tick/sched: Ensure quiet_vmstat() is called when the idle tick was stopped too

This patchset contains enhancements on top of Aaron's -v6 of the series (see the changelog below).

It fixes the following problems two problems:

1) A customer provided some evidence which indicates that the idle tick was stopped; albeit, CPU-specific vmstat counters still remained populated. Thus one can only assume quiet_vmstat() was not invoked on return to the idle loop.

If I understand correctly, I suspect this divergence might erroneously prevent a reclaim attempt by kswapd. If the number of zone specific free pages are below their per-cpu drift value then zone_page_state_snapshot() is used to compute a more accurate view of the aforementioned statistic. Thus any task blocked on the NUMA node specific pfmemalloc_wait queue will be unable to make significant progress via direct reclaim unless it is killed after being woken up by kswapd (see throttle_direct_reclaim()).

2) With a SCHED_FIFO task that busy loops on a given CPU, and kworker for that CPU at SCHED_OTHER priority, queuing work to sync per-vmstats will either cause that work to never execute, or stalld boosts kworker priority which causes a latency violation.

Follows the v6 cover letter, with updated changelog. The numbers, for the test program attached at the end of this cover letter, executed inside a KVM VM, are:

Vanilla Patch

cycles per idle loop 151858 153258 (+1.0%)

cycles per syscall 8461 8690 (+2.6%)

--------

I have incorporated an idea from Marcelo's patch [1] where a CPU-specific variable is used to indicate if a vmstat differential/or imbalance is present for a given CPU. So, at the appropriate time, vmstat processing can be initiated. The hope is that this particular approach is "cheaper" when compared to need_update() - used currently; in the context of nohz_full and the scheduling-clock tick being stopped, we would now with this patch, check if a CPU-specific vmstat imbalance is present before exiting user-mode (see tick_nohz_user_enter_prepare()).

This trivial test program [2] was used to determine the somewhat impact under vanilla and with the proposed changes; mlock(2) and munlock(2) was used solely to modify vmstat item 'NR_MLOCK'. The following is an average count of CPU-cycles across the aforementioned system calls and the idle loop, respectively. I believe these results are negligible:

	  Modified		   |  		Vanilla
                                   |
  cycles per syscall: 7399         | 	cycles per syscall: 4150
  cycles per idle loop: 141048     |	cycles per idle loop: 144730
                                   |

Any feedback would be appreciated. Thanks.

Changes since v6 [6]:
- sync vmstats independently of whether vmstat_update work is queued or not
- clean vmstat_dirty before differential sync loop
- cancel pending work if tick stopped
- do not queue work to remote CPU if tick stopped

Changes since v5 [3]:

- Introduced __tick_nohz_user_enter_prepare()
- Switched to EXPORT_SYMBOL_GPL()

Changes since v4 [4]:

- Moved vmstat_dirty specific changes into a separate patch (Marcelo Tosatti)

Changes since v3 [5]:

- Used EXPORT_SYMBOL() on tick_nohz_user_enter_prepare()
- Replaced need_update()
- Introduced CPU-specific variable namely vmstat_dirty and mark_vmstat_dirty()

[1]: https://lore.kernel.org/lkml/[email protected]/
[2]: https://pastebin.com/8AtzSAuK
[3]: https://lore.kernel.org/lkml/[email protected]/
[4]: https://lore.kernel.org/lkml/[email protected]/
[5]: https://lore.kernel.org/lkml/[email protected]/
[6]: https://lore.kernel.org/linux-mm/[email protected]/

 include/linux/tick.h     |    5 +++--
 kernel/time/tick-sched.c |   19 ++++++++++++++++++-
 mm/vmstat.c              |   74 ++++++++++++++++++++++++++++++++++++--------------------------------------
 3 files changed, 57 insertions(+), 41 deletions(-)

--- test-vmstat-overhead.c ---

#include <stdio.h>
#include <stdlib.h>
#include <sys/mman.h>
#include <unistd.h>
#include <string.h>

typedef unsigned long long cycles_t;
typedef unsigned long long usecs_t;
typedef unsigned long long u64;

#ifdef __x86_64__

#define DECLARE_ARGS(val, low, high)    unsigned long low, high
#define EAX_EDX_VAL(val, low, high)     ((low) | ((u64)(high) << 32))
#define EAX_EDX_ARGS(val, low, high)    "a" (low), "d" (high)
#define EAX_EDX_RET(val, low, high)     "=a" (low), "=d" (high)
#else
#define DECLARE_ARGS(val, low, high)    unsigned long long val
#define EAX_EDX_VAL(val, low, high)     (val)
#define EAX_EDX_ARGS(val, low, high)    "A" (val)
#define EAX_EDX_RET(val, low, high)     "=A" (val)

