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, 170 insertions, 0 deletions

diff --git a/tools/testing/selftests/rlimits/.gitignore b/tools/testing/selftests/rlimits/.gitignore
new file mode 100644
index 000000000..091021f25
--- /dev/null
+++ b/tools/testing/selftests/rlimits/.gitignore
@@ -0,0 +1,2 @@
+# SPDX-License-Identifier: GPL-2.0-only
+rlimits-per-userns
diff --git a/tools/testing/selftests/rlimits/Makefile b/tools/testing/selftests/rlimits/Makefile
new file mode 100644
index 000000000..03aadb406
--- /dev/null
+++ b/tools/testing/selftests/rlimits/Makefile
@@ -0,0 +1,6 @@
+# SPDX-License-Identifier: GPL-2.0-or-later
+
+CFLAGS += -Wall -O2 -g
+TEST_GEN_PROGS := rlimits-per-userns
+
+include ../lib.mk
diff --git a/tools/testing/selftests/rlimits/config b/tools/testing/selftests/rlimits/config
new file mode 100644
index 000000000..416bd53ce
--- /dev/null
+++ b/tools/testing/selftests/rlimits/config
@@ -0,0 +1 @@
+CONFIG_USER_NS=y
diff --git a/tools/testing/selftests/rlimits/rlimits-per-userns.c b/tools/testing/selftests/rlimits/rlimits-per-userns.c
new file mode 100644
index 000000000..26dc949e9
--- /dev/null
+++ b/tools/testing/selftests/rlimits/rlimits-per-userns.c
@@ -0,0 +1,161 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Author: Alexey Gladkov <gladkov.alexey@gmail.com>
+ */
+#define _GNU_SOURCE
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <sys/time.h>
+#include <sys/resource.h>
+#include <sys/prctl.h>
+#include <sys/stat.h>
+
+#include <unistd.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <sched.h>
+#include <signal.h>
+#include <limits.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <err.h>
+
+#define NR_CHILDS 2
+
+static char *service_prog;
+static uid_t user   = 60000;
+static uid_t group  = 60000;
+
+static void setrlimit_nproc(rlim_t n)
+{
+	pid_t pid = getpid();
+	struct rlimit limit = {
+		.rlim_cur = n,
+		.rlim_max = n
+	};
+
+	warnx("(pid=%d): Setting RLIMIT_NPROC=%ld", pid, n);
+
+	if (setrlimit(RLIMIT_NPROC, &limit) < 0)
+		err(EXIT_FAILURE, "(pid=%d): setrlimit(RLIMIT_NPROC)", pid);
+}
+
+static pid_t fork_child(void)
+{
+	pid_t pid = fork();
+
+	if (pid < 0)
+		err(EXIT_FAILURE, "fork");
+
+	if (pid > 0)
+		return pid;
+
+	pid = getpid();
+
+	warnx("(pid=%d): New process starting ...", pid);
+
+	if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
+		err(EXIT_FAILURE, "(pid=%d): prctl(PR_SET_PDEATHSIG)", pid);
+
+	signal(SIGUSR1, SIG_DFL);
+
+	warnx("(pid=%d): Changing to uid=%d, gid=%d", pid, user, group);
+
+	if (setgid(group) < 0)
+		err(EXIT_FAILURE, "(pid=%d): setgid(%d)", pid, group);
+	if (setuid(user) < 0)
+		err(EXIT_FAILURE, "(pid=%d): setuid(%d)", pid, user);
+
+	warnx("(pid=%d): Service running ...", pid);
+
+	warnx("(pid=%d): Unshare user namespace", pid);
+	if (unshare(CLONE_NEWUSER) < 0)
+		err(EXIT_FAILURE, "unshare(CLONE_NEWUSER)");
+
+	char *const argv[] = { "service", NULL };
+	char *const envp[] = { "I_AM_SERVICE=1", NULL };
+
+	warnx("(pid=%d): Executing real service ...", pid);
+
+	execve(service_prog, argv, envp);
+	err(EXIT_FAILURE, "(pid=%d): execve", pid);
+}
+
+int main(int argc, char **argv)
+{
+	size_t i;
+	pid_t child[NR_CHILDS];
+	int wstatus[NR_CHILDS];
+	int childs = NR_CHILDS;
+	pid_t pid;
+
+	if (getenv("I_AM_SERVICE")) {
+		pause();
+		exit(EXIT_SUCCESS);
+	}
+
+	service_prog = argv[0];
+	pid = getpid();
+
+	warnx("(pid=%d) Starting testcase", pid);
+
+	/*
+	 * This rlimit is not a problem for root because it can be exceeded.
+	 */
+	setrlimit_nproc(1);
+
+	for (i = 0; i < NR_CHILDS; i++) {
+		child[i] = fork_child();
+		wstatus[i] = 0;
+		usleep(250000);
+	}
+
+	while (1) {
+		for (i = 0; i < NR_CHILDS; i++) {
+			if (child[i] <= 0)
+				continue;
+
+			errno = 0;
+			pid_t ret = waitpid(child[i], &wstatus[i], WNOHANG);
+
+			if (!ret || (!WIFEXITED(wstatus[i]) && !WIFSIGNALED(wstatus[i])))
+				continue;
+
+			if (ret < 0 && errno != ECHILD)
+				warn("(pid=%d): waitpid(%d)", pid, child[i]);
+
+			child[i] *= -1;
+			childs -= 1;
+		}
+
+		if (!childs)
+			break;
+
+		usleep(250000);
+
+		for (i = 0; i < NR_CHILDS; i++) {
+			if (child[i] <= 0)
+				continue;
+			kill(child[i], SIGUSR1);
+		}
+	}
+
+	for (i = 0; i < NR_CHILDS; i++) {
+		if (WIFEXITED(wstatus[i]))
+			warnx("(pid=%d): pid %d exited, status=%d",
+				pid, -child[i], WEXITSTATUS(wstatus[i]));
+		else if (WIFSIGNALED(wstatus[i]))
+			warnx("(pid=%d): pid %d killed by signal %d",
+				pid, -child[i], WTERMSIG(wstatus[i]));
+
+		if (WIFSIGNALED(wstatus[i]) && WTERMSIG(wstatus[i]) == SIGUSR1)
+			continue;
+
+		warnx("(pid=%d): Test failed", pid);
+		exit(EXIT_FAILURE);
+	}
+
+	warnx("(pid=%d): Test passed", pid);
+	exit(EXIT_SUCCESS);
+}