On Thu, Oct 14, 1999 at 03:45:13PM -0500, David Thompson wrote:
> Shawn McKee wrote:
> >Hi Everyone
> >
> >I saw a reply from Jacob on this list indicating the 2.2.10 version of
> >the modules work with 2.2.12 kernels. I have gotten the libafs-2.2.10.*
> >modules
> >and the modified insmod and put them in /usr/vice/etc/modload and
> >/sbin respectively. I have also modified the afs (.rc) script to hardcode
> >the VERSION to 2.2.10 and use a -f on the insmod line to force the load.
> >
> >I am unable to get the afs client to start under Redhat 6.1 (kernel
> >2.2.12-20 SMP). The /usr/vice/etc/modload/libafs has MANY unresolved
> >references (attached to the end of this message is the libafs.map file).
> >
> >My question is whether or not anyone has run AFS under Redhat 6.1 with the
> >2.2.12 kernel? If anyone has any information about AFS under Redhat 6.1, I
> >would appreciate hearing about it. (Including suggestions for resolving the
> >unresolved
> >references below) Thanks for your time!
>
> Add the option '-P smp' to the insmod command line. The smp version
> prefix changed from 2.2.10 to 2.2.12.
>
> I haven't run into any problems with the most recent Transarc build on
> 2.2.12, but I haven't stressed it much.
I didn't stressed it much but got the following when running the afs 2.2.10
module on RedHat-6.1. The only change I made was that I recompiled the
kernel RPM package without symbols.
Exceeded pool of AFS vnodes(VLRU cycle?)<1>Unable to handle kernel paging request at
virtual address ffffffff
current->tss.cr3 = 03f81000, %cr3 = 03f81000
*pde = 00281067
*pte = 00000000
Oops: 0002
CPU: 0
EIP: 0010:[<c8111e6f>]
EFLAGS: 00010292
eax: 00000028 ebx: 00000000 ecx: c65aa000 edx: c65aa000
esi: c65aa000 edi: c246f400 ebp: c4b52610 esp: c65abcec
ds: 0018 es: 0018 ss: 0018
Process updatedb (pid: 21978, process nr: 88, stackpage=c65ab000)
Stack: c65aa000 c04aab7c c4b52610 c80ed6f6 c812c54f 00000000 c65aa000 c04aab7c
c0d8fa80 c65aa000 c811acae c3738e00 00000000 00000001 c04aab7c 00004000
00000800 00000000 c246f400 c04aabd0 c3738e00 00000000 00000004 00000fa0
Call Trace: [<c80ed6f6>] [<c812c54f>] [<c811acae>] [<c8139750>] [<c80f730d>]
[<c80f7389>] [<c80f0e35>]
[<c811a9e2>] [<c80ef5ed>] [<c80e417a>] [<c80e3eea>] [<c80f8b79>] [<c80f8bda>]
[<c80f1dc4>] [<c811c3db>]
[<c01312f6>] [<c01314d4>] [<c01315c8>] [<c012f4ea>] [<c01093e4>]
Code: c6 05 ff ff ff ff 2a 83 c4 10 5b c3 90 57 56 53 8b 74 24 10
Result: all afsd processes are hung. The kernel report a load equal
to the number of afsd processes and /etc/rc.d/init.d/afs stop just hang.
I guess the only cure is a reboot which will be likely to fail to umount
properly the file systems due to /afs being hung :-(
Of course since the source is not open I'm just unable to debug
this and or provide more appropriate information. Binary only kernel
modules are by essence a Bad Thing.
I report the bug to make clear to others that running 2.2.10 module
on RH-6.1 SMP is likely not to be stable enough for production setups.
I didn't tried arla lately, I will probably do so soon.
Daniel
--
[EMAIL PROTECTED] | W3C, INRIA Rhone-Alpes | Today's Bookmarks :
Tel : +33 476 615 257 | 655, avenue de l'Europe | Linux, WWW, rpmfind,
Fax : +33 476 615 207 | 38330 Montbonnot FRANCE | rpm2html, XML,
http://www.w3.org/People/W3Cpeople.html#Veillard | badminton, and Kaffe.
