Building symbol table

Jan Smets jan.smets@alcatel-lucent.com
Wed Nov 14 13:08:00 GMT 2012


Hi,

I'm looking for a better way to build my symbol table.

I'm running an embedded (vxworks-alike) OS and I currently use objcopy 
-O binary to dump the text/data sections and push them to the board.

Currently my symbol table is built like this:
  1) I have partially linked my executable
  2) dump symbols with nm
  3) parse the nm output with a script to build a new C file with 
basically a ["symbol_name_string", symbol_name] table.
  4) link the compiled symbol table C file with the partially linked 
executable from 1) and let the linker resolve the addresses.
This is very time consuming and I would like to get rid of the double link.

I was exploring different ideas:
  a) include the .symtab/etc sections in my data segment and use a 
linker script to PROVIDE() pointers to the beginning/end of the tables.
     This didn't work.

  b) Make a binary dump of the .shstrtab/.symtab/.strtab sections of my 
ELF executable and push that binary data to my board in a similar way as 
my text/data segments.

   I was hoping that  objcopy  -j .strtab -j .symtab <in> <out> -O 
binary would get my that.  Similarly readelf -x .strtab <file> works, 
but that is not the output format I want.

Ideally I prefer to get option 1) working, alternatively option 2). Are 
there any other options available? Or is the "double-link" still the 
best way to do it?

Maybe (?) the DWARF unwind tables contain similar information, but I'm 
using SJLJ exception handling so I can't easily switch. If that is a 
better solution, then maybe I can make the switch.

Thanks for your feedback.

- Jan







More information about the Binutils mailing list