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, 59 insertions, 0 deletions

diff --git a/arch/alpha/lib/strlen.S b/arch/alpha/lib/strlen.S
new file mode 100644
index 000000000..4fc6a6ff2
--- /dev/null
+++ b/arch/alpha/lib/strlen.S
@@ -0,0 +1,59 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * strlen.S (c) 1995 David Mosberger (davidm@cs.arizona.edu)
+ *
+ * Finds length of a 0-terminated string.  Optimized for the
+ * Alpha architecture:
+ *
+ *	- memory accessed as aligned quadwords only
+ *	- uses bcmpge to compare 8 bytes in parallel
+ *	- does binary search to find 0 byte in last
+ *	  quadword (HAKMEM needed 12 instructions to
+ *	  do this instead of the 9 instructions that
+ *	  binary search needs).
+ */
+#include <linux/export.h>
+	.set noreorder
+	.set noat
+
+	.align 3
+
+	.globl	strlen
+	.ent	strlen
+
+strlen:
+	ldq_u	$1, 0($16)	# load first quadword ($16  may be misaligned)
+	lda	$2, -1($31)
+	insqh	$2, $16, $2
+	andnot	$16, 7, $0
+	or	$2, $1, $1
+	cmpbge	$31, $1, $2	# $2  <- bitmask: bit i == 1 <==> i-th byte == 0
+	bne	$2, found
+
+loop:	ldq	$1, 8($0)
+	addq	$0, 8, $0	# addr += 8
+	nop			# helps dual issue last two insns
+	cmpbge	$31, $1, $2
+	beq	$2, loop
+
+found:	blbs	$2, done	# make aligned case fast
+	negq	$2, $3
+	and	$2, $3, $2
+
+	and	$2, 0x0f, $1
+	addq	$0, 4, $3
+	cmoveq	$1, $3, $0
+
+	and	$2, 0x33, $1
+	addq	$0, 2, $3
+	cmoveq	$1, $3, $0
+
+	and	$2, 0x55, $1
+	addq	$0, 1, $3
+	cmoveq	$1, $3, $0
+
+done:	subq	$0, $16, $0
+	ret	$31, ($26)
+
+	.end	strlen
+	EXPORT_SYMBOL(strlen)