Skip to contentSkip to conversation

More VM cleanups

PATCH Superseded Matthew Brost 3 participants 11 messages Jul 20, 2023

Subsystems (2)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Matthew Brost

author maintainer
#0 lore raw

v2 cover More VM cleanups

Continuation of [1]. These are the last changes in [1] that will not break the uAPI.

Matt

[1] https://patchwork.freedesktop.org/series/117156/

Matthew Brost (6):
drm/xe: Avoid doing rebinds
drm/xe: Reduce the number list links in xe_vma drm/xe: Replace list_del_init with list_del for
userptr.invalidate_link cleanup
drm/xe: Change tile masks from u64 to u8
drm/xe: Combine destroy_cb and destroy_work in xe_vma into union drm/xe: Only alloc userptr part of xe_vma for userptrs

 drivers/gpu/drm/xe/xe_bo.c       |   6 +-
 drivers/gpu/drm/xe/xe_exec.c     |   2 +-
 drivers/gpu/drm/xe/xe_pt.c       |   3 +-
 drivers/gpu/drm/xe/xe_vm.c       | 135 ++++++++++++++++++++++---------
 drivers/gpu/drm/xe/xe_vm_types.h | 131 +++++++++++++++++-------------
 5 files changed, 178 insertions(+), 99 deletions(-)
Signature · 1 line
2.34.1
author maintainer
#1 lore raw

PATCH v2 1/6 drm/xe: Avoid doing rebinds

If we dont change page sizes we can avoid doing rebinds rather just do a partial unbind. The algorithm to determine its page size is greedy as we assume all pages in the removed VMA are the largest page used in the VMA.

v2: Don't exceed 100 lines
v3: struct xe_vma_op_unmap remove in different patch, remove XXX comment

---
 drivers/gpu/drm/xe/xe_pt.c       |  1 +
 drivers/gpu/drm/xe/xe_vm.c       | 72 +++++++++++++++++++++++++++-----
 drivers/gpu/drm/xe/xe_vm_types.h |  7 ++++
 3 files changed, 70 insertions(+), 10 deletions(-)
drivers/gpu/drm/xe/xe_pt.c +1 -0
index 851ea7c01b91..bbf37e5c8abb 100644
@@ -668,6 +668,7 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
 		if (!is_null)
 			xe_res_next(curs, next - addr);
 		xe_walk->va_curs_start = next;
+		xe_walk->vma->gpuva.flags |= (XE_VMA_PTE_4K << level);
 		*action = ACTION_CONTINUE;
 
 		return ret;
drivers/gpu/drm/xe/xe_vm.c +62 -10

113 lines not on this page · raw message

drivers/gpu/drm/xe/xe_vm_types.h +7 -0
index dbe6aed6d2cf..29ec04fa1e26 100644
@@ -29,6 +29,9 @@ struct xe_vm;
 #define XE_VMA_ATOMIC_PTE_BIT	(DRM_GPUVA_USERBITS << 2)
 #define XE_VMA_FIRST_REBIND	(DRM_GPUVA_USERBITS << 3)
 #define XE_VMA_LAST_REBIND	(DRM_GPUVA_USERBITS << 4)
+#define XE_VMA_PTE_4K		(DRM_GPUVA_USERBITS << 5)
+#define XE_VMA_PTE_2M		(DRM_GPUVA_USERBITS << 6)
+#define XE_VMA_PTE_1G		(DRM_GPUVA_USERBITS << 7)
 
 struct xe_vma {
 	/** @gpuva: Base GPUVA object */
@@ -336,6 +339,10 @@ struct xe_vma_op_remap {
 	u64 start;
 	/** @range: range of the VMA unmap */
 	u64 range;
+	/** @skip_prev: skip prev rebind */
+	bool skip_prev;
+	/** @skip_next: skip next rebind */
+	bool skip_next;
 	/** @unmap_done: unmap operation in done */
 	bool unmap_done;
 };
Signature · 1 line
2.34.1
maintainer Reviewed-by on v2 1/6
#2 lore raw

Re: [Intel-xe] [PATCH v2 1/6] drm/xe: Avoid doing rebinds

Quoted 7 lines · Matthew Brost

If we dont change page sizes we can avoid doing rebinds rather just do a partial unbind. The algorithm to determine its page size is greedy as we assume all pages in the removed VMA are the largest page used in the VMA.

v2: Don't exceed 100 lines
v3: struct xe_vma_op_unmap remove in different patch, remove XXX comment

thanks

Quoted 171 lines
---
 drivers/gpu/drm/xe/xe_pt.c       |  1 +
 drivers/gpu/drm/xe/xe_vm.c       | 72 +++++++++++++++++++++++++++-----
 drivers/gpu/drm/xe/xe_vm_types.h |  7 ++++
 3 files changed, 70 insertions(+), 10 deletions(-)
drivers/gpu/drm/xe/xe_pt.c +1 -0
index 851ea7c01b91..bbf37e5c8abb 100644
@@ -668,6 +668,7 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
 		if (!is_null)
 			xe_res_next(curs, next - addr);
 		xe_walk->va_curs_start = next;
+		xe_walk->vma->gpuva.flags |= (XE_VMA_PTE_4K << level);
 		*action = ACTION_CONTINUE;
 
 		return ret;
drivers/gpu/drm/xe/xe_vm.c +62 -10
index 87907a97a8c3..19ce6b0c6564 100644
@@ -2384,6 +2384,16 @@ static struct xe_vma *new_vma(struct xe_vm *vm, struct drm_gpuva_op_map *op,
 	return vma;
 }
 
+static u64 xe_vma_max_pte_size(struct xe_vma *vma)
+{
+	if (vma->gpuva.flags & XE_VMA_PTE_1G)
+		return SZ_1G;
+	else if (vma->gpuva.flags & XE_VMA_PTE_2M)
+		return SZ_2M;
+
+	return SZ_4K;
+}
+
 /*
  * Parse operations list and create any resources needed for the operations
  * prior to fully committing to the operations. This setup can fail.
@@ -2460,6 +2470,13 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_engine *e,
 				break;
 			}
 			case DRM_GPUVA_OP_REMAP:
+			{
+				struct xe_vma *old =
+					gpuva_to_vma(op->base.remap.unmap->va);
+
+				op->remap.start = xe_vma_start(old);
+				op->remap.range = xe_vma_size(old);
+
 				if (op->base.remap.prev) {
 					struct xe_vma *vma;
 					bool read_only =
@@ -2478,6 +2495,20 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_engine *e,
 					}
 
 					op->remap.prev = vma;
+
+					/*
+					 * Userptr creates a new SG mapping so
+					 * we must also rebind.
+					 */
+					op->remap.skip_prev = !xe_vma_is_userptr(old) &&
+						IS_ALIGNED(xe_vma_end(vma),
+							   xe_vma_max_pte_size(old));
+					if (op->remap.skip_prev) {
+						op->remap.range -=
+							xe_vma_end(vma) -
+							xe_vma_start(old);
+						op->remap.start = xe_vma_end(vma);
+					}
 				}
 
 				if (op->base.remap.next) {
@@ -2499,14 +2530,21 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_engine *e,
 					}
 
