Re: [PATCH] KVM: selftests: Check guest memfd validity with flags

[Bibo Mao <maobibo@loongson.cn>]

On 2026/5/13 上午7:41, Sean Christopherson wrote:
> On Fri, May 08, 2026, Bibo Mao wrote:
>> The type of guest_memfd in structure kvm_userspace_memory_region2
>> is __u32, it is not correct to assign it with -1 and check whether
>> it is smaller than 0. Here check flags with KVM_MEM_GUEST_MEMFD
>> set.
>>
>> Signed-off-by: Bibo Mao <maobibo@loongson.cn>
>> ---
>>   tools/testing/selftests/kvm/lib/kvm_util.c | 4 +---
>>   1 file changed, 1 insertion(+), 3 deletions(-)
>>
>> diff --git a/tools/testing/selftests/kvm/lib/kvm_util.c b/tools/testing/selftests/kvm/lib/kvm_util.c
>> index 2a76eca7029d..9d3553f7e6a5 100644
>> --- a/tools/testing/selftests/kvm/lib/kvm_util.c
>> +++ b/tools/testing/selftests/kvm/lib/kvm_util.c
>> @@ -817,7 +817,7 @@ static void __vm_mem_region_delete(struct kvm_vm *vm,
>>   		kvm_munmap(region->mmap_alias, region->mmap_size);
>>   		close(region->fd);
>>   	}
>> -	if (region->region.guest_memfd >= 0)
>> +	if (region->region.flags & KVM_MEM_GUEST_MEMFD)
> 
> Hmm, it's a bit gross, but this is probably more robust?
> 
> 	if ((int)region->region.guest_memfd < 0)
yes, this is more direct, only that some guys in the community do not 
like type conversion. Both are ok for me.

> 
> E.g. if we somehow end up in a state where KVM_MEM_GUEST_MEMFD is either stale
> or the guest_memfd file was already closed.  I highly doubt either of those things
> will happen, but logically it's the correct fix (the only reason guest_memfd is
> a u32 is being it's part of the kernel's uAPI).
Actually it probably will happen, how about something like this:
-       if (region->region.guest_memfd >= 0)
+       if ((int)region->region.guest_memfd >= 0) {
                 close(region->region.guest_memfd);
+               region->region.guest_memfd = -1;
+       }

Regards
Bibo Mao