[openib-general] disable_irq WARN_ON

[Grant Grundler]

On Thu, Jan 13, 2005 at 07:57:52AM -0800, Roland Dreier wrote:
>     Andras> However, when I put a card in an rx4640 (four-way thing,
>     Andras> same ZX1 chipset as the 2600 ) I got an oops upon
>     Andras> modprobe. This is 2.6.10-mm2 (I've compared some diffs
>     Andras> with 11-rc1 and nothing relevant seems to have changed).
...
> I'm not sure what's going on.  mthca_init_one() is bailing out and
> disabling the device on the way out, and iosapic_unregister_intr() is
> warning (not actually an oops, just WARN_ON(1)) because it can't find
> a valid IRQ number.

I suspect this is a result of mthca_init_one bailing WITHOUT
having succesfully called request_irq().
request_irq() implies a call to enable_irq().
The WARN_ON() may be unneccessary since I would expect it to be safe
to call disable_irq() multiple times for a given device. I'll ask
if that's a reasonable expectation (or not).


> Does this happen even without MSI turned on?

If MSI were turned on, I would expect the system to MCA (crash hard).
He only has firmware rev 3.0.2 on the HCA.
As I learned, one needs at least 3.3.2.

My advice is to get your vendor to tell you which *mellanox*
firmware is safe to use on your HCA if they can't offer you anything
newer than 3.0.2.

Keep in mind that openib.org is "experimental" and no vendor
will officially support it....however they would be smart to
informally provide "appropriate resources" (ie newer firmware)
so you can do some informal testing using their firmware.

grant