llvm dev - Jan 2018 - FYI, we've posted a component of Spectre mitigation on llvm-commits

On Thu, Jan 4, 2018 at 12:31 PM Stephen Checkoway via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
>
> > On Jan 4, 2018, at 04:23, Chandler Carruth via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
> >
> > Sending a note here as this seems likely to be of relatively broad
> interest.
>
> It looks like this is producing code of the following form.
>
>         call    next
> loop:
>         pause
>         jmp     loop
> next:
>         mov     [rsp], r11
>         ret
>
> As I understand it, the busy loop is to cause the speculative execution to
> be trapped in the loop. Was something like ud2 considered? I presume that
> would stop the speculative execution without involving any of the execution
> units the way the busy loop does.
>
The pause instruction will also avoid tying up execution resources in
speculative contexts, so I wouldn't expect it to be significantly different.

>
> --
> Stephen Checkoway
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20180104/e51f6309/attachment.html>

On Jan 4, 2018, at 11:52, Chandler Carruth via llvm-dev <llvm-dev at
lists.llvm.org> wrote:
> On Thu, Jan 4, 2018 at 12:31 PM Stephen Checkoway via llvm-dev <llvm-dev
at lists.llvm.org> wrote:
> 
>> As I understand it, the busy loop is to cause the speculative execution
to be trapped in the loop. Was something like ud2 considered? I presume that
would stop the speculative execution without involving any of the execution
units the way the busy loop does.
>> 
> The pause instruction will also avoid tying up execution resources in
speculative contexts, so I wouldn't expect it to be significantly different.
Got it. The Software Developer Manual isn't entirely clear on this point (to
me at least) and IACA shows a number of ports in use during the 4 or 5 Uops
pause takes.

Thank you,

Steve

-- 
Stephen Checkoway

On Thu, Jan 4, 2018 at 1:05 PM Stephen Checkoway via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
>
> On Jan 4, 2018, at 11:52, Chandler Carruth via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
> > On Thu, Jan 4, 2018 at 12:31 PM Stephen Checkoway via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
> >
> >> As I understand it, the busy loop is to cause the speculative
execution
> to be trapped in the loop. Was something like ud2 considered? I presume
> that would stop the speculative execution without involving any of the
> execution units the way the busy loop does.
> >>
> > The pause instruction will also avoid tying up execution resources in
> speculative contexts, so I wouldn't expect it to be significantly
different.
>
> Got it. The Software Developer Manual isn't entirely clear on this
point
> (to me at least) and IACA shows a number of ports in use during the 4 or 5
> Uops pause takes.
>
That's if it executes. But pause doesn't get speculatively executed at
all.

Sadly, this makes IACA and other tools challenging to use. Instead you want
to construct a tight careful benchmark, control frequency and everything
else, and look at exact retire rates of instructions etc.

But if you find instruction sequences that reliably benchmark as faster,
please share this on the review thread. =D

>
> Thank you,
>
> Steve
>
> --
> Stephen Checkoway
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20180104/68b6fbaf/attachment.html>