On Thu, 2008-10-30 at 08:45 +0100, Peter Korsgaard wrote:

I would write a dedicated driver for something like that instead of
using gpiolib.

Gpiolib has quite some overhead compared to the actual work for
changing a SoC gpio pin, but it also has some very nice
advantages. For most stuff people use GPIO pins for, the overhead is
not an issue (SPI chip selects, leds, keys, reset signals, ..). It's
very handy that we nowadays have generic drivers that work with any
GPIO (being SoC gpios or stuff on spi/i2c).

For what it's worth, I think it could be useful to have an optimally
fast gpiolib driver.  This would let you have truly generic bit-bang
drivers.  Take a look at drivers/spi for example -- there's ~7 bit-bang
drivers in there that use spi_bitbang.c for a library of functions.  I
am not familiar with these drivers, but at a glance it seems feasible
that with fast gpiolib implementations you could reduce all of those to
a single spi_bitbang.c driver.

Trent, it looks like the big performance wins for you were obtained by
deviating from the gpiolib interface?  In that case a generic SPI/JTAG
bit bang driver wouldn't benefit from the speed boosts in your MPC8572
GPIO driver (without explicitly calling mpc8572_gpio_lock and
_mpc8572_gpio_set, etc).  Is that correct?  In any case, I appreciate
what you did with your driver, so please don't give up on getting some
of your bits merged in :)

-- 
Nate Case [off-list ref]