Re: [PATCH v4 2/5] LoongArch: Use la.pcrel instead of la.abs for exception handlers

[Youling Tang]

On 02/16/2023 07:18 PM, Youling Tang wrote:
> On 02/16/2023 04:03 PM, Youling Tang wrote:
> > On 02/16/2023 03:10 PM, Xi Ruoyao wrote:
> > > On Thu, 2023-02-16 at 14:59 +0800, Jinyang He wrote:
> > > > +.macro la.abs reg, sym
> > > > +766:
> > > > +    nop
> > > > +    nop
> > > > +    nop
> > > > +    nop
> > >
> > > In the "formal" version we can code
> > >
> > > lu12i.w        reg, 0
> > > ori        reg, reg, 0
> > > lu32i.d        reg, 0
> > > lu52i.d        reg, reg, 0
> > >
> > > here.  Then we only need to fixup the immediate slot so we can avoid
> > > using parse_r.
> > >
> > >
> > > > +    .pushsection ".laabs", "aw", %progbits
> > > > +768:
> > > > +    .word 768b-766b
> > > > +    parse_r regno, \reg
> > > > +    .word regno
> > > > +    .dword \sym
> > > > +    .popsection
> > > > +.endm
> >
> > I will try to modify a version for testing, using the following
> > definition, when the RELOCATABLE is turned on, the "la.abs macro" is
> > used, otherwise the "la.abs pseudo instruction" is still used as before.
> >
> > #ifdef CONFIG_RELOCATABLE
> > .macro la.abs reg, sym
> > lu12i.w        reg, 0
> > ori        reg, reg, 0
> > lu32i.d        reg, 0
> > lu52i.d        reg, reg, 0
> > .endm
> > #endif
>
> On the basis of the v4 patch set, remove patch2, and then add the
> following patches, and the test is successful on qemu.
>
> If this method is more acceptable to everyone, I will send v5.
>
> diff --git a/arch/loongarch/include/asm/asmmacro.h
> b/arch/loongarch/include/asm/asmmacro.h
> index 328bb956f241..adb04ae6b208 100644
> --- a/arch/loongarch/include/asm/asmmacro.h
> +++ b/arch/loongarch/include/asm/asmmacro.h
> @@ -667,4 +667,19 @@
>         nor     \dst, \src, zero
>  .endm
>
> +#ifdef CONFIG_RELOCATABLE
> +.macro la.abs reg, sym
> +766:
> +       lu12i.w \reg, 0
> +       ori     \reg, \reg, 0
> +       lu32i.d \reg, 0
> +       lu52i.d \reg, \reg, 0
> +       .pushsection ".laabs", "aw", %progbits
> +768:
> +       .dword 768b-766b
> +       .dword \sym
> +       .popsection
> +.endm
> +#endif
> +
>  #endif /* _ASM_ASMMACRO_H */
> diff --git a/arch/loongarch/kernel/relocate.c
> b/arch/loongarch/kernel/relocate.c
> index 7d19cc0d2185..7ad327a554f9 100644
> --- a/arch/loongarch/kernel/relocate.c
> +++ b/arch/loongarch/kernel/relocate.c
> @@ -12,6 +12,7 @@
>  #include <linux/start_kernel.h>
>  #include <asm/bootinfo.h>
>  #include <asm/early_ioremap.h>
> +#include <asm/inst.h>
>  #include <asm/sections.h>
>
>  #define RELOCATED(x) ((void *)((long)x + reloc_offset))
> @@ -45,6 +46,32 @@ static inline __init void relocate_relative(void)
>         }
>  }
>
> +static inline void __init relocate_laabs(long offset)
> +{
> +       extern void *__laabs_begin;
> +       extern void *__laabs_end;
> +       struct laabs {
> +               long offset;
> +               long symvalue;
> +       } *p;
> +
> +       for (p = (void *)&__laabs_begin; (void *)p < (void
> *)&__laabs_end; p++) {
> +               long v = p->symvalue + reloc_offset;
> +               union loongarch_instruction *insn = (void *)p -
> p->offset + offset;
> +               u32 lu12iw, ori, lu32id, lu52id;
> +
> +               lu12iw = (v >> 12) & 0xfffff;
> +               ori = v & 0xfff;
> +               lu32id = (v >> 32) & 0xfffff;
> +               lu52id = v >> 52;
> +
> +               insn[0].reg1i20_format.immediate = lu12iw;
> +               insn[1].reg2i12_format.immediate = ori;
> +               insn[2].reg1i20_format.immediate = lu32id;
> +               insn[3].reg2i12_format.immediate = lu52id;
> +    }
> +}
> +
>  #ifdef CONFIG_RANDOMIZE_BASE
>  static inline __init unsigned long rotate_xor(unsigned long hash,
>                                               const void *area, size_t
> size)
> @@ -168,8 +195,10 @@ void *__init do_kaslr(void)
>                 update_reloc_offset(&reloc_offset, offset);
>         }
>
> -       if (reloc_offset)
> +       if (reloc_offset) {
>                 relocate_relative();
> +               relocate_laabs(offset);
> +       }

Self review:

        if (reloc_offset)
                relocate_relative();

        relocate_laabs(offset);

>         return kernel_entry;
>  }
> @@ -181,6 +210,8 @@ void __init relocate_kernel(void)
>
>         if (reloc_offset)
>                 relocate_relative();
> +
> +       relocate_laabs(0);
>  }
>
>  /*
> diff --git a/arch/loongarch/kernel/vmlinux.lds.S
> b/arch/loongarch/kernel/vmlinux.lds.S
> index aec0b6567d24..0e58c68bf427 100644
> --- a/arch/loongarch/kernel/vmlinux.lds.S
> +++ b/arch/loongarch/kernel/vmlinux.lds.S
> @@ -66,6 +66,13 @@ SECTIONS
>                 __alt_instructions_end = .;
>         }
>
> +       . = ALIGN(8);
> +       .laabs : AT(ADDR(.laabs) - LOAD_OFFSET) {
> +               __laabs_begin = .;
> +               *(.laabs)
> +               __laabs_end = .;
> +       }
> +
>         .got : ALIGN(16) { *(.got) }
>
>
> Youling.