Add new relocation type support #57

cloudspurs · 2022-07-17T08:54:07Z

No description provided.

xen0n · 2022-07-17T14:20:20Z

Isn't there already #50? You may want to consider my suggestions there (and I submitted #56 for official approval) as naming is very important.

xry111 · 2022-07-20T08:15:59Z

docs/LoongArch-ELF-ABI-CN.adoc

+|67
+|R_LARCH_ABS_HI20
+|32 位绝对地址的[12:31]位
+|`(*(uint32_t *) PC) [24 ... 5] = (S+A) [31 ... 12]+`


What does these extra + mean?

have modified

xry111 · 2022-07-20T08:23:44Z

docs/LoongArch-ELF-ABI-CN.adoc

+
+|71
+|R_LARCH_PCALA_HI20
+|32位相对PC偏移的[12:31]位


This description seems different than the symbolic definition below. Say PC is 0x0f00 and (S + A) is 0x1000, (((S+A) & ~0xfff) - (PC & ~0xfff)) [31 ... 12] will be 0x1000 but the PC-relative offset will be 0x0100, and "the [12:31] bits of PC-relative offset" will be zero.

Added additional explanation: "all PC-relative offset calculations do not include the lowest 12 bits".

xry111 · 2022-07-20T08:30:26Z

docs/LoongArch-ELF-ABI-CN.adoc

+
+|100
+|R_LARCH_RELAX
+|暂未实现


Even if it's not implemented, you should give it a definition. Otherwise you should just leave it undefined.

This relocation is expected to do some optimizations, do or not won't affect correctness.
The linker has implemented some specific instructions, but the assembler still has some support that has not been completed.
If it is deleted, the current version linker can not to link next version object (maybe soon to be patched).

Then maybe a simple description like "mark a instruction suitable for link-time relaxation" or something (does @xen0n have a better idea?)

if sym - pc >= 0xffe00000 && sym -pc <= 0x1ffffc
can do relax:
pcalau12i $r12, sym
addi.d $r12, $r12, sym
=>
pcaddi $r12, sym
nop

BTW: make sure you all read MaskRay's excellent writeups on the topic of linker relaxations before implementing any of these. ;-)

xry111

Even if we ignore the debateful naming part, and we don't care "human ego" of reviewers outside Loongson, I don't think we can merge this in status quo. If you allow me to speak candidly: this just seems like a scratch paper where somebody wrote some random thoughts, instead of a formal technical documentation.

xry111 · 2022-07-20T08:36:42Z

docs/LoongArch-ELF-ABI-CN.adoc

+|72
+|R_LARCH_PCALA_LO12
+|32位相对PC偏移的[0:11]位
+|`(*(uint32_t *) PC) [21 ... 10] = (S+A) [11 ... 0]+`


The text description is completely different than the symbolic definition. The symbolic description is exactly same as R_LARCH_ABS_LO12.

I guess you want to use something like:

pcalau12i $t0, %pcala_hi20(something) ld.d $t0, $t0, %lo12(something)

to load something, if I interpret @SixWeining's comment in https://reviews.llvm.org/D128427 correctly. Then, (1) you don't need a separate "%pcala_lo12" (it's same as "%abs_lo12" so only one "%lo12" is enough); (2) you need to alter the definition of %pcala_hi20 to account the carry from %lo12 into consideration. Like:

R_LARCH_PCALA_HI20 ::= (*(u32 *)PC) [24..5] = (((S+A+0x800) & ~0xfff) - (PC & ~0xfff))[31..12]

R_LARCH_ABS_LO12 and R_LARCH_PCALA_LO12 are calculated in the same way. Only take the lower 12 bits without other operations. In earlier versions it was the same.

This is an error and has been modified.

xry111 · 2022-07-20T08:58:39Z

It's better to consistently use [a..b] or [a:b] notion, instead of mixing up Python-style [a:b] notion with Rust-style [a..b] notion.

ChenghuaXu · 2022-07-20T09:48:58Z

It's better to consistently use [a..b] or [a:b] notion, instead of mixing up Python-style [a:b] notion with Rust-style [a..b] notion.

@cloudspurs

xen0n · 2022-07-21T08:25:16Z

On a related note, what does the "ALA" in "PCALA" mean? I don't think we've seen any full form anywhere for this abbreviation.

xry111 · 2022-07-21T09:01:38Z

docs/LoongArch-ELF-ABI-EN.adoc

+|[31 ... 12] bits of 32/64-bit PC-relative offset
+|`+(*(uint32_t *) PC) [24 ... 5] = (((S+A) & ~0xfff) - (PC & ~0xfff)) [31 ... 12]+`


Suggested change

|[31 ... 12] bits of 32/64-bit PC-relative offset

|`+(*(uint32_t *) PC) [24 ... 5] = (((S+A) & ~0xfff) - (PC & ~0xfff)) [31 ... 12]+`

|[31 ... 12] bits of 32/64-bit PC-relative offset to the nearest 4KB bounary for the target

|`+(*(uint32_t *) PC) [24 ... 5] = (((S+A+0x800) & ~0xfff) - (PC & ~0xfff)) [31 ... 12]+`

