Message-ID: <e473a5e8-590d-3c7a-5d56-5cd0351ccba1@gmail.com>
Date:   Tue, 26 Mar 2019 15:45:02 -0700
From:   Florian Fainelli <f.fainelli@...il.com>
To:     Vladimir Oltean <olteanv@...il.com>, davem@...emloft.net,
        netdev@...r.kernel.org
Cc:     andrew@...n.ch, vivien.didelot@...il.com, linus.walleij@...aro.org
Subject: Re: [RFC PATCH net-next 10/13] net: dsa: sja1105: Add support for
 traffic through standalone ports

On 3/26/19 3:38 PM, Vladimir Oltean wrote:
> On 3/27/19 12:13 AM, Florian Fainelli wrote:
>> On 3/26/19 3:03 PM, Vladimir Oltean wrote:
>>> On 3/26/19 4:31 AM, Florian Fainelli wrote:
>>>>
>>>>
>>>> On 3/23/2019 8:23 PM, Vladimir Oltean wrote:
>>>>> In order to support this, we are creating a make-shift switch tag
>>>>> out of
>>>>> a VLAN trunk configured on the CPU port. Termination on switch ports
>>>>> only works when not under a vlan_filtering bridge. We are making
>>>>> use of
>>>>> the generic CONFIG_NET_DSA_TAG_8021Q code and leveraging it from
>>>>> our own
>>>>> CONFIG_NET_DSA_TAG_SJA1105.
>>>>>
>>>>> There are two types of traffic: regular and link-local.
>>>>> The link-local traffic received on the CPU port is trapped from the
>>>>> switch's regular forwarding decisions because it matched one of the
>>>>> two
>>>>> DMAC filters for management traffic.
>>>>> On transmission, the switch requires special massaging for these
>>>>> link-local frames. Due to a weird implementation of the switching
>>>>> IP, by
>>>>> default it drops link-local frames that originate on the CPU port. It
>>>>> needs to be told where to forward them to, through an SPI command
>>>>> ("management route") that is valid for only a single frame.
>>>>> So when we're sending link-local traffic, we need to clone skb's from
>>>>> DSA and send them in our custom xmit worker that also performs SPI
>>>>> access.
>>>>>
>>>>> For that purpose, the DSA xmit handler and the xmit worker communicate
>>>>> through a per-port "skb ring" software structure, with a producer
>>>>> and a
>>>>> consumer index. At the moment this structure is rather fragile
>>>>> (ping-flooding to a link-local DMAC would cause most of the frames to
>>>>> get dropped). I would like to move the management traffic on a
>>>>> separate
>>>>> netdev queue that I can stop when the skb ring got full and
>>>>> hardware is
>>>>> busy processing, so that we are not forced to drop traffic.
>>>>>
>>>>> Signed-off-by: Vladimir Oltean <olteanv@...il.com>
>>>>
>>>> Reviewed-by: Florian Fainelli <f.fainelli@...il.com>
>>>>
>>>> I do like the idea of setting up specific management queue later on,
>>>> although it is not clear to me how you would go about integrating it as
>>>> a network device, given the DSA slave and master devices, do you know
>>>> roughly how you would proceed?
>>>>
>>>
>>> Actually I was thinking about leveraging the multiqueue support that you
>>> added in 55199df6d2af ("net: dsa: Allow switch drivers to indicate
>>> number of TX queues") and expose the slave netdev .ndo_select_queue
>>> callback towards DSA ports. There I would return queue #0 if
>>> sja1105_is_link_local(skb), and queue #1 otherwise.
>>> Are there any complications that I'm missing?
>>
>> So that queue could be used to steer management traffic, but it would
>> still attempt to perform a dev_queue_xmit() using the master DSA network
>> device unless you somehow change that and/or parent that queue to a
>> different network device that the sja1105 switch driver creates (which
>> is doable).
>>
> 
> But the problem I'm trying to solve with the management queue is not
> congestion on the master port or inside the switch, but a problem that I
> myself have created by putting some skb's in a ring that is finite (and
> small) in size: the DSA xmit racing with my xmit worker.

Oh I understood that part, which is why I was wondering if it even made
sense to make use of a particular queue and the flow control that is
offered with that given that this is already quite an ad-hoc solution
and what you proposed seems to do the job alright.

> Congestion management on the switch is a much ampler issue that I don't
> yet know how to handle. The MACs don't appear to generate pause frames,
> and the pause frames that they receive are trapped to the CPU as
> link-local traffic (DMAC 01-80-C2-00-00-01) where they are simply
> consumed by the master's MAC.

Woah, okay :) I suppose this can be made to work if you accept loading
your host CPU a little bit and have it perform flow control instead of
the switch itself. There is no way to have the switch's internal
buffering automatically deal with pause frames?
-- 
Florian

Powered by blists - more mailing lists