sched/cache: Keep nr_pref_llc_running in the runnable domain, to fix LLC mis-scheduling buggrafted

alb_break_llc() decides whether to break LLC preference during active
load balance. It does so by testing that every runnable fair task on the
source rq prefers its LLC:

	env->src_rq->nr_pref_llc_running == env->src_rq->cfs.h_nr_runnable

But the two counters cover different sets. nr_pref_llc_running is updated
in account_llc_enqueue()/account_llc_dequeue(), next to cfs_rq->nr_queued,
so it follows queued tasks. h_nr_runnable is updated in set_delayed()/
clear_delayed() and drops delay-dequeued tasks.

So under DELAY_DEQUEUE, a preferring task that goes to sleep stays counted
in nr_pref_llc_running while h_nr_runnable falls. The equality then breaks,
alb_break_llc() returns false, and active balance is free to pull a task
off its preferred LLC. Active balance only moves runnable tasks, and this
is the only LLC check it consults: once the stopper runs, LBF_ACTIVE_LB
skips the per-task test in can_migrate_task(). The runnable set is the one
we want.

Fix it on the counter side. A task should be counted in
nr_pref_llc_running exactly while it is both queued on its preferred LLC
(pref_llc_queued) and runnable (!sched_delayed). Define that membership
once in task_pref_llc_runnable(), and adjust the counter only through
pref_llc_running_inc()/pref_llc_running_dec() from the four sites that
change either input: account_llc_enqueue(), account_llc_dequeue(),
set_delayed() and clear_delayed(). Gating every update on the same
predicate keeps the delay, wake and dequeue paths from double-counting
or underflowing; see the comments at those sites for the ordering.

nr_llc_running and sd->llc_counts are not touched and stay on queued
semantics.

Fixes: 714059f79ff0 ("sched/cache: Handle moving single tasks to/from their preferred LLC")
Closes: https://lore.kernel.org/lkml/20260827135000.735138-1-zhanxusheng@xiaomi.com/
Reported-by: Zhan Xusheng <zhanxusheng@xiaomi.com>
Suggested-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Kayra Cizmeci <kayracizmeci@gmail.com>
Cc: <stable@kernel.org> # v7.2.x
Link: https://patch.msgid.link/06af61afedac32e6477f57feb4d658f6c411c3af.1790035273.git.tim.c.chen@linux.intel.com

