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, 354 insertions, 0 deletions

diff --git a/net/sctp/chunk.c b/net/sctp/chunk.c
new file mode 100644
index 000000000..5b889e89e
--- /dev/null
+++ b/net/sctp/chunk.c
@@ -0,0 +1,354 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* SCTP kernel implementation
+ * (C) Copyright IBM Corp. 2003, 2004
+ *
+ * This file is part of the SCTP kernel implementation
+ *
+ * This file contains the code relating the chunk abstraction.
+ *
+ * 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:
+ *    Jon Grimm             <jgrimm@us.ibm.com>
+ *    Sridhar Samudrala     <sri@us.ibm.com>
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/net.h>
+#include <linux/inet.h>
+#include <linux/skbuff.h>
+#include <linux/slab.h>
+#include <net/sock.h>
+#include <net/sctp/sctp.h>
+#include <net/sctp/sm.h>
+
+/* This file is mostly in anticipation of future work, but initially
+ * populate with fragment tracking for an outbound message.
+ */
+
+/* Initialize datamsg from memory. */
+static void sctp_datamsg_init(struct sctp_datamsg *msg)
+{
+	refcount_set(&msg->refcnt, 1);
+	msg->send_failed = 0;
+	msg->send_error = 0;
+	msg->can_delay = 1;
+	msg->abandoned = 0;
+	msg->expires_at = 0;
+	INIT_LIST_HEAD(&msg->chunks);
+}
+
+/* Allocate and initialize datamsg. */
+static struct sctp_datamsg *sctp_datamsg_new(gfp_t gfp)
+{
+	struct sctp_datamsg *msg;
+	msg = kmalloc_obj(struct sctp_datamsg, gfp);
+	if (msg) {
+		sctp_datamsg_init(msg);
+		SCTP_DBG_OBJCNT_INC(datamsg);
+	}
+	return msg;
+}
+
+void sctp_datamsg_free(struct sctp_datamsg *msg)
+{
+	struct sctp_chunk *chunk;
+
+	/* This doesn't have to be a _safe vairant because
+	 * sctp_chunk_free() only drops the refs.
+	 */
+	list_for_each_entry(chunk, &msg->chunks, frag_list)
+		sctp_chunk_free(chunk);
+
+	sctp_datamsg_put(msg);
+}
+
+/* Final destructruction of datamsg memory. */
+static void sctp_datamsg_destroy(struct sctp_datamsg *msg)
+{
+	struct sctp_association *asoc = NULL;
+	struct list_head *pos, *temp;
+	struct sctp_chunk *chunk;
+	struct sctp_ulpevent *ev;
+	int error, sent;
+
+	/* Release all references. */
+	list_for_each_safe(pos, temp, &msg->chunks) {
+		list_del_init(pos);
+		chunk = list_entry(pos, struct sctp_chunk, frag_list);
+
+		if (!msg->send_failed) {
+			sctp_chunk_put(chunk);
+			continue;
+		}
+
+		asoc = chunk->asoc;
+		error = msg->send_error ?: asoc->outqueue.error;
+		sent = chunk->has_tsn ? SCTP_DATA_SENT : SCTP_DATA_UNSENT;
+
+		if (sctp_ulpevent_type_enabled(asoc->subscribe,
+					       SCTP_SEND_FAILED)) {
+			ev = sctp_ulpevent_make_send_failed(asoc, chunk, sent,
+							    error, GFP_ATOMIC);
+			if (ev)
+				asoc->stream.si->enqueue_event(&asoc->ulpq, ev);
+		}
+
+		if (sctp_ulpevent_type_enabled(asoc->subscribe,
+					       SCTP_SEND_FAILED_EVENT)) {
+			ev = sctp_ulpevent_make_send_failed_event(asoc, chunk,
+								  sent, error,
+								  GFP_ATOMIC);
+			if (ev)
+				asoc->stream.si->enqueue_event(&asoc->ulpq, ev);
+		}
+
+		sctp_chunk_put(chunk);
+	}
+
+	SCTP_DBG_OBJCNT_DEC(datamsg);
+	kfree(msg);
+}
+
+/* Hold a reference. */
+static void sctp_datamsg_hold(struct sctp_datamsg *msg)
+{
+	refcount_inc(&msg->refcnt);
+}
+
+/* Release a reference. */
+void sctp_datamsg_put(struct sctp_datamsg *msg)
+{
+	if (refcount_dec_and_test(&msg->refcnt))
+		sctp_datamsg_destroy(msg);
+}
+
+/* Assign a chunk to this datamsg. */
+static void sctp_datamsg_assign(struct sctp_datamsg *msg, struct sctp_chunk *chunk)
+{
+	sctp_datamsg_hold(msg);
+	chunk->msg = msg;
+}
+
+
+/* A data chunk can have a maximum payload of (2^16 - 20).  Break
+ * down any such message into smaller chunks.  Opportunistically, fragment
+ * the chunks down to the current MTU constraints.  We may get refragmented
+ * later if the PMTU changes, but it is _much better_ to fragment immediately
+ * with a reasonable guess than always doing our fragmentation on the
+ * soft-interrupt.
