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, 170 insertions, 0 deletions

diff --git a/Documentation/driver-api/regulator.rst b/Documentation/driver-api/regulator.rst
new file mode 100644
index 000000000..b43c78eb2
--- /dev/null
+++ b/Documentation/driver-api/regulator.rst
@@ -0,0 +1,170 @@
+.. Copyright 2007-2008 Wolfson Microelectronics
+
+..   This documentation is free software; you can redistribute
+..   it and/or modify it under the terms of the GNU General Public
+..   License version 2 as published by the Free Software Foundation.
+
+=================================
+Voltage and current regulator API
+=================================
+
+:Author: Liam Girdwood
+:Author: Mark Brown
+
+Introduction
+============
+
+This framework is designed to provide a standard kernel interface to
+control voltage and current regulators.
+
+The intention is to allow systems to dynamically control regulator power
+output in order to save power and prolong battery life. This applies to
+both voltage regulators (where voltage output is controllable) and
+current sinks (where current limit is controllable).
+
+Note that additional (and currently more complete) documentation is
+available in the Linux kernel source under
+``Documentation/power/regulator``.
+
+Glossary
+--------
+
+The regulator API uses a number of terms which may not be familiar:
+
+Regulator
+
+    Electronic device that supplies power to other devices. Most regulators
+    can enable and disable their output and some can also control their
+    output voltage or current.
+
+Consumer
+
+    Electronic device which consumes power provided by a regulator. These
+    may either be static, requiring only a fixed supply, or dynamic,
+    requiring active management of the regulator at runtime.
+
+Power Domain
+
+    The electronic circuit supplied by a given regulator, including the
+    regulator and all consumer devices. The configuration of the regulator
+    is shared between all the components in the circuit.
+
+Power Management Integrated Circuit (PMIC)
+
+    An IC which contains numerous regulators and often also other
+    subsystems. In an embedded system the primary PMIC is often equivalent
+    to a combination of the PSU and southbridge in a desktop system.
+
+Consumer driver interface
+=========================
+
+This offers a similar API to the kernel clock framework. Consumer
+drivers use `get <#API-regulator-get>`__ and
+`put <#API-regulator-put>`__ operations to acquire and release
+regulators. Functions are provided to `enable <#API-regulator-enable>`__
+and `disable <#API-regulator-disable>`__ the regulator and to get and
+set the runtime parameters of the regulator.
+
+When requesting regulators consumers use symbolic names for their
+supplies, such as "Vcc", which are mapped into actual regulator devices
+by the machine interface.
+
+A stub version of this API is provided when the regulator framework is
+not in use in order to minimise the need to use ifdefs.
+
+Enabling and disabling
+----------------------
+
+The regulator API provides reference counted enabling and disabling of
+regulators. Consumer devices use the :c:func:`regulator_enable()` and
+:c:func:`regulator_disable()` functions to enable and disable
+regulators. Calls to the two functions must be balanced.
+
+Note that since multiple consumers may be using a regulator and machine
+constraints may not allow the regulator to be disabled there is no
+guarantee that calling :c:func:`regulator_disable()` will actually
+cause the supply provided by the regulator to be disabled. Consumer
+drivers should assume that the regulator may be enabled at all times.
+
+Configuration
+-------------
+
+Some consumer devices may need to be able to dynamically configure their
+supplies. For example, MMC drivers may need to select the correct
+operating voltage for their cards. This may be done while the regulator
+is enabled or disabled.
+
+The :c:func:`regulator_set_voltage()` and
+:c:func:`regulator_set_current_limit()` functions provide the primary
+interface for this. Both take ranges of voltages and currents, supporting
+drivers that do not require a specific value (eg, CPU frequency scaling
+normally permits the CPU to use a wider range of supply voltages at lower
+frequencies but does not require that the supply voltage be lowered). Where
+an exact value is required both minimum and maximum values should be
+identical.
+
+Callbacks
+---------
+
+Callbacks may also be registered for events such as regulation failures.
+
+Regulator driver interface
+==========================
+
+Drivers for regulator chips register the regulators with the regulator
+core, providing operations structures to the core. A notifier interface
+allows error conditions to be reported to the core.
+
+Registration should be triggered by explicit setup done by the platform,
+supplying a struct regulator_init_data for the regulator
+containing constraint and supply information.
+
+Machine interface
+=================
+
+This interface provides a way to define how regulators are connected to
+consumers on a given system and what the valid operating parameters are
+for the system.
+
+Supplies
+--------
+
+Regulator supplies are specified using struct
+:c:type:`regulator_consumer_supply`. This is done at driver registration
+time as part of the machine constraints.
+
+Constraints
+-----------
+
+As well as defining the connections the machine interface also provides
+constraints defining the operations that clients are allowed to perform
+and the parameters that may be set. This is required since generally
+regulator devices will offer more flexibility than it is safe to use on
+a given system, for example supporting higher supply voltages than the
+consumers are rated for.
+
+This is done at driver registration time` by providing a
+struct regulation_constraints.
+
+The constraints may also specify an initial configuration for the
+regulator in the constraints, which is particularly useful for use with
+static consumers.
+
+API reference
+=============
+
+Due to limitations of the kernel documentation framework and the
+existing layout of the source code the entire regulator API is
+documented here.
+
+.. kernel-doc:: include/linux/regulator/consumer.h
+   :internal:
+
+.. kernel-doc:: include/linux/regulator/machine.h
+   :internal:
+
+.. kernel-doc:: include/linux/regulator/driver.h
+   :internal:
+
+.. kernel-doc:: drivers/regulator/core.c
+   :export: