Re: [PATCH 2/3] livepatch/rcu: Warn when system consistency is broken in RCU code

[Paul E. McKenney]

On Thu, May 11, 2017 at 03:52:46PM +0200, Petr Mladek wrote:
> On Mon 2017-05-08 15:13:22, Steven Rostedt wrote:
> > On Mon, 8 May 2017 11:51:08 -0500
> > Josh Poimboeuf <jpoimboe@xxxxxxxxxx> wrote:
> > 
> > > Another idea would be to figure out a way to stop using RCU in
> > > klp_ftrace_handler() altogether.
> > > 
> > 
> > That may work if rcu_enter_irq() doesn't. But that's how NMIs use rcu.
> 
> I am a bit confused by the above. Does it mean that RCU could not be
> used in NMI handlers?

Only RCU readers can be used in NMI handlers, that is, rcu_read_lock(),
rcu_read_unlock(), rcu_dereference(), and so on.

							Thanx, Paul

> Anyway, a crazy idea is to use the livepatch consistency model instead
> of RCU to protect the function stack. The model makes sure that all
> tasks, including the idle ones, were not running any patched function
> (and their ftrace handlers) at some point. It should be safe
> but I am not sure if it is worth it.
> 
> Alternatively, it might be enough to use the probably more lightwight
> solution that is used when ftrace handlers are deregistered, I mean:
> 
> 	/*
> 	 * We need to do a hard force of sched synchronization.
> 	 * This is because we use preempt_disable() to do RCU, but
> 	 * the function tracers can be called where RCU is not watching
> 	 * (like before user_exit()). We can not rely on the RCU
> 	 * infrastructure to do the synchronization, thus we must do it
> 	 * ourselves.
> 	 */
> 	schedule_on_each_cpu(ftrace_sync);
> 
> 	/*
> 	 * When the kernel is preeptive, tasks can be preempted
> 	 * while on a ftrace trampoline. Just scheduling a task on
> 	 * a CPU is not good enough to flush them. Calling
> 	 * synchornize_rcu_tasks() will wait for those tasks to
> 	 * execute and either schedule voluntarily or enter user space.
> 	 */
> 	if (IS_ENABLED(CONFIG_PREEMPT))
> 		synchronize_rcu_tasks();
> 
> 
> 
> Best  Regards,
> Petr
>