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, 88 insertions, 0 deletions

diff --git a/drivers/md/dm-cache-background-tracker.h b/drivers/md/dm-cache-background-tracker.h
new file mode 100644
index 000000000..47156c14a
--- /dev/null
+++ b/drivers/md/dm-cache-background-tracker.h
@@ -0,0 +1,88 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2017 Red Hat. All rights reserved.
+ *
+ * This file is released under the GPL.
+ */
+
+#ifndef DM_CACHE_BACKGROUND_WORK_H
+#define DM_CACHE_BACKGROUND_WORK_H
+
+#include <linux/vmalloc.h>
+#include "dm-cache-policy.h"
+
+/*----------------------------------------------------------------*/
+
+/*
+ * The cache policy decides what background work should be performed,
+ * such as promotions, demotions and writebacks. The core cache target
+ * is in charge of performing the work, and does so when it sees fit.
+ *
+ * The background_tracker acts as a go between. Keeping track of future
+ * work that the policy has decided upon, and handing (issuing) it to
+ * the core target when requested.
+ *
+ * There is no locking in this, so calls will probably need to be
+ * protected with a spinlock.
+ */
+
+struct bt_work {
+	struct list_head list;
+	struct rb_node node;
+	struct policy_work work;
+};
+
+extern struct kmem_cache *btracker_work_cache;
+
+struct background_work;
+struct background_tracker;
+
+/*
+ * Create a new tracker, it will not be able to queue more than
+ * 'max_work' entries.
+ */
+struct background_tracker *btracker_create(unsigned int max_work);
+
+/*
+ * Destroy the tracker. No issued, but not complete, work should
+ * exist when this is called. It is fine to have queued but unissued
+ * work.
+ */
+void btracker_destroy(struct background_tracker *b);
+
+unsigned int btracker_nr_demotions_queued(struct background_tracker *b);
+
+/*
+ * Queue some work within the tracker. 'work' should point to the work
+ * to queue, this will be copied (ownership doesn't pass).  If pwork
+ * is not NULL then it will be set to point to the tracker's internal
+ * copy of the work.
+ *
+ * returns -EINVAL iff the work is already queued.  -ENOMEM if the work
+ * couldn't be queued for another reason.
+ */
+int btracker_queue(struct background_tracker *b,
+		   struct policy_work *work,
+		   struct policy_work **pwork);
+
+/*
+ * Hands out the next piece of work to be performed.
+ * Returns -ENODATA if there's no work.
+ */
+int btracker_issue(struct background_tracker *b, struct policy_work **work);
+
+/*
+ * Informs the tracker that the work has been completed and it may forget
+ * about it.
+ */
+void btracker_complete(struct background_tracker *b, struct policy_work *op);
+
+/*
+ * Predicate to see if an origin block is already scheduled for promotion.
+ */
+bool btracker_promotion_already_present(struct background_tracker *b,
+					dm_oblock_t oblock);
+
+/*----------------------------------------------------------------*/
+
+#endif