Re: [PATCH 1/4] linux-user/ppc: Fix XER access in save/restore_user_regs

[Richard Henderson]

On 10/14/21 3:32 PM, matheus.ferst@eldorado.org.br wrote:
> From: Matheus Ferst <matheus.ferst@eldorado.org.br>
>
> We should use cpu_read_xer/cpu_write_xer to save/restore the complete
> register since some of its bits are in other fields of CPUPPCState. A
> test is added to prevent future regressions.
>
> Fixes: da91a00f191f ("target-ppc: Split out SO, OV, CA fields from XER")
> Signed-off-by: Matheus Ferst <matheus.ferst@eldorado.org.br>
> ---
>   linux-user/ppc/signal.c                     |  9 +++--
>   tests/tcg/ppc64/Makefile.target             |  2 +
>   tests/tcg/ppc64le/Makefile.target           |  2 +
>   tests/tcg/ppc64le/signal_save_restore_xer.c | 42 +++++++++++++++++++++
>   4 files changed, 52 insertions(+), 3 deletions(-)
>   create mode 100644 tests/tcg/ppc64le/signal_save_restore_xer.c

The code is good so,
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>

> +    sigaction(SIGILL, &sa, NULL);
> +
> +    asm("mtspr 1, %1\n\t"
> +        ".long 0x0\n\t"

While Appendix B does guarantee that "0" is and always will be an invalid instruction, I 
wonder if the test itself would be clearer (i.e. self-documenting the intent) using 
SIGTRAP and "trap".


r~