Re: [RFA 3/5] New port: CR16: gdb port

[Joel Brobecker <brobecker at adacore dot com>]

> I have extracted a cvs diff below. A read only CVS did not allow me to add
> new files, so I had to add them to CVS/Entries. The diff appears to be 
> against /dev/null. Please let me know if this is OK.

Yes, this is fine. Or you could have simply attached the new file,
instead of making a diff version /dev/null. Either way works for me.

> +/* Target-dependent code for GNU/Linux on the Sitel CR16 processors.
> +
> +   Copyright (C) 2013 Free Software Foundation, Inc.

The copyright years should cover the years that the work was committed
to a medium (hard drive). IIRC, you submitted this work in 2012, so
it should contain at least 2012:

   Copyright (C) 2012-2013 Free Softward Foundation, Inc.

> +static struct type *
> +cr16_register_type (struct gdbarch *gdbarch, int reg_nr)
> +{
> +  switch (reg_nr)
> +    {
> +    case CR16_PC_REGNUM:	/* Note: PC in CR16 is of 24 bits.  */
> +      return builtin_type (gdbarch)->builtin_func_ptr;
> +
> +    case CR16_RA_REGNUM:	/* Return address reg.  */
> +      return builtin_type (gdbarch)->builtin_data_ptr;
> +      break;

Pedro remarked on one of my patches that the "break" statement
after a return is useless, and he's right. Let's remove all of
them (more of those in the same function).

> +static void
> +check_for_saved (void *result_untyped, pv_t addr, CORE_ADDR size, pv_t value)
> +{
> +  struct cr16_prologue *result = (struct cr16_prologue *) result_untyped;
> +
> +  if (value.kind == pvk_register
> +      && value.k == 0
> +      && pv_is_register (addr, CR16_SP_REGNUM)
> +      && size == register_size (target_gdbarch(), value.reg))

Missing space after "target_gdbarch".

> +  stack = make_pv_area (CR16_SP_REGNUM, gdbarch_addr_bit (target_gdbarch()));

Same here.

> +      /* Read 6 bytes, max 48 bit opcode.  */
> +      target_read_memory (pc, buf, 6);
> +      cr16_words[0] = buf[1] << 8 | buf[0];
> +      cr16_words[1] = buf[3] << 8 | buf[2];
> +      cr16_words[2] = buf[5] << 8 | buf[4];
> +      cr16_allWords = (((ULONGLONG) cr16_words[0] << 32)
> +		      + ((ULONGLONG) cr16_words[1] << 16)
> +		      + cr16_words[2]);
> +
> +      /* Find a matching opcode in table.
> +         Nonzero means instruction has a match.  */
> +      is_decoded = cr16_match_opcode ();
> +      cr16_make_instruction ();
> +      length = cr16_currInsn.size;

It hurts every time I read this code... Nothing you can do short
of improving opcode, but this is really awful :-(.

> +  /* Functions with a variable number of arguments have all of their
> +     variable arguments and the last non-variable argument passed
> +     on the stack.
> +
> +     Otherwise, we can pass up to four arguments on the stack.

Up to 4 arguments on the stack? Or by register?

> +static const gdb_byte *
> +cr16_breakpoint_from_pc (struct gdbarch *gdbarch, CORE_ADDR * pcptr,
> +			 int *lenptr)
> +{
> +  /* We use different breakpoint instructions for ELF and uClinux.
> +     See cr16-linux-tdep.c for more details.  */
> +  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
> +
> +  *lenptr = 2;
> +  if (tdep == NULL || tdep->breakpoint == NULL)
> +    return breakpoint_elf;

The above does not look right. tdep should never be NULL, and
I suggest that tdep->breakpoint always be initialized (by
cr16_gdbarch_init).

> +  return tdep->breakpoint;

So the contents of the function should be limited to this last line.

The rest still looks pretty good to me :)

-- 
Joel