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, 189 insertions, 0 deletions

diff --git a/Documentation/networking/ipvlan.rst b/Documentation/networking/ipvlan.rst
new file mode 100644
index 000000000..895d0ccfd
--- /dev/null
+++ b/Documentation/networking/ipvlan.rst
@@ -0,0 +1,189 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+===================
+IPVLAN Driver HOWTO
+===================
+
+Initial Release:
+	Mahesh Bandewar <maheshb AT google.com>
+
+1. Introduction:
+================
+This is conceptually very similar to the macvlan driver with one major
+exception of using L3 for mux-ing /demux-ing among slaves. This property makes
+the master device share the L2 with its slave devices. I have developed this
+driver in conjunction with network namespaces and not sure if there is use case
+outside of it.
+
+
+2. Building and Installation:
+=============================
+
+In order to build the driver, please select the config item CONFIG_IPVLAN.
+The driver can be built into the kernel (CONFIG_IPVLAN=y) or as a module
+(CONFIG_IPVLAN=m).
+
+
+3. Configuration:
+=================
+
+There are no module parameters for this driver and it can be configured
+using IProute2/ip utility.
+::
+
+    ip link add link <master> name <slave> type ipvlan [ mode MODE ] [ FLAGS ]
+       where
+	 MODE: l3 (default) | l3s | l2
+	 FLAGS: bridge (default) | private | vepa
+
+e.g.
+
+    (a) Following will create IPvlan link with eth0 as master in
+	L3 bridge mode::
+
+	  bash# ip link add link eth0 name ipvl0 type ipvlan
+    (b) This command will create IPvlan link in L2 bridge mode::
+
+	  bash# ip link add link eth0 name ipvl0 type ipvlan mode l2 bridge
+
+    (c) This command will create an IPvlan device in L2 private mode::
+
+	  bash# ip link add link eth0 name ipvlan type ipvlan mode l2 private
+
+    (d) This command will create an IPvlan device in L2 vepa mode::
+
+	  bash# ip link add link eth0 name ipvlan type ipvlan mode l2 vepa
+
+
+4. Operating modes:
+===================
+
+IPvlan has two modes of operation - L2 and L3. For a given master device,
+you can select one of these two modes and all slaves on that master will
+operate in the same (selected) mode. The RX mode is almost identical except
+that in L3 mode the slaves won't receive any multicast / broadcast traffic.
+L3 mode is more restrictive since routing is controlled from the other (mostly)
+default namespace.
+
+4.1 L2 mode:
+------------
+
+In this mode TX processing happens on the stack instance attached to the
+slave device and packets are switched and queued to the master device to send
+out. In this mode the slaves will RX/TX multicast and broadcast (if applicable)
+as well.
+
+4.2 L3 mode:
+------------
+
+In this mode TX processing up to L3 happens on the stack instance attached
+to the slave device and packets are switched to the stack instance of the
+master device for the L2 processing and routing from that instance will be
+used before packets are queued on the outbound device. In this mode the slaves
+will not receive nor can send multicast / broadcast traffic.
+
+4.3 L3S mode:
+-------------
+
+This is very similar to the L3 mode except that iptables (conn-tracking)
+works in this mode and hence it is L3-symmetric (L3s). This will have slightly less
+performance but that shouldn't matter since you are choosing this mode over plain-L3
+mode to make conn-tracking work.
+
+5. Mode flags:
+==============
+
+At this time following mode flags are available
+
+5.1 bridge:
+-----------
+This is the default option. To configure the IPvlan port in this mode,
+user can choose to either add this option on the command-line or don't specify
+anything. This is the traditional mode where slaves can cross-talk among
+themselves apart from talking through the master device.
+
+5.2 private:
+------------
+If this option is added to the command-line, the port is set in private
+mode. i.e. port won't allow cross communication between slaves.
+
+5.3 vepa:
+---------
+If this is added to the command-line, the port is set in VEPA mode.
+i.e. port will offload switching functionality to the external entity as
+described in 802.1Qbg
+Note: VEPA mode in IPvlan has limitations. IPvlan uses the mac-address of the
+master-device, so the packets which are emitted in this mode for the adjacent
+neighbor will have source and destination mac same. This will make the switch /
+router send the redirect message.
+
+6. What to choose (macvlan vs. ipvlan)?
+=======================================
+
+These two devices are very similar in many regards and the specific use
+case could very well define which device to choose. if one of the following
+situations defines your use case then you can choose to use ipvlan:
+
+
+(a) The Linux host that is connected to the external switch / router has
+    policy configured that allows only one mac per port.
+(b) No of virtual devices created on a master exceed the mac capacity and
+    puts the NIC in promiscuous mode and degraded performance is a concern.
+(c) If the slave device is to be put into the hostile / untrusted network
+    namespace where L2 on the slave could be changed / misused.
+
+
+6. Example configuration:
+=========================
+
+::
+
+  +=============================================================+
+  |  Host: host1                                                |
+  |                                                             |
+  |   +----------------------+      +----------------------+    |
+  |   |   NS:ns0             |      |  NS:ns1              |    |
+  |   |                      |      |                      |    |
+  |   |                      |      |                      |    |
+  |   |        ipvl0         |      |         ipvl1        |    |
+  |   +----------#-----------+      +-----------#----------+    |
+  |              #                              #               |
+  |              ################################               |
+  |                              # eth0                         |
+  +==============================#==============================+
+
+
+(a) Create two network namespaces - ns0, ns1::
+
+	ip netns add ns0
+	ip netns add ns1
+
+(b) Create two ipvlan slaves on eth0 (master device)::
+
+	ip link add link eth0 ipvl0 type ipvlan mode l2
+	ip link add link eth0 ipvl1 type ipvlan mode l2
+
+(c) Assign slaves to the respective network namespaces::
+
+	ip link set dev ipvl0 netns ns0
+	ip link set dev ipvl1 netns ns1
+
+(d) Now switch to the namespace (ns0 or ns1) to configure the slave devices
+
+	- For ns0::
+
+		(1) ip netns exec ns0 bash
+		(2) ip link set dev ipvl0 up
+		(3) ip link set dev lo up
+		(4) ip -4 addr add 127.0.0.1 dev lo
+		(5) ip -4 addr add $IPADDR dev ipvl0
+		(6) ip -4 route add default via $ROUTER dev ipvl0
+
+	- For ns1::
+
+		(1) ip netns exec ns1 bash
+		(2) ip link set dev ipvl1 up
+		(3) ip link set dev lo up
+		(4) ip -4 addr add 127.0.0.1 dev lo
+		(5) ip -4 addr add $IPADDR dev ipvl1
+		(6) ip -4 route add default via $ROUTER dev ipvl1