Re: [PATCH] xen/vpci: do not use pci_sanitize_bar_memory for domU

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

On Wed, Feb 25, 2026 at 04:44:09PM +0100, Jan Beulich wrote:
> On 25.02.2026 00:12, Ariadne Conill wrote:
> > From: Steven Noonan <steven@xxxxxxxxx>
> > 
> > This function should only be used for the hardware domain, because it
> > compares addresses against the host e820 map.
> 
> The same is true for its sibling function, pci_check_bar(). For both the
> question is whether skipping is the right thing, or whether for DomU-s
> checking against their memory map is what is needed instead.

Well, the usage of pci_check_bar() from modify_bars() is slightly
different: it does get called with the BAR host addresses (not the
guest ones).  So the check will work as expected, albeit given a domU
cannot change the BAR host addresses the check is likely redundant,
but not strictly wrong.

Checking against the memory map for domUs would be complicated, as the
guest can play many games with that memory map.

> > --- a/xen/drivers/vpci/header.c
> > +++ b/xen/drivers/vpci/header.c
> > @@ -394,13 +394,15 @@ static int modify_bars(const struct pci_dev *pdev, 
> > uint16_t cmd, bool rom_only)
> >              }
> >          }
> >  
> > -        rc = pci_sanitize_bar_memory(bar->mem);
> > -        if ( rc )
> > -        {
> > -            gprintk(XENLOG_WARNING,
> > -                    "%pp: failed to sanitize BAR#%u memory: %d\n",
> > -                    &pdev->sbdf, i, rc);
> > -            return rc;
> > +        if (is_hardware_domain(pdev->domain)) {
> 
> Nit: Style (see surrounding code you alter).
> 
> > +            rc = pci_sanitize_bar_memory(mem);
> > +            if ( rc )
> > +            {
> > +                gprintk(XENLOG_WARNING,
> > +                        "%pp: failed to sanitize BAR#%u memory: %d\n",
> > +                        &pdev->sbdf, i, rc);
> > +                return rc;
> > +            }
> 
> To avoid the need for re-indentation here (reducing churn) you may want
> to leverage that rc is 0 ahead of the call to pci_sanitize_bar_memory().
> I.e. you could make just the call conditional, without touching anything
> else.

You could also introduce an early continue, and leave the check(s)
below applicable to the hardware domain only.

Thanks, Roger.