Re: [PATCH v2] gdb/MicroBlaze: Add support for native linux gdb

[Michael Eager <eager@eagercon.com>]

Committed d62f64e17

On 6/10/26 6:19 PM, Gopi Kumar Bulusu wrote:
> 
> namaskaaram
> 
> On Thu, Jun 11, 2026, 6:09 AM Michael Eager <eager@eagercon.com 
> <mailto:eager@eagercon.com>> wrote:
> 
>     Hi Gopi --
> 
>     I (or more accurately, git) fixed several whitespace issues.
> 
> 
> Please share. I did check all the whitespace issues flagged by git-diff.
> 
> 
>     Since this is a revised patch, has it been retested?
> 
> 
> 
> Yes, it has been retested.
> 
> dhanyavaadaaha
> gopi
> 
> 
>     On 6/1/26 10:05 PM, Gopi Kumar Bulusu wrote:
>      > namaskaaram
>      >
>      > Attached is the updated patch. All the changes suggested on PATCH v1
>      > have been made and tested.
>      >
>      > I did not remove the (sal.line != 0) check as it seemed trivial.
>      > However, I can make that a separate patch if necessary.
>      >
>      > Our sysadm In the process of setting up a git compatible email - the
>      > process may take more time; until then appreciate your patience with
>      > the attachments.
>      >
>      > Important Note:
>      >
>      > 1) gdb appears to have an elaborate scheme to handle C++ exceptions
>      > and "tunnel" them through exception unaware C code. This code is with
>      > readline. Anytime
>      > an error is discovered (even a syntax error) by a module,
>      > gdb_exception_error() thrown is tunneled through readline using the
>      > TRY_SJLJ/CATCH_SJLJ macros
>      > implemented using setjmp/longjmp.
>      >
>      > 2) There is an issue that I came across while testing; native gdb
>     goes
>      > into an infinite loop when handling errors. I analyzed it and
>     found it
>      > to be an obvious glibc
>      > sysdeps/unix/sysv/linux/microblaze/____longjmp_chk.S issue and
>      > reported it to AMD. It is possible I missed out a glibc patch; At
>     this
>      > time - this is not a gdb issue.
>      >
>      > 3) Plan to follow through [2] and ensure a gcc/glibc defect report is
>      > created if needed.
>      >
>      > dhanyavaadaaha
>      > gopi
>      >
>      > On Tue, Jun 2, 2026 at 5:58 AM Michael Eager <eager@eagerm.com
>     <mailto:eager@eagerm.com>> wrote:
>      >>
>      >> Please post an updated patch for review.
>      >>
>      >> On 5/31/26 11:38 PM, Gopi Kumar Bulusu wrote:
>      >>> On Fri, May 29, 2026 at 8:26 PM Tom Tromey <tom@tromey.com
>     <mailto:tom@tromey.com>> wrote:
>      >>>>
>      >>>>>>>>> Gopi Kumar Bulusu <gopi@sankhya.com
>     <mailto:gopi@sankhya.com>> writes:
>      >>>>
>      >>>>> Attached is a patch to add native linux support for gdb including
>      >>>>> cache target support..
>      >>>>
>      >>>> Thanks for the patch.
>      >>>>
>      >>>> I read through it and have some nits.  There's nothing very
>     serious
>      >>>> except this:
>      >>>>
>      >>>>> The original sources for the patch come from Xilinx/AMD Yocto git
>      >>>>> repository (2025.2)
>      >>>>
>      >>>> Who wrote them and are they covered by the copyright assignment?
>      >>>>
>      >>>> This is the most important consideration, the patch can't land
>     without
>      >>>> this being clear.
>      >>>
>      >>> This work is being performed under a contract with Xilinx/AMD as
>      >>> confirmed in a different response to
>      >>> Simon and therefore covered by the copyright assignment.
>      >>>
>      >>> I will make the changes as described below and retest.
>      >>>
>      >>> Approved for commit ?
>      >>>
>      >>> dhanyavaadaaha
>      >>> gopi
>      >>>
>      >>>>
>      >>>>>   From a3328b4cdfc58783b5f290f22255b36e26a6fb14 Mon Sep 17
>     00:00:00 2001
>      >>>>> From: Gopi Kumar Bulusu <gopi@sankhya.com
>     <mailto:gopi@sankhya.com>>
>      >>>>> Date: Thu, 7 May 2026 17:47:40 +0530
>      >>>>> Subject: [PATCH] gdb/MicroBlaze: Add support for native linux gdb
>      >>>>
>      >>>> It's best to just git send-email rather than using an attachment.
>      >>>>
>      >>>>>       bfd/ChangeLog:
>      >>>>
>      >>>>>       * elf32-microblaze.c (microblaze_elf_grok_prstatus):
>     New function.
>      >>>>>         (microblaze_elf_grok_psinfo): Likewise.
>      >>>>
>      >>>>>       gdb/ChangeLog:
>      >>>>
>      >>>>>       * Makefile.in (HFILES_NO_SRCDIR): Add microblaze-linux-
>     tdep.h
>      >>>>>         ALLDEPFILES: Add microblaze-linux-nat.c
>      >>>>
>      >>>> gdb doesn't use ChangeLogs any more and this text should all
>     be removed.
>      >>>>
>      >>>>> +#define MICROBLAZE_TARGET_HAS_GETREGS 0
>      >>>>> +#define MICROBLAZE_TARGET_HAS_SETREGS 0
>      >>>>
>      >>>> This seems weird:
>      >>>>
>      >>>>> +/* Wrapper function around ptrace.  */
>      >>>>> +
>      >>>>> +static void
