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, 155 insertions, 0 deletions

diff --git a/arch/xtensa/lib/memset.S b/arch/xtensa/lib/memset.S
new file mode 100644
index 000000000..262c3f39f
--- /dev/null
+++ b/arch/xtensa/lib/memset.S
@@ -0,0 +1,155 @@
+/*
+ *  arch/xtensa/lib/memset.S
+ *
+ *  ANSI C standard library function memset
+ *  (Well, almost.  .fixup code might return zero.)
+ *
+ *  This file is subject to the terms and conditions of the GNU General
+ *  Public License.  See the file "COPYING" in the main directory of
+ *  this archive for more details.
+ *
+ *  Copyright (C) 2002 Tensilica Inc.
+ */
+
+#include <linux/linkage.h>
+#include <asm/asmmacro.h>
+#include <asm/core.h>
+
+/*
+ * void *memset(void *dst, int c, size_t length)
+ *
+ * The algorithm is as follows:
+ *   Create a word with c in all byte positions
+ *   If the destination is aligned,
+ *     do 16B chucks with a loop, and then finish up with
+ *     8B, 4B, 2B, and 1B stores conditional on the length.
+ *   If destination is unaligned, align it by conditionally
+ *     setting 1B and 2B and then go to aligned case.
+ *   This code tries to use fall-through branches for the common
+ *     case of an aligned destination (except for the branches to
+ *     the alignment labels).
+ */
+
+.text
+ENTRY(__memset)
+WEAK(memset)
+
+	abi_entry_default
+	# a2/ dst, a3/ c, a4/ length
+	extui	a3, a3, 0, 8	# mask to just 8 bits
+	slli	a7, a3, 8	# duplicate character in all bytes of word
+	or	a3, a3, a7	# ...
+	slli	a7, a3, 16	# ...
+	or	a3, a3, a7	# ...
+	mov	a5, a2		# copy dst so that a2 is return value
+	movi	a6, 3		# for alignment tests
+	bany	a2, a6, .Ldstunaligned # if dst is unaligned
+.L0:	# return here from .Ldstunaligned when dst is aligned
+	srli	a7, a4, 4	# number of loop iterations with 16B
+				# per iteration
+	bnez	a4, .Laligned
+	abi_ret_default
+
+/*
+ * Destination is word-aligned.
+ */
+	# set 16 bytes per iteration for word-aligned dst
+	.align	4		# 1 mod 4 alignment for LOOPNEZ
+	.byte	0		# (0 mod 4 alignment for LBEG)
+.Laligned:
+#if XCHAL_HAVE_LOOPS
+	loopnez	a7, .Loop1done
+#else /* !XCHAL_HAVE_LOOPS */
+	beqz	a7, .Loop1done
+	slli	a6, a7, 4
+	add	a6, a6, a5	# a6 = end of last 16B chunk
+#endif /* !XCHAL_HAVE_LOOPS */
+.Loop1:
+EX(10f) s32i	a3, a5,  0
+EX(10f) s32i	a3, a5,  4
+EX(10f) s32i	a3, a5,  8
+EX(10f) s32i	a3, a5, 12
+	addi	a5, a5, 16
+#if !XCHAL_HAVE_LOOPS
+	blt	a5, a6, .Loop1
+#endif /* !XCHAL_HAVE_LOOPS */
+.Loop1done:
+	bbci.l	a4, 3, .L2
+	# set 8 bytes
+EX(10f) s32i	a3, a5,  0
+EX(10f) s32i	a3, a5,  4
+	addi	a5, a5,  8
+.L2:
+	bbci.l	a4, 2, .L3
+	# set 4 bytes
+EX(10f) s32i	a3, a5,  0
+	addi	a5, a5,  4
+.L3:
+	bbci.l	a4, 1, .L4
+	# set 2 bytes
+EX(10f) s16i	a3, a5,  0
+	addi	a5, a5,  2
+.L4:
+	bbci.l	a4, 0, .L5
+	# set 1 byte
+EX(10f) s8i	a3, a5,  0
+.L5:
+.Lret1:
+	abi_ret_default
+
+/*
+ * Destination is unaligned
+ */
+
+.Ldstunaligned:
+	bltui	a4, 8, .Lbyteset	# do short copies byte by byte
+	bbci.l	a5, 0, .L20		# branch if dst alignment half-aligned
+	# dst is only byte aligned
+	# set 1 byte
+EX(10f) s8i	a3, a5,  0
+	addi	a5, a5,  1
+	addi	a4, a4, -1
+	# now retest if dst aligned
+	bbci.l	a5, 1, .L0	# if now aligned, return to main algorithm
+.L20:
+	# dst half-aligned
+	# set 2 bytes
+EX(10f) s16i	a3, a5,  0
+	addi	a5, a5,  2
+	addi	a4, a4, -2
+	j	.L0		# dst is now aligned, return to main algorithm
+
+/*
+ * Byte by byte set
+ */
+	.align	4
+	.byte	0		# 1 mod 4 alignment for LOOPNEZ
+				# (0 mod 4 alignment for LBEG)
+.Lbyteset:
+#if XCHAL_HAVE_LOOPS
+	loopnez	a4, .Lbytesetdone
+#else /* !XCHAL_HAVE_LOOPS */
+	beqz	a4, .Lbytesetdone
+	add	a6, a5, a4	# a6 = ending address
+#endif /* !XCHAL_HAVE_LOOPS */
+.Lbyteloop:
+EX(10f) s8i	a3, a5, 0
+	addi	a5, a5, 1
+#if !XCHAL_HAVE_LOOPS
+	blt	a5, a6, .Lbyteloop
+#endif /* !XCHAL_HAVE_LOOPS */
+.Lbytesetdone:
+	abi_ret_default
+
+ENDPROC(__memset)
+EXPORT_SYMBOL(__memset)
+EXPORT_SYMBOL(memset)
+
+	.section .fixup, "ax"
+	.align	4
+
+/* We return zero if a failure occurred. */
+
+10:
+	movi	a2, 0
+	abi_ret_default