Re: [PATCH for-6.2 14/43] target/sparc: Set fault address in sparc_cpu_do_unaligned_access

[Peter Maydell]

On Thu, 29 Jul 2021 at 02:01, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> We ought to have been recording the virtual address for reporting
> to the guest trap handler.  Mirror the SFSR FIXME from the sparc64
> version of get_physical_address_data.
>
> Cc: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  target/sparc/ldst_helper.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
>
> diff --git a/target/sparc/ldst_helper.c b/target/sparc/ldst_helper.c
> index 974afea041..7367b48c8b 100644
> --- a/target/sparc/ldst_helper.c
> +++ b/target/sparc/ldst_helper.c
> @@ -1963,6 +1963,14 @@ void QEMU_NORETURN 
> sparc_cpu_do_unaligned_access(CPUState *cs, vaddr addr,
>      SPARCCPU *cpu = SPARC_CPU(cs);
>      CPUSPARCState *env = &cpu->env;
>
> +#ifdef TARGET_SPARC64
> +    /* FIXME: ASI field in SFSR must be set */
> +    env->dmmu.sfsr = SFSR_VALID_BIT; /* Fault status register */
> +    env->dmmu.sfar = addr;           /* Fault address register */
> +#else
> +    env->mmuregs[4] = addr;
> +#endif
> +
>      cpu_raise_exception_ra(env, TT_UNALIGNED, retaddr);
>  }
>  #endif

The architecture manual seems to be gratuitously opaque about
whether and where the fault address for an alignment fault gets
recorded, but Linux at least for 64-bit seems to pull it out of the
sfar, so I guess that's right.

We probably ought to have the "if SFSR_VALID_BIT already set,
set the OW bit". MMUAccessType and mmu_idx give us enough to
set the CT bits and the WRITE bit in the same way we do at
the start of get_physical_address_data().

-- PMM