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, 179 insertions, 0 deletions

diff --git a/include/uapi/cxl/features.h b/include/uapi/cxl/features.h
new file mode 100644
index 000000000..490606d76
--- /dev/null
+++ b/include/uapi/cxl/features.h
@@ -0,0 +1,179 @@
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
+/*
+ * Copyright (c) 2024,2025, Intel Corporation
+ *
+ * These are definitions for the mailbox command interface of CXL subsystem.
+ */
+#ifndef _UAPI_CXL_FEATURES_H_
+#define _UAPI_CXL_FEATURES_H_
+
+#include <linux/types.h>
+
+typedef unsigned char __uapi_uuid_t[16];
+
+#ifdef __KERNEL__
+#include <linux/uuid.h>
+/*
+ * Note, __uapi_uuid_t is 1-byte aligned on modern compilers and 4-byte
+ * aligned on others. Ensure that __uapi_uuid_t in a struct is placed at
+ * a 4-byte aligned offset, or the structure is packed, to ensure
+ * consistent padding.
+ */
+static_assert(sizeof(__uapi_uuid_t) == sizeof(uuid_t));
+#define __uapi_uuid_t uuid_t
+#endif
+
+/*
+ * struct cxl_mbox_get_sup_feats_in - Get Supported Features input
+ *
+ * @count: bytes of Feature data to return in output
+ * @start_idx: index of first requested Supported Feature Entry, 0 based.
+ * @reserved: reserved field, must be 0s.
+ *
+ * Get Supported Features (0x500h) CXL r3.2 8.2.9.6.1 command.
+ * Input block for Get support Feature
+ */
+struct cxl_mbox_get_sup_feats_in {
+	__le32 count;
+	__le16 start_idx;
+	__u8 reserved[2];
+} __attribute__ ((__packed__));
+
+/* CXL spec r3.2 Table 8-87 command effects */
+#define CXL_CMD_CONFIG_CHANGE_COLD_RESET	BIT(0)
+#define CXL_CMD_CONFIG_CHANGE_IMMEDIATE		BIT(1)
+#define CXL_CMD_DATA_CHANGE_IMMEDIATE		BIT(2)
+#define CXL_CMD_POLICY_CHANGE_IMMEDIATE		BIT(3)
+#define CXL_CMD_LOG_CHANGE_IMMEDIATE		BIT(4)
+#define CXL_CMD_SECURITY_STATE_CHANGE		BIT(5)
+#define CXL_CMD_BACKGROUND			BIT(6)
+#define CXL_CMD_BGCMD_ABORT_SUPPORTED		BIT(7)
+#define CXL_CMD_EFFECTS_VALID			BIT(9)
+#define CXL_CMD_CONFIG_CHANGE_CONV_RESET	BIT(10)
+#define CXL_CMD_CONFIG_CHANGE_CXL_RESET		BIT(11)
+#define CXL_CMD_EFFECTS_RESERVED		GENMASK(15, 12)
+
+/*
+ * struct cxl_feat_entry - Supported Feature Entry
+ * @uuid: UUID of the Feature
+ * @id: id to identify the feature. 0 based
+ * @get_feat_size: max bytes required for Get Feature command for this Feature
+ * @set_feat_size: max bytes required for Set Feature command for this Feature
+ * @flags: attribute flags
+ * @get_feat_ver: Get Feature version
+ * @set_feat_ver: Set Feature version
+ * @effects: Set Feature command effects
+ * @reserved: reserved, must be 0
+ *
+ * CXL spec r3.2 Table 8-109
+ * Get Supported Features Supported Feature Entry
+ */
+struct cxl_feat_entry {
+	__uapi_uuid_t uuid;
+	__le16 id;
+	__le16 get_feat_size;
+	__le16 set_feat_size;
+	__le32 flags;
+	__u8 get_feat_ver;
+	__u8 set_feat_ver;
+	__le16 effects;
+	__u8 reserved[18];
+} __attribute__ ((__packed__));
+
+/* @flags field for 'struct cxl_feat_entry' */
+#define CXL_FEATURE_F_CHANGEABLE		BIT(0)
+#define CXL_FEATURE_F_PERSIST_FW_UPDATE		BIT(4)
+#define CXL_FEATURE_F_DEFAULT_SEL		BIT(5)
+#define CXL_FEATURE_F_SAVED_SEL			BIT(6)
+
+/*
+ * struct cxl_mbox_get_sup_feats_out - Get Supported Features output
+ * @num_entries: number of Supported Feature Entries returned
+ * @supported_feats: number of supported Features
+ * @reserved: reserved, must be 0s.
