[PATCH v6] selftests/cgroup: add PSI pressure trigger and validation tests

[PATCH v6] selftests/cgroup: add PSI pressure trigger and validation tests

[Tao Cui]

From: Tao Cui <cuitao@kylinos.cn>

The cgroup selftests have no PSI coverage. Add test_psi.c: per-resource
trigger smoke tests (one trigger per fd, IRQ full-only), a
cgroup.pressure hide/show toggle test, and two CPU-pressure trigger
tests using over-subscription: one expects the trigger to fire at a
tiny threshold, the other expects it not to fire at a whole-window
threshold. Skips when PSI is disabled or a resource is absent.

Signed-off-by: Tao Cui <cuitao@kylinos.cn>

---
Changes since v5 (Michal Koutny review):
- Terminate trigger writes with a newline instead of a NUL. The parser
  (psi_write()) overwrites the last byte of the write with a NUL, so a
  plain strlen()-sized write silently truncates the last digit for
  privileged users (e.g. a 2000000us window becomes 200000us) and
  fails with EINVAL for unprivileged ones. The newline keeps the
  payload intact and drops the reliance on the undocumented NUL
  convention.
- Keep pressure_open() silent: the SKIP message already reports why a
  resource is unavailable and strace accompanies these tests anyway.
  This replaces the ksft_perror() added on review of v4, which
  duplicated the SKIP reason.
- Treat a test cgroup creation failure in FIXTURE_SETUP() as a failure
  instead of a SKIP.
- Drop the unprivileged-minimum remark from the fire test comment and
  the duplicate sysconf() error report.
- Add cgroup_trigger_no_fire, the false-trigger side of the firing
  test: with the threshold equal to the whole window, ncpus+1 hogs must
  not fire it. Under this oversubscription the measured cpu.pressure
  "some" stays in the 2-17% range on 2- and 16-CPU machines, so the
  margin is comfortable.

Changes since v4 (Suren Baghdasaryan review):
- Report pressure_open() errors with ksft_perror() instead of a raw
  fprintf.
- Create the test cgroup in FIXTURE_SETUP() so that setup and teardown
  are symmetric, and drop the per-test creation and the NULL guard in
  the teardown.
- Factor the identical bodies of proc_trigger_io/memory/cpu into
  test_psi_write(); the helper takes _metadata so that its ASSERTs
  attribute to the calling test, following the seccomp selftests
  precedent.
- Drop the now-unneeded NULL init of self->cg and the stale stdbool.h
  include left over from the restructuring.

Changes since v3 (Suren Baghdasaryan review):
- Convert to the kselftest harness: each case is a TEST_F(psi, ...)
  with the cgroup root/PSI availability checks in FIXTURE_SETUP() and
  the teardown (kill hogs, destroy cgroup) in FIXTURE_TEARDOWN(),
  which also removes the "ret"/"created" bookkeeping.
- A trigger that does not fire within the poll timeout is now a FAIL
  instead of a SKIP: ncpus+1 hogs with a 1usec threshold must stall,
  so a timeout indicates a real problem.
- Treat a poll() timeout and a poll() error uniformly via ASSERT.
- Check sysconf(_SC_NPROCESSORS_ONLN) only for -1 and report
  strerror(errno); declare variables one per line; for(;;) {}.
- Make hog_cpu() die with the runner via PR_SET_PDEATHSIG so an
  interrupted run does not leave orphaned hogs pinning every CPU.

Changes since v2 (Suren Baghdasaryan, Michal Koutny review):
- Restructure the trigger test into per-resource cases (io, memory, cpu,
  irq) so a failure points at the specific resource; irq is skipped when
  /proc/pressure/irq is absent.
- Spawn the CPU hogs with cg_run_nowait() instead of open-coding fork(),
  and arm the trigger with a 2s window so unprivileged users can set it.
- Address the remaining review comments on cleanup and robustness:
  guard teardown with a "created" flag, use cg_read_strcmp() instead of
  atoi(), report strerror() on errors, and fix the unused-parameter and
  sign-compare nits.

Changes since v1 (Michal Koutny, sashiko review):
- Keep trigger tests smoke-level; switch the firing test from memory to
  CPU pressure; drop churn_memory().
- Keep the runner out of the cgroup; add PSI/IRQ skip-guards and a
  .gitignore entry.

v1: https://lore.kernel.org/all/20260724025826.504586-1-cui.tao@linux.dev/
v2: https://lore.kernel.org/all/20260728083742.2359320-1-cui.tao@linux.dev/
v3: https://lore.kernel.org/all/20260813133723.1663605-1-cui.tao@linux.dev/
v4: https://lore.kernel.org/all/20260824085913.546741-1-cui.tao@linux.dev/
v5: https://lore.kernel.org/all/20260902040725.877155-1-cui.tao@linux.dev/
---
 tools/testing/selftests/cgroup/.gitignore |   1 +
 tools/testing/selftests/cgroup/Makefile   |   2 +
 tools/testing/selftests/cgroup/config     |   1 +
 tools/testing/selftests/cgroup/test_psi.c | 219 ++++++++++++++++++++++
 4 files changed, 223 insertions(+)
 create mode 100644 tools/testing/selftests/cgroup/test_psi.c

diff --git a/tools/testing/selftests/cgroup/.gitignore b/tools/testing/selftests/cgroup/.gitignore
index 952e4448bf070..ce2b907c57ea3 100644
--- a/tools/testing/selftests/cgroup/.gitignore
+++ b/tools/testing/selftests/cgroup/.gitignore
@@ -8,5 +8,6 @@ test_kill
 test_kmem
 test_memcontrol
 test_pids
+test_psi
 test_zswap
 wait_inotify
diff --git a/tools/testing/selftests/cgroup/Makefile b/tools/testing/selftests/cgroup/Makefile
index e01584c2189ac..a8c69e37332a4 100644
--- a/tools/testing/selftests/cgroup/Makefile
+++ b/tools/testing/selftests/cgroup/Makefile
@@ -16,6 +16,7 @@ TEST_GEN_PROGS += test_kill
 TEST_GEN_PROGS += test_kmem
 TEST_GEN_PROGS += test_memcontrol
 TEST_GEN_PROGS += test_pids
+TEST_GEN_PROGS += test_psi
 TEST_GEN_PROGS += test_zswap
 
