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, 119 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-block-device b/Documentation/ABI/testing/sysfs-block-device
new file mode 100644
index 000000000..2d543cfa4
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-block-device
@@ -0,0 +1,119 @@
+What:		/sys/block/*/device/sw_activity
+Date:		Jun, 2008
+KernelVersion:	v2.6.27
+Contact:	linux-ide@vger.kernel.org
+Description:
+		(RW) Used by drivers which support software controlled activity
+		LEDs.
+
+		It has the following valid values:
+
+		==	========================================================
+		0	OFF - the LED is not activated on activity
+		1	BLINK_ON - the LED blinks on every 10ms when activity is
+			detected.
+		2	BLINK_OFF - the LED is on when idle, and blinks off
+			every 10ms when activity is detected.
+		==	========================================================
+
+		Note that the user must turn sw_activity OFF it they wish to
+		control the activity LED via the em_message file.
+
+
+What:		/sys/block/*/device/unload_heads
+Date:		Sep, 2008
+KernelVersion:	v2.6.28
+Contact:	linux-ide@vger.kernel.org
+Description:
+		(RW) Hard disk shock protection
+
+		Writing an integer value to this file will take the heads of the
+		respective drive off the platter and block all I/O operations
+		for the specified number of milliseconds.
+
+		- If the device does not support the unload heads feature,
+		  access is denied with -EOPNOTSUPP.
+		- The maximal value accepted for a timeout is 30000
+		  milliseconds.
+		- A previously set timeout can be cancelled and disk can resume
+		  normal operation immediately by specifying a timeout of 0.
+		- Some hard drives only comply with an earlier version of the
+		  ATA standard, but support the unload feature nonetheless.
+		  There is no safe way Linux can detect these devices, so this
+		  is not enabled by default. If it is known that your device
+		  does support the unload feature, then you can tell the kernel
+		  to enable it by writing -1. It can be disabled again by
+		  writing -2.
+		- Values below -2 are rejected with -EINVAL
+
+		For more information, see
+		Documentation/admin-guide/laptops/disk-shock-protection.rst
+
+
+What:		/sys/block/*/device/ncq_prio_enable
+Date:		Oct, 2016
+KernelVersion:	v4.10
+Contact:	linux-ide@vger.kernel.org
+Description:
+		(RW) Write to the file to turn on or off the SATA NCQ (native
+		command queueing) priority support. By default this feature is
+		turned off. If the device does not support the SATA NCQ
+		priority feature, writing "1" to this file results in an error
+		(see ncq_prio_supported).
+
+
+What:		/sys/block/*/device/sas_ncq_prio_enable
+Date:		Oct, 2016
+KernelVersion:	v4.10
+Contact:	linux-ide@vger.kernel.org
+Description:
+		(RW) This is the equivalent of the ncq_prio_enable attribute
+		file for SATA devices connected to a SAS host-bus-adapter
+		(HBA) implementing support for the SATA NCQ priority feature.
+		This file does not exist if the HBA driver does not implement
+		support for the SATA NCQ priority feature, regardless of the
+		device support for this feature (see sas_ncq_prio_supported).
+
+
+What:		/sys/block/*/device/ncq_prio_supported
+Date:		Aug, 2021
+KernelVersion:	v5.15
+Contact:	linux-ide@vger.kernel.org
+Description:
+		(RO) Indicates if the device supports the SATA NCQ (native
+		command queueing) priority feature.
+
+
+What:		/sys/block/*/device/sas_ncq_prio_supported
+Date:		Aug, 2021
+KernelVersion:	v5.15
+Contact:	linux-ide@vger.kernel.org
+Description:
+		(RO) This is the equivalent of the ncq_prio_supported attribute
+		file for SATA devices connected to a SAS host-bus-adapter
+		(HBA) implementing support for the SATA NCQ priority feature.
+		This file does not exist if the HBA driver does not implement
+		support for the SATA NCQ priority feature, regardless of the
+		device support for this feature.
+
+
+What:		/sys/block/*/device/cdl_supported
+Date:		May, 2023
+KernelVersion:	v6.5
+Contact:	linux-scsi@vger.kernel.org
+Description:
+		(RO) Indicates if the device supports the command duration
+		limits feature found in some ATA and SCSI devices.
+
+
+What:		/sys/block/*/device/cdl_enable
+Date:		May, 2023
+KernelVersion:	v6.5
+Contact:	linux-scsi@vger.kernel.org
+Description:
+		(RW) For a device supporting the command duration limits
+		feature, write to the file to turn on or off the feature.
+		By default this feature is turned off.
+		Writing "1" to this file enables the use of command duration
+		limits for read and write commands in the kernel and turns on
+		the feature on the device. Writing "0" disables the feature.