svn commit: r334880 - head/sys/dev/vnic

[Ian Lepore]

On Sun, 2018-07-08 at 12:14 -0400, Mark Johnston wrote:
> On Sun, Jul 08, 2018 at 09:58:35AM -0600, Warner Losh wrote:
> > 
> > On Sun, Jul 8, 2018 at 9:55 AM, Sean Bruno <sbruno at freebsd.org>
> > wrote:
> > 
> > > 
> > > 
> > > 
> > > On 07/08/18 09:26, Mark Johnston wrote:
> > > > 
> > > > On Sun, Jul 08, 2018 at 09:10:27AM -0600, Sean Bruno wrote:
> > > > > 
> > > > > 
> > > > > 
> > > > > On 07/07/18 11:43, Mark Johnston wrote:
> > > > > > 
> > > > > > On Sat, Jun 09, 2018 at 02:47:49PM +0000, Andrew Turner
> > > > > > wrote:
> > > > > > > 
> > > > > > > Author: andrew
> > > > > > > Date: Sat Jun  9 14:47:49 2018
> > > > > > > New Revision: 334880
> > > > > > > URL: https://svnweb.freebsd.org/changeset/base/334880
> > > > > > > 
> > > > > > > Log:
> > > > > > >   In the ThunderX BGX network driver we were skipping the
> > > > > > > NULL
> > > terminator
> > > > 
> > > > > 
> > > > > > 
> > > > > > > 
> > > > > > >   when parsing the phy type, however this is included in
> > > > > > > the length
> > > returned
> > > > 
> > > > > 
> > > > > > 
> > > > > > > 
> > > > > > >   by OF_getprop. To fix this stop ignoring the
> > > > > > > terminator.
> > > > > > > 
> > > > > > >   PR:              228828
> > > > > > >   Reported by:     sbruno
> > > > > > >   Sponsored by:    DARPA, AFRL
> > > > > > This seems to break vnic on packet.net ThunderXs.  In
> > > > > > particular, VF
> > > > > > creation fails.  It seems the problem in my case is that
> > > > > > there are
> > > > > > multiple PHY devices in the device tree, e.g., xfi at 0, xfi at 1
> > > > > > .  With
> > > this
> > > > 
> > > > > 
> > > > > > 
> > > > > > change, bgx_fdt_phy_name_match() fails to match against any
> > > > > > device
> > > > > > containing a unit address in the node name.
> > > > > > 
> > > > > 
> > > > > Huh ... this was *required* to get the ThunderXs we have in
> > > > > the FreeBSD
> > > > > cluster to work at all.  o.0
> > > > > 
> > > > > I guess "someone" needs to contact "someone" to figure out
> > > > > which is
> > > > > correct or we need to replace our FreeBSD cluster machines
> > > > > with ones
> > > > > that work like the packet.net machines?
> > > > I think the current code works fine if there's only one PHY
> > > > device, so
> > > > my problem is probably just the result of having a different
> > > > hardware
> > > > setup.  We can probably fix the code to handle both
> > > > cases.  Could you
> > > > mail me the output of "ofwdump -ap" from the cluster machine?
> > > > 
> > > > 
> > > 
> > > I dropped the output here:
> > > https://people.freebsd.org/~sbruno/ofwdump.txt
> > 
> > Ian's method is better. But Ian's question is better: are there not
> > phandles to find this stuff? Names in FDT are kinda meaningless
> > most of the
> > time (I say kinda here to gloss over a laundry list of exceptions
> > that PHY
> > finding typically does not fall into).
> Sean's output shows why the current code works for him.  In my case
> the
> bgx subnodes don't contain a qlm-mode property, so we're falling back
> to
> name matching:
> 
> Node 0x891c: bgx0
>   #address-cells:
>     00 00 00 01 
>   #size-cells:
>     00 00 00 00 
>   reg:
>     00 00 80 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 
>   Node 0x8ad8: xfi at 0
>     reg:
>       00 00 00 00 
>     local-mac-address:
>       fc 15 b4 97 48 b7 
>     phy-handle:
>       00 00 00 75 
>   Node 0x8b34: xfi at 1
>     reg:
>       00 00 00 01 
>     local-mac-address:
>       fc 15 b4 97 48 b8 
>     phy-handle:
>       00 00 00 76 
> 
> Being unfamiliar with FDT, could I ask you to explain how the code
> could
> be using phandles to find the PHYs?
> 

In Sean's output, the phy nodes had multiple strings encoded in their
qlm-mode properties (note the embedded 0 about 6-8 bytes in). Based on
pure guesswork (because I can't find published fdt-bindings docs for
any of this stuff), maybe the phy supports multiple modes, so the mac
node has to have a property to say which mode is used. It may be that
the property in the mac node is optional when the phy node only lists
one qlm-mode, and maybe in that case the thing to do is retrieve the
property from the phy node after using the xref-phandle to get to that
node.

-- Ian