On 2016-12-01 03:55, Joe Ayers wrote:
> We think this is an SOC level problem given the at803x and defect is
> embedded in the AR93xx SOC. This GPIO pin is believed to be in
> silicon. If so, the same GPIO across all hardware that uses this SOC.
> Specifically, for the Ubiquiti "AirMax" line, this SOC is relevant
> AirGrid HP M5 XW
> NanoBridge M5 XW 16/19
> NanoStation M5 Loco XW
> PowerBeam M2 XW 400
> PowerBeam M5 XW 300/400
> PowerBeam M5 XW 620 (Rocket XW image but has newer flash chip and
> needs write protect fix still)
> Rocket M5 XW
> Rocket Titanium M5 XW (for completeness, but we're not supporting this)
> NanoStation M5 XW (this has an external switch chip somehow avoiding
> the defect or other).
You're wrong about that. I just opened up A NanoStation M5 Loco XW, and
the 8032 PHY is an external chip. The GPIO pin is external and routed to
the reset pin of that chip. That also explain why pretty much only UBNT
products are affected, and other AR934x based devices aren't.
So far I've only applied the fix to the Loco, if other devices need it,
we need to explicitly change the mach file for it. Please update to the
latest version of my staging tree and test my fix on the Loco.
Please also let me know which other devices use the 8032 PHY - you can
check that in the boot log based on the PHY ID.
> We have only been able to reproduce this failure on NanoBridge M5 XW
> 19 and the Loco NS M5 XW (in CC release). A ~year ago the symptoms
> occurred in CC builds in all models except NSM5-XW. Presumably, you
> had since committed a fix within the at803x driver for the ubnt
> devices that appear to be working now? Regardless, all these devices
> are in scope (as could be other manufacturer's devices) to be included
> in the fix at the SOC level. Will the Lede staged commit catch all
> these devices?
There were other SoC related ethernet driver issues that affect other
devices as well. They were unrelated to the 8032 PHY.
Lede-dev mailing list