GNU Binutils 2.38 has been released

GNU Binutils 2.38 has been released

[Nick Clifton]

Hi Everyone,

We are pleased to announce that version 2.38 of the GNU Binutils project
sources have been released and are now available for download at:

  https://ftp.gnu.org/gnu/binutils
  https://sourceware.org/pub/binutils/releases/

The SHA256 checksums are as follows:

  070ec71cf077a6a58e0b959f05a09a35015378c2d8a51e90f3aeabfe30590ef8  binutils-2.38.tar.bz2
  6e54170356709d401f1cb781f86bb59656b99b398412047f5c44f0fbc633fa4d  binutils-2.38.tar.bz2.sig
  b3f1dc5b17e75328f19bd88250bee2ef9f91fc8cbb7bd48bdb31390338636052  binutils-2.38.tar.gz
  864d330b71f2b40120d96e68ebed43a07c5286d9c28f3a3893ae4ae7ec25ede9  binutils-2.38.tar.gz.sig
  807ccfa77ccfc3e09b1f760cd73e13839180c6950a37843df1c1e7a58c777dc9  binutils-2.38.tar.lz
  49ee3c3b5803dea2acb15a3fa4c8ad0daebf142ba284a8b3d3524a52823a0ceb  binutils-2.38.tar.lz.sig
  e316477a914f567eccc34d5d29785b8b0f5a10208d36bbacedcc39048ecfe024  binutils-2.38.tar.xz
  41301d67da78df1ad6df04aefe9e7bea8235484b0323cee52caa8f7435385014  binutils-2.38.tar.xz.sig

This release contains numerous bug fixes and improvements, along with
the following new features:

  Assembler:
    General:
      * Add support for the LoongArch architecture.

      * Add an option to control how multibyte characters are handled in
        the assembler.  Using the option warnings can be generated when
        such characters are encountered in symbol names, or anywhere in
        the input source file(s).

    AArch64 and ARM:
      * Add support for more system registers.
      * Add support for Scalable Matrix Extension.
      * Add support for Cortex-R52+, Cortex-A510, Cortex-A710,
        Cortex-X2, Cortex-A710 cores.
      * Add support for 'v8.7-a', 'v8.8-a', 'v9-a', 'v9.1-a',
        'armv9.2-a' and 'armv9.3-a' architecture extensions.

    X86:
      * Add a command-line option to encode aligned vector move as
        unaligned vector move. 
      * Add support for Intel AVX512_FP16 instructions.
      * The outputs of .ds.x directive and .tfloat directive with hex
        input have been reduced from 12 bytes to 10 bytes to match the
        output of .tfloat directive.

  Linker:
    * Add support for the LoongArch architecture.

    * Add -z pack-relative-relocs/-z no pack-relative-relocs to x86 ELF
     linker to pack relative relocations in the DT_RELR section.

    * Add -z indirect-extern-access/-z noindirect-extern-access to x86
      ELF linker to control canonical function pointers and copy
      relocation.

  Other Binary Tools:

    * elfedit: Add --output-abiversion option to update ABIVERSION.

    * Tools which display symbols or strings (readelf, strings, nm,
      objdump) have a new command line option which controls how unicode
      characters are handled.  By default they are treated as normal for
      the tool.  Using --unicode=locale will display them according to
      the current locale.  Using --unicode=hex will display them as hex
      byte values, whilst --unicode=escape will display them as escape
      sequences.  In addition using --unicode=highlight will display
      them as unicode escape sequences highlighted in red (if supported
      by the output device).

    * readelf -r dumps RELR relative relocations now.

    * Support for efi-app-aarch64, efi-rtdrv-aarch64 and
      efi-bsdrv-aarch64 has been added to objcopy in order to enable
      UEFI development using binutils. 

    * ar: Add --thin for creating thin archives. -T is a deprecated
      alias without diagnostics. In many ar implementations -T has a
      different meaning, as specified by X/Open System Interface.

Our thanks go out to all of the binutils contributors, past and
present, for helping to make this release possible.

Cheers
  Nick Clifton
  GNU Binutils Chief Maintainer

Re: GNU Binutils 2.38 has been released

[Waldemar Brodkorb]

Hi Nick,
Nick Clifton via Binutils wrote,

> Hi Everyone,
> 
> We are pleased to announce that version 2.38 of the GNU Binutils project
> sources have been released and are now available for download at:
> 
>   https://ftp.gnu.org/gnu/binutils
>   https://sourceware.org/pub/binutils/releases/

I am having problems with following commit
aed44286efa8ae8717a77d94b51ac3614e2ca6dc. 

When trying to build the Linux kernel I get a lot of:
./arch/riscv/include/asm/irqflags.h:34: Error: unrecognized opcode
`csrrc s2,0x100,2'
./arch/riscv/include/asm/irqflags.h:52: Error: unrecognized opcode
`csrs 0x100,s2'

The kernel is compiled with gcc 11.2 and  -mabi=lp64 -march=rv64ima.

