On Thu, 5 Jan 2006, Manuel Bouyer wrote:

> I had a quick look at the sparc64's bus_space(9) implementation,
> and could'nt figure out what kind of barriers are enforced by
> bus_space_read and bus_space_write, if any.
> That is, if a driver do 2 consecutive bus_space_write, are we sure that
> the writes will reach the device in order ?
> Also it seems the implementation is different for 32 and 64bit kernels,
> so the behavior may be different.

sparc64 bus_space operations have no implicit barriers.
You need to use explicit barriers.

Eduardo