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, 137 insertions, 0 deletions

diff --git a/Documentation/sphinx/kernel_feat.py b/Documentation/sphinx/kernel_feat.py
new file mode 100644
index 000000000..bdc0fef5c
--- /dev/null
+++ b/Documentation/sphinx/kernel_feat.py
@@ -0,0 +1,137 @@
+# coding=utf-8
+# SPDX-License-Identifier: GPL-2.0
+#
+"""
+    kernel-feat
+    ~~~~~~~~~~~
+
+    Implementation of the ``kernel-feat`` reST-directive.
+
+    :copyright:  Copyright (C) 2016  Markus Heiser
+    :copyright:  Copyright (C) 2016-2019  Mauro Carvalho Chehab
+    :maintained-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
+    :license:    GPL Version 2, June 1991 see Linux/COPYING for details.
+
+    The ``kernel-feat`` (:py:class:`KernelFeat`) directive calls the
+    tools/docs/get_feat.pl script to parse the Kernel ABI files.
+
+    Overview of directive's argument and options.
+
+    .. code-block:: rst
+
+        .. kernel-feat:: <ABI directory location>
+            :debug:
+
+    The argument ``<ABI directory location>`` is required. It contains the
+    location of the ABI files to be parsed.
+
+    ``debug``
+      Inserts a code-block with the *raw* reST. Sometimes it is helpful to see
+      what reST is generated.
+
+"""
+
+import codecs
+import os
+import re
+import sys
+
+from docutils import nodes, statemachine
+from docutils.statemachine import ViewList
+from docutils.parsers.rst import directives, Directive
+from sphinx.util.docutils import switch_source_input
+
+srctree = os.path.abspath(os.environ["srctree"])
+sys.path.insert(0, os.path.join(srctree, "tools/lib/python"))
+
+from feat.parse_features import ParseFeature                # pylint: disable=C0413
+
+def ErrorString(exc):  # Shamelessly stolen from docutils
+    return f'{exc.__class__.__name}: {exc}'
+
+__version__  = '1.0'
+
+def setup(app):
+
+    app.add_directive("kernel-feat", KernelFeat)
+    return dict(
+        version = __version__
+        , parallel_read_safe = True
+        , parallel_write_safe = True
+    )
+
+class KernelFeat(Directive):
+
+    """KernelFeat (``kernel-feat``) directive"""
+
+    required_arguments = 1
+    optional_arguments = 2
+    has_content = False
+    final_argument_whitespace = True
+
+    option_spec = {
+        "debug"     : directives.flag
+    }
+
+    def warn(self, message, **replace):
+        replace["fname"]   = self.state.document.current_source
+        replace["line_no"] = replace.get("line_no", self.lineno)
+        message = ("%(fname)s:%(line_no)s: [kernel-feat WARN] : " + message) % replace
+        self.state.document.settings.env.app.warn(message, prefix="")
+
+    def run(self):
+        doc = self.state.document
+        if not doc.settings.file_insertion_enabled:
+            raise self.warning("docutils: file insertion disabled")
+
+        env = doc.settings.env
+
+        srctree = os.path.abspath(os.environ["srctree"])
+
+        feature_dir = os.path.join(srctree, 'Documentation', self.arguments[0])
+
+        feat = ParseFeature(feature_dir, False, True)
+        feat.parse()
+
+        if len(self.arguments) > 1:
+            arch = self.arguments[1]
+            lines = feat.output_arch_table(arch)
+        else:
+            lines = feat.output_matrix()
+
+        line_regex = re.compile(r"^\.\. FILE (\S+)$")
+
+        out_lines = ""
+
+        for line in lines.split("\n"):
+            match = line_regex.search(line)
+            if match:
+                fname = match.group(1)
+
+                # Add the file to Sphinx build dependencies
+                env.note_dependency(os.path.abspath(fname))
+            else:
+                out_lines += line + "\n"
+
+        nodeList = self.nestedParse(out_lines, self.arguments[0])
+        return nodeList
+
+    def nestedParse(self, lines, fname):
+        content = ViewList()
+        node    = nodes.section()
+
+        if "debug" in self.options:
+            code_block = "\n\n.. code-block:: rst\n    :linenos:\n"
+            for l in lines.split("\n"):
+                code_block += "\n    " + l
+            lines = code_block + "\n\n"
+
+        for c, l in enumerate(lines.split("\n")):
+            content.append(l, fname, c)
+
+        buf  = self.state.memo.title_styles, self.state.memo.section_level, self.state.memo.reporter
+
+        with switch_source_input(self.state, content):
+            self.state.nested_parse(content, 0, node, match_titles=1)
+
+        return node.children