>      >>>>> +fetch_target_gp_regs (int tid, elf_gregset_t *gregs)
>      >>>>> +{
>      >>>>> +  elf_greg_t *gregp = *gregs;
>      >>>>> +
>      >>>>> +#if MICROBLAZE_TARGET_HAS_GETREGS
>      >>>>> +  if (ptrace (PTRACE_GETREGS, tid, 0, (long) gregp) < 0)
>      >>>>> +    {
>      >>>>> +      perror_with_name (_("Couldn't get registers"));
>      >>>>> +      return;
>      >>>>> +    }
>      >>>>> +#error "this configuration did not work during testing"
>      >>>>
>      >>>> Normally we don't put new dead code into gdb.
>      >>>>
>      >>>> I think this and the defines could just be replaced with a comment
>      >>>> explaining that PTRACE_GETREGS doesn't work on this platform.
>      >>>>
>      >>>>> +++ b/gdb/microblaze-linux-tdep.h
>      >>>>> @@ -0,0 +1,24 @@
>      >>>>> +/* Target-dependent code for GNU/Linux on MicroBlaze.
>      >>>>> +
>      >>>>> +   Copyright (C) 2021-2026 Free Software Foundation, Inc.
>      >>>>> +
>      >>>>> +   This file is part of GDB.
>      >>>>> +
>      >>>>> +   This program is free software; you can redistribute it
>     and/or modify
>      >>>>> +   it under the terms of the GNU General Public License as
>     published by
>      >>>>> +   the Free Software Foundation; either version 3 of the
>     License, or
>      >>>>> +   (at your option) any later version.
>      >>>>> +
>      >>>>> +   This program is distributed in the hope that it will be
>     useful,
>      >>>>> +   but WITHOUT ANY WARRANTY; without even the implied
>     warranty of
>      >>>>> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
>      >>>>> +   GNU General Public License for more details.
>      >>>>> +
>      >>>>> +   You should have received a copy of the GNU General Public
>     License
>      >>>>> +   along with this program.  If not, see <http://
>     www.gnu.org/licenses/ <http://www.gnu.org/licenses/>>.  */
>      >>>>> +#ifndef MICROBLAZE_LINUX_TDEP_H
>      >>>>> +#define MICROBLAZE_LINUX_TDEP_H
>      >>>>> + /* Target descriptions.  */
>      >>>>> + extern struct target_desc *tdesc_microblaze_linux;
>      >>>>
>      >>>> Blank line between the #define and the comment.
>      >>>>
>      >>>> Also this define should have a slightly different name, see
>      >>>> check-include-guards.py.
>      >>>>
>      >>>>>           non_stack_instruction_found = 0;
>      >>>>> +          cache->register_offsets[rd] = -imm;
>      >>>>>           continue;
>      >>>>
>      >>>> Indentation looks wrong, probably not using tabs.
>      >>>>
>      >>>>> @@ -399,8 +400,7 @@ microblaze_skip_prologue (struct gdbarch
>     *gdbarch, CORE_ADDR start_pc)
>      >>>>>        {
>      >>>>>          sal = find_sal_for_pc (func_start, 0);
>      >>>>
>      >>>>> -      if (sal.end < func_end
>      >>>>> -       && start_pc <= sal.end)
>      >>>>> +      if (sal.line != 0 && sal.end <= func_end  && start_pc
>     <= sal.end)
>      >>>>>         start_pc = sal.end;
>      >>>>
>      >>>> No objection from me but this seems somewhat unrelated.
>      >>>>
>      >>>>
>      >>>>>      /* Call for side effects.  */
>      >>>>> -  get_frame_func (next_frame);
>      >>>>> +  cache->pc = get_frame_func (next_frame);
>      >>>>
>      >>>> Please remove that comment as it is now wrong.
>      >>>>
>      >>>>> +  if (regnum == MICROBLAZE_SP_REGNUM)
>      >>>>> +        regnum = 1;
>      >>>>
>      >>>> Wrong indentation.
>      >>>>
>      >>>>> +  if (regnum == -1) {
>      >>>>> +    int i;
>      >>>>> +
>      >>>>> +    for (i = 0; i < MICROBLAZE_REDR_REGNUM; i++) {
>      >>>>> +      regcache->raw_supply (i, regs + i *
>     MICROBLAZE_REGISTER_SIZE);
>      >>>>> +    }
>      >>>>
>      >>>> Wrong brace placement in this hunk.
>      >>>>
>      >>>>> +  cb (".reg", tdep->sizeof_gregset, tdep->sizeof_gregset,
>     tdep->gregset, NULL,
>      >>>>> +      cb_data);
>      >>>>
>      >>>> Normally in new code we're using 'nullptr' now.  Best would be
>     to go
>      >>>> through the whole patch and check for this.  You don't have to fix
>      >>>> pre-existing code that is using NULL though.
>      >>>>
>      >>>>>      gdbarch *gdbarch
>      >>>>>        = gdbarch_alloc (&info, gdbarch_tdep_up (new
>     microblaze_gdbarch_tdep));
>      >>>>
>      >>>>> +  microblaze_gdbarch_tdep *tdep
>      >>>>> +    = gdbarch_tdep<microblaze_gdbarch_tdep> (gdbarch);
>      >>>>> +
>      >>>>> +  tdep->gregset = NULL;
>      >>>>> +  tdep->sizeof_gregset = 0;
>      >>>>> +  tdep->fpregset = NULL;
>      >>>>> +  tdep->sizeof_fpregset = 0;
>      >>>>
>      >>>> Here it would be better to just add inline initializers to the new
>      >>>> fields in microblaze_gdbarch_tdep.  Then they'll automatically be
>      >>>> initialized by the 'new'.
>      >>>>
>      >>>> Tom
>      >>>
>      >>
> 
>     -- 
>     Michael Eager
> 

-- 
Michael Eager