[RFC] RISC-V: Fix the 32-bit --enable-targets=all build breakage.

[RFC] RISC-V: Fix the 32-bit --enable-targets=all build breakage.

[Nelson Chu]

This should be the same problem as the commit,
ba9b3ef5ee666467b67780e81f868c432f4fc56d.

Generally, the elfxx-* file should be compiled both for 32-bit and 64-bit
machine, since xx can be 32 and 64.  However, seems like the elfxx-riscv.c
and elf32-riscv.c are missing for BFD32_BACKENDS_CFILES.

This patch try to add the missing linker files for the BFD32_BACKENDS, but
the dst_mask of R_RISCV_CALL may be overflow for the host 32-bit machine.
Therefore, we rewrite the R_RISCV_CALL stuff in the perform_relocation, use
the two 32-bit bfd_vma values to encode it, since the R_RISCV_CALL should be
two instructions, rather than one address.

bfd/
	* Makefile.am: Added elfxx-riscv.* and elf32-riscv.* to
	BFD32_BACKENDS and BFD32_BACKENDS_CFILES.
	* Makefile.in: Regernated.
	* elfnn-riscv.c (perform_relocation): Relocate the R_RISCV_CALL and
	R_RISCV_CALL_PLT relocs by two bfd_vma values, since they are two
	insturctions in fact.  So that the rv32 riscv toolchain built on
	the host 32-bit machine should work.
	* elfxx-riscv.c (howto_table): Let elfnn-riscv.c:perform_relocation
	to handle the dst_mask of R_RISCV_CALL and R_RISCV_CALL_PLT specially.
	Set dst_mask to MINUS_ONE.
---
 bfd/Makefile.am   |  4 ++++
 bfd/Makefile.in   |  4 ++++
 bfd/elfnn-riscv.c | 24 +++++++++++++++++++++---
 bfd/elfxx-riscv.c |  6 ++----
 4 files changed, 31 insertions(+), 7 deletions(-)

diff --git a/bfd/Makefile.am b/bfd/Makefile.am
index 097177b..1a935d6 100644
--- a/bfd/Makefile.am
+++ b/bfd/Makefile.am
@@ -336,6 +336,8 @@ BFD32_BACKENDS = \
 	elf32-pj.lo \
 	elf32-ppc.lo \
 	elf32-pru.lo \
+	elf32-riscv.lo \
+	elfxx-riscv.lo \
 	elf32-rl78.lo \
 	elf32-rx.lo \
 	elf32-s390.lo \
@@ -469,6 +471,8 @@ BFD32_BACKENDS_CFILES = \
 	elf32-pj.c \
 	elf32-ppc.c \
 	elf32-pru.c \
+	elf32-riscv.c \
+	elfxx-riscv.c \
 	elf32-rl78.c \
 	elf32-rx.c \
 	elf32-s390.c \
diff --git a/bfd/Makefile.in b/bfd/Makefile.in
index a76b653..3817167 100644
--- a/bfd/Makefile.in
+++ b/bfd/Makefile.in
@@ -762,6 +762,8 @@ BFD32_BACKENDS = \
 	elf32-pj.lo \
 	elf32-ppc.lo \
 	elf32-pru.lo \
+	elf32-riscv.lo \
+	elfxx-riscv.lo \
 	elf32-rl78.lo \
 	elf32-rx.lo \
 	elf32-s390.lo \
@@ -895,6 +897,8 @@ BFD32_BACKENDS_CFILES = \
 	elf32-pj.c \
 	elf32-ppc.c \
 	elf32-pru.c \
+	elf32-riscv.lo \
+	elfxx-riscv.c \
 	elf32-rl78.c \
 	elf32-rx.c \
 	elf32-s390.c \
diff --git a/bfd/elfnn-riscv.c b/bfd/elfnn-riscv.c
index 36cbf1e..024dd32 100644
--- a/bfd/elfnn-riscv.c
+++ b/bfd/elfnn-riscv.c
@@ -1607,6 +1607,9 @@ perform_relocation (const reloc_howto_type *howto,
 		    bfd *input_bfd,
 		    bfd_byte *contents)
 {
+  bool handle_call_special = false;
+  bfd_vma value_call_auipc, value_call_jalr;
+
   if (howto->pc_relative)
     value -= sec_addr (input_section) + rel->r_offset;
   value += rel->r_addend;
@@ -1644,8 +1647,9 @@ perform_relocation (const reloc_howto_type *howto,
     case R_RISCV_CALL_PLT:
       if (ARCH_SIZE > 32 && !VALID_UTYPE_IMM (RISCV_CONST_HIGH_PART (value)))
 	return bfd_reloc_overflow;
-      value = ENCODE_UTYPE_IMM (RISCV_CONST_HIGH_PART (value))
-	      | (ENCODE_ITYPE_IMM (value) << 32);
+      value_call_auipc = ENCODE_UTYPE_IMM (RISCV_CONST_HIGH_PART (value));
+      value_call_jalr = ENCODE_ITYPE_IMM (value);
+      handle_call_special = true;
       break;
 
