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, 110 insertions, 0 deletions

diff --git a/Documentation/admin-guide/media/vimc.rst b/Documentation/admin-guide/media/vimc.rst
new file mode 100644
index 000000000..29d843a8d
--- /dev/null
+++ b/Documentation/admin-guide/media/vimc.rst
@@ -0,0 +1,110 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+The Virtual Media Controller Driver (vimc)
+==========================================
+
+The vimc driver emulates complex video hardware using the V4L2 API and the Media
+API. It has a capture device and three subdevices: sensor, debayer and scaler.
+
+Topology
+--------
+
+The topology is hardcoded, although you could modify it in vimc-core and
+recompile the driver to achieve your own topology. This is the default topology:
+
+.. _vimc_topology_graph:
+
+.. kernel-figure:: vimc.dot
+    :alt:   Diagram of the default media pipeline topology
+    :align: center
+
+    Media pipeline graph on vimc
+
+Configuring the topology
+~~~~~~~~~~~~~~~~~~~~~~~~
+
+Each subdevice will come with its default configuration (pixelformat, height,
+width, ...). One needs to configure the topology in order to match the
+configuration on each linked subdevice to stream frames through the pipeline.
+If the configuration doesn't match, the stream will fail. The ``v4l-utils``
+package is a bundle of user-space applications, that comes with ``media-ctl`` and
+``v4l2-ctl`` that can be used to configure the vimc configuration. This sequence
+of commands fits for the default topology:
+
+.. code-block:: bash
+
+        media-ctl -d platform:vimc -V '"Sensor A":0[fmt:SBGGR8_1X8/640x480]'
+        media-ctl -d platform:vimc -V '"Debayer A":0[fmt:SBGGR8_1X8/640x480]'
+        media-ctl -d platform:vimc -V '"Scaler":0[fmt:RGB888_1X24/640x480]'
+        media-ctl -d platform:vimc -V '"Scaler":0[crop:(100,50)/400x150]'
+        media-ctl -d platform:vimc -V '"Scaler":1[fmt:RGB888_1X24/300x700]'
+        v4l2-ctl -z platform:vimc -d "RGB/YUV Capture" -v width=300,height=700
+        v4l2-ctl -z platform:vimc -d "Raw Capture 0" -v pixelformat=BA81
+
+Subdevices
+----------
+
+Subdevices define the behavior of an entity in the topology. Depending on the
+subdevice, the entity can have multiple pads of type source or sink.
+
+vimc-sensor:
+	Generates images in several formats using video test pattern generator.
+	Exposes:
+
+	* 1 Pad source
+
+vimc-lens:
+	Ancillary lens for a sensor. Supports auto focus control. Linked to
+	a vimc-sensor using an ancillary link. The lens supports FOCUS_ABSOLUTE
+	control.
+
+.. code-block:: bash
+
+	media-ctl -p
+	...
+	- entity 28: Lens A (0 pad, 0 link)
+			type V4L2 subdev subtype Lens flags 0
+			device node name /dev/v4l-subdev6
+	- entity 29: Lens B (0 pad, 0 link)
+			type V4L2 subdev subtype Lens flags 0
+			device node name /dev/v4l-subdev7
+	v4l2-ctl -d /dev/v4l-subdev7 -C focus_absolute
+	focus_absolute: 0
+
+
+vimc-debayer:
+	Transforms images in bayer format into a non-bayer format.
+	Exposes:
+
+	* 1 Pad sink
+	* 1 Pad source
+
+vimc-scaler:
+	Re-size the image to meet the source pad resolution. E.g.: if the sync
+	pad is configured to 360x480 and the source to 1280x720, the image will
+	be stretched to fit the source resolution. Works for any resolution
+	within the vimc limitations (even shrinking the image if necessary).
+	Exposes:
+
+	* 1 Pad sink
+	* 1 Pad source
+
+vimc-capture:
+	Exposes node /dev/videoX to allow userspace to capture the stream.
+	Exposes:
+
+	* 1 Pad sink
+	* 1 Pad source
+
+Module options
+--------------
+
+Vimc has a module parameter to configure the driver.
+
+* ``allocator=<unsigned int>``
+
+	memory allocator selection, default is 0. It specifies the way buffers
+	will be allocated.
+
+		- 0: vmalloc
+		- 1: dma-contig