma->vma;
-	struct xe_vm *vm = xe_vma_vm(vma);
-	struct xe_device *xe = vm->xe;
-
-	lockdep_assert_held(&vm->lock);
-	xe_assert(xe, xe_vma_is_userptr(vma));
-
-	return xe_hmm_userptr_populate_range(uvma, false);
-}
-
 static bool preempt_fences_waiting(struct xe_vm *vm)
 {
 	struct xe_exec_queue *q;
@@ -300,25 +270,6 @@ void xe_vm_remove_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q)
 	up_write(&vm->lock);
 }
 
-/**
- * __xe_vm_userptr_needs_repin() - Check whether the VM does have userptrs
- * that need repinning.
- * @vm: The VM.
- *
- * This function checks for whether the VM has userptrs that need repinning,
- * and provides a release-type barrier on the userptr.notifier_lock after
- * checking.
- *
- * Return: 0 if there are no userptrs needing repinning, -EAGAIN if there are.
- */
-int __xe_vm_userptr_needs_repin(struct xe_vm *vm)
-{
-	lockdep_assert_held_read(&vm->userptr.notifier_lock);
-
-	return (list_empty(&vm->userptr.repin_list) &&
-		list_empty(&vm->userptr.invalidated)) ? 0 : -EAGAIN;
-}
-
 #define XE_VM_REBIND_RETRY_TIMEOUT_MS 1000
 
 /**
@@ -584,203 +535,6 @@ static void preempt_rebind_work_func(struct work_struct *w)
 	trace_xe_vm_rebind_worker_exit(vm);
 }
 
-static void __vma_userptr_invalidate(struct xe_vm *vm, struct xe_userptr_vma *uvma)
-{
-	struct xe_userptr *userptr = &uvma->userptr;
-	struct xe_vma *vma = &uvma->vma;
-	struct dma_resv_iter cursor;
-	struct dma_fence *fence;
-	long err;
-
-	/*
-	 * Tell exec and rebind worker they need to repin and rebind this
-	 * userptr.
-	 */
-	if (!xe_vm_in_fault_mode(vm) &&
-	    !(vma->gpuva.flags & XE_VMA_DESTROYED)) {
-		spin_lock(&vm->userptr.invalidated_lock);
-		list_move_tail(&userptr->invalidate_link,
-			       &vm->userptr.invalidated);
-		spin_unlock(&vm->userptr.invalidated_lock);
-	}
-
-	/*
-	 * Preempt fences turn into schedule disables, pipeline these.
-	 * Note that even in fault mode, we need to wait for binds and
-	 * unbinds to complete, and those are attached as BOOKMARK fences
-	 * to the vm.
-	 */
-	dma_resv_iter_begin(&cursor, xe_vm_resv(vm),
-			    DMA_RESV_USAGE_BOOKKEEP);
-	dma_resv_for_each_fence_unlocked(&cursor, fence)
-		dma_fence_enable_sw_signaling(fence);
-	dma_resv_iter_end(&cursor);
-
-	err = dma_resv_wait_timeout(xe_vm_resv(vm),
-				    DMA_RESV_USAGE_BOOKKEEP,
-				    false, MAX_SCHEDULE_TIMEOUT);
-	XE_WARN_ON(err <= 0);
-
-	if (xe_vm_in_fault_mode(vm) && userptr->initial_bind) {
-		err = xe_vm_invalidate_vma(vma);
-		XE_WARN_ON(err);
-	}
-
-	xe_hmm_userptr_unmap(uvma);
-}
-
-static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni,
-				   const struct mmu_notifier_range *range,
-				   unsigned long cur_seq)
-{
-	struct xe_userptr_vma *uvma = container_of(mni, typeof(*uvma), userptr.notifier);
-	struct xe_vma *vma = &uvma->vma;
-	struct xe_vm *vm = xe_vma_vm(vma);
-
-	xe_assert(vm->xe, xe_vma_is_userptr(vma));
-	trace_xe_vma_userptr_invalidate(vma);
-
-	if (!mmu_notifier_range_blockable(range))
-		return false;
-
-	vm_dbg(&xe_vma_vm(vma)->xe->drm,
-	       "NOTIFIER: addr=0x%016llx, range=0x%016llx",
-		xe_vma_start(vma), xe_vma_size(vma));
-
-	down_write(&vm->userptr.notifier_lock);
-	mmu_interval_set_seq(mni, cur_seq);
-
-	__vma_userptr_invalidate(vm, uvma);
-	up_write(&vm->userptr.notifier_lock);
-	trace_xe_vma_userptr_invalidate_complete(vma);
-
-	return true;
-}
-
-static const struct mmu_interval_notifier_ops vma_userptr_notifier_ops = {
-	.invalidate = vma_userptr_invalidate,
-};
-
-#if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
-/**
- * xe_vma_userptr_force_invalidate() - force invalidate a userptr
- * @uvma: The userptr vma to invalidate
- *
- * Perform a forced userptr invalidation for testing purposes.
- */
-void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma)
-{
-	struct xe_vm *vm = xe_vma_vm(&uvma->vma);
-
-	/* Protect against concurrent userptr pinning */
-	lockdep_assert_held(&vm->lock);
-	/* Protect against concurrent notifiers */
-	lockdep_assert_held(&vm->userptr.notifier_lock);
-	/*
-	 * Protect against concurrent instances of this function and
-	 * the critical exec sections
-	 */
-	xe_vm_assert_held(vm);
-
-	if (!mmu_interval_read_retry(&uvma->userptr.notifier,
-				     uvma->userptr.notifier_seq))
-		uvma->userptr.notifier_seq -= 2;
-	__vma_userptr_invalidate(vm, uvma);
-}
-#endif
-
-int xe_vm_userptr_pin(struct xe_vm *vm)
-{
-	struct xe_userptr_vma *uvma, *next;
-	int err = 0;
-
-	xe_assert(vm->xe, !xe_vm_in_fault_mode(vm));
-	lockdep_assert_held_write(&vm->lock);
-
-	/* Collect invalidated userptrs */
-	spin_lock(&vm->userptr.invalidated_lock);
-	xe_assert(vm->xe, list_empty(&vm->userptr.repin_list));
-	list_for_each_entry_safe(uvma, next, &vm->userptr.invalidated,
-				 userptr.invalidate_link) {
-		list_del_init(&uvma->userptr.invalidate_link);
-		list_add_tail(&uvma->userptr.repin_link,
-			      &vm->userptr.repin_list);
-	}
-	spin_unlock(&vm->userptr.invalidated_lock);
-
-	/* Pin and move to bind list */
-	list_for_each_entry_safe(uvma, next, &vm->userptr.repin_list,
-				 userptr.repin_link) {
-		err = xe_vma_userptr_pin_pages(uvma);
-		if (err == -EFAULT) {
-			list_del_init(&uvma->userptr.repin_link);
-			/*
-			 * We might have already done the pin once already, but
-			 * then had to retry before the re-bind happened, due
-			 * some other condition in the caller, but in the
-			 * meantime the userptr got dinged by the notifier such
-			 * that we need to revalidate here, but this time we hit
-			 * the EFAULT. In such a case make sure we remove
-			 * ourselves from the rebind list to avoid going down in
-			 * flames.
-			 */
-			if (!list_empty(&uvma->vma.combined_links.rebind))
-				list_del_init(&uvma->vma.combined_links.rebind);
-
-			/* Wait for pending binds */
-			xe_vm_lock(vm, false);
-			dma_resv_wait_timeout(xe_vm_resv(vm),
-					      DMA_RESV_USAGE_BOOKKEEP,
-					      false, MAX_SCHEDULE_TIMEOUT);
-
-			down_read(&vm->userptr.notifier_lock);
-			err = xe_vm_invalidate_vma(&uvma->vma);
-			up_read(&vm->userptr.notifier_lock);
-			xe_vm_unlock(vm);
-			if (err)
-				break;
-		} else {
-			if (err)
-				break;
-
-			list_del_init(&uvma->userptr.repin_link);
-			list_move_tail(&uvma->vma.combined_links.rebind,
-				       &vm->rebind_list);
-		}
-	}
-
-	if (err) {
-		down_write(&vm->userptr.notifier_lock);
-		spin_lock(&vm->userptr.invalidated_lock);
-		list_for_each_entry_safe(uvma, next, &vm->userptr.repin_list,
-					 userptr.repin_link) {
-			list_del_init(&uvma->userptr.repin_link);
-			list_move_tail(&uvma->userptr.invalidate_link,
-				       &vm->userptr.invalidated);
-		}
-		spin_unlock(&vm->userptr.invalidated_lock);
-		up_write(&vm->userptr.notifier_lock);
-	}
-	return err;
-}
-
-/**
- * xe_vm_userptr_check_repin() - Check whether the VM might have userptrs
- * that need repinning.
- * @vm: The VM.
- *
- * This function does an advisory check for whether the VM has userptrs that
- * need repinning.
- *
- * Return: 0 if there are no indications of userptrs needing repinning,
- * -EAGAIN if there are.
- */
-int xe_vm_userptr_check_repin(struct xe_vm *vm)
-{
-	return (list_empty_careful(&vm->userptr.repin_list) &&
-		list_empty_careful(&vm->userptr.invalidated)) ? 0 : -EAGAIN;
-}
-
 static int xe_vma_ops_alloc(struct xe_vma_ops *vops, bool array_of_binds)
 {
 	int i;
@@ -1245,25 +999,17 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm,
 		drm_gpuvm_bo_put(vm_bo);
 	} else /* userptr or null */ {
 		if (!is_null && !is_cpu_addr_mirror) {
-			struct xe_userptr *userptr = &to_userptr_vma(vma)->userptr;
+			struct xe_userptr_vma *uvma = to_userptr_vma(vma);
 			u64 size = end - start + 1;
 			int err;
 
-			INIT_LIST_HEAD(&userptr->invalidate_link);
-			INIT_LIST_HEAD(&userptr->repin_link);
 			vma->gpuva.gem.offset = bo_offset_or_userptr;
-			mutex_init(&userptr->unmap_mutex);
 
-			err = mmu_interval_notifier_insert(&userptr->notifier,
-							   current->mm,
-							   xe_vma_userptr(vma), size,
-							   &vma_userptr_notifier_ops);
+			err = xe_userptr_setup(uvma, xe_vma_userptr(vma), size);
 			if (err) {
 				xe_vma_free(vma);
 				return ERR_PTR(err);
 			}
-
-			userptr->notifier_seq = LONG_MAX;
 		}
 
 		xe_vm_get(vm);
@@ -1283,18 +1029,8 @@ static void xe_vma_destroy_late(struct xe_vma *vma)
 
 	if (xe_vma_is_userptr(vma)) {
 		struct xe_userptr_vma *uvma = to_userptr_vma(vma);
-		struct xe_userptr *userptr = &uvma->userptr;
 
-		if (userptr->sg)
-			xe_hmm_userptr_free_sg(uvma);
-
-		/*
-		 * Since userptr pages are not pinned, we can't remove
-		 * the notifier until we're sure the GPU is not accessing
-		 * them anymore
-		 */
-		mmu_interval_notifier_remove(&userptr->notifier);
-		mutex_destroy(&userptr->unmap_mutex);
+		xe_userptr_remove(uvma);
 		xe_vm_put(vm);
 	} else if (xe_vma_is_null(vma) || xe_vma_is_cpu_addr_mirror(vma)) {
 		xe_vm_put(vm);
@@ -1331,11 +1067,7 @@ static void xe_vma_destroy(struct xe_vma *vma, struct dma_fence *fence)
 
 	if (xe_vma_is_userptr(vma)) {
 		xe_assert(vm->xe, vma->gpuva.flags & XE_VMA_DESTROYED);
-
-		spin_lock(&vm->userptr.invalidated_lock);
-		xe_assert(vm->xe, list_empty(&to_userptr_vma(vma)->userptr.repin_link));
-		list_del(&to_userptr_vma(vma)->userptr.invalidate_link);
-		spin_unlock(&vm->userptr.invalidated_lock);
+		xe_userptr_destroy(to_userptr_vma(vma));
 	} else if (!xe_vma_is_null(vma) && !xe_vma_is_cpu_addr_mirror(vma)) {
 		xe_bo_assert_held(xe_vma_bo(vma));
 
diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h
index b3e5bec0fa58..bec9bf89d937 100644
--- a/drivers/gpu/drm/xe/xe_vm.h
+++ b/drivers/gpu/drm/xe/xe_vm.h
@@ -220,12 +220,6 @@ static inline bool xe_vm_in_preempt_fence_mode(struct xe_vm *vm)
 int xe_vm_add_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q);
 void xe_vm_remove_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q);
 
-int xe_vm_userptr_pin(struct xe_vm *vm);
-
-int __xe_vm_userptr_needs_repin(struct xe_vm *vm);
-
-int xe_vm_userptr_check_repin(struct xe_vm *vm);
-
 int xe_vm_rebind(struct xe_vm *vm, bool rebind_worker);
 struct dma_fence *xe_vma_rebind(struct xe_vm *vm, struct xe_vma *vma,
 				u8 tile_mask);
@@ -266,10 +260,6 @@ static inline void xe_vm_reactivate_rebind(struct xe_vm *vm)
 	}
 }
 
-int xe_vma_userptr_pin_pages(struct xe_userptr_vma *uvma);
-
-int xe_vma_userptr_check_repin(struct xe_userptr_vma *uvma);
-
 bool xe_vm_validate_should_retry(struct drm_exec *exec, int err, ktime_t *end);
 
 int xe_vm_lock_vma(struct drm_exec *exec, struct xe_vma *vma);
@@ -391,11 +381,4 @@ static inline bool xe_vm_is_validating(struct xe_vm *vm)
 #define xe_vm_has_valid_gpu_mapping(tile, tile_present, tile_invalidated)	\
 	((READ_ONCE(tile_present) & ~READ_ONCE(tile_invalidated)) & BIT((tile)->id))
 
-#if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
-void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma);
-#else
-static inline void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma)
-{
-}
-#endif
 #endif
diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
index b5108d010786..aee050c295db 100644
--- a/drivers/gpu/drm/xe/xe_vm_types.h
+++ b/drivers/gpu/drm/xe/xe_vm_types.h
@@ -17,6 +17,7 @@
 #include "xe_device_types.h"
 #include "xe_pt_types.h"
 #include "xe_range_fence.h"
+#include "xe_userptr.h"
 
 struct xe_bo;
 struct xe_svm_range;
@@ -46,37 +47,6 @@ struct xe_vm_pgtable_update_op;
 #define XE_VMA_DUMPABLE		(DRM_GPUVA_USERBITS << 8)
 #define XE_VMA_SYSTEM_ALLOCATOR	(DRM_GPUVA_USERBITS << 9)
 
-/** struct xe_userptr - User pointer */
-struct xe_userptr {
-	/** @invalidate_link: Link for the vm::userptr.invalidated list */
-	struct list_head invalidate_link;
-	/** @userptr: link into VM repin list if userptr. */
-	struct list_head repin_link;
-	/**
-	 * @notifier: MMU notifier for user pointer (invalidation call back)
-	 */
-	struct mmu_interval_notifier notifier;
-	/** @sgt: storage for a scatter gather table */
-	struct sg_table sgt;
-	/** @sg: allocated scatter gather table */
-	struct sg_table *sg;
-	/** @notifier_seq: notifier sequence number */
-	unsigned long notifier_seq;
-	/** @unmap_mutex: Mutex protecting dma-unmapping */
-	struct mutex unmap_mutex;
-	/**
-	 * @initial_bind: user pointer has been bound at least once.
-	 * write: vm->userptr.notifier_lock in read mode and vm->resv held.
-	 * read: vm->userptr.notifier_lock in write mode or vm->resv held.
-	 */
-	bool initial_bind;
-	/** @mapped: Whether the @sgt sg-table is dma-mapped. Protected by @unmap_mutex. */
-	bool mapped;
-#if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
-	u32 divisor;
-#endif
-};
-
 /**
  * struct xe_vma_mem_attr - memory attributes associated with vma
  */
@@ -289,33 +259,7 @@ struct xe_vm {
 	const struct xe_pt_ops *pt_ops;
 
 	/** @userptr: user pointer state */
-	struct {
-		/**
-		 * @userptr.repin_list: list of VMAs which are user pointers,
-		 * and needs repinning. Protected by @lock.
-		 */
-		struct list_head repin_list;
-		/**
-		 * @notifier_lock: protects notifier in write mode and
-		 * submission in read mode.
-		 */
-		struct rw_semaphore notifier_lock;
-		/**
-		 * @userptr.invalidated_lock: Protects the
-		 * @userptr.invalidated list.
-		 */
-		spinlock_t invalidated_lock;
-		/**
-		 * @userptr.invalidated: List of invalidated userptrs, not yet
-		 * picked
-		 * up for revalidation. Protected from access with the
-		 * @invalidated_lock. Removing items from the list
-		 * additionally requires @lock in write mode, and adding
-		 * items to the list requires either the @userptr.notifier_lock in
-		 * write mode, OR @lock in write mode.
-		 */
-		struct list_head invalidated;
-	} userptr;
+	struct xe_userptr_vm userptr;
 
 	/** @preempt: preempt state */
 	struct {
-- 
2.51.0

[PATCH v6 7/8] drm/xe/userptr: replace xe_hmm with gpusvm

[Matthew Auld]

Goal here is cut over to gpusvm and remove xe_hmm, relying instead on
common code. The core facilities we need are get_pages(), unmap_pages()
and free_pages() for a given useptr range, plus a vm level notifier
lock, which is now provided by gpusvm.

v2:
  - Reuse the same SVM vm struct we use for full SVM, that way we can
    use the same lock (Matt B & Himal)
v3:
  - Re-use svm_init/fini for userptr.
v4:
  - Allow building xe without userptr if we are missing DRM_GPUSVM
    config. (Matt B)
  - Always make .read_only match xe_vma_read_only() for the ctx. (Dafna)
v5:
  - Fix missing conversion with CONFIG_DRM_XE_USERPTR_INVAL_INJECT
v6:
  - Convert the new user in xe_vm_madise.

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Dafna Hirschfeld <dafna.hirschfeld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
---
 drivers/gpu/drm/xe/Kconfig         |   2 +-
 drivers/gpu/drm/xe/Makefile        |   3 +-
 drivers/gpu/drm/xe/xe_exec.c       |   5 +-
 drivers/gpu/drm/xe/xe_hmm.c        | 325 -----------------------------
 drivers/gpu/drm/xe/xe_hmm.h        |  18 --
 drivers/gpu/drm/xe/xe_pt.c         |  24 +--
 drivers/gpu/drm/xe/xe_svm.c        |  32 +--
 drivers/gpu/drm/xe/xe_svm.h        |  56 +++--
 drivers/gpu/drm/xe/xe_userptr.c    |  54 +++--
 drivers/gpu/drm/xe/xe_userptr.h    |  46 ++--
 drivers/gpu/drm/xe/xe_vm.c         |  48 ++---
 drivers/gpu/drm/xe/xe_vm_madvise.c |  24 +--
 drivers/gpu/drm/xe/xe_vm_types.h   |   8 +-
 13 files changed, 171 insertions(+), 474 deletions(-)
 delete mode 100644 drivers/gpu/drm/xe/xe_hmm.c
 delete mode 100644 drivers/gpu/drm/xe/xe_hmm.h

diff --git a/drivers/gpu/drm/xe/Kconfig b/drivers/gpu/drm/xe/Kconfig
index 714d5702dfd7..7219f6b884b6 100644
--- a/drivers/gpu/drm/xe/Kconfig
+++ b/drivers/gpu/drm/xe/Kconfig
@@ -40,12 +40,12 @@ config DRM_XE
 	select DRM_TTM
 	select DRM_TTM_HELPER
 	select DRM_EXEC
+	select DRM_GPUSVM if !UML && DEVICE_PRIVATE
 	select DRM_GPUVM
 	select DRM_SCHED
 	select MMU_NOTIFIER
 	select WANT_DEV_COREDUMP
 	select AUXILIARY_BUS
-	select HMM_MIRROR
 	select REGMAP if I2C
 	help
 	  Driver for Intel Xe2 series GPUs and later. Experimental support
diff --git a/drivers/gpu/drm/xe/Makefile b/drivers/gpu/drm/xe/Makefile
index 85ead33f69ab..f57e3d4dc6fe 100644
--- a/drivers/gpu/drm/xe/Makefile
+++ b/drivers/gpu/drm/xe/Makefile
@@ -130,7 +130,6 @@ xe-y += xe_bb.o \
 	xe_tuning.o \
 	xe_uc.o \
 	xe_uc_fw.o \
-	xe_userptr.o \
 	xe_vm.o \
 	xe_vm_madvise.o \
 	xe_vram.o \
@@ -141,8 +140,8 @@ xe-y += xe_bb.o \
 	xe_wopcm.o
 
 xe-$(CONFIG_I2C)	+= xe_i2c.o
-xe-$(CONFIG_HMM_MIRROR) += xe_hmm.o
 xe-$(CONFIG_DRM_XE_GPUSVM) += xe_svm.o
+xe-$(CONFIG_DRM_GPUSVM) += xe_userptr.o
 
 # graphics hardware monitoring (HWMON) support
 xe-$(CONFIG_HWMON) += xe_hwmon.o
diff --git a/drivers/gpu/drm/xe/xe_exec.c b/drivers/gpu/drm/xe/xe_exec.c
index 44364c042ad7..25a59b6934f6 100644
--- a/drivers/gpu/drm/xe/xe_exec.c
+++ b/drivers/gpu/drm/xe/xe_exec.c
@@ -19,6 +19,7 @@
 #include "xe_ring_ops_types.h"
 #include "xe_sched_job.h"
 #include "xe_sync.h"
+#include "xe_svm.h"
 #include "xe_vm.h"
 
 /**
@@ -294,7 +295,7 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 		if (err)
 			goto err_put_job;
 
-		err = down_read_interruptible(&vm->userptr.notifier_lock);
+		err = xe_svm_notifier_lock_interruptible(vm);
 		if (err)
 			goto err_put_job;
 
@@ -336,7 +337,7 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 
 err_repin:
 	if (!xe_vm_in_lr_mode(vm))
-		up_read(&vm->userptr.notifier_lock);
+		xe_svm_notifier_unlock(vm);
 err_put_job:
 	if (err)
 		xe_sched_job_put(job);
diff --git a/drivers/gpu/drm/xe/xe_hmm.c b/drivers/gpu/drm/xe/xe_hmm.c
deleted file mode 100644
index 57b71956ddf4..000000000000
--- a/drivers/gpu/drm/xe/xe_hmm.c
+++ /dev/null
@@ -1,325 +0,0 @@
-// SPDX-License-Identifier: MIT
-/*
- * Copyright © 2024 Intel Corporation
- */
-
-#include <linux/scatterlist.h>
-#include <linux/mmu_notifier.h>
-#include <linux/dma-mapping.h>
-#include <linux/memremap.h>
-#include <linux/swap.h>
-#include <linux/hmm.h>
-#include <linux/mm.h>
-#include "xe_hmm.h"
-#include "xe_vm.h"
-#include "xe_bo.h"
-
-static u64 xe_npages_in_range(unsigned long start, unsigned long end)
-{
-	return (end - start) >> PAGE_SHIFT;
-}
-
-static int xe_alloc_sg(struct xe_device *xe, struct sg_table *st,
-		       struct hmm_range *range, struct rw_semaphore *notifier_sem)
-{
-	unsigned long i, npages, hmm_pfn;
-	unsigned long num_chunks = 0;
-	int ret;
-
-	/* HMM docs says this is needed. */
-	ret = down_read_interruptible(notifier_sem);
-	if (ret)
-		return ret;
-
-	if (mmu_interval_read_retry(range->notifier, range->notifier_seq)) {
-		up_read(notifier_sem);
-		return -EAGAIN;
-	}
-
-	npages = xe_npages_in_range(range->start, range->end);
-	for (i = 0; i < npages;) {
-		unsigned long len;
-
-		hmm_pfn = range->hmm_pfns[i];
-		xe_assert(xe, hmm_pfn & HMM_PFN_VALID);
-
-		len = 1UL << hmm_pfn_to_map_order(hmm_pfn);
-
-		/* If order > 0 the page may extend beyond range->start */
-		len -= (hmm_pfn & ~HMM_PFN_FLAGS) & (len - 1);
-		i += len;
-		num_chunks++;
-	}
-	up_read(notifier_sem);
-
-	return sg_alloc_table(st, num_chunks, GFP_KERNEL);
-}
-
-/**
- * xe_build_sg() - build a scatter gather table for all the physical pages/pfn
- * in a hmm_range. dma-map pages if necessary. dma-address is save in sg table
- * and will be used to program GPU page table later.
- * @xe: the xe device who will access the dma-address in sg table
- * @range: the hmm range that we build the sg table from. range->hmm_pfns[]
- * has the pfn numbers of pages that back up this hmm address range.
- * @st: pointer to the sg table.
- * @notifier_sem: The xe notifier lock.
- * @write: whether we write to this range. This decides dma map direction
- * for system pages. If write we map it bi-diretional; otherwise
- * DMA_TO_DEVICE
- *
- * All the contiguous pfns will be collapsed into one entry in
- * the scatter gather table. This is for the purpose of efficiently
- * programming GPU page table.
- *
- * The dma_address in the sg table will later be used by GPU to
- * access memory. So if the memory is system memory, we need to
- * do a dma-mapping so it can be accessed by GPU/DMA.
- *
- * FIXME: This function currently only support pages in system
- * memory. If the memory is GPU local memory (of the GPU who
- * is going to access memory), we need gpu dpa (device physical
- * address), and there is no need of dma-mapping. This is TBD.
- *
- * FIXME: dma-mapping for peer gpu device to access remote gpu's
- * memory. Add this when you support p2p
- *
- * This function allocates the storage of the sg table. It is
- * caller's responsibility to free it calling sg_free_table.
- *
- * Returns 0 if successful; -ENOMEM if fails to allocate memory
- */
-static int xe_build_sg(struct xe_device *xe, struct hmm_range *range,
-		       struct sg_table *st,
-		       struct rw_semaphore *notifier_sem,
-		       bool write)
-{
-	unsigned long npages = xe_npages_in_range(range->start, range->end);
-	struct device *dev = xe->drm.dev;
-	struct scatterlist *sgl;
-	struct page *page;
-	unsigned long i, j;
-
-	lockdep_assert_held(notifier_sem);
-
-	i = 0;
-	for_each_sg(st->sgl, sgl, st->nents, j) {
-		unsigned long hmm_pfn, size;
-
-		hmm_pfn = range->hmm_pfns[i];
-		page = hmm_pfn_to_page(hmm_pfn);
-		xe_assert(xe, !is_device_private_page(page));
-
-		size = 1UL << hmm_pfn_to_map_order(hmm_pfn);
-		size -= page_to_pfn(page) & (size - 1);
-		i += size;
-
-		if (unlikely(j == st->nents - 1)) {
-			xe_assert(xe, i >= npages);
-			if (i > npages)
-				size -= (i - npages);
-
-			sg_mark_end(sgl);
-		} else {
-			xe_assert(xe, i < npages);
-		}
-
-		sg_set_page(sgl, page, size << PAGE_SHIFT, 0);
-	}
-
-	return dma_map_sgtable(dev, st, write ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE,
-			       DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_NO_KERNEL_MAPPING);
-}
-
-static void xe_hmm_userptr_set_mapped(struct xe_userptr_vma *uvma)
-{
-	struct xe_userptr *userptr = &uvma->userptr;
-	struct xe_vm *vm = xe_vma_vm(&uvma->vma);
-
-	lockdep_assert_held_write(&vm->lock);
-	lockdep_assert_held(&vm->userptr.notifier_lock);
-
-	mutex_lock(&userptr->unmap_mutex);
-	xe_assert(vm->xe, !userptr->mapped);
-	userptr->mapped = true;
-	mutex_unlock(&userptr->unmap_mutex);
-}
-
-void xe_hmm_userptr_unmap(struct xe_userptr_vma *uvma)
-{
-	struct xe_userptr *userptr = &uvma->userptr;
-	struct xe_vma *vma = &uvma->vma;
-	bool write = !xe_vma_read_only(vma);
-	struct xe_vm *vm = xe_vma_vm(vma);
-	struct xe_device *xe = vm->xe;
-
-	if (!lockdep_is_held_type(&vm->userptr.notifier_lock, 0) &&
-	    !lockdep_is_held_type(&vm->lock, 0) &&
-	    !(vma->gpuva.flags & XE_VMA_DESTROYED)) {
-		/* Don't unmap in exec critical section. */
-		xe_vm_assert_held(vm);
-		/* Don't unmap while mapping the sg. */
-		lockdep_assert_held(&vm->lock);
-	}
-
-	mutex_lock(&userptr->unmap_mutex);
-	if (userptr->sg && userptr->mapped)
-		dma_unmap_sgtable(xe->drm.dev, userptr->sg,
-				  write ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE, 0);
-	userptr->mapped = false;
-	mutex_unlock(&userptr->unmap_mutex);
-}
-
-/**
- * xe_hmm_userptr_free_sg() - Free the scatter gather table of userptr
- * @uvma: the userptr vma which hold the scatter gather table
- *
- * With function xe_userptr_populate_range, we allocate storage of
- * the userptr sg table. This is a helper function to free this
- * sg table, and dma unmap the address in the table.
- */
-void xe_hmm_userptr_free_sg(struct xe_userptr_vma *uvma)
-{
-	struct xe_userptr *userptr = &uvma->userptr;
-
-	xe_assert(xe_vma_vm(&uvma->vma)->xe, userptr->sg);
-	xe_hmm_userptr_unmap(uvma);
-	sg_free_table(userptr->sg);
-	userptr->sg = NULL;
-}
-
-/**
- * xe_hmm_userptr_populate_range() - Populate physical pages of a virtual
- * address range
- *
- * @uvma: userptr vma which has information of the range to populate.
- * @is_mm_mmap_locked: True if mmap_read_lock is already acquired by caller.
- *
- * This function populate the physical pages of a virtual
- * address range. The populated physical pages is saved in
- * userptr's sg table. It is similar to get_user_pages but call
- * hmm_range_fault.
- *
- * This function also read mmu notifier sequence # (
- * mmu_interval_read_begin), for the purpose of later
- * comparison (through mmu_interval_read_retry).
- *
- * This must be called with mmap read or write lock held.
- *
- * This function allocates the storage of the userptr sg table.
- * It is caller's responsibility to free it calling sg_free_table.
- *
- * returns: 0 for success; negative error no on failure
- */
-int xe_hmm_userptr_populate_range(struct xe_userptr_vma *uvma,
-				  bool is_mm_mmap_locked)
-{
-	unsigned long timeout =
-		jiffies + msecs_to_jiffies(HMM_RANGE_DEFAULT_TIMEOUT);
-	unsigned long *pfns;
-	struct xe_userptr *userptr;
-	struct xe_vma *vma = &uvma->vma;
-	u64 userptr_start = xe_vma_userptr(vma);
-	u64 userptr_end = userptr_start + xe_vma_size(vma);
-	struct xe_vm *vm = xe_vma_vm(vma);
-	struct hmm_range hmm_range = {
-		.pfn_flags_mask = 0, /* ignore pfns */
-		.default_flags = HMM_PFN_REQ_FAULT,
-		.start = userptr_start,
-		.end = userptr_end,
-		.notifier = &uvma->userptr.notifier,
-		.dev_private_owner = vm->xe,
-	};
-	bool write = !xe_vma_read_only(vma);
-	unsigned long notifier_seq;
-	u64 npages;
-	int ret;
-
-	userptr = &uvma->userptr;
-
-	if (is_mm_mmap_locked)
-		mmap_assert_locked(userptr->notifier.mm);
-
-	if (vma->gpuva.flags & XE_VMA_DESTROYED)
-		return 0;
-
-	notifier_seq = mmu_interval_read_begin(&userptr->notifier);
-	if (notifier_seq == userptr->notifier_seq)
-		return 0;
-
-	if (userptr->sg)
-		xe_hmm_userptr_free_sg(uvma);
-
-	npages = xe_npages_in_range(userptr_start, userptr_end);
-	pfns = kvmalloc_array(npages, sizeof(*pfns), GFP_KERNEL);
-	if (unlikely(!pfns))
-		return -ENOMEM;
-
-	if (write)
-		hmm_range.default_flags |= HMM_PFN_REQ_WRITE;
-
-	if (!mmget_not_zero(userptr->notifier.mm)) {
-		ret = -EFAULT;
-		goto free_pfns;
-	}
-
-	hmm_range.hmm_pfns = pfns;
-
-	while (true) {
-		hmm_range.notifier_seq = mmu_interval_read_begin(&userptr->notifier);
-
-		if (!is_mm_mmap_locked)
-			mmap_read_lock(userptr->notifier.mm);
-
-		ret = hmm_range_fault(&hmm_range);
-
-		if (!is_mm_mmap_locked)
-			mmap_read_unlock(userptr->notifier.mm);
-
-		if (ret == -EBUSY) {
-			if (time_after(jiffies, timeout))
-				break;
-
-			continue;
-		}
-		break;
-	}
-
-	mmput(userptr->notifier.mm);
-
-	if (ret)
-		goto free_pfns;
-
-	ret = xe_alloc_sg(vm->xe, &userptr->sgt, &hmm_range, &vm->userptr.notifier_lock);
-	if (ret)
-		goto free_pfns;
-
-	ret = down_read_interruptible(&vm->userptr.notifier_lock);
-	if (ret)
-		goto free_st;
-
-	if (mmu_interval_read_retry(hmm_range.notifier, hmm_range.notifier_seq)) {
-		ret = -EAGAIN;
-		goto out_unlock;
-	}
-
-	ret = xe_build_sg(vm->xe, &hmm_range, &userptr->sgt,
-			  &vm->userptr.notifier_lock, write);
-	if (ret)
-		goto out_unlock;
-
-	userptr->sg = &userptr->sgt;
-	xe_hmm_userptr_set_mapped(uvma);
-	userptr->notifier_seq = hmm_range.notifier_seq;
-	up_read(&vm->userptr.notifier_lock);
-	kvfree(pfns);
-	return 0;
-
-out_unlock:
-	up_read(&vm->userptr.notifier_lock);
-free_st:
-	sg_free_table(&userptr->sgt);
-free_pfns:
-	kvfree(pfns);
-	return ret;
-}
diff --git a/drivers/gpu/drm/xe/xe_hmm.h b/drivers/gpu/drm/xe/xe_hmm.h
deleted file mode 100644
index 0ea98d8e7bbc..000000000000
--- a/drivers/gpu/drm/xe/xe_hmm.h
+++ /dev/null
@@ -1,18 +0,0 @@
-/* SPDX-License-Identifier: MIT
- *
- * Copyright © 2024 Intel Corporation
- */
-
-#ifndef _XE_HMM_H_
-#define _XE_HMM_H_
-
-#include <linux/types.h>
-
-struct xe_userptr_vma;
-
-int xe_hmm_userptr_populate_range(struct xe_userptr_vma *uvma, bool is_mm_mmap_locked);
-
-void xe_hmm_userptr_free_sg(struct xe_userptr_vma *uvma);
-
-void xe_hmm_userptr_unmap(struct xe_userptr_vma *uvma);
-#endif
diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
index 9eac29cbdc48..382b25798bc3 100644
--- a/drivers/gpu/drm/xe/xe_pt.c
+++ b/drivers/gpu/drm/xe/xe_pt.c
@@ -761,8 +761,8 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
 
 	if (!xe_vma_is_null(vma) && !range) {
 		if (xe_vma_is_userptr(vma))
-			xe_res_first_sg(to_userptr_vma(vma)->userptr.sg, 0,
-					xe_vma_size(vma), &curs);
+			xe_res_first_dma(to_userptr_vma(vma)->userptr.pages.dma_addr, 0,
+					 xe_vma_size(vma), &curs);
 		else if (xe_bo_is_vram(bo) || xe_bo_is_stolen(bo))
 			xe_res_first(bo->ttm.resource, xe_vma_bo_offset(vma),
 				     xe_vma_size(vma), &curs);
@@ -915,7 +915,7 @@ bool xe_pt_zap_ptes(struct xe_tile *tile, struct xe_vma *vma)
 	if (xe_vma_bo(vma))
 		xe_bo_assert_held(xe_vma_bo(vma));
 	else if (xe_vma_is_userptr(vma))
-		lockdep_assert_held(&xe_vma_vm(vma)->userptr.notifier_lock);
+		lockdep_assert_held(&xe_vma_vm(vma)->svm.gpusvm.notifier_lock);
 
 	if (!(pt_mask & BIT(tile->id)))
 		return false;
@@ -1050,7 +1050,7 @@ static void xe_pt_commit_locks_assert(struct xe_vma *vma)
 	xe_pt_commit_prepare_locks_assert(vma);
 
 	if (xe_vma_is_userptr(vma))
-		lockdep_assert_held_read(&vm->userptr.notifier_lock);
+		xe_svm_assert_held_read(vm);
 }
 
 static void xe_pt_commit(struct xe_vma *vma,
@@ -1407,7 +1407,7 @@ static int vma_check_userptr(struct xe_vm *vm, struct xe_vma *vma,
 	struct xe_userptr_vma *uvma;
 	unsigned long notifier_seq;
 
-	lockdep_assert_held_read(&vm->userptr.notifier_lock);
+	xe_svm_assert_held_read(vm);
 
 	if (!xe_vma_is_userptr(vma))
 		return 0;
@@ -1416,7 +1416,7 @@ static int vma_check_userptr(struct xe_vm *vm, struct xe_vma *vma,
 	if (xe_pt_userptr_inject_eagain(uvma))
 		xe_vma_userptr_force_invalidate(uvma);
 
-	notifier_seq = uvma->userptr.notifier_seq;
+	notifier_seq = uvma->userptr.pages.notifier_seq;
 
 	if (!mmu_interval_read_retry(&uvma->userptr.notifier,
 				     notifier_seq))
@@ -1437,7 +1437,7 @@ static int op_check_userptr(struct xe_vm *vm, struct xe_vma_op *op,
 {
 	int err = 0;
 
-	lockdep_assert_held_read(&vm->userptr.notifier_lock);
+	xe_svm_assert_held_read(vm);
 
 	switch (op->base.op) {
 	case DRM_GPUVA_OP_MAP:
@@ -1478,12 +1478,12 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)
 	if (err)
 		return err;
 
-	down_read(&vm->userptr.notifier_lock);
+	down_read(&vm->svm.gpusvm.notifier_lock);
 
 	list_for_each_entry(op, &vops->list, link) {
 		err = op_check_userptr(vm, op, pt_update_ops);
 		if (err) {
-			up_read(&vm->userptr.notifier_lock);
+			up_read(&vm->svm.gpusvm.notifier_lock);
 			break;
 		}
 	}
@@ -2200,7 +2200,7 @@ static void bind_op_commit(struct xe_vm *vm, struct xe_tile *tile,
 			   vma->tile_invalidated & ~BIT(tile->id));
 	vma->tile_staged &= ~BIT(tile->id);
 	if (xe_vma_is_userptr(vma)) {
-		lockdep_assert_held_read(&vm->userptr.notifier_lock);
+		xe_svm_assert_held_read(vm);
 		to_userptr_vma(vma)->userptr.initial_bind = true;
 	}
 
@@ -2236,7 +2236,7 @@ static void unbind_op_commit(struct xe_vm *vm, struct xe_tile *tile,
 	if (!vma->tile_present) {
 		list_del_init(&vma->combined_links.rebind);
 		if (xe_vma_is_userptr(vma)) {
-			lockdep_assert_held_read(&vm->userptr.notifier_lock);
+			xe_svm_assert_held_read(vm);
 
 			spin_lock(&vm->userptr.invalidated_lock);
 			list_del_init(&to_userptr_vma(vma)->userptr.invalidate_link);
@@ -2535,7 +2535,7 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops)
 	if (pt_update_ops->needs_svm_lock)
 		xe_svm_notifier_unlock(vm);
 	if (pt_update_ops->needs_userptr_lock)
-		up_read(&vm->userptr.notifier_lock);
+		up_read(&vm->svm.gpusvm.notifier_lock);
 
 	xe_tlb_inval_job_put(mjob);
 	xe_tlb_inval_job_put(ijob);
diff --git a/drivers/gpu/drm/xe/xe_svm.c b/drivers/gpu/drm/xe/xe_svm.c
index d6ef0cd8a02c..e8e7d6365dc4 100644
--- a/drivers/gpu/drm/xe/xe_svm.c
+++ b/drivers/gpu/drm/xe/xe_svm.c
@@ -630,22 +630,26 @@ int xe_svm_init(struct xe_vm *vm)
 {
 	int err;
 
