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, 45 insertions, 0 deletions

diff --git a/Documentation/ABI/testing/debugfs-wilco-ec b/Documentation/ABI/testing/debugfs-wilco-ec
new file mode 100644
index 000000000..682e3c09e
--- /dev/null
+++ b/Documentation/ABI/testing/debugfs-wilco-ec
@@ -0,0 +1,45 @@
+What:		/sys/kernel/debug/wilco_ec/h1_gpio
+Date:		April 2019
+KernelVersion:	5.2
+Description:
+		As part of Chrome OS's FAFT (Fully Automated Firmware Testing)
+		tests, we need to ensure that the H1 chip is properly setting
+		some GPIO lines. The h1_gpio attribute exposes the state
+		of the lines:
+		- ENTRY_TO_FACT_MODE in BIT(0)
+		- SPI_CHROME_SEL in BIT(1)
+
+		Output will formatted with "0x%02x\n".
+
+What:		/sys/kernel/debug/wilco_ec/raw
+Date:		January 2019
+KernelVersion:	5.1
+Description:
+		Write and read raw mailbox commands to the EC.
+
+		You can write a hexadecimal sentence to raw, and that series of
+		bytes will be sent to the EC. Then, you can read the bytes of
+		response by reading from raw.
+
+		For writing, bytes 0-1 indicate the message type, one of enum
+		wilco_ec_msg_type. Byte 2+ consist of the data passed in the
+		request, starting at MBOX[0]. At least three bytes are required
+		for writing, two for the type and at least a single byte of
+		data.
+
+		Example::
+
+		    // Request EC info type 3 (EC firmware build date)
+		    // Corresponds with sending type 0x00f0 with
+		    // MBOX = [38, 00, 03, 00]
+		    $ echo 00 f0 38 00 03 00 > /sys/kernel/debug/wilco_ec/raw
+		    // View the result. The decoded ASCII result "12/21/18" is
+		    // included after the raw hex.
+		    // Corresponds with MBOX = [00, 00, 31, 32, 2f, 32, 31, 38, ...]
+		    $ cat /sys/kernel/debug/wilco_ec/raw
+		    00 00 31 32 2f 32 31 2f 31 38 00 38 00 01 00 2f 00  ..12/21/18.8...
+
+		Note that the first 16 bytes of the received MBOX[] will be
+		printed, even if some of the data is junk, and skipping bytes
+		17 to 32. It is up to you to know how many of the first bytes of
+		data are the actual response.