Re: [GIT PULL] adaptive spinning mutexes

On Wed, 14 Jan 2009 22:14:58 +0100
Ingo Molnar <mingo@elte.hu> wrote:

> 
> * Andrew Morton <akpm@linux-foundation.org> wrote:
> 
> > On Wed, 14 Jan 2009 21:51:22 +0100
> > Ingo Molnar <mingo@elte.hu> wrote:
> > 
> > > 
> > > * Andrew Morton <akpm@linux-foundation.org> wrote:
> > > 
> > > > > > Do people enable CONFIG_SCHED_DEBUG?
> > > > > 
> > > > > If they suspect performance problems and want to analyze them?
> > > > 
> > > > The vast majority of users do not and usually cannot compile their own 
> > > > kernels.
> > > 
> > > ... which they derive from distro kernels or some old .config they always 
> > > used, via 'make oldconfig'. You are arguing against well-established facts 
> > > here.
> > > 
> > > If you dont believe my word for it, here's an analysis of all kernel 
> > > configs posted to lkml in the past 8 months:
> > > 
> > >    $ grep ^CONFIG_SCHED_DEBUG linux-kernel | wc -l
> > >    424
> > > 
> > >    $ grep 'CONFIG_SCHED_DEBUG is not' linux-kernel | wc -l
> > >    109
> > > 
> > > i.e. CONFIG_SCHED_DEBUG=y is set in 80% of the configs. A large majority 
> > > of testers has it enabled and /sys/debug/sched_features was always a good 
> > > mechanism that we used for runtime toggles.
> > 
> > You just disproved your own case :(
> 
> how so? 80% is not enough?

No.

It really depends on what distros do.

> I also checked Fedora and it has SCHED_DEBUG=y 
> in its kernel rpms.

If all distros set SCHED_DEBUG=y then fine.

But if they do this then we should do this at the kernel.org level, and
make it a hard-to-turn-off thing via CONFIG_EMBEDDED=y.

> note that there's also a performance issue here: we generally _dont want_ 
> a debug sysctl overhead in the mutex code or in any fastpath for that 
> matter. So making it depend on SCHED_DEBUG is useful.
> 
> sched_feat() features get optimized out at build time when SCHED_DEBUG is 
> disabled. So it gives us the best of two worlds: the utility of sysctls in 
> the SCHED_DEBUG=y, and they get compiled out in the !SCHED_DEBUG case.

I'm not detecting here a sufficient appreciation of the number of
sched-related regressions we've seen in recent years, nor of the
difficulty encountered in diagnosing and fixing them.  Let alone
the difficulty getting those fixes propagated out a *long* time
after the regression was added.

You're taking a whizzy new feature which drastically changes a critical
core kernel feature and jamming it into mainline with a vestigial
amount of testing coverage without giving sufficient care and thought
to the practical lessons which we have learned from doing this in the
past.

This is a highly risky change.  It's not that the probability of
failure is high - the problem is that the *cost* of the improbable
failure is high.  We should seek to minimize that cost.