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

3 files changed, 198 insertions, 0 deletions

diff --git a/samples/watch_queue/.gitignore b/samples/watch_queue/.gitignore
new file mode 100644
index 000000000..823b351d3
--- /dev/null
+++ b/samples/watch_queue/.gitignore
@@ -0,0 +1,2 @@
+# SPDX-License-Identifier: GPL-2.0-only
+/watch_test
diff --git a/samples/watch_queue/Makefile b/samples/watch_queue/Makefile
new file mode 100644
index 000000000..c0db3a6bc
--- /dev/null
+++ b/samples/watch_queue/Makefile
@@ -0,0 +1,4 @@
+# SPDX-License-Identifier: GPL-2.0-only
+userprogs-always-y += watch_test
+
+userccflags += -I usr/include
diff --git a/samples/watch_queue/watch_test.c b/samples/watch_queue/watch_test.c
new file mode 100644
index 000000000..24cf7d7a1
--- /dev/null
+++ b/samples/watch_queue/watch_test.c
@@ -0,0 +1,192 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Use watch_queue API to watch for notifications.
+ *
+ * Copyright (C) 2020 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#define _GNU_SOURCE
+#include <stdbool.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <signal.h>
+#include <unistd.h>
+#include <errno.h>
+#include <sys/ioctl.h>
+#include <limits.h>
+
+// Work around glibc header silliness
+#undef AT_RENAME_NOREPLACE
+#undef AT_RENAME_EXCHANGE
+#undef AT_RENAME_WHITEOUT
+
+#include <linux/watch_queue.h>
+#include <linux/unistd.h>
+#include <linux/keyctl.h>
+
+#ifndef KEYCTL_WATCH_KEY
+#define KEYCTL_WATCH_KEY -1
+#endif
+#ifndef __NR_keyctl
+#define __NR_keyctl -1
+#endif
+
+#define BUF_SIZE 256
+
+static long keyctl_watch_key(int key, int watch_fd, int watch_id)
+{
+	return syscall(__NR_keyctl, KEYCTL_WATCH_KEY, key, watch_fd, watch_id);
+}
+
+static const char *key_subtypes[256] = {
+	[NOTIFY_KEY_INSTANTIATED]	= "instantiated",
+	[NOTIFY_KEY_UPDATED]		= "updated",
+	[NOTIFY_KEY_LINKED]		= "linked",
+	[NOTIFY_KEY_UNLINKED]		= "unlinked",
+	[NOTIFY_KEY_CLEARED]		= "cleared",
+	[NOTIFY_KEY_REVOKED]		= "revoked",
+	[NOTIFY_KEY_INVALIDATED]	= "invalidated",
+	[NOTIFY_KEY_SETATTR]		= "setattr",
+};
+
+static void saw_key_change(struct watch_notification *n, size_t len)
+{
+	struct key_notification *k = (struct key_notification *)n;
+
+	if (len != sizeof(struct key_notification)) {
+		fprintf(stderr, "Incorrect key message length\n");
+		return;
+	}
+
+	printf("KEY %08x change=%u[%s] aux=%u\n",
+	       k->key_id, n->subtype, key_subtypes[n->subtype], k->aux);
+}
+
+/*
+ * Consume and display events.
+ */
+static void consumer(int fd)
+{
+	unsigned char buffer[433], *p, *end;
+	union {
+		struct watch_notification n;
+		unsigned char buf1[128];
+	} n;
+	ssize_t buf_len;
+
+	for (;;) {
+		buf_len = read(fd, buffer, sizeof(buffer));
+		if (buf_len == -1) {
+			perror("read");
+			exit(1);
+		}
+
+		if (buf_len == 0) {
+			printf("-- END --\n");
+			return;
+		}
+
+		if (buf_len > sizeof(buffer)) {
+			fprintf(stderr, "Read buffer overrun: %zd\n", buf_len);
+			return;
+		}
+
+		printf("read() = %zd\n", buf_len);
+
+		p = buffer;
+		end = buffer + buf_len;
+		while (p < end) {
+			size_t largest, len;
+
+			largest = end - p;
+			if (largest > 128)
+				largest = 128;
+			if (largest < sizeof(struct watch_notification)) {
+				fprintf(stderr, "Short message header: %zu\n", largest);
+				return;
+			}
+			memcpy(&n, p, largest);
+
+			printf("NOTIFY[%03zx]: ty=%06x sy=%02x i=%08x\n",
+			       p - buffer, n.n.type, n.n.subtype, n.n.info);
+
+			len = n.n.info & WATCH_INFO_LENGTH;
+			if (len < sizeof(n.n) || len > largest) {
+				fprintf(stderr, "Bad message length: %zu/%zu\n", len, largest);
+				exit(1);
+			}
+
+			switch (n.n.type) {
+			case WATCH_TYPE_META:
+				switch (n.n.subtype) {
+				case WATCH_META_REMOVAL_NOTIFICATION:
+					printf("REMOVAL of watchpoint %08x\n",
+					       (n.n.info & WATCH_INFO_ID) >>
+					       WATCH_INFO_ID__SHIFT);
+					break;
+				case WATCH_META_LOSS_NOTIFICATION:
+					printf("-- LOSS --\n");
+					break;
+				default:
+					printf("other meta record\n");
+					break;
+				}
+				break;
+			case WATCH_TYPE_KEY_NOTIFY:
+				saw_key_change(&n.n, len);
+				break;
+			default:
+				printf("other type\n");
+				break;
+			}
+
+			p += len;
+		}
+	}
+}
+
+static struct watch_notification_filter filter = {
+	.nr_filters	= 1,
+	.filters = {
+		[0]	= {
+			.type			= WATCH_TYPE_KEY_NOTIFY,
+			.subtype_filter[0]	= UINT_MAX,
+		},
+	},
+};
+
+int main(int argc, char **argv)
+{
+	int pipefd[2], fd;
+
+	if (pipe2(pipefd, O_NOTIFICATION_PIPE) == -1) {
+		perror("pipe2");
+		exit(1);
+	}
+	fd = pipefd[0];
+
+	if (ioctl(fd, IOC_WATCH_QUEUE_SET_SIZE, BUF_SIZE) == -1) {
+		perror("watch_queue(size)");
+		exit(1);
+	}
+
+	if (ioctl(fd, IOC_WATCH_QUEUE_SET_FILTER, &filter) == -1) {
+		perror("watch_queue(filter)");
+		exit(1);
+	}
+
+	if (keyctl_watch_key(KEY_SPEC_SESSION_KEYRING, fd, 0x01) == -1) {
+		perror("keyctl");
+		exit(1);
+	}
+
+	if (keyctl_watch_key(KEY_SPEC_USER_KEYRING, fd, 0x02) == -1) {
+		perror("keyctl");
+		exit(1);
+	}
+
+	consumer(fd);
+	exit(0);
+}