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, 335 insertions, 0 deletions

diff --git a/lib/test_sysctl.c b/lib/test_sysctl.c
new file mode 100644
index 000000000..909cfcf76
--- /dev/null
+++ b/lib/test_sysctl.c
@@ -0,0 +1,335 @@
+// SPDX-License-Identifier: GPL-2.0-or-later OR copyleft-next-0.3.1
+/*
+ * proc sysctl test driver
+ *
+ * Copyright (C) 2017 Luis R. Rodriguez <mcgrof@kernel.org>
+ */
+
+/*
+ * This module provides an interface to the proc sysctl interfaces.  This
+ * driver requires CONFIG_SYSCTL. It will not normally be loaded by the
+ * system unless explicitly requested by name. You can also build this driver
+ * into your kernel.
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/init.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/printk.h>
+#include <linux/fs.h>
+#include <linux/miscdevice.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+#include <linux/async.h>
+#include <linux/delay.h>
+#include <linux/vmalloc.h>
+
+static int i_zero;
+static int i_one_hundred = 100;
+static int match_int_ok = 1;
+
+enum {
+	TEST_H_SETUP_NODE,
+	TEST_H_MNT,
+	TEST_H_MNTERROR,
+	TEST_H_EMPTY_ADD,
+	TEST_H_EMPTY,
+	TEST_H_U8,
+	TEST_H_SIZE /* Always at the end */
+};
+
+static struct ctl_table_header *ctl_headers[TEST_H_SIZE] = {};
+struct test_sysctl_data {
+	int int_0001;
+	int int_0002;
+	int int_0003[4];
+
+	int boot_int;
+
+	unsigned int uint_0001;
+
+	char string_0001[65];
+
+#define SYSCTL_TEST_BITMAP_SIZE	65536
+	unsigned long *bitmap_0001;
+};
+
+static struct test_sysctl_data test_data = {
+	.int_0001 = 60,
+	.int_0002 = 1,
+
+	.int_0003[0] = 0,
+	.int_0003[1] = 1,
+	.int_0003[2] = 2,
+	.int_0003[3] = 3,
+
+	.boot_int = 0,
+
+	.uint_0001 = 314,
+
+	.string_0001 = "(none)",
+};
+
+/* These are all under /proc/sys/debug/test_sysctl/ */
+static const struct ctl_table test_table[] = {
+	{
+		.procname	= "int_0001",
+		.data		= &test_data.int_0001,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &i_zero,
+		.extra2         = &i_one_hundred,
+	},
+	{
+		.procname	= "int_0002",
+		.data		= &test_data.int_0002,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
+	{
+		.procname	= "int_0003",
+		.data		= &test_data.int_0003,
+		.maxlen		= sizeof(test_data.int_0003),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
+	{
+		.procname	= "match_int",
+		.data		= &match_int_ok,
+		.maxlen		= sizeof(match_int_ok),
+		.mode		= 0444,
+		.proc_handler	= proc_dointvec,
+	},
+	{
+		.procname	= "boot_int",
+		.data		= &test_data.boot_int,
+		.maxlen		= sizeof(test_data.boot_int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+		.extra1		= SYSCTL_ZERO,
+		.extra2         = SYSCTL_ONE,
+	},
+	{
+		.procname	= "uint_0001",
+		.data		= &test_data.uint_0001,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_douintvec,
+	},
+	{
+		.procname	= "string_0001",
+		.data		= &test_data.string_0001,
+		.maxlen		= sizeof(test_data.string_0001),
+		.mode		= 0644,
+		.proc_handler	= proc_dostring,
+	},
+	{
+		.procname	= "bitmap_0001",
+		.data		= &test_data.bitmap_0001,
+		.maxlen		= SYSCTL_TEST_BITMAP_SIZE,
+		.mode		= 0644,
+		.proc_handler	= proc_do_large_bitmap,
+	},
+};
+
+static void test_sysctl_calc_match_int_ok(void)
+{
+	int i;
+
+	struct {
+		int defined;
+		int wanted;
+	} match_int[] = {
+		{.defined = *(int *)SYSCTL_ZERO,	.wanted = 0},
+		{.defined = *(int *)SYSCTL_ONE,		.wanted = 1},
+		{.defined = *(int *)SYSCTL_TWO,		.wanted = 2},
+		{.defined = *(int *)SYSCTL_THREE,	.wanted = 3},
+		{.defined = *(int *)SYSCTL_FOUR,	.wanted = 4},
+		{.defined = *(int *)SYSCTL_ONE_HUNDRED, .wanted = 100},
+		{.defined = *(int *)SYSCTL_TWO_HUNDRED,	.wanted = 200},
+		{.defined = *(int *)SYSCTL_ONE_THOUSAND, .wanted = 1000},
+		{.defined = *(int *)SYSCTL_THREE_THOUSAND, .wanted = 3000},
+		{.defined = *(int *)SYSCTL_INT_MAX,	.wanted = INT_MAX},
+		{.defined = *(int *)SYSCTL_MAXOLDUID,	.wanted = 65535},
+		{.defined = *(int *)SYSCTL_NEG_ONE,	.wanted = -1},
+	};
+
+	for (i = 0; i < ARRAY_SIZE(match_int); i++)
+		if (match_int[i].defined != match_int[i].wanted)
+			match_int_ok = 0;
+}
+
+static int test_sysctl_setup_node_tests(void)
+{
+	test_sysctl_calc_match_int_ok();
+	test_data.bitmap_0001 = kzalloc(SYSCTL_TEST_BITMAP_SIZE/8, GFP_KERNEL);
+	if (!test_data.bitmap_0001)
+		return -ENOMEM;
+	ctl_headers[TEST_H_SETUP_NODE] = register_sysctl("debug/test_sysctl", test_table);
+	if (!ctl_headers[TEST_H_SETUP_NODE]) {
+		kfree(test_data.bitmap_0001);
+		return -ENOMEM;
+	}
+
+	return 0;
+}
+
+/* Used to test that unregister actually removes the directory */
+static const struct ctl_table test_table_unregister[] = {
+	{
+		.procname	= "unregister_error",
+		.data		= &test_data.int_0001,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+	},
+};
+
+static int test_sysctl_run_unregister_nested(void)
+{
+	struct ctl_table_header *unregister;
+
+	unregister = register_sysctl("debug/test_sysctl/unregister_error",
+				   test_table_unregister);
+	if (!unregister)
+		return -ENOMEM;
+
+	unregister_sysctl_table(unregister);
+	return 0;
+}
+
+static int test_sysctl_run_register_mount_point(void)
+{
+	ctl_headers[TEST_H_MNT]
+		= register_sysctl_mount_point("debug/test_sysctl/mnt");
+	if (!ctl_headers[TEST_H_MNT])
+		return -ENOMEM;
+
+	ctl_headers[TEST_H_MNTERROR]
+		= register_sysctl("debug/test_sysctl/mnt/mnt_error",
+				  test_table_unregister);
+	/*
+	 * Don't check the result.:
+	 * If it fails (expected behavior), return 0.
