There's as slight bug in switch chip in IPQ4xxx which bit me and MT doesn't have a solution (yet).
What is the bug? Could you share some information?
A slightly complicated scenario: xDSL connection with modem in bridge mode. VLANs all over, so modem connected to hybrid port (untagged, which is used for PPPoE, gets tagged to e.g VLAN ID = 2 on ingress, tagged VLAN, which is used to carry IPTV multicasts, remains tagged), ether port member of common bridge, vlan interface with vlan id=2 created on top of bridge, pppoe client on top of vlan device.
Result: ppoe client never connects.
Answer from MT:
... we are working on a fix though, but we are not sure when will a fix be available. This is only specific to devices using the IPQ4000 CPU.
The problem is that PPPoE discovery stage packets don't get tagged as soon as they are sent to the VLAN interface, it is only this very specific packet (EtherType=0x8863).
Sadly there is no workaround for this issue for now other than using bridge VLAN filtering instead.
Proposed workaround solves the issue however it means no HW offload.
I'm not really sure if the wording of MT is correct. It seems to imply that the problem lies in tagging of packets on vlan interface (the SW one). My workaround is to use another VLAN switch to host the hybrid port for connection towards xDSL modem and get VLAN with ID 2 to hAP ac2 device via trunk. As my workaround works it indicates that the problem problem lies in untagging of particular packets on egress from IPQ4xxx switch (which is the step my workaround ... works around).