On Fri, Feb 17, 2017 at 1:32 PM, Phil Reid [off-list ref] wrote:

On 17/02/2017 17:23, Andy Shevchenko wrote:

quoted

On Fri, Feb 17, 2017 at 11:12 AM, Phil Reid [off-list ref]
wrote:

quoted

When a threaded irq handler is chained attached to one of the gpio
pins when configure for level irq the altera_gpio_irq_leveL_high_handler
does not mask the interrupt while being handled by the chained irq.
This resulting in the threaded irq not getting enough cycles to complete
quickly enough before the irq was disabled as faulty.
It looks like handle_level_irq should be used in this situation
instead of handle_simple_irq.

quoted

@@ -310,7 +310,8 @@ static int altera_gpio_probe(struct platform_device

*pdev)
        altera_gc->interrupt_trigger = reg;

        ret = gpiochip_irqchip_add(&altera_gc->mmchip.gc,
&altera_irq_chip, 0,
-               handle_simple_irq, IRQ_TYPE_NONE);
+               altera_gc->interrupt_trigger == IRQ_TYPE_LEVEL_HIGH ?
+               handle_level_irq : handle_simple_irq, IRQ_TYPE_NONE);

AFAIK, handle_bad_irq() should be used here.

G'day Andy

Grepping drivers/gpio find a combination of
handle_simple_irq
handle_level_irq
handle_edge_irq
handle_bad_irq
used in gpiochip_irqchip_add

Try to add date of change and amount of use.
I bet handle_bad_irq() would be the winner.

The ones which use handle_bad_irq call irq_set_handler_locked in their
irq_type callback to either handle_level_irq / handle_edge_irq

Yep.

So I think in this case it's correct. But I'm no expert.

I dunno.

-- 
With Best Regards,
Andy Shevchenko