4 files changed, 279 insertions, 0 deletions

diff --git a/tools/testing/selftests/ipc/.gitignore b/tools/testing/selftests/ipc/.gitignore
new file mode 100644
index 000000000..9ed280e4c
--- /dev/null
+++ b/tools/testing/selftests/ipc/.gitignore
@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: GPL-2.0-only
+msgque_test
+msgque
diff --git a/tools/testing/selftests/ipc/Makefile b/tools/testing/selftests/ipc/Makefile
new file mode 100644
index 000000000..fad10f2bb
--- /dev/null
+++ b/tools/testing/selftests/ipc/Makefile
@@ -0,0 +1,18 @@
+# SPDX-License-Identifier: GPL-2.0
+ARCH ?= $(shell uname -m 2>/dev/null || echo not)
+override ARCH := $(shell echo $(ARCH) | sed -e s/i.86/i386/)
+ifeq ($(ARCH),i386)
+        override ARCH := x86
+	CFLAGS := -DCONFIG_X86_32 -D__i386__
+endif
+ifeq ($(ARCH),x86_64)
+        override ARCH := x86
+	CFLAGS := -DCONFIG_X86_64 -D__x86_64__
+endif
+
+CFLAGS += $(KHDR_INCLUDES)
+
+TEST_GEN_PROGS := msgque
+
+include ../lib.mk
+
diff --git a/tools/testing/selftests/ipc/config b/tools/testing/selftests/ipc/config
new file mode 100644
index 000000000..070244710
--- /dev/null
+++ b/tools/testing/selftests/ipc/config
@@ -0,0 +1,2 @@
+CONFIG_EXPERT=y
+CONFIG_CHECKPOINT_RESTORE=y
diff --git a/tools/testing/selftests/ipc/msgque.c b/tools/testing/selftests/ipc/msgque.c
new file mode 100644
index 000000000..4b4c1abfa
--- /dev/null
+++ b/tools/testing/selftests/ipc/msgque.c
@@ -0,0 +1,256 @@
+// SPDX-License-Identifier: GPL-2.0
+#define _GNU_SOURCE
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <errno.h>
+#include <sys/msg.h>
+#include <fcntl.h>
+
+#include "kselftest.h"
+
+#define MAX_MSG_SIZE		32
+
+struct msg1 {
+	int msize;
+	long mtype;
+	char mtext[MAX_MSG_SIZE];
+};
+
+#define TEST_STRING "Test sysv5 msg"
+#define MSG_TYPE 1
+
+#define ANOTHER_TEST_STRING "Yet another test sysv5 msg"
+#define ANOTHER_MSG_TYPE 26538
+
+struct msgque_data {
+	key_t key;
+	int msq_id;
+	int qbytes;
+	int qnum;
+	int mode;
+	struct msg1 *messages;
+};
+
+int restore_queue(struct msgque_data *msgque)
+{
+	int fd, ret, id, i;
+	char buf[32];
+
+	fd = open("/proc/sys/kernel/msg_next_id", O_WRONLY);
+	if (fd == -1) {
+		ksft_test_result_fail("Failed to open /proc/sys/kernel/msg_next_id\n");
+		return -errno;
+	}
+	sprintf(buf, "%d", msgque->msq_id);
+
+	ret = write(fd, buf, strlen(buf));
+	if (ret != strlen(buf)) {
+		ksft_test_result_fail("Failed to write to /proc/sys/kernel/msg_next_id\n");
+		return -errno;
+	}
+
+	id = msgget(msgque->key, msgque->mode | IPC_CREAT | IPC_EXCL);
+	if (id == -1) {
+		ksft_test_result_fail("Failed to create queue\n");
+		return -errno;
+	}
+
+	if (id != msgque->msq_id) {
+		ksft_test_result_fail("Restored queue has wrong id (%d instead of %d)\n"
+								, id, msgque->msq_id);
+		ret = -EFAULT;
+		goto destroy;
+	}
+
+	for (i = 0; i < msgque->qnum; i++) {
+		if (msgsnd(msgque->msq_id, &msgque->messages[i].mtype,
+			   msgque->messages[i].msize, IPC_NOWAIT) != 0) {
+			ksft_test_result_fail("msgsnd failed (%m)\n");
+			ret = -errno;
+			goto destroy;
+		}
+	}
+	return 0;
+
+destroy:
+	if (msgctl(id, IPC_RMID, NULL))
+		printf("Failed to destroy queue: %d\n", -errno);
+	return ret;
+}
+
+int check_and_destroy_queue(struct msgque_data *msgque)
+{
+	struct msg1 message;
+	int cnt = 0, ret;
+
+	while (1) {
+		ret = msgrcv(msgque->msq_id, &message.mtype, MAX_MSG_SIZE,
+				0, IPC_NOWAIT);
+		if (ret < 0) {
+			if (errno == ENOMSG)
+				break;
+			ksft_test_result_fail("Failed to read IPC message: %m\n");
+			ret = -errno;
+			goto err;
+		}
+		if (ret != msgque->messages[cnt].msize) {
+			ksft_test_result_fail("Wrong message size: %d (expected %d)\n", ret, msgque->messages[cnt].msize);
+			ret = -EINVAL;
+			goto err;
+		}
