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, 72 insertions, 0 deletions

diff --git a/arch/alpha/lib/strchr.S b/arch/alpha/lib/strchr.S
new file mode 100644
index 000000000..68c54ff50
--- /dev/null
+++ b/arch/alpha/lib/strchr.S
@@ -0,0 +1,72 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * arch/alpha/lib/strchr.S
+ * Contributed by Richard Henderson (rth@tamu.edu)
+ *
+ * Return the address of a given character within a null-terminated
+ * string, or null if it is not found.
+ */
+#include <linux/export.h>
+#include <asm/regdef.h>
+
+	.set noreorder
+	.set noat
+
+	.align 3
+	.globl strchr
+	.ent strchr
+strchr:
+	.frame sp, 0, ra
+	.prologue 0
+
+	zapnot	a1, 1, a1	# e0    : zero extend the search character
+	ldq_u   t0, 0(a0)	# .. e1 : load first quadword
+	sll	a1, 8, t5	# e0    : replicate the search character
+	andnot  a0, 7, v0	# .. e1 : align our loop pointer
+	or	t5, a1, a1	# e0    :
+	lda	t4, -1		# .. e1 : build garbage mask
+	sll	a1, 16, t5	# e0    :
+	cmpbge  zero, t0, t2	# .. e1 : bits set iff byte == zero
+	mskqh	t4, a0, t4	# e0    :
+	or	t5, a1, a1	# .. e1 :
+	sll	a1, 32, t5	# e0    :
+	cmpbge	zero, t4, t4	# .. e1 : bits set iff byte is garbage
+	or	t5, a1, a1	# e0    :
+	xor	t0, a1, t1	# .. e1 : make bytes == c zero
+	cmpbge  zero, t1, t3	# e0    : bits set iff byte == c
+	or	t2, t3, t0	# e1    : bits set iff char match or zero match
+	andnot	t0, t4, t0	# e0    : clear garbage bits
+	bne	t0, $found	# .. e1 (zdb)
+
+$loop:	ldq	t0, 8(v0)	# e0    :
+	addq	v0, 8, v0	# .. e1 :
+	nop			# e0    :
+	xor	t0, a1, t1	# .. e1 (ev5 data stall)
+	cmpbge	zero, t0, t2	# e0    : bits set iff byte == 0
+	cmpbge	zero, t1, t3	# .. e1 : bits set iff byte == c
+	or	t2, t3, t0	# e0    :
+	beq	t0, $loop	# .. e1 (zdb)
+
+$found:	negq    t0, t1		# e0    : clear all but least set bit
+	and     t0, t1, t0	# e1 (stall)
+
+	and	t0, t3, t1	# e0    : bit set iff byte was the char
+	beq	t1, $retnull	# .. e1 (zdb)
+
+	and     t0, 0xf0, t2	# e0    : binary search for that set bit
+	and	t0, 0xcc, t3	# .. e1 :
+	and	t0, 0xaa, t4	# e0    :
+	cmovne	t2, 4, t2	# .. e1 :
+	cmovne	t3, 2, t3	# e0    :
+	cmovne	t4, 1, t4	# .. e1 :
+	addq	t2, t3, t2	# e0    :
+	addq	v0, t4, v0	# .. e1 :
+	addq	v0, t2, v0	# e0    :
+	ret			# .. e1 :
+
+$retnull:
+	mov	zero, v0	# e0    :
+	ret			# .. e1 :
+
+	.end strchr
+	EXPORT_SYMBOL(strchr)