For example, S+A = 0x12001800 and PC = 0x12000010. We want:

pcalau12i $t0, 0x2 # 0x2000 >> 12 addi.d $t0, $t0, 0x800 # -2048

to get S+A. Without +0x800 we get pcalau12i $t0, $0x1 then we can't use addi.d (ori is needed for a positive offset >= 2048). But the binutils patch uses addi.d, not ori.

pcalau12i $t0, 0x2 # (S+A+0x800)&(~0xfff) - (PC & (~0xfff))

yetist · 2022-07-23T10:42:46Z

docs/LoongArch-ELF-ABI-CN.adoc

+|79
+|R_LARCH_GOT_HI20
+|GOT 表项 32/64 位绝对地址的 [31 ... 12] 位
+|`+(*(uint32_t *) PC) [24 ... 5] = (GP+G) [31 ... 12]+`
+
+|80
+|R_LARCH_GOT_LO12
+|GOT 表项 32/64 位绝对地址的 [11 ... 0] 位
+|`+(*(uint32_t *) PC) [21 ... 10] = (GP+G) [11 ... 0]+`


The names of 79 and 80 are not the same as in the code, which is correct?

The names in the binutils code are:

R_LARCH_GOT64_HI20 79 R_LARCH_GOT64_LO12 80

https://github.com/loongson/binutils-gdb/blob/291cb57c2947868201be6c2b0b903d318a7df90d/include/elf/loongarch.h#L161

Please see https://sourceware.org/pipermail/binutils/2022-July/121937.html

ChenghuaXu

Ok.

xen0n · 2022-07-25T02:18:19Z

@xry111 hasn't approved, why merge?

xry111 · 2022-07-25T06:35:48Z

See #58 for a fix of "details" for PC-relative relocs.

gcc-13 has been pretty stable over the weeks I have been using it, so keyword one latest gcc-13 snapshot for wider testing. This should bring the entire loong world to make use of the new-style relocs [1] [2]. Currently the loong kernel cannot be natively built with such a toolchain combination, but a fix is in the works [3]. Because no Gentoo kernel source package has been keyworded ~loong, and the current userbase can be effectively notified in other ways (the Telegram Loongson user group for example), I decided to proceed with the keywording to enable widespread testing of the fresh loong code. The various toolchain projects all cut their releases just recently so we should have plenty of time to polish things before the next big release. [1]: loongson/LoongArch-Documentation#57 [2]: https://sourceware.org/git/?p=binutils-gdb.git;a=commitdiff;h=6d13722a97cee3fd397e116bde3bcedbb1e220be;hp=5d471bd907be60e9858b22cdf4fd10ddc0f6ee1a [3]: https://lore.kernel.org/loongarch/CAAhV-H7oQkGVqZV_s7tXjDGBL9yQ+CsUkCv7tC+pWuW2MXe9LA@mail.gmail.com/T/#me74ccdafb0d98d5d13fa2a82356ec5dff1a49a89 Signed-off-by: WANG Xuerui <[email protected]>

Eldeloro1 · 2022-08-11T19:56:54Z

No description provided.

These relocation types are used by GNU binutils >= 2.40 and GCC >= 13. Add their definitions so we will be able to use them in later patches. Link: loongson/LoongArch-Documentation#57 Signed-off-by: Xi Ruoyao <[email protected]>

These relocation types are used by GNU binutils >= 2.40 and GCC >= 13. Add their definitions so we will be able to use them in later patches. Link: loongson/LoongArch-Documentation#57 Tested-by: WANG Xuerui <[email protected]> Signed-off-by: Xi Ruoyao <[email protected]> Signed-off-by: Huacai Chen <[email protected]>

Link: loongson/LoongArch-Documentation#57 Signed-off-by: Xiaotian Wu <[email protected]>

These relocation types are used by GNU binutils >= 2.40 and GCC >= 13. Add their definitions so we will be able to use them in later patches. Link: loongson/LoongArch-Documentation#57 Tested-by: WANG Xuerui <[email protected]> Signed-off-by: Xi Ruoyao <[email protected]> Signed-off-by: Huacai Chen <[email protected]>

LoongArch inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I5OHOB -------------------------------- These relocation types are used by GNU binutils >= 2.40 and GCC >= 13. Add their definitions so we will be able to use them in later patches. Link: loongson/LoongArch-Documentation#57 Signed-off-by: Xi Ruoyao <[email protected]> Signed-off-by: Huacai Chen <[email protected]>

These relocation types are used by GNU binutils >= 2.40 and GCC >= 13. Add their definitions so we will be able to use them in later patches. Link: loongson/LoongArch-Documentation#57 Tested-by: WANG Xuerui <[email protected]> Signed-off-by: Xi Ruoyao <[email protected]> Signed-off-by: Huacai Chen <[email protected]>

LoongArch inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I5OHOB -------------------------------- These relocation types are used by GNU binutils >= 2.40 and GCC >= 13. Add their definitions so we will be able to use them in later patches. Link: loongson/LoongArch-Documentation#57 Signed-off-by: Xi Ruoyao <[email protected]> Signed-off-by: Huacai Chen <[email protected]>

xry111 reviewed Jul 20, 2022

View reviewed changes

xry111 suggested changes Jul 20, 2022

View reviewed changes

xry111 mentioned this pull request Jul 20, 2022

Add new relocation type support #50

Closed

cloudspurs force-pushed the new-reloc branch 3 times, most recently from 0056438 to 2800362 Compare July 21, 2022 08:14

cloudspurs force-pushed the new-reloc branch from 2800362 to f403cbd Compare July 21, 2022 08:50

xry111 reviewed Jul 21, 2022

View reviewed changes

Add new relocation type support

54cf2b1

cloudspurs force-pushed the new-reloc branch from f403cbd to 54cf2b1 Compare July 23, 2022 02:46

yetist reviewed Jul 23, 2022

View reviewed changes

ChenghuaXu approved these changes Jul 25, 2022

View reviewed changes

lzshhxx approved these changes Jul 25, 2022

View reviewed changes

chenglulu326 approved these changes Jul 25, 2022

View reviewed changes

lzshhxx approved these changes Jul 25, 2022

View reviewed changes

yetist approved these changes Jul 25, 2022

View reviewed changes

ChenghuaXu merged commit 363cc96 into loongson:main Jul 25, 2022

xry111 mentioned this pull request Jul 29, 2022

LoongArch: add new relocs introduced by LoongArch psABI v2 gimli-rs/object#450

Merged

xen0n mentioned this pull request Aug 3, 2022

cmd/link: need to handle new-style LoongArch relocs golang/go#54222

Closed

ChenghuaXu mentioned this pull request Aug 16, 2022

Replace R_LARCH_SOP_PUSH_* with regular bit extraction relocation types #9

Closed

wojiushixiaobai mentioned this pull request Aug 25, 2022

cmd/compile, cmd/link: c-shared not supported on linux/loong64 golang/go#53301

Closed

yetist added a commit to loongarch64-archive/grub that referenced this pull request Sep 9, 2022

LoongArch: Support new relocation types

a59f417

Link: loongson/LoongArch-Documentation#57 Signed-off-by: Xiaotian Wu <[email protected]>

yetist added a commit to loongarch64-archive/grub that referenced this pull request Sep 9, 2022

LoongArch: Support new relocation types in v2.00 ABI

cbc9d6b

Link: loongson/LoongArch-Documentation#57 Signed-off-by: Xiaotian Wu <[email protected]>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add new relocation type support #57

Add new relocation type support #57

cloudspurs commented Jul 17, 2022

xen0n commented Jul 17, 2022

xry111 Jul 20, 2022 •

edited

Loading

cloudspurs Jul 21, 2022

xry111 Jul 20, 2022

cloudspurs Jul 21, 2022

xry111 Jul 20, 2022

lzshhxx Jul 20, 2022

xry111 Jul 20, 2022

lzshhxx Jul 20, 2022

xen0n Jul 21, 2022

xry111 left a comment

xry111 Jul 20, 2022

xry111 Jul 20, 2022

lzshhxx Jul 21, 2022

cloudspurs Jul 21, 2022

xry111 commented Jul 20, 2022

ChenghuaXu commented Jul 20, 2022

xen0n commented Jul 21, 2022

xry111 Jul 21, 2022

lzshhxx Jul 21, 2022 •

edited

Loading

yetist Jul 23, 2022

ChenghuaXu Jul 25, 2022

ChenghuaXu left a comment

xen0n commented Jul 25, 2022

xry111 commented Jul 25, 2022

Eldeloro1 commented Aug 11, 2022

		\|[31 ... 12] bits of 32/64-bit PC-relative offset
		\|`+((uint32_t ) PC) [24 ... 5] = (((S+A) & ~0xfff) - (PC & ~0xfff)) [31 ... 12]+`

Add new relocation type support #57

Add new relocation type support #57

Conversation

cloudspurs commented Jul 17, 2022

xen0n commented Jul 17, 2022

xry111 Jul 20, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

xry111 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

xry111 commented Jul 20, 2022

ChenghuaXu commented Jul 20, 2022

xen0n commented Jul 21, 2022

Choose a reason for hiding this comment

lzshhxx Jul 21, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ChenghuaXu left a comment

Choose a reason for hiding this comment

xen0n commented Jul 25, 2022

xry111 commented Jul 25, 2022

Eldeloro1 commented Aug 11, 2022

xry111 Jul 20, 2022 •

edited

Loading

lzshhxx Jul 21, 2022 •

edited

Loading