Re: [rfc][2/3] gdbserver bi-arch support: core s390x part

[Daniel Jacobowitz <drow@false.org>]

On Mon, Jan 21, 2008 at 06:45:52PM +0100, Ulrich Weigand wrote:
> @@ -548,6 +548,7 @@ status_pending_p (struct inferior_list_e
>  static void
>  linux_wait_for_process (struct process_info **childp, int *wstatp)
>  {
> +  static int arch_setup_done = 0;
>    int ret;
>    int to_wait_for = -1;
>  

This isn't quite good enough.  It's already possible for the
architecture to change during a single gdbserver run, so I think we
have to support re-calling arch_setup.  Right now you have to run in
extended mode and replace a 32-bit binary with a 64-bit one before
restarting it, but I'll be checking in the support for "set remote
exec-file" shortly if I catch up on patch review :-)

All it takes is hopefully a global flag that we can reset when
switching to a new inferior.

> diff -urNp gdb-orig/gdb/gdbserver/linux-s390-low.c gdb-head/gdb/gdbserver/linux-s390-low.c
> --- gdb-orig/gdb/gdbserver/linux-s390-low.c	2008-01-18 00:57:40.000000000 +0100
> +++ gdb-head/gdb/gdbserver/linux-s390-low.c	2008-01-18 01:00:18.000000000 +0100
> @@ -102,24 +102,61 @@ static const unsigned char s390_breakpoi
>  static CORE_ADDR
>  s390_get_pc ()
>  {
> -  unsigned long pc;
> -  collect_register_by_name ("pswa", &pc);
> +  if (register_size (0) == 4)
> +    {
> +      unsigned int pc;
> +      collect_register_by_name ("pswa", &pc);
>  #ifndef __s390x__
> -  pc &= 0x7fffffff;
> +      pc &= 0x7fffffff;
>  #endif
> -  return pc;
> +      return pc;
> +    }
> +  else
> +    {
> +      unsigned long pc;
> +      collect_register_by_name ("pswa", &pc);
> +      return pc;
> +    }
>  }

This is harmlessly if dead if gdbserver is 32-bit, right?

> +static void
> +s390_arch_setup (void)
> +{
> +  /* Assume 31-bit inferior process.  */
> +  init_registers_s390 ();
> +
> +  /* On a 64-bit host, check the low bit of the (31-bit) PSWM
> +     -- if this is one, we actually have a 64-bit inferior.  */
> +#ifdef __s390x__
> +  {
> +    unsigned int pswm;
> +    collect_register_by_name ("pswm", &pswm);
> +    if (pswm & 1)
> +      init_registers_s390x ();
> +  }
> +#endif

This makes my head hurt quite a lot.  You're fetching registers into
the cache before you know their size for sure?  I think using
ptrace directly in this case might be more obvious.

-- 
Daniel Jacobowitz
CodeSourcery