[PATCH v2 0/4] PCI: Rework error reporting with PCIe failed link retraining

From: Maciej W. Rozycki
Date: Fri Aug 09 2024 - 09:24:58 EST


Hi,

This is v2 superseding a patch series originally posted here:
<https://lore.kernel.org/r/alpine.DEB.2.21.2402092125070.2376@xxxxxxxxxxxxxxxxx/>.

This patch series addresses issues observed by Ilpo as reported here:
<https://lore.kernel.org/r/aa2d1c4e-9961-d54a-00c7-ddf8e858a9b0@xxxxxxxxxxxxxxx/>,
one with excessive delays happening when `pcie_failed_link_retrain' is
called, but link retraining has not been actually attempted, and another
one with an API misuse caused by a merge mistake.

It also addresses an issue observed by Matthew as discussed here:
<https://lore.kernel.org/r/20240806000659.30859-1-mattc@xxxxxxxxxxxxxxx/>
and here:
<https://lore.kernel.org/r/20240722193407.23255-1-mattc@xxxxxxxxxxxxxxx/>.
where a stale LBMS bit state causes `pcie_failed_link_retrain', in the
absence of a downstream device, to leave the link stuck at the 2.5GT/s
speed rate, which then negatively impacts devices plugged in in the
future.

See individual change description for further details; 1/4 and 2/4 are
new changes, 3/4 supersedes:
<https://patchwork.kernel.org/project/linux-pci/patch/alpine.DEB.2.21.2402100045590.2376@xxxxxxxxxxxxxxxxx/>,
and 4/4 supersedes:
<https://patchwork.kernel.org/project/linux-pci/patch/alpine.DEB.2.21.2402100048440.2376@xxxxxxxxxxxxxxxxx/>.

These changes have been verified with a SiFive HiFive Unmatched system,
also using a small debug change to verify that the state of the LBMS bit
is clear at the exit from `pcie_failed_link_retrain'.

Ilpo, since 3/4 and 4/4 have only been trivially updated and their
combined effect is not changed even I chose to retain your Reviewed-by
tags from v1. Let me know if you disagree and what to do so you don't.

I apologise to take so long, it's been a tough period to me load-wise.

Maciej