Re: [PATCH 1/2] sched: eliminate bandwidth race between throttling and distribution

[Phil Auld <pauld@redhat.com>]

Hi Josh,

On Sat, Apr 11, 2020 at 07:00:17PM -0700 Josh Don wrote:
> +[1]pauld@redhat.com
> 
> On Fri, Apr 10, 2020 at 3:52 PM Josh Don <[2]joshdon@google.com> wrote:
> 
>     From: Paul Turner <[3]pjt@google.com>
> 
>     There is a race window in which an entity begins throttling before quota
>     is added to the pool, but does not finish throttling until after we have
>     finished with distribute_cfs_runtime(). This entity is not observed by
>     distribute_cfs_runtime() because it was not on the throttled list at the
>     time that distribution was running. This race manifests as rare
>     period-length statlls for such entities.
> 
>     Rather than heavy-weight the synchronization with the progress of
>     distribution, we can fix this by aborting throttling if bandwidth has
>     become available. Otherwise, we immediately add the entity to the
>     throttled list so that it can be observed by a subsequent distribution.
> 
>     Additionally, we can remove the case of adding the throttled entity to
>     the head of the throttled list, and simply always add to the tail.
>     Thanks to 26a8b12747c97, distribute_cfs_runtime() no longer holds onto
>     its own pool of runtime. This means that if we do hit the !assign and
>     distribute_running case, we know that distribution is about to end.
> 
>     Signed-off-by: Paul Turner <[4]pjt@google.com>
>     Co-developed-by: Ben Segall <[5]bsegall@google.com>
>     Signed-off-by: Ben Segall <[6]bsegall@google.com>
>     Signed-off-by: Josh Don <[7]joshdon@google.com>

This looks good to me, thanks for the cc.

