padani filesystemu

[Jiri Kosina]

On Wed, 7 Jun 2006, Martin Dostal wrote:

> ruzne casto se mi stava (nekdy za tyden, nekdy za mesic), ze server castecne
> umre a na terminalu je tohle:
> ---------------------------------------
> kernel BUG at block/ll_rw_blk.c:2891!
> invalid operand: 0000 [#1]
> CPU:    0
> EIP:    0060:[<c019ab13>]    Not tainted VLI
> EFLAGS: 00010246   (2.6.15-1-k7)
> EIP is at submit_bio+0x18/0xaa
> eax: 00000000   ebx: eaf6df40   ecx: 00000000   edx: 00000001
> esi: eaf6df40   edi: 0046cd2f   ebp: 00000000   esp: c194dc5c
> ds: 007b   es: 007b   ss: 0068
> Process pdflush (pid: 160, threadinfo=c194c000 task=dffdf030)
> Stack: c014b813 dfe47cc0 00000010 00000000 c194dd50 00000019 c194dcec dd13f398
>        c014b87f 00000000 c014b27c 00000001 eaf6df40 00000010 00000001 00000000
>        00000000 c194dd50 00000019 c194dcec dd13f398 f8930c19 00000001 df0045b4
> ---------------------------------------

V podobnych pripadech je jeste dobre prilozit unwinduty stack ktery 
ukazuje callchain kterym k chybe doslo.

> Problem mam na stroji s Debian Etch na AMDcku, deska NForce4, na tom RAID5 na
> SATA discich (softwarove, bootovaci partition na RAID1), LVM a ReiserFS. Nezda
> se ale, ze by problem zpusobovala prave tahle kombinace. Jadro je 2.6.15-1-K7,
> ale delalo to i na 2.6.12-i386 uplne stejne.

Pouzivate cciss driver? Pokud ano, tak by vam mel pomoci patch zverejneny 
v tomto threadu: 
http://www.ussg.iu.edu/hypermail/linux/kernel/0604.1/0448.html

Pokud nechcete mergovat cely ten patch, jeho podstata tkvi v:

@@ -1249,6 +1296,8 @@ static void cciss_update_drive_info(int

                blk_queue_max_sectors(disk->queue, 512);
+               blk_queue_softirq_done(disk->queue, cciss_softirq_done);
                disk->queue->queuedata = hba[ctlr];
                blk_queue_hardsect_size(disk->queue,

Bylo to opraveno myslim v 2.6.17-rc2.

-- 
JiKos.