netdev - Re: Link down reasons

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: Windows password security audit tool. GUI, reports in PDF.

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

Message-ID: <87zh9stocb.fsf@mellanox.com>
Date:   Thu, 28 May 2020 11:12:36 +0200
From:   Petr Machata <petrm@...lanox.com>
To:     Amit Cohen <amitc@...lanox.com>
Cc:     Andrew Lunn <andrew@...n.ch>, mlxsw <mlxsw@...lanox.com>,
        "netdev\@vger.kernel.org" <netdev@...r.kernel.org>,
        "o.rempel\@pengutronix.de" <o.rempel@...gutronix.de>
Subject: Re: Link down reasons

Amit Cohen <amitc@...lanox.com> writes:

> Andrew Lunn <andrew@...n.ch> writes:
>
>>On Wed, May 27, 2020 at 03:41:22PM +0000, Amit Cohen wrote:
>>> Hi Andrew,
>>>
>>> We are planning to send a set that exposes link-down reason in ethtool.
>>>
>>> It seems that the ability of your set “Ethernet cable test support”
>>> can be integrated with link-down reason.
>>>
>>>
>>>
>>> The idea is to expose reason and subreason (if there is):
>>>
>>> $ ethtool ethX
>>>
>>> …
>>>
>>> Link detected: no (No cable) // No sub reason
>>>
>>>
>>>
>>> $ ethtool ethY
>>>
>>> Link detected: no (Autoneg failure, No partner detected)
>>>
>>>
>>>
>>> Currently we have reason “cable issue” and subreasons “unsupported
>>> cable” and “shorted cable”.
>>>
>>> The mechanism of cable test can be integrated and allow us report “cable issue”
>>> reason and “shorted cable” subreason.
>>
>>I don't really see them being combinable. First off, your API seems
>>too limiting. How do you say which pair is broken, or at what
>>distance? What about open cable, as opposed to shorted cable?

Under the proposed API, open cable and shorted cable would be two
different subreasons of "cable issue" reason.

>>So i would suggest:
>>
>>Link detected: no (cable issue)
>>
>>And then recommend the user uses ethtool --cable-test to get all the
>>details, and you have a much more flexible API to provide as much or
>>as little information as you have.

Yeah, the API that we are thinking of makes this possible.

Andrew, pardon my ignorance in these matters, can a PHY driver in
general determine that the issue is with the cable, even without running
the fairly expensive cable test? I.e. is it reasonable to expect that
the driver can tell us, yeah, the cable is baked (=> cable issue), but
please run the cable test to learn the details?

Ideally we would just use the cable test when trying to determine the
link-down reason, but the peculiar reporting mechanism and the fact that
it is a lengthy operation prevent this.

> Link-down reason has to consider cable-test or not? In order to report
> "cable issue", we assume that the driver implemented link-down reason
> in addition to cable-test?

Note that the "driver" here can refer both to a MAC driver as well as a
PHY driver. Ethtool can in principle do a cascade of callbacks to
different subsystems when trying to figure out why the link is down.

> I'm asking about PHY driver for example that implemented cable-test
> and not link-down reason, so according to cable-test we should report
> "cable issue" as a link-down reason or do not expose reason here?