On 01/05/2009 07:46 AM, Arnd Bergmann wrote:

On Sunday 28 December 2008, malc wrote:

quoted

Now to the Christmas cheer, i've tried v2.6.28 and couldn't help but
notice that the problem is gone, bisecting v2.6.27 (which funnily i
had to mark good) to v2.6.28 (which has to be marked bad) wasn't fun
but eventually converged at ab598b6680f1e74c267d1547ee352f3e1e530f89

commit ab598b6680f1e74c267d1547ee352f3e1e530f89
Author: Paul Mackerras [off-list ref]
Date:   Sun Nov 30 11:49:45 2008 +0000

     powerpc: Fix system calls on Cell entered with XER.SO=1

Now the lock-up is gone, however the code never exercises the path
taken during the lock-up so i guess it, at least, deserves a better
look by PPC64 care takers.

Yes, this change was suspected to help with Mono as well, not just
Java, because both of them use their own syscall path rather than
going through glibc. The reason why you see the lock-up in a different
place is because the bug manifested in getting incorrect syscall
return codes, which probably made mono go into a normally unused
error handling case.

Just FYI, I looked into a problem of Mono that was reported during
its build.  During the build Mono itself is used, and it failed due
to lack of memory.  Mono kept using more and more memory until all
was consumed.  I didn't look at why, I just saw it with the Gnome
system monitor.

-Geoff