#endif

static inline unsigned long long __rdtscll(void)

{
        DECLARE_ARGS(val, low, high);

        asm volatile("cpuid; rdtsc" : EAX_EDX_RET(val, low, high));

        return EAX_EDX_VAL(val, low, high);
}

#define rdtscll(val) do { (val) = __rdtscll(); } while (0)

#define NRSYSCALLS 30000
#define NRSLEEPS 100000

void main(int argc, char *argv[])

{
	unsigned long a, b, cycles;
	int i, syscall = 0;
	void *page = malloc(4096);

	if (mlock(page, 4096))
		perror("mlock");
	if (munlock(page, 4096))
		perror("munlock");

	if (argc != 2) {
		printf("usage: %s {idle,syscall}\n", argv[0]);
		exit(1);
	}

        rdtscll(a);

	if (strncmp("idle", argv[1], 4) == 0)
		syscall = 0;
	else if (strncmp("syscall", argv[1], 7) == 0)
		syscall = 1;
	else {
		printf("usage: %s {idle,syscall}\n", argv[0]);
		exit(1);
	}
	
	if (syscall == 1) {
        	for (i = 0; i < NRSYSCALLS; i++) {
			if (mlock(page, 4096))
				perror("mlock");
			if (munlock(page, 4096))
				perror("munlock");
		}
	} else {
        	for (i = 0; i < NRSLEEPS; i++)
		 	usleep(10);
	}

        rdtscll(b);

        cycles = b - a;

	if (syscall == 1)
        	printf("cycles per syscall: %d\n", (b-a)/(NRSYSCALLS*2));
	else
		printf("cycles per idle loop: %d\n", (b-a)/NRSLEEPS);
}

From: Aaron Tomlin <[email protected]>

Add CPU-specific variable namely vmstat_dirty to indicate if a vmstat imbalance is present for a given CPU. Therefore, at the appropriate time, we can fold all the remaining differentials.

This speeds up quiet_vmstat in case no per-CPU differentials exist.

Based on
https://lore.kernel.org/lkml/[email protected]/

---
 mm/vmstat.c |   54 ++++++++++++++++++++----------------------------------
 1 file changed, 20 insertions(+), 34 deletions(-)

Index: linux-2.6/mm/vmstat.c
===================================================================
@@ -195,6 +195,12 @@ void fold_vm_numa_events(void)
 #endif
 
 #ifdef CONFIG_SMP
+static DEFINE_PER_CPU_ALIGNED(bool, vmstat_dirty);
+
+static inline void mark_vmstat_dirty(void)
+{
+	this_cpu_write(vmstat_dirty, true);
+}
 
 int calculate_pressure_threshold(struct zone *zone)
 {

@@ -367,6 +373,7 @@ void __mod_zone_page_state(struct zone *
 		x = 0;
 	}
 	__this_cpu_write(*p, x);
+	mark_vmstat_dirty();
 
 	if (IS_ENABLED(CONFIG_PREEMPT_RT))
 		preempt_enable();

@@ -405,6 +412,7 @@ void __mod_node_page_state(struct pglist
 		x = 0;
 	}
 	__this_cpu_write(*p, x);
+	mark_vmstat_dirty();
 
 	if (IS_ENABLED(CONFIG_PREEMPT_RT))
 		preempt_enable();

@@ -603,6 +611,7 @@ static inline void mod_zone_state(struct
 
 	if (z)
 		zone_page_state_add(z, zone, item);
+	mark_vmstat_dirty();
 }
 
 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,

@@ -671,6 +680,7 @@ static inline void mod_node_state(struct
 
 	if (z)
 		node_page_state_add(z, pgdat, item);
+	mark_vmstat_dirty();
 }
 
 void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,

@@ -825,6 +835,14 @@ static int refresh_cpu_vm_stats(bool do_
 	int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, };
 	int changes = 0;
 
+	/*
+	 * Clear vmstat_dirty before clearing the percpu vmstats.
+	 * If interrupts are enabled, it is possible that an interrupt
+	 * or another task modifies a percpu vmstat, which will
+	 * set vmstat_dirty to true.
+	 */
+	this_cpu_write(vmstat_dirty, false);
+
 	for_each_populated_zone(zone) {
 		struct per_cpu_zonestat __percpu *pzstats = zone->per_cpu_zonestats;
 #ifdef CONFIG_NUMA

