[OpenAFS] GCPAGs not working? Kernel panic in afs_pag_destroy

[Bob Hoffman]

I'm having much the same problem that Mike Polek reported recently.  We 
are suffering frequent kernel panics that appear to be the result of 
PAGs not being GC'd.

Version info:  Red Hat Enterprise 4 Linux, x86_64 architecture, kernel 
2.6.9-42.0.8.ELsmp
OpenAFS 1.4.6 installed from the openafs-1.4.6-el4.2.* RPMs.
The startup script has "sysctl -w afs.GCPAGs=1"

The kernel panic is not written to any log file, so I took a photo of 
the screen and transcribed the last 24 lines here:

R10: 0000000100000000 R11: 0000ffff803fd180 R12: 00000000ffffffff
R13: 0000000000000286 R14: 0000000000000000 R15: ffffffff801cae8c
FS:  0000000000000000(0000) GS:ffffffff804e5900(0000) knlGS:0000000000000000
CS:  0010 DS: 0018 ES: 0018 CR0:000000008005003b
CR2: 000000315f48fae0 CR3: 0000000037ea8000 CR4: 00000000000006e0
Process events /1 (pid: 11, threadinfo 00000108cfed4000, task 00000100cfea3030)
Stack: 0000000000000000 ffffffffa01fd724 00000104132dcf08 00000104132dcf00
       0000010037e61440 ffffffffa02355ea 00000104132dcf08 ffffffff801caf53
       0000000000000000 ffffffff803f15e0
Call Trace:<ffffffffa01fd724>{:libafs:afs_FindUser+212} <ffffffffa02355ea>{:liba
fs:afs_pag_destroy+26}
       <ffffffff801caf53>{key_cleanup+199} <ffffffff80147856>{worker_thread+419}

       <ffffffff80133dad>{default_wake_function+0} <ffffffff80133dfe>{__wake_up_
common+67}
       <ffffffff80133dad>{default_wake_function+0} <ffffffff801476b3>{worker_thr
ead+0}
       <ffffffff8014b4cb>{kthread+200} <ffffffff80110f47>{child_rip+8}
       <ffffffff8014b403>{kthread+0} <ffffffff80110f3f>{child_rip+0}


Code: 0f 0b 4b c0 24 a0 ff ff ff ff c6 00 0f b6 03 a8 01 74 12 48
RIP <ffffffffa01f41fb>{:libafs:Afs_Lock_ReleaseR+59} RSP <00000100cfed5e18>
<0>Kernel panic - not syncing: Oops


Like Mike, I am also seeing /proc/sys/afs/GCPAGs being set to 8.  Should 
I set up a cron job that forces it back to 1 if it sees it change?  Is 
there anything I can instrument to help debug this?

Thanks in advance,

---Bob.