drm/xe: Convert pagefaulting code to use drm_exec

Replace the calls into ttm_eu_reserve_buffers with the drm_exec helpers.
Also reuse some code.

v4:
- Kerneldoc xe_vm_prepare_vma().
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230908091716.36984-6-thomas.hellstrom@linux.intel.comSigned-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drivers/gpu/drm/xe/xe_gt_pagefault.c

@@ -8,6 +8,7 @@
 #include <linux/bitfield.h>
 #include <linux/circ_buf.h>
 
+#include <drm/drm_exec.h>
 #include <drm/drm_managed.h>
 #include <drm/ttm/ttm_execbuf_util.h>
 
@@ -84,11 +85,6 @@ static bool vma_matches(struct xe_vma *vma, u64 page_addr)
 	return true;
 }
 
-static bool only_needs_bo_lock(struct xe_bo *bo)
-{
-	return bo && bo->vm;
-}
-
 static struct xe_vma *lookup_vma(struct xe_vm *vm, u64 page_addr)
 {
 	struct xe_vma *vma = NULL;
@@ -103,17 +99,45 @@ static struct xe_vma *lookup_vma(struct xe_vm *vm, u64 page_addr)
 	return vma;
 }
 
+static int xe_pf_begin(struct drm_exec *exec, struct xe_vma *vma,
+		       bool atomic, unsigned int id)
+{
+	struct xe_bo *bo = xe_vma_bo(vma);
+	struct xe_vm *vm = xe_vma_vm(vma);
+	unsigned int num_shared = 2; /* slots for bind + move */
+	int err;
+
+	err = xe_vm_prepare_vma(exec, vma, num_shared);
+	if (err)
+		return err;
+
+	if (atomic) {
+		if (xe_vma_is_userptr(vma)) {
+			err = -EACCES;
+			return err;
+		}
+
+		/* Migrate to VRAM, move should invalidate the VMA first */
+		err = xe_bo_migrate(bo, XE_PL_VRAM0 + id);
+		if (err)
+			return err;
+	} else if (bo) {
+		/* Create backing store if needed */
+		err = xe_bo_validate(bo, vm, true);
+		if (err)
+			return err;
+	}
+
+	return 0;
+}
+
 static int handle_pagefault(struct xe_gt *gt, struct pagefault *pf)
 {
 	struct xe_device *xe = gt_to_xe(gt);
 	struct xe_tile *tile = gt_to_tile(gt);
+	struct drm_exec exec;
 	struct xe_vm *vm;
 	struct xe_vma *vma = NULL;
-	struct xe_bo *bo;
-	LIST_HEAD(objs);
-	LIST_HEAD(dups);
-	struct ttm_validate_buffer tv_bo, tv_vm;
-	struct ww_acquire_ctx ww;
 	struct dma_fence *fence;
 	bool write_locked;
 	int ret = 0;
@@ -170,35 +194,10 @@ static int handle_pagefault(struct xe_gt *gt, struct pagefault *pf)
 	}
 
 	/* Lock VM and BOs dma-resv */
-	bo = xe_vma_bo(vma);
-	if (!only_needs_bo_lock(bo)) {
-		tv_vm.num_shared = xe->info.tile_count;
-		tv_vm.bo = xe_vm_ttm_bo(vm);
-		list_add(&tv_vm.head, &objs);
-	}
-	if (bo) {
-		tv_bo.bo = &bo->ttm;
-		tv_bo.num_shared = xe->info.tile_count;
-		list_add(&tv_bo.head, &objs);
-	}
-
-	ret = ttm_eu_reserve_buffers(&ww, &objs, false, &dups);
-	if (ret)
-		goto unlock_vm;
-
-	if (atomic) {
-		if (xe_vma_is_userptr(vma)) {
-			ret = -EACCES;
-			goto unlock_dma_resv;
-		}
-
-		/* Migrate to VRAM, move should invalidate the VMA first */
-		ret = xe_bo_migrate(bo, XE_PL_VRAM0 + tile->id);
-		if (ret)
-			goto unlock_dma_resv;
-	} else if (bo) {
-		/* Create backing store if needed */
-		ret = xe_bo_validate(bo, vm, true);
+	drm_exec_init(&exec, 0);
+	drm_exec_until_all_locked(&exec) {
+		ret = xe_pf_begin(&exec, vma, atomic, tile->id);
+		drm_exec_retry_on_contention(&exec);
 		if (ret)
 			goto unlock_dma_resv;
 	}
@@ -225,7 +224,7 @@ static int handle_pagefault(struct xe_gt *gt, struct pagefault *pf)
 	vma->usm.tile_invalidated &= ~BIT(tile->id);
 
 unlock_dma_resv:
-	ttm_eu_backoff_reservation(&ww, &objs);
+	drm_exec_fini(&exec);
 unlock_vm:
 	if (!ret)
 		vm->usm.last_fault_vma = vma;
@@ -490,13 +489,9 @@ static int handle_acc(struct xe_gt *gt, struct acc *acc)
 {
 	struct xe_device *xe = gt_to_xe(gt);
 	struct xe_tile *tile = gt_to_tile(gt);
+	struct drm_exec exec;
 	struct xe_vm *vm;
 	struct xe_vma *vma;
-	struct xe_bo *bo;
-	LIST_HEAD(objs);
-	LIST_HEAD(dups);
-	struct ttm_validate_buffer tv_bo, tv_vm;
-	struct ww_acquire_ctx ww;
 	int ret = 0;
 
