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, 217 insertions, 0 deletions

diff --git a/tools/testing/selftests/fchmodat2/.gitignore b/tools/testing/selftests/fchmodat2/.gitignore
new file mode 100644
index 000000000..82a4846cb
--- /dev/null
+++ b/tools/testing/selftests/fchmodat2/.gitignore
@@ -0,0 +1,2 @@
+# SPDX-License-Identifier: GPL-2.0-only
+/*_test
diff --git a/tools/testing/selftests/fchmodat2/Makefile b/tools/testing/selftests/fchmodat2/Makefile
new file mode 100644
index 000000000..4373cea79
--- /dev/null
+++ b/tools/testing/selftests/fchmodat2/Makefile
@@ -0,0 +1,15 @@
+# SPDX-License-Identifier: GPL-2.0-or-later
+
+CFLAGS += -Wall -O2 -g -fsanitize=address -fsanitize=undefined $(KHDR_INCLUDES)
+
+# gcc requires -static-libasan in order to ensure that Address Sanitizer's
+# library is the first one loaded. However, clang already statically links the
+# Address Sanitizer if -fsanitize is specified. Therefore, simply omit
+# -static-libasan for clang builds.
+ifeq ($(LLVM),)
+    CFLAGS += -static-libasan
+endif
+
+TEST_GEN_PROGS := fchmodat2_test
+
+include ../lib.mk
diff --git a/tools/testing/selftests/fchmodat2/fchmodat2_test.c b/tools/testing/selftests/fchmodat2/fchmodat2_test.c
new file mode 100644
index 000000000..8e1b6384b
--- /dev/null
+++ b/tools/testing/selftests/fchmodat2/fchmodat2_test.c
@@ -0,0 +1,200 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#define _GNU_SOURCE
+#include <fcntl.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <syscall.h>
+#include <unistd.h>
+
+#include "kselftest.h"
+
+struct testdir {
+	char *dirname;
+	int dfd;
+};
+
+int sys_fchmodat2(int dfd, const char *filename, mode_t mode, int flags)
+{
+	int ret = syscall(__NR_fchmodat2, dfd, filename, mode, flags);
+
+	return ret >= 0 ? ret : -errno;
+}
+
+static void setup_testdir(struct testdir *testdir)
+{
+	int ret, dfd;
+	char dirname[] = "/tmp/ksft-fchmodat2.XXXXXX";
+
+	/* Make the top-level directory. */
+	if (!mkdtemp(dirname))
+		ksft_exit_fail_msg("%s: failed to create tmpdir\n", __func__);
+
+	dfd = open(dirname, O_PATH | O_DIRECTORY);
+	if (dfd < 0) {
+		ksft_perror("failed to open tmpdir");
+		goto err;
+	}
+
+	ret = openat(dfd, "regfile", O_CREAT | O_WRONLY | O_TRUNC, 0644);
+	if (ret < 0) {
+		ksft_perror("failed to create file in tmpdir");
+		goto err;
+	}
+	close(ret);
+
+	ret = symlinkat("regfile", dfd, "symlink");
+	if (ret < 0) {
+		ksft_perror("symlinkat() failed");
+		goto err_regfile;
+	}
+
+	testdir->dirname = strdup(dirname);
+	if (!testdir->dirname) {
+		ksft_perror("Out of memory");
+		goto err_symlink;
+	}
+	testdir->dfd = dfd;
+
+	return;
+
+err_symlink:
+	unlinkat(testdir->dfd, "symlink", 0);
+err_regfile:
+	unlinkat(testdir->dfd, "regfile", 0);
+err:
+	unlink(dirname);
+	ksft_exit_fail();
+}
+
+static void cleanup_testdir(struct testdir *testdir)
+{
+	unlinkat(testdir->dfd, "regfile", 0);
