interrupt storm on MSI IXP600 based motherboards

[Dan Langille]

On Jan 23, 2009, at 2:34 PM, Marat N.Afanasyev wrote:

> Dan Langille wrote:
>> On Jan 22, 2009, at 11:38 AM, Dan Langille wrote:
>>> Victor Balada Diaz wrote:
>>>> On Wed, Jan 21, 2009 at 07:22:06PM +0300, Marat N.Afanasyev wrote:
>>>>>>> trouble with onboard re(4) was resolved in -CURRENT and - 
>>>>>>> STABLE, but storms are not bound to ethernet only. storm may  
>>>>>>> appear on any device. if any device generates enough  
>>>>>>> interrupts rate, storm will arrive.
>>>>>> Yes, I just got another storm, on my ATA controller this time. Ah
>>>>>> well, so much for the idea of disabling unneeded devices!
>>>>>>
>>>>>> -pete.
>>>>>>
>>>>> it's a kind of magic, really. I built a new kernel with KDB and  
>>>>> DDB and after 1 day, 13:15 I'm still waiting for storm to  
>>>>> arrive. And I added
>>>>> hw.acpi.osname="Linux" to /boot/loader.conf.
>>>> Try doing lots of IO and you will get the problem soon. You might
>>>> want to try:
>>>> while true; do dd if=/dev/zero of=BAH bs=1M count=1024; sync; done
>>>
>>> FWIW, last night I changed the address of the comm port IO in my  
>>> BIOS. Then I ran the Bacula regression test suite (lots of IO).   
>>> For my machine, once the interrupt storm starts, it continues.  I  
>>> do not know if that happens to everyone.
>>>
>>> Since changing the address, I have had no interrupt storms. I have  
>>> been running the above IO loop for about ten minutes.
>>>
>>> No storm yet (knock on wood).
>> And it's back:
>> Jan 22 17:21:46 polo kernel: interrupt storm detected on "irq22:";  
>> throttling interrupt source
>> Jan 22 17:23:19 polo kernel: interrupt storm detected on "irq22:";  
>> throttling interrupt source
>> Jan 22 17:28:20 polo kernel: interrupt storm detected on "irq22:";  
>> throttling interrupt source
>> Jan 22 17:33:20 polo kernel: interrupt storm detected on "irq22:";  
>> throttling interrupt source
>> Jan 22 17:38:20 polo kernel: interrupt storm detected on "irq22:";  
>> throttling interrupt source
>> I shall try the hw.acpi.osname="Linux" option now.
>> From dmsg: Jan 22 18:10:07 polo kernel: ACPI: Overriding _OS  
>> definition with "Linux"
>
> it works for me for 3 days, 16:27 and still no sign of interrupt  
> storm. and emu10kx0 generates as many as 93 interrupt per second  
> without trouble.
>
> What is your situation?

The box has rebooted twice tonight.  The first time, it was running  
the "while true; do dd..." script.  The second time, it was not.

The box is now up responding to pings, but I cannot ssh to it.  I  
can't get to the console until Monday.

-- 
Dan Langille
http://langille.org/