There is a bug in glibc with calloc() not returning zeroed memory when the process address space is locked, for instance by a call to mlockall(). Versions of glibc affected are:
* RHEL-4.6 and OEL-4.6: glibc-2.3.4-2.39, all architectures.
* RHEL-5.1 and OEL-5.1: glibc-2.5-18, all achitectures.
* SLES10 SP2: glibc-2.4-31.54 , all architectures
This produces various symptoms including segmentation faults. Oracle CRS, the cssd process, as of 10.2.0.4 and higher, use the mlockall() call, and can fail. CRS versions prior to 10.2.0.4 do not use this call, and aren't effected.
Details on the problem are posted here:
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812
https://bugzilla.redhat.com/show_bug.cgi?id=405781
This bug has been fixed in RHEL5.2 glibc and OEL5.2 (glibc-2.5-54). It is also fixed in RHEL4.7 and OEL4.7 (glibc-2.3.4-2.41). We are currently testing a fix from Novell for SLES10. The bug was not present in earlier versions of glibc and it is not present in RHEL3.
Please see Metalink Note 732847.1
Comments (2)
can you post the metalink doc please? it is not published
Posted by thomas flatley | August 21, 2008 8:26 AM
Posted on August 21, 2008 08:26
The info in the Metalink note is the same as in my blog entry. You do need to log in to read Metalink notes.
Only additional information is the Oracle bug number, which is: 6790001
Posted by Elena Zannoni | August 22, 2008 9:23 AM
Posted on August 22, 2008 09:23