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, 148 insertions, 0 deletions

diff --git a/arch/arm/lib/memset.S b/arch/arm/lib/memset.S
new file mode 100644
index 000000000..de75ae4d5
--- /dev/null
+++ b/arch/arm/lib/memset.S
@@ -0,0 +1,148 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ *  linux/arch/arm/lib/memset.S
+ *
+ *  Copyright (C) 1995-2000 Russell King
+ *
+ *  ASM optimised string functions
+ */
+#include <linux/linkage.h>
+#include <asm/assembler.h>
+#include <asm/unwind.h>
+
+	.text
+	.align	5
+
+ENTRY(__memset)
+ENTRY(mmioset)
+WEAK(memset)
+UNWIND( .fnstart         )
+	and	r1, r1, #255		@ cast to unsigned char
+	ands	r3, r0, #3		@ 1 unaligned?
+	mov	ip, r0			@ preserve r0 as return value
+	bne	6f			@ 1
+/*
+ * we know that the pointer in ip is aligned to a word boundary.
+ */
+1:	orr	r1, r1, r1, lsl #8
+	orr	r1, r1, r1, lsl #16
+	mov	r3, r1
+7:	cmp	r2, #16
+	blt	4f
+UNWIND( .fnend              )
+
+#if ! CALGN(1)+0
+
+/*
+ * We need 2 extra registers for this loop - use r8 and the LR
+ */
+UNWIND( .fnstart            )
+UNWIND( .save {r8, lr}      )
+	stmfd	sp!, {r8, lr}
+	mov	r8, r1
+	mov	lr, r3
+
+2:	subs	r2, r2, #64
+	stmiage	ip!, {r1, r3, r8, lr}	@ 64 bytes at a time.
+	stmiage	ip!, {r1, r3, r8, lr}
+	stmiage	ip!, {r1, r3, r8, lr}
+	stmiage	ip!, {r1, r3, r8, lr}
+	bgt	2b
+	ldmfdeq	sp!, {r8, pc}		@ Now <64 bytes to go.
+/*
+ * No need to correct the count; we're only testing bits from now on
+ */
+	tst	r2, #32
+	stmiane	ip!, {r1, r3, r8, lr}
+	stmiane	ip!, {r1, r3, r8, lr}
+	tst	r2, #16
+	stmiane	ip!, {r1, r3, r8, lr}
+	ldmfd	sp!, {r8, lr}
+UNWIND( .fnend              )
+
+#else
+
+/*
+ * This version aligns the destination pointer in order to write
+ * whole cache lines at once.
+ */
+
+UNWIND( .fnstart               )
+UNWIND( .save {r4-r8, lr}      )
+	stmfd	sp!, {r4-r8, lr}
+	mov	r4, r1
+	mov	r5, r3
+	mov	r6, r1
+	mov	r7, r3
+	mov	r8, r1
+	mov	lr, r3
+
+	cmp	r2, #96
+	tstgt	ip, #31
+	ble	3f
+
+	and	r8, ip, #31
+	rsb	r8, r8, #32
+	sub	r2, r2, r8
+	movs	r8, r8, lsl #(32 - 4)
+	stmiacs	ip!, {r4, r5, r6, r7}
+	stmiami	ip!, {r4, r5}
+	tst	r8, #(1 << 30)
+	mov	r8, r1
+	strne	r1, [ip], #4
+
+3:	subs	r2, r2, #64
+	stmiage	ip!, {r1, r3-r8, lr}
+	stmiage	ip!, {r1, r3-r8, lr}
+	bgt	3b
+	ldmfdeq	sp!, {r4-r8, pc}
+
+	tst	r2, #32
+	stmiane	ip!, {r1, r3-r8, lr}
+	tst	r2, #16
+	stmiane	ip!, {r4-r7}
+	ldmfd	sp!, {r4-r8, lr}
+UNWIND( .fnend                 )
+
+#endif
+
+UNWIND( .fnstart            )
+4:	tst	r2, #8
+	stmiane	ip!, {r1, r3}
+	tst	r2, #4
+	strne	r1, [ip], #4
+/*
+ * When we get here, we've got less than 4 bytes to set.  We
+ * may have an unaligned pointer as well.
+ */
+5:	tst	r2, #2
+	strbne	r1, [ip], #1
+	strbne	r1, [ip], #1
+	tst	r2, #1
+	strbne	r1, [ip], #1
+	ret	lr
+
+6:	subs	r2, r2, #4		@ 1 do we have enough
+	blt	5b			@ 1 bytes to align with?
+	cmp	r3, #2			@ 1
+	strblt	r1, [ip], #1		@ 1
+	strble	r1, [ip], #1		@ 1
+	strb	r1, [ip], #1		@ 1
+	add	r2, r2, r3		@ 1 (r2 = r2 - (4 - r3))
+	b	1b
+UNWIND( .fnend   )
+ENDPROC(memset)
+ENDPROC(mmioset)
+ENDPROC(__memset)
+
+ENTRY(__memset32)
+UNWIND( .fnstart         )
+	mov	r3, r1			@ copy r1 to r3 and fall into memset64
+UNWIND( .fnend   )
+ENDPROC(__memset32)
+ENTRY(__memset64)
+UNWIND( .fnstart         )
+	mov	ip, r0			@ preserve r0 as return value
+	b	7b			@ jump into the middle of memset
+UNWIND( .fnend   )
+ENDPROC(__memset64)