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, 140 insertions, 0 deletions

diff --git a/include/sound/timer.h b/include/sound/timer.h
new file mode 100644
index 000000000..e549eab71
--- /dev/null
+++ b/include/sound/timer.h
@@ -0,0 +1,140 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+#ifndef __SOUND_TIMER_H
+#define __SOUND_TIMER_H
+
+/*
+ *  Timer abstract layer
+ *  Copyright (c) by Jaroslav Kysela <perex@perex.cz>,
+ *		     Abramo Bagnara <abramo@alsa-project.org>
+ */
+
+#include <sound/asound.h>
+#include <linux/interrupt.h>
+
+#define snd_timer_chip(timer) ((timer)->private_data)
+
+#define SNDRV_TIMER_DEVICES	16
+
+#define SNDRV_TIMER_DEV_FLG_PCM	0x10000000
+
+#define SNDRV_TIMER_HW_AUTO	0x00000001	/* auto trigger is supported */
+#define SNDRV_TIMER_HW_STOP	0x00000002	/* call stop before start */
+#define SNDRV_TIMER_HW_SLAVE	0x00000004	/* only slave timer (variable resolution) */
+#define SNDRV_TIMER_HW_FIRST	0x00000008	/* first tick can be incomplete */
+#define SNDRV_TIMER_HW_WORK	0x00000010	/* timer is called from work */
+
+#define SNDRV_TIMER_IFLG_SLAVE	  0x00000001
+#define SNDRV_TIMER_IFLG_RUNNING  0x00000002
+#define SNDRV_TIMER_IFLG_START	  0x00000004
+#define SNDRV_TIMER_IFLG_AUTO	  0x00000008	/* auto restart */
+#define SNDRV_TIMER_IFLG_FAST	  0x00000010	/* fast callback (do not use work) */
+#define SNDRV_TIMER_IFLG_CALLBACK 0x00000020	/* timer callback is active */
+#define SNDRV_TIMER_IFLG_EXCLUSIVE 0x00000040	/* exclusive owner - no more instances */
+#define SNDRV_TIMER_IFLG_EARLY_EVENT 0x00000080	/* write early event to the poll queue */
+
+#define SNDRV_TIMER_FLG_CHANGE	0x00000001
+#define SNDRV_TIMER_FLG_RESCHED	0x00000002	/* need reschedule */
+
+struct snd_timer;
+
+struct snd_timer_hardware {
+	/* -- must be filled with low-level driver */
+	unsigned int flags;		/* various flags */
+	unsigned long resolution;	/* average timer resolution for one tick in nsec */
+	unsigned long resolution_min;	/* minimal resolution */
+	unsigned long resolution_max;	/* maximal resolution */
+	unsigned long ticks;		/* max timer ticks per interrupt */
+	/* -- low-level functions -- */
+	int (*open) (struct snd_timer * timer);
+	int (*close) (struct snd_timer * timer);
+	unsigned long (*c_resolution) (struct snd_timer * timer);
+	int (*start) (struct snd_timer * timer);
+	int (*stop) (struct snd_timer * timer);
+	int (*set_period) (struct snd_timer * timer, unsigned long period_num, unsigned long period_den);
+	int (*precise_resolution) (struct snd_timer * timer, unsigned long *num, unsigned long *den);
+};
+
+struct snd_timer {
+	int tmr_class;
+	struct snd_card *card;
+	struct module *module;
+	int tmr_device;
+	int tmr_subdevice;
+	char id[64];
+	char name[80];
+	unsigned int flags;
+	int running;			/* running instances */
+	unsigned long sticks;		/* schedule ticks */
+	void *private_data;
+	void (*private_free) (struct snd_timer *timer);
+	struct snd_timer_hardware hw;
+	spinlock_t lock;
+	struct list_head device_list;
+	struct list_head open_list_head;
+	struct list_head active_list_head;
+	struct list_head ack_list_head;
+	struct list_head sack_list_head; /* slow ack list head */
+	struct work_struct task_work;
+	struct kref kref;
+	int max_instances;	/* upper limit of timer instances */
+	int num_instances;	/* current number of timer instances */
+};
+
+struct snd_timer_instance {
+	struct snd_timer *timer;
+	char *owner;
+	unsigned int flags;
+	void *private_data;
+	void (*private_free) (struct snd_timer_instance *ti);
+	void (*callback) (struct snd_timer_instance *timeri,
+			  unsigned long ticks, unsigned long resolution);
+	void (*ccallback) (struct snd_timer_instance * timeri,
+			   int event,
+			   struct timespec64 * tstamp,
+			   unsigned long resolution);
+	void (*disconnect)(struct snd_timer_instance *timeri);
+	void *callback_data;
+	unsigned long ticks;		/* auto-load ticks when expired */
+	unsigned long cticks;		/* current ticks */
+	unsigned long pticks;		/* accumulated ticks for callback */
+	unsigned long resolution;	/* current resolution for work */
+	unsigned long lost;		/* lost ticks */
+	int slave_class;
+	unsigned int slave_id;
+	struct list_head open_list;
+	struct list_head active_list;
+	struct list_head master_list;
+	struct list_head ack_list;
+	struct list_head slave_list_head;
+	struct list_head slave_active_head;
+	struct snd_timer_instance *master;
+};
+
+/*
+ *  Registering
+ */
+
+int snd_timer_new(struct snd_card *card, char *id, struct snd_timer_id *tid, struct snd_timer **rtimer);
+void snd_timer_notify(struct snd_timer *timer, int event, struct timespec64 *tstamp);
+int snd_timer_global_new(char *id, int device, struct snd_timer **rtimer);
+int snd_timer_global_free(struct snd_timer *timer);
+int snd_timer_global_register(struct snd_timer *timer);
+
+struct snd_timer_instance *snd_timer_instance_new(const char *owner);
+void snd_timer_instance_free(struct snd_timer_instance *timeri);
+int snd_timer_open(struct snd_timer_instance *timeri, struct snd_timer_id *tid, unsigned int slave_id);
+void snd_timer_close(struct snd_timer_instance *timeri);
+unsigned long snd_timer_resolution(struct snd_timer_instance *timeri);
+int snd_timer_start(struct snd_timer_instance *timeri, unsigned int ticks);
+int snd_timer_stop(struct snd_timer_instance *timeri);
+int snd_timer_continue(struct snd_timer_instance *timeri);
+int snd_timer_pause(struct snd_timer_instance *timeri);
+
+void snd_timer_interrupt(struct snd_timer *timer, unsigned long ticks_left);
+
+struct snd_timer *snd_timeri_timer_get(struct snd_timer_instance *timeri);
+void snd_timeri_timer_put(struct snd_timer *timer);
+
+DEFINE_FREE(snd_timeri_timer, struct snd_timer *, if (_T) snd_timeri_timer_put(_T))
+
+#endif /* __SOUND_TIMER_H */