On Thu, 2013-01-03 at 22:37 +0100, Franck Jullien wrote:
> Hi,
>
> I ported GDB to the or1k toolchain. It is now GDB 7.5.
>
> I you want to test it, it's here:
> https://github.com/Franck79/or1k-src/commit/prepare_patch
Hi Franck,
Great work - sorry for the slow reply. Comments below:
> I basically copied or32-tdep files from GDB 7.2....
> I replaced all or32 names with or1k.
>
> This port support or1k and or1knd target. The default target is or1k.
> To change this, use:
>
> (gdb) set architecture or1knd
> The target architecture is assumed to be or1knd
>
> (the no delay slot target has not been tested).
As noted in my private email to you, the user guide needs updating, and
it would be good to run regression to check that nothing new has broken.
One other thought. It would be good to merge any relevant changes in to
or32-tdep.c to allow legacy users to benefit.
> This are the only (functional) changes:
>
> + Notes on the GDB 7.5 version
> + ============================
> +
> + This version is just an upgrade of the previous port. It does use CGEN
> + for instruction lookup in or1k_single_step_through_delay as the
> new toolchain
> + is CGEN based.
> +
> + This version is compatible with or1knd target (no delay slot version of
> the
> + toolchain). A check in bfd_arch_info mach value is done (bfd_mach_or1k or
> + bfd_mach_or1knd) to choose if or1k_single_step_through_delay must be
> + implemented.
> +
>
> ----> Please comment, correct this.
>
> @@ -541,20 +573,22 @@
>
> /*--------------------------------------------------------------------------*/
>
> static int
> -or32_single_step_through_delay( struct gdbarch *gdbarch,
> +or1k_single_step_through_delay( struct gdbarch *gdbarch,
> struct frame_info *this_frame )
> {
> ........
>
> + struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
>
> ........
>
> - /* Decode the previous instruction to see if it was a branch or a jump, and
> - hence we are in a delay slot. */
> - index = insn_decode (or32_fetch_instruction (gdbarch, ppc));
> - return or32_opcodes[index].flags & OR32_IF_DELAY;
>
> ----> Don't use the automata thing anymore
Looks reasonable to me.
> + insns = cgen_lookup_insn (tdep->gdb_cgen_cpu_desc,
> + NULL,
> + or1k_fetch_instruction (gdbarch, ppc),
> + NULL,
> + 32,
> + &tmp_fields,
> + 0);
> +
> + /* TODO: we should add a delay slot flag to the CGEN_INSN and remove
> + * this hard coded test. */
>
> ----> Should we do that ? (other arch are doing it like I did here).
>
> + return ((CGEN_INSN_NUM(insns) == OR1K_INSN_L_J) ||
> + (CGEN_INSN_NUM(insns) == OR1K_INSN_L_JAL) ||
> + (CGEN_INSN_NUM(insns) == OR1K_INSN_L_JR) ||
> + (CGEN_INSN_NUM(insns) == OR1K_INSN_L_JALR) ||
> + (CGEN_INSN_NUM(insns) == OR1K_INSN_L_BNF) ||
> + (CGEN_INSN_NUM(insns) == OR1K_INSN_L_BF));
This works, but I agree it would be much better to encode it in the CGEN
semantic information.
> .....
>
> - switch (gdbarch_byte_order (gdbarch))
> - {
> - case BFD_ENDIAN_BIG:
> - set_gdbarch_print_insn (gdbarch, print_insn_big_or32);
> - break;
> -
> - case BFD_ENDIAN_LITTLE:
> - set_gdbarch_print_insn (gdbarch, print_insn_little_or32);
> - break;
>
> - case BFD_ENDIAN_UNKNOWN:
> - error ("or32_gdbarch_init: Unknown endianness");
> - break;
> - }
> + set_gdbarch_print_insn (gdbarch, print_insn_or1k);
>
> ----> print_insn_or1k is now endian aware
Excellent - the old version was always a real nuisance.
> .....
>
> + /* Get a CGEN CPU descriptor for this architecture. */
> + {
> +
> + const char *mach_name = binfo->printable_name;
> + enum cgen_endian endian = (info.byte_order == BFD_ENDIAN_BIG
> + ? CGEN_ENDIAN_BIG
> + : CGEN_ENDIAN_LITTLE);
> +
> + tdep->gdb_cgen_cpu_desc = or1k_cgen_cpu_open
> (CGEN_CPU_OPEN_BFDMACH, mach_name,
> + CGEN_CPU_OPEN_ENDIAN, endian,
> + CGEN_CPU_OPEN_END);
> +
> + or1k_cgen_init_asm (tdep->gdb_cgen_cpu_desc);
> + }
> +
> + /* If this mach as delay slot */
> + if (binfo->mach == bfd_mach_or1k)
> + {
> + set_gdbarch_single_step_through_delay
> + (gdbarch,
> or1k_single_step_through_delay);
> + }
>
> ----> Don't use the automata thing anymore
>
> ......
>
> - /* Initialize the automata for the assembler */
> - build_automata();
>
> ----> Don't use the automata thing anymore
I haven't gone back to the original to look in detail, but I guess with
a CGEN assembler/disassembler you don't need this.
> .......
>
> +static const struct frame_unwind or1k_frame_unwind = {
> .type = NORMAL_FRAME,
> - .this_id = or32_frame_this_id,
> - .prev_register = or32_frame_prev_register,
> + .stop_reason = default_frame_unwind_stop_reason,
> + .this_id = or1k_frame_this_id,
> + .prev_register = or1k_frame_prev_register,
> .unwind_data = NULL,
> .sniffer = default_frame_sniffer,
> .dealloc_cache = NULL,
>
> ----> there is a new handler, stop_reason. Assign default handler to it.
Seems the obvious thing to do.
Best wishes,
Jeremy
--
Tel: +44 (1590) 610184
Cell: +44 (7970) 676050
SkypeID: jeremybennett
Email: [email protected]
Web: www.embecosm.com
_______________________________________________
OpenRISC mailing list
[email protected]
http://lists.openrisc.net/listinfo/openrisc
