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, 133 insertions, 0 deletions

diff --git a/arch/alpha/lib/memset.S b/arch/alpha/lib/memset.S
new file mode 100644
index 000000000..9075d6918
--- /dev/null
+++ b/arch/alpha/lib/memset.S
@@ -0,0 +1,133 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * linux/arch/alpha/lib/memset.S
+ *
+ * This is an efficient (and small) implementation of the C library "memset()"
+ * function for the alpha.
+ *
+ *	(C) Copyright 1996 Linus Torvalds
+ *
+ * This routine is "moral-ware": you are free to use it any way you wish, and
+ * the only obligation I put on you is a moral one: if you make any improvements
+ * to the routine, please send me your improvements for me to use similarly.
+ *
+ * The scheduling comments are according to the EV5 documentation (and done by 
+ * hand, so they might well be incorrect, please do tell me about it..)
+ */
+#include <linux/export.h>
+	.set noat
+	.set noreorder
+.text
+	.globl memset
+	.globl __memset
+	.globl ___memset
+	.globl __memset16
+	.globl __constant_c_memset
+
+	.ent ___memset
+.align 5
+___memset:
+	.frame $30,0,$26,0
+	.prologue 0
+
+	and $17,255,$1		/* E1 */
+	insbl $17,1,$17		/* .. E0 */
+	bis $17,$1,$17		/* E0 (p-c latency, next cycle) */
+	sll $17,16,$1		/* E1 (p-c latency, next cycle) */
+
+	bis $17,$1,$17		/* E0 (p-c latency, next cycle) */
+	sll $17,32,$1		/* E1 (p-c latency, next cycle) */
+	bis $17,$1,$17		/* E0 (p-c latency, next cycle) */
+	ldq_u $31,0($30)	/* .. E1 */
+
+.align 5
+__constant_c_memset:
+	addq $18,$16,$6		/* E0 */
+	bis $16,$16,$0		/* .. E1 */
+	xor $16,$6,$1		/* E0 */
+	ble $18,end		/* .. E1 */
+
+	bic $1,7,$1		/* E0 */
+	beq $1,within_one_quad	/* .. E1 (note EV5 zero-latency forwarding) */
+	and $16,7,$3		/* E0 */
+	beq $3,aligned		/* .. E1 (note EV5 zero-latency forwarding) */
+
+	ldq_u $4,0($16)		/* E0 */
+	bis $16,$16,$5		/* .. E1 */
+	insql $17,$16,$2	/* E0 */
+	subq $3,8,$3		/* .. E1 */
+
+	addq $18,$3,$18		/* E0	$18 is new count ($3 is negative) */
+	mskql $4,$16,$4		/* .. E1 (and possible load stall) */
+	subq $16,$3,$16		/* E0 	$16 is new aligned destination */
+	bis $2,$4,$1		/* .. E1 */
+
+	bis $31,$31,$31		/* E0 */
+	ldq_u $31,0($30)	/* .. E1 */
+	stq_u $1,0($5)		/* E0 */
+	bis $31,$31,$31		/* .. E1 */
+
+.align 4
+aligned:
+	sra $18,3,$3		/* E0 */
+	and $18,7,$18		/* .. E1 */
+	bis $16,$16,$5		/* E0 */
+	beq $3,no_quad		/* .. E1 */
+
+.align 3
+loop:
+	stq $17,0($5)		/* E0 */
+	subq $3,1,$3		/* .. E1 */
+	addq $5,8,$5		/* E0 */
+	bne $3,loop		/* .. E1 */
+
+no_quad:
+	bis $31,$31,$31		/* E0 */
+	beq $18,end		/* .. E1 */
+	ldq $7,0($5)		/* E0 */
+	mskqh $7,$6,$2		/* .. E1 (and load stall) */
+
+	insqh $17,$6,$4		/* E0 */
+	bis $2,$4,$1		/* .. E1 */
+	stq $1,0($5)		/* E0 */
+	ret $31,($26),1		/* .. E1 */
+
+.align 3
+within_one_quad:
+	ldq_u $1,0($16)		/* E0 */
+	insql $17,$16,$2	/* E1 */
+	mskql $1,$16,$4		/* E0 (after load stall) */
+	bis $2,$4,$2		/* E0 */
+
+	mskql $2,$6,$4		/* E0 */
+	mskqh $1,$6,$2		/* .. E1 */
+	bis $2,$4,$1		/* E0 */
+	stq_u $1,0($16)		/* E0 */
+
+end:
+	ret $31,($26),1		/* E1 */
+	.end ___memset
+EXPORT_SYMBOL(___memset)
+EXPORT_SYMBOL(__constant_c_memset)
+
+	.align 5
+	.ent __memset16
+__memset16:
+	.prologue 0
+
+	inswl $17,0,$1		/* E0 */
+	inswl $17,2,$2		/* E0 */
+	inswl $17,4,$3		/* E0 */
+	or $1,$2,$1		/* .. E1 */
+	inswl $17,6,$4		/* E0 */
+	or $1,$3,$1		/* .. E1 */
+	or $1,$4,$17		/* E0 */
+	br __constant_c_memset	/* .. E1 */
+
+	.end __memset16
+EXPORT_SYMBOL(__memset16)
+
+memset = ___memset
+__memset = ___memset
+	EXPORT_SYMBOL(memset)
+	EXPORT_SYMBOL(__memset)