Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge tag 'sched_urgent_for_v6.3_rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull scheduler fix from Borislav Petkov:
- Fix a corner case where vruntime of a task is not being sanitized
* tag 'sched_urgent_for_v6.3_rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched/fair: Sanitize vruntime of entity being migrated
+46
-10
+3
kernel/sched/core.c
+3
kernel/sched/core.c
+43
-10
kernel/sched/fair.c
+43
-10
kernel/sched/fair.c
···
4648
4648
#endif
4649
4649
}
4650
4650
4651
+
static inline bool entity_is_long_sleeper(struct sched_entity *se)
4652
+
{
4653
+
struct cfs_rq *cfs_rq;
4654
+
u64 sleep_time;
4655
+
4656
+
if (se->exec_start == 0)
4657
+
return false;
4658
+
4659
+
cfs_rq = cfs_rq_of(se);
4660
+
4661
+
sleep_time = rq_clock_task(rq_of(cfs_rq));
4662
+
4663
+
/* Happen while migrating because of clock task divergence */
4664
+
if (sleep_time <= se->exec_start)
4665
+
return false;
4666
+
4667
+
sleep_time -= se->exec_start;
4668
+
if (sleep_time > ((1ULL << 63) / scale_load_down(NICE_0_LOAD)))
4669
+
return true;
4670
+
4671
+
return false;
4672
+
}
4673
+
4651
4674
static void
4652
4675
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
4653
4676
{
4654
4677
u64 vruntime = cfs_rq->min_vruntime;
4655
-
u64 sleep_time;
4656
4678
4657
4679
/*
4658
4680
* The 'current' period is already promised to the current tasks,
···
4706
4684
4707
4685
/*
4708
4686
* Pull vruntime of the entity being placed to the base level of
4709
-
* cfs_rq, to prevent boosting it if placed backwards. If the entity
4710
-
* slept for a long time, don't even try to compare its vruntime with
4711
-
* the base as it may be too far off and the comparison may get
4712
-
* inversed due to s64 overflow.
4687
+
* cfs_rq, to prevent boosting it if placed backwards.
4688
+
* However, min_vruntime can advance much faster than real time, with
4689
+
* the extreme being when an entity with the minimal weight always runs
4690
+
* on the cfs_rq. If the waking entity slept for a long time, its
4691
+
* vruntime difference from min_vruntime may overflow s64 and their
4692
+
* comparison may get inversed, so ignore the entity's original
4693
+
* vruntime in that case.
4694
+
* The maximal vruntime speedup is given by the ratio of normal to
4695
+
* minimal weight: scale_load_down(NICE_0_LOAD) / MIN_SHARES.
4696
+
* When placing a migrated waking entity, its exec_start has been set
4697
+
* from a different rq. In order to take into account a possible
4698
+
* divergence between new and prev rq's clocks task because of irq and
4699
+
* stolen time, we take an additional margin.
4700
+
* So, cutting off on the sleep time of
4701
+
* 2^63 / scale_load_down(NICE_0_LOAD) ~ 104 days
4702
+
* should be safe.
4713
4703
*/
4714
-
sleep_time = rq_clock_task(rq_of(cfs_rq)) - se->exec_start;
4715
-
if ((s64)sleep_time > 60LL * NSEC_PER_SEC)
4704
+
if (entity_is_long_sleeper(se))
4716
4705
se->vruntime = vruntime;
4717
4706
else
4718
4707
se->vruntime = max_vruntime(se->vruntime, vruntime);
···
4803
4770
4804
4771
if (flags & ENQUEUE_WAKEUP)
4805
4772
place_entity(cfs_rq, se, 0);
4773
+
/* Entity has migrated, no longer consider this task hot */
4774
+
if (flags & ENQUEUE_MIGRATED)
4775
+
se->exec_start = 0;
4806
4776
4807
4777
check_schedstat_required();
4808
4778
update_stats_enqueue_fair(cfs_rq, se, flags);
···
7692
7656
7693
7657
/* Tell new CPU we are migrated */
7694
7658
se->avg.last_update_time = 0;
7695
-
7696
-
/* We have migrated, no longer consider this task hot */
7697
-
se->exec_start = 0;
7698
7659
7699
7660
update_scan_period(p, new_cpu);
7700
7661
}