+ */
+struct sctp_datamsg *sctp_datamsg_from_user(struct sctp_association *asoc,
+					    struct sctp_sndrcvinfo *sinfo,
+					    struct iov_iter *from)
+{
+	size_t len, first_len, max_data, remaining;
+	size_t msg_len = iov_iter_count(from);
+	struct sctp_shared_key *shkey = NULL;
+	struct list_head *pos, *temp;
+	struct sctp_chunk *chunk;
+	struct sctp_datamsg *msg;
+	int err;
+
+	msg = sctp_datamsg_new(GFP_KERNEL);
+	if (!msg)
+		return ERR_PTR(-ENOMEM);
+
+	/* Note: Calculate this outside of the loop, so that all fragments
+	 * have the same expiration.
+	 */
+	if (asoc->peer.prsctp_capable && sinfo->sinfo_timetolive &&
+	    (SCTP_PR_TTL_ENABLED(sinfo->sinfo_flags) ||
+	     !SCTP_PR_POLICY(sinfo->sinfo_flags)))
+		msg->expires_at = jiffies +
+				  msecs_to_jiffies(sinfo->sinfo_timetolive);
+
+	/* This is the biggest possible DATA chunk that can fit into
+	 * the packet
+	 */
+	max_data = asoc->frag_point;
+	if (unlikely(!max_data)) {
+		max_data = sctp_min_frag_point(sctp_sk(asoc->base.sk),
+					       sctp_datachk_len(&asoc->stream));
+		pr_warn_ratelimited("%s: asoc:%p frag_point is zero, forcing max_data to default minimum (%zu)",
+				    __func__, asoc, max_data);
+	}
+
+	/* If the peer requested that we authenticate DATA chunks
+	 * we need to account for bundling of the AUTH chunks along with
+	 * DATA.
+	 */
+	if (sctp_auth_send_cid(SCTP_CID_DATA, asoc)) {
+		const struct sctp_hmac *hmac_desc =
+			sctp_auth_asoc_get_hmac(asoc);
+
+		if (hmac_desc)
+			max_data -= SCTP_PAD4(sizeof(struct sctp_auth_chunk) +
+					      hmac_desc->hmac_len);
+
+		if (sinfo->sinfo_tsn &&
+		    sinfo->sinfo_ssn != asoc->active_key_id) {
+			shkey = sctp_auth_get_shkey(asoc, sinfo->sinfo_ssn);
+			if (!shkey) {
+				err = -EINVAL;
+				goto errout;
+			}
+		} else {
+			shkey = asoc->shkey;
+		}
+	}
+
+	/* Set first_len and then account for possible bundles on first frag */
+	first_len = max_data;
+
+	/* Check to see if we have a pending SACK and try to let it be bundled
+	 * with this message.  Do this if we don't have any data queued already.
+	 * To check that, look at out_qlen and retransmit list.
+	 * NOTE: we will not reduce to account for SACK, if the message would
+	 * not have been fragmented.
+	 */
+	if (timer_pending(&asoc->timers[SCTP_EVENT_TIMEOUT_SACK]) &&
+	    asoc->outqueue.out_qlen == 0 &&
+	    list_empty(&asoc->outqueue.retransmit) &&
+	    msg_len > max_data)
+		first_len -= SCTP_PAD4(sizeof(struct sctp_sack_chunk));
+
+	/* Encourage Cookie-ECHO bundling. */
+	if (asoc->state < SCTP_STATE_COOKIE_ECHOED)
+		first_len -= SCTP_ARBITRARY_COOKIE_ECHO_LEN;
+
+	/* Account for a different sized first fragment */
+	if (msg_len >= first_len) {
+		msg->can_delay = 0;
+		if (msg_len > first_len)
+			SCTP_INC_STATS(asoc->base.net,
+				       SCTP_MIB_FRAGUSRMSGS);
+	} else {
+		/* Which may be the only one... */
+		first_len = msg_len;
+	}
+
+	/* Create chunks for all DATA chunks. */
+	for (remaining = msg_len; remaining; remaining -= len) {
+		u8 frag = SCTP_DATA_MIDDLE_FRAG;
+
+		if (remaining == msg_len) {
+			/* First frag, which may also be the last */
+			frag |= SCTP_DATA_FIRST_FRAG;
+			len = first_len;
+		} else {
+			/* Middle frags */
+			len = max_data;
+		}
+
+		if (len >= remaining) {
+			/* Last frag, which may also be the first */
+			len = remaining;
+			frag |= SCTP_DATA_LAST_FRAG;
+
+			/* The application requests to set the I-bit of the
+			 * last DATA chunk of a user message when providing
+			 * the user message to the SCTP implementation.
