Re: [PATCH v5 7/9] vfio/pci: Clean up BAR zap and revocation

On Thu, 30 Jul 2026 15:47:13 +0100
Matt Evans <[email protected]> wrote:

> Hi Alex,
> 
> On 29/07/2026 18:52, Alex Williamson wrote:
> > On Wed, 15 Jul 2026 18:47:30 +0100
> > Matt Evans <[email protected]> wrote:  
> >> diff --git a/include/linux/vfio_pci_core.h b/include/linux/vfio_pci_core.h
> >> index 9a1674c152aa..e2b4252e7c3f 100644
> >> --- a/include/linux/vfio_pci_core.h
> >> +++ b/include/linux/vfio_pci_core.h
> >> @@ -134,6 +134,7 @@ struct vfio_pci_core_device {
> >>  	bool			pm_intx_masked;
> >>  	bool			pm_runtime_engaged;
> >>  	bool			sriov_active;
> >> +	bool			zap_bars_on_revoke;
> >>  	struct pci_saved_state	*pci_saved_state;
> >>  	struct pci_saved_state	*pm_save;
> >>  	int			ioeventfds_nr;  
> > 
> > This should be in the bitfield usage group since it's only modified at
> > init time.  
> 
> This was intentional, but happy to change it if you're certain ofc.  Is
> it inconceivable that a sub-driver could set it after init?  I'd say
> they _shouldn't_, but only review will stop them and this placement
> intended to be cautious.  It seemed a low cost way to avoid issues
> around synchronisation on the bitfield.

I'd agree with the statement that they shouldn't, it would be difficult
to synchronize setting the flag once there are any active mappings of
the BARs.  Also, if we put it in the bitfield category under the
comment that the value is only modified at setup/release, it documents
the intentions, hopefully to the extent the author or reviewers notice.
An argument can always be made to change it if there's a worthwhile use
case.  Thanks,

Alex