Re: [PULL 24/28] hpet: switch to fine-grained device locking

[Igor Mammedov <imammedo@redhat.com>]

On Wed, 10 Sep 2025 13:23:10 +0200
Paolo Bonzini <pbonzini@redhat.com> wrote:

> On Wed, Sep 10, 2025 at 1:16 PM Igor Mammedov <imammedo@redhat.com> wrote:
> >
> > On Mon, 8 Sep 2025 15:30:19 +0100
> > Daniel P. Berrangé <berrange@redhat.com> wrote:
> >  
> > > Hi,
> > >
> > > This patches causes a regression making QEMU  abort in the KVM Xen
> > > functional test.
> > >
> > > To reproduce please run 'make check-functional-x86_64', or more
> > > specifically run this single test:
> > >
> > >  QEMU_TEST_QEMU_BINARY=./build/qemu-system-x86_64 PYTHONPATH=./python:./tests/functional ./tests/functional/x86_64/test_kvm_xen.py
> > >
> > > though I recommend you first add this series:
> > >
> > >   https://lists.nongnu.org/archive/html/qemu-devel/2025-09/msg01540.html
> > >
> > > as that fixes an infinite loop in the functional test console
> > > code on QEMU abnormal premature exit.  
> >
> > I'll try to reproduce and see where it goes wrong  
> 
> I've only looked at it briefly, but it fails on the bql_locked()
> assertion of xen_evtchn_set_gsi().
> 
> I think the assertion can be moved inside "if (gsi && gsi == s->callback_gsi)".

given it's manipulating interrupts,
shouldn't we take BQL when doing so?

I see 2 options here
  1: guard qemu_irq_foo calls inside of hpet.c with BQL_LOCK_GUARD
  2: drop new hpet:lock and use BQL instead of it.

> 
> Paolo
>