Hi, On Thu, Oct 30, 2008 at 06:13:05PM +0100, Florian Fainelli wrote: > Le Wednesday 29 October 2008 22:10:46 Phil Sutter, vous avez écrit : > > Yes it does, but that's not part of gpiolib itself. Accessing them needs > > a combination of gpio_to_chip() and container_of() to be used, which I > > doubt makes sense on a device with a single, platform gpio chip. > > Yes, that makes it unexportable the way it is done yet. What I suggest is not > overriding the struct rb532_gpio_chip with thoses callbacks, but do like you > suggested initially. > > > I'm not sure if this is absolutely true. The original CompactFlash > > driver e.g. clears interrupt level in cf_irq_handler() and sets it in > > prepare_cf_irq(). The latter function is called more than once. > > This should be moved the IRQ handler, where a specific check for the IRQ being > a GPIO one should set the interrupt status and level accordingly. Sounds reasonable to me. So I'll prepare a patch which: * removes the function pointers from rb532_gpio_chip * exports getters and setters for interrupt status and level The GPIO config and function registers should indeed only be accessed at bootup, so the corresponding getters/setters will be only locally accessible in arch/mips/rb532/gpio.c. Doing it this way also prevents any driver from breaking others, as the complete GPIO configuration will be done at a single place (i.e., inside gpio.c). IIRC pata-rb532-cf did use them only at initialisation state and to prevent the described situation. I got the pata driver working by the way. I just wanted to get a solution for the GPIO problem first, as it needs to access interrupt level and status at least. Greetings, Phil
