WiWi shares the peripheral layout, not the PCI transport, with the legacy
Celestica entry. Only the WiWi match uses these offsets, vector mapping
and gate.
I have not verified whether this FPGA replays a masked event. Instead, the
proposed change refuses requests to arm a timestamp, PPS or generator
until probe reads back the open gate. A request made too early gets
-EAGAIN and can be retried. Disable remains available for cleanup. There
is no speculative acknowledgement after opening the gate. The mock-MMIO
test covers early refusal and a successful retry; it is not an MSI-X
delivery test.
v3:
https://lore.kernel.org/r/20261116-wiwi-v3-87afa10a8-0-ahmadexp@gmail.com (local)