     case R_RISCV_JAL:
@@ -1717,6 +1721,21 @@ perform_relocation (const reloc_howto_type *howto,
       return bfd_reloc_notsupported;
     }
 
+  /* Handle R_RISCV_CALL and R_RISCV_CALL_PLT specially, in case the
+     host machine is 32-bit.  */
+  if (handle_call_special)
+    {
+      bfd_vma insn_auipc = riscv_get_insn (32, contents + rel->r_offset);
+      bfd_vma insn_jalr = riscv_get_insn (32, contents + rel->r_offset + 0x4);
+      bfd_vma dst_mask_auipc = ENCODE_UTYPE_IMM (-1U);
+      bfd_vma dst_mask_jalr = ENCODE_ITYPE_IMM (-1U);
+      insn_auipc = (insn_auipc & ~dst_mask_auipc) | (value_call_auipc & dst_mask_auipc);
+      insn_jalr = (insn_jalr & ~dst_mask_jalr) | (value_call_jalr & dst_mask_jalr);
+      riscv_put_insn (32, insn_auipc, contents + rel->r_offset);
+      riscv_put_insn (32, insn_jalr, contents + rel->r_offset + 0x4);
+      return bfd_reloc_ok;
+    }
+
   bfd_vma word;
   if (riscv_is_insn_reloc (howto))
     word = riscv_get_insn (howto->bitsize, contents + rel->r_offset);
@@ -1727,7 +1746,6 @@ perform_relocation (const reloc_howto_type *howto,
     riscv_put_insn (howto->bitsize, word, contents + rel->r_offset);
   else
     bfd_put (howto->bitsize, input_bfd, word, contents + rel->r_offset);
-
   return bfd_reloc_ok;
 }
 
diff --git a/bfd/elfxx-riscv.c b/bfd/elfxx-riscv.c
index 3ffbaad..858d29e 100644
--- a/bfd/elfxx-riscv.c
+++ b/bfd/elfxx-riscv.c
@@ -264,8 +264,7 @@ static reloc_howto_type howto_table[] =
 	 "R_RISCV_CALL",		/* name */
 	 false,				/* partial_inplace */
 	 0,				/* src_mask */
-	 ENCODE_UTYPE_IMM (-1U) | ((bfd_vma) ENCODE_ITYPE_IMM (-1U) << 32),
-					/* dst_mask */
+	 MINUS_ONE,			/* dst_mask */
 	 true),				/* pcrel_offset */
 
   /* Like R_RISCV_CALL, but not locally binding.  */
@@ -280,8 +279,7 @@ static reloc_howto_type howto_table[] =
 	 "R_RISCV_CALL_PLT",		/* name */
 	 false,				/* partial_inplace */
 	 0,				/* src_mask */
-	 ENCODE_UTYPE_IMM (-1U) | ((bfd_vma) ENCODE_ITYPE_IMM (-1U) << 32),
-					/* dst_mask */
+	 MINUS_ONE,			/* dst_mask */
 	 true),				/* pcrel_offset */
 
   /* High 20 bits of 32-bit PC-relative GOT access.  */
-- 
2.7.4

Re: [RFC] RISC-V: Fix the 32-bit --enable-targets=all build breakage.

[Nelson Chu]

Hi Alan,

I see you have a gcc patch to fix the problem of libopcodes, so should
we still need this patch to fix the --target-all on the 32-bit host
machine?