Reviewed-by: Phil Auld <pauld@redhat.com>

>     ---
>      kernel/sched/fair.c | 78 ++++++++++++++++++++++++++-------------------
>      1 file changed, 46 insertions(+), 32 deletions(-)
> 
>     diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>     index 02f323b85b6d..3fb986c52825 100644
>     --- a/kernel/sched/fair.c
>     +++ b/kernel/sched/fair.c
>     @@ -4587,17 +4587,15 @@ static inline struct cfs_bandwidth
>     *tg_cfs_bandwidth(struct task_group *tg)
>             return &tg->cfs_bandwidth;
>      }
> 
>     -/* returns 0 on failure to allocate runtime */
>     -static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
>     +/* returns 0 on failure to allocate runtime, called with cfs_b->lock held
>     */
>     +static int __assign_cfs_rq_runtime(struct cfs_bandwidth *cfs_b,
>     +                                  struct cfs_rq *cfs_rq, u64
>     target_runtime)
>      {
>     -       struct task_group *tg = cfs_rq->tg;
>     -       struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
>             u64 amount = 0, min_amount;
> 
>             /* note: this is a positive sum as runtime_remaining <= 0 */
>     -       min_amount = sched_cfs_bandwidth_slice() - cfs_rq->
>     runtime_remaining;
>     +       min_amount = target_runtime - cfs_rq->runtime_remaining;
> 
>     -       raw_spin_lock(&cfs_b->lock);
>             if (cfs_b->quota == RUNTIME_INF)
>                     amount = min_amount;
>             else {
>     @@ -4609,13 +4607,26 @@ static int assign_cfs_rq_runtime(struct cfs_rq
>     *cfs_rq)
>                             cfs_b->idle = 0;
>                     }
>             }
>     -       raw_spin_unlock(&cfs_b->lock);
> 
>             cfs_rq->runtime_remaining += amount;
> 
>             return cfs_rq->runtime_remaining > 0;
>      }
> 
>     +/* returns 0 on failure to allocate runtime */
>     +static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
>     +{
>     +       int ret;
>     +       struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
>     +
>     +       raw_spin_lock(&cfs_b->lock);
>     +       ret = __assign_cfs_rq_runtime(cfs_b, cfs_rq,
>     +                                     sched_cfs_bandwidth_slice());
>     +       raw_spin_unlock(&cfs_b->lock);
>     +
>     +       return ret;
>     +}
>     +
>      static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64
>     delta_exec)
>      {
>             /* dock delta_exec before expiring quota (as it could span periods)
>     */
>     @@ -4704,13 +4715,33 @@ static int tg_throttle_down(struct task_group *tg,
>     void *data)
>             return 0;
>      }
> 
>     -static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
>     +static bool throttle_cfs_rq(struct cfs_rq *cfs_rq)
>      {
>             struct rq *rq = rq_of(cfs_rq);
>             struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
>             struct sched_entity *se;
>             long task_delta, idle_task_delta, dequeue = 1;
>     -       bool empty;
>     +
>     +       raw_spin_lock(&cfs_b->lock);
>     +       /* This will start the period timer if necessary */
>     +       if (__assign_cfs_rq_runtime(cfs_b, cfs_rq, 1)) {
>     +               /*
>     +                * We have raced with bandwidth becoming available, and if
>     we
>     +                * actually throttled the timer might not unthrottle us for
>     an
>     +                * entire period. We additionally needed to make sure that
>     any
>     +                * subsequent check_cfs_rq_runtime calls agree not to
>     throttle
>     +                * us, as we may commit to do cfs put_prev+pick_next, so we
>     ask
>     +                * for 1ns of runtime rather than just check cfs_b.
>     +                */
>     +               dequeue = 0;
>     +       } else {
>     +               list_add_tail_rcu(&cfs_rq->throttled_list,
>     +                                 &cfs_b->throttled_cfs_rq);
>     +       }
>     +       raw_spin_unlock(&cfs_b->lock);
>     +
>     +       if (!dequeue)
>     +               return false;  /* Throttle no longer required. */
> 
>             se = cfs_rq->tg->se[cpu_of(rq_of(cfs_rq))];
> 
>     @@ -4744,29 +4775,13 @@ static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
>             if (!se)
>                     sub_nr_running(rq, task_delta);
> 
>     -       cfs_rq->throttled = 1;
>     -       cfs_rq->throttled_clock = rq_clock(rq);
>     -       raw_spin_lock(&cfs_b->lock);
>     -       empty = list_empty(&cfs_b->throttled_cfs_rq);
>     -
>             /*
>     -        * Add to the _head_ of the list, so that an already-started
>     -        * distribute_cfs_runtime will not see us. If disribute_cfs_runtime
>     is
>     -        * not running add to the tail so that later runqueues don't get
>     starved.
>     +        * Note: distribution will already see us throttled via the
>     +        * throttled-list.  rq->lock protects completion.
>              */
>     -       if (cfs_b->distribute_running)
>     -               list_add_rcu(&cfs_rq->throttled_list, &cfs_b->
>     throttled_cfs_rq);
>     -       else
>     -               list_add_tail_rcu(&cfs_rq->throttled_list, &cfs_b->
>     throttled_cfs_rq);
>     -
>     -       /*
>     -        * If we're the first throttled task, make sure the bandwidth
>     -        * timer is running.
>     -        */
>     -       if (empty)
>     -               start_cfs_bandwidth(cfs_b);
>     -
>     -       raw_spin_unlock(&cfs_b->lock);
>     +       cfs_rq->throttled = 1;
>     +       cfs_rq->throttled_clock = rq_clock(rq);
>     +       return true;
>      }
> 
>      void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
>     @@ -5121,8 +5136,7 @@ static bool check_cfs_rq_runtime(struct cfs_rq
>     *cfs_rq)
>             if (cfs_rq_throttled(cfs_rq))
>                     return true;
> 
>     -       throttle_cfs_rq(cfs_rq);
>     -       return true;
>     +       return throttle_cfs_rq(cfs_rq);
>      }
> 
>      static enum hrtimer_restart sched_cfs_slack_timer(struct hrtimer *timer)
>     --
>     2.26.0.110.g2183baf09c-goog
> 
> 
> 
> References:
> 
> [1] mailto:pauld@redhat.com
> [2] mailto:joshdon@google.com
> [3] mailto:pjt@google.com
> [4] mailto:pjt@google.com
> [5] mailto:bsegall@google.com
> [6] mailto:bsegall@google.com
> [7] mailto:joshdon@google.com

--