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, 194 insertions, 0 deletions

diff --git a/Documentation/userspace-api/netlink/netlink-raw.rst b/Documentation/userspace-api/netlink/netlink-raw.rst
new file mode 100644
index 000000000..aae296c17
--- /dev/null
+++ b/Documentation/userspace-api/netlink/netlink-raw.rst
@@ -0,0 +1,194 @@
+.. SPDX-License-Identifier: BSD-3-Clause
+
+======================================================
+Netlink specification support for raw Netlink families
+======================================================
+
+This document describes the additional properties required by raw Netlink
+families such as ``NETLINK_ROUTE`` which use the ``netlink-raw`` protocol
+specification.
+
+Specification
+=============
+
+The netlink-raw schema extends the :doc:`genetlink-legacy <genetlink-legacy>`
+schema with properties that are needed to specify the protocol numbers and
+multicast IDs used by raw netlink families. See :ref:`classic_netlink` for more
+information. The raw netlink families also make use of type-specific
+sub-messages.
+
+Globals
+-------
+
+protonum
+~~~~~~~~
+
+The ``protonum`` property is used to specify the protocol number to use when
+opening a netlink socket.
+
+.. code-block:: yaml
+
+  # SPDX-License-Identifier: ((GPL-2.0 WITH Linux-syscall-note) OR BSD-3-Clause)
+
+  name: rt-addr
+  protocol: netlink-raw
+  protonum: 0             # part of the NETLINK_ROUTE protocol
+
+
+Multicast group properties
+--------------------------
+
+value
+~~~~~
+
+The ``value`` property is used to specify the group ID to use for multicast
+group registration.
+
+.. code-block:: yaml
+
+  mcast-groups:
+    list:
+      -
+        name: rtnlgrp-ipv4-ifaddr
+        value: 5
+      -
+        name: rtnlgrp-ipv6-ifaddr
+        value: 9
+      -
+        name: rtnlgrp-mctp-ifaddr
+        value: 34
+
+Sub-messages
+------------
+
+Several raw netlink families such as
+:ref:`rt-link<netlink-rt-link>` and
+:ref:`tc<netlink-tc>` use attribute nesting as an
+abstraction to carry module specific information.
+
+Conceptually it looks as follows::
+
+    [OUTER NEST OR MESSAGE LEVEL]
+      [GENERIC ATTR 1]
+      [GENERIC ATTR 2]
+      [GENERIC ATTR 3]
+      [GENERIC ATTR - wrapper]
+        [MODULE SPECIFIC ATTR 1]
+        [MODULE SPECIFIC ATTR 2]
+
+The ``GENERIC ATTRs`` at the outer level are defined in the core (or rt_link or
+core TC), while specific drivers, TC classifiers, qdiscs etc. can carry their
+own information wrapped in the ``GENERIC ATTR - wrapper``. Even though the
+example above shows attributes nesting inside the wrapper, the modules generally
+have full freedom to define the format of the nest. In practice the payload of
+the wrapper attr has very similar characteristics to a netlink message. It may
+contain a fixed header / structure, netlink attributes, or both. Because of
+those shared characteristics we refer to the payload of the wrapper attribute as
+a sub-message.
+
+A sub-message attribute uses the value of another attribute as a selector key to
+choose the right sub-message format. For example if the following attribute has
+already been decoded:
+
+.. code-block:: json
+
+  { "kind": "gre" }
+
+and we encounter the following attribute spec:
+
+.. code-block:: yaml
+
+  -
+    name: data
+    type: sub-message
+    sub-message: linkinfo-data-msg
+    selector: kind
+
+Then we look for a sub-message definition called ``linkinfo-data-msg`` and use
+the value of the ``kind`` attribute i.e. ``gre`` as the key to choose the
+correct format for the sub-message:
+
+.. code-block:: yaml
+
+  sub-messages:
+    name: linkinfo-data-msg
+    formats:
+      -
+        value: bridge
+        attribute-set: linkinfo-bridge-attrs
+      -
+        value: gre
+        attribute-set: linkinfo-gre-attrs
+      -
+        value: geneve
+        attribute-set: linkinfo-geneve-attrs
+
+This would decode the attribute value as a sub-message with the attribute-set
+called ``linkinfo-gre-attrs`` as the attribute space.
+
+A sub-message can have an optional ``fixed-header`` followed by zero or more
+attributes from an ``attribute-set``. For example the following
+``tc-options-msg`` sub-message defines message formats that use a mixture of
+``fixed-header``, ``attribute-set`` or both together:
+
+.. code-block:: yaml
+
+  sub-messages:
+    -
+      name: tc-options-msg
+      formats:
+        -
+          value: bfifo
+          fixed-header: tc-fifo-qopt
+        -
+          value: cake
+          attribute-set: tc-cake-attrs
+        -
+          value: netem
+          fixed-header: tc-netem-qopt
+          attribute-set: tc-netem-attrs
+
+Note that a selector attribute must appear in a netlink message before any
+sub-message attributes that depend on it.
+
+If an attribute such as ``kind`` is defined at more than one nest level, then a
+sub-message selector will be resolved using the value 'closest' to the selector.
+For example, if the same attribute name is defined in a nested ``attribute-set``
+alongside a sub-message selector and also in a top level ``attribute-set``, then
+the selector will be resolved using the value 'closest' to the selector. If the
+value is not present in the message at the same level as defined in the spec
+then this is an error.
+
+Nested struct definitions
+-------------------------
+
+Many raw netlink families such as :ref:`tc<netlink-tc>`
+make use of nested struct definitions. The ``netlink-raw`` schema makes it
+possible to embed a struct within a struct definition using the ``struct``
+property. For example, the following struct definition embeds the
+``tc-ratespec`` struct definition for both the ``rate`` and the ``peakrate``
+members of ``struct tc-tbf-qopt``.
+
+.. code-block:: yaml
+
+  -
+    name: tc-tbf-qopt
+    type: struct
+    members:
+      -
+        name: rate
+        type: binary
+        struct: tc-ratespec
+      -
+        name: peakrate
+        type: binary
+        struct: tc-ratespec
+      -
+        name: limit
+        type: u32
+      -
+        name: buffer
+        type: u32
+      -
+        name: mtu
+        type: u32