Re: [PATCH v4] MIPS: Replace setup_irq() by request_irq()

From: Thomas Bogendoerfer
Date: Wed Mar 11 2020 - 12:03:40 EST


On Wed, Mar 11, 2020 at 06:42:10PM +0530, afzal mohammed wrote:
> Hi Thomas,
>
> On Wed, Mar 11, 2020 at 11:42:17AM +0100, Thomas Bogendoerfer wrote:
> > On Wed, Mar 11, 2020 at 02:33:08PM +0530, afzal mohammed wrote:
>
> > > diff --git a/arch/mips/kernel/cevt-r4k.c b/arch/mips/kernel/cevt-r4k.c
>
> > > int r4k_clockevent_init(void)
> > > {
> > > - unsigned long flags = IRQF_PERCPU | IRQF_TIMER | IRQF_SHARED;
> > > + unsigned long flags = IRQF_PERCPU | IRQF_TIMER;
>
> > I don't see why this should help. In my tree only sgi-ip30 removes
> > IRQF_SHARED from flags, but then it uses setup_percpu_irq().
> > What do I miss ?
>
> You did not miss anything. Though it works, i took a wrong route
> following the tags & arrived at that solution in a hurry.
> (struct irqaction used in sgi-ip30 was used here earlier w/ setup_irq).
>
> The problem is sanity checks in request_irq() [ rather in
> request_thread_iq() ]
>
>
> if (((irqflags & IRQF_SHARED) && !dev_id) ||
> (!(irqflags & IRQF_SHARED) && (irqflags & IRQF_COND_SUSPEND)) ||
> ((irqflags & IRQF_NO_SUSPEND) && (irqflags & IRQF_COND_SUSPEND)))
> return -EINVAL;
>
> If IRQF_SHARED is passed, it exepcts a non-NULL dev_id, here it is
> NULL, setup_irq() doesn't have any check like that.

so request_irq() is not a 1:1 replacement for our current setup_irq().
Or put it the another way our setup_irq() might be buggy, when used for
shared interrupts.

> So i think proper solution is to add a non NULL dev_id, as removing
> IRQF_SHARED might affect some platforms that might be using that
> interrupt line shared.
>
> Patch with non-NULL dev_id below, it works w/ Nathan's test case.

I'm not sure, I like the adding of string pointers as dev_id arguments
in your patch. How can we make sure they are unique enough for the use
case ? I guess using handler as dev_id does a better job here.

And before doing that, lets clean up some of the IRQF_SHARED usage first.
All sni IRQF_SHARED can go away, the interrupt lines are exclusive there.

loongson2ef/lemote-2f/irq.c: looks like the only user of
LOONGSON_NORTH_BRIDGE_IRQ, so IRQF_SHARED could go as well.
Could someone confirm that ?

All other need to stay, IMHO.

And v4 is already in mips-next, so I need an incremental patch please.

Thomas.

--
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea. [ RFC1925, 2.3 ]