Source-Changes archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: CVS commit: syssrc



On Sat, 29 Apr 2000, Soren S. Jorvang wrote:

> On Sat, Apr 29, 2000 at 02:25:16PM +1000, Simon Burge wrote:
> > Umm, is the logic right here?  It appears that uvm_pageidlezero is
> > only called _after_ a runnable process appears on whichqs, and then
> > uvm_pageidlezero will only clear one page because it checks whichqs at
> > the end of it's loop.  Does "vmstat -s | egrep 'zero|pages free'" show
> > the zero page count at the number of free pages?
> 
> The stats happen to look right under some types of benchmarking,
> but you are right.
> 
> > With my earlier experiments at calling uvm_pageidlezero _before_ the
> > busy loop checking whichqs I was getting lockups on a pmax during the
> > SCSI drive probe (after the "waiting for 2 seconds....".
> 
> Hmm, seems to work okay for me. Could you try again on your pmax
> with what's there now?

  This fails miserably on my 5000/25.  Trying to run programs get segment
violations very frequently.  My guess at this point is that using
uncached address space to zero the page leaves non-zeroed data in the
cache.  If that cached data is then used instead of the zeroed memory,
things that expect the data to be zero will fail.

--
Michael L. Hitch                        mhitch%montana.edu@localhost
Computer Consultant
Information Technology Center
Montana State University        Bozeman, MT     USA




Home | Main Index | Thread Index | Old Index