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, 180 insertions, 0 deletions

diff --git a/net/sysctl_net.c b/net/sysctl_net.c
new file mode 100644
index 000000000..e190a639e
--- /dev/null
+++ b/net/sysctl_net.c
@@ -0,0 +1,180 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* -*- linux-c -*-
+ * sysctl_net.c: sysctl interface to net subsystem.
+ *
+ * Begun April 1, 1996, Mike Shaver.
+ * Added /proc/sys/net directories for each protocol family. [MS]
+ *
+ * Revision 1.2  1996/05/08  20:24:40  shaver
+ * Added bits for NET_BRIDGE and the NET_IPV4_ARP stuff and
+ * NET_IPV4_IP_FORWARD.
+ *
+ *
+ */
+
+#include <linux/mm.h>
+#include <linux/export.h>
+#include <linux/sysctl.h>
+#include <linux/nsproxy.h>
+
+#include <net/sock.h>
+
+#ifdef CONFIG_INET
+#include <net/ip.h>
+#endif
+
+#ifdef CONFIG_NET
+#include <linux/if_ether.h>
+#endif
+
+static struct ctl_table_set *
+net_ctl_header_lookup(struct ctl_table_root *root)
+{
+	return &current->nsproxy->net_ns->sysctls;
+}
+
+static int is_seen(struct ctl_table_set *set)
+{
+	return &current->nsproxy->net_ns->sysctls == set;
+}
+
+/* Return standard mode bits for table entry. */
+static int net_ctl_permissions(struct ctl_table_header *head,
+			       const struct ctl_table *table)
+{
+	struct net *net = container_of(head->set, struct net, sysctls);
+
+	/* Allow network administrator to have same access as root. */
+	if (ns_capable_noaudit(net->user_ns, CAP_NET_ADMIN)) {
+		int mode = (table->mode >> 6) & 7;
+		return (mode << 6) | (mode << 3) | mode;
+	}
+
+	return table->mode;
+}
+
+static void net_ctl_set_ownership(struct ctl_table_header *head,
+				  kuid_t *uid, kgid_t *gid)
+{
+	struct net *net = container_of(head->set, struct net, sysctls);
+	kuid_t ns_root_uid;
+	kgid_t ns_root_gid;
+
+	ns_root_uid = make_kuid(net->user_ns, 0);
+	if (uid_valid(ns_root_uid))
+		*uid = ns_root_uid;
+
+	ns_root_gid = make_kgid(net->user_ns, 0);
+	if (gid_valid(ns_root_gid))
+		*gid = ns_root_gid;
+}
+
+static struct ctl_table_root net_sysctl_root = {
+	.lookup = net_ctl_header_lookup,
+	.permissions = net_ctl_permissions,
+	.set_ownership = net_ctl_set_ownership,
+};
+
+static int __net_init sysctl_net_init(struct net *net)
+{
+	setup_sysctl_set(&net->sysctls, &net_sysctl_root, is_seen);
+	return 0;
+}
+
+static void __net_exit sysctl_net_exit(struct net *net)
+{
+	retire_sysctl_set(&net->sysctls);
+}
+
+static struct pernet_operations sysctl_pernet_ops = {
+	.init = sysctl_net_init,
+	.exit = sysctl_net_exit,
+};
+
+static struct ctl_table_header *net_header;
+__init int net_sysctl_init(void)
+{
+	static struct ctl_table empty[1];
+	int ret = -ENOMEM;
+	/* Avoid limitations in the sysctl implementation by
+	 * registering "/proc/sys/net" as an empty directory not in a
+	 * network namespace.
+	 */
+	net_header = register_sysctl_sz("net", empty, 0);
+	if (!net_header)
+		goto out;
+	ret = register_pernet_subsys(&sysctl_pernet_ops);
+	if (ret)
+		goto out1;
+out:
+	return ret;
+out1:
+	unregister_sysctl_table(net_header);
+	net_header = NULL;
+	goto out;
+}
+
+/* Return error when sysctls for non-init netns are unsafe by verifying:
+ * 1) being read-only, or
+ * 2) having a data pointer which points outside of the global kernel/module
+ *    data segment, and rather into the heap where a per-net object was
+ *    allocated.
+ */
+static int ensure_safe_net_sysctl(struct net *net, const char *path,
+				  const struct ctl_table *table,
+				  size_t table_size)
+{
+	const struct ctl_table *ent;
+
+	pr_debug("Registering net sysctl (net %p): %s\n", net, path);
+	ent = table;
+	for (size_t i = 0; i < table_size; ent++, i++) {
+		unsigned long addr;
+		const char *where;
+
+		pr_debug("  procname=%s mode=%o proc_handler=%ps data=%p\n",
+			 ent->procname, ent->mode, ent->proc_handler, ent->data);
+
+		/* If it's not writable inside the netns, then it can't hurt. */
+		if ((ent->mode & 0222) == 0) {
+			pr_debug("    Not writable by anyone\n");
+			continue;
+		}
+
+		/* Where does data point? */
+		addr = (unsigned long)ent->data;
+		if (is_module_address(addr))
+			where = "module";
+		else if (is_kernel_core_data(addr))
+			where = "kernel";
+		else
+			continue;
+
+		/* Warn on netns leak. */
+		WARN(1, "sysctl %s/%s: data points to %s global data: %ps\n",
+		     path, ent->procname, where, ent->data);
+
+		return -EACCES;
+	}
+
+	return 0;
+}
+
+struct ctl_table_header *register_net_sysctl_sz(struct net *net,
+						const char *path,
+						const struct ctl_table *table,
+						size_t table_size)
+{
+	if (!net_eq(net, &init_net))
+		if (ensure_safe_net_sysctl(net, path, table, table_size))
+			return NULL;
+
+	return __register_sysctl_table(&net->sysctls, path, table, table_size);
+}
+EXPORT_SYMBOL_GPL(register_net_sysctl_sz);
+
+void unregister_net_sysctl_table(struct ctl_table_header *header)
+{
+	unregister_sysctl_table(header);
+}
+EXPORT_SYMBOL_GPL(unregister_net_sysctl_table);