For isolated applications that busy loop (packet processing with DPDK, for example), workqueue functions either stall (if the -rt app priority is higher than kworker thread priority) or interrupt the -rt app (if the -rt app priority is lower than kworker thread priority. For those cases, it is ideal to avoid the workqueue function in the first place. For the vmstat_update function, which syncs per-cpu statistics to global stats, one can modify the -rt app to explicitly flush before the realtime loop starts. Suggested by Christoph Lameter at https://lkml.org/lkml/2017/5/12/364. Suggested-by: Christoph Lameter <cl@xxxxxxxxx> Signed-off-by: Marcelo Tosatti <mtosatti@xxxxxxxxxx> diff --git a/Documentation/admin-guide/sysctl/vm.rst b/Documentation/admin-guide/sysctl/vm.rst index f455fa00c00f..ea4ad1f9f82f 100644 --- a/Documentation/admin-guide/sysctl/vm.rst +++ b/Documentation/admin-guide/sysctl/vm.rst @@ -67,6 +67,7 @@ Currently, these files are in /proc/sys/vm: - percpu_pagelist_fraction - stat_interval - stat_refresh +- quiet_vmstat - numa_stat - swappiness - unprivileged_userfaultfd @@ -827,6 +828,21 @@ as 0) and "fails" with EINVAL if any are found, with a warning in dmesg. (At time of writing, a few stats are known sometimes to be found negative, with no ill effects: errors and warnings on these stats are suppressed.) +quiet_vmstat +============ + +Any read or write flushes the current cpu's per-cpu vm statistics into +their global totals, and flushes pending updates. + +This is useful for isolated realtime applications that, in the setup +phase, use services that modify per-cpu vm statistics (mlock, for example). + +Performing an explicit flush before starting the time sensitive loop removes +the need for it to happen at workqueue time (which would interrupt the +application). + +Note that without pinning the task to a particular CPU, which CPU has its +per-cpu vm statistics flushed is not well defined. numa_stat ========= diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h index 322dcbfcc933..da98cc28b3ca 100644 --- a/include/linux/vmstat.h +++ b/include/linux/vmstat.h @@ -290,6 +290,8 @@ void refresh_zone_stat_thresholds(void); struct ctl_table; int vmstat_refresh(struct ctl_table *, int write, void *buffer, size_t *lenp, loff_t *ppos); +int proc_quiet_vmstat(struct ctl_table *table, int write, void *buffer, + size_t *lenp, loff_t *ppos); void drain_zonestat(struct zone *zone, struct per_cpu_pageset *); diff --git a/kernel/sysctl.c b/kernel/sysctl.c index afad085960b8..71b28311b950 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -3016,6 +3016,13 @@ static struct ctl_table vm_table[] = { .mode = 0600, .proc_handler = vmstat_refresh, }, + { + .procname = "quiet_vmstat", + .data = NULL, + .maxlen = 0, + .mode = 0600, + .proc_handler = proc_quiet_vmstat, + }, #endif #ifdef CONFIG_MMU { diff --git a/mm/vmstat.c b/mm/vmstat.c index 698bc0bc18d1..cec8e5c32f97 100644 --- a/mm/vmstat.c +++ b/mm/vmstat.c @@ -1936,6 +1936,24 @@ void quiet_vmstat(void) refresh_cpu_vm_stats(false); } +int proc_quiet_vmstat(struct ctl_table *table, int write, + void *buffer, size_t *lenp, loff_t *ppos) +{ + if (write) + *ppos += *lenp; + else + *lenp = 0; + + preempt_disable(); + if (need_update(smp_processor_id()) == true) + refresh_cpu_vm_stats(false); + preempt_enable(); + + flush_delayed_work(this_cpu_ptr(&vmstat_work)); + + return 0; +} + /* * Shepherd worker thread that checks the * differentials of processors that have their worker
