Re: [PATCH] RISC-V: Check if the merged ISA string is conflicting

Yep, you're right. I've already sent another patch series to address this:

https://sourceware.org/pipermail/binutils/2026-July/150010.html

Please take a look when you have a chance.

On Tue, Jun 30, 2026 at 3:15 AM Jeffrey Law
<[email protected]> wrote:
>
>
>
> On 1/12/2025 11:29 PM, Yu-Cheng Liang wrote:
> > In the current implementation, the linker does not check if the merged
> > ISA string is conflicting when attempting to merge two ISA strings.
> > For instance, no error message is generated when the linker links
> > objectA (rv64i_zfinx) with objectB (rv64if).
> >
> > Signed-off-by: Yu-Cheng Liang <[email protected]>
> > ---
> >   bfd/elfnn-riscv.c | 8 ++++++++
> >   1 file changed, 8 insertions(+)
> >
> > diff --git a/bfd/elfnn-riscv.c b/bfd/elfnn-riscv.c
> > index 3aba9c85b0f..0b8c7c33117 100644
> > --- a/bfd/elfnn-riscv.c
> > +++ b/bfd/elfnn-riscv.c
> > @@ -3965,6 +3965,14 @@ riscv_merge_arch_attr_info (bfd *ibfd, char
> > *in_arch, char *out_arch)
> >
> >     merged_arch_str = riscv_arch_str (ARCH_SIZE, &merged_subsets);
> >
> > +  /* Check if the merged ISA string is conflicting.  */
> > +  riscv_subset_list_t subsets = {NULL, NULL, NULL};
> > +  unsigned xlen;
> > +  riscv_parse_subset_t riscv_rps_merged =
> > +    {&subsets, _bfd_error_handler, &xlen, NULL, false};
> > +  if (!riscv_parse_subset (&riscv_rps_merged, merged_arch_str))
> > +    return NULL;
> > +
> >     /* Release the subset lists.  */
> >     riscv_release_subset_list (&in_subsets);
> >     riscv_release_subset_list (&out_subsets);
> So this code was recently moved into elfxx-riscv.c.
>
> The patch applies cleanly over in that context and Kito already gave
> approval around multilib questions that I think Nelson raised.
>
> The question I have is don't we possibly leak memory for various objects
> allocated by riscv_parse_subset and its children and stored into the
> subsets and perhaps other fields of riscv_rps_merged?
>
> Jeff
>