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, 117 insertions, 0 deletions

diff --git a/net/ipv4/fou_bpf.c b/net/ipv4/fou_bpf.c
new file mode 100644
index 000000000..54984f317
--- /dev/null
+++ b/net/ipv4/fou_bpf.c
@@ -0,0 +1,117 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Unstable Fou Helpers for TC-BPF hook
+ *
+ * These are called from SCHED_CLS BPF programs. Note that it is
+ * allowed to break compatibility for these functions since the interface they
+ * are exposed through to BPF programs is explicitly unstable.
+ */
+
+#include <linux/bpf.h>
+#include <linux/btf_ids.h>
+
+#include <net/dst_metadata.h>
+#include <net/fou.h>
+
+struct bpf_fou_encap {
+	__be16 sport;
+	__be16 dport;
+};
+
+enum bpf_fou_encap_type {
+	FOU_BPF_ENCAP_FOU,
+	FOU_BPF_ENCAP_GUE,
+};
+
+__bpf_kfunc_start_defs();
+
+/* bpf_skb_set_fou_encap - Set FOU encap parameters
+ *
+ * This function allows for using GUE or FOU encapsulation together with an
+ * ipip device in collect-metadata mode.
+ *
+ * It is meant to be used in BPF tc-hooks and after a call to the
+ * bpf_skb_set_tunnel_key helper, responsible for setting IP addresses.
+ *
+ * Parameters:
+ * @skb_ctx	Pointer to ctx (__sk_buff) in TC program. Cannot be NULL
+ * @encap	Pointer to a `struct bpf_fou_encap` storing UDP src and
+ * 		dst ports. If sport is set to 0 the kernel will auto-assign a
+ * 		port. This is similar to using `encap-sport auto`.
+ * 		Cannot be NULL
+ * @type	Encapsulation type for the packet. Their definitions are
+ * 		specified in `enum bpf_fou_encap_type`
+ */
+__bpf_kfunc int bpf_skb_set_fou_encap(struct __sk_buff *skb_ctx,
+				      struct bpf_fou_encap *encap, int type)
+{
+	struct sk_buff *skb = (struct sk_buff *)skb_ctx;
+	struct ip_tunnel_info *info = skb_tunnel_info(skb);
+
+	if (unlikely(!encap))
+		return -EINVAL;
+
+	if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX)))
+		return -EINVAL;
+
+	switch (type) {
+	case FOU_BPF_ENCAP_FOU:
+		info->encap.type = TUNNEL_ENCAP_FOU;
+		break;
+	case FOU_BPF_ENCAP_GUE:
+		info->encap.type = TUNNEL_ENCAP_GUE;
+		break;
+	default:
+		info->encap.type = TUNNEL_ENCAP_NONE;
+	}
+
+	if (test_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags))
+		info->encap.flags |= TUNNEL_ENCAP_FLAG_CSUM;
+
+	info->encap.sport = encap->sport;
+	info->encap.dport = encap->dport;
+
+	return 0;
+}
+
+/* bpf_skb_get_fou_encap - Get FOU encap parameters
+ *
+ * This function allows for reading encap metadata from a packet received
+ * on an ipip device in collect-metadata mode.
+ *
+ * Parameters:
+ * @skb_ctx	Pointer to ctx (__sk_buff) in TC program. Cannot be NULL
+ * @encap	Pointer to a struct bpf_fou_encap storing UDP source and
+ * 		destination port. Cannot be NULL
+ */
+__bpf_kfunc int bpf_skb_get_fou_encap(struct __sk_buff *skb_ctx,
+				      struct bpf_fou_encap *encap)
+{
+	struct sk_buff *skb = (struct sk_buff *)skb_ctx;
+	struct ip_tunnel_info *info = skb_tunnel_info(skb);
+
+	if (unlikely(!info))
+		return -EINVAL;
+
+	encap->sport = info->encap.sport;
+	encap->dport = info->encap.dport;
+
+	return 0;
+}
+
+__bpf_kfunc_end_defs();
+
+BTF_KFUNCS_START(fou_kfunc_set)
+BTF_ID_FLAGS(func, bpf_skb_set_fou_encap)
+BTF_ID_FLAGS(func, bpf_skb_get_fou_encap)
+BTF_KFUNCS_END(fou_kfunc_set)
+
+static const struct btf_kfunc_id_set fou_bpf_kfunc_set = {
+	.owner = THIS_MODULE,
+	.set   = &fou_kfunc_set,
+};
+
+int register_fou_bpf(void)
+{
+	return register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS,
+					 &fou_bpf_kfunc_set);
+}