On Sun, 7 Dec 2014, Joe Perches wrote:

On Sun, 2014-12-07 at 11:44 +0100, Julia Lawall wrote:

quoted

quoted

A negative to that approach is inlined functions would
take the function name of the parent not keep the
inlined function name.

I tried the following:

#include <stdio.h>

inline int foo() {
  printf("%s %x\n",__func__,0x12345);
}

int main () {
  foo();
}

The assembly code generated for main is:

0000000000400470 <main>:
  400470:       b9 45 23 01 00          mov    $0x12345,%ecx
  400475:       ba 4b 06 40 00          mov    $0x40064b,%edx
  40047a:       be 44 06 40 00          mov    $0x400644,%esi
  40047f:       bf 01 00 00 00          mov    $0x1,%edi
  400484:       31 c0                   xor    %eax,%eax
  400486:       e9 d5 ff ff ff          jmpq   400460 <__printf_chk@plt>

That is, the call to foo seems tom be inlined.

But the output is:

foo 12345

So it seems that __func__ is determined before inlining.

True, and that's what I intended to describe.

If you did that with a kernel module and replaced
"%s, __func__" with "%pf, __builtin_return_address(0)"
when built with kallsyms you should get:

"modname 12345" when most would expect "foo 12345"

when built without kallsyms, that output should be
"<address> 12345"

but the object code should be smaller.

OK.  But the semantic patch is only using __func__ and only in cases where 
the string wanted is similar to the name of the current function, so I 
think it should be OK?

julia