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, 78 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/debugfs-cros-ec b/Documentation/ABI/testing/debugfs-cros-ec
new file mode 100644
index 000000000..9a040c6f5
--- /dev/null
+++ b/Documentation/ABI/testing/debugfs-cros-ec
@@ -0,0 +1,78 @@
+What:		/sys/kernel/debug/<cros-ec-device>/console_log
+Date:		September 2017
+KernelVersion:	4.13
+Description:
+		If the EC supports the CONSOLE_READ command type, this file
+		can be used to grab the EC logs. The kernel polls for the log
+		and keeps its own buffer but userspace should grab this and
+		write it out to some logs.
+
+What:		/sys/kernel/debug/<cros-ec-device>/panicinfo
+Date:		September 2017
+KernelVersion:	4.13
+Description:
+		This file dumps the EC panic information from the previous
+		reboot. This file will only exist if the PANIC_INFO command
+		type is supported by the EC.
+
+What:		/sys/kernel/debug/<cros-ec-device>/pdinfo
+Date:		June 2018
+KernelVersion:	4.17
+Description:
+		This file provides the port role, muxes and power debug
+		information for all the USB PD/type-C ports available. If
+		the are no ports available, this file will be just an empty
+		file.
+
+What:		/sys/kernel/debug/<cros-ec-device>/uptime
+Date:		June 2019
+KernelVersion:	5.3
+Description:
+		A u32 providing the time since EC booted in ms. This is
+		is used for synchronizing the AP host time with the EC
+		log. An error is returned if the command is not supported
+		by the EC or there is a communication problem.
+
+What:		/sys/kernel/debug/<cros-ec-device>/last_resume_result
+Date:		June 2019
+KernelVersion:	5.3
+Description:
+		Some ECs have a feature where they will track transitions to
+		the (Intel) processor's SLP_S0 line, in order to detect cases
+		where a system failed to go into S0ix. When the system resumes,
+		an EC with this feature will return a summary of SLP_S0
+		transitions that occurred. The last_resume_result file returns
+		the most recent response from the AP's resume message to the EC.
+
+		The bottom 31 bits contain a count of the number of SLP_S0
+		transitions that occurred since the suspend message was
+		received. Bit 31 is set if the EC attempted to wake the
+		system due to a timeout when watching for SLP_S0 transitions.
+		Callers can use this to detect a wake from the EC due to
+		S0ix timeouts. The result will be zero if no suspend
+		transitions have been attempted, or the EC does not support
+		this feature.
+
+		Output will be in the format: "0x%08x\n".
+
+What:		/sys/kernel/debug/<cros-ec-device>/suspend_timeout_ms
+Date:		August 2022
+KernelVersion:	6.1
+Description:
+		Some ECs have a feature where they will track transitions of
+		a hardware-controlled sleep line, such as Intel's SLP_S0 line,
+		in order to detect cases where a system failed to go into deep
+		sleep states. The suspend_timeout_ms file controls the amount of
+		time in milliseconds the EC will wait before declaring a sleep
+		timeout event and attempting to wake the system.
+
+		Supply 0 to use the default value coded into EC firmware. Supply
+		65535 (EC_HOST_SLEEP_TIMEOUT_INFINITE) to disable the EC sleep
+		failure detection mechanism. Values in between 0 and 65535
+		indicate the number of milliseconds the EC should wait after a
+		sleep transition before declaring a timeout. This includes both
+		the duration after a sleep command was received but before the
+		hardware line changed, as well as the duration between when the
+		hardware line changed and the kernel sent an EC resume command.
+
+		Output will be in the format: "%u\n".