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/include/net/cls_cgroup.h b/include/net/cls_cgroup.h
new file mode 100644
index 000000000..668aeee9b
--- /dev/null
+++ b/include/net/cls_cgroup.h
@@ -0,0 +1,88 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * cls_cgroup.h			Control Group Classifier
+ *
+ * Authors:	Thomas Graf <tgraf@suug.ch>
+ */
+
+#ifndef _NET_CLS_CGROUP_H
+#define _NET_CLS_CGROUP_H
+
+#include <linux/cgroup.h>
+#include <linux/hardirq.h>
+#include <linux/rcupdate.h>
+#include <net/sock.h>
+#include <net/inet_sock.h>
+
+#ifdef CONFIG_CGROUP_NET_CLASSID
+struct cgroup_cls_state {
+	struct cgroup_subsys_state css;
+	u32 classid;
+};
+
+struct cgroup_cls_state *task_cls_state(struct task_struct *p);
+
+static inline u32 task_cls_classid(struct task_struct *p)
+{
+	u32 classid;
+
+	if (in_interrupt())
+		return 0;
+
+	rcu_read_lock();
+	classid = container_of(task_css(p, net_cls_cgrp_id),
+			       struct cgroup_cls_state, css)->classid;
+	rcu_read_unlock();
+
+	return classid;
+}
+
+static inline void sock_update_classid(struct sock_cgroup_data *skcd)
+{
+	u32 classid;
+
+	classid = task_cls_classid(current);
+	sock_cgroup_set_classid(skcd, classid);
+}
+
+static inline u32 __task_get_classid(struct task_struct *task)
+{
+	return task_cls_state(task)->classid;
+}
+
+static inline u32 task_get_classid(const struct sk_buff *skb)
+{
+	u32 classid = __task_get_classid(current);
+
+	/* Due to the nature of the classifier it is required to ignore all
+	 * packets originating from softirq context as accessing `current'
+	 * would lead to false results.
+	 *
+	 * This test assumes that all callers of dev_queue_xmit() explicitly
+	 * disable bh. Knowing this, it is possible to detect softirq based
+	 * calls by looking at the number of nested bh disable calls because
+	 * softirqs always disables bh.
+	 */
+	if (softirq_count()) {
+		struct sock *sk = skb_to_full_sk(skb);
+
+		/* If there is an sock_cgroup_classid we'll use that. */
+		if (!sk || !sk_fullsock(sk))
+			return 0;
+
+		classid = sock_cgroup_classid(&sk->sk_cgrp_data);
+	}
+
+	return classid;
+}
+#else /* !CONFIG_CGROUP_NET_CLASSID */
+static inline void sock_update_classid(struct sock_cgroup_data *skcd)
+{
+}
+
+static inline u32 task_get_classid(const struct sk_buff *skb)
+{
+	return 0;
+}
+#endif /* CONFIG_CGROUP_NET_CLASSID */
+#endif  /* _NET_CLS_CGROUP_H */