+	unlinkat(testdir->dfd, "symlink", 0);
+	rmdir(testdir->dirname);
+	free(testdir->dirname);
+}
+
+int expect_mode(int dfd, const char *filename, mode_t expect_mode)
+{
+	struct stat st;
+	int ret = fstatat(dfd, filename, &st, AT_SYMLINK_NOFOLLOW);
+
+	if (ret) {
+		ksft_perror("fstatat() failed\n");
+		return 0;
+	}
+
+	return (st.st_mode == expect_mode);
+}
+
+void test_regfile(void)
+{
+	struct testdir testdir;
+	int ret;
+
+	setup_testdir(&testdir);
+
+	ret = sys_fchmodat2(testdir.dfd, "regfile", 0640, 0);
+
+	if (ret < 0) {
+		ksft_perror("fchmodat2(noflag) failed");
+		goto out;
+	}
+
+	if (!expect_mode(testdir.dfd, "regfile", 0100640)) {
+		ksft_print_msg("%s: wrong file mode bits after fchmodat2\n",
+				__func__);
+		ret = 1;
+		goto out;
+	}
+
+	ret = sys_fchmodat2(testdir.dfd, "regfile", 0600, AT_SYMLINK_NOFOLLOW);
+
+	if (ret < 0) {
+		ksft_perror("fchmodat2(AT_SYMLINK_NOFOLLOW) failed");
+		goto out;
+	}
+
+	if (!expect_mode(testdir.dfd, "regfile", 0100600)) {
+		ksft_print_msg("%s: wrong file mode bits after fchmodat2 with nofollow\n",
+			       __func__);
+		ret = 1;
+	}
+
+out:
+	ksft_test_result(ret == 0, "fchmodat2(regfile)\n");
+	cleanup_testdir(&testdir);
+}
+
+void test_symlink(void)
+{
+	struct testdir testdir;
+	int ret;
+
+	setup_testdir(&testdir);
+
+	ret = sys_fchmodat2(testdir.dfd, "symlink", 0640, 0);
+
+	if (ret < 0) {
+		ksft_perror("fchmodat2(noflag) failed");
+		goto err;
+	}
+
+	if (!expect_mode(testdir.dfd, "regfile", 0100640)) {
+		ksft_print_msg("%s: wrong file mode bits after fchmodat2\n",
+			       __func__);
+		goto err;
+	}
+
+	if (!expect_mode(testdir.dfd, "symlink", 0120777)) {
+		ksft_print_msg("%s: wrong symlink mode bits after fchmodat2\n",
+			       __func__);
+		goto err;
+	}
+
+	ret = sys_fchmodat2(testdir.dfd, "symlink", 0600, AT_SYMLINK_NOFOLLOW);
+
+	/*
+	 * On certain filesystems (xfs or btrfs), chmod operation fails. So we
+	 * first check the symlink target but if the operation fails we mark the
+	 * test as skipped.
+	 *
+	 * https://sourceware.org/legacy-ml/libc-alpha/2020-02/msg00467.html
+	 */
+	if (ret == 0 && !expect_mode(testdir.dfd, "symlink", 0120600)) {
+		ksft_print_msg("%s: wrong symlink mode bits after fchmodat2 with nofollow\n",
+				__func__);
+		ret = 1;
+		goto err;
+	}
+
+	if (!expect_mode(testdir.dfd, "regfile", 0100640)) {
+		ksft_print_msg("%s: wrong file mode bits after fchmodat2 with nofollow\n",
+			       __func__);
+	}
+
+	if (ret != 0)
+		ksft_test_result_skip("fchmodat2(symlink)\n");
+	else
+		ksft_test_result_pass("fchmodat2(symlink)\n");
+	cleanup_testdir(&testdir);
+	return;
+
+err:
+	ksft_test_result_fail("fchmodat2(symlink)\n");
+	cleanup_testdir(&testdir);
+}
+
+#define NUM_TESTS 2
+
+int main(int argc, char **argv)
+{
+	ksft_print_header();
+	ksft_set_plan(NUM_TESTS);
+
+	test_regfile();
+	test_symlink();
+
+	ksft_finished();
+}