Re: [RT git pull] fixes for rt/sched

[Gregory Haskins]

[ Forgot to CC John the first time around ]

Gregory Haskins wrote:
> The following changes since commit cfe6498c6a03e574bc9384cc95d73dfd20a13f05:
>   Steven Rostedt (1):
>         rt sched: add rt overload stats
>
> are available in the git repository at:
>
>   git://git.kernel.org/pub/scm/linux/kernel/git/ghaskins/linux-2.6-hacks.git rt/sched
>
> Gregory Haskins (1):
>       rt/sched: ensure interrupts are disabled during post-schedule() processing
>
>  kernel/sched_rt.c |    6 ++++--
>  1 files changed, 4 insertions(+), 2 deletions(-)
>
> ---------------
>
> rt/sched: ensure interrupts are disabled during post-schedule() processing
>
> Impact: Fix run-time opps at boot up.
>
> post-schedule() is required to run with interrupts disabled, but we
> currently rely on the caller to have left them disabled for us.  This
> has apparently changed recently such that interrupts are now on and
> we thus crash.  Fix this once and for all by using the irqsave/restore
> variants of spin_lock(rq). 
>
> Signed-off-by: Gregory Haskins <ghaskins@xxxxxxxxxx>
> CC: John Stultz <johnstul@xxxxxxxxxx>
> ---
>
>  kernel/sched_rt.c |    6 ++++--
>  1 files changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c
> index 856f6fb..0cd6b07 100644
> --- a/kernel/sched_rt.c
> +++ b/kernel/sched_rt.c
> @@ -1348,14 +1348,16 @@ static int needs_post_schedule_rt(struct rq *rq)
>  
>  static void post_schedule_rt(struct rq *rq)
>  {
> +	unsigned long flags;
> +
>  	/*
>  	 * This is only called if needs_post_schedule_rt() indicates that
>  	 * we need to push tasks away
>  	 */
> -	spin_lock(&rq->lock);
> +	spin_lock_irqsave(&rq->lock, flags);
>  	push_rt_tasks(rq);
>  	schedstat_inc(rq, rto_schedule_tail);
> -	spin_unlock(&rq->lock);
> +	spin_unlock_irqrestore(&rq->lock, flags);
>  }
>  
>  /*
>
>

Attachment: signature.asc
Description: OpenPGP digital signature