Re: [PATCH 2/4 v2] irqdomain: augment add_simple() to allocate descs

[Rob Herring <robherring2@gmail.com>]

On 11/26/2012 06:13 PM, Linus Walleij wrote:
> On Mon, Nov 26, 2012 at 9:26 PM, Grant Likely <grant.likely@secretlab.ca> wrote:
> 
>>> +                     if (irq_base < 0) {
>>> +                             WARN(1, "Cannot allocate irq_descs @ IRQ%d, assuming pre-allocated\n",
>>> +                                  first_irq);
>>> +                             irq_base = first_irq;
>>
>> As I just commented on the previous version, WARN() is probably too
>> verbose (and scary). Make it an informational.
> 
> So the discussion began with me removing exactly that kind of WARN()
> from arch/arm/common/gic.c:
> http://marc.info/?l=linux-arm-kernel&m=134860088710574&w=2
> 
> Which was NACKed by Rob:
> http://marc.info/?l=linux-arm-kernel&m=134860136515611&w=2
> Who prefered to leave it in to encourage platforms to get fixed.
> 
> This code just follows exactly that pattern.
> 
> I'm happy to patch out *both* (or rather patch gic.c to use
> irq_domain_add_simple()) because I never quite liked
> it in the first place.
> 
>> However, I see another problem. What is the requested range straddles
>> the boundary between reserved and non-reserved IRQs? It would be good to
>> give some information about which irq range was requested and maybe
>> report which ones were available.... or check to see if the request is
>> inside or partially inside the reserved region?
> 
> Right now the usual symptom of that is that the system hangs.
> 
> Do you mean we should probe around a bit with
> irq_get_next_irq() to figure out more precisely what the
> problem is, or did you have something more elegant
> in mind?

My objection was removing completely (which a pr_debug effectively
does). I think Grant is saying just make the warning more informative
about why it failed which is fine with me. nr_irqs is already printed
out, so that provides some info already (although it is pretty terse).

Rob