+		if (message.mtype != msgque->messages[cnt].mtype) {
+			ksft_test_result_fail("Wrong message type\n");
+			ret = -EINVAL;
+			goto err;
+		}
+		if (memcmp(message.mtext, msgque->messages[cnt].mtext, ret)) {
+			ksft_test_result_fail("Wrong message content\n");
+			ret = -EINVAL;
+			goto err;
+		}
+		cnt++;
+	}
+
+	if (cnt != msgque->qnum) {
+		ksft_test_result_fail("Wrong message number\n");
+		ret = -EINVAL;
+		goto err;
+	}
+
+	ret = 0;
+err:
+	if (msgctl(msgque->msq_id, IPC_RMID, NULL)) {
+		printf("Failed to destroy queue: %d\n", -errno);
+		return -errno;
+	}
+	return ret;
+}
+
+int dump_queue(struct msgque_data *msgque)
+{
+	struct msqid_ds ds;
+	int kern_id;
+	int i, ret;
+
+	for (kern_id = 0; kern_id < 256; kern_id++) {
+		ret = msgctl(kern_id, MSG_STAT, &ds);
+		if (ret < 0) {
+			if (errno == EINVAL)
+				continue;
+			ksft_test_result_fail("Failed to get stats for IPC queue with id %d\n",
+					kern_id);
+			return -errno;
+		}
+
+		if (ret == msgque->msq_id)
+			break;
+	}
+
+	msgque->messages = malloc(sizeof(struct msg1) * ds.msg_qnum);
+	if (msgque->messages == NULL) {
+		ksft_test_result_fail("Failed to get stats for IPC queue\n");
+		return -ENOMEM;
+	}
+
+	msgque->qnum = ds.msg_qnum;
+	msgque->mode = ds.msg_perm.mode;
+	msgque->qbytes = ds.msg_qbytes;
+
+	for (i = 0; i < msgque->qnum; i++) {
+		ret = msgrcv(msgque->msq_id, &msgque->messages[i].mtype,
+				MAX_MSG_SIZE, i, IPC_NOWAIT | MSG_COPY);
+		if (ret < 0) {
+			if (errno == EOPNOTSUPP)
+				ksft_exit_skip("MSG_COPY not supported\n");
+
+			ksft_test_result_fail("Failed to copy IPC message: %m (%d)\n", errno);
+			return -errno;
+		}
+		msgque->messages[i].msize = ret;
+	}
+	return 0;
+}
+
+int fill_msgque(struct msgque_data *msgque)
+{
+	struct msg1 msgbuf;
+
+	msgbuf.mtype = MSG_TYPE;
+	memcpy(msgbuf.mtext, TEST_STRING, sizeof(TEST_STRING));
+	if (msgsnd(msgque->msq_id, &msgbuf.mtype, sizeof(TEST_STRING),
+				IPC_NOWAIT) != 0) {
+		ksft_test_result_fail("First message send failed (%m)\n");
+		return -errno;
+	}
+
+	msgbuf.mtype = ANOTHER_MSG_TYPE;
+	memcpy(msgbuf.mtext, ANOTHER_TEST_STRING, sizeof(ANOTHER_TEST_STRING));
+	if (msgsnd(msgque->msq_id, &msgbuf.mtype, sizeof(ANOTHER_TEST_STRING),
+				IPC_NOWAIT) != 0) {
+		ksft_test_result_fail("Second message send failed (%m)\n");
+		return -errno;
+	}
+	return 0;
+}
+
+int main(int argc, char **argv)
+{
+	int err;
+	struct msgque_data msgque;
+
+	if (getuid() != 0)
+		ksft_exit_skip("Please run the test as root - Exiting.\n");
+
+	msgque.key = ftok(argv[0], 822155650);
+	if (msgque.key == -1) {
+		ksft_test_result_fail("Can't make key: %d\n", -errno);
+		ksft_exit_fail();
+	}
+
+	msgque.msq_id = msgget(msgque.key, IPC_CREAT | IPC_EXCL | 0666);
+	if (msgque.msq_id == -1) {
+		err = -errno;
+		ksft_test_result_fail("Can't create queue: %d\n", err);
+		goto err_out;
+	}
+
+	err = fill_msgque(&msgque);
+	if (err) {
+		ksft_test_result_fail("Failed to fill queue: %d\n", err);
+		goto err_destroy;
+	}
+
+	err = dump_queue(&msgque);
+	if (err) {
+		ksft_test_result_fail("Failed to dump queue: %d\n", err);
+		goto err_destroy;
+	}
+
+	err = check_and_destroy_queue(&msgque);
+	if (err) {
+		ksft_test_result_fail("Failed to check and destroy queue: %d\n", err);
+		goto err_out;
+	}
+
+	err = restore_queue(&msgque);
+	if (err) {
+		ksft_test_result_fail("Failed to restore queue: %d\n", err);
+		goto err_destroy;
+	}
+
+	err = check_and_destroy_queue(&msgque);
+	if (err) {
+		ksft_test_result_fail("Failed to test queue: %d\n", err);
+		goto err_out;
+	}
+	ksft_exit_pass();
+
+err_destroy:
+	if (msgctl(msgque.msq_id, IPC_RMID, NULL)) {
+		printf("Failed to destroy queue: %d\n", -errno);
+		ksft_exit_fail();
+	}
+err_out:
+	ksft_exit_fail();
+}