Re: [sparc64] Strange interaction between 2.6 kernel and 2.5 (and 2.6) glibc

BERTRAND Joël <joel.bertrand@xxxxxxxxxxx> · Thu, 12 Jul 2007 11:11:32 +0200

BERTRAND Joël wrote:
David Miller wrote:
From: BERTRAND_Joël <joel.bertrand@xxxxxxxxxxx>
Date: Wed, 11 Jul 2007 12:24:07 +0200

Root rayleigh:[/etc/mail] > ps -eLf | grep clamd
clamav    3502     1  3502  1    2 10:53 ?        00:01:23 
/usr/sbin/clamd
clamav    3502     1  7226  0    2 12:00 ?        00:00:00 
/usr/sbin/clamd
root      8091  5893  8091  0    1 12:18 pts/0    00:00:00 grep clamd

    clamd remains in S mode.

    I have tried to obtain mode information with sysrq+w without any 
success, sysrq+w returns no blocked task. In this example, clamd is 
blocked, but some other tasks can be blocked.

This is not enough information to debug this problem, sorry.

You'll need to do some thinking about what kind of other information
you can fetch from this stuck process.  Perhaps running
clamd under strace will provide a good debugging trace so we can
see exactly what kind of socket it is stuck on, and in what way.

    OK. Now, I run clamd under strace. I hope that it will quickly hang 
;-) When,clamd hangs, I will post here its strace output. Are there 
significative differences between futex management on sbus and pci 
sparc64 ? I have tried to reproduce this bug on sbus workstation without 
any success, but I can see it on all sparc64/PCI I have (U60, U80, U420)...

	I'm lucky ;-) First mail in clamd and clamd sleeps... You can find in 
attachement complete strace log (strace -ff ...).

	Regards,

	JKB
Attachment:
clamav-strace.bz2

Description: Binary data