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, 179 insertions, 0 deletions

diff --git a/drivers/char/hangcheck-timer.c b/drivers/char/hangcheck-timer.c
new file mode 100644
index 000000000..231cbf7b3
--- /dev/null
+++ b/drivers/char/hangcheck-timer.c
@@ -0,0 +1,179 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * hangcheck-timer.c
+ *
+ * Driver for a little io fencing timer.
+ *
+ * Copyright (C) 2002, 2003 Oracle.  All rights reserved.
+ *
+ * Author: Joel Becker <joel.becker@oracle.com>
+ */
+
+/*
+ * The hangcheck-timer driver uses the TSC to catch delays that
+ * jiffies does not notice.  A timer is set.  When the timer fires, it
+ * checks whether it was delayed and if that delay exceeds a given
+ * margin of error.  The hangcheck_tick module parameter takes the timer
+ * duration in seconds.  The hangcheck_margin parameter defines the
+ * margin of error, in seconds.  The defaults are 60 seconds for the
+ * timer and 180 seconds for the margin of error.  IOW, a timer is set
+ * for 60 seconds.  When the timer fires, the callback checks the
+ * actual duration that the timer waited.  If the duration exceeds the
+ * allotted time and margin (here 60 + 180, or 240 seconds), the machine
+ * is restarted.  A healthy machine will have the duration match the
+ * expected timeout very closely.
+ */
+
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/reboot.h>
+#include <linux/init.h>
+#include <linux/delay.h>
+#include <linux/uaccess.h>
+#include <linux/sysrq.h>
+#include <linux/timer.h>
+#include <linux/hrtimer.h>
+
+#define VERSION_STR "0.9.1"
+
+#define DEFAULT_IOFENCE_MARGIN 60	/* Default fudge factor, in seconds */
+#define DEFAULT_IOFENCE_TICK 180	/* Default timer timeout, in seconds */
+
+static int hangcheck_tick = DEFAULT_IOFENCE_TICK;
+static int hangcheck_margin = DEFAULT_IOFENCE_MARGIN;
+static int hangcheck_reboot;  /* Defaults to not reboot */
+static int hangcheck_dump_tasks;  /* Defaults to not dumping SysRQ T */
+
+/* options - modular */
+module_param(hangcheck_tick, int, 0);
+MODULE_PARM_DESC(hangcheck_tick, "Timer delay.");
+module_param(hangcheck_margin, int, 0);
+MODULE_PARM_DESC(hangcheck_margin, "If the hangcheck timer has been delayed more than hangcheck_margin seconds, the driver will fire.");
+module_param(hangcheck_reboot, int, 0);
+MODULE_PARM_DESC(hangcheck_reboot, "If nonzero, the machine will reboot when the timer margin is exceeded.");
+module_param(hangcheck_dump_tasks, int, 0);
+MODULE_PARM_DESC(hangcheck_dump_tasks, "If nonzero, the machine will dump the system task state when the timer margin is exceeded.");
+
+MODULE_AUTHOR("Oracle");
+MODULE_DESCRIPTION("Hangcheck-timer detects when the system has gone out to lunch past a certain margin.");
+MODULE_LICENSE("GPL");
+MODULE_VERSION(VERSION_STR);
+
+/* options - nonmodular */
+#ifndef MODULE
+
+static int __init hangcheck_parse_tick(char *str)
+{
+	int par;
+
+	if (get_option(&str, &par))
+		hangcheck_tick = par;
+	return 1;
+}
+
+static int __init hangcheck_parse_margin(char *str)
+{
+	int par;
+
+	if (get_option(&str, &par))
+		hangcheck_margin = par;
+	return 1;
+}
+
+static int __init hangcheck_parse_reboot(char *str)
+{
+	int par;
+
+	if (get_option(&str, &par))
+		hangcheck_reboot = par;
+	return 1;
+}
+
+static int __init hangcheck_parse_dump_tasks(char *str)
+{
+	int par;
+
+	if (get_option(&str, &par))
+		hangcheck_dump_tasks = par;
+	return 1;
+}
+
+__setup("hcheck_tick", hangcheck_parse_tick);
+__setup("hcheck_margin", hangcheck_parse_margin);
+__setup("hcheck_reboot", hangcheck_parse_reboot);
+__setup("hcheck_dump_tasks", hangcheck_parse_dump_tasks);
+#endif /* not MODULE */
+
+#define TIMER_FREQ 1000000000ULL
+
+/* Last time scheduled */
+static unsigned long long hangcheck_tsc, hangcheck_tsc_margin;
+
+static void hangcheck_fire(struct timer_list *);
+
+static DEFINE_TIMER(hangcheck_ticktock, hangcheck_fire);
+
+static void hangcheck_fire(struct timer_list *unused)
+{
+	unsigned long long cur_tsc, tsc_diff;
+
+	cur_tsc = ktime_get_ns();
+
+	if (cur_tsc > hangcheck_tsc)
+		tsc_diff = cur_tsc - hangcheck_tsc;
+	else
+		tsc_diff = (cur_tsc + (~0ULL - hangcheck_tsc)); /* or something */
+
+	if (tsc_diff > hangcheck_tsc_margin) {
+		if (hangcheck_dump_tasks) {
+			pr_crit("Hangcheck: Task state:\n");
+#ifdef CONFIG_MAGIC_SYSRQ
+			handle_sysrq('t');
+#endif  /* CONFIG_MAGIC_SYSRQ */
+		}
+		if (hangcheck_reboot) {
+			pr_crit("Hangcheck: hangcheck is restarting the machine.\n");
+			emergency_restart();
+		} else {
+			pr_crit("Hangcheck: hangcheck value past margin!\n");
+		}
+	}
+#if 0
+	/*
+	 * Enable to investigate delays in detail
+	 */
+	pr_debug("Hangcheck: called %lld ns since last time (%lld ns overshoot)\n",
+			tsc_diff, tsc_diff - hangcheck_tick*TIMER_FREQ);
+#endif
+	mod_timer(&hangcheck_ticktock, jiffies + (hangcheck_tick*HZ));
+	hangcheck_tsc = ktime_get_ns();
+}
+
+
+static int __init hangcheck_init(void)
+{
+	pr_debug("Hangcheck: starting hangcheck timer %s (tick is %d seconds, margin is %d seconds).\n",
+	       VERSION_STR, hangcheck_tick, hangcheck_margin);
+	hangcheck_tsc_margin =
+		(unsigned long long)hangcheck_margin + hangcheck_tick;
+	hangcheck_tsc_margin *= TIMER_FREQ;
+
+	hangcheck_tsc = ktime_get_ns();
+	mod_timer(&hangcheck_ticktock, jiffies + (hangcheck_tick*HZ));
+
+	return 0;
+}
+
+
+static void __exit hangcheck_exit(void)
+{
+	timer_delete_sync(&hangcheck_ticktock);
+	pr_debug("Hangcheck: Stopped hangcheck timer.\n");
+}
+
+module_init(hangcheck_init);
+module_exit(hangcheck_exit);