Thanks
Nelson

On Fri, Nov 12, 2021 at 4:35 PM Nelson Chu <nelson.chu@sifive.com> wrote:
>
> This should be the same problem as the commit,
> ba9b3ef5ee666467b67780e81f868c432f4fc56d.
>
> Generally, the elfxx-* file should be compiled both for 32-bit and 64-bit
> machine, since xx can be 32 and 64.  However, seems like the elfxx-riscv.c
> and elf32-riscv.c are missing for BFD32_BACKENDS_CFILES.
>
> This patch try to add the missing linker files for the BFD32_BACKENDS, but
> the dst_mask of R_RISCV_CALL may be overflow for the host 32-bit machine.
> Therefore, we rewrite the R_RISCV_CALL stuff in the perform_relocation, use
> the two 32-bit bfd_vma values to encode it, since the R_RISCV_CALL should be
> two instructions, rather than one address.
>
> bfd/
>         * Makefile.am: Added elfxx-riscv.* and elf32-riscv.* to
>         BFD32_BACKENDS and BFD32_BACKENDS_CFILES.
>         * Makefile.in: Regernated.
>         * elfnn-riscv.c (perform_relocation): Relocate the R_RISCV_CALL and
>         R_RISCV_CALL_PLT relocs by two bfd_vma values, since they are two
>         insturctions in fact.  So that the rv32 riscv toolchain built on
>         the host 32-bit machine should work.
>         * elfxx-riscv.c (howto_table): Let elfnn-riscv.c:perform_relocation
>         to handle the dst_mask of R_RISCV_CALL and R_RISCV_CALL_PLT specially.
>         Set dst_mask to MINUS_ONE.
> ---
>  bfd/Makefile.am   |  4 ++++
>  bfd/Makefile.in   |  4 ++++
>  bfd/elfnn-riscv.c | 24 +++++++++++++++++++++---
>  bfd/elfxx-riscv.c |  6 ++----
>  4 files changed, 31 insertions(+), 7 deletions(-)
>
> diff --git a/bfd/Makefile.am b/bfd/Makefile.am
> index 097177b..1a935d6 100644
> --- a/bfd/Makefile.am
> +++ b/bfd/Makefile.am
> @@ -336,6 +336,8 @@ BFD32_BACKENDS = \
>         elf32-pj.lo \
>         elf32-ppc.lo \
>         elf32-pru.lo \
> +       elf32-riscv.lo \
> +       elfxx-riscv.lo \
>         elf32-rl78.lo \
>         elf32-rx.lo \
>         elf32-s390.lo \
> @@ -469,6 +471,8 @@ BFD32_BACKENDS_CFILES = \
>         elf32-pj.c \
>         elf32-ppc.c \
>         elf32-pru.c \
> +       elf32-riscv.c \
> +       elfxx-riscv.c \
>         elf32-rl78.c \
>         elf32-rx.c \
>         elf32-s390.c \
> diff --git a/bfd/Makefile.in b/bfd/Makefile.in
> index a76b653..3817167 100644
> --- a/bfd/Makefile.in
> +++ b/bfd/Makefile.in
> @@ -762,6 +762,8 @@ BFD32_BACKENDS = \
>         elf32-pj.lo \
>         elf32-ppc.lo \
>         elf32-pru.lo \
> +       elf32-riscv.lo \
> +       elfxx-riscv.lo \
>         elf32-rl78.lo \
>         elf32-rx.lo \
>         elf32-s390.lo \
> @@ -895,6 +897,8 @@ BFD32_BACKENDS_CFILES = \
>         elf32-pj.c \
>         elf32-ppc.c \
>         elf32-pru.c \
> +       elf32-riscv.lo \
> +       elfxx-riscv.c \
>         elf32-rl78.c \
>         elf32-rx.c \
>         elf32-s390.c \
> diff --git a/bfd/elfnn-riscv.c b/bfd/elfnn-riscv.c
> index 36cbf1e..024dd32 100644
> --- a/bfd/elfnn-riscv.c
> +++ b/bfd/elfnn-riscv.c
> @@ -1607,6 +1607,9 @@ perform_relocation (const reloc_howto_type *howto,
>                     bfd *input_bfd,
>                     bfd_byte *contents)
>  {
> +  bool handle_call_special = false;
> +  bfd_vma value_call_auipc, value_call_jalr;
> +
>    if (howto->pc_relative)
>      value -= sec_addr (input_section) + rel->r_offset;
>    value += rel->r_addend;
> @@ -1644,8 +1647,9 @@ perform_relocation (const reloc_howto_type *howto,
>      case R_RISCV_CALL_PLT:
>        if (ARCH_SIZE > 32 && !VALID_UTYPE_IMM (RISCV_CONST_HIGH_PART (value)))
>         return bfd_reloc_overflow;
> -      value = ENCODE_UTYPE_IMM (RISCV_CONST_HIGH_PART (value))
> -             | (ENCODE_ITYPE_IMM (value) << 32);
> +      value_call_auipc = ENCODE_UTYPE_IMM (RISCV_CONST_HIGH_PART (value));
> +      value_call_jalr = ENCODE_ITYPE_IMM (value);
> +      handle_call_special = true;
>        break;
>
>      case R_RISCV_JAL:
> @@ -1717,6 +1721,21 @@ perform_relocation (const reloc_howto_type *howto,
>        return bfd_reloc_notsupported;
>      }
>
> +  /* Handle R_RISCV_CALL and R_RISCV_CALL_PLT specially, in case the
> +     host machine is 32-bit.  */
> +  if (handle_call_special)
> +    {
> +      bfd_vma insn_auipc = riscv_get_insn (32, contents + rel->r_offset);
> +      bfd_vma insn_jalr = riscv_get_insn (32, contents + rel->r_offset + 0x4);
> +      bfd_vma dst_mask_auipc = ENCODE_UTYPE_IMM (-1U);
> +      bfd_vma dst_mask_jalr = ENCODE_ITYPE_IMM (-1U);
> +      insn_auipc = (insn_auipc & ~dst_mask_auipc) | (value_call_auipc & dst_mask_auipc);
> +      insn_jalr = (insn_jalr & ~dst_mask_jalr) | (value_call_jalr & dst_mask_jalr);
> +      riscv_put_insn (32, insn_auipc, contents + rel->r_offset);
> +      riscv_put_insn (32, insn_jalr, contents + rel->r_offset + 0x4);
> +      return bfd_reloc_ok;
> +    }
> +
>    bfd_vma word;
>    if (riscv_is_insn_reloc (howto))
>      word = riscv_get_insn (howto->bitsize, contents + rel->r_offset);
> @@ -1727,7 +1746,6 @@ perform_relocation (const reloc_howto_type *howto,
>      riscv_put_insn (howto->bitsize, word, contents + rel->r_offset);
>    else
>      bfd_put (howto->bitsize, input_bfd, word, contents + rel->r_offset);
> -
>    return bfd_reloc_ok;
>  }
>
> diff --git a/bfd/elfxx-riscv.c b/bfd/elfxx-riscv.c
> index 3ffbaad..858d29e 100644
> --- a/bfd/elfxx-riscv.c
> +++ b/bfd/elfxx-riscv.c
> @@ -264,8 +264,7 @@ static reloc_howto_type howto_table[] =
>          "R_RISCV_CALL",                /* name */
>          false,                         /* partial_inplace */
>          0,                             /* src_mask */
> -        ENCODE_UTYPE_IMM (-1U) | ((bfd_vma) ENCODE_ITYPE_IMM (-1U) << 32),
> -                                       /* dst_mask */
> +        MINUS_ONE,                     /* dst_mask */
>          true),                         /* pcrel_offset */
>
>    /* Like R_RISCV_CALL, but not locally binding.  */
> @@ -280,8 +279,7 @@ static reloc_howto_type howto_table[] =
>          "R_RISCV_CALL_PLT",            /* name */
>          false,                         /* partial_inplace */
>          0,                             /* src_mask */
> -        ENCODE_UTYPE_IMM (-1U) | ((bfd_vma) ENCODE_ITYPE_IMM (-1U) << 32),
> -                                       /* dst_mask */
> +        MINUS_ONE,                     /* dst_mask */
>          true),                         /* pcrel_offset */
>
>    /* High 20 bits of 32-bit PC-relative GOT access.  */
> --
> 2.7.4
>

Re: [RFC] RISC-V: Fix the 32-bit --enable-targets=all build breakage.

[Alan Modra]

On Fri, Nov 12, 2021 at 04:38:08PM +0800, Nelson Chu wrote:
> Hi Alan,
> 
> I see you have a gcc patch to fix the problem of libopcodes, so should
> we still need this patch to fix the --target-all on the 32-bit host
> machine?

No, the patch I just committed now fixed the build problem.

Whether you want to enable riscv32 support with a 32-bit bfd is really
a separate issue.

-- 
Alan Modra
Australia Development Lab, IBM

Re: [RFC] RISC-V: Fix the 32-bit --enable-targets=all build breakage.

[Nelson Chu]

On Fri, Nov 12, 2021 at 5:35 PM Alan Modra <amodra@gmail.com> wrote:
>
> On Fri, Nov 12, 2021 at 04:38:08PM +0800, Nelson Chu wrote:
> > Hi Alan,
> >
> > I see you have a gcc patch to fix the problem of libopcodes, so should
> > we still need this patch to fix the --target-all on the 32-bit host
> > machine?
>
> No, the patch I just committed now fixed the build problem.

Thanks for helping to fix this :)

> Whether you want to enable riscv32 support with a 32-bit bfd is really
> a separate issue.

OK, I will figure this out, thanks.

Nelson

> --
> Alan Modra
> Australia Development Lab, IBM

Re: [RFC] RISC-V: Fix the 32-bit --enable-targets=all build breakage.

[Luis Machado]

Hi,

I'm still seeing the following build failure for 32-bit ARM with 
--enable-targets=all:

--

libsim.a(sim-main.o): In function `store_rd':
/builds/binutils-gdb-armhf-bionic/sim/riscv/../../../../repos/binutils-gdb/sim/riscv/sim-main.c:73: 
undefined reference to `riscv_gpr_names_abi'
/builds/binutils-gdb-armhf-bionic/sim/riscv/../../../../repos/binutils-gdb/sim/riscv/sim-main.c:71: 
undefined reference to `riscv_gpr_names_abi'
libsim.a(sim-main.o): In function `initialize_cpu':
/builds/binutils-gdb-armhf-bionic/sim/riscv/../../../../repos/binutils-gdb/sim/riscv/sim-main.c:1109: 
undefined reference to `riscv_opcodes'
collect2: error: ld returned 1 exit status
Makefile:253: recipe for target 'run' failed
make[4]: *** [run] Error 1
make[4]: Leaving directory '/builds/binutils-gdb-armhf-bionic/sim/riscv'
Makefile:2279: recipe for target 'all-recursive' failed
make[3]: *** [all-recursive] Error 1
make[3]: Leaving directory '/builds/binutils-gdb-armhf-bionic/sim'
Makefile:1604: recipe for target 'all' failed
make[2]: *** [all] Error 2
make[2]: Leaving directory '/builds/binutils-gdb-armhf-bionic/sim'
Makefile:11218: recipe for target 'all-sim' failed
make[1]: *** [all-sim] Error 2
make[1]: Leaving directory '/builds/binutils-gdb-armhf-bionic'
Makefile:999: recipe for target 'all' failed
make: *** [all] Error 2

--

I thought Alan's fix would address this, but it looks like it doesn't. 
It seems these symbols are not available when doing an all-targets build.

On 11/12/21 6:42 AM, Nelson Chu wrote:
> On Fri, Nov 12, 2021 at 5:35 PM Alan Modra <amodra@gmail.com> wrote:
>>
>> On Fri, Nov 12, 2021 at 04:38:08PM +0800, Nelson Chu wrote:
>>> Hi Alan,
>>>
>>> I see you have a gcc patch to fix the problem of libopcodes, so should
>>> we still need this patch to fix the --target-all on the 32-bit host
>>> machine?
>>
>> No, the patch I just committed now fixed the build problem.
> 
> Thanks for helping to fix this :)
> 
>> Whether you want to enable riscv32 support with a 32-bit bfd is really
>> a separate issue.
> 
> OK, I will figure this out, thanks.
> 
> Nelson
> 
>> --
>> Alan Modra
>> Australia Development Lab, IBM

Re: [RFC] RISC-V: Fix the 32-bit --enable-targets=all build breakage.

[Andrew Burgess]

* Nelson Chu <nelson.chu@sifive.com> [2021-11-12 17:42:42 +0800]:

> On Fri, Nov 12, 2021 at 5:35 PM Alan Modra <amodra@gmail.com> wrote:
> >
> > On Fri, Nov 12, 2021 at 04:38:08PM +0800, Nelson Chu wrote:
> > > Hi Alan,
> > >
> > > I see you have a gcc patch to fix the problem of libopcodes, so should
> > > we still need this patch to fix the --target-all on the 32-bit host
> > > machine?
> >
> > No, the patch I just committed now fixed the build problem.
> 
> Thanks for helping to fix this :)
> 
> > Whether you want to enable riscv32 support with a 32-bit bfd is really
> > a separate issue.
> 
> OK, I will figure this out, thanks.

