We have been waiting for the CM5 for a while, because PTP support on the CM4 is partially broken and somewhat lacking. Specifically, the Broadcom PHY doesn't support point releases in the packet version field, i.e. 2.0 works but 2.1 does not. There is a patch to force linuxptp to put 2.0 on the packet, but the maintainer rejected it saying they "won't support broken hardware".
It's a big issue because it means that we are stuck with older versions of linuxptp, or maintaining a fork that implements a dubious hack.
There also seems to be an issue where the SYNC_IN pin doesn't work, so you can only have PPS in or out, not both.
The other thing is you can't use your own clock for the PHY. It uses some internal SoC clock which it steers, but it's not as stable as we require. Since there is no public datasheet for Broadcom PHY, we can't even find which pin to inject one into.
I have a feeling that the CM5 won't fix all of that, but maybe it will fix enough that we can use it. There really aren't any other options at the moment.
It's a big issue because it means that we are stuck with older versions of linuxptp, or maintaining a fork that implements a dubious hack.
There also seems to be an issue where the SYNC_IN pin doesn't work, so you can only have PPS in or out, not both.
The other thing is you can't use your own clock for the PHY. It uses some internal SoC clock which it steers, but it's not as stable as we require. Since there is no public datasheet for Broadcom PHY, we can't even find which pin to inject one into.
I have a feeling that the CM5 won't fix all of that, but maybe it will fix enough that we can use it. There really aren't any other options at the moment.
Statistics: Posted by aestetix — Mon Jun 24, 2024 8:25 am