Re: [EXT] Re: [Intel-wired-lan] Intel I350 regression 5.10 -> 5.14 ("The NVM Checksum Is Not Valid") [8086:1521]

On Tue, Oct 12, 2021 at 7:42 PM Paul Menzel [off-list ref] wrote:

[Cc: +ACPI maintainers]

Am 12.10.21 um 18:34 schrieb Andreas K. Huettel:

quoted

quoted

quoted

The messages easily identifiable are:

huettel@pinacolada ~/tmp $ cat kernel-messages-5.10.59.txt |grep igb
Oct  5 15:11:18 dilfridge kernel: [    2.090675] igb: Intel(R) Gigabit Ethernet Network Driver
Oct  5 15:11:18 dilfridge kernel: [    2.090676] igb: Copyright (c) 2007-2014 Intel Corporation.
Oct  5 15:11:18 dilfridge kernel: [    2.090728] igb 0000:01:00.0: enabling device (0000 -> 0002)

This line is missing below, it indicates that the kernel couldn't or
didn't power up the PCIe for some reason. We're looking for something
like ACPI or PCI patches (possibly PCI-Power management) to be the
culprit here.

So I did a git bisect from linux-v5.10 (good) to linux-v5.14.11 (bad).

The result was:

dilfridge /usr/src/linux-git # git bisect bad
6381195ad7d06ef979528c7452f3ff93659f86b1 is the first bad commit
commit 6381195ad7d06ef979528c7452f3ff93659f86b1
Author: Rafael J. Wysocki [off-list ref]
Date:   Mon May 24 17:26:16 2021 +0200

     ACPI: power: Rework turning off unused power resources
[...]

I tried naive reverting of this commit on top of 5.14.11. That applies nearly cleanly,
and after a reboot the additional ethernet interfaces show up with their MAC in the
boot messages.

(Not knowing how safe that experiment was, I did not go further than single mode and
immediately rebooted into 5.10 afterwards.)

Reverting this is rather not an option, because the code before it was
a one-off fix of an earlier issue, but it should be fixable given some
more information.

Basically, I need a boot log from both the good and bad cases and the
acpidump output from the affected machine.