Re: [2.6.18-rc7-mm1] slow boot

[Andi Kleen <ak@suse.de>]

On Thursday 28 September 2006 19:00, Tilman Schmidt wrote:

missing context here, but ...

> On 24.09.2006 23:53, Andrew Morton wrote:
> > Do you have the time to go through the
> > http://www.zip.com.au/~akpm/linux/patches/stuff/bisecting-mm-trees.txt
> > process?
> 
> Phew, it's done. And the winner is:
> 
> x86_64-mm-i386-stacktrace-unwinder.patch
> --------8<--------8<--------8<--------8<--------8<--------8<--------
> i386: Do stacktracer conversion too
> 
> Following x86-64 patches. Reuses code from them in fact.
> 
> Convert the standard backtracer to do all output using
> callbacks.   Use the x86-64 stack tracer implementation
> that uses these callbacks to implement the stacktrace interface.
> 
> This allows to use the new dwarf2 unwinder for stacktrace
> and get better backtraces.
> 
> Cc: mingo@elte.hu
> 
> Signed-off-by: Andi Kleen <ak@suse.de>
> -------->8-------->8-------->8-------->8-------->8-------->8--------
> 
> Backing out just this patch from 2.6.18-mm1 (and resolving conflicts
> manually the obvious way) gets the boot time back to normal (ie. as
> fast as 2.6.18 mainline) on my
> Linux gx110 2.6.18-mm1-noinitrd #2 PREEMPT Thu Sep 28 18:48:32 CEST 2006 i686 i686 i386 GNU/Linux
> machine.


Hmm, i assume you have lockdep on. The new backtracer is of course slower
than the old one and it will slow down lockdep which takes a lot of backtraces. 
But it shouldn't be a significant slowdown.

Can you perhaps boot with profile=1 and then send readprofile output after
boot?

-Andi