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

3 files changed, 180 insertions, 0 deletions

diff --git a/tools/testing/selftests/kcmp/.gitignore b/tools/testing/selftests/kcmp/.gitignore
new file mode 100644
index 000000000..38ccdfe80
--- /dev/null
+++ b/tools/testing/selftests/kcmp/.gitignore
@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: GPL-2.0-only
+kcmp_test
+kcmp-test-file
diff --git a/tools/testing/selftests/kcmp/Makefile b/tools/testing/selftests/kcmp/Makefile
new file mode 100644
index 000000000..59a1e5379
--- /dev/null
+++ b/tools/testing/selftests/kcmp/Makefile
@@ -0,0 +1,9 @@
+# SPDX-License-Identifier: GPL-2.0-only
+CFLAGS += $(KHDR_INCLUDES)
+
+TEST_GEN_PROGS := kcmp_test
+
+EXTRA_CLEAN := $(OUTPUT)/kcmp-test-file
+
+include ../lib.mk
+
diff --git a/tools/testing/selftests/kcmp/kcmp_test.c b/tools/testing/selftests/kcmp/kcmp_test.c
new file mode 100644
index 000000000..79aa438b7
--- /dev/null
+++ b/tools/testing/selftests/kcmp/kcmp_test.c
@@ -0,0 +1,168 @@
+// SPDX-License-Identifier: GPL-2.0
+#define _GNU_SOURCE
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <signal.h>
+#include <limits.h>
+#include <unistd.h>
+#include <errno.h>
+#include <string.h>
+#include <fcntl.h>
+#include <linux/unistd.h>
+#include <linux/kcmp.h>
+
+#include <sys/syscall.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/wait.h>
+#include <sys/epoll.h>
+
+#include "kselftest.h"
+
+static long sys_kcmp(int pid1, int pid2, int type, unsigned long fd1, unsigned long fd2)
+{
+	return syscall(__NR_kcmp, pid1, pid2, type, fd1, fd2);
+}
+
+static const unsigned int duped_num = 64;
+
+int main(int argc, char **argv)
+{
+	const char kpath[] = "kcmp-test-file";
+	struct kcmp_epoll_slot epoll_slot;
+	struct epoll_event ev;
+	int pid1, pid2;
+	int pipefd[2];
+	int fd1, fd2;
+	int epollfd;
+	int status;
+	int fddup;
+
+	fd1 = open(kpath, O_RDWR | O_CREAT | O_TRUNC, 0644);
+	pid1 = getpid();
+
+	if (fd1 < 0) {
+		perror("Can't create file");
+		ksft_exit_fail();
+	}
+
+	if (pipe(pipefd)) {
+		perror("Can't create pipe");
+		ksft_exit_fail();
+	}
+
+	epollfd = epoll_create1(0);
+	if (epollfd < 0) {
+		perror("epoll_create1 failed");
+		ksft_exit_fail();
+	}
+
+	memset(&ev, 0xff, sizeof(ev));
+	ev.events = EPOLLIN | EPOLLOUT;
+
+	if (epoll_ctl(epollfd, EPOLL_CTL_ADD, pipefd[0], &ev)) {
+		perror("epoll_ctl failed");
+		ksft_exit_fail();
+	}
+
+	fddup = dup2(pipefd[1], duped_num);
+	if (fddup < 0) {
+		perror("dup2 failed");
+		ksft_exit_fail();
+	}
+
+	if (epoll_ctl(epollfd, EPOLL_CTL_ADD, fddup, &ev)) {
+		perror("epoll_ctl failed");
+		ksft_exit_fail();
+	}
+	close(fddup);
+
+	pid2 = fork();
+	if (pid2 < 0) {
+		perror("fork failed");
+		ksft_exit_fail();
+	}
+
+	if (!pid2) {
+		int pid2 = getpid();
+		int ret;
+
+		ksft_print_header();
+		ksft_set_plan(3);
+
+		fd2 = open(kpath, O_RDWR);
+		if (fd2 < 0) {
+			perror("Can't open file");
+			ksft_exit_fail();
+		}
+
+		/* An example of output and arguments */
+		printf("pid1: %6d pid2: %6d FD: %2ld FILES: %2ld VM: %2ld "
+		       "FS: %2ld SIGHAND: %2ld IO: %2ld SYSVSEM: %2ld "
+		       "INV: %2ld\n",
+		       pid1, pid2,
+		       sys_kcmp(pid1, pid2, KCMP_FILE,		fd1, fd2),
+		       sys_kcmp(pid1, pid2, KCMP_FILES,		0, 0),
+		       sys_kcmp(pid1, pid2, KCMP_VM,		0, 0),
+		       sys_kcmp(pid1, pid2, KCMP_FS,		0, 0),
+		       sys_kcmp(pid1, pid2, KCMP_SIGHAND,	0, 0),
+		       sys_kcmp(pid1, pid2, KCMP_IO,		0, 0),
+		       sys_kcmp(pid1, pid2, KCMP_SYSVSEM,	0, 0),
+
+			/* This one should fail */
+		       sys_kcmp(pid1, pid2, KCMP_TYPES + 1,	0, 0));
+
+		/* This one should return same fd */
+		ret = sys_kcmp(pid1, pid2, KCMP_FILE, fd1, fd1);
+		if (ret) {
+			printf("FAIL: 0 expected but %d returned (%s)\n",
+				ret, strerror(errno));
+			ksft_inc_fail_cnt();
+			ret = -1;
+		} else {
+			printf("PASS: 0 returned as expected\n");
+			ksft_inc_pass_cnt();
+		}
+
+		/* Compare with self */
+		ret = sys_kcmp(pid1, pid1, KCMP_VM, 0, 0);
+		if (ret) {
+			printf("FAIL: 0 expected but %d returned (%s)\n",
+				ret, strerror(errno));
+			ksft_inc_fail_cnt();
+			ret = -1;
+		} else {
+			printf("PASS: 0 returned as expected\n");
+			ksft_inc_pass_cnt();
+		}
+
+		/* Compare epoll target */
+		epoll_slot = (struct kcmp_epoll_slot) {
+			.efd	= epollfd,
+			.tfd	= duped_num,
+			.toff	= 0,
+		};
+		ret = sys_kcmp(pid1, pid1, KCMP_EPOLL_TFD, pipefd[1],
+			       (unsigned long)(void *)&epoll_slot);
+		if (ret) {
+			printf("FAIL: 0 expected but %d returned (%s)\n",
+				ret, strerror(errno));
+			ksft_inc_fail_cnt();
+			ret = -1;
+		} else {
+			printf("PASS: 0 returned as expected\n");
+			ksft_inc_pass_cnt();
+		}
+
+
+		if (ret)
+			ksft_exit_fail();
+		else
+			ksft_exit_pass();
+	}
+
+	waitpid(pid2, &status, P_ALL);
+
+	return 0;
+}