-	spin_lock_init(&vm->svm.garbage_collector.lock);
-	INIT_LIST_HEAD(&vm->svm.garbage_collector.range_list);
-	INIT_WORK(&vm->svm.garbage_collector.work,
-		  xe_svm_garbage_collector_work_func);
+	if (vm->flags & XE_VM_FLAG_FAULT_MODE) {
+		spin_lock_init(&vm->svm.garbage_collector.lock);
+		INIT_LIST_HEAD(&vm->svm.garbage_collector.range_list);
+		INIT_WORK(&vm->svm.garbage_collector.work,
+			  xe_svm_garbage_collector_work_func);
 
-	err = drm_gpusvm_init(&vm->svm.gpusvm, "Xe SVM", &vm->xe->drm,
-			      current->mm, xe_svm_devm_owner(vm->xe), 0,
-			      vm->size, xe_modparam.svm_notifier_size * SZ_1M,
-			      &gpusvm_ops, fault_chunk_sizes,
-			      ARRAY_SIZE(fault_chunk_sizes));
-	if (err)
-		return err;
+		err = drm_gpusvm_init(&vm->svm.gpusvm, "Xe SVM", &vm->xe->drm,
+				      current->mm, xe_svm_devm_owner(vm->xe), 0,
+				      vm->size,
+				      xe_modparam.svm_notifier_size * SZ_1M,
+				      &gpusvm_ops, fault_chunk_sizes,
+				      ARRAY_SIZE(fault_chunk_sizes));
+		drm_gpusvm_driver_set_lock(&vm->svm.gpusvm, &vm->lock);
+	} else {
+		err = drm_gpusvm_init(&vm->svm.gpusvm, "Xe SVM (simple)",
+				      &vm->xe->drm, NULL, NULL, 0, 0, 0, NULL,
+				      NULL, 0);
+	}
 
-	drm_gpusvm_driver_set_lock(&vm->svm.gpusvm, &vm->lock);
-
-	return 0;
+	return err;
 }
 
 /**
diff --git a/drivers/gpu/drm/xe/xe_svm.h b/drivers/gpu/drm/xe/xe_svm.h
index f0e0e009c7fc..9d791b0bc6c2 100644
--- a/drivers/gpu/drm/xe/xe_svm.h
+++ b/drivers/gpu/drm/xe/xe_svm.h
@@ -155,15 +155,6 @@ static inline unsigned long xe_svm_range_size(struct xe_svm_range *range)
 	return drm_gpusvm_range_size(&range->base);
 }
 
-#define xe_svm_assert_in_notifier(vm__) \
-	lockdep_assert_held_write(&(vm__)->svm.gpusvm.notifier_lock)
-
-#define xe_svm_notifier_lock(vm__)	\
-	drm_gpusvm_notifier_lock(&(vm__)->svm.gpusvm)
-
-#define xe_svm_notifier_unlock(vm__)	\
-	drm_gpusvm_notifier_unlock(&(vm__)->svm.gpusvm)
-
 void xe_svm_flush(struct xe_vm *vm);
 
 #else
@@ -206,12 +197,21 @@ int xe_devm_add(struct xe_tile *tile, struct xe_vram_region *vr)
 static inline
 int xe_svm_init(struct xe_vm *vm)
 {
+#if IS_ENABLED(CONFIG_DRM_GPUSVM)
+	return drm_gpusvm_init(&vm->svm.gpusvm, "Xe SVM (simple)", &vm->xe->drm,
+			       NULL, NULL, 0, 0, 0, NULL, NULL, 0);
+#else
 	return 0;
+#endif
 }
 
 static inline
 void xe_svm_fini(struct xe_vm *vm)
 {
+#if IS_ENABLED(CONFIG_DRM_GPUSVM)
+	xe_assert(vm->xe, xe_vm_is_closed(vm));
+	drm_gpusvm_fini(&vm->svm.gpusvm);
+#endif
 }
 
 static inline
@@ -328,19 +328,47 @@ struct drm_pagemap *xe_vma_resolve_pagemap(struct xe_vma *vma, struct xe_tile *t
 	return NULL;
 }
 
-#define xe_svm_assert_in_notifier(...) do {} while (0)
+static inline void xe_svm_flush(struct xe_vm *vm)
+{
+}
 #define xe_svm_range_has_dma_mapping(...) false
+#endif /* CONFIG_DRM_XE_GPUSVM */
+
+#if IS_ENABLED(CONFIG_DRM_GPUSVM) /* Need to support userptr without XE_GPUSVM */
+#define xe_svm_assert_in_notifier(vm__) \
+	lockdep_assert_held_write(&(vm__)->svm.gpusvm.notifier_lock)
+
+#define xe_svm_assert_held_read(vm__) \
+	lockdep_assert_held_read(&(vm__)->svm.gpusvm.notifier_lock)
+
+#define xe_svm_notifier_lock(vm__)	\
+	drm_gpusvm_notifier_lock(&(vm__)->svm.gpusvm)
+
+#define xe_svm_notifier_lock_interruptible(vm__)	\
+	down_read_interruptible(&(vm__)->svm.gpusvm.notifier_lock)
+
+#define xe_svm_notifier_unlock(vm__)	\
+	drm_gpusvm_notifier_unlock(&(vm__)->svm.gpusvm)
+
+#else
+#define xe_svm_assert_in_notifier(...) do {} while (0)
+
+static inline void xe_svm_assert_held_read(struct xe_vm *vm)
+{
+}
 
 static inline void xe_svm_notifier_lock(struct xe_vm *vm)
 {
 }
 
+static inline int xe_svm_notifier_lock_interruptible(struct xe_vm *vm)
+{
+	return 0;
+}
+
 static inline void xe_svm_notifier_unlock(struct xe_vm *vm)
 {
 }
+#endif /* CONFIG_DRM_GPUSVM */
 
-static inline void xe_svm_flush(struct xe_vm *vm)
-{
-}
-#endif
 #endif
diff --git a/drivers/gpu/drm/xe/xe_userptr.c b/drivers/gpu/drm/xe/xe_userptr.c
index 17ecff17c993..91d09af71ced 100644
--- a/drivers/gpu/drm/xe/xe_userptr.c
+++ b/drivers/gpu/drm/xe/xe_userptr.c
@@ -7,7 +7,6 @@
 
 #include <linux/mm.h>
 
-#include "xe_hmm.h"
 #include "xe_trace_bo.h"
 
 /**
@@ -25,7 +24,7 @@
 int xe_vma_userptr_check_repin(struct xe_userptr_vma *uvma)
 {
 	return mmu_interval_check_retry(&uvma->userptr.notifier,
-					uvma->userptr.notifier_seq) ?
+					uvma->userptr.pages.notifier_seq) ?
 		-EAGAIN : 0;
 }
 
@@ -35,14 +34,14 @@ int xe_vma_userptr_check_repin(struct xe_userptr_vma *uvma)
  * @vm: The VM.
  *
  * This function checks for whether the VM has userptrs that need repinning,
- * and provides a release-type barrier on the userptr.notifier_lock after
+ * and provides a release-type barrier on the svm.gpusvm.notifier_lock after
  * checking.
  *
  * Return: 0 if there are no userptrs needing repinning, -EAGAIN if there are.
  */
 int __xe_vm_userptr_needs_repin(struct xe_vm *vm)
 {
-	lockdep_assert_held_read(&vm->userptr.notifier_lock);
+	lockdep_assert_held_read(&vm->svm.gpusvm.notifier_lock);
 
 	return (list_empty(&vm->userptr.repin_list) &&
 		list_empty(&vm->userptr.invalidated)) ? 0 : -EAGAIN;
@@ -53,11 +52,22 @@ int xe_vma_userptr_pin_pages(struct xe_userptr_vma *uvma)
 	struct xe_vma *vma = &uvma->vma;
 	struct xe_vm *vm = xe_vma_vm(vma);
 	struct xe_device *xe = vm->xe;
+	struct drm_gpusvm_ctx ctx = {
+		.read_only = xe_vma_read_only(vma),
+	};
 
 	lockdep_assert_held(&vm->lock);
 	xe_assert(xe, xe_vma_is_userptr(vma));
 
-	return xe_hmm_userptr_populate_range(uvma, false);
+	if (vma->gpuva.flags & XE_VMA_DESTROYED)
+		return 0;
+
+	return drm_gpusvm_get_pages(&vm->svm.gpusvm, &uvma->userptr.pages,
+				    uvma->userptr.notifier.mm,
+				    &uvma->userptr.notifier,
+				    xe_vma_userptr(vma),
+				    xe_vma_userptr(vma) + xe_vma_size(vma),
+				    &ctx);
 }
 
 static void __vma_userptr_invalidate(struct xe_vm *vm, struct xe_userptr_vma *uvma)
@@ -66,6 +76,10 @@ static void __vma_userptr_invalidate(struct xe_vm *vm, struct xe_userptr_vma *uv
 	struct xe_vma *vma = &uvma->vma;
 	struct dma_resv_iter cursor;
 	struct dma_fence *fence;
+	struct drm_gpusvm_ctx ctx = {
+		.in_notifier = true,
+		.read_only = xe_vma_read_only(vma),
+	};
 	long err;
 
 	/*
@@ -102,7 +116,8 @@ static void __vma_userptr_invalidate(struct xe_vm *vm, struct xe_userptr_vma *uv
 		XE_WARN_ON(err);
 	}
 
-	xe_hmm_userptr_unmap(uvma);
+	drm_gpusvm_unmap_pages(&vm->svm.gpusvm, &uvma->userptr.pages,
+			       xe_vma_size(vma) >> PAGE_SHIFT, &ctx);
 }
 
 static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni,
@@ -123,11 +138,11 @@ static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni,
 	       "NOTIFIER: addr=0x%016llx, range=0x%016llx",
 		xe_vma_start(vma), xe_vma_size(vma));
 
-	down_write(&vm->userptr.notifier_lock);
+	down_write(&vm->svm.gpusvm.notifier_lock);
 	mmu_interval_set_seq(mni, cur_seq);
 
 	__vma_userptr_invalidate(vm, uvma);
-	up_write(&vm->userptr.notifier_lock);
+	up_write(&vm->svm.gpusvm.notifier_lock);
 	trace_xe_vma_userptr_invalidate_complete(vma);
 
 	return true;
@@ -151,7 +166,7 @@ void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma)
 	/* Protect against concurrent userptr pinning */
 	lockdep_assert_held(&vm->lock);
 	/* Protect against concurrent notifiers */
-	lockdep_assert_held(&vm->userptr.notifier_lock);
+	lockdep_assert_held(&vm->svm.gpusvm.notifier_lock);
 	/*
 	 * Protect against concurrent instances of this function and
 	 * the critical exec sections
@@ -159,8 +174,8 @@ void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma)
 	xe_vm_assert_held(vm);
 
 	if (!mmu_interval_read_retry(&uvma->userptr.notifier,
-				     uvma->userptr.notifier_seq))
-		uvma->userptr.notifier_seq -= 2;
+				     uvma->userptr.pages.notifier_seq))
+		uvma->userptr.pages.notifier_seq -= 2;
 	__vma_userptr_invalidate(vm, uvma);
 }
 #endif
@@ -209,9 +224,9 @@ int xe_vm_userptr_pin(struct xe_vm *vm)
 					      DMA_RESV_USAGE_BOOKKEEP,
 					      false, MAX_SCHEDULE_TIMEOUT);
 
-			down_read(&vm->userptr.notifier_lock);
+			down_read(&vm->svm.gpusvm.notifier_lock);
 			err = xe_vm_invalidate_vma(&uvma->vma);
-			up_read(&vm->userptr.notifier_lock);
+			up_read(&vm->svm.gpusvm.notifier_lock);
 			xe_vm_unlock(vm);
 			if (err)
 				break;
@@ -226,7 +241,7 @@ int xe_vm_userptr_pin(struct xe_vm *vm)
 	}
 
 	if (err) {
-		down_write(&vm->userptr.notifier_lock);
+		down_write(&vm->svm.gpusvm.notifier_lock);
 		spin_lock(&vm->userptr.invalidated_lock);
 		list_for_each_entry_safe(uvma, next, &vm->userptr.repin_list,
 					 userptr.repin_link) {
@@ -235,7 +250,7 @@ int xe_vm_userptr_pin(struct xe_vm *vm)
 				       &vm->userptr.invalidated);
 		}
 		spin_unlock(&vm->userptr.invalidated_lock);
-		up_write(&vm->userptr.notifier_lock);
+		up_write(&vm->svm.gpusvm.notifier_lock);
 	}
 	return err;
 }
@@ -265,7 +280,6 @@ int xe_userptr_setup(struct xe_userptr_vma *uvma, unsigned long start,
 
 	INIT_LIST_HEAD(&userptr->invalidate_link);
 	INIT_LIST_HEAD(&userptr->repin_link);
-	mutex_init(&userptr->unmap_mutex);
 
 	err = mmu_interval_notifier_insert(&userptr->notifier, current->mm,
 					   start, range,
@@ -273,17 +287,18 @@ int xe_userptr_setup(struct xe_userptr_vma *uvma, unsigned long start,
 	if (err)
 		return err;
 
-	userptr->notifier_seq = LONG_MAX;
+	userptr->pages.notifier_seq = LONG_MAX;
 
 	return 0;
 }
 
 void xe_userptr_remove(struct xe_userptr_vma *uvma)
 {
+	struct xe_vm *vm = xe_vma_vm(&uvma->vma);
 	struct xe_userptr *userptr = &uvma->userptr;
 
-	if (userptr->sg)
-		xe_hmm_userptr_free_sg(uvma);
+	drm_gpusvm_free_pages(&vm->svm.gpusvm, &uvma->userptr.pages,
+			      xe_vma_size(&uvma->vma) >> PAGE_SHIFT);
 
 	/*
 	 * Since userptr pages are not pinned, we can't remove
@@ -291,7 +306,6 @@ void xe_userptr_remove(struct xe_userptr_vma *uvma)
 	 * them anymore
 	 */
 	mmu_interval_notifier_remove(&userptr->notifier);
-	mutex_destroy(&userptr->unmap_mutex);
 }
 
 void xe_userptr_destroy(struct xe_userptr_vma *uvma)
diff --git a/drivers/gpu/drm/xe/xe_userptr.h b/drivers/gpu/drm/xe/xe_userptr.h
index dc49348ac42d..ef801234991e 100644
--- a/drivers/gpu/drm/xe/xe_userptr.h
+++ b/drivers/gpu/drm/xe/xe_userptr.h
@@ -12,6 +12,8 @@
 #include <linux/scatterlist.h>
 #include <linux/spinlock.h>
 
+#include <drm/drm_gpusvm.h>
+
 struct xe_vm;
 struct xe_vma;
 struct xe_userptr_vma;
@@ -23,11 +25,6 @@ struct xe_userptr_vm {
 	 * and needs repinning. Protected by @lock.
 	 */
 	struct list_head repin_list;
-	/**
-	 * @notifier_lock: protects notifier in write mode and
-	 * submission in read mode.o
-	 */
-	struct rw_semaphore notifier_lock;
 	/**
 	 * @userptr.invalidated_lock: Protects the
 	 * @userptr.invalidated list.
@@ -39,7 +36,7 @@ struct xe_userptr_vm {
 	 * up for revalidation. Protected from access with the
 	 * @invalidated_lock. Removing items from the list
 	 * additionally requires @lock in write mode, and adding
-	 * items to the list requires either the @userptr.notifier_lock in
+	 * items to the list requires either the @svm.gpusvm.notifier_lock in
 	 * write mode, OR @lock in write mode.
 	 */
 	struct list_head invalidated;
@@ -51,31 +48,27 @@ struct xe_userptr {
 	struct list_head invalidate_link;
 	/** @userptr: link into VM repin list if userptr. */
 	struct list_head repin_link;
+	/**
+	 * @pages: gpusvm pages for this user pointer.
+	 */
+	struct drm_gpusvm_pages pages;
 	/**
 	 * @notifier: MMU notifier for user pointer (invalidation call back)
 	 */
 	struct mmu_interval_notifier notifier;
-	/** @sgt: storage for a scatter gather table */
-	struct sg_table sgt;
-	/** @sg: allocated scatter gather table */
-	struct sg_table *sg;
-	/** @notifier_seq: notifier sequence number */
-	unsigned long notifier_seq;
-	/** @unmap_mutex: Mutex protecting dma-unmapping */
-	struct mutex unmap_mutex;
+
 	/**
 	 * @initial_bind: user pointer has been bound at least once.
-	 * write: vm->userptr.notifier_lock in read mode and vm->resv held.
-	 * read: vm->userptr.notifier_lock in write mode or vm->resv held.
+	 * write: vm->svm.gpusvm.notifier_lock in read mode and vm->resv held.
+	 * read: vm->svm.gpusvm.notifier_lock in write mode or vm->resv held.
 	 */
 	bool initial_bind;
-	/** @mapped: Whether the @sgt sg-table is dma-mapped. Protected by @unmap_mutex. */
-	bool mapped;
 #if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
 	u32 divisor;
 #endif
 };
 
+#if IS_ENABLED(CONFIG_DRM_GPUSVM)
 void xe_userptr_remove(struct xe_userptr_vma *uvma);
 int xe_userptr_setup(struct xe_userptr_vma *uvma, unsigned long start,
 		     unsigned long range);
@@ -86,6 +79,23 @@ int __xe_vm_userptr_needs_repin(struct xe_vm *vm);
 int xe_vm_userptr_check_repin(struct xe_vm *vm);
 int xe_vma_userptr_pin_pages(struct xe_userptr_vma *uvma);
 int xe_vma_userptr_check_repin(struct xe_userptr_vma *uvma);
+#else
+static inline void xe_userptr_remove(struct xe_userptr_vma *uvma) {}
+
+static inline int xe_userptr_setup(struct xe_userptr_vma *uvma,
+				   unsigned long start, unsigned long range)
+{
+	return -ENODEV;
+}
+
+static inline void xe_userptr_destroy(struct xe_userptr_vma *uvma) {}
+
+static inline int xe_vm_userptr_pin(struct xe_vm *vm) { return 0; }
+static inline int __xe_vm_userptr_needs_repin(struct xe_vm *vm) { return 0; }
+static inline int xe_vm_userptr_check_repin(struct xe_vm *vm) { return 0; }
+static inline int xe_vma_userptr_pin_pages(struct xe_userptr_vma *uvma) { return -ENODEV; }
+static inline int xe_vma_userptr_check_repin(struct xe_userptr_vma *uvma) { return -ENODEV; };
+#endif
 
 #if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
 void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma);
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index 57afe9819842..49a0cc0ab6c7 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -40,9 +40,7 @@
 #include "xe_tile.h"
 #include "xe_tlb_inval.h"
 #include "xe_trace_bo.h"
-#include "xe_userptr.h"
 #include "xe_wa.h"
-#include "xe_hmm.h"
 
 static struct drm_gem_object *xe_vm_obj(struct xe_vm *vm)
 {
@@ -220,7 +218,7 @@ int xe_vm_add_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q)
 	++vm->preempt.num_exec_queues;
 	q->lr.pfence = pfence;
 
-	down_read(&vm->userptr.notifier_lock);
+	xe_svm_notifier_lock(vm);
 
 	drm_gpuvm_resv_add_fence(&vm->gpuvm, exec, pfence,
 				 DMA_RESV_USAGE_BOOKKEEP, DMA_RESV_USAGE_BOOKKEEP);
@@ -234,7 +232,7 @@ int xe_vm_add_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q)
 	if (wait)
 		dma_fence_enable_sw_signaling(pfence);
 
-	up_read(&vm->userptr.notifier_lock);
+	xe_svm_notifier_unlock(vm);
 
 out_fini:
 	drm_exec_fini(exec);
@@ -498,9 +496,9 @@ static void preempt_rebind_work_func(struct work_struct *w)
 	(!(__tries)++ || __xe_vm_userptr_needs_repin(__vm)) : \
 	__xe_vm_userptr_needs_repin(__vm))
 
-	down_read(&vm->userptr.notifier_lock);
+	xe_svm_notifier_lock(vm);
 	if (retry_required(tries, vm)) {
-		up_read(&vm->userptr.notifier_lock);
+		xe_svm_notifier_unlock(vm);
 		err = -EAGAIN;
 		goto out_unlock;
 	}
@@ -514,7 +512,7 @@ static void preempt_rebind_work_func(struct work_struct *w)
 	/* Point of no return. */
 	arm_preempt_fences(vm, &preempt_fences);
 	resume_and_reinstall_preempt_fences(vm, &exec);
-	up_read(&vm->userptr.notifier_lock);
+	xe_svm_notifier_unlock(vm);
 
 out_unlock:
 	drm_exec_fini(&exec);
@@ -1449,7 +1447,6 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags, struct xe_file *xef)
 
