The new option is CONFIG_CC_DISABLE_AUTO_INLINE and it's
located at:
* Kernel hacking
* Disable gcc automatic inlining
This option appends this options CFLAGS:
-fno-default-inline
-fno-inline
-fno-inline-small-functions
-fno-indirect-inlining
-fno-inline-functions-called-once
It's still not clear which of these really work and which doesn't.
Just in case, just disable all of them.
Signed-off-by: Ezequiel Garcia <[email protected]>
---
Changes from v1:
* Change define name to CONFIG_CC_XXX
* Disable all options that refer to inlining
This work is part of CELF Workgroup Project:
"Kernel_dynamic_memory_allocation_tracking_and_reduction":
http://elinux.org/Kernel_dynamic_memory_allocation_tracking_and_reduction
Makefile | 8 ++++++++
kernel/trace/Kconfig | 4 ++++
2 files changed, 12 insertions(+), 0 deletions(-)
diff --git a/Makefile b/Makefile
index ddf5be9..3169603 100644
--- a/Makefile
+++ b/Makefile
@@ -561,6 +561,14 @@ else
KBUILD_CFLAGS += -O2
endif
+ifdef CONFIG_CC_DISABLE_AUTO_INLINE
+KBUILD_CFLAGS += -fno-default-inline \
+ -fno-inline \
+ -fno-inline-small-functions \
+ -fno-indirect-inlining \
+ -fno-inline-functions-called-once
+endif
+
include $(srctree)/arch/$(SRCARCH)/Makefile
ifdef CONFIG_READABLE_ASM
diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig
index 8c4c070..144da36 100644
--- a/kernel/trace/Kconfig
+++ b/kernel/trace/Kconfig
@@ -128,6 +128,10 @@ config TRACING_SUPPORT
depends on STACKTRACE_SUPPORT
default y
+config CC_DISABLE_AUTO_INLINE
+ bool "Disable gcc automatic inlining"
+ default n
+
if TRACING_SUPPORT
menuconfig FTRACE
--
1.7.8.6
Ezequiel Garcia <[email protected]> writes:
> The new option is CONFIG_CC_DISABLE_AUTO_INLINE and it's
> located at:
> * Kernel hacking
> * Disable gcc automatic inlining
You don't state anywhere why you want that?
-Andi
--
[email protected] -- Speaking for myself only
On Sun, Oct 7, 2012 at 9:18 PM, Ezequiel Garcia <[email protected]> wrote:
> The new option is CONFIG_CC_DISABLE_AUTO_INLINE and it's
> located at:
> * Kernel hacking
> * Disable gcc automatic inlining
Can you guarantee that this will still produce a valid kernel?
AFAIK we have some functions which have to be inlined in any case.
--
Thanks,
//richard
On Sun, Oct 7, 2012 at 7:21 PM, Andi Kleen <[email protected]> wrote:
> Ezequiel Garcia <[email protected]> writes:
>
>> The new option is CONFIG_CC_DISABLE_AUTO_INLINE and it's
>> located at:
>> * Kernel hacking
>> * Disable gcc automatic inlining
>
> You don't state anywhere why you want that?
>
This is important -if not necessary- to trace memory allocations with
ftrace kmem
events. Allocations are traced through _RET_IP_ macro, which in turn
only works accurately if gcc don't inline.
Without this patch you will trace callsites like kstrdup, instead of the real
caller.
Perhaps I should explain this better in the commit message?
Thanks for reviewing,
Ezequiel
On Sun, Oct 7, 2012 at 7:30 PM, richard -rw- weinberger
<[email protected]> wrote:
> On Sun, Oct 7, 2012 at 9:18 PM, Ezequiel Garcia <[email protected]> wrote:
>> The new option is CONFIG_CC_DISABLE_AUTO_INLINE and it's
>> located at:
>> * Kernel hacking
>> * Disable gcc automatic inlining
>
> Can you guarantee that this will still produce a valid kernel?
> AFAIK we have some functions which have to be inlined in any case.
>
Not sure what you mean by "valid" kernel. I'm using a kernel with
this, of course.
If you mean always_inline attribute: that's a whole different story.
This option only disables *automatic* inlining.
Ezequiel
Hey Richard,
On Sun, Oct 7, 2012 at 7:30 PM, richard -rw- weinberger
<[email protected]> wrote:
> On Sun, Oct 7, 2012 at 9:18 PM, Ezequiel Garcia <[email protected]> wrote:
>> The new option is CONFIG_CC_DISABLE_AUTO_INLINE and it's
>> located at:
>> * Kernel hacking
>> * Disable gcc automatic inlining
>
> Can you guarantee that this will still produce a valid kernel?
> AFAIK we have some functions which have to be inlined in any case.
>
As a matter of fact, you're right... this breaks compilation on my
desktop machine.
I guess I'll have to give it a respin :-(
Thanks for the feedback.
Ezequiel