[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20200602122954.0c35072b@kicinski-fedora-PC1C0HJN.hsd1.ca.comcast.net>
Date: Tue, 2 Jun 2020 12:29:54 -0700
From: Jakub Kicinski <kuba@...nel.org>
To: Victor Julien <victor@...iniac.net>
Cc: Willem de Bruijn <willemdebruijn.kernel@...il.com>,
Network Development <netdev@...r.kernel.org>,
"David S. Miller" <davem@...emloft.net>,
Jonathan Corbet <corbet@....net>,
Eric Dumazet <edumazet@...gle.com>,
Mao Wenan <maowenan@...wei.com>, Arnd Bergmann <arnd@...db.de>,
Neil Horman <nhorman@...driver.com>, linux-doc@...r.kernel.org,
linux-kernel <linux-kernel@...r.kernel.org>,
Alexander Drozdov <al.drozdov@...il.com>,
Tom Herbert <tom@...bertland.com>
Subject: Re: [PATCH net-next v2] af-packet: new flag to indicate all csums
are good
On Tue, 2 Jun 2020 21:22:11 +0200 Victor Julien wrote:
> - receiver uses nfp (netronome) driver: TP_STATUS_CSUM_VALID set for
> every packet, including the bad TCP ones
> - receiver uses ixgbe driver: TP_STATUS_CSUM_VALID not set for the bad
> packets.
>
> Again purely based on 'git grep' it seems nfp does not support
> UNNECESSARY, while ixgbe does.
>
> (my original testing was with the nfp only, so now I at least understand
> my original thinking)
FWIW nfp defaults to CHECKSUM_COMPLETE if the device supports it (see
if you have RXCSUM_COMPLETE in the probe logs). It supports UNNECESSARY
as well, but IDK if there is a way to choose the preferred checksum
types in the stack :( You'd have to edit the driver and remove the
NFP_NET_CFG_CTRL_CSUM_COMPLETE from the NFP_NET_CFG_CTRL_RXCSUM_ANY
mask to switch to using UNNECESSARY.
Powered by blists - more mailing lists