Re: [Bug #11308] tbench regression on each kernel release from 2.6.22 -> 2.6.28

[Ingo Molnar <mingo-X9Un+BFzKDI@public.gmane.org>]

* Mike Galbraith <efault-Mmb7MZpHnFY@public.gmane.org> wrote:

> > It looks like a potentially bogus bisection result, but _maybe_ it 
> > has relevance: changes the size of "struct security_operations", 
> > which could have alignment and layout effects on all sorts of kernel 
> > variables, kmalloc sizes, etc.
> 
> This may well be a mythical creature infestation for all I know ;-), 
> but it's address is somewhere in the 2069f45..847106f block, 316 
> commits, none of which look like they should be the least bit 
> interesting to netperf.  I reverted this particular commit in 27.git, 
> got the expected result.  Looks like I'll keep poking at it, can't 
> seem to resist.  Grr.

are you sure it's 2069f45..847106f? Filtering out the 
likely-uninteresting commits:

git log --pretty=format:"%h: %s" 2069f45..847106f | grep -viE \
'block|alsa|pcmcia|sound|Merge|iosched|blk|DAC960|scsi|s390|paride|pktcdvd|filter|cdrom|drm'

gives us:

 7daf705: Start using the new '%pS' infrastructure to print symbols
 6f0f0fd: security: remove register_security hook
 93cbace: security: remove dummy module fix
 5915eb5: security: remove dummy module
 b478a9f: security: remove unused sb_get_mnt_opts hook
 32502b8: splice: fix generic_file_splice_read() race with page invalidation
 8b3d356: ramfs: enable splice write
 a144ff0: xen: Avoid allocations causing swap activity on the resume path

which really only leaves that security commit your bisection fingered. 
Which _slightly_ raises its likelyhood of being implicated. Structure 
size changes can move two formerly far-apart netperf-relevant symbols on 
the same cacheline, which can start cache ping-pong-ing badly.

It wouldnt be the first such incident - alignment changes impacting 
macro benchmarks. (and it's hard to find it as the thing that changes 
alignment/size/sharedness might be something totally unrelated)

It's still a bit too early to say this for sure though ...

	Ingo