Re: [GIT PULL] Fix lost interrupt race in Xen event channels

["Jan Beulich" <JBeulich@novell.com>]

>>> On 30.08.10 at 10:48, Keir Fraser <keir.fraser@eu.citrix.com> wrote:
> Well, in 2.6.18 it was at least very unlikely that unmask_evtchn() would be
> called on an already-unmasked port. And the implementation of
> unmask-evtchn() is safe in the case that it is called for an
> already-unmasked port -- it just does a bit more work than necessary in that
> case.

... plus possibly causes an unnecessary upcall.

However, do you also think that pirq_unmask_and_notify() is safe
to be called twice? I would think the double EOI potentially sent to
Xen could lead to an interrupt getting ack-ed that didn't even get
started to be serviced yet. And this, afaict, can happen in 2.6.18
as well (ack_pirq() -> move_native_irq() -> disable_pirq()/
enable_pirq() -> pirq_unmask_and_notify() followed by end_pirq()
-> pirq_unmask_and_notify()). Here, however, you couldn't even
use the mask bit to detect the situation, since the masking only
happens after already having called move_native_irq() (i.e. the
event channel will be masked when you get into
pirq_unmask_and_notify() the second time).

Jan