[syslinux] elf.c32 plus it's library

[Geert Stappers]

Op 25-09-2007 om 12:39 schreef H. Peter Anvin:
> Geert Stappers wrote:
> > 
~ ~  RTEMS bootable    |           |               |
~ ~--------------------+-----------+---------------+----------------
> >~ CPU + Clock       | P4 3GHz   | P4  2GHz      | P3 866MHz
> >> memory            |  2048 M   |   512 M       |  256 M
> >> Vendor            |  none     | DOLCH FlexPAC | HP Kayak XU800
> >> BIOS              | Intel     |  Award        |    HP
> >              versions dropped, latest available assumed
> >> pxelinux.0        |   OK      |    OK         |    OK
> >> elf.c32           |   OK      |   FAIL        |   FAIL
> >> linux.c32         |  FAIL     |   FAIL        |   FAIL
> > 
> > All linux.c32 configs failing, strange ...

>From another posting:
| | With linux.c32, there is the following error message:
| |
| | loadind /kernel/hello.exe... ok
| | kernel load failure (insuficient memory?)
| | boot:
|
| You can't load your hello.exe with linux.c32,
| you have to load a Linux kernel...

Could be elaborated how the linux.c32 test was done?

> > Has any one a working configuration for linux.c32?
> > ( Is a special kernel needed? Or another special configuration? )
> 
> No, but I think this strengthens my belief that there is something
> subtly wrong with the shuffle library (used by elf.c32 and linux.c32.)
> 
> To explain in more detail: after loading the kernel into memory, the
> shuffle library takes a list of memory segments and a list representing
> the final configuration, and computes a sequence of moves that makes it
> possible to get from point A to point B.  This depends in part on the
> memory configuration of the system, as well, of course, as on random events.
> 
> It always works on my test machine, of course (2 GB RAM), but that's
> probably as much selection effect as anything else.  It's also possible
> that having lots of memory makes it less likely to provoke failure.
> 
> It would be *extremely* interesting if you could recompile elf.c32 with
> "#define DEBUG 0" in com32/lib/syslinux/movebits.c changed to "#define
> DEBUG 1" and see what prints out.

I wonder if it would make sense to run linux.c32 with the same modification.


Looking forward to see an update of the RTEMS bootable table.

Cheers
Geert Stappers
-- 
stappers at alpaca:/usr/src/syslinux/com32/modules
$ grep movebits *
elf.c:#include <syslinux/movebits.h>
Binair bestand elf.elf komt overeen
Binair bestand linux.elf komt overeen
stappers at alpaca:/usr/src/syslinux/com32/modules
$ LANG=C grep movebits *
elf.c:#include <syslinux/movebits.h>
Binary file elf.elf matches
Binary file linux.elf matches
stappers at alpaca:/usr/src/syslinux/com32/modules
$