Re: [PATCH] xen: use PHYSDEVOP_get_free_pirq to implement find_unbound_pirq

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Sat, Nov 20, 2010 at 01:19:31AM +0000, Stefano Stabellini wrote:
> On Fri, 19 Nov 2010, Konrad Rzeszutek Wilk wrote:
> > On Fri, Nov 19, 2010 at 01:58:03PM +0000, Stefano Stabellini wrote:
> > > Use PHYSDEVOP_get_free_pirq to implement find_unbound_pirq
> > > 
> > > Use the new hypercall PHYSDEVOP_get_free_pirq to ask Xen to allocate a
> > > pirq. Remove the unsupported PHYSDEVOP_get_nr_pirqs hypercall to get the
> > > amount of pirq available.
> > > 
> > > Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
> > > 
> > > diff --git a/drivers/xen/events.c b/drivers/xen/events.c
> > > index 321a0c8..ffd286e 100644
> > > --- a/drivers/xen/events.c
> > > +++ b/drivers/xen/events.c
> > > @@ -382,12 +382,17 @@ static int get_nr_hw_irqs(void)
> > >  	return ret;
> > >  }
> > >  
> > > -/* callers of this function should make sure that PHYSDEVOP_get_nr_pirqs
> > > - * succeeded otherwise nr_pirqs won't hold the right value */
> > > -static int find_unbound_pirq(void)
> > > +static int find_unbound_pirq(int type)
> > >  {
> > > -	int i;
> > > -	for (i = nr_pirqs-1; i >= 0; i--) {
> > > +	int rc, i;
> > > +	struct physdev_get_free_pirq op_get_free_pirq;
> > > +	op_get_free_pirq.type = type;
> > > +
> > > +	rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
> > > +	if (!rc)
> > > +		return op_get_free_pirq.pirq;
> > > +
> > > +	for (i = 16; i <= nr_pirqs-1; i++) {
> > 
> > 16? No no. Why not re-use the old loop, like so:
> > 
> >  	for (i = nr_pirqs-1; i >= 0; i--) {
> > 
> 
> Because we don't know the real nr_pirqs anymore (PHYSDEVOP_get_nr_pirqs
> has been removed), so it is highly possible that starting from the top
> down would give us pirq numbers out of range in Xen.  Therefore we need
> to start from the bottom up, and the bottom for Xen is 16.

Right, I forgot about the hypercall call. How about using LEGACY_IRQ instead then?

> 
> 
> > >  		if (pirq_to_irq[i] < 0)
> > >  			return i;
> > >  	}
> > > @@ -669,7 +674,7 @@ void xen_allocate_pirq_msi(char *name, int *irq, int *pirq)
> > >  	if (*irq == -1)
> > >  		goto out;
> > >  
> > > -	*pirq = find_unbound_pirq();
> > > +	*pirq = find_unbound_pirq(MAP_PIRQ_TYPE_MSI);
> > >  	if (*pirq == -1)
> > >  		goto out;
> > >  
> > > @@ -1504,23 +1509,12 @@ void xen_callback_vector(void) {}
> > >  void __init xen_init_IRQ(void)
> > >  {
> > >  	int i, rc;
> > > -	struct physdev_nr_pirqs op_nr_pirqs;
> > >  
> > >  	cpu_evtchn_mask_p = kcalloc(nr_cpu_ids, sizeof(struct cpu_evtchn_s),
> > >  				    GFP_KERNEL);
> > >  	irq_info = kcalloc(nr_irqs, sizeof(*irq_info), GFP_KERNEL);
> > >  
> > > -	rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_nr_pirqs, &op_nr_pirqs);
> > > -	if (rc < 0) {
> > > -		nr_pirqs = nr_irqs;
> > > -		if (rc != -ENOSYS)
> > > -			printk(KERN_WARNING "PHYSDEVOP_get_nr_pirqs returned rc=%d\n", rc);
> > > -	} else {
> > > -		if (xen_pv_domain() && !xen_initial_domain())
> > > -			nr_pirqs = max((int)op_nr_pirqs.nr_pirqs, nr_irqs);
> > > -		else
> > > -			nr_pirqs = op_nr_pirqs.nr_pirqs;
> > > -	}
> > > +	nr_pirqs = nr_irqs;
> > 
> > Why not just get rid of nr_pirgs altogether then? And use 'nr_irqs' instead?
> > 
> 
> Yeah, I guess we could do that. I kept it around just to make it more
> obvious that the max pirq number is different from nr_irqs and we don't
> know what the exact value is.


But with this change it is the same, is it not?

	"nr_pirgs = nr_irqs;"

Or when you say "max pirq" you are referring to something else altogether?