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/tests/backward-ring-buffer.c b/tools/perf/tests/backward-ring-buffer.c
new file mode 100644
index 000000000..2735cc26d
--- /dev/null
+++ b/tools/perf/tests/backward-ring-buffer.c
@@ -0,0 +1,170 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Test backward bit in event attribute, read ring buffer from end to
+ * beginning
+ */
+
+#include <evlist.h>
+#include <sys/prctl.h>
+#include "record.h"
+#include "tests.h"
+#include "debug.h"
+#include "parse-events.h"
+#include "util/mmap.h"
+#include <errno.h>
+#include <linux/string.h>
+#include <perf/mmap.h>
+
+#define NR_ITERS 111
+
+static void testcase(void)
+{
+	int i;
+
+	for (i = 0; i < NR_ITERS; i++) {
+		char proc_name[15];
+
+		snprintf(proc_name, sizeof(proc_name), "p:%d\n", i);
+		prctl(PR_SET_NAME, proc_name);
+	}
+}
+
+static int count_samples(struct evlist *evlist, int *sample_count,
+			 int *comm_count)
+{
+	int i;
+
+	for (i = 0; i < evlist__core(evlist)->nr_mmaps; i++) {
+		struct mmap *map = &evlist__overwrite_mmap(evlist)[i];
+		union perf_event *event;
+
+		perf_mmap__read_init(&map->core);
+		while ((event = perf_mmap__read_event(&map->core)) != NULL) {
+			const u32 type = event->header.type;
+
+			switch (type) {
+			case PERF_RECORD_SAMPLE:
+				(*sample_count)++;
+				break;
+			case PERF_RECORD_COMM:
+				(*comm_count)++;
+				break;
+			default:
+				pr_err("Unexpected record of type %d\n", type);
+				return TEST_FAIL;
+			}
+		}
+		perf_mmap__read_done(&map->core);
+	}
+	return TEST_OK;
+}
+
+static int do_test(struct evlist *evlist, int mmap_pages,
+		   int *sample_count, int *comm_count)
+{
+	int err;
+	char sbuf[STRERR_BUFSIZE];
+
+	err = evlist__do_mmap(evlist, mmap_pages);
+	if (err < 0) {
+		pr_debug("evlist__mmap: %s\n",
+			 str_error_r(errno, sbuf, sizeof(sbuf)));
+		return TEST_FAIL;
+	}
+
+	evlist__enable(evlist);
+	testcase();
+	evlist__disable(evlist);
+
+	err = count_samples(evlist, sample_count, comm_count);
+	evlist__do_munmap(evlist);
+	return err;
+}
+
+
+static int test__backward_ring_buffer(struct test_suite *test __maybe_unused, int subtest __maybe_unused)
+{
+	int ret = TEST_SKIP, err, sample_count = 0, comm_count = 0;
+	char pid[16], sbuf[STRERR_BUFSIZE];
+	struct evlist *evlist;
+	struct evsel *evsel __maybe_unused;
+	struct parse_events_error parse_error;
+	struct record_opts opts = {
+		.target = {
+			.uses_mmap = true,
+		},
+		.freq	      = 0,
+		.mmap_pages   = 256,
+		.default_interval = 1,
+	};
+
+	snprintf(pid, sizeof(pid), "%d", getpid());
+	pid[sizeof(pid) - 1] = '\0';
+	opts.target.tid = opts.target.pid = pid;
+
+	evlist = evlist__new();
+	if (!evlist) {
+		pr_debug("Not enough memory to create evlist\n");
+		return TEST_FAIL;
+	}
+
+	err = evlist__create_maps(evlist, &opts.target);
+	if (err < 0) {
+		pr_debug("Not enough memory to create thread/cpu maps\n");
+		goto out_put_evlist;
+	}
+
+	parse_events_error__init(&parse_error);
+	/*
+	 * Set backward bit, ring buffer should be writing from end. Record
+	 * it in aux evlist
+	 */
+	err = parse_events(evlist, "syscalls:sys_enter_prctl/overwrite/", &parse_error);
+	parse_events_error__exit(&parse_error);
+	if (err) {
+		pr_debug("Failed to parse tracepoint event, try use root\n");
+		ret = TEST_SKIP;
+		goto out_put_evlist;
+	}
+
+	evlist__config(evlist, &opts, NULL);
+
+	err = evlist__open(evlist);
+	if (err < 0) {
+		pr_debug("perf_evlist__open: %s\n",
+			 str_error_r(errno, sbuf, sizeof(sbuf)));
+		goto out_put_evlist;
+	}
+
+	ret = TEST_FAIL;
+	err = do_test(evlist, opts.mmap_pages, &sample_count,
+		      &comm_count);
+	if (err != TEST_OK)
+		goto out_put_evlist;
+
+	if ((sample_count != NR_ITERS) || (comm_count != NR_ITERS)) {
+		pr_err("Unexpected counter: sample_count=%d, comm_count=%d\n",
+		       sample_count, comm_count);
+		goto out_put_evlist;
+	}
+
+	evlist__close(evlist);
+
+	err = evlist__open(evlist);
+	if (err < 0) {
+		pr_debug("perf_evlist__open: %s\n",
+			 str_error_r(errno, sbuf, sizeof(sbuf)));
+		goto out_put_evlist;
+	}
+
+	err = do_test(evlist, 1, &sample_count, &comm_count);
+	if (err != TEST_OK)
+		goto out_put_evlist;
+
+	ret = TEST_OK;
+out_put_evlist:
+	evlist__put(evlist);
+	return ret;
+}
+
+DEFINE_SUITE("Read backward ring buffer", backward_ring_buffer);