Re: [PATCH V4 04/14] genirq: Generic chip: Add big endian I/O accessors

[Boris Brezillon <boris.brezillon@free-electrons.com>]

Hi,

On Tue, 11 Nov 2014 00:03:01 +0100
Alexandre Belloni <alexandre.belloni@free-electrons.com> wrote:

> Adding Boris in Cc: as he wrote that part.

Thanks for putting me in the loop.

> 
> On 10/11/2014 at 14:11:44 -0800, Kevin Cernekee wrote :
> > On Mon, Nov 10, 2014 at 2:00 PM, Kevin Hilman <khilman@kernel.org> wrote:
> > > Kevin Cernekee <cernekee@gmail.com> writes:
> > >
> > >> Use io{read,write}32be if the caller specified IRQ_GC_BE_IO when creating
> > >> the irqchip.
> > >>
> > >> Signed-off-by: Kevin Cernekee <cernekee@gmail.com>
> > >
> > > I bisected a couple ARM boot failures in next-20141110 on atmel sama5 platforms down to
> > > this patch, though I'm not quite yet sure how it's causing the failure.
> > > I'm not getting any console output, so haven't been able to dig deeper
> > > yet.  Maybe the atmel maintainers (Cc'd) can help dig.
> > >
> > > I've confirmed that reverting $SUBJECT patch (commit
> > > b79055952badbd73710685643bab44104f2509ea2) on top of next-20141110 gets
> > > things booting again.
> > >
> > > Also, it only happens with sama5_defconfig, not with multi_v7_defconfig.
> > 
> > In drivers/irqchip/irq-atmel-aic-common.c I see:
> > 
> >         ret = irq_alloc_domain_generic_chips(domain, 32, 1, name,
> >                                              handle_level_irq, 0, 0,
> >                                              IRQCHIP_SKIP_SET_WAKE);
> > 
> > and IRQCHIP_SKIP_SET_WAKE is (1 << 4), same as IRQ_GC_BE_IO.
> > 
> > Is it possible that the caller is passing values intended for
> > irq_chip->flags into a function expecting
> > irq_domain_chip_generic->gc_flags ?

Indeed, I don't know what I tried to do in the first place but this is
completely wrong.
First because the last argument is not a valid flag as you pointed out,
then because I clearly have set irq_set_wake and thus setting
IRQCHIP_SKIP_SET_WAKE makes no sense.

I also realized I should directly pass handle_fasteoi_irq and not
handle_level_irq for the handler, that clr flags (IRQ_NOREQUEST |
IRQ_NOPROBE | IRQ_NOAUTOEN) are missing and that IRQ_GC_INIT_MASK_CACHE
is missing too.

I'll propose a patch fixing all those bugs.

Sorry for the inconvenience :-(.

Regards,

Boris

-- 
Boris Brezillon, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com