cvs commit: src/sys/amd64/amd64 mp_machdep.c src/sys/i386/i386 mp_machdep.c

[Kris Kennaway]

Colin Percival wrote:
> Kris Kennaway wrote:
>> Colin Percival wrote:
>>>   Change the default for hyperthreading (or, generally speaking, cases
>>>   where the L1 cache is shared between CPUs) to disabled for security
>>>   reasons.  As in earlier releases, this can be changed by setting
>>>   machdep.hyperthreading_allowed=1 in /boot/loader.conf.
>>>     This is not an MFC -- no seatbelts in CURRENT.
>> What are you waiting for to happen in HEAD, 
> 
> To quote core@, whom I agree with on this point:
>   We think this decision should be revisited once at least one of the
>   following occur: new crypto code is made available by crypto vendors to
>   address cache-related attacks, or sufficient work is performed on
>   scheduling and protection mechanisms to prevent the attack from being
>   exploited.
> 
>> and what work are you doing
>> to expedite that?
> 
> When I have time, I'm working on a cryptographic library which is designed
> from the ground up to be immune to timing and cache-based side channel attacks.
> Unfortunately my time has been rather limited lately due to the pressing need
> to eat, but if you know any companies which would be interested in sponsoring
> this work I'd be very happy to hear from them.

Sounds possibly infeasible and at the very least a long way off.  Is it 
appropriate to continue to leave hyperthreading disabled for the 
indefinite future pending the completion of a library you hope to one 
day write?

I suspect that the fact that there has been little interest in 
sponsoring your development of this project should suggest to you that 
this hyperthreading issue is not widely considered to be a real security 
problem.

Kris