| 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
| ||
|
Message-ID: <a554a5cd-541c-ebe5-c70e-953d0b8b1b87@enneenne.com>
Date: Thu, 26 May 2022 09:45:39 +0200
From: Rodolfo Giometti <giometti@...eenne.com>
To: Vladimir Oltean <olteanv@...il.com>
Cc: Jakub Kicinski <kuba@...nel.org>, Andrew Lunn <andrew@...n.ch>,
Matej Zachar <zachar.matej@...il.com>, netdev@...r.kernel.org
Subject: Re: [DSA] fallback PTP to master port when switch does not support it
On 25/05/22 17:55, Vladimir Oltean wrote:
> On Wed, May 25, 2022 at 05:00:24PM +0200, Rodolfo Giometti wrote:
>> On 07/04/22 11:44, Vladimir Oltean wrote:
>>> On Tue, Apr 05, 2022 at 12:48:51PM -0700, Jakub Kicinski wrote:
>>>> On Tue, 5 Apr 2022 00:04:30 +0200 Andrew Lunn wrote:
>>>>> What i don't like about your proposed fallback is that it gives the
>>>>> impression the slave ports actually support PTP, when they do not.
>>>>
>>>> +1, running PTP on the master means there is a non-PTP-aware switch
>>>> in the path, which should not be taken lightly.
>>>
>>> +2, the change could probably be technically done, and there are aspects
>>> worth discussing, but the goal presented here is questionable and it's
>>> best to not fool ourselves into thinking that the variable queuing delays
>>> of the switch are taken into account when reporting the timestamps,
>>> which they aren't.
>>>
>>> I think that by the time you realize that you need PTP hardware
>>> timestamping on switch ports but you have a PTP-unaware switch
>>> integrated *into* your system, you need to go back to the drawing board.
>>
>> IMHO this patch is a great hack but what you say sounds good to me.
>
> How many Ethernet connections are there between the switch and the host?
It depends how the hardware is designed. However usually the host has an
ethernet connected to a switch's port named CPU port, so just one.
> One alternative which requires no code changes is to connect one more
> switch port and run PTP at your own risk on the attached FEC port
> (not DSA master).
I see, but here we are talking about of not-so-well designed boards :( whose
have a switch that can't manage PTP and we still need to have a sort of time
sync. The trick can be to forward PTP packets to the host's ethernet (and
viceversa).
> What switch driver is it? There are 2 paths to be discussed.
> On TX, does the switch forward DSA-untagged packets from the host port? Where to?
> On RX, does the switch tag all packets with a DSA header towards the
> host? I guess yes,
Of course it's in charge of the DSA to properly setup the switch in order to
abstract all switch's ports as host's ethernet ports, so all packets go where
they have to go.
> but in that case, be aware that not many Ethernet
> controllers can timestamp non-PTP packets. And you need anyway to demote
> e.g. HWTSTAMP_FILTER_PTP_V2_EVENT to HWTSTAMP_FILTER_ALL when you pass
> the request to the master to account for that, which you are not doing.
Mmm... I can't see problems here... can you please explain it?
>> However we can modify the patch in order to leave the default behavior as-is
>> but adding the ability to enable this hack via DTS flag as follow:
>>
>> ports {
>> #address-cells = <1>;
>> #size-cells = <0>;
>>
>> port@0 {
>> reg = <0>;
>> label = "lan1";
>> allow-ptp-fallback;
>> };
>>
>> port@1 {
>> reg = <1>;
>> label = "lan2";
>> };
>>
>> ...
>>
>> port@5 {
>> reg = <5>;
>> label = "cpu";
>> ethernet = <&fec>;
>>
>> fixed-link {
>> speed = <1000>;
>> full-duplex;
>> };
>> };
>> };
>>
>> Then the code can do as follow:
>>
>> static int dsa_slave_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
>> {
>> struct dsa_slave_priv *p = netdev_priv(dev);
>> struct dsa_switch *ds = p->dp->ds;
>> int port = p->dp->index;
>> struct net_device *master = dsa_slave_to_master(dev);
>>
>> /* Pass through to switch driver if it supports timestamping */
>> switch (cmd) {
>> case SIOCGHWTSTAMP:
>> if (ds->ops->port_hwtstamp_get)
>> return ds->ops->port_hwtstamp_get(ds, port, ifr);
>> if (p->dp->allow_ptp_fallback && master->netdev_ops->ndo_do_ioctl)
>> return master->netdev_ops->ndo_do_ioctl(master, ifr, cmd);
>> break;
>> case SIOCSHWTSTAMP:
>> if (ds->ops->port_hwtstamp_set)
>> return ds->ops->port_hwtstamp_set(ds, port, ifr);
>> if (p->dp->allow_ptp_fallback && master->netdev_ops->ndo_do_ioctl)
>> return master->netdev_ops->ndo_do_ioctl(master, ifr, cmd);
>> break;
>> }
>>
>> return phylink_mii_ioctl(p->dp->pl, ifr, cmd);
>> }
>>
>> In this manner the default behavior is to return error if the switch doesn't
>> support the PTP functions, but developers can intentionally enable the PTP
>> fallback on specific ports only in order to be able to use PTP on buggy
>> hardware.
>>
>> Can this solution be acceptable?
>
> Generally we don't allow policy configuration through the device tree.
I agree, but here we have to signal an hardware that can't do something and
(IMHO) the device tree is a good point to address it. :)
Ciao,
Rodolfo
--
GNU/Linux Solutions e-mail: giometti@...eenne.com
Linux Device Driver giometti@...ux.it
Embedded Systems phone: +39 349 2432127
UNIX programming skype: rodolfo.giometti
Powered by blists - more mailing lists