Re: [PATCH 2/3] x86/cpu: fix unbootable VMs by inlining memcmp in hypervisor_cpuid_base

[Thomas Gleixner <tglx@xxxxxxxxxxxxx>]

On Fri, Aug 02 2024 at 16:25, Nikolay Borisov wrote:
> On 2.08.24 г. 11:50 ч., Alexey Dobriyan wrote:
>> If this memcmp() is not inlined then PVH early boot code can call
>> into KASAN-instrumented memcmp() which results in unbootable VMs:
>> 
>>      pvh_start_xen
>>      xen_prepare_pvh
>>      xen_cpuid_base
>>      hypervisor_cpuid_base
>>      memcmp
>> 
>> Ubuntu's gcc version 11.4.0 (Ubuntu 11.4.0-1ubuntu1~22.04) inlines
>> memcmp with patch and the bug is partially fixed.
>> 
>> Leave FIXME just in case someone cares enough to compare 3 pairs of
>> integers like 3 pairs of integers.
>> 
>> Signed-off-by: Alexey Dobriyan <adobriyan@xxxxxxxxx>
>> ---
>> 
>>   arch/x86/include/asm/cpuid.h | 15 ++++++++++++++-
>>   1 file changed, 14 insertions(+), 1 deletion(-)
>> 
>> diff --git a/arch/x86/include/asm/cpuid.h b/arch/x86/include/asm/cpuid.h
>> index 6b122a31da06..3eca7824430e 100644
>> --- a/arch/x86/include/asm/cpuid.h
>> +++ b/arch/x86/include/asm/cpuid.h
>> @@ -196,7 +196,20 @@ static inline uint32_t hypervisor_cpuid_base(const char 
>> *sig, uint32_t leaves)
>>      for_each_possible_hypervisor_cpuid_base(base) {
>>              cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
>>   
>> -            if (!memcmp(sig, signature, 12) &&
>> +            /*
>> +             * FIXME rewrite cpuid comparators to accept uint32_t[3].
>> +             *
>> +             * This memcmp()
>> +             * a) is called from PVH early boot code
>> +             *    before instrumentation is set up,
>> +             * b) may be compiled to "call memcmp" (not inlined),
>> +             * c) memcmp() itself may be instrumented.
>> +             *
>> +             * Any combination of 2 is fine, but all 3 aren't.
>> +             *
>> +             * Force inline this function call.
>> +             */
>> +            if (!__builtin_memcmp(sig, signature, 12) &&
>
> Instead of putting this giant FIXME, why not simply do the comparison as 
> ints, i.e ((uint32_t)&sig[0]) == signature1 && ((uitn32_t)&sig[4]) == 
> signature2 && ((uint32_t)&sig[8] == signature_3  and be done with it?

Because a smart compiler might turn it into a memcmp() :