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, 137 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-class-rc b/Documentation/ABI/testing/sysfs-class-rc
new file mode 100644
index 000000000..84e46d70d
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-class-rc
@@ -0,0 +1,137 @@
+What:		/sys/class/rc/
+Date:		Apr 2010
+KernelVersion:	2.6.35
+Contact:	Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
+Description:
+		The rc/ class sub-directory belongs to the Remote Controller
+		core and provides a sysfs interface for configuring infrared
+		remote controller receivers.
+
+What:		/sys/class/rc/rc<N>/
+Date:		Apr 2010
+KernelVersion:	2.6.35
+Contact:	Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
+Description:
+		A /sys/class/rc/rcN directory is created for each remote
+		control receiver device where N is the number of the receiver.
+
+What:		/sys/class/rc/rc<N>/protocols
+Date:		Jun 2010
+KernelVersion:	2.6.36
+Contact:	Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
+Description:
+		Reading this file returns a list of available protocols,
+		something like::
+
+		    "rc5 [rc6] nec jvc [sony]"
+
+		Enabled protocols are shown in [] brackets.
+
+		Writing "+proto" will add a protocol to the list of enabled
+		protocols.
+
+		Writing "-proto" will remove a protocol from the list of enabled
+		protocols.
+
+		Writing "proto" will enable only "proto".
+
+		Writing "none" will disable all protocols.
+
+		Write fails with EINVAL if an invalid protocol combination or
+		unknown protocol name is used.
+
+What:		/sys/class/rc/rc<N>/filter
+Date:		Jan 2014
+KernelVersion:	3.15
+Contact:	Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
+Description:
+		Sets the scancode filter expected value.
+
+		Use in combination with /sys/class/rc/rcN/filter_mask to set the
+		expected value of the bits set in the filter mask.
+		If the hardware supports it then scancodes which do not match
+		the filter will be ignored. Otherwise the write will fail with
+		an error.
+
+		This value may be reset to 0 if the current protocol is altered.
+
+What:		/sys/class/rc/rc<N>/filter_mask
+Date:		Jan 2014
+KernelVersion:	3.15
+Contact:	Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
+Description:
+		Sets the scancode filter mask of bits to compare.
+		Use in combination with /sys/class/rc/rcN/filter to set the bits
+		of the scancode which should be compared against the expected
+		value. A value of 0 disables the filter to allow all valid
+		scancodes to be processed.
+
+		If the hardware supports it then scancodes which do not match
+		the filter will be ignored. Otherwise the write will fail with
+		an error.
+
+		This value may be reset to 0 if the current protocol is altered.
+
+What:		/sys/class/rc/rc<N>/wakeup_protocols
+Date:		Feb 2017
+KernelVersion:	4.11
+Contact:	Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
+Description:
+		Reading this file returns a list of available protocols to use
+		for the wakeup filter, something like::
+
+		    "rc-5 nec nec-x rc-6-0 rc-6-6a-24 [rc-6-6a-32] rc-6-mce"
+
+		Note that protocol variants are listed, so "nec", "sony",
+		"rc-5", "rc-6" have their different bit length encodings
+		listed if available.
+
+		The enabled wakeup protocol is shown in [] brackets.
+
+		Only one protocol can be selected at a time.
+
+		Writing "proto" will use "proto" for wakeup events.
+
+		Writing "none" will disable wakeup.
+
+		Write fails with EINVAL if an invalid protocol combination or
+		unknown protocol name is used, or if wakeup is not supported by
+		the hardware.
+
+What:		/sys/class/rc/rc<N>/wakeup_filter
+Date:		Jan 2014
+KernelVersion:	3.15
+Contact:	Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
+Description:
+		Sets the scancode wakeup filter expected value.
+
+		Use in combination with /sys/class/rc/rcN/wakeup_filter_mask to
+		set the expected value of the bits set in the wakeup filter mask
+		to trigger a system wake event.
+
+		If the hardware supports it and wakeup_filter_mask is not 0 then
+		scancodes which match the filter will wake the system from e.g.
+		suspend to RAM or power off.
+
+		Otherwise the write will fail with an error.
+
+		This value may be reset to 0 if the wakeup protocol is altered.
+
+What:		/sys/class/rc/rc<N>/wakeup_filter_mask
+Date:		Jan 2014
+KernelVersion:	3.15
+Contact:	Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
+Description:
+		Sets the scancode wakeup filter mask of bits to compare.
+
+		Use in combination with /sys/class/rc/rcN/wakeup_filter to set
+		the bits of the scancode which should be compared against the
+		expected value to trigger a system wake event.
+
+		If the hardware supports it and wakeup_filter_mask is not 0 then
+		scancodes which match the filter will wake the system from e.g.
+		suspend to RAM or power off.
+
+		Otherwise the write will fail with an error.
+
+		This value may be reset to 0 if the wakeup protocol is altered.