[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <b111cb87-3f30-860a-640b-2f7264a16838@loongson.cn>
Date: Thu, 3 Aug 2023 19:36:24 +0800
From: Tiezhu Yang <yangtiezhu@...ngson.cn>
To: Peter Zijlstra <peterz@...radead.org>
Cc: Josh Poimboeuf <jpoimboe@...nel.org>,
Huacai Chen <chenhuacai@...nel.org>, loongarch@...ts.linux.dev,
linux-kernel@...r.kernel.org, loongson-kernel@...ts.loongnix.cn
Subject: Re: [PATCH v1 5/6] objtool: Add skipped member in struct reloc
On 07/25/2023 07:59 PM, Peter Zijlstra wrote:
> On Tue, Jul 25, 2023 at 04:15:09PM +0800, Tiezhu Yang wrote:
>> There exist multiple relocation types in one location, such as a pair of
>> R_LARCH_ADD32 and R_LARCH_SUB32 in section .rela.discard.unreachable and
>> .rela.discard.reachable on LoongArch.
>>
>> Here is an example:
>>
>> $ readelf -rW init/main.o
>>
>> Relocation section '.rela.discard.unreachable' at offset 0x3e20 contains 2 entries:
>> Offset Info Type Symbol's Value Symbol's Name + Addend
>> 0000000000000000 0000000a00000032 R_LARCH_ADD32 0000000000000000 .init.text + 230
>> 0000000000000000 0000001a00000037 R_LARCH_SUB32 0000000000000000 L0^A + 0
>>
>
> Please explain; why is this?
>
> How does:
>
> #define __annotate_unreachable(c) ({ \
> asm volatile(__stringify_label(c) ":\n\t" \
> ".pushsection .discard.unreachable\n\t" \
> ".long " __stringify_label(c) "b - .\n\t" \
> ".popsection\n\t" : : "i" (c)); \
> })
> #define annotate_unreachable() __annotate_unreachable(__COUNTER__)
>
> Manage to generate this..
>
Sorry for the late reply, let me try to explain it.
R_LARCH_ADD32 relocation type is 32-bit label addition,
R_LARCH_SUB32 relocation type is 32-bit label subtraction,
they are intended for local labels, the label difference
will be calculated as a constant before linking, preserve
a pair of R_LARCH_ADD32 and R_LARCH_SUB32 to fix the label
difference.
Here is a simple example:
-------------------------------
.L0
// do something
.L1
.data
.word .L0 - .L1
-------------------------------
NewDiff = 0
R_LARCH_ADD32:
NewDiff = NewDiff + .L0_Addr
R_LARCH_SUB32:
NewDiff = NewDiff - .L1_Addr
I discussed offline with the engineers who are familiar with gnu
assembler of LoongArch, maybe it can modify the gnu assembler
to use R_LARCH_32_PCREL to replace a pair of R_LARCH_ADD32 and
R_LARCH_SUB32, then I will test it again and drop this change if
possible.
Here is LoongArch ELF Relocations spec:
https://github.com/loongson/la-abi-specs/blob/release/laelf.adoc
R_LARCH_ADD32
32-bit in-place addition
*(int32_t *) PC += S + A
R_LARCH_SUB32
32-bit in-place subtraction
*(int32_t *) PC -= S + A
R_LARCH_32_PCREL
32-bit PC relative
(*(uint32_t *) PC) = (S+A-PC) [31 ... 0]
Thanks very much for your review comments, it is helpful to
make progress in the right direction.
Thanks,
Tiezhu
Powered by blists - more mailing lists