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, 229 insertions, 0 deletions

diff --git a/net/openvswitch/vport-netdev.c b/net/openvswitch/vport-netdev.c
new file mode 100644
index 000000000..44808cd0f
--- /dev/null
+++ b/net/openvswitch/vport-netdev.c
@@ -0,0 +1,229 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2007-2012 Nicira, Inc.
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/if_arp.h>
+#include <linux/if_bridge.h>
+#include <linux/if_vlan.h>
+#include <linux/kernel.h>
+#include <linux/llc.h>
+#include <linux/rtnetlink.h>
+#include <linux/skbuff.h>
+#include <linux/openvswitch.h>
+#include <linux/export.h>
+
+#include <net/ip_tunnels.h>
+#include <net/rtnetlink.h>
+
+#include "datapath.h"
+#include "vport.h"
+#include "vport-internal_dev.h"
+#include "vport-netdev.h"
+
+static struct vport_ops ovs_netdev_vport_ops;
+
+/* Must be called with rcu_read_lock. */
+static void netdev_port_receive(struct sk_buff *skb)
+{
+	struct vport *vport;
+
+	vport = ovs_netdev_get_vport(skb->dev);
+	if (unlikely(!vport))
+		goto error;
+
+	if (unlikely(skb_warn_if_lro(skb)))
+		goto error;
+
+	/* Make our own copy of the packet.  Otherwise we will mangle the
+	 * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
+	 */
+	skb = skb_share_check(skb, GFP_ATOMIC);
+	if (unlikely(!skb))
+		return;
+
+	if (skb->dev->type == ARPHRD_ETHER)
+		skb_push_rcsum(skb, ETH_HLEN);
+
+	ovs_vport_receive(vport, skb, skb_tunnel_info(skb));
+	return;
+error:
+	kfree_skb(skb);
+}
+
+/* Called with rcu_read_lock and bottom-halves disabled. */
+static rx_handler_result_t netdev_frame_hook(struct sk_buff **pskb)
+{
+	struct sk_buff *skb = *pskb;
+
+	if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
+		return RX_HANDLER_PASS;
+
+	netdev_port_receive(skb);
+	return RX_HANDLER_CONSUMED;
+}
+
+static struct net_device *get_dpdev(const struct datapath *dp)
+{
+	struct vport *local;
+
+	local = ovs_vport_ovsl(dp, OVSP_LOCAL);
+	return local->dev;
+}
+
+static struct vport *ovs_netdev_link(struct vport *vport)
+{
+	int err;
+
+	if (WARN_ON_ONCE(!vport->dev)) {
+		err = -ENODEV;
+		goto error_free_vport;
+	}
+
+	rtnl_lock();
+	/* Do not link devices that are not registered to avoid a potential
+	 * race with the NETDEV_UNREGISTER notification in dp_device_event().
+	 */
+	if (vport->dev->reg_state != NETREG_REGISTERED) {
+		err = -ENODEV;
+		goto error_put_unlock;
+	}
+
+	err = netdev_master_upper_dev_link(vport->dev,
+					   get_dpdev(vport->dp),
+					   NULL, NULL, NULL);
+	if (err)
+		goto error_put_unlock;
+
+	err = netdev_rx_handler_register(vport->dev, netdev_frame_hook,
+					 vport);
+	if (err)
+		goto error_master_upper_dev_unlink;
+
+	dev_disable_lro(vport->dev);
+	dev_set_promiscuity(vport->dev, 1);
+	vport->dev->priv_flags |= IFF_OVS_DATAPATH;
+	rtnl_unlock();
+
+	return vport;
+
+error_master_upper_dev_unlink:
+	netdev_upper_dev_unlink(vport->dev, get_dpdev(vport->dp));
+error_put_unlock:
+	netdev_put(vport->dev, &vport->dev_tracker);
+	rtnl_unlock();
+error_free_vport:
+	ovs_vport_free(vport);
+	return ERR_PTR(err);
+}
+
+static struct vport *netdev_create(const struct vport_parms *parms)
+{
+	struct vport *vport;
+	int err;
+
+	vport = ovs_vport_alloc(0, &ovs_netdev_vport_ops, parms);
+	if (IS_ERR(vport))
+		return vport;
+
+	vport->dev = dev_get_by_name(ovs_dp_get_net(vport->dp), parms->name);
+	if (!vport->dev) {
+		err = -ENODEV;
+		goto error_free_vport;
+	}
+	netdev_tracker_alloc(vport->dev, &vport->dev_tracker, GFP_KERNEL);
+
+	/* Ensure that the provided name is not an alias. */
+	if (strcmp(parms->name, ovs_vport_name(vport))) {
+		err = -ENODEV;
+		goto error_put;
+	}
+
+	if (vport->dev->flags & IFF_LOOPBACK ||
+	    (vport->dev->type != ARPHRD_ETHER &&
+	     vport->dev->type != ARPHRD_NONE) ||
+	    ovs_is_internal_dev(vport->dev)) {
+		err = -EINVAL;
+		goto error_put;
+	}
+
+	return ovs_netdev_link(vport);
+error_put:
+	netdev_put(vport->dev, &vport->dev_tracker);
+error_free_vport:
+	ovs_vport_free(vport);
+	return ERR_PTR(err);
+}
+
+static void vport_netdev_free(struct rcu_head *rcu)
+{
+	struct vport *vport = container_of(rcu, struct vport, rcu);
+
+	netdev_put(vport->dev, &vport->dev_tracker);
+	ovs_vport_free(vport);
+}
+
+void ovs_netdev_detach_dev(struct vport *vport)
+{
+	ASSERT_RTNL();
+	netdev_rx_handler_unregister(vport->dev);
+	netdev_upper_dev_unlink(vport->dev,
+				netdev_master_upper_dev_get(vport->dev));
+	dev_set_promiscuity(vport->dev, -1);
+
+	/* paired with smp_mb() in netdev_destroy() */
+	smp_wmb();
+
+	vport->dev->priv_flags &= ~IFF_OVS_DATAPATH;
+}
+
+static void netdev_destroy(struct vport *vport)
+{
+	/* When called from ovs_db_notify_wq() after a dp_device_event(), the
+	 * port has already been detached, so we can avoid taking the RTNL by
+	 * checking this first.
+	 */
+	if (netif_is_ovs_port(vport->dev)) {
+		rtnl_lock();
+		/* Check again while holding the lock to ensure we don't race
+		 * with the netdev notifier and detach twice.
+		 */
+		if (netif_is_ovs_port(vport->dev))
+			ovs_netdev_detach_dev(vport);
+		rtnl_unlock();
+	}
+
+	/* paired with smp_wmb() in ovs_netdev_detach_dev() */
+	smp_mb();
+
+	call_rcu(&vport->rcu, vport_netdev_free);
+}
+
+/* Returns null if this device is not attached to a datapath. */
+struct vport *ovs_netdev_get_vport(struct net_device *dev)
+{
+	if (likely(netif_is_ovs_port(dev)))
+		return (struct vport *)
+			rcu_dereference_rtnl(dev->rx_handler_data);
+	else
+		return NULL;
+}
+
+static struct vport_ops ovs_netdev_vport_ops = {
+	.type		= OVS_VPORT_TYPE_NETDEV,
+	.create		= netdev_create,
+	.destroy	= netdev_destroy,
+	.send		= dev_queue_xmit,
+};
+
+int __init ovs_netdev_init(void)
+{
+	return ovs_vport_ops_register(&ovs_netdev_vport_ops);
+}
+
+void ovs_netdev_exit(void)
+{
+	ovs_vport_ops_unregister(&ovs_netdev_vport_ops);
+}