Re: [Xen-devel] [PATCH] x86: correct CPUID output for out of bounds input

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 01.09.16 at 13:23, <andrew.cooper3@xxxxxxxxxx> wrote:
> On 24/08/16 16:31, Jan Beulich wrote:
>> Another place where we should try to behave like real hardware; see
>> the code comments.
>>
>> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
>>
>> --- a/xen/arch/x86/hvm/hvm.c
>> +++ b/xen/arch/x86/hvm/hvm.c
>> @@ -3358,6 +3358,31 @@ void hvm_cpuid(unsigned int input, unsig
>>      if ( !edx )
>>          edx = &dummy;
>>  
>> +    if ( input & 0xffff )
>> +    {
>> +        /*
>> +         * Requests beyond the highest supported leaf within a group return
>> +         * zero on AMD and the highest basic leaf output on others.
>> +         */
>> +        unsigned int lvl;
>> +
>> +        hvm_cpuid(input & 0xffff0000, &lvl, NULL, NULL, NULL);
> 
> I have specifically deferred fixing this issue so far, because I don't
> want to increase the quantity of recursion with hvm_cpuid().
> 
> Also, because of the poor datastructure for domain cpuid, this adds 1
> and possibly 2 extra loops over the unordered list.
> 
> 
> On the way back from Toronto, I started experimenting with my
> full-policy plans, including a structured information layout so
> cpuid.basic.max_leaf can be found directly, and starting a guest_cpuid()
> function intending to replace both pv_cpuid() and hvm_cpuid() in due course.
> 
> Would you be amenable to leaving this issue as-is for now, until there
> is a more efficient way of fixing it?

If you get this ready for 4.8, yes. Otherwise I think the variant here
is better than nothing until yours arrives.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel