netdev - Re: Kernel oops with 2.6.26, padlock and ipsec: probably problem with fpu state changes

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <200808090443.00597.wolfgang.walter@stwm.de>
Date:	Sat, 9 Aug 2008 04:43:00 +0200
From:	Wolfgang Walter <wolfgang.walter@...m.de>
To:	Herbert Xu <herbert@...dor.apana.org.au>
Cc:	"H. Peter Anvin" <hpa@...or.com>,
	Suresh Siddha <suresh.b.siddha@...el.com>,
	"netdev@...r.kernel.org" <netdev@...r.kernel.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	Ingo Molnar <mingo@...e.hu>, viro@...iv.linux.org.uk,
	vegard.nossum@...il.com
Subject: Re: Kernel oops with 2.6.26, padlock and ipsec: probably problem with fpu state changes

On Saturday 09 August 2008, Herbert Xu wrote:
> On Fri, Aug 08, 2008 at 12:01:15PM -0700, H. Peter Anvin wrote:
> >
> > It's technically overkill, if (and only if!) these instructions don't 
> > actually touch the SSE state (most likely they're using the SSE 
> > pipeline, and need this stuff to deal with power management issues.)
> 
> Yes the PadLock uses the SSE pipeline, but doesn't touch any
> of the state.
> 
> > However, overkill is a good way to make sure something is dead. 
> > Applying the patch will make sure we fix the regression, and we can 
> > worry about optimizing this further post-2.6.27.
> 
> Do we really need the FPU changes right now? I'd prefer for that
> to be backed out until a proper solution is found.  Disabling
> preemption around crypto is really bad for scheduling latency.
> 

These FPU changes are already in 2.6.26. Undoing them, would that be accepted 
for 2.6.26 stable?

Maybe the following solution would be possible: if a processor with padlock is 
detected the memory for xstate is always allocated when the thread is created 
instead "lazy"?

Or would it be possible to change __switch_to(): bevor calling __unlazy_fpu() 
check if xstate is NULL, if yes, clear TS_USEDFPU and math-state?

As I wrote in my other mail: 2.6.26 with the patch seems to perform rather 
well with ipsec. Network latency and bandwidth seem completely unchanged 
compared to 2.6.15.13 as far as I can see yet.

But this is of course not a good test for kernel latency itself. As we don't 
have any destops based on VIA C3 (and these using padlock) I can't really 
test this (even if I set one up I don't know how it used to feel with 
2.6.25 :-) ). 

But I doubt that a lot of people use it as a desktop together with a VIA C3 
and ipsec. If they would they must see this crash.

Regards,
-- 
Wolfgang Walter
Studentenwerk München
Anstalt des öffentlichen Rechts
Leiter EDV
Leopoldstraße 15
80802 München
Tel: +49 89 38196-276
Fax: +49 89 38196-144
wolfgang.walter@...m.de
http://www.studentenwerk-muenchen.de/
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html