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, 91 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-class-rtc b/Documentation/ABI/testing/sysfs-class-rtc
new file mode 100644
index 000000000..95984289a
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-class-rtc
@@ -0,0 +1,91 @@
+What:		/sys/class/rtc/
+Date:		March 2006
+KernelVersion:	2.6.17
+Contact:	linux-rtc@vger.kernel.org
+Description:
+		The rtc/ class subdirectory belongs to the RTC subsystem.
+
+What:		/sys/class/rtc/rtcX/
+Date:		March 2006
+KernelVersion:	2.6.17
+Contact:	linux-rtc@vger.kernel.org
+Description:
+		The /sys/class/rtc/rtc{0,1,2,3,...} directories correspond
+		to each RTC device.
+
+What:		/sys/class/rtc/rtcX/date
+Date:		March 2006
+KernelVersion:	2.6.17
+Contact:	linux-rtc@vger.kernel.org
+Description:
+		(RO) RTC-provided date in YYYY-MM-DD format
+
+What:		/sys/class/rtc/rtcX/hctosys
+Date:		September 2009
+KernelVersion:	2.6.32
+Contact:	linux-rtc@vger.kernel.org
+Description:
+		(RO) 1 if the RTC provided the system time at boot via the
+		CONFIG_RTC_HCTOSYS kernel option, 0 otherwise
+
+What:		/sys/class/rtc/rtcX/max_user_freq
+Date:		October 2007
+KernelVersion:	2.6.24
+Contact:	linux-rtc@vger.kernel.org
+Description:
+		(RW) The maximum interrupt rate an unprivileged user may request
+		from this RTC.
+
+What:		/sys/class/rtc/rtcX/name
+Date:		March 2006
+KernelVersion:	2.6.17
+Contact:	linux-rtc@vger.kernel.org
+Description:
+		(RO) The name of the RTC corresponding to this sysfs directory
+
+What:		/sys/class/rtc/rtcX/range
+Date:		January 2018
+KernelVersion:	4.16
+Contact:	linux-rtc@vger.kernel.org
+Description:
+		Valid time range for the RTC, as seconds from epoch, formatted
+		as [min, max]
+
+What:		/sys/class/rtc/rtcX/since_epoch
+Date:		March 2006
+KernelVersion:	2.6.17
+Contact:	linux-rtc@vger.kernel.org
+Description:
+		(RO) RTC-provided time as the number of seconds since the epoch
+
+What:		/sys/class/rtc/rtcX/time
+Date:		March 2006
+KernelVersion:	2.6.17
+Contact:	linux-rtc@vger.kernel.org
+Description:
+		(RO) RTC-provided time in 24-hour notation (hh:mm:ss)
+
+What:		/sys/class/rtc/rtcX/offset
+Date:		February 2016
+KernelVersion:	4.6
+Contact:	linux-rtc@vger.kernel.org
+Description:
+		(RW) The amount which the rtc clock has been adjusted in
+		firmware. Visible only if the driver supports clock offset
+		adjustment. The unit is parts per billion, i.e. The number of
+		clock ticks which are added to or removed from the rtc's base
+		clock per billion ticks. A positive value makes a day pass more
+		slowly, longer, and a negative value makes a day pass more
+		quickly.
+
+What:		/sys/class/rtc/rtcX/wakealarm
+Date:		February 2007
+KernelVersion:	2.6.20
+Contact:	linux-rtc@vger.kernel.org
+Description:
+		(RW) The time at which the clock will generate a system wakeup
+		event. This is a one shot wakeup event, so must be reset after
+		wake if a daily wakeup is required. Format is seconds since the
+		epoch by default, or if there's a leading +, seconds in the
+		future, or if there is a leading +=, seconds ahead of the
+		current alarm.