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, 76 insertions, 0 deletions

diff --git a/scripts/misc-check b/scripts/misc-check
new file mode 100755
index 000000000..40e5a4b01
--- /dev/null
+++ b/scripts/misc-check
@@ -0,0 +1,76 @@
+#!/bin/sh
+# SPDX-License-Identifier: GPL-2.0-only
+
+set -e
+
+# Detect files that are tracked but ignored by git.
+check_tracked_ignored_files () {
+	git -C "${srctree:-.}" ls-files -i -c --exclude-per-directory=.gitignore 2>/dev/null |
+		sed 's/$/: warning: ignored by one of the .gitignore files/' >&2
+}
+
+# Check for missing #include <linux/export.h>
+#
+# The rule for including <linux/export.h> is very simple:
+# Include <linux/export.h> only when you use EXPORT_SYMBOL(). That's it.
+#
+# However, some headers include <linux/export.h> even though they are completely
+# unrelated to EXPORT_SYMBOL().
+#
+# One example is include/linux/module.h. Please note <linux/module.h> and
+# <linux/export.h> are orthogonal. <linux/module.h> should be included by files
+# that can be compiled as modules. In other words, <linux/module.h> should be
+# included by EXPORT_SYMBOL consumers. In contrast, <linux/export.h> should be
+# included from EXPORT_SYMBOL providers, which may or may not be modular.
+# Hence, include/linux/module.h should *not* include <linux/export.h>.
+#
+# Another example is include/linux/linkage.h, which is completely unrelated to
+# EXPORT_SYMBOL(). Worse, it is included by most C files, which means, most C
+# files end up including <linux/export.h>, even though only some of them
+# actually export symbols. Hence, include/linux/linkage.h should *not* include
+# <linux/export.h>.
+#
+# Before fixing such headers, we must ensure that C files using EXPORT_SYMBOL()
+# include <linux/export.h> directly, since many C files currently rely on
+# <linux/export.h> being included indirectly (likely, via <linux/linkage> etc.).
+#
+# Therefore, this check.
+#
+# The problem is simple - the warned files use EXPORT_SYMBOL(), but do not
+# include <linux/export.h>. Please add #include <linux/export.h> to them.
+#
+# If the included headers are sorted alphabetically, please insert
+# <linux/export.h> in the appropriate position to maintain the sort order.
+# For this reason, this script only checks missing <linux/export.h>, but
+# does not automatically fix it.
+check_missing_include_linux_export_h () {
+
+	git -C "${srctree:-.}" grep --files-with-matches -E 'EXPORT_SYMBOL((_NS)?(_GPL)?|_FOR_MODULES)\(.*\)' \
+	    -- '*.[ch]' :^tools/ :^include/linux/export.h |
+	xargs -r git -C "${srctree:-.}" grep --files-without-match '#include[[:space:]]*<linux/export\.h>' |
+	xargs -r printf "%s: warning: EXPORT_SYMBOL() is used, but #include <linux/export.h> is missing\n" >&2
+}
+
+# If you do not use EXPORT_SYMBOL(), please do not include <linux/export.h>.
+# Currently, this is checked for *.c files, but not for *.h files, because some
+# *.c files rely on <linux/export.h> being included indirectly.
+check_unnecessary_include_linux_export_h () {
+
+	git -C "${srctree:-.}" grep --files-with-matches '#include[[:space:]]*<linux/export\.h>' \
+	    -- '*.[c]' :^tools/ |
+	xargs -r git -C "${srctree:-.}" grep --files-without-match -E 'EXPORT_SYMBOL((_NS)?(_GPL)?|_FOR_MODULES)\(.*\)' |
+	xargs -r printf "%s: warning: EXPORT_SYMBOL() is not used, but #include <linux/export.h> is present\n" >&2
+}
+
+case "${KBUILD_EXTRA_WARN}" in
+*1*)
+	check_tracked_ignored_files
+	;;
+esac
+
+case "${KBUILD_EXTRA_WARN}" in
+*2*)
+	check_missing_include_linux_export_h
+	check_unnecessary_include_linux_export_h
+	;;
+esac