 					op->remap.next = vma;
-				}
 
-				/* XXX: Support no doing remaps */
-				op->remap.start =
-					xe_vma_start(gpuva_to_vma(op->base.remap.unmap->va));
-				op->remap.range =
-					xe_vma_size(gpuva_to_vma(op->base.remap.unmap->va));
+					/*
+					 * Userptr creates a new SG mapping so
+					 * we must also rebind.
+					 */
+					op->remap.skip_next = !xe_vma_is_userptr(old) &&
+						IS_ALIGNED(xe_vma_start(vma),
+							   xe_vma_max_pte_size(old));
+					if (op->remap.skip_next)
+						op->remap.range -=
+							xe_vma_end(old) -
+							xe_vma_start(vma);
+				}
 				break;
+			}
 			case DRM_GPUVA_OP_UNMAP:
 			case DRM_GPUVA_OP_PREFETCH:
 				/* Nothing to do */
@@ -2549,10 +2587,23 @@ static int xe_vma_op_commit(struct xe_vm *vm, struct xe_vma_op *op)
 	case DRM_GPUVA_OP_REMAP:
 		prep_vma_destroy(vm, gpuva_to_vma(op->base.remap.unmap->va),
 				 true);
-		if (op->remap.prev)
+
+		if (op->remap.prev) {
 			err |= xe_vm_insert_vma(vm, op->remap.prev);
-		if (op->remap.next)
+			if (!err && op->remap.skip_prev)
+				op->remap.prev = NULL;
+		}
+		if (op->remap.next) {
 			err |= xe_vm_insert_vma(vm, op->remap.next);
+			if (!err && op->remap.skip_next)
+				op->remap.next = NULL;
+		}
+
+		/* Adjust for partial unbind after removin VMA from VM */
+		if (!err) {
+			op->base.remap.unmap->va->va.addr = op->remap.start;
+			op->base.remap.unmap->va->va.range = op->remap.range;
+		}
 		break;
 	case DRM_GPUVA_OP_UNMAP:
 		prep_vma_destroy(vm, gpuva_to_vma(op->base.unmap.va), true);
@@ -2622,9 +2673,10 @@ static int __xe_vma_op_execute(struct xe_vm *vm, struct xe_vma *vma,
 		bool next = !!op->remap.next;
 
 		if (!op->remap.unmap_done) {
-			vm->async_ops.munmap_rebind_inflight = true;
-			if (prev || next)
+			if (prev || next) {
+				vm->async_ops.munmap_rebind_inflight = true;
 				vma->gpuva.flags |= XE_VMA_FIRST_REBIND;
+			}
 			err = xe_vm_unbind(vm, vma, op->engine, op->syncs,
 					   op->num_syncs,
 					   !prev && !next ? op->fence : NULL,
drivers/gpu/drm/xe/xe_vm_types.h +7 -0
index dbe6aed6d2cf..29ec04fa1e26 100644
@@ -29,6 +29,9 @@ struct xe_vm;
 #define XE_VMA_ATOMIC_PTE_BIT	(DRM_GPUVA_USERBITS << 2)
 #define XE_VMA_FIRST_REBIND	(DRM_GPUVA_USERBITS << 3)
 #define XE_VMA_LAST_REBIND	(DRM_GPUVA_USERBITS << 4)
+#define XE_VMA_PTE_4K		(DRM_GPUVA_USERBITS << 5)
+#define XE_VMA_PTE_2M		(DRM_GPUVA_USERBITS << 6)
+#define XE_VMA_PTE_1G		(DRM_GPUVA_USERBITS << 7)
 
 struct xe_vma {
 	/** @gpuva: Base GPUVA object */
@@ -336,6 +339,10 @@ struct xe_vma_op_remap {
 	u64 start;
 	/** @range: range of the VMA unmap */
 	u64 range;
+	/** @skip_prev: skip prev rebind */
+	bool skip_prev;
+	/** @skip_next: skip next rebind */
+	bool skip_next;
 	/** @unmap_done: unmap operation in done */
 	bool unmap_done;
 };
Signature · 1 line
2.34.1
author maintainer
#3 lore raw

PATCH v2 2/6 drm/xe: Reduce the number list links in xe_vma

In reply to Matthew Brost

Combine the userptr, rebind, and destroy links into a union as the lists these links belong to are mutually exclusive.

v2: Adjust which lists are combined (Thomas H)
v3: Add kernel doc why this is safe (Thomas H), remove related change of list_del_init -> list_del (Rodrigo)

