On 13 May 2006, at 13:27, Andrew Morton wrote:
>
>> + init_evtchn_cpu_bindings();
>> +
>> + /* No VIRQ or IPI bindings. */
>> + for (cpu = 0; cpu < NR_CPUS; cpu++) {
>
> Using NR_CPUS is a little... old-fashioned. I'd suggest a sweep
> through
> all the Xen code, look for places where it should be using
> for_each_foo_cpu().
Actually that's a particularly good catch in this case, since we use
per_cpu() inside the loop and that's only well defined for
cpu_possible_map. Oops.
The elusive users of ring.h are our split device drivers. It hides a
bunch of details about muxing requests and responses on the same ring,
and notification thresholds. There are a few other places we have ring
buffers but they are sufficiently simple that implementing in place is
clearer.
Thanks,
Keir
