This is the second step which introduces a tunable interface that allow
numa stats configurable for optimizing zone_statistics(), as suggested by
Dave Hansen and Ying Huang.
=========================================================================
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.
ChangeLog:
V2->V3:
a) Propose a better way to use jump label to eliminate the overhead of
branch selection in zone_statistics(), as inspired by Ying Huang;
b) Add a paragraph in commit log to describe the way for branch target
selection;
c) Use a more descriptive name numa_stats_mode instead of vmstat_mode,
and change the description accordingly, as suggested by Michal Hocko;
d) Make this functionality NUMA-specific via ifdef
V1->V2:
a) Merge to one patch;
b) Use jump label to eliminate the overhead of branch selection;
c) Add a single-time log message at boot time to help tell users what
happened.
Reported-by: Jesper Dangaard Brouer <[email protected]>
Suggested-by: Dave Hansen <[email protected]>
Suggested-by: Ying Huang <[email protected]>
Signed-off-by: Kemi Wang <[email protected]>
---
Documentation/sysctl/vm.txt | 24 +++++++++
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 | 129 ++++++++++++++++++++++++++++++++++++++++++++
7 files changed, 200 insertions(+)
diff --git a/Documentation/sysctl/vm.txt b/Documentation/sysctl/vm.txt
index 9baf66a..e310e69 100644
--- a/Documentation/sysctl/vm.txt
+++ b/Documentation/sysctl/vm.txt
@@ -61,6 +61,7 @@ Currently, these files are in /proc/sys/vm:
- swappiness
- user_reserve_kbytes
- vfs_cache_pressure
+- numa_stats_mode
- watermark_scale_factor
- zone_reclaim_mode
@@ -843,6 +844,29 @@ ten times more freeable objects than there are.
=============================================================
+numa_stats_mode
+
+This interface allows numa statistics 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
+
+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 does not affect system's decision and it is very rarely
+consumed. you can do:
+ echo [A|a]uto > /proc/sys/vm/numa_stats_mode
+This is also system default configuration, with this default setting, numa
+counters update is skipped unless the counter is *read* by users at least
+once.
+
+==============================================================
+
watermark_scale_factor:
This factor controls the aggressiveness of kswapd. It defines the
diff --git a/drivers/base/node.c b/drivers/base/node.c
index 3855902..b57b5622 100644
--- a/drivers/base/node.c
+++ b/drivers/base/node.c
@@ -153,6 +153,8 @@ static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
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"
@@ -186,6 +188,8 @@ static ssize_t node_read_vmstat(struct device *dev,
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 --git a/include/linux/vmstat.h b/include/linux/vmstat.h
index ade7cb5..d52e882 100644
--- a/include/linux/vmstat.h
+++ b/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_state(int node,
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 --git a/init/main.c b/init/main.c
index 0ee9c686..1e300a8 100644
--- a/init/main.c
+++ b/init/main.c
@@ -567,6 +567,9 @@ asmlinkage __visible void __init start_kernel(void)
sort_main_extable();
trap_init();
mm_init();
+#ifdef CONFIG_NUMA
+ pr_info("vmstat: NUMA stats is skipped unless it has been consumed\n");
+#endif
ftrace_init();
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index 6648fbb..0678668 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -1374,6 +1374,13 @@ 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 --git a/mm/page_alloc.c b/mm/page_alloc.c
index c841af8..6d7ea18 100644
--- a/mm/page_alloc.c
+++ b/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.
