Thx Alex, that allowed me to fill the gaps and I have consolidated the SRU Justification and added it to the bug description.
** Description changed: SRU Justification: ================== [Impact] * Since older firmware may not support the uplink state setting, this can lead to problems. * Now expose firmware indication that it supports setting eswitch uplink state to follow the physical link. * If a kernel without the backport is used on an adapter which does not have the latest adapter firmware, the adapter silently drops outgoing traffic. * This is a regression which was introduced with kernel 5.4.0-48. [Fix] * upstream fix (as in 5.11): - 9c9be85f6b59d80efe4705109c0396df18d4e11d 9c9be85f6b59 "net/mlx5e: Add missing capability check for uplink follow" + 9c9be85f6b59d80efe4705109c0396df18d4e11d 9c9be85f6b59 "net/mlx5e: Add missing capability check for uplink follow" * backport for focal: https://launchpadlibrarian.net/529543695/0001 -Backport-net-mlx5e-Add-missing-capability-check-for-.patch * backport for groovy: https://launchpadlibrarian.net/529775887/0001 -Backport-groovy-net-mlx5e-Add-missing-capability-che.patch [Test Case] - * It requires an IBM Z or LinuxONE system, with groovy/focal installed - in LPAR and RoCE Express 2.x adapters attached. + * Two IBM Z or LinuxONE systems, installed with Ubuntu Server 20.04 or + 20.10 on LPAR, are needed. - * ... + * Each with RoCE Express 2.x adapters (Mellanox ConnectX4/5) attached + and firmware 16.29.1006 or earlier. + + * Assign an IP address to the adapters on both systems and try to ping + one node from the other. + + * The ping will just fail with the stock Ubuntu kernels (not having the + patch), but will succeed with kernels that incl. the patches (like the + test builds from the PPA mentioned below). * Due to the lack of hardware this needs to be verified by IBM. [Regression Potential] * Undesired / erroneous behavior in case the modified if condition is assembled in a wrong way. * Again wrong behavior in case the modification of the capability bits in mlx5_ifc_cmd_hca_cap_bits are wrong. * All modification are limited to the mlx5 driver only. * The changes are relatively limited with effectively two lines removed and 4 added (three of them adjustments of the capability bits only). * The modifications were done and tested by IBM and reviewed by Mellanox (see LP comments), based on a PPA test build. [Other] * The above patch/commit was upstream accepted with kernel 5.11. * Hence the patch is not needed for hirsute, just needs to be SRUed for groovy and focal. * The commit couldn't be cleanly cherry-picked, mainly due to changed context, hence the backport(s). __________ Expose firmware indication that it supports setting eswitch uplink state to follow (follow the physical link). Condition setting the eswitch uplink admin-state with this capability bit. Older FW may not support the uplink state setting. Available fix with kernel 5.11. https://github.com/torvalds/linux/commit/9c9be85f6b59d80efe4705109c0396df18d4e11d Now required for Ubuntu 20.04 via backport patch. -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1921104 Title: net/mlx5e: Add missing capability check for uplink follow for Ubuntu 20.04 To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu-z-systems/+bug/1921104/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