@@ -1949,35 +1967,6 @@ static void vmstat_update(struct work_st
 }
 
 /*
- * Check if the diffs for a certain cpu indicate that
- * an update is needed.
- */
-static bool need_update(int cpu)
-{
-	pg_data_t *last_pgdat = NULL;
-	struct zone *zone;
-
-	for_each_populated_zone(zone) {
-		struct per_cpu_zonestat *pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
-		struct per_cpu_nodestat *n;
-
-		/*
-		 * The fast way of checking if there are any vmstat diffs.
-		 */
-		if (memchr_inv(pzstats->vm_stat_diff, 0, sizeof(pzstats->vm_stat_diff)))
-			return true;
-
-		if (last_pgdat == zone->zone_pgdat)
-			continue;
-		last_pgdat = zone->zone_pgdat;
-		n = per_cpu_ptr(zone->zone_pgdat->per_cpu_nodestats, cpu);
-		if (memchr_inv(n->vm_node_stat_diff, 0, sizeof(n->vm_node_stat_diff)))
-			return true;
-	}
-	return false;
-}
-
-/*
  * Switch off vmstat processing and then fold all the remaining differentials
  * until the diffs stay at zero. The function is used by NOHZ and can only be
  * invoked when tick processing is not active.

@@ -1987,10 +1976,7 @@ void quiet_vmstat(void)
 	if (system_state != SYSTEM_RUNNING)
 		return;
 
-	if (!delayed_work_pending(this_cpu_ptr(&vmstat_work)))
-		return;
-
-	if (!need_update(smp_processor_id()))
+	if (!__this_cpu_read(vmstat_dirty))
 		return;
 
 	/*

@@ -2021,7 +2007,7 @@ static void vmstat_shepherd(struct work_
 	for_each_online_cpu(cpu) {
 		struct delayed_work *dw = &per_cpu(vmstat_work, cpu);
 
-		if (!delayed_work_pending(dw) && need_update(cpu))
+		if (!delayed_work_pending(dw) && per_cpu(vmstat_dirty, cpu))
 			queue_delayed_work_on(cpu, mm_percpu_wq, dw, 0);
 
 		cond_resched();

From: Aaron Tomlin <[email protected]>

In the context of the idle task and an adaptive-tick mode/or a nohz_full CPU, quiet_vmstat() can be called: before stopping the idle tick, entering an idle state and on exit. In particular, for the latter case, when the idle task is required to reschedule, the idle tick can remain stopped and the timer expiration time endless i.e., KTIME_MAX. Now, indeed before a nohz_full CPU enters an idle state, CPU-specific vmstat counters should be processed to ensure the respective values have been reset and folded into the zone specific 'vm_stat[]'. That being said, it can only occur when: the idle tick was previously stopped, and reprogramming of the timer is not required.

A customer provided some evidence which indicates that the idle tick was stopped; albeit, CPU-specific vmstat counters still remained populated. Thus one can only assume quiet_vmstat() was not invoked on return to the idle loop.

If I understand correctly, I suspect this divergence might erroneously prevent a reclaim attempt by kswapd. If the number of zone specific free pages are below their per-cpu drift value then zone_page_state_snapshot() is used to compute a more accurate view of the aforementioned statistic. Thus any task blocked on the NUMA node specific pfmemalloc_wait queue will be unable to make significant progress via direct reclaim unless it is killed after being woken up by kswapd (see throttle_direct_reclaim()).

Consider the following theoretical scenario:

1. CPU Y migrated running task A to CPU X that was
in an idle state i.e. waiting for an IRQ - not polling; marked the current task on CPU X to need/or require a reschedule i.e., set TIF_NEED_RESCHED and invoked a reschedule IPI to CPU X (see sched_move_task())

2. CPU X acknowledged the reschedule IPI from CPU Y;
generic idle loop code noticed the TIF_NEED_RESCHED flag against the idle task and attempts to exit of the loop and calls the main scheduler function i.e. __schedule().

