[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <661cc33a-5f65-2769-cc1a-65791cb4b131@pengutronix.de>
Date: Wed, 8 Jan 2020 15:55:34 +0100
From: Ahmad Fatoum <a.fatoum@...gutronix.de>
To: linux-can@...r.kernel.org, netdev@...r.kernel.org,
pabeni@...hat.com
Cc: Sascha Hauer <kernel@...gutronix.de>
Subject: [BUG] pfifo_fast may cause out-of-order CAN frame transmission
Hello,
I've run into an issue of CAN frames being sent out-of-order on an i.MX6 Dual
with Linux v5.5-rc5. Bisecting has lead me down to this commit:
ba27b4cdaaa ("net: dev: introduce support for sch BYPASS for lockless qdisc")
With it, using pfifo_fast, every few hundred frames, FlexCAN's .ndo_start_xmit is
passed frames in an order different from how userspace stuffed them into the same
socket.
Reverting it fixes the issue as does booting with maxcpus=1 or using pfifo
instead of pfifo_fast.
According to [1], such reordering shouldn't be happening.
Details on my setup:
Kernel version: v5.5-rc5, (occurs much more often with LOCKDEP turned on)
CAN-Bitrate: 250 kbit/s
CAN frames are generated with:
cangen canX -I2 -L1 -Di -i -g0.12 -p 100
which keeps polling after ENOBUFS until socket is writable, sends out a CAN
frame with one incrementing payload byte and then waits 120 usec before repeating.
Please let me know if any additional info is needed.
Cheers
Ahmad
[1]: http://linux-tc-notes.sourceforge.net/tc/doc/sch_pfifo_fast.txt
--
Pengutronix e.K. | |
Steuerwalder Str. 21 | http://www.pengutronix.de/ |
31137 Hildesheim, Germany | Phone: +49-5121-206917-0 |
Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
Powered by blists - more mailing lists