I've been running 3.22 on Solaris 7 (built with WorkShop 5 C) since February
and have never seen this problem. Perhaps my traffic is too low but
if this is a 2.6 specific problem, it's worth noting that
there are bugs with malloc on Solaris 2.6. Would an exim built with
gcc use the standard libc malloc?
Sun patch 105210-38 for Solaris 2.6 fixes the following (among others)...
4225913 malloc does not set errno to ENOMEM when not enough memory
4193683 malloc() with very big requested array size works incorrectly
4314913 fix for Bug 4042527 leaves part of heap unusable, so frees can not
combine.
4053258 calloc does not set errno to ENOMEM when not enough memory
4042527 fix for Bug 4042527 leaves part of heap unusable, so frees can not
combine.
malloc does not comply with standard
4100561 watchmalloc.so.1 malfunctions when applied to a process using valloc()
Some of those are vague enough to mean anything.
Do any of the systems with this problem have this patch applied? (The above
are fixed in 105210-28, the next 10 versions don't seem to affect malloc).
Robert Evans
--
mailto:Robert.Evans@cs.cf.ac.uk Tel:+44(0)29 2087 5518
http://www.cs.cf.ac.uk/People/Robert.Evans.html Fax:+44(0)29 2087 4598
Dept of Computer Science, Cardiff University, PO Box 916, Cardiff, CF24 3XF
Cardiff University is the public name of the University of Wales, Cardiff,
a constituent institution of the University of Wales.
