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, 291 insertions, 0 deletions

diff --git a/drivers/md/dm-exception-store.c b/drivers/md/dm-exception-store.c
new file mode 100644
index 000000000..58cbc5e54
--- /dev/null
+++ b/drivers/md/dm-exception-store.c
@@ -0,0 +1,291 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
+ * Copyright (C) 2006-2008 Red Hat GmbH
+ *
+ * This file is released under the GPL.
+ */
+
+#include "dm-exception-store.h"
+
+#include <linux/ctype.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include <linux/vmalloc.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+
+#define DM_MSG_PREFIX "snapshot exception stores"
+
+static LIST_HEAD(_exception_store_types);
+static DEFINE_SPINLOCK(_lock);
+
+static struct dm_exception_store_type *__find_exception_store_type(const char *name)
+{
+	struct dm_exception_store_type *type;
+
+	list_for_each_entry(type, &_exception_store_types, list)
+		if (!strcmp(name, type->name))
+			return type;
+
+	return NULL;
+}
+
+static struct dm_exception_store_type *_get_exception_store_type(const char *name)
+{
+	struct dm_exception_store_type *type;
+
+	spin_lock(&_lock);
+
+	type = __find_exception_store_type(name);
+
+	if (type && !try_module_get(type->module))
+		type = NULL;
+
+	spin_unlock(&_lock);
+
+	return type;
+}
+
+/*
+ * get_type
+ * @type_name
+ *
+ * Attempt to retrieve the dm_exception_store_type by name.  If not already
+ * available, attempt to load the appropriate module.
+ *
+ * Exstore modules are named "dm-exstore-" followed by the 'type_name'.
+ * Modules may contain multiple types.
+ * This function will first try the module "dm-exstore-<type_name>",
+ * then truncate 'type_name' on the last '-' and try again.
+ *
+ * For example, if type_name was "clustered-shared", it would search
+ * 'dm-exstore-clustered-shared' then 'dm-exstore-clustered'.
+ *
+ * 'dm-exception-store-<type_name>' is too long of a name in my
+ * opinion, which is why I've chosen to have the files
+ * containing exception store implementations be 'dm-exstore-<type_name>'.
+ * If you want your module to be autoloaded, you will follow this
+ * naming convention.
+ *
+ * Returns: dm_exception_store_type* on success, NULL on failure
+ */
+static struct dm_exception_store_type *get_type(const char *type_name)
+{
+	char *p, *type_name_dup;
+	struct dm_exception_store_type *type;
+
+	type = _get_exception_store_type(type_name);
+	if (type)
+		return type;
+
+	type_name_dup = kstrdup(type_name, GFP_KERNEL);
+	if (!type_name_dup) {
+		DMERR("No memory left to attempt load for \"%s\"", type_name);
+		return NULL;
+	}
+
+	while (request_module("dm-exstore-%s", type_name_dup) ||
+	       !(type = _get_exception_store_type(type_name))) {
+		p = strrchr(type_name_dup, '-');
+		if (!p)
+			break;
+		p[0] = '\0';
+	}
+
+	if (!type)
+		DMWARN("Module for exstore type \"%s\" not found.", type_name);
+
+	kfree(type_name_dup);
+
+	return type;
+}
+
+static void put_type(struct dm_exception_store_type *type)
+{
+	spin_lock(&_lock);
+	module_put(type->module);
+	spin_unlock(&_lock);
+}
+
+int dm_exception_store_type_register(struct dm_exception_store_type *type)
+{
+	int r = 0;
+
+	spin_lock(&_lock);
+	if (!__find_exception_store_type(type->name))
+		list_add(&type->list, &_exception_store_types);
+	else
+		r = -EBUSY;
+	spin_unlock(&_lock);
+
+	return r;
+}
+EXPORT_SYMBOL(dm_exception_store_type_register);
+
+int dm_exception_store_type_unregister(struct dm_exception_store_type *type)
+{
+	spin_lock(&_lock);
+
+	if (!__find_exception_store_type(type->name)) {
