Re: [Qemu-ppc] [RFC PATCH v2 01/21] ppc/xive: introduce a skeleton for the sPAPR XIVE interrupt controller

[David Gibson]

On Fri, Sep 22, 2017 at 02:42:07PM +0200, Cédric Le Goater wrote:
> On 09/22/2017 01:00 PM, David Gibson wrote:
> > On Tue, Sep 19, 2017 at 03:15:44PM +0200, Cédric Le Goater wrote:
> >> On 09/19/2017 04:27 AM, David Gibson wrote:
> >>> On Mon, Sep 11, 2017 at 07:12:15PM +0200, Cédric Le Goater wrote:
> >>>> Start with a couple of attributes for the XIVE sPAPR controller
> >>>> model. The number of provisionned IRQ is necessary to size the
> >>>> different internal XIVE tables, the number of CPUs is also.
> >>>>
> >>>> Signed-off-by: Cédric Le Goater <address@hidden>
> >>>
> >>> [snip]
> >>>
> >>>> +static void spapr_xive_realize(DeviceState *dev, Error **errp)
> >>>> +{
> >>>> +    sPAPRXive *xive = SPAPR_XIVE(dev);
> >>>> +
> >>>> +    if (!xive->nr_targets) {
> >>>> +        error_setg(errp, "Number of interrupt targets needs to be 
> >>>> greater 0");
> >>>> +        return;
> >>>> +    }
> >>>> +    /* We need to be able to allocate at least the IPIs */
> >>>> +    if (!xive->nr_irqs || xive->nr_irqs < xive->nr_targets) {
> >>>> +        error_setg(errp, "Number of interrupts too small");
> >>>> +        return;
> >>>> +    }
> >>>> +}
> >>>> +
> >>>> +static Property spapr_xive_properties[] = {
> >>>> +    DEFINE_PROP_UINT32("nr-irqs", sPAPRXive, nr_irqs, 0),
> >>>> +    DEFINE_PROP_UINT32("nr-targets", sPAPRXive, nr_targets, 0),
> >>>
> >>> I'm a bit uneasy about the number of targets having to be set in
> >>> advance: this can make life awkward when CPUs are hotplugged.  I know
> >>> there's something similar in xics, but it has caused some hassles, and
> >>> we're starting to move away from it.
> >>>
> >>> Do you really need this?
> >>>
> >>
> >> Some of the internal table size depend on the number of cpus 
> >> defined for the machine.
> > 
> > Which ones?  My impression was that there needed to be at least #cpus
> > * #priority-levels EQs, but there could be more than that, 
> 
> euh no, not in spapr mode at least. There are 8 queues per cpu.

Ok.

> > so it was no longer as tightly bound to the number if "interrupt servers"> 
> > as xics.
> 
> ah. I think I see what you mean, that we could allocate them on the 
> fly when needed by some hcalls ?

Not at hcall time, no, but at cpu hot(un)plug time I was wondering if we
could (de)allocate them then.

> The other place where I use the nr_targets is to provision the 
> IRQ numbers for the IPIs but that could probably be done in some 
> other way, specially it there is a IRQ allocator at the machine
> level.

Hm, ok.

> 
> C.  
> >> When the sPAPRXive object is instantiated, 
> >> we use xics_max_server_number() to get the max number of cpus
> >> provisioned.
> >>
> >> C.
> >>
> > 
> 

-- 
David Gibson                    | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
                                | _way_ _around_!
http://www.ozlabs.org/~dgibson

signature.asc
Description: PGP signature