Re: [PATCH net-next v10 08/12] net: dsa: realtek: rtl8365mb: use raw VID for PVID on family D

From: Mieczyslaw Nalewaj

Date: Tue Nov 17 2026 - 01:29:07 EST


Hi,

Thanks for the review. Replying to the two PVID points.

> Is there something that prevents the family C MCIDX fields and VLANMC entries from being loaded from EEPROM as well?

Nothing prevents it in principle, but I would like to keep it out of this series. The series is meant to leave family C behaviour untouched, and the family C regression runs (RTL8367S on Archer AX55 v1) were done with the existing setup path. Writing the PVID registers of every family C port at setup would be a behaviour change on chips this series does not otherwise touch, so it is better done as a separate patch with its own justification and testing.

The family D case is different in kind: there the register holds a raw VID with no VLANMC indirection, so a non-zero hardware or EEPROM default is directly used as the PVID by rtl8365mb_vlan_port_get_pvid() and rtl8365mb_port_vlan_filtering(). I could not rule that out for a chip whose reset state I cannot verify against the vendor API, hence the explicit clear.

I agree the kernel-doc of rtl8365mb_c_vlan_port_reset_pvid() should not claim more than is known. I will reword it in the next revision to say that family C relies on the reset value of the register and does not defend against an EEPROM-preloaded MC index. If you or Luiz would rather have family C cleared as well, I can send that as a separate follow-up.

> Luiz raised these points on v7 ... I don't see these points addressed or answered in this version.

This is addressed in v11. The rtl8365mb_is_d() branches and the _direct helpers are gone, and PVID get/set/clear/reset now go through struct rtl8365mb_family_info, with separate family C (8/13) and family D (9/13) implementations. VLANMC index 0 is described as a driver convention, not as a hardware neutral PVID.

Best regards,
Mieczyslaw Nalewaj