megaraid irq disable after days

Hello,

Here a mail, which I have posted to linux-kernel yesterday, which describe a problem with a kernel 2.6.10 which disable IRQ of my megraid after days of work. A very good friend advise me to forward this mail here.

Sorry for the cross-posting, but I don't know neither how to solve the problem, nor even if it's a problem of my configuration or of the kernel.

Regards,
Julien

-------- Message original --------
Sujet: megaraid irq disable after days
Date: Tue, 24 May 2005 16:13:31 +0200
De: julien.combes.i-carre.net <[email protected]>
Pour: [email protected]

Hello,

I have a problem with a kernel 2.6.10 (sources from debian) which disable IRQ of my megraid (driver megaraid_mbox) on several servers after days of work and several millions of interuptions. When the IRQ is disable, the servers do that :

May 22 03:02:20 relternet-01 kernel: irq 17: nobody cared!
May 22 03:02:20 relternet-01 kernel:  [__report_bad_irq+42/160]
__report_bad_irq+0x2a/0xa0
May 22 03:02:20 relternet-01 kernel:  [handle_IRQ_event+48/112]
handle_IRQ_event+0x30/0x70
May 22 03:02:20 relternet-01 kernel:  [note_interrupt+112/176]
note_interrupt+0x70/0xb0
May 22 03:02:20 relternet-01 kernel:  [__do_IRQ+304/320]
__do_IRQ+0x130/0x140
May 22 03:02:20 relternet-01 kernel:  [do_IRQ+25/48] do_IRQ+0x19/0x30
May 22 03:02:20 relternet-01 kernel:  [common_interrupt+26/32]
common_interrupt+0x1a/0x20
May 22 03:02:20 relternet-01 kernel:  [mwait_idle+51/80]
mwait_idle+0x33/0x50
May 22 03:02:20 relternet-01 kernel:  [cpu_idle+59/80] cpu_idle+0x3b/0x50
May 22 03:02:20 relternet-01 kernel: handlers:
May 22 03:02:20 relternet-01 kernel: [pg0+944120576/1069794304]
(megaraid_isr+0x0/0x1e0 [megaraid_mbox])
May 22 03:02:20 relternet-01 kernel: Disabling IRQ #17

I haven't noticed something else curious on the servers.

hardware of these servers :
- bi Intel Xeon 2.4Ghz
- 4.5 GB of RAM
- MegaRaid SCSI 320-2 [1]

The kernel use SMP, HT, high memory support 64GB, megaraid_mbox driver (v2.20.4.1, in module with initrd until yesterday) and don't use preempt.

As I didn't find anything that match IRQ disable and megaraid under internet, I have tried several ways :
- at the beginning, the IRQ of eth0 and megaraid was shared. I have
corrected this [2] but the problem is staying.
- I have try newer version of the kernel 2.6.9 to 2.6.10. No benefic
result.
- I have try the boot option "acpi=ht". No benefic result.
- the firmeware of the megaraid has been upgraded (to the 1L37
version). No benefic result.

Since yesterday, I am trying, on all server which have the problem, kernel 2.6.11.10 (source take kernel.org) with megaraid_mbox built-in (v2.20.4.5) and with differents boot options on servers :
- "noirqdebug" and "acpi=ht"
- "noirqdebug" and "acpi=off"
- "acpi=off"
- "acpi=ht"

I have this problem since I installed them few weeks ago with my new FAI (Debian Fully autmatic Installation). I have a lot of difficulty for solving the probleme because servers can run without problem one or two weeks. These servers are not yet in production but still in tests of charge; they should be in production in one or two weeks... if I find a way to correct this problem !

Do you have any ideas of which way I can search ?

Regards,
Julien

[1] 0000:03:08.0 RAID bus controller: LSI Logic / Symbios Logic MegaRAID (rev 01)
Subsystem: LSI Logic / Symbios Logic MegaRAID 518 SCSI 320-2
Controller
Flags: bus master, fast Back2Back, 66MHz, slow devsel, latency
32, IRQ 17
Memory at d0500000 (32-bit, prefetchable) [size=64K]
Capabilities: [80] Power Management version 2

[2]

irq  0:  88018734 timer                 irq 16:  21659113 eth0
irq  1:         9 i8042                 irq 17:  12532775 megaraid
irq  2:         0 cascade [4]           irq 18:        30 aic79xx
irq 12:         3                       irq 19:        30 aic79xx
irq 14:         1 ide0                  irq 23:         0 ehci_hcd