Re: [RFC PATCH 01/22 -v2] Add basic support for gcc profiler instrumentation

[Steven Rostedt <rostedt@goodmis.org>]

On Thu, 10 Jan 2008, Jan Kiszka wrote:
> > ===================================================================
> > --- /dev/null	1970-01-01 00:00:00.000000000 +0000
> > +++ linux-compile-i386.git/arch/x86/kernel/mcount-wrapper.S	2008-01-09 14:10:07.000000000 -0500
> > @@ -0,0 +1,25 @@
> > +/*
> > + *  linux/arch/x86/mcount-wrapper.S
> > + *
> > + *  Copyright (C) 2004 Ingo Molnar
> > + */
> > +
> > +.globl mcount
> > +mcount:
> > +	cmpl $0, mcount_enabled
> > +	jz out
> > +
> > +	push %ebp
> > +	mov %esp, %ebp
>
> What is the benefit of having a call frame in this trampoline? We used
> to carry this in the i386 mcount tracer for Adeos/I-pipe too (it was
> derived from the -rt code), but I just successfully tested a removal
> patch. Also glibc [1] doesn't include it.

OK, I just tried this out on i386, and it works fine.

>
> > +	pushl %eax
> > +	pushl %ecx
> > +	pushl %edx
> > +
> > +	call __mcount
>
> I think this indirection should be avoided, just like the 64-bit version
> and glibc do.

I also did this too.

>
> > +
> > +	popl %edx
> > +	popl %ecx
> > +	popl %eax
> > +	popl %ebp
> > +out:
> > +	ret
>

I'll go try the updates on x86_64 now.

Thanks for the tips!

-- Steve