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

2 files changed, 151 insertions, 0 deletions

diff --git a/Documentation/driver-api/issei/index.rst b/Documentation/driver-api/issei/index.rst
new file mode 100644
index 000000000..604267463
--- /dev/null
+++ b/Documentation/driver-api/issei/index.rst
@@ -0,0 +1,16 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+.. include:: <isonum.txt>
+
+=========================================================
+The Intel Silicon Security Engine Interface (Intel SSEI)
+=========================================================
+
+**Copyright** |copy| 2026 Intel Corporation
+
+
+.. toctree::
+   :caption: Table of Contents
+   :maxdepth: 3
+
+   issei
diff --git a/Documentation/driver-api/issei/issei.rst b/Documentation/driver-api/issei/issei.rst
new file mode 100644
index 000000000..a5e99e92e
--- /dev/null
+++ b/Documentation/driver-api/issei/issei.rst
@@ -0,0 +1,135 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+Introduction
+============
+
+The Intel Silicon Security Engine (Intel SSE) is an isolated and
+protected computing resource (Co-processor) residing inside
+Intel client chipsets released in 2024 (Lunar Lake) or later.
+The Intel SSE provide security support and platform boot orchestration.
+The actual feature set depends on the Intel chipset SKU.
+
+The Intel Silicon Security Engine Interface (Intel SSEI)
+is the interface between the Host and Intel SSE.
+This interface is exposed to the host as one or more PCI devices.
+The Intel SSEI Driver is in charge of the communication channel between
+a host application and the Intel SSE features.
+
+Each Intel SSE feature, or Intel SSE Client is addressed by a unique UUID and
+each client has its own protocol. The protocol is message-based with a
+header and payload up to maximal number of bytes advertised by the client,
+upon connection.
+
+Intel SSEI Driver
+=================
+
+The driver exposes a character device with device nodes /dev/isseiX.
+
+An application maintains communication with an Intel SSE feature while
+/dev/isseiX is open. The binding to a specific feature is performed by calling
+:c:macro:`IOCTL_ISSEI_CONNECT_CLIENT`, which passes the desired UUID.
+The number of instances of an Intel SSE feature that can be opened
+at the same time is limited to single instance.
+
+The driver is transparent to data that are passed between firmware feature
+and host application.
+
+Because some of the Intel SSE features can change the system
+configuration, the driver by default allows only a privileged
+user to access it.
+
+The connection termination is performed by calling
+:c:macro:`IOCTL_ISSEI_DISCONNECT_CLIENT`.
+
+The session is terminated calling :c:expr:`close(fd)`.
+
+A code snippet for an application communicating with SPDM client:
+
+.. code-block:: C
+
+        struct issei_connect_client_data data = {.in_client_uuid =
+                {0xe8, 0x51, 0x49, 0xdf, 0x94, 0x47, 0x4C,
+                 0x9A, 0x83, 0x67, 0xC4, 0xE3, 0x34, 0x64, 0xF1, 0xB4}};
+        __u8 req_data[] = {0x10, 0x84, 0x00, 0x00}; /* SPDM Get Version */
+        size_t req_data_len = sizeof(req_data);
+        __u8 res_data[256];
+        size_t res_data_len = sizeof(res_data);
+        int fd = open("/dev/issei0", O_RDWR);
+
+        ioctl(fd, IOCTL_ISSEI_CONNECT_CLIENT, &data);
+
+        printf("Ver=%d, MaxLen=%u, Flags=0x%08X\n",
+               data.out_client_properties.protocol_version,
+               data.out_client_properties.max_msg_length,
+               data.out_client_properties.flags);
+
+        [...]
+
+        write(fd, req_data, req_data_len);
+
+        [...]
+
+        read(fd, res_data, res_data_len);
+
+        printf("SPDM version count %u, version[0]=%02X%02X\n",
+               res_data[5], res_data[6], res_data[7]);
+
+        [...]
+
+        ioctl(fd, IOCTL_ISSEI_DISCONNECT_CLIENT, &data);
+
+        [...]
+
+        close(fd);
+
+
+User space API ioctl
+====================
+
+The Intel SSEI Driver supports the following ioctl commands:
+
+IOCTL_ISSEI_CONNECT_CLIENT
+--------------------------
+Connect to firmware Feature/Client.
+
+.. code-block:: none
+
+        Usage:
+
+        struct issei_connect_client_data client_data;
+
+        ioctl(fd, IOCTL_ISSEI_CONNECT_CLIENT, &client_data);
+
+        struct issei_connect_client_data - contain the following
+        Inputs:
+                in_client_uuid        - UUID of the FW Feature that needs to connect to.
+        Outputs:
+                out_client_properties - Client Properties: MTU, Protocol Version and Flags.
+
+        Error returns:
+                ENOTTY  No such client (i.e. wrong UUID) or connection is not allowed.
+                EINVAL  Wrong IOCTL Number
+                ENODEV  Device or Connection is not initialized or ready.
+                ENOMEM  Unable to allocate memory to client internal data.
+                EFAULT  Fatal Error (e.g. Unable to access user input data)
+                EBUSY   Connection Already Open
+
+:Note:
+        max_msg_length (MTU) in client properties describes the maximum
+        data that can be sent or received. (e.g. with MTU=2K, can send
+        requests up to bytes 2k and received responses up to 2k bytes).
+
+IOCTL_ISSEI_DISCONNECT_CLIENT
+-----------------------------
+Disconnect from firmware Feature/Client.
+
+.. code-block:: none
+
+        Usage:
+
+        ioctl(fd, IOCTL_ISSEI_DISCONNECT_CLIENT, NULL);
+
+        Error returns:
+                EINVAL    Wrong IOCTL Number
+                ENODEV    Device or Connection is not initialized or ready.
+                ENOTCONN  Feature/Client is not connected.