[RFC v2 3/7] drm/xe/svm: Clear CPU_AUTORESET_ACTIVE on first GPU fault

[Arvind Yadav <arvind.yadav@intel.com>]

CPU address mirror VMAs start with cpu_autoreset_active set, indicating
they are still CPU-only.

Clear cpu_autoreset_active after the first successful GPU fault so
subsequent munmap follows the SVM path instead of the autoreset path.

Do this in xe_svm_handle_pagefault() on the success path only.
Prefetch faults that install no PTEs must not transition this state.

v2:
  - Move xe_vma_gpu_touch() to the success path in
    xe_svm_handle_pagefault() so prefetch faults that find no range do
    not transition the state. (Matt)
  - Add xe_vma_gpu_touch() helper in xe_vm.h and use
    vma->cpu_autoreset_active instead of vma->gpuva.flags. (Matt)

Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
---
 drivers/gpu/drm/xe/xe_svm.c | 11 +++++++++++
 drivers/gpu/drm/xe/xe_vm.h  | 13 +++++++++++++
 2 files changed, 24 insertions(+)

diff --git a/drivers/gpu/drm/xe/xe_svm.c b/drivers/gpu/drm/xe/xe_svm.c
index 5933b2b6392b..fd57c9d41db8 100644
--- a/drivers/gpu/drm/xe/xe_svm.c
+++ b/drivers/gpu/drm/xe/xe_svm.c
@@ -1388,6 +1388,9 @@ int xe_svm_handle_pagefault(struct xe_vm *vm, struct xe_vma *vma,
 			    bool atomic)
 {
 	int need_vram, ret;
+
+	lockdep_assert_held_write(&vm->lock);
+
 retry:
 	need_vram = xe_vma_need_vram_for_atomic(vm->xe, vma, atomic);
 	if (need_vram < 0)
@@ -1406,6 +1409,14 @@ int xe_svm_handle_pagefault(struct xe_vm *vm, struct xe_vma *vma,
 
 		goto retry;
 	}
+
+	/*
+	 * Mark the VMA as GPU-touched only after a successful fault-in.
+	 * Prefetch faults that find no range must not transition this state.
+	 */
+	if (!ret && xe_vma_has_cpu_autoreset_active(vma))
+		xe_vma_gpu_touch(vma);
+
 	return ret;
 }
 
diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h
index bdf42083da86..8d45f896f90b 100644
--- a/drivers/gpu/drm/xe/xe_vm.h
+++ b/drivers/gpu/drm/xe/xe_vm.h
@@ -441,4 +441,17 @@ static inline struct drm_exec *xe_vm_validation_exec(struct xe_vm *vm)
 	((READ_ONCE(tile_present) & ~READ_ONCE(tile_invalidated)) & BIT((tile)->id))
 
 void xe_vma_mem_attr_copy(struct xe_vma_mem_attr *to, struct xe_vma_mem_attr *from);
+
+/**
+ * xe_vma_gpu_touch() - Mark a VMA as no longer CPU-only
+ * @vma: VMA to update
+ *
+ * Clear cpu_autoreset_active after the first successful GPU fault-in.
+ * Caller must hold vm->lock in write mode.
+ */
+static inline void xe_vma_gpu_touch(struct xe_vma *vma)
+{
+	lockdep_assert_held_write(&xe_vma_vm(vma)->lock);
+	vma->cpu_autoreset_active = false;
+}
 #endif
-- 
2.43.0