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, 192 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-class-led-trigger-netdev b/Documentation/ABI/testing/sysfs-class-led-trigger-netdev
new file mode 100644
index 000000000..ed46b37ab
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-class-led-trigger-netdev
@@ -0,0 +1,192 @@
+What:		/sys/class/leds/<led>/device_name
+Date:		Dec 2017
+KernelVersion:	4.16
+Contact:	linux-leds@vger.kernel.org
+Description:
+		Specifies the network device name to monitor.
+
+What:		/sys/class/leds/<led>/interval
+Date:		Dec 2017
+KernelVersion:	4.16
+Contact:	linux-leds@vger.kernel.org
+Description:
+		Specifies the duration of the LED blink in milliseconds.
+		Defaults to 50 ms.
+
+		When offloaded is true, the interval value MUST be set to the
+		default value and cannot be changed.
+		Trying to set any value in this specific mode will return
+		an EINVAL error.
+
+What:		/sys/class/leds/<led>/link
+Date:		Dec 2017
+KernelVersion:	4.16
+Contact:	linux-leds@vger.kernel.org
+Description:
+		Signal the link state of the named network device.
+
+		If set to 0 (default), the LED's normal state is off.
+
+		If set to 1, the LED's normal state reflects the link state
+		of the named network device.
+		Setting this value also immediately changes the LED state.
+
+
+What:		/sys/class/leds/<led>/tx
+Date:		Dec 2017
+KernelVersion:	4.16
+Contact:	linux-leds@vger.kernel.org
+Description:
+		Signal transmission of data on the named network device.
+
+		If set to 0 (default), the LED will not blink on transmission.
+
+		If set to 1, the LED will blink for the milliseconds specified
+		in interval to signal transmission.
+
+		When offloaded is true, the blink interval is controlled by
+		hardware and won't reflect the value set in interval.
+
+What:		/sys/class/leds/<led>/rx
+Date:		Dec 2017
+KernelVersion:	4.16
+Contact:	linux-leds@vger.kernel.org
+Description:
+		Signal reception of data on the named network device.
+
+		If set to 0 (default), the LED will not blink on reception.
+
+		If set to 1, the LED will blink for the milliseconds specified
+		in interval to signal reception.
+
+		When offloaded is true, the blink interval is controlled by
+		hardware and won't reflect the value set in interval.
+
+What:		/sys/class/leds/<led>/offloaded
+Date:		Jun 2023
+KernelVersion:	6.5
+Contact:	linux-leds@vger.kernel.org
+Description:
+		Communicate whether the LED trigger modes are offloaded to
+		hardware or whether software fallback is used.
+
+		If 0, the LED is using software fallback to blink.
+
+		If 1, the LED blinking in requested mode is offloaded to
+		hardware.
+
+What:		/sys/class/leds/<led>/link_10
+Date:		Jun 2023
+KernelVersion:	6.5
+Contact:	linux-leds@vger.kernel.org
+Description:
+		Signal the link speed state of 10Mbps of the named network device.
+
+		If set to 0 (default), the LED's normal state is off.
+
+		If set to 1, the LED's normal state reflects the link state
+		speed of 10MBps of the named network device.
+		Setting this value also immediately changes the LED state.
+
+		Present only if the named network device supports 10Mbps link speed.
+
+What:		/sys/class/leds/<led>/link_100
+Date:		Jun 2023
+KernelVersion:	6.5
+Contact:	linux-leds@vger.kernel.org
+Description:
+		Signal the link speed state of 100Mbps of the named network device.
+
+		If set to 0 (default), the LED's normal state is off.
+
+		If set to 1, the LED's normal state reflects the link state
+		speed of 100Mbps of the named network device.
+		Setting this value also immediately changes the LED state.
+
+		Present only if the named network device supports 100Mbps link speed.
+
+What:		/sys/class/leds/<led>/link_1000
+Date:		Jun 2023
+KernelVersion:	6.5
+Contact:	linux-leds@vger.kernel.org
+Description:
+		Signal the link speed state of 1000Mbps of the named network device.
+
+		If set to 0 (default), the LED's normal state is off.
+
+		If set to 1, the LED's normal state reflects the link state
+		speed of 1000Mbps of the named network device.
+		Setting this value also immediately changes the LED state.
+
+		Present only if the named network device supports 1000Mbps link speed.
+
+What:		/sys/class/leds/<led>/link_2500
+Date:		Nov 2023
+KernelVersion:	6.8
+Contact:	linux-leds@vger.kernel.org
+Description:
+		Signal the link speed state of 2500Mbps of the named network device.
+
+		If set to 0 (default), the LED's normal state is off.
+
+		If set to 1, the LED's normal state reflects the link state
+		speed of 2500Mbps of the named network device.
+		Setting this value also immediately changes the LED state.
+
+		Present only if the named network device supports 2500Mbps link speed.
+
+What:		/sys/class/leds/<led>/link_5000
+Date:		Nov 2023
+KernelVersion:	6.8
+Contact:	linux-leds@vger.kernel.org
+Description:
+		Signal the link speed state of 5000Mbps of the named network device.
+
+		If set to 0 (default), the LED's normal state is off.
+
+		If set to 1, the LED's normal state reflects the link state
+		speed of 5000Mbps of the named network device.
+		Setting this value also immediately changes the LED state.
+
+		Present only if the named network device supports 5000Mbps link speed.
+
+What:		/sys/class/leds/<led>/link_10000
+Date:		Nov 2023
+KernelVersion:	6.8
+Contact:	linux-leds@vger.kernel.org
+Description:
+		Signal the link speed state of 10000Mbps of the named network device.
+
+		If set to 0 (default), the LED's normal state is off.
+
+		If set to 1, the LED's normal state reflects the link state
+		speed of 10000Mbps of the named network device.
+		Setting this value also immediately changes the LED state.
+
+		Present only if the named network device supports 10000Mbps link speed.
+
+What:		/sys/class/leds/<led>/half_duplex
+Date:		Jun 2023
+KernelVersion:	6.5
+Contact:	linux-leds@vger.kernel.org
+Description:
+		Signal the link half duplex state of the named network device.
+
+		If set to 0 (default), the LED's normal state is off.
+
+		If set to 1, the LED's normal state reflects the link half
+		duplex state of the named network device.
+		Setting this value also immediately changes the LED state.
+
+What:		/sys/class/leds/<led>/full_duplex
+Date:		Jun 2023
+KernelVersion:	6.5
+Contact:	linux-leds@vger.kernel.org
+Description:
+		Signal the link full duplex state of the named network device.
+
+		If set to 0 (default), the LED's normal state is off.
+
+		If set to 1, the LED's normal state reflects the link full
+		duplex state of the named network device.
+		Setting this value also immediately changes the LED state.