On Wed, 24 Feb 2021 10:27:13 +0100, Jaroslav Kysela wrote: > > Dne 24. 02. 21 v 9:52 Takashi Iwai napsal(a): > > On Wed, 24 Feb 2021 09:14:41 +0100, > > Jaroslav Kysela wrote: > >> > >> Dne 24. 02. 21 v 8:12 Takashi Iwai napsal(a): > >>> On Tue, 23 Feb 2021 21:56:16 +0100, > >>> Jaroslav Kysela wrote: > >>>> > >>>> Dne 23. 02. 21 v 17:20 Takashi Iwai napsal(a): > >>>>>>> Of course, this implementation would make the integration much easier, > >>>>>>> and that's a big benefit. So I have a mixed feeling and not decided > >>>>>>> yet whether we should go for it right now... > >>>>>> > >>>>>> I think that we can reconsider the LED handling implementation later, when > >>>>>> someone brings something better on the table. > >>>>> > >>>>> What worried me is the plan to expose this capability to user-space. > >>>>> If it's only a kernel-internal, we can fix it in the kernel and > >>>>> nothing else broken, but if it's a part of API, that's not easy. > >>>>> > >>>>> So, if any, I'd like to avoid exposing to the user-space at first. > >>>>> (But then it comes to the question how to deal with a case like AMD > >>>>> ACP...) > >>>> > >>>> I tried to propose a complete solution and the ACP was one strong reason for > >>>> this kernel / user space API. So without the user space support, it's just > >>>> a half solution for known issues. > >>>> > >>>> Frankly, I don't see any drawback or a problem even if we remove this API > >>>> later. > >>> > >>> Removing the user-space API is absolutely no-go. The only exception > >>> would be either the case really no one uses it or it's too buggy and > >>> unfixable. > >> > >> This is a special case. Even if those LED bits are ignored by kernel in > >> future, we expect to be replaced with another layer. Thus the functionality > >> must be retained. > > > > Well, we cannot know whether the replacement really happens or > > happened, and hence we never kill the old one. That's the problem. > > > >>>> The LED group bits are just informal for the user space and it's > >>>> expected to create the user controls tied to this LED functionality only in > >>>> alsa-lib/plugins at the moment. The kernel may return an error when the user > >>>> space tries to set those new bits when the API is deprecated and I believe > >>>> that the hardware design faults like AMD ACP (without the hardware mute) are rare. > >>> > >>> The experience tells us that users are creative enough to (ab)use a > >>> new ABI in any unexpected ways, and we have no control for it. So > >>> it's not about how alsa-lib is implemented but rather how ABI could be > >>> abused :) > >> > >> Ok, I don't have other ideas. I don't agree with your argumentation for this > >> particular case, where the functionality is marginal. Ideally, the AMD driver > >> may be recoded to use double-buffering and software mute switch, so we should > >> handle everything in the kernel space. > > > > My argument is that we're trying to add too much freedom just for this > > "marginal" problem. Honestly speaking, I would feel rather more > > comfortable if it were a kernel control element that just does trigger > > the LED like the original patch from AMD guys. Then you cannot do > > much wrong. OTOH, creating a virtual capture switch and let alsa-lib > > handling the software mute, while PA should ignores the soft-mute but > > We can force the softvol even if PA set the skip flag for this particular PCM > stream. > > > dealing only with the assigned mute LED... Sounds too complex to me. > > It seems that you misunderstood the number of issues which my code is trying > to resolve: > > 1) set LED based on state from multiple cards (so you cannot trigger LED > inside single driver / single control element); we need one arbiter; this is > the main argument > 2) unifies the audio LED interface > 3) reduce the hardware driver code Those purposes are all fine. But they don't need to be exposed for user controls that can be abused. That's the very concern. Takashi