Adding Frederic,
https://lkml.org/lkml/2018/10/11/304
On (12/11/18 16:32), Feng Tang wrote:
Here is the v1 patch: https://lkml.org/lkml/2018/10/11/304
And actually no one ruled out the v1 patch :), I don't have HW of other archs like arm/ppc, so I just read some of the arch code, and found most of them use the similar flow like x86, that's why I chosed to finding a soluton inside panic.c itself.
Interesting. So if the problem is that we need to clear cpu bit in several cpumaks (e.g. nohz.idle_cpus_mask) when we stop_this_cpu(), then I'd say let's clear cpumasks which are needed to be clear (doing some of the things which sched_cpu_dying() does, except that we need it on !CONFIG_HOTPLUG_CPU systems too). The idea of notifiers also looks interesting.
x86 and sched gurus, can you please help?
-ss