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, 183 insertions, 0 deletions

diff --git a/arch/alpha/lib/memmove.S b/arch/alpha/lib/memmove.S
new file mode 100644
index 000000000..3a27689e3
--- /dev/null
+++ b/arch/alpha/lib/memmove.S
@@ -0,0 +1,183 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * arch/alpha/lib/memmove.S
+ *
+ * Barely optimized memmove routine for Alpha EV5.
+ *
+ * This is hand-massaged output from the original memcpy.c.  We defer to
+ * memcpy whenever possible; the backwards copy loops are not unrolled.
+ */
+#include <linux/export.h>
+	.set noat
+	.set noreorder
+	.text
+
+	.align 4
+	.globl memmove
+	.ent memmove
+memmove:
+	ldgp $29, 0($27)
+	unop
+	nop
+	.prologue 1
+
+	addq $16,$18,$4
+	addq $17,$18,$5
+	cmpule $4,$17,$1		/*  dest + n <= src  */
+	cmpule $5,$16,$2		/*  dest >= src + n  */
+
+	bis $1,$2,$1
+	mov $16,$0
+	xor $16,$17,$2
+	bne $1,memcpy			!samegp
+
+	and $2,7,$2			/* Test for src/dest co-alignment.  */
+	and $16,7,$1
+	cmpule $16,$17,$3
+	bne $3,$memmove_up		/* dest < src */
+
+	and $4,7,$1
+	bne $2,$misaligned_dn
+	unop
+	beq $1,$skip_aligned_byte_loop_head_dn
+
+$aligned_byte_loop_head_dn:
+	lda $4,-1($4)
+	lda $5,-1($5)
+	unop
+	ble $18,$egress
+
+	ldq_u $3,0($5)
+	ldq_u $2,0($4)
+	lda $18,-1($18)
+	extbl $3,$5,$1
+
+	insbl $1,$4,$1
+	mskbl $2,$4,$2
+	bis $1,$2,$1
+	and $4,7,$6
+
+	stq_u $1,0($4)
+	bne $6,$aligned_byte_loop_head_dn
+
+$skip_aligned_byte_loop_head_dn:
+	lda $18,-8($18)
+	blt $18,$skip_aligned_word_loop_dn
+
+$aligned_word_loop_dn:
+	ldq $1,-8($5)
+	nop
+	lda $5,-8($5)
+	lda $18,-8($18)
+
+	stq $1,-8($4)
+	nop
+	lda $4,-8($4)
+	bge $18,$aligned_word_loop_dn
+
+$skip_aligned_word_loop_dn:
+	lda $18,8($18)
+	bgt $18,$byte_loop_tail_dn
+	unop
+	ret $31,($26),1
+
+	.align 4
+$misaligned_dn:
+	nop
+	fnop
+	unop
+	beq $18,$egress
+
+$byte_loop_tail_dn:
+	ldq_u $3,-1($5)
+	ldq_u $2,-1($4)
+	lda $5,-1($5)
+	lda $4,-1($4)
+
+	lda $18,-1($18)
+	extbl $3,$5,$1
+	insbl $1,$4,$1
+	mskbl $2,$4,$2
+
+	bis $1,$2,$1
+	stq_u $1,0($4)
+	bgt $18,$byte_loop_tail_dn
+	br $egress
+
+$memmove_up:
+	mov $16,$4
+	mov $17,$5
+	bne $2,$misaligned_up
+	beq $1,$skip_aligned_byte_loop_head_up
+
+$aligned_byte_loop_head_up:
+	unop
+	ble $18,$egress
+	ldq_u $3,0($5)
+	ldq_u $2,0($4)
+
+	lda $18,-1($18)
+	extbl $3,$5,$1
+	insbl $1,$4,$1
+	mskbl $2,$4,$2
+
+	bis $1,$2,$1
+	lda $5,1($5)
+	stq_u $1,0($4)
+	lda $4,1($4)
+
+	and $4,7,$6
+	bne $6,$aligned_byte_loop_head_up
+
+$skip_aligned_byte_loop_head_up:
+	lda $18,-8($18)
+	blt $18,$skip_aligned_word_loop_up
+
+$aligned_word_loop_up:
+	ldq $1,0($5)
+	nop
+	lda $5,8($5)
+	lda $18,-8($18)
+
+	stq $1,0($4)
+	nop
+	lda $4,8($4)
+	bge $18,$aligned_word_loop_up
+
+$skip_aligned_word_loop_up:
+	lda $18,8($18)
+	bgt $18,$byte_loop_tail_up
+	unop
+	ret $31,($26),1
+
+	.align 4
+$misaligned_up:
+	nop
+	fnop
+	unop
+	beq $18,$egress
+
+$byte_loop_tail_up:
+	ldq_u $3,0($5)
+	ldq_u $2,0($4)
+	lda $18,-1($18)
+	extbl $3,$5,$1
+
+	insbl $1,$4,$1
+	mskbl $2,$4,$2
+	bis $1,$2,$1
+	stq_u $1,0($4)
+
+	lda $5,1($5)
+	lda $4,1($4)
+	nop
+	bgt $18,$byte_loop_tail_up
+
+$egress:
+	ret $31,($26),1
+	nop
+	nop
+	nop
+
+	.end memmove
+	EXPORT_SYMBOL(memmove)