+			 */
+			if ((sinfo->sinfo_flags & SCTP_EOF) ||
+			    (sinfo->sinfo_flags & SCTP_SACK_IMMEDIATELY))
+				frag |= SCTP_DATA_SACK_IMM;
+		}
+
+		chunk = asoc->stream.si->make_datafrag(asoc, sinfo, len, frag,
+						       GFP_KERNEL);
+		if (!chunk) {
+			err = -ENOMEM;
+			goto errout;
+		}
+
+		err = sctp_user_addto_chunk(chunk, len, from);
+		if (err < 0)
+			goto errout_chunk_free;
+
+		chunk->shkey = shkey;
+
+		/* Put the chunk->skb back into the form expected by send.  */
+		__skb_pull(chunk->skb, (__u8 *)chunk->chunk_hdr -
+				       chunk->skb->data);
+
+		sctp_datamsg_assign(msg, chunk);
+		list_add_tail(&chunk->frag_list, &msg->chunks);
+	}
+
+	return msg;
+
+errout_chunk_free:
+	sctp_chunk_free(chunk);
+
+errout:
+	list_for_each_safe(pos, temp, &msg->chunks) {
+		list_del_init(pos);
+		chunk = list_entry(pos, struct sctp_chunk, frag_list);
+		sctp_chunk_free(chunk);
+	}
+	sctp_datamsg_put(msg);
+
+	return ERR_PTR(err);
+}
+
+/* Check whether this message has expired. */
+int sctp_chunk_abandoned(struct sctp_chunk *chunk)
+{
+	if (!chunk->asoc->peer.prsctp_capable)
+		return 0;
+
+	if (chunk->msg->abandoned)
+		return 1;
+
+	if (!chunk->has_tsn &&
+	    !(chunk->chunk_hdr->flags & SCTP_DATA_FIRST_FRAG))
+		return 0;
+
+	if (SCTP_PR_TTL_ENABLED(chunk->sinfo.sinfo_flags) &&
+	    time_after(jiffies, chunk->msg->expires_at)) {
+		struct sctp_stream_out *streamout =
+			SCTP_SO(&chunk->asoc->stream,
+				chunk->sinfo.sinfo_stream);
+
+		if (chunk->sent_count) {
+			chunk->asoc->abandoned_sent[SCTP_PR_INDEX(TTL)]++;
+			streamout->ext->abandoned_sent[SCTP_PR_INDEX(TTL)]++;
+		} else {
+			chunk->asoc->abandoned_unsent[SCTP_PR_INDEX(TTL)]++;
+			streamout->ext->abandoned_unsent[SCTP_PR_INDEX(TTL)]++;
+		}
+		chunk->msg->abandoned = 1;
+		return 1;
+	} else if (SCTP_PR_RTX_ENABLED(chunk->sinfo.sinfo_flags) &&
+		   chunk->sent_count > chunk->sinfo.sinfo_timetolive) {
+		struct sctp_stream_out *streamout =
+			SCTP_SO(&chunk->asoc->stream,
+				chunk->sinfo.sinfo_stream);
+
+		chunk->asoc->abandoned_sent[SCTP_PR_INDEX(RTX)]++;
+		streamout->ext->abandoned_sent[SCTP_PR_INDEX(RTX)]++;
+		chunk->msg->abandoned = 1;
+		return 1;
+	} else if (!SCTP_PR_POLICY(chunk->sinfo.sinfo_flags) &&
+		   chunk->msg->expires_at &&
+		   time_after(jiffies, chunk->msg->expires_at)) {
+		chunk->msg->abandoned = 1;
+		return 1;
+	}
+	/* PRIO policy is processed by sendmsg, not here */
+
+	return 0;
+}
+
+/* This chunk (and consequently entire message) has failed in its sending. */
+void sctp_chunk_fail(struct sctp_chunk *chunk, int error)
+{
+	chunk->msg->send_failed = 1;
+	chunk->msg->send_error = error;
+}