Re: [PATCH 3/4] tcg/aarch64: Emit BTI insns at jump landing pads

[Peter Maydell]

On Wed, 16 Aug 2023 at 15:27, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> The prologue is entered via "call"; the epilogue, each tb,
> and each goto_tb continuation point are all reached via "jump".
>
> As tcg_out_goto_long is only used by tcg_out_exit_tb, merge
> the two functions.  Change the indirect register used to
> TCG_REG_TMP1, aka X16, so that the BTI condition created
> is "jump" instead of "jump or call".

TCG_REG_TMP1 is X17, not X16...

> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

>  static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
>  {
> +    const tcg_insn_unit *target;
> +    ptrdiff_t offset;
> +
>      /* Reuse the zeroing that exists for goto_ptr.  */
>      if (a0 == 0) {
> -        tcg_out_goto_long(s, tcg_code_gen_epilogue);
> +        target = tcg_code_gen_epilogue;
>      } else {
>          tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_X0, a0);
> -        tcg_out_goto_long(s, tb_ret_addr);
> +        target = tb_ret_addr;
> +    }
> +
> +    offset = tcg_pcrel_diff(s, target) >> 2;
> +    if (offset == sextract64(offset, 0, 26)) {
> +        tcg_out_insn(s, 3206, B, offset);
> +    } else {
> +        tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_TMP1, (intptr_t)target);
> +        tcg_out_insn(s, 3207, BR, TCG_REG_TMP1);

Since it's now important that the tempreg we have here is
one of X16/X17 in order to get the right BTI behaviour,
I think a build-time assert of this would be helpful.
That will catch the possibility that we forget about this
and decide to rearrange which registers we use as
tempregs later.

>      }
>  }
>
> @@ -1970,6 +1984,7 @@ static void tcg_out_goto_tb(TCGContext *s, int which)
>      tcg_out32(s, I3206_B);
>      tcg_out_insn(s, 3207, BR, TCG_REG_TMP0);
>      set_jmp_reset_offset(s, which);
> +    tcg_out_bti(s, BTI_J);
>  }

Otherwise
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>

"Did we miss anywhere that should have had a BTI insn"
is a bit tricky to review, but I assume you've done enough
testing on a BTI-enabled host to catch that.

thanks
-- PMM