Re: [PATCH] Built kernel without -O2 option

[Hui Zhu <teawater@gmail.com>]

On Mon, Nov 29, 2010 at 16:52, Américo Wang <xiyou.wangcong@gmail.com> wrote:
> On Mon, Nov 29, 2010 at 04:24:42PM +0800, Hui Zhu wrote:
>>On Mon, Nov 29, 2010 at 16:16, Américo Wang <xiyou.wangcong@gmail.com> wrote:
>>> On Mon, Nov 29, 2010 at 11:56:15AM +0800, Hui Zhu wrote:
>>>>Hi,
>>>>
>>>>Now, there are a lot of ways to debug the Linux kernel with GDB, like
>>>>qemu, kgtp or kgdb and so on.
>>>>But the developer more like add a printk. It have a lot of reason, a big one is:
>>>>(gdb) p ret
>>>>$3 = <value optimized out>
>>>>And the code execution order is not right.
>>>>
>>>>This is becuase the Kernel is bult with gcc -O2.  Gcc will not
>>>>generate enough debug message with file with -O2.
>>>>So GDB cannot work very well with Linux kernel.
>>>>
>>>>So I make a patch that add a option in "Kernel hacking" called "Close
>>>>GCC optimization".  It will make kernel be built without -O2.
>>>>
>>>>I built and use it in i386 and x86_64.  I will try to make it OK in other arch.
>>>>
>>>
>>> The problem is that some functions _have to_ be inlined and gcc without -O2
>>> doesn't inline them. Have check all the cases? I doubt.
>>
>>If they really need O2, I set  them to O2.
>>Actually, this is the main work, find out the file that need the O2.  :)
>>
>>For example:
>>ifdef CONFIG_CC_CLOSE_OPTIMIZATION
>>CFLAGS_fpu.o                           += -O2
>>CFLAGS_aesni-intel_glue.o              += -O2
>>CFLAGS_ghash-clmulni-intel_glue.o      += -O2
>>endif
>
> No, I didn't mean this, I meant some function that have to be inlined
> are those who only work when being inlined, e.g. current_text_addr().
>
> I think it is not alone. :)
>
> Also, since many inline functions sit in hot-path, are there any
> performance regressions with your patch applied?
>

I did some nm, looks each file that have inline have a special code
for inline function.

> BTW, you need to Cc kbuild for makefile changes like this.
>

Thanks for you remind me.

Best,
Hui