netdev - Re: [PATCH net-next v2 4/5] ptp: Support late timestamp determination

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]

Date:   Sat, 16 Apr 2022 00:04:20 +0200
From:   Gerhard Engleder <gerhard@...leder-embedded.com>
To:     Richard Cochran <richardcochran@...il.com>
Cc:     Vinicius Costa Gomes <vinicius.gomes@...el.com>, yangbo.lu@....com,
        David Miller <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>, mlichvar@...hat.com,
        netdev <netdev@...r.kernel.org>
Subject: Re: [PATCH net-next v2 4/5] ptp: Support late timestamp determination

> > For igc and tsnep the 16 bytes in front of the RX frame exist anyway.
> > So this would be a minimal solution with best performance as a first
> > step. A lookup for netdev/phc can be added in the future if there is
> > a driver, which needs that.
>
> It is a design mistake to base new kernel interfaces on hardware
> quirks.
>
> > Is it worth posting an implementation in that direction?
>
> Sure, but please make thoughts about how this would work for the
> non-igc world.
>
> IIRC one of the nxp switches also has such counters?  You can start
> with that.

I did some measurements (A53, 1.2GHz): netdev lookup and call to
my driver takes ~400ns. ptp_convert_timestamp() takes ~6000ns, I
assume because of class_find_device_by_name() call. So eliminating
the netdev lookup is the wrong optimisation target.

I will try to do it like that:

- normal, driver without cycles support:   use hwtstamps->hwtstamp
directly (not changed)
- driver with cycles support:   netdev lookup for address/cookie
conversion to hwtstamp (newly implemented)
- vclock:   slow path with phc lookup (not changed)

Gerhard