quoted

Interrupt handler (hardirq context) at CPU0, and process at CPU1, eg
role switch, unload module, etc.

the process at CPU1 would need to disable interrupts (spin_lock_irq() or
spin_lock_irqsave()), not the hardirq on CPU0 as that already runs with interrrupts
disabled.

https://www.kernel.org/doc/html/latest/kernel-hacking/locking.html#table-of-
minimum-requirements

I see, thanks.

Peter