Here is a first RFC for the new GPIO interface. It's terribly overdue, and in order to avoid delaying it further I decided to send it as-is instead of being even more perfectionist. The main issue I have with it is that it highlights some points that could be simplified in gpiolib, some of which require a refactoring of gpiolib's clients. I wasn't sure which, of the cleanup or new interface, should come first and stayed in this state of confusion for several months ; but in the end I believe it's more appropriate to clearly lay the interfaces we want to use first, and then refactor the lib accordingly. So please be mindful that another round of cleanup, both of gpiolib itself and its consumers, should come once a later version of these patches is validated (and hopefully merged) and that this series should be considered just another step towards gpiolib's sanitization (although I also secretly hope the new interface will be deemed useful by itself). Another point that definitely needs more attention is the documentation. I am not sure whether the new interface should be described as a couple of sections in the existing GPIO documentation (the current approach) or as a new documentation file of its own. The new gpiod public interface is just an export of how gpiolib now works internally. By representing GPIOs as opaque descriptors intead of integers, it aims at making GPIO usage safer and more coherent with respect to other subsystems by using a proper get/put mechanism and ensuring all GPIOs in use at a given time are valid. It also makes GPIO acquisition easier for consumers, by offering a standard way to bind GPIOs to devices. Finally, it tries to fix some of the inconsistencies gpiolib acquired over time, like the incoherent handling of active low GPIOs which was only valid when using the sysfs interface. The old GPIO interface is still accessible and works the same way as before ; I do not expect any existing code to break as a result of applying this series (which is not extensively tested yet BTW). TODO (preferably once this batch is validated): - Change gpio_get()'s behavior to allow it to return specifiers other than the first one for DT-declared GPIOs? - Update as many users of the legacy interface as possible to use the new interface - Simplify gpiolib by removing redundant code and corner-cases (gpio_request must die) - Try to implement the GPIO fast-path method with gpiod, maybe using the LSB of GPIO descriptors as a marker to forge fake descriptors? - If possible, completely deprecate and remove the old gpio interface Alexandre Courbot (5): gpiolib: factorize gpiod_get/set functions gpiolib: export descriptor-based GPIO interface gpiolib: port of_ functions to use gpiod gpiolib: add gpiod_get() and gpiod_put() functions gpiolib: update documentation Documentation/gpio.txt | 119 ++++++++++ drivers/gpio/gpiolib-of.c | 28 ++- drivers/gpio/gpiolib.c | 525 ++++++++++++++++++++++++++---------------- include/asm-generic/gpio.h | 227 ++++++------------ include/linux/gpio.h | 11 +- include/linux/gpio/consumer.h | 201 ++++++++++++++++ include/linux/gpio/driver.h | 163 +++++++++++++ include/linux/of_gpio.h | 29 ++- 8 files changed, 920 insertions(+), 383 deletions(-) create mode 100644 include/linux/gpio/consumer.h create mode 100644 include/linux/gpio/driver.h -- 1.8.4 -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
