On Sat, Nov 06, 2004 at 07:14:20AM +1100, matthew green wrote:
   > 
   >    On Fri, Nov 05, 2004 at 08:58:35PM +1100, matthew green wrote:
   >    
   >    > the bug fix may have helped - but it hasn't fixed everything :-(
   >    
   >    I'm not surprized.  Did you attempt to analyze the hang?
   > 
   > 
   > it's not hung.  it's still able to act as an NFS server... so i haven't
   > bothered rebooting.  and because it's busy, i haven't gone poking again
   > yet... last i knew hardclock() called softintr_schedule(softclock_si)
   > but the softclock intr is never seen.. hardclock() does this at 100hz as
   > one would expect but once it's losing, softclock() itself is never
   > called again...
   
   Hm.  I'm surprized the system is not hung.  You sure that softclock() 
   is never called?  How does NFS function if the timers are not working?
   Have you looked at the callout buckets to see if any of them are being
   processed?  Are you sur hardclock() is running?


softclock() is never called.  NFS eventually did hang...once it got
an error.  i haven't looked at the callout buckets.  hardclock() is
definately running.  it's a year or more since i found out that much,
then the problem went away for me... and espcially with the previous
intr fix for 0 vs -1 it seems much harder to reproduce again.


.mrg.