+		spin_unlock(&_lock);
+		return -EINVAL;
+	}
+
+	list_del(&type->list);
+
+	spin_unlock(&_lock);
+
+	return 0;
+}
+EXPORT_SYMBOL(dm_exception_store_type_unregister);
+
+static int set_chunk_size(struct dm_exception_store *store,
+			  const char *chunk_size_arg, char **error)
+{
+	unsigned int chunk_size;
+
+	if (kstrtouint(chunk_size_arg, 10, &chunk_size)) {
+		*error = "Invalid chunk size";
+		return -EINVAL;
+	}
+
+	if (!chunk_size) {
+		store->chunk_size = store->chunk_mask = store->chunk_shift = 0;
+		return 0;
+	}
+
+	return dm_exception_store_set_chunk_size(store, chunk_size, error);
+}
+
+int dm_exception_store_set_chunk_size(struct dm_exception_store *store,
+				      unsigned int chunk_size,
+				      char **error)
+{
+	/* Check chunk_size is a power of 2 */
+	if (!is_power_of_2(chunk_size)) {
+		*error = "Chunk size is not a power of 2";
+		return -EINVAL;
+	}
+
+	/* Validate the chunk size against the device block size */
+	if (chunk_size %
+	    (bdev_logical_block_size(dm_snap_cow(store->snap)->bdev) >> 9) ||
+	    chunk_size %
+	    (bdev_logical_block_size(dm_snap_origin(store->snap)->bdev) >> 9)) {
+		*error = "Chunk size is not a multiple of device blocksize";
+		return -EINVAL;
+	}
+
+	if (chunk_size > INT_MAX >> SECTOR_SHIFT) {
+		*error = "Chunk size is too high";
+		return -EINVAL;
+	}
+
+	store->chunk_size = chunk_size;
+	store->chunk_mask = chunk_size - 1;
+	store->chunk_shift = __ffs(chunk_size);
+
+	return 0;
+}
+
+int dm_exception_store_create(struct dm_target *ti, int argc, char **argv,
+			      struct dm_snapshot *snap,
+			      unsigned int *args_used,
+			      struct dm_exception_store **store)
+{
+	int r = 0;
+	struct dm_exception_store_type *type = NULL;
+	struct dm_exception_store *tmp_store;
+	char persistent;
+
+	if (argc < 2) {
+		ti->error = "Insufficient exception store arguments";
+		return -EINVAL;
+	}
+
+	tmp_store = kzalloc_obj(*tmp_store);
+	if (!tmp_store) {
+		ti->error = "Exception store allocation failed";
+		return -ENOMEM;
+	}
+
+	persistent = toupper(*argv[0]);
+	if (persistent == 'P')
+		type = get_type("P");
+	else if (persistent == 'N')
+		type = get_type("N");
+	else {
+		ti->error = "Exception store type is not P or N";
+		r = -EINVAL;
+		goto bad_type;
+	}
+
+	if (!type) {
+		ti->error = "Exception store type not recognised";
+		r = -EINVAL;
+		goto bad_type;
+	}
+
+	tmp_store->type = type;
+	tmp_store->snap = snap;
+
+	r = set_chunk_size(tmp_store, argv[1], &ti->error);
+	if (r)
+		goto bad;
+
+	r = type->ctr(tmp_store, (strlen(argv[0]) > 1 ? &argv[0][1] : NULL));
+	if (r) {
+		ti->error = "Exception store type constructor failed";
+		goto bad;
+	}
+
+	*args_used = 2;
+	*store = tmp_store;
+	return 0;
+
+bad:
+	put_type(type);
+bad_type:
+	kfree(tmp_store);
+	return r;
+}
+EXPORT_SYMBOL(dm_exception_store_create);
+
+void dm_exception_store_destroy(struct dm_exception_store *store)
+{
+	store->type->dtr(store);
+	put_type(store->type);
+	kfree(store);
+}
+EXPORT_SYMBOL(dm_exception_store_destroy);
+
+int dm_exception_store_init(void)
+{
+	int r;
+
+	r = dm_transient_snapshot_init();
+	if (r) {
+		DMERR("Unable to register transient exception store type.");
+		goto transient_fail;
+	}
+
+	r = dm_persistent_snapshot_init();
+	if (r) {
+		DMERR("Unable to register persistent exception store type");
+		goto persistent_fail;
+	}
+
+	return 0;
+
+persistent_fail:
+	dm_transient_snapshot_exit();
+transient_fail:
+	return r;
+}
+
+void dm_exception_store_exit(void)
+{
+	dm_persistent_snapshot_exit();
+	dm_transient_snapshot_exit();
+}