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, 142 insertions, 0 deletions

diff --git a/include/uapi/fwctl/fwctl.h b/include/uapi/fwctl/fwctl.h
new file mode 100644
index 000000000..2d6d4049c
--- /dev/null
+++ b/include/uapi/fwctl/fwctl.h
@@ -0,0 +1,142 @@
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
+/* Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES.
+ */
+#ifndef _UAPI_FWCTL_H
+#define _UAPI_FWCTL_H
+
+#include <linux/types.h>
+#include <linux/ioctl.h>
+
+#define FWCTL_TYPE 0x9A
+
+/**
+ * DOC: General ioctl format
+ *
+ * The ioctl interface follows a general format to allow for extensibility. Each
+ * ioctl is passed a structure pointer as the argument providing the size of
+ * the structure in the first u32. The kernel checks that any structure space
+ * beyond what it understands is 0. This allows userspace to use the backward
+ * compatible portion while consistently using the newer, larger, structures.
+ *
+ * ioctls use a standard meaning for common errnos:
+ *
+ *  - ENOTTY: The IOCTL number itself is not supported at all
+ *  - E2BIG: The IOCTL number is supported, but the provided structure has
+ *    non-zero in a part the kernel does not understand.
+ *  - EOPNOTSUPP: The IOCTL number is supported, and the structure is
+ *    understood, however a known field has a value the kernel does not
+ *    understand or support.
+ *  - EINVAL: Everything about the IOCTL was understood, but a field is not
+ *    correct.
+ *  - ENOMEM: Out of memory.
+ *  - ENODEV: The underlying device has been hot-unplugged and the FD is
+ *            orphaned.
+ *
+ * As well as additional errnos, within specific ioctls.
+ */
+enum {
+	FWCTL_CMD_BASE = 0,
+	FWCTL_CMD_INFO = 0,
+	FWCTL_CMD_RPC = 1,
+};
+
+enum fwctl_device_type {
+	FWCTL_DEVICE_TYPE_ERROR = 0,
+	FWCTL_DEVICE_TYPE_MLX5 = 1,
+	FWCTL_DEVICE_TYPE_CXL = 2,
+	FWCTL_DEVICE_TYPE_BNXT = 3,
+	FWCTL_DEVICE_TYPE_PDS = 4,
+};
+
+/**
+ * struct fwctl_info - ioctl(FWCTL_INFO)
+ * @size: sizeof(struct fwctl_info)
+ * @flags: Must be 0
+ * @out_device_type: Returns the type of the device from enum fwctl_device_type
+ * @device_data_len: On input the length of the out_device_data memory. On
+ *	output the size of the kernel's device_data which may be larger or
+ *	smaller than the input. Maybe 0 on input.
+ * @out_device_data: Pointer to a memory of device_data_len bytes. Kernel will
+ *	fill the entire memory, zeroing as required.
+ *
+ * Returns basic information about this fwctl instance, particularly what driver
+ * is being used to define the device_data format.
+ */
+struct fwctl_info {
+	__u32 size;
+	__u32 flags;
+	__u32 out_device_type;
+	__u32 device_data_len;
+	__aligned_u64 out_device_data;
+};
+#define FWCTL_INFO _IO(FWCTL_TYPE, FWCTL_CMD_INFO)
+
+/**
+ * enum fwctl_rpc_scope - Scope of access for the RPC
+ *
+ * Refer to fwctl.rst for a more detailed discussion of these scopes.
+ */
+enum fwctl_rpc_scope {
+	/**
+	 * @FWCTL_RPC_CONFIGURATION: Device configuration access scope
+	 *
+	 * Read/write access to device configuration. When configuration
+	 * is written to the device it remains in a fully supported state.
+	 */
+	FWCTL_RPC_CONFIGURATION = 0,
+	/**
+	 * @FWCTL_RPC_DEBUG_READ_ONLY: Read only access to debug information
+	 *
+	 * Readable debug information. Debug information is compatible with
+	 * kernel lockdown, and does not disclose any sensitive information. For
+	 * instance exposing any encryption secrets from this information is
+	 * forbidden.
+	 */
+	FWCTL_RPC_DEBUG_READ_ONLY = 1,
+	/**
+	 * @FWCTL_RPC_DEBUG_WRITE: Writable access to lockdown compatible debug information
+	 *
+	 * Allows write access to data in the device which may leave a fully
+	 * supported state. This is intended to permit intensive and possibly
+	 * invasive debugging. This scope will taint the kernel.
+	 */
+	FWCTL_RPC_DEBUG_WRITE = 2,
+	/**
+	 * @FWCTL_RPC_DEBUG_WRITE_FULL: Write access to all debug information
+	 *
+	 * Allows read/write access to everything. Requires CAP_SYS_RAW_IO, so
+	 * it is not required to follow lockdown principals. If in doubt
+	 * debugging should be placed in this scope. This scope will taint the
+	 * kernel.
+	 */
+	FWCTL_RPC_DEBUG_WRITE_FULL = 3,
+};
+
+/**
+ * struct fwctl_rpc - ioctl(FWCTL_RPC)
+ * @size: sizeof(struct fwctl_rpc)
+ * @scope: One of enum fwctl_rpc_scope, required scope for the RPC
+ * @in_len: Length of the in memory
+ * @out_len: Length of the out memory
+ * @in: Request message in device specific format
+ * @out: Response message in device specific format
+ *
+ * Deliver a Remote Procedure Call to the device FW and return the response. The
+ * call's parameters and return are marshaled into linear buffers of memory. Any
+ * errno indicates that delivery of the RPC to the device failed. Return status
+ * originating in the device during a successful delivery must be encoded into
+ * out.
+ *
+ * The format of the buffers matches the out_device_type from FWCTL_INFO.
+ */
+struct fwctl_rpc {
+	__u32 size;
+	__u32 scope;
+	__u32 in_len;
+	__u32 out_len;
+	__aligned_u64 in;
+	__aligned_u64 out;
+};
+#define FWCTL_RPC _IO(FWCTL_TYPE, FWCTL_CMD_RPC)
+
+#endif