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, 142 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-ptp b/Documentation/ABI/testing/sysfs-ptp
new file mode 100644
index 000000000..9c317ac7c
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-ptp
@@ -0,0 +1,142 @@
+What:		/sys/class/ptp/
+Date:		September 2010
+Contact:	Richard Cochran <richardcochran@gmail.com>
+Description:
+		This directory contains files and directories
+		providing a standardized interface to the ancillary
+		features of PTP hardware clocks.
+
+What:		/sys/class/ptp/ptp<N>/
+Date:		September 2010
+Contact:	Richard Cochran <richardcochran@gmail.com>
+Description:
+		This directory contains the attributes of the Nth PTP
+		hardware clock registered into the PTP class driver
+		subsystem.
+
+What:		/sys/class/ptp/ptp<N>/clock_name
+Date:		September 2010
+Contact:	Richard Cochran <richardcochran@gmail.com>
+Description:
+		This file contains the name of the PTP hardware clock
+		as a human readable string. The purpose of this
+		attribute is to provide the user with a "friendly
+		name" and to help distinguish PHY based devices from
+		MAC based ones. The string does not necessarily have
+		to be any kind of unique id.
+
+What:		/sys/class/ptp/ptp<N>/max_adjustment
+Date:		September 2010
+Contact:	Richard Cochran <richardcochran@gmail.com>
+Description:
+		This file contains the PTP hardware clock's maximum
+		frequency adjustment value (a positive integer) in
+		parts per billion.
+
+What:		/sys/class/ptp/ptp<N>/max_vclocks
+Date:		May 2021
+Contact:	Yangbo Lu <yangbo.lu@nxp.com>
+Description:
+		This file contains the maximum number of ptp vclocks.
+		Write integer to re-configure it.
+
+What:		/sys/class/ptp/ptp<N>/n_alarms
+Date:		September 2010
+Contact:	Richard Cochran <richardcochran@gmail.com>
+Description:
+		This file contains the number of periodic or one shot
+		alarms offer by the PTP hardware clock.
+
+What:		/sys/class/ptp/ptp<N>/n_external_timestamps
+Date:		September 2010
+Contact:	Richard Cochran <richardcochran@gmail.com>
+Description:
+		This file contains the number of external timestamp
+		channels offered by the PTP hardware clock.
+
+What:		/sys/class/ptp/ptp<N>/n_periodic_outputs
+Date:		September 2010
+Contact:	Richard Cochran <richardcochran@gmail.com>
+Description:
+		This file contains the number of programmable periodic
+		output channels offered by the PTP hardware clock.
+
+What:		/sys/class/ptp/ptp<N>/n_pins
+Date:		March 2014
+Contact:	Richard Cochran <richardcochran@gmail.com>
+Description:
+		This file contains the number of programmable pins
+		offered by the PTP hardware clock.
+
+What:		/sys/class/ptp/ptp<N>/n_vclocks
+Date:		May 2021
+Contact:	Yangbo Lu <yangbo.lu@nxp.com>
+Description:
+		This file contains the number of virtual PTP clocks in
+		use.  By default, the value is 0 meaning that only the
+		physical clock is in use.  Setting the value creates
+		the corresponding number of virtual clocks and causes
+		the physical clock to become free running.  Setting the
+		value back to 0 deletes the virtual clocks and
+		switches the physical clock back to normal, adjustable
+		operation.
+
+What:		/sys/class/ptp/ptp<N>/pins
+Date:		March 2014
+Contact:	Richard Cochran <richardcochran@gmail.com>
+Description:
+		This directory contains one file for each programmable
+		pin offered by the PTP hardware clock. The file name
+		is the hardware dependent pin name. Reading from this
+		file produces two numbers, the assigned function (see
+		the `PTP_PF_` enumeration values in linux/ptp_clock.h)
+		and the channel number. The function and channel
+		assignment may be changed by two writing numbers into
+		the file.
+
+What:		/sys/class/ptp/ptp<N>/pps_available
+Date:		September 2010
+Contact:	Richard Cochran <richardcochran@gmail.com>
+Description:
+		This file indicates whether the PTP hardware clock
+		supports a Pulse Per Second to the host CPU. Reading
+		"1" means that the PPS is supported, while "0" means
+		not supported.
+
+What:		/sys/class/ptp/ptp<N>/extts_enable
+Date:		September 2010
+Contact:	Richard Cochran <richardcochran@gmail.com>
+Description:
+		This write-only file enables or disables external
+		timestamps. To enable external timestamps, write the
+		channel index followed by a "1" into the file.
+		To disable external timestamps, write the channel
+		index followed by a "0" into the file.
+
+What:		/sys/class/ptp/ptp<N>/fifo
+Date:		September 2010
+Contact:	Richard Cochran <richardcochran@gmail.com>
+Description:
+		This file provides timestamps on external events, in
+		the form of three integers: channel index, seconds,
+		and nanoseconds.
+
+What:		/sys/class/ptp/ptp<N>/period
+Date:		September 2010
+Contact:	Richard Cochran <richardcochran@gmail.com>
+Description:
+		This write-only file enables or disables periodic
+		outputs. To enable a periodic output, write five
+		integers into the file: channel index, start time
+		seconds, start time nanoseconds, period seconds, and
+		period nanoseconds. To disable a periodic output, set
+		all the seconds and nanoseconds values to zero.
+
+What:		/sys/class/ptp/ptp<N>/pps_enable
+Date:		September 2010
+Contact:	Richard Cochran <richardcochran@gmail.com>
+Description:
+		This write-only file enables or disables delivery of
+		PPS events to the Linux PPS subsystem. To enable PPS
+		events, write a "1" into the file. To disable events,
+		write a "0" into the file.