On 20.12.2023 01:23, Sébastien Chaumat wrote:
> I had to triple check:
>
> The first call is from xen_register_pirq()  and seem to originate from
> early_irq_init()  : triggering is 1
> in this first call the HYPERVISOR_physdev_ops is called with triggering 1
> shareable 0
>
> The second call is from xen_register_pirq() called from xen_register_gsi() :
>   trigger=0 polarity 1 at the start of  xen_register_pirq()
> but then trigger=1 polarity=1 just before the call to PHYSDEVOPS_setup_gsi

I'm sorry, but this is getting extremely confusing. Can you please supply
a complete kernel log, where also other potentially relevant messages
would be present (like those issued when PHYSDEVOP_setup_gsi has failed,
as ought to happen when invoked the 2nd time for the same GSI)? It's fine
if you want to leave your extra printk()s in place for this, but please
be sure (perhaps by way of also supplying the patch you use) to make clear
where exactly they live, so we can make sense of those extra log messages.

Jan