@@ -2743,6 +2745,14 @@ static inline void zone_statistics(struct zone *preferred_zone, struct zone *z)
#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 --git a/mm/vmstat.c b/mm/vmstat.c
index 4bb13e7..469599c 100644
--- a/mm/vmstat.c
+++ b/mm/vmstat.c
@@ -32,6 +32,91 @@
#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)
+{
+ const char *str = s;
+
+ if (strcmp(str, "auto") == 0 || strcmp(str, "Auto") == 0)
+ vm_numa_stats_mode = VM_NUMA_STAT_AUTO_MODE;
+ else if (strcmp(str, "strict") == 0 || strcmp(str, "Strict") == 0)
+ vm_numa_stats_mode = VM_NUMA_STAT_STRICT_MODE;
+ else if (strcmp(str, "coarse") == 0 || strcmp(str, "Coarse") == 0)
+ 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 changes from %s mode 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 mode 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 mode to coarse mode\n",
+ vm_numa_stats_mode_name[oldval]);
+ } else
+ pr_warn("invalid vm_numa_stats_mode:%d\n", vm_numa_stats_mode);
+ }
+
+ 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 +999,42 @@ unsigned long sum_zone_numa_state(int node,
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 +1703,10 @@ static int zoneinfo_show(struct seq_file *m, void *arg)
{
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 +1803,10 @@ static int vmstat_show(struct seq_file *m, void *arg)
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;
}
--
2.7.4
On Thu, 28 Sep 2017 14:11:41 +0800 Kemi Wang <[email protected]> wrote:
> This is the second step which introduces a tunable interface that allow
> numa stats configurable for optimizing zone_statistics(), as suggested by
> Dave Hansen and Ying Huang.
Looks OK I guess.
I fiddled with it a lot. Please consider:
From: Andrew Morton <[email protected]>
Subject: mm-sysctl-make-numa-stats-configurable-fix
- tweak documentation
- move advisory message from start_kernel() into mm_init() (I'm not sure
we really need this message)
- use strcasecmp() in __parse_vm_numa_stats_mode()
- clean up coding style amd nessages in sysctl_vm_numa_stats_mode_handler()
Cc: Aaron Lu <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: Christopher Lameter <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Jesper Dangaard Brouer <[email protected]>
Cc: Johannes Weiner <[email protected]>
Cc: Jonathan Corbet <[email protected]>
Cc: Kees Cook <[email protected]>
Cc: Kemi Wang <[email protected]>
Cc: "Luis R . Rodriguez" <[email protected]>
Cc: Mel Gorman <[email protected]>
Cc: Michal Hocko <[email protected]>
Cc: Sebastian Andrzej Siewior <[email protected]>
Cc: Tim Chen <[email protected]>
Cc: Vlastimil Babka <[email protected]>
Cc: Ying Huang <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
---
Documentation/sysctl/vm.txt | 15 ++++++-------
init/main.c | 6 ++---
mm/vmstat.c | 39 +++++++++++++++-------------------
3 files changed, 29 insertions(+), 31 deletions(-)
diff -puN Documentation/sysctl/vm.txt~mm-sysctl-make-numa-stats-configurable-fix Documentation/sysctl/vm.txt
--- a/Documentation/sysctl/vm.txt~mm-sysctl-make-numa-stats-configurable-fix
+++ a/Documentation/sysctl/vm.txt
@@ -853,7 +853,7 @@ ten times more freeable objects than the
numa_stats_mode
-This interface allows numa statistics configurable.
+This interface allows runtime configuration or numa statistics.
When page allocation performance becomes a bottleneck and you can tolerate
some possible tool breakage and decreased numa counter precision, you can
@@ -864,13 +864,14 @@ When page allocation performance is not
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 does not affect system's decision and it is very rarely
-consumed. you can do:
+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 also system default configuration, with this default setting, numa
-counters update is skipped unless the counter is *read* by users at least
-once.
+
+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.
==============================================================
diff -puN drivers/base/node.c~mm-sysctl-make-numa-stats-configurable-fix drivers/base/node.c
diff -puN include/linux/vmstat.h~mm-sysctl-make-numa-stats-configurable-fix include/linux/vmstat.h
diff -puN init/main.c~mm-sysctl-make-numa-stats-configurable-fix init/main.c
--- a/init/main.c~mm-sysctl-make-numa-stats-configurable-fix
+++ 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)
@@ -567,9 +570,6 @@ asmlinkage __visible void __init start_k
sort_main_extable();
trap_init();
mm_init();
-#ifdef CONFIG_NUMA
- pr_info("vmstat: NUMA stats is skipped unless it has been consumed\n");
-#endif
ftrace_init();
diff -puN kernel/sysctl.c~mm-sysctl-make-numa-stats-configurable-fix kernel/sysctl.c
diff -puN mm/page_alloc.c~mm-sysctl-make-numa-stats-configurable-fix mm/page_alloc.c
diff -puN mm/vmstat.c~mm-sysctl-make-numa-stats-configurable-fix mm/vmstat.c
--- a/mm/vmstat.c~mm-sysctl-make-numa-stats-configurable-fix
+++ a/mm/vmstat.c
@@ -40,13 +40,11 @@ static DEFINE_MUTEX(vm_numa_stats_mode_l
static int __parse_vm_numa_stats_mode(char *s)
{
- const char *str = s;
-
- if (strcmp(str, "auto") == 0 || strcmp(str, "Auto") == 0)
+ if (strcasecmp(s, "auto"))
vm_numa_stats_mode = VM_NUMA_STAT_AUTO_MODE;
- else if (strcmp(str, "strict") == 0 || strcmp(str, "Strict") == 0)
+ else if (strcasecmp(s, "strict") == 0)
vm_numa_stats_mode = VM_NUMA_STAT_STRICT_MODE;
- else if (strcmp(str, "coarse") == 0 || strcmp(str, "Coarse") == 0)
+ 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);
@@ -86,30 +84,29 @@ int sysctl_vm_numa_stats_mode_handler(st
/* no change */
mutex_unlock(&vm_numa_stats_mode_lock);
return 0;
- } else if (vm_numa_stats_mode == VM_NUMA_STAT_AUTO_MODE)
+ } 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.
+ * Keep the branch selection in last time when numa
+ * stats is changed to auto mode.
*/
- pr_info("numa stats changes from %s mode to auto mode\n",
- vm_numa_stats_mode_name[oldval]);
- else if (vm_numa_stats_mode == VM_NUMA_STAT_STRICT_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 mode to strict mode\n",
- vm_numa_stats_mode_name[oldval]);
+ 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.
+ * 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 mode to coarse mode\n",
- vm_numa_stats_mode_name[oldval]);
- } else
- pr_warn("invalid vm_numa_stats_mode:%d\n", vm_numa_stats_mode);
+ pr_info("numa stats changes from %s to coarse mode\n",
+ vm_numa_stats_mode_name[oldval]);
+ }
}
mutex_unlock(&vm_numa_stats_mode_lock);
_
On 2017年09月29日 05:29, Andrew Morton wrote:
> On Thu, 28 Sep 2017 14:11:41 +0800 Kemi Wang <[email protected]> wrote:
>
>> This is the second step which introduces a tunable interface that allow
>> numa stats configurable for optimizing zone_statistics(), as suggested by
>> Dave Hansen and Ying Huang.
>
> Looks OK I guess.
>
> I fiddled with it a lot. Please consider:
>
Thanks for your help to make it more graceful! I will be more careful next time.
There may be a typo error in Documentation/sysctl/vm.txt, see comment below.
> From: Andrew Morton <[email protected]>
> Subject: mm-sysctl-make-numa-stats-configurable-fix
>
> - tweak documentation
>
> - move advisory message from start_kernel() into mm_init() (I'm not sure
> we really need this message)
>
> - use strcasecmp() in __parse_vm_numa_stats_mode()
>
> - clean up coding style amd nessages in sysctl_vm_numa_stats_mode_handler()
>
> Cc: Aaron Lu <[email protected]>
> Cc: Andi Kleen <[email protected]>
> Cc: Christopher Lameter <[email protected]>
> Cc: Dave Hansen <[email protected]>
> Cc: Jesper Dangaard Brouer <[email protected]>
> Cc: Johannes Weiner <[email protected]>
> Cc: Jonathan Corbet <[email protected]>
> Cc: Kees Cook <[email protected]>
> Cc: Kemi Wang <[email protected]>
> Cc: "Luis R . Rodriguez" <[email protected]>
> Cc: Mel Gorman <[email protected]>
> Cc: Michal Hocko <[email protected]>
> Cc: Sebastian Andrzej Siewior <[email protected]>
> Cc: Tim Chen <[email protected]>
> Cc: Vlastimil Babka <[email protected]>
> Cc: Ying Huang <[email protected]>
> Signed-off-by: Andrew Morton <[email protected]>
> ---
>
> Documentation/sysctl/vm.txt | 15 ++++++-------
> init/main.c | 6 ++---
> mm/vmstat.c | 39 +++++++++++++++-------------------
> 3 files changed, 29 insertions(+), 31 deletions(-)
>
> diff -puN Documentation/sysctl/vm.txt~mm-sysctl-make-numa-stats-configurable-fix Documentation/sysctl/vm.txt
> --- a/Documentation/sysctl/vm.txt~mm-sysctl-make-numa-stats-configurable-fix
> +++ a/Documentation/sysctl/vm.txt
> @@ -853,7 +853,7 @@ ten times more freeable objects than the
>
> numa_stats_mode
>
> -This interface allows numa statistics configurable.
> +This interface allows runtime configuration *or* numa statistics.
>
typo? or->of/for?
> When page allocation performance becomes a bottleneck and you can tolerate
> some possible tool breakage and decreased numa counter precision, you can
> @@ -864,13 +864,14 @@ When page allocation performance is not
> 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 does not affect system's decision and it is very rarely
> -consumed. you can do:
> +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 also system default configuration, with this default setting, numa
> -counters update is skipped unless the counter is *read* by users at least
> -once.
> +
> +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.
>
> ==============================================================
>
> diff -puN drivers/base/node.c~mm-sysctl-make-numa-stats-configurable-fix drivers/base/node.c
> diff -puN include/linux/vmstat.h~mm-sysctl-make-numa-stats-configurable-fix include/linux/vmstat.h
> diff -puN init/main.c~mm-sysctl-make-numa-stats-configurable-fix init/main.c
> --- a/init/main.c~mm-sysctl-make-numa-stats-configurable-fix
> +++ 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)
> @@ -567,9 +570,6 @@ asmlinkage __visible void __init start_k
> sort_main_extable();
> trap_init();
> mm_init();
> -#ifdef CONFIG_NUMA
> - pr_info("vmstat: NUMA stats is skipped unless it has been consumed\n");
> -#endif
>
> ftrace_init();
>
> diff -puN kernel/sysctl.c~mm-sysctl-make-numa-stats-configurable-fix kernel/sysctl.c
> diff -puN mm/page_alloc.c~mm-sysctl-make-numa-stats-configurable-fix mm/page_alloc.c
> diff -puN mm/vmstat.c~mm-sysctl-make-numa-stats-configurable-fix mm/vmstat.c
> --- a/mm/vmstat.c~mm-sysctl-make-numa-stats-configurable-fix
> +++ a/mm/vmstat.c
> @@ -40,13 +40,11 @@ static DEFINE_MUTEX(vm_numa_stats_mode_l
>
> static int __parse_vm_numa_stats_mode(char *s)
> {
> - const char *str = s;
> -
> - if (strcmp(str, "auto") == 0 || strcmp(str, "Auto") == 0)
> + if (strcasecmp(s, "auto"))
> vm_numa_stats_mode = VM_NUMA_STAT_AUTO_MODE;
> - else if (strcmp(str, "strict") == 0 || strcmp(str, "Strict") == 0)
> + else if (strcasecmp(s, "strict") == 0)
> vm_numa_stats_mode = VM_NUMA_STAT_STRICT_MODE;
> - else if (strcmp(str, "coarse") == 0 || strcmp(str, "Coarse") == 0)
> + 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);
> @@ -86,30 +84,29 @@ int sysctl_vm_numa_stats_mode_handler(st
> /* no change */
> mutex_unlock(&vm_numa_stats_mode_lock);
> return 0;
> - } else if (vm_numa_stats_mode == VM_NUMA_STAT_AUTO_MODE)
> + } 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.
> + * Keep the branch selection in last time when numa
> + * stats is changed to auto mode.
> */
> - pr_info("numa stats changes from %s mode to auto mode\n",
> - vm_numa_stats_mode_name[oldval]);
> - else if (vm_numa_stats_mode == VM_NUMA_STAT_STRICT_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 mode to strict mode\n",
> - vm_numa_stats_mode_name[oldval]);
> + 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.
> + * 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 mode to coarse mode\n",
> - vm_numa_stats_mode_name[oldval]);
> - } else
> - pr_warn("invalid vm_numa_stats_mode:%d\n", vm_numa_stats_mode);
> + pr_info("numa stats changes from %s to coarse mode\n",
> + vm_numa_stats_mode_name[oldval]);
> + }
> }
>
> mutex_unlock(&vm_numa_stats_mode_lock);
> _
>
On 09/28/2017 08:11 AM, Kemi Wang wrote:
> This is the second step which introduces a tunable interface that allow
> numa stats configurable for optimizing zone_statistics(), as suggested by
> Dave Hansen and Ying Huang.
>
> =========================================================================
> 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.
>
> ChangeLog:
> V2->V3:
> a) Propose a better way to use jump label to eliminate the overhead of
> branch selection in zone_statistics(), as inspired by Ying Huang;
> b) Add a paragraph in commit log to describe the way for branch target
> selection;
> c) Use a more descriptive name numa_stats_mode instead of vmstat_mode,
> and change the description accordingly, as suggested by Michal Hocko;
> d) Make this functionality NUMA-specific via ifdef
>
> V1->V2:
> a) Merge to one patch;
> b) Use jump label to eliminate the overhead of branch selection;
> c) Add a single-time log message at boot time to help tell users what
> happened.
>
> Reported-by: Jesper Dangaard Brouer <[email protected]>
> Suggested-by: Dave Hansen <[email protected]>
> Suggested-by: Ying Huang <[email protected]>
> Signed-off-by: Kemi Wang <[email protected]>
> ---
> Documentation/sysctl/vm.txt | 24 +++++++++
> 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 | 129 ++++++++++++++++++++++++++++++++++++++++++++
> 7 files changed, 200 insertions(+)
>
> diff --git a/Documentation/sysctl/vm.txt b/Documentation/sysctl/vm.txt
> index 9baf66a..e310e69 100644
> --- a/Documentation/sysctl/vm.txt
> +++ b/Documentation/sysctl/vm.txt
> @@ -61,6 +61,7 @@ Currently, these files are in /proc/sys/vm:
> - swappiness
> - user_reserve_kbytes
> - vfs_cache_pressure
> +- numa_stats_mode
> - watermark_scale_factor
> - zone_reclaim_mode
>
> @@ -843,6 +844,29 @@ ten times more freeable objects than there are.
>
> =============================================================
>
> +numa_stats_mode
> +
> +This interface allows numa statistics 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
> +
> +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 does not affect system's decision and it is very rarely
> +consumed. you can do:
> + echo [A|a]uto > /proc/sys/vm/numa_stats_mode
> +This is also system default configuration, with this default setting, numa
> +counters update is skipped unless the counter is *read* by users at least
> +once.
It says "the counter", but it seems multiple files in /proc and /sys are
triggering this, so perhaps list them?
Also, is it possible that with contemporary userspace/distros (systemd
etc.) there will always be something that will read one of those upon boot?
> +
> +==============================================================
> +
> watermark_scale_factor:
>
> This factor controls the aggressiveness of kswapd. It defines the
> diff --git a/drivers/base/node.c b/drivers/base/node.c
> index 3855902..b57b5622 100644
> --- a/drivers/base/node.c
> +++ b/drivers/base/node.c
> @@ -153,6 +153,8 @@ static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
> 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"
> @@ -186,6 +188,8 @@ static ssize_t node_read_vmstat(struct device *dev,
> 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 --git a/include/linux/vmstat.h b/include/linux/vmstat.h
> index ade7cb5..d52e882 100644
> --- a/include/linux/vmstat.h
> +++ b/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_state(int node,
> 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);
These seem to be called only from within mm/vmstat.c where they live, so
I'd suggest removing these extern declarations, and making them static
in vmstat.c.
...
> #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)
> +{
> + const char *str = s;
> +
> + if (strcmp(str, "auto") == 0 || strcmp(str, "Auto") == 0)
> + vm_numa_stats_mode = VM_NUMA_STAT_AUTO_MODE;
> + else if (strcmp(str, "strict") == 0 || strcmp(str, "Strict") == 0)
> + vm_numa_stats_mode = VM_NUMA_STAT_STRICT_MODE;
> + else if (strcmp(str, "coarse") == 0 || strcmp(str, "Coarse") == 0)
> + 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;
Do we need to restore vm_numa_stats_mode? AFAICS it didn't change. Also,
should the EINVAL be returned also to userspace? (not sure what's the
API here, hmm man 2 sysctl doesn't list EINVAL...)
On 09/28/2017 11:29 PM, Andrew Morton wrote:
> On Thu, 28 Sep 2017 14:11:41 +0800 Kemi Wang <[email protected]> wrote:
>
>> This is the second step which introduces a tunable interface that allow
>> numa stats configurable for optimizing zone_statistics(), as suggested by
>> Dave Hansen and Ying Huang.
>
> Looks OK I guess.
>
> I fiddled with it a lot. Please consider:
>
> From: Andrew Morton <[email protected]>
> Subject: mm-sysctl-make-numa-stats-configurable-fix
>
> - tweak documentation
>
> - move advisory message from start_kernel() into mm_init() (I'm not sure
> we really need this message)
Actually, I'm not sure we need any of the current messages, or to have
them at higher priority than pr_debug()? They are all triggered by admin
action, or unconditionally upon boot.
OTOH I think that an useful message that's currently missing would be
when the static_key_enable() is triggered in auto mode. Bonus points for
including the name of the process and the stat file that was read.
However static_key_enable() returns void and not whether it actually
flipped the switch, so it's not trivial.
[+CC linux-api]
On 09/28/2017 08:11 AM, Kemi Wang wrote:
> This is the second step which introduces a tunable interface that allow
> numa stats configurable for optimizing zone_statistics(), as suggested by
> Dave Hansen and Ying Huang.
>
> =========================================================================
> 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.
>
> ChangeLog:
> V2->V3:
> a) Propose a better way to use jump label to eliminate the overhead of
> branch selection in zone_statistics(), as inspired by Ying Huang;
> b) Add a paragraph in commit log to describe the way for branch target
> selection;
> c) Use a more descriptive name numa_stats_mode instead of vmstat_mode,
> and change the description accordingly, as suggested by Michal Hocko;
> d) Make this functionality NUMA-specific via ifdef
>
> V1->V2:
> a) Merge to one patch;
> b) Use jump label to eliminate the overhead of branch selection;
> c) Add a single-time log message at boot time to help tell users what
> happened.
>
> Reported-by: Jesper Dangaard Brouer <[email protected]>
> Suggested-by: Dave Hansen <[email protected]>
> Suggested-by: Ying Huang <[email protected]>
> Signed-off-by: Kemi Wang <[email protected]>
> ---
> Documentation/sysctl/vm.txt | 24 +++++++++
> 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 | 129 ++++++++++++++++++++++++++++++++++++++++++++
> 7 files changed, 200 insertions(+)
>
> diff --git a/Documentation/sysctl/vm.txt b/Documentation/sysctl/vm.txt
> index 9baf66a..e310e69 100644
> --- a/Documentation/sysctl/vm.txt
> +++ b/Documentation/sysctl/vm.txt
> @@ -61,6 +61,7 @@ Currently, these files are in /proc/sys/vm:
> - swappiness
> - user_reserve_kbytes
> - vfs_cache_pressure
> +- numa_stats_mode
> - watermark_scale_factor
> - zone_reclaim_mode
>
> @@ -843,6 +844,29 @@ ten times more freeable objects than there are.
