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, 62 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/sysfs-devices-mapping b/Documentation/ABI/testing/sysfs-devices-mapping
new file mode 100644
index 000000000..2eee1446a
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-devices-mapping
@@ -0,0 +1,62 @@
+What:           /sys/devices/uncore_iio_x/dieX
+Date:           February 2020
+Contact:        Alexander Antonov <alexander.antonov@linux.intel.com>
+Description:
+                Each IIO stack (PCIe root port) has its own IIO PMON block, so
+                each dieX file (where X is die number) holds "Segment:Root Bus"
+                for PCIe root port, which can be monitored by that IIO PMON
+                block.
+                For example, on 4-die Xeon platform with up to 6 IIO stacks per
+                die and, therefore, 6 IIO PMON blocks per die, the mapping of
+                IIO PMON block 0 exposes as the following::
+
+		    $ ls /sys/devices/uncore_iio_0/die*
+		    -r--r--r-- /sys/devices/uncore_iio_0/die0
+		    -r--r--r-- /sys/devices/uncore_iio_0/die1
+		    -r--r--r-- /sys/devices/uncore_iio_0/die2
+		    -r--r--r-- /sys/devices/uncore_iio_0/die3
+
+		    $ tail /sys/devices/uncore_iio_0/die*
+		    ==> /sys/devices/uncore_iio_0/die0 <==
+		    0000:00
+		    ==> /sys/devices/uncore_iio_0/die1 <==
+		    0000:40
+		    ==> /sys/devices/uncore_iio_0/die2 <==
+		    0000:80
+		    ==> /sys/devices/uncore_iio_0/die3 <==
+		    0000:c0
+
+                Which means::
+
+		    IIO PMU 0 on die 0 belongs to PCI RP on bus 0x00, domain 0x0000
+		    IIO PMU 0 on die 1 belongs to PCI RP on bus 0x40, domain 0x0000
+		    IIO PMU 0 on die 2 belongs to PCI RP on bus 0x80, domain 0x0000
+		    IIO PMU 0 on die 3 belongs to PCI RP on bus 0xc0, domain 0x0000
+
+What:           /sys/devices/uncore_upi_x/dieX
+Date:           March 2022
+Contact:        Alexander Antonov <alexander.antonov@linux.intel.com>
+Description:
+                Each /sys/devices/uncore_upi_X/dieY file holds "upi_Z,die_W"
+                value that means UPI link number X on die Y is connected to UPI
+                link Z on die W and this link between sockets can be monitored
+                by UPI PMON block.
+                For example, 4-die Sapphire Rapids platform has the following
+                UPI 0 topology::
+
+		    # tail /sys/devices/uncore_upi_0/die*
+		    ==> /sys/devices/uncore_upi_0/die0 <==
+		    upi_1,die_1
+		    ==> /sys/devices/uncore_upi_0/die1 <==
+		    upi_0,die_3
+		    ==> /sys/devices/uncore_upi_0/die2 <==
+		    upi_1,die_3
+		    ==> /sys/devices/uncore_upi_0/die3 <==
+		    upi_0,die_1
+
+                Which means::
+
+		    UPI link 0 on die 0 is connected to UPI link 1 on die 1
+		    UPI link 0 on die 1 is connected to UPI link 0 on die 3
+		    UPI link 0 on die 2 is connected to UPI link 1 on die 3
+		    UPI link 0 on die 3 is connected to UPI link 0 on die 1
\ No newline at end of file