Re: [Xen-devel] xen-unstable warn_on in msi.c:636

["Jan Beulich" <JBeulich@xxxxxxxxxx>]

>>> On 30.06.11 at 15:31, George Dunlap <George.Dunlap@xxxxxxxxxxxxx> wrote:
> On Thu, Jun 30, 2011 at 1:17 PM, Jan Beulich <JBeulich@xxxxxxxxxx> wrote:
>>>>> On 30.06.11 at 14:03, George Dunlap <George.Dunlap@xxxxxxxxxxxxx> wrote:
>>> Olaf / Jan / all, what do you think of the attached patch, which
>>> teaches the MSI code how to deal with 0 pbas?
>>
>> No, definitely not an option. They simply must not be zero. The
>> problem just is that the BARs for virtual functions all read as zero
>> (and the real value must be calculated from information from the
>> PF's config space).
> 
> But at the moment it IS zero, and this "real value" is simply not
> being calculated.
> 
> I realize this needs to be fixed the Right Way eventually, but at the
> moment anyone who uses SRIOV will have a bunch of scary warnings which
> they can't do anything about except ignore; and it's never a good idea
> to condition people to ignore this kind of warning.
> 
> If the current situation is safe enough that fixing it is a low
> priority, then it's safe enough to remove the warnings for the time
> being.  If it's not safe enough to remove the warnings, then it's not
> safe enough to leave and a fix needs to be a priority.

It really is important to fix (and should have been fixed long ago),
as it's part of a larger security problem we're having. But the
(mainly Intel) people who wrote the MSI and passthrough code
didn't get around to it, and I'm finding only minimal amount of
time to work on this (complicated by not having readily available
any device I could actually try this out with).

A temporary solution I would be less opposed to would be to
warn only for physical functions - that information is available
(in struct pci_dev_info).

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel