Il 13/03/2014 18:08, Radim Krčmář ha scritto:
> I agree that old code is wrong and the patch looks correct, but I only
> see how the bug may cause pending IRR to not be delivered in time,
> not how interrupt can disrupt a higher priority task.
Right. Also, on SMP guests the effect would likely be just a deadlock
if a lower-priority ISR interrupted a higher priority task and accessed
shared data (since you need anyway a spinlock in addition to raising the
IRQL).
A more likely explanation is that if the remote processor delays an IPI
too much, it will have a stable TLB entry. The resulting random
corruption of paged memory is compatible with the BAD_POOL_HEADER error
codes that Radim observed.
Paolo, can you change the last sentence to ", which means we don't
inject pending IRR immediately."? (or do we just forget it?)
It's already in Linus's tree.
Paolo
--
To unsubscribe from this list: send the line "unsubscribe stable" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html