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, 152 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-class-powercap b/Documentation/ABI/testing/sysfs-class-powercap
new file mode 100644
index 000000000..ca491ec4e
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-class-powercap
@@ -0,0 +1,152 @@
+What:		/sys/class/powercap/
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		The powercap/ class sub directory belongs to the power cap
+		subsystem. Refer to
+		Documentation/power/powercap/powercap.rst for details.
+
+What:		/sys/class/powercap/<control type>
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		A <control type> is a unique name under /sys/class/powercap.
+		Here <control type> determines how the power is going to be
+		controlled. A <control type> can contain multiple power zones.
+
+What:		/sys/class/powercap/<control type>/enabled
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		This allows to enable/disable power capping for a "control type".
+		This status affects every power zone using this "control_type.
+
+What:		/sys/class/powercap/<control type>/<power zone>
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		A power zone is a single or a collection of devices, which can
+		be independently monitored and controlled. A power zone sysfs
+		entry is qualified with the name of the <control type>.
+		E.g. intel-rapl:0:1:1.
+
+What:		/sys/class/powercap/<control type>/<power zone>/<child power zone>
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Power zones may be organized in a hierarchy in which child
+		power zones provide monitoring and control for a subset of
+		devices under the parent. For example, if there is a parent
+		power zone for a whole CPU package, each CPU core in it can
+		be a child power zone.
+
+What:		/sys/class/powercap/.../<power zone>/name
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Specifies the name of this power zone.
+
+What:		/sys/class/powercap/.../<power zone>/energy_uj
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Current energy counter in micro-joules. Write "0" to reset.
+		If the counter can not be reset, then this attribute is
+		read-only.
+
+What:		/sys/class/powercap/.../<power zone>/max_energy_range_uj
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Range of the above energy counter in micro-joules.
+
+
+What:		/sys/class/powercap/.../<power zone>/power_uw
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Current power in micro-watts.
+
+What:		/sys/class/powercap/.../<power zone>/max_power_range_uw
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Range of the above power value in micro-watts.
+
+What:		/sys/class/powercap/.../<power zone>/constraint_X_name
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Each power zone can define one or more constraints. Each
+		constraint can have an optional name. Here "X" can have values
+		from 0 to max integer.
+
+What:		/sys/class/powercap/.../<power zone>/constraint_X_power_limit_uw
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Power limit in micro-watts should be applicable for
+		the time window specified by "constraint_X_time_window_us".
+		Here "X" can have values from 0 to max integer.
+
+What:		/sys/class/powercap/.../<power zone>/constraint_X_time_window_us
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Time window in micro seconds. This is used along with
+		constraint_X_power_limit_uw to define a power constraint.
+		Here "X" can have values from 0 to max integer.
+
+
+What:		/sys/class/powercap/<control type>/.../constraint_X_max_power_uw
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Maximum allowed power in micro watts for this constraint.
+		Here "X" can have values from 0 to max integer.
+
+What:		/sys/class/powercap/<control type>/.../constraint_X_min_power_uw
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Minimum allowed power in micro watts for this constraint.
+		Here "X" can have values from 0 to max integer.
+
+What:		/sys/class/powercap/.../<power zone>/constraint_X_max_time_window_us
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Maximum allowed time window in micro seconds for this
+		constraint. Here "X" can have values from 0 to max integer.
+
+What:		/sys/class/powercap/.../<power zone>/constraint_X_min_time_window_us
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Minimum allowed time window in micro seconds for this
+		constraint. Here "X" can have values from 0 to max integer.
+
+What:		/sys/class/powercap/.../<power zone>/enabled
+Date:		September 2013
+KernelVersion:	3.13
+Contact:	linux-pm@vger.kernel.org
+Description:
+		This allows to enable/disable power capping at power zone level.
+		This applies to current power zone and its children.