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, 130 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-class-devlink b/Documentation/ABI/testing/sysfs-class-devlink
new file mode 100644
index 000000000..8a21ce515
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-class-devlink
@@ -0,0 +1,130 @@
+What:		/sys/class/devlink/.../
+Date:		May 2020
+Contact:	Saravana Kannan <saravanak@google.com>
+Description:
+		Provide a place in sysfs for the device link objects in the
+		kernel at any given time.  The name of a device link directory,
+		denoted as ... above, is of the form <supplier>--<consumer>
+		where <supplier> is the supplier bus:device name and <consumer>
+		is the consumer bus:device name.
+
+What:		/sys/class/devlink/.../auto_remove_on
+Date:		May 2020
+Contact:	Saravana Kannan <saravanak@google.com>
+Description:
+		This file indicates if the device link will ever be
+		automatically removed by the driver core when the consumer and
+		supplier devices themselves are still present.
+
+		This will be one of the following strings:
+
+		- 'consumer unbind'
+		- 'supplier unbind'
+		- 'never'
+
+		'consumer unbind' means the device link will be removed when
+		the consumer's driver is unbound from the consumer device.
+
+		'supplier unbind' means the device link will be removed when
+		the supplier's driver is unbound from the supplier device.
+
+		'never' means the device link will not be automatically removed
+		when as long as the supplier and consumer devices themselves
+		are still present.
+
+What:		/sys/class/devlink/.../consumer
+Date:		May 2020
+Contact:	Saravana Kannan <saravanak@google.com>
+Description:
+		This file is a symlink to the consumer device's sysfs directory.
+
+What:		/sys/class/devlink/.../runtime_pm
+Date:		May 2020
+Contact:	Saravana Kannan <saravanak@google.com>
+Description:
+		This file indicates if the device link has any impact on the
+		runtime power management behavior of the consumer and supplier
+		devices. For example: Making sure the supplier doesn't enter
+		runtime suspend while the consumer is active.
+
+		This will be one of the following strings:
+
+		===   ========================================
+		'0'   Does not affect runtime power management
+		'1'   Affects runtime power management
+		===   ========================================
+
+What:		/sys/class/devlink/.../status
+Date:		May 2020
+Contact:	Saravana Kannan <saravanak@google.com>
+Description:
+		This file indicates the status of the device link. The status
+		of a device link is affected by whether the supplier and
+		consumer devices have been bound to their corresponding
+		drivers. The status of a device link also affects the binding
+		and unbinding of the supplier and consumer devices with their
+		drivers and also affects whether the software state of the
+		supplier device is synced with the hardware state of the
+		supplier device after boot up.
+		See also: sysfs-devices-state_synced.
+
+		This will be one of the following strings:
+
+		- 'not tracked'
+		- 'dormant'
+		- 'available'
+		- 'consumer probing'
+		- 'active'
+		- 'supplier unbinding'
+		- 'unknown'
+
+		'not tracked' means this device link does not track the status
+		and has no impact on the binding, unbinding and syncing the
+		hardware and software device state.
+
+		'dormant' means the supplier and the consumer devices have not
+		bound to their driver.
+
+		'available' means the supplier has bound to its driver and is
+		available to supply resources to the consumer device.
+
+		'consumer probing' means the consumer device is currently
+		trying to bind to its driver.
+
+		'active' means the supplier and consumer devices have both
+		bound successfully to their drivers.
+
+		'supplier unbinding' means the supplier devices is currently in
+		the process of unbinding from its driver.
+
+		'unknown' means the state of the device link is not any of the
+		above. If this is ever the value, there's a bug in the kernel.
+
+What:		/sys/class/devlink/.../supplier
+Date:		May 2020
+Contact:	Saravana Kannan <saravanak@google.com>
+Description:
+		This file is a symlink to the supplier device's sysfs directory.
+
+What:		/sys/class/devlink/.../sync_state_only
+Date:		May 2020
+Contact:	Saravana Kannan <saravanak@google.com>
+Description:
+		This file indicates if the device link is limited to only
+		affecting the syncing of the hardware and software state of the
+		supplier device.
+
+		This will be one of the following strings:
+
+		===  ================================
+		'0'
+		'1'  Affects runtime power management
+		===  ================================
+
+		'0' means the device link can affect other device behaviors
+		like binding/unbinding, suspend/resume, runtime power
+		management, etc.
+
+		'1' means the device link will only affect the syncing of
+		hardware and software state of the supplier device after boot
+		up and doesn't not affect other behaviors of the devices.