OpenVZ / Virtuozzo kernel development (devel@openvz.org)
 help / color / mirror / Atom feed
* [Devel] [PATCH hci-8.1] selftests/cgroup: Pin the hogs of test_cpucg_weight_overprovisioned
@ 2026-11-12  4:23 Eva Kurchatova
  2026-11-12  4:27 ` no-reply
                   ` (2 more replies)
  0 siblings, 3 replies; 4+ messages in thread
From: Eva Kurchatova @ 2026-11-12  4:23 UTC (permalink / raw)
  To: khorenko; +Cc: devel

The test runs one hog thread per CPU in each of three cgroups weighted
1:2:3 and expects their CPU time in the same ratio. That holds when
each CPU runs one hog of every cgroup. On a NUMA machine the load
balancer leaves them less evenly spread, and the heaviest cgroup can
fall short of the tolerance.

Pin the hogs of each cgroup one per CPU: the test checks the weights,
not the load balancer.

Fixes: 6376b22cd0a3 ("cgroup: Add test_cpucg_weight_overprovisioned() testcase")

https://virtuozzo.atlassian.net/browse/VSTOR-152674
Feature: fix selftests

Signed-off-by: Eva Kurchatova <eva.kurchatova@virtuozzo.com>
---
 tools/testing/selftests/cgroup/test_cpu.c | 31 +++++++++++++++++++----
 1 file changed, 26 insertions(+), 5 deletions(-)

diff --git a/tools/testing/selftests/cgroup/test_cpu.c b/tools/testing/selftests/cgroup/test_cpu.c
index a2b50af8e9ee..3385e75eed81 100644
--- a/tools/testing/selftests/cgroup/test_cpu.c
+++ b/tools/testing/selftests/cgroup/test_cpu.c
@@ -6,6 +6,7 @@
 #include <sys/wait.h>
 #include <errno.h>
 #include <pthread.h>
+#include <sched.h>
 #include <stdio.h>
 #include <time.h>
 #include <unistd.h>
@@ -30,6 +31,7 @@ struct cpu_hog_func_param {
 	int nprocs;
 	struct timespec ts;
 	enum hog_clock_type clock_type;
+	bool pin;
 };
 
 /*
@@ -136,7 +138,13 @@ static int hog_cpus_timed(const char *cgroup, void *arg)
 	struct timespec ts_run = param->ts;
 	struct timespec ts_remaining = ts_run;
 	struct timespec ts_start;
-	int i, ret;
+	pthread_attr_t attr;
+	cpu_set_t cpus, one;
+	int i, ret, cpu = -1;
+	/* Unpinned, as before, where the mask does not fit in a cpu_set_t */
+	bool pin = param->pin && !sched_getaffinity(0, sizeof(cpus), &cpus);
+
+	pthread_attr_init(&attr);
 
 	ret = clock_gettime(CLOCK_MONOTONIC, &ts_start);
 	if (ret != 0)
@@ -145,7 +153,18 @@ static int hog_cpus_timed(const char *cgroup, void *arg)
 	for (i = 0; i < param->nprocs; i++) {
 		pthread_t tid;
 
-		ret = pthread_create(&tid, NULL, &hog_cpu_thread_func, NULL);
+		/* One hog per allowed CPU, in turn */
+		if (pin) {
+			do {
+				cpu = (cpu + 1) % CPU_SETSIZE;
+			} while (!CPU_ISSET(cpu, &cpus));
+			CPU_ZERO(&one);
+			CPU_SET(cpu, &one);
+			ret = pthread_attr_setaffinity_np(&attr, sizeof(one), &one);
+			if (ret != 0)
+				return ret;
+		}
+		ret = pthread_create(&tid, &attr, &hog_cpu_thread_func, NULL);
 		if (ret != 0)
 			return ret;
 	}
@@ -372,7 +391,8 @@ run_cpucg_weight_test(
 	return ret;
 }
 
-static pid_t weight_hog_ncpus(const struct cpu_hogger *child, int ncpus)
+static pid_t weight_hog_ncpus(const struct cpu_hogger *child, int ncpus,
+			      bool pin)
 {
 	long usage_seconds = 10;
 	struct cpu_hog_func_param param = {
@@ -382,13 +402,14 @@ static pid_t weight_hog_ncpus(const struct cpu_hogger *child, int ncpus)
 			.tv_nsec = 0,
 		},
 		.clock_type = CPU_HOG_CLOCK_WALL,
+		.pin = pin,
 	};
 	return cg_run_nowait(child->cgroup, hog_cpus_timed, (void *)&param);
 }
 
 static pid_t weight_hog_all_cpus(const struct cpu_hogger *child)
 {
-	return weight_hog_ncpus(child, get_nprocs());
+	return weight_hog_ncpus(child, get_nprocs(), true);
 }
 
 static int
@@ -434,7 +455,7 @@ static int test_cpucg_weight_overprovisioned(const char *root)
 
 static pid_t weight_hog_one_cpu(const struct cpu_hogger *child)
 {
-	return weight_hog_ncpus(child, 1);
+	return weight_hog_ncpus(child, 1, false);
 }
 
 static int
-- 
2.56.0

_______________________________________________
Devel mailing list
Devel@openvz.org
https://lists.openvz.org/mailman/listinfo/devel

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [Devel] [PATCH hci-8.1] selftests/cgroup: Pin the hogs of test_cpucg_weight_overprovisioned
  2026-11-12  4:23 [Devel] [PATCH hci-8.1] selftests/cgroup: Pin the hogs of test_cpucg_weight_overprovisioned Eva Kurchatova
@ 2026-11-12  4:27 ` no-reply
  2026-11-12 11:49 ` Konstantin Khorenko
  2026-11-12 11:49 ` Konstantin Khorenko
  2 siblings, 0 replies; 4+ messages in thread
From: no-reply @ 2026-11-12  4:27 UTC (permalink / raw)
  To: Eva Kurchatova, khorenko; +Cc: devel

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] selftests/cgroup: hog_cpus_timed() never calls pthread_attr_destroy()
--

commit 3e0ad2c93c22b3e844f853e154c385c491c1c5ca
Author: Eva Kurchatova <eva.kurchatova@virtuozzo.com>

selftests/cgroup: Pin the hogs of test_cpucg_weight_overprovisioned

This patch pins each hog thread in test_cpucg_weight_overprovisioned() to
one CPU. The threads of each cgroup are spread one per allowed CPU, so the
test checks the cgroup weights and not the load balancer's placement.

> diff --git a/tools/testing/selftests/cgroup/test_cpu.c b/tools/testing/selftests/cgroup/test_cpu.c
> --- a/tools/testing/selftests/cgroup/test_cpu.c
> +++ b/tools/testing/selftests/cgroup/test_cpu.c

[ ... ]

> @@ -136,7 +138,13 @@ static int hog_cpus_timed(const char *cgroup, void *arg)
>  	struct timespec ts_run = param->ts;
>  	struct timespec ts_remaining = ts_run;
>  	struct timespec ts_start;
> -	int i, ret;
> +	pthread_attr_t attr;
> +	cpu_set_t cpus, one;
> +	int i, ret, cpu = -1;
> +	/* Unpinned, as before, where the mask does not fit in a cpu_set_t */
> +	bool pin = param->pin && !sched_getaffinity(0, sizeof(cpus), &cpus);
> +
> +	pthread_attr_init(&attr);

[Severity: Low]
Is the pthread_attr_t initialized here ever released?  hog_cpus_timed()
calls pthread_attr_init(&attr) unconditionally, but none of its exit paths
call pthread_attr_destroy(&attr).

This covers the final "return 0", the early return after a failed
pthread_attr_setaffinity_np(), the return after a failed pthread_create(),
and the clock_gettime()/nanosleep() failure returns.

With pin set, glibc's pthread_attr_setaffinity_np() allocates a copy of the
cpuset inside attr, and pthread_attr_destroy() is what frees it.

The return value of pthread_attr_init() is also not checked.

The impact is small, since hog_cpus_timed() runs in a short-lived child
forked by cg_run_nowait() and the memory is reclaimed on exit.  Would it be
worth adding a pthread_attr_destroy(&attr) on the exit paths?

> @@ -145,7 +153,18 @@ static int hog_cpus_timed(const char *cgroup, void *arg)
>  	for (i = 0; i < param->nprocs; i++) {
>  		pthread_t tid;
>  
> -		ret = pthread_create(&tid, NULL, &hog_cpu_thread_func, NULL);
> +		/* One hog per allowed CPU, in turn */
> +		if (pin) {
> +			do {
> +				cpu = (cpu + 1) % CPU_SETSIZE;
> +			} while (!CPU_ISSET(cpu, &cpus));
> +			CPU_ZERO(&one);
> +			CPU_SET(cpu, &one);
> +			ret = pthread_attr_setaffinity_np(&attr, sizeof(one), &one);
> +			if (ret != 0)
> +				return ret;
> +		}
> +		ret = pthread_create(&tid, &attr, &hog_cpu_thread_func, NULL);
>  		if (ret != 0)
>  			return ret;
>  	}

[ ... ]

-- 
Sashiko AI review · https://sashiko.vzint.dev/#/patchset/20261112042325.399633-1-eva.kurchatova@virtuozzo.com?part=1

_______________________________________________
Devel mailing list
Devel@openvz.org
https://lists.openvz.org/mailman/listinfo/devel

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [Devel] [PATCH hci-8.1] selftests/cgroup: Pin the hogs of test_cpucg_weight_overprovisioned
  2026-11-12  4:23 [Devel] [PATCH hci-8.1] selftests/cgroup: Pin the hogs of test_cpucg_weight_overprovisioned Eva Kurchatova
  2026-11-12  4:27 ` no-reply
@ 2026-11-12 11:49 ` Konstantin Khorenko
  2026-11-12 11:49 ` Konstantin Khorenko
  2 siblings, 0 replies; 4+ messages in thread
From: Konstantin Khorenko @ 2026-11-12 11:49 UTC (permalink / raw)
  To: Eva Kurchatova; +Cc: devel

Applying with these fixes on my side:
- subject: "Pin the hogs" -> "pin the hogs", lowercase after the
  prefix.

Asks, not needed for this patch:
- Send it to mainline too, see the task below.
- Say whether test_cpucg_nested_weight_overprovisioned passes on the
  same NUMA host; if it does not, pin its hogs in a separate patch.

On Thu, 12 Nov 2026 06:23:20 +0200, Eva Kurchatova wrote:
> Pin the hogs of each cgroup one per CPU: the test checks the weights,
> not the load balancer.

This needs to go to mainline too, I created a task for that:
https://virtuozzo.atlassian.net/browse/VSTOR-154829

Mainline v7.4-rc1 has the same unpinned hog_cpus_timed(), so the test
fails there on the same hosts. For the mainline version, take Sashiko's
pthread_attr_destroy() remark: init the attr only when pinning, pass
NULL otherwise and destroy it after the loop. Upstream reviewers will
ask for it.

> -	return weight_hog_ncpus(child, get_nprocs());
> +	return weight_hog_ncpus(child, get_nprocs(), true);

run_cpucg_nested_weight_test() with overprovisioned set starts
get_nprocs() unpinned hogs in each of its three leaves through the same
hog_cpus_timed(), so the reasoning of this patch applies to it as well.
Its tolerance is wider in practice, which may be why it has not failed
yet. ".pin = overprovisioned" in its param would cover it; the
underprovisioned cases must stay unpinned, or all hogs land on the
first CPUs of the mask.

-- 
Konstantin Khorenko,
Virtuozzo Linux Kernel Team

_______________________________________________
Devel mailing list
Devel@openvz.org
https://lists.openvz.org/mailman/listinfo/devel

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [Devel] [PATCH hci-8.1] selftests/cgroup: Pin the hogs of test_cpucg_weight_overprovisioned
  2026-11-12  4:23 [Devel] [PATCH hci-8.1] selftests/cgroup: Pin the hogs of test_cpucg_weight_overprovisioned Eva Kurchatova
  2026-11-12  4:27 ` no-reply
  2026-11-12 11:49 ` Konstantin Khorenko
@ 2026-11-12 11:49 ` Konstantin Khorenko
  2 siblings, 0 replies; 4+ messages in thread
From: Konstantin Khorenko @ 2026-11-12 11:49 UTC (permalink / raw)
  To: Eva Kurchatova; +Cc: devel

On Thu, 12 Nov 2026 06:23:20 +0200, Eva Kurchatova wrote:
> selftests/cgroup: Pin the hogs of test_cpucg_weight_overprovisioned

Applied to branch-rh10-6.12.0-211.52.1.22.x.vz10-ovz, will be released in rh10-6.12.0-211.52.1.22.21.vz10, thanks!

[1/1] selftests/cgroup: Pin the hogs of test_cpucg_weight_overprovisioned
      https://bitbucket.org/virtuozzocore/vzkernel.vzs/commits/a56b30724f24

-- 
Konstantin Khorenko,
Virtuozzo Linux Kernel Team


_______________________________________________
Devel mailing list
Devel@openvz.org
https://lists.openvz.org/mailman/listinfo/devel

^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2026-11-12 11:50 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-12  4:23 [Devel] [PATCH hci-8.1] selftests/cgroup: Pin the hogs of test_cpucg_weight_overprovisioned Eva Kurchatova
2026-11-12  4:27 ` no-reply
2026-11-12 11:49 ` Konstantin Khorenko
2026-11-12 11:49 ` Konstantin Khorenko

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox