2020-04-14 13:20:04

by Fangrui Song

[permalink] [raw]
Subject: Re: [PATCH v4] MIPS: Truncate link address into 32bit for 32bit kernel

On 2020-04-13, Jiaxun Yang wrote:
>On Mon, 13 Apr 2020 07:59:29 +0100 (BST)
>"Maciej W. Rozycki" <[email protected]> wrote:
>
>> On Mon, 13 Apr 2020, Jiaxun Yang wrote:
>>
>> > LLD failed to link vmlinux with 64bit load address for 32bit ELF
>> > while bfd will strip 64bit address into 32bit silently.
>> > To fix LLD build, we should truncate load address provided by
>> > platform into 32bit for 32bit kernel.
>>
>> Reviewed-by: Maciej W. Rozycki <[email protected]>
>>
>> > diff --git a/arch/mips/kernel/vmlinux.lds.S
>> > b/arch/mips/kernel/vmlinux.lds.S index a5f00ec73ea6..5226cd8e4bee
>> > 100644 --- a/arch/mips/kernel/vmlinux.lds.S
>> > +++ b/arch/mips/kernel/vmlinux.lds.S
>> > @@ -55,7 +55,7 @@ SECTIONS
>> > /* . = 0xa800000000300000; */
>> > . = 0xffffffff80300000;
>> > #endif
>> > - . = VMLINUX_LOAD_ADDRESS;
>> > + . = VMLINUX_LINK_ADDRESS;
>>
>> The CONFIG_BOOT_ELF64 cruft right above it looks interesting to me,
>> never have ever been used. We have had the current arrangement since:
>
>It confused me either.
>It's only used by SGI so probably it's time to rename it as
>BOOT_SEG_ELF64.
>
>Wish someone could clarify what is it.
>
>Thanks.

Agreed that -Ttext in

arch/mips/boot/compressed/Makefile
100: cmd_zld = $(LD) $(KBUILD_LDFLAGS) -Ttext $(VMLINUZ_LOAD_ADDRESS) -T $< $(vmlinuzobjs-y) -o $@

and a few other places are brittle. They need to be replaced with Output Section Address:
(https://sourceware.org/binutils/docs/ld/Output-Section-Address.html
https://github.com/llvm/llvm-project/blob/master/lld/docs/ELF/linker_script.rst#output-section-address)

-Ttext changes the address of .text . This can lead to the change of the
address of the text segment (RX), but this is not guaranteed (many
sections can be placed before .text and they are not affected).
See https://reviews.llvm.org/D70468 "[ELF] Error if -Ttext-segment is specified"


Reviewed-by: Fangrui Song <[email protected]>

>>
>> commit 923ec3d20eef9e36456868b590873ce39f17fe71
>> Author: Ralf Baechle <[email protected]>
>> Date: Wed Nov 6 22:16:38 2002 +0000
>>
>> Define load address in linker script instead of relying on the
>> deprecated and notoriously unreliable option -Ttext.
>>
>> and previously `-Ttext' was used with this script anyway, though not
>> very long, as the script was entirely ignored until:
>>
>> commit 7a782968041ffc4c2d89816238e2f8ea5cceddba
>> Author: Ralf Baechle <[email protected]>
>> Date: Thu Oct 31 23:54:21 2002 +0000
>>
>> Merge with Linux 2.5.36.
>>
>> Maciej
>
>--
>Jiaxun Yang


2020-05-13 22:26:53

by Maciej W. Rozycki

[permalink] [raw]
Subject: Re: [PATCH v4] MIPS: Truncate link address into 32bit for 32bit kernel

On Mon, 13 Apr 2020, Fangrui Song wrote:

> >> > diff --git a/arch/mips/kernel/vmlinux.lds.S
> >> > b/arch/mips/kernel/vmlinux.lds.S index a5f00ec73ea6..5226cd8e4bee
> >> > 100644 --- a/arch/mips/kernel/vmlinux.lds.S
> >> > +++ b/arch/mips/kernel/vmlinux.lds.S
> >> > @@ -55,7 +55,7 @@ SECTIONS
> >> > /* . = 0xa800000000300000; */
> >> > . = 0xffffffff80300000;
> >> > #endif
> >> > - . = VMLINUX_LOAD_ADDRESS;
> >> > + . = VMLINUX_LINK_ADDRESS;
> >>
> >> The CONFIG_BOOT_ELF64 cruft right above it looks interesting to me,
> >> never have ever been used. We have had the current arrangement since:
> >
> >It confused me either.
> >It's only used by SGI so probably it's time to rename it as
> >BOOT_SEG_ELF64.
> >
> >Wish someone could clarify what is it.
> >
> >Thanks.
>
> Agreed that -Ttext in
>
> arch/mips/boot/compressed/Makefile
> 100: cmd_zld = $(LD) $(KBUILD_LDFLAGS) -Ttext $(VMLINUZ_LOAD_ADDRESS) -T
> $< $(vmlinuzobjs-y) -o $@
>
> and a few other places are brittle. They need to be replaced with Output
> Section Address:
> (https://sourceware.org/binutils/docs/ld/Output-Section-Address.html
> https://github.com/llvm/llvm-project/blob/master/lld/docs/ELF/linker_script.rst#output-section-address)
>
> -Ttext changes the address of .text . This can lead to the change of the
> address of the text segment (RX), but this is not guaranteed (many
> sections can be placed before .text and they are not affected).

That is unrelated, but you're free to clean it up of course.

Maciej