[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20190729190015.GA31413@splinter>
Date: Mon, 29 Jul 2019 22:00:15 +0300
From: Ido Schimmel <idosch@...sch.org>
To: Petr Machata <petrm@...lanox.com>
Cc: "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
Jiri Pirko <jiri@...lanox.com>,
Ido Schimmel <idosch@...lanox.com>,
"davem@...emloft.net" <davem@...emloft.net>
Subject: Re: [PATCH net v2] mlxsw: spectrum_ptp: Increase parsing depth when
PTP is enabled
On Mon, Jul 29, 2019 at 06:26:14PM +0000, Petr Machata wrote:
> Spectrum systems have a configurable limit on how far into the packet they
> parse. By default, the limit is 96 bytes.
>
> An IPv6 PTP packet is layered as Ethernet/IPv6/UDP (14+40+8 bytes), and
> sequence ID of a PTP event is only available 32 bytes into payload, for a
> total of 94 bytes. When an additional 802.1q header is present as
> well (such as when ptp4l is running on a VLAN port), the parsing limit is
> exceeded. Such packets are not recognized as PTP, and are not timestamped.
>
> Therefore generalize the current VXLAN-specific parsing depth setting to
> allow reference-counted requests from other modules as well. Keep it in the
> VXLAN module, because the MPRS register also configures UDP destination
> port number used for VXLAN, and is thus closely tied to the VXLAN code
> anyway.
>
> Then invoke the new interfaces from both VXLAN (in obvious places), as well
> as from PTP code, when the (global) timestamping configuration changes from
> disabled to enabled or vice versa.
>
> Fixes: 8748642751ed ("mlxsw: spectrum: PTP: Support SIOCGHWTSTAMP, SIOCSHWTSTAMP ioctls")
> Signed-off-by: Petr Machata <petrm@...lanox.com>
Reviewed-by: Ido Schimmel <idosch@...lanox.com>
Powered by blists - more mailing lists