On Sat, Mar 12, 2022 at 08:36:20PM +0100, Allan W. Nielsen wrote:
> With this in mind, I do agree with you that it does not make much sense
> to compensate they few cm of PCB tracks without also calibrating for
> differences from packet to packet.
The PCB traces AFTER the PHY are part of the network, and if they
contribute to path asymmetry, then that can and should be corrected
using the delayAsymmetry configuration variable.
> If we do not offer default delays directly in the driver, everybody will
> have to calibrate all boards just to have decent results, we will not
> have a good way to provide default delay numbers, and this will be
> different from what is done in other drivers.
Who says the other drivers are even remotely reasonable? Not me.
I've been fighting this voodoo engineering all along, but people seem
to ignore me.
> I do understand that you have a concern that these numbers may change in
> future updates. But this has not been a problem in other drivers doing
> the same.
Wrong. See the git history of the i210 driver. Also the data sheets.
> But if this is still a concern, we can add a comment to say
> that these numbers must be treated as UAPI, and chancing them, may
> cause regressions on calibrated PHYs.
Comments will be ignored. And when the next batch of developers comes
along, they will ignore your prohibition.
Thanks,
Richard