On Wed, 2011-06-08 at 18:27 -0700, Tim Vaillancourt wrote:
> After some more thought, I understand the DMA and IOMMU portions of
> this issue, and why DMA cannot be setup, what I don't understand is
> why this would cause a loss of the aacraid driver and a system crash.
> Shouldn't this be expected? If not, what is not doing it's job
> correctly?
The aacraid driver needs to handle the error, but it does not.
> I guess what I am looking for is a way to run Xen in a stable way. It
> seems wrong to me that this brings my system down, even if it is a
> mainline bug.
It is wrong. This driver won't be reliable with Xen.
> Another side to this issue is we have this occur only on less than 10%
> of our identical systems. Any idea why that would be?
The recent change in the way you run backups presumably can cause a
sudden increase in memory usage, whereas on other systems this never
happens. Disk drivers (among others) cannot wait for data to be swapped
out when try to allocate memory, because they may themselves be used for
swapping. Generally the kernel tries to ensure there is some physical
memory available for immediate allocation, by swapping out data early.
But if there is a sudden increase in memory usage then the kernel may
not start swapping soon enough to avoid an allocation failure in the
disk driver.
Ben.
--
Ben Hutchings
Lowery's Law:
If it jams, force it. If it breaks, it needed replacing anyway.
Attachment:
signature.asc
Description: This is a digitally signed message part