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, 101 insertions, 0 deletions

diff --git a/include/rdma/ib_cache.h b/include/rdma/ib_cache.h
new file mode 100644
index 000000000..eed46d966
--- /dev/null
+++ b/include/rdma/ib_cache.h
@@ -0,0 +1,101 @@
+/* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ * Copyright (c) 2005 Intel Corporation. All rights reserved.
+ * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
+ */
+
+#ifndef _IB_CACHE_H
+#define _IB_CACHE_H
+
+#include <rdma/ib_verbs.h>
+
+int rdma_query_gid(struct ib_device *device, u32 port_num, int index,
+		   union ib_gid *gid);
+void *rdma_read_gid_hw_context(const struct ib_gid_attr *attr);
+const struct ib_gid_attr *rdma_find_gid(struct ib_device *device,
+					const union ib_gid *gid,
+					enum ib_gid_type gid_type,
+					struct net_device *ndev);
+const struct ib_gid_attr *rdma_find_gid_by_port(struct ib_device *ib_dev,
+						const union ib_gid *gid,
+						enum ib_gid_type gid_type,
+						u32 port,
+						struct net_device *ndev);
+const struct ib_gid_attr *rdma_find_gid_by_filter(
+	struct ib_device *device, const union ib_gid *gid, u32 port_num,
+	bool (*filter)(const union ib_gid *gid, const struct ib_gid_attr *,
+		       void *),
+	void *context);
+
+int rdma_read_gid_l2_fields(const struct ib_gid_attr *attr,
+			    u16 *vlan_id, u8 *smac);
+struct net_device *rdma_read_gid_attr_ndev_rcu(const struct ib_gid_attr *attr);
+
+/**
+ * ib_get_cached_pkey - Returns a cached PKey table entry
+ * @device_handle: The device to query.
+ * @port_num: The port number of the device to query.
+ * @index: The index into the cached PKey table to query.
+ * @pkey: The PKey value found at the specified index.
+ *
+ * ib_get_cached_pkey() fetches the specified PKey table entry stored in
+ * the local software cache.
+ */
+int ib_get_cached_pkey(struct ib_device    *device_handle,
+		       u32                  port_num,
+		       int                  index,
+		       u16                 *pkey);
+
+/**
+ * ib_find_cached_pkey - Returns the PKey table index where a specified
+ *   PKey value occurs.
+ * @device: The device to query.
+ * @port_num: The port number of the device to search for the PKey.
+ * @pkey: The PKey value to search for.
+ * @index: The index into the cached PKey table where the PKey was found.
+ *
+ * ib_find_cached_pkey() searches the specified PKey table in
+ * the local software cache.
+ */
+int ib_find_cached_pkey(struct ib_device    *device,
+			u32                  port_num,
+			u16                  pkey,
+			u16                 *index);
+
+/**
+ * ib_get_cached_lmc - Returns a cached lmc table entry
+ * @device: The device to query.
+ * @port_num: The port number of the device to query.
+ * @lmc: The lmc value for the specified port for that device.
+ *
+ * ib_get_cached_lmc() fetches the specified lmc table entry stored in
+ * the local software cache.
+ */
+int ib_get_cached_lmc(struct ib_device *device,
+		      u32               port_num,
+		      u8                *lmc);
+
+/**
+ * ib_get_cached_port_state - Returns a cached port state table entry
+ * @device: The device to query.
+ * @port_num: The port number of the device to query.
+ * @port_active: port_state for the specified port for that device.
+ *
+ * ib_get_cached_port_state() fetches the specified port_state table entry stored in
+ * the local software cache.
+ */
+int ib_get_cached_port_state(struct ib_device *device,
+			     u32               port_num,
+			      enum ib_port_state *port_active);
+
+bool rdma_is_zero_gid(const union ib_gid *gid);
+const struct ib_gid_attr *rdma_get_gid_attr(struct ib_device *device,
+					    u32 port_num, int index);
+void rdma_put_gid_attr(const struct ib_gid_attr *attr);
+void rdma_hold_gid_attr(const struct ib_gid_attr *attr);
+ssize_t rdma_query_gid_table(struct ib_device *device,
+			     struct ib_uverbs_gid_entry *entries,
+			     size_t max_entries);
+
+#endif /* _IB_CACHE_H */