We are facing a service-impacting issue on the Juniper MX304, where the downstream AE interface is admin up but link down.

Apr 26 10:08:14 Router fpc0 MQSS(1): WI-0: OGE 0 interface FIFO overflow

Apr 26 10:08:13 Router-fpc0 ytchip-luss[18362]: ppe_error_interrupt(8320): YT[0].slice[1]_PPE 120 Errors thread timeout error

As noted in internal PR 1866145,

> FIFO is getting full due to some unknown reasons, which we could not figure out from the customer logs.

> It is also true that FIFO getting full can't lead to PPEs getting busy.

> However, the flow control (FC) needs to be enabled for OGE used for host path.

> There is a 16 entries deep credit FIFO front ending this which doesn't have any configuration control due to ASIC design.

> If FC frames are not honoured by Marvell switch, it can jam the credit FIFO which can lead to OGE FIFO overflow error reported.

> It is not clear if there was a bursty host path traffic coming to OGE. But, the OGE MAC statistics showed that pause frames were sent to Marvell switch.

> If bursty incoming host path traffic caused the pause frames to Marvell switch and if it was not honored, it can cause OGE FIFO overflow error.

So, enabling FC in Marvell switch will guarantee one level of protection. The FC needs to be enabled for OGE for all YT PFEs.

So, the suggestion to enable FC in Marvell switch to address this customer escalation. With this, we can provide one level of protection and monitor.

This was enabled via internal PR 1869373, in below releases (all releases after this will also have the fix).

junos:23.4R2-S4-J6 junos:24.4R2 junos:25.2R1 junos:25.3R1

There is also a workaround available fix this temporarily, but the same will get removed when FPC/Chassis reboots.

Kindly reach out to JTAC for the workaround.