brk syscall in libgloss/rs6000/simulator.S and libgloss/m68k/simulator.S

[Jeff Johnston]

Rory,

   You should not be using nor changing the simulator.S code.  It is 
designed to work with the existing generic testing simulator (found in 
gdb package last time I checked).  The simulator is used for running the 
gcc testsuite among other things.  For such a simulator, the syscalls 
are arbitrary - i.e. the simulator can define any syscall number system 
it wants to.  It simply translates these into local syscalls on the 
system running the simulator.  In this case, the powerpc simulator has 
specified syscall 17 as brk so it translates this to a local brk call 
when it sees it.  The simulator also has the liberty of arbitrarily 
defining how a syscall is even implemented.  It may use a rarely used 
insn, a special trap, etc.., whatever it wants to.  This may or may not 
correspond to what is actually used in a linux system.  You cannot 
expect to run that executable on a real powerpc system or a third-party 
simulator/emulator.

   The correct way to handle this is to create a new libgloss library 
under libgloss/rs6000.  You will have to create an ld script that is to 
be used instead of -msim (the compiler won't internally know to link 
with your new libgloss library).  There was at one time a linux library 
in rs6000 but it no longer exists and I can't say whether or not it ever 
worked.

-- Jeff J.

Rory Yorke wrote:
> Hi,
> 
> After successfully building a PowerPC cross-compiling binutils and
> gcc+newlib, I had the following problem when running PowerPC binaries
> under qemu:
> 
>   rory@segserver:~/hack/c/hello$ cat simplesum.c
>   #include <stdio.h>
> 
>   int
>   main()
>   {
>     double y=1+2;
>     printf("1+2 = %g\n",y);
>     return 0;
>   }
>   rory@segserver:~/hack/c/hello$ powerpc-linux-elf-gcc -msim simplesum.c
> -o simplesum
>   rory@segserver:~/hack/c/hello$ qemu-ppc simplesum
>   qemu: Unsupported syscall: 17
>   qemu: Unsupported syscall: 17
>   Invalid data memory access: 0x00000014
>   Invalid segfault errno (2000000)
>   qemu: uncaught target signal 11 (Segmentation fault) - exiting
> 
> According to the /usr/include/asm-i386/unistd.h on my machine, system
> call 17 is 'break'; according to 'man 2 break', this system call is
> unimplemented.  I guessed that what was really wanted was brk, which
> is system call 45.  A corresponding change to the function brk in
> libgloss/rs6000/simulator.S gave me a working result, i.e.,
> 
>   rory@segserver:~/hack/c/hello$ powerpc-linux-elf-gcc -msim simplesum.c
> -o simplesum
>   rory@segserver:~/hack/c/hello$ qemu-ppc simplesum
> 1+2 = 3
> 
> I don't really understand what is going on here: I don't know what the
> break system call is (Google didn't help much), or why it's
> unimplemented, or why it's called in libgloss and not brk.  Perhaps
> there is some other target for which this is the correct behaviour.
> Regardless, this fix Works On My Machine, and I thought it might be
> handy for others to know.
> 
> The patch attached is against newlib 1.12.0; I checked the CVS trunk,
> and
> libgloss/rs6000/simulator.S still calls syscall 17.  The patch alters
> libgloss/m68k/simulator.S in a similar manner; I haven't tested with
> an m68k target.  I grepped for brk in the entire libgloss tree, but I
> didn't understand the other hits.
> 
> Finally, some details on my cross-compile setup; the cross-compile
> tools are:
> 
>   binutils 2.17.50.0.15.20070418
>   gcc 2.95.3
>   newlib 1.12.0
> 
> Host is an Ubuntu 6.06 system, with Ubuntu-provided gcc 4.0.3 and
> binutils 2.16.91 20060118.
> 
> binutils config:
> 
>   ../../binutils-2.17.50.0.15/configure --target=powerpc-linux-elf
> --prefix=/home/rory/powerpc-cross
> 
> gcc/newlib config:
> 
>   ../../gcc-2.95.3/configure
> --with-gcc-version-trigger=/home/rory/work/adarter/cross-compile2/gcc-2.
> 95.3/gcc/version.c --host=i686-pc-linux-gnu --target=powerpc-linux-elf
> --prefix=/home/rory/powerpc-cross --with-newlib
> --program-prefix=powerpc-linux-elf- --norecursion
> 
> (that last is from config.status; my invocation was simpler).
> 
> gcc was built with newlib and libgloss symlinks in the gcc top-level
> source directory.
> 
> I hope this is of some use.
> 
> Regards,
> 
> Rory
>