---
 drivers/gpu/drm/xe/xe_bo.c       |  6 +++--
 drivers/gpu/drm/xe/xe_exec.c     |  2 +-
 drivers/gpu/drm/xe/xe_pt.c       |  2 +-
 drivers/gpu/drm/xe/xe_vm.c       | 43 ++++++++++++++++----------------
 drivers/gpu/drm/xe/xe_vm_types.h | 37 ++++++++++++++++-----------
 5 files changed, 49 insertions(+), 41 deletions(-)
drivers/gpu/drm/xe/xe_bo.c +4 -2
index 828a9fea420b..538501c46b8b 100644
@@ -474,8 +474,10 @@ static int xe_bo_trigger_rebind(struct xe_device *xe, struct xe_bo *bo,
 			}
 
 			xe_vm_assert_held(vm);
-			if (list_empty(&vma->rebind_link) && vma->tile_present)
-				list_add_tail(&vma->rebind_link, &vm->rebind_list);
+			if (list_empty(&vma->combined_links.rebind) &&
+			    vma->tile_present)
+				list_add_tail(&vma->combined_links.rebind,
+					      &vm->rebind_list);
 
 			if (vm_resv_locked)
 				dma_resv_unlock(&vm->resv);
drivers/gpu/drm/xe/xe_exec.c +1 -1
index 1f64d64a94fa..fff4a9d9d12a 100644
@@ -120,7 +120,7 @@ static int xe_exec_begin(struct xe_engine *e, struct ww_acquire_ctx *ww,
 	 * BOs have valid placements possibly moving an evicted BO back
 	 * to a location where the GPU can access it).
 	 */
-	list_for_each_entry(vma, &vm->rebind_list, rebind_link) {
+	list_for_each_entry(vma, &vm->rebind_list, combined_links.rebind) {
 		XE_WARN_ON(xe_vma_is_null(vma));
 
 		if (xe_vma_is_userptr(vma))
drivers/gpu/drm/xe/xe_pt.c +1 -1
index bbf37e5c8abb..303313adf3ef 100644
@@ -1690,7 +1690,7 @@ __xe_pt_unbind_vma(struct xe_tile *tile, struct xe_vma *vma, struct xe_engine *e
 	}
 
 	if (!vma->tile_present)
-		list_del_init(&vma->rebind_link);
+		list_del_init(&vma->combined_links.rebind);
 
 	if (unbind_pt_update.locked) {
 		XE_WARN_ON(!xe_vma_is_userptr(vma));
drivers/gpu/drm/xe/xe_vm.c +21 -22

123 lines not on this page · raw message

drivers/gpu/drm/xe/xe_vm_types.h +22 -15
index 29ec04fa1e26..f1f3b619d996 100644
@@ -49,21 +49,28 @@ struct xe_vma {
 	 */
 	u64 tile_present;
 
-	/** @userptr_link: link into VM repin list if userptr */
-	struct list_head userptr_link;
-
-	/**
-	 * @rebind_link: link into VM if this VMA needs rebinding, and
-	 * if it's a bo (not userptr) needs validation after a possible
-	 * eviction. Protected by the vm's resv lock.
-	 */
-	struct list_head rebind_link;
-
-	/**
-	 * @unbind_link: link or list head if an unbind of multiple VMAs, in
-	 * single unbind op, is being done.
-	 */
-	struct list_head unbind_link;
+	/** @combined_links: links into lists which are mutually exclusive */
+	union {
+		/**
+		 * @userptr: link into VM repin list if userptr. Protected by
+		 * vm->lock in write mode.
+		 */
+		struct list_head userptr;
+		/**
+		 * @rebind: link into VM if this VMA needs rebinding, and
+		 * if it's a bo (not userptr) needs validation after a possible
+		 * eviction. Protected by the vm's resv lock and typically
+		 * vm->lock is also held in write mode. The only place where
+		 * vm->lock isn't held is the BO eviction path which has
+		 * mutually exclusive execution with userptr.
+		 */
+		struct list_head rebind;
+		/**
+		 * @destroy: link to contested list when VM is being closed.
+		 * Protected by vm->lock in write mode and vm's resv lock.
+		 */
+		struct list_head destroy;
+	} combined_links;
 
 	/** @destroy_cb: callback to destroy VMA when unbind job is done */
 	struct dma_fence_cb destroy_cb;
Signature · 1 line
2.34.1
author maintainer
#4 lore raw

PATCH v2 3/6 drm/xe: Replace list_del_init with list_del for userptr.invalidate_link cleanup

In reply to Matthew Brost

This list isn't used again, list_del is the proper call.

