Bug#700015: libx11-6: crashes on systems using Mali graphics

[Сергей И. Королев <exception0x0d@gmail.com>]

good news! i make quick and dirty hacks and this work system wide.

details:
  my libMali.so have one nop instruction in function _init and two
unused bytes in section .init;
  i disassemble libMali.so to find addresses of XInitThreads thunk and
function called from _init;
  i increase section .init lenght to +2;
  i make simple dummy asm code:

.thumb

/*
        fake _init
*/
.org 0x02476c
        nop
        blx XInitThread
        bl sub_x
        nop

/*
        fake XInitThreads thunk
*/
.org 0x024d14
XInitThread:
        nop

/*
        fake subfunction for _init
*/
.org 0x0268e0
sub_x:
        nop

  i compile this code and make objdump -d of result and have:


a.out:     file format elf32-littlearm


Disassembly of section .text:

00000000 <XInitThread-0x24d14>:
        ...
   2476c:       46c0            nop                     ; (mov r8, r8)
   2476e:       f000 ead2       blx     24d14 <XInitThread>
   24772:       f002 f8b5       bl      268e0 <sub_x>
   24776:       46c0            nop                     ; (mov r8, r8)
        ...

00024d14 <XInitThread>:
   24d14:       46c0            nop                     ; (mov r8, r8)
        ...

000268e0 <sub_x>:
   268e0:       46c0            nop                     ; (mov r8, r8)

  i replace call to subfunction and nop in _init to generated code;
  and in result i have needed - XInitThreads call after libMali.so load;


2013/3/4 Michal Suchanek <hramrach@gmail.com>:
> On 4 March 2013 09:17, Сергей И. Королев <exception0x0d@gmail.com> wrote:
>> in my instalation gnome-shell, empathy, evolution, emerillon and some
>> other application affected to this bug. i think about system way
>> workaround. how about patch libMali.so to call XInitThreads in init
>> after library loaded?
>>
>
> That's presumably possible but you will have to binary-patch the
> library. There is no source.
>
> Also it's good idea to try the workaround with gdb to make sure you
> really have the same problem.
>
> Thanks
>
> Michal