------- Comment From alexs...@de.ibm.com 2021-04-27 07:51 EDT-------
Also verified on the groovy build

root@pok1-qz1-sr1-rk011-s21:~# uname -a
Linux pok1-qz1-sr1-rk011-s21 5.8.0-51-generic #57~lp1925452-Ubuntu SMP Mon Apr 
26 15:27:41 UTC 2021 s390x s390x s390x GNU/Linux

root@pok1-qz1-sr1-rk011-s21:~# ping -c 5 172.31.22.42
PING 172.31.22.42 (172.31.22.42) 56(84) bytes of data.
64 bytes from 172.31.22.42: icmp_seq=2 ttl=0 time=0.117 ms
64 bytes from 172.31.22.42: icmp_seq=3 ttl=0 time=0.080 ms
64 bytes from 172.31.22.42: icmp_seq=4 ttl=0 time=0.066 ms

-- 
You received this bug notification because you are a member of Kernel
Packages, which is subscribed to linux in Ubuntu.
https://bugs.launchpad.net/bugs/1925452

Title:
  [Ubuntu 21.04] net/mlx5: Fix HW spec violation configuring uplink

Status in Ubuntu on IBM z Systems:
  In Progress
Status in linux package in Ubuntu:
  In Progress
Status in linux source package in Focal:
  In Progress
Status in linux source package in Groovy:
  In Progress
Status in linux source package in Hirsute:
  In Progress

Bug description:
  SRU Justification:
  ==================

  [Impact]

  * In addition to 9c9be85f6b59 "net/mlx5e: Add missing capability check
  for uplink follow" (handled in LP#1921104) another fix 1a73704c82ed
  "Fix HW spec violation configuring uplink" (handled in this bug,
  LP#1925452) is needed to fix issues that were introduced with commit
  7d0314b11cdd "net/mlx5e: Modify uplink state on interface up/down".

  * Commit 1a73704c82ed "Fix HW spec violation configuring uplink" fixes
  a regression for mlx5 adapters required to operate in switchdev mode.

  * This fix makes sure that the uplink port is modified to follow only
  if the uplink_follow capability if it's set as required by the hw
  specification.

  * Failure cause traffic to the uplink representer net device to cease
  after switching to switchdev mode.

  [Fix]

  * upstream fix (upstream with v5.12-rc7)
  1a73704c82ed4ee95532ac04645d02075bd1ce3d 1a73704c82ed "Fix HW spec violation 
configuring uplink"

  * can be cleanly cherry picked from hirsute master-next.

  * a backport for groovy:
  
https://launchpadlibrarian.net/534888680/groovy-0001-net-mlx5-Fix-HW-spec-violation-configuring-uplink.patch

  * a backport for focal:
  
https://launchpadlibrarian.net/534847308/focal-0001-net-mlx5-Fix-HW-spec-violation-configuring-uplink.patch

  [Test Case]

  * Two servers, installed with Ubuntu Server 20.04 or 20.10 are needed.

  * Each server needs to have a Mellanox ConnectX4/5 adapter, attached
  to the same switch

  * Adapters must be running adapter firmware level 16.29.1006 or
  earlier.

  * enable SRIOV and switchdev mode on one adapter:
    echo 0 > /sys/bus/pci/devices/0100\:00\:00.0/sriov_drivers_autoprobe
    echo 0 > /sys/bus/pci/devices/0100\:00\:00.1/sriov_drivers_autoprobe
    echo 64 > /sys/bus/pci/devices/0100\:00\:00.0/sriov_numvfs
    echo 64 > /sys/bus/pci/devices/0100\:00\:00.1/sriov_numvfs
    devlink dev eswitch set pci/0100:00:00.0 mode switchdev
    devlink dev eswitch set pci/0100:00:00.1 mode switchdev

  * Assign an IP address to the physical function device of the adapters
  on both systems

  * IP communication will fail

  * With the fix, IP communication can be established.

  [Regression Potential]

  * There is always at least some potential for regression. In this case
  the new code can go wrong (or might become worse than before) in case
  the new if statement is wrong.

  * It checks for the condition of "MLX5_CAP_GEN(mdev, uplink_follow)"
  and in case MLX5_CAP_GEN is calculated erroneous or mdev is other than
  expected, the mlx5_modify_vport_admin_state call might go wrong, too.

  * But since only the If clause was added, the changes are pretty
  minimal and therefore well traceable.

  [Other]

  * Since the patch/commit is upstream with 5.12-rc7, it will be in Impish.
  __________

  With the 5.4.0-48 update for Ubuntu 20.04, the commit

  "net/mlx5e: Modify uplink state on interface up/down"

  was integrated which contains a regression for mlx5 adapters with
  older adapter firmware.

  This is the second bugfix to repair the regression, required to operate the 
adapter in switchdev mode.
  The first part was integrated via Bug 192185 - LP1921104.

  This is the upstream fix required to be backported to Ubuntu 20.04:
  
https://github.com/torvalds/linux/commit/1a73704c82ed4ee95532ac04645d02075bd1ce3d

  ---Additional Hardware Info---
  Mellanox ConnectX network adapter

  ---uname output---
  5.4.0-73-generic

  ---Steps to Reproduce---
   enable switchdev mode
  try to ping external host

  Backports need to be provide for 21.04, 20.10 and 20.04 (if not
  cleanly applyable)!

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu-z-systems/+bug/1925452/+subscriptions

-- 
Mailing list: https://launchpad.net/~kernel-packages
Post to     : kernel-packages@lists.launchpad.net
Unsubscribe : https://launchpad.net/~kernel-packages
More help   : https://help.launchpad.net/ListHelp

Reply via email to