Our custom nr_unint logic follows the kernel's one about the global nr_uninterruptible one. One invariant that both share is that values of nr_unint can go negative when a task is migrated from a CPU. What's important is that when summed across all task groups they either result in a positive number (we have sleeping tasks in uninterruptible state) or 0 (no uninterruptible processes).
Current code is broken in that the math in calc_load_ve always deals with unsigned integers values which leads to enormous values of 'nr_active' when ->nr_unint is negative on a tg. The proper way to work with nr_unint is to always cast it to an signed integer and also make nr_active a signed values as well to properly deal with negative values. Signed-off-by: Nikolay Borisov <nikolay.bori...@virtuozzo.com> https://jira.sw.ru/browse/PSBM-143355 --- kernel/sched/loadavg.c | 14 +++++++++++--- 1 file changed, 11 insertions(+), 3 deletions(-) diff --git a/kernel/sched/loadavg.c b/kernel/sched/loadavg.c index 8a5fa39b8b7b..c0c6b89a4958 100644 --- a/kernel/sched/loadavg.c +++ b/kernel/sched/loadavg.c @@ -114,7 +114,7 @@ extern raw_spinlock_t load_ve_lock; void calc_load_ve(void) { - unsigned long nr_unint, nr_active; + long nr_unint, nr_active; struct task_group *tg; int i; @@ -134,13 +134,21 @@ void calc_load_ve(void) * overhead for activate/deactivate operations. So, we * don't account child cgroup unint tasks here. */ - nr_active += tg->cfs_rq[i]->nr_unint; + nr_active += (int)tg->cfs_rq[i]->nr_unint; #endif #ifdef CONFIG_RT_GROUP_SCHED nr_active += tg->rt_rq[i]->rt_nr_running; #endif } - nr_active *= FIXED_1; + /* + * Unless we screwed somewhere, nr_active must always be >= 0, because + * we can have no runnable tasks and no taks in uninterruptible sleep + */ + if (WARN_ON_ONCE(nr_active < 0)) { + nr_active = 0; + } else { + nr_active *= FIXED_1; + } tg->avenrun[0] = calc_load(tg->avenrun[0], EXP_1, nr_active); tg->avenrun[1] = calc_load(tg->avenrun[1], EXP_5, nr_active); -- 2.31.1 _______________________________________________ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel
