[PATCH] [RISCV] Support subtraction of .uleb128.

classic Classic list List threaded Threaded
13 messages Options
Reply | Threaded
Open this post in threaded view
|

[PATCH] [RISCV] Support subtraction of .uleb128.

Kuan-Lin Chen
The data length of uleb128 is variable.  So linker must recalculate the
value of the subtraction.  The patch leave relocations in object files
so that linker can relocate again after relaxation.

bfd/ChangeLog:
* bfd-in2.h: Regenerated.
* elfnn-riscv.c (write_uleb128): New function.
(perform_relocation): Perform R_RISCV_SUB_ULEB128 and
R_RISCV_SET_ULEB128 relocation.
(riscv_elf_relocate_section): Likewise.
* elfxx-riscv.c (howto_table): Add R_RISCV_SUB_ULEB128 and
R_RISCV_SET_ULEB128.
(riscv_elf_ignore_reloc): New function.
* libbfd.h: Regenerated.
* reloc.c (BFD_RELOC_RISCV_SET_ULEB128, BFD_RELOC_RISCV_SUB_ULEB128):
New relocations to support .uleb128.

gas/ChangeLog:
* config/tc-riscv.c (md_apply_fix): Add BFD_RELOC_RISCV_SET_ULEB128 and
BFD_RELOC_RISCV_SUB_ULEB128.
(riscv_insert_uleb128_fixes): New function.
(riscv_md_end): Scan rs_leb128 fragments.
(riscv_pseudo_table): Remove uleb128.

include/ChangeLog:
* elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.

ld/ChangeLog:
* testsuite/ld-riscv-elf/ld-riscv-elf.exp: Add uleb128.
* testsuite/ld-riscv-elf/uleb128.d: New test.
* testsuite/ld-riscv-elf/uleb128.s: New file.

OK to commit?
--
Best regards,
Kuan-Lin Chen.
[hidden email]

0001-RISC-V-Support-subtraction-of-.uleb128.patch (21K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] [RISCV] Support subtraction of .uleb128.

Andreas Schwab-2
On Nov 27 2019, Kuan-Lin Chen wrote:

> diff --git a/include/ChangeLog b/include/ChangeLog
> index 47bb86cf71..c290b245c6 100644
> --- a/include/ChangeLog
> +++ b/include/ChangeLog
> @@ -1,3 +1,7 @@
> +2019-11-27  Kuan-Lin Chen  <[hidden email]>
> +
> + * elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.

There are too many open parens here.

Andreas.

--
Andreas Schwab, [hidden email]
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
"And now for something completely different."
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] [RISCV] Support subtraction of .uleb128.

Kuan-Lin Chen
Fixed in the patch.
Thanks.

OK to commit?

Andreas Schwab <[hidden email]> 於 2019年11月27日 週三 下午6:36寫道:

>
> On Nov 27 2019, Kuan-Lin Chen wrote:
>
> > diff --git a/include/ChangeLog b/include/ChangeLog
> > index 47bb86cf71..c290b245c6 100644
> > --- a/include/ChangeLog
> > +++ b/include/ChangeLog
> > @@ -1,3 +1,7 @@
> > +2019-11-27  Kuan-Lin Chen  <[hidden email]>
> > +
> > +     * elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.
>
> There are too many open parens here.
>
> Andreas.
>
> --
> Andreas Schwab, [hidden email]
> GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
> "And now for something completely different."


--
Best regards,
Kuan-Lin Chen.
[hidden email]

0001-RISC-V-Support-subtraction-of-.uleb128.patch (21K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] [RISCV] Support subtraction of .uleb128.

Jim Wilson-2
In reply to this post by Kuan-Lin Chen
On 11/27/19 12:11 AM, Kuan-Lin Chen wrote:
> The data length of uleb128 is variable.  So linker must recalculate the
> value of the subtraction.  The patch leave relocations in object files
> so that linker can relocate again after relaxation.

I think we should have some agreement on the psABI change before we
accept binutils patches.  We should give the LLVM/lld folks a chance to
comment.  I see you did file a psABI pull request, and that there is no
discussion on that yet.
https://github.com/riscv/riscv-elf-psabi-doc/pull/124

I haven't reviewed the patch yet.  I took a week off, and am now trying
to catch up on the backlog.

Jim
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] [RISCV] Support subtraction of .uleb128.

Sourceware - binutils list mailing list
In reply to this post by Kuan-Lin Chen
On Wed, 27 Nov 2019 00:11:50 PST (-0800), [hidden email] wrote:

> The data length of uleb128 is variable.  So linker must recalculate the
> value of the subtraction.  The patch leave relocations in object files
> so that linker can relocate again after relaxation.
>
> bfd/ChangeLog:
> * bfd-in2.h: Regenerated.
> * elfnn-riscv.c (write_uleb128): New function.
> (perform_relocation): Perform R_RISCV_SUB_ULEB128 and
> R_RISCV_SET_ULEB128 relocation.
> (riscv_elf_relocate_section): Likewise.
> * elfxx-riscv.c (howto_table): Add R_RISCV_SUB_ULEB128 and
> R_RISCV_SET_ULEB128.
> (riscv_elf_ignore_reloc): New function.
> * libbfd.h: Regenerated.
> * reloc.c (BFD_RELOC_RISCV_SET_ULEB128, BFD_RELOC_RISCV_SUB_ULEB128):
> New relocations to support .uleb128.
>
> gas/ChangeLog:
> * config/tc-riscv.c (md_apply_fix): Add BFD_RELOC_RISCV_SET_ULEB128 and
> BFD_RELOC_RISCV_SUB_ULEB128.
> (riscv_insert_uleb128_fixes): New function.
> (riscv_md_end): Scan rs_leb128 fragments.
> (riscv_pseudo_table): Remove uleb128.
>
> include/ChangeLog:
> * elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.
>
> ld/ChangeLog:
> * testsuite/ld-riscv-elf/ld-riscv-elf.exp: Add uleb128.
> * testsuite/ld-riscv-elf/uleb128.d: New test.
> * testsuite/ld-riscv-elf/uleb128.s: New file.
>
> OK to commit?

I'm having a lot of trouble trying to figure out how this works.  LMK if I'm
missing something, there's some comments in line:

> From 9f7f0aaa484dee2b32c550f8cca4de18959f01e1 Mon Sep 17 00:00:00 2001
> From: Kuan-Lin Chen <[hidden email]>
> Date: Thu, 14 Nov 2019 14:24:22 +0800
> Subject: [PATCH] RISC-V: Support subtraction of .uleb128.
>
> The data length of uleb128 is variable.  So linker must recalculate the
> value of the subtraction.  The patch leave relocations in object files
> so that linker can relocate again after relaxation.
>
> bfd/ChangeLog:
>         * bfd-in2.h: Regenerated.
>         * elfnn-riscv.c (write_uleb128): New function.
>         (perform_relocation): Perform R_RISCV_SUB_ULEB128 and
>          R_RISCV_SET_ULEB128 relocation.
>         (riscv_elf_relocate_section): Likewise.
>         * elfxx-riscv.c (howto_table): Add R_RISCV_SUB_ULEB128 and
>         R_RISCV_SET_ULEB128.
>         (riscv_elf_ignore_reloc): New function.
>         * libbfd.h: Regenerated.
>         * reloc.c (BFD_RELOC_RISCV_SET_ULEB128, BFD_RELOC_RISCV_SUB_ULEB128):
>         New relocations to support .uleb128.
>
> gas/ChangeLog:
>         * config/tc-riscv.c (md_apply_fix): Add BFD_RELOC_RISCV_SET_ULEB128 and
>         BFD_RELOC_RISCV_SUB_ULEB128.
>         (riscv_insert_uleb128_fixes): New function.
>         (riscv_md_end): Scan rs_leb128 fragments.
>         (riscv_pseudo_table): Remove uleb128.
>
> include/ChangeLog:
>         * elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.
>
> ld/ChangeLog:
>         * testsuite/ld-riscv-elf/ld-riscv-elf.exp: Add uleb128.
>         * testsuite/ld-riscv-elf/uleb128.d: New test.
>         * testsuite/ld-riscv-elf/uleb128.s: New file.
> ---
>  bfd/ChangeLog                              | 14 ++++
>  bfd/bfd-in2.h                              |  2 +
>  bfd/elfnn-riscv.c                          | 82 ++++++++++++++++++++++
>  bfd/elfxx-riscv.c                          | 51 ++++++++++++++
>  bfd/libbfd.h                               |  2 +
>  bfd/reloc.c                                |  4 ++
>  gas/ChangeLog                              |  8 +++
>  gas/config/tc-riscv.c                      | 47 ++++++++++++-
>  include/ChangeLog                          |  4 ++
>  include/elf/riscv.h                        |  2 +
>  ld/ChangeLog                               |  6 ++
>  ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp |  1 +
>  ld/testsuite/ld-riscv-elf/uleb128.d        | 18 +++++
>  ld/testsuite/ld-riscv-elf/uleb128.s        | 18 +++++
>  14 files changed, 258 insertions(+), 1 deletion(-)
>  create mode 100644 ld/testsuite/ld-riscv-elf/uleb128.d
>  create mode 100644 ld/testsuite/ld-riscv-elf/uleb128.s
>
> diff --git a/bfd/ChangeLog b/bfd/ChangeLog
> index 4a0852e577..d11ffde74c 100644
> --- a/bfd/ChangeLog
> +++ b/bfd/ChangeLog
> @@ -1,3 +1,17 @@
> +2019-11-27  Kuan-Lin Chen  <[hidden email]>
> +
> +        * bfd-in2.h: Regenerated.
> +        * elfnn-riscv.c (write_uleb128): New function.
> +        (perform_relocation): Perform R_RISCV_SUB_ULEB128 and
> +         R_RISCV_SET_ULEB128 relocation.
> +        (riscv_elf_relocate_section): Likewise.
> +        * elfxx-riscv.c (howto_table): Add R_RISCV_SUB_ULEB128 and
> +        R_RISCV_SET_ULEB128.
> +        (riscv_elf_ignore_reloc): New function.
> +        * libbfd.h: Regenerated.
> +        * reloc.c (BFD_RELOC_RISCV_SET_ULEB128, BFD_RELOC_RISCV_SUB_ULEB128):
> +        New relocations to support .uleb128.
> +
>  2019-11-27  Alan Modra  <[hidden email]>
>  
>          PR 23652
> diff --git a/bfd/bfd-in2.h b/bfd/bfd-in2.h
> index 44902fc8d0..a32708ebc2 100644
> --- a/bfd/bfd-in2.h
> +++ b/bfd/bfd-in2.h
> @@ -4380,6 +4380,8 @@ number for the SBIC, SBIS, SBI and CBI instructions  */
>    BFD_RELOC_RISCV_SET16,
>    BFD_RELOC_RISCV_SET32,
>    BFD_RELOC_RISCV_32_PCREL,
> +  BFD_RELOC_RISCV_SET_ULEB128,
> +  BFD_RELOC_RISCV_SUB_ULEB128,
>  
>  /* Renesas RL78 Relocations.  */
>    BFD_RELOC_RL78_NEG8,
> diff --git a/bfd/elfnn-riscv.c b/bfd/elfnn-riscv.c
> index 997f786602..075ef6b82d 100644
> --- a/bfd/elfnn-riscv.c
> +++ b/bfd/elfnn-riscv.c
> @@ -1419,6 +1419,26 @@ riscv_global_pointer_value (struct bfd_link_info *info)
>    return h->u.def.value + sec_addr (h->u.def.section);
>  }
>  
> +/* Write VAL in uleb128 format to P, returning a pointer to the
> +   following byte.
> +   This code is copied from elf-attr.c.  */
> +
> +static bfd_byte *
> +write_uleb128 (bfd_byte *p, unsigned int val)
> +{
> +  bfd_byte c;
> +  do
> +    {
> +      c = val & 0x7f;
> +      val >>= 7;
> +      if (val)
> +        c |= 0x80;
> +      *(p++) = c;
> +    }
> +  while (val);
> +  return p;
> +}
> +
>  /* Emplace a static relocation.  */
>  
>  static bfd_reloc_status_type
> @@ -1512,6 +1532,25 @@ perform_relocation (const reloc_howto_type *howto,
>          value = ENCODE_RVC_LUI_IMM (RISCV_CONST_HIGH_PART (value));
>        break;
>  
> +    case R_RISCV_SET_ULEB128:
> +    case R_RISCV_SUB_ULEB128:
> +      {
> +        unsigned int len = 0;
> +        bfd_byte *endp, *p;
> +
> +        _bfd_read_unsigned_leb128 (input_bfd, contents + rel->r_offset, &len);
> +
> +        /* Clean the contents value to zero.  Do not reduce the length.  */
> +        p = contents + rel->r_offset;
> +        endp = p + len -1;
> +        memset (p, 0x80, len);
> +        *(endp) = 0;
> +        p = write_uleb128 (p, value) - 1;
> +        if (p < endp)
> +          *p |= 0x80;
> +        return bfd_reloc_ok;
> +      }
> +
>      case R_RISCV_32:
>      case R_RISCV_64:
>      case R_RISCV_ADD8:
> @@ -1767,6 +1806,8 @@ riscv_elf_relocate_section (bfd *output_bfd,
>    struct elf_link_hash_entry **sym_hashes = elf_sym_hashes (input_bfd);
>    bfd_vma *local_got_offsets = elf_local_got_offsets (input_bfd);
>    bfd_boolean absolute;
> +  bfd_vma uleb128_vma;
> +  Elf_Internal_Rela *uleb128_rel = NULL;
>  
>    if (!riscv_init_pcrel_relocs (&pcrel_relocs))
>      return FALSE;
> @@ -1871,6 +1912,47 @@ riscv_elf_relocate_section (bfd *output_bfd,
>          case R_RISCV_DELETE:
>            /* These require no special handling beyond perform_relocation.  */
>            break;
> +        case R_RISCV_SET_ULEB128:
> +          if (!uleb128_rel)
> +            {
> +              /* Save the minuend to use later.  */
> +              uleb128_vma = relocation;
> +              uleb128_rel = rel;
> +              continue;
> +            }
> +          else
> +            {
> +              if (uleb128_rel->r_offset != rel->r_offset)
> +                {
> +                  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
> +                                         input_bfd, howto->name);
> +                  bfd_set_error (bfd_error_bad_value);
> +                }
> +              relocation = relocation - uleb128_vma;
> +              uleb128_rel = NULL;
> +              break;
> +            }
> +
> +        case R_RISCV_SUB_ULEB128:
> +          if (uleb128_rel)
> +            {
> +              if (uleb128_rel->r_offset != rel->r_offset)
> +                {
> +                  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
> +                                         input_bfd, howto->name);
> +                  bfd_set_error (bfd_error_bad_value);
> +                }
> +              relocation = uleb128_vma - relocation;

GCC gives me -Wmaybe-uninitialized for both of these.

> +              uleb128_rel = NULL;
> +              break;
> +            }
> +          else
> +            {
> +              /* Save the subtrahend to use later.  */
> +              uleb128_vma = relocation;
> +              uleb128_rel = rel;
> +              continue;
> +            }
>  
>          case R_RISCV_GOT_HI20:
>            if (h != NULL)
> diff --git a/bfd/elfxx-riscv.c b/bfd/elfxx-riscv.c
> index 245717f70f..c136ba2207 100644
> --- a/bfd/elfxx-riscv.c
> +++ b/bfd/elfxx-riscv.c
> @@ -38,6 +38,8 @@
>     relocations for the debug info.  */
>  static bfd_reloc_status_type riscv_elf_add_sub_reloc
>    (bfd *, arelent *, asymbol *, void *, asection *, bfd *, char **);
> +static bfd_reloc_status_type riscv_elf_ignore_reloc
> +  (bfd *, arelent *, asymbol *, void *, asection *, bfd *, char **);
>  
>  /* The relocation table used for SHT_RELA sections.  */
>  
> @@ -855,8 +857,41 @@ static reloc_howto_type howto_table[] =
>           0,                                /* src_mask */
>           MINUS_ONE,                        /* dst_mask */
>           FALSE),                        /* pcrel_offset */
> +
> +  /* The length of unsigned-leb128 is variant, just assume the
> +     size is one byte here.  */
> +  HOWTO (R_RISCV_SET_ULEB128,                /* type */
> +         0,                                /* rightshift */
> +         0,                                /* size */
> +         0,                                /* bitsize */
> +         FALSE,                                /* pc_relative */
> +         0,                                /* bitpos */
> +         complain_overflow_dont,        /* complain_on_overflow */
> +         riscv_elf_ignore_reloc,        /* special_function */
> +         "R_RISCV_SET_ULEB128",                /* name */
> +         FALSE,                                /* partial_inplace */
> +         0,                                /* src_mask */
> +         0,                                /* dst_mask */
> +         FALSE),                        /* pcrel_offset */

I don't understand how this could possibly work: uleb128s can be long, so we
can't just treat them as a single byte.  I'm having trouble testing it, though,
as I can't figure out how to emit R_RISCV_SET_ULEB128 alone.  For example, when
I try to insert a uleb128 in the assembler I'm not even getting a relocation:

    .text
    .globl _start
    _start:
            nop
    .L0:
            nop
    .L1:
    uleb_l0:
            addi x0, x0, 1
            .uleb128 .L0
            # Manually fix up the alignment, as .align is broken.
            .rept 3
            .byte 0xAA
            .endr
            addi x0, x0, 2
   
    $ objdump -dr
    test.o:     file format elf64-littleriscv
   
   
    Disassembly of section .text:
   
    0000000000000000 <_start>:
       0:   00000013                nop
       4:   00000013                nop
   
    0000000000000008 <uleb_l0>:
       8:   00100013                li      zero,1
       c:   aa04                    fsd     fs1,16(a2)
       e:   aaaa                    fsd     fa0,336(sp)
      10:   00200013                li      zero,2

so then when I link I'm just getting the wrong value entirely.

    test:     file format elf64-littleriscv
   
   
    Disassembly of section .text:
   
    0000000000010078 <_start>:
       10078:       00000013                nop
       1007c:       00000013                nop
   
    0000000000010080 <uleb_l0>:
       10080:       00100013                li      zero,1
       10084:       aa04                    fsd     fs1,16(a2)
       10086:       aaaa                    fsd     fa0,336(sp)
       10088:       00200013                li      zero,2

> +  /* The length of unsigned-leb128 is variant, just assume the
> +     size is one byte here.  */
> +  HOWTO (R_RISCV_SUB_ULEB128,                /* type */
> +         0,                                /* rightshift */
> +         0,                                /* size */
> +         0,                                /* bitsize */
> +         FALSE,                                /* pc_relative */
> +         0,                                /* bitpos */
> +         complain_overflow_dont,        /* complain_on_overflow */
> +         riscv_elf_ignore_reloc,        /* special_function */
> +         "R_RISCV_SUB_ULEB128",                /* name */
> +         FALSE,                                /* partial_inplace */
> +         0,                                /* src_mask */
> +         0,                                /* dst_mask */
> +         FALSE),                        /* pcrel_offset */
>  };

I can get a R_RISCV_SUB_ULEB128, but it appears to be both producing the wrong
value and garbaging up bytes after the relocation.  For example:

    .text
    .globl _start
    _start:
            nop
    .L0:
            nop
    .L1:
    uleb_l1_l0:
            addi x0, x0, 1
            .uleb128 .L1 - .L0
            # Manually fix up the alignment, as .align is broken
            .rept 3
            .byte 0xAA
            .endr
            addi x0, x0, 2

produces the relocation

    test.o:     file format elf64-littleriscv
   
   
    Disassembly of section .text:
   
    0000000000000000 <_start>:
       0:   00000013                nop
   
    0000000000000004 <.L0>:
       4:   00000013                nop
   
    0000000000000008 <uleb_l1_l0>:
       8:   00100013                li      zero,1
       c:   aa04                    fsd     fs1,16(a2)
                            c: R_RISCV_SET_ULEB128  .L1
                            c: R_RISCV_SUB_ULEB128  .L0
       e:   aaaa                    fsd     fa0,336(sp)
      10:   00200013                li      zero,2

but then goes and corrupts the resulting executable


    test:     file format elf64-littleriscv
    Disassembly of section .text:
   
    0000000000010078 <_start>:
       10078:       00000013                nop
       1007c:       00000013                nop
   
    0000000000010080 <uleb_l1_l0>:
       10080:       00100013                li      zero,1
       10084:       ff84                    sd      s1,56(a5)
       10086:       000ffffb                0xffffb
       1008a:       0020                    addi    s0,sp,8

I can't think of a way to make this work aside from just emitting uleb128s at
their full length, which would make them work but would really defeat the
purpose of the format.  I guess we could then relax these, but that seems like
a lot of work.

> +
>  /* A mapping from BFD reloc types to RISC-V ELF reloc types.  */
>  
>  struct elf_reloc_map
> @@ -917,6 +952,8 @@ static const struct elf_reloc_map riscv_reloc_map[] =
>    { BFD_RELOC_RISCV_SET16, R_RISCV_SET16 },
>    { BFD_RELOC_RISCV_SET32, R_RISCV_SET32 },
>    { BFD_RELOC_RISCV_32_PCREL, R_RISCV_32_PCREL },
> +  { BFD_RELOC_RISCV_SET_ULEB128, R_RISCV_SET_ULEB128 },
> +  { BFD_RELOC_RISCV_SUB_ULEB128, R_RISCV_SUB_ULEB128 },
>  };
>  
>  /* Given a BFD reloc type, return a howto structure.  */
> @@ -1011,6 +1048,20 @@ riscv_elf_add_sub_reloc (bfd *abfd,
>    return bfd_reloc_ok;
>  }
>  
> +/* Special handler for relocations which don't have to be relocated.
> +   This function just simply return bfd_reloc_ok.  */
> +static bfd_reloc_status_type
> +riscv_elf_ignore_reloc (bfd *abfd ATTRIBUTE_UNUSED, arelent *reloc_entry,
> +                        asymbol *symbol ATTRIBUTE_UNUSED,
> +                        void *data ATTRIBUTE_UNUSED, asection *input_section,
> +                        bfd *output_bfd, char **error_message ATTRIBUTE_UNUSED)
> +{
> +  if (output_bfd != NULL)
> +    reloc_entry->address += input_section->output_offset;
> +
> +  return bfd_reloc_ok;
> +}
> +
>  /* Parsing subset version.
>  
>     Return Value:
> diff --git a/bfd/libbfd.h b/bfd/libbfd.h
> index 77b732ee4b..7aa8f91504 100644
> --- a/bfd/libbfd.h
> +++ b/bfd/libbfd.h
> @@ -2327,6 +2327,8 @@ static const char *const bfd_reloc_code_real_names[] = { "@@uninitialized@@",
>    "BFD_RELOC_RISCV_SET16",
>    "BFD_RELOC_RISCV_SET32",
>    "BFD_RELOC_RISCV_32_PCREL",
> +  "BFD_RELOC_RISCV_SET_ULEB128",
> +  "BFD_RELOC_RISCV_SUB_ULEB128",
>    "BFD_RELOC_RL78_NEG8",
>    "BFD_RELOC_RL78_NEG16",
>    "BFD_RELOC_RL78_NEG24",
> diff --git a/bfd/reloc.c b/bfd/reloc.c
> index b00b79f319..9b8b75ad21 100644
> --- a/bfd/reloc.c
> +++ b/bfd/reloc.c
> @@ -5212,6 +5212,10 @@ ENUMX
>    BFD_RELOC_RISCV_SET32
>  ENUMX
>    BFD_RELOC_RISCV_32_PCREL
> +ENUMX
> +  BFD_RELOC_RISCV_SET_ULEB128
> +ENUMX
> +  BFD_RELOC_RISCV_SUB_ULEB128
>  ENUMDOC
>    RISC-V relocations.
>  
> diff --git a/gas/ChangeLog b/gas/ChangeLog
> index 09991524da..e53ac0895b 100644
> --- a/gas/ChangeLog
> +++ b/gas/ChangeLog
> @@ -1,3 +1,11 @@
> +2019-11-27  Kuan-Lin Chen  <[hidden email]>
> +
> +        * config/tc-riscv.c (md_apply_fix): Add BFD_RELOC_RISCV_SET_ULEB128 and
> +        BFD_RELOC_RISCV_SUB_ULEB128.
> +        (riscv_insert_uleb128_fixes): New function.
> +        (riscv_md_end): Scan rs_leb128 fragments.
> +        (riscv_pseudo_table): Remove uleb128.
> +
>  2019-11-25  Andrew Pinski  <[hidden email]>
>  
>          * config/tc-aarch64.c (md_begin): Use correct
> diff --git a/gas/config/tc-riscv.c b/gas/config/tc-riscv.c
> index e50505138e..c35c591c0a 100644
> --- a/gas/config/tc-riscv.c
> +++ b/gas/config/tc-riscv.c
> @@ -2392,6 +2392,9 @@ md_apply_fix (fixS *fixP, valueT *valP, segT seg ATTRIBUTE_UNUSED)
>      case BFD_RELOC_RISCV_SUB32:
>      case BFD_RELOC_RISCV_SUB64:
>      case BFD_RELOC_RISCV_RELAX:
> +    /* cvt_frag_to_fill () has called output_leb128 ().  */
> +    case BFD_RELOC_RISCV_SET_ULEB128:
> +    case BFD_RELOC_RISCV_SUB_ULEB128:
>        break;
>  
>      case BFD_RELOC_RISCV_TPREL_HI20:
> @@ -3127,11 +3130,54 @@ riscv_set_public_attributes (void)
>      riscv_write_out_arch_attr ();
>  }
>  
> +/* Scan uleb128 subtraction expressions and insert fixups for them.
> +   e.g., .uleb128 .L1 - .L0
> +   Becuase relaxation may change the value of the subtraction, we
> +   must resolve them in link-time.  */
> +
> +static void
> +riscv_insert_uleb128_fixes (bfd *abfd ATTRIBUTE_UNUSED,
> +                            asection *sec, void *xxx ATTRIBUTE_UNUSED)
> +{
> +  segment_info_type *seginfo = seg_info (sec);
> +  struct frag *fragP;
> +
> +  subseg_set (sec, 0);
> +
> +  for (fragP = seginfo->frchainP->frch_root;
> +       fragP; fragP = fragP->fr_next)
> +    {
> +      expressionS *exp, *exp_dup;
> +
> +      if (fragP->fr_type != rs_leb128  || fragP->fr_symbol == NULL)
> +        continue;
> +
> +      exp = symbol_get_value_expression (fragP->fr_symbol);
> +
> +      if (exp->X_op != O_subtract)
> +        continue;

Presumably that's why I can't get a _SET alone.

> +
> +      /* Only unsigned leb128 can be handle.  */
> +      gas_assert (fragP->fr_subtype == 0);
> +      exp_dup = xmemdup (exp, sizeof (*exp), sizeof (*exp));
> +      exp_dup->X_op = O_symbol;
> +      exp_dup->X_op_symbol = NULL;
> +
> +      /* Insert relocations to resolve the subtraction in link-time.  */
> +      fix_new_exp (fragP, fragP->fr_fix, 0,
> +                   exp_dup, 0, BFD_RELOC_RISCV_SET_ULEB128);
> +      exp_dup->X_add_symbol = exp->X_op_symbol;
> +      fix_new_exp (fragP, fragP->fr_fix, 0,
> +                   exp_dup, 0, BFD_RELOC_RISCV_SUB_ULEB128);
> +    }
> +}
> +
>  /* Called after all assembly has been done.  */
>  
>  void
>  riscv_md_end (void)
>  {
> +  bfd_map_over_sections (stdoutput, riscv_insert_uleb128_fixes, NULL);
>    riscv_set_public_attributes ();
>  }
>  
> @@ -3215,7 +3261,6 @@ static const pseudo_typeS riscv_pseudo_table[] =
>    {"dtprelword", s_dtprel, 4},
>    {"dtpreldword", s_dtprel, 8},
>    {"bss", s_bss, 0},
> -  {"uleb128", s_riscv_leb128, 0},
>    {"sleb128", s_riscv_leb128, 1},
>    {"insn", s_riscv_insn, 0},
>    {"attribute", s_riscv_attribute, 0},
> diff --git a/include/ChangeLog b/include/ChangeLog
> index 47bb86cf71..c290b245c6 100644
> --- a/include/ChangeLog
> +++ b/include/ChangeLog
> @@ -1,3 +1,7 @@
> +2019-11-27  Kuan-Lin Chen  <[hidden email]>
> +
> +        * elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.
> +
>  2019-11-25  Alan Modra  <[hidden email]>
>  
>          * coff/ti.h (GET_SCNHDR_SIZE, PUT_SCNHDR_SIZE, GET_SCN_SCNLEN),
> diff --git a/include/elf/riscv.h b/include/elf/riscv.h
> index 2f98aa4a3e..030679b35f 100644
> --- a/include/elf/riscv.h
> +++ b/include/elf/riscv.h
> @@ -88,6 +88,8 @@ START_RELOC_NUMBERS (elf_riscv_reloc_type)
>    RELOC_NUMBER (R_RISCV_SET16, 55)
>    RELOC_NUMBER (R_RISCV_SET32, 56)
>    RELOC_NUMBER (R_RISCV_32_PCREL, 57)
> +  RELOC_NUMBER (R_RISCV_SET_ULEB128, 58)
> +  RELOC_NUMBER (R_RISCV_SUB_ULEB128, 59)
>  END_RELOC_NUMBERS (R_RISCV_max)
>  
>  /* Processor specific flags for the ELF header e_flags field.  */
> diff --git a/ld/ChangeLog b/ld/ChangeLog
> index 969ab78035..5c2e406101 100644
> --- a/ld/ChangeLog
> +++ b/ld/ChangeLog
> @@ -1,3 +1,9 @@
> +2019-11-27  Kuan-Lin Chen  <[hidden email]>
> +
> +        * testsuite/ld-riscv-elf/ld-riscv-elf.exp: Add uleb128.
> +        * testsuite/ld-riscv-elf/uleb128.d: New test.
> +        * testsuite/ld-riscv-elf/uleb128.s: New file.
> +
>  2019-11-26  Martin Liska  <[hidden email]>
>  
>          * scripttempl/arclinux.sc: Add .text.sorted.* which is sorted
> diff --git a/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp b/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
> index 7aabbdd641..809e40f08d 100644
> --- a/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
> +++ b/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
> @@ -38,6 +38,7 @@ if [istarget "riscv*-*-*"] {
>      run_dump_test "attr-merge-priv-spec"
>      run_dump_test "attr-merge-arch-failed-01"
>      run_dump_test "attr-merge-stack-align-failed"
> +    run_dump_test "uleb128"
>      run_ld_link_tests {
>          { "Weak reference 32" "-T weakref.ld -melf32lriscv" ""
>              "-march=rv32i -mabi=ilp32" {weakref32.s}
> diff --git a/ld/testsuite/ld-riscv-elf/uleb128.d b/ld/testsuite/ld-riscv-elf/uleb128.d
> new file mode 100644
> index 0000000000..a921478e98
> --- /dev/null
> +++ b/ld/testsuite/ld-riscv-elf/uleb128.d
> @@ -0,0 +1,18 @@
> +#source: uleb128.s
> +#as: -march=rv32ic
> +#ld: -melf32lriscv
> +#objdump: -d
> +
> +.*:[         ]+file format .*
> +
> +Disassembly of section .text:
> +
> +.* <_start>:
> +.*jal.*<bar>
> +.*jal.*<bar>
> +.*jal.*<bar>
> +.*jal.*<bar>
> +.*jal.*<bar>
> +.*jal.*<bar>
> +.*:[         ]+0e0c.*
> +#pass
> diff --git a/ld/testsuite/ld-riscv-elf/uleb128.s b/ld/testsuite/ld-riscv-elf/uleb128.s
> new file mode 100644
> index 0000000000..f7d23be163
> --- /dev/null
> +++ b/ld/testsuite/ld-riscv-elf/uleb128.s
> @@ -0,0 +1,18 @@
> +.text
> +.globl bar
> +.globl _start
> +.option rvc
> +.align 2
> +_start:
> +.L0:
> +        .rept 6
> +        call bar
> +        .endr
> +.align 2
> +.L1:
> +        .uleb128 .L1 - .L0
> +        .uleb128 .L2 - .L0
> +.L2:
> +.align 2
> +bar:
> +        nop

It looks like alignment is broken here after uleb128s, but it's not actually
triggering an error in all cases.  Having two of them in this test case doesn't
trigger the issue, but an odd number does.  For example, the following source

    .text
    .globl _start
    .align 2
    _start:
            .uleb128 _start
    .align 2
    _end:
            nop

links _end to a misaligned symbol

    Disassembly of section .text:
   
    0000000000010078 <_start>:
            ...
   
    0000000000010079 <_end>:
       10079:       00000013                nop
       1007d:       0000                    unimp
            ...

I'm assuming that's because norvc assumes 4-byte alignment in text sections, so
it's just eliding the alingment directives.  With RVC I do get an error message

    .text
    .globl _start
    .option rvc
    .align 2
    _start:
            .uleb128 _start
    .align 2
    _end:
            nop

    ./install/bin/riscv64-unknown-linux-gnu-ld: test.o(.text+0x1): 3 bytes required for alignment to 4-byte boundary, but only 2 present
    ./install/bin/riscv64-unknown-linux-gnu-ld: can't relax section: bad value


Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] [RISCV] Support subtraction of .uleb128.

Kuan-Lin Chen
Hi Palmer,

> GCC gives me -Wmaybe-uninitialized for both of these.
May I know your gcc version?
I tried gcc-7.3.0 and gcc-8.2.0, and I didn't get the warning.

> I don't understand how this could possibly work: uleb128s can be long, so we
> can't just treat them as a single byte.
Yes, uleb128s can be long so I implement riscv_elf_ignore_reloc to
ignore the overflow checking.
And I relocate uleb128s by myself in perform_relocation.

> I'm having trouble testing it, though,
> as I can't figure out how to emit R_RISCV_SET_ULEB128 alone.  For example, when
> I try to insert a uleb128 in the assembler I'm not even getting a relocation:
.uleb128 only support subtract format. (.uleb128 .L1 - .L0)
It is impossible to get final the value and the length for ".uleb128
symbol" in assemble-time.
I think ".uleb128 symbol" is meaningless.
Therefore, all targets just fill the value as assemble-time value
without relocations.

>I can get a R_RISCV_SUB_ULEB128, but it appears to be both producing the wrong
>value and garbaging up bytes after the relocation.  For example:

I get the same result in the object file but the different result in executable.
pal.o:     file format elf64-littleriscv

Disassembly of section .text:

0000000000000000 <_start>:
   0:   00000013                nop

0000000000000004 <.L0>:
   4:   00000013                nop

0000000000000008 <uleb_l1_l0>:
   8:   00100013                li      zero,1
   c:   aa04                    fsd     fs1,16(a2)
                        c: R_RISCV_SET_ULEB128  .L1
                        c: R_RISCV_SUB_ULEB128  .L0
   e:   aaaa                    fsd     fa0,336(sp)
  10:   00200013                li      zero,2


a.out:     file format elf64-littleriscv


Disassembly of section .text:

0000000000010078 <_start>:
   10078:       00000013                nop
   1007c:       00000013                nop

0000000000010080 <uleb_l1_l0>:
   10080:       00100013                li      zero,1
   10084:       aa04                    fsd     fs1,16(a2)
   10086:       aaaa                    fsd     fa0,336(sp)
   10088:       00200013                li      zero,2

Maybe there is something mismatched. But I don't have any idea currently.
Could you please give me more information about your flow?
Thanks.

> Presumably that's why I can't get a _SET alone.
I think it's meaningless to use _SET alone in assembly.

> It looks like alignment is broken here after uleb128s, but it's not actually
> triggering an error in all cases.  Having two of them in this test case doesn't
> trigger the issue, but an odd number does.  For example, the following source
I think this alignment issue can be fixed by the following patch:
https://sourceware.org/ml/binutils/2019-12/msg00024.html

Thanks for your review.

