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, 224 insertions, 0 deletions

diff --git a/net/ipv6/ipcomp6.c b/net/ipv6/ipcomp6.c
new file mode 100644
index 000000000..b340d67eb
--- /dev/null
+++ b/net/ipv6/ipcomp6.c
@@ -0,0 +1,224 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * IP Payload Compression Protocol (IPComp) for IPv6 - RFC3173
+ *
+ * Copyright (C)2003 USAGI/WIDE Project
+ *
+ * Author	Mitsuru KANDA  <mk@linux-ipv6.org>
+ */
+/*
+ * [Memo]
+ *
+ * Outbound:
+ *  The compression of IP datagram MUST be done before AH/ESP processing,
+ *  fragmentation, and the addition of Hop-by-Hop/Routing header.
+ *
+ * Inbound:
+ *  The decompression of IP datagram MUST be done after the reassembly,
+ *  AH/ESP processing.
+ */
+
+#define pr_fmt(fmt) "IPv6: " fmt
+
+#include <linux/module.h>
+#include <net/ip.h>
+#include <net/xfrm.h>
+#include <net/ipcomp.h>
+#include <linux/crypto.h>
+#include <linux/err.h>
+#include <linux/pfkeyv2.h>
+#include <linux/random.h>
+#include <linux/percpu.h>
+#include <linux/smp.h>
+#include <linux/list.h>
+#include <linux/vmalloc.h>
+#include <linux/rtnetlink.h>
+#include <net/ip6_route.h>
+#include <net/icmp.h>
+#include <net/ipv6.h>
+#include <net/protocol.h>
+#include <linux/ipv6.h>
+#include <linux/icmpv6.h>
+#include <linux/mutex.h>
+
+static int ipcomp6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
+				u8 type, u8 code, int offset, __be32 info)
+{
+	struct net *net = dev_net(skb->dev);
+	__be32 spi;
+	const struct ipv6hdr *iph = (const struct ipv6hdr *)skb->data;
+	struct ip_comp_hdr *ipcomph =
+		(struct ip_comp_hdr *)(skb->data + offset);
+	struct xfrm_state *x;
+
+	if (type != ICMPV6_PKT_TOOBIG &&
+	    type != NDISC_REDIRECT)
+		return 0;
+
+	spi = htonl(ntohs(ipcomph->cpi));
+	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
+			      spi, IPPROTO_COMP, AF_INET6);
+	if (!x)
+		return 0;
+
+	if (type == NDISC_REDIRECT)
+		ip6_redirect(skb, net, skb->dev->ifindex, 0,
+			     sock_net_uid(net, NULL));
+	else
+		ip6_update_pmtu(skb, net, info, 0, 0, sock_net_uid(net, NULL));
+	xfrm_state_put(x);
+
+	return 0;
+}
+
+static struct lock_class_key xfrm_state_lock_key;
+static struct xfrm_state *ipcomp6_tunnel_create(struct xfrm_state *x)
+{
+	struct net *net = xs_net(x);
+	struct xfrm_state *t = NULL;
+
+	t = xfrm_state_alloc(net);
+	if (!t)
+		goto out;
+	lockdep_set_class(&t->lock, &xfrm_state_lock_key);
+
+	t->id.proto = IPPROTO_IPV6;
+	t->id.spi = xfrm6_tunnel_alloc_spi(net, (xfrm_address_t *)&x->props.saddr);
+	if (!t->id.spi)
+		goto error;
+
+	memcpy(t->id.daddr.a6, x->id.daddr.a6, sizeof(struct in6_addr));
+	memcpy(&t->sel, &x->sel, sizeof(t->sel));
+	t->props.family = AF_INET6;
+	t->props.mode = x->props.mode;
+	memcpy(t->props.saddr.a6, x->props.saddr.a6, sizeof(struct in6_addr));
+	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;
+}
+
+static int ipcomp6_tunnel_attach(struct xfrm_state *x)
+{
+	struct net *net = xs_net(x);
+	int err = 0;
+	struct xfrm_state *t = NULL;
+	__be32 spi;
+	u32 mark = x->mark.m & x->mark.v;
+
+	spi = xfrm6_tunnel_spi_lookup(net, (xfrm_address_t *)&x->props.saddr);
+	if (spi)
+		t = xfrm_state_lookup(net, mark, (xfrm_address_t *)&x->id.daddr,
+					      spi, IPPROTO_IPV6, AF_INET6);
+	if (!t) {
+		t = ipcomp6_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 ipcomp6_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 ipv6hdr);
+		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 = ipcomp6_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 ipcomp6_rcv_cb(struct sk_buff *skb, int err)
+{
+	return 0;
+}
+
+static const struct xfrm_type ipcomp6_type = {
+	.owner		= THIS_MODULE,
+	.proto		= IPPROTO_COMP,
+	.init_state	= ipcomp6_init_state,
+	.destructor	= ipcomp_destroy,
+	.input		= ipcomp_input,
+	.output		= ipcomp_output,
+};
+
+static struct xfrm6_protocol ipcomp6_protocol = {
+	.handler	= xfrm6_rcv,
+	.input_handler	= xfrm_input,
+	.cb_handler	= ipcomp6_rcv_cb,
+	.err_handler	= ipcomp6_err,
+	.priority	= 0,
+};
+
+static int __init ipcomp6_init(void)
+{
+	if (xfrm_register_type(&ipcomp6_type, AF_INET6) < 0) {
+		pr_info("%s: can't add xfrm type\n", __func__);
+		return -EAGAIN;
+	}
+	if (xfrm6_protocol_register(&ipcomp6_protocol, IPPROTO_COMP) < 0) {
+		pr_info("%s: can't add protocol\n", __func__);
+		xfrm_unregister_type(&ipcomp6_type, AF_INET6);
+		return -EAGAIN;
+	}
+	return 0;
+}
+
+static void __exit ipcomp6_fini(void)
+{
+	if (xfrm6_protocol_deregister(&ipcomp6_protocol, IPPROTO_COMP) < 0)
+		pr_info("%s: can't remove protocol\n", __func__);
+	xfrm_unregister_type(&ipcomp6_type, AF_INET6);
+}
+
+module_init(ipcomp6_init);
+module_exit(ipcomp6_fini);
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) for IPv6 - RFC3173");
+MODULE_AUTHOR("Mitsuru KANDA <mk@linux-ipv6.org>");
+
+MODULE_ALIAS_XFRM_TYPE(AF_INET6, XFRM_PROTO_COMP);