bug-glibc
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Problem in dynamic linker?


From: Thomas,Stephen
Subject: Problem in dynamic linker?
Date: Mon, 19 May 2003 14:53:58 +0100

Hi,

I am working on a port of Linux (+ glibc etc.) to the new SuperH SH5 
architecture. We are currently basing our glibc port on V2.2.5. I have been 
having trouble with the dynamic linker when using non-lazy relocations. The 
problem is that under some circumstances the dynamic linker segfaults in 
elf_dynamic_do_rela in elf/do-rel.h, line 99:

    ElfW(Half) ndx = version[ELFW(R_SYM) (r->r_info)];

Where 'r' has the value 11 (decimal).

The problem appears to lie in macro _ELF_DYNAMIC_DO_RELOC, defined in 
elf/dynamic-link.h. This gets used in ELF_DYNAMIC_DO_RELA, which is used in 
ELF_DYNAMIC_RELOCATE, which in turn gets called from elf/dl-reloc.c, function 
_dl_relocate_object.

_ELF_DYNAMIC_DO_RELOC defines an array 'ranges' of size 2. It always 
initializes ranges[0], but only initializes ranges[1] if non-lazy & some other 
condition. But it then calls elf_dynamic_do_rela using both ranges[0] and [1], 
so the 2nd call causes a problem.

I downloaded V2.3.2, & noticed that there has been a change made in this area. 
The code now looks something like this:

    initialize ranges[0]
    if (! ELF_DURING_STARTUP && ((do_lazy) || (....))
    {
        initialize ranges[1]...
    }
    else
        ...

(*) if (ELF_DURING_STARTUP)
        elf_dynamic_do_##reloc (..., ranges[0].start,...);
    else
        for (ranges_index = 0; ranges_index < 2; ++ranges_index)
              elf_dynamic_do_##reloc (..., ranges [ranges_index].start, ...);

However, I do not believe this is adequate - when ELF_DURING_STARTUP is 0 (as 
in this case), and lazy relocation is used, elf_dynamic_do_<xxx> still gets 
called with ranges[1].

I modified the line marked (*) to:

    if (ELF_DURING_STARTUP && (!do_lazy))

& sure enough, my problem went away!

Since I don't really understand much of this code, can anyone confirm that my 
analysis is correct? Or have I missed something?

Thanks,

Steve Thomas
SuperH (UK) Ltd.




reply via email to

[Prev in Thread] Current Thread [Next in Thread]