On 3/12/26 15:39, Tvrtko Ursulin wrote:
>
> On 12/03/2026 14:13, Srinivasan Shanmugam wrote:
>> amdgpu_vm_bo_update() allows bo_va->base.bo to be NULL in some paths,
>> such as PRT-only updates.
>>
>> Although amdgpu_vm_is_bo_always_valid() already returns false for a NULL
>> BO, Smatch still warns that bo may be NULL before it is dereferenced
>> later in the block.
>>
>> Add an explicit `bo &&` check before calling
>> amdgpu_vm_is_bo_always_valid() to make the non-NULL condition clear and
>> fixes the below smatch error
>>
>> drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c:1353 amdgpu_vm_bo_update() error: we
>> previously assumed 'bo' could be null (see line 1292)
>>
>> Fixes: 26e20235ce00 ("drm/amdgpu: Add amdgpu_bo_is_vm_bo helper")
>> Cc: Dan Carpenter <[email protected]>
>> Cc: Tvrtko Ursulin <[email protected]>
>> Cc: Christian König <[email protected]>
>> Cc: Alex Deucher <[email protected]>
>> Signed-off-by: Srinivasan Shanmugam <[email protected]>
>> ---
>> drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>> index b89013a6aa0b..0d26346178d4 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>> @@ -1349,7 +1349,7 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev,
>> struct amdgpu_bo_va *bo_va,
>> * the evicted list so that it gets validated again on the
>> * next command submission.
>> */
>> - if (amdgpu_vm_is_bo_always_valid(vm, bo)) {
>> + if (bo && amdgpu_vm_is_bo_always_valid(vm, bo)) {
>
> That would be unfortunate:
>
> bool amdgpu_vm_is_bo_always_valid(struct amdgpu_vm *vm, struct amdgpu_bo *bo)
> {
> return bo && bo->tbo.base.resv == vm->root.bo->tbo.base.resv;
> }
>
> Maybe Dan can make smatch smarter? :) Because I don't think papering randomly
> at a single call site is great. It is even in the same compilation unit. Hmm
> does the order matter to smatch? Should we maybe move
> amdgpu_vm_is_bo_always_valid() to be earlier in the file?
Yeah that patch doesn't make much sense.
We explicitly added the amdgpu_vm_is_bo_always_valid() function to avoid such
NULL checks.
Regards,
Christian.
>
> Regards,
>
> Tvrtko
>
>> if (bo->tbo.resource &&
>> !(bo->preferred_domains &
>> amdgpu_mem_type_to_domain(bo->tbo.resource->mem_type)))
>
