Messages in this thread Patch in this message | | | Subject | [patch] workqueue: schedule WORK_CPU_UNBOUND work on wq_unbound_cpumask CPUs | From | Mike Galbraith <> | Date | Wed, 22 Jul 2015 07:24:46 +0200 |
| |
On Tue, 2015-07-21 at 10:55 +0200, Mike Galbraith wrote: > On Sun, 2015-07-19 at 10:02 +0200, Mike Galbraith wrote: > > > Why do we do nothing about these allegedly unbound work items? > > My box seems to think the answer is: no reason other than nobody having > asked the source to please not do that. Guess I'll go ask a NUMA box.
My [128] socket boxen show zero signs of caring, and it's dirt simple, so it's no longer an experiment. Fly or die little patchlet...
WORK_CPU_UNBOUND work items queued to a bound workqueue always run locally. This is a good thing normally, but not when the user has asked us to keep unbound work away from certain CPUs. Round robin these to wq_unbound_cpumask CPUs instead, as perturbation avoidance trumps performance.
Signed-off-by: Mike Galbraith <umgwanakikbuti@gmail.com> --- kernel/workqueue.c | 27 +++++++++++++++++++++++++-- 1 file changed, 25 insertions(+), 2 deletions(-)
--- a/kernel/workqueue.c +++ b/kernel/workqueue.c @@ -301,6 +301,9 @@ static bool workqueue_freezing; /* PL: static cpumask_var_t wq_unbound_cpumask; /* PL: low level cpumask for all unbound wqs */ +/* CPU where WORK_CPU_UNBOUND work was last round robin scheduled from this CPU */ +static DEFINE_PER_CPU(unsigned int, wq_unbound_rr_cpu_last); + /* the per-cpu worker pools */ static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS], cpu_worker_pools); @@ -1294,6 +1297,24 @@ static bool is_chained_work(struct workq return worker && worker->current_pwq->wq == wq; } +/* + * When queueing WORK_CPU_UNBOUND work to a !WQ_UNBOUND queue, round + * robin among wq_unbound_cpumask to avoid perturbing sensitive tasks. + */ +static unsigned int select_round_robin_cpu(unsigned int cpu) +{ + if (cpumask_test_cpu(cpu, wq_unbound_cpumask)) + return cpu; + if (cpumask_empty(wq_unbound_cpumask)) + return cpu; + cpu = __this_cpu_read(wq_unbound_rr_cpu_last); + cpu = cpumask_next_and(cpu, wq_unbound_cpumask, cpu_online_mask); + if (cpu >= nr_cpu_ids) + cpu = 0; + __this_cpu_write(wq_unbound_rr_cpu_last, cpu); + return cpu; +} + static void __queue_work(int cpu, struct workqueue_struct *wq, struct work_struct *work) { @@ -1322,9 +1343,11 @@ static void __queue_work(int cpu, struct cpu = raw_smp_processor_id(); /* pwq which will be used unless @work is executing elsewhere */ - if (!(wq->flags & WQ_UNBOUND)) + if (!(wq->flags & WQ_UNBOUND)) { + if (req_cpu == WORK_CPU_UNBOUND) + cpu = select_round_robin_cpu(cpu); pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); - else + } else pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu)); /*
| |