Re: [tip:sched/core] [sched/smp] 06ddd17521: BUG:using_smp_processor_id()in_preemptible

["Chen, Yu C" <yu.c.chen@intel.com>]

Hi Sebastian,

On 7/10/2025 2:25 PM, Sebastian Andrzej Siewior wrote:
> On 2025-07-10 11:18:29 [+0800], Chen, Yu C wrote:
>>> [ 25.235357][ T1] dump_stack_lvl (lib/dump_stack.c:123 (discriminator 1))
>>> [ 25.235357][ T1] check_preemption_disabled (arch/x86/include/asm/preempt.h:85 lib/smp_processor_id.c:53)
>>> [ 25.235357][ T1] __kvm_is_vmx_supported (arch/x86/include/asm/cpuid/api.h:74 arch/x86/include/asm/cpuid/api.h:113 arch/x86/kvm/vmx/vmx.c:2789)
>>> [ 25.235357][ T1] vmx_init (arch/x86/kvm/vmx/vmx.c:2808 arch/x86/kvm/vmx/vmx.c:8653)
>>> [ 25.235357][ T1] vt_init (arch/x86/kvm/vmx/main.c:1072)
> …
>> I took a glance at the warning, before this patch,
>> is_percpu_thread() always return true when CONFIG_SMP is not set.
>> After this patch,
>> is_percpu_thread() checks the current task's CPU affinity.
>> So debug_smp_processor_id() -> check_preemption_disabled() ->
>> is_percpu_thread() might not always return true anymore, which caused
>> the warning.
>>
>> Actually the issue is in __kvm_is_vmx_supported(), should
>> we use something like this below:
> 
> No, it should not. If you look closely you will see that the call chains
> is
>     vmx_init() -> kvm_is_vmx_supported() -> __kvm_is_vmx_supported()
> 
> There is a migrate_disable() around __kvm_is_vmx_supported(). So why
> does this warning trigger then?
> 

migrate_disable() disables the task migration between CPUs by restricting
the task's affinity, but it does not disable the preemption on single
CPUs IMO. The scope of guard(preempt)() in migrate_disable() is just
within the migrate_disable(). debug_smp_processor_id() warns when the
preemption is enabled.

Thanks,
Chenyu