Re: [PATCH v2 31/32] sh: support a 2-byte smp_store_mb

["Michael S. Tsirkin" <mst@redhat.com>]

On Wed, Jan 06, 2016 at 12:40:23PM +0100, Peter Zijlstra wrote:
> On Wed, Jan 06, 2016 at 01:19:44PM +0200, Michael S. Tsirkin wrote:
> > On Tue, Jan 05, 2016 at 06:27:35PM -0500, Rich Felker wrote:
> > > On Thu, Dec 31, 2015 at 09:09:47PM +0200, Michael S. Tsirkin wrote:
> > > > At the moment, xchg on sh only supports 4 and 1 byte values, so using it
> > > > from smp_store_mb means attempts to store a 2 byte value using this
> > > > macro fail.
> > > > 
> > > > And happens to be exactly what virtio drivers want to do.
> > > > 
> > > > Check size and fall back to a slower, but safe, WRITE_ONCE+smp_mb.
> > > 
> > > Can you please do this for size 1 as well (i.e. all sizes != 4)? If
> > > you check the source, the code for size-1 xchg in sh cmpxchg-llsc.h is
> > > completely wrong and operates on a 32-bit object at the address passed
> > > to it. This code is presently unused anyway and I plan to submit a
> > > patch to remove the size 1 case.
> > > 
> > > Rich
> > 
> > Ouch. And PeterZ says I should write a 2-byte xchg in asm instead,
> > and Fedora can't even build a full kernel for this arch at the moment :(
> 
> Does the kernel.org hosted cross compiler work?

I'll test, thanks for the hint.

> > Peter, what do you think? How about I leave this patch as is for now?
> 
> No, and I object to removing the single byte implementation too. Either
> remove the full arch or fix xchg() to conform. xchg() should work on all
> native word sizes, for SH that would be 1,2 and 4 bytes.

Rick, maybe you could explain how is current 1 byte xchg on llsc wrong?

It does use 4 byte accesses but IIUC that is all that exists on
this architecture.


-- 
MST