Palmer Dabbelt <[hidden email]> 於 2019年12月7日 週六 上午7:47寫道:

>
> On Wed, 27 Nov 2019 00:11:50 PST (-0800), [hidden email] wrote:
> > The data length of uleb128 is variable.  So linker must recalculate the
> > value of the subtraction.  The patch leave relocations in object files
> > so that linker can relocate again after relaxation.
> >
> > bfd/ChangeLog:
> > * bfd-in2.h: Regenerated.
> > * elfnn-riscv.c (write_uleb128): New function.
> > (perform_relocation): Perform R_RISCV_SUB_ULEB128 and
> > R_RISCV_SET_ULEB128 relocation.
> > (riscv_elf_relocate_section): Likewise.
> > * elfxx-riscv.c (howto_table): Add R_RISCV_SUB_ULEB128 and
> > R_RISCV_SET_ULEB128.
> > (riscv_elf_ignore_reloc): New function.
> > * libbfd.h: Regenerated.
> > * reloc.c (BFD_RELOC_RISCV_SET_ULEB128, BFD_RELOC_RISCV_SUB_ULEB128):
> > New relocations to support .uleb128.
> >
> > gas/ChangeLog:
> > * config/tc-riscv.c (md_apply_fix): Add BFD_RELOC_RISCV_SET_ULEB128 and
> > BFD_RELOC_RISCV_SUB_ULEB128.
> > (riscv_insert_uleb128_fixes): New function.
> > (riscv_md_end): Scan rs_leb128 fragments.
> > (riscv_pseudo_table): Remove uleb128.
> >
> > include/ChangeLog:
> > * elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.
> >
> > ld/ChangeLog:
> > * testsuite/ld-riscv-elf/ld-riscv-elf.exp: Add uleb128.
> > * testsuite/ld-riscv-elf/uleb128.d: New test.
> > * testsuite/ld-riscv-elf/uleb128.s: New file.
> >
> > OK to commit?
>
> I'm having a lot of trouble trying to figure out how this works.  LMK if I'm
> missing something, there's some comments in line:
>
> > From 9f7f0aaa484dee2b32c550f8cca4de18959f01e1 Mon Sep 17 00:00:00 2001
> > From: Kuan-Lin Chen <[hidden email]>
> > Date: Thu, 14 Nov 2019 14:24:22 +0800
> > Subject: [PATCH] RISC-V: Support subtraction of .uleb128.
> >
> > The data length of uleb128 is variable.  So linker must recalculate the
> > value of the subtraction.  The patch leave relocations in object files
> > so that linker can relocate again after relaxation.
> >
> > bfd/ChangeLog:
> >         * bfd-in2.h: Regenerated.
> >         * elfnn-riscv.c (write_uleb128): New function.
> >         (perform_relocation): Perform R_RISCV_SUB_ULEB128 and
> >          R_RISCV_SET_ULEB128 relocation.
> >         (riscv_elf_relocate_section): Likewise.
> >         * elfxx-riscv.c (howto_table): Add R_RISCV_SUB_ULEB128 and
> >         R_RISCV_SET_ULEB128.
> >         (riscv_elf_ignore_reloc): New function.
> >         * libbfd.h: Regenerated.
> >         * reloc.c (BFD_RELOC_RISCV_SET_ULEB128, BFD_RELOC_RISCV_SUB_ULEB128):
> >         New relocations to support .uleb128.
> >
> > gas/ChangeLog:
> >         * config/tc-riscv.c (md_apply_fix): Add BFD_RELOC_RISCV_SET_ULEB128 and
> >         BFD_RELOC_RISCV_SUB_ULEB128.
> >         (riscv_insert_uleb128_fixes): New function.
> >         (riscv_md_end): Scan rs_leb128 fragments.
> >         (riscv_pseudo_table): Remove uleb128.
> >
> > include/ChangeLog:
> >         * elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.
> >
> > ld/ChangeLog:
> >         * testsuite/ld-riscv-elf/ld-riscv-elf.exp: Add uleb128.
> >         * testsuite/ld-riscv-elf/uleb128.d: New test.
> >         * testsuite/ld-riscv-elf/uleb128.s: New file.
> > ---
> >  bfd/ChangeLog                              | 14 ++++
> >  bfd/bfd-in2.h                              |  2 +
> >  bfd/elfnn-riscv.c                          | 82 ++++++++++++++++++++++
> >  bfd/elfxx-riscv.c                          | 51 ++++++++++++++
> >  bfd/libbfd.h                               |  2 +
> >  bfd/reloc.c                                |  4 ++
> >  gas/ChangeLog                              |  8 +++
> >  gas/config/tc-riscv.c                      | 47 ++++++++++++-
> >  include/ChangeLog                          |  4 ++
> >  include/elf/riscv.h                        |  2 +
> >  ld/ChangeLog                               |  6 ++
> >  ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp |  1 +
> >  ld/testsuite/ld-riscv-elf/uleb128.d        | 18 +++++
> >  ld/testsuite/ld-riscv-elf/uleb128.s        | 18 +++++
> >  14 files changed, 258 insertions(+), 1 deletion(-)
> >  create mode 100644 ld/testsuite/ld-riscv-elf/uleb128.d
> >  create mode 100644 ld/testsuite/ld-riscv-elf/uleb128.s
> >
> > diff --git a/bfd/ChangeLog b/bfd/ChangeLog
> > index 4a0852e577..d11ffde74c 100644
> > --- a/bfd/ChangeLog
> > +++ b/bfd/ChangeLog
> > @@ -1,3 +1,17 @@
> > +2019-11-27  Kuan-Lin Chen  <[hidden email]>
> > +
> > +        * bfd-in2.h: Regenerated.
> > +        * elfnn-riscv.c (write_uleb128): New function.
> > +        (perform_relocation): Perform R_RISCV_SUB_ULEB128 and
> > +         R_RISCV_SET_ULEB128 relocation.
> > +        (riscv_elf_relocate_section): Likewise.
> > +        * elfxx-riscv.c (howto_table): Add R_RISCV_SUB_ULEB128 and
> > +        R_RISCV_SET_ULEB128.
> > +        (riscv_elf_ignore_reloc): New function.
> > +        * libbfd.h: Regenerated.
> > +        * reloc.c (BFD_RELOC_RISCV_SET_ULEB128, BFD_RELOC_RISCV_SUB_ULEB128):
> > +        New relocations to support .uleb128.
> > +
> >  2019-11-27  Alan Modra  <[hidden email]>
> >
> >          PR 23652
> > diff --git a/bfd/bfd-in2.h b/bfd/bfd-in2.h
> > index 44902fc8d0..a32708ebc2 100644
> > --- a/bfd/bfd-in2.h
> > +++ b/bfd/bfd-in2.h
> > @@ -4380,6 +4380,8 @@ number for the SBIC, SBIS, SBI and CBI instructions  */
> >    BFD_RELOC_RISCV_SET16,
> >    BFD_RELOC_RISCV_SET32,
> >    BFD_RELOC_RISCV_32_PCREL,
> > +  BFD_RELOC_RISCV_SET_ULEB128,
> > +  BFD_RELOC_RISCV_SUB_ULEB128,
> >
> >  /* Renesas RL78 Relocations.  */
> >    BFD_RELOC_RL78_NEG8,
> > diff --git a/bfd/elfnn-riscv.c b/bfd/elfnn-riscv.c
> > index 997f786602..075ef6b82d 100644
> > --- a/bfd/elfnn-riscv.c
> > +++ b/bfd/elfnn-riscv.c
> > @@ -1419,6 +1419,26 @@ riscv_global_pointer_value (struct bfd_link_info *info)
> >    return h->u.def.value + sec_addr (h->u.def.section);
> >  }
> >
> > +/* Write VAL in uleb128 format to P, returning a pointer to the
> > +   following byte.
> > +   This code is copied from elf-attr.c.  */
> > +
> > +static bfd_byte *
> > +write_uleb128 (bfd_byte *p, unsigned int val)
> > +{
> > +  bfd_byte c;
> > +  do
> > +    {
> > +      c = val & 0x7f;
> > +      val >>= 7;
> > +      if (val)
> > +        c |= 0x80;
> > +      *(p++) = c;
> > +    }
> > +  while (val);
> > +  return p;
> > +}
> > +
> >  /* Emplace a static relocation.  */
> >
> >  static bfd_reloc_status_type
> > @@ -1512,6 +1532,25 @@ perform_relocation (const reloc_howto_type *howto,
> >          value = ENCODE_RVC_LUI_IMM (RISCV_CONST_HIGH_PART (value));
> >        break;
> >
> > +    case R_RISCV_SET_ULEB128:
> > +    case R_RISCV_SUB_ULEB128:
> > +      {
> > +        unsigned int len = 0;
> > +        bfd_byte *endp, *p;
> > +
> > +        _bfd_read_unsigned_leb128 (input_bfd, contents + rel->r_offset, &len);
> > +
> > +        /* Clean the contents value to zero.  Do not reduce the length.  */
> > +        p = contents + rel->r_offset;
> > +        endp = p + len -1;
> > +        memset (p, 0x80, len);
> > +        *(endp) = 0;
> > +        p = write_uleb128 (p, value) - 1;
> > +        if (p < endp)
> > +          *p |= 0x80;
> > +        return bfd_reloc_ok;
> > +      }
> > +
> >      case R_RISCV_32:
> >      case R_RISCV_64:
> >      case R_RISCV_ADD8:
> > @@ -1767,6 +1806,8 @@ riscv_elf_relocate_section (bfd *output_bfd,
> >    struct elf_link_hash_entry **sym_hashes = elf_sym_hashes (input_bfd);
> >    bfd_vma *local_got_offsets = elf_local_got_offsets (input_bfd);
> >    bfd_boolean absolute;
> > +  bfd_vma uleb128_vma;
> > +  Elf_Internal_Rela *uleb128_rel = NULL;
> >
> >    if (!riscv_init_pcrel_relocs (&pcrel_relocs))
> >      return FALSE;
> > @@ -1871,6 +1912,47 @@ riscv_elf_relocate_section (bfd *output_bfd,
> >          case R_RISCV_DELETE:
> >            /* These require no special handling beyond perform_relocation.  */
> >            break;
> > +        case R_RISCV_SET_ULEB128:
> > +          if (!uleb128_rel)
> > +            {
> > +              /* Save the minuend to use later.  */
> > +              uleb128_vma = relocation;
> > +              uleb128_rel = rel;
> > +              continue;
> > +            }
> > +          else
> > +            {
> > +              if (uleb128_rel->r_offset != rel->r_offset)
> > +                {
> > +                  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
> > +                                         input_bfd, howto->name);
> > +                  bfd_set_error (bfd_error_bad_value);
> > +                }
> > +              relocation = relocation - uleb128_vma;
> > +              uleb128_rel = NULL;
> > +              break;
> > +            }
> > +
> > +        case R_RISCV_SUB_ULEB128:
> > +          if (uleb128_rel)
> > +            {
> > +              if (uleb128_rel->r_offset != rel->r_offset)
> > +                {
> > +                  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
> > +                                         input_bfd, howto->name);
> > +                  bfd_set_error (bfd_error_bad_value);
> > +                }
> > +              relocation = uleb128_vma - relocation;
>
> GCC gives me -Wmaybe-uninitialized for both of these.
>
> > +              uleb128_rel = NULL;
> > +              break;
> > +            }
> > +          else
> > +            {
> > +              /* Save the subtrahend to use later.  */
> > +              uleb128_vma = relocation;
> > +              uleb128_rel = rel;
> > +              continue;
> > +            }
> >
> >          case R_RISCV_GOT_HI20:
> >            if (h != NULL)
> > diff --git a/bfd/elfxx-riscv.c b/bfd/elfxx-riscv.c
> > index 245717f70f..c136ba2207 100644
> > --- a/bfd/elfxx-riscv.c
> > +++ b/bfd/elfxx-riscv.c
> > @@ -38,6 +38,8 @@
> >     relocations for the debug info.  */
> >  static bfd_reloc_status_type riscv_elf_add_sub_reloc
> >    (bfd *, arelent *, asymbol *, void *, asection *, bfd *, char **);
> > +static bfd_reloc_status_type riscv_elf_ignore_reloc
> > +  (bfd *, arelent *, asymbol *, void *, asection *, bfd *, char **);
> >
> >  /* The relocation table used for SHT_RELA sections.  */
> >
> > @@ -855,8 +857,41 @@ static reloc_howto_type howto_table[] =
> >           0,                                /* src_mask */
> >           MINUS_ONE,                        /* dst_mask */
> >           FALSE),                        /* pcrel_offset */
> > +
> > +  /* The length of unsigned-leb128 is variant, just assume the
> > +     size is one byte here.  */
> > +  HOWTO (R_RISCV_SET_ULEB128,                /* type */
> > +         0,                                /* rightshift */
> > +         0,                                /* size */
> > +         0,                                /* bitsize */
> > +         FALSE,                                /* pc_relative */
> > +         0,                                /* bitpos */
> > +         complain_overflow_dont,        /* complain_on_overflow */
> > +         riscv_elf_ignore_reloc,        /* special_function */
> > +         "R_RISCV_SET_ULEB128",                /* name */
> > +         FALSE,                                /* partial_inplace */
> > +         0,                                /* src_mask */
> > +         0,                                /* dst_mask */
> > +         FALSE),                        /* pcrel_offset */
>
> I don't understand how this could possibly work: uleb128s can be long, so we
> can't just treat them as a single byte.  I'm having trouble testing it, though,
> as I can't figure out how to emit R_RISCV_SET_ULEB128 alone.  For example, when
> I try to insert a uleb128 in the assembler I'm not even getting a relocation:
>
>     .text
>     .globl _start
>     _start:
>             nop
>     .L0:
>             nop
>     .L1:
>     uleb_l0:
>             addi x0, x0, 1
>             .uleb128 .L0
>             # Manually fix up the alignment, as .align is broken.
>             .rept 3
>             .byte 0xAA
>             .endr
>             addi x0, x0, 2
>
>     $ objdump -dr
>     test.o:     file format elf64-littleriscv
>
>
>     Disassembly of section .text:
>
>     0000000000000000 <_start>:
>        0:   00000013                nop
>        4:   00000013                nop
>
>     0000000000000008 <uleb_l0>:
>        8:   00100013                li      zero,1
>        c:   aa04                    fsd     fs1,16(a2)
>        e:   aaaa                    fsd     fa0,336(sp)
>       10:   00200013                li      zero,2
>
> so then when I link I'm just getting the wrong value entirely.
>
>     test:     file format elf64-littleriscv
>
>
>     Disassembly of section .text:
>
>     0000000000010078 <_start>:
>        10078:       00000013                nop
>        1007c:       00000013                nop
>
>     0000000000010080 <uleb_l0>:
>        10080:       00100013                li      zero,1
>        10084:       aa04                    fsd     fs1,16(a2)
>        10086:       aaaa                    fsd     fa0,336(sp)
>        10088:       00200013                li      zero,2
>
> > +  /* The length of unsigned-leb128 is variant, just assume the
> > +     size is one byte here.  */
> > +  HOWTO (R_RISCV_SUB_ULEB128,                /* type */
> > +         0,                                /* rightshift */
> > +         0,                                /* size */
> > +         0,                                /* bitsize */
> > +         FALSE,                                /* pc_relative */
> > +         0,                                /* bitpos */
> > +         complain_overflow_dont,        /* complain_on_overflow */
> > +         riscv_elf_ignore_reloc,        /* special_function */
> > +         "R_RISCV_SUB_ULEB128",                /* name */
> > +         FALSE,                                /* partial_inplace */
> > +         0,                                /* src_mask */
> > +         0,                                /* dst_mask */
> > +         FALSE),                        /* pcrel_offset */
> >  };
>
> I can get a R_RISCV_SUB_ULEB128, but it appears to be both producing the wrong
> value and garbaging up bytes after the relocation.  For example:
>
>     .text
>     .globl _start
>     _start:
>             nop
>     .L0:
>             nop
>     .L1:
>     uleb_l1_l0:
>             addi x0, x0, 1
>             .uleb128 .L1 - .L0
>             # Manually fix up the alignment, as .align is broken
>             .rept 3
>             .byte 0xAA
>             .endr
>             addi x0, x0, 2
>
> produces the relocation
>
>     test.o:     file format elf64-littleriscv
>
>
>     Disassembly of section .text:
>
>     0000000000000000 <_start>:
>        0:   00000013                nop
>
>     0000000000000004 <.L0>:
>        4:   00000013                nop
>
>     0000000000000008 <uleb_l1_l0>:
>        8:   00100013                li      zero,1
>        c:   aa04                    fsd     fs1,16(a2)
>                             c: R_RISCV_SET_ULEB128  .L1
>                             c: R_RISCV_SUB_ULEB128  .L0
>        e:   aaaa                    fsd     fa0,336(sp)
>       10:   00200013                li      zero,2
>
> but then goes and corrupts the resulting executable
>
>
>     test:     file format elf64-littleriscv
>     Disassembly of section .text:
>
>     0000000000010078 <_start>:
>        10078:       00000013                nop
>        1007c:       00000013                nop
>
>     0000000000010080 <uleb_l1_l0>:
>        10080:       00100013                li      zero,1
>        10084:       ff84                    sd      s1,56(a5)
>        10086:       000ffffb                0xffffb
>        1008a:       0020                    addi    s0,sp,8
>
> I can't think of a way to make this work aside from just emitting uleb128s at
> their full length, which would make them work but would really defeat the
> purpose of the format.  I guess we could then relax these, but that seems like
> a lot of work.
>
> > +
> >  /* A mapping from BFD reloc types to RISC-V ELF reloc types.  */
> >
> >  struct elf_reloc_map
> > @@ -917,6 +952,8 @@ static const struct elf_reloc_map riscv_reloc_map[] =
> >    { BFD_RELOC_RISCV_SET16, R_RISCV_SET16 },
> >    { BFD_RELOC_RISCV_SET32, R_RISCV_SET32 },
> >    { BFD_RELOC_RISCV_32_PCREL, R_RISCV_32_PCREL },
> > +  { BFD_RELOC_RISCV_SET_ULEB128, R_RISCV_SET_ULEB128 },
> > +  { BFD_RELOC_RISCV_SUB_ULEB128, R_RISCV_SUB_ULEB128 },
> >  };
> >
> >  /* Given a BFD reloc type, return a howto structure.  */
> > @@ -1011,6 +1048,20 @@ riscv_elf_add_sub_reloc (bfd *abfd,
> >    return bfd_reloc_ok;
> >  }
> >
> > +/* Special handler for relocations which don't have to be relocated.
> > +   This function just simply return bfd_reloc_ok.  */
> > +static bfd_reloc_status_type
> > +riscv_elf_ignore_reloc (bfd *abfd ATTRIBUTE_UNUSED, arelent *reloc_entry,
> > +                        asymbol *symbol ATTRIBUTE_UNUSED,
> > +                        void *data ATTRIBUTE_UNUSED, asection *input_section,
> > +                        bfd *output_bfd, char **error_message ATTRIBUTE_UNUSED)
> > +{
> > +  if (output_bfd != NULL)
> > +    reloc_entry->address += input_section->output_offset;
> > +
> > +  return bfd_reloc_ok;
> > +}
> > +
> >  /* Parsing subset version.
> >
> >     Return Value:
> > diff --git a/bfd/libbfd.h b/bfd/libbfd.h
> > index 77b732ee4b..7aa8f91504 100644
> > --- a/bfd/libbfd.h
> > +++ b/bfd/libbfd.h
> > @@ -2327,6 +2327,8 @@ static const char *const bfd_reloc_code_real_names[] = { "@@uninitialized@@",
> >    "BFD_RELOC_RISCV_SET16",
> >    "BFD_RELOC_RISCV_SET32",
> >    "BFD_RELOC_RISCV_32_PCREL",
> > +  "BFD_RELOC_RISCV_SET_ULEB128",
> > +  "BFD_RELOC_RISCV_SUB_ULEB128",
> >    "BFD_RELOC_RL78_NEG8",
> >    "BFD_RELOC_RL78_NEG16",
> >    "BFD_RELOC_RL78_NEG24",
> > diff --git a/bfd/reloc.c b/bfd/reloc.c
> > index b00b79f319..9b8b75ad21 100644
> > --- a/bfd/reloc.c
> > +++ b/bfd/reloc.c
> > @@ -5212,6 +5212,10 @@ ENUMX
> >    BFD_RELOC_RISCV_SET32
> >  ENUMX
> >    BFD_RELOC_RISCV_32_PCREL
> > +ENUMX
> > +  BFD_RELOC_RISCV_SET_ULEB128
> > +ENUMX
> > +  BFD_RELOC_RISCV_SUB_ULEB128
> >  ENUMDOC
> >    RISC-V relocations.
> >
> > diff --git a/gas/ChangeLog b/gas/ChangeLog
> > index 09991524da..e53ac0895b 100644
> > --- a/gas/ChangeLog
> > +++ b/gas/ChangeLog
> > @@ -1,3 +1,11 @@
> > +2019-11-27  Kuan-Lin Chen  <[hidden email]>
> > +
> > +        * config/tc-riscv.c (md_apply_fix): Add BFD_RELOC_RISCV_SET_ULEB128 and
> > +        BFD_RELOC_RISCV_SUB_ULEB128.
> > +        (riscv_insert_uleb128_fixes): New function.
> > +        (riscv_md_end): Scan rs_leb128 fragments.
> > +        (riscv_pseudo_table): Remove uleb128.
> > +
> >  2019-11-25  Andrew Pinski  <[hidden email]>
> >
> >          * config/tc-aarch64.c (md_begin): Use correct
> > diff --git a/gas/config/tc-riscv.c b/gas/config/tc-riscv.c
> > index e50505138e..c35c591c0a 100644
> > --- a/gas/config/tc-riscv.c
> > +++ b/gas/config/tc-riscv.c
> > @@ -2392,6 +2392,9 @@ md_apply_fix (fixS *fixP, valueT *valP, segT seg ATTRIBUTE_UNUSED)
> >      case BFD_RELOC_RISCV_SUB32:
> >      case BFD_RELOC_RISCV_SUB64:
> >      case BFD_RELOC_RISCV_RELAX:
> > +    /* cvt_frag_to_fill () has called output_leb128 ().  */
> > +    case BFD_RELOC_RISCV_SET_ULEB128:
> > +    case BFD_RELOC_RISCV_SUB_ULEB128:
> >        break;
> >
> >      case BFD_RELOC_RISCV_TPREL_HI20:
> > @@ -3127,11 +3130,54 @@ riscv_set_public_attributes (void)
> >      riscv_write_out_arch_attr ();
> >  }
> >
> > +/* Scan uleb128 subtraction expressions and insert fixups for them.
> > +   e.g., .uleb128 .L1 - .L0
> > +   Becuase relaxation may change the value of the subtraction, we
> > +   must resolve them in link-time.  */
> > +
> > +static void
> > +riscv_insert_uleb128_fixes (bfd *abfd ATTRIBUTE_UNUSED,
> > +                            asection *sec, void *xxx ATTRIBUTE_UNUSED)
> > +{
> > +  segment_info_type *seginfo = seg_info (sec);
> > +  struct frag *fragP;
> > +
> > +  subseg_set (sec, 0);
> > +
> > +  for (fragP = seginfo->frchainP->frch_root;
> > +       fragP; fragP = fragP->fr_next)
> > +    {
> > +      expressionS *exp, *exp_dup;
> > +
> > +      if (fragP->fr_type != rs_leb128  || fragP->fr_symbol == NULL)
> > +        continue;
> > +
> > +      exp = symbol_get_value_expression (fragP->fr_symbol);
> > +
> > +      if (exp->X_op != O_subtract)
> > +        continue;
>
> Presumably that's why I can't get a _SET alone.
>
> > +
> > +      /* Only unsigned leb128 can be handle.  */
> > +      gas_assert (fragP->fr_subtype == 0);
> > +      exp_dup = xmemdup (exp, sizeof (*exp), sizeof (*exp));
> > +      exp_dup->X_op = O_symbol;
> > +      exp_dup->X_op_symbol = NULL;
> > +
> > +      /* Insert relocations to resolve the subtraction in link-time.  */
> > +      fix_new_exp (fragP, fragP->fr_fix, 0,
> > +                   exp_dup, 0, BFD_RELOC_RISCV_SET_ULEB128);
> > +      exp_dup->X_add_symbol = exp->X_op_symbol;
> > +      fix_new_exp (fragP, fragP->fr_fix, 0,
> > +                   exp_dup, 0, BFD_RELOC_RISCV_SUB_ULEB128);
> > +    }
> > +}
> > +
> >  /* Called after all assembly has been done.  */
> >
> >  void
> >  riscv_md_end (void)
> >  {
> > +  bfd_map_over_sections (stdoutput, riscv_insert_uleb128_fixes, NULL);
> >    riscv_set_public_attributes ();
> >  }
> >
> > @@ -3215,7 +3261,6 @@ static const pseudo_typeS riscv_pseudo_table[] =
> >    {"dtprelword", s_dtprel, 4},
> >    {"dtpreldword", s_dtprel, 8},
> >    {"bss", s_bss, 0},
> > -  {"uleb128", s_riscv_leb128, 0},
> >    {"sleb128", s_riscv_leb128, 1},
> >    {"insn", s_riscv_insn, 0},
> >    {"attribute", s_riscv_attribute, 0},
> > diff --git a/include/ChangeLog b/include/ChangeLog
> > index 47bb86cf71..c290b245c6 100644
> > --- a/include/ChangeLog
> > +++ b/include/ChangeLog
> > @@ -1,3 +1,7 @@
> > +2019-11-27  Kuan-Lin Chen  <[hidden email]>
> > +
> > +        * elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.
> > +
> >  2019-11-25  Alan Modra  <[hidden email]>
> >
> >          * coff/ti.h (GET_SCNHDR_SIZE, PUT_SCNHDR_SIZE, GET_SCN_SCNLEN),
> > diff --git a/include/elf/riscv.h b/include/elf/riscv.h
> > index 2f98aa4a3e..030679b35f 100644
> > --- a/include/elf/riscv.h
> > +++ b/include/elf/riscv.h
> > @@ -88,6 +88,8 @@ START_RELOC_NUMBERS (elf_riscv_reloc_type)
> >    RELOC_NUMBER (R_RISCV_SET16, 55)
> >    RELOC_NUMBER (R_RISCV_SET32, 56)
> >    RELOC_NUMBER (R_RISCV_32_PCREL, 57)
> > +  RELOC_NUMBER (R_RISCV_SET_ULEB128, 58)
> > +  RELOC_NUMBER (R_RISCV_SUB_ULEB128, 59)
> >  END_RELOC_NUMBERS (R_RISCV_max)
> >
> >  /* Processor specific flags for the ELF header e_flags field.  */
> > diff --git a/ld/ChangeLog b/ld/ChangeLog
> > index 969ab78035..5c2e406101 100644
> > --- a/ld/ChangeLog
> > +++ b/ld/ChangeLog
> > @@ -1,3 +1,9 @@
> > +2019-11-27  Kuan-Lin Chen  <[hidden email]>
> > +
> > +        * testsuite/ld-riscv-elf/ld-riscv-elf.exp: Add uleb128.
> > +        * testsuite/ld-riscv-elf/uleb128.d: New test.
> > +        * testsuite/ld-riscv-elf/uleb128.s: New file.
> > +
> >  2019-11-26  Martin Liska  <[hidden email]>
> >
> >          * scripttempl/arclinux.sc: Add .text.sorted.* which is sorted
> > diff --git a/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp b/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
> > index 7aabbdd641..809e40f08d 100644
> > --- a/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
> > +++ b/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
> > @@ -38,6 +38,7 @@ if [istarget "riscv*-*-*"] {
> >      run_dump_test "attr-merge-priv-spec"
> >      run_dump_test "attr-merge-arch-failed-01"
> >      run_dump_test "attr-merge-stack-align-failed"
> > +    run_dump_test "uleb128"
> >      run_ld_link_tests {
> >          { "Weak reference 32" "-T weakref.ld -melf32lriscv" ""
> >              "-march=rv32i -mabi=ilp32" {weakref32.s}
> > diff --git a/ld/testsuite/ld-riscv-elf/uleb128.d b/ld/testsuite/ld-riscv-elf/uleb128.d
> > new file mode 100644
> > index 0000000000..a921478e98
> > --- /dev/null
> > +++ b/ld/testsuite/ld-riscv-elf/uleb128.d
> > @@ -0,0 +1,18 @@
> > +#source: uleb128.s
> > +#as: -march=rv32ic
> > +#ld: -melf32lriscv
> > +#objdump: -d
> > +
> > +.*:[         ]+file format .*
> > +
> > +Disassembly of section .text:
> > +
> > +.* <_start>:
> > +.*jal.*<bar>
> > +.*jal.*<bar>
> > +.*jal.*<bar>
> > +.*jal.*<bar>
> > +.*jal.*<bar>
> > +.*jal.*<bar>
> > +.*:[         ]+0e0c.*
> > +#pass
> > diff --git a/ld/testsuite/ld-riscv-elf/uleb128.s b/ld/testsuite/ld-riscv-elf/uleb128.s
> > new file mode 100644
> > index 0000000000..f7d23be163
> > --- /dev/null
> > +++ b/ld/testsuite/ld-riscv-elf/uleb128.s
> > @@ -0,0 +1,18 @@
> > +.text
> > +.globl bar
> > +.globl _start
> > +.option rvc
> > +.align 2
> > +_start:
> > +.L0:
> > +        .rept 6
> > +        call bar
> > +        .endr
> > +.align 2
> > +.L1:
> > +        .uleb128 .L1 - .L0
> > +        .uleb128 .L2 - .L0
> > +.L2:
> > +.align 2
> > +bar:
> > +        nop
>
> It looks like alignment is broken here after uleb128s, but it's not actually
> triggering an error in all cases.  Having two of them in this test case doesn't
> trigger the issue, but an odd number does.  For example, the following source
>
>     .text
>     .globl _start
>     .align 2
>     _start:
>             .uleb128 _start
>     .align 2
>     _end:
>             nop
>
> links _end to a misaligned symbol
>
>     Disassembly of section .text:
>
>     0000000000010078 <_start>:
>             ...
>
>     0000000000010079 <_end>:
>        10079:       00000013                nop
>        1007d:       0000                    unimp
>             ...
>
> I'm assuming that's because norvc assumes 4-byte alignment in text sections, so
> it's just eliding the alingment directives.  With RVC I do get an error message
>
>     .text
>     .globl _start
>     .option rvc
>     .align 2
>     _start:
>             .uleb128 _start
>     .align 2
>     _end:
>             nop
>
>     ./install/bin/riscv64-unknown-linux-gnu-ld: test.o(.text+0x1): 3 bytes required for alignment to 4-byte boundary, but only 2 present
>     ./install/bin/riscv64-unknown-linux-gnu-ld: can't relax section: bad value
>
>


--
Best regards,
Kuan-Lin Chen.
[hidden email]
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] [RISCV] Support subtraction of .uleb128.

Jim Wilson-2
I tried the patch, and it works for me.  It even fixes one gcc
testsuite failure oddly enough, gcc.dg/debug/dwarf2/inline5.c, I
didn't look at the details of why.

I see that there is no actual relaxation of the uleb128.  We just pick
the size at assembly time via a variant frag, and then assuming that
the value can never increase during relaxation, we just need to add
enough leading zeros at link time to ensure that the size of the
uleb128 doesn't change if the value gets smaller.  It is a little
worrying that we can only support uleb128 and not sleb128 here, since
this trick won't work for sleb128, but I don't think that is a
problem.  Just handling uleb128 is good enough to allow gcc to use
uleb128/sleb128, as we only use subtraction on addresses, and they are
always treated as unsigned.  I see that object files get a bit larger
due to the extra relocations, but executables end up smaller because
the gcc_except_table is now smaller.  A simple rv32 C++ hello world
program linked against a static libstdc++ reduces in size by about
2.4% with this patch.

The assumption that the uleb128 value always decreases could be wrong
if we are subtracting a text section address from a data section
address, because the size of alignment padding between the two can
increase if the text section size decreases.  This code could fail in
this case.  I don't believe that gcc ever does this, but an end user
could conceivably write assembly code that does this which could be a
problem.  We could handle this in the assembler by forcing the uleb128
size to the pointer size if we are subtracting two pointers that
aren't in the same section.  I think that being in the same section is
enough, and that we don't have to require them to be in the same frag
because of how we handle alignment when relaxing.  That would make the
patch safer, and shouldn't change the size reduction we get with it.
This may require some inconvenient code duplication though, since you
are currently using the generic s_leb128 support which doesn't do
this.  Unless maybe we can add a target dependent hook there to avoid
copying it.

There is the issue I mentioned before that I'd like to see the psABI
updated before we modify binutils to use the new relocs.  Letting gcc
emit code that llvm can't link will cause problems.  Once this patch
goes in, gcc will automatically start emitting the new relocs because
a configure test that currently fails will start working.  I added
some info to the psABI issue to help, but I think we really need some
feedback from the LLVM folks before we commit this.

Otherwise, I don't see anything wrong with the patch other than a few
style issues.  These are all minor problems.

+       endp = p + len -1;
+       memset (p, 0x80, len);
+       *(endp) = 0;

Seem like the memset should be len - 1 but it seems harmless because we
just set the *endp value twice.

+  /* The length of unsigned-leb128 is variant, just assume the
+     size is one byte here.  */

I'd use "variable" instead of "variant", but that is just a style
issue.  This happens in two places.

+   Becuase relaxation may change the value of the subtraction, we
+   must resolve them in link-time.  */

Becuase -> Because
in -> at

+      /* Only unsigned leb128 can be handle.  */

handle -> handled

+      /* Insert relocations to resolve the subtraction in link-time.  */

in -> at

Jim
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] [RISCV] Support subtraction of .uleb128.

Nelson Chu-2
There is one minor thing.  Maciej has pointed out before that it's
better done with one the percent-codes to `_bfd_error_handler' rather
than aborting the link right away, so that any further link errors are
also reported and you don't have to shake them out one by one.  So
report the relocation error via linker's callback function seems to be
better.  I think maybe we can report a dangerous relocation for the
mismatched R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128.  What I meant
is that,

+ case R_RISCV_SET_ULEB128:
...
+  else
+    {
+      if (uleb128_rel->r_offset != rel->r_offset)
+ {
-  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
- input_bfd, howto->name);
-  bfd_set_error (bfd_error_bad_value);
+ msg = ("R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128 are mismatched. ");
+ r = bfd_reloc_dangerous;
+ break;
+ }
+      relocation = relocation - uleb128_vma;
+      uleb128_rel = NULL;
+      break;
+    }
+
+ case R_RISCV_SUB_ULEB128:
+  if (uleb128_rel)
+    {
+      if (uleb128_rel->r_offset != rel->r_offset)
+ {
-  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
- input_bfd, howto->name);
-  bfd_set_error (bfd_error_bad_value);
+ msg = ("R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128 are mismatched. ");
+ r = bfd_reloc_dangerous;
+ break;
+ }
+      relocation = uleb128_vma - relocation;
+      uleb128_rel = NULL;
+      break;
+    }

Thanks and regards
Nelson
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] [RISCV] Support subtraction of .uleb128.

Kuan-Lin Chen
Hi,

The patch is fixed some problems as your comments in previous mails.

>The assumption that the uleb128 address size can only decrease with relaxation is true only if both addresses are in the same section.
I found this kind of subtract is banned in cvt_frag_to_fill for
rs_leb128.  Therefore, it doesn't have to restrict by ourselves.

Thanks for your reviews.

Nelson Chu <[hidden email]> 於 2019年12月18日 週三 上午9:50寫道:

>
> There is one minor thing.  Maciej has pointed out before that it's
> better done with one the percent-codes to `_bfd_error_handler' rather
> than aborting the link right away, so that any further link errors are
> also reported and you don't have to shake them out one by one.  So
> report the relocation error via linker's callback function seems to be
> better.  I think maybe we can report a dangerous relocation for the
> mismatched R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128.  What I meant
> is that,
>
> + case R_RISCV_SET_ULEB128:
> ...
> +  else
> +    {
> +      if (uleb128_rel->r_offset != rel->r_offset)
> + {
> -  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
> - input_bfd, howto->name);
> -  bfd_set_error (bfd_error_bad_value);
> + msg = ("R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128 are mismatched. ");
> + r = bfd_reloc_dangerous;
> + break;
> + }
> +      relocation = relocation - uleb128_vma;
> +      uleb128_rel = NULL;
> +      break;
> +    }
> +
> + case R_RISCV_SUB_ULEB128:
> +  if (uleb128_rel)
> +    {
> +      if (uleb128_rel->r_offset != rel->r_offset)
> + {
> -  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
> - input_bfd, howto->name);
> -  bfd_set_error (bfd_error_bad_value);
> + msg = ("R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128 are mismatched. ");
> + r = bfd_reloc_dangerous;
> + break;
> + }
> +      relocation = uleb128_vma - relocation;
> +      uleb128_rel = NULL;
> +      break;
> +    }
>
> Thanks and regards
> Nelson


--
Best regards,
Kuan-Lin Chen.
[hidden email]

0001-RISC-V-Support-subtraction-of-.uleb128.patch (19K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] [RISCV] Support subtraction of .uleb128.

Kito Cheng-2
Hi Kuan-Lin:

> * testsuite/ld-riscv-elf/uleb128.d: New test.
> * testsuite/ld-riscv-elf/uleb128.s: New file.
Those two files are not included in the patch, maybe you forgot add?

On Mon, Jan 6, 2020 at 4:18 PM Kuan-Lin Chen <[hidden email]>
wrote:

> Hi,
>
> The patch is fixed some problems as your comments in previous mails.
>
> >The assumption that the uleb128 address size can only decrease with
> relaxation is true only if both addresses are in the same section.
> I found this kind of subtract is banned in cvt_frag_to_fill for
> rs_leb128.  Therefore, it doesn't have to restrict by ourselves.
>
> Thanks for your reviews.
>
> Nelson Chu <[hidden email]> 於 2019年12月18日 週三 上午9:50寫道:
> >
> > There is one minor thing.  Maciej has pointed out before that it's
> > better done with one the percent-codes to `_bfd_error_handler' rather
> > than aborting the link right away, so that any further link errors are
> > also reported and you don't have to shake them out one by one.  So
> > report the relocation error via linker's callback function seems to be
> > better.  I think maybe we can report a dangerous relocation for the
> > mismatched R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128.  What I meant
> > is that,
> >
> > + case R_RISCV_SET_ULEB128:
> > ...
> > +  else
> > +    {
> > +      if (uleb128_rel->r_offset != rel->r_offset)
> > + {
> > -  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
> > - input_bfd, howto->name);
> > -  bfd_set_error (bfd_error_bad_value);
> > + msg = ("R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128 are mismatched. ");
> > + r = bfd_reloc_dangerous;
> > + break;
> > + }
> > +      relocation = relocation - uleb128_vma;
> > +      uleb128_rel = NULL;
> > +      break;
> > +    }
> > +
> > + case R_RISCV_SUB_ULEB128:
> > +  if (uleb128_rel)
> > +    {
> > +      if (uleb128_rel->r_offset != rel->r_offset)
> > + {
> > -  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
> > - input_bfd, howto->name);
> > -  bfd_set_error (bfd_error_bad_value);
> > + msg = ("R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128 are mismatched. ");
> > + r = bfd_reloc_dangerous;
> > + break;
> > + }
> > +      relocation = uleb128_vma - relocation;
> > +      uleb128_rel = NULL;
> > +      break;
> > +    }
> >
> > Thanks and regards
> > Nelson
>
>
>
> --
> Best regards,
> Kuan-Lin Chen.
> [hidden email]
>
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] [RISCV] Support subtraction of .uleb128.

Kuan-Lin Chen
In reply to this post by Kuan-Lin Chen
Hi,

Sorry, I miss the test case in the previous email.
So I resend this mail.

Kuan-Lin Chen <[hidden email]> 於 2020年1月6日 週一 下午4:18寫道:

>
> Hi,
>
> The patch is fixed some problems as your comments in previous mails.
>
> >The assumption that the uleb128 address size can only decrease with relaxation is true only if both addresses are in the same section.
> I found this kind of subtract is banned in cvt_frag_to_fill for
> rs_leb128.  Therefore, it doesn't have to restrict by ourselves.
>
> Thanks for your reviews.
>
> Nelson Chu <[hidden email]> 於 2019年12月18日 週三 上午9:50寫道:
> >
> > There is one minor thing.  Maciej has pointed out before that it's
> > better done with one the percent-codes to `_bfd_error_handler' rather
> > than aborting the link right away, so that any further link errors are
> > also reported and you don't have to shake them out one by one.  So
> > report the relocation error via linker's callback function seems to be
> > better.  I think maybe we can report a dangerous relocation for the
> > mismatched R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128.  What I meant
> > is that,
> >
> > + case R_RISCV_SET_ULEB128:
> > ...
> > +  else
> > +    {
> > +      if (uleb128_rel->r_offset != rel->r_offset)
> > + {
> > -  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
> > - input_bfd, howto->name);
> > -  bfd_set_error (bfd_error_bad_value);
> > + msg = ("R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128 are mismatched. ");
> > + r = bfd_reloc_dangerous;
> > + break;
> > + }
> > +      relocation = relocation - uleb128_vma;
> > +      uleb128_rel = NULL;
> > +      break;
> > +    }
> > +
> > + case R_RISCV_SUB_ULEB128:
> > +  if (uleb128_rel)
> > +    {
> > +      if (uleb128_rel->r_offset != rel->r_offset)
> > + {
> > -  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
> > - input_bfd, howto->name);
> > -  bfd_set_error (bfd_error_bad_value);
> > + msg = ("R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128 are mismatched. ");
> > + r = bfd_reloc_dangerous;
> > + break;
> > + }
> > +      relocation = uleb128_vma - relocation;
> > +      uleb128_rel = NULL;
> > +      break;
> > +    }
> >
> > Thanks and regards
> > Nelson
>
>
>
> --
> Best regards,
> Kuan-Lin Chen.
> [hidden email]


--
Best regards,
Kuan-Lin Chen.
[hidden email]

0001-RISC-V-Support-subtraction-of-.uleb128.patch (21K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] [RISCV] Support subtraction of .uleb128.

Sourceware - binutils list mailing list
In reply to this post by Kuan-Lin Chen
On Mon, 09 Dec 2019 21:42:58 PST (-0800), [hidden email] wrote:
> Hi Palmer,
>
>> GCC gives me -Wmaybe-uninitialized for both of these.
> May I know your gcc version?
> I tried gcc-7.3.0 and gcc-8.2.0, and I didn't get the warning.

$ gcc --version
gcc (Debian 8.3.0-6) 8.3.0
Copyright (C) 2018 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

>> I don't understand how this could possibly work: uleb128s can be long, so we
>> can't just treat them as a single byte.
> Yes, uleb128s can be long so I implement riscv_elf_ignore_reloc to
> ignore the overflow checking.
> And I relocate uleb128s by myself in perform_relocation.
>
>> I'm having trouble testing it, though,
>> as I can't figure out how to emit R_RISCV_SET_ULEB128 alone.  For example, when
>> I try to insert a uleb128 in the assembler I'm not even getting a relocation:
> .uleb128 only support subtract format. (.uleb128 .L1 - .L0)
> It is impossible to get final the value and the length for ".uleb128
> symbol" in assemble-time.

Well, you know symbols are at most 64-bit on rv64 and 32-bit on rv32, so you
can just emit enough space to encode any symbol.  That's better than emitting 7
bits, which is unlikely to encode any symbol.

> I think ".uleb128 symbol" is meaningless.
> Therefore, all targets just fill the value as assemble-time value
> without relocations.

If the assembler can't handle something then it needs to provide an error, it
can't just generate incorrect code.

>>I can get a R_RISCV_SUB_ULEB128, but it appears to be both producing the wrong
>>value and garbaging up bytes after the relocation.  For example:
>
> I get the same result in the object file but the different result in executable.
> pal.o:     file format elf64-littleriscv
>
> Disassembly of section .text:
>
> 0000000000000000 <_start>:
>    0:   00000013                nop
>
> 0000000000000004 <.L0>:
>    4:   00000013                nop
>
> 0000000000000008 <uleb_l1_l0>:
>    8:   00100013                li      zero,1
>    c:   aa04                    fsd     fs1,16(a2)
>                         c: R_RISCV_SET_ULEB128  .L1
>                         c: R_RISCV_SUB_ULEB128  .L0
>    e:   aaaa                    fsd     fa0,336(sp)
>   10:   00200013                li      zero,2
>
>
> a.out:     file format elf64-littleriscv
>
>
> Disassembly of section .text:
>
> 0000000000010078 <_start>:
>    10078:       00000013                nop
>    1007c:       00000013                nop
>
> 0000000000010080 <uleb_l1_l0>:
>    10080:       00100013                li      zero,1
>    10084:       aa04                    fsd     fs1,16(a2)
>    10086:       aaaa                    fsd     fa0,336(sp)
>    10088:       00200013                li      zero,2
>
> Maybe there is something mismatched. But I don't have any idea currently.
> Could you please give me more information about your flow?

IIRC I just applied the patch on top of whatever was HEAD of binutils-gdb at
the time, but I don't have the working directory around any more.

> Thanks.
>
>> Presumably that's why I can't get a _SET alone.
> I think it's meaningless to use _SET alone in assembly.
>
>> It looks like alignment is broken here after uleb128s, but it's not actually
>> triggering an error in all cases.  Having two of them in this test case doesn't
>> trigger the issue, but an odd number does.  For example, the following source
> I think this alignment issue can be fixed by the following patch:
> https://sourceware.org/ml/binutils/2019-12/msg00024.html

Ya, thanks -- I think there's some issues there as well, IIRC I commented on
it.

>
> Thanks for your review.
>
> Palmer Dabbelt <[hidden email]> 於 2019年12月7日 週六 上午7:47寫道:
>>
>> On Wed, 27 Nov 2019 00:11:50 PST (-0800), [hidden email] wrote:
>> > The data length of uleb128 is variable.  So linker must recalculate the
>> > value of the subtraction.  The patch leave relocations in object files
>> > so that linker can relocate again after relaxation.
>> >
>> > bfd/ChangeLog:
>> > * bfd-in2.h: Regenerated.
>> > * elfnn-riscv.c (write_uleb128): New function.
>> > (perform_relocation): Perform R_RISCV_SUB_ULEB128 and
>> > R_RISCV_SET_ULEB128 relocation.
>> > (riscv_elf_relocate_section): Likewise.
>> > * elfxx-riscv.c (howto_table): Add R_RISCV_SUB_ULEB128 and
>> > R_RISCV_SET_ULEB128.
>> > (riscv_elf_ignore_reloc): New function.
>> > * libbfd.h: Regenerated.
>> > * reloc.c (BFD_RELOC_RISCV_SET_ULEB128, BFD_RELOC_RISCV_SUB_ULEB128):
>> > New relocations to support .uleb128.
>> >
>> > gas/ChangeLog:
>> > * config/tc-riscv.c (md_apply_fix): Add BFD_RELOC_RISCV_SET_ULEB128 and
>> > BFD_RELOC_RISCV_SUB_ULEB128.
>> > (riscv_insert_uleb128_fixes): New function.
>> > (riscv_md_end): Scan rs_leb128 fragments.
>> > (riscv_pseudo_table): Remove uleb128.
>> >
>> > include/ChangeLog:
>> > * elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.
>> >
>> > ld/ChangeLog:
>> > * testsuite/ld-riscv-elf/ld-riscv-elf.exp: Add uleb128.
>> > * testsuite/ld-riscv-elf/uleb128.d: New test.
>> > * testsuite/ld-riscv-elf/uleb128.s: New file.
>> >
>> > OK to commit?
>>
>> I'm having a lot of trouble trying to figure out how this works.  LMK if I'm
>> missing something, there's some comments in line:
>>
>> > From 9f7f0aaa484dee2b32c550f8cca4de18959f01e1 Mon Sep 17 00:00:00 2001
>> > From: Kuan-Lin Chen <[hidden email]>
>> > Date: Thu, 14 Nov 2019 14:24:22 +0800
>> > Subject: [PATCH] RISC-V: Support subtraction of .uleb128.
>> >
>> > The data length of uleb128 is variable.  So linker must recalculate the
>> > value of the subtraction.  The patch leave relocations in object files
>> > so that linker can relocate again after relaxation.
>> >
>> > bfd/ChangeLog:
>> >         * bfd-in2.h: Regenerated.
>> >         * elfnn-riscv.c (write_uleb128): New function.
>> >         (perform_relocation): Perform R_RISCV_SUB_ULEB128 and
>> >          R_RISCV_SET_ULEB128 relocation.
>> >         (riscv_elf_relocate_section): Likewise.
>> >         * elfxx-riscv.c (howto_table): Add R_RISCV_SUB_ULEB128 and
>> >         R_RISCV_SET_ULEB128.
>> >         (riscv_elf_ignore_reloc): New function.
>> >         * libbfd.h: Regenerated.
>> >         * reloc.c (BFD_RELOC_RISCV_SET_ULEB128, BFD_RELOC_RISCV_SUB_ULEB128):
>> >         New relocations to support .uleb128.
>> >
>> > gas/ChangeLog:
>> >         * config/tc-riscv.c (md_apply_fix): Add BFD_RELOC_RISCV_SET_ULEB128 and
>> >         BFD_RELOC_RISCV_SUB_ULEB128.
>> >         (riscv_insert_uleb128_fixes): New function.
>> >         (riscv_md_end): Scan rs_leb128 fragments.
>> >         (riscv_pseudo_table): Remove uleb128.
>> >
>> > include/ChangeLog:
>> >         * elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.
>> >
>> > ld/ChangeLog:
>> >         * testsuite/ld-riscv-elf/ld-riscv-elf.exp: Add uleb128.
>> >         * testsuite/ld-riscv-elf/uleb128.d: New test.
>> >         * testsuite/ld-riscv-elf/uleb128.s: New file.
>> > ---
>> >  bfd/ChangeLog                              | 14 ++++
>> >  bfd/bfd-in2.h                              |  2 +
>> >  bfd/elfnn-riscv.c                          | 82 ++++++++++++++++++++++
>> >  bfd/elfxx-riscv.c                          | 51 ++++++++++++++
>> >  bfd/libbfd.h                               |  2 +
>> >  bfd/reloc.c                                |  4 ++
>> >  gas/ChangeLog                              |  8 +++
>> >  gas/config/tc-riscv.c                      | 47 ++++++++++++-
>> >  include/ChangeLog                          |  4 ++
>> >  include/elf/riscv.h                        |  2 +
>> >  ld/ChangeLog                               |  6 ++
>> >  ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp |  1 +
>> >  ld/testsuite/ld-riscv-elf/uleb128.d        | 18 +++++
>> >  ld/testsuite/ld-riscv-elf/uleb128.s        | 18 +++++
>> >  14 files changed, 258 insertions(+), 1 deletion(-)
>> >  create mode 100644 ld/testsuite/ld-riscv-elf/uleb128.d
>> >  create mode 100644 ld/testsuite/ld-riscv-elf/uleb128.s
>> >
>> > diff --git a/bfd/ChangeLog b/bfd/ChangeLog
>> > index 4a0852e577..d11ffde74c 100644
>> > --- a/bfd/ChangeLog
>> > +++ b/bfd/ChangeLog
>> > @@ -1,3 +1,17 @@
>> > +2019-11-27  Kuan-Lin Chen  <[hidden email]>
>> > +
>> > +        * bfd-in2.h: Regenerated.
>> > +        * elfnn-riscv.c (write_uleb128): New function.
>> > +        (perform_relocation): Perform R_RISCV_SUB_ULEB128 and
>> > +         R_RISCV_SET_ULEB128 relocation.
>> > +        (riscv_elf_relocate_section): Likewise.
>> > +        * elfxx-riscv.c (howto_table): Add R_RISCV_SUB_ULEB128 and
>> > +        R_RISCV_SET_ULEB128.
>> > +        (riscv_elf_ignore_reloc): New function.
>> > +        * libbfd.h: Regenerated.
>> > +        * reloc.c (BFD_RELOC_RISCV_SET_ULEB128, BFD_RELOC_RISCV_SUB_ULEB128):
>> > +        New relocations to support .uleb128.
>> > +
>> >  2019-11-27  Alan Modra  <[hidden email]>
>> >
>> >          PR 23652
>> > diff --git a/bfd/bfd-in2.h b/bfd/bfd-in2.h
>> > index 44902fc8d0..a32708ebc2 100644
>> > --- a/bfd/bfd-in2.h
>> > +++ b/bfd/bfd-in2.h
>> > @@ -4380,6 +4380,8 @@ number for the SBIC, SBIS, SBI and CBI instructions  */
>> >    BFD_RELOC_RISCV_SET16,
>> >    BFD_RELOC_RISCV_SET32,
>> >    BFD_RELOC_RISCV_32_PCREL,
>> > +  BFD_RELOC_RISCV_SET_ULEB128,
>> > +  BFD_RELOC_RISCV_SUB_ULEB128,
>> >
>> >  /* Renesas RL78 Relocations.  */
>> >    BFD_RELOC_RL78_NEG8,
>> > diff --git a/bfd/elfnn-riscv.c b/bfd/elfnn-riscv.c
>> > index 997f786602..075ef6b82d 100644
>> > --- a/bfd/elfnn-riscv.c
>> > +++ b/bfd/elfnn-riscv.c
>> > @@ -1419,6 +1419,26 @@ riscv_global_pointer_value (struct bfd_link_info *info)
>> >    return h->u.def.value + sec_addr (h->u.def.section);
>> >  }
>> >
>> > +/* Write VAL in uleb128 format to P, returning a pointer to the
>> > +   following byte.
>> > +   This code is copied from elf-attr.c.  */
>> > +
>> > +static bfd_byte *
>> > +write_uleb128 (bfd_byte *p, unsigned int val)
>> > +{
>> > +  bfd_byte c;
>> > +  do
>> > +    {
>> > +      c = val & 0x7f;
>> > +      val >>= 7;
>> > +      if (val)
>> > +        c |= 0x80;
>> > +      *(p++) = c;
>> > +    }
>> > +  while (val);
>> > +  return p;
>> > +}
>> > +
>> >  /* Emplace a static relocation.  */
>> >
>> >  static bfd_reloc_status_type
>> > @@ -1512,6 +1532,25 @@ perform_relocation (const reloc_howto_type *howto,
>> >          value = ENCODE_RVC_LUI_IMM (RISCV_CONST_HIGH_PART (value));
>> >        break;
>> >
>> > +    case R_RISCV_SET_ULEB128:
>> > +    case R_RISCV_SUB_ULEB128:
>> > +      {
>> > +        unsigned int len = 0;
>> > +        bfd_byte *endp, *p;
>> > +
>> > +        _bfd_read_unsigned_leb128 (input_bfd, contents + rel->r_offset, &len);
>> > +
>> > +        /* Clean the contents value to zero.  Do not reduce the length.  */
>> > +        p = contents + rel->r_offset;
>> > +        endp = p + len -1;
>> > +        memset (p, 0x80, len);
>> > +        *(endp) = 0;
>> > +        p = write_uleb128 (p, value) - 1;
>> > +        if (p < endp)
>> > +          *p |= 0x80;
>> > +        return bfd_reloc_ok;
>> > +      }
>> > +
>> >      case R_RISCV_32:
>> >      case R_RISCV_64:
>> >      case R_RISCV_ADD8:
>> > @@ -1767,6 +1806,8 @@ riscv_elf_relocate_section (bfd *output_bfd,
>> >    struct elf_link_hash_entry **sym_hashes = elf_sym_hashes (input_bfd);
>> >    bfd_vma *local_got_offsets = elf_local_got_offsets (input_bfd);
>> >    bfd_boolean absolute;
>> > +  bfd_vma uleb128_vma;
>> > +  Elf_Internal_Rela *uleb128_rel = NULL;
>> >
>> >    if (!riscv_init_pcrel_relocs (&pcrel_relocs))
>> >      return FALSE;
>> > @@ -1871,6 +1912,47 @@ riscv_elf_relocate_section (bfd *output_bfd,
>> >          case R_RISCV_DELETE:
>> >            /* These require no special handling beyond perform_relocation.  */
>> >            break;
>> > +        case R_RISCV_SET_ULEB128:
>> > +          if (!uleb128_rel)
>> > +            {
>> > +              /* Save the minuend to use later.  */
>> > +              uleb128_vma = relocation;
>> > +              uleb128_rel = rel;
>> > +              continue;
>> > +            }
>> > +          else
>> > +            {
>> > +              if (uleb128_rel->r_offset != rel->r_offset)
>> > +                {
>> > +                  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
>> > +                                         input_bfd, howto->name);
>> > +                  bfd_set_error (bfd_error_bad_value);
>> > +                }
>> > +              relocation = relocation - uleb128_vma;
>> > +              uleb128_rel = NULL;
>> > +              break;
>> > +            }
>> > +
>> > +        case R_RISCV_SUB_ULEB128:
>> > +          if (uleb128_rel)
>> > +            {
>> > +              if (uleb128_rel->r_offset != rel->r_offset)
>> > +                {
>> > +                  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
>> > +                                         input_bfd, howto->name);
>> > +                  bfd_set_error (bfd_error_bad_value);
>> > +                }
>> > +              relocation = uleb128_vma - relocation;
>>
>> GCC gives me -Wmaybe-uninitialized for both of these.
>>
>> > +              uleb128_rel = NULL;
>> > +              break;
>> > +            }
>> > +          else
>> > +            {
>> > +              /* Save the subtrahend to use later.  */
>> > +              uleb128_vma = relocation;
>> > +              uleb128_rel = rel;
>> > +              continue;
>> > +            }
>> >
>> >          case R_RISCV_GOT_HI20:
>> >            if (h != NULL)
>> > diff --git a/bfd/elfxx-riscv.c b/bfd/elfxx-riscv.c
>> > index 245717f70f..c136ba2207 100644
>> > --- a/bfd/elfxx-riscv.c
>> > +++ b/bfd/elfxx-riscv.c
>> > @@ -38,6 +38,8 @@
>> >     relocations for the debug info.  */
>> >  static bfd_reloc_status_type riscv_elf_add_sub_reloc
>> >    (bfd *, arelent *, asymbol *, void *, asection *, bfd *, char **);
>> > +static bfd_reloc_status_type riscv_elf_ignore_reloc
>> > +  (bfd *, arelent *, asymbol *, void *, asection *, bfd *, char **);
>> >
>> >  /* The relocation table used for SHT_RELA sections.  */
>> >
>> > @@ -855,8 +857,41 @@ static reloc_howto_type howto_table[] =
>> >           0,                                /* src_mask */
>> >           MINUS_ONE,                        /* dst_mask */
>> >           FALSE),                        /* pcrel_offset */
>> > +
>> > +  /* The length of unsigned-leb128 is variant, just assume the
>> > +     size is one byte here.  */
>> > +  HOWTO (R_RISCV_SET_ULEB128,                /* type */
>> > +         0,                                /* rightshift */
>> > +         0,                                /* size */
>> > +         0,                                /* bitsize */
>> > +         FALSE,                                /* pc_relative */
>> > +         0,                                /* bitpos */
>> > +         complain_overflow_dont,        /* complain_on_overflow */
>> > +         riscv_elf_ignore_reloc,        /* special_function */
>> > +         "R_RISCV_SET_ULEB128",                /* name */
>> > +         FALSE,                                /* partial_inplace */
>> > +         0,                                /* src_mask */
>> > +         0,                                /* dst_mask */
>> > +         FALSE),                        /* pcrel_offset */
>>
>> I don't understand how this could possibly work: uleb128s can be long, so we
>> can't just treat them as a single byte.  I'm having trouble testing it, though,
>> as I can't figure out how to emit R_RISCV_SET_ULEB128 alone.  For example, when
>> I try to insert a uleb128 in the assembler I'm not even getting a relocation:
>>
>>     .text
>>     .globl _start
>>     _start:
>>             nop
>>     .L0:
>>             nop
>>     .L1:
>>     uleb_l0:
>>             addi x0, x0, 1
>>             .uleb128 .L0
>>             # Manually fix up the alignment, as .align is broken.
>>             .rept 3
>>             .byte 0xAA
>>             .endr
>>             addi x0, x0, 2
>>
>>     $ objdump -dr
>>     test.o:     file format elf64-littleriscv
>>
>>
>>     Disassembly of section .text:
>>
>>     0000000000000000 <_start>:
>>        0:   00000013                nop
>>        4:   00000013                nop
>>
>>     0000000000000008 <uleb_l0>:
>>        8:   00100013                li      zero,1
>>        c:   aa04                    fsd     fs1,16(a2)
>>        e:   aaaa                    fsd     fa0,336(sp)
>>       10:   00200013                li      zero,2
>>
>> so then when I link I'm just getting the wrong value entirely.
>>
>>     test:     file format elf64-littleriscv
>>
>>
>>     Disassembly of section .text:
>>
>>     0000000000010078 <_start>:
>>        10078:       00000013                nop
>>        1007c:       00000013                nop
>>
>>     0000000000010080 <uleb_l0>:
>>        10080:       00100013                li      zero,1
>>        10084:       aa04                    fsd     fs1,16(a2)
>>        10086:       aaaa                    fsd     fa0,336(sp)
>>        10088:       00200013                li      zero,2
>>
>> > +  /* The length of unsigned-leb128 is variant, just assume the
>> > +     size is one byte here.  */
>> > +  HOWTO (R_RISCV_SUB_ULEB128,                /* type */
>> > +         0,                                /* rightshift */
>> > +         0,                                /* size */
>> > +         0,                                /* bitsize */
>> > +         FALSE,                                /* pc_relative */
>> > +         0,                                /* bitpos */
>> > +         complain_overflow_dont,        /* complain_on_overflow */
>> > +         riscv_elf_ignore_reloc,        /* special_function */
>> > +         "R_RISCV_SUB_ULEB128",                /* name */
>> > +         FALSE,                                /* partial_inplace */
>> > +         0,                                /* src_mask */
>> > +         0,                                /* dst_mask */
>> > +         FALSE),                        /* pcrel_offset */
>> >  };
>>
>> I can get a R_RISCV_SUB_ULEB128, but it appears to be both producing the wrong
>> value and garbaging up bytes after the relocation.  For example:
>>
>>     .text
>>     .globl _start
>>     _start:
>>             nop
>>     .L0:
>>             nop
>>     .L1:
>>     uleb_l1_l0:
>>             addi x0, x0, 1
>>             .uleb128 .L1 - .L0
>>             # Manually fix up the alignment, as .align is broken
>>             .rept 3
>>             .byte 0xAA
>>             .endr
>>             addi x0, x0, 2
>>
>> produces the relocation
>>
>>     test.o:     file format elf64-littleriscv
>>
>>
>>     Disassembly of section .text:
>>
>>     0000000000000000 <_start>:
>>        0:   00000013                nop
>>
>>     0000000000000004 <.L0>:
>>        4:   00000013                nop
>>
>>     0000000000000008 <uleb_l1_l0>:
>>        8:   00100013                li      zero,1
>>        c:   aa04                    fsd     fs1,16(a2)
>>                             c: R_RISCV_SET_ULEB128  .L1
>>                             c: R_RISCV_SUB_ULEB128  .L0
>>        e:   aaaa                    fsd     fa0,336(sp)
>>       10:   00200013                li      zero,2
>>
>> but then goes and corrupts the resulting executable
>>
>>
>>     test:     file format elf64-littleriscv
>>     Disassembly of section .text:
>>
>>     0000000000010078 <_start>:
>>        10078:       00000013                nop
>>        1007c:       00000013                nop
>>
>>     0000000000010080 <uleb_l1_l0>:
>>        10080:       00100013                li      zero,1
>>        10084:       ff84                    sd      s1,56(a5)
>>        10086:       000ffffb                0xffffb
>>        1008a:       0020                    addi    s0,sp,8
>>
>> I can't think of a way to make this work aside from just emitting uleb128s at
>> their full length, which would make them work but would really defeat the
>> purpose of the format.  I guess we could then relax these, but that seems like
>> a lot of work.
>>
>> > +
>> >  /* A mapping from BFD reloc types to RISC-V ELF reloc types.  */
>> >
>> >  struct elf_reloc_map
>> > @@ -917,6 +952,8 @@ static const struct elf_reloc_map riscv_reloc_map[] =
>> >    { BFD_RELOC_RISCV_SET16, R_RISCV_SET16 },
>> >    { BFD_RELOC_RISCV_SET32, R_RISCV_SET32 },
>> >    { BFD_RELOC_RISCV_32_PCREL, R_RISCV_32_PCREL },
>> > +  { BFD_RELOC_RISCV_SET_ULEB128, R_RISCV_SET_ULEB128 },
>> > +  { BFD_RELOC_RISCV_SUB_ULEB128, R_RISCV_SUB_ULEB128 },
>> >  };
>> >
>> >  /* Given a BFD reloc type, return a howto structure.  */
>> > @@ -1011,6 +1048,20 @@ riscv_elf_add_sub_reloc (bfd *abfd,
>> >    return bfd_reloc_ok;
>> >  }
>> >
>> > +/* Special handler for relocations which don't have to be relocated.
>> > +   This function just simply return bfd_reloc_ok.  */
>> > +static bfd_reloc_status_type
>> > +riscv_elf_ignore_reloc (bfd *abfd ATTRIBUTE_UNUSED, arelent *reloc_entry,
>> > +                        asymbol *symbol ATTRIBUTE_UNUSED,
>> > +                        void *data ATTRIBUTE_UNUSED, asection *input_section,
>> > +                        bfd *output_bfd, char **error_message ATTRIBUTE_UNUSED)
>> > +{
>> > +  if (output_bfd != NULL)
>> > +    reloc_entry->address += input_section->output_offset;
>> > +
>> > +  return bfd_reloc_ok;
>> > +}
>> > +
>> >  /* Parsing subset version.
>> >
>> >     Return Value:
>> > diff --git a/bfd/libbfd.h b/bfd/libbfd.h
>> > index 77b732ee4b..7aa8f91504 100644
>> > --- a/bfd/libbfd.h
>> > +++ b/bfd/libbfd.h
>> > @@ -2327,6 +2327,8 @@ static const char *const bfd_reloc_code_real_names[] = { "@@uninitialized@@",
>> >    "BFD_RELOC_RISCV_SET16",
>> >    "BFD_RELOC_RISCV_SET32",
>> >    "BFD_RELOC_RISCV_32_PCREL",
>> > +  "BFD_RELOC_RISCV_SET_ULEB128",
>> > +  "BFD_RELOC_RISCV_SUB_ULEB128",
>> >    "BFD_RELOC_RL78_NEG8",
>> >    "BFD_RELOC_RL78_NEG16",
>> >    "BFD_RELOC_RL78_NEG24",
>> > diff --git a/bfd/reloc.c b/bfd/reloc.c
>> > index b00b79f319..9b8b75ad21 100644
>> > --- a/bfd/reloc.c
>> > +++ b/bfd/reloc.c
>> > @@ -5212,6 +5212,10 @@ ENUMX
>> >    BFD_RELOC_RISCV_SET32
>> >  ENUMX
>> >    BFD_RELOC_RISCV_32_PCREL
>> > +ENUMX
>> > +  BFD_RELOC_RISCV_SET_ULEB128
>> > +ENUMX
>> > +  BFD_RELOC_RISCV_SUB_ULEB128
>> >  ENUMDOC
>> >    RISC-V relocations.
>> >
>> > diff --git a/gas/ChangeLog b/gas/ChangeLog
>> > index 09991524da..e53ac0895b 100644
>> > --- a/gas/ChangeLog
>> > +++ b/gas/ChangeLog
>> > @@ -1,3 +1,11 @@
>> > +2019-11-27  Kuan-Lin Chen  <[hidden email]>
>> > +
>> > +        * config/tc-riscv.c (md_apply_fix): Add BFD_RELOC_RISCV_SET_ULEB128 and
>> > +        BFD_RELOC_RISCV_SUB_ULEB128.
>> > +        (riscv_insert_uleb128_fixes): New function.
>> > +        (riscv_md_end): Scan rs_leb128 fragments.
>> > +        (riscv_pseudo_table): Remove uleb128.
>> > +
>> >  2019-11-25  Andrew Pinski  <[hidden email]>
>> >
>> >          * config/tc-aarch64.c (md_begin): Use correct
>> > diff --git a/gas/config/tc-riscv.c b/gas/config/tc-riscv.c
>> > index e50505138e..c35c591c0a 100644
>> > --- a/gas/config/tc-riscv.c
>> > +++ b/gas/config/tc-riscv.c
>> > @@ -2392,6 +2392,9 @@ md_apply_fix (fixS *fixP, valueT *valP, segT seg ATTRIBUTE_UNUSED)
>> >      case BFD_RELOC_RISCV_SUB32:
>> >      case BFD_RELOC_RISCV_SUB64:
>> >      case BFD_RELOC_RISCV_RELAX:
>> > +    /* cvt_frag_to_fill () has called output_leb128 ().  */
>> > +    case BFD_RELOC_RISCV_SET_ULEB128:
>> > +    case BFD_RELOC_RISCV_SUB_ULEB128:
>> >        break;
>> >
>> >      case BFD_RELOC_RISCV_TPREL_HI20:
>> > @@ -3127,11 +3130,54 @@ riscv_set_public_attributes (void)
>> >      riscv_write_out_arch_attr ();
>> >  }
>> >
>> > +/* Scan uleb128 subtraction expressions and insert fixups for them.
>> > +   e.g., .uleb128 .L1 - .L0
>> > +   Becuase relaxation may change the value of the subtraction, we
>> > +   must resolve them in link-time.  */
>> > +
>> > +static void
>> > +riscv_insert_uleb128_fixes (bfd *abfd ATTRIBUTE_UNUSED,
>> > +                            asection *sec, void *xxx ATTRIBUTE_UNUSED)
>> > +{
>> > +  segment_info_type *seginfo = seg_info (sec);
>> > +  struct frag *fragP;
>> > +
>> > +  subseg_set (sec, 0);
>> > +
>> > +  for (fragP = seginfo->frchainP->frch_root;
>> > +       fragP; fragP = fragP->fr_next)
>> > +    {
>> > +      expressionS *exp, *exp_dup;
>> > +
>> > +      if (fragP->fr_type != rs_leb128  || fragP->fr_symbol == NULL)
>> > +        continue;
>> > +
>> > +      exp = symbol_get_value_expression (fragP->fr_symbol);
>> > +
>> > +      if (exp->X_op != O_subtract)
>> > +        continue;
>>
>> Presumably that's why I can't get a _SET alone.
>>
>> > +
>> > +      /* Only unsigned leb128 can be handle.  */
>> > +      gas_assert (fragP->fr_subtype == 0);
>> > +      exp_dup = xmemdup (exp, sizeof (*exp), sizeof (*exp));
>> > +      exp_dup->X_op = O_symbol;
>> > +      exp_dup->X_op_symbol = NULL;
>> > +
>> > +      /* Insert relocations to resolve the subtraction in link-time.  */
>> > +      fix_new_exp (fragP, fragP->fr_fix, 0,
>> > +                   exp_dup, 0, BFD_RELOC_RISCV_SET_ULEB128);
>> > +      exp_dup->X_add_symbol = exp->X_op_symbol;
>> > +      fix_new_exp (fragP, fragP->fr_fix, 0,
>> > +                   exp_dup, 0, BFD_RELOC_RISCV_SUB_ULEB128);
>> > +    }
>> > +}
>> > +
>> >  /* Called after all assembly has been done.  */
>> >
>> >  void
>> >  riscv_md_end (void)
>> >  {
>> > +  bfd_map_over_sections (stdoutput, riscv_insert_uleb128_fixes, NULL);
>> >    riscv_set_public_attributes ();
>> >  }
>> >
>> > @@ -3215,7 +3261,6 @@ static const pseudo_typeS riscv_pseudo_table[] =
>> >    {"dtprelword", s_dtprel, 4},
>> >    {"dtpreldword", s_dtprel, 8},
>> >    {"bss", s_bss, 0},
>> > -  {"uleb128", s_riscv_leb128, 0},
>> >    {"sleb128", s_riscv_leb128, 1},
>> >    {"insn", s_riscv_insn, 0},
>> >    {"attribute", s_riscv_attribute, 0},
>> > diff --git a/include/ChangeLog b/include/ChangeLog
>> > index 47bb86cf71..c290b245c6 100644
>> > --- a/include/ChangeLog
>> > +++ b/include/ChangeLog
>> > @@ -1,3 +1,7 @@
>> > +2019-11-27  Kuan-Lin Chen  <[hidden email]>
>> > +
>> > +        * elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.
>> > +
>> >  2019-11-25  Alan Modra  <[hidden email]>
>> >
>> >          * coff/ti.h (GET_SCNHDR_SIZE, PUT_SCNHDR_SIZE, GET_SCN_SCNLEN),
>> > diff --git a/include/elf/riscv.h b/include/elf/riscv.h
>> > index 2f98aa4a3e..030679b35f 100644
>> > --- a/include/elf/riscv.h
>> > +++ b/include/elf/riscv.h
>> > @@ -88,6 +88,8 @@ START_RELOC_NUMBERS (elf_riscv_reloc_type)
>> >    RELOC_NUMBER (R_RISCV_SET16, 55)
>> >    RELOC_NUMBER (R_RISCV_SET32, 56)
>> >    RELOC_NUMBER (R_RISCV_32_PCREL, 57)
>> > +  RELOC_NUMBER (R_RISCV_SET_ULEB128, 58)
>> > +  RELOC_NUMBER (R_RISCV_SUB_ULEB128, 59)
>> >  END_RELOC_NUMBERS (R_RISCV_max)
>> >
>> >  /* Processor specific flags for the ELF header e_flags field.  */
>> > diff --git a/ld/ChangeLog b/ld/ChangeLog
>> > index 969ab78035..5c2e406101 100644
>> > --- a/ld/ChangeLog
>> > +++ b/ld/ChangeLog
>> > @@ -1,3 +1,9 @@
>> > +2019-11-27  Kuan-Lin Chen  <[hidden email]>
>> > +
>> > +        * testsuite/ld-riscv-elf/ld-riscv-elf.exp: Add uleb128.
>> > +        * testsuite/ld-riscv-elf/uleb128.d: New test.
>> > +        * testsuite/ld-riscv-elf/uleb128.s: New file.
>> > +
>> >  2019-11-26  Martin Liska  <[hidden email]>
>> >
>> >          * scripttempl/arclinux.sc: Add .text.sorted.* which is sorted
>> > diff --git a/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp b/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
>> > index 7aabbdd641..809e40f08d 100644
>> > --- a/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
>> > +++ b/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
>> > @@ -38,6 +38,7 @@ if [istarget "riscv*-*-*"] {
>> >      run_dump_test "attr-merge-priv-spec"
>> >      run_dump_test "attr-merge-arch-failed-01"
>> >      run_dump_test "attr-merge-stack-align-failed"
>> > +    run_dump_test "uleb128"
>> >      run_ld_link_tests {
>> >          { "Weak reference 32" "-T weakref.ld -melf32lriscv" ""
>> >              "-march=rv32i -mabi=ilp32" {weakref32.s}
>> > diff --git a/ld/testsuite/ld-riscv-elf/uleb128.d b/ld/testsuite/ld-riscv-elf/uleb128.d
>> > new file mode 100644
>> > index 0000000000..a921478e98
>> > --- /dev/null
>> > +++ b/ld/testsuite/ld-riscv-elf/uleb128.d
>> > @@ -0,0 +1,18 @@
>> > +#source: uleb128.s
>> > +#as: -march=rv32ic
>> > +#ld: -melf32lriscv
>> > +#objdump: -d
>> > +
>> > +.*:[         ]+file format .*
>> > +
>> > +Disassembly of section .text:
>> > +
>> > +.* <_start>:
>> > +.*jal.*<bar>
>> > +.*jal.*<bar>
>> > +.*jal.*<bar>
>> > +.*jal.*<bar>
>> > +.*jal.*<bar>
>> > +.*jal.*<bar>
>> > +.*:[         ]+0e0c.*
>> > +#pass
>> > diff --git a/ld/testsuite/ld-riscv-elf/uleb128.s b/ld/testsuite/ld-riscv-elf/uleb128.s
>> > new file mode 100644
>> > index 0000000000..f7d23be163
>> > --- /dev/null
>> > +++ b/ld/testsuite/ld-riscv-elf/uleb128.s
>> > @@ -0,0 +1,18 @@
>> > +.text
>> > +.globl bar
>> > +.globl _start
>> > +.option rvc
>> > +.align 2
>> > +_start:
>> > +.L0:
>> > +        .rept 6
>> > +        call bar
>> > +        .endr
>> > +.align 2
>> > +.L1:
>> > +        .uleb128 .L1 - .L0
>> > +        .uleb128 .L2 - .L0
>> > +.L2:
>> > +.align 2
>> > +bar:
>> > +        nop
>>
>> It looks like alignment is broken here after uleb128s, but it's not actually
>> triggering an error in all cases.  Having two of them in this test case doesn't
>> trigger the issue, but an odd number does.  For example, the following source
>>
>>     .text
>>     .globl _start
>>     .align 2
>>     _start:
>>             .uleb128 _start
>>     .align 2
>>     _end:
>>             nop
>>
>> links _end to a misaligned symbol
>>
>>     Disassembly of section .text:
>>
>>     0000000000010078 <_start>:
>>             ...
>>
>>     0000000000010079 <_end>:
>>        10079:       00000013                nop
>>        1007d:       0000                    unimp
>>             ...
>>
>> I'm assuming that's because norvc assumes 4-byte alignment in text sections, so
>> it's just eliding the alingment directives.  With RVC I do get an error message
>>
>>     .text
>>     .globl _start
>>     .option rvc
>>     .align 2
>>     _start:
>>             .uleb128 _start
>>     .align 2
>>     _end:
>>             nop
>>
>>     ./install/bin/riscv64-unknown-linux-gnu-ld: test.o(.text+0x1): 3 bytes required for alignment to 4-byte boundary, but only 2 present
>>     ./install/bin/riscv64-unknown-linux-gnu-ld: can't relax section: bad value
>>
>>
>
>
> --
> Best regards,
> Kuan-Lin Chen.
> [hidden email]
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] [RISCV] Support subtraction of .uleb128.

Kuan-Lin Chen
Hi,

This version patch updates as follows:
Fix the warning about uninitialized variable.
Set a limit to .uleb128 expressions as Palmer's suggestion.
Thanks.

