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, 89 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-class-led b/Documentation/ABI/testing/sysfs-class-led
new file mode 100644
index 000000000..d4c918cc1
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-class-led
@@ -0,0 +1,89 @@
+What:		/sys/class/leds/<led>/brightness
+Date:		March 2006
+KernelVersion:	2.6.17
+Contact:	Richard Purdie <rpurdie@rpsys.net>
+Description:
+		Set the brightness of the LED.
+
+		Most LEDs don't have hardware brightness support, so will
+		just be turned on for non-zero brightness settings.
+
+		.. Note::
+
+		  For multicolor LEDs, writing to this file will update all
+		  LEDs within the group to a calculated percentage of what
+		  each color LED intensity is set to.
+
+		  The percentage is calculated for each grouped LED via
+		  the equation below::
+
+		    led_brightness = brightness * multi_intensity/max_brightness
+
+		  For additional details please refer to
+		  Documentation/leds/leds-class-multicolor.rst.
+
+		The value is between 0 and /sys/class/leds/<led>/max_brightness
+		and is represented by as a decimal.
+
+		Writing 0 to this file clears active trigger.
+
+		Writing non-zero to this file while trigger is active changes the
+		top brightness trigger is going to use.
+
+
+
+What:		/sys/class/leds/<led>/max_brightness
+Date:		March 2006
+KernelVersion:	2.6.17
+Contact:	Richard Purdie <rpurdie@rpsys.net>
+Description:
+		Maximum brightness level for this LED, default is 255 (LED_FULL).
+
+		If the LED does not support different brightness levels, this
+		should be 1.
+
+What:		/sys/class/leds/<led>/brightness_hw_changed
+Date:		January 2017
+KernelVersion:	4.11
+Description:
+		Last hardware set brightness level for this LED. Some LEDs
+		may be changed autonomously by hardware/firmware. Only LEDs
+		where this happens and the driver can detect this, will have
+		this file.
+
+		This file supports poll() to detect when the hardware changes
+		the brightness.
+
+		Reading this file will return the last brightness level set
+		by the hardware, this may be different from the current
+		brightness. Reading this file when no hw brightness change
+		event has happened will return an ENODATA error.
+
+What:		/sys/class/leds/<led>/trigger
+Date:		March 2006
+KernelVersion:	2.6.17
+Contact:	Richard Purdie <rpurdie@rpsys.net>
+Description:
+		Set the trigger for this LED. A trigger is a kernel based source
+		of LED events.
+
+		You can change triggers in a similar manner to the way an IO
+		scheduler is chosen. Trigger specific parameters can appear in
+		/sys/class/leds/<led> once a given trigger is selected. For
+		their documentation see `sysfs-class-led-trigger-*`.
+
+		Writing "none" removes the trigger for this LED.
+
+		Writing "default" sets the trigger to the LED's default trigger
+		(which would often be configured in the device tree for the
+		hardware).
+
+What:		/sys/class/leds/<led>/inverted
+Date:		January 2011
+KernelVersion:	2.6.38
+Contact:	Richard Purdie <rpurdie@rpsys.net>
+Description:
+		Invert the LED on/off state. This parameter is specific to
+		gpio and backlight triggers. In case of the backlight trigger,
+		it is useful when driving a LED which is intended to indicate
+		a device in a standby like state.