syslogd and bzip2 hogs after 6.3 to 7.0 upgrade

[Neil Darlow]

Hi,

I have recently upgraded from RELENG_6_3 to RELENG_7_0 via the source 
method. I followed the upgrade instructions in the 7.0 release notes and 
rebuilt all ports successfully.

My hardware is a VIA EPIA PD10000 (VIA Centerhauls CPU) with gmirrored 
IDE disks and GELI encrypted swap (using padlock).

My kernel is GENERIC with removal of 386, 486 and 585 CPU support and 
addition of options for:

DEVICE_POLLING, HZ=1000, ALTQ, PFLOG, PFSYNC and QUOTA.

I use PF for my firewalling and device polling on my two integrated Via 
Rhine NICs.

This setup worked great under RELENG_6_2 and RELENG_6_3 (only for a 
short period due to RELENG_7_0 becoming available) with absolutely no 
issues whatever.

Since upgrading to RELENG_7_0 I have a problem when the nightly rotation 
of logfiles occurs. A number of bzip2 processes appear which, with 
syslogd, consume huge amounts of CPU.

Looking at the output of "top" I see that my interrupt CPU usage has 
increased to 46% and userland CPU usage to 44%. The remainder is system 
and there is no idle CPU time left.

The bzip2 processes appear to be running very slowly. Looking at 
/var/log/ you can see the logfiles have been renamed but not compressed 
although, given a long time, I believe they will be.

I must point out that I'm using SCHED_4BSD and not SCHED_ULE on this 
system. I am aware that ULE has been reported to suffer from this type 
of problem.

I'm at a bit of a loss as to where to look next. At present I am 
rebuilding the kernel with i586 support just in case gcc-4.2.1 has 
introduced a code generation incompatibility that might affect my VIA 
CPU somehow.

Any suggestions gratefully received. Please CC me as I'm not subscribed 
to the list.

Regards,
Neil Darlow