Palmer Dabbelt <[hidden email]> 於 2020年1月10日 週五 上午7:19寫道:

>
> On Mon, 09 Dec 2019 21:42:58 PST (-0800), [hidden email] wrote:
> > Hi Palmer,
> >
> >> GCC gives me -Wmaybe-uninitialized for both of these.
> > May I know your gcc version?
> > I tried gcc-7.3.0 and gcc-8.2.0, and I didn't get the warning.
>
> $ gcc --version
> gcc (Debian 8.3.0-6) 8.3.0
> Copyright (C) 2018 Free Software Foundation, Inc.
> This is free software; see the source for copying conditions.  There is NO
> warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
>
> >> I don't understand how this could possibly work: uleb128s can be long, so we
> >> can't just treat them as a single byte.
> > Yes, uleb128s can be long so I implement riscv_elf_ignore_reloc to
> > ignore the overflow checking.
> > And I relocate uleb128s by myself in perform_relocation.
> >
> >> I'm having trouble testing it, though,
> >> as I can't figure out how to emit R_RISCV_SET_ULEB128 alone.  For example, when
> >> I try to insert a uleb128 in the assembler I'm not even getting a relocation:
> > .uleb128 only support subtract format. (.uleb128 .L1 - .L0)
> > It is impossible to get final the value and the length for ".uleb128
> > symbol" in assemble-time.
>
> Well, you know symbols are at most 64-bit on rv64 and 32-bit on rv32, so you
> can just emit enough space to encode any symbol.  That's better than emitting 7
> bits, which is unlikely to encode any symbol.
>
> > I think ".uleb128 symbol" is meaningless.
> > Therefore, all targets just fill the value as assemble-time value
> > without relocations.
>
> If the assembler can't handle something then it needs to provide an error, it
> can't just generate incorrect code.
>
> >>I can get a R_RISCV_SUB_ULEB128, but it appears to be both producing the wrong
> >>value and garbaging up bytes after the relocation.  For example:
> >
> > I get the same result in the object file but the different result in executable.
> > pal.o:     file format elf64-littleriscv
> >
> > Disassembly of section .text:
> >
> > 0000000000000000 <_start>:
> >    0:   00000013                nop
> >
> > 0000000000000004 <.L0>:
> >    4:   00000013                nop
> >
> > 0000000000000008 <uleb_l1_l0>:
> >    8:   00100013                li      zero,1
> >    c:   aa04                    fsd     fs1,16(a2)
> >                         c: R_RISCV_SET_ULEB128  .L1
> >                         c: R_RISCV_SUB_ULEB128  .L0
> >    e:   aaaa                    fsd     fa0,336(sp)
> >   10:   00200013                li      zero,2
> >
> >
> > a.out:     file format elf64-littleriscv
> >
> >
> > Disassembly of section .text:
> >
> > 0000000000010078 <_start>:
> >    10078:       00000013                nop
> >    1007c:       00000013                nop
> >
> > 0000000000010080 <uleb_l1_l0>:
> >    10080:       00100013                li      zero,1
> >    10084:       aa04                    fsd     fs1,16(a2)
> >    10086:       aaaa                    fsd     fa0,336(sp)
> >    10088:       00200013                li      zero,2
> >
> > Maybe there is something mismatched. But I don't have any idea currently.
> > Could you please give me more information about your flow?
>
> IIRC I just applied the patch on top of whatever was HEAD of binutils-gdb at
> the time, but I don't have the working directory around any more.
>
> > Thanks.
> >
> >> Presumably that's why I can't get a _SET alone.
> > I think it's meaningless to use _SET alone in assembly.
> >
> >> It looks like alignment is broken here after uleb128s, but it's not actually
> >> triggering an error in all cases.  Having two of them in this test case doesn't
> >> trigger the issue, but an odd number does.  For example, the following source
> > I think this alignment issue can be fixed by the following patch:
> > https://sourceware.org/ml/binutils/2019-12/msg00024.html
>
> Ya, thanks -- I think there's some issues there as well, IIRC I commented on
> it.
>
> >
> > Thanks for your review.
> >
> > Palmer Dabbelt <[hidden email]> 於 2019年12月7日 週六 上午7:47寫道:
> >>
> >> On Wed, 27 Nov 2019 00:11:50 PST (-0800), [hidden email] wrote:
> >> > The data length of uleb128 is variable.  So linker must recalculate the
> >> > value of the subtraction.  The patch leave relocations in object files
> >> > so that linker can relocate again after relaxation.
> >> >
> >> > bfd/ChangeLog:
> >> > * bfd-in2.h: Regenerated.
> >> > * elfnn-riscv.c (write_uleb128): New function.
> >> > (perform_relocation): Perform R_RISCV_SUB_ULEB128 and
> >> > R_RISCV_SET_ULEB128 relocation.
> >> > (riscv_elf_relocate_section): Likewise.
> >> > * elfxx-riscv.c (howto_table): Add R_RISCV_SUB_ULEB128 and
> >> > R_RISCV_SET_ULEB128.
> >> > (riscv_elf_ignore_reloc): New function.
> >> > * libbfd.h: Regenerated.
> >> > * reloc.c (BFD_RELOC_RISCV_SET_ULEB128, BFD_RELOC_RISCV_SUB_ULEB128):
> >> > New relocations to support .uleb128.
> >> >
> >> > gas/ChangeLog:
> >> > * config/tc-riscv.c (md_apply_fix): Add BFD_RELOC_RISCV_SET_ULEB128 and
> >> > BFD_RELOC_RISCV_SUB_ULEB128.
> >> > (riscv_insert_uleb128_fixes): New function.
> >> > (riscv_md_end): Scan rs_leb128 fragments.
> >> > (riscv_pseudo_table): Remove uleb128.
> >> >
> >> > include/ChangeLog:
> >> > * elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.
> >> >
> >> > ld/ChangeLog:
> >> > * testsuite/ld-riscv-elf/ld-riscv-elf.exp: Add uleb128.
> >> > * testsuite/ld-riscv-elf/uleb128.d: New test.
> >> > * testsuite/ld-riscv-elf/uleb128.s: New file.
> >> >
> >> > OK to commit?
> >>
> >> I'm having a lot of trouble trying to figure out how this works.  LMK if I'm
> >> missing something, there's some comments in line:
> >>
> >> > From 9f7f0aaa484dee2b32c550f8cca4de18959f01e1 Mon Sep 17 00:00:00 2001
> >> > From: Kuan-Lin Chen <[hidden email]>
> >> > Date: Thu, 14 Nov 2019 14:24:22 +0800
> >> > Subject: [PATCH] RISC-V: Support subtraction of .uleb128.
> >> >
> >> > The data length of uleb128 is variable.  So linker must recalculate the
> >> > value of the subtraction.  The patch leave relocations in object files
> >> > so that linker can relocate again after relaxation.
> >> >
> >> > bfd/ChangeLog:
> >> >         * bfd-in2.h: Regenerated.
> >> >         * elfnn-riscv.c (write_uleb128): New function.
> >> >         (perform_relocation): Perform R_RISCV_SUB_ULEB128 and
> >> >          R_RISCV_SET_ULEB128 relocation.
> >> >         (riscv_elf_relocate_section): Likewise.
> >> >         * elfxx-riscv.c (howto_table): Add R_RISCV_SUB_ULEB128 and
> >> >         R_RISCV_SET_ULEB128.
> >> >         (riscv_elf_ignore_reloc): New function.
> >> >         * libbfd.h: Regenerated.
> >> >         * reloc.c (BFD_RELOC_RISCV_SET_ULEB128, BFD_RELOC_RISCV_SUB_ULEB128):
> >> >         New relocations to support .uleb128.
> >> >
> >> > gas/ChangeLog:
> >> >         * config/tc-riscv.c (md_apply_fix): Add BFD_RELOC_RISCV_SET_ULEB128 and
> >> >         BFD_RELOC_RISCV_SUB_ULEB128.
> >> >         (riscv_insert_uleb128_fixes): New function.
> >> >         (riscv_md_end): Scan rs_leb128 fragments.
> >> >         (riscv_pseudo_table): Remove uleb128.
> >> >
> >> > include/ChangeLog:
> >> >         * elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.
> >> >
> >> > ld/ChangeLog:
> >> >         * testsuite/ld-riscv-elf/ld-riscv-elf.exp: Add uleb128.
> >> >         * testsuite/ld-riscv-elf/uleb128.d: New test.
> >> >         * testsuite/ld-riscv-elf/uleb128.s: New file.
> >> > ---
> >> >  bfd/ChangeLog                              | 14 ++++
> >> >  bfd/bfd-in2.h                              |  2 +
> >> >  bfd/elfnn-riscv.c                          | 82 ++++++++++++++++++++++
> >> >  bfd/elfxx-riscv.c                          | 51 ++++++++++++++
> >> >  bfd/libbfd.h                               |  2 +
> >> >  bfd/reloc.c                                |  4 ++
> >> >  gas/ChangeLog                              |  8 +++
> >> >  gas/config/tc-riscv.c                      | 47 ++++++++++++-
> >> >  include/ChangeLog                          |  4 ++
> >> >  include/elf/riscv.h                        |  2 +
> >> >  ld/ChangeLog                               |  6 ++
> >> >  ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp |  1 +
> >> >  ld/testsuite/ld-riscv-elf/uleb128.d        | 18 +++++
> >> >  ld/testsuite/ld-riscv-elf/uleb128.s        | 18 +++++
> >> >  14 files changed, 258 insertions(+), 1 deletion(-)
> >> >  create mode 100644 ld/testsuite/ld-riscv-elf/uleb128.d
> >> >  create mode 100644 ld/testsuite/ld-riscv-elf/uleb128.s
> >> >
> >> > diff --git a/bfd/ChangeLog b/bfd/ChangeLog
> >> > index 4a0852e577..d11ffde74c 100644
> >> > --- a/bfd/ChangeLog
> >> > +++ b/bfd/ChangeLog
> >> > @@ -1,3 +1,17 @@
> >> > +2019-11-27  Kuan-Lin Chen  <[hidden email]>
> >> > +
> >> > +        * bfd-in2.h: Regenerated.
> >> > +        * elfnn-riscv.c (write_uleb128): New function.
> >> > +        (perform_relocation): Perform R_RISCV_SUB_ULEB128 and
> >> > +         R_RISCV_SET_ULEB128 relocation.
> >> > +        (riscv_elf_relocate_section): Likewise.
> >> > +        * elfxx-riscv.c (howto_table): Add R_RISCV_SUB_ULEB128 and
> >> > +        R_RISCV_SET_ULEB128.
> >> > +        (riscv_elf_ignore_reloc): New function.
> >> > +        * libbfd.h: Regenerated.
> >> > +        * reloc.c (BFD_RELOC_RISCV_SET_ULEB128, BFD_RELOC_RISCV_SUB_ULEB128):
> >> > +        New relocations to support .uleb128.
> >> > +
> >> >  2019-11-27  Alan Modra  <[hidden email]>
> >> >
> >> >          PR 23652
> >> > diff --git a/bfd/bfd-in2.h b/bfd/bfd-in2.h
> >> > index 44902fc8d0..a32708ebc2 100644
> >> > --- a/bfd/bfd-in2.h
> >> > +++ b/bfd/bfd-in2.h
> >> > @@ -4380,6 +4380,8 @@ number for the SBIC, SBIS, SBI and CBI instructions  */
> >> >    BFD_RELOC_RISCV_SET16,
> >> >    BFD_RELOC_RISCV_SET32,
> >> >    BFD_RELOC_RISCV_32_PCREL,
> >> > +  BFD_RELOC_RISCV_SET_ULEB128,
> >> > +  BFD_RELOC_RISCV_SUB_ULEB128,
> >> >
> >> >  /* Renesas RL78 Relocations.  */
> >> >    BFD_RELOC_RL78_NEG8,
> >> > diff --git a/bfd/elfnn-riscv.c b/bfd/elfnn-riscv.c
> >> > index 997f786602..075ef6b82d 100644
> >> > --- a/bfd/elfnn-riscv.c
> >> > +++ b/bfd/elfnn-riscv.c
> >> > @@ -1419,6 +1419,26 @@ riscv_global_pointer_value (struct bfd_link_info *info)
> >> >    return h->u.def.value + sec_addr (h->u.def.section);
> >> >  }
> >> >
> >> > +/* Write VAL in uleb128 format to P, returning a pointer to the
> >> > +   following byte.
> >> > +   This code is copied from elf-attr.c.  */
> >> > +
> >> > +static bfd_byte *
> >> > +write_uleb128 (bfd_byte *p, unsigned int val)
> >> > +{
> >> > +  bfd_byte c;
> >> > +  do
> >> > +    {
> >> > +      c = val & 0x7f;
> >> > +      val >>= 7;
> >> > +      if (val)
> >> > +        c |= 0x80;
> >> > +      *(p++) = c;
> >> > +    }
> >> > +  while (val);
> >> > +  return p;
> >> > +}
> >> > +
> >> >  /* Emplace a static relocation.  */
> >> >
> >> >  static bfd_reloc_status_type
> >> > @@ -1512,6 +1532,25 @@ perform_relocation (const reloc_howto_type *howto,
> >> >          value = ENCODE_RVC_LUI_IMM (RISCV_CONST_HIGH_PART (value));
> >> >        break;
> >> >
> >> > +    case R_RISCV_SET_ULEB128:
> >> > +    case R_RISCV_SUB_ULEB128:
> >> > +      {
> >> > +        unsigned int len = 0;
> >> > +        bfd_byte *endp, *p;
> >> > +
> >> > +        _bfd_read_unsigned_leb128 (input_bfd, contents + rel->r_offset, &len);
> >> > +
> >> > +        /* Clean the contents value to zero.  Do not reduce the length.  */
> >> > +        p = contents + rel->r_offset;
> >> > +        endp = p + len -1;
> >> > +        memset (p, 0x80, len);
> >> > +        *(endp) = 0;
> >> > +        p = write_uleb128 (p, value) - 1;
> >> > +        if (p < endp)
> >> > +          *p |= 0x80;
> >> > +        return bfd_reloc_ok;
> >> > +      }
> >> > +
> >> >      case R_RISCV_32:
> >> >      case R_RISCV_64:
> >> >      case R_RISCV_ADD8:
> >> > @@ -1767,6 +1806,8 @@ riscv_elf_relocate_section (bfd *output_bfd,
> >> >    struct elf_link_hash_entry **sym_hashes = elf_sym_hashes (input_bfd);
> >> >    bfd_vma *local_got_offsets = elf_local_got_offsets (input_bfd);
> >> >    bfd_boolean absolute;
> >> > +  bfd_vma uleb128_vma;
> >> > +  Elf_Internal_Rela *uleb128_rel = NULL;
> >> >
> >> >    if (!riscv_init_pcrel_relocs (&pcrel_relocs))
> >> >      return FALSE;
> >> > @@ -1871,6 +1912,47 @@ riscv_elf_relocate_section (bfd *output_bfd,
> >> >          case R_RISCV_DELETE:
> >> >            /* These require no special handling beyond perform_relocation.  */
> >> >            break;
> >> > +        case R_RISCV_SET_ULEB128:
> >> > +          if (!uleb128_rel)
> >> > +            {
> >> > +              /* Save the minuend to use later.  */
> >> > +              uleb128_vma = relocation;
> >> > +              uleb128_rel = rel;
> >> > +              continue;
> >> > +            }
> >> > +          else
> >> > +            {
> >> > +              if (uleb128_rel->r_offset != rel->r_offset)
> >> > +                {
> >> > +                  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
> >> > +                                         input_bfd, howto->name);
> >> > +                  bfd_set_error (bfd_error_bad_value);
> >> > +                }
> >> > +              relocation = relocation - uleb128_vma;
> >> > +              uleb128_rel = NULL;
> >> > +              break;
> >> > +            }
> >> > +
> >> > +        case R_RISCV_SUB_ULEB128:
> >> > +          if (uleb128_rel)
> >> > +            {
> >> > +              if (uleb128_rel->r_offset != rel->r_offset)
> >> > +                {
> >> > +                  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
> >> > +                                         input_bfd, howto->name);
> >> > +                  bfd_set_error (bfd_error_bad_value);
> >> > +                }
> >> > +              relocation = uleb128_vma - relocation;
> >>
> >> GCC gives me -Wmaybe-uninitialized for both of these.
> >>
> >> > +              uleb128_rel = NULL;
> >> > +              break;
> >> > +            }
> >> > +          else
> >> > +            {
> >> > +              /* Save the subtrahend to use later.  */
> >> > +              uleb128_vma = relocation;
> >> > +              uleb128_rel = rel;
> >> > +              continue;
> >> > +            }
> >> >
> >> >          case R_RISCV_GOT_HI20:
> >> >            if (h != NULL)
> >> > diff --git a/bfd/elfxx-riscv.c b/bfd/elfxx-riscv.c
> >> > index 245717f70f..c136ba2207 100644
> >> > --- a/bfd/elfxx-riscv.c
> >> > +++ b/bfd/elfxx-riscv.c
> >> > @@ -38,6 +38,8 @@
> >> >     relocations for the debug info.  */
> >> >  static bfd_reloc_status_type riscv_elf_add_sub_reloc
> >> >    (bfd *, arelent *, asymbol *, void *, asection *, bfd *, char **);
> >> > +static bfd_reloc_status_type riscv_elf_ignore_reloc
> >> > +  (bfd *, arelent *, asymbol *, void *, asection *, bfd *, char **);
> >> >
> >> >  /* The relocation table used for SHT_RELA sections.  */
> >> >
> >> > @@ -855,8 +857,41 @@ static reloc_howto_type howto_table[] =
> >> >           0,                                /* src_mask */
> >> >           MINUS_ONE,                        /* dst_mask */
> >> >           FALSE),                        /* pcrel_offset */
> >> > +
> >> > +  /* The length of unsigned-leb128 is variant, just assume the
> >> > +     size is one byte here.  */
> >> > +  HOWTO (R_RISCV_SET_ULEB128,                /* type */
> >> > +         0,                                /* rightshift */
> >> > +         0,                                /* size */
> >> > +         0,                                /* bitsize */
> >> > +         FALSE,                                /* pc_relative */
> >> > +         0,                                /* bitpos */
> >> > +         complain_overflow_dont,        /* complain_on_overflow */
> >> > +         riscv_elf_ignore_reloc,        /* special_function */
> >> > +         "R_RISCV_SET_ULEB128",                /* name */
> >> > +         FALSE,                                /* partial_inplace */
> >> > +         0,                                /* src_mask */
> >> > +         0,                                /* dst_mask */
> >> > +         FALSE),                        /* pcrel_offset */
> >>
> >> I don't understand how this could possibly work: uleb128s can be long, so we
