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, 250 insertions, 0 deletions

diff --git a/net/sctp/inqueue.c b/net/sctp/inqueue.c
new file mode 100644
index 000000000..d666cec6b
--- /dev/null
+++ b/net/sctp/inqueue.c
@@ -0,0 +1,250 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* SCTP kernel implementation
+ * Copyright (c) 1999-2000 Cisco, Inc.
+ * Copyright (c) 1999-2001 Motorola, Inc.
+ * Copyright (c) 2002 International Business Machines, Corp.
+ *
+ * This file is part of the SCTP kernel implementation
+ *
+ * These functions are the methods for accessing the SCTP inqueue.
+ *
+ * An SCTP inqueue is a queue into which you push SCTP packets
+ * (which might be bundles or fragments of chunks) and out of which you
+ * pop SCTP whole chunks.
+ *
+ * Please send any bug reports or fixes you make to the
+ * email address(es):
+ *    lksctp developers <linux-sctp@vger.kernel.org>
+ *
+ * Written or modified by:
+ *    La Monte H.P. Yarroll <piggy@acm.org>
+ *    Karl Knutson <karl@athena.chicago.il.us>
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <net/sctp/sctp.h>
+#include <net/sctp/sm.h>
+#include <linux/interrupt.h>
+#include <linux/slab.h>
+
+/* Initialize an SCTP inqueue.  */
+void sctp_inq_init(struct sctp_inq *queue)
+{
+	INIT_LIST_HEAD(&queue->in_chunk_list);
+	queue->in_progress = NULL;
+
+	/* Create a task for delivering data.  */
+	INIT_WORK(&queue->immediate, NULL);
+}
+
+/* Properly release the chunk which is being worked on. */
+static inline void sctp_inq_chunk_free(struct sctp_chunk *chunk)
+{
+	if (chunk->head_skb)
+		chunk->skb = chunk->head_skb;
+	sctp_chunk_free(chunk);
+}
+
+/* Release the memory associated with an SCTP inqueue.  */
+void sctp_inq_free(struct sctp_inq *queue)
+{
+	struct sctp_chunk *chunk, *tmp;
+
+	/* Empty the queue.  */
+	list_for_each_entry_safe(chunk, tmp, &queue->in_chunk_list, list) {
+		list_del_init(&chunk->list);
+		sctp_chunk_free(chunk);
+	}
+
+	/* If there is a packet which is currently being worked on,
+	 * free it as well.
+	 */
+	if (queue->in_progress) {
+		sctp_inq_chunk_free(queue->in_progress);
+		queue->in_progress = NULL;
+	}
+}
+
+/* Put a new packet in an SCTP inqueue.
+ * We assume that packet->sctp_hdr is set and in host byte order.
+ */
+void sctp_inq_push(struct sctp_inq *q, struct sctp_chunk *chunk)
+{
+	/* Directly call the packet handling routine.  Drop the chunk if the
+	 * receiver or the transport it was looked up on is gone.
+	 */
+	if (chunk->rcvr->dead ||
+	    (chunk->transport && chunk->transport->dead)) {
+		sctp_chunk_free(chunk);
+		return;
+	}
+
+	/* We are now calling this either from the soft interrupt
+	 * or from the backlog processing.
+	 * Eventually, we should clean up inqueue to not rely
+	 * on the BH related data structures.
+	 */
+	list_add_tail(&chunk->list, &q->in_chunk_list);
+	if (chunk->asoc)
+		chunk->asoc->stats.ipackets++;
+	q->immediate.func(&q->immediate);
+}
+
+/* Peek at the next chunk on the inqeue. */
+struct sctp_chunkhdr *sctp_inq_peek(struct sctp_inq *queue)
+{
+	struct sctp_chunk *chunk;
+	struct sctp_chunkhdr *ch = NULL;
+
+	chunk = queue->in_progress;
+	/* If there is no more chunks in this packet, say so */
+	if (chunk->singleton ||
+	    chunk->end_of_packet ||
+	    chunk->pdiscard)
+		    return NULL;
+
+	ch = (struct sctp_chunkhdr *)chunk->chunk_end;
+
+	return ch;
+}
+
+
+/* Extract a chunk from an SCTP inqueue.
+ *
+ * WARNING:  If you need to put the chunk on another queue, you need to
+ * make a shallow copy (clone) of it.
+ */
+struct sctp_chunk *sctp_inq_pop(struct sctp_inq *queue)
+{
+	struct sctp_chunk *chunk;
+	struct sctp_chunkhdr *ch = NULL;
+
+	/* The assumption is that we are safe to process the chunks
+	 * at this time.
+	 */
+
+	chunk = queue->in_progress;
+	if (chunk) {
+		/* There is a packet that we have been working on.
+		 * Any post processing work to do before we move on?
+		 */
+		if (chunk->singleton ||
+		    chunk->end_of_packet ||
+		    chunk->pdiscard) {
+			if (chunk->head_skb == chunk->skb) {
+				chunk->skb = skb_shinfo(chunk->skb)->frag_list;
+				goto new_skb;
+			}
+			if (chunk->skb->next) {
+				chunk->skb = chunk->skb->next;
+				goto new_skb;
+			}
+
+			sctp_inq_chunk_free(chunk);
+			chunk = queue->in_progress = NULL;
+		} else {
+			/* Nothing to do. Next chunk in the packet, please. */
+			ch = (struct sctp_chunkhdr *)chunk->chunk_end;
+			/* Force chunk->skb->data to chunk->chunk_end.  */
+			skb_pull(chunk->skb, chunk->chunk_end - chunk->skb->data);
+			/* We are guaranteed to pull a SCTP header. */
+		}
+	}
+
+	/* Do we need to take the next packet out of the queue to process? */
+	if (!chunk) {
+		struct list_head *entry;
+
+next_chunk:
+		/* Is the queue empty?  */
+		entry = sctp_list_dequeue(&queue->in_chunk_list);
+		if (!entry)
+			return NULL;
+
+		chunk = list_entry(entry, struct sctp_chunk, list);
+
+		if (skb_is_gso(chunk->skb) && skb_is_gso_sctp(chunk->skb)) {
+			/* GSO-marked skbs but without frags, handle
+			 * them normally
+			 */
+			if (skb_shinfo(chunk->skb)->frag_list)
+				chunk->head_skb = chunk->skb;
+
+			/* skbs with "cover letter" */
+			if (chunk->head_skb && chunk->skb->data_len == chunk->skb->len) {
+				if (WARN_ON(!skb_shinfo(chunk->skb)->frag_list)) {
+					__SCTP_INC_STATS(dev_net(chunk->skb->dev),
+							 SCTP_MIB_IN_PKT_DISCARDS);
+					sctp_chunk_free(chunk);
+					goto next_chunk;
+				}
+				chunk->skb = skb_shinfo(chunk->skb)->frag_list;
+			}
+		}
+
+		if (chunk->asoc)
+			sock_rps_save_rxhash(chunk->asoc->base.sk, chunk->skb);
+
+		queue->in_progress = chunk;
+
+new_skb:
+		/* This is the first chunk in the packet.  */
+		ch = (struct sctp_chunkhdr *)chunk->skb->data;
+		chunk->singleton = 1;
+		chunk->data_accepted = 0;
+		chunk->pdiscard = 0;
+		chunk->auth = 0;
+		chunk->has_asconf = 0;
+		chunk->end_of_packet = 0;
+		if (chunk->head_skb) {
+			struct sctp_input_cb
+				*cb = SCTP_INPUT_CB(chunk->skb),
+				*head_cb = SCTP_INPUT_CB(chunk->head_skb);
+
+			cb->chunk = head_cb->chunk;
+			cb->af = head_cb->af;
+			cb->encap_port = head_cb->encap_port;
+		}
+	}
+
+	chunk->chunk_hdr = ch;
+	chunk->chunk_end = ((__u8 *)ch) + SCTP_PAD4(ntohs(ch->length));
+	skb_pull(chunk->skb, sizeof(*ch));
+	chunk->subh.v = NULL; /* Subheader is no longer valid.  */
+	if (unlikely(ntohs(ch->length) < sizeof(*ch))) {
+		chunk->pdiscard = 1;
+	} else if (chunk->chunk_end + sizeof(*ch) <=
+		   skb_tail_pointer(chunk->skb)) {
+		/* This is not a singleton */
+		chunk->singleton = 0;
+	} else if (chunk->chunk_end > skb_tail_pointer(chunk->skb)) {
+		/* Discard inside state machine. */
+		chunk->pdiscard = 1;
+		chunk->chunk_end = skb_tail_pointer(chunk->skb);
+	} else {
+		/* We are at the end of the packet, so mark the chunk
+		 * in case we need to send a SACK.
+		 */
+		chunk->end_of_packet = 1;
+	}
+
+	pr_debug("+++sctp_inq_pop+++ chunk:%p[%s], length:%d, skb->len:%d\n",
+		 chunk, sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)),
+		 ntohs(chunk->chunk_hdr->length), chunk->skb->len);
+
+	return chunk;
+}
+
+/* Set a top-half handler.
+ *
+ * Originally, we the top-half handler was scheduled as a BH.  We now
+ * call the handler directly in sctp_inq_push() at a time that
+ * we know we are lock safe.
+ * The intent is that this routine will pull stuff out of the
+ * inqueue and process it.
+ */
+void sctp_inq_set_th_handler(struct sctp_inq *q, work_func_t callback)
+{
+	INIT_WORK(&q->immediate, callback);
+}