The patch titled
Subject: mm, sysctl: make NUMA stats configurable
has been removed from the -mm tree. Its filename was
mm-sysctl-make-numa-stats-configurable.patch
This patch was dropped because an updated version will be merged
------------------------------------------------------
From: Kemi Wang <kemi.wang@xxxxxxxxx>
Subject: mm, sysctl: make NUMA stats configurable
When page allocation performance becomes a bottleneck and you can tolerate
some possible tool breakage and decreased numa counter precision, you can
do:
echo [C|c]oarse > /proc/sys/vm/numa_stats_mode
In this case, numa counter update is ignored.
We can see about *4.8%*(185->176) drop of cpu cycles per single page
allocation and reclaim on Jesper's page_bench01 (single thread) and
*8.1%*(343->315) drop of cpu cycles per single page allocation and reclaim
on Jesper's page_bench03 (88 threads) running on a 2-Socket
Broadwell-based server (88 threads, 126G memory).
Benchmark link provided by Jesper D Brouer(increase loop times to
10000000):
https://github.com/netoptimizer/prototype-kernel/tree/master/kernel/mm/bench
=========================================================================
When page allocation performance is not a bottleneck and you want all
tooling to work, you can do:
echo [S|s]trict > /proc/sys/vm/numa_stats_mode
=========================================================================
We recommend automatic detection of numa statistics by system, this is also
system default configuration, you can do:
echo [A|a]uto > /proc/sys/vm/numa_stats_mode
In this case, numa counter update is skipped unless it has been read by
users at least once, e.g. cat /proc/zoneinfo.
Branch target selection with jump label:
a) When numa_stats_mode is changed to *strict*, jump to the branch for
numa counters update.
b) When numa_stats_mode is changed to *coarse*, return back directly.
c) When numa_stats_mode is changed to *auto*, the branch target used in
last time is kept, and the branch target is changed to the branch for
numa counters update once numa counters are *read* by users.
Therefore, with the help of jump label, the page allocation performance is
hardly affected when numa counters are updated with a call in
zone_statistics(). Meanwhile, the auto mode can give people benefit
without manual tuning.
Many thanks to Michal Hocko, Dave Hansen and Ying Huang for comments to
help improve the original patch.
[akpm@xxxxxxxxxxxxxxxxxxxx: various fixes]
[akpm@xxxxxxxxxxxxxxxxxxxx: fix typo]
Link: http://lkml.kernel.org/r/1506579101-5457-1-git-send-email-kemi.wang@xxxxxxxxx
Signed-off-by: Kemi Wang <kemi.wang@xxxxxxxxx>
Reported-by: Jesper Dangaard Brouer <brouer@xxxxxxxxxx>
Suggested-by: Dave Hansen <dave.hansen@xxxxxxxxx>
Suggested-by: Ying Huang <ying.huang@xxxxxxxxx>
Cc: "Luis R . Rodriguez" <mcgrof@xxxxxxxxxx>
Cc: Kees Cook <keescook@xxxxxxxxxxxx>
Cc: Jonathan Corbet <corbet@xxxxxxx>
Cc: Michal Hocko <mhocko@xxxxxxxx>
Cc: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx>
Cc: Johannes Weiner <hannes@xxxxxxxxxxx>
Cc: Christopher Lameter <cl@xxxxxxxxx>
Cc: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
Cc: Vlastimil Babka <vbabka@xxxxxxx>
Cc: Tim Chen <tim.c.chen@xxxxxxxxx>
Cc: Andi Kleen <andi.kleen@xxxxxxxxx>
Cc: Aaron Lu <aaron.lu@xxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
---
Documentation/sysctl/vm.txt | 25 ++++++
drivers/base/node.c | 4 +
include/linux/vmstat.h | 23 ++++++
init/main.c | 3
kernel/sysctl.c | 7 +
mm/page_alloc.c | 10 ++
mm/vmstat.c | 126 ++++++++++++++++++++++++++++++++++
7 files changed, 198 insertions(+)
diff -puN Documentation/sysctl/vm.txt~mm-sysctl-make-numa-stats-configurable Documentation/sysctl/vm.txt
--- a/Documentation/sysctl/vm.txt~mm-sysctl-make-numa-stats-configurable
+++ a/Documentation/sysctl/vm.txt
@@ -61,6 +61,7 @@ Currently, these files are in /proc/sys/
- swappiness
- user_reserve_kbytes
- vfs_cache_pressure
+- numa_stats_mode
- watermark_scale_factor
- zone_reclaim_mode
@@ -850,6 +851,30 @@ ten times more freeable objects than the
=============================================================
+numa_stats_mode
+
+This interface allows runtime configuration of numa statistics.
+
+When page allocation performance becomes a bottleneck and you can tolerate
+some possible tool breakage and decreased numa counter precision, you can
+do:
+ echo [C|c]oarse > /proc/sys/vm/numa_stats_mode
+
+When page allocation performance is not a bottleneck and you want all
+tooling to work, you can do:
+ echo [S|s]trict > /proc/sys/vm/numa_stat_mode
+
+We recommend automatic detection of numa statistics by system, because
+numa statistics do not affect system decisions and it is very rarely
+consumed. In this case you can do:
+ echo [A|a]uto > /proc/sys/vm/numa_stats_mode
+
+This is the system default configuration. With this default setting, numa
+counter updates are skipped until the counter is *read* by userspace at
+least once.
+
+==============================================================
+
watermark_scale_factor:
This factor controls the aggressiveness of kswapd. It defines the
diff -puN drivers/base/node.c~mm-sysctl-make-numa-stats-configurable drivers/base/node.c
--- a/drivers/base/node.c~mm-sysctl-make-numa-stats-configurable
+++ a/drivers/base/node.c
@@ -161,6 +161,8 @@ static DEVICE_ATTR(meminfo, S_IRUGO, nod
static ssize_t node_read_numastat(struct device *dev,
struct device_attribute *attr, char *buf)
{
+ if (vm_numa_stats_mode == VM_NUMA_STAT_AUTO_MODE)
+ static_branch_enable(&vm_numa_stats_mode_key);
return sprintf(buf,
"numa_hit %lu\n"
"numa_miss %lu\n"
@@ -194,6 +196,8 @@ static ssize_t node_read_vmstat(struct d
n += sprintf(buf+n, "%s %lu\n",
vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
sum_zone_numa_state(nid, i));
+ if (vm_numa_stats_mode == VM_NUMA_STAT_AUTO_MODE)
+ static_branch_enable(&vm_numa_stats_mode_key);
#endif
for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
diff -puN include/linux/vmstat.h~mm-sysctl-make-numa-stats-configurable include/linux/vmstat.h
--- a/include/linux/vmstat.h~mm-sysctl-make-numa-stats-configurable
+++ a/include/linux/vmstat.h
@@ -6,9 +6,28 @@
#include <linux/mmzone.h>
#include <linux/vm_event_item.h>
#include <linux/atomic.h>
+#include <linux/static_key.h>
extern int sysctl_stat_interval;
+#ifdef CONFIG_NUMA
+DECLARE_STATIC_KEY_FALSE(vm_numa_stats_mode_key);
+/*
+ * vm_numa_stats_mode:
+ * 0 = auto mode of NUMA stats, automatic detection of NUMA statistics.
+ * 1 = strict mode of NUMA stats, keep NUMA statistics.
+ * 2 = coarse mode of NUMA stats, ignore NUMA statistics.
+ */
+#define VM_NUMA_STAT_AUTO_MODE 0
+#define VM_NUMA_STAT_STRICT_MODE 1
+#define VM_NUMA_STAT_COARSE_MODE 2
+#define VM_NUMA_STAT_MODE_LEN 16
+extern int vm_numa_stats_mode;
+extern char sysctl_vm_numa_stats_mode[];
+extern int sysctl_vm_numa_stats_mode_handler(struct ctl_table *table, int write,
+ void __user *buffer, size_t *length, loff_t *ppos);
+#endif
+
#ifdef CONFIG_VM_EVENT_COUNTERS
/*
* Light weight per cpu counter implementation.