Just to keep you in the loop, I ran into GDB build problems caused by
the riscv (and mips) libopcodes and bfd libraries not being built,
when building bfd in 32-bit mode.  There's a dependency between GDB
and libopcodes.  I posted this patch:

  https://sourceware.org/pipermail/gdb-patches/2021-December/184365.html

I'll probably push the patch next week unless someone gives a
compelling reason not too, but I'm more than happy to revert the riscv
parts if/when the riscv libopcodes is built in 32-bit mode.

Thanks,
Andrew

Re: [RFC] RISC-V: Fix the 32-bit --enable-targets=all build breakage.

[Luis Machado]

Hi,

On 12/10/21 7:40 AM, Andrew Burgess via Binutils wrote:
> * Nelson Chu <nelson.chu@sifive.com> [2021-11-12 17:42:42 +0800]:
> 
>> On Fri, Nov 12, 2021 at 5:35 PM Alan Modra <amodra@gmail.com> wrote:
>>>
>>> On Fri, Nov 12, 2021 at 04:38:08PM +0800, Nelson Chu wrote:
>>>> Hi Alan,
>>>>
>>>> I see you have a gcc patch to fix the problem of libopcodes, so should
>>>> we still need this patch to fix the --target-all on the 32-bit host
>>>> machine?
>>>
>>> No, the patch I just committed now fixed the build problem.
>>
>> Thanks for helping to fix this :)
>>
>>> Whether you want to enable riscv32 support with a 32-bit bfd is really
>>> a separate issue.
>>
>> OK, I will figure this out, thanks.
> 
> Just to keep you in the loop, I ran into GDB build problems caused by
> the riscv (and mips) libopcodes and bfd libraries not being built,
> when building bfd in 32-bit mode.  There's a dependency between GDB
> and libopcodes.  I posted this patch:
> 
>    https://sourceware.org/pipermail/gdb-patches/2021-December/184365.html
> 
> I'll probably push the patch next week unless someone gives a
> compelling reason not too, but I'm more than happy to revert the riscv
> parts if/when the riscv libopcodes is built in 32-bit mode.
> 
> Thanks,
> Andrew
> 

I suppose the binutils issue still remains in this case? As reported here:

https://sourceware.org/pipermail/binutils/2021-December/118736.html

Re: [RFC] RISC-V: Fix the 32-bit --enable-targets=all build breakage.

[Andrew Burgess]

* Luis Machado via Binutils <binutils@sourceware.org> [2021-12-10 09:36:12 -0300]:

> Hi,
> 
> On 12/10/21 7:40 AM, Andrew Burgess via Binutils wrote:
> > * Nelson Chu <nelson.chu@sifive.com> [2021-11-12 17:42:42 +0800]:
> > 
> > > On Fri, Nov 12, 2021 at 5:35 PM Alan Modra <amodra@gmail.com> wrote:
> > > > 
> > > > On Fri, Nov 12, 2021 at 04:38:08PM +0800, Nelson Chu wrote:
> > > > > Hi Alan,
> > > > > 
> > > > > I see you have a gcc patch to fix the problem of libopcodes, so should
> > > > > we still need this patch to fix the --target-all on the 32-bit host
> > > > > machine?
> > > > 
> > > > No, the patch I just committed now fixed the build problem.
> > > 
> > > Thanks for helping to fix this :)
> > > 
> > > > Whether you want to enable riscv32 support with a 32-bit bfd is really
> > > > a separate issue.
> > > 
> > > OK, I will figure this out, thanks.
> > 
> > Just to keep you in the loop, I ran into GDB build problems caused by
> > the riscv (and mips) libopcodes and bfd libraries not being built,
> > when building bfd in 32-bit mode.  There's a dependency between GDB
> > and libopcodes.  I posted this patch:
> > 
> >    https://sourceware.org/pipermail/gdb-patches/2021-December/184365.html
> > 
> > I'll probably push the patch next week unless someone gives a
> > compelling reason not too, but I'm more than happy to revert the riscv
> > parts if/when the riscv libopcodes is built in 32-bit mode.
> > 
> > Thanks,
> > Andrew
> > 
> 
> I suppose the binutils issue still remains in this case? As reported here:
> 
> https://sourceware.org/pipermail/binutils/2021-December/118736.html

I guess so.  I didn't see that in the 32-bit build I was doing, but
I've done nothing that would solve that problem.

