Re: [PATCH v2 07/55] target/mips: Implement do_unaligned_access for user-only

[Peter Maydell]

On Tue, 3 Aug 2021 at 05:18, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> Cc: Philippe Mathieu-Daudé <f4bug@amsat.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  linux-user/mips/cpu_loop.c        | 20 ++++++++++++++++----
>  target/mips/cpu.c                 |  2 +-
>  target/mips/tcg/op_helper.c       |  3 +--
>  target/mips/tcg/user/tlb_helper.c | 23 +++++++++++------------
>  4 files changed, 29 insertions(+), 19 deletions(-)
>
> diff --git a/linux-user/mips/cpu_loop.c b/linux-user/mips/cpu_loop.c
> index 9d813ece4e..51f4eb65a6 100644
> --- a/linux-user/mips/cpu_loop.c
> +++ b/linux-user/mips/cpu_loop.c
> @@ -158,12 +158,24 @@ done_syscall:
>              break;
>          case EXCP_TLBL:
>          case EXCP_TLBS:
> -        case EXCP_AdEL:
> -        case EXCP_AdES:
>              info.si_signo = TARGET_SIGSEGV;
>              info.si_errno = 0;
> -            /* XXX: check env->error_code */
> -            info.si_code = TARGET_SEGV_MAPERR;
> +            info.si_code = (env->error_code & EXCP_TLB_NOMATCH
> +                            ? TARGET_SEGV_MAPERR : TARGET_SEGV_ACCERR);
> +            info._sifields._sigfault._addr = env->CP0_BadVAddr;
> +            queue_signal(env, info.si_signo, QEMU_SI_FAULT, &info);
> +            break;
> +        case EXCP_AdEL:
> +        case EXCP_AdES:
> +            /*
> +             * Note that on real hw AdE is also raised for access to a
> +             * kernel address from user mode instead of a TLB error.
> +             * For simplicity, we do not distinguish this in the user
> +             * version of mips_cpu_tlb_fill so only unaligned comes here.
> +             */
> +            info.si_signo = TARGET_SIGBUS;
> +            info.si_errno = 0;
> +            info.si_code = TARGET_BUS_ADRALN;

The MIPS kernel code doesn't use si_code BUS_ADRALN for alignment
errors, as far as I can see -- it just calls force_sig(SIGBUS):
https://elixir.bootlin.com/linux/latest/source/arch/mips/kernel/unaligned.c#L557

which gets you the same thing our force_sig() does: a code
of SI_KERNEL. So I think we should be calling
   force_sig(TARGET_SIGBUS);
here rather than filling in a target_siginfo_t and calling
queue_signal().

>              info._sifields._sigfault._addr = env->CP0_BadVAddr;
>              queue_signal(env, info.si_signo, QEMU_SI_FAULT, &info);
>              break;

thanks
-- PMM