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, 224 insertions, 0 deletions

diff --git a/net/llc/llc_input.c b/net/llc/llc_input.c
new file mode 100644
index 000000000..8eb3d73c3
--- /dev/null
+++ b/net/llc/llc_input.c
@@ -0,0 +1,224 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * llc_input.c - Minimal input path for LLC
+ *
+ * Copyright (c) 1997 by Procom Technology, Inc.
+ * 		 2001-2003 by Arnaldo Carvalho de Melo <acme@conectiva.com.br>
+ */
+#include <linux/netdevice.h>
+#include <linux/slab.h>
+#include <linux/export.h>
+#include <net/net_namespace.h>
+#include <net/llc.h>
+#include <net/llc_pdu.h>
+#include <net/llc_sap.h>
+
+#if 0
+#define dprintk(args...) printk(KERN_DEBUG args)
+#else
+#define dprintk(args...)
+#endif
+
+/*
+ * Packet handler for the station, registerable because in the minimal
+ * LLC core that is taking shape only the very minimal subset of LLC that
+ * is needed for things like IPX, Appletalk, etc will stay, with all the
+ * rest in the llc1 and llc2 modules.
+ */
+static void (*llc_station_handler)(struct sk_buff *skb);
+
+/*
+ * Packet handlers for LLC_DEST_SAP and LLC_DEST_CONN.
+ */
+static void (*llc_type_handlers[2])(struct llc_sap *sap,
+				    struct sk_buff *skb);
+
+void llc_add_pack(int type, void (*handler)(struct llc_sap *sap,
+					    struct sk_buff *skb))
+{
+	smp_wmb(); /* ensure initialisation is complete before it's called */
+	if (type == LLC_DEST_SAP || type == LLC_DEST_CONN)
+		llc_type_handlers[type - 1] = handler;
+}
+
+void llc_remove_pack(int type)
+{
+	if (type == LLC_DEST_SAP || type == LLC_DEST_CONN)
+		llc_type_handlers[type - 1] = NULL;
+	synchronize_net();
+}
+
+void llc_set_station_handler(void (*handler)(struct sk_buff *skb))
+{
+	/* Ensure initialisation is complete before it's called */
+	if (handler)
+		smp_wmb();
+
+	llc_station_handler = handler;
+
+	if (!handler)
+		synchronize_net();
+}
+
+/**
+ *	llc_pdu_type - returns which LLC component must handle for PDU
+ *	@skb: input skb
+ *
+ *	This function returns which LLC component must handle this PDU.
+ */
+static __inline__ int llc_pdu_type(struct sk_buff *skb)
+{
+	int type = LLC_DEST_CONN; /* I-PDU or S-PDU type */
+	struct llc_pdu_sn *pdu = llc_pdu_sn_hdr(skb);
+
+	if ((pdu->ctrl_1 & LLC_PDU_TYPE_MASK) != LLC_PDU_TYPE_U)
+		goto out;
+	switch (LLC_U_PDU_CMD(pdu)) {
+	case LLC_1_PDU_CMD_XID:
+	case LLC_1_PDU_CMD_UI:
+	case LLC_1_PDU_CMD_TEST:
+		type = LLC_DEST_SAP;
+		break;
+	case LLC_2_PDU_CMD_SABME:
+	case LLC_2_PDU_CMD_DISC:
+	case LLC_2_PDU_RSP_UA:
+	case LLC_2_PDU_RSP_DM:
+	case LLC_2_PDU_RSP_FRMR:
+		break;
+	default:
+		type = LLC_DEST_INVALID;
+		break;
+	}
+out:
+	return type;
+}
+
+/**
+ *	llc_fixup_skb - initializes skb pointers
+ *	@skb: This argument points to incoming skb
+ *
+ *	Initializes internal skb pointer to start of network layer by deriving
+ *	length of LLC header; finds length of LLC control field in LLC header
+ *	by looking at the two lowest-order bits of the first control field
+ *	byte; field is either 3 or 4 bytes long.
