Re: 2.6.25-rc7-git2: Reported regressions from 2.6.24

[Adrian Bunk <bunk@kernel.org>]

On Fri, Mar 28, 2008 at 05:17:17PM +0100, Rafael J. Wysocki wrote:
> On Friday, 28 of March 2008, Adrian Bunk wrote:
> > On Fri, Mar 28, 2008 at 12:16:03PM +0100, Thomas Gleixner wrote:
> > > On Fri, 28 Mar 2008, Adrian Bunk wrote:
> > > > On Fri, Mar 28, 2008 at 12:00:25PM +0100, Peter Zijlstra wrote:
> > > > > On Fri, 2008-03-28 at 11:58 +0100, Thomas Gleixner wrote:
> > > > > > On Fri, 28 Mar 2008, Thomas Gleixner wrote:
> > > > > > > > Bug-Entry	: http://bugzilla.kernel.org/show_bug.cgi?id=9969
> > > > > > > > Subject		: 2.6.24-git15 Keyboard Issue?
> > > > > > > > Submitter	: Chris Holvenstot <cholvenstot@comcast.net>
> > > > > > > > Date		: 2008-02-06 14:02 (51 days old)
> > > > > > > > References	: http://lkml.org/lkml/2008/2/6/100
> > > > > > > > 		  http://lkml.org/lkml/2008/2/13/82
> > > > > > > > Handled-By	: Thomas Gleixner <tglx@linutronix.de>
> > > > > > > > Patch		: http://lkml.org/lkml/2008/2/15/343
> > > > > > >  
> > > > > > > asked the bug reporter for an update.
> > > > > > 
> > > > > > Now that lkml.org is working again, I checked the patch which is
> > > > > > referenced above and I have a hard time to connect it even remotely to
> > > > > > that bug.
> > > > > > 
> > > > > > As far as I can tell the discussion on lkml identified GROUP_SCHED=y
> > > > > > as the culprit, but I have no idea whether there was any resolution
> > > > > > other than disabling GROUP_SCHED.
> > > > > > 
> > > > > > Peter ??
> > > > > 
> > > > > Correct, I am working on the issue but that's not going to be .25 stuff.
> > > > 
> > > > How are we going to get 2.6.25 working no worse than 2.6.24?
> > > > 
> > > > Letting GROUP_SCHED depend on BROKEN would sound logical, but that would 
> > > > also kill FAIR_GROUP_SCHED.
> > > 
> > > I think default it to off should be sufficient.
> > 
> > A system administrator saying yes to GROUP_SCHED when running
> > "make oldconfig" can hardly be blamed for doing so, and "makes your 
> > mouse and keyboard unusable" mustn't be the result.
> > 
> > Avoiding unexpected breakages when toggling an option is _much_ more 
> > important than e.g. whether some exotic randconfig configuration 
> > compiles.
> 
> Perhaps add something like "(UNRELIABLE)" to the Kconfig option?

That would be called "depends on BROKEN".

There seem to be two choices:
- either FAIR_GROUP_SCHED=y brought in some situations real advantages
  with 2.6.24, then the solution is to revert commits until 2.6.25 is no 
  worse than 2.6.24 (no matter how many and how big they are) or
- we can simply let GROUP_SCHED depend on BROKEN in 2.6.25.

What we need are not hints for users how to avoid known problems, 
what we need are big cluebats for kernel developers who caused
regressions.

> Rafael

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed