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, 68 insertions, 0 deletions

diff --git a/include/net/sctp/ulpqueue.h b/include/net/sctp/ulpqueue.h
new file mode 100644
index 000000000..60f664129
--- /dev/null
+++ b/include/net/sctp/ulpqueue.h
@@ -0,0 +1,68 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/* SCTP kernel implementation
+ * (C) Copyright IBM Corp. 2001, 2004
+ * Copyright (c) 1999-2000 Cisco, Inc.
+ * Copyright (c) 1999-2001 Motorola, Inc.
+ * Copyright (c) 2001 Intel Corp.
+ * Copyright (c) 2001 Nokia, Inc.
+ * Copyright (c) 2001 La Monte H.P. Yarroll
+ *
+ * These are the definitions needed for the sctp_ulpq type.  The
+ * sctp_ulpq is the interface between the Upper Layer Protocol, or ULP,
+ * and the core SCTP state machine.  This is the component which handles
+ * reassembly and ordering.
+ *
+ * Please send any bug reports or fixes you make to the
+ * email addresses:
+ *    lksctp developers <linux-sctp@vger.kernel.org>
+ *
+ * Written or modified by:
+ *   Jon Grimm             <jgrimm@us.ibm.com>
+ *   La Monte H.P. Yarroll <piggy@acm.org>
+ *   Sridhar Samudrala     <sri@us.ibm.com>
+ */
+
+#ifndef __sctp_ulpqueue_h__
+#define __sctp_ulpqueue_h__
+
+/* A structure to carry information to the ULP (e.g. Sockets API) */
+struct sctp_ulpq {
+	char pd_mode;
+	struct sctp_association *asoc;
+	struct sk_buff_head reasm;
+	struct sk_buff_head reasm_uo;
+	struct sk_buff_head lobby;
+};
+
+/* Prototypes. */
+void sctp_ulpq_init(struct sctp_ulpq *ulpq, struct sctp_association *asoc);
+void sctp_ulpq_flush(struct sctp_ulpq *ulpq);
+void sctp_ulpq_free(struct sctp_ulpq *);
+
+/* Add a new DATA chunk for processing. */
+int sctp_ulpq_tail_data(struct sctp_ulpq *, struct sctp_chunk *, gfp_t);
+
+/* Add a new event for propagation to the ULP. */
+int sctp_ulpq_tail_event(struct sctp_ulpq *, struct sk_buff_head *skb_list);
+
+/* Renege previously received chunks.  */
+void sctp_ulpq_renege(struct sctp_ulpq *, struct sctp_chunk *, gfp_t);
+
+/* Perform partial delivery. */
+void sctp_ulpq_partial_delivery(struct sctp_ulpq *, gfp_t);
+
+/* Abort the partial delivery. */
+void sctp_ulpq_abort_pd(struct sctp_ulpq *, gfp_t);
+
+/* Clear the partial data delivery condition on this socket. */
+int sctp_clear_pd(struct sock *sk, struct sctp_association *asoc);
+
+/* Skip over an SSN. */
+void sctp_ulpq_skip(struct sctp_ulpq *ulpq, __u16 sid, __u16 ssn);
+
+void sctp_ulpq_reasm_flushtsn(struct sctp_ulpq *, __u32);
+
+__u16 sctp_ulpq_renege_list(struct sctp_ulpq *ulpq,
+			    struct sk_buff_head *list, __u16 needed);
+
+#endif /* __sctp_ulpqueue_h__ */