When I revert the commit, everything is fine.

What can we do about it?

best regards
 Waldemar

Re: GNU Binutils 2.38 has been released

[Jan Beulich]

On 14.02.2022 11:58, Waldemar Brodkorb wrote:
> I am having problems with following commit
> aed44286efa8ae8717a77d94b51ac3614e2ca6dc. 
> 
> When trying to build the Linux kernel I get a lot of:
> ./arch/riscv/include/asm/irqflags.h:34: Error: unrecognized opcode
> `csrrc s2,0x100,2'
> ./arch/riscv/include/asm/irqflags.h:52: Error: unrecognized opcode
> `csrs 0x100,s2'
> 
> The kernel is compiled with gcc 11.2 and  -mabi=lp64 -march=rv64ima.

Isn't "ima" too little? I don't think that includes CSR accesses, that's
its own extension (Zicsr).

Jan

Re: GNU Binutils 2.38 has been released

[Waldemar Brodkorb]

Hi Jan,
Jan Beulich wrote,

> On 14.02.2022 11:58, Waldemar Brodkorb wrote:
> > I am having problems with following commit
> > aed44286efa8ae8717a77d94b51ac3614e2ca6dc. 
> > 
> > When trying to build the Linux kernel I get a lot of:
> > ./arch/riscv/include/asm/irqflags.h:34: Error: unrecognized opcode
> > `csrrc s2,0x100,2'
> > ./arch/riscv/include/asm/irqflags.h:52: Error: unrecognized opcode
> > `csrs 0x100,s2'
> > 
> > The kernel is compiled with gcc 11.2 and  -mabi=lp64 -march=rv64ima.
> 
> Isn't "ima" too little? I don't think that includes CSR accesses, that's
> its own extension (Zicsr).

Before this commit it was enough to build the Linux kernel.
What letter needs to be added to get Zicsr extensions, though?
 
 best regards
  Waldemar

Re: GNU Binutils 2.38 has been released

[Jan Beulich]

On 23.02.2022 14:48, Waldemar Brodkorb wrote:
> Jan Beulich wrote,
>> On 14.02.2022 11:58, Waldemar Brodkorb wrote:
>>> I am having problems with following commit
>>> aed44286efa8ae8717a77d94b51ac3614e2ca6dc. 
>>>
>>> When trying to build the Linux kernel I get a lot of:
>>> ./arch/riscv/include/asm/irqflags.h:34: Error: unrecognized opcode
>>> `csrrc s2,0x100,2'
>>> ./arch/riscv/include/asm/irqflags.h:52: Error: unrecognized opcode
>>> `csrs 0x100,s2'
>>>
>>> The kernel is compiled with gcc 11.2 and  -mabi=lp64 -march=rv64ima.
>>
>> Isn't "ima" too little? I don't think that includes CSR accesses, that's
>> its own extension (Zicsr).
> 
> Before this commit it was enough to build the Linux kernel.

Which is an apparently valid observation, but not an excuse.

> What letter needs to be added to get Zicsr extensions, though?

That's not a single letter iirc, and I'm afraid I can't tell you
the right spelling off the top of my head. I'm pretty sure gas
doc has it. Might be something like rv64ima_zicsr.

Jan

Re: GNU Binutils 2.38 has been released

[Waldemar Brodkorb]

Hi Jan,
Jan Beulich wrote,

> On 23.02.2022 14:48, Waldemar Brodkorb wrote:
> > Jan Beulich wrote,
> >> On 14.02.2022 11:58, Waldemar Brodkorb wrote:
> >>> I am having problems with following commit
> >>> aed44286efa8ae8717a77d94b51ac3614e2ca6dc. 
> >>>
> >>> When trying to build the Linux kernel I get a lot of:
> >>> ./arch/riscv/include/asm/irqflags.h:34: Error: unrecognized opcode
> >>> `csrrc s2,0x100,2'
> >>> ./arch/riscv/include/asm/irqflags.h:52: Error: unrecognized opcode
> >>> `csrs 0x100,s2'
> >>>
> >>> The kernel is compiled with gcc 11.2 and  -mabi=lp64 -march=rv64ima.
> >>
> >> Isn't "ima" too little? I don't think that includes CSR accesses, that's
> >> its own extension (Zicsr).
> > 
> > Before this commit it was enough to build the Linux kernel.
> 
> Which is an apparently valid observation, but not an excuse.

At least some comment in the commit log would have been nice for
stupid people like me.

> > What letter needs to be added to get Zicsr extensions, though?
> 
> That's not a single letter iirc, and I'm afraid I can't tell you
> the right spelling off the top of my head. I'm pretty sure gas
> doc has it. Might be something like rv64ima_zicsr.
 
