On Sat, 23 Jun 2007 09:09:51 -0700, David Brownell [off-list ref] wrote:

quoted

And for mmiowb() issue, I put it just only I was not sure whether
gpio_set_value() guarantee I/O barrier.  Now I see i2c-gpio.c, etc. do
not have such barriers.  I will remove these barriers and fix platform
gpio codes.

I don't think this is a case where there'd be a benefit to
allowing non-barriered implementations, and thus requiring
all portable code to include platform-neutral I/O barriers.
I don't know that such neutral primitives actually exist...

I'll update the GPIO docs to make that clear, unless you
have some strong argument to the contrary.

No contrary argument.  Some guide to writers of GPIO implementations
about the I/O barriers should be appreciated.

---
Atsushi Nemoto