Re: [PATCH] tracing: Have stack tracer call rcu_nmi_enter() only when RCU is not watching

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Thu, Sep 21, 2017 at 01:10:38PM -0400, Steven Rostedt wrote:
> 
> Testing some other code, this RCU splat triggered:
> 
>  =============================
>  WARNING: suspicious RCU usage
>  4.13.0-rc7-test+ #117 Tainted: G        W
>  -----------------------------
>  arch/x86/kernel/traps.c:305 entry code didn't wake RCU!
> 
>  other info that might help us debug this:
> 
>  RCU used illegally from idle CPU!
>  rcu_scheduler_active = 2, debug_locks = 1
>  RCU used illegally from extended quiescent state!
>  no locks held by swapper/1/0.
> 
>  stack backtrace:
>  CPU: 1 PID: 0 Comm: swapper/1 Tainted: G        W       4.13.0-rc7-test+ #117
>  Hardware name: Hewlett-Packard HP Compaq Pro 6300 SFF/339A, BIOS K01 v03.03 07/14/2016
>  Call Trace:
>   dump_stack+0x86/0xcf
>   lockdep_rcu_suspicious+0xc5/0x100
>   do_error_trap+0x125/0x130
>   ? do_error_trap+0x5/0x130
>   ? trace_hardirqs_off_thunk+0x1a/0x1c
>   ? do_invalid_op+0x5/0x30
>   do_invalid_op+0x20/0x30
>   invalid_op+0x1e/0x30
>  RIP: 0010:module_assert_mutex_or_preempt+0x34/0x40
>  RSP: 0018:ffffc900006abc58 EFLAGS: 00010046
>  RAX: 0000000000000000 RBX: ffffffffa000a077 RCX: 0000000000000002
>  RDX: 0000000000000000 RSI: 00000000ffffffff RDI: 0000000000000046
>  RBP: ffffc900006abc58 R08: ffffc900006abf40 R09: 0000000000000000
>  R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000
>  R13: 0000000000000000 R14: ffff8801188d8040 R15: ffffffff81ed5720
>   ? 0xffffffffa000a077
>   ? module_assert_mutex_or_preempt+0x30/0x40
>   __module_address+0x2c/0xf0
>   ? 0xffffffffa000a077
>   __module_text_address+0x12/0x60
>   ? 0xffffffffa000a077
>   is_module_text_address+0x1f/0x50
>   ? 0xffffffffa000a077
>   __kernel_text_address+0x30/0x90
>   unwind_get_return_address+0x1f/0x30
>   __save_stack_trace+0x83/0xd0
>   ? 0xffffffffa000a077
>   ? rcu_dynticks_eqs_exit+0x5/0x40
>   save_stack_trace+0x1b/0x20
>   check_stack+0xf8/0x2f0
>   ? rcu_dynticks_eqs_enter+0x30/0x30
>   stack_trace_call+0x6e/0x80
>   0xffffffffa000a077
>   ? ftrace_graph_caller+0x78/0xa8
>   ? rcu_dynticks_eqs_exit+0x5/0x40
>   rcu_dynticks_eqs_exit+0x5/0x40
>   rcu_idle_exit+0xdf/0xf0
>   ? rcu_dynticks_eqs_exit+0x5/0x40
>   ? rcu_idle_exit+0xdf/0xf0
>   do_idle+0x128/0x200
>   cpu_startup_entry+0x1d/0x20
>   start_secondary+0x108/0x130
>   secondary_startup_64+0x9f/0x9f
> 
> What happened, was that the CPU was coming out of idle, and
> rcu_dynticks_eqs_exit() was traced by the stack tracer. The stack was bigger
> than the current recorded max stack so the stack tracer triggered a stack
> trace calling save_stack_trace(). As save_stack_trace() performs RCU logic,
> the stack tracer called rcu_irq_enter() before calling save_stack_trace().
> There is two problems here.
> 
> 1) rcu_irq_enter() is called unconditionally
> 
> Talking with Paul McKenney, it should only be called if rcu_is_watching()
> returns false.
> 
> 2) rcu_irq_enter() appears to not function well in this location of the RCU
>    internal logic (tracing rcu_dynticks_eqs_exit()), and causes this bug.
> 
> The solution is two fold.
> 
> 1) check if RCU is watching, and only wake RCU if it is not
>    (solves the first issue)
> 
> 2) call rcu_nmi_enter() instead of rcu_irq_enter(). It makes more sense as
>    tracing acts more like NMIs and not IRQs (can be called when interrupts
>    are disabled. (solves the second issue).
> 
> Suggested-by: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
> Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>

Looks good!

Acked-by: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>

> ---
>  kernel/trace/trace_stack.c | 10 ++++++++--
>  1 file changed, 8 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/trace/trace_stack.c b/kernel/trace/trace_stack.c
> index a4df67c..2462e11 100644
> --- a/kernel/trace/trace_stack.c
> +++ b/kernel/trace/trace_stack.c
> @@ -77,6 +77,7 @@ check_stack(unsigned long ip, unsigned long *stack)
>  {
>  	unsigned long this_size, flags; unsigned long *p, *top, *start;
>  	static int tracer_frame;
> +	bool no_rcu;
>  	int frame_size = ACCESS_ONCE(tracer_frame);
>  	int i, x;
> 
> @@ -104,14 +105,18 @@ check_stack(unsigned long ip, unsigned long *stack)
>  	if (unlikely(rcu_irq_enter_disabled()))
>  		return;
> 
> +	no_rcu = !rcu_is_watching();
> +
>  	local_irq_save(flags);
>  	arch_spin_lock(&stack_trace_max_lock);
> 
>  	/*
>  	 * RCU may not be watching, make it see us.
>  	 * The stack trace code uses rcu_sched.
> +	 * Note, tracing acts more like an NMI than an IRQ.
>  	 */
> -	rcu_irq_enter();
> +	if (no_rcu)
> +		rcu_nmi_enter();
> 
>  	/* In case another CPU set the tracer_frame on us */
>  	if (unlikely(!frame_size))
> @@ -205,7 +210,8 @@ check_stack(unsigned long ip, unsigned long *stack)
>  	}
> 
>   out:
> -	rcu_irq_exit();
> +	if (no_rcu)
> +		rcu_nmi_exit();
>  	arch_spin_unlock(&stack_trace_max_lock);
>  	local_irq_restore(flags);
>  }
> -- 
> 2.9.5
>