From: Lan Tianyu Tianyu.Lan@microsoft.com
When add "nosmt" parameter, kernel still boots up all logical cpus once and set CR4.MCE on each CPU. This is to avoid shutting down machine when a broadacasted MCE is observed CR4.MCE=0b. (Detail please see comment in the cpu_smt_allowed()). Smt cpus will bring up and bring down during kernel boot with "nosmt" parameter.
When CONFIG_HOTPLUG_CPU=Y, CPU_DYING callbacks will be called inside stop-machine and irq is disabled. This happens in the take_cpu_down() callback. When CONFIG_HOTPLUG_CPU=N,CPU_DYING callbacks will be called with irq enabled.
smpcfd_dying_cpu() is one of CPU_DYING callbacks and it assumes to be called when irq is disabled. smpcfd_dying_cpu() calls flush_smp_call_ function_queue() which requires to be called with irq disabled.
When CONFIG_HOTPLUG_CPU=N and add "nosmt" parameter, smpcfd_dying_cpu() is called with irq enalbed and this triggers BUG_ON(!irqs_disabled()) in the irq_work_run_list(). This patch is to fix the issue.
Fixes: 0cc3cd21657b ("cpu/hotplug: Boot HT siblings at least once") Signed-off-by: Lan Tianyu Tianyu.Lan@microsoft.com --- kernel/smp.c | 5 +++++ 1 file changed, 5 insertions(+)
diff --git a/kernel/smp.c b/kernel/smp.c index f4cf1b0..33f1970 100644 --- a/kernel/smp.c +++ b/kernel/smp.c @@ -73,6 +73,8 @@ int smpcfd_dead_cpu(unsigned int cpu)
int smpcfd_dying_cpu(unsigned int cpu) { + unsigned long flags; + /* * The IPIs for the smp-call-function callbacks queued by other * CPUs might arrive late, either due to hardware latencies or @@ -82,7 +84,10 @@ int smpcfd_dying_cpu(unsigned int cpu) * ensure that the outgoing CPU doesn't go offline with work * still pending. */ + local_irq_save(flags); flush_smp_call_function_queue(false); + local_irq_restore(flags); + return 0; }