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, 84 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-module b/Documentation/ABI/testing/sysfs-module
new file mode 100644
index 000000000..d5b7d19bd
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-module
@@ -0,0 +1,84 @@
+What:		/sys/module/pch_phub/drivers/.../pch_mac
+Date:		August 2010
+KernelVersion:	2.6.35
+Contact:	masa-korg@dsn.okisemi.com
+Description:	Write/read GbE MAC address.
+
+What:		/sys/module/pch_phub/drivers/.../pch_firmware
+Date:		August 2010
+KernelVersion:	2.6.35
+Contact:	masa-korg@dsn.okisemi.com
+Description:	Write/read Option ROM data.
+
+
+What:		/sys/module/ehci_hcd/drivers/.../uframe_periodic_max
+Date:		July 2011
+KernelVersion:	3.1
+Contact:	Kirill Smelkov <kirr@mns.spb.ru>
+Description:	Maximum time allowed for periodic transfers per microframe (μs)
+
+		Note:
+		  USB 2.0 sets maximum allowed time for periodic transfers per
+		  microframe to be 80%, that is 100 microseconds out of 125
+		  microseconds (full microframe).
+
+		  However there are cases, when 80% max isochronous bandwidth is
+		  too limiting. For example two video streams could require 110
+		  microseconds of isochronous bandwidth per microframe to work
+		  together.
+
+		Through this setting it is possible to raise the limit so that
+		the host controller would allow allocating more than 100
+		microseconds of periodic bandwidth per microframe.
+
+		Beware, non-standard modes are usually not thoroughly tested by
+		hardware designers, and the hardware can malfunction when this
+		setting differ from default 100.
+
+What:		/sys/module/*/{coresize,initsize}
+Date:		Jan 2012
+KernelVersion:	3.3
+Contact:	Kay Sievers <kay.sievers@vrfy.org>
+Description:	Module size in bytes.
+
+What:		/sys/module/*/initstate
+Date:		Nov 2006
+KernelVersion:	2.6.19
+Contact:	Kay Sievers <kay.sievers@vrfy.org>
+Description:	Show the initialization state(live, coming, going) of
+		the module.
+
+What:		/sys/module/*/import_ns
+Date:		January 2026
+KernelVersion:	7.1
+Contact:	linux-modules@vger.kernel.org
+Description:	List of symbol namespaces imported by this module via
+		MODULE_IMPORT_NS(). Each namespace appears on a separate line.
+		This file only exists for modules that import at least one
+		namespace.
+
+What:		/sys/module/*/taint
+Date:		Jan 2012
+KernelVersion:	3.3
+Contact:	Kay Sievers <kay.sievers@vrfy.org>
+Description:	Module taint flags:
+			==  =====================
+			P   proprietary module
+			O   out-of-tree module
+			F   force-loaded module
+			C   staging driver module
+			E   unsigned module
+			K   livepatch module
+			N   in-kernel test module
+			==  =====================
+
+What:		/sys/module/grant_table/parameters/free_per_iteration
+Date:		July 2023
+KernelVersion:	6.5 but backported to all supported stable branches
+Contact:	Xen developer discussion <xen-devel@lists.xenproject.org>
+Description:	Read and write number of grant entries to attempt to free per iteration.
+
+		Note: Future versions of Xen and Linux may provide a better
+		interface for controlling the rate of deferred grant reclaim
+		or may not need it at all.
+Users:		Qubes OS (https://www.qubes-os.org)