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, 108 insertions, 0 deletions

diff --git a/arch/arc/Makefile b/arch/arc/Makefile
new file mode 100644
index 000000000..868805ffc
--- /dev/null
+++ b/arch/arc/Makefile
@@ -0,0 +1,108 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
+#
+
+KBUILD_DEFCONFIG := haps_hs_smp_defconfig
+
+ifeq ($(CROSS_COMPILE),)
+CROSS_COMPILE := $(call cc-cross-prefix, arc-linux- arceb-linux- arc-linux-gnu-)
+endif
+
+cflags-y	+= -fno-common -pipe -fno-builtin -mmedium-calls -mswape -D__linux__
+
+tune-mcpu-def-$(CONFIG_ISA_ARCOMPACT)	:= -mcpu=arc700
+tune-mcpu-def-$(CONFIG_ISA_ARCV2)	:= -mcpu=hs38
+
+ifeq ($(CONFIG_ARC_TUNE_MCPU),)
+cflags-y				+= $(tune-mcpu-def-y)
+else
+tune-mcpu				:= $(CONFIG_ARC_TUNE_MCPU)
+ifneq ($(call cc-option,$(tune-mcpu)),)
+cflags-y				+= $(tune-mcpu)
+else
+# The flag provided by 'CONFIG_ARC_TUNE_MCPU' option isn't known by this compiler
+# (probably the compiler is too old). Use ISA default mcpu flag instead as a safe option.
+$(warning ** WARNING ** CONFIG_ARC_TUNE_MCPU flag '$(tune-mcpu)' is unknown, fallback to '$(tune-mcpu-def-y)')
+cflags-y				+= $(tune-mcpu-def-y)
+endif
+endif
+
+ifdef CONFIG_ARC_CURR_IN_REG
+# For a global register definition, make sure it gets passed to every file
+# We had a customer reported bug where some code built in kernel was NOT using
+# any kernel headers, and missing the global register
+# Can't do unconditionally because of recursive include issues
+# due to <linux/thread_info.h>
+LINUXINCLUDE	+=  -include $(srctree)/arch/arc/include/asm/current.h
+cflags-y	+= -ffixed-gp
+endif
+
+cflags-y				+= -fsection-anchors
+
+cflags-$(CONFIG_ARC_HAS_LLSC)		+= -mlock
+
+ifdef CONFIG_ISA_ARCV2
+
+ifdef CONFIG_ARC_USE_UNALIGNED_MEM_ACCESS
+cflags-y				+= -munaligned-access
+else
+cflags-y				+= -mno-unaligned-access
+endif
+
+ifndef CONFIG_ARC_HAS_LL64
+cflags-y				+= -mno-ll64
+endif
+
+ifndef CONFIG_ARC_HAS_DIV_REM
+cflags-y				+= -mno-div-rem
+endif
+
+endif
+
+cfi := $(call as-instr,.cfi_startproc\n.cfi_endproc,-DARC_DW2_UNWIND_AS_CFI)
+cflags-$(CONFIG_ARC_DW2_UNWIND)		+= -fasynchronous-unwind-tables $(cfi)
+
+# small data is default for elf32 tool-chain. If not usable, disable it
+# This also allows repurposing GP as scratch reg to gcc reg allocator
+disable_small_data := y
+cflags-$(disable_small_data)		+= -mno-sdata
+
+cflags-$(CONFIG_CPU_BIG_ENDIAN)		+= -mbig-endian
+ldflags-$(CONFIG_CPU_BIG_ENDIAN)	+= -EB
+
+LIBGCC	= $(shell $(CC) $(cflags-y) --print-libgcc-file-name)
+
+# Modules with short calls might break for calls into builtin-kernel
+KBUILD_CFLAGS_MODULE	+= -mlong-calls -mno-millicode
+
+# Finally dump eveything into kernel build system
+KBUILD_CFLAGS	+= $(cflags-y)
+KBUILD_AFLAGS	+= $(KBUILD_CFLAGS)
+KBUILD_LDFLAGS	+= $(ldflags-y)
+
+core-y				+= arch/arc/plat-sim/
+core-$(CONFIG_ARC_PLAT_TB10X)	+= arch/arc/plat-tb10x/
+core-$(CONFIG_ARC_PLAT_AXS10X)	+= arch/arc/plat-axs10x/
+core-$(CONFIG_ARC_SOC_HSDK)	+= arch/arc/plat-hsdk/
+
+libs-y		+= arch/arc/lib/ $(LIBGCC)
+
+boot		:= arch/arc/boot
+
+boot_targets := uImage.bin uImage.gz uImage.lzma
+
+PHONY += $(boot_targets)
+$(boot_targets): vmlinux
+	$(Q)$(MAKE) $(build)=$(boot) $(boot)/$@
+
+uimage-default-y			:= uImage.bin
+uimage-default-$(CONFIG_KERNEL_GZIP)	:= uImage.gz
+uimage-default-$(CONFIG_KERNEL_LZMA)	:= uImage.lzma
+
+PHONY += uImage
+uImage: $(uimage-default-y)
+	@ln -sf $< $(boot)/uImage
+	@$(kecho) '  Image $(boot)/uImage is ready'
+
+CLEAN_FILES += $(boot)/uImage