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, 74 insertions, 0 deletions

diff --git a/Documentation/driver-api/sm501.rst b/Documentation/driver-api/sm501.rst
new file mode 100644
index 000000000..882507453
--- /dev/null
+++ b/Documentation/driver-api/sm501.rst
@@ -0,0 +1,74 @@
+.. include:: <isonum.txt>
+
+============
+SM501 Driver
+============
+
+:Copyright: |copy| 2006, 2007 Simtec Electronics
+
+The Silicon Motion SM501 multimedia companion chip is a multifunction device
+which may provide numerous interfaces including USB host controller USB gadget,
+asynchronous serial ports, audio functions, and a dual display video interface.
+The device may be connected by PCI or local bus with varying functions enabled.
+
+Core
+----
+
+The core driver in drivers/mfd provides common services for the
+drivers which manage the specific hardware blocks. These services
+include locking for common registers, clock control and resource
+management.
+
+The core registers drivers for both PCI and generic bus based
+chips via the platform device and driver system.
+
+On detection of a device, the core initialises the chip (which may
+be specified by the platform data) and then exports the selected
+peripheral set as platform devices for the specific drivers.
+
+The core re-uses the platform device system as the platform device
+system provides enough features to support the drivers without the
+need to create a new bus-type and the associated code to go with it.
+
+
+Resources
+---------
+
+Each peripheral has a view of the device which is implicitly narrowed to
+the specific set of resources that peripheral requires in order to
+function correctly.
+
+The centralised memory allocation allows the driver to ensure that the
+maximum possible resource allocation can be made to the video subsystem
+as this is by-far the most resource-sensitive of the on-chip functions.
+
+The primary issue with memory allocation is that of moving the video
+buffers once a display mode is chosen. Indeed when a video mode change
+occurs the memory footprint of the video subsystem changes.
+
+Since video memory is difficult to move without changing the display
+(unless sufficient contiguous memory can be provided for the old and new
+modes simultaneously) the video driver fully utilises the memory area
+given to it by aligning fb0 to the start of the area and fb1 to the end
+of it. Any memory left over in the middle is used for the acceleration
+functions, which are transient and thus their location is less critical
+as it can be moved.
+
+
+Configuration
+-------------
+
+The platform device driver uses a set of platform data to pass
+configurations through to the core and the subsidiary drivers
+so that there can be support for more than one system carrying
+an SM501 built into a single kernel image.
+
+The PCI driver assumes that the PCI card behaves as per the Silicon
+Motion reference design.
+
+There is an errata (AB-5) affecting the selection of the
+of the M1XCLK and M1CLK frequencies. These two clocks
+must be sourced from the same PLL, although they can then
+be divided down individually. If this is not set, then SM501 may
+lock and hang the whole system. The driver will refuse to
+attach if the PLL selection is different.