Hi, > Does this code introduce an open-coded trylock pattern?
I will keep the existing synchronization in v8. channel_changing is a vport-admission reservation protected by vport_mutex, not an unprotected replacement for the serialization of Ethernet reconfiguration. RTNL serializes the Ethernet reconfiguration and reset writers. The reservation remains set while the rebuild releases vport_mutex between individual vport operations, so concurrent RDMA admission through mana_cfg_vport(check=true) can reject that interval under vport_mutex. The success and failure cleanup paths clear the reservation under the same mutex. Holding vport_mutex over the whole rebuild is not a drop-in replacement: the vport configuration/unconfiguration helpers acquire it themselves. The report does not identify a missing exclusion edge in the existing protocol, which is also present in v6. I am not redesigning that protocol solely on the basis of its resemblance to a trylock. Thanks, Wei
