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

1 files changed, 97 insertions, 0 deletions

diff --git a/scripts/package/mkspec b/scripts/package/mkspec
new file mode 100755
index 000000000..c604f8c17
--- /dev/null
+++ b/scripts/package/mkspec
@@ -0,0 +1,97 @@
+#!/bin/sh
+#
+#	Output a simple RPM spec file.
+#	This version assumes a minimum of RPM 4.13
+#
+#	The only gothic bit here is redefining install_post to avoid
+#	stripping the symbols from files in the kernel which we want
+#
+#	Patched for non-x86 by Opencon (L) 2002 <opencon@rio.skydome.net>
+#
+
+set -eu
+
+output=$1
+
+mkdir -p "$(dirname "${output}")"
+
+exec >"${output}"
+
+if grep -q CONFIG_MODULES=y include/config/auto.conf; then
+echo '%define with_devel %{?_without_devel: 0} %{?!_without_devel: 1}'
+else
+echo '%define with_devel 0'
+fi
+
+# use %{debug_package} machinery to generate -debuginfo
+with_debuginfo_rpm=0
+# manually generate -debuginfo package
+with_debuginfo_manual=0
+# debuginfo package generation uses find-debuginfo.sh under the hood,
+# which only works on uncompressed modules that contain debuginfo
+if grep -q CONFIG_DEBUG_INFO=y include/config/auto.conf &&
+   (! grep -q CONFIG_MODULE_COMPRESS=y include/config/auto.conf) &&
+   (! grep -q CONFIG_DEBUG_INFO_SPLIT=y include/config/auto.conf); then
+	# If module signing is enabled (which may be required to boot with
+	# lockdown enabled), the find-debuginfo.sh machinery cannot be used
+	# because the signatures will be stripped off the modules. However, due
+	# to an rpm bug in versions prior to 4.20.0
+	#
+	#     https://github.com/rpm-software-management/rpm/issues/3057
+	#     https://github.com/rpm-software-management/rpm/commit/49f906998f3cf1f4152162ca61ac0869251c380f
+	#
+	# We cannot provide our own debuginfo package because it does not listen
+	# to our custom files list, failing the build due to unpackaged files.
+	# Manually generate the debug info package if using rpm 4.20.0. If not
+	# using rpm 4.20.0, avoid generating a -debuginfo package altogether,
+	# as it is not safe.
+	if grep -q CONFIG_MODULE_SIG=y include/config/auto.conf; then
+		rpm_ver_str=$(rpm --version 2>/dev/null)
+		# Split the version on spaces
+		IFS=' '
+		set -- $rpm_ver_str
+		if [ "${1:-}" = RPM -a "${2:-}" = version ]; then
+			IFS=.
+			set -- $3
+			rpm_ver=$(( 1000000 * $1 + 10000 * $2 + 100 * $3 + ${4:-0} ))
+			if [ "$rpm_ver" -ge 4200000 ]; then
+				with_debuginfo_manual='%{?_without_debuginfo:0}%{?!_without_debuginfo:1}'
+			fi
+		fi
+	else
+		with_debuginfo_rpm='%{?_without_debuginfo:0}%{?!_without_debuginfo:1}'
+	fi
+fi
+echo "%define with_debuginfo_manual $with_debuginfo_manual"
+echo "%define with_debuginfo_rpm $with_debuginfo_rpm"
+
+cat<<EOF
+%define ARCH ${ARCH}
+%define KERNELRELEASE ${KERNELRELEASE}
+%define pkg_release $("${srctree}/scripts/build-version")
+EOF
+
+cat "${srctree}/scripts/package/kernel.spec"
+
+# collect the user's name and email address for the changelog entry
+if [ "$(command -v git)" ]; then
+	name=$(git config user.name) || true
+	email=$(git config user.email) || true
+fi
+
+if [ ! "${name:+set}" ]; then
+	name=${KBUILD_BUILD_USER:-$(id -nu)}
+fi
+
+if [ ! "${email:+set}" ]; then
+	buildhost=${KBUILD_BUILD_HOST:-$(hostname -f 2>/dev/null || hostname)}
+	builduser=${KBUILD_BUILD_USER:-$(id -nu)}
+	email="${builduser}@${buildhost}"
+fi
+
+cat << EOF
+
+%changelog
+* $(LC_ALL=C date +'%a %b %d %Y') ${name} <${email}>
+- Custom built Linux kernel.
+EOF