Re: [BUG] WARNING: at kernel/trace/ring_buffer.c:3420

[Steven Rostedt]

On Tue, 2010-03-09 at 11:35 +0800, Lai Jiangshan wrote:
> Steven Rostedt wrote:
> > On Mon, 2010-03-08 at 10:03 -0500, Steven Rostedt wrote:
> > 
> >>> ringbuffer resizing and reseting will increase the ->record_disabled
> >>> and then wait until a rcu_shced grace period passes.
> >>>
> >>> Contrarily, testing ->record_disabled should be at the same
> >>> preempt disabled critical region as writing into ringbuffer, otherwise
> >>> it will leave a window break ringbuffer resizing or reseting.
> >> So the resizing and the resetting need a synchronize_sched() after the
> >> disabling of the buffers, right?
> > 
> > Looking at the code, the synchronize_sched() is already done in
> > ring_buffer_resize, and the caller (trace.c:tracing_reset() ) also
> > disables the ring buffer and calls synchronize_sched().
> > 
> > With that, what other window is still opened (after this fix)?
> > 
> 
> This window is still opened: (RCU vs IDLE vs Tracing)
> 
> synchronize_sched() does not protect preempt_disable()/enable() for
> idle process. But tracing(function_graph, function) introduce more
> preempt_disable()/enable() for idle process. It brings windows.

It most definitely should. If it does not, then we have major problems.
>From the comment above synchronize_sched(void):

/**
 * synchronize_sched - wait until an rcu-sched grace period has elapsed.
 *
 * Control will return to the caller some time after a full rcu-sched
 * grace period has elapsed, in other words after all currently executing
 * rcu-sched read-side critical sections have completed.   These read-side
 * critical sections are delimited by rcu_read_lock_sched() and
 * rcu_read_unlock_sched(), and may be nested.  Note that preempt_disable(),
 * local_irq_disable(), and so on may be used in place of
 * rcu_read_lock_sched().

I do not see how idle can be ignored here.

> 
> I bet that this bug is not come from this window.
> (I added some strict code to RCU and did stress test,
> bug was still occurred.)
> 

What exactly were you doing when this happened? Was this from Li's test?

-- Steve


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/