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, 176 insertions, 0 deletions

diff --git a/net/x25/x25_timer.c b/net/x25/x25_timer.c
new file mode 100644
index 000000000..7896cd43f
--- /dev/null
+++ b/net/x25/x25_timer.c
@@ -0,0 +1,176 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ *	X.25 Packet Layer release 002
+ *
+ *	This is ALPHA test software. This code may break your machine,
+ *	randomly fail to work with new releases, misbehave and/or generally
+ *	screw up. It might even work.
+ *
+ *	This code REQUIRES 2.1.15 or higher
+ *
+ *	History
+ *	X.25 001	Jonathan Naylor	Started coding.
+ *	X.25 002	Jonathan Naylor	New timer architecture.
+ *					Centralised disconnection processing.
+ */
+
+#include <linux/errno.h>
+#include <linux/jiffies.h>
+#include <linux/timer.h>
+#include <net/sock.h>
+#include <net/tcp_states.h>
+#include <net/x25.h>
+
+static void x25_heartbeat_expiry(struct timer_list *t);
+static void x25_timer_expiry(struct timer_list *t);
+
+void x25_init_timers(struct sock *sk)
+{
+	struct x25_sock *x25 = x25_sk(sk);
+
+	timer_setup(&x25->timer, x25_timer_expiry, 0);
+
+	/* initialized by sock_init_data */
+	sk->sk_timer.function = x25_heartbeat_expiry;
+}
+
+void x25_start_heartbeat(struct sock *sk)
+{
+	sk_reset_timer(sk, &sk->sk_timer, jiffies + 5 * HZ);
+}
+
+void x25_stop_heartbeat(struct sock *sk)
+{
+	sk_stop_timer(sk, &sk->sk_timer);
+}
+
+void x25_start_t2timer(struct sock *sk)
+{
+	struct x25_sock *x25 = x25_sk(sk);
+
+	sk_reset_timer(sk, &x25->timer, jiffies + x25->t2);
+}
+
+void x25_start_t21timer(struct sock *sk)
+{
+	struct x25_sock *x25 = x25_sk(sk);
+
+	sk_reset_timer(sk, &x25->timer, jiffies + x25->t21);
+}
+
+void x25_start_t22timer(struct sock *sk)
+{
+	struct x25_sock *x25 = x25_sk(sk);
+
+	sk_reset_timer(sk, &x25->timer, jiffies + x25->t22);
+}
+
+void x25_start_t23timer(struct sock *sk)
+{
+	struct x25_sock *x25 = x25_sk(sk);
+
+	sk_reset_timer(sk, &x25->timer, jiffies + x25->t23);
+}
+
+void x25_stop_timer(struct sock *sk)
+{
+	sk_stop_timer(sk, &x25_sk(sk)->timer);
+}
+
+unsigned long x25_display_timer(struct sock *sk)
+{
+	struct x25_sock *x25 = x25_sk(sk);
+
+	if (!timer_pending(&x25->timer))
+		return 0;
+
+	return x25->timer.expires - jiffies;
+}
+
+static void x25_heartbeat_expiry(struct timer_list *t)
+{
+	struct sock *sk = timer_container_of(sk, t, sk_timer);
+
+	bh_lock_sock(sk);
+	if (sock_owned_by_user(sk)) /* can currently only occur in state 3 */
+		goto restart_heartbeat;
+
+	switch (x25_sk(sk)->state) {
+
+		case X25_STATE_0:
+			/*
+			 * Magic here: If we listen() and a new link dies
+			 * before it is accepted() it isn't 'dead' so doesn't
+			 * get removed.
+			 */
+			if (sock_flag(sk, SOCK_DESTROY) ||
+			    (sk->sk_state == TCP_LISTEN &&
+			     sock_flag(sk, SOCK_DEAD))) {
+				bh_unlock_sock(sk);
+				x25_destroy_socket_from_timer(sk);
+				goto out;
+			}
+			break;
+
+		case X25_STATE_3:
+			/*
+			 * Check for the state of the receive buffer.
+			 */
+			x25_check_rbuf(sk);
+			break;
+	}
+restart_heartbeat:
+	/* Do not rearm once __x25_destroy_socket() has unlinked the socket:
+	 * it is past its cancel point and owns the teardown from there on.
+	 */
+	if (sk_hashed(sk))
+		x25_start_heartbeat(sk);
+	bh_unlock_sock(sk);
+out:
+	sock_put(sk);
+}
+
+/*
+ *	Timer has expired, it may have been T2, T21, T22, or T23. We can tell
+ *	by the state machine state.
+ */
+static inline void x25_do_timer_expiry(struct sock * sk)
+{
+	struct x25_sock *x25 = x25_sk(sk);
+
+	switch (x25->state) {
+
+		case X25_STATE_3:	/* T2 */
+			if (x25->condition & X25_COND_ACK_PENDING) {
+				x25->condition &= ~X25_COND_ACK_PENDING;
+				x25_enquiry_response(sk);
+			}
+			break;
+
+		case X25_STATE_1:	/* T21 */
+		case X25_STATE_4:	/* T22 */
+			x25_write_internal(sk, X25_CLEAR_REQUEST);
+			x25->state = X25_STATE_2;
+			x25_start_t23timer(sk);
+			break;
+
+		case X25_STATE_2:	/* T23 */
+			x25_disconnect(sk, ETIMEDOUT, 0, 0);
+			break;
+	}
+}
+
+static void x25_timer_expiry(struct timer_list *t)
+{
+	struct x25_sock *x25 = timer_container_of(x25, t, timer);
+	struct sock *sk = &x25->sk;
+
+	bh_lock_sock(sk);
+	if (sock_owned_by_user(sk)) { /* can currently only occur in state 3 */
+		if (x25_sk(sk)->state == X25_STATE_3)
+			x25_start_t2timer(sk);
+	} else
+		x25_do_timer_expiry(sk);
+	bh_unlock_sock(sk);
+	sock_put(sk);
+}