Since the idle tick was previously stopped no scheduling-clock tick would occur. So, no deferred timers would be handled

3. Post transition to kernel execution Task A
running on CPU Y, indirectly released a few pages (e.g. see __free_one_page()); CPU Y's 'vm_stat_diff[NR_FREE_PAGES]' was updated and zone specific 'vm_stat[]' update was deferred as per the CPU-specific stat threshold

4. Task A does invoke exit(2) and the kernel does
remove the task from the run-queue; the idle task was selected to execute next since there are no other runnable tasks assigned to the given CPU (see pick_next_task() and pick_next_task_idle())

5. On return to the idle loop since the idle tick
was already stopped and can remain so (see [1] below) e.g. no pending soft IRQs, no attempt is made to zero and fold CPU Y's vmstat counters since reprogramming of the scheduling-clock tick is not required/or needed (see [2])

...
do_idle
{

__current_set_polling()
tick_nohz_idle_enter()

while (!need_resched()) {

local_irq_disable()

...

/* No polling or broadcast event */ cpuidle_idle_call()
{

if (cpuidle_not_available(drv, dev)) {
tick_nohz_idle_stop_tick()
__tick_nohz_idle_stop_tick(this_cpu_ptr(&tick_cpu_sched)) {
int cpu = smp_processor_id()

if (ts->timer_expires_base)
expires = ts->timer_expires
else if (can_stop_idle_tick(cpu, ts))
(1) -------> expires = tick_nohz_next_event(ts, cpu)
else
return

ts->idle_calls++

if (expires > 0LL) {

tick_nohz_stop_tick(ts, cpu) {

if (ts->tick_stopped && (expires == ts->next_tick)) {
(2) -------> if (tick == KTIME_MAX || ts->next_tick ==
hrtimer_get_expires(&ts->sched_timer)) return
}
...
}

So the idea of with this patch is to ensure refresh_cpu_vm_stats(false) is called, when it is appropriate, on return to the idle loop when the idle tick was previously stopped too. Additionally, in the context of nohz_full, when the scheduling-tick is stopped and before exiting to user-mode, ensure no CPU-specific vmstat differentials remain.

---
 include/linux/tick.h     |    5 +++--
 kernel/time/tick-sched.c |   19 ++++++++++++++++++-
 2 files changed, 21 insertions(+), 3 deletions(-)

Index: linux-2.6/include/linux/tick.h
===================================================================
@@ -11,7 +11,6 @@
 #include <linux/context_tracking_state.h>
 #include <linux/cpumask.h>
 #include <linux/sched.h>
-#include <linux/rcupdate.h>
 
 #ifdef CONFIG_GENERIC_CLOCKEVENTS
 extern void __init tick_init(void);

@@ -272,6 +271,7 @@ static inline void tick_dep_clear_signal
 
 extern void tick_nohz_full_kick_cpu(int cpu);
 extern void __tick_nohz_task_switch(void);
+void __tick_nohz_user_enter_prepare(void);
 extern void __init tick_nohz_full_setup(cpumask_var_t cpumask);
 #else
 static inline bool tick_nohz_full_enabled(void) { return false; }

@@ -296,6 +296,7 @@ static inline void tick_dep_clear_signal
 
 static inline void tick_nohz_full_kick_cpu(int cpu) { }
 static inline void __tick_nohz_task_switch(void) { }
+static inline void __tick_nohz_user_enter_prepare(void) { }
 static inline void tick_nohz_full_setup(cpumask_var_t cpumask) { }
 #endif
 

@@ -308,7 +309,7 @@ static inline void tick_nohz_task_switch
 static inline void tick_nohz_user_enter_prepare(void)
 {
 	if (tick_nohz_full_cpu(smp_processor_id()))
-		rcu_nocb_flush_deferred_wakeup();
+		__tick_nohz_user_enter_prepare();
 }
 
 #endif

Index: linux-2.6/kernel/time/tick-sched.c
===================================================================
@@ -26,6 +26,7 @@
 #include <linux/posix-timers.h>
 #include <linux/context_tracking.h>
 #include <linux/mm.h>
+#include <linux/rcupdate.h>
 
 #include <asm/irq_regs.h>
 

@@ -519,6 +520,20 @@ void __tick_nohz_task_switch(void)
 	}
 }
 
+void __tick_nohz_user_enter_prepare(void)
+{
+	struct tick_sched *ts;
+
+	if (tick_nohz_full_cpu(smp_processor_id())) {
+		ts = this_cpu_ptr(&tick_cpu_sched);
+
+		if (ts->tick_stopped)
+			quiet_vmstat();
+		rcu_nocb_flush_deferred_wakeup();
+	}
+}
+EXPORT_SYMBOL_GPL(__tick_nohz_user_enter_prepare);
+
 /* Get the boot-time nohz CPU list from the kernel parameters. */
 void __init tick_nohz_full_setup(cpumask_var_t cpumask)
 {