 	INIT_LIST_HEAD(&vm->userptr.repin_list);
 	INIT_LIST_HEAD(&vm->userptr.invalidated);
-	init_rwsem(&vm->userptr.notifier_lock);
 	spin_lock_init(&vm->userptr.invalidated_lock);
 
 	ttm_lru_bulk_move_init(&vm->lru_bulk_move);
@@ -1475,11 +1472,9 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags, struct xe_file *xef)
 		xe_pm_runtime_get_noresume(xe);
 	}
 
-	if (flags & XE_VM_FLAG_FAULT_MODE) {
-		err = xe_svm_init(vm);
-		if (err)
-			goto err_no_resv;
-	}
+	err = xe_svm_init(vm);
+	if (err)
+		goto err_no_resv;
 
 	vm_resv_obj = drm_gpuvm_resv_object_alloc(&xe->drm);
 	if (!vm_resv_obj) {
@@ -1680,9 +1675,9 @@ void xe_vm_close_and_put(struct xe_vm *vm)
 		vma = gpuva_to_vma(gpuva);
 
 		if (xe_vma_has_no_bo(vma)) {
-			down_read(&vm->userptr.notifier_lock);
+			xe_svm_notifier_lock(vm);
 			vma->gpuva.flags |= XE_VMA_DESTROYED;
-			up_read(&vm->userptr.notifier_lock);
+			xe_svm_notifier_unlock(vm);
 		}
 
 		xe_vm_remove_vma(vm, vma);
@@ -1726,8 +1721,7 @@ void xe_vm_close_and_put(struct xe_vm *vm)
 		xe_vma_destroy_unlocked(vma);
 	}
 
-	if (xe_vm_in_fault_mode(vm))
-		xe_svm_fini(vm);
+	xe_svm_fini(vm);
 
 	up_write(&vm->lock);
 
@@ -2068,9 +2062,9 @@ static const u32 region_to_mem_type[] = {
 static void prep_vma_destroy(struct xe_vm *vm, struct xe_vma *vma,
 			     bool post_commit)
 {
-	down_read(&vm->userptr.notifier_lock);
+	xe_svm_notifier_lock(vm);
 	vma->gpuva.flags |= XE_VMA_DESTROYED;
-	up_read(&vm->userptr.notifier_lock);
+	xe_svm_notifier_unlock(vm);
 	if (post_commit)
 		xe_vm_remove_vma(vm, vma);
 }
@@ -2713,9 +2707,9 @@ static void xe_vma_op_unwind(struct xe_vm *vm, struct xe_vma_op *op,
 		struct xe_vma *vma = gpuva_to_vma(op->base.unmap.va);
 
 		if (vma) {
-			down_read(&vm->userptr.notifier_lock);
+			xe_svm_notifier_lock(vm);
 			vma->gpuva.flags &= ~XE_VMA_DESTROYED;
-			up_read(&vm->userptr.notifier_lock);
+			xe_svm_notifier_unlock(vm);
 			if (post_commit)
 				xe_vm_insert_vma(vm, vma);
 		}
@@ -2734,9 +2728,9 @@ static void xe_vma_op_unwind(struct xe_vm *vm, struct xe_vma_op *op,
 			xe_vma_destroy_unlocked(op->remap.next);
 		}
 		if (vma) {
-			down_read(&vm->userptr.notifier_lock);
+			xe_svm_notifier_lock(vm);
 			vma->gpuva.flags &= ~XE_VMA_DESTROYED;
-			up_read(&vm->userptr.notifier_lock);
+			xe_svm_notifier_unlock(vm);
 			if (post_commit)
 				xe_vm_insert_vma(vm, vma);
 		}
@@ -3311,6 +3305,8 @@ static int vm_bind_ioctl_check_args(struct xe_device *xe, struct xe_vm *vm,
 				 op == DRM_XE_VM_BIND_OP_MAP_USERPTR) ||
 		    XE_IOCTL_DBG(xe, coh_mode == XE_COH_NONE &&
 				 op == DRM_XE_VM_BIND_OP_MAP_USERPTR) ||
+		    XE_IOCTL_DBG(xe, op == DRM_XE_VM_BIND_OP_MAP_USERPTR &&
+				 !IS_ENABLED(CONFIG_DRM_GPUSVM)) ||
 		    XE_IOCTL_DBG(xe, obj &&
 				 op == DRM_XE_VM_BIND_OP_PREFETCH) ||
 		    XE_IOCTL_DBG(xe, prefetch_region &&
@@ -3856,13 +3852,13 @@ int xe_vm_invalidate_vma(struct xe_vma *vma)
 	 */
 	if (IS_ENABLED(CONFIG_PROVE_LOCKING)) {
 		if (xe_vma_is_userptr(vma)) {
-			lockdep_assert(lockdep_is_held_type(&vm->userptr.notifier_lock, 0) ||
-				       (lockdep_is_held_type(&vm->userptr.notifier_lock, 1) &&
+			lockdep_assert(lockdep_is_held_type(&vm->svm.gpusvm.notifier_lock, 0) ||
+				       (lockdep_is_held_type(&vm->svm.gpusvm.notifier_lock, 1) &&
 					lockdep_is_held(&xe_vm_resv(vm)->lock.base)));
 
 			WARN_ON_ONCE(!mmu_interval_check_retry
 				     (&to_userptr_vma(vma)->userptr.notifier,
-				      to_userptr_vma(vma)->userptr.notifier_seq));
+				      to_userptr_vma(vma)->userptr.pages.notifier_seq));
 			WARN_ON_ONCE(!dma_resv_test_signaled(xe_vm_resv(vm),
 							     DMA_RESV_USAGE_BOOKKEEP));
 
diff --git a/drivers/gpu/drm/xe/xe_vm_madvise.c b/drivers/gpu/drm/xe/xe_vm_madvise.c
index 09c5783ee523..9862f2498f18 100644
--- a/drivers/gpu/drm/xe/xe_vm_madvise.c
+++ b/drivers/gpu/drm/xe/xe_vm_madvise.c
@@ -18,9 +18,8 @@ struct xe_vmas_in_madvise_range {
 	u64 range;
 	struct xe_vma **vmas;
 	int num_vmas;
-	bool has_svm_vmas;
 	bool has_bo_vmas;
-	bool has_userptr_vmas;
+	bool has_svm_userptr_vmas;
 };
 
 static int get_vmas(struct xe_vm *vm, struct xe_vmas_in_madvise_range *madvise_range)
@@ -46,10 +45,8 @@ static int get_vmas(struct xe_vm *vm, struct xe_vmas_in_madvise_range *madvise_r
 
 		if (xe_vma_bo(vma))
 			madvise_range->has_bo_vmas = true;
-		else if (xe_vma_is_cpu_addr_mirror(vma))
-			madvise_range->has_svm_vmas = true;
-		else if (xe_vma_is_userptr(vma))
-			madvise_range->has_userptr_vmas = true;
+		else if (xe_vma_is_cpu_addr_mirror(vma) || xe_vma_is_userptr(vma))
+			madvise_range->has_svm_userptr_vmas = true;
 
 		if (madvise_range->num_vmas == max_vmas) {
 			max_vmas <<= 1;
@@ -409,16 +406,10 @@ int xe_vm_madvise_ioctl(struct drm_device *dev, void *data, struct drm_file *fil
 		}
 	}
 
-	if (madvise_range.has_userptr_vmas) {
-		err = down_read_interruptible(&vm->userptr.notifier_lock);
-		if (err)
-			goto err_fini;
-	}
-
-	if (madvise_range.has_svm_vmas) {
+	if (madvise_range.has_svm_userptr_vmas) {
 		err = down_read_interruptible(&vm->svm.gpusvm.notifier_lock);
 		if (err)
-			goto unlock_userptr;
+			goto err_fini;
 	}
 
 	attr_type = array_index_nospec(args->type, ARRAY_SIZE(madvise_funcs));
@@ -426,12 +417,9 @@ int xe_vm_madvise_ioctl(struct drm_device *dev, void *data, struct drm_file *fil
 
 	err = xe_vm_invalidate_madvise_range(vm, args->start, args->start + args->range);
 
-	if (madvise_range.has_svm_vmas)
+	if (madvise_range.has_svm_userptr_vmas)
 		xe_svm_notifier_unlock(vm);
 
-unlock_userptr:
-	if (madvise_range.has_userptr_vmas)
-		up_read(&vm->userptr.notifier_lock);
 err_fini:
 	if (madvise_range.has_bo_vmas)
 		drm_exec_fini(&exec);
diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
index aee050c295db..f6dd964be42f 100644
--- a/drivers/gpu/drm/xe/xe_vm_types.h
+++ b/drivers/gpu/drm/xe/xe_vm_types.h
@@ -110,10 +110,10 @@ struct xe_vma {
 
 	/**
 	 * @tile_invalidated: Tile mask of binding are invalidated for this VMA.
-	 * protected by BO's resv and for userptrs, vm->userptr.notifier_lock in
-	 * write mode for writing or vm->userptr.notifier_lock in read mode and
+	 * protected by BO's resv and for userptrs, vm->svm.gpusvm.notifier_lock in
+	 * write mode for writing or vm->svm.gpusvm.notifier_lock in read mode and
 	 * the vm->resv. For stable reading, BO's resv or userptr
-	 * vm->userptr.notifier_lock in read mode is required. Can be
+	 * vm->svm.gpusvm.notifier_lock in read mode is required. Can be
 	 * opportunistically read with READ_ONCE outside of locks.
 	 */
 	u8 tile_invalidated;
@@ -124,7 +124,7 @@ struct xe_vma {
 	/**
 	 * @tile_present: Tile mask of binding are present for this VMA.
 	 * protected by vm->lock, vm->resv and for userptrs,
-	 * vm->userptr.notifier_lock for writing. Needs either for reading,
+	 * vm->svm.gpusvm.notifier_lock for writing. Needs either for reading,
 	 * but if reading is done under the vm->lock only, it needs to be held
 	 * in write mode.
 	 */
-- 
2.51.0

[PATCH v6 8/8] drm/xe/pt: unify xe_pt_svm_pre_commit with userptr

[Matthew Auld]

We now use the same notifier lock for SVM and userptr, with that we can
combine xe_pt_userptr_pre_commit and xe_pt_svm_pre_commit.

v2: (Matt B)
  - Re-use xe_svm_notifier_lock/unlock for userptr.
  - Combine svm/userptr handling further down into op_check_svm_userptr.
v3:
  - Only hide the ops if we lack DRM_GPUSVM, since we also need them for
    userptr.

