Re: mutex lock panic on halt/shutdown unmounting GPT/RAID filesystem

[Greg Oster <oster%cs.usask.ca@localhost>]

On Tue, 5 Jul 2011 10:57:48 -0500 (CDT)
"John D. Baker" <jdbaker%mylinuxisp.com@localhost> wrote:

> I'm building (built, rather, now testing) a machine to be my new file
> server.  I'm using a NetBSD-5.1_STABLE/i386 kernel made from GENERIC
> plus "options RF_RAID5_RS=1" to get RAID-5 w/rotated sparing.  The
> raid is built across 8 1TB Hitachi SATA drives using a 4-port siisata
> (sii3124-based) PCIe card and the onboard ahcisata ports of an intel
> D945GCL board.  The system disk is on the parallel ATA bus.
> 
> The machine has 4GB of RAM, although only a little over 3.5GB is
> actually visible, naturally.  As such, I only defined a token 2GB
> of swap space.
> 
> I have the RAID in a single filesystem defined using GPT and wedges.
> All file systems are mounted with the "log" option.
> 
> Prior to having the RAID intialized, newfs'ed and mounted, nothing
> appeared amiss.
> 
> Once the big filesystem was online, the system will panic when
> shutting down (e.g.: shutdown -r now) with "mutex lock error: locking
> against myself".  Unfortunately it all goes by too fast to read.  It
> saves the crashdump, but I think the subsequent savecore has
> problems--the memory image is there, but the kernel image is only 10
> bytes in size.  It displays "(null) bad address".
> 
> This appears to happen AFTER the big filesystem is successfully
> unmounted but BEFORE the system disk's filesystems have been
> unmounted.  Console and dmesg indicates that the OS partitions get
> the log-replay treatment when starting up again.  The RAID and its
> filesystem are intact.
> 
> I've determined that if I manually unmount the GPT/RAID filesystem
> before shutting down it doesn't panic.
> 
> Any clues?

It may not be related, but I suspect you may be the first person to be
testing the RF_RAID5_RS bits :) It wouldn't surprise me if there were a
few wrinkles with that code and NetBSD.
 
> Need more data?

Probably... (But I won't have a chance to look at this for at least 3
weeks... )

Later...

Greg Oster