Den fre 5 jan. 2024 kl 11:26 skrev Shinichiro Kawasaki
<shinichiro.kawasaki@xxxxxxx>:
>
> On Jan 05, 2024 / 09:44, Lukas Wunner wrote:
> > On Fri, Jan 05, 2024 at 08:18:05AM +0000, Shinichiro Kawasaki wrote:
> > > --- a/drivers/platform/x86/p2sb.c
> > > +++ b/drivers/platform/x86/p2sb.c
> > > @@ -150,6 +153,14 @@ static int p2sb_cache_resources(void)
> > > if (!bus)
> > > return -ENODEV;
> > >
> > > + /*
> > > + * When a device with same devfn exists and it is not P2SB, do not
> > > + * touch it.
> > > + */
> > > + pci_bus_read_config_dword(bus, devfn_p2sb, PCI_CLASS_REVISION, &class);
> > > + if (!PCI_POSSIBLE_ERROR(class) && class >> 8 != P2SB_CLASS_CODE)
> > > + return -ENODEV;
> > > +
> >
> > The function should probably return if PCI_POSSIBLE_ERROR() is true.
>
> At this point, the P2SB device can be still hidden and PCI_POSSIBLE_ERROR() can
> be true. In that case, the function should not return.
>
> > Also I think you can use PCI_CLASS_MEMORY_OTHER, so how about:
> >
> > if (PCI_POSSIBLE_ERROR(class) || class >> 16 != PCI_CLASS_MEMORY_OTHER)
> > return -ENODEV;
> >
> > Can alternatively use "class >> 8 != PCI_CLASS_MEMORY_OTHER << 8" if you
> > want to ensure the lowest byte is 0x00.
>
> Thanks, it looks the better to use PCI_CLASS_MEMORY_OTHER. Will reflect it when
> I create the formal fix patch.
Both of the variants seem to work for me.
I tried the first patch on its own (059b825c5234), with
if (!PCI_POSSIBLE_ERROR(class) && class >> 8 != P2SB_CLASS_CODE)
return -ENODEV;
Then Lukas' suggestion (b97584391ea7), with
if (PCI_POSSIBLE_ERROR(class) || class >> 16 != PCI_CLASS_MEMORY_OTHER)
return -ENODEV;
Tested-by: Klara Modin <klarasmodin@xxxxxxxxx>
Attachment:
dmesg1.gz
Description: application/gzip
Attachment:
dmesg2.gz
Description: application/gzip
