drm/amdkfd: handle VMA remove race

VMA may be removed before unmap notifier callback, and deferred list
work remove range, return success for this special case as we are
handling stale retry fault.

Signed-off-by: Philip Yang <[email protected]>
Reviewed-by: Felix Kuehling <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>

Author: PhilipYangA

drivers/gpu/drm/amd/amdkfd/kfd_svm.c

@@ -2548,20 +2548,13 @@ svm_range_count_fault(struct amdgpu_device *adev, struct kfd_process *p,
 }
 
 static bool
-svm_fault_allowed(struct mm_struct *mm, uint64_t addr, bool write_fault)
+svm_fault_allowed(struct vm_area_struct *vma, bool write_fault)
 {
 	unsigned long requested = VM_READ;
-	struct vm_area_struct *vma;
 
 	if (write_fault)
 		requested |= VM_WRITE;
 
-	vma = find_vma(mm, addr << PAGE_SHIFT);
-	if (!vma || (addr << PAGE_SHIFT) < vma->vm_start) {
-		pr_debug("address 0x%llx VMA is removed\n", addr);
-		return true;
-	}
-
 	pr_debug("requested 0x%lx, vma permission flags 0x%lx\n", requested,
 		vma->vm_flags);
 	return (vma->vm_flags & requested) == requested;
@@ -2579,6 +2572,7 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid,
 	int32_t best_loc;
 	int32_t gpuidx = MAX_GPU_INSTANCE;
 	bool write_locked = false;
+	struct vm_area_struct *vma;
 	int r = 0;
 
 	if (!KFD_IS_SVM_API_SUPPORTED(adev->kfd.dev)) {
@@ -2654,7 +2648,17 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid,
 		goto out_unlock_range;
 	}
 
-	if (!svm_fault_allowed(mm, addr, write_fault)) {
+	/* __do_munmap removed VMA, return success as we are handling stale
+	 * retry fault.
+	 */
+	vma = find_vma(mm, addr << PAGE_SHIFT);
+	if (!vma || (addr << PAGE_SHIFT) < vma->vm_start) {
+		pr_debug("address 0x%llx VMA is removed\n", addr);
+		r = 0;
+		goto out_unlock_range;
+	}
+
+	if (!svm_fault_allowed(vma, write_fault)) {
 		pr_debug("fault addr 0x%llx no %s permission\n", addr,
 			write_fault ? "write" : "read");
 		r = -EPERM;