 LOCAL_HDRS += $(selfdir)/clone3/clone3_selftests.h $(selfdir)/pidfd/pidfd.h
@@ -32,4 +33,5 @@ $(OUTPUT)/test_kill: $(LIBCGROUP_O)
 $(OUTPUT)/test_kmem: $(LIBCGROUP_O)
 $(OUTPUT)/test_memcontrol: $(LIBCGROUP_O)
 $(OUTPUT)/test_pids: $(LIBCGROUP_O)
+$(OUTPUT)/test_psi: $(LIBCGROUP_O)
 $(OUTPUT)/test_zswap: $(LIBCGROUP_O)
diff --git a/tools/testing/selftests/cgroup/config b/tools/testing/selftests/cgroup/config
index 39f979690dd3b..8a3ef479e83d3 100644
--- a/tools/testing/selftests/cgroup/config
+++ b/tools/testing/selftests/cgroup/config
@@ -4,3 +4,4 @@ CONFIG_CGROUP_FREEZER=y
 CONFIG_CGROUP_SCHED=y
 CONFIG_MEMCG=y
 CONFIG_PAGE_COUNTER=y
+CONFIG_PSI=y
diff --git a/tools/testing/selftests/cgroup/test_psi.c b/tools/testing/selftests/cgroup/test_psi.c
new file mode 100644
index 0000000000000..95dcb8a9f1843
--- /dev/null
+++ b/tools/testing/selftests/cgroup/test_psi.c
@@ -0,0 +1,219 @@
+// SPDX-License-Identifier: GPL-2.0
+#define _GNU_SOURCE
+#include <errno.h>
+#include <fcntl.h>
+#include <poll.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+#include <sys/prctl.h>
+#include <linux/limits.h>
+
+#include "../kselftest_harness.h"
+#include "cgroup_util.h"
+
+#define PSI_POLL_TIMEOUT_MS	5000
+
+/*
+ * The kernel parser overwrites the last byte of a trigger write with a
+ * NUL, so terminate with a newline to keep the payload intact.
+ */
+static ssize_t write_trigger(int fd, const char *trigger)
+{
+	char buf[64];
+
+	snprintf(buf, sizeof(buf), "%s\n", trigger);
+
+	return write(fd, buf, strlen(buf));
+}
+
+static int pressure_open(const char *resource)
+{
+	char path[PATH_MAX];
+
+	snprintf(path, sizeof(path), "/proc/pressure/%s", resource);
+	return open(path, O_RDWR);
+}
+
+FIXTURE(psi)
+{
+	char root[PATH_MAX];
+	char *cg;
+};
+
+FIXTURE_SETUP(psi)
+{
+	int psi_fd;
+
+	if (cg_find_unified_root(self->root, sizeof(self->root), NULL))
+		SKIP(return, "cgroup v2 isn't mounted");
+
+	/* PSI must be enabled (CONFIG_PSI=y, not disabled on the cmdline). */
+	psi_fd = open("/proc/pressure/memory", O_RDONLY);
+	if (psi_fd < 0)
+		SKIP(return, "PSI unavailable (CONFIG_PSI=n or psi=0)");
+	close(psi_fd);
+
+	self->cg = cg_name(self->root, "psi_trigger_test");
+	ASSERT_NE(NULL, self->cg);
+	ASSERT_EQ(0, cg_create(self->cg));
+}
+
+FIXTURE_TEARDOWN(psi)
+{
+	cg_killall(self->cg);
+	cg_destroy(self->cg);
+	free(self->cg);
+}
+
+/*
+ * /proc/pressure/<resource> accepts exactly one trigger per file
+ * descriptor. Verify that a "some" trigger arms and that a second
+ * trigger on the same fd is rejected with EBUSY.
+ */
+static void test_psi_write(struct __test_metadata *_metadata,
+			   const char *filename)
+{
+	int fd;
+
+	fd = pressure_open(filename);
+	ASSERT_GE(fd, 0);
+	ASSERT_GT(write_trigger(fd, "some 150000 2000000"), 0);
+	ASSERT_EQ(-1, write_trigger(fd, "full 150000 2000000"));
+	ASSERT_EQ(EBUSY, errno);
+	close(fd);
+}
+
+TEST_F(psi, proc_trigger_io)
+{
+	test_psi_write(_metadata, "io");
+}
+
+TEST_F(psi, proc_trigger_memory)
+{
+	test_psi_write(_metadata, "memory");
+}
+
+TEST_F(psi, proc_trigger_cpu)
+{
+	test_psi_write(_metadata, "cpu");
+}
+
+/*
+ * irq only tracks "full", so a "some" trigger must be rejected while a
+ * "full" trigger arms. irq is optional -- it only exists with IRQ-time
+ * accounting -- so a missing /proc/pressure/irq is SKIP, not FAIL.
+ */
+TEST_F(psi, proc_trigger_irq)
+{
+	int fd;
+
+	fd = pressure_open("irq");
+	if (fd < 0)
+		SKIP(return, "/proc/pressure/irq unavailable");
+
+	ASSERT_EQ(-1, write_trigger(fd, "some 150000 2000000"));
+	ASSERT_GT(write_trigger(fd, "full 150000 2000000"), 0);
+	close(fd);
+}
+
+/*
+ * cgroup.pressure gates visibility of the per-resource *.pressure files
+ * inside a cgroup: writing 0 hides them, writing 1 shows them again.
+ * Drive one hide/show cycle and check that memory.pressure appears and
+ * disappears along with it.
+ */
+TEST_F(psi, cgroup_pressure_toggle)
+{
+	char buf[BUF_SIZE];
+
+	ASSERT_EQ(0, cg_write(self->cg, "cgroup.pressure", "0"));
+	ASSERT_EQ(0, cg_read_strcmp(self->cg, "cgroup.pressure", "0\n"));
+	ASSERT_LT(cg_read(self->cg, "memory.pressure", buf, sizeof(buf)), 0);
+
+	ASSERT_EQ(0, cg_write(self->cg, "cgroup.pressure", "1"));
+	ASSERT_EQ(0, cg_read_strcmp(self->cg, "cgroup.pressure", "1\n"));
+	ASSERT_GE(cg_read(self->cg, "memory.pressure", buf, sizeof(buf)), 0);
+}
+
+/*
+ * A child that burns CPU forever; stopped by cg_killall() on teardown.
+ * It also dies with the runner, so an interrupted run (e.g. Ctrl-C
+ * during poll()) does not leave orphaned hogs pinning every CPU.
+ */
+static int hog_cpu(const char *cgroup, void *arg)
+{
+	prctl(PR_SET_PDEATHSIG, SIGKILL);
+	for (;;) {}
+	return 0;
+}
+
+/*
+ * Arm a "some" trigger on a cgroup's cpu.pressure, oversubscribe the
+ * cgroup with more spinning hogs than there are CPUs, and check that the
+ * trigger fires once the cgroup stalls on CPU.
+ */
+TEST_F(psi, cgroup_trigger_fire)
+{
+	char *cpupress;
+	struct pollfd pfd = { .events = POLLPRI };
+	long ncpus;
+	int fd;
+	int i;
+
+	cpupress = cg_control(self->cg, "cpu.pressure");
+	ASSERT_NE(NULL, cpupress);
+	fd = open(cpupress, O_RDWR);
+	free(cpupress);
+	ASSERT_GE(fd, 0);
+	pfd.fd = fd;
+
+	/* 1usec threshold over a 2s window: any CPU stall fires it. */
+	ASSERT_GT(write_trigger(fd, "some 1 2000000"), 0);
+
+	ncpus = sysconf(_SC_NPROCESSORS_ONLN);
+	ASSERT_NE(-1, ncpus);
+
+	/* ncpus+1 hogs guarantee CPU contention inside the cgroup. */
+	for (i = 0; i < ncpus + 1; i++)
+		ASSERT_GE(cg_run_nowait(self->cg, hog_cpu, NULL), 0);
+
+	ASSERT_EQ(1, poll(&pfd, 1, PSI_POLL_TIMEOUT_MS));
+	ASSERT_NE(0, pfd.revents & POLLPRI);
+	close(fd);
+}
+
+/*
+ * The flip side of cgroup_trigger_fire: with the threshold equal to the
+ * whole window, even ncpus+1 hogs must not accumulate enough stall to
+ * fire the trigger.
+ */
+TEST_F(psi, cgroup_trigger_no_fire)
+{
+	char *cpupress;
+	struct pollfd pfd = { .events = POLLPRI };
+	long ncpus;
+	int fd;
+	int i;
+
+	cpupress = cg_control(self->cg, "cpu.pressure");
+	ASSERT_NE(NULL, cpupress);
+	fd = open(cpupress, O_RDWR);
+	free(cpupress);
+	ASSERT_GE(fd, 0);
+	pfd.fd = fd;
+
+	ASSERT_GT(write_trigger(fd, "some 2000000 2000000"), 0);
+
+	ncpus = sysconf(_SC_NPROCESSORS_ONLN);
+	ASSERT_NE(-1, ncpus);
+
+	for (i = 0; i < ncpus + 1; i++)
+		ASSERT_GE(cg_run_nowait(self->cg, hog_cpu, NULL), 0);
+
+	ASSERT_EQ(0, poll(&pfd, 1, PSI_POLL_TIMEOUT_MS));
+	close(fd);
+}
+
+TEST_HARNESS_MAIN
-- 
2.43.0

