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, 129 insertions, 0 deletions

diff --git a/scripts/show_delta b/scripts/show_delta
new file mode 100755
index 000000000..3755b6c6e
--- /dev/null
+++ b/scripts/show_delta
@@ -0,0 +1,129 @@
+#!/usr/bin/env python3
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# show_deltas: Read list of printk messages instrumented with
+# time data, and format with time deltas.
+#
+# Also, you can show the times relative to a fixed point.
+#
+# Copyright 2003 Sony Corporation
+#
+
+import sys
+import string
+
+def usage():
+	print ("""usage: show_delta [<options>] <filename>
+
+This program parses the output from a set of printk message lines which
+have time data prefixed because the CONFIG_PRINTK_TIME option is set, or
+the kernel command line option "time" is specified. When run with no
+options, the time information is converted to show the time delta between
+each printk line and the next.  When run with the '-b' option, all times
+are relative to a single (base) point in time.
+
+Options:
+  -h            Show this usage help.
+  -b <base>	Specify a base for time references.
+		<base> can be a number or a string.
+		If it is a string, the first message line
+		which matches (at the beginning of the
+		line) is used as the time reference.
+
+ex: $ dmesg >timefile
+    $ show_delta -b NET4 timefile
+
+will show times relative to the line in the kernel output
+starting with "NET4".
+""")
+	sys.exit(1)
+
+# returns a tuple containing the seconds and text for each message line
+# seconds is returned as a float
+# raise an exception if no timing data was found
+def get_time(line):
+	if line[0]!="[":
+		raise ValueError
+
+	# split on closing bracket
+	(time_str, rest) = string.split(line[1:],']',1)
+	time = string.atof(time_str)
+
+	#print "time=", time
+	return (time, rest)
+
+
+# average line looks like:
+# [    0.084282] VFS: Mounted root (romfs filesystem) readonly
+# time data is expressed in seconds.useconds,
+# convert_line adds a delta for each line
+last_time = 0.0
+def convert_line(line, base_time):
+	global last_time
+
+	try:
+		(time, rest) = get_time(line)
+	except:
+		# if any problem parsing time, don't convert anything
+		return line
+
+	if base_time:
+		# show time from base
+		delta = time - base_time
+	else:
+		# just show time from last line
+		delta = time - last_time
+		last_time = time
+
+	return ("[%5.6f < %5.6f >]" % (time, delta)) + rest
+
+def main():
+	base_str = ""
+	filein = ""
+	for arg in sys.argv[1:]:
+		if arg=="-b":
+			base_str = sys.argv[sys.argv.index("-b")+1]
+		elif arg=="-h":
+			usage()
+		else:
+			filein = arg
+
+	if not filein:
+		usage()
+
+	try:
+		lines = open(filein,"r").readlines()
+	except:
+		print ("Problem opening file: %s" % filein)
+		sys.exit(1)
+
+	if base_str:
+		print ('base= "%s"' % base_str)
+		# assume a numeric base.  If that fails, try searching
+		# for a matching line.
+		try:
+			base_time = float(base_str)
+		except:
+			# search for line matching <base> string
+			found = 0
+			for line in lines:
+				try:
+					(time, rest) = get_time(line)
+				except:
+					continue
+				if string.find(rest, base_str)==1:
+					base_time = time
+					found = 1
+					# stop at first match
+					break
+			if not found:
+				print ('Couldn\'t find line matching base pattern "%s"' % base_str)
+				sys.exit(1)
+	else:
+		base_time = 0.0
+
+	for line in lines:
+		print (convert_line(line, base_time),)
+
+if __name__ == "__main__":
+	main()