 	/* We only support ACC_TRIGGER at the moment */
@@ -528,23 +523,15 @@ static int handle_acc(struct xe_gt *gt, struct acc *acc)
 		goto unlock_vm;
 
 	/* Lock VM and BOs dma-resv */
-	bo = xe_vma_bo(vma);
-	if (!only_needs_bo_lock(bo)) {
-		tv_vm.num_shared = xe->info.tile_count;
-		tv_vm.bo = xe_vm_ttm_bo(vm);
-		list_add(&tv_vm.head, &objs);
+	drm_exec_init(&exec, 0);
+	drm_exec_until_all_locked(&exec) {
+		ret = xe_pf_begin(&exec, vma, true, tile->id);
+		drm_exec_retry_on_contention(&exec);
+		if (ret)
+			break;
 	}
-	tv_bo.bo = &bo->ttm;
-	tv_bo.num_shared = xe->info.tile_count;
-	list_add(&tv_bo.head, &objs);
-	ret = ttm_eu_reserve_buffers(&ww, &objs, false, &dups);
-	if (ret)
-		goto unlock_vm;
-
-	/* Migrate to VRAM, move should invalidate the VMA first */
-	ret = xe_bo_migrate(bo, XE_PL_VRAM0 + tile->id);
 
-	ttm_eu_backoff_reservation(&ww, &objs);
+	drm_exec_fini(&exec);
 unlock_vm:
 	up_read(&vm->lock);
 	xe_vm_put(vm);

drivers/gpu/drm/xe/xe_vm.c

@@ -35,6 +35,11 @@
 
 #define TEST_VM_ASYNC_OPS_ERROR
 
+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
  * @vma: The userptr vma
@@ -422,7 +427,7 @@ int xe_vm_lock_dma_resv(struct xe_vm *vm, struct drm_exec *exec,
 	lockdep_assert_held(&vm->lock);
 
 	if (lock_vm) {
-		err = drm_exec_prepare_obj(exec, &xe_vm_ttm_bo(vm)->base, num_shared);
+		err = drm_exec_prepare_obj(exec, xe_vm_obj(vm), num_shared);
 		if (err)
 			return err;
 	}
@@ -514,7 +519,7 @@ static int xe_preempt_work_begin(struct drm_exec *exec, struct xe_vm *vm,
 	struct xe_vma *vma;
 	int err;
 
-	err = drm_exec_prepare_obj(exec, &xe_vm_ttm_bo(vm)->base,
+	err = drm_exec_prepare_obj(exec, xe_vm_obj(vm),
 				   vm->preempt.num_exec_queues);
 	if (err)
 		return err;
@@ -1095,6 +1100,33 @@ static void xe_vma_destroy(struct xe_vma *vma, struct dma_fence *fence)
 	}
 }
 
+/**
+ * xe_vm_prepare_vma() - drm_exec utility to lock a vma
+ * @exec: The drm_exec object we're currently locking for.
+ * @vma: The vma for witch we want to lock the vm resv and any attached
+ * object's resv.
+ * @num_shared: The number of dma-fence slots to pre-allocate in the
+ * objects' reservation objects.
+ *
+ * Return: 0 on success, negative error code on error. In particular
+ * may return -EDEADLK on WW transaction contention and -EINTR if
+ * an interruptible wait is terminated by a signal.
+ */
+int xe_vm_prepare_vma(struct drm_exec *exec, struct xe_vma *vma,
+		      unsigned int num_shared)
+{
+	struct xe_vm *vm = xe_vma_vm(vma);
+	struct xe_bo *bo = xe_vma_bo(vma);
+	int err;
+
+	XE_WARN_ON(!vm);
+	err = drm_exec_prepare_obj(exec, xe_vm_obj(vm), num_shared);
+	if (!err && bo && !bo->vm)
+		err = drm_exec_prepare_obj(exec, &bo->ttm.base, num_shared);
+
+	return err;
+}
+
 static void xe_vma_destroy_unlocked(struct xe_vma *vma)
 {
 	struct ttm_validate_buffer tv[2];

drivers/gpu/drm/xe/xe_vm.h

@@ -219,6 +219,9 @@ void xe_vm_fence_all_extobjs(struct xe_vm *vm, struct dma_fence *fence,
 
 int xe_analyze_vm(struct drm_printer *p, struct xe_vm *vm, int gt_id);
 
+int xe_vm_prepare_vma(struct drm_exec *exec, struct xe_vma *vma,
+		      unsigned int num_shared);
+
 /**
  * xe_vm_resv() - Return's the vm's reservation object
  * @vm: The vm