---
 drivers/gpu/drm/xe/xe_vm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/xe/xe_vm.c +1 -1
index 87faf4b723b3..1ed3bc2541f2 100644
@@ -1065,7 +1065,7 @@ static void xe_vma_destroy(struct xe_vma *vma, struct dma_fence *fence)
 		XE_WARN_ON(!(vma->gpuva.flags & XE_VMA_DESTROYED));
 
 		spin_lock(&vm->userptr.invalidated_lock);
-		list_del_init(&vma->userptr.invalidate_link);
+		list_del(&vma->userptr.invalidate_link);
 		spin_unlock(&vm->userptr.invalidated_lock);
 	} else if (!xe_vma_is_null(vma)) {
 		xe_bo_assert_held(xe_vma_bo(vma));
Signature · 1 line
2.34.1
maintainer Reviewed-by on v2 3/6
#5 lore raw

Re: [Intel-xe] [PATCH v2 3/6] drm/xe: Replace list_del_init with list_del for userptr.invalidate_link cleanup

Quoted 3 lines · Matthew Brost

This list isn't used again, list_del is the proper call.

Quoted 20 lines
---
 drivers/gpu/drm/xe/xe_vm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/xe/xe_vm.c +1 -1
index 87faf4b723b3..1ed3bc2541f2 100644
@@ -1065,7 +1065,7 @@ static void xe_vma_destroy(struct xe_vma *vma, struct dma_fence *fence)
 		XE_WARN_ON(!(vma->gpuva.flags & XE_VMA_DESTROYED));
 
 		spin_lock(&vm->userptr.invalidated_lock);
-		list_del_init(&vma->userptr.invalidate_link);
+		list_del(&vma->userptr.invalidate_link);
 		spin_unlock(&vm->userptr.invalidated_lock);
 	} else if (!xe_vma_is_null(vma)) {
 		xe_bo_assert_held(xe_vma_bo(vma));
Signature · 1 line
2.34.1
author maintainer
#6 lore raw

PATCH v2 4/6 drm/xe: Change tile masks from u64 to u8

In reply to Matthew Brost

This will save us a few bytes in the xe_vma structure.

v2: Use hweight8 rather than hweight_long (Rodrigo)

---
 drivers/gpu/drm/xe/xe_vm.c       | 12 ++++++------
 drivers/gpu/drm/xe/xe_vm_types.h | 28 ++++++++++++++--------------
 2 files changed, 20 insertions(+), 20 deletions(-)
drivers/gpu/drm/xe/xe_vm.c +6 -6
index 1ed3bc2541f2..92d97d347a44 100644
@@ -871,7 +871,7 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm,
 				    u64 start, u64 end,
 				    bool read_only,
 				    bool is_null,
-				    u64 tile_mask)
+				    u8 tile_mask)
 {
 	struct xe_vma *vma;
 	struct xe_tile *tile;
@@ -1574,7 +1574,7 @@ xe_vm_unbind_vma(struct xe_vma *vma, struct xe_engine *e,
 	struct dma_fence_array *cf = NULL;
 	struct xe_vm *vm = xe_vma_vm(vma);
 	int cur_fence = 0, i;
-	int number_tiles = hweight_long(vma->tile_present);
+	int number_tiles = hweight8(vma->tile_present);
 	int err;
 	u8 id;
 
