netdev - Re: [PATCH net-next 3/3] selftests: net: Add port split test

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20200520134340.GA1050786@splinter>
Date:   Wed, 20 May 2020 16:43:40 +0300
From:   Ido Schimmel <idosch@...sch.org>
To:     Andrew Lunn <andrew@...n.ch>
Cc:     netdev@...r.kernel.org, davem@...emloft.net, kuba@...nel.org,
        jiri@...lanox.com, danieller@...lanox.com, mlxsw@...lanox.com,
        michael.chan@...adcom.com, jeffrey.t.kirsher@...el.com,
        saeedm@...lanox.com, leon@...nel.org, snelson@...sando.io,
        drivers@...sando.io, vivien.didelot@...il.com,
        f.fainelli@...il.com, Ido Schimmel <idosch@...lanox.com>
Subject: Re: [PATCH net-next 3/3] selftests: net: Add port split test

On Tue, May 19, 2020 at 09:33:06PM +0200, Andrew Lunn wrote:
> > It's basically the number of lanes
> 
> Then why not call it lanes? It makes it clearer how this maps to the
> hardware?

I'm not against it. We discussed it and decided to go with width. Jiri /
Danielle, are you OK with changing to lanes?

> 
> > > Is it well defined that all splits of the for 2, 4, 8 have to be
> > > supported?
> > 
> > That I don't actually know. It is true for Mellanox and I can only
> > assume it holds for other vendors. So far beside mlxsw only nfp
> > implemented port_split() callback. I see it has this check:
> > 
> > ```
> >         if (eth_port.is_split || eth_port.port_lanes % count) {
> >                 ret = -EINVAL;
> >                 goto out;
> >         }
> > ```
> > 
> > So it seems to be consistent with mlxsw. Jakub will hopefully chime in
> > and keep me honest.
> > 
> > > Must all 40Gbps ports with a width of 4, be splitable to 2x
> > > 20Mps? It seems like some hardware might only allow 4x 10G?
> > 
> > Possible. There are many vendor-specific quirks in this area, as I'm
> > sure you know :)
> 
> So this makes me wonder if the API is sufficient. Do we actually want
> to enumerate what is possible, rather than leave the user to guess,
> trial and error?

The API is merely a read-only number which represents the number of
lanes. It's useful on its own without relation to split because it
allows you to debug issues such as "why port X does not support speed
Y". I actually wrote an ugly drgn script to pull this info from mlxsw
once:

```
#!/usr/bin/env drgn

import drgn
from drgn.helpers.linux import list_first_entry

devlink = list_first_entry(prog['devlink_list'].address_of_(),
                           'struct devlink', 'list')
mlxsw_core = drgn.reinterpret(prog.type("struct mlxsw_core"), devlink.priv)
mlxsw_sp = drgn.reinterpret(prog.type("struct mlxsw_sp"),
                            mlxsw_core.driver_priv)

max_ports = mlxsw_core.max_ports.value_() - 1
for i in range(0, max_ports):
    if mlxsw_sp.ports[i].value_() == 0:
        continue

    print("local_port=%d module=%d width=%d" %
          (mlxsw_sp.ports[i].local_port.value_(),
          mlxsw_sp.ports[i].mapping.module.value_(),
          mlxsw_sp.ports[i].mapping.width.value_()))
```

> 
> > I assume you're asking because you are trying to see if the test is not
> > making some vendor-specific assumptions?
> 
> Not just the test, but also the API itself. Is the API generic enough?
> Should we actually be able to indicate a 40G port cannot be used as 2x
> 20G? But 4x 10G is O.K?

We can do it, but I prefer to wait with new uAPI until we actually see a
need for it. We currently only expose width/lanes which is useful on its
own and infer from it how a port can be split. If this assumption turns
out to be wrong, we can add new attributes to answer the questions you
posed.

> The PDF you gave a link to actually says nothing about 2x 50G, or 2x
> 20G. There is a cable which does support 2x 50G. Does the firmware do
> any sanity checking and return errors if you ask it to do something
> which does not make sense with the cable currently inserted in the
> SFP cage?

Wait, I linked to a 4x splitter, not to 2x. This is a 2x:
https://www.mellanox.com/related-docs/prod_cables/PB_MCP7H00-G0xxxxxxx_100GbE_QSFP28_to_2x50GbE_2xQSFP28_DAC_Splitter.pdf

Complete list is here:
https://www.mellanox.com/products/interconnect/ethernet-copper-splitter

The firmware does not really care if it's a split or not. Software just
maps {module, width/lanes} -> local_port where each local_port is
represented by a netdev.

The supported speeds are set according to the width/lanes of the port.
If you requested a speed that is not supported or cannot be handled by
the cable you are using, then you will not be able to negotiate a link,
but it's the same regardless if the port is split or not.