07:07 <mmind00> lag: do mfd devices say anything about exclusive use of sub-devices (one or the other, never at the same time) ?

07:07 <mmind00> code in question is the new Rockchip pwm block that can do either pwm-output or counter-input and it looks like the current proposed patch just reimplements a mfd ( https://lore.kernel.org/linux-rockchip/20250602-rk3576-pwm-v2-4-a6434b0ce60c@collabora.com/ )

07:07 <mmind00> in v1 of the series a rationale was given for that ( https://lore.kernel.org/all/13790724.uLZWGnKmhe@workhorse/ ) - though still this looks like mostly a MFD to me :-)

07:14 <lag> mmind00: Looks like an MFD to me

07:14 <lag> mmind00: Yes, you can use an MFD core driver to control state / manage single-use resources

07:16 <lag> "The most apt choice seems to be undocumented, so I'll opt to hand-roll my own implementation instead, and not document it" - nice!

07:20 <mmind00> lag: thanks a lot for the confirmation

09:39 <linusw__> paulk: OK I will get to it.

12:59 <CounterPillow> lag: I did not know it's the most apt choice, as it is undocumented.

13:01 <CounterPillow> if you want people to use your shit maybe explain it somewhere other than a mailing list.

13:10 <CounterPillow> lag: I've even Cc'd you to ask about auxbus/platform bus/mfd in the cover letter with a big "AUXBUS/MFD/PLATFORM BUS DISCUSSION" paragraph as the second paragraph, since it appears you prefer answering the same questions over e-mail repeatedly for decades instead of documenting it in-tree. However, you did not respond.

13:17 <ukleinek> CounterPillow: he just explained it in irc! :-D *duck*

13:18 <CounterPillow> Yeah, to someone other than me ;)

13:39 <lag> a) It's not "my shit" - it's a community project - I just help to look after it

13:39 <lag> b) I do not WANT you to use it - the less inept people I have to deal with, the better

13:39 <lag> c) This 20 year old subsystem is probably undocumented for a reason, most likely because it's simple enough to understand that documentation is not required d) If you _really_ can't live without documentation, write some - patches always welcome

13:40 <lag> CounterPillow: ^

13:41 <lag> s/ d)/\n d)/

14:24 <lag> CounterPillow: Also, I just looked up your mail - a) what makes you think I would find a hidden paragraph inside the cover letter of a random patch-set that looked like it had nothing to do with me? b) You added that paragraph to a submission containing your already authored solution - if you were confused about which API use, it wouldn't it have been better to send a mail with an appropriate subject line _before_ choosing?

14:50 <ukleinek> linusw__: would you support this patch even though it changes the return value of a direction_input request?: https://paste.debian.net/hidden/e3e09dff/

14:53 <ukleinek> ah, that triggers a WARN_ON, so probably not a good idea to drop it

19:58 <CounterPillow> lag: the code does not explain which of auxbus, platform bus and mfd is the right choice for a given device. That's the kind of thing documentation is good at: explaining high-level concepts to stop abuse of subsystems to fit a certain other thing. Pissing on my leg for not writing said documentation is funny because 1. I was in fact the very person who needed this documentation, so not in a good spot to be the one making authoritative

19:58 <CounterPillow> statements on the truth here, 2. I've already been spending a significant amount of time writing docs patches lately because none of you do it. I did not send you an e-mail with the appropriate subject line before choosing because I'd rather not be blocked on work by the common occurrence of people not responding.

21:09 <Forty-Bot> I did a mutually-exclusive pwm/timer and I just differentiated the driver by making them check for #pwm-cells

21:09 <Forty-Bot> but I agree we are sorely in need of better docs

21:14 <Forty-Bot> IMO auxbus is more flexible than MFD, so if you need to coordinate between devices or have some special probe sequence it can adapt to that

21:17 <Forty-Bot> but it's... crustier

21:18 <CounterPillow> isn't auxbus newer?

21:18 <Forty-Bot> they're both at least a decade old at this point

21:19 <Forty-Bot> it's just that every new SoC comes with a set of registers where the hardware designers stuck all the miscellaneous stuff

21:19 <Forty-Bot> and that maps well onto the MFD abstraction

21:19 <Forty-Bot> so MFDs get more attention