Re: [BUG] sched: BUG() in select_fallback_rq() at boot with isolcpus=domain on the boot CPU and CONFIG_CPUSETS=n

From: Mathieu Desnoyers

Date: Wed Nov 18 2026 - 12:35:20 EST


Hi Peter,

On 2026-11-17 10:02, Peter Zijlstra wrote:
> Arguably, marking the boot cpu as isolated is asking for it. Is this
> really worth fixing? Or should we look for a better failure message?

A clear failure would be enough for me. Today the kernel hits a BUG()
in select_fallback_rq(), and nothing in the log points at isolcpus.

It is also a regression. The same command line boots with
CONFIG_CPUSETS=y. With CONFIG_CPUSETS=n, it boots with bf5b0c27fad2
("sched: Switch the fallback task allowed cpumask to HK_TYPE_DOMAIN")
reverted.

Waiman's plan to refuse domain isolation of the boot CPU would cover
both, if the refusal prints a message and the boot goes on. It changes
one setup that works today: isolcpus=domain,0,1 with CONFIG_CPUSETS=y.

Waiman, I can test a patch in the same VMs, with and without
CONFIG_CPUSETS.

Thanks,

Mathieu

--
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com