On Thu, 29 Jul 2021 at 02:09, Richard Henderson
<richard.hender...@linaro.org> wrote:
>
> Use the newly exposed do_unaligned_access hook from atomic_mmu_lookup,
> which has access to complete alignment info from the TCGMemOpIdx arg.
>
> Signed-off-by: Richard Henderson <richard.hender...@linaro.org>
> ---
> accel/tcg/user-exec.c | 23 ++++++++++++++++++++++-
> 1 file changed, 22 insertions(+), 1 deletion(-)
>
> diff --git a/accel/tcg/user-exec.c b/accel/tcg/user-exec.c
> index 90d1a2d327..dd77e90789 100644
> --- a/accel/tcg/user-exec.c
> +++ b/accel/tcg/user-exec.c
> @@ -852,6 +852,16 @@ int cpu_signal_handler(int host_signum, void *pinfo,
>
> /* The softmmu versions of these helpers are in cputlb.c. */
>
> +static void cpu_unaligned_access(CPUState *cpu, vaddr addr,
> + MMUAccessType access_type,
> + int mmu_idx, uintptr_t ra)
> +{
> + CPUClass *cc = CPU_GET_CLASS(cpu);
> +
> + cc->tcg_ops->do_unaligned_access(cpu, addr, access_type, mmu_idx, ra);
> + g_assert_not_reached();
> +}
The softmmu version doesn't g_assert_not_reached(), I think
perhaps with the intent that a CPU implementation could
in some cases return without raising an exception to
mean "continue with the unaligned access". We should decide
whether we want the API to permit that, or else consistently
have both softmmu and useronly versions be marked noreturn
and with an assert, and we should document whichever we choose.
-- PMM
