cvs commit: src/sys/i386/i386 pmap.c

[Andrew Gallatin]

Kris Kennaway writes:
 > On Thu, Aug 07, 2003 at 10:48:45AM -0400, Andrew Gallatin wrote:
 > > Andrew Gallatin [gallatin at cs.duke.edu] wrote:
 > > > 
 > > > Bosko Milekic writes:
 > > >  > 
 > > >  >   If you're building with -j, can you try without it?  I seem to recall
 > > >  >   -j builds on i386 giving me similar results but two weeks ago.  I
 > > >  >   haven't tried with -j again since then, though.
 > > > 
 > > > -j16 works with a kernels from July 4th and June 1, June 7 and June 10th.
 > > > This is holding the src tree constant.
 > > > 
 > > > I'm going to try with the SMALL_PIPE_SIZE fix and see if that helps.
 > > 
 > > It changed the failure somewhat.  Now make segfaults, and syntax
 > > errors which may not actually exist are found when building libstdc++.
 > > 
 > > I'm going to back off to a known good kernel and make sure my srcs are
 > > not hosed.
 > 
 > I've was also seeing that behaviour on the alpha machines before the
 > SMALL_PIPE_SIZE change (don't have statistics about whether it is
 > still persisting).  Perhaps there were two or more problems here.
 > 

Basically, the way thing stand now is that merging the pmap fix from
i386 fixes the fault-on-nofault-entry panics, but there remains a
memory corruption problem.

Most times the corruption its a mysterious failure of install when
building libc, sometimes its a segv of make.  Silby claims to have
seen this (install failure) on x86, so at least its not alpha specific.


Drew