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, 96 insertions, 0 deletions

diff --git a/net/batman-adv/bitarray.c b/net/batman-adv/bitarray.c
new file mode 100644
index 000000000..f814756d9
--- /dev/null
+++ b/net/batman-adv/bitarray.c
@@ -0,0 +1,96 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (C) B.A.T.M.A.N. contributors:
+ *
+ * Simon Wunderlich, Marek Lindner
+ */
+
+#include "bitarray.h"
+#include "main.h"
+
+#include <linux/bitmap.h>
+
+#include "log.h"
+
+/**
+ * batadv_bitmap_shift_left() - shift the sequence number bitmap left
+ * @seq_bits: the sequence number bitmap to shift
+ * @n: number of positions to shift left
+ *
+ * Shift @seq_bits by @n positions. No-op if @n is not within the bounds of
+ * the bitmap.
+ */
+static void batadv_bitmap_shift_left(unsigned long *seq_bits, s32 n)
+{
+	if (n <= 0 || n >= BATADV_TQ_LOCAL_WINDOW_SIZE)
+		return;
+
+	bitmap_shift_left(seq_bits, seq_bits, n, BATADV_TQ_LOCAL_WINDOW_SIZE);
+}
+
+/**
+ * batadv_bit_get_packet() - receive and process one packet within the sequence
+ *  number window
+ * @priv: the bat priv with all the mesh interface information
+ * @seq_bits: pointer to the sequence number bitmap of received packets
+ * @seq_num_diff: difference between the current/received sequence number and
+ *  the last sequence number
+ * @set_mark: whether this packet should be marked in seq_bits
+ *
+ * Return: true if the window was moved (either new or very old),
+ *  false if the window was not moved/shifted.
+ */
+bool batadv_bit_get_packet(void *priv, unsigned long *seq_bits,
+			   s32 seq_num_diff, int set_mark)
+{
+	struct batadv_priv *bat_priv = priv;
+
+	/* sequence number is slightly older. We already got a sequence number
+	 * higher than this one, so we just mark it.
+	 */
+	if (seq_num_diff <= 0 && seq_num_diff > -BATADV_TQ_LOCAL_WINDOW_SIZE) {
+		if (set_mark)
+			batadv_set_bit(seq_bits, -seq_num_diff);
+		return false;
+	}
+
+	/* sequence number is slightly newer, so we shift the window and
+	 * set the mark if required
+	 */
+	if (seq_num_diff > 0 && seq_num_diff < BATADV_TQ_LOCAL_WINDOW_SIZE) {
+		batadv_bitmap_shift_left(seq_bits, seq_num_diff);
+
+		if (set_mark)
+			batadv_set_bit(seq_bits, 0);
+		return true;
+	}
+
+	/* sequence number is much newer, probably missed a lot of packets */
+	if (seq_num_diff >= BATADV_TQ_LOCAL_WINDOW_SIZE &&
+	    seq_num_diff < BATADV_EXPECTED_SEQNO_RANGE) {
+		batadv_dbg(BATADV_DBG_BATMAN, bat_priv,
+			   "We missed a lot of packets (%i) !\n",
+			   seq_num_diff - 1);
+		bitmap_zero(seq_bits, BATADV_TQ_LOCAL_WINDOW_SIZE);
+		if (set_mark)
+			batadv_set_bit(seq_bits, 0);
+		return true;
+	}
+
+	/* received a much older packet. The other host either restarted
+	 * or the old packet got delayed somewhere in the network. The
+	 * packet should be dropped without calling this function if the
+	 * seqno window is protected.
+	 *
+	 * seq_num_diff <= -BATADV_TQ_LOCAL_WINDOW_SIZE
+	 * or
+	 * seq_num_diff >= BATADV_EXPECTED_SEQNO_RANGE
+	 */
+	batadv_dbg(BATADV_DBG_BATMAN, bat_priv,
+		   "Other host probably restarted!\n");
+
+	bitmap_zero(seq_bits, BATADV_TQ_LOCAL_WINDOW_SIZE);
+	if (set_mark)
+		batadv_set_bit(seq_bits, 0);
+
+	return true;
+}