Re: [PATCH 0/3] ACPI: Initialize GPEs before the initial namespace scan

[Mika Westerberg]

On Thu, Aug 10, 2017 at 12:29:12AM +0200, Rafael J. Wysocki wrote:
> Hi,
> 
> This is exceptional for at least two reasons.
> 
> First, we need it now to work around boot problems on multiple platforms
> already seen in the field (which are shipping products).
> 
> Second, the ACPICA changes in this series are kind of Linux-specific, because
> they are related to how Linux carries out the initialization of devices, which
> very well may be different from what the other OSes using ACPICA do.
> 
> For these reasons, I'd like to make the following ACPICA changes in Linux
> only for the time being and then decide whether or not to take them into
> the upstream.  If there are major concerns about that, please let me know.
> 
> The issue at hand is that some platforms with Thunderbolt controllers won't
> boot if there are any Thunderbolt devices connected to them at boot time (if
> the devices are connected later, everything works as expected).  That turns
> out to be related to a complicated sequence of events involving the platform
> firmware which needs to happen in exactly the right order at the right time
> for things to work and that requires GPEs to be enabled before enumerating
> the PCI bus.
> 
> The first patch changes ACPICA to check the status of runtime GPEs before
> enabling them for the first time in order to avoid missing events (that
> is key for edge-triggered GPEs) and to process them early enough.
> 
> The second one makes it possible to change the ordering of initialization
> between GPEs and devices, and the third one actually changes that ordering.
> 
> The patches are based on Mika's work, so kudos to him.

Thank you for taking care of this! I've tested the series on Intel Skull
Canyon NUC, Dell XPS 9350, 9550 and 9365, and it fixes the issue on all
of them.

Tested-by: Mika Westerberg <mika.westerberg@xxxxxxxxxxxxxxx>