@@ -890,6 +905,9 @@ static void tick_nohz_stop_tick(struct t
 		ts->do_timer_last = 0;
 	}
 
+	/* Attempt to fold when the idle tick is stopped or not */
+	quiet_vmstat();
+
 	/* Skip reprogram of event if its not changed */
 	if (ts->tick_stopped && (expires == ts->next_tick)) {
 		/* Sanity check: make sure clockevent is actually programmed */

@@ -911,7 +929,6 @@ static void tick_nohz_stop_tick(struct t
 	 */
 	if (!ts->tick_stopped) {
 		calc_load_nohz_start();
-		quiet_vmstat();
 
 		ts->last_tick = hrtimer_get_expires(&ts->sched_timer);
 		ts->tick_stopped = 1;

From the vmstat shepherd, for CPUs that have the tick stopped, do not queue local work to flush the per-CPU vmstats, since in that case the flush is performed on return to userspace or when entering idle.

Per-cpu pages can be freed remotely from housekeeping CPUs.

Move the quiet_vmstat call after ts->tick_stopped = 1 assignment.

---
 kernel/time/tick-sched.c |    6 +++---
 mm/vmstat.c              |   22 +++++++++++++++++-----
 2 files changed, 20 insertions(+), 8 deletions(-)

Index: linux-2.6/mm/vmstat.c
===================================================================
@@ -29,6 +29,7 @@
 #include <linux/page_ext.h>
 #include <linux/page_owner.h>
 #include <linux/migrate.h>
+#include <linux/tick.h>
 
 #include "internal.h"
 

@@ -1973,19 +1974,27 @@ static void vmstat_update(struct work_st
  */
 void quiet_vmstat(void)
 {
+	struct delayed_work *dw;
+
 	if (system_state != SYSTEM_RUNNING)
 		return;
 
 	if (!__this_cpu_read(vmstat_dirty))
 		return;
 
+	refresh_cpu_vm_stats(false);
+
 	/*
-	 * Just refresh counters and do not care about the pending delayed
-	 * vmstat_update. It doesn't fire that often to matter and canceling
-	 * it would be too expensive from this path.
-	 * vmstat_shepherd will take care about that for us.
+	 * If the tick is stopped, cancel any delayed work to avoid
+	 * interruptions to this CPU in the future.
+	 *
+	 * Otherwise just refresh counters and do not care about the pending
+	 * delayed vmstat_update. It doesn't fire that often to matter
+	 * and canceling it would be too expensive from this path.
 	 */
-	refresh_cpu_vm_stats(false);
+	dw = &per_cpu(vmstat_work, smp_processor_id());
+	if (delayed_work_pending(dw) && tick_nohz_tick_stopped())
+		cancel_delayed_work(dw);
 }
 
 /*

@@ -2007,6 +2016,9 @@ static void vmstat_shepherd(struct work_
 	for_each_online_cpu(cpu) {
 		struct delayed_work *dw = &per_cpu(vmstat_work, cpu);
 
+		if (tick_nohz_tick_stopped_cpu(cpu))
+			continue;
+
 		if (!delayed_work_pending(dw) && per_cpu(vmstat_dirty, cpu))
 			queue_delayed_work_on(cpu, mm_percpu_wq, dw, 0);
 

Index: linux-2.6/kernel/time/tick-sched.c
===================================================================
@@ -905,9 +905,6 @@ static void tick_nohz_stop_tick(struct t
 		ts->do_timer_last = 0;
 	}
 
-	/* Attempt to fold when the idle tick is stopped or not */
-	quiet_vmstat();
-
 	/* Skip reprogram of event if its not changed */
 	if (ts->tick_stopped && (expires == ts->next_tick)) {
 		/* Sanity check: make sure clockevent is actually programmed */

@@ -935,6 +932,9 @@ static void tick_nohz_stop_tick(struct t
 		trace_tick_stop(1, TICK_DEP_MASK_NONE);
 	}
 
+	/* Attempt to fold when the idle tick is stopped or not */
+	quiet_vmstat();
+
 	ts->next_tick = tick;
 
 	/*

Please ignore this series. Resent series with proper "[PATCH -v7]" subject.