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, 90 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-auxdisplay-linedisp b/Documentation/ABI/testing/sysfs-auxdisplay-linedisp
new file mode 100644
index 000000000..55f1b559e
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-auxdisplay-linedisp
@@ -0,0 +1,90 @@
+What:		/sys/.../message
+Date:		October 2021
+KernelVersion:	5.16
+Description:
+		Controls the text message displayed on character line displays.
+
+		Reading returns the current message with a trailing newline.
+		Writing updates the displayed message. Messages longer than the
+		display width will automatically scroll. Trailing newlines in
+		input are automatically trimmed.
+
+		Writing an empty string clears the display.
+
+		Example:
+		  echo "Hello World" > message
+		  cat message			# Returns "Hello World\n"
+
+What:		/sys/.../num_chars
+Date:		November 2025
+KernelVersion:	6.18
+Contact:	Jean-François Lessard <jefflessard3@gmail.com>
+Description:
+		Read-only attribute showing the character width capacity of
+		the line display device. Messages longer than this will scroll.
+
+		Example:
+		  cat num_chars		# Returns "16\n" for 16-char display
+
+What:		/sys/.../scroll_step_ms
+Date:		October 2021
+KernelVersion:	5.16
+Description:
+		Controls the scrolling speed for messages longer than the display
+		width, specified in milliseconds per scroll step.
+
+		Setting to 0 disables scrolling. Default is 500ms.
+
+		Example:
+		  echo "250" > scroll_step_ms	# 4Hz scrolling
+		  cat scroll_step_ms		# Returns "250\n"
+
+What:		/sys/.../map_seg7
+Date:		January 2024
+KernelVersion:	6.9
+Description:
+		Read/write binary blob representing the ASCII-to-7-segment
+		display conversion table used by the linedisp driver, as defined
+		by struct seg7_conversion_map in <linux/map_to_7segment.h>.
+
+		Only visible on displays with 7-segment capability.
+
+		This attribute is not human-readable. Writes must match the
+		struct size exactly, else -EINVAL is returned; reads return the
+		entire mapping as a binary blob.
+
+		This interface and its implementation match existing conventions
+		used in segment-mapped display drivers since 2005.
+
+		ABI note: This style of binary sysfs attribute *is an exception*
+		to current "one value per file, text only" sysfs rules, for
+		historical compatibility and driver uniformity. New drivers are
+		discouraged from introducing additional binary sysfs ABIs.
+
+		Reference interface guidance:
+		- include/uapi/linux/map_to_7segment.h
+
+What:		/sys/.../map_seg14
+Date:		January 2024
+KernelVersion:	6.9
+Description:
+		Read/write binary blob representing the ASCII-to-14-segment
+		display conversion table used by the linedisp driver, as defined
+		by struct seg14_conversion_map in <linux/map_to_14segment.h>.
+
+		Only visible on displays with 14-segment capability.
+
+		This attribute is not human-readable. Writes must match the
+		struct size exactly, else -EINVAL is returned; reads return the
+		entire mapping as a binary blob.
+
+		This interface and its implementation match existing conventions
+		used by segment-mapped display drivers since 2005.
+
+		ABI note: This style of binary sysfs attribute *is an exception*
+		to current "one value per file, text only" sysfs rules, for
+		historical compatibility and driver uniformity. New drivers are
+		discouraged from introducing additional binary sysfs ABIs.
+
+		Reference interface guidance:
+		- include/uapi/linux/map_to_14segment.h