Re: [PATCH] RISC-V: Initialize the jvt CSR

On Mon, Sep 25, 2023 at 1:37 PM Hau Hsu via Newlib
<[email protected]> wrote:
>
> We set symbol '__jvt_base$' as weak. So if the symbol is not set in the
> linker script, the address would be 0. We initialize jvt CSR only if
> the address is not 0.
>
> psabi reference:
> https://github.com/riscv-non-isa/riscv-elf-psabi-doc/blob/2d770815dc9a8b11e61ea1abd487cb25ee56ad5e/riscv-elf.adoc#table-jump-relaxation
> ---
>  libgloss/riscv/crt0.S | 10 ++++++++++
>  1 file changed, 10 insertions(+)
>
> diff --git a/libgloss/riscv/crt0.S b/libgloss/riscv/crt0.S
> index 7a4e4e32c..4744ba039 100644
> --- a/libgloss/riscv/crt0.S
> +++ b/libgloss/riscv/crt0.S
> @@ -26,6 +26,16 @@ _start:
>    addi  gp, gp, %pcrel_lo(1b)
>  .option pop
>
> +  /* Initialize jvt CSR */
> +  .weak __jvt_base$
> +.option push
> +.option arch, +zcmt, +zicsr

I guess I am little concerned about the backward compatibility, zcmt
isn't supported by trunk binutils even, so maybe we should just +zicsr
to make sure csrw is usable, and use csr number directly instead of
using symbolic name to prevent the backward incompatible issue.

> +  lla a0, __jvt_base$
> +  beqz a0, .Ljvt_init_end
> +  csrw jvt, a0
> +.Ljvt_init_end:
> +.option pop
> +
>    # Clear the bss segment
>    la      a0, __bss_start
>    la      a2, _end
> --
> 2.42.0
>