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, 138 insertions, 0 deletions

diff --git a/drivers/md/dm-vdo/wait-queue.h b/drivers/md/dm-vdo/wait-queue.h
new file mode 100644
index 000000000..7e8ee6afe
--- /dev/null
+++ b/drivers/md/dm-vdo/wait-queue.h
@@ -0,0 +1,138 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright 2023 Red Hat
+ */
+
+#ifndef VDO_WAIT_QUEUE_H
+#define VDO_WAIT_QUEUE_H
+
+#include <linux/compiler.h>
+#include <linux/types.h>
+
+/**
+ * A vdo_wait_queue is a circular singly linked list of entries waiting to be notified
+ * of a change in a condition. Keeping a circular list allows the vdo_wait_queue
+ * structure to simply be a pointer to the tail (newest) entry, supporting
+ * constant-time enqueue and dequeue operations. A null pointer is an empty waitq.
+ *
+ *   An empty waitq:
+ *     waitq0.last_waiter -> NULL
+ *
+ *   A singleton waitq:
+ *     waitq1.last_waiter -> entry1 -> entry1 -> [...]
+ *
+ *   A three-element waitq:
+ *     waitq2.last_waiter -> entry3 -> entry1 -> entry2 -> entry3 -> [...]
+ *
+ *   linux/wait.h's wait_queue_head is _not_ used because vdo_wait_queue's
+ *   interface is much less complex (doesn't need locking, priorities or timers).
+ *   Made possible by vdo's thread-based resource allocation and locking; and
+ *   the polling nature of vdo_wait_queue consumers.
+ *
+ *   FIXME: could be made to use a linux/list.h's list_head but its extra barriers
+ *   really aren't needed. Nor is a doubly linked list, but vdo_wait_queue could
+ *   make use of __list_del_clearprev() -- but that would compromise the ability
+ *   to make full use of linux's list interface.
+ */
+
+struct vdo_waiter;
+
+struct vdo_wait_queue {
+	/* The tail of the queue, the last (most recently added) entry */
+	struct vdo_waiter *last_waiter;
+	/* The number of waiters currently in the queue */
+	size_t length;
+};
+
+/**
+ * vdo_waiter_callback_fn - Callback type that will be called to resume processing
+ *                          of a waiter after it has been removed from its wait queue.
+ */
+typedef void (*vdo_waiter_callback_fn)(struct vdo_waiter *waiter, void *context);
+
+/**
+ * vdo_waiter_match_fn - Method type for waiter matching methods.
+ *
+ * Returns false if the waiter does not match.
+ */
+typedef bool (*vdo_waiter_match_fn)(struct vdo_waiter *waiter, void *context);
+
+/* The structure for entries in a vdo_wait_queue. */
+struct vdo_waiter {
+	/*
+	 * The next waiter in the waitq. If this entry is the last waiter, then this
+	 * is actually a pointer back to the head of the waitq.
+	 */
+	struct vdo_waiter *next_waiter;
+
+	/* Optional waiter-specific callback to invoke when dequeuing this waiter. */
+	vdo_waiter_callback_fn callback;
+};
+
+/**
+ * vdo_waiter_is_waiting() - Check whether a waiter is waiting.
+ * @waiter: The waiter to check.
+ *
+ * Return: true if the waiter is on some vdo_wait_queue.
+ */
+static inline bool vdo_waiter_is_waiting(struct vdo_waiter *waiter)
+{
+	return (waiter->next_waiter != NULL);
+}
+
+/**
+ * vdo_waitq_init() - Initialize a vdo_wait_queue.
+ * @waitq: The vdo_wait_queue to initialize.
+ */
+static inline void vdo_waitq_init(struct vdo_wait_queue *waitq)
+{
+	*waitq = (struct vdo_wait_queue) {
+		.last_waiter = NULL,
+		.length = 0,
+	};
+}
+
+/**
+ * vdo_waitq_has_waiters() - Check whether a vdo_wait_queue has any entries waiting.
+ * @waitq: The vdo_wait_queue to query.
+ *
+ * Return: true if there are any waiters in the waitq.
+ */
+static inline bool __must_check vdo_waitq_has_waiters(const struct vdo_wait_queue *waitq)
+{
+	return (waitq->last_waiter != NULL);
+}
+
+void vdo_waitq_enqueue_waiter(struct vdo_wait_queue *waitq,
+			      struct vdo_waiter *waiter);
+
+struct vdo_waiter *vdo_waitq_dequeue_waiter(struct vdo_wait_queue *waitq);
+
+void vdo_waitq_notify_all_waiters(struct vdo_wait_queue *waitq,
+				  vdo_waiter_callback_fn callback, void *context);
+
+bool vdo_waitq_notify_next_waiter(struct vdo_wait_queue *waitq,
+				  vdo_waiter_callback_fn callback, void *context);
+
+void vdo_waitq_transfer_all_waiters(struct vdo_wait_queue *from_waitq,
+				    struct vdo_wait_queue *to_waitq);
+
+struct vdo_waiter *vdo_waitq_get_first_waiter(const struct vdo_wait_queue *waitq);
+
+void vdo_waitq_dequeue_matching_waiters(struct vdo_wait_queue *waitq,
+					vdo_waiter_match_fn waiter_match,
+					void *match_context,
+					struct vdo_wait_queue *matched_waitq);
+
+/**
+ * vdo_waitq_num_waiters() - Return the number of waiters in a vdo_wait_queue.
+ * @waitq: The vdo_wait_queue to query.
+ *
+ * Return: The number of waiters in the waitq.
+ */
+static inline size_t __must_check vdo_waitq_num_waiters(const struct vdo_wait_queue *waitq)
+{
+	return waitq->length;
+}
+
+#endif /* VDO_WAIT_QUEUE_H */