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

8 files changed, 217 insertions, 0 deletions

diff --git a/tools/testing/selftests/splice/.gitignore b/tools/testing/selftests/splice/.gitignore
new file mode 100644
index 000000000..be8266f5d
--- /dev/null
+++ b/tools/testing/selftests/splice/.gitignore
@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: GPL-2.0-only
+default_file_splice_read
+splice_read
diff --git a/tools/testing/selftests/splice/Makefile b/tools/testing/selftests/splice/Makefile
new file mode 100644
index 000000000..541cd826d
--- /dev/null
+++ b/tools/testing/selftests/splice/Makefile
@@ -0,0 +1,5 @@
+# SPDX-License-Identifier: GPL-2.0
+TEST_PROGS := default_file_splice_read.sh short_splice_read.sh
+TEST_GEN_PROGS_EXTENDED := default_file_splice_read splice_read
+
+include ../lib.mk
diff --git a/tools/testing/selftests/splice/config b/tools/testing/selftests/splice/config
new file mode 100644
index 000000000..058c92836
--- /dev/null
+++ b/tools/testing/selftests/splice/config
@@ -0,0 +1 @@
+CONFIG_TEST_LKM=m
diff --git a/tools/testing/selftests/splice/default_file_splice_read.c b/tools/testing/selftests/splice/default_file_splice_read.c
new file mode 100644
index 000000000..a3c6e5672
--- /dev/null
+++ b/tools/testing/selftests/splice/default_file_splice_read.c
@@ -0,0 +1,9 @@
+// SPDX-License-Identifier: GPL-2.0
+#define _GNU_SOURCE
+#include <fcntl.h>
+
+int main(int argc, char **argv)
+{
+        splice(0, 0, 1, 0, 1<<30, 0);
+	return 0;
+}
diff --git a/tools/testing/selftests/splice/default_file_splice_read.sh b/tools/testing/selftests/splice/default_file_splice_read.sh
new file mode 100755
index 000000000..490db5a2e
--- /dev/null
+++ b/tools/testing/selftests/splice/default_file_splice_read.sh
@@ -0,0 +1,8 @@
+#!/bin/sh
+# SPDX-License-Identifier: GPL-2.0
+n=`./default_file_splice_read </dev/null | wc -c`
+
+test "$n" = 0 && exit 0
+
+echo "default_file_splice_read broken: leaked $n"
+exit 1
diff --git a/tools/testing/selftests/splice/settings b/tools/testing/selftests/splice/settings
new file mode 100644
index 000000000..89cedfc0d
--- /dev/null
+++ b/tools/testing/selftests/splice/settings
@@ -0,0 +1 @@
+timeout=5
diff --git a/tools/testing/selftests/splice/short_splice_read.sh b/tools/testing/selftests/splice/short_splice_read.sh
new file mode 100755
index 000000000..4710e09f4
--- /dev/null
+++ b/tools/testing/selftests/splice/short_splice_read.sh
@@ -0,0 +1,133 @@
+#!/bin/sh
+# SPDX-License-Identifier: GPL-2.0
+#
+# Test for mishandling of splice() on pseudofilesystems, which should catch
+# bugs like 11990a5bd7e5 ("module: Correctly truncate sysfs sections output")
+#
+# Since splice fallback was removed as part of the set_fs() rework, many of these
+# tests expect to fail now. See https://lore.kernel.org/lkml/202009181443.C2179FB@keescook/
+set -e
+
+DIR=$(dirname "$0")
+
+ret=0
+
+expect_success()
+{
+	title="$1"
+	shift
+
+	echo "" >&2
+	echo "$title ..." >&2
+
+	set +e
+	"$@"
+	rc=$?
+	set -e
+
+	case "$rc" in
+	0)
+		echo "ok: $title succeeded" >&2
+		;;
+	1)
+		echo "FAIL: $title should work" >&2
+		ret=$(( ret + 1 ))
+		;;
+	*)
+		echo "FAIL: something else went wrong" >&2
+		ret=$(( ret + 1 ))
+		;;
+	esac
+}
+
+expect_failure()
+{
+	title="$1"
+	shift
+
+	echo "" >&2
+	echo "$title ..." >&2
+
+	set +e
+	"$@"
+	rc=$?
