Re: [Xen-devel] [PATCH 1/2] xen: allow on_selected_cpus with interrupts disabled

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

On Fri, 2013-05-03 at 16:57 +0100, Keir Fraser wrote:
> On 03/05/2013 16:21, "Ian Campbell" <Ian.Campbell@xxxxxxxxxx> wrote:
> 
> > On Fri, 2013-05-03 at 15:58 +0100, Stefano Stabellini wrote:
> >> Allow on_selected_cpus with interrupts disabled, use it with care.
> > 
> > This is a deadlock waiting to happen. Can we not find a way to do cross
> > CPU EOI without it? If we can guarantee that we only need to EOI on one
> > CPU then does that make a specialised SGI vector more plausible?
> > 
> > Can the IPI call not be moved outside the lock? i.e. remove it from the
> > list under the lock and then IPI outside?
> > 
> > Or could you queue the IRQ on a per-pcpu list of IRQs to EOI and then
> > outside the lock send an IPI to the other CPU to check the list.
> > 
> > At the least this should assert that he current cpu isn't in the mask
> > when wait == 1.
> 
> There's little chance of me being flexible on changing the
> on_selected_cpus() interface.

I hadn't noticed this was common code, but in any case I agree with you.

>  This may be better handled under arch/arm, or
> with a new interface, or just as you sugegst rethinking the higher-level
> problem so you don't get painted into this corner in the first place.

ACK.

Ian.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel