> I think this is getting overly complex, so let's rewind a bit. > > I believe Abhishek mentioned in a previous review what the differences > are between what the PHY reports when read, and what it actually > supports, and the result was that there was not a single bit in the > supported mask that was correct. I was hopeful that maybe Andrew would > respond to that, but seems not to, so I'm putting this statement here. > More on this below. Yes, i did not really realise how wrong Marvell got this. As you point out, it is more wrong than right. My thinking with calling the usual feature discovery mechanism and then fixing them up, is that we keep extending them. BaseT1 has been added etc. If a PHY is mostly getting it right, we might in the future get new features implemented for free, if the hardware correctly declares them. But in this case, if it cannot get even the basics mostly correct, there is little hope it will get more exotic features correct. So, i agree in Russell. Forget about asking the hardware, just hard code the correct features. Sorry for making you do extra work which you now need to discard. However, please do keep it as two patches. It makes it easier to deal with regressions on the device you cannot test if we can just revert one patch. Andrew
