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, 168 insertions, 0 deletions

diff --git a/drivers/scsi/arm/msgqueue.c b/drivers/scsi/arm/msgqueue.c
new file mode 100644
index 000000000..581158313
--- /dev/null
+++ b/drivers/scsi/arm/msgqueue.c
@@ -0,0 +1,168 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ *  linux/drivers/acorn/scsi/msgqueue.c
+ *
+ *  Copyright (C) 1997-1998 Russell King
+ *
+ *  message queue handling
+ */
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/stddef.h>
+#include <linux/init.h>
+
+#include "msgqueue.h"
+
+/*
+ * Function: struct msgqueue_entry *mqe_alloc(MsgQueue_t *msgq)
+ * Purpose : Allocate a message queue entry
+ * Params  : msgq - message queue to claim entry for
+ * Returns : message queue entry or NULL.
+ */
+static struct msgqueue_entry *mqe_alloc(MsgQueue_t *msgq)
+{
+	struct msgqueue_entry *mq;
+
+	if ((mq = msgq->free) != NULL)
+		msgq->free = mq->next;
+
+	return mq;
+}
+
+/*
+ * Function: void mqe_free(MsgQueue_t *msgq, struct msgqueue_entry *mq)
+ * Purpose : free a message queue entry
+ * Params  : msgq - message queue to free entry from
+ *	     mq   - message queue entry to free
+ */
+static void mqe_free(MsgQueue_t *msgq, struct msgqueue_entry *mq)
+{
+	if (mq) {
+		mq->next = msgq->free;
+		msgq->free = mq;
+	}
+}
+
+/*
+ * Function: void msgqueue_initialise(MsgQueue_t *msgq)
+ * Purpose : initialise a message queue
+ * Params  : msgq - queue to initialise
+ */
+void msgqueue_initialise(MsgQueue_t *msgq)
+{
+	int i;
+
+	msgq->qe = NULL;
+	msgq->free = &msgq->entries[0];
+
+	for (i = 0; i < NR_MESSAGES; i++)
+		msgq->entries[i].next = &msgq->entries[i + 1];
+
+	msgq->entries[NR_MESSAGES - 1].next = NULL;
+}
+
+
+/*
+ * Function: void msgqueue_free(MsgQueue_t *msgq)
+ * Purpose : free a queue
+ * Params  : msgq - queue to free
+ */
+void msgqueue_free(MsgQueue_t *msgq)
+{
+}
+
+/*
+ * Function: int msgqueue_msglength(MsgQueue_t *msgq)
+ * Purpose : calculate the total length of all messages on the message queue
+ * Params  : msgq - queue to examine
+ * Returns : number of bytes of messages in queue
+ */
+int msgqueue_msglength(MsgQueue_t *msgq)
+{
+	struct msgqueue_entry *mq = msgq->qe;
+	int length = 0;
+
+	for (mq = msgq->qe; mq; mq = mq->next)
+		length += mq->msg.length;
+
+	return length;
+}
+
+/*
+ * Function: struct message *msgqueue_getmsg(MsgQueue_t *msgq, int msgno)
+ * Purpose : return a message
+ * Params  : msgq   - queue to obtain message from
+ *	   : msgno  - message number
+ * Returns : pointer to message string, or NULL
+ */
+struct message *msgqueue_getmsg(MsgQueue_t *msgq, int msgno)
+{
+	struct msgqueue_entry *mq;
+
+	for (mq = msgq->qe; mq && msgno; mq = mq->next, msgno--);
+
+	return mq ? &mq->msg : NULL;
+}
+
+/*
+ * Function: int msgqueue_addmsg(MsgQueue_t *msgq, int length, ...)
+ * Purpose : add a message onto a message queue
+ * Params  : msgq   - queue to add message on
+ *	     length - length of message
+ *	     ...    - message bytes
+ * Returns : != 0 if successful
+ */
+int msgqueue_addmsg(MsgQueue_t *msgq, int length, ...)
+{
+	struct msgqueue_entry *mq = mqe_alloc(msgq);
+	va_list ap;
+
+	if (mq) {
+		struct msgqueue_entry **mqp;
+		int i;
+
+		va_start(ap, length);
+		for (i = 0; i < length; i++)
+			mq->msg.msg[i] = va_arg(ap, unsigned int);
+		va_end(ap);
+
+		mq->msg.length = length;
+		mq->msg.fifo = 0;
+		mq->next = NULL;
+
+		mqp = &msgq->qe;
+		while (*mqp)
+			mqp = &(*mqp)->next;
+
+		*mqp = mq;
+	}
+
+	return mq != NULL;
+}
+
+/*
+ * Function: void msgqueue_flush(MsgQueue_t *msgq)
+ * Purpose : flush all messages from message queue
+ * Params  : msgq - queue to flush
+ */
+void msgqueue_flush(MsgQueue_t *msgq)
+{
+	struct msgqueue_entry *mq, *mqnext;
+
+	for (mq = msgq->qe; mq; mq = mqnext) {
+		mqnext = mq->next;
+		mqe_free(msgq, mq);
+	}
+	msgq->qe = NULL;
+}
+
+EXPORT_SYMBOL(msgqueue_initialise);
+EXPORT_SYMBOL(msgqueue_free);
+EXPORT_SYMBOL(msgqueue_msglength);
+EXPORT_SYMBOL(msgqueue_getmsg);
+EXPORT_SYMBOL(msgqueue_addmsg);
+EXPORT_SYMBOL(msgqueue_flush);
+
+MODULE_AUTHOR("Russell King");
+MODULE_DESCRIPTION("SCSI message queue handling");
+MODULE_LICENSE("GPL");