Re: [Qemu-devel] [PATCH 2/7] target/moxie: Fix tlb_fill

[Peter Maydell]

On 15 December 2017 at 17:07, Richard Henderson
<address@hidden> wrote:
> We should not exit unless moxie_cpu_handle_mmu_fault has failed.
>
> Signed-off-by: Richard Henderson <address@hidden>
> ---
>  target/moxie/helper.c | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/target/moxie/helper.c b/target/moxie/helper.c
> index 2ecee89f11..6890ffd71c 100644
> --- a/target/moxie/helper.c
> +++ b/target/moxie/helper.c
> @@ -36,9 +36,8 @@ void tlb_fill(CPUState *cs, target_ulong addr, 
> MMUAccessType access_type,
>
>      ret = moxie_cpu_handle_mmu_fault(cs, addr, access_type, mmu_idx);
>      if (unlikely(ret)) {
> -        cpu_restore_state(cs, retaddr);
> +        cpu_loop_exit_restore(cs, retaddr);
>      }
> -    cpu_loop_exit(cs);
>  }
>
>  void helper_raise_exception(CPUMoxieState *env, int ex)
> --

Reviewed-by: Peter Maydell <address@hidden>

I suspect this codepath has never been tested. There's another
bug in moxie_cpu_handle_mmu_fault() where it will call
tlb_set_page() even in the "mmu lookup missed, we're
going to throw an exception" codepath.

Is moxie even worth bothering to keep in QEMU? As far as I can tell:
 * the only changes to target/moxie since the target was added in
   2013 have been generic cross-tree changes and cleanups and
   minor bugs found by static analysis etc
 * the last commit signed-off-by the moxie maintainer was
   in March 2013
 * last email to qemu-devel by the maintainer was December 2013
 * as far as I can tell nobody's ever reported a bug to us,
   which is suggestive that it has no users

This target is costing us maintenance effort -- is it actually
useful to anybody ?

thanks
-- PMM