Re: [PATCH 1/7] rcu: Fix preempt-unsafe debug check of rcu extendedquiescent state

From: Pavel Ivanov
Date: Mon Sep 26 2011 - 18:04:41 EST


On Mon, Sep 26, 2011 at 6:19 AM, Frederic Weisbecker <fweisbec@xxxxxxxxx> wrote:
> In the rcu_check_extended_qs() function that is used to check
> illegal uses of RCU under extended quiescent states, we look
> at the local value of dynticks that is even if an extended
> quiescent state or odd otherwise.
>
> We are looking at it without disabling the preemption though
> and this opens a race window where we may read the state of
> a remote CPU instead, like in the following scenario:
>
> CPU 1                                                        CPU 2
>
> bool rcu_check_extended_qs(void)
> {
>        struct rcu_dynticks *rdtp;
>
>        rdtp = &per_cpu(rcu_dynticks,
>                        raw_smp_processor_id());
>
>        < ---- Task is migrated here ---- >
>
>        // CPU 1 goes idle and increase rdtp->dynticks
>                                                        /* Here we are reading the value for
>                                                        the remote CPU 1 instead of the local CPU */
>                                                        if (atomic_read(&rdtp->dynticks) & 0x1)
>                                                                return false;
>
>                                                        return true;
>                                                        }
>
> The possible result of this is false positive return value of that
> function, suggesting we are in an extended quiescent state in random
> places.

How is this different from what your patch allows?

CPU 1                                               CPU 2

bool rcu_check_extended_qs(void)
{
       struct rcu_dynticks *rdtp =
&get_cpu_var(rcu_dynticks);
bool ext_qs = true;

       if (atomic_read(&rdtp->dynticks) & 0x1)
ext_qs = false;

put_cpu_var(rcu_dynticks);

       < ---- Task is migrated here ---- >

                                                 /* Here we return true/false
based on the value for the
remote CPU 1 instead of the
local CPU */

                                                 return ext_qs;
                                                 }


Looks like it can result in the same false positive result.

Pavel


> Fix this by disabling preemption while reading that value.
>
> Signed-off-by: Frederic Weisbecker <fweisbec@xxxxxxxxx>
> Cc: Peter Zijlstra <a.p.zijlstra@xxxxxxxxx>
> Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> Cc: Lai Jiangshan <laijs@xxxxxxxxxxxxxx>
> Cc: Ingo Molnar <mingo@xxxxxxxxxx>
> ---
>  kernel/rcutree.c |   10 ++++++----
>  1 files changed, 6 insertions(+), 4 deletions(-)
>
> diff --git a/kernel/rcutree.c b/kernel/rcutree.c
> index c9b4adf..234dca3 100644
> --- a/kernel/rcutree.c
> +++ b/kernel/rcutree.c
> @@ -469,13 +469,15 @@ void rcu_irq_exit(void)
>
>  bool rcu_check_extended_qs(void)
>  {
> -       struct rcu_dynticks *rdtp;
> +       struct rcu_dynticks *rdtp = &get_cpu_var(rcu_dynticks);
> +       bool ext_qs = true;
>
> -       rdtp = &per_cpu(rcu_dynticks, raw_smp_processor_id());
>        if (atomic_read(&rdtp->dynticks) & 0x1)
> -               return false;
> +               ext_qs = false;
> +
> +       put_cpu_var(rcu_dynticks);
>
> -       return true;
> +       return ext_qs;
>  }
>  EXPORT_SYMBOL_GPL(rcu_check_extended_qs);
>
> --
> 1.7.5.4
--
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/