+	set -e
+
+	case "$rc" in
+	0)
+		echo "FAIL: $title unexpectedly worked" >&2
+		ret=$(( ret + 1 ))
+		;;
+	1)
+		echo "ok: $title correctly failed" >&2
+		;;
+	*)
+		echo "FAIL: something else went wrong" >&2
+		ret=$(( ret + 1 ))
+		;;
+	esac
+}
+
+do_splice()
+{
+	filename="$1"
+	bytes="$2"
+	expected="$3"
+	report="$4"
+
+	out=$("$DIR"/splice_read "$filename" "$bytes" | cat)
+	if [ "$out" = "$expected" ] ; then
+		echo "      matched $report" >&2
+		return 0
+	else
+		echo "      no match: '$out' vs $report" >&2
+		return 1
+	fi
+}
+
+test_splice()
+{
+	filename="$1"
+
+	echo "  checking $filename ..." >&2
+
+	full=$(cat "$filename")
+	rc=$?
+	if [ $rc -ne 0 ] ; then
+		return 2
+	fi
+
+	two=$(echo "$full" | grep -m1 . | cut -c-2)
+
+	# Make sure full splice has the same contents as a standard read.
+	echo "    splicing 4096 bytes ..." >&2
+	if ! do_splice "$filename" 4096 "$full" "full read" ; then
+		return 1
+	fi
+
+	# Make sure a partial splice see the first two characters.
+	echo "    splicing 2 bytes ..." >&2
+	if ! do_splice "$filename" 2 "$two" "'$two'" ; then
+		return 1
+	fi
+
+	return 0
+}
+
+### /proc/$pid/ has no splice interface; these should all fail.
+expect_failure "proc_single_open(), seq_read() splice" test_splice /proc/$$/limits
+expect_failure "special open(), seq_read() splice" test_splice /proc/$$/comm
+
+### /proc/sys/ has a splice interface; these should all succeed.
+expect_success "proc_handler: proc_dointvec_minmax() splice" test_splice /proc/sys/fs/nr_open
+expect_success "proc_handler: proc_dostring() splice" test_splice /proc/sys/kernel/modprobe
+expect_success "proc_handler: special read splice" test_splice /proc/sys/kernel/version
+
+### /sys/ has no splice interface; these should all fail.
+if ! [ -d /sys/module/test_module/sections ] ; then
+	expect_success "test_module kernel module load" modprobe test_module
+fi
+expect_success "kernfs attr splice" test_splice /sys/module/test_module/coresize
+expect_success "kernfs binattr splice" test_splice /sys/module/test_module/sections/.init.text
+
+exit $ret
diff --git a/tools/testing/selftests/splice/splice_read.c b/tools/testing/selftests/splice/splice_read.c
new file mode 100644
index 000000000..46dae6a25
--- /dev/null
+++ b/tools/testing/selftests/splice/splice_read.c
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: GPL-2.0
+#define _GNU_SOURCE
+#include <errno.h>
+#include <fcntl.h>
+#include <limits.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+
+int main(int argc, char *argv[])
+{
+	int fd;
+	size_t size;
+	ssize_t spliced;
+
+	if (argc < 2) {
+		fprintf(stderr, "Usage: %s INPUT [BYTES]\n", argv[0]);
+		return EXIT_FAILURE;
+	}
+
+	fd = open(argv[1], O_RDONLY);
+	if (fd < 0) {
+		perror(argv[1]);
+		return EXIT_FAILURE;
+	}
+
+	if (argc == 3)
+		size = atol(argv[2]);
+	else {
+		struct stat statbuf;
+
+		if (fstat(fd, &statbuf) < 0) {
+			perror(argv[1]);
+			return EXIT_FAILURE;
+		}
+
+		if (statbuf.st_size > INT_MAX) {
+			fprintf(stderr, "%s: Too big\n", argv[1]);
+			return EXIT_FAILURE;
+		}
+
+		size = statbuf.st_size;
+	}
+
+	/* splice(2) file to stdout. */
+	spliced = splice(fd, NULL, STDOUT_FILENO, NULL,
+		      size, SPLICE_F_MOVE);
+	if (spliced < 0) {
+		perror("splice");
+		return EXIT_FAILURE;
+	}
+
+	close(fd);
+	return EXIT_SUCCESS;
+}