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, 135 insertions, 0 deletions

diff --git a/Documentation/admin-guide/kernel-parameters.rst b/Documentation/admin-guide/kernel-parameters.rst
new file mode 100644
index 000000000..02a725536
--- /dev/null
+++ b/Documentation/admin-guide/kernel-parameters.rst
@@ -0,0 +1,135 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+.. _kernelparameters:
+
+The kernel's command-line parameters
+====================================
+
+The following is a consolidated list of the kernel parameters as implemented
+by the __setup(), early_param(), core_param() and module_param() macros
+and sorted into English Dictionary order (defined as ignoring all
+punctuation and sorting digits before letters in a case insensitive
+manner), and with descriptions where known.
+
+The kernel parses parameters from the kernel command line up to "``--``";
+if it doesn't recognize a parameter and it doesn't contain a '.', the
+parameter gets passed to init: parameters with '=' go into init's
+environment, others are passed as command line arguments to init.
+Everything after "``--``" is passed as an argument to init.
+
+Module parameters can be specified in two ways: via the kernel command
+line with a module name prefix, or via modprobe, e.g.::
+
+	(kernel command line) usbcore.blinkenlights=1
+	(modprobe command line) modprobe usbcore blinkenlights=1
+
+Parameters for modules which are built into the kernel need to be
+specified on the kernel command line.  modprobe looks through the
+kernel command line (/proc/cmdline) and collects module parameters
+when it loads a module, so the kernel command line can be used for
+loadable modules too.
+
+This document may not be entirely up to date and comprehensive. The command
+"modinfo -p ${modulename}" shows a current list of all parameters of a loadable
+module. Loadable modules, after being loaded into the running kernel, also
+reveal their parameters in /sys/module/${modulename}/parameters/. Some of these
+parameters may be changed at runtime by the command
+``echo -n ${value} > /sys/module/${modulename}/parameters/${parm}``.
+
+Special handling
+----------------
+
+Hyphens (dashes) and underscores are equivalent in parameter names, so::
+
+	log_buf_len=1M print-fatal-signals=1
+
+can also be entered as::
+
+	log-buf-len=1M print_fatal_signals=1
+
+Double-quotes can be used to protect spaces in values, e.g.::
+
+	param="spaces in here"
+
+cpu lists
+~~~~~~~~~
+
+Some kernel parameters take a list of CPUs as a value, e.g.  isolcpus,
+nohz_full, irqaffinity, rcu_nocbs.  The format of this list is:
+
+	<cpu number>,...,<cpu number>
+
+or
+
+	<cpu number>-<cpu number>
+	(must be a positive range in ascending order)
+
+or a mixture
+
+<cpu number>,...,<cpu number>-<cpu number>
+
+Note that for the special case of a range one can split the range into equal
+sized groups and for each group use some amount from the beginning of that
+group:
+
+	<cpu number>-<cpu number>:<used size>/<group size>
+
+For example one can add to the command line following parameter:
+
+	isolcpus=1,2,10-20,100-2000:2/25
+
+where the final item represents CPUs 100,101,125,126,150,151,...
+
+The value "N" can be used to represent the numerically last CPU on the system,
+i.e "foo_cpus=16-N" would be equivalent to "16-31" on a 32 core system.
+
+Keep in mind that "N" is dynamic, so if system changes cause the bitmap width
+to change, such as less cores in the CPU list, then N and any ranges using N
+will also change.  Use the same on a small 4 core system, and "16-N" becomes
+"16-3" and now the same boot input will be flagged as invalid (start > end).
+
+The special case-tolerant group name "all" has a meaning of selecting all CPUs,
+so that "nohz_full=all" is the equivalent of "nohz_full=0-N".
+
+The semantics of "N" and "all" is supported on a level of bitmaps and holds for
+all users of bitmap_parselist().
+
+Metric suffixes
+~~~~~~~~~~~~~~~
+
+The [KMG] suffix is commonly described after a number of kernel
+parameter values. 'K', 'M', 'G', 'T', 'P', and 'E' suffixes are allowed.
+These letters represent the _binary_ multipliers 'Kilo', 'Mega', 'Giga',
+'Tera', 'Peta', and 'Exa', equaling 2^10, 2^20, 2^30, 2^40, 2^50, and
+2^60 bytes respectively. Such letter suffixes can also be entirely omitted.
+
+Kernel Build Options
+--------------------
+
+The parameters listed below are only valid if certain kernel build options
+were enabled and if respective hardware is present. This list should be kept
+in alphabetical order. The text in square brackets at the beginning
+of each description states the restrictions within which a parameter
+is applicable.
+
+Parameters denoted with BOOT are actually interpreted by the boot
+loader, and have no meaning to the kernel directly.
+Do not modify the syntax of boot loader parameters without extreme
+need or coordination with <Documentation/arch/x86/boot.rst>.
+
+There are also arch-specific kernel-parameters not documented here.
+
+Note that ALL kernel parameters listed below are CASE SENSITIVE, and that
+a trailing = on the name of any parameter states that the parameter will
+be entered as an environment variable, whereas its absence indicates that
+it will appear as a kernel argument readable via /proc/cmdline by programs
+running once the system is up.
+
+The number of kernel parameters is not limited, but the length of the
+complete command line (parameters including spaces etc.) is limited to
+a fixed number of characters. This limit depends on the architecture
+and is between 256 and 4096 characters. It is defined in the file
+./include/uapi/asm-generic/setup.h as COMMAND_LINE_SIZE.
+
+.. include:: kernel-parameters.txt
+   :literal: