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, 106 insertions, 0 deletions

diff --git a/scripts/Makefile.kasan b/scripts/Makefile.kasan
new file mode 100644
index 000000000..91504e812
--- /dev/null
+++ b/scripts/Makefile.kasan
@@ -0,0 +1,106 @@
+# SPDX-License-Identifier: GPL-2.0
+
+ifdef CONFIG_CC_HAS_KASAN_MEMINTRINSIC_PREFIX
+# Safe for compiler to generate meminstrinsic calls in uninstrumented files.
+CFLAGS_KASAN_NOSANITIZE :=
+else
+# Don't let compiler generate memintrinsic calls in uninstrumented files
+# because they are instrumented.
+CFLAGS_KASAN_NOSANITIZE := -fno-builtin
+endif
+
+KASAN_SHADOW_OFFSET ?= $(CONFIG_KASAN_SHADOW_OFFSET)
+
+cc-param = $(call cc-option, -mllvm -$(1), $(call cc-option, --param $(1)))
+rustc-param = $(call rustc-option, -Cllvm-args=-$(1),)
+
+check-args = $(foreach arg,$(2),$(call $(1),$(arg)))
+
+kasan_params :=
+
+ifdef CONFIG_KASAN_STACK
+	stack_enable := 1
+else
+	stack_enable := 0
+endif
+
+ifdef CONFIG_KASAN_GENERIC
+
+ifdef CONFIG_KASAN_INLINE
+	# When the number of memory accesses in a function is less than this
+	# call threshold number, the compiler will use inline instrumentation.
+	# 10000 is chosen offhand as a sufficiently large number to make all
+	# kernel functions to be instrumented inline.
+	call_threshold := 10000
+else
+	call_threshold := 0
+endif
+
+# First, enable -fsanitize=kernel-address together with providing the shadow
+# mapping offset, as for GCC, -fasan-shadow-offset fails without -fsanitize
+# (GCC accepts the shadow mapping offset via -fasan-shadow-offset instead of
+# a --param like the other KASAN parameters).
+# Instead of ifdef-checking the compiler, rely on cc-option.
+CFLAGS_KASAN := $(call cc-option, -fsanitize=kernel-address \
+		-fasan-shadow-offset=$(KASAN_SHADOW_OFFSET), \
+		$(call cc-option, -fsanitize=kernel-address \
+		-mllvm -asan-mapping-offset=$(KASAN_SHADOW_OFFSET)))
+
+# The minimum supported `rustc` version has a minimum supported LLVM
+# version late enough that we can assume support for -asan-mapping-offset.
+RUSTFLAGS_KASAN := -Zsanitizer=kernel-address \
+		   -Zsanitizer-recover=kernel-address \
+		   -Cllvm-args=-asan-mapping-offset=$(KASAN_SHADOW_OFFSET)
+
+# Now, add other parameters enabled similarly in GCC, Clang, and rustc.
+# As some of them are not supported by older compilers, these will be filtered
+# through `cc-param` or `rust-param` as applicable.
+kasan_params += asan-instrumentation-with-call-threshold=$(call_threshold) \
+		asan-stack=$(stack_enable) \
+		asan-instrument-allocas=1 \
+		asan-globals=1
+
+# Instrument memcpy/memset/memmove calls by using instrumented __asan_mem*()
+# instead. With compilers that don't support this option, compiler-inserted
+# memintrinsics won't be checked by KASAN on GENERIC_ENTRY architectures.
+kasan_params += asan-kernel-mem-intrinsic-prefix=1
+
+endif # CONFIG_KASAN_GENERIC
+
+ifdef CONFIG_KASAN_SW_TAGS
+
+CFLAGS_KASAN := -fsanitize=kernel-hwaddress
+
+RUSTFLAGS_KASAN := -Zsanitizer=kernel-hwaddress \
+		   -Zsanitizer-recover=kernel-hwaddress
+
+ifdef CONFIG_KASAN_INLINE
+	kasan_params += hwasan-mapping-offset=$(KASAN_SHADOW_OFFSET)
+else
+	kasan_params += hwasan-instrument-with-calls=1
+endif
+
+kasan_params += hwasan-instrument-stack=$(stack_enable) \
+		hwasan-use-short-granules=0 \
+		hwasan-inline-all-checks=0
+
+# Instrument memcpy/memset/memmove calls by using instrumented __(hw)asan_mem*().
+ifdef CONFIG_CC_HAS_KASAN_MEMINTRINSIC_PREFIX
+	ifdef CONFIG_CC_IS_GCC
+		kasan_params += asan-kernel-mem-intrinsic-prefix=1
+	else
+		kasan_params += hwasan-kernel-mem-intrinsic-prefix=1
+	endif
+endif # CONFIG_CC_HAS_KASAN_MEMINTRINSIC_PREFIX
+
+endif # CONFIG_KASAN_SW_TAGS
+
+# Add all as-supported KASAN LLVM parameters requested by the configuration.
+CFLAGS_KASAN += $(call check-args, cc-param, $(kasan_params))
+
+ifdef CONFIG_RUST
+	# Avoid calling `rustc-param` unless Rust is enabled.
+	RUSTFLAGS_KASAN += $(call check-args, rustc-param, $(kasan_params))
+endif # CONFIG_RUST
+
+export CFLAGS_KASAN CFLAGS_KASAN_NOSANITIZE RUSTFLAGS_KASAN