Commit
53ae3acd (arm64: Only enable local interrupts after the CPU
is marked online) moved the enabling of the GIC after the CPUs are
marked online.
This has some interesting effect:
[...]
[<
ffffffc0002eefd8>] gic_raise_softirq+0xf8/0x160
[<
ffffffc000088f58>] smp_send_reschedule+0x38/0x40
[<
ffffffc0000c8728>] resched_task+0x84/0xc0
[<
ffffffc0000c8cdc>] check_preempt_curr+0x58/0x98
[<
ffffffc0000c8d38>] ttwu_do_wakeup+0x1c/0xf4
[<
ffffffc0000c8f90>] ttwu_do_activate.constprop.84+0x64/0x70
[<
ffffffc0000cad30>] try_to_wake_up+0x1d4/0x2b4
[<
ffffffc0000cae6c>] default_wake_function+0x10/0x18
[<
ffffffc0000c5ca4>] __wake_up_common+0x60/0xa0
[<
ffffffc0000c7784>] complete+0x48/0x64
[<
ffffffc000088bec>] secondary_start_kernel+0xe8/0x110
[...]
Here, we end-up calling gic_raise_softirq without having initialized
the interrupt controller for this CPU. While this goes unnoticed
with GICv2 (the distributor is always accessible), it explodes with
GICv3.
The fix is to move the call to notify_cpu_starting before we set
the secondary CPU online.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
if (cpu_ops[cpu]->cpu_postboot)
cpu_ops[cpu]->cpu_postboot();
+ /*
+ * Enable GIC and timers.
+ */
+ notify_cpu_starting(cpu);
+
/*
* OK, now it's safe to let the boot CPU continue. Wait for
* the CPU migration code to notice that the CPU is online
set_cpu_online(cpu, true);
complete(&cpu_running);
- /*
- * Enable GIC and timers.
- */
- notify_cpu_starting(cpu);
-
local_irq_enable();
local_fiq_enable();