+ * @ents: Supported Feature Entries array
+ *
+ * CXL spec r3.2 Table 8-108
+ * Get supported Features Output Payload
+ */
+struct cxl_mbox_get_sup_feats_out {
+	__struct_group(cxl_mbox_get_sup_feats_out_hdr, hdr, /* no attrs */,
+		__le16 num_entries;
+		__le16 supported_feats;
+		__u8 reserved[4];
+	);
+	struct cxl_feat_entry ents[] __counted_by_le(num_entries);
+} __attribute__ ((__packed__));
+
+/*
+ * Get Feature CXL spec r3.2 Spec 8.2.9.6.2
+ */
+
+/*
+ * struct cxl_mbox_get_feat_in - Get Feature input
+ * @uuid: UUID for Feature
+ * @offset: offset of the first byte in Feature data for output payload
+ * @count: count in bytes of Feature data returned
+ * @selection: 0 current value, 1 default value, 2 saved value
+ *
+ * CXL spec r3.2 section 8.2.9.6.2 Table 8-99
+ */
+struct cxl_mbox_get_feat_in {
+	__uapi_uuid_t uuid;
+	__le16 offset;
+	__le16 count;
+	__u8 selection;
+} __attribute__ ((__packed__));
+
+/*
+ * enum cxl_get_feat_selection - selection field of Get Feature input
+ */
+enum cxl_get_feat_selection {
+	CXL_GET_FEAT_SEL_CURRENT_VALUE,
+	CXL_GET_FEAT_SEL_DEFAULT_VALUE,
+	CXL_GET_FEAT_SEL_SAVED_VALUE,
+	CXL_GET_FEAT_SEL_MAX
+};
+
+/*
+ * Set Feature CXL spec r3.2  8.2.9.6.3
+ */
+
+/*
+ * struct cxl_mbox_set_feat_in - Set Features input
+ * @uuid: UUID for Feature
+ * @flags: set feature flags
+ * @offset: byte offset of Feature data to update
+ * @version: Feature version of the data in Feature Data
+ * @rsvd: reserved, must be 0s.
+ * @feat_data: raw byte stream of Features data to update
+ *
+ * CXL spec r3.2 section 8.2.9.6.3 Table 8-101
+ */
+struct cxl_mbox_set_feat_in {
+	__struct_group(cxl_mbox_set_feat_hdr, hdr, /* no attrs */,
+		__uapi_uuid_t uuid;
+		__le32 flags;
+		__le16 offset;
+		__u8 version;
+		__u8 rsvd[9];
+	);
+	__u8 feat_data[];
+}  __packed;
+
+/*
+ * enum cxl_set_feat_flag_data_transfer - Set Feature flags field
+ */
+enum cxl_set_feat_flag_data_transfer {
+	CXL_SET_FEAT_FLAG_FULL_DATA_TRANSFER = 0,
+	CXL_SET_FEAT_FLAG_INITIATE_DATA_TRANSFER,
+	CXL_SET_FEAT_FLAG_CONTINUE_DATA_TRANSFER,
+	CXL_SET_FEAT_FLAG_FINISH_DATA_TRANSFER,
+	CXL_SET_FEAT_FLAG_ABORT_DATA_TRANSFER,
+	CXL_SET_FEAT_FLAG_DATA_TRANSFER_MAX
+};
+
+#define CXL_SET_FEAT_FLAG_DATA_TRANSFER_MASK		GENMASK(2, 0)
+#define CXL_SET_FEAT_FLAG_DATA_SAVED_ACROSS_RESET	BIT(3)
+
+#endif