]> git.baikalelectronics.ru Git - kernel.git/commit
Merge branch 'make-dsa-work-with-bonding-s-arp-monitor'
authorJakub Kicinski <kuba@kernel.org>
Thu, 4 Aug 2022 02:20:17 +0000 (19:20 -0700)
committerJakub Kicinski <kuba@kernel.org>
Thu, 4 Aug 2022 02:20:18 +0000 (19:20 -0700)
commit76006d840c354303f5e447a0865469bf0c4d3afe
tree68ce7f01a394516f08b9d36579bbc3f1fb81bf64
parenta29c02028fe5452bfd19cdd3e4407aecfeb45ae6
parent42b35f50d773eb33dfca3bc8a7d3c1481e0af3d0
Merge branch 'make-dsa-work-with-bonding-s-arp-monitor'

Vladimir Oltean says:

====================
Make DSA work with bonding's ARP monitor

Since commit 8b8afca777b8 ("net: dsa: declare lockless TX feature for
slave ports") in v5.7, DSA breaks the ARP monitoring logic from the
bonding driver, fact which was pointed out by Brian Hutchinson who uses
a linux-5.10.y stable kernel.

Initially I got lured by other similar hacks introduced for other
NETIF_F_LLTX drivers, which, inspired by the bonding documentation,
update the trans_start of their TX queues by hand.

However Jakub pointed out that this simply isn't a proper solution, and
after coming to think more about it, I agree, and it doesn't work
properly with DSA nor is it maintainable for the future changes I plan
for it (multiple DSA masters in a LAG).

I've tested these changes using a DSA-based setup and a veth-based
setup, using the active-backup mode and ARP monitoring, with and without
arp_validate.

Link to v1:
https://patchwork.kernel.org/project/netdevbpf/patch/20220715232641.952532-1-vladimir.oltean@nxp.com/

Link to v2:
https://patchwork.kernel.org/project/netdevbpf/patch/20220727152000.3616086-1-vladimir.oltean@nxp.com/
====================

Link: https://lore.kernel.org/r/20220731124108.2810233-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>