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, 208 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/debugfs-scmi-raw b/Documentation/ABI/testing/debugfs-scmi-raw
new file mode 100644
index 000000000..5847b96b3
--- /dev/null
+++ b/Documentation/ABI/testing/debugfs-scmi-raw
@@ -0,0 +1,208 @@
+What:		/sys/kernel/debug/scmi/<n>/raw/message
+Date:		March 2023
+KernelVersion:	6.3
+Contact:	cristian.marussi@arm.com
+Description:	SCMI Raw synchronous message injection/snooping facility; write
+		a complete SCMI synchronous command message (header included)
+		in little-endian binary format to have it sent to the configured
+		backend SCMI server for instance <n>.
+		Any subsequently received response can be read from this same
+		entry if it arrived within the configured timeout.
+		Each write to the entry causes one command request to be built
+		and sent while the replies are read back one message at time
+		(receiving an EOF at each message boundary).
+Users:		Debugging, any userspace test suite
+
+What:		/sys/kernel/debug/scmi/<n>/raw/message_async
+Date:		March 2023
+KernelVersion:	6.3
+Contact:	cristian.marussi@arm.com
+Description:	SCMI Raw asynchronous message injection/snooping facility; write
+		a complete SCMI asynchronous command message (header included)
+		in little-endian binary format to have it sent to the configured
+		backend SCMI server for instance <n>.
+		Any subsequently received response can be read from this same
+		entry if it arrived within the configured timeout.
+		Any additional delayed response received afterwards can be read
+		from this same entry too if it arrived within the configured
+		timeout.
+		Each write to the entry causes one command request to be built
+		and sent while the replies are read back one message at time
+		(receiving an EOF at each message boundary).
+Users:		Debugging, any userspace test suite
+
+What:		/sys/kernel/debug/scmi/<n>/raw/message_poll
+Date:		June 2025
+KernelVersion:	6.16
+Contact:	cristian.marussi@arm.com
+Description:	SCMI Raw message injection/snooping facility using polling mode;
+		write a complete SCMI command message (header included) in
+		little-endian binary format to have it sent to the configured
+		backend SCMI server for instance <n>, using polling mode on
+		the reception path. (if transport is polling capable)
+		Any subsequently received response can be read from this same
+		entry if it arrived within the configured timeout.
+		Each write to the entry causes one command request to be built
+		and sent while the replies are read back one message at time
+		(receiving an EOF at each message boundary).
+Users:		Debugging, any userspace test suite
+
+What:		/sys/kernel/debug/scmi/<n>/raw/message_poll_async
+Date:		June 2025
+KernelVersion:	6.16
+Contact:	cristian.marussi@arm.com
+Description:	SCMI Raw asynchronous message injection/snooping facility using
+		polling-mode; write a complete SCMI asynchronous command message
+		(header included) in little-endian binary format to have it sent
+		to the configured backend SCMI server for instance <n>, using
+		polling-mode on the reception path of the immediate part of the
+		asynchronous command. (if transport is polling capable)
+		Any subsequently received response can be read from this same
+		entry if it arrived within the configured timeout.
+		Any additional delayed response received afterwards can be read
+		from this same entry too if it arrived within the configured
+		timeout.
+		Each write to the entry causes one command request to be built
+		and sent while the replies are read back one message at time
+		(receiving an EOF at each message boundary).
+Users:		Debugging, any userspace test suite
+
+What:		/sys/kernel/debug/scmi/<n>/raw/errors
+Date:		March 2023
+KernelVersion:	6.3
+Contact:	cristian.marussi@arm.com
+Description:	SCMI Raw message errors facility; any kind of timed-out or
+		generally unexpectedly received SCMI message, for instance <n>,
+		can be read from this entry.
+		Each read gives back one message at time (receiving an EOF at
+		each message boundary).
+Users:		Debugging, any userspace test suite
+
+What:		/sys/kernel/debug/scmi/<n>/raw/notification
+Date:		March 2023
+KernelVersion:	6.3
+Contact:	cristian.marussi@arm.com
+Description:	SCMI Raw notification snooping facility; any notification
+		emitted by the backend SCMI server, for instance <n>, can be
+		read from this entry.
+		Each read gives back one message at time (receiving an EOF at
+		each message boundary).
+Users:		Debugging, any userspace test suite
+
+What:		/sys/kernel/debug/scmi/<n>/raw/reset
+Date:		March 2023
+KernelVersion:	6.3
+Contact:	cristian.marussi@arm.com
+Description:	SCMI Raw stack reset facility; writing a value to this entry
+		causes the internal queues of any kind of received message,
+		still pending to be read out for instance <n>, to be immediately
+		flushed.
