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, 170 insertions, 0 deletions

diff --git a/tools/perf/util/strbuf.c b/tools/perf/util/strbuf.c
new file mode 100644
index 000000000..a64a37628
--- /dev/null
+++ b/tools/perf/util/strbuf.c
@@ -0,0 +1,170 @@
+// SPDX-License-Identifier: GPL-2.0
+#include "cache.h"
+#include "debug.h"
+#include "strbuf.h"
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include <linux/zalloc.h>
+#include <errno.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+
+/*
+ * Used as the default ->buf value, so that people can always assume
+ * buf is non NULL and ->buf is NUL terminated even for a freshly
+ * initialized strbuf.
+ */
+char strbuf_slopbuf[1];
+
+int strbuf_init(struct strbuf *sb, ssize_t hint)
+{
+	sb->alloc = sb->len = 0;
+	sb->buf = strbuf_slopbuf;
+	if (hint)
+		return strbuf_grow(sb, hint);
+	return 0;
+}
+
+void strbuf_release(struct strbuf *sb)
+{
+	if (sb->alloc) {
+		zfree(&sb->buf);
+		strbuf_init(sb, 0);
+	}
+}
+
+char *strbuf_detach(struct strbuf *sb, size_t *sz)
+{
+	char *res = sb->alloc ? sb->buf : NULL;
+	if (sz)
+		*sz = sb->len;
+	strbuf_init(sb, 0);
+	return res;
+}
+
+int strbuf_grow(struct strbuf *sb, size_t extra)
+{
+	char *buf;
+	size_t nr = sb->len + extra + 1;
+
+	if (nr < sb->alloc)
+		return 0;
+
+	if (nr <= sb->len)
+		return -E2BIG;
+
+	if (alloc_nr(sb->alloc) > nr)
+		nr = alloc_nr(sb->alloc);
+
+	/*
+	 * Note that sb->buf == strbuf_slopbuf if sb->alloc == 0, and it is
+	 * a static variable. Thus we have to avoid passing it to realloc.
+	 */
+	buf = realloc(sb->alloc ? sb->buf : NULL, nr * sizeof(*buf));
+	if (!buf)
+		return -ENOMEM;
+
+	sb->buf = buf;
+	sb->alloc = nr;
+	return 0;
+}
+
+int strbuf_addch(struct strbuf *sb, int c)
+{
+	int ret = strbuf_grow(sb, 1);
+	if (ret)
+		return ret;
+
+	sb->buf[sb->len++] = c;
+	sb->buf[sb->len] = '\0';
+	return 0;
+}
+
+int strbuf_add(struct strbuf *sb, const void *data, size_t len)
+{
+	int ret = strbuf_grow(sb, len);
+	if (ret)
+		return ret;
+
+	memcpy(sb->buf + sb->len, data, len);
+	return strbuf_setlen(sb, sb->len + len);
+}
+
+static int strbuf_addv(struct strbuf *sb, const char *fmt, va_list ap)
+{
+	int len, ret;
+	va_list ap_saved;
+
+	if (!strbuf_avail(sb)) {
+		ret = strbuf_grow(sb, 64);
+		if (ret)
+			return ret;
+	}
+
+	va_copy(ap_saved, ap);
+	len = vsnprintf(sb->buf + sb->len, sb->alloc - sb->len, fmt, ap);
+	if (len < 0) {
+		va_end(ap_saved);
+		return len;
+	}
+	if (len > strbuf_avail(sb)) {
+		ret = strbuf_grow(sb, len);
+		if (ret) {
+			va_end(ap_saved);
+			return ret;
+		}
+		len = vsnprintf(sb->buf + sb->len, sb->alloc - sb->len, fmt, ap_saved);
+		if (len > strbuf_avail(sb)) {
+			pr_debug("this should not happen, your vsnprintf is broken");
+			va_end(ap_saved);
+			return -EINVAL;
+		}
+	}
+	va_end(ap_saved);
+	return strbuf_setlen(sb, sb->len + len);
+}
+
+int strbuf_addf(struct strbuf *sb, const char *fmt, ...)
+{
+	va_list ap;
+	int ret;
+
+	va_start(ap, fmt);
+	ret = strbuf_addv(sb, fmt, ap);
+	va_end(ap);
+	return ret;
+}
+
+ssize_t strbuf_read(struct strbuf *sb, int fd, ssize_t hint)
+{
+	size_t oldlen = sb->len;
+	size_t oldalloc = sb->alloc;
+	int ret;
+
+	ret = strbuf_grow(sb, hint ? hint : 8192);
+	if (ret)
+		return ret;
+
+	for (;;) {
+		ssize_t cnt;
+
+		cnt = read(fd, sb->buf + sb->len, sb->alloc - sb->len - 1);
+		if (cnt < 0) {
+			if (oldalloc == 0)
+				strbuf_release(sb);
+			else
+				strbuf_setlen(sb, oldlen);
+			return cnt;
+		}
+		if (!cnt)
+			break;
+		sb->len += cnt;
+		ret = strbuf_grow(sb, 8192);
+		if (ret)
+			return ret;
+	}
+
+	sb->buf[sb->len] = '\0';
+	return sb->len - oldlen;
+}