In article <mtu8ylx49ob.fsf@contents-vnder-pressvre.mit.edu>
nathanw@wasabisystems.com wrote:

> However, the bus_space_*() routines are supposed to take care of all
> of this (I believe that Alpha sparse-space complexity was part of the
> motivation for bus_space, in fact), so the suggestions of drivers
> failing because of side-effect reads on an Alpha that don't happen on
> other systems smells fishy to me.

I think the problem of fxp(4) is a race on updating uint16_t members
in a TX DMA descriptor, but it could be resolved by a patch I posted
before which was based on OpenBSD's one.
(FreeBSD uses atomic_clear_16() so it would not work without BWX.)
---
Izumi Tsutsui
tsutsui@ceres.dti.ne.jp