svn commit: r301197 - head/sys/dev/xen/netfront

[Roger Pau Monné]

On Thu, Jun 02, 2016 at 03:01:03PM +0200, Hans Petter Selasky wrote:
> On 06/02/16 14:54, Roger Pau Monné wrote:
> > On Thu, Jun 02, 2016 at 01:19:56PM +0200, Hans Petter Selasky wrote:
> > > On 06/02/16 13:14, Roger Pau Monné wrote:
> > > > +		callout_reset(&rxq->rx_refill, hz/10, xn_alloc_rx_buffers_callout,
> > > > +		    rxq);
> > > 
> > > Maybe use callout_reset_curcpu() to take advantage of callout's SMP
> > > capabilities ?
> > 
> > Yes, that's fine. But what's the benefit of it? I don't really care whether
> > the callout is run on the current CPU or not. Is callout_reset_curcpu
> > cheaper than callout_reset?
> > 
> 
> Hi,
> 
> It is maybe not cheaper, but it will distribute the load of the
> xn_alloc_rx_buffers_callout() callback, to the current CPU calling
> callout_reset_curcpu(). Else xn_alloc_rx_buffers_callout() will always be
> called from callback thread zero.

Thanks for the clarification. I did get the impression that callout_reset 
already distributed the callbacks across the number of available CPUs, maybe 
the man page should be expanded to explain this?

I've committed the change as r301204.

Roger.