Re: [PATCH v3 5/6] x86/PCI: avoid re-evaluation of extended config space accessibility

[Jan Beulich <jbeulich@xxxxxxxx>]

On 29.01.2026 16:32, Roger Pau Monné wrote:
> On Thu, Jan 29, 2026 at 02:10:01PM +0100, Jan Beulich wrote:
>> When, during boot, we have already correctly determined availability of
>> the MMCFG access method for a given bus range, there's then no need to
>> invoke pci_check_extcfg() again for every of the devices. This in
>> particular avoids ->ext_cfg to transiently indicate the wrong state.
>>
>> Switch to using Xen style on lines being touched and immediately adjacent
>> ones.
>>
>> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
> 
> Acked-by: Roger Pau Monné <roger.pau@xxxxxxxxxx>

Thanks.

> One suggestion for a further addition below.
> 
>> ---
>> v3: New.
>>
>> --- a/xen/arch/x86/physdev.c
>> +++ b/xen/arch/x86/physdev.c
>> @@ -528,6 +528,8 @@ ret_t do_physdev_op(int cmd, XEN_GUEST_H
>>          if ( !ret )
>>              ret = pci_segment_iterate(info.segment, 
>> physdev_check_pci_extcfg,
>>                                        &info);
>> +        else if ( ret > 0 ) /* Indication of "no change". */
>> +            ret = 0;
>>  
>>          if ( !ret && has_vpci(currd) && (info.flags & 
>> XEN_PCI_MMCFG_RESERVED) )
> 
> Maybe it doesn't need to be strictly done here, but now that
> pci_mmcfg_reserved() signals whether the MMCFG was already registered,
> could you also restrict the register_vpci_mmcfg_handler() call to ret
> == 0?

Possibly; you know vPCI better than I do.

> That will also simplify the logic in register_vpci_mmcfg_handler()
> since we no longer need to return 0 when the region is already
> registered, returning -EEXIST should be fine if the caller is
> adjusted.

I think this then will want to be a separate change, with its own
justification.

Jan