Am 31.07.2013 10:35, schrieb Javier Martinez Canillas:

The problem is that board files and drivers that has not not been completed
migrated to DT assumes (at least for OMAP) that *every* GPIO line is mapped as
an IRQ and they just do:

gpio_request(gpio,...);
gpio_direction_input()
request[_threaded]_irq(gpio_to_irq(gpio), ...);

My patch-set changed the gpio-omap driver to not map all GPIO lines but only the
ones that were really used as an IRQ and let the DT core to do the mapping from
irq_create_of_mapping(). The first problem reported with the OMAP patch was that
a driver was using the above sequence and that the GPIO had not been mapped.
This user was booting with DT and so this showed a bug in the driver and a DT
that did not conform with the standard schema used in mainline but this shows a
potential issue.

There must have been a bug in the patch too. I've also added that 
iinterrupt-parent stuff (with the same flags as used by the driver) and 
just have let the driver call

request_threaded_irq(gpio_to_irq(gpio), flags);

without the gpio_request()/input() before. And request_threaded_irq() 
returned -EBUSY. I'm pretty sure nothing else did use that gpio, but I 
haven't looked at why request_threaded_irq() returned -EBUSY. I assume 
the new mapping stuff did reserve the irq in such a way, that the driver 
couldn't request the IRQ.

Otherwise I wouldn't have had a problem by just adding the necessary 
entries to the DT.

But I have to say I didn't like the syntax too, and it wasn't obvious 
how the syntax is and how to conclude from a gpio number to an 
irq-number and the patch didn't really include some documentation or 
useful example.

Regards,

Alexander Holler