+ */
+static inline int llc_fixup_skb(struct sk_buff *skb)
+{
+	u8 llc_len = 2;
+	struct llc_pdu_un *pdu;
+
+	if (unlikely(!pskb_may_pull(skb, sizeof(*pdu))))
+		return 0;
+
+	pdu = (struct llc_pdu_un *)skb->data;
+	if ((pdu->ctrl_1 & LLC_PDU_TYPE_MASK) == LLC_PDU_TYPE_U)
+		llc_len = 1;
+	llc_len += 2;
+
+	if (unlikely(!pskb_may_pull(skb, llc_len)))
+		return 0;
+
+	skb_pull(skb, llc_len);
+	skb_reset_transport_header(skb);
+	if (skb->protocol == htons(ETH_P_802_2)) {
+		__be16 pdulen;
+		s32 data_size;
+
+		if (skb->mac_len < ETH_HLEN)
+			return 0;
+
+		pdulen = eth_hdr(skb)->h_proto;
+		data_size = ntohs(pdulen) - llc_len;
+
+		if (data_size < 0 ||
+		    !pskb_may_pull(skb, data_size))
+			return 0;
+		if (unlikely(pskb_trim_rcsum(skb, data_size)))
+			return 0;
+	}
+	return 1;
+}
+
+/**
+ *	llc_rcv - 802.2 entry point from net lower layers
+ *	@skb: received pdu
+ *	@dev: device that receive pdu
+ *	@pt: packet type
+ *	@orig_dev: the original receive net device
+ *
+ *	When the system receives a 802.2 frame this function is called. It
+ *	checks SAP and connection of received pdu and passes frame to
+ *	llc_{station,sap,conn}_rcv for sending to proper state machine. If
+ *	the frame is related to a busy connection (a connection is sending
+ *	data now), it queues this frame in the connection's backlog.
+ */
+int llc_rcv(struct sk_buff *skb, struct net_device *dev,
+	    struct packet_type *pt, struct net_device *orig_dev)
+{
+	struct llc_sap *sap;
+	struct llc_pdu_sn *pdu;
+	int dest;
+	int (*rcv)(struct sk_buff *, struct net_device *,
+		   struct packet_type *, struct net_device *);
+	void (*sta_handler)(struct sk_buff *skb);
+	void (*sap_handler)(struct llc_sap *sap, struct sk_buff *skb);
+
+	/*
+	 * When the interface is in promisc. mode, drop all the crap that it
+	 * receives, do not try to analyse it.
+	 */
+	if (unlikely(skb->pkt_type == PACKET_OTHERHOST)) {
+		dprintk("%s: PACKET_OTHERHOST\n", __func__);
+		goto drop;
+	}
+	skb = skb_share_check(skb, GFP_ATOMIC);
+	if (unlikely(!skb))
+		goto out;
+	if (unlikely(!llc_fixup_skb(skb)))
+		goto drop;
+	pdu = llc_pdu_sn_hdr(skb);
+	if (unlikely(!pdu->dsap)) /* NULL DSAP, refer to station */
+	       goto handle_station;
+	sap = llc_sap_find(pdu->dsap);
+	if (unlikely(!sap)) {/* unknown SAP */
+		dprintk("%s: llc_sap_find(%02X) failed!\n", __func__,
+			pdu->dsap);
+		goto drop;
+	}
+	/*
+	 * First the upper layer protocols that don't need the full
+	 * LLC functionality
+	 */
+	rcv = rcu_dereference(sap->rcv_func);
+	dest = llc_pdu_type(skb);
+	sap_handler = dest ? READ_ONCE(llc_type_handlers[dest - 1]) : NULL;
+	if (unlikely(!sap_handler)) {
+		if (rcv)
+			rcv(skb, dev, pt, orig_dev);
+		else
+			kfree_skb(skb);
+	} else {
+		if (rcv) {
+			struct sk_buff *cskb = skb_clone(skb, GFP_ATOMIC);
+			if (cskb)
+				rcv(cskb, dev, pt, orig_dev);
+		}
+		sap_handler(sap, skb);
+	}
+	llc_sap_put(sap);
+out:
+	return 0;
+drop:
+	kfree_skb(skb);
+	goto out;
+handle_station:
+	sta_handler = READ_ONCE(llc_station_handler);
+	if (!sta_handler)
+		goto drop;
+	sta_handler(skb);
+	goto out;
+}
+
+EXPORT_SYMBOL(llc_add_pack);
+EXPORT_SYMBOL(llc_remove_pack);
+EXPORT_SYMBOL(llc_set_station_handler);