search

riscvarchive / riscv-binutils-gdb

RISC-V backports for binutils-gdb. Development is done upstream at the FSF.
GNU General Public License v2.0
148 stars 233 forks source link

ld: cannot find default versions of the ISA extension `i', when linking rv32e #261

Closed kenta2 closed 3 years ago

kenta2 commented 3 years ago

I built riscv-gnu-toolchain (commit b715e4f01b43efef487166f75d5d85d3c33fa7ef , tag: 2021.04.23) with --with-arch=rv32e --with-abi=ilp32e .

Then, compiling and linking the simplest possible C program main(){} results in the following error when linking:

.../lib/gcc/riscv32-unknown-elf/10.2.0/../../../../riscv32-unknown-elf/bin/ld: cannot find default versions of the ISA extension `i'

The problem goes away if I roll back riscv-gnu-toolchain to (commit d45cfc68be6ce0a2f69daf66e64fc446224b3416 , tag: 2021.04.07), so the problem was likely introduced in "Bump Binutils to 2.36.1".

Drilling down into riscv-binutils, the error message was added in commit 7ef19aa616aee98f0a5c7c948d0db56a9f3c5e57 to bfd/elfxx-riscv.c .

kito-cheng commented 3 years ago

IIRC @Nelson1225 has fixed this issue on upstream? Nelson could you provide commit id for that?

Nelson1225 commented 3 years ago

Probably three commits in binutils mainline need to be applied,

  1. RISC-V: Add i-ext as the implicit extension when e-ext is set. https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=f0bae2552db1dd4f1995608fbf6648fcee4e9e0c

  2. RISC-V: The version of i-ext should be RISCV_UNKNOWN_VERSION when expanding g-ext. https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=f5b10973531cdda7e1aba7496f70476b5146fcf8

  3. RISC-V: Don't report the mismatched version warning for the implicit extensions. https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=e3839c100fdb15d05e6a2fd44557ffdd6fb921b8

The commit ids are mentioned in the links.

Nelson1225 commented 3 years ago

@kito-cheng Should we create two more branches to track the binutils mainline and integration branch here?

Error1000 commented 3 years ago

Bump? I am still seeing this issue ...

jim-wilson commented 3 years ago

Fixed. Ideally we should have rv32e targets in the regression testsuite to catch problems like this, but there are too many testsuite failures for rv32e, so this will be a fair amount of work to fix.