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

2 files changed, 236 insertions, 0 deletions

diff --git a/tools/testing/selftests/cpu-hotplug/Makefile b/tools/testing/selftests/cpu-hotplug/Makefile
new file mode 100644
index 000000000..8b66c4738
--- /dev/null
+++ b/tools/testing/selftests/cpu-hotplug/Makefile
@@ -0,0 +1,11 @@
+# SPDX-License-Identifier: GPL-2.0
+all:
+
+TEST_PROGS := cpu-on-off-test.sh
+
+include ../lib.mk
+
+run_full_test:
+	@/bin/bash ./cpu-on-off-test.sh -a && echo "cpu-hotplug selftests: [PASS]" || echo "cpu-hotplug selftests: [FAIL]"
+
+clean:
diff --git a/tools/testing/selftests/cpu-hotplug/cpu-on-off-test.sh b/tools/testing/selftests/cpu-hotplug/cpu-on-off-test.sh
new file mode 100755
index 000000000..287cfd580
--- /dev/null
+++ b/tools/testing/selftests/cpu-hotplug/cpu-on-off-test.sh
@@ -0,0 +1,225 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-2.0
+
+SYSFS=
+# Kselftest framework requirement - SKIP code is 4.
+ksft_skip=4
+retval=0
+
+prerequisite()
+{
+	msg="skip all tests:"
+
+	if [ $UID != 0 ]; then
+		echo $msg must be run as root >&2
+		exit $ksft_skip
+	fi
+
+	taskset -p 01 $$
+
+	SYSFS=`mount -t sysfs | head -1 | awk '{ print $3 }'`
+
+	if [ ! -d "$SYSFS" ]; then
+		echo $msg sysfs is not mounted >&2
+		exit $ksft_skip
+	fi
+
+	if ! ls $SYSFS/devices/system/cpu/cpu*/online > /dev/null 2>&1; then
+		echo $msg cpu hotplug is not supported >&2
+		exit $ksft_skip
+	fi
+
+	echo "CPU online/offline summary:"
+	online_cpus=`cat $SYSFS/devices/system/cpu/online`
+	online_max=${online_cpus##*-}
+
+	if [[ "$online_cpus" = "$online_max" ]]; then
+		echo "$msg: since there is only one cpu: $online_cpus"
+		exit $ksft_skip
+	fi
+
+	present_cpus=`cat $SYSFS/devices/system/cpu/present`
+	present_max=${present_cpus##*-}
+	echo "present_cpus = $present_cpus present_max = $present_max"
+
+	echo -e "\t Cpus in online state: $online_cpus"
+
+	offline_cpus=`cat $SYSFS/devices/system/cpu/offline`
+	if [[ "a$offline_cpus" = "a" ]]; then
+		offline_cpus=0
+	else
+		offline_max=${offline_cpus##*-}
+	fi
+	echo -e "\t Cpus in offline state: $offline_cpus"
+}
+
+#
+# list all hot-pluggable CPUs
+#
+hotpluggable_cpus()
+{
+	local state=${1:-.\*}
+
+	for cpu in $SYSFS/devices/system/cpu/cpu*; do
+		if [ -f $cpu/online ] && grep -q $state $cpu/online; then
+			echo ${cpu##/*/cpu}
+		fi
+	done
+}
+
+hotpluggable_offline_cpus()
+{
+	hotpluggable_cpus 0
+}
+
+hotpluggable_online_cpus()
+{
+	hotpluggable_cpus 1
+}
+
+cpu_is_online()
+{
+	grep -q 1 $SYSFS/devices/system/cpu/cpu$1/online
+}
+
+cpu_is_offline()
+{
+	grep -q 0 $SYSFS/devices/system/cpu/cpu$1/online
+}
+
+online_cpu()
+{
+	echo 1 > $SYSFS/devices/system/cpu/cpu$1/online
+}
+
+offline_cpu()
+{
+	echo 0 > $SYSFS/devices/system/cpu/cpu$1/online
+}
+
+online_cpu_expect_success()
+{
+	local cpu=$1
+
+	if ! online_cpu $cpu; then
+		echo $FUNCNAME $cpu: unexpected fail >&2
+		retval=1
+	elif ! cpu_is_online $cpu; then
+		echo $FUNCNAME $cpu: unexpected offline >&2
+		retval=1
+	fi
+}
+
+online_cpu_expect_fail()
+{
+	local cpu=$1
+
+	if online_cpu $cpu 2> /dev/null; then
+		echo $FUNCNAME $cpu: unexpected success >&2
+		retval=1
+	elif ! cpu_is_offline $cpu; then
+		echo $FUNCNAME $cpu: unexpected online >&2
+		retval=1
+	fi
+}
+
+offline_cpu_expect_success()
+{
+	local cpu=$1
+
+	if ! offline_cpu $cpu; then
+		echo $FUNCNAME $cpu: unexpected fail >&2
+		retval=1
+	elif ! cpu_is_offline $cpu; then
+		echo $FUNCNAME $cpu: unexpected offline >&2
+		retval=1
+	fi
+}
+
+offline_cpu_expect_fail()
+{
+	local cpu=$1
+
+	if offline_cpu $cpu 2> /dev/null; then
+		echo $FUNCNAME $cpu: unexpected success >&2
+		retval=1
+	elif ! cpu_is_online $cpu; then
+		echo $FUNCNAME $cpu: unexpected offline >&2
+		retval=1
+	fi
+}
+
+online_all_hot_pluggable_cpus()
+{
+	for cpu in `hotpluggable_offline_cpus`; do
+		online_cpu_expect_success $cpu
+	done
+}
+
+offline_all_hot_pluggable_cpus()
+{
+	local reserve_cpu=$online_max
+	for cpu in `hotpluggable_online_cpus`; do
+		# Reserve one cpu oneline at least.
+		if [ $cpu -eq $reserve_cpu ];then
+			continue
+		fi
+		offline_cpu_expect_success $cpu
+	done
+}
+
+allcpus=0
+online_cpus=0
+online_max=0
+offline_cpus=0
+offline_max=0
+present_cpus=0
+present_max=0
+
+while getopts ah opt; do
+	case $opt in
+	a)
+		allcpus=1
+		;;
+	h)
+		echo "Usage $0 [ -a ]"
+		echo -e "\t default offline one cpu"
+		echo -e "\t run with -a option to offline all cpus"
+		exit
+		;;
+	esac
+done
+
+prerequisite
+
+#
+# Safe test (default) - offline and online one cpu
+#
+if [ $allcpus -eq 0 ]; then
+	echo "Limited scope test: one hotplug cpu"
+	echo -e "\t (leaves cpu in the original state):"
+	echo -e "\t online to offline to online: cpu $online_max"
+	offline_cpu_expect_success $online_max
+	online_cpu_expect_success $online_max
+
+	if [[ $offline_cpus -gt 0 ]]; then
+		echo -e "\t online to offline to online: cpu $present_max"
+		online_cpu_expect_success $present_max
+		offline_cpu_expect_success $present_max
+		online_cpu $present_max
+	fi
+	exit $retval
+else
+	echo "Full scope test: all hotplug cpus"
+	echo -e "\t online all offline cpus"
+	echo -e "\t offline all online cpus"
+	echo -e "\t online all offline cpus"
+fi
+
+online_all_hot_pluggable_cpus
+
+offline_all_hot_pluggable_cpus
+
+online_all_hot_pluggable_cpus
+
+exit $retval