On 30/09/12 11:35, Stijn Devriendt wrote: > In our kernel tree we have similar code. If you like I can request > permission > to share. I can, however, already give you an update on the basic > structure, perhaps > it's useful now. > > For the first part, the drivers need to implement a the gpio interface > for groups. > gpio_set_multi, gpio_get_multi, gpio_direction_input_multi, > gpio_direction_output_multi. Each of them gets a 'u32 mask'. > > Secondly, gpiolib gets some new code to handle groups: > groups are requested via a list of gpio ids. Mind that order is respected: > request( [1, 5, 2, 4] ) followed by a set(0x5) will translate to > gpio_set_multi( 0x18 ). An opaque gpio_group struct is used to keep track. > This means the gpiolib interface also has a u32 mask, but translation is > done for the gpio-drivers. > > There is some code to request groups via device-tree (again respecting > order) > and there are also platform driver structures. > > gpiolib was also extended to export groups into sysfs, respecting policy > (input, output, user-selectable) and to make softlinks to groups in other > driver's subdir. (One driver we use this in is a power-sequencer with 2 > gpios selecting a margining profile, this driver then has the gpio_group > exported in it's sysfs dir as .../profile, allowing H/W engineers to select > the profile without voltage glitches) > > There's also a separate driver, that does nothing more than exporting > both individual pins and groups to userspace based on platform description > or devicetree. This is probably less interesting for mainline, since we're > abusing device-tree to do away with some init script that can do the same. > > The rationale behind a 32bit mask is that typical processors can at most > set one processor-word worth of GPIOs at once and there are probably > few chips with over 32GPIOs on a single gpio_chip anyway. > Nevertheless, in the era of 64bit, it's definitely possible to go for > u64 instead.
Hi Stijn, thank you for your notes! Besides what I discussed with JC and Linus, I find the unsigned int (i.e. u32 or u64, depending on the arch) quite appealing. It is a nice compromise between my general bit mapped data model (variable size *u8 array) and the bool *values list. Even maps easily onto a single sysfs entry for values, by abstracting a gpio list to an actual data word. What do others think? JC? Linus? I'm considering this (unsigned int data) a valid option. One question: How did you solve the one-value-per-file in the sysfs interface? Thanks in advance! Roland -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/