@@ -1649,7 +1649,7 @@ xe_vm_bind_vma(struct xe_vma *vma, struct xe_engine *e,
 	struct dma_fence_array *cf = NULL;
 	struct xe_vm *vm = xe_vma_vm(vma);
 	int cur_fence = 0, i;
-	int number_tiles = hweight_long(vma->tile_mask);
+	int number_tiles = hweight8(vma->tile_mask);
 	int err;
 	u8 id;
 
@@ -2245,7 +2245,7 @@ static void print_op(struct xe_device *xe, struct drm_gpuva_op *op)
 static struct drm_gpuva_ops *
 vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo,
 			 u64 bo_offset_or_userptr, u64 addr, u64 range,
-			 u32 operation, u64 tile_mask, u32 region)
+			 u32 operation, u8 tile_mask, u32 region)
 {
 	struct drm_gem_object *obj = bo ? &bo->ttm.base : NULL;
 	struct ww_acquire_ctx ww;
@@ -2342,7 +2342,7 @@ vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo,
 }
 
 static struct xe_vma *new_vma(struct xe_vm *vm, struct drm_gpuva_op_map *op,
-			      u64 tile_mask, bool read_only, bool is_null)
+			      u8 tile_mask, bool read_only, bool is_null)
 {
 	struct xe_bo *bo = op->gem.obj ? gem_to_xe_bo(op->gem.obj) : NULL;
 	struct xe_vma *vma;
@@ -3327,7 +3327,7 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 		u64 addr = bind_ops[i].addr;
 		u32 op = bind_ops[i].op;
 		u64 obj_offset = bind_ops[i].obj_offset;
-		u64 tile_mask = bind_ops[i].tile_mask;
+		u8 tile_mask = bind_ops[i].tile_mask;
 		u32 region = bind_ops[i].region;
 
 		ops[i] = vm_bind_ioctl_ops_create(vm, bos[i], obj_offset,
drivers/gpu/drm/xe/xe_vm_types.h +14 -14
index f1f3b619d996..268d2475f2ae 100644
@@ -37,18 +37,6 @@ struct xe_vma {
 	/** @gpuva: Base GPUVA object */
 	struct drm_gpuva gpuva;
 
-	/** @tile_mask: Tile mask of where to create binding for this VMA */
-	u64 tile_mask;
-
-	/**
-	 * @tile_present: GT 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,
-	 * but if reading is done under the vm->lock only, it needs to be held
-	 * in write mode.
-	 */
-	u64 tile_present;
-
 	/** @combined_links: links into lists which are mutually exclusive */
 	union {
 		/**
@@ -106,9 +94,21 @@ struct xe_vma {
 	/** @usm: unified shared memory state */
 	struct {
 		/** @tile_invalidated: VMA has been invalidated */
-		u64 tile_invalidated;
+		u8 tile_invalidated;
 	} usm;
 
+	/** @tile_mask: Tile mask of where to create binding for this VMA */
+	u8 tile_mask;
+
+	/**
+	 * @tile_present: GT 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,
+	 * but if reading is done under the vm->lock only, it needs to be held
+	 * in write mode.
+	 */
+	u8 tile_present;
+
 	struct {
 		struct list_head rebind_link;
 	} notifier;
@@ -395,7 +395,7 @@ struct xe_vma_op {
 	 */
 	struct async_op_fence *fence;
 	/** @tile_mask: gt mask for this operation */
-	u64 tile_mask;
+	u8 tile_mask;
 	/** @flags: operation flags */
 	enum xe_vma_op_flags flags;
 
Signature · 1 line
2.34.1
maintainer Reviewed-by on v2 4/6
#7 lore raw

Re: [Intel-xe] [PATCH v2 4/6] drm/xe: Change tile masks from u64 to u8

Quoted 5 lines · Matthew Brost

This will save us a few bytes in the xe_vma structure.

v2: Use hweight8 rather than hweight_long (Rodrigo)

