Re: [PATCH 1/7] lockdep: Print a nice description of an irq locking issue

[Yong Zhang]

On Thu, Apr 21, 2011 at 9:41 AM, Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> Â---
>
> The above is the case when the unsafe lock is taken while holding
> a lock taken in irq context. But when a lock is taken that also
> grabs a unsafe lock, the call chain is shown:
>
> Â---
> other info that might help us debug this:
>
> Chain exists of:
> Â&rq->lock --> lockA --> lockC
>
> ÂPossible interrupt unsafe locking scenario:
>
> Â Â Â CPU0 Â Â Â Â Â Â Â Â Â ÂCPU1
> Â Â Â ---- Â Â Â Â Â Â Â Â Â Â----
> Âlock(lockC);
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â local_irq_disable();
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â lock(&rq->lock);
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â lock(lockA);
> Â<Interrupt>
> Â Âlock(&rq->lock);
>
> Â*** DEADLOCK ***

But this is not a real deadlock, right? Or maybe you can teach me :)

I just assume your scenario should show a more real one, But for
a possible irq-dependence deadlock, it seems not easy to find out.

Thanks,
Yong

>
> Acked-by: Peter Zijlstra <a.p.zijlstra@xxxxxxxxx>
> Signed-off-by: Steven Rostedt <rostedt@xxxxxxxxxxx>
> ---
> Âkernel/lockdep.c | Â 70 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
> Â1 files changed, 70 insertions(+), 0 deletions(-)
>
> diff --git a/kernel/lockdep.c b/kernel/lockdep.c
> index 0d2058d..bb77c030 100644
> --- a/kernel/lockdep.c
> +++ b/kernel/lockdep.c
> @@ -490,6 +490,18 @@ void get_usage_chars(struct lock_class *class, char usage[LOCK_USAGE_CHARS])
> Â Â Â Âusage[i] = '\0';
> Â}
>
> +static int __print_lock_name(struct lock_class *class)
> +{
> + Â Â Â char str[KSYM_NAME_LEN];
> + Â Â Â const char *name;
> +
> + Â Â Â name = class->name;
> + Â Â Â if (!name)
> + Â Â Â Â Â Â Â name = __get_key_name(class->key, str);
> +
> + Â Â Â return printk("%s", name);
> +}
> +
> Âstatic void print_lock_name(struct lock_class *class)
> Â{
> Â Â Â Âchar str[KSYM_NAME_LEN], usage[LOCK_USAGE_CHARS];
> @@ -1325,6 +1337,62 @@ print_shortest_lock_dependencies(struct lock_list *leaf,
> Â Â Â Âreturn;
> Â}
>
> +static void
> +print_irq_lock_scenario(struct lock_list *safe_entry,
> + Â Â Â Â Â Â Â Â Â Â Â struct lock_list *unsafe_entry,
> + Â Â Â Â Â Â Â Â Â Â Â struct held_lock *prev,
> + Â Â Â Â Â Â Â Â Â Â Â struct held_lock *next)
> +{
> + Â Â Â struct lock_class *safe_class = safe_entry->class;
> + Â Â Â struct lock_class *unsafe_class = unsafe_entry->class;
> + Â Â Â struct lock_class *middle_class = hlock_class(prev);
> +
> + Â Â Â if (middle_class == safe_class)
> + Â Â Â Â Â Â Â middle_class = hlock_class(next);
> +
> + Â Â Â /*
> + Â Â Â Â* A direct locking problem where unsafe_class lock is taken
> + Â Â Â Â* directly by safe_class lock, then all we need to show
> + Â Â Â Â* is the deadlock scenario, as it is obvious that the
> + Â Â Â Â* unsafe lock is taken under the safe lock.
> + Â Â Â Â*
> + Â Â Â Â* But if there is a chain instead, where the safe lock takes
> + Â Â Â Â* an intermediate lock (middle_class) where this lock is
> + Â Â Â Â* not the same as the safe lock, then the lock chain is
> + Â Â Â Â* used to describe the problem. Otherwise we would need
> + Â Â Â Â* to show a different CPU case for each link in the chain
> + Â Â Â Â* from the safe_class lock to the unsafe_class lock.
> + Â Â Â Â*/
> + Â Â Â if (middle_class != unsafe_class) {
> + Â Â Â Â Â Â Â printk("Chain exists of:\n Â");
> + Â Â Â Â Â Â Â __print_lock_name(safe_class);
> + Â Â Â Â Â Â Â printk(" --> ");
> + Â Â Â Â Â Â Â __print_lock_name(middle_class);
> + Â Â Â Â Â Â Â printk(" --> ");
> + Â Â Â Â Â Â Â __print_lock_name(unsafe_class);
> + Â Â Â Â Â Â Â printk("\n\n");
> + Â Â Â }
> +
> + Â Â Â printk(" Possible interrupt unsafe locking scenario:\n\n");
> + Â Â Â printk(" Â Â Â CPU0 Â Â Â Â Â Â Â Â Â ÂCPU1\n");
> + Â Â Â printk(" Â Â Â ---- Â Â Â Â Â Â Â Â Â Â----\n");
> + Â Â Â printk(" Âlock(");
> + Â Â Â __print_lock_name(unsafe_class);
> + Â Â Â printk(");\n");
> + Â Â Â printk(" Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â local_irq_disable();\n");
> + Â Â Â printk(" Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â lock(");
> + Â Â Â __print_lock_name(safe_class);
> + Â Â Â printk(");\n");
> + Â Â Â printk(" Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â lock(");
> + Â Â Â __print_lock_name(middle_class);
> + Â Â Â printk(");\n");
> + Â Â Â printk(" Â<Interrupt>\n");
> + Â Â Â printk(" Â Âlock(");
> + Â Â Â __print_lock_name(safe_class);
> + Â Â Â printk(");\n");
> + Â Â Â printk("\n *** DEADLOCK ***\n\n");
> +}
> +
> Âstatic int
> Âprint_bad_irq_dependency(struct task_struct *curr,
> Â Â Â Â Â Â Â Â Â Â Â Â struct lock_list *prev_root,
> @@ -1376,6 +1444,8 @@ print_bad_irq_dependency(struct task_struct *curr,
> Â Â Â Âprint_stack_trace(forwards_entry->class->usage_traces + bit2, 1);
>
> Â Â Â Âprintk("\nother info that might help us debug this:\n\n");
> + Â Â Â print_irq_lock_scenario(backwards_entry, forwards_entry, prev, next);
> +
> Â Â Â Âlockdep_print_held_locks(curr);
>
> Â Â Â Âprintk("\nthe dependencies between %s-irq-safe lock", irqclass);
> --
> 1.7.2.3
>
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at Âhttp://vger.kernel.org/majordomo-info.html
> Please read the FAQ at Âhttp://www.tux.org/lkml/
>



-- 
Only stand for myself
--
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/