[MODERATED] Re: spectrev1+

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Thu, May 31, 2018 at 02:41:46PM -0400, speck for Jon Masters wrote:
> On 05/31/2018 08:50 AM, speck for Jiri Kosina wrote:
> 
> > so, according to the information I have, this is likely to go public on 
> > 2018-06-12 (it's the one referred to as "Bounds Check Bypass Store" in the 
> > documents).
> 
> Correct. We've been asking Intel to extend this while tooling is
> confirmed to be identifying all the potential sites, and are in
> discussions with other vendors to coordinate that ask.
> 
> > I've specifically asked Intel for permission to allow Dan Carpenter to be 
> > briefed on this issue, so that he could try to teach smatch about 
> > detecting such patterns (IIUC it should be in principle as simple as 
> > changing his spectrev1 check to also look for stores and not just reads) 
> > some time ago.
> 
> Dan is covered under an agreement between Oracle and Intel and is
> already aware of the issue. He's also been working on tooling. I defer
> to Konrad on whether he can/should be on the list for other issues.

Hi!

Sorry for the late response. Went on vacation and email backlog.

Dan has been informed about this disclosure, I will double check whether
he got all or just this one. 

And also figuring out whether Dan's understanding of the disclosure and what his
tool currently does - is exactly what the disclosure speaks off.

> 
> > Has anyone here tried to teach any semantic analysis tool about this 
> > code pattern?
> 
> We have a binary scanner tool but as Josh said, it's not yet quite
> right. Btw, for the store variant, it's essentially acknowledging that a
> store implies a load. I think the existing tools are probably
> insufficient to catch all instances of this, which is why I've asked for
> this deadline (which isn't dependent on a third party) to be extended.
> 
> Jon.
> 
> -- 
> Computer Architect | Sent from my Fedora powered laptop
>