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, 206 insertions, 0 deletions

diff --git a/net/ipv4/ipcomp.c b/net/ipv4/ipcomp.c
new file mode 100644
index 000000000..b1ea2d37e
--- /dev/null
+++ b/net/ipv4/ipcomp.c
@@ -0,0 +1,206 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * IP Payload Compression Protocol (IPComp) - RFC3173.
+ *
+ * Copyright (c) 2003 James Morris <jmorris@intercode.com.au>
+ *
+ * Todo:
+ *   - Tunable compression parameters.
+ *   - Compression stats.
+ *   - Adaptive compression.
+ */
+#include <linux/module.h>
+#include <linux/err.h>
+#include <linux/rtnetlink.h>
+#include <net/ip.h>
+#include <net/xfrm.h>
+#include <net/icmp.h>
+#include <net/ipcomp.h>
+#include <net/protocol.h>
+#include <net/sock.h>
+
+static int ipcomp4_err(struct sk_buff *skb, u32 info)
+{
+	struct net *net = dev_net(skb->dev);
+	__be32 spi;
+	const struct iphdr *iph = (const struct iphdr *)skb->data;
+	struct ip_comp_hdr *ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
+	struct xfrm_state *x;
+
+	switch (icmp_hdr(skb)->type) {
+	case ICMP_DEST_UNREACH:
+		if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
+			return 0;
+		break;
+	case ICMP_REDIRECT:
+		break;
+	default:
+		return 0;
+	}
+
+	spi = htonl(ntohs(ipch->cpi));
+	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
+			      spi, IPPROTO_COMP, AF_INET);
+	if (!x)
+		return 0;
+
+	if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
+		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_COMP);
+	else
+		ipv4_redirect(skb, net, 0, IPPROTO_COMP);
+	xfrm_state_put(x);
+
+	return 0;
+}
+
+/* We always hold one tunnel user reference to indicate a tunnel */
+static struct lock_class_key xfrm_state_lock_key;
+static struct xfrm_state *ipcomp_tunnel_create(struct xfrm_state *x)
+{
+	struct net *net = xs_net(x);
+	struct xfrm_state *t;
+
+	t = xfrm_state_alloc(net);
+	if (!t)
+		goto out;
+	lockdep_set_class(&t->lock, &xfrm_state_lock_key);
+
+	t->id.proto = IPPROTO_IPIP;
+	t->id.spi = x->props.saddr.a4;
+	t->id.daddr.a4 = x->id.daddr.a4;
+	memcpy(&t->sel, &x->sel, sizeof(t->sel));
+	t->props.family = AF_INET;
+	t->props.mode = x->props.mode;
+	t->props.saddr.a4 = x->props.saddr.a4;
+	t->props.flags = x->props.flags;
+	t->props.extra_flags = x->props.extra_flags;
+	memcpy(&t->mark, &x->mark, sizeof(t->mark));
+	t->if_id = x->if_id;
+
+	if (xfrm_init_state(t, NULL))
+		goto error;
+
+	atomic_set(&t->tunnel_users, 1);
+out:
+	return t;
+
+error:
+	t->km.state = XFRM_STATE_DEAD;
+	xfrm_state_put(t);
+	t = NULL;
+	goto out;
+}
+
+/*
+ * Must be protected by xfrm_cfg_mutex.  State and tunnel user references are
+ * always incremented on success.
+ */
+static int ipcomp_tunnel_attach(struct xfrm_state *x)
+{
+	struct net *net = xs_net(x);
+	int err = 0;
+	struct xfrm_state *t;
+	u32 mark = x->mark.v & x->mark.m;
+
+	t = xfrm_state_lookup(net, mark, (xfrm_address_t *)&x->id.daddr.a4,
+			      x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
+	if (!t) {
+		t = ipcomp_tunnel_create(x);
+		if (!t) {
+			err = -EINVAL;
+			goto out;
+		}
+		xfrm_state_insert(t);
+		xfrm_state_hold(t);
+	}
+	x->tunnel = t;
+	atomic_inc(&t->tunnel_users);
+out:
+	return err;
+}
+
+static int ipcomp4_init_state(struct xfrm_state *x,
+			      struct netlink_ext_ack *extack)
+{
+	int err = -EINVAL;
+
+	x->props.header_len = 0;
+	switch (x->props.mode) {
+	case XFRM_MODE_TRANSPORT:
+		break;
+	case XFRM_MODE_TUNNEL:
+		x->props.header_len += sizeof(struct iphdr);
+		break;
+	default:
+		NL_SET_ERR_MSG(extack, "Unsupported XFRM mode for IPcomp");
+		goto out;
+	}
+
+	err = ipcomp_init_state(x, extack);
+	if (err)
+		goto out;
+
+	if (x->props.mode == XFRM_MODE_TUNNEL) {
+		err = ipcomp_tunnel_attach(x);
+		if (err) {
+			NL_SET_ERR_MSG(extack, "Kernel error: failed to initialize the associated state");
+			goto out;
+		}
+	}
+
+	err = 0;
+out:
+	return err;
+}
+
+static int ipcomp4_rcv_cb(struct sk_buff *skb, int err)
+{
+	return 0;
+}
+
+static const struct xfrm_type ipcomp_type = {
+	.owner		= THIS_MODULE,
+	.proto	     	= IPPROTO_COMP,
+	.init_state	= ipcomp4_init_state,
+	.destructor	= ipcomp_destroy,
+	.input		= ipcomp_input,
+	.output		= ipcomp_output
+};
+
+static struct xfrm4_protocol ipcomp4_protocol = {
+	.handler	=	xfrm4_rcv,
+	.input_handler	=	xfrm_input,
+	.cb_handler	=	ipcomp4_rcv_cb,
+	.err_handler	=	ipcomp4_err,
+	.priority	=	0,
+};
+
+static int __init ipcomp4_init(void)
+{
+	if (xfrm_register_type(&ipcomp_type, AF_INET) < 0) {
+		pr_info("%s: can't add xfrm type\n", __func__);
+		return -EAGAIN;
+	}
+	if (xfrm4_protocol_register(&ipcomp4_protocol, IPPROTO_COMP) < 0) {
+		pr_info("%s: can't add protocol\n", __func__);
+		xfrm_unregister_type(&ipcomp_type, AF_INET);
+		return -EAGAIN;
+	}
+	return 0;
+}
+
+static void __exit ipcomp4_fini(void)
+{
+	if (xfrm4_protocol_deregister(&ipcomp4_protocol, IPPROTO_COMP) < 0)
+		pr_info("%s: can't remove protocol\n", __func__);
+	xfrm_unregister_type(&ipcomp_type, AF_INET);
+}
+
+module_init(ipcomp4_init);
+module_exit(ipcomp4_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp/IPv4) - RFC3173");
+MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
+
+MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_COMP);