Matthew Wilcox wrote:
On Thu, Nov 20, 2008 at 11:24:07AM +1100, Vincent Rizza wrote:
I've made some more changes based on the feedback. I figured if the
sysfs entries were removed before the msi_desc structs were deleted
from msi_list then we wouldn't have to worry about synchronization.
I'm not sure that's true. Fortunately, there's an easy way to find out.
(sleep 10; cat) < /sys/bus/pci/..../msix_irq/0
then rmmod the driver (assuming it takes less than ten seconds to unload
... if it takes longer, increase the sleep above).
If you get an oops, then you have some more worrying to do ;-)
For bonus points, if the above doesn't oops, try hot-unplugging the
device as well (if you don't have the hardware to do this, use fakephp
to simulate it).
Thanks, I'll give these a go.
What I did initially was:
while(1); do cat < /sys/bus/pci/..../msix_irq/0; done
while(1); do rmmod ib_mthca; modprobe ib_mthca msi_x=1; done
I let that run for about 30 minutes.
How does it look? What's the decision on the directory name?
I still don't like it being called 'msix_irq'. As I said, we could also
put MSI interrupts in the same directory. What's wrong with just
'interrupts'? or 'irqs' if you want to be picky. If your problem with
tht is that 'irq' isn't in there, we could call it 'aux_irqs' or
'extra_irqs' or something, i
"interrupts" should do the trick. :)
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
