arlad dumps core on NetBSD-1.4.2/i386

[Dr. Lex Wennmacher]

On Apr 25, 10:15pm, Love wrote:

> > blue# /usr/arla/bin/mount_xfs /dev/xfs0 /afs
> >
> > succeedes, but starting arlad not (details at the end of my mail). It seems
> > that the probing for the presence of AFS syscalls in libkafs fails.
>
> Can you edit xfs/bsd/xfs_deb.c and change x to X for XDEBSYS to make the
> syscall being printed out. Then you can try set the environment-variable
> AFS_SYSCALL to whatever the number the debugging outpot printed in dmesg.

O.K., did that. The syscall number is 210, as I expected. So, the LKM loaded
syscall gets called from arlad, but arlad still dies from a segmentation
violation.

I then enabled debugging for everything (by setting X(XDEB*) in xfs_deb.c).
Here is what gets syslogged:

On modload (from dmesg):
xfs_modload
before initq(messageq and sleepq)
before initq(messageq and sleepq)
xfs_syscall installed at 210
xfs_init

On mount_xfs:
xfs_mount: struct mount mp = 0xc05fd600 path = '/afs' data = '/dev/xfs0'
dev = 30.0
xfs_statfs: mp = 0xc05fd600, sbp = 0xc05fd640, proc = 0xc7f7bb04
xfs_start mp = 0xc05fd600, flags = 0, proc = 0xc7f7bb04
xfs_sync: mp = 0xc05fd600, waitfor = 2, cred = 0xc051e480, proc = 0xc7d02c30

On starting arlad:
xfs_syscall(20, 0x0, 800c5603, 0xbfbfd0f8, 0)
a_opcode = 800c5603
xfs_message_rpc opcode = 22

(arlad dying in user space)

So, it seems to me that arlad chokes on an incoming message from xfs_mod.

Btw., I just checked: arla-0.32 runs fine on NetBSD-1.4.1, so this problem
seems really only be 1.4.2-related.


Cheers
Lex Wennmacher