On Fri, Oct 09, 2020 at 01:54:34PM -0400, Qian Cai wrote: > On Fri, 2020-10-09 at 13:36 -0400, Qian Cai wrote: > > Back to x86, we have: > > > > start_secondary() > > smp_callin() > > apic_ap_setup() > > setup_local_APIC() > > printk() in certain conditions. > > > > which is before smp_store_cpu_info(). > > > > Can't we add a rcu_cpu_starting() at the very top for each start_secondary(), > > secondary_start_kernel(), smp_start_secondary() etc, so we don't worry about > > any printk() later? > > This is rather irony. rcu_cpu_starting() is taking a lock and then reports > itself. > > [ 8.826732][ T0] __lock_acquire.cold.76+0x2ad/0x3e0 > [ 8.826732][ T0] lock_acquire+0x1c8/0x820 > [ 8.826732][ T0] _raw_spin_lock_irqsave+0x30/0x50 > [ 8.826732][ T0] rcu_cpu_starting+0xd0/0x2c0 > [ 8.826732][ T0] start_secondary+0x10/0x2a0 > [ 8.826732][ T0] secondary_startup_64_no_verify+0xb8/0xbb Fun!!! There should be some way around this. I cannot safely record the offline-to-online transition without acquiring a lock. I suppose I could trick lockdep into thinking that it was a recursive lockdep report. Any other approaches? Thanx, Paul