@@ -229,6 +248,10 @@ extern unsigned long sum_zone_node_page_
extern unsigned long sum_zone_numa_state(int node, enum numa_stat_item item);
extern unsigned long node_page_state(struct pglist_data *pgdat,
enum node_stat_item item);
+extern void zero_zone_numa_counters(struct zone *zone);
+extern void zero_zones_numa_counters(void);
+extern void zero_global_numa_counters(void);
+extern void invalid_numa_statistics(void);
#else
#define sum_zone_node_page_state(node, item) global_zone_page_state(item)
#define node_page_state(node, item) global_node_page_state(item)
diff -puN init/main.c~mm-sysctl-make-numa-stats-configurable init/main.c
--- a/init/main.c~mm-sysctl-make-numa-stats-configurable
+++ a/init/main.c
@@ -504,6 +504,9 @@ static void __init mm_init(void)
pgtable_init();
vmalloc_init();
ioremap_huge_init();
+#ifdef CONFIG_NUMA
+ pr_info("vmstat: NUMA stat updates are skipped unless they have been used\n");
+#endif
}
asmlinkage __visible void __init start_kernel(void)
diff -puN kernel/sysctl.c~mm-sysctl-make-numa-stats-configurable kernel/sysctl.c
--- a/kernel/sysctl.c~mm-sysctl-make-numa-stats-configurable
+++ a/kernel/sysctl.c
@@ -1371,6 +1371,13 @@ static struct ctl_table vm_table[] = {
.mode = 0644,
.proc_handler = &hugetlb_mempolicy_sysctl_handler,
},
+ {
+ .procname = "numa_stats_mode",
+ .data = sysctl_vm_numa_stats_mode,
+ .maxlen = VM_NUMA_STAT_MODE_LEN,
+ .mode = 0644,
+ .proc_handler = sysctl_vm_numa_stats_mode_handler,
+ },
#endif
{
.procname = "hugetlb_shm_group",
diff -puN mm/page_alloc.c~mm-sysctl-make-numa-stats-configurable mm/page_alloc.c
--- a/mm/page_alloc.c~mm-sysctl-make-numa-stats-configurable
+++ a/mm/page_alloc.c
@@ -83,6 +83,8 @@ DEFINE_PER_CPU(int, numa_node);
EXPORT_PER_CPU_SYMBOL(numa_node);
#endif
+DEFINE_STATIC_KEY_FALSE(vm_numa_stats_mode_key);
+
#ifdef CONFIG_HAVE_MEMORYLESS_NODES
/*
* N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
@@ -2739,6 +2741,14 @@ static inline void zone_statistics(struc
#ifdef CONFIG_NUMA
enum numa_stat_item local_stat = NUMA_LOCAL;
+ /*
+ * skip zone_statistics() if NUMA stats is set to coarse mode or
+ * NUMA stats is never consumed in auto mode.
+ */
+
+ if (!static_branch_unlikely(&vm_numa_stats_mode_key))
+ return;
+
if (z->node != numa_node_id())
local_stat = NUMA_OTHER;
diff -puN mm/vmstat.c~mm-sysctl-make-numa-stats-configurable mm/vmstat.c
--- a/mm/vmstat.c~mm-sysctl-make-numa-stats-configurable
+++ a/mm/vmstat.c
@@ -32,6 +32,88 @@
#define NUMA_STATS_THRESHOLD (U16_MAX - 2)
+#ifdef CONFIG_NUMA
+int vm_numa_stats_mode = VM_NUMA_STAT_AUTO_MODE;
+char sysctl_vm_numa_stats_mode[VM_NUMA_STAT_MODE_LEN] = "auto";
+static const char *vm_numa_stats_mode_name[3] = {"auto", "strict", "coarse"};
+static DEFINE_MUTEX(vm_numa_stats_mode_lock);
+
+static int __parse_vm_numa_stats_mode(char *s)
+{
+ if (strcasecmp(s, "auto"))
+ vm_numa_stats_mode = VM_NUMA_STAT_AUTO_MODE;
+ else if (strcasecmp(s, "strict") == 0)
+ vm_numa_stats_mode = VM_NUMA_STAT_STRICT_MODE;
+ else if (strcasecmp(s, "coarse"))
+ vm_numa_stats_mode = VM_NUMA_STAT_COARSE_MODE;
+ else {
+ pr_warn("Ignoring invalid vm_numa_stats_mode value: %s\n", s);
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+int sysctl_vm_numa_stats_mode_handler(struct ctl_table *table, int write,
+ void __user *buffer, size_t *length, loff_t *ppos)
+{
+ char old_string[VM_NUMA_STAT_MODE_LEN];
+ int ret, oldval;
+
+ mutex_lock(&vm_numa_stats_mode_lock);
+ if (write)
+ strncpy(old_string, (char *)table->data, VM_NUMA_STAT_MODE_LEN);
+ ret = proc_dostring(table, write, buffer, length, ppos);
+ if (ret || !write) {
+ mutex_unlock(&vm_numa_stats_mode_lock);
+ return ret;
+ }
+
+ oldval = vm_numa_stats_mode;
+ if (__parse_vm_numa_stats_mode((char *)table->data)) {
+ /*
+ * invalid sysctl_vm_numa_stats_mode value, restore saved string
+ */
+ strncpy((char *)table->data, old_string, VM_NUMA_STAT_MODE_LEN);
+ vm_numa_stats_mode = oldval;
+ } else {
+ /*
+ * check whether numa stats mode changes or not
+ */
+ if (vm_numa_stats_mode == oldval) {
+ /* no change */
+ mutex_unlock(&vm_numa_stats_mode_lock);
+ return 0;
+ } else if (vm_numa_stats_mode == VM_NUMA_STAT_AUTO_MODE) {
+ /*
+ * Keep the branch selection in last time when numa
+ * stats is changed to auto mode.
+ */
+ pr_info("numa stats changed from %s to auto mode\n",
+ vm_numa_stats_mode_name[oldval]);
+ } else if (vm_numa_stats_mode == VM_NUMA_STAT_STRICT_MODE) {
+ static_branch_enable(&vm_numa_stats_mode_key);
+ pr_info("numa stats changes from %s to strict mode\n",
+ vm_numa_stats_mode_name[oldval]);
+ } else if (vm_numa_stats_mode == VM_NUMA_STAT_COARSE_MODE) {
+ static_branch_disable(&vm_numa_stats_mode_key);
+ /*
+ * Invalidate numa counters when vmstat mode is set to
+ * coarse mode, because users can't tell the difference
+ * between the dead state and when allocator activity is
+ * quiet once zone_statistics() is turned off.
+ */
+ invalid_numa_statistics();
+ pr_info("numa stats changes from %s to coarse mode\n",
+ vm_numa_stats_mode_name[oldval]);
+ }
+ }
+
+ mutex_unlock(&vm_numa_stats_mode_lock);
+ return 0;
+}
+#endif
+
#ifdef CONFIG_VM_EVENT_COUNTERS
DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
EXPORT_PER_CPU_SYMBOL(vm_event_states);
@@ -914,6 +996,42 @@ unsigned long sum_zone_numa_state(int no
return count;
}
+/* zero numa counters within a zone */
+void zero_zone_numa_counters(struct zone *zone)
+{
+ int item, cpu;
+
+ for (item = 0; item < NR_VM_NUMA_STAT_ITEMS; item++) {
+ atomic_long_set(&zone->vm_numa_stat[item], 0);
+ for_each_online_cpu(cpu)
+ per_cpu_ptr(zone->pageset, cpu)->vm_numa_stat_diff[item] = 0;
+ }
+}
+
+/* zero numa counters of all the populated zones */
+void zero_zones_numa_counters(void)
+{
+ struct zone *zone;
+
+ for_each_populated_zone(zone)
+ zero_zone_numa_counters(zone);
+}
+
+/* zero global numa counters */
+void zero_global_numa_counters(void)
+{
+ int item;
+
+ for (item = 0; item < NR_VM_NUMA_STAT_ITEMS; item++)
+ atomic_long_set(&vm_numa_stat[item], 0);
+}
+
+void invalid_numa_statistics(void)
+{
+ zero_zones_numa_counters();
+ zero_global_numa_counters();
+}
+
/*
* Determine the per node value of a stat item.
*/
@@ -1582,6 +1700,10 @@ static int zoneinfo_show(struct seq_file
{
pg_data_t *pgdat = (pg_data_t *)arg;
walk_zones_in_node(m, pgdat, false, false, zoneinfo_show_print);
+#ifdef CONFIG_NUMA
+ if (vm_numa_stats_mode == VM_NUMA_STAT_AUTO_MODE)
+ static_branch_enable(&vm_numa_stats_mode_key);
+#endif
return 0;
}
@@ -1678,6 +1800,10 @@ static int vmstat_show(struct seq_file *
static void vmstat_stop(struct seq_file *m, void *arg)
{
+#ifdef CONFIG_NUMA
+ if (vm_numa_stats_mode == VM_NUMA_STAT_AUTO_MODE)
+ static_branch_enable(&vm_numa_stats_mode_key);
+#endif
kfree(m->private);
m->private = NULL;
}
_
Patches currently in -mm which might be from kemi.wang@xxxxxxxxx are
--
To unsubscribe from this list: send the line "unsubscribe mm-commits" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
