On Thu, Apr 08, 2021 at 05:56:00PM +0300, Andy Shevchenko wrote: > It seems that Xilinx GPIO driver operates with bit arrays longer than 32 and > thus can leverage bitmap APIs for that. It makes code better to understand. > > The ->probe() function is modified to try read properties for both channels > since is_dual check makes only sense for the amount of pins used for the > second > channel. On top of that kzalloc() guarantees zero initial values for the > fields > in the private data structure, hence drop unneeded conditionals and > assignments. > > The change is inspired by Syed Nayyar Waris' ideas about bitmap API extension.
As I was afraid in the cover letter, I found some mistakes already. In any case, I'll wait for the comments and test of other patches if possible. Out of curiosity, below I point out the issues. ... > + return bitmap_bitremap(gpio, chip->sw_map, chip->hw_map, > chip->gc.ngpio); Seems we have to use 64 instead of ngpio here. ... > + bitmap_replace(state, chip->state, hw_bits, hw_mask, gc->ngpio); Ditto. ... > + bitmap_copy(chip->state, state, gc->ngpio); Ditto. ... > + for_each_set_bit(bit, all, 64) > + generic_handle_irq(irq_find_mapping(gc->irq.domain, bit)); Here should be used gpio actually and ngpio IIUC. -- With Best Regards, Andy Shevchenko
