Re: 2.6.25-rc4 rcupreempt.h WARNINGs while suspend/resume

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Fri, Mar 07, 2008 at 12:35:26PM +0800, Dave Young wrote:
> On Fri, Mar 7, 2008 at 12:19 PM, Paul E. McKenney
> <paulmck@linux.vnet.ibm.com> wrote:
> > On Fri, Mar 07, 2008 at 11:07:48AM +0800, Dave Young wrote:
> >  > On Fri, Mar 7, 2008 at 12:27 AM, Paul E. McKenney
> >  > <paulmck@linux.vnet.ibm.com> wrote:
> >  > > On Thu, Mar 06, 2008 at 07:08:55PM +0800, Dave Young wrote:
> >
> > > >  > My syslog became a 2G size big file yestoday due to the warnings.
> >  > >  > How about change the WARN_ON to WARN_ON_ONCE?
> >  > >
> >  > >  Hello, Dave,
> >  > >
> >  > >  I might be convinced to make this change for 2.6.26, but the condition
> >  > >  that the WARN_ON() is complaining about is quite serious, so I don't
> >  > >  want to take a chance on it getting lost in the noise in the 2.6.25
> >  > >  series.
> >  > >
> >  > >  Seem reasonable?
> >  >
> >  > IMHO, WARN_ON_ONCE is enough for my eyes :)
> >
> >  I could believe that, but my experience has been that many others
> >  need the condition to be obvious...
> >
> >
> >  > >  Better yet, is there some sort of time-limited WARN_ON that kicks out
> >  > >  a message at most once per second or some such?  Enough to definitely
> >  > >  be noticed, but not enough to bring the machine to its knees?
> >  >
> >  > Seems there's no such functions/macros, but is is really needed?
> >
> >  If everyone reports errors when they see isolated WARN_ON()s in their
> >  logfiles, then no.  But...
> 
> Ok, I agree with you.
> 
> Maybe something like WARN_ON_HZ(condition) or
> WARN_ON_PERIOD(condition, period_value)?

Makes sense to me!  The other benefit of this sort of thing is that
it lets you know whether the problem was a one-off or whether it
continued happening -- but without too much log bloat.

I was thinking in terms of once every ten seconds, but am not all
that hung up on the exact value of the period.

Thoughts?

						Thanx, Paul