Skip to contentSkip to conversation

mm/damon/core: Avoid fake load reports due to uninterruptible sleeps

PATCH In review SeongJae Park 2 participants 3 messages Nov 24, 2021 – Nov 25, 2021

Subsystems (3)
Stable branch Memory management DAMON
lore
Share & tools
lore

SeongJae Park

author maintainer
#0 lore raw

PATCH 1/1 mm/damon/core: Avoid fake load reports due to uninterruptible sleeps

Because DAMON sleeps in uninterruptible mode, /proc/loadavg reports fake load while DAMON is turned on, though it is doing nothing. This can confuse users[1]. To avoid the case, this commit makes DAMON sleeps in idle mode.

[1] https://lore.kernel.org/all/[email protected]/

Fixes 2224d8485492 ("mm: introduce Data Access MONitor (DAMON)")
Reported-by Oleksandr Natalenko <[email protected]>
Signed-off-by SeongJae Park <[email protected]>
Cc <[email protected]> # 5.15.x

---
I think this needs to be applied on v5.15.y, but this cannot cleanly applied there as is. I will back-port this on v5.15.y and post later once this is merged in the mainline.

 mm/damon/core.c | 21 ++++++++++++++++++---
 1 file changed, 18 insertions(+), 3 deletions(-)
mm/damon/core.c +18 -3
index daacd9536c7c..7813f47aadc9 100644
@@ -12,6 +12,8 @@
 #include <linux/kthread.h>
 #include <linux/mm.h>
 #include <linux/random.h>
+#include <linux/sched.h>
+#include <linux/sched/debug.h>
 #include <linux/slab.h>
 #include <linux/string.h>
 

@@ -976,12 +978,25 @@ static unsigned long damos_wmark_wait_us(struct damos *scheme)
 	return 0;
 }
 
+/* sleep for @usecs in idle mode */
+static void __sched damon_usleep_idle(unsigned long usecs)
+{
+	ktime_t exp = ktime_add_us(ktime_get(), usecs);
+	u64 delta = usecs * NSEC_PER_USEC / 100;	/* allow 1% error */
+
+	for (;;) {
+		__set_current_state(TASK_IDLE);
+		if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
+			break;
+	}
+}
+
 static void kdamond_usleep(unsigned long usecs)
 {
 	if (usecs > 100 * 1000)
-		schedule_timeout_interruptible(usecs_to_jiffies(usecs));
+		schedule_timeout_idle(usecs_to_jiffies(usecs));
 	else
-		usleep_range(usecs, usecs + 1);
+		damon_usleep_idle(usecs);
 }
 
 /* Returns negative error code if it's not activated but should return */

@@ -1036,7 +1051,7 @@ static int kdamond_fn(void *data)
 				ctx->callback.after_sampling(ctx))
 			done = true;
 
-		usleep_range(ctx->sample_interval, ctx->sample_interval + 1);
+		kdamond_usleep(ctx->sample_interval);
 
 		if (ctx->primitive.check_accesses)
 			max_nr_accesses = ctx->primitive.check_accesses(ctx);
Signature · 1 line
2.17.1
maintainer
#1 lore raw
Quoted 22 lines · On Wed, 24 Nov 2021 14:52:19 +0000 SeongJae Park <[email protected]> wrote:

Because DAMON sleeps in uninterruptible mode, /proc/loadavg reports fake load while DAMON is turned on, though it is doing nothing. This can confuse users[1]. To avoid the case, this commit makes DAMON sleeps in idle mode.

mm/damon/core.c +4 -0
@@ -12,6 +12,8 @@
 #include <linux/kthread.h>
 #include <linux/mm.h>
 #include <linux/random.h>
+#include <linux/sched.h>
+#include <linux/sched/debug.h>
 #include <linux/slab.h>
 #include <linux/string.h>
 

@@ -976,12 +978,25 @@ static unsigned long damos_wmark_wait_us(struct damos *scheme)
 	return 0;
 }
 
+/* sleep for @usecs in idle mode */
+static void __sched damon_usleep_idle(unsigned long usecs)

mm/damon/core.c static unsigned long damos_wmark_wait_us(struct damos *scheme) · line 992

Quoted 10 lines
mm/damon/core.c +10 -0
+{
+	ktime_t exp = ktime_add_us(ktime_get(), usecs);
+	u64 delta = usecs * NSEC_PER_USEC / 100;	/* allow 1% error */
+
+	for (;;) {
+		__set_current_state(TASK_IDLE);
+		if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
+			break;
+	}
+}

Let's not copy-n-paste usleep_range() into damon code?

A new usleep_idle_range() in kernel/time/timer.c seems like a worthwhile addition. Perhaps usleep_idle_range() and usleep_range() can be static inline wrappers against a new, more general function in kernel/time/timer.c - usleep_range_state(min, max, state)?

author maintainer
#2 lore raw
Quoted 37 lines · On Wed, 24 Nov 2021 13:42:10 -0800 Andrew Morton <[email protected]>…
Quoted 22 lines · On Wed, 24 Nov 2021 14:52:19 +0000 SeongJae Park <[email protected]> wrote:

Because DAMON sleeps in uninterruptible mode, /proc/loadavg reports fake load while DAMON is turned on, though it is doing nothing. This can confuse users[1]. To avoid the case, this commit makes DAMON sleeps in idle mode.

mm/damon/core.c +4 -0
@@ -12,6 +12,8 @@
 #include <linux/kthread.h>
 #include <linux/mm.h>
 #include <linux/random.h>
+#include <linux/sched.h>
+#include <linux/sched/debug.h>
 #include <linux/slab.h>
 #include <linux/string.h>
 

@@ -976,12 +978,25 @@ static unsigned long damos_wmark_wait_us(struct damos *scheme)
 	return 0;
 }
 
+/* sleep for @usecs in idle mode */
+static void __sched damon_usleep_idle(unsigned long usecs)

mm/damon/core.c static unsigned long damos_wmark_wait_us(struct damos *scheme) · line 992

Quoted 10 lines
mm/damon/core.c +10 -0
+{
+	ktime_t exp = ktime_add_us(ktime_get(), usecs);
+	u64 delta = usecs * NSEC_PER_USEC / 100;	/* allow 1% error */
+
+	for (;;) {
+		__set_current_state(TASK_IDLE);
+		if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
+			break;
+	}
+}

Let's not copy-n-paste usleep_range() into damon code?

Quoted 4 lines

A new usleep_idle_range() in kernel/time/timer.c seems like a worthwhile addition. Perhaps usleep_idle_range() and usleep_range() can be static inline wrappers against a new, more general function in kernel/time/timer.c - usleep_range_state(min, max, state)?

Makes sense, agreed. I will post next version soon.

Thanks,
SJ