5.3-BETA1 hard lockups (HP DL380 G3)

[Andrew Gallatin]

Chris Dillon writes:
 > 
 > Anybody else experiencing hard lockups with a recent 5.3-BETA?  I can 
 > manage to build and install a kernel if I hold my breath while doing 
 > it, but I can't make it through a buildworld without the system 
 > locking up hard enough to require a manual reset (thank goodness for 
 > iLO and its virtual reset button :-).
 > 
 > The hardware I'm working on is a HP Proliant DL380 G3, with dual 
 > 2.8GHz Xeons, 2GB RAM, and using the on-board SmartArray 5i+ 
 > controller with six 72GB 10KRPM disks in a RAID5 configuration. 
 > Everything worked fine at first, even went through a few 
 > kernels+worlds and plenty of port builds over the last month, but 
 > within the last week or so these hard lockups started happening during 
 > kernel/world builds (probably any moderate load would do).  I thought 
 > maybe something might have gone wrong with the hardware so I ran the 
 > HP Diagnostics on it for 24 hours in a continuous loop and everything 
 > went fine.  Any clues?

Me too.  4x opteron 844 w/AMD64 kernel, 4GB ram, mpt scsi.  Building
kernels is fine, but a buildworld always locks up hard until I backed
down to a UP kernel.  Do you have Doug White's IPI deadlock patch
applied?  (I do..).

You might also have a greater chance to get into ddb if you
set debug.kdb.stop_cpus=0

There's also MP_WATCHDOG for you, since you're on i386.


Drew