2019-06-03 11:33:25

by Anders Roxell

[permalink] [raw]
Subject: [PATCH] mm: kasan: mark file report so ftrace doesn't trace it

__kasan_report() triggers ftrace and the preempt_count() in ftrace
causes a call to __asan_load4(), breaking the circular dependency by
making report as no trace for ftrace.

Signed-off-by: Anders Roxell <[email protected]>
---
mm/kasan/Makefile | 3 +++
1 file changed, 3 insertions(+)

diff --git a/mm/kasan/Makefile b/mm/kasan/Makefile
index 08b43de2383b..2b2da731483c 100644
--- a/mm/kasan/Makefile
+++ b/mm/kasan/Makefile
@@ -3,12 +3,14 @@ KASAN_SANITIZE := n
UBSAN_SANITIZE_common.o := n
UBSAN_SANITIZE_generic.o := n
UBSAN_SANITIZE_generic_report.o := n
+UBSAN_SANITIZE_report.o := n
UBSAN_SANITIZE_tags.o := n
KCOV_INSTRUMENT := n

CFLAGS_REMOVE_common.o = $(CC_FLAGS_FTRACE)
CFLAGS_REMOVE_generic.o = $(CC_FLAGS_FTRACE)
CFLAGS_REMOVE_generic_report.o = $(CC_FLAGS_FTRACE)
+CFLAGS_REMOVE_report.o = $(CC_FLAGS_FTRACE)
CFLAGS_REMOVE_tags.o = $(CC_FLAGS_FTRACE)

# Function splitter causes unnecessary splits in __asan_load1/__asan_store1
@@ -17,6 +19,7 @@ CFLAGS_REMOVE_tags.o = $(CC_FLAGS_FTRACE)
CFLAGS_common.o := $(call cc-option, -fno-conserve-stack -fno-stack-protector)
CFLAGS_generic.o := $(call cc-option, -fno-conserve-stack -fno-stack-protector)
CFLAGS_generic_report.o := $(call cc-option, -fno-conserve-stack -fno-stack-protector)
+CFLAGS_report.o := $(call cc-option, -fno-conserve-stack -fno-stack-protector)
CFLAGS_tags.o := $(call cc-option, -fno-conserve-stack -fno-stack-protector)

obj-$(CONFIG_KASAN) := common.o init.o report.o
--
2.20.1


2019-06-03 12:07:36

by Dmitry Vyukov

[permalink] [raw]
Subject: Re: [PATCH] mm: kasan: mark file report so ftrace doesn't trace it

On Mon, Jun 3, 2019 at 11:11 AM Anders Roxell <[email protected]> wrote:
>
> __kasan_report() triggers ftrace and the preempt_count() in ftrace
> causes a call to __asan_load4(), breaking the circular dependency by
> making report as no trace for ftrace.
>
> Signed-off-by: Anders Roxell <[email protected]>
> ---
> mm/kasan/Makefile | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/mm/kasan/Makefile b/mm/kasan/Makefile
> index 08b43de2383b..2b2da731483c 100644
> --- a/mm/kasan/Makefile
> +++ b/mm/kasan/Makefile
> @@ -3,12 +3,14 @@ KASAN_SANITIZE := n
> UBSAN_SANITIZE_common.o := n
> UBSAN_SANITIZE_generic.o := n
> UBSAN_SANITIZE_generic_report.o := n
> +UBSAN_SANITIZE_report.o := n
> UBSAN_SANITIZE_tags.o := n
> KCOV_INSTRUMENT := n
>
> CFLAGS_REMOVE_common.o = $(CC_FLAGS_FTRACE)
> CFLAGS_REMOVE_generic.o = $(CC_FLAGS_FTRACE)
> CFLAGS_REMOVE_generic_report.o = $(CC_FLAGS_FTRACE)
> +CFLAGS_REMOVE_report.o = $(CC_FLAGS_FTRACE)
> CFLAGS_REMOVE_tags.o = $(CC_FLAGS_FTRACE)
>
> # Function splitter causes unnecessary splits in __asan_load1/__asan_store1
> @@ -17,6 +19,7 @@ CFLAGS_REMOVE_tags.o = $(CC_FLAGS_FTRACE)
> CFLAGS_common.o := $(call cc-option, -fno-conserve-stack -fno-stack-protector)
> CFLAGS_generic.o := $(call cc-option, -fno-conserve-stack -fno-stack-protector)
> CFLAGS_generic_report.o := $(call cc-option, -fno-conserve-stack -fno-stack-protector)
> +CFLAGS_report.o := $(call cc-option, -fno-conserve-stack -fno-stack-protector)
> CFLAGS_tags.o := $(call cc-option, -fno-conserve-stack -fno-stack-protector)
>
> obj-$(CONFIG_KASAN) := common.o init.o report.o


Acked-by: Dmitry Vyukov <[email protected]>

Is it needed in all section? Or you just followed the pattern?
Different flag changes were initially done on very specific files for
specific reasons. E.g. -fno-conserve-stack is only for performance
reasons, so report* should not be there. But I see Peter already added
generic_report.o there. Perhaps we need to give up on selective
per-file changes, because this causes constant flow of new bugs in the
absence of testing and just do something like:

KASAN_SANITIZE := n
KCOV_INSTRUMENT := n
UBSAN_SANITIZE := n
CFLAGS_REMOVE = $(CC_FLAGS_FTRACE)
CFLAGS := $(call cc-option, -fno-conserve-stack -fno-stack-protector)