Re: Linux irq subsys i2c interaction question

[Thomas Gleixner <tglx@linutronix.de>]

Hans,

On Tue, 7 Mar 2017, Hans de Goede wrote:

> I've an "interesting" irq problem. I've an i2c pmic (Intel Cherry Trail
> Whiskey Cove) which itself contains an i2c controller (adapter in Linux
> terms) and has a pin dedicated for raising irqs by the external battery
> charger ic attached to its i2c-adapter.
> 
> To be able to use the irq for the external-charger, the driver for the
> PMIC needs to implement an irqchip and here things get interesting. This
> irqchip can NOT use handle_nested_irq, because the i2c-client driver's
> irq-handler will want to read/write to the external-charger which uses
> the i2c-controller embedded in the PMIC which requires handling of new
> (not arrived when started) PMIC irqs, which cannot be done if the client
> irq-handler is running in handle_nested_irq, because then the PMIC's irq
> handler is already / still running and blocked in the i2c-client's
> irq-handler which is waiting for the new interrupt(s) to get processed to
> signal completion of the i2c-transaction(s) it is doing.

Cute circular dependency.

> I've solved this the following way, which works but
> I wonder if it is the right way to solve this ?
> 
> Note this sits inside the threaded interrupt handler
> for the PMIC irq (after reading and acking the irqs):
> 
>         /*
>          * Do NOT use handle_nested_irq here, the client irq handler will
>          * likely want to do i2c transfers and the i2c controller uses this
>          * interrupt handler as well, so running the client irq handler from
>          * this thread will cause things to lock up.
>          */
>         if (reg & CHT_WC_EXTCHGRIRQ_CLIENT_IRQ) {
>                 /*
>                  * generic_handle_irq expects local irqs to be disabled
>                  * as normally it is called from interrupt context.
>                  */
>                 local_irq_disable();
>                 generic_handle_irq(adap->client_irq);
>                 local_irq_enable();
>         }
> 
> Not really pretty, but it works well enough. I can
> live with this if you can live with it too :)

I'm a bit worried about this being hardcoded for that particular use
case. That also means that you cannot use the generic regmap irq handling
stuff and need to have your own irq magic there. Wouldn't it be smarter to
mark that interrupt in some way and have that as a generic option?

Can you point me at the relevant drivers/patches?

Thanks,

	tglx