Re: [Xen-devel] [PATCH v2] x86/viridian: Add Partition Reference Time enlightenment

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 14.10.14 at 12:04, <ian.campbell@xxxxxxxxxx> wrote:
> On Tue, 2014-10-14 at 10:56 +0100, Jan Beulich wrote:
>> >>> On 14.10.14 at 09:45, <ian.campbell@xxxxxxxxxx> wrote:
>> > On Mon, 2014-10-13 at 09:10 +0100, Jan Beulich wrote:
>> >> >>> On 10.10.14 at 18:36, <msw@xxxxxxxxx> wrote:
>> >> > On Mon, Sep 29, 2014 at 11:28:44AM +0100, Paul Durrant wrote:
>> >> >> +    /*
>> >> >> +     * The guest will calculate reference time according to the 
>> >> >> following
>> >> >> +     * formula:
>> >> >> +     *
>> >> >> +     * ReferenceTime = ((RDTSC() * TscScale) >> 64) + TscOffset
>> >> >> +     *
>> >> >> +     * Windows uses a 100ns tick, so we need a scale which is cpu
>> >> >> +     * ticks per 100ns shifted left by 64.
>> >> >> +     */
>> >> >> +    p->TscScale = ((10000ul << 32) / d->arch.tsc_khz) << 32;
>> >> >> +
>> >> >> +    do {
>> >> >> +        p->TscSequence++;
>> >> >> +    } while ( p->TscSequence == 0xFFFFFFFF ||
>> >> >> +              p->TscSequence == 0 ); /* Avoid both 'invalid' values */
>> >> > 
>> >> > Anthony Liguori and I were looking this over today and he pointed
>> >> > something out: couldn't a second vCPU of the guest write 0 or
>> >> > 0xffffffff in a tight loop to cause a hypervisor DoS?
>> >> 
>> >> Yes, this is at least a theoretical issue that should be fixed. I don't
>> >> think it's a practical issue though: I'd expect the compiler to eliminate
>> >> the two reads of the field and instead directly use the result of the
>> >> increment.
>> > 
>> > Wouldn't that just mean the attacker needs to write fffffffe or ffffffff
>> > instead?
>> 
>> No. The effect of what I said would amount to
>> 
>>      x = p->TscSequence;
>>      do {
>>              x++;
>>      } while ( !(x + 1) || !x )
>>      p->TscSequence = x;
>> 
>> (or something equivalent without using a loop).
> 
> Ah right. Perhaps it would better to write it that way and use some sort
> of ACCESS_ONCE like macrot enforce it actually ends up that way rather
> than rely on the vagaries of the compiler?

Of course - that's why I said it's at least a theoretical issue and
needs fixing. But the v3 Paul sent deals with this differently
anyway, so not much point in continuing finding another clean
solution.

Jan


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