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, 118 insertions, 0 deletions

diff --git a/tools/testing/selftests/fpu/.gitignore b/tools/testing/selftests/fpu/.gitignore
new file mode 100644
index 000000000..d6d12ac1d
--- /dev/null
+++ b/tools/testing/selftests/fpu/.gitignore
@@ -0,0 +1,2 @@
+# SPDX-License-Identifier: GPL-2.0+
+test_fpu
diff --git a/tools/testing/selftests/fpu/Makefile b/tools/testing/selftests/fpu/Makefile
new file mode 100644
index 000000000..ea62c176e
--- /dev/null
+++ b/tools/testing/selftests/fpu/Makefile
@@ -0,0 +1,9 @@
+# SPDX-License-Identifier: GPL-2.0+
+
+LDLIBS := -lm
+
+TEST_GEN_PROGS := test_fpu
+
+TEST_PROGS := run_test_fpu.sh
+
+include ../lib.mk
diff --git a/tools/testing/selftests/fpu/run_test_fpu.sh b/tools/testing/selftests/fpu/run_test_fpu.sh
new file mode 100755
index 000000000..d77be93ec
--- /dev/null
+++ b/tools/testing/selftests/fpu/run_test_fpu.sh
@@ -0,0 +1,46 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-2.0
+#
+# Load kernel module for FPU tests
+
+uid=$(id -u)
+if [ $uid -ne 0 ]; then
+	echo "$0: Must be run as root"
+	exit 1
+fi
+
+if ! which modprobe > /dev/null 2>&1; then
+	echo "$0: You need modprobe installed"
+        exit 4
+fi
+
+if ! modinfo test_fpu > /dev/null 2>&1; then
+	echo "$0: You must have the following enabled in your kernel:"
+	echo "CONFIG_TEST_FPU=m"
+	exit 4
+fi
+
+NR_CPUS=$(getconf _NPROCESSORS_ONLN)
+if [ ! $NR_CPUS ]; then
+	NR_CPUS=1
+fi
+
+modprobe test_fpu
+
+if [ ! -e /sys/kernel/debug/selftest_helpers/test_fpu ]; then
+	mount -t debugfs none /sys/kernel/debug
+
+	if [ ! -e /sys/kernel/debug/selftest_helpers/test_fpu ]; then
+		echo "$0: Error mounting debugfs"
+		exit 4
+	fi
+fi
+
+echo "Running 1000 iterations on all CPUs... "
+for i in $(seq 1 1000); do
+	for c in $(seq 1 $NR_CPUS); do
+		./test_fpu &
+	done
+done
+
+rmmod test_fpu
diff --git a/tools/testing/selftests/fpu/test_fpu.c b/tools/testing/selftests/fpu/test_fpu.c
new file mode 100644
index 000000000..200238522
--- /dev/null
+++ b/tools/testing/selftests/fpu/test_fpu.c
@@ -0,0 +1,61 @@
+// SPDX-License-Identifier: GPL-2.0+
+/* This testcase operates with the test_fpu kernel driver.
+ * It modifies the FPU control register in user mode and calls the kernel
+ * module to perform floating point operations in the kernel. The control
+ * register value should be independent between kernel and user mode.
+ */
+
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <errno.h>
+#include <string.h>
+#include <fenv.h>
+#include <unistd.h>
+#include <fcntl.h>
+
+const char *test_fpu_path = "/sys/kernel/debug/selftest_helpers/test_fpu";
+
+int main(void)
+{
+	char dummy[1];
+	int fd = open(test_fpu_path, O_RDONLY);
+
+	if (fd < 0) {
+		printf("[SKIP]\tcan't access %s: %s\n",
+		       test_fpu_path, strerror(errno));
+		return 0;
+	}
+
+	if (read(fd, dummy, 1) < 0) {
+		printf("[FAIL]\taccess with default rounding mode failed\n");
+		return 1;
+	}
+
+	fesetround(FE_DOWNWARD);
+	if (read(fd, dummy, 1) < 0) {
+		printf("[FAIL]\taccess with downward rounding mode failed\n");
+		return 2;
+	}
+	if (fegetround() != FE_DOWNWARD) {
+		printf("[FAIL]\tusermode rounding mode clobbered\n");
+		return 3;
+	}
+
+	/* Note: the tests up to this point are quite safe and will only return
+	 * an error. But the exception mask setting can cause misbehaving kernel
+	 * to crash.
+	 */
+	feclearexcept(FE_ALL_EXCEPT);
+	feenableexcept(FE_ALL_EXCEPT);
+	if (read(fd, dummy, 1) < 0) {
+		printf("[FAIL]\taccess with fpu exceptions unmasked failed\n");
+		return 4;
+	}
+	if (fegetexcept() != FE_ALL_EXCEPT) {
+		printf("[FAIL]\tusermode fpu exception mask clobbered\n");
+		return 5;
+	}
+
+	printf("[OK]\ttest_fpu\n");
+	return 0;
+}