drm/amdgpu: use atomic functions with memory barriers for vm fault info

commit 6df8e84 upstream.

The atomic variable vm_fault_info_updated is used to synchronize access to
adev->gmc.vm_fault_info between the interrupt handler and
get_vm_fault_info().

The default atomic functions like atomic_set() and atomic_read() do not
provide memory barriers. This allows for CPU instruction reordering,
meaning the memory accesses to vm_fault_info and the vm_fault_info_updated
flag are not guaranteed to occur in the intended order. This creates a
race condition that can lead to inconsistent or stale data being used.

The previous implementation, which used an explicit mb(), was incomplete
and inefficient. It failed to account for all potential CPU reorderings,
such as the access of vm_fault_info being reordered before the atomic_read
of the flag. This approach is also more verbose and less performant than
using the proper atomic functions with acquire/release semantics.

Fix this by switching to atomic_set_release() and atomic_read_acquire().
These functions provide the necessary acquire and release semantics,
which act as memory barriers to ensure the correct order of operations.
It is also more efficient and idiomatic than using explicit full memory
barriers.

Fixes: b97dfa2 ("drm/amdgpu: save vm fault information for amdkfd")
Cc: [email protected]
Signed-off-by: Gui-Dong Han <[email protected]>
Signed-off-by: Felix Kuehling <[email protected]>
Reviewed-by: Felix Kuehling <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>

Author: 2045gemini

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c

@@ -2353,10 +2353,9 @@ void amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(struct kgd_mem *mem)
 int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct amdgpu_device *adev,
 					  struct kfd_vm_fault_info *mem)
 {
-	if (atomic_read(&adev->gmc.vm_fault_info_updated) == 1) {
+	if (atomic_read_acquire(&adev->gmc.vm_fault_info_updated) == 1) {
 		*mem = *adev->gmc.vm_fault_info;
-		mb(); /* make sure read happened */
-		atomic_set(&adev->gmc.vm_fault_info_updated, 0);
+		atomic_set_release(&adev->gmc.vm_fault_info_updated, 0);
 	}
 	return 0;
 }

drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c

@@ -1055,7 +1055,7 @@ static int gmc_v7_0_sw_init(void *handle)
 					GFP_KERNEL);
 	if (!adev->gmc.vm_fault_info)
 		return -ENOMEM;
-	atomic_set(&adev->gmc.vm_fault_info_updated, 0);
+	atomic_set_release(&adev->gmc.vm_fault_info_updated, 0);
 
 	return 0;
 }
@@ -1287,7 +1287,7 @@ static int gmc_v7_0_process_interrupt(struct amdgpu_device *adev,
 	vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
 			     VMID);
 	if (amdgpu_amdkfd_is_kfd_vmid(adev, vmid)
-		&& !atomic_read(&adev->gmc.vm_fault_info_updated)) {
+		&& !atomic_read_acquire(&adev->gmc.vm_fault_info_updated)) {
 		struct kfd_vm_fault_info *info = adev->gmc.vm_fault_info;
 		u32 protections = REG_GET_FIELD(status,
 					VM_CONTEXT1_PROTECTION_FAULT_STATUS,
@@ -1303,8 +1303,7 @@ static int gmc_v7_0_process_interrupt(struct amdgpu_device *adev,
 		info->prot_read = protections & 0x8 ? true : false;
 		info->prot_write = protections & 0x10 ? true : false;
 		info->prot_exec = protections & 0x20 ? true : false;
-		mb();
-		atomic_set(&adev->gmc.vm_fault_info_updated, 1);
+		atomic_set_release(&adev->gmc.vm_fault_info_updated, 1);
 	}
 
 	return 0;

drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c

@@ -1168,7 +1168,7 @@ static int gmc_v8_0_sw_init(void *handle)
 					GFP_KERNEL);
 	if (!adev->gmc.vm_fault_info)
 		return -ENOMEM;
-	atomic_set(&adev->gmc.vm_fault_info_updated, 0);
+	atomic_set_release(&adev->gmc.vm_fault_info_updated, 0);
 
 	return 0;
 }
@@ -1468,7 +1468,7 @@ static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
 	vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
 			     VMID);
 	if (amdgpu_amdkfd_is_kfd_vmid(adev, vmid)
-		&& !atomic_read(&adev->gmc.vm_fault_info_updated)) {
+		&& !atomic_read_acquire(&adev->gmc.vm_fault_info_updated)) {
 		struct kfd_vm_fault_info *info = adev->gmc.vm_fault_info;
 		u32 protections = REG_GET_FIELD(status,
 					VM_CONTEXT1_PROTECTION_FAULT_STATUS,
@@ -1484,8 +1484,7 @@ static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
 		info->prot_read = protections & 0x8 ? true : false;
 		info->prot_write = protections & 0x10 ? true : false;
 		info->prot_exec = protections & 0x20 ? true : false;
-		mb();
-		atomic_set(&adev->gmc.vm_fault_info_updated, 1);
+		atomic_set_release(&adev->gmc.vm_fault_info_updated, 1);
 	}
 
 	return 0;