Re: [PATCH v5 1/3] dt-bindings: interrupt-controller: Add Actions SIRQ controller binding

From: Marc Zyngier
Date: Thu Aug 27 2020 - 11:42:12 EST


Cristian,

On 2020-08-27 16:24, Cristian Ciocaltea wrote:
Hi Marc,

On Thu, Aug 27, 2020 at 11:35:06AM +0100, Marc Zyngier wrote:
On 2020-08-27 11:06, Cristian Ciocaltea wrote:
> On Wed, Aug 26, 2020 at 04:48:38PM -0600, Rob Herring wrote:
> > On Wed, Aug 26, 2020 at 3:42 PM Cristian Ciocaltea
> > <cristian.ciocaltea@xxxxxxxxx> wrote:

[...]

> > Ultimately the GIC trigger type has to be
> > something. Is it fixed or passed thru? If the latter, just use 0
> > (IRQ_TYPE_NONE) if the GIC trigger mode is not fixed. Having some sort
> > of translation of the trigger is pretty common.
>
> Yes, as explained above, the SIRQ controller performs indeed the
> translation of the incoming signal. So if I understand correctly, your
> suggestion would be to use the following inside the sirq node:
>
> interrupts = <GIC_SPI 13 IRQ_TYPE_NONE>, /* SIRQ0 */
> [...]

Please don't. If you are describing a GIC interrupt, use a
trigger that actually exists. Given that you have a 1:1
mapping between input and output, just encode the output
trigger that matches the input.

Understood, the only remark here is that internally, the driver will
not use this information and instead will continue to rely on the input
to properly set the trigger type for the output.

It's fine. The binding has to be consistent on its own, but
doesn't dictate the way the driver does thing.

The question is if the driver should also emit a warning (or error?)
when the trigger type supplied via DT doesn't match the expected value.

Rob will tell you that the kernel isn't a validation tool for broken
DTs. Shout if you want, but you are allowed to simply ignore the
output trigger for example

If yes, we should also clarify what the user is supposed to provide in
the controller node: the trigger type before the conversion (the input)
or the one after the conversion (the output).

The output of a SIRQ should be compatible with the GIC input it is
attached to. You can have:

device (LEVEL_LOW) -> SIRQ (LEVEL_HIGH) -> GIC

but you can't have:

device (LEVEL_LOW) -> SIRQ (EDGE_RISING) -> GIC

because that's not an acceptable transformation for the SIRQ,
nor can you have:

device (EDGE_FALLING) -> SIRQ (EDGE_FALLING) -> GIC

because EDGE_FALLING isn't a valid input for the GIC.

In both of the invalid cases, you would be free to apply
which ever transformation actually makes sense, and shout
at the user if you want to help them debugging their turf.
The later part is definitely optional.

Hope this helps,

M.
--
Jazz is not dead. It just smells funny...