Re: [PATCH v6] selftests/cgroup: add PSI pressure trigger and validation tests

[Suren Baghdasaryan]

On Thu, Sep 3, 2026 at 12:35 AM Tao Cui <cui.tao@linux.dev> wrote:
>
> From: Tao Cui <cuitao@kylinos.cn>
>
> The cgroup selftests have no PSI coverage. Add test_psi.c: per-resource
> trigger smoke tests (one trigger per fd, IRQ full-only), a
> cgroup.pressure hide/show toggle test, and two CPU-pressure trigger
> tests using over-subscription: one expects the trigger to fire at a
> tiny threshold, the other expects it not to fire at a whole-window
> threshold. Skips when PSI is disabled or a resource is absent.
>
> Signed-off-by: Tao Cui <cuitao@kylinos.cn>

Thanks!

Reviewed-by: Suren Baghdasaryan <surenb@google.com>

>
> ---
> Changes since v5 (Michal Koutny review):
> - Terminate trigger writes with a newline instead of a NUL. The parser
>   (psi_write()) overwrites the last byte of the write with a NUL, so a
>   plain strlen()-sized write silently truncates the last digit for
>   privileged users (e.g. a 2000000us window becomes 200000us) and
>   fails with EINVAL for unprivileged ones. The newline keeps the
>   payload intact and drops the reliance on the undocumented NUL
>   convention.
> - Keep pressure_open() silent: the SKIP message already reports why a
>   resource is unavailable and strace accompanies these tests anyway.
>   This replaces the ksft_perror() added on review of v4, which
>   duplicated the SKIP reason.
> - Treat a test cgroup creation failure in FIXTURE_SETUP() as a failure
>   instead of a SKIP.
> - Drop the unprivileged-minimum remark from the fire test comment and
>   the duplicate sysconf() error report.
> - Add cgroup_trigger_no_fire, the false-trigger side of the firing
>   test: with the threshold equal to the whole window, ncpus+1 hogs must
>   not fire it. Under this oversubscription the measured cpu.pressure
>   "some" stays in the 2-17% range on 2- and 16-CPU machines, so the
>   margin is comfortable.
>
> Changes since v4 (Suren Baghdasaryan review):
> - Report pressure_open() errors with ksft_perror() instead of a raw
>   fprintf.
> - Create the test cgroup in FIXTURE_SETUP() so that setup and teardown
>   are symmetric, and drop the per-test creation and the NULL guard in
>   the teardown.
> - Factor the identical bodies of proc_trigger_io/memory/cpu into
>   test_psi_write(); the helper takes _metadata so that its ASSERTs
>   attribute to the calling test, following the seccomp selftests
>   precedent.
> - Drop the now-unneeded NULL init of self->cg and the stale stdbool.h
>   include left over from the restructuring.
>
> Changes since v3 (Suren Baghdasaryan review):
> - Convert to the kselftest harness: each case is a TEST_F(psi, ...)
>   with the cgroup root/PSI availability checks in FIXTURE_SETUP() and
>   the teardown (kill hogs, destroy cgroup) in FIXTURE_TEARDOWN(),
>   which also removes the "ret"/"created" bookkeeping.
> - A trigger that does not fire within the poll timeout is now a FAIL
>   instead of a SKIP: ncpus+1 hogs with a 1usec threshold must stall,
>   so a timeout indicates a real problem.
> - Treat a poll() timeout and a poll() error uniformly via ASSERT.
> - Check sysconf(_SC_NPROCESSORS_ONLN) only for -1 and report
>   strerror(errno); declare variables one per line; for(;;) {}.
> - Make hog_cpu() die with the runner via PR_SET_PDEATHSIG so an
>   interrupted run does not leave orphaned hogs pinning every CPU.
>
> Changes since v2 (Suren Baghdasaryan, Michal Koutny review):
> - Restructure the trigger test into per-resource cases (io, memory, cpu,
>   irq) so a failure points at the specific resource; irq is skipped when
>   /proc/pressure/irq is absent.
> - Spawn the CPU hogs with cg_run_nowait() instead of open-coding fork(),
>   and arm the trigger with a 2s window so unprivileged users can set it.
> - Address the remaining review comments on cleanup and robustness:
>   guard teardown with a "created" flag, use cg_read_strcmp() instead of
>   atoi(), report strerror() on errors, and fix the unused-parameter and
>   sign-compare nits.
>
> Changes since v1 (Michal Koutny, sashiko review):
> - Keep trigger tests smoke-level; switch the firing test from memory to
>   CPU pressure; drop churn_memory().
> - Keep the runner out of the cgroup; add PSI/IRQ skip-guards and a
>   .gitignore entry.
>
> v1: https://lore.kernel.org/all/20260724025826.504586-1-cui.tao@linux.dev/
> v2: https://lore.kernel.org/all/20260728083742.2359320-1-cui.tao@linux.dev/
> v3: https://lore.kernel.org/all/20260813133723.1663605-1-cui.tao@linux.dev/
> v4: https://lore.kernel.org/all/20260824085913.546741-1-cui.tao@linux.dev/
> v5: https://lore.kernel.org/all/20260902040725.877155-1-cui.tao@linux.dev/
> ---
>  tools/testing/selftests/cgroup/.gitignore |   1 +
>  tools/testing/selftests/cgroup/Makefile   |   2 +
>  tools/testing/selftests/cgroup/config     |   1 +
>  tools/testing/selftests/cgroup/test_psi.c | 219 ++++++++++++++++++++++
>  4 files changed, 223 insertions(+)
>  create mode 100644 tools/testing/selftests/cgroup/test_psi.c
>
> diff --git a/tools/testing/selftests/cgroup/.gitignore b/tools/testing/selftests/cgroup/.gitignore
> index 952e4448bf070..ce2b907c57ea3 100644
> --- a/tools/testing/selftests/cgroup/.gitignore
> +++ b/tools/testing/selftests/cgroup/.gitignore
> @@ -8,5 +8,6 @@ test_kill
>  test_kmem
>  test_memcontrol
>  test_pids
> +test_psi
>  test_zswap
>  wait_inotify
> diff --git a/tools/testing/selftests/cgroup/Makefile b/tools/testing/selftests/cgroup/Makefile
> index e01584c2189ac..a8c69e37332a4 100644
