How to quickly resolve the IOMMU regression that currently plagues a lot of people in 5.19.y

[Thorsten Leemhuis]

[resend with proper subject, sorry for the noise]

[note to self: don't get distracted when writing the subject]

On 21.09.22 08:53, Thorsten Leemhuis wrote:
> Hi Greg! As you likely heard already, 5.19.9 introduced a regression
> that breaks Thunderbolt and USB-C docks (and apparently Wifi in some
> cases as well) on quite a few (many?) modern systems. It's one of those
> problems where I think "hey, we ideally should fix this in stable as
> fast as possible" we briefly talked about last week on the LPC hallways.
> That made me wonder how to actually archive that in this particular case
> while keeping all involved parties happy and not skipping any CI testing
> queues considered important.
> 
> FWIW, here are a few few reports about the issue (I assume there are
> some for Arch Linux and openSUSE Tumbleweed as well, but didn't check).
> 
> https://lore.kernel.org/linux-iommu/485A6EA5-6D58-42EA-B298-8571E97422DE@xxxxxxxxxxxxxxxxx/
> https://bugzilla.kernel.org/show_bug.cgi?id=216497
> https://bugzilla.redhat.com/show_bug.cgi?id=2128458
> https://bugzilla.redhat.com/show_bug.cgi?id=2127753
> 
> A revert of the culprit (9cd4f1434479f ("iommu/vt-d: Fix possible
> recursive locking in intel_iommu_init()"); in 5.19.y it's 	9516acba29e3)
> for mainline is here:
> https://lore.kernel.org/lkml/20220920081701.3453504-1-baolu.lu@xxxxxxxxxxxxxxx/
> 
> A few hours ago the revert was queued to get send to Joerg:
> https://lore.kernel.org/linux-iommu/20220921024054.3570256-1-baolu.lu@xxxxxxxxxxxxxxx/
> 
> I fear it could easily take another week to get this fixed in stable
> depending on how fast the patch makes it to mainline and the timing of
> the next 5.19.y release and its -rc phase. That to me sounds like way
> too long for a problem like this that apparently plagues quite a few
> people.
> 
> That made me wonder: would you in cases like this be willing to start
> the -rc phase for a interim 5.19.y release with just that revert while
> it's still heading towards mainline? Then the CI systems that test
> stable -rcs could chew on things already; and the new stable release
> could go out right after the revert landed in mainline (unless the
> testing finds any problems, of course).
> 
> Ciao, Thorsten
> 
>