lkml.org 
[lkml]   [2021]   [Dec]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH 5/6] irqchip/apple-aic: Support multiple dies
On Sat, 18 Dec 2021 05:39:04 +0000,
Hector Martin <marcan@marcan.st> wrote:
>
> On 14/12/2021 01.10, Marc Zyngier wrote:
> >> switch (fwspec->param[0]) {
> >> case AIC_IRQ:
> >> - if (fwspec->param[1] >= ic->nr_irq)
> >> + if (die >= ic->nr_die)
> >> + return -EINVAL;
> >> + if (args[0] >= ic->nr_irq)
> >> return -EINVAL;
> >> - *hwirq = (FIELD_PREP(AIC_EVENT_TYPE, AIC_EVENT_TYPE_HW) |
> >> - FIELD_PREP(AIC_EVENT_NUM, fwspec->param[1]));
> >> + *hwirq = AIC_IRQ_HWIRQ(die, args[0]);
> >> break;
> >
> > A side issue with this is that it breaks MSIs, due to the way we
> > construct the intspec (I did hit that when upgrading the M1 intspec to
> > 4 cells for the PMU). I have the following hack locally:
> >
> > diff --git a/drivers/pci/controller/pcie-apple.c b/drivers/pci/controller/pcie-apple.c
> > index b090924b41fe..f7b4a67b13cf 100644
> > --- a/drivers/pci/controller/pcie-apple.c
> > +++ b/drivers/pci/controller/pcie-apple.c
> > @@ -218,7 +218,7 @@ static int apple_msi_domain_alloc(struct irq_domain *domain, unsigned int virq,
> > if (hwirq < 0)
> > return -ENOSPC;
> >
> > - fwspec.param[1] += hwirq;
> > + fwspec.param[1 + (fwspec.param_count == 4)] += hwirq;
> >
> > ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, &fwspec);
> > if (ret)
> >
>
> Heh, I never noticed; probably because I guess the SD card reader on the
> machine I've been testing this on doesn't use MSIs, and I haven't tried
> WiFi yet.
>
> Perhaps (fwspec.param_count - 2)? It's probably a safer long-term
> assumption that the last two cells will always be leaf IRQ number and type.

Yup, that'd work as well, as long as we make this assumption explicit.

M.

--
Without deviation from the norm, progress is not possible.

\
 
 \ /
  Last update: 2021-12-20 14:39    [W:0.057 / U:0.200 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site