xmj@hellogcc

1、最近编译glibc库，在链接时遇到一个问题，现象如下：

crtn.S:20: multiple definition of `dummy’ crtn.S:37: multiple definition of `_init’ crtn.S:88: multiple definition of `_fini’ crti.S:58: undefined reference to `i_am_not_a_leaf’ crtn.S:110: undefined reference to `i_am_not_a_leaf’ crtn.S:111: undefined reference to `i_am_not_a_leaf’

搜了下，按照网上的提示把问题解决了。顺便看下initfini.c的代码，感觉挺有意思。

2、这个C文件（initfini.c），首先会被编译成s文件（initfini.s），然后再使用sed程序将其分割成两个S文件（crti.S和crtn.S），最后分别编译成crti.o和crtn.o。文件开头注释里介绍了crti.o和crtn.o的作用：

# This directory contains the C startup code (that which calls main). This # consists of the startfile, built from start.c and installed as crt0.o # (traditionally) or crt1.o (for ELF). In ELF we also install crti.o and # crtn.o, special “initializer” and “finalizer” files used in the link # to make the .init and .fini sections work right; both these files are # built (in an arcane manner) from initfini.c.

Makefile里，有对应的sed程序：

# We only have one kind of startup code files. Static binaries and # shared libraries are build using the PIC version. $(objpfx)crti.S: $(objpfx)initfini.s sed -n -e ‘1,/@HEADER_ENDS/p’
-e ‘/@_.*_PROLOG_BEGINS/,/@_.*_PROLOG_ENDS/p’
-e ‘/@TRAILER_BEGINS/,\(p' $< \> $@   $(objpfx)crtn.S: $(objpfx)initfini.s sed \-n \-e '1,/@HEADER\_ENDS/p' \\ \-e '/@\_.\*\_EPILOG\_BEGINS/,/@\_.\*\_EPILOG\_ENDS/p' \\ \-e '/@TRAILER\_BEGINS/,\)p’ $< > $@

对照着C代码和生成的汇编代码，就可以看出，它的实现方法是，用sed通过C代码中特定的注释来提取、分割汇编代码的。比如“@HEADER_ENDS”，在C文件中的代码如下：

/* The initial common code ends here. */ asm (“\n/*@HEADER_ENDS*/”);

生成的汇编代码如下：

/*@HEADER_ENDS*/

3、显然，这些嵌入的注释，在生成的汇编代码里需要保持原有的顺序，这样才能保证sed可以正确的提取汇编代码。

问题就出在gcc恰恰把生成汇编代码中的注释顺序打乱了，导致sed没能正确提取汇编代码，生成了错误的crti.S和crtn.S。网上的解决方法是，关掉一个优化选项。将Makefile中的

CFLAGS-initfini.s = -g0 -fPIC -fno-inline-functions

改为

CFLAGS-initfini.s = -g0 -fPIC -fno-inline-functions -fno-unit-at-a-time

即可。