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, 139 insertions, 0 deletions

diff --git a/arch/arm/lib/findbit.S b/arch/arm/lib/findbit.S
new file mode 100644
index 000000000..b7ac2d3c0
--- /dev/null
+++ b/arch/arm/lib/findbit.S
@@ -0,0 +1,139 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ *  linux/arch/arm/lib/findbit.S
+ *
+ *  Copyright (C) 1995-2000 Russell King
+ *
+ * 16th March 2001 - John Ripley <jripley@sonicblue.com>
+ *   Fixed so that "size" is an exclusive not an inclusive quantity.
+ *   All users of these functions expect exclusive sizes, and may
+ *   also call with zero size.
+ * Reworked by rmk.
+ */
+#include <linux/linkage.h>
+#include <asm/assembler.h>
+#include <asm/unwind.h>
+                .text
+
+#ifdef __ARMEB__
+#define SWAB_ENDIAN le
+#else
+#define SWAB_ENDIAN be
+#endif
+
+		.macro	find_first, endian, set, name
+ENTRY(_find_first_\name\()bit_\endian)
+	UNWIND(	.fnstart)
+		teq	r1, #0
+		beq	3f
+		mov	r2, #0
+1:		ldr	r3, [r0], #4
+		.ifeq \set
+		mvns	r3, r3			@ invert/test bits
+		.else
+		movs	r3, r3			@ test bits
+		.endif
+		.ifc \endian, SWAB_ENDIAN
+		bne	.L_found_swab
+		.else
+		bne	.L_found		@ found the bit?
+		.endif
+		add	r2, r2, #32		@ next index
+2:		cmp	r2, r1			@ any more?
+		blo	1b
+3:		mov	r0, r1			@ no more bits
+		ret	lr
+	UNWIND(	.fnend)
+ENDPROC(_find_first_\name\()bit_\endian)
+		.endm
+
+		.macro	find_next, endian, set, name
+ENTRY(_find_next_\name\()bit_\endian)
+	UNWIND(	.fnstart)
+		cmp	r2, r1
+		bhs	3b
+		mov	ip, r2, lsr #5		@ word index
+		add	r0, r0, ip, lsl #2
+		ands	ip, r2, #31		@ bit position
+		beq	1b
+		ldr	r3, [r0], #4
+		.ifeq \set
+		mvn	r3, r3			@ invert bits
+		.endif
+		.ifc \endian, SWAB_ENDIAN
+		rev_l	r3, ip
+		.if	.Lrev_l_uses_tmp
+		@ we need to recompute ip because rev_l will have overwritten
+		@ it.
+		and	ip, r2, #31		@ bit position
+		.endif
+		.endif
+		movs	r3, r3, lsr ip		@ shift off unused bits
+		bne	.L_found
+		orr	r2, r2, #31		@ no zero bits
+		add	r2, r2, #1		@ align bit pointer
+		b	2b			@ loop for next bit
+	UNWIND(	.fnend)
+ENDPROC(_find_next_\name\()bit_\endian)
+		.endm
+
+		.macro	find_bit, endian, set, name
+		find_first \endian, \set, \name
+		find_next  \endian, \set, \name
+		.endm
+
+/* _find_first_zero_bit_le and _find_next_zero_bit_le */
+		find_bit le, 0, zero_
+
+/* _find_first_bit_le and _find_next_bit_le */
+		find_bit le, 1
+
+#ifdef __ARMEB__
+
+/* _find_first_zero_bit_be and _find_next_zero_bit_be */
+		find_bit be, 0, zero_
+
+/* _find_first_bit_be and _find_next_bit_be */
+		find_bit be, 1
+
+#endif
+
+/*
+ * One or more bits in the LSB of r3 are assumed to be set.
+ */
+.L_found_swab:
+	UNWIND(	.fnstart)
+		rev_l	r3, ip
+.L_found:
+#if __LINUX_ARM_ARCH__ >= 7
+		rbit	r3, r3			@ reverse bits
+		clz	r3, r3			@ count high zero bits
+		add	r0, r2, r3		@ add offset of first set bit
+#elif __LINUX_ARM_ARCH__ >= 5
+		rsb	r0, r3, #0
+		and	r3, r3, r0		@ mask out lowest bit set
+		clz	r3, r3			@ count high zero bits
+		rsb	r3, r3, #31		@ offset of first set bit
+		add	r0, r2, r3		@ add offset of first set bit
+#else
+		mov	ip, #~0
+		tst	r3, ip, lsr #16		@ test bits 0-15
+		addeq	r2, r2, #16
+		moveq	r3, r3, lsr #16
+		tst	r3, #0x00ff
+		addeq	r2, r2, #8
+		moveq	r3, r3, lsr #8
+		tst	r3, #0x000f
+		addeq	r2, r2, #4
+		moveq	r3, r3, lsr #4
+		tst	r3, #0x0003
+		addeq	r2, r2, #2
+		moveq	r3, r3, lsr #2
+		tst	r3, #0x0001
+		addeq	r2, r2, #1
+		mov	r0, r2
+#endif
+		cmp	r1, r0			@ Clamp to maxbit
+		movlo	r0, r1
+		ret	lr
+	UNWIND(	.fnend)