Re: [RFC][PATCH] lockdep: Print a nice description of an irq locking issue

[Steven Rostedt <rostedt@goodmis.org>]

On Tue, 2011-03-29 at 13:14 -0400, Steven Rostedt wrote:
>    
>      ---
>     other info that might help us debug this:
>     
>     Chain exists of:
>       &rq->lock --> lockA --> lockC
>     
>      Possible interrupt unsafe locking scenario:
>     
>            CPU0                    CPU1
>            ----                    ----
>       lock(lockC);
>                                    local_irq_disable();
>                                    lock(&rq->lock);
>                                    lock(lockA);
>       <Interrupt>
>         lock(&rq->lock)
>     
>      *** DEADLOCK ***
>     

Note, the above output is real. To produce this output, I wrote a module
that created a "lockA", "lockB" and "lockC" and had the following:

        spin_lock_irq(&lockA);
        spin_lock(&lockB);
        spin_unlock(&lockB);
        spin_unlock_irq(&lockA);

        spin_lock_irq(&lockB);
        spin_lock(&lockC);
        spin_unlock(&lockC);
        spin_unlock_irq(&lockB);

        spin_lock(&lockC);
        spin_unlock(&lockC);
        ret = register_trace_sched_switch(probe_switch, NULL);

static void
probe_switch(void *ignore, struct task_struct *p, struct task_struct *n)
{
        unsigned long flags;

        spin_lock_irqsave(&lockA, flags);
        spin_unlock_irqrestore(&lockA, flags);
}


probe_switch is called via the trace_sched_switch() trace point that is
called with the rq lock held, producing the call chain that will trigger
lockdep to produce a dump.

-- Steve