[PATCH] bfd: check for truncation with R_RISCV_32 relocations

[Nelson Chu]

It's been a long time so I forgot the details, if we add this overflow
checks for ADD/SUB/SET relocations, then the regression will fail since we
may generate 32-bit debug information for the rv64 toolchain in gcc, and
that's why we always silent the truncated error.  Not sure if that also
fails for now, we need to make sure of that at first.

Nelson

On Wed, Jan 31, 2024 at 2:21 AM Joseph Faulls <Joseph.Faulls@imgtec.com>
wrote:

> Relevant bug report: https://sourceware.org/bugzilla/show_bug.cgi?id=31318
>
>
>
> With high addresses, these relocations can be truncated. In these cases,
>
> throw an error instead of silently truncating.
>
>
>
> bfd/
>
>         * elfnn-riscv.c (perform_relocation): Check for overflow.
>
>
>
> ld/
>
>         * ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp: Updated.
>
>         * ld/testsuite/ld-riscv-elf/data-reloc-rv64-abs32-truncation.d:
>
>           New test case.
>
> ---
>
> bfd/elfnn-riscv.c                                            | 3 +++
>
> ld/testsuite/ld-riscv-elf/data-reloc-rv64-abs32-truncation.d | 4 ++++
>
> ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp                   | 1 +
>
> 3 files changed, 8 insertions(+)
>
> create mode 100644
> ld/testsuite/ld-riscv-elf/data-reloc-rv64-abs32-truncation.d
>
>
>
> diff --git a/bfd/elfnn-riscv.c b/bfd/elfnn-riscv.c
>
> index 8b27e3b8d6a..975b40e5f53 100644
>
> --- a/bfd/elfnn-riscv.c
>
> +++ b/bfd/elfnn-riscv.c
>
> @@ -1865,6 +1865,9 @@ perform_relocation (const reloc_howto_type *howto,
>
>        }
>
>
>
>      case R_RISCV_32:
>
> +      if (value & ~howto->dst_mask)
>
> +       return bfd_reloc_overflow;
>
> +      break;
>
>      case R_RISCV_64:
>
>      case R_RISCV_ADD8:
>
>      case R_RISCV_ADD16:
>
> diff --git a/ld/testsuite/ld-riscv-elf/data-reloc-rv64-abs32-truncation.d
> b/ld/testsuite/ld-riscv-elf/data-reloc-rv64-abs32-truncation.d
>
> new file mode 100644
>
> index 00000000000..925d1cdb662
>
> --- /dev/null
>
> +++ b/ld/testsuite/ld-riscv-elf/data-reloc-rv64-abs32-truncation.d
>
> @@ -0,0 +1,4 @@
>
> +#source: data-reloc.s
>
> +#as: -march=rv64i -mabi=lp64 -defsym __abs__=1
>
> +#ld: -m[riscv_choose_lp64_emul] -Ttext 0x8000 --defsym _start=0x0
> --defsym abs=0xc00000100 --defsym abs_local=0x200
>
> +#error: .*relocation truncated to fit: R_RISCV_32
>
> diff --git a/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
> b/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
>
> index 7e1281d826b..5a4d8728366 100644
>
> --- a/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
>
> +++ b/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
>
> @@ -175,6 +175,7 @@ if [istarget "riscv*-*-*"] {
>
>      run_dump_test "uleb128"
>
>      run_dump_test "pr31179"
>
>      run_dump_test "pr31179-r"
>
> +    run_dump_test "data-reloc-rv64-abs32-truncation"
>
>      run_ld_link_tests [list \
>
>         [list "Weak reference 32" "-T weakref.ld
> -m[riscv_choose_ilp32_emul]" "" \
>
>             "-march=rv32i -mabi=ilp32" {weakref32.s} \
>
> --
>
> 2
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://sourceware.org/pipermail/binutils/attachments/20240131/2a4e465d/attachment-0001.htm>