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, 260 insertions, 0 deletions

diff --git a/scripts/cleanpatch b/scripts/cleanpatch
new file mode 100755
index 000000000..9f1755271
--- /dev/null
+++ b/scripts/cleanpatch
@@ -0,0 +1,260 @@
+#!/usr/bin/env perl
+# SPDX-License-Identifier: GPL-2.0
+#
+# Clean a patch file -- or directory of patch files -- of stealth whitespace.
+# WARNING: this can be a highly destructive operation.  Use with caution.
+#
+
+use warnings;
+use bytes;
+use File::Basename;
+
+# Default options
+$max_width = 79;
+
+# Clean up space-tab sequences, either by removing spaces or
+# replacing them with tabs.
+sub clean_space_tabs($)
+{
+    no bytes;			# Tab alignment depends on characters
+
+    my($li) = @_;
+    my($lo) = '';
+    my $pos = 0;
+    my $nsp = 0;
+    my($i, $c);
+
+    for ($i = 0; $i < length($li); $i++) {
+	$c = substr($li, $i, 1);
+	if ($c eq "\t") {
+	    my $npos = ($pos+$nsp+8) & ~7;
+	    my $ntab = ($npos >> 3) - ($pos >> 3);
+	    $lo .= "\t" x $ntab;
+	    $pos = $npos;
+	    $nsp = 0;
+	} elsif ($c eq "\n" || $c eq "\r") {
+	    $lo .= " " x $nsp;
+	    $pos += $nsp;
+	    $nsp = 0;
+	    $lo .= $c;
+	    $pos = 0;
+	} elsif ($c eq " ") {
+	    $nsp++;
+	} else {
+	    $lo .= " " x $nsp;
+	    $pos += $nsp;
+	    $nsp = 0;
+	    $lo .= $c;
+	    $pos++;
+	}
+    }
+    $lo .= " " x $nsp;
+    return $lo;
+}
+
+# Compute the visual width of a string
+sub strwidth($) {
+    no bytes;			# Tab alignment depends on characters
+
+    my($li) = @_;
+    my($c, $i);
+    my $pos = 0;
+    my $mlen = 0;
+
+    for ($i = 0; $i < length($li); $i++) {
+	$c = substr($li,$i,1);
+	if ($c eq "\t") {
+	    $pos = ($pos+8) & ~7;
+	} elsif ($c eq "\n") {
+	    $mlen = $pos if ($pos > $mlen);
+	    $pos = 0;
+	} else {
+	    $pos++;
+	}
+    }
+
+    $mlen = $pos if ($pos > $mlen);
+    return $mlen;
+}
+
+$name = basename($0);
+
+@files = ();
+
+while (defined($a = shift(@ARGV))) {
+    if ($a =~ /^-/) {
+	if ($a eq '-width' || $a eq '-w') {
+	    $max_width = shift(@ARGV)+0;
+	} else {
+	    print STDERR "Usage: $name [-width #] files...\n";
+	    exit 1;
+	}
+    } else {
+	push(@files, $a);
+    }
+}
+
+foreach $f ( @files ) {
+    print STDERR "$name: $f\n";
+
+    if (! -f $f) {
+	print STDERR "$f: not a file\n";
+	next;
+    }
+
+    if (!open(FILE, '+<', $f)) {
+	print STDERR "$name: Cannot open file: $f: $!\n";
+	next;
+    }
+
+    binmode FILE;
+
+    # First, verify that it is not a binary file; consider any file
+    # with a zero byte to be a binary file.  Is there any better, or
+    # additional, heuristic that should be applied?