Quoted 121 lines
---
 drivers/gpu/drm/xe/xe_vm.c       | 12 ++++++------
 drivers/gpu/drm/xe/xe_vm_types.h | 28 ++++++++++++++--------------
 2 files changed, 20 insertions(+), 20 deletions(-)
drivers/gpu/drm/xe/xe_vm.c +6 -6
index 1ed3bc2541f2..92d97d347a44 100644
@@ -871,7 +871,7 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm,
 				    u64 start, u64 end,
 				    bool read_only,
 				    bool is_null,
-				    u64 tile_mask)
+				    u8 tile_mask)
 {
 	struct xe_vma *vma;
 	struct xe_tile *tile;
@@ -1574,7 +1574,7 @@ xe_vm_unbind_vma(struct xe_vma *vma, struct xe_engine *e,
 	struct dma_fence_array *cf = NULL;
 	struct xe_vm *vm = xe_vma_vm(vma);
 	int cur_fence = 0, i;
-	int number_tiles = hweight_long(vma->tile_present);
+	int number_tiles = hweight8(vma->tile_present);
 	int err;
 	u8 id;
 
@@ -1649,7 +1649,7 @@ xe_vm_bind_vma(struct xe_vma *vma, struct xe_engine *e,
 	struct dma_fence_array *cf = NULL;
 	struct xe_vm *vm = xe_vma_vm(vma);
 	int cur_fence = 0, i;
-	int number_tiles = hweight_long(vma->tile_mask);
+	int number_tiles = hweight8(vma->tile_mask);
 	int err;
 	u8 id;
 
@@ -2245,7 +2245,7 @@ static void print_op(struct xe_device *xe, struct drm_gpuva_op *op)
 static struct drm_gpuva_ops *
 vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo,
 			 u64 bo_offset_or_userptr, u64 addr, u64 range,
-			 u32 operation, u64 tile_mask, u32 region)
+			 u32 operation, u8 tile_mask, u32 region)
 {
 	struct drm_gem_object *obj = bo ? &bo->ttm.base : NULL;
 	struct ww_acquire_ctx ww;
@@ -2342,7 +2342,7 @@ vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo,
 }
 
 static struct xe_vma *new_vma(struct xe_vm *vm, struct drm_gpuva_op_map *op,
-			      u64 tile_mask, bool read_only, bool is_null)
+			      u8 tile_mask, bool read_only, bool is_null)
 {
 	struct xe_bo *bo = op->gem.obj ? gem_to_xe_bo(op->gem.obj) : NULL;
 	struct xe_vma *vma;
@@ -3327,7 +3327,7 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 		u64 addr = bind_ops[i].addr;
 		u32 op = bind_ops[i].op;
 		u64 obj_offset = bind_ops[i].obj_offset;
-		u64 tile_mask = bind_ops[i].tile_mask;
+		u8 tile_mask = bind_ops[i].tile_mask;
 		u32 region = bind_ops[i].region;
 
 		ops[i] = vm_bind_ioctl_ops_create(vm, bos[i], obj_offset,
drivers/gpu/drm/xe/xe_vm_types.h +14 -14
index f1f3b619d996..268d2475f2ae 100644
@@ -37,18 +37,6 @@ struct xe_vma {
 	/** @gpuva: Base GPUVA object */
 	struct drm_gpuva gpuva;
 
-	/** @tile_mask: Tile mask of where to create binding for this VMA */
-	u64 tile_mask;
-
-	/**
-	 * @tile_present: GT 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,
-	 * but if reading is done under the vm->lock only, it needs to be held
-	 * in write mode.
-	 */
-	u64 tile_present;
-
 	/** @combined_links: links into lists which are mutually exclusive */
 	union {
 		/**
@@ -106,9 +94,21 @@ struct xe_vma {
 	/** @usm: unified shared memory state */
 	struct {
 		/** @tile_invalidated: VMA has been invalidated */
-		u64 tile_invalidated;
+		u8 tile_invalidated;
 	} usm;
 
