[Intel-xe] [PATCH v4 9/9] drm/xe: Avoid doing rebinds

[Matthew Brost <matthew.brost@intel.com>]

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

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
---
 drivers/gpu/drm/xe/xe_pt.c       |  1 +
 drivers/gpu/drm/xe/xe_vm.c       | 73 +++++++++++++++++++++++++-------
 drivers/gpu/drm/xe/xe_vm_types.h | 17 +++-----
 3 files changed, 66 insertions(+), 25 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
index 00855681c0d5..a8d96cbd53e3 100644
--- a/drivers/gpu/drm/xe/xe_pt.c
+++ b/drivers/gpu/drm/xe/xe_pt.c
@@ -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;
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index 1da61beb1765..4359ce858fb1 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -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 +2495,21 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_engine *e,
 					}
 
 					op->remap.prev = vma;
+
+					/*
+					 * XXX: Not sure why userptr doesn't
+					 * work but really shouldn't be a use
+					 * case.
+					 */
+					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) {
@@ -2499,20 +2531,17 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_engine *e,
 					}
 
 					op->remap.next = vma;
+					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);
 				}
-
-				/* 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));
 				break;
+			}
 			case DRM_GPUVA_OP_UNMAP:
-				op->unmap.start =
-					xe_vma_start(gpuva_to_vma(op->base.unmap.va));
-				op->unmap.range =
-					xe_vma_size(gpuva_to_vma(op->base.unmap.va));
-				break;
 			case DRM_GPUVA_OP_PREFETCH:
 				/* Nothing to do */
 				break;
@@ -2554,10 +2583,23 @@ 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);
@@ -2627,9 +2669,10 @@ 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,
diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
index cad806bafbfc..b517b711929c 100644
--- a/drivers/gpu/drm/xe/xe_vm_types.h
+++ b/drivers/gpu/drm/xe/xe_vm_types.h
@@ -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 */
@@ -323,14 +326,6 @@ struct xe_vma_op_map {
 	bool is_null;
 };
 
-/** struct xe_vma_op_unmap - VMA unmap operation */
-struct xe_vma_op_unmap {
-	/** @start: start of the VMA unmap */
-	u64 start;
-	/** @range: range of the VMA unmap */
-	u64 range;
-};
-
 /** struct xe_vma_op_remap - VMA remap operation */
 struct xe_vma_op_remap {
 	/** @prev: VMA preceding part of a split mapping */
@@ -341,6 +336,10 @@ 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;
 };
@@ -398,8 +397,6 @@ struct xe_vma_op {
 	union {
 		/** @map: VMA map operation specific data */
 		struct xe_vma_op_map map;
-		/** @unmap: VMA unmap operation specific data */
-		struct xe_vma_op_unmap unmap;
 		/** @remap: VMA remap operation specific data */
 		struct xe_vma_op_remap remap;
 		/** @prefetch: VMA prefetch operation specific data */
-- 
2.34.1