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, 243 insertions, 0 deletions

diff --git a/drivers/block/swim_asm.S b/drivers/block/swim_asm.S
new file mode 100644
index 000000000..3d7a2d875
--- /dev/null
+++ b/drivers/block/swim_asm.S
@@ -0,0 +1,243 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * low-level functions for the SWIM floppy controller
+ *
+ * needs assembly language because is very timing dependent
+ * this controller exists only on macintosh 680x0 based
+ *
+ * Copyright (C) 2004,2008 Laurent Vivier <Laurent@lvivier.info>
+ *
+ * based on Alastair Bridgewater SWIM analysis, 2001
+ * based on netBSD IWM driver (c) 1997, 1998 Hauke Fath.
+ *
+ * 2004-08-21 (lv) - Initial implementation
+ * 2008-11-05 (lv) - add get_swim_mode
+ */
+
+	.equ	write_data,	0x0000
+	.equ	write_mark,	0x0200
+	.equ	write_CRC,	0x0400
+	.equ	write_parameter,0x0600
+	.equ	write_phase,	0x0800
+	.equ	write_setup,	0x0a00
+	.equ	write_mode0,	0x0c00
+	.equ	write_mode1,	0x0e00
+	.equ	read_data,	0x1000
+	.equ	read_mark,	0x1200
+	.equ	read_error,	0x1400
+	.equ	read_parameter,	0x1600
+	.equ	read_phase,	0x1800
+	.equ	read_setup,	0x1a00
+	.equ	read_status,	0x1c00
+	.equ	read_handshake,	0x1e00
+
+	.equ	o_side, 0
+	.equ	o_track, 1
+	.equ	o_sector, 2
+	.equ	o_size, 3
+	.equ	o_crc0, 4
+	.equ	o_crc1, 5
+
+	.equ	seek_time, 30000
+	.equ	max_retry, 40
+	.equ	sector_size, 512
+
+	.global swim_read_sector_header
+swim_read_sector_header:
+	link	%a6, #0
+	moveml	%d1-%d5/%a0-%a4,%sp@-
+	movel	%a6@(0x0c), %a4
+	bsr	mfm_read_addrmark
+	moveml	%sp@+, %d1-%d5/%a0-%a4
+	unlk	%a6
+	rts
+
+sector_address_mark:
+	.byte	0xa1, 0xa1, 0xa1, 0xfe
+sector_data_mark:
+	.byte	0xa1, 0xa1, 0xa1, 0xfb
+
+mfm_read_addrmark:
+	movel	%a6@(0x08), %a3
+	lea	%a3@(read_handshake), %a2
+	lea	%a3@(read_mark), %a3
+	moveq	#-1, %d0
+	movew	#seek_time, %d2
+
+wait_header_init:
+	tstb	%a3@(read_error - read_mark)
+	moveb	#0x18, %a3@(write_mode0 - read_mark)
+	moveb	#0x01, %a3@(write_mode1 - read_mark)
+	moveb	#0x01, %a3@(write_mode0 - read_mark)
+	tstb	%a3@(read_error - read_mark)
+	moveb	#0x08, %a3@(write_mode1 - read_mark)
+
+	lea	sector_address_mark, %a0
+	moveq	#3, %d1
+
+wait_addr_mark_byte:
+
+	tstb	%a2@
+	dbmi	%d2, wait_addr_mark_byte
+	bpl	header_exit
+
+	moveb	%a3@, %d3
+	cmpb	%a0@+, %d3
+	dbne	%d1, wait_addr_mark_byte
+	bne	wait_header_init
+
+	moveq	#max_retry, %d2
+
+amark0:	tstb	%a2@
+	dbmi	%d2, amark0
+	bpl	signal_nonyb
+
+	moveb	%a3@, %a4@(o_track)
+
+	moveq	#max_retry, %d2
+
+amark1:	tstb	%a2@