+	/** @tile_mask: Tile mask of where to create binding for this VMA */
+	u8 tile_mask;
+
+	/**
+	 * @tile_present: GT 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,
+	 * but if reading is done under the vm->lock only, it needs to be held
+	 * in write mode.
+	 */
+	u8 tile_present;
+
 	struct {
 		struct list_head rebind_link;
 	} notifier;
@@ -395,7 +395,7 @@ struct xe_vma_op {
 	 */
 	struct async_op_fence *fence;
 	/** @tile_mask: gt mask for this operation */
-	u64 tile_mask;
+	u8 tile_mask;
 	/** @flags: operation flags */
 	enum xe_vma_op_flags flags;
 
Signature · 1 line
2.34.1
author maintainer
#8 lore raw

PATCH v2 5/6 drm/xe: Combine destroy_cb and destroy_work in xe_vma into union

In reply to Matthew Brost

The callback kicks the worker thus mutually exclusive execution, combining saves a bit of space in xe_vma.

---
 drivers/gpu/drm/xe/xe_vm_types.h | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
drivers/gpu/drm/xe/xe_vm_types.h +6 -5
index 268d2475f2ae..809cfe734d43 100644
@@ -60,11 +60,12 @@ struct xe_vma {
 		struct list_head destroy;
 	} combined_links;
 
-	/** @destroy_cb: callback to destroy VMA when unbind job is done */
-	struct dma_fence_cb destroy_cb;
-
-	/** @destroy_work: worker to destroy this BO */
-	struct work_struct destroy_work;
+	union {
+		/** @destroy_cb: callback to destroy VMA when unbind job is done */
+		struct dma_fence_cb destroy_cb;
+		/** @destroy_work: worker to destroy this BO */
+		struct work_struct destroy_work;
+	};
 
 	/** @userptr: user pointer state */
 	struct {
Signature · 1 line
2.34.1
author maintainer
#9 lore raw

PATCH v2 6/6 drm/xe: Only alloc userptr part of xe_vma for userptrs

In reply to Matthew Brost

Only alloc userptr part of xe_vma for userptrs, this will save on space in the common BO case.

---
 drivers/gpu/drm/xe/xe_vm.c       |  6 +++-
 drivers/gpu/drm/xe/xe_vm_types.h | 56 ++++++++++++++++++--------------
 2 files changed, 36 insertions(+), 26 deletions(-)
drivers/gpu/drm/xe/xe_vm.c +5 -1
index 92d97d347a44..766830d21e01 100644
@@ -880,7 +880,11 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm,
 	XE_BUG_ON(start >= end);
 	XE_BUG_ON(end >= vm->size);
 
-	vma = kzalloc(sizeof(*vma), GFP_KERNEL);
+	if (!bo && !is_null)	/* userptr */
+		vma = kzalloc(sizeof(*vma), GFP_KERNEL);
+	else
+		vma = kzalloc(sizeof(*vma) - sizeof(struct xe_userptr),
+			      GFP_KERNEL);
 	if (!vma) {
 		vma = ERR_PTR(-ENOMEM);
 		return vma;
drivers/gpu/drm/xe/xe_vm_types.h +31 -25

74 lines not on this page · raw message

Signature · 1 line
2.34.1
#10 lore raw

[Intel-xe] ✗ CI.Patch_applied: failure for More VM cleanups (rev2)

In reply to Matthew Brost
== Series Details ==

== Series Details ==

Series: More VM cleanups (rev2)
URL : https://patchwork.freedesktop.org/series/120574/
State : failure

== Summary ==

=== Applying kernel patches on branch 'drm-xe-next' with base: === Base commit: f5459dca4 drm/xe: Cleanup style warnings and errors === git am output follows ===
error: patch failed: drivers/gpu/drm/xe/xe_vm.c:2499
error: drivers/gpu/drm/xe/xe_vm.c: patch does not apply
hint: Use 'git am --show-current-patch' to see the failed patch
Applying: drm/xe: Avoid doing rebinds
Patch failed at 0001 drm/xe: Avoid doing rebinds
When you have resolved this problem, run "git am --continue". If you prefer to skip this patch, run "git am --skip" instead. To restore the original branch and stop patching, run "git am --abort".