> -----Original Message----- > From: Mark Brown <broonie@xxxxxxxxxx> > Sent: Saturday, February 24, 2024 1:20 AM > To: Ding, Shenghao <shenghao-ding@xxxxxx> > Cc: Pierre-Louis Bossart <pierre-louis.bossart@xxxxxxxxxxxxxxx>; > andriy.shevchenko@xxxxxxxxxxxxxxx; lgirdwood@xxxxxxxxx; perex@xxxxxxxx; > 13916275206@xxxxxxx; alsa-devel@xxxxxxxxxxxxxxxx; linux- > kernel@xxxxxxxxxxxxxxx; liam.r.girdwood@xxxxxxxxx; bard.liao@xxxxxxxxx; > mengdong.lin@xxxxxxxxx; yung-chuan.liao@xxxxxxxxxxxxxxx; Xu, Baojun > <baojun.xu@xxxxxx>; Lu, Kevin <kevin-lu@xxxxxx>; tiwai@xxxxxxx; > soyer@xxxxxx > Subject: Re: [EXTERNAL] Re: [PATCH v9] ASoc: tas2783: Add tas2783 codec > driver > > On Fri, Feb 23, 2024 at 10:12:49AM +0000, Ding, Shenghao wrote: > > Hi Pierre-Louis > > > > > In the SoundWire spec, the unique_id is *LINK SPECIFIC*, and only > > > used at the bus level within the context of a link to help avoid > > > enumeration conflicts > > > > If you are using the unique_id as a SYSTEM-UNIQUE value to lookup > > > EFI data, this is a TI-specific requirement that needs to be documented. > > > That also means you need to double-check for errors so make sure > > > there are no board configurations where the same unique_id is used > > > in multiple links, or by devices other than tas2783. > > > This code only covers the tas2783s sitting in the same bus link. As to > > cases of the different SWD links, customer will be required to have > > the secondary development on current code. I'm sure my customers have > much knowledge to handle this. > > As Pierre says I think we really should have some sort of defensive > programming here, even if you're going to leave multi-link systems to future > work people will still have older versions in distributions or whtaever. While > I'm not sure the consequences of getting things wrong are likely to be that > bad (I'm expecting bad quality audio) it's also going to be kind of hard to > figure out if we just silently pick the wrong calibration, especially if it's > actually a valid calibration for another device in the system. Other vendors > (eg, Cirrus) seem to have figured out a scheme here? Thanks for your comments, Mark & Pierre. I will discuss with my customers on how to find a compromise between new solution and current solution already released to market.