+		Can be used to reset and clean the SCMI Raw stack between to
+		different test-run.
+Users:		Debugging, any userspace test suite
+
+What:		/sys/kernel/debug/scmi/<n>/raw/channels/<m>/message
+Date:		March 2023
+KernelVersion:	6.3
+Contact:	cristian.marussi@arm.com
+Description:	SCMI Raw synchronous message injection/snooping facility; write
+		a complete SCMI synchronous command message (header included)
+		in little-endian binary format to have it sent to the configured
+		backend SCMI server for instance <n> through the <m> transport
+		channel.
+		Any subsequently received response can be read from this same
+		entry if it arrived on channel <m> within the configured
+		timeout.
+		Each write to the entry causes one command request to be built
+		and sent while the replies are read back one message at time
+		(receiving an EOF at each message boundary).
+		Channel identifier <m> matches the SCMI protocol number which
+		has been associated with this transport channel in the DT
+		description, with base protocol number 0x10 being the default
+		channel for this instance.
+		Note that these per-channel entries rooted at <..>/channels
+		exist only if the transport is configured to have more than
+		one default channel.
+Users:		Debugging, any userspace test suite
+
+What:		/sys/kernel/debug/scmi/<n>/raw/channels/<m>/message_async
+Date:		March 2023
+KernelVersion:	6.3
+Contact:	cristian.marussi@arm.com
+Description:	SCMI Raw asynchronous message injection/snooping facility; write
+		a complete SCMI asynchronous command message (header included)
+		in little-endian binary format to have it sent to the configured
+		backend SCMI server for instance <n> through the <m> transport
+		channel.
+		Any subsequently received response can be read from this same
+		entry if it arrived on channel <m> within the configured
+		timeout.
+		Any additional delayed response received afterwards can be read
+		from this same entry too if it arrived within the configured
+		timeout.
+		Each write to the entry causes one command request to be built
+		and sent while the replies are read back one message at time
+		(receiving an EOF at each message boundary).
+		Channel identifier <m> matches the SCMI protocol number which
+		has been associated with this transport channel in the DT
+		description, with base protocol number 0x10 being the default
+		channel for this instance.
+		Note that these per-channel entries rooted at <..>/channels
+		exist only if the transport is configured to have more than
+		one default channel.
+Users:		Debugging, any userspace test suite
+
+
+What:		/sys/kernel/debug/scmi/<n>/raw/channels/<m>/message_poll
+Date:		June 2025
+KernelVersion:	6.16
+Contact:	cristian.marussi@arm.com
+Description:	SCMI Raw message injection/snooping facility using polling mode;
+		write a complete SCMI command message (header included) in
+		little-endian binary format to have it sent to the configured
+		backend SCMI server for instance <n> through the <m> transport
+		channel, using polling mode on the reception path.
+		(if transport is polling capable)
+		Any subsequently received response can be read from this same
+		entry if it arrived on channel <m> within the configured
+		timeout.
+		Each write to the entry causes one command request to be built
+		and sent while the replies are read back one message at time
+		(receiving an EOF at each message boundary).
+		Channel identifier <m> matches the SCMI protocol number which
+		has been associated with this transport channel in the DT
+		description, with base protocol number 0x10 being the default
+		channel for this instance.
+		Note that these per-channel entries rooted at <..>/channels
+		exist only if the transport is configured to have more than
+		one default channel.
+Users:		Debugging, any userspace test suite
+
+What:		/sys/kernel/debug/scmi/<n>/raw/channels/<m>/message_poll_async
+Date:		June 2025
+KernelVersion:	6.16
+Contact:	cristian.marussi@arm.com
+Description:	SCMI Raw asynchronous message injection/snooping facility using
+		polling-mode; write a complete SCMI asynchronous command message
+		(header included) in little-endian binary format to have it sent
+		to the configured backend SCMI server for instance <n> through
+		the <m> transport channel, using polling mode on the reception
+		path of the immediate part of the asynchronous command.
+		(if transport is polling capable)
+		Any subsequently received response can be read from this same
+		entry if it arrived on channel <m> within the configured
+		timeout.
+		Any additional delayed response received afterwards can be read
+		from this same entry too if it arrived within the configured
+		timeout.
+		Each write to the entry causes one command request to be built
+		and sent while the replies are read back one message at time
+		(receiving an EOF at each message boundary).
+		Channel identifier <m> matches the SCMI protocol number which
+		has been associated with this transport channel in the DT
+		description, with base protocol number 0x10 being the default
+		channel for this instance.
+		Note that these per-channel entries rooted at <..>/channels
+		exist only if the transport is configured to have more than
+		one default channel.
+Users:		Debugging, any userspace test suite