+	 * If successful (missbehavior of register mount point), we want to see
+	 * mnt_error when we run the sysctl test script
+	 */
+
+	return 0;
+}
+
+static const struct ctl_table test_table_empty[] = { };
+
+static int test_sysctl_run_register_empty(void)
+{
+	/* Tets that an empty dir can be created */
+	ctl_headers[TEST_H_EMPTY_ADD]
+		= register_sysctl("debug/test_sysctl/empty_add", test_table_empty);
+	if (!ctl_headers[TEST_H_EMPTY_ADD])
+		return -ENOMEM;
+
+	/* Test that register on top of an empty dir works */
+	ctl_headers[TEST_H_EMPTY]
+		= register_sysctl("debug/test_sysctl/empty_add/empty", test_table_empty);
+	if (!ctl_headers[TEST_H_EMPTY])
+		return -ENOMEM;
+
+	return 0;
+}
+
+static const struct ctl_table table_u8_over[] = {
+	{
+		.procname	= "u8_over",
+		.data		= &test_data.uint_0001,
+		.maxlen		= sizeof(u8),
+		.mode		= 0644,
+		.proc_handler	= proc_dou8vec_minmax,
+		.extra1		= SYSCTL_FOUR,
+		.extra2		= SYSCTL_ONE_THOUSAND,
+	},
+};
+
+static const struct ctl_table table_u8_under[] = {
+	{
+		.procname	= "u8_under",
+		.data		= &test_data.uint_0001,
+		.maxlen		= sizeof(u8),
+		.mode		= 0644,
+		.proc_handler	= proc_dou8vec_minmax,
+		.extra1		= SYSCTL_NEG_ONE,
+		.extra2		= SYSCTL_ONE_HUNDRED,
+	},
+};
+
+static const struct ctl_table table_u8_valid[] = {
+	{
+		.procname	= "u8_valid",
+		.data		= &test_data.uint_0001,
+		.maxlen		= sizeof(u8),
+		.mode		= 0644,
+		.proc_handler	= proc_dou8vec_minmax,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= SYSCTL_TWO_HUNDRED,
+	},
+};
+
+static int test_sysctl_register_u8_extra(void)
+{
+	/* should fail because it's over */
+	ctl_headers[TEST_H_U8]
+		= register_sysctl("debug/test_sysctl", table_u8_over);
+	if (ctl_headers[TEST_H_U8])
+		return -ENOMEM;
+
+	/* should fail because it's under */
+	ctl_headers[TEST_H_U8]
+		= register_sysctl("debug/test_sysctl", table_u8_under);
+	if (ctl_headers[TEST_H_U8])
+		return -ENOMEM;
+
+	/* should not fail because it's valid */
+	ctl_headers[TEST_H_U8]
+		= register_sysctl("debug/test_sysctl", table_u8_valid);
+	if (!ctl_headers[TEST_H_U8])
+		return -ENOMEM;
+
+	return 0;
+}
+
+static int __init test_sysctl_init(void)
+{
+	int err = 0;
+
+	int (*func_array[])(void) = {
+		test_sysctl_setup_node_tests,
+		test_sysctl_run_unregister_nested,
+		test_sysctl_run_register_mount_point,
+		test_sysctl_run_register_empty,
+		test_sysctl_register_u8_extra
+	};
+
+	for (int i = 0; !err && i < ARRAY_SIZE(func_array); i++)
+		err = func_array[i]();
+
+	return err;
+}
+module_init(test_sysctl_init);
+
+static void __exit test_sysctl_exit(void)
+{
+	kfree(test_data.bitmap_0001);
+	for (int i = 0; i < TEST_H_SIZE; i++) {
+		if (ctl_headers[i])
+			unregister_sysctl_table(ctl_headers[i]);
+	}
+}
+
+module_exit(test_sysctl_exit);
+
+MODULE_AUTHOR("Luis R. Rodriguez <mcgrof@kernel.org>");
+MODULE_DESCRIPTION("proc sysctl test driver");
+MODULE_LICENSE("GPL");