Re: [RFC PATCH v2 07/10] irqchip: atmel-aic: document new dt properties and children nodes

[Boris BREZILLON]

Hello Thomas,

Le 29/03/2014 10:19, Thomas Petazzoni a écrit :
> Dear Boris BREZILLON,
>
> On Fri, 28 Mar 2014 18:59:05 +0100, Boris BREZILLON wrote:
>
> > +Optional children nodes:
> > +- muxed irq entries:
> > +  Required properties:
> > +   * compatible: Shall be
> > +     "atmel,aic-mux-1reg-irq": irq enable/disable/retrieve-status is done by
> > +     setting/clearing/reading flags in a specific register
> > +     or
> > +     "atmel,aic-mux-3reg-irq": irq enable/disable/retrieve-status is done
> > +     by writing/reading flags in specific enable/disable/mask registers
> > +   * reg: encode the interrupt control register.
> > +     The first cell encode the irq line.
> > +     The second cell encode the offset register within its iomem range
> > +     The last cell encode the iomem region size (should always be set to 0x4).
> > +   * atmel,aic-mux-reg-mask: define the mask used to disable the interrupts
> > +     generated by the muxed entry.
> Can you describe in more details what are these muxed irqs? Are they
> interrupts raised to the AIC that may actually be related to several
> devices, like a shared interrupt?

Exactly, muxed irqs are shared irqs.

> If that's the case, then what you want is to implement separate
> interrupt controller drivers to handle those shared interrupts, and
> demux them into multiple separate interrupts.

It may work (I'll take a look), but we may have to modify several 
drivers (and
it may break the DT ABI).

> Note that the way you use the "ranges" property seems wrong to me:
> you're using it as a "hack" to define the base address of some
> peripherals that are outside the AIC, while the ranges property is
> normally used to describe the address translations between a child bus
> and a parent bus. Which is not what you have here, as far as I can
> understand.

This is clearly not a real bus, but more a virtual/conceptual bus where each
irq source is a bus device.

> So could you give more details about the design of the AIC and these
> muxed interrupts, to see if the DT binding you're proposing is actually
> the right way of representing the hardware?

The AIC controller muxes several peripheral irqs to one irq line.
If a given peripheral has a pending irq, and the driver requesting this 
irq is
not loaded (either because it was not enabled or because it is not 
loaded yet)
it may generate spurious interrupts (or even hang forever), waiting for 
someone
clearing/reading the interrupt flag(s).

See this thread for an example of what can happen if we don't disable 
all muxed
interrupts before enabling an IRQ line:

https://lkml.org/lkml/2013/3/8/176


In this series' implementation, I define all the muxed (or shared) interrupt
entries (using the DT) and let the AIC driver disable all the interrupt 
sources
when shutting down an interrupt line or before starting the AIC controller.

This provides a generic solution to avoid these spurious interrupt issues
instead of adding several hooks in the machine specific code (one for each
impacted peripheral).


Please tell me if you see a better solution (but keep in mind, this 
should be
done during early init, because the at91 init timer is using a muxed irq 
line
on almost all at91 SoCs).

Best Regards,

Boris
> Thomas

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/