I have not found it in the gas docs. GCC complains about it while
trying to use it:

Traceback (most recent call last):
File
 "/home/wbx/openadk/toolchain_build_qemu-riscv64_uclibc-ng/w-gcc-11.2.0-1/gcc-11.2.0/gcc/config/riscv/arch-canonicalize",
 line 114, in <module>
 print (arch_canonicalize(arg))
 File
 "/home/wbx/openadk/toolchain_build_qemu-riscv64_uclibc-ng/w-gcc-11.2.0-1/gcc-11.2.0/gcc/config/riscv/arch-canonicalize",
 line 89, in arch_canonicalize
 long_exts = list(sorted(filter(lambda x:len(x) != 1,
 long_exts),
 File
 "/home/wbx/openadk/toolchain_build_qemu-riscv64_uclibc-ng/w-gcc-11.2.0-1/gcc-11.2.0/gcc/config/riscv/arch-canonicalize",
 line 81, in longext_sort
 raise Exception("Unsupported extension `%s`" % exts)
 Exception: Unsupported extension `zicsr`
 --with-abi=lp64 is not supported for ISA 
 gmake[6]: *** [Makefile:4353: configure-gcc] Error 1

Any other idea where to find this information. There is no
mentioning of zicsr here:
https://gcc.gnu.org/git/?p=gcc.git;a=blob;f=gcc/config/riscv/t-linux-multilib;h=298547fee38b2b4062e7854c539aabd292927857;hb=39c0b8f1ac4be2fdaa309733bbae11fedc6b8b3d

best regards
 Waldemar
 

Re: GNU Binutils 2.38 has been released

[Jim Wilson]

On Wed, Feb 23, 2022 at 9:31 AM Waldemar Brodkorb <wbx@openadk.org> wrote:

> Any other idea where to find this information. There is no
> mentioning of zicsr here:
>
> https://gcc.gnu.org/git/?p=gcc.git;a=blob;f=gcc/config/riscv/t-linux-multilib;h=298547fee38b2b4062e7854c539aabd292927857;hb=39c0b8f1ac4be2fdaa309733bbae11fedc6b8b3d


Support for the CSR instructions were removed from the base architecture
over 3 years ago.  We didn't make the toolchain change at the time because
we recognized that it would be a mess.  We decided to fix the mess later.
However, over 3 years have gone by now, and no one volunteered to fix the
mess, so we unfortunately had to force the issue.  This was standing in the
way for the platform/profile changes which will be coming soon, and
required this change. This change was announced on the various toolchain
and linux mailing lists, including linux-kernel and linux-riscv, but
obviously there is no easy way to notify everyone.

You can find info about zicsr in the ISA manuals.
https://riscv.org/technical/specifications/
See chapter 9.

Here is a pointer to the announcement for the change.
http://lists.infradead.org/pipermail/linux-riscv/2022-January/011728.html

I don't follow the linux kernel mailing lists, but I believe patches to fix
this problem have already been posted.  Maybe they haven't been merged
yet.  They will need to be backported to older kernel versions if you want
to build them with newer toolchain versions.

Jim

Re: GNU Binutils 2.38 has been released

[Jim Wilson]

On Wed, Feb 23, 2022 at 9:31 AM Waldemar Brodkorb <wbx@openadk.org> wrote:

> Traceback (most recent call last):
> File
>
>  "/home/wbx/openadk/toolchain_build_qemu-riscv64_uclibc-ng/w-gcc-11.2.0-1/gcc-11.2.0/gcc/config/riscv/arch-canonicalize",
>  line 114, in <module>
>  print (arch_canonicalize(arg))
>  File
>
>  "/home/wbx/openadk/toolchain_build_qemu-riscv64_uclibc-ng/w-gcc-11.2.0-1/gcc-11.2.0/gcc/config/riscv/arch-canonicalize",
>  line 89, in arch_canonicalize
>  long_exts = list(sorted(filter(lambda x:len(x) != 1,
>  long_exts),
>  File
>
>  "/home/wbx/openadk/toolchain_build_qemu-riscv64_uclibc-ng/w-gcc-11.2.0-1/gcc-11.2.0/gcc/config/riscv/arch-canonicalize",
>  line 81, in longext_sort
>  raise Exception("Unsupported extension `%s`" % exts)
>  Exception: Unsupported extension `zicsr`
>  --with-abi=lp64 is not supported for ISA
>  gmake[6]: *** [Makefile:4353: configure-gcc] Error 1
>

The GCC arch_canonicalize script was missed in the initial round of fixes.
There is a patch for that too.  And that patch may have to be backported to
older versions also.  This is going to be a mess for a while until we get
everything fixed.  Unfortunately us software guys can't do much about the
problem, as it is the RISC-V International ISA guys that are causing the
trouble.  They keep making changes without considering how it is going to
affect the software support.

Jim