>
> =============================================================
>
> +numa_stats_mode
> +
> +This interface allows numa statistics 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
> +
> +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 does not affect system's decision and it is very rarely
> +consumed. you can do:
> + echo [A|a]uto > /proc/sys/vm/numa_stats_mode
> +This is also system default configuration, with this default setting, numa
> +counters update is skipped unless the counter is *read* by users at least
> +once.
> +
> +==============================================================
> +
> watermark_scale_factor:
>
> This factor controls the aggressiveness of kswapd. It defines the
> diff --git a/drivers/base/node.c b/drivers/base/node.c
> index 3855902..b57b5622 100644
> --- a/drivers/base/node.c
> +++ b/drivers/base/node.c
> @@ -153,6 +153,8 @@ static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
> 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"
> @@ -186,6 +188,8 @@ static ssize_t node_read_vmstat(struct device *dev,
> 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 --git a/include/linux/vmstat.h b/include/linux/vmstat.h
> index ade7cb5..d52e882 100644
> --- a/include/linux/vmstat.h
> +++ b/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_state(int node,
> 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 --git a/init/main.c b/init/main.c
> index 0ee9c686..1e300a8 100644
> --- a/init/main.c
> +++ b/init/main.c
> @@ -567,6 +567,9 @@ asmlinkage __visible void __init start_kernel(void)
> sort_main_extable();
> trap_init();
> mm_init();
> +#ifdef CONFIG_NUMA
> + pr_info("vmstat: NUMA stats is skipped unless it has been consumed\n");
> +#endif
>
> ftrace_init();
>
> diff --git a/kernel/sysctl.c b/kernel/sysctl.c
> index 6648fbb..0678668 100644
> --- a/kernel/sysctl.c
> +++ b/kernel/sysctl.c
> @@ -1374,6 +1374,13 @@ 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 --git a/mm/page_alloc.c b/mm/page_alloc.c
> index c841af8..6d7ea18 100644
> --- a/mm/page_alloc.c
> +++ b/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.
> @@ -2743,6 +2745,14 @@ static inline void zone_statistics(struct zone *preferred_zone, struct zone *z)
> #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 --git a/mm/vmstat.c b/mm/vmstat.c
> index 4bb13e7..469599c 100644
> --- a/mm/vmstat.c
> +++ b/mm/vmstat.c
> @@ -32,6 +32,91 @@
>
> #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)
> +{
> + const char *str = s;
> +
> + if (strcmp(str, "auto") == 0 || strcmp(str, "Auto") == 0)
> + vm_numa_stats_mode = VM_NUMA_STAT_AUTO_MODE;
> + else if (strcmp(str, "strict") == 0 || strcmp(str, "Strict") == 0)
> + vm_numa_stats_mode = VM_NUMA_STAT_STRICT_MODE;
> + else if (strcmp(str, "coarse") == 0 || strcmp(str, "Coarse") == 0)
> + 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 changes from %s mode 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 mode 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 mode to coarse mode\n",
> + vm_numa_stats_mode_name[oldval]);
> + } else
> + pr_warn("invalid vm_numa_stats_mode:%d\n", vm_numa_stats_mode);
> + }
> +
> + 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 +999,42 @@ unsigned long sum_zone_numa_state(int node,
> 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 +1703,10 @@ static int zoneinfo_show(struct seq_file *m, void *arg)
> {
> 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 +1803,10 @@ static int vmstat_show(struct seq_file *m, void *arg)
>
> 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;
> }
>