Re: [RFC] Reorganizing HD-audio driver code?

[Takashi Iwai <tiwai@suse.de>]

On Fri, 27 Jun 2025 15:22:20 +0200,
Richard Fitzgerald wrote:
> 
> On 27/06/2025 1:04 pm, Takashi Iwai wrote:
> > Hi,
> > 
> > HD-audio driver is known to be quite messy in both file structures and
> > its design, but until now I haven't touched its files paths so much
> > because I set a higher priority for the easiness of backport to stable
> > kernels.  But, you can't leave garbages forever, it's been already
> > high time for a large clean up.
> > 
> > So I tried a quick code reorganization, and put the result in
> > test/hda-reorg branch of sound.git tree.
> >    https://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound.git/log/?h=test/hda-reorg
> > 
> > The basic idea is to move the code from sound/pci/hda/* into different
> > subdirectories in sound/hda/ per functionality, as most of the stuff
> > are independent from PCI, but rather HD-audio bus specific.
> > 
> 
> This all seems reasonable to me. I always thought it strange that there
> is a sound/hda directory but most of the HDA support isn't in that
> directory.

Thanks for your review!

It's a long story: at the time of ASoC Intel HD-audio support, we
thought to implement more ASoC-friendly way of HD-audio controller and
codec drivers, e.g. adapting DAPM and others.  So we began with the
factoring out the HD-audio basic core stuff to the common directory
sound/hda/*, while keeping the rest legacy stuff almost as is.  But
ASoC implementation didn't fly in the end from various reasons, and
the legacy HD-audio stuff was good enough for the actual use cases;
it's too bit to fail, after all.

> > The Realtek codec is split further to smaller pieces (which was really
> > huge).
> 
> Yes, that file was very confusing having support and quirks for so many
> Realtek parts all in one file.
> 
> > The Cirrus and TI sub-codec drivers are moved to codecs/side-codecs
> > subdirectory:
> 
> That's ok
> 
> > They can be put to each own directory and drop the file name prefix,
> > if we want, too.  Let me know if Cirrus and TI people would like to
> > split to more subdirectories.
> 
> I don't mind either way.
> The hda_component* files are common to the amps and the realtek driver
> so I wonder whether they belong in helpers. They are only utility
> wrappers around the kernel component-binding APIs.

Right.  So far, just because it's basically only binding with
side-codecs, I put into side-codecs subdirectory.

> > *HOWEVER* the biggest question is: whether it's worth?
> > 
> > Essentially, this makes almost impossible to make a patch for stable
> > trees from the original commit as is; one has to translate the file
> > paths and adjust manually in each patch.
> 
> Depends which file you are patching and how much it has changed.
> Git can figure out file renames (and changing directory is a rename).

I'm afraid that the Realtek codec changes might be hard to track
automatically.  Maybe the Cirrus side-codecs stuff would work.

> > Also, of course, if anyone is working on HD-audio stuff right now, the
> > work had to be adjusted to the new file path.  It'd be one-off action,
> > though.
> 
> Speaking only for Cirrus, I don't think this makes much extra effort
> for us. Our amp drivers have only changed location, so that should be
> trivial. The other file we change is patch_realtek.c but that typically
> is only 1-line quirk entries.

OK, thanks for confirmation!


Takashi