[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20221207154826.5477008b@kernel.org>
Date: Wed, 7 Dec 2022 15:48:26 -0800
From: Jakub Kicinski <kuba@...nel.org>
To: Oleksij Rempel <o.rempel@...gutronix.de>
Cc: Woojung Huh <woojung.huh@...rochip.com>,
Andrew Lunn <andrew@...n.ch>, Arun.Ramadoss@...rochip.com,
Florian Fainelli <f.fainelli@...il.com>,
"David S. Miller" <davem@...emloft.net>, netdev@...r.kernel.org,
linux-kernel@...r.kernel.org, UNGLinuxDriver@...rochip.com,
Eric Dumazet <edumazet@...gle.com>, kernel@...gutronix.de,
Paolo Abeni <pabeni@...hat.com>,
Vladimir Oltean <olteanv@...il.com>,
Vivien Didelot <vivien.didelot@...il.com>
Subject: Re: [PATCH net-next v1 1/1] net: dsa: microchip: add stats64
support for ksz8 series of switches
On Wed, 7 Dec 2022 07:16:30 +0100 Oleksij Rempel wrote:
> > FWIW for normal netdevs / NICs the rtnl_link_stat pkts do not include
> > pause frames, normally. Otherwise one can't maintain those stats in SW
> > (and per-ring stats, if any, don't add up to the full link stats).
> > But if you have a good reason to do this - I won't nack..
>
> Pause frames are accounted by rx/tx_bytes by HW. Since pause frames may
> have different size, it is not possible to correct byte counters, so I
> need to add them to the packet counters.
I have embarrassed myself with my lack of understanding of pause frames
before but nonetheless - are you sure? I thought they are always 64B.
Quick look at the standard seems to agree:
31C.3.1 Receive state diagram (INITIATE MAC CONTROL FUNCTION) for
EXTENSION operation
shows a 64 octet frame.
Sending long pause frames seems self-defeating as we presumably want
the receiver to react ASAP.
Powered by blists - more mailing lists