Re: [Qemu-devel] [PATCH 2/4] target-lm32: add semihosting support

[Peter Maydell]

On 22 April 2014 19:18, Michael Walle <address@hidden> wrote:
> Intercept certain system calls if semihosting is enabled. This should
> behave like the GDB simulator.
> @@ -608,8 +609,20 @@ static void dec_scall(DisasContext *dc)
>          break;
>      case 7:
>          LOG_DIS("scall\n");
> -        tcg_gen_movi_tl(cpu_pc, dc->pc);
> -        t_gen_raise_exception(dc, EXCP_SYSTEMCALL);
> +        if (unlikely(semihosting_enabled)) {
> +            TCGv t0 = tcg_temp_new();
> +            int l1 = gen_new_label();
> +
> +            gen_helper_semihosting(t0, cpu_env);
> +            tcg_gen_brcondi_tl(TCG_COND_EQ, t0, 0, l1);
> +            tcg_gen_movi_tl(cpu_pc, dc->pc);
> +            t_gen_raise_exception(dc, EXCP_SYSTEMCALL);
> +            gen_set_label(l1);
> +            tcg_temp_free(t0);
> +        } else {
> +            tcg_gen_movi_tl(cpu_pc, dc->pc);
> +            t_gen_raise_exception(dc, EXCP_SYSTEMCALL);
> +        }
>          break;
>      default:
>          qemu_log_mask(LOG_GUEST_ERROR, "invalid opcode @0x%x", dc->pc);

This seems a bit odd to me -- if semihosting is enabled,
we generate a call to a helper. If that helper returns
failure (eg attempt to open nonexistent file) we raise
an exception. Why not just do what the ARM semihosting
does and put the call to the semihosting handling code
in the lm32_cpu_do_interrupt() function?

(It's a bit sad that the semihosting API for lm32
doesn't seem to provide any way of distinguishing
"syscall for semihosting" from "other syscall",
incidentally.)

thanks
-- PMM