On 23.12.2024 3:47 AM, Cheng Jiang (IOE) wrote: > Hi Konrad, > > On 12/20/2024 9:46 PM, Konrad Dybcio wrote: >> On 13.12.2024 8:05 AM, Cheng Jiang (IOE) wrote: >> >> [...] >> >>>> /* >>>> * If the board-specific file is missing, try loading the default >>>> * one, unless that was attempted already >>>> */ >>>> >>>> But, even more importantly: >>>> >>>> a) do we want to load the "incorrect" file? >>> Normally, there is a default NVM file ending with .bin, which is suitable >>> for most boards. However, some boards have different configurations that >>> require a specific NVM. If a board-specific NVM is not found, a default >>> NVM is preferred over not loading any NVM. >> >> So, if one is specified, but not found, this should either be a loud error, >> or a very loud warning & fallback. Otherwise, the device may provide subpar >> user experience without the user getting a chance to know the reason. >> >> I think failing is better here, as that sends a clearer message, and would >> only happen if the DT has a specific path (meaning the user put some >> intentions behind that choice) >> > In the existing BT driver implementation, even if the rampatch/nvm are not found, > BT still works with ROM code only. No fails, just a warning in the dmesg. For this > new approach, we use the similar logic. > > The fallback to load a default nvm file is due to each board has a unique board > id, it's not necessary to upstream all the board-specific nvm since most of them > may be the same, the default nvm file is suitable for them. But we can't set the > default nvm file name in the DT, since the platform can attach different > connectivity boards. This is a reasonable way to approach this. Okay, let's do it this way then. >>>> b) why would we want to specify the .bin file if it's the default anyway? >>> The default NVM directory is the root of qca. The 'firmware-name' property >>> can specify an NVM file in another directory. This can be either a default >>> NVM like 'QCA6698/hpnv21.bin' or a board-specific NVM like 'QCA6698/hpnv21.b205'. >> >> Do we expect QCA6698/hpnv21.bin and QCAabcd/hpnv21.bin to be compatible? >> > No. It may be different. That's a bit disappointing considering the filename implies it's suitable for a family of chips.. But I guess there's nothing we can change here. Konrad
