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, 161 insertions, 0 deletions

diff --git a/tools/perf/tests/task-exit.c b/tools/perf/tests/task-exit.c
new file mode 100644
index 000000000..95393edbf
--- /dev/null
+++ b/tools/perf/tests/task-exit.c
@@ -0,0 +1,161 @@
+// SPDX-License-Identifier: GPL-2.0
+#include "debug.h"
+#include "evlist.h"
+#include "evsel.h"
+#include "target.h"
+#include "thread_map.h"
+#include "tests.h"
+#include "util/mmap.h"
+
+#include <errno.h>
+#include <signal.h>
+#include <linux/string.h>
+#include <perf/cpumap.h>
+#include <perf/evlist.h>
+#include <perf/mmap.h>
+
+static int exited;
+static int nr_exit;
+
+static void sig_handler(int sig __maybe_unused)
+{
+	exited = 1;
+}
+
+/*
+ * evlist__prepare_workload will send a SIGUSR1 if the fork fails, since
+ * we asked by setting its exec_error to this handler.
+ */
+static void workload_exec_failed_signal(int signo __maybe_unused,
+					siginfo_t *info __maybe_unused,
+					void *ucontext __maybe_unused)
+{
+	exited	= 1;
+	nr_exit = -1;
+}
+
+/*
+ * This test will start a workload that does nothing then it checks
+ * if the number of exit event reported by the kernel is 1 or not
+ * in order to check the kernel returns correct number of event.
+ */
+static int test__task_exit(struct test_suite *test __maybe_unused, int subtest __maybe_unused)
+{
+	int err = -1;
+	union perf_event *event;
+	struct evsel *evsel;
+	struct evlist *evlist;
+	struct target target = {
+		.uses_mmap	= true,
+	};
+	const char *argv[] = { "true", NULL };
+	char sbuf[STRERR_BUFSIZE];
+	struct perf_cpu_map *cpus;
+	struct perf_thread_map *threads;
+	struct mmap *md;
+	int retry_count = 0;
+
+	signal(SIGCHLD, sig_handler);
+
+	evlist = evlist__new_dummy();
+	if (evlist == NULL) {
+		pr_debug("evlist__new_dummy\n");
+		return -1;
+	}
+
+	/*
+	 * Create maps of threads and cpus to monitor. In this case
+	 * we start with all threads and cpus (-1, -1) but then in
+	 * evlist__prepare_workload we'll fill in the only thread
+	 * we're monitoring, the one forked there.
+	 */
+	cpus = perf_cpu_map__new_any_cpu();
+	threads = thread_map__new_by_tid(-1);
+	if (!cpus || !threads) {
+		err = -ENOMEM;
+		pr_debug("Not enough memory to create thread/cpu maps\n");
+		goto out_put_evlist;
+	}
+
+	perf_evlist__set_maps(evlist__core(evlist), cpus, threads);
+
+	err = evlist__prepare_workload(evlist, &target, argv, false, workload_exec_failed_signal);
+	if (err < 0) {
+		pr_debug("Couldn't run the workload!\n");
+		goto out_put_evlist;
+	}
+
+	evsel = evlist__first(evlist);
+	evsel->core.attr.task = 1;
+#ifdef __s390x__
+	evsel->core.attr.sample_freq = 1000000;
+#else
+	evsel->core.attr.sample_freq = 1;
+#endif
+	evsel->core.attr.inherit = 0;
+	evsel->core.attr.watermark = 0;
+	evsel->core.attr.wakeup_events = 1;
+	evsel->core.attr.exclude_kernel = 1;
+
+	err = evlist__open(evlist);
+	if (err < 0) {
+		pr_debug("Couldn't open the evlist: %s\n",
+			 str_error_r(-err, sbuf, sizeof(sbuf)));
+		goto out_put_evlist;
+	}
+
+	if (evlist__do_mmap(evlist, 128) < 0) {
+		pr_debug("failed to mmap events: %d (%s)\n", errno,
+			 str_error_r(errno, sbuf, sizeof(sbuf)));
+		err = -1;
+		goto out_put_evlist;
+	}
+
+	evlist__start_workload(evlist);
+
+retry:
+	md = &evlist__mmap(evlist)[0];
+	if (perf_mmap__read_init(&md->core) < 0)
+		goto out_init;
+
+	while ((event = perf_mmap__read_event(&md->core)) != NULL) {
+		if (event->header.type == PERF_RECORD_EXIT)
+			nr_exit++;
+
+		perf_mmap__consume(&md->core);
+	}
+	perf_mmap__read_done(&md->core);
+
+out_init:
+	if (!exited || !nr_exit) {
+		evlist__poll(evlist, -1);
+
+		if (retry_count++ > 1000) {
+			pr_debug("Failed after retrying 1000 times\n");
+			err = -1;
+			goto out_put_evlist;
+		}
+
+		goto retry;
+	}
+
+	if (nr_exit != 1) {
+		pr_debug("received %d EXIT records\n", nr_exit);
+		err = -1;
+	}
+
+out_put_evlist:
+	perf_cpu_map__put(cpus);
+	perf_thread_map__put(threads);
+	evlist__put(evlist);
+	return err;
+}
+
+struct test_case tests__task_exit[] = {
+	TEST_CASE_EXCLUSIVE("Number of exit events of a simple workload", task_exit),
+	{	.name = NULL, }
+};
+struct test_suite suite__task_exit = {
+	.desc = "Number of exit events of a simple workload",
+	.test_cases = tests__task_exit,
+};