Re: Regression: kernels since 2.6.26 are unusably slow

[Arjan van de Ven <arjan@infradead.org>]

On Fri, 25 Sep 2009 23:09:32 +0100
Aneurin Price <aneurin.price@gmail.com> wrote:

> Hi,
> Every kernel since 2.6.26 has been unusably slow for me - to the
> extent that I initially thought they were hanging on boot. Exactly
> how bad it is seems to vary, but it could take several minutes to
> boot, and then another minute or two to log in (with bash pegging the
> CPU as it loads). Everything seems to need vastly more CPU time than
> usual - it feels a little like I'm using a 486.

means part of memory is uncached.

> 
> I've bisected the problem down to commit
> 42651f15824d003e8357693ab72c4dbb3e280836 (x86: fix trimming e820 with
> MTRR holes). Having basically no idea what that means, I thought I'd
> try building a kernel with MTRRs disabled, to see if that would make
> any difference, but no joy.

b0rked MTRRs will do that to you ;-)

> 
> Can anyone give me some idea of where to go next, or let me know what
> further information I should provide?
> 

one thing to try is enabling CONFIG_MTRR_SANITIZER as config option;
that lets linux rewrite all MTRRs to see if it can puzzle a combination
together that fits all your memory...

-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org