Thanks,
Andrew

Re: [RFC] RISC-V: Fix the 32-bit --enable-targets=all build breakage.

[Luis Machado]

On 12/10/21 11:57 AM, Andrew Burgess wrote:
> * Luis Machado via Binutils <binutils@sourceware.org> [2021-12-10 09:36:12 -0300]:
> 
>> Hi,
>>
>> On 12/10/21 7:40 AM, Andrew Burgess via Binutils wrote:
>>> * Nelson Chu <nelson.chu@sifive.com> [2021-11-12 17:42:42 +0800]:
>>>
>>>> On Fri, Nov 12, 2021 at 5:35 PM Alan Modra <amodra@gmail.com> wrote:
>>>>>
>>>>> On Fri, Nov 12, 2021 at 04:38:08PM +0800, Nelson Chu wrote:
>>>>>> Hi Alan,
>>>>>>
>>>>>> I see you have a gcc patch to fix the problem of libopcodes, so should
>>>>>> we still need this patch to fix the --target-all on the 32-bit host
>>>>>> machine?
>>>>>
>>>>> No, the patch I just committed now fixed the build problem.
>>>>
>>>> Thanks for helping to fix this :)
>>>>
>>>>> Whether you want to enable riscv32 support with a 32-bit bfd is really
>>>>> a separate issue.
>>>>
>>>> OK, I will figure this out, thanks.
>>>
>>> Just to keep you in the loop, I ran into GDB build problems caused by
>>> the riscv (and mips) libopcodes and bfd libraries not being built,
>>> when building bfd in 32-bit mode.  There's a dependency between GDB
>>> and libopcodes.  I posted this patch:
>>>
>>>     https://sourceware.org/pipermail/gdb-patches/2021-December/184365.html
>>>
>>> I'll probably push the patch next week unless someone gives a
>>> compelling reason not too, but I'm more than happy to revert the riscv
>>> parts if/when the riscv libopcodes is built in 32-bit mode.
>>>
>>> Thanks,
>>> Andrew
>>>
>>
>> I suppose the binutils issue still remains in this case? As reported here:
>>
>> https://sourceware.org/pipermail/binutils/2021-December/118736.html
> 
> I guess so.  I didn't see that in the 32-bit build I was doing, but
> I've done nothing that would solve that problem.

Thanks for clarifying it. It's great that we can get the gdb builds 
going, so thanks for that.

Re: [RFC] RISC-V: Fix the 32-bit --enable-targets=all build breakage.

[Nelson Chu]

On Fri, Dec 10, 2021 at 6:40 PM Andrew Burgess <aburgess@redhat.com> wrote:
>
> * Nelson Chu <nelson.chu@sifive.com> [2021-11-12 17:42:42 +0800]:
>
> > On Fri, Nov 12, 2021 at 5:35 PM Alan Modra <amodra@gmail.com> wrote:
> > >
> > > On Fri, Nov 12, 2021 at 04:38:08PM +0800, Nelson Chu wrote:
> > > > Hi Alan,
> > > >
> > > > I see you have a gcc patch to fix the problem of libopcodes, so should
> > > > we still need this patch to fix the --target-all on the 32-bit host
> > > > machine?
> > >
> > > No, the patch I just committed now fixed the build problem.
> >
> > Thanks for helping to fix this :)
> >
> > > Whether you want to enable riscv32 support with a 32-bit bfd is really
> > > a separate issue.
> >
> > OK, I will figure this out, thanks.
>
> Just to keep you in the loop, I ran into GDB build problems caused by
> the riscv (and mips) libopcodes and bfd libraries not being built,
> when building bfd in 32-bit mode.  There's a dependency between GDB
> and libopcodes.  I posted this patch:

Yeah, thanks.  I have been paying attention to this problem, but I
don’t have good solutions and ideas like you, Alan, Jim, Mike and
Luis, so I can’t give some comments that I think are helpful...

>   https://sourceware.org/pipermail/gdb-patches/2021-December/184365.html
>
> I'll probably push the patch next week unless someone gives a
> compelling reason not too, but I'm more than happy to revert the riscv
> parts if/when the riscv libopcodes is built in 32-bit mode.

