Re: Problem with MSI-X on IA64

Bernhard Walle <bwalle@xxxxxxx> · Mon, 19 Mar 2007 19:19:22 +0100

Hello,

* Vivek Goyal <vgoyal@xxxxxxxxxx> [2007-03-15 06:37]:
> On Wed, Mar 14, 2007 at 12:44:03PM +0100, Bernhard Walle wrote:
> > 
> > * Vivek Goyal <vgoyal@xxxxxxxxxx> [2007-03-01 13:50]:
> > > 
> > > Do you have more details about why MSI initialization failed? That only
> > > can explain what should be the right solution.
> > 
> > It's not that initialisation fails (I'm booting the kdump kernel with
> > pci=nomsi, however, I also tried without).
> 
> In previous post the kernel message say MSI-X initialization failed while
> initializing cciss driver. Are you saying that if you boot kdump kernel
> with pci=nomsi, then you run into another problem?

No, you got me wrong, the MSI-X initialisation of the driver doesn't
fail (it fails because pci=nomsi, but that's intended). The problem
was ...

> Kernel getting the interrupt is not a problem. Kernel will invoke all the
> interrupt handlers and if no one handles it then after some time kernel
> will disable that irq line.

Exactly. :)

> But it will not crash.

Right, but it hangs.

> Later "irqpoll" will ensure that your driver can still work properly
> despite the fact that driver's irq line is disabled.

Yes, and irqpoll is broken on IA64, at least on the platform I tested.
See the patch below that I CC'd to this mailing list which addresses
that issue.

With my patch enabled, the cciss driver is working with irqpoll in the
kdump kernel.

> > The firmware vendor seems to work on a fix to support the reset
> > message. So I think we can ignore this issue here for now.
> 
> Is somebody working on this?

Yes, the firmware people (to not send interrupts after kexec, I think
on IA64 where the firmware has more influence as on x86_64 this is
possible, I'm not sure).

Regards,
Bernhard


Re: Problem with MSI-X on IA64
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]