There is a greater discussion going on about whether and how CSIT should be setting maximum frame size in various tests.

Regardless of the outcome, the fact is the hw_interface_set_mtu API call started failing between 2110 and 2202.

Focusing on performance testbed (say 2n-skx), physical function used, dpdk_plugin, xxv710 Intel NIC, and base L2 cross-connect test, the value of 9200 started failing with 34839 [0], and value 1500 started failing with 34843.

At the time of 2202 RC1 I am not aware of any situation where setting any value passes.
Even if usual use cases should not rely on the API call (the default should already be as large as possible according to the driver), the fact an API call now seems to be always failing is a bug.

[0] https://gerrit.fd.io/r/c/vpp/+/34839/4#message-d5d0c80dc32b1aa29a142fb48b6c132ca04ff332
[1] https://gerrit.fd.io/r/c/vpp/+/34843/12#message-6c01b85caf7709c80067a3c8ae1e35fc706ac0e7