Re: [PATCH for-8.0 v2 12/13] target/s390x: Use tcg_gen_atomic_cmpxchg_i128 for CDSG

[Ilya Leoshkevich]

On Fri, Nov 11, 2022 at 06:08:19PM +1000, Richard Henderson wrote:
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  target/s390x/helper.h            |  2 --
>  target/s390x/tcg/mem_helper.c    | 52 ---------------------------
>  target/s390x/tcg/translate.c     | 60 ++++++++++++++++++++------------
>  target/s390x/tcg/insn-data.h.inc |  2 +-
>  4 files changed, 38 insertions(+), 78 deletions(-)

Acked-by: Ilya Leoshkevich <iii@linux.ibm.com>

I was wondering what assembly this would generate in parallel mode and
wrote a small test. On my x86_64 machine it ended up being
helper_atomic_cmpxchgo_be() -> cpu_atomic_cmpxchgo_be_mmu() ->
lock cmpxchg16b, nothing surprising.

On an s390x host we fall back to cpu_exec_step_atomic(), because in the
configure test:

      int main(void)
      {
        unsigned __int128 x = 0, y = 0;
        __sync_val_compare_and_swap_16(&x, y, x);
        return 0;
      }

x and y are not aligned. I guess that's working as intended as well,
even though it would be nice to eventually make use of cdsg there.

I will post the test shortly.