Re: [Xen-devel] [PATCH 6/7] mini-os: Added rmb to xenbus code

[Samuel Thibault <samuel.thibault@xxxxxxxxxxxx>]

Ian Campbell, le Fri 06 Jun 2014 15:15:42 +0100, a écrit :
> On Fri, 2014-06-06 at 05:59 +0100, karim.allah.ahmed@xxxxxxxxx wrote:
> > In any case, by looking at the patch it seems like rmb was required to
> > make sure that the reply message is visible to the current processor
> > when this thread wakes up and starts reading it.
> 
> Can we assume (or arrange) that schedule() gives us this guarantee? Do
> we want to?

I'd tend to think something like that, yes. More precisely, thread
wait/wake primitives usually provide a barrier: events before the wake
are supposed to complete before the wait returns. So both should have a
memory barrier.

Samuel

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