On Fri, 2011-11-18 at 08:58 -0800, Greg Rose wrote:
> On 11/17/2011 4:44 PM, Ben Hutchings wrote:
> > On Thu, 2011-11-17 at 16:32 -0800, Greg Rose wrote:
> >> On 11/17/2011 4:15 PM, Ben Hutchings wrote:
> >>> Sorry to come to this rather late.
> >>>
> >>> On Tue, 2011-11-08 at 23:55 -0800, Roopa Prabhu wrote:
> >>> [...]
> >>>> v2 -> v3
> >>>> - Moved set and get filter ops from rtnl_link_ops to netdev_ops
> >>>> - Support for SRIOV VFs.
> >>>> [Note: The get filters msg (in the way current get rtnetlink handles
> >>>> it) might get too big for SRIOV vfs. This patch follows existing sriov
> >>>> vf get code and tries to accomodate filters for all VF's in a PF.
> >>>> And for the SRIOV case I have only tested the fact that the VF
> >>>> arguments are getting delivered to rtnetlink correctly. The code
> >>>> follows existing sriov vf handling code so rest of it should work fine]
> >>> [...]
> >>>
> >>> This is already broken for large numbers of VFs, and increasing the
> >>> amount of information per VF is going to make the situation worse. I am
> >>> no netlink expert but I think that the current approach of bundling all
> >>> information about an interface in a single message may not be
> >>> sustainable.
> >>>
> >>> Also, I'm unclear on why this interface is to be used to set filtering
> >>> for the (PF) net device as well as for related VFs. Doesn't that
> >>> duplicate the functionality of ndo_set_rx_mode and
> >>> ndo_vlan_rx_{add,kill}_vid?
> >>
> >> Functionally yes but contextually no. This allows the PF driver to know
> >> that it is setting these filters in the context of the existence of VFs,
> >> allowing it to take appropriate action. The other two functions may be
> >> called without the presence of SR-IOV enablement and the existence of VFs.
> >>
> >> Anyway, that's why I asked Roopa to add that capability.
> >
> > I don't follow. The PF driver already knows whether it has enabled VFs.
> >
> > How do filters set this way interact with filters set through the
> > existing operations? Should they override promiscuous mode? None of
> > this has been specified.
>
> Promiscuous mode is exactly the issue this feature is intended for. I'm
> not familiar with the solarflare device but Intel HW promiscuous mode is
> only promiscuous on the physical port, not on the VEB. So a packet sent
> from a VF will not be captured by the PF across the VEB unless the MAC
> and VLAN filters have been programmed into the HW.
Yes, I get it. The hardware bridge needs to know more about the address
configuration on the host than the driver is getting at the moment.
> So you may not need
> the feature for your devices but it is required for Intel devices.
Well we don't have the hardware bridge but that means each VF driver
needs to know whether to fall back to the software bridge. The net
driver needs much the same additional information.
> And
> it's a fairly simple request, just allow -1 to indicate that the target
> of the filter requests is for the PF itself. Using the already existing
> set_rx_mode function wont' work because the PF driver will look at it
> and figure it's in promiscuous mode anyway, so it won't set the filters
> into the HW. At least that is how it is in the case of our HW and
> driver. Again, the behavior of your HW and driver is unknown to me and
> thus you may not require this feature.
What concerns me is that this seems to be a workaround rather than a fix
for over-use of promiscuous mode, and it changes the semantics of
filtering modes in ways that haven't been well-specified.
What if there's a software bridge between two net devices corresponding
to separate physical ports, so that they really need to be promiscuous?
What if the administrator runs tcpdump and really wants the (PF) net
device to be promiscuous?
These cases shouldn't break because of VF acceleration. Or at least we
should make a conscious and documented decision that 'promiscuous'
doesn't mean that if you enable it on your network adapter.
Ben.
--
Ben Hutchings, Staff Engineer, Solarflare
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