> --- a/tools/testing/selftests/cgroup/Makefile
> +++ b/tools/testing/selftests/cgroup/Makefile
> @@ -16,6 +16,7 @@ TEST_GEN_PROGS += test_kill
>  TEST_GEN_PROGS += test_kmem
>  TEST_GEN_PROGS += test_memcontrol
>  TEST_GEN_PROGS += test_pids
> +TEST_GEN_PROGS += test_psi
>  TEST_GEN_PROGS += test_zswap
>
>  LOCAL_HDRS += $(selfdir)/clone3/clone3_selftests.h $(selfdir)/pidfd/pidfd.h
> @@ -32,4 +33,5 @@ $(OUTPUT)/test_kill: $(LIBCGROUP_O)
>  $(OUTPUT)/test_kmem: $(LIBCGROUP_O)
>  $(OUTPUT)/test_memcontrol: $(LIBCGROUP_O)
>  $(OUTPUT)/test_pids: $(LIBCGROUP_O)
> +$(OUTPUT)/test_psi: $(LIBCGROUP_O)
>  $(OUTPUT)/test_zswap: $(LIBCGROUP_O)
> diff --git a/tools/testing/selftests/cgroup/config b/tools/testing/selftests/cgroup/config
> index 39f979690dd3b..8a3ef479e83d3 100644
> --- a/tools/testing/selftests/cgroup/config
> +++ b/tools/testing/selftests/cgroup/config
> @@ -4,3 +4,4 @@ CONFIG_CGROUP_FREEZER=y
>  CONFIG_CGROUP_SCHED=y
>  CONFIG_MEMCG=y
>  CONFIG_PAGE_COUNTER=y
> +CONFIG_PSI=y
> diff --git a/tools/testing/selftests/cgroup/test_psi.c b/tools/testing/selftests/cgroup/test_psi.c
> new file mode 100644
> index 0000000000000..95dcb8a9f1843
> --- /dev/null
> +++ b/tools/testing/selftests/cgroup/test_psi.c
> @@ -0,0 +1,219 @@
> +// SPDX-License-Identifier: GPL-2.0
> +#define _GNU_SOURCE
> +#include <errno.h>
> +#include <fcntl.h>
> +#include <poll.h>
> +#include <stdio.h>
> +#include <stdlib.h>
> +#include <string.h>
> +#include <unistd.h>
> +#include <sys/prctl.h>
> +#include <linux/limits.h>
> +
> +#include "../kselftest_harness.h"
> +#include "cgroup_util.h"
> +
> +#define PSI_POLL_TIMEOUT_MS    5000
> +
> +/*
> + * The kernel parser overwrites the last byte of a trigger write with a
> + * NUL, so terminate with a newline to keep the payload intact.
> + */
> +static ssize_t write_trigger(int fd, const char *trigger)
> +{
> +       char buf[64];
> +
> +       snprintf(buf, sizeof(buf), "%s\n", trigger);
> +
> +       return write(fd, buf, strlen(buf));
> +}
> +
> +static int pressure_open(const char *resource)
> +{
> +       char path[PATH_MAX];
> +
> +       snprintf(path, sizeof(path), "/proc/pressure/%s", resource);
> +       return open(path, O_RDWR);
> +}
> +
> +FIXTURE(psi)
> +{
> +       char root[PATH_MAX];
> +       char *cg;
> +};
> +
> +FIXTURE_SETUP(psi)
> +{
> +       int psi_fd;
> +
> +       if (cg_find_unified_root(self->root, sizeof(self->root), NULL))
> +               SKIP(return, "cgroup v2 isn't mounted");
> +
> +       /* PSI must be enabled (CONFIG_PSI=y, not disabled on the cmdline). */
> +       psi_fd = open("/proc/pressure/memory", O_RDONLY);
> +       if (psi_fd < 0)
> +               SKIP(return, "PSI unavailable (CONFIG_PSI=n or psi=0)");
> +       close(psi_fd);
> +
> +       self->cg = cg_name(self->root, "psi_trigger_test");
> +       ASSERT_NE(NULL, self->cg);
> +       ASSERT_EQ(0, cg_create(self->cg));
> +}
> +
> +FIXTURE_TEARDOWN(psi)
> +{
> +       cg_killall(self->cg);
> +       cg_destroy(self->cg);
> +       free(self->cg);
> +}
> +
> +/*
> + * /proc/pressure/<resource> accepts exactly one trigger per file
> + * descriptor. Verify that a "some" trigger arms and that a second
> + * trigger on the same fd is rejected with EBUSY.
> + */
> +static void test_psi_write(struct __test_metadata *_metadata,
> +                          const char *filename)
> +{
> +       int fd;
> +
> +       fd = pressure_open(filename);
> +       ASSERT_GE(fd, 0);
> +       ASSERT_GT(write_trigger(fd, "some 150000 2000000"), 0);
> +       ASSERT_EQ(-1, write_trigger(fd, "full 150000 2000000"));
> +       ASSERT_EQ(EBUSY, errno);
> +       close(fd);
> +}
> +
> +TEST_F(psi, proc_trigger_io)
> +{
> +       test_psi_write(_metadata, "io");
> +}
> +
> +TEST_F(psi, proc_trigger_memory)
> +{
> +       test_psi_write(_metadata, "memory");
> +}
> +
> +TEST_F(psi, proc_trigger_cpu)
> +{
> +       test_psi_write(_metadata, "cpu");
> +}
> +
> +/*
> + * irq only tracks "full", so a "some" trigger must be rejected while a
> + * "full" trigger arms. irq is optional -- it only exists with IRQ-time
> + * accounting -- so a missing /proc/pressure/irq is SKIP, not FAIL.
> + */
> +TEST_F(psi, proc_trigger_irq)
> +{
> +       int fd;
> +
> +       fd = pressure_open("irq");
> +       if (fd < 0)
> +               SKIP(return, "/proc/pressure/irq unavailable");
> +
> +       ASSERT_EQ(-1, write_trigger(fd, "some 150000 2000000"));
> +       ASSERT_GT(write_trigger(fd, "full 150000 2000000"), 0);
> +       close(fd);
> +}
> +
> +/*
> + * cgroup.pressure gates visibility of the per-resource *.pressure files
> + * inside a cgroup: writing 0 hides them, writing 1 shows them again.
> + * Drive one hide/show cycle and check that memory.pressure appears and
> + * disappears along with it.
> + */
> +TEST_F(psi, cgroup_pressure_toggle)
> +{
> +       char buf[BUF_SIZE];
> +
> +       ASSERT_EQ(0, cg_write(self->cg, "cgroup.pressure", "0"));
> +       ASSERT_EQ(0, cg_read_strcmp(self->cg, "cgroup.pressure", "0\n"));
> +       ASSERT_LT(cg_read(self->cg, "memory.pressure", buf, sizeof(buf)), 0);
> +
> +       ASSERT_EQ(0, cg_write(self->cg, "cgroup.pressure", "1"));
> +       ASSERT_EQ(0, cg_read_strcmp(self->cg, "cgroup.pressure", "1\n"));
> +       ASSERT_GE(cg_read(self->cg, "memory.pressure", buf, sizeof(buf)), 0);
> +}
> +
> +/*
> + * A child that burns CPU forever; stopped by cg_killall() on teardown.
> + * It also dies with the runner, so an interrupted run (e.g. Ctrl-C
> + * during poll()) does not leave orphaned hogs pinning every CPU.

Did you mean to say "orphaned hogs spinning on every CPU"?

> + */
> +static int hog_cpu(const char *cgroup, void *arg)
> +{
> +       prctl(PR_SET_PDEATHSIG, SIGKILL);
> +       for (;;) {}
> +       return 0;
> +}
> +
> +/*
> + * Arm a "some" trigger on a cgroup's cpu.pressure, oversubscribe the
> + * cgroup with more spinning hogs than there are CPUs, and check that the
> + * trigger fires once the cgroup stalls on CPU.
> + */
> +TEST_F(psi, cgroup_trigger_fire)
> +{
> +       char *cpupress;
> +       struct pollfd pfd = { .events = POLLPRI };
> +       long ncpus;
> +       int fd;
> +       int i;
> +
> +       cpupress = cg_control(self->cg, "cpu.pressure");
> +       ASSERT_NE(NULL, cpupress);
> +       fd = open(cpupress, O_RDWR);
> +       free(cpupress);
> +       ASSERT_GE(fd, 0);
> +       pfd.fd = fd;
> +
> +       /* 1usec threshold over a 2s window: any CPU stall fires it. */
> +       ASSERT_GT(write_trigger(fd, "some 1 2000000"), 0);
> +
> +       ncpus = sysconf(_SC_NPROCESSORS_ONLN);
> +       ASSERT_NE(-1, ncpus);
> +
> +       /* ncpus+1 hogs guarantee CPU contention inside the cgroup. */
> +       for (i = 0; i < ncpus + 1; i++)
> +               ASSERT_GE(cg_run_nowait(self->cg, hog_cpu, NULL), 0);
> +
> +       ASSERT_EQ(1, poll(&pfd, 1, PSI_POLL_TIMEOUT_MS));
> +       ASSERT_NE(0, pfd.revents & POLLPRI);
> +       close(fd);
> +}
> +
> +/*
> + * The flip side of cgroup_trigger_fire: with the threshold equal to the
> + * whole window, even ncpus+1 hogs must not accumulate enough stall to
> + * fire the trigger.
> + */
> +TEST_F(psi, cgroup_trigger_no_fire)
> +{
> +       char *cpupress;
> +       struct pollfd pfd = { .events = POLLPRI };
> +       long ncpus;
> +       int fd;
> +       int i;
> +
> +       cpupress = cg_control(self->cg, "cpu.pressure");
> +       ASSERT_NE(NULL, cpupress);
> +       fd = open(cpupress, O_RDWR);
> +       free(cpupress);
> +       ASSERT_GE(fd, 0);
> +       pfd.fd = fd;
> +
> +       ASSERT_GT(write_trigger(fd, "some 2000000 2000000"), 0);
> +
> +       ncpus = sysconf(_SC_NPROCESSORS_ONLN);
> +       ASSERT_NE(-1, ncpus);
> +
> +       for (i = 0; i < ncpus + 1; i++)
> +               ASSERT_GE(cg_run_nowait(self->cg, hog_cpu, NULL), 0);
> +
> +       ASSERT_EQ(0, poll(&pfd, 1, PSI_POLL_TIMEOUT_MS));
> +       close(fd);
> +}
> +
> +TEST_HARNESS_MAIN
> --
> 2.43.0
>

Re: [PATCH v6] selftests/cgroup: add PSI pressure trigger and validation tests

[Tao Cui]

Hello Suren,

在 2026/9/25 14:56, Suren Baghdasaryan 写道:
> On Thu, Sep 3, 2026 at 12:35 AM Tao Cui <cui.tao@linux.dev> wrote:
>>
>> From: Tao Cui <cuitao@kylinos.cn>
>>
>> The cgroup selftests have no PSI coverage. Add test_psi.c: per-resource
>> trigger smoke tests (one trigger per fd, IRQ full-only), a
>> cgroup.pressure hide/show toggle test, and two CPU-pressure trigger
>> tests using over-subscription: one expects the trigger to fire at a
>> tiny threshold, the other expects it not to fire at a whole-window
>> threshold. Skips when PSI is disabled or a resource is absent.
>>
>> Signed-off-by: Tao Cui <cuitao@kylinos.cn>
> 
> Thanks!
> 
> Reviewed-by: Suren Baghdasaryan <surenb@google.com>
> 

Thank you!

>>
>> ---
>> Changes since v5 (Michal Koutny review):
>> - Terminate trigger writes with a newline instead of a NUL. The parser
>>   (psi_write()) overwrites the last byte of the write with a NUL, so a
>>   plain strlen()-sized write silently truncates the last digit for
>>   privileged users (e.g. a 2000000us window becomes 200000us) and
>>   fails with EINVAL for unprivileged ones. The newline keeps the
>>   payload intact and drops the reliance on the undocumented NUL
>>   convention.
>> - Keep pressure_open() silent: the SKIP message already reports why a
>>   resource is unavailable and strace accompanies these tests anyway.
>>   This replaces the ksft_perror() added on review of v4, which
>>   duplicated the SKIP reason.
>> - Treat a test cgroup creation failure in FIXTURE_SETUP() as a failure
>>   instead of a SKIP.
>> - Drop the unprivileged-minimum remark from the fire test comment and
>>   the duplicate sysconf() error report.
>> - Add cgroup_trigger_no_fire, the false-trigger side of the firing
>>   test: with the threshold equal to the whole window, ncpus+1 hogs must
>>   not fire it. Under this oversubscription the measured cpu.pressure
>>   "some" stays in the 2-17% range on 2- and 16-CPU machines, so the
>>   margin is comfortable.
>>
>> Changes since v4 (Suren Baghdasaryan review):
>> - Report pressure_open() errors with ksft_perror() instead of a raw
>>   fprintf.
>> - Create the test cgroup in FIXTURE_SETUP() so that setup and teardown
>>   are symmetric, and drop the per-test creation and the NULL guard in
>>   the teardown.
>> - Factor the identical bodies of proc_trigger_io/memory/cpu into
>>   test_psi_write(); the helper takes _metadata so that its ASSERTs
>>   attribute to the calling test, following the seccomp selftests
>>   precedent.
>> - Drop the now-unneeded NULL init of self->cg and the stale stdbool.h
>>   include left over from the restructuring.
>>
>> Changes since v3 (Suren Baghdasaryan review):
>> - Convert to the kselftest harness: each case is a TEST_F(psi, ...)
>>   with the cgroup root/PSI availability checks in FIXTURE_SETUP() and
>>   the teardown (kill hogs, destroy cgroup) in FIXTURE_TEARDOWN(),
>>   which also removes the "ret"/"created" bookkeeping.
>> - A trigger that does not fire within the poll timeout is now a FAIL
>>   instead of a SKIP: ncpus+1 hogs with a 1usec threshold must stall,
>>   so a timeout indicates a real problem.
>> - Treat a poll() timeout and a poll() error uniformly via ASSERT.
>> - Check sysconf(_SC_NPROCESSORS_ONLN) only for -1 and report
>>   strerror(errno); declare variables one per line; for(;;) {}.
>> - Make hog_cpu() die with the runner via PR_SET_PDEATHSIG so an
>>   interrupted run does not leave orphaned hogs pinning every CPU.
>>
>> Changes since v2 (Suren Baghdasaryan, Michal Koutny review):
>> - Restructure the trigger test into per-resource cases (io, memory, cpu,
>>   irq) so a failure points at the specific resource; irq is skipped when
>>   /proc/pressure/irq is absent.
>> - Spawn the CPU hogs with cg_run_nowait() instead of open-coding fork(),
>>   and arm the trigger with a 2s window so unprivileged users can set it.
>> - Address the remaining review comments on cleanup and robustness:
>>   guard teardown with a "created" flag, use cg_read_strcmp() instead of
>>   atoi(), report strerror() on errors, and fix the unused-parameter and
>>   sign-compare nits.
>>
>> Changes since v1 (Michal Koutny, sashiko review):
>> - Keep trigger tests smoke-level; switch the firing test from memory to
>>   CPU pressure; drop churn_memory().
>> - Keep the runner out of the cgroup; add PSI/IRQ skip-guards and a
>>   .gitignore entry.
>>
>> v1: https://lore.kernel.org/all/20260724025826.504586-1-cui.tao@linux.dev/
>> v2: https://lore.kernel.org/all/20260728083742.2359320-1-cui.tao@linux.dev/
>> v3: https://lore.kernel.org/all/20260813133723.1663605-1-cui.tao@linux.dev/
>> v4: https://lore.kernel.org/all/20260824085913.546741-1-cui.tao@linux.dev/
>> v5: https://lore.kernel.org/all/20260902040725.877155-1-cui.tao@linux.dev/
>> ---
>>  tools/testing/selftests/cgroup/.gitignore |   1 +
>>  tools/testing/selftests/cgroup/Makefile   |   2 +
>>  tools/testing/selftests/cgroup/config     |   1 +
>>  tools/testing/selftests/cgroup/test_psi.c | 219 ++++++++++++++++++++++
>>  4 files changed, 223 insertions(+)
>>  create mode 100644 tools/testing/selftests/cgroup/test_psi.c
>>
>> diff --git a/tools/testing/selftests/cgroup/.gitignore b/tools/testing/selftests/cgroup/.gitignore
>> index 952e4448bf070..ce2b907c57ea3 100644
>> --- a/tools/testing/selftests/cgroup/.gitignore
>> +++ b/tools/testing/selftests/cgroup/.gitignore
>> @@ -8,5 +8,6 @@ test_kill
>>  test_kmem
>>  test_memcontrol
>>  test_pids
>> +test_psi
>>  test_zswap
>>  wait_inotify
>> diff --git a/tools/testing/selftests/cgroup/Makefile b/tools/testing/selftests/cgroup/Makefile
>> index e01584c2189ac..a8c69e37332a4 100644
>> --- a/tools/testing/selftests/cgroup/Makefile
>> +++ b/tools/testing/selftests/cgroup/Makefile
>> @@ -16,6 +16,7 @@ TEST_GEN_PROGS += test_kill
>>  TEST_GEN_PROGS += test_kmem
>>  TEST_GEN_PROGS += test_memcontrol
>>  TEST_GEN_PROGS += test_pids
>> +TEST_GEN_PROGS += test_psi
>>  TEST_GEN_PROGS += test_zswap
>>
>>  LOCAL_HDRS += $(selfdir)/clone3/clone3_selftests.h $(selfdir)/pidfd/pidfd.h
>> @@ -32,4 +33,5 @@ $(OUTPUT)/test_kill: $(LIBCGROUP_O)
>>  $(OUTPUT)/test_kmem: $(LIBCGROUP_O)
>>  $(OUTPUT)/test_memcontrol: $(LIBCGROUP_O)
>>  $(OUTPUT)/test_pids: $(LIBCGROUP_O)
>> +$(OUTPUT)/test_psi: $(LIBCGROUP_O)
>>  $(OUTPUT)/test_zswap: $(LIBCGROUP_O)
>> diff --git a/tools/testing/selftests/cgroup/config b/tools/testing/selftests/cgroup/config
>> index 39f979690dd3b..8a3ef479e83d3 100644
>> --- a/tools/testing/selftests/cgroup/config
>> +++ b/tools/testing/selftests/cgroup/config
>> @@ -4,3 +4,4 @@ CONFIG_CGROUP_FREEZER=y
>>  CONFIG_CGROUP_SCHED=y
>>  CONFIG_MEMCG=y
>>  CONFIG_PAGE_COUNTER=y
>> +CONFIG_PSI=y
>> diff --git a/tools/testing/selftests/cgroup/test_psi.c b/tools/testing/selftests/cgroup/test_psi.c
>> new file mode 100644
>> index 0000000000000..95dcb8a9f1843
>> --- /dev/null
>> +++ b/tools/testing/selftests/cgroup/test_psi.c
>> @@ -0,0 +1,219 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +#define _GNU_SOURCE
>> +#include <errno.h>
>> +#include <fcntl.h>
>> +#include <poll.h>
>> +#include <stdio.h>
>> +#include <stdlib.h>
>> +#include <string.h>
>> +#include <unistd.h>
>> +#include <sys/prctl.h>
>> +#include <linux/limits.h>
>> +
>> +#include "../kselftest_harness.h"
>> +#include "cgroup_util.h"
>> +
>> +#define PSI_POLL_TIMEOUT_MS    5000
>> +
>> +/*
>> + * The kernel parser overwrites the last byte of a trigger write with a
>> + * NUL, so terminate with a newline to keep the payload intact.
>> + */
>> +static ssize_t write_trigger(int fd, const char *trigger)
>> +{
>> +       char buf[64];
>> +
>> +       snprintf(buf, sizeof(buf), "%s\n", trigger);
>> +
>> +       return write(fd, buf, strlen(buf));
>> +}
>> +
>> +static int pressure_open(const char *resource)
>> +{
>> +       char path[PATH_MAX];
>> +
>> +       snprintf(path, sizeof(path), "/proc/pressure/%s", resource);
>> +       return open(path, O_RDWR);
>> +}
>> +
>> +FIXTURE(psi)
>> +{
>> +       char root[PATH_MAX];
>> +       char *cg;
>> +};
>> +
>> +FIXTURE_SETUP(psi)
>> +{
>> +       int psi_fd;
>> +
>> +       if (cg_find_unified_root(self->root, sizeof(self->root), NULL))
>> +               SKIP(return, "cgroup v2 isn't mounted");
>> +
>> +       /* PSI must be enabled (CONFIG_PSI=y, not disabled on the cmdline). */
>> +       psi_fd = open("/proc/pressure/memory", O_RDONLY);
>> +       if (psi_fd < 0)
>> +               SKIP(return, "PSI unavailable (CONFIG_PSI=n or psi=0)");
>> +       close(psi_fd);
>> +
>> +       self->cg = cg_name(self->root, "psi_trigger_test");
>> +       ASSERT_NE(NULL, self->cg);
>> +       ASSERT_EQ(0, cg_create(self->cg));
>> +}
>> +
>> +FIXTURE_TEARDOWN(psi)
>> +{
>> +       cg_killall(self->cg);
>> +       cg_destroy(self->cg);
>> +       free(self->cg);
>> +}
>> +
>> +/*
>> + * /proc/pressure/<resource> accepts exactly one trigger per file
>> + * descriptor. Verify that a "some" trigger arms and that a second
>> + * trigger on the same fd is rejected with EBUSY.
>> + */
>> +static void test_psi_write(struct __test_metadata *_metadata,
>> +                          const char *filename)
>> +{
>> +       int fd;
>> +
>> +       fd = pressure_open(filename);
>> +       ASSERT_GE(fd, 0);
>> +       ASSERT_GT(write_trigger(fd, "some 150000 2000000"), 0);
>> +       ASSERT_EQ(-1, write_trigger(fd, "full 150000 2000000"));
>> +       ASSERT_EQ(EBUSY, errno);
>> +       close(fd);
>> +}
>> +
>> +TEST_F(psi, proc_trigger_io)
>> +{
>> +       test_psi_write(_metadata, "io");
>> +}
>> +
>> +TEST_F(psi, proc_trigger_memory)
>> +{
>> +       test_psi_write(_metadata, "memory");
>> +}
>> +
>> +TEST_F(psi, proc_trigger_cpu)
>> +{
>> +       test_psi_write(_metadata, "cpu");
>> +}
>> +
>> +/*
>> + * irq only tracks "full", so a "some" trigger must be rejected while a
>> + * "full" trigger arms. irq is optional -- it only exists with IRQ-time
>> + * accounting -- so a missing /proc/pressure/irq is SKIP, not FAIL.
>> + */
>> +TEST_F(psi, proc_trigger_irq)
>> +{
>> +       int fd;
>> +
>> +       fd = pressure_open("irq");
>> +       if (fd < 0)
>> +               SKIP(return, "/proc/pressure/irq unavailable");
>> +
>> +       ASSERT_EQ(-1, write_trigger(fd, "some 150000 2000000"));
>> +       ASSERT_GT(write_trigger(fd, "full 150000 2000000"), 0);
>> +       close(fd);
>> +}
>> +
>> +/*
>> + * cgroup.pressure gates visibility of the per-resource *.pressure files
>> + * inside a cgroup: writing 0 hides them, writing 1 shows them again.
>> + * Drive one hide/show cycle and check that memory.pressure appears and
>> + * disappears along with it.
>> + */
>> +TEST_F(psi, cgroup_pressure_toggle)
>> +{
>> +       char buf[BUF_SIZE];
>> +
>> +       ASSERT_EQ(0, cg_write(self->cg, "cgroup.pressure", "0"));
>> +       ASSERT_EQ(0, cg_read_strcmp(self->cg, "cgroup.pressure", "0\n"));
>> +       ASSERT_LT(cg_read(self->cg, "memory.pressure", buf, sizeof(buf)), 0);
>> +
>> +       ASSERT_EQ(0, cg_write(self->cg, "cgroup.pressure", "1"));
>> +       ASSERT_EQ(0, cg_read_strcmp(self->cg, "cgroup.pressure", "1\n"));
>> +       ASSERT_GE(cg_read(self->cg, "memory.pressure", buf, sizeof(buf)), 0);
>> +}
>> +
>> +/*
>> + * A child that burns CPU forever; stopped by cg_killall() on teardown.
>> + * It also dies with the runner, so an interrupted run (e.g. Ctrl-C
>> + * during poll()) does not leave orphaned hogs pinning every CPU.
> 
> Did you mean to say "orphaned hogs spinning on every CPU"?
> 

