Il 24/04/2012 18:08, Michael S. Tsirkin ha scritto:
> On Tue, Apr 24, 2012 at 05:40:07PM +0200, Paolo Bonzini wrote:
>> I would be grateful if, instead of fixing the qemu-barrier.h parts of
>> the patches, you picked up the (sole) patch in the atomics branch of
>> git://github.com/bonzi
On Tue, Apr 24, 2012 at 05:40:07PM +0200, Paolo Bonzini wrote:
> I would be grateful if, instead of fixing the qemu-barrier.h parts of
> the patches, you picked up the (sole) patch in the atomics branch of
> git://github.com/bonzini/qemu.git. The constructs there are more
> com
Il 24/04/2012 17:11, Michael S. Tsirkin ha scritto:
> On Tue, Apr 24, 2012 at 05:01:07PM +0200, Paolo Bonzini wrote:
>> Il 24/04/2012 16:38, Michael S. Tsirkin ha scritto:
> I think it is prudent to address this theoretical race condition.
I think your fix is right, but it is not need
On Tue, Apr 24, 2012 at 05:01:07PM +0200, Paolo Bonzini wrote:
> Il 24/04/2012 16:38, Michael S. Tsirkin ha scritto:
> >>> I think it is prudent to address this theoretical race condition.
> >>
> >> I think your fix is right, but it is not needed on x86. lfence is only
> >> required for weakly-ord
Il 24/04/2012 16:38, Michael S. Tsirkin ha scritto:
>>> I think it is prudent to address this theoretical race condition.
>>
>> I think your fix is right, but it is not needed on x86. lfence is only
>> required for weakly-ordered memory types, so says Intel,
>
> I see this in spec:
>
> The LFENC
On Tue, Apr 24, 2012 at 03:48:27PM +0200, Paolo Bonzini wrote:
> Il 23/04/2012 15:19, Michael S. Tsirkin ha scritto:
> > virtio has the equivalent of:
> >
> > if (vq->last_avail_index != vring_avail_idx(vq)) {
> > read descriptor head at vq->last_avail_index;
> > }
> >
> > In
Il 23/04/2012 15:19, Michael S. Tsirkin ha scritto:
> virtio has the equivalent of:
>
> if (vq->last_avail_index != vring_avail_idx(vq)) {
> read descriptor head at vq->last_avail_index;
> }
>
> In theory, processor can reorder descriptor head
> read to happen speculativ
virtio has the equivalent of:
if (vq->last_avail_index != vring_avail_idx(vq)) {
read descriptor head at vq->last_avail_index;
}
In theory, processor can reorder descriptor head
read to happen speculatively before the index read.
this would trigger the following ra
