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, 109 insertions, 0 deletions

diff --git a/include/net/gso.h b/include/net/gso.h
new file mode 100644
index 000000000..29975440c
--- /dev/null
+++ b/include/net/gso.h
@@ -0,0 +1,109 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+
+#ifndef _NET_GSO_H
+#define _NET_GSO_H
+
+#include <linux/skbuff.h>
+
+/* Keeps track of mac header offset relative to skb->head.
+ * It is useful for TSO of Tunneling protocol. e.g. GRE.
+ * For non-tunnel skb it points to skb_mac_header() and for
+ * tunnel skb it points to outer mac header.
+ * Keeps track of level of encapsulation of network headers.
+ */
+struct skb_gso_cb {
+	union {
+		int	mac_offset;
+		int	data_offset;
+	};
+	int	encap_level;
+	__wsum	csum;
+	__u16	csum_start;
+};
+#define SKB_GSO_CB_OFFSET	32
+#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET))
+
+static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
+{
+	return (skb_mac_header(inner_skb) - inner_skb->head) -
+		SKB_GSO_CB(inner_skb)->mac_offset;
+}
+
+static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
+{
+	int new_headroom, headroom;
+	int ret;
+
+	headroom = skb_headroom(skb);
+	ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
+	if (ret)
+		return ret;
+
+	new_headroom = skb_headroom(skb);
+	SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
+	return 0;
+}
+
+static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
+{
+	/* Do not update partial checksums if remote checksum is enabled. */
+	if (skb->remcsum_offload)
+		return;
+
+	SKB_GSO_CB(skb)->csum = res;
+	SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
+}
+
+/* Compute the checksum for a gso segment. First compute the checksum value
+ * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
+ * then add in skb->csum (checksum from csum_start to end of packet).
+ * skb->csum and csum_start are then updated to reflect the checksum of the
+ * resultant packet starting from the transport header-- the resultant checksum
+ * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
+ * header.
+ */
+static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
+{
+	unsigned char *csum_start = skb_transport_header(skb);
+	int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
+	__wsum partial = SKB_GSO_CB(skb)->csum;
+
+	SKB_GSO_CB(skb)->csum = res;
+	SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
+
+	return csum_fold(csum_partial(csum_start, plen, partial));
+}
+
+struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
+				  netdev_features_t features, bool tx_path);
+
+static inline struct sk_buff *skb_gso_segment(struct sk_buff *skb,
+					      netdev_features_t features)
+{
+	return __skb_gso_segment(skb, features, true);
+}
+
+struct sk_buff *skb_eth_gso_segment(struct sk_buff *skb,
+				    netdev_features_t features, __be16 type);
+
+struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
+				    netdev_features_t features);
+
+bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
+
+bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
+
+static inline void skb_gso_error_unwind(struct sk_buff *skb, __be16 protocol,
+					int pulled_hlen, u16 mac_offset,
+					int mac_len)
+{
+	skb->protocol = protocol;
+	skb->encapsulation = 1;
+	skb_push(skb, pulled_hlen);
+	skb_reset_transport_header(skb);
+	skb->mac_header = mac_offset;
+	skb->network_header = skb->mac_header + mac_len;
+	skb->mac_len = mac_len;
+}
+
+#endif /* _NET_GSO_H */