Yes, "spinning" it is -- that's a typo in the hog_cpu() comment,
introduced when the PDEATHSIG part was added. If there is another
revision of this patch I'll fold the fix in, otherwise feel free to fix it up on the way in.

Thanks,
Tao

>> + */
>> +static int hog_cpu(const char *cgroup, void *arg)
>> +{
>> +       prctl(PR_SET_PDEATHSIG, SIGKILL);
>> +       for (;;) {}
>> +       return 0;
>> +}
>> +
>> +/*
>> + * Arm a "some" trigger on a cgroup's cpu.pressure, oversubscribe the
>> + * cgroup with more spinning hogs than there are CPUs, and check that the
>> + * trigger fires once the cgroup stalls on CPU.
>> + */
>> +TEST_F(psi, cgroup_trigger_fire)
>> +{
>> +       char *cpupress;
>> +       struct pollfd pfd = { .events = POLLPRI };
>> +       long ncpus;
>> +       int fd;
>> +       int i;
>> +
>> +       cpupress = cg_control(self->cg, "cpu.pressure");
>> +       ASSERT_NE(NULL, cpupress);
>> +       fd = open(cpupress, O_RDWR);
>> +       free(cpupress);
>> +       ASSERT_GE(fd, 0);
>> +       pfd.fd = fd;
>> +
>> +       /* 1usec threshold over a 2s window: any CPU stall fires it. */
>> +       ASSERT_GT(write_trigger(fd, "some 1 2000000"), 0);
>> +
>> +       ncpus = sysconf(_SC_NPROCESSORS_ONLN);
>> +       ASSERT_NE(-1, ncpus);
>> +
>> +       /* ncpus+1 hogs guarantee CPU contention inside the cgroup. */
>> +       for (i = 0; i < ncpus + 1; i++)
>> +               ASSERT_GE(cg_run_nowait(self->cg, hog_cpu, NULL), 0);
>> +
>> +       ASSERT_EQ(1, poll(&pfd, 1, PSI_POLL_TIMEOUT_MS));
>> +       ASSERT_NE(0, pfd.revents & POLLPRI);
>> +       close(fd);
>> +}
>> +
>> +/*
>> + * The flip side of cgroup_trigger_fire: with the threshold equal to the
>> + * whole window, even ncpus+1 hogs must not accumulate enough stall to
>> + * fire the trigger.
>> + */
>> +TEST_F(psi, cgroup_trigger_no_fire)
>> +{
>> +       char *cpupress;
>> +       struct pollfd pfd = { .events = POLLPRI };
>> +       long ncpus;
>> +       int fd;
>> +       int i;
>> +
>> +       cpupress = cg_control(self->cg, "cpu.pressure");
>> +       ASSERT_NE(NULL, cpupress);
>> +       fd = open(cpupress, O_RDWR);
>> +       free(cpupress);
>> +       ASSERT_GE(fd, 0);
>> +       pfd.fd = fd;
>> +
>> +       ASSERT_GT(write_trigger(fd, "some 2000000 2000000"), 0);
>> +
>> +       ncpus = sysconf(_SC_NPROCESSORS_ONLN);
>> +       ASSERT_NE(-1, ncpus);
>> +
>> +       for (i = 0; i < ncpus + 1; i++)
>> +               ASSERT_GE(cg_run_nowait(self->cg, hog_cpu, NULL), 0);
>> +
>> +       ASSERT_EQ(0, poll(&pfd, 1, PSI_POLL_TIMEOUT_MS));
>> +       close(fd);
>> +}
>> +
>> +TEST_HARNESS_MAIN
>> --
>> 2.43.0
>>