NAK/Cmnt: [SRU][J:linux-bluefield][PATCH v1 0/1] UBUNTU: SAUCE: mlxbf-gige: Vitesse PHY stuck in a bad state during reboot test

[Bartlomiej Zolnierkiewicz]

Hi Asmaa,

This patch fails to build:

/build/jammy/drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_main.c: In func
tion 'mlxbf_gige_shutdown':
/build/jammy/drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_main.c:569:33:
error: 'MLXBF_GIGE_BLUEFIELD3' undeclared (first use in this function)
  569 |         if (priv->hw_version == MLXBF_GIGE_BLUEFIELD3)
      |                                 ^~~~~~~~~~~~~~~~~~~~~

--
Best regards,
Bartlomiej

On Mon, Apr 29, 2024 at 10:02 PM Asmaa Mnebhi <asmaa at nvidia.com> wrote:
>
> BugLink: https://bugs.launchpad.net/bugs/2062384
>
> SRU Justification:
>
> [Impact]
>
> During the QA reboot test, the BF3 Vitesse PHY gets stuck in a bad state, resulting in no ip provisioning. The only way to recover is to powercycle.
> We might have found a software workaround to avoid getting in this state in the first place: suspend the PHY during graceful shutdown. Suspend the PHY = Power down = set bit 11 to 1 in reg 0 of the PHY. This WA passed 1800 reboots on QA's setup.
>
> [Fix]
>
> * During reboot, the mlxbf_gige_shutdown() function makes a call to phy_stop(). phy_stop() calls phy_suspend().
> * Certain Linux PHY drivers, like the Vitesse PHY, don't support suspend() to power down the PHY during shutdown.
> * Our Hardware also does not toggle the hard reset signal of the PHY during reboot.
> * Hence, when the PHY is in a bad state, it stays in its bad state until powercycle.
> * We have found a way to prevent the PHY from entering this bad state by suspending the PHY in the case of reboot.
>
> [Test Case]
>
> * do the reboot test (at least 2000 reboots): run 'reboot' from linux.
> * Check that the oob_net0 interface is up and the ip is assigned.
> * please note that if the the OOB doesn't get an ip, try reloading the driver (rmmod/modprobe). it that solves the issue, that would be a different bug. In the bug at stake, nothing recovers the OOB ip except power cycle.
>
> [Regression Potential]
>
> * Make sure the redfish DHCP is still working during the reboot test
> * Make sure the OOB gets an ip
>
> [Other]
>
> These changes were made both in the mlxbf-gige driver and UEFI
>