On Wed, Jun 29, 2016 at 10:24:16AM +0200, Alexander Gordeev wrote:
> On Mon, Jun 13, 2016 at 04:17:19PM +0200, Andrew Jones wrote:
> > On Mon, Jun 13, 2016 at 04:03:53PM +0200, Paolo Bonzini wrote:
> > > Are the barriers really necessary? I.e. why do we need
> > > __raw_{read,write}* vs just {read,write}*.
> >
> > We could drop the wrappers, and require unit tests to use barriers
> > explicitly, when needed. I'm OK with that, because I think I mostly
> > forget that we have them in the wrappers and use them explicitly
> > anyway... I'll need to double check all the uses though.
>
> According to my findings read*/write* io accessors are used on
> ioremap-ed areas in lib/arm/io.c and lib/virtio-mmio.c at the moment.
>
> Even if read*/write* wrappers are stripped of SMP barriers,
> removing compiler barriers from them would be a bad idea AFAICT.
>
> So if we need compiler barriers in read*/write* then why not
> just leave SMP barriers there?
Andrew, Paolo,
I am paraphrasing myself. So read*/write* are used extensively
by lib/virtio-mmio.c to access memory mapped registers. AFAICT
we do want serialize accesses to devices and need barriers for
that. The same applies to lib/pci-testdev.c when/if it accepted.
> > Thanks,
> > drew
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