> >> can't just treat them as a single byte.  I'm having trouble testing it, though,
> >> as I can't figure out how to emit R_RISCV_SET_ULEB128 alone.  For example, when
> >> I try to insert a uleb128 in the assembler I'm not even getting a relocation:
> >>
> >>     .text
> >>     .globl _start
> >>     _start:
> >>             nop
> >>     .L0:
> >>             nop
> >>     .L1:
> >>     uleb_l0:
> >>             addi x0, x0, 1
> >>             .uleb128 .L0
> >>             # Manually fix up the alignment, as .align is broken.
> >>             .rept 3
> >>             .byte 0xAA
> >>             .endr
> >>             addi x0, x0, 2
> >>
> >>     $ objdump -dr
> >>     test.o:     file format elf64-littleriscv
> >>
> >>
> >>     Disassembly of section .text:
> >>
> >>     0000000000000000 <_start>:
> >>        0:   00000013                nop
> >>        4:   00000013                nop
> >>
> >>     0000000000000008 <uleb_l0>:
> >>        8:   00100013                li      zero,1
> >>        c:   aa04                    fsd     fs1,16(a2)
> >>        e:   aaaa                    fsd     fa0,336(sp)
> >>       10:   00200013                li      zero,2
> >>
> >> so then when I link I'm just getting the wrong value entirely.
> >>
> >>     test:     file format elf64-littleriscv
> >>
> >>
> >>     Disassembly of section .text:
> >>
> >>     0000000000010078 <_start>:
> >>        10078:       00000013                nop
> >>        1007c:       00000013                nop
> >>
> >>     0000000000010080 <uleb_l0>:
> >>        10080:       00100013                li      zero,1
> >>        10084:       aa04                    fsd     fs1,16(a2)
> >>        10086:       aaaa                    fsd     fa0,336(sp)
> >>        10088:       00200013                li      zero,2
> >>
> >> > +  /* The length of unsigned-leb128 is variant, just assume the
> >> > +     size is one byte here.  */
> >> > +  HOWTO (R_RISCV_SUB_ULEB128,                /* type */
> >> > +         0,                                /* rightshift */
> >> > +         0,                                /* size */
> >> > +         0,                                /* bitsize */
> >> > +         FALSE,                                /* pc_relative */
> >> > +         0,                                /* bitpos */
> >> > +         complain_overflow_dont,        /* complain_on_overflow */
> >> > +         riscv_elf_ignore_reloc,        /* special_function */
> >> > +         "R_RISCV_SUB_ULEB128",                /* name */
> >> > +         FALSE,                                /* partial_inplace */
> >> > +         0,                                /* src_mask */
> >> > +         0,                                /* dst_mask */
> >> > +         FALSE),                        /* pcrel_offset */
> >> >  };
> >>
> >> I can get a R_RISCV_SUB_ULEB128, but it appears to be both producing the wrong
> >> value and garbaging up bytes after the relocation.  For example:
> >>
> >>     .text
> >>     .globl _start
> >>     _start:
> >>             nop
> >>     .L0:
> >>             nop
> >>     .L1:
> >>     uleb_l1_l0:
> >>             addi x0, x0, 1
> >>             .uleb128 .L1 - .L0
> >>             # Manually fix up the alignment, as .align is broken
> >>             .rept 3
> >>             .byte 0xAA
> >>             .endr
> >>             addi x0, x0, 2
> >>
> >> produces the relocation
> >>
> >>     test.o:     file format elf64-littleriscv
> >>
> >>
> >>     Disassembly of section .text:
> >>
> >>     0000000000000000 <_start>:
> >>        0:   00000013                nop
> >>
> >>     0000000000000004 <.L0>:
> >>        4:   00000013                nop
> >>
> >>     0000000000000008 <uleb_l1_l0>:
> >>        8:   00100013                li      zero,1
> >>        c:   aa04                    fsd     fs1,16(a2)
> >>                             c: R_RISCV_SET_ULEB128  .L1
> >>                             c: R_RISCV_SUB_ULEB128  .L0
> >>        e:   aaaa                    fsd     fa0,336(sp)
> >>       10:   00200013                li      zero,2
> >>
> >> but then goes and corrupts the resulting executable
> >>
> >>
> >>     test:     file format elf64-littleriscv
> >>     Disassembly of section .text:
> >>
> >>     0000000000010078 <_start>:
> >>        10078:       00000013                nop
> >>        1007c:       00000013                nop
> >>
> >>     0000000000010080 <uleb_l1_l0>:
> >>        10080:       00100013                li      zero,1
> >>        10084:       ff84                    sd      s1,56(a5)
> >>        10086:       000ffffb                0xffffb
> >>        1008a:       0020                    addi    s0,sp,8
> >>
> >> I can't think of a way to make this work aside from just emitting uleb128s at
> >> their full length, which would make them work but would really defeat the
> >> purpose of the format.  I guess we could then relax these, but that seems like
> >> a lot of work.
> >>
> >> > +
> >> >  /* A mapping from BFD reloc types to RISC-V ELF reloc types.  */
> >> >
> >> >  struct elf_reloc_map
> >> > @@ -917,6 +952,8 @@ static const struct elf_reloc_map riscv_reloc_map[] =
> >> >    { BFD_RELOC_RISCV_SET16, R_RISCV_SET16 },
> >> >    { BFD_RELOC_RISCV_SET32, R_RISCV_SET32 },
> >> >    { BFD_RELOC_RISCV_32_PCREL, R_RISCV_32_PCREL },
> >> > +  { BFD_RELOC_RISCV_SET_ULEB128, R_RISCV_SET_ULEB128 },
> >> > +  { BFD_RELOC_RISCV_SUB_ULEB128, R_RISCV_SUB_ULEB128 },
> >> >  };
> >> >
> >> >  /* Given a BFD reloc type, return a howto structure.  */
> >> > @@ -1011,6 +1048,20 @@ riscv_elf_add_sub_reloc (bfd *abfd,
> >> >    return bfd_reloc_ok;
> >> >  }
> >> >
> >> > +/* Special handler for relocations which don't have to be relocated.
> >> > +   This function just simply return bfd_reloc_ok.  */
> >> > +static bfd_reloc_status_type
> >> > +riscv_elf_ignore_reloc (bfd *abfd ATTRIBUTE_UNUSED, arelent *reloc_entry,
> >> > +                        asymbol *symbol ATTRIBUTE_UNUSED,
> >> > +                        void *data ATTRIBUTE_UNUSED, asection *input_section,
> >> > +                        bfd *output_bfd, char **error_message ATTRIBUTE_UNUSED)
> >> > +{
> >> > +  if (output_bfd != NULL)
> >> > +    reloc_entry->address += input_section->output_offset;
> >> > +
> >> > +  return bfd_reloc_ok;
> >> > +}
> >> > +
> >> >  /* Parsing subset version.
> >> >
> >> >     Return Value:
> >> > diff --git a/bfd/libbfd.h b/bfd/libbfd.h
> >> > index 77b732ee4b..7aa8f91504 100644
> >> > --- a/bfd/libbfd.h
> >> > +++ b/bfd/libbfd.h
> >> > @@ -2327,6 +2327,8 @@ static const char *const bfd_reloc_code_real_names[] = { "@@uninitialized@@",
> >> >    "BFD_RELOC_RISCV_SET16",
> >> >    "BFD_RELOC_RISCV_SET32",
> >> >    "BFD_RELOC_RISCV_32_PCREL",
> >> > +  "BFD_RELOC_RISCV_SET_ULEB128",
> >> > +  "BFD_RELOC_RISCV_SUB_ULEB128",
> >> >    "BFD_RELOC_RL78_NEG8",
> >> >    "BFD_RELOC_RL78_NEG16",
> >> >    "BFD_RELOC_RL78_NEG24",
> >> > diff --git a/bfd/reloc.c b/bfd/reloc.c
> >> > index b00b79f319..9b8b75ad21 100644
> >> > --- a/bfd/reloc.c
> >> > +++ b/bfd/reloc.c
> >> > @@ -5212,6 +5212,10 @@ ENUMX
> >> >    BFD_RELOC_RISCV_SET32
> >> >  ENUMX
> >> >    BFD_RELOC_RISCV_32_PCREL
> >> > +ENUMX
> >> > +  BFD_RELOC_RISCV_SET_ULEB128
> >> > +ENUMX
> >> > +  BFD_RELOC_RISCV_SUB_ULEB128
> >> >  ENUMDOC
> >> >    RISC-V relocations.
> >> >
> >> > diff --git a/gas/ChangeLog b/gas/ChangeLog
> >> > index 09991524da..e53ac0895b 100644
> >> > --- a/gas/ChangeLog
> >> > +++ b/gas/ChangeLog
> >> > @@ -1,3 +1,11 @@
> >> > +2019-11-27  Kuan-Lin Chen  <[hidden email]>
> >> > +
> >> > +        * config/tc-riscv.c (md_apply_fix): Add BFD_RELOC_RISCV_SET_ULEB128 and
> >> > +        BFD_RELOC_RISCV_SUB_ULEB128.
> >> > +        (riscv_insert_uleb128_fixes): New function.
> >> > +        (riscv_md_end): Scan rs_leb128 fragments.
> >> > +        (riscv_pseudo_table): Remove uleb128.
> >> > +
> >> >  2019-11-25  Andrew Pinski  <[hidden email]>
> >> >
> >> >          * config/tc-aarch64.c (md_begin): Use correct
> >> > diff --git a/gas/config/tc-riscv.c b/gas/config/tc-riscv.c
> >> > index e50505138e..c35c591c0a 100644
> >> > --- a/gas/config/tc-riscv.c
> >> > +++ b/gas/config/tc-riscv.c
> >> > @@ -2392,6 +2392,9 @@ md_apply_fix (fixS *fixP, valueT *valP, segT seg ATTRIBUTE_UNUSED)
> >> >      case BFD_RELOC_RISCV_SUB32:
> >> >      case BFD_RELOC_RISCV_SUB64:
> >> >      case BFD_RELOC_RISCV_RELAX:
> >> > +    /* cvt_frag_to_fill () has called output_leb128 ().  */
> >> > +    case BFD_RELOC_RISCV_SET_ULEB128:
> >> > +    case BFD_RELOC_RISCV_SUB_ULEB128:
> >> >        break;
> >> >
> >> >      case BFD_RELOC_RISCV_TPREL_HI20:
> >> > @@ -3127,11 +3130,54 @@ riscv_set_public_attributes (void)
> >> >      riscv_write_out_arch_attr ();
> >> >  }
> >> >
> >> > +/* Scan uleb128 subtraction expressions and insert fixups for them.
> >> > +   e.g., .uleb128 .L1 - .L0
> >> > +   Becuase relaxation may change the value of the subtraction, we
> >> > +   must resolve them in link-time.  */
> >> > +
> >> > +static void
> >> > +riscv_insert_uleb128_fixes (bfd *abfd ATTRIBUTE_UNUSED,
> >> > +                            asection *sec, void *xxx ATTRIBUTE_UNUSED)
> >> > +{
> >> > +  segment_info_type *seginfo = seg_info (sec);
> >> > +  struct frag *fragP;
> >> > +
> >> > +  subseg_set (sec, 0);
> >> > +
> >> > +  for (fragP = seginfo->frchainP->frch_root;
> >> > +       fragP; fragP = fragP->fr_next)
> >> > +    {
> >> > +      expressionS *exp, *exp_dup;
> >> > +
> >> > +      if (fragP->fr_type != rs_leb128  || fragP->fr_symbol == NULL)
> >> > +        continue;
> >> > +
> >> > +      exp = symbol_get_value_expression (fragP->fr_symbol);
> >> > +
> >> > +      if (exp->X_op != O_subtract)
> >> > +        continue;
> >>
> >> Presumably that's why I can't get a _SET alone.
> >>
> >> > +
> >> > +      /* Only unsigned leb128 can be handle.  */
> >> > +      gas_assert (fragP->fr_subtype == 0);
> >> > +      exp_dup = xmemdup (exp, sizeof (*exp), sizeof (*exp));
> >> > +      exp_dup->X_op = O_symbol;
> >> > +      exp_dup->X_op_symbol = NULL;
> >> > +
> >> > +      /* Insert relocations to resolve the subtraction in link-time.  */
> >> > +      fix_new_exp (fragP, fragP->fr_fix, 0,
> >> > +                   exp_dup, 0, BFD_RELOC_RISCV_SET_ULEB128);
> >> > +      exp_dup->X_add_symbol = exp->X_op_symbol;
> >> > +      fix_new_exp (fragP, fragP->fr_fix, 0,
> >> > +                   exp_dup, 0, BFD_RELOC_RISCV_SUB_ULEB128);
> >> > +    }
> >> > +}
> >> > +
> >> >  /* Called after all assembly has been done.  */
> >> >
> >> >  void
> >> >  riscv_md_end (void)
> >> >  {
> >> > +  bfd_map_over_sections (stdoutput, riscv_insert_uleb128_fixes, NULL);
> >> >    riscv_set_public_attributes ();
> >> >  }
> >> >
> >> > @@ -3215,7 +3261,6 @@ static const pseudo_typeS riscv_pseudo_table[] =
> >> >    {"dtprelword", s_dtprel, 4},
> >> >    {"dtpreldword", s_dtprel, 8},
> >> >    {"bss", s_bss, 0},
> >> > -  {"uleb128", s_riscv_leb128, 0},
> >> >    {"sleb128", s_riscv_leb128, 1},
> >> >    {"insn", s_riscv_insn, 0},
> >> >    {"attribute", s_riscv_attribute, 0},
> >> > diff --git a/include/ChangeLog b/include/ChangeLog
> >> > index 47bb86cf71..c290b245c6 100644
> >> > --- a/include/ChangeLog
> >> > +++ b/include/ChangeLog
> >> > @@ -1,3 +1,7 @@
> >> > +2019-11-27  Kuan-Lin Chen  <[hidden email]>
> >> > +
> >> > +        * elf/riscv.h ((R_RISCV_SET_ULEB128, (R_RISCV_SUB_ULEB128): Define.
> >> > +
> >> >  2019-11-25  Alan Modra  <[hidden email]>
> >> >
> >> >          * coff/ti.h (GET_SCNHDR_SIZE, PUT_SCNHDR_SIZE, GET_SCN_SCNLEN),
> >> > diff --git a/include/elf/riscv.h b/include/elf/riscv.h
> >> > index 2f98aa4a3e..030679b35f 100644
> >> > --- a/include/elf/riscv.h
> >> > +++ b/include/elf/riscv.h
> >> > @@ -88,6 +88,8 @@ START_RELOC_NUMBERS (elf_riscv_reloc_type)
> >> >    RELOC_NUMBER (R_RISCV_SET16, 55)
> >> >    RELOC_NUMBER (R_RISCV_SET32, 56)
> >> >    RELOC_NUMBER (R_RISCV_32_PCREL, 57)
> >> > +  RELOC_NUMBER (R_RISCV_SET_ULEB128, 58)
> >> > +  RELOC_NUMBER (R_RISCV_SUB_ULEB128, 59)
> >> >  END_RELOC_NUMBERS (R_RISCV_max)
> >> >
> >> >  /* Processor specific flags for the ELF header e_flags field.  */
> >> > diff --git a/ld/ChangeLog b/ld/ChangeLog
> >> > index 969ab78035..5c2e406101 100644
> >> > --- a/ld/ChangeLog
> >> > +++ b/ld/ChangeLog
> >> > @@ -1,3 +1,9 @@
> >> > +2019-11-27  Kuan-Lin Chen  <[hidden email]>
> >> > +
> >> > +        * testsuite/ld-riscv-elf/ld-riscv-elf.exp: Add uleb128.
> >> > +        * testsuite/ld-riscv-elf/uleb128.d: New test.
> >> > +        * testsuite/ld-riscv-elf/uleb128.s: New file.
> >> > +
> >> >  2019-11-26  Martin Liska  <[hidden email]>
> >> >
> >> >          * scripttempl/arclinux.sc: Add .text.sorted.* which is sorted
> >> > diff --git a/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp b/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
> >> > index 7aabbdd641..809e40f08d 100644
> >> > --- a/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
> >> > +++ b/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
> >> > @@ -38,6 +38,7 @@ if [istarget "riscv*-*-*"] {
> >> >      run_dump_test "attr-merge-priv-spec"
> >> >      run_dump_test "attr-merge-arch-failed-01"
> >> >      run_dump_test "attr-merge-stack-align-failed"
> >> > +    run_dump_test "uleb128"
> >> >      run_ld_link_tests {
> >> >          { "Weak reference 32" "-T weakref.ld -melf32lriscv" ""
> >> >              "-march=rv32i -mabi=ilp32" {weakref32.s}
> >> > diff --git a/ld/testsuite/ld-riscv-elf/uleb128.d b/ld/testsuite/ld-riscv-elf/uleb128.d
> >> > new file mode 100644
> >> > index 0000000000..a921478e98
> >> > --- /dev/null
> >> > +++ b/ld/testsuite/ld-riscv-elf/uleb128.d
> >> > @@ -0,0 +1,18 @@
> >> > +#source: uleb128.s
> >> > +#as: -march=rv32ic
> >> > +#ld: -melf32lriscv
> >> > +#objdump: -d
> >> > +
> >> > +.*:[         ]+file format .*
> >> > +
> >> > +Disassembly of section .text:
> >> > +
> >> > +.* <_start>:
> >> > +.*jal.*<bar>
> >> > +.*jal.*<bar>
> >> > +.*jal.*<bar>
> >> > +.*jal.*<bar>
> >> > +.*jal.*<bar>
> >> > +.*jal.*<bar>
> >> > +.*:[         ]+0e0c.*
> >> > +#pass
> >> > diff --git a/ld/testsuite/ld-riscv-elf/uleb128.s b/ld/testsuite/ld-riscv-elf/uleb128.s
> >> > new file mode 100644
> >> > index 0000000000..f7d23be163
> >> > --- /dev/null
> >> > +++ b/ld/testsuite/ld-riscv-elf/uleb128.s
> >> > @@ -0,0 +1,18 @@
> >> > +.text
> >> > +.globl bar
> >> > +.globl _start
> >> > +.option rvc
> >> > +.align 2
> >> > +_start:
> >> > +.L0:
> >> > +        .rept 6
> >> > +        call bar
> >> > +        .endr
> >> > +.align 2
> >> > +.L1:
> >> > +        .uleb128 .L1 - .L0
> >> > +        .uleb128 .L2 - .L0
> >> > +.L2:
> >> > +.align 2
> >> > +bar:
> >> > +        nop
> >>
> >> It looks like alignment is broken here after uleb128s, but it's not actually
> >> triggering an error in all cases.  Having two of them in this test case doesn't
> >> trigger the issue, but an odd number does.  For example, the following source
> >>
> >>     .text
> >>     .globl _start
> >>     .align 2
> >>     _start:
> >>             .uleb128 _start
> >>     .align 2
> >>     _end:
> >>             nop
> >>
> >> links _end to a misaligned symbol
> >>
> >>     Disassembly of section .text:
> >>
> >>     0000000000010078 <_start>:
> >>             ...
> >>
> >>     0000000000010079 <_end>:
> >>        10079:       00000013                nop
> >>        1007d:       0000                    unimp
> >>             ...
> >>
> >> I'm assuming that's because norvc assumes 4-byte alignment in text sections, so
> >> it's just eliding the alingment directives.  With RVC I do get an error message
> >>
> >>     .text
> >>     .globl _start
> >>     .option rvc
> >>     .align 2
> >>     _start:
> >>             .uleb128 _start
> >>     .align 2
> >>     _end:
> >>             nop
> >>
> >>     ./install/bin/riscv64-unknown-linux-gnu-ld: test.o(.text+0x1): 3 bytes required for alignment to 4-byte boundary, but only 2 present
> >>     ./install/bin/riscv64-unknown-linux-gnu-ld: can't relax section: bad value
> >>
> >>
> >
> >
> > --
> > Best regards,
> > Kuan-Lin Chen.
> > [hidden email]


--
Best regards,
Kuan-Lin Chen.
[hidden email]

0001-RISC-V-Support-subtraction-of-.uleb128.patch (21K) Download Attachment