[PATCH v5 0/5] Add support for TLS Descriptors (TLSDESC)

Fangrui Song i@maskray.me
Sat Jan 27 19:18:17 GMT 2024


On Fri, Jan 26, 2024 at 11:26 PM Lulu Cai <cailulu@loongson.cn> wrote:
>
> On 1/27/24 10:54 AM, Fangrui Song wrote:
> > On Fri, Dec 22, 2023 at 3:42 AM Lulu Cai <cailulu@loongson.cn> wrote:
> >> v5 mainly changes the relocation name and adds relocation
> >> for immediate value field of ld.w/d.
> >>
> >> 1. R_LARCH_TLS_DESC_ADD_PC_LO12 -> R_LARCH_TLS_DESC_PC_LO12.
> >>     R_LARCH_TLS_DESC_LD(113)     -> R_LARCH_TLS_DESC_LD(119)
> >>
> >> 2. Added 32 abs test case.
> >>
> >> 3. Added R_LARCH_TLS_DESC_LD relocation to the immediate field of
> >>     the ld.w/d instruction.
> >>
> >>     Such as:
> >>     pcalau12i $r4,%desc_pc_hi20(var)
> >>     addi.d $r4,$r4,%desc_pc_lo12(var)
> >>     ld.d $r1,$r4,0                    -> ld.d $r1,$r4,%desc_ld(var)
> >>     jirl $r1,$r1,%desc_call(var)
> >>     addi.d $a0,$a0,$tp
> >>
> >> Lulu Cai (4):
> >>    LoongArch: Add new relocs and macro for TLSDESC.
> >>    LoongArch: Add support for TLSDESC in ld.
> >>    LoongArch: Add tls transition support.
> >>    LoongArch: Add testsuit for DESC and tls transition and tls
> >>      relaxation.
> >>
> >> mengqinggang (1):
> >>    LoongArch: Add support for TLS LD/GD/DESC relaxation
> >>
> >>   bfd/bfd-in2.h                                 |  15 +
> >>   bfd/elfnn-loongarch.c                         | 552 +++++++++++++++++-
> >>   bfd/elfxx-loongarch.c                         | 270 ++++++++-
> >>   bfd/libbfd.h                                  |  15 +
> >>   bfd/reloc.c                                   |  36 ++
> >>   gas/config/tc-loongarch.c                     |  20 +-
> >>   gas/testsuite/gas/loongarch/macro_op.d        | 128 ++--
> >>   gas/testsuite/gas/loongarch/macro_op_32.d     | 120 ++--
> >>   .../gas/loongarch/macro_op_large_abs.d        | 160 ++---
> >>   .../gas/loongarch/macro_op_large_pc.d         | 160 ++---
> >>   gas/testsuite/gas/loongarch/tlsdesc_32.d      |  28 +
> >>   gas/testsuite/gas/loongarch/tlsdesc_32.s      |  12 +
> >>   gas/testsuite/gas/loongarch/tlsdesc_32_abs.d  |  26 +
> >>   gas/testsuite/gas/loongarch/tlsdesc_32_abs.s  |   8 +
> >>   gas/testsuite/gas/loongarch/tlsdesc_64.d      |  28 +
> >>   gas/testsuite/gas/loongarch/tlsdesc_64.s      |  12 +
> >>   .../gas/loongarch/tlsdesc_large_abs.d         |  34 ++
> >>   .../gas/loongarch/tlsdesc_large_abs.s         |  12 +
> >>   .../gas/loongarch/tlsdesc_large_pc.d          |  38 ++
> >>   .../gas/loongarch/tlsdesc_large_pc.s          |  17 +
> >>   include/elf/loongarch.h                       |  26 +-
> >>   include/opcode/loongarch.h                    |   6 +
> >>   ld/testsuite/ld-loongarch-elf/desc-ie.d       |  16 +
> >>   ld/testsuite/ld-loongarch-elf/desc-ie.s       |  18 +
> >>   ld/testsuite/ld-loongarch-elf/desc-le.d       |  15 +
> >>   ld/testsuite/ld-loongarch-elf/desc-le.s       |  14 +
> >>   ld/testsuite/ld-loongarch-elf/desc-norelax.d  |  16 +
> >>   ld/testsuite/ld-loongarch-elf/desc-norelax.s  |   5 +
> >>   ld/testsuite/ld-loongarch-elf/desc-relax.d    |  15 +
> >>   ld/testsuite/ld-loongarch-elf/desc-relax.s    |   5 +