I'm OK with this, we could probably revert it when my previous patch
does work.  Thanks for all your work and Jim's suggestion.  Thanks for
Alan's help, I think his idea is really great, but as I said, I can’t
give some comments that I think are helpful for now...  So I create a
bugzilla as follows, in case I forgot the details in the future, or
someone is interested to fix the problem,
https://sourceware.org/bugzilla/show_bug.cgi?id=28684

Thanks for all
Nelson

> Thanks,
> Andrew
>

Re: [RFC] RISC-V: Fix the 32-bit --enable-targets=all build breakage.

[Luis Machado]

Hi Alan,

On 11/12/21 6:34 AM, Alan Modra via Binutils wrote:
> On Fri, Nov 12, 2021 at 04:38:08PM +0800, Nelson Chu wrote:
>> Hi Alan,
>>
>> I see you have a gcc patch to fix the problem of libopcodes, so should
>> we still need this patch to fix the --target-all on the 32-bit host
>> machine?
> 
> No, the patch I just committed now fixed the build problem.
> 

Unfortunately it seems to have side effects. The bpf sim still gets 
built for a 32-bit host, but its opcodes dependencies are no longer 
there, resulting in undefined references:

libsim.a(sim-close.o): In function `sim_close':
binutils-gdb/sim/bpf/../common/sim-close.c:43: undefined reference to 
`bpf_cgen_cpu_close'
libsim.a(sim-if.o): In function `sim_open':
binutils-gdb/sim/bpf/sim-if.c:166: undefined reference to 
`bpf_cgen_cpu_open_1'
binutils-gdb/sim/bpf/sim-if.c:179: undefined reference to 
`bpf_cgen_init_dis'

The references are contained in hese files:

opcodes/bpf-desc.c:bpf_cgen_cpu_close (CGEN_CPU_DESC cd)

opcodes/bpf-desc.c:bpf_cgen_cpu_open_1 (const char *mach_name, enum 
cgen_endian endian)

opcodes/bpf-dis.c:bpf_cgen_init_dis (CGEN_CPU_DESC cd)


> Whether you want to enable riscv32 support with a 32-bit bfd is really
> a separate issue.
> 

Re: [RFC] RISC-V: Fix the 32-bit --enable-targets=all build breakage.

[Alan Modra]

On Wed, Nov 17, 2021 at 10:52:26AM -0300, Luis Machado wrote:
> Hi Alan,
> 
> On 11/12/21 6:34 AM, Alan Modra via Binutils wrote:
> > On Fri, Nov 12, 2021 at 04:38:08PM +0800, Nelson Chu wrote:
> > > Hi Alan,
> > > 
> > > I see you have a gcc patch to fix the problem of libopcodes, so should
> > > we still need this patch to fix the --target-all on the 32-bit host
> > > machine?
> > 
> > No, the patch I just committed now fixed the build problem.
> > 
> 
> Unfortunately it seems to have side effects. The bpf sim still gets built
> for a 32-bit host, but its opcodes dependencies are no longer there,
> resulting in undefined references:

Did such a sim configuration actually work prior to my patch?  If it
did, I'm willing to move the bpf files from TARGET64_LIBOPCODES_CFILES
to TARGET32_LIBOPCODES_CFILES.

-- 
Alan Modra
Australia Development Lab, IBM

Re: [RFC] RISC-V: Fix the 32-bit --enable-targets=all build breakage.

[Mike Frysinger]

[-- Attachment #1: Type: text/plain, Size: 992 bytes --]

On 18 Nov 2021 11:32, Alan Modra via Binutils wrote:
> On Wed, Nov 17, 2021 at 10:52:26AM -0300, Luis Machado wrote:
> > On 11/12/21 6:34 AM, Alan Modra via Binutils wrote:
> > > On Fri, Nov 12, 2021 at 04:38:08PM +0800, Nelson Chu wrote:
> > > > Hi Alan,
> > > > 
> > > > I see you have a gcc patch to fix the problem of libopcodes, so should
> > > > we still need this patch to fix the --target-all on the 32-bit host
> > > > machine?
> > > 
> > > No, the patch I just committed now fixed the build problem.
> > > 
> > 
> > Unfortunately it seems to have side effects. The bpf sim still gets built
> > for a 32-bit host, but its opcodes dependencies are no longer there,
> > resulting in undefined references:
> 
> Did such a sim configuration actually work prior to my patch?  If it
> did, I'm willing to move the bpf files from TARGET64_LIBOPCODES_CFILES
> to TARGET32_LIBOPCODES_CFILES.

`--host=i686-linux-gnu --enable-targets=all --enable-sim` was working
-mike

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]