Re: Bogus names on vmware

Re: Bogus names on vmware

[Stephen Hemminger]

On Wed, 21 Oct 2015 18:17:56 +0200
Lennart Poettering <lennart@poettering.net> wrote:

> On Wed, 21.10.15 08:41, Stephen Hemminger (stephen@networkplumber.org) wrote:
> 
> > Do you have any recommendations on how to deal with VMware borkage
> > of the ACPI firmware index?
> 
> Hmm, shouldn't the kernel filter such crappy data out already? I mean,
> if we work around crappy firmware, why should that happen in userspace? So
> far that was always a kernel job...
> 
> Any specific reason you assume that 0xffff is a good cutoff point?
> 
> > https://communities.vmware.com/message/2378360
> > 
> > My proposal is to just add special case:
> > 
> > diff --git a/src/udev/udev-builtin-net_id.c b/src/udev/udev-builtin-net_id.c
> > index 6e7e127..b8c6f47 100644
> > --- a/src/udev/udev-builtin-net_id.c
> > +++ b/src/udev/udev-builtin-net_id.c
> > @@ -133,7 +133,7 @@ static int dev_pci_onboard(struct udev_device *dev, struct netnames *names) {
> >          size_t l;
> >          char *s;
> >          const char *attr;
> > -        int idx;
> > +        unsigned long idx;
> >  
> >          /* ACPI _DSM  -- device specific method for naming a PCI or PCI Express device */
> >          attr = udev_device_get_sysattr_value(names->pcidev, "acpi_index");
> > @@ -143,8 +143,9 @@ static int dev_pci_onboard(struct udev_device *dev, struct netnames *names) {
> >          if (!attr)
> >                  return -ENOENT;
> >  
> > +        /* vmware returns bogus values */
> >          idx = strtoul(attr, NULL, 0);
> > -        if (idx <= 0)
> > +        if (idx == 0 || idx > 0xffff)
> >                  return -EINVAL;
> >  
> >          /* kernel provided port index for multiple ports on a single PCI function */
> > 

Sorry for the long latency, for some reason this ended up in Spam.

The kernel could filter but I suspect the ACPI developers would consider
it a valid value.  I made up 0xffff as a mask assuming 64K slots is more than
any normal ACPI would return.

VMware has other issues, there concept of PCI addresses is equally borked (unique)
 http://goo.gl/uhGEj8

Copying the ACPI kernel list on this.