Hi Jacek and Pavel, On 13 May 2018 at 04:44, Jacek Anaszewski <jacek.anaszewski@xxxxxxxxx> wrote: > Hi Pavel, > > > On 05/12/2018 10:35 AM, Pavel Machek wrote: >> >> Hi! >> >>>>> I disagree here. We already had the same discussion at the occasion >>>>> of the patch [0] and it turned out to be a dead-end [1]. Now we have >>>>> neither the driver nor the generic pattern interface. >>>>> >>>>> We also already have some older LED class drivers that implement custom >>>>> pattern interfaces (e.g. drivers/leds/leds-lm3533.c) and the same >>>>> approach can be applied in this case. >>>> >>>> >>>> Please don't. It was mistake to implement custom pattern interfaces >>>> back then, it is still mistake now. >>> >>> >>> It turned out to be really hard to cover all known pattern generator >>> implementations with generic interface. Sure, it would be nice to have >>> one, but the whole discussion around [0] only unveiled the diversity of >>> parameters to cover. And still new devices appear on the market. >>> >>> We would have to propose a set of pattern schemes and allow to >>> add new ones to it. >> >> >> I believe that what I'm proposing below is close enough to universal. >> >>>> If we really need solution now, I'd recommend "pattern" file with >>>> >>>> "<delta time> <brightness> <delta time> <brightness>". >>>> >>>> In this specific case, hardware only supports patterns in this format: >>>> >>>> low_time 0 rise_time 255 high_time 255 fall_time 0 >>>> >>>> so driver would simply -EINVAL on anything else. >>> >>> >>> I'm fine with the pattern file, but the pattern format would have >>> to be defined in the per-driver ABI documentation. It wouldn't much >>> differ from the custom pattern approach though, unless I'm missing some >>> gain of having pattern setting in a uniformly named single sysfs file >>> (with semantics differing from driver to driver). >> >> >> I'm proposing "<delta time> <brightness> ..." sysfs file. It certainly >> covers this hardware, it would be enough to cover the Qualcomm Pulse >> generator (IIRC), and it would cover most uses cases of Nokia N900's >> LED. >> >> Yes, we would need to document limitations of each chip. But it should >> be easily possible to run pattern designed for Spreadtrum on N900, >> even if it would not work the other way around. >> >> (If someone really wants to run complex patterns on simple hardware, >> we can provide software emulation using same file format. I believe I >> still have that patch somewhere.) > > > OK, I've revised the discussion under Qualcomm LPG patch set and > it seems that we have almost ready solution in [0], except the > pattern_repeat file you mention in [1]. So probably Baolin could > address your remarks from [1] and add pattern_repeat file to the > patch that begins thread [0]. > > [0] https://lkml.org/lkml/2017/11/15/27 > [1] https://lkml.org/lkml/2017/12/8/470 Thanks for your suggestion. So I will remove the sysfs part from the new driver, then send incremental patches when introducing some common LED interfaces. -- Baolin.wang Best Regards -- 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
