[K42-discussion] crash in autoconf's mmap test
Bryan S Rosenburg
rosnbrg at us.ibm.com
Sat Sep 10 04:52:59 EST 2005
Livio is correct. I wish I'd remembered his bug report from a few months
back. Andrew's test program mmap's a 1-page file on top of a page
obtained from malloc. We should be splitting the region backing the
malloc pool into three regions, but instead we're simply deleting the tail
of the original pool from the address space. A subsequent malloc
segfaults when it touches that part of the pool.
Livio's "fix" (which he published only to prove the nature of the bug) is
to fill the tail with a new region. That solution effectively zeros part
of the malloc pool and is clearly bogus.
The suggested correct solution involves a "splitRegion" service. I find
that we have a method with that name implemented in ProcessReplicated, but
it hasn't been exported or used. I suspect it hasn't been tested. I'll
ask Jonathan for more of the story when he gets back from vacation (on
Monday, I think).
- Bryan
Livio Soares <livio at eecg.toronto.edu>
Sent by: k42-discussion-bounces at ozlabs.org
09/09/2005 11:40 AM
To
Andrew Baumann <andrewb at cse.unsw.edu.au>
cc
K42-discussion at ozlabs.org
Subject
Re: [K42-discussion] crash in autoconf's mmap test
Hey Andrew,
I filed a bug report, and a "fix" (well, not really) to get past the
mmap
crash in autoconf a few months ago. Check it out here:
http://ozlabs.org/bugzilla/show_bug.cgi?id=14
I eventually talked to Jonathan about it and we had some ideas about
fixing
the issue permanently (it included changing how Regions/sub-Regions
are
created/managed). Alas, life has been crazy, so none of the things
discussed
went forward.
Hope this helps,
Livio
_______________________________________________
K42-discussion mailing list
K42-discussion at ozlabs.org
https://ozlabs.org/mailman/listinfo/k42-discussion
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://ozlabs.org/pipermail/k42-discussion/attachments/20050909/76f3ffb2/attachment.htm
More information about the K42-discussion
mailing list