+	dbmi	%d2, amark1
+	bpl	signal_nonyb
+
+	moveb	%a3@, %a4@(o_side)
+
+	moveq	#max_retry, %d2
+
+amark2:	tstb	%a2@
+	dbmi	%d2, amark2
+	bpl	signal_nonyb
+
+	moveb	%a3@, %a4@(o_sector)
+
+	moveq	#max_retry, %d2
+
+amark3:	tstb	%a2@
+	dbmi	%d2, amark3
+	bpl	signal_nonyb
+
+	moveb	%a3@, %a4@(o_size)
+
+	moveq	#max_retry, %d2
+
+crc0:	tstb	%a2@
+	dbmi	%d2, crc0
+	bpl	signal_nonyb
+
+	moveb	%a3@, %a4@(o_crc0)
+
+	moveq	#max_retry, %d2
+
+crc1:	tstb	%a2@
+	dbmi	%d2, crc1
+	bpl	signal_nonyb
+
+	moveb	%a3@, %a4@(o_crc1)
+
+	tstb	%a3@(read_error - read_mark)
+
+header_exit:
+	moveq	#0, %d0
+	moveb	#0x18, %a3@(write_mode0 - read_mark)
+	rts
+signal_nonyb:
+	moveq	#-1, %d0
+	moveb	#0x18, %a3@(write_mode0 - read_mark)
+	rts
+
+	.global swim_read_sector_data
+swim_read_sector_data:
+	link	%a6, #0
+	moveml	%d1-%d5/%a0-%a5,%sp@-
+	movel	%a6@(0x0c), %a4
+	bsr	mfm_read_data
+	moveml	%sp@+, %d1-%d5/%a0-%a5
+	unlk	%a6
+	rts
+
+mfm_read_data:
+	movel	%a6@(0x08), %a3
+	lea	%a3@(read_handshake), %a2
+	lea	%a3@(read_data), %a5
+	lea	%a3@(read_mark), %a3
+	movew	#seek_time, %d2
+
+wait_data_init:
+	tstb	%a3@(read_error - read_mark)
+	moveb	#0x18, %a3@(write_mode0 - read_mark)
+	moveb	#0x01, %a3@(write_mode1 - read_mark)
+	moveb	#0x01, %a3@(write_mode0 - read_mark)
+	tstb	%a3@(read_error - read_mark)
+	moveb	#0x08, %a3@(write_mode1 - read_mark)
+
+	lea	sector_data_mark, %a0
+	moveq	#3, %d1
+
+	/* wait data address mark */
+
+wait_data_mark_byte:
+
+	tstb	%a2@
+	dbmi	%d2, wait_data_mark_byte
+	bpl	data_exit
+
+	moveb	%a3@, %d3
+	cmpb	%a0@+, %d3
+	dbne	%d1, wait_data_mark_byte
+	bne	wait_data_init
+
+	/* read data */
+
+	tstb	%a3@(read_error - read_mark)
+
+	movel	#sector_size-1, %d4		/* sector size */
+read_new_data:
+	movew	#max_retry, %d2
+read_data_loop:
+	moveb	%a2@, %d5
+	andb	#0xc0, %d5
+	dbne	%d2, read_data_loop
+	beq	data_exit
+	moveb	%a5@, %a4@+
+	andb	#0x40, %d5
+	dbne	%d4, read_new_data
+	beq	exit_loop
+	moveb	%a5@, %a4@+
+	dbra	%d4, read_new_data
+exit_loop:
+
+	/* read CRC */
+
+	movew	#max_retry, %d2
+data_crc0:
+
+	tstb	%a2@
+	dbmi	%d2, data_crc0
+	bpl	data_exit
+
+	moveb	%a3@, %d5
+
+	moveq	#max_retry, %d2
+
+data_crc1:
+
+	tstb	%a2@
+	dbmi	%d2, data_crc1
+	bpl	data_exit
+
+	moveb	%a3@, %d5
+
+	tstb	%a3@(read_error - read_mark)
+
+	moveb	#0x18, %a3@(write_mode0 - read_mark)
+
+	/* return number of bytes read */
+
+	movel	#sector_size, %d0
+	addw	#1, %d4
+	subl	%d4, %d0
+	rts
+data_exit:
+	moveb	#0x18, %a3@(write_mode0 - read_mark)
+	moveq	#-1, %d0
+	rts