+    $is_binary = 0;
+
+    while (read(FILE, $data, 65536) > 0) {
+	if ($data =~ /\0/) {
+	    $is_binary = 1;
+	    last;
+	}
+    }
+
+    if ($is_binary) {
+	print STDERR "$name: $f: binary file\n";
+	next;
+    }
+
+    seek(FILE, 0, 0);
+
+    $in_bytes = 0;
+    $out_bytes = 0;
+    $lineno = 0;
+
+    @lines  = ();
+
+    $in_hunk = 0;
+    $err = 0;
+
+    while ( defined($line = <FILE>) ) {
+	$lineno++;
+	$in_bytes += length($line);
+
+	if (!$in_hunk) {
+	    if ($line =~
+		/^\@\@\s+\-([0-9]+),([0-9]+)\s+\+([0-9]+),([0-9]+)\s\@\@/) {
+		$minus_lines = $2;
+		$plus_lines = $4;
+		if ($minus_lines || $plus_lines) {
+		    $in_hunk = 1;
+		    @hunk_lines = ($line);
+		}
+	    } else {
+		push(@lines, $line);
+		$out_bytes += length($line);
+	    }
+	} else {
+	    # We're in a hunk
+
+	    if ($line =~ /^\+/) {
+		$plus_lines--;
+
+		$text = substr($line, 1);
+		$text =~ s/[ \t\r]*$//;		# Remove trailing spaces
+		$text = clean_space_tabs($text);
+
+		$l_width = strwidth($text);
+		if ($max_width && $l_width > $max_width) {
+		    print STDERR
+			"$f:$lineno: adds line exceeds $max_width ",
+			"characters ($l_width)\n";
+		}
+
+		push(@hunk_lines, '+'.$text);
+	    } elsif ($line =~ /^\-/) {
+		$minus_lines--;
+		push(@hunk_lines, $line);
+	    } elsif ($line =~ /^ /) {
+		$plus_lines--;
+		$minus_lines--;
+		push(@hunk_lines, $line);
+	    } else {
+		print STDERR "$name: $f: malformed patch\n";
+		$err = 1;
+		last;
+	    }
+
+	    if ($plus_lines < 0 || $minus_lines < 0) {
+		print STDERR "$name: $f: malformed patch\n";
+		$err = 1;
+		last;
+	    } elsif ($plus_lines == 0 && $minus_lines == 0) {
+		# End of a hunk.  Process this hunk.
+		my $i;
+		my $l;
+		my @h = ();
+		my $adj = 0;
+		my $done = 0;
+
+		for ($i = scalar(@hunk_lines)-1; $i > 0; $i--) {
+		    $l = $hunk_lines[$i];
+		    if (!$done && $l eq "+\n") {
+			$adj++; # Skip this line
+		    } elsif ($l =~ /^[ +]/) {
+			$done = 1;
+			unshift(@h, $l);
+		    } else {
+			unshift(@h, $l);
+		    }
+		}
+
+		$l = $hunk_lines[0];  # Hunk header
+		undef @hunk_lines;    # Free memory
+
+		if ($adj) {
+		    die unless
+			($l =~ /^\@\@\s+\-([0-9]+),([0-9]+)\s+\+([0-9]+),([0-9]+)\s\@\@(.*)$/);
+		    my $mstart = $1;
+		    my $mlin = $2;
+		    my $pstart = $3;
+		    my $plin = $4;
+		    my $tail = $5; # doesn't include the final newline
+
+		    $l = sprintf("@@ -%d,%d +%d,%d @@%s\n",
+				 $mstart, $mlin, $pstart, $plin-$adj,
+				 $tail);
+		}
+		unshift(@h, $l);
+
+		# Transfer to the output array
+		foreach $l (@h) {
+		    $out_bytes += length($l);
+		    push(@lines, $l);
+		}
+
+		$in_hunk = 0;
+	    }
+	}
+    }
+
+    if ($in_hunk) {
+	print STDERR "$name: $f: malformed patch\n";
+	$err = 1;
+    }
+
+    if (!$err) {
+	if ($in_bytes != $out_bytes) {
+	    # Only write to the file if changed
+	    seek(FILE, 0, 0);
+	    print FILE @lines;
+
+	    if ( !defined($where = tell(FILE)) ||
+		 !truncate(FILE, $where) ) {
+		die "$name: Failed to truncate modified file: $f: $!\n";
+	    }
+	}
+    }
+
+    close(FILE);
+}