RCU is not watching idle threads because they are not scheduled
on busy CPUs and might block finishing grace periods. As a result,
the livepatch ftrace handler might see ops->func_stack and other
flags in a wrong state. Then a livepatch might make the system
unstable.
Note that there might be serious consequences only when the livepatch
modifies semantic of functions used by idle kthreads. We are safe when
none of the patched functions is used by the idle kthreads. Also
everything is good when the functions might be changed one by one
(using the immediate flag). See Documentation/livepatch/livepatch.txt
for more details about the consistency model.
This patch makes sure that even the idle threads see the critical
section by calling rcu_irq_enter_irqson(). The same fix was used
also for the stack tracer, see the commit a2d7629048322ae62b
("tracing: Have stack tracer force RCU to be watching").
Signed-off-by: Petr Mladek <pmladek@xxxxxxxx>
---
kernel/livepatch/patch.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/kernel/livepatch/patch.c b/kernel/livepatch/patch.c
index f8269036bf0b..4c4fbe409008 100644
--- a/kernel/livepatch/patch.c
+++ b/kernel/livepatch/patch.c
@@ -59,6 +59,9 @@ static void notrace klp_ftrace_handler(unsigned long ip,
ops = container_of(fops, struct klp_ops, fops);
+ /* RCU may not be watching, make it see us. */
+ rcu_irq_enter_irqson();
+
rcu_read_lock();
func = list_first_or_null_rcu(&ops->func_stack, struct klp_func,
@@ -116,6 +119,7 @@ static void notrace klp_ftrace_handler(unsigned long ip,
klp_arch_set_pc(regs, (unsigned long)func->new_func);
unlock:
rcu_read_unlock();
+ rcu_irq_exit_irqson();
}
/*
--
1.8.5.6
--
To unsubscribe from this list: send the line "unsubscribe live-patching" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