> >>   ld/testsuite/ld-loongarch-elf/ie-le.d         |  13 +
> >>   ld/testsuite/ld-loongarch-elf/ie-le.s         |  11 +
> >>   .../ld-loongarch-elf/ld-loongarch-elf.exp     |   9 +
> >>   ld/testsuite/ld-loongarch-elf/macro_op.d      | 391 +++++++------
> >>   ld/testsuite/ld-loongarch-elf/macro_op_32.d   | 120 ++--
> >>   ld/testsuite/ld-loongarch-elf/tlsdesc-dso.d   |  56 ++
> >>   ld/testsuite/ld-loongarch-elf/tlsdesc-dso.s   |  65 +++
> >>   opcodes/loongarch-opc.c                       |  54 ++
> >>   38 files changed, 1980 insertions(+), 566 deletions(-)
> >>   create mode 100644 gas/testsuite/gas/loongarch/tlsdesc_32.d
> >>   create mode 100644 gas/testsuite/gas/loongarch/tlsdesc_32.s
> >>   create mode 100644 gas/testsuite/gas/loongarch/tlsdesc_32_abs.d
> >>   create mode 100644 gas/testsuite/gas/loongarch/tlsdesc_32_abs.s
> >>   create mode 100644 gas/testsuite/gas/loongarch/tlsdesc_64.d
> >>   create mode 100644 gas/testsuite/gas/loongarch/tlsdesc_64.s
> >>   create mode 100644 gas/testsuite/gas/loongarch/tlsdesc_large_abs.d
> >>   create mode 100644 gas/testsuite/gas/loongarch/tlsdesc_large_abs.s
> >>   create mode 100644 gas/testsuite/gas/loongarch/tlsdesc_large_pc.d
> >>   create mode 100644 gas/testsuite/gas/loongarch/tlsdesc_large_pc.s
> >>   create mode 100644 ld/testsuite/ld-loongarch-elf/desc-ie.d
> >>   create mode 100644 ld/testsuite/ld-loongarch-elf/desc-ie.s
> >>   create mode 100644 ld/testsuite/ld-loongarch-elf/desc-le.d
> >>   create mode 100644 ld/testsuite/ld-loongarch-elf/desc-le.s
> >>   create mode 100644 ld/testsuite/ld-loongarch-elf/desc-norelax.d
> >>   create mode 100644 ld/testsuite/ld-loongarch-elf/desc-norelax.s
> >>   create mode 100644 ld/testsuite/ld-loongarch-elf/desc-relax.d
> >>   create mode 100644 ld/testsuite/ld-loongarch-elf/desc-relax.s
> >>   create mode 100644 ld/testsuite/ld-loongarch-elf/ie-le.d
> >>   create mode 100644 ld/testsuite/ld-loongarch-elf/ie-le.s
> >>   create mode 100644 ld/testsuite/ld-loongarch-elf/tlsdesc-dso.d
> >>   create mode 100644 ld/testsuite/ld-loongarch-elf/tlsdesc-dso.s
> >>
> >> --
> >> 2.43.0
> >>
> > Technically you could define R_LARCH_TLS_DESC to be shared by 32-bit
> > and 64-bit arches, like RELATIVE/IRELATIVE/JUMP_SLOT.
> > That said, I see that the change has been made.
>
>
> Yes that's totally fine. The main reason here is because other TLS types of
>
> dynamic relocation have 32-bit and 64-bit, so DESC is distinguished.
>
>
> > It seems that la.tls.desc expands to 4 instructions with 2
> > R_LARCH_RELAX relocations?
> > I think one R_LARCH_RELAX is sufficient to implement relaxation.
> >
> > RISC-V has a similar problem and I have suggested
> > https://github.com/riscv-non-isa/riscv-elf-psabi-doc/pull/421
>
>
> This is done because the current psABI stipulates that the instruction
>
> to be deleted must be followed by an R_LARCH_RELAX relocation.
>
> In the future, we will try to implement la.tls.desc with four instructions
>
> using only one R_LARCH_RELAX implementation.

Thanks for the answer. I am looking forward to one fewer R_LARCH_RELAX.


More information about the Binutils mailing list