Hello,
Recently, Naresh reported that the function-graph tracer on the latest
kernel crashes on arm. I could reproduce it and bisected. I finally found
the commit f9b58e8c7d03 ("ARM: 8800/1: use choice for kernel unwinders")
was the first bad commit.
Actually, this commit is just changing Kconfig for making kernel unwinder
choosable. However, as a side effect, this makes CONFIG_FRAME_POINTER=n
by default even if CONFIG_FUNCTION_GRAPH_TRACER=y.
(Note that function-graph tracer implementation on arm depends on
FRAME_POINTER.)
This seems odd, because the commit introduced below code
+choice
+ prompt "Choose kernel unwinder"
+ default UNWINDER_ARM if AEABI && !FUNCTION_GRAPH_TRACER
+ default UNWINDER_FRAME_POINTER if !AEABI || FUNCTION_GRAPH_TRACER
+ help
So, it seems UNWINDER_FRAME_POINTER is used if FUNCTION_GRAPH_TRACER=y.
However, it seems not working. (I guess this is a wrong syntax for Kconfig)
Moreover, since this just setting default value, there seems no direct
dependency to FRAME_POINTER from FUNCTION_GRAPH_TRACER. I guess user can
change it...
I made a fix below (which I tested). This is ugly (arch specific dependency
in generic part) but works.
It enables both of FRAME_POINTER and UNWINDER_ARM.
However I still have some questions.
- Is above choice+default a correct syntax for Kconfig? (Masahiro?)
- Can UNWINDER_ARM work with FRAME_POINTER? (Arnd?)
diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig
index 5e3de28c7677..f79c54680f3b 100644
--- a/kernel/trace/Kconfig
+++ b/kernel/trace/Kconfig
@@ -160,6 +160,7 @@ config FUNCTION_GRAPH_TRACER
depends on HAVE_FUNCTION_GRAPH_TRACER
depends on FUNCTION_TRACER
depends on !X86_32 || !CC_OPTIMIZE_FOR_SIZE
+ select ARCH_WANT_FRAME_POINTERS if ARM
default y
help
Enable the kernel to trace a function at both its return
Thank you,
--
Masami Hiramatsu <mhira...@kernel.org>
