Re: [REGRESSION][BISECTED] Early boot hangs since f0551af02130

[Rob <rob@durendal.co.uk>]

>So in theory the same problem should happen with a pre 6.9 kernel and 
>'nr_cpus=4' on the kernel command line. Can you please verify?

Confirmed, 6.6.40-2-lts with 'nr_cpus=4' hangs

>Another data point to be evaluated is whether booting the unmodified 
>6.10+ kernel with 'nointremap' instead of 'iommu=off' on the command 
>line makes the machine boot as well.

linux-mainline 6.10-1.1 with 'nointremap' boots

>... connect that machine over serial ...

Unfortunately I don't have the hardware for this

Kind regards,

Rob


* Thomas Gleixner (tglx@linutronix.de) wrote:
>On Fri, Jul 26 2024 at 00:01, Christian Heusel wrote:
>> Sorry for causing some confusion on that part and I hope the above
>> outputs help with debugging this issue!
>
>Not really. The issue is completely unclear and nothing makes any sense
>here.
>
>The only difference is the resulting number of possible CPUs.
>
>[    0.026516] CPU topo: Allowing 4 present CPUs plus 12 hotplug CPUs
>
>versus:
>
>[    0.025251] CPU topo: Allowing 4 present CPUs plus 0 hotplug CPUs
>
>which really should not make any difference because the 1300X has only 4
>CPUs.
>
>So the bisected commit is the messenger for some other problem
>presumably in the IOMMU area which is just papered over when there are
>12 possible hotplug CPUs enumerated.
>
>So in theory the same problem should happen with a pre 6.9 kernel and
>'nr_cpus=4' on the kernel command line. Can you please verify?
>
>Another data point to be evaluated is whether booting the unmodified
>6.10+ kernel with 'nointremap' instead of 'iommu=off' on the command
>line makes the machine boot as well.
>
>Thanks,
>
>        tglx

-- 
Rob Newcater