Suggested-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
---
 drivers/gpu/drm/xe/xe_pt.c       | 123 ++++++++++++-------------------
 drivers/gpu/drm/xe/xe_pt_types.h |   2 -
 2 files changed, 46 insertions(+), 79 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
index 382b25798bc3..74e33153c827 100644
--- a/drivers/gpu/drm/xe/xe_pt.c
+++ b/drivers/gpu/drm/xe/xe_pt.c
@@ -1432,8 +1432,8 @@ static int vma_check_userptr(struct xe_vm *vm, struct xe_vma *vma,
 	return 0;
 }
 
-static int op_check_userptr(struct xe_vm *vm, struct xe_vma_op *op,
-			    struct xe_vm_pgtable_update_ops *pt_update)
+static int op_check_svm_userptr(struct xe_vm *vm, struct xe_vma_op *op,
+				struct xe_vm_pgtable_update_ops *pt_update)
 {
 	int err = 0;
 
@@ -1455,9 +1455,40 @@ static int op_check_userptr(struct xe_vm *vm, struct xe_vma_op *op,
 	case DRM_GPUVA_OP_UNMAP:
 		break;
 	case DRM_GPUVA_OP_PREFETCH:
-		err = vma_check_userptr(vm, gpuva_to_vma(op->base.prefetch.va),
-					pt_update);
+		if (xe_vma_is_cpu_addr_mirror(gpuva_to_vma(op->base.prefetch.va))) {
+			struct xe_svm_range *range = op->map_range.range;
+			unsigned long i;
+
+			xe_assert(vm->xe,
+				  xe_vma_is_cpu_addr_mirror(gpuva_to_vma(op->base.prefetch.va)));
+			xa_for_each(&op->prefetch_range.range, i, range) {
+				xe_svm_range_debug(range, "PRE-COMMIT");
+
+				if (!xe_svm_range_pages_valid(range)) {
+					xe_svm_range_debug(range, "PRE-COMMIT - RETRY");
+					return -ENODATA;
+				}
+			}
+		} else {
+			err = vma_check_userptr(vm, gpuva_to_vma(op->base.prefetch.va), pt_update);
+		}
 		break;
+#if IS_ENABLED(CONFIG_DRM_XE_GPUSVM)
+	case DRM_GPUVA_OP_DRIVER:
+		if (op->subop == XE_VMA_SUBOP_MAP_RANGE) {
+			struct xe_svm_range *range = op->map_range.range;
+
+			xe_assert(vm->xe, xe_vma_is_cpu_addr_mirror(op->map_range.vma));
+
+			xe_svm_range_debug(range, "PRE-COMMIT");
+
+			if (!xe_svm_range_pages_valid(range)) {
+				xe_svm_range_debug(range, "PRE-COMMIT - RETRY");
+				return -EAGAIN;
+			}
+		}
+		break;
+#endif
 	default:
 		drm_warn(&vm->xe->drm, "NOT POSSIBLE");
 	}
@@ -1465,7 +1496,7 @@ static int op_check_userptr(struct xe_vm *vm, struct xe_vma_op *op,
 	return err;
 }
 
-static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)
+static int xe_pt_svm_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)
 {
 	struct xe_vm *vm = pt_update->vops->vm;
 	struct xe_vma_ops *vops = pt_update->vops;
@@ -1478,12 +1509,12 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)
 	if (err)
 		return err;
 
-	down_read(&vm->svm.gpusvm.notifier_lock);
+	xe_svm_notifier_lock(vm);
 
 	list_for_each_entry(op, &vops->list, link) {
-		err = op_check_userptr(vm, op, pt_update_ops);
+		err = op_check_svm_userptr(vm, op, pt_update_ops);
 		if (err) {
-			up_read(&vm->svm.gpusvm.notifier_lock);
+			xe_svm_notifier_unlock(vm);
 			break;
 		}
 	}
@@ -1491,58 +1522,6 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)
 	return err;
 }
 
-#if IS_ENABLED(CONFIG_DRM_XE_GPUSVM)
-static int xe_pt_svm_pre_commit(struct xe_migrate_pt_update *pt_update)
-{
-	struct xe_vm *vm = pt_update->vops->vm;
-	struct xe_vma_ops *vops = pt_update->vops;
-	struct xe_vma_op *op;
-	unsigned long i;
-	int err;
-
-	err = xe_pt_pre_commit(pt_update);
-	if (err)
-		return err;
-
-	xe_svm_notifier_lock(vm);
-
-	list_for_each_entry(op, &vops->list, link) {
-		struct xe_svm_range *range = NULL;
-
-		if (op->subop == XE_VMA_SUBOP_UNMAP_RANGE)
-			continue;
-
-		if (op->base.op == DRM_GPUVA_OP_PREFETCH) {
-			xe_assert(vm->xe,
-				  xe_vma_is_cpu_addr_mirror(gpuva_to_vma(op->base.prefetch.va)));
-			xa_for_each(&op->prefetch_range.range, i, range) {
-				xe_svm_range_debug(range, "PRE-COMMIT");
-
-				if (!xe_svm_range_pages_valid(range)) {
-					xe_svm_range_debug(range, "PRE-COMMIT - RETRY");
-					xe_svm_notifier_unlock(vm);
-					return -ENODATA;
-				}
-			}
-		} else {
-			xe_assert(vm->xe, xe_vma_is_cpu_addr_mirror(op->map_range.vma));
-			xe_assert(vm->xe, op->subop == XE_VMA_SUBOP_MAP_RANGE);
-			range = op->map_range.range;
-
-			xe_svm_range_debug(range, "PRE-COMMIT");
-
-			if (!xe_svm_range_pages_valid(range)) {
-				xe_svm_range_debug(range, "PRE-COMMIT - RETRY");
-				xe_svm_notifier_unlock(vm);
-				return -EAGAIN;
-			}
-		}
-	}
-
-	return 0;
-}
-#endif
-
 struct xe_pt_stage_unbind_walk {
 	/** @base: The pagewalk base-class. */
 	struct xe_pt_walk base;
@@ -1844,7 +1823,7 @@ static int bind_op_prepare(struct xe_vm *vm, struct xe_tile *tile,
 						 xe_vma_start(vma),
 						 xe_vma_end(vma));
 		++pt_update_ops->current_op;
-		pt_update_ops->needs_userptr_lock |= xe_vma_is_userptr(vma);
+		pt_update_ops->needs_svm_lock |= xe_vma_is_userptr(vma);
 
 		/*
 		 * If rebind, we have to invalidate TLB on !LR vms to invalidate
@@ -1952,7 +1931,7 @@ static int unbind_op_prepare(struct xe_tile *tile,
 	xe_pt_update_ops_rfence_interval(pt_update_ops, xe_vma_start(vma),
 					 xe_vma_end(vma));
 	++pt_update_ops->current_op;
-	pt_update_ops->needs_userptr_lock |= xe_vma_is_userptr(vma);
+	pt_update_ops->needs_svm_lock |= xe_vma_is_userptr(vma);
 	pt_update_ops->needs_invalidation = true;
 
 	xe_pt_commit_prepare_unbind(vma, pt_op->entries, pt_op->num_entries);
@@ -2339,20 +2318,14 @@ static const struct xe_migrate_pt_update_ops migrate_ops = {
 	.pre_commit = xe_pt_pre_commit,
 };
 
-static const struct xe_migrate_pt_update_ops userptr_migrate_ops = {
+#if IS_ENABLED(CONFIG_DRM_GPUSVM)
+static const struct xe_migrate_pt_update_ops svm_userptr_migrate_ops = {
 	.populate = xe_vm_populate_pgtable,
 	.clear = xe_migrate_clear_pgtable_callback,
-	.pre_commit = xe_pt_userptr_pre_commit,
-};
-
-#if IS_ENABLED(CONFIG_DRM_XE_GPUSVM)
-static const struct xe_migrate_pt_update_ops svm_migrate_ops = {
-	.populate = xe_vm_populate_pgtable,
-	.clear = xe_migrate_clear_pgtable_callback,
-	.pre_commit = xe_pt_svm_pre_commit,
+	.pre_commit = xe_pt_svm_userptr_pre_commit,
 };
 #else
-static const struct xe_migrate_pt_update_ops svm_migrate_ops;
+static const struct xe_migrate_pt_update_ops svm_userptr_migrate_ops;
 #endif
 
 static struct xe_dep_scheduler *to_dep_scheduler(struct xe_exec_queue *q,
@@ -2390,9 +2363,7 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops)
 	int err = 0, i;
 	struct xe_migrate_pt_update update = {
 		.ops = pt_update_ops->needs_svm_lock ?
-			&svm_migrate_ops :
-			pt_update_ops->needs_userptr_lock ?
-			&userptr_migrate_ops :
+			&svm_userptr_migrate_ops :
 			&migrate_ops,
 		.vops = vops,
 		.tile_id = tile->id,
@@ -2534,8 +2505,6 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops)
 
 	if (pt_update_ops->needs_svm_lock)
 		xe_svm_notifier_unlock(vm);
-	if (pt_update_ops->needs_userptr_lock)
-		up_read(&vm->svm.gpusvm.notifier_lock);
 
 	xe_tlb_inval_job_put(mjob);
 	xe_tlb_inval_job_put(ijob);
diff --git a/drivers/gpu/drm/xe/xe_pt_types.h b/drivers/gpu/drm/xe/xe_pt_types.h
index 17cdd7c7e9f5..881f01e14db8 100644
--- a/drivers/gpu/drm/xe/xe_pt_types.h
+++ b/drivers/gpu/drm/xe/xe_pt_types.h
@@ -105,8 +105,6 @@ struct xe_vm_pgtable_update_ops {
 	u32 current_op;
 	/** @needs_svm_lock: Needs SVM lock */
 	bool needs_svm_lock;
-	/** @needs_userptr_lock: Needs userptr lock */
-	bool needs_userptr_lock;
 	/** @needs_invalidation: Needs invalidation */
 	bool needs_invalidation;
 	/**
-- 
2.51.0