[PATCH AUTOSEL 5.15 58/68] drm/amdkfd: handle VMA remove race

[Sasha Levin]

From: Philip Yang <Philip.Yang at amd.com>

[ Upstream commit 0cc53cb450669cf1def4ff89e8cbcd8ec3c62380 ]

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 <Philip.Yang at amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling at amd.com>
Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
Signed-off-by: Sasha Levin <sashal at kernel.org>
---
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 22 +++++++++++++---------
 1 file changed, 13 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index 179080329af89..4e933fb0fc698 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -2448,20 +2448,13 @@ 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;
@@ -2479,6 +2472,7 @@ 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)) {
@@ -2552,7 +2546,17 @@ 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;
-- 
2.33.0