Re: [PATCH v5 10/22] target/arm: Define arm_cpu_do_unaligned_access for CONFIG_USER_ONLY

[Peter Maydell]

On Fri, 11 Oct 2019 at 14:50, Richard Henderson
<address@hidden> wrote:
>
> We will need this to raise unaligned exceptions from user mode.
>
> Signed-off-by: Richard Henderson <address@hidden>
> ---
>  target/arm/tlb_helper.c | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/target/arm/tlb_helper.c b/target/arm/tlb_helper.c
> index 5feb312941..29b92a1149 100644
> --- a/target/arm/tlb_helper.c
> +++ b/target/arm/tlb_helper.c
> @@ -10,8 +10,6 @@
>  #include "internals.h"
>  #include "exec/exec-all.h"
>
> -#if !defined(CONFIG_USER_ONLY)
> -
>  static inline uint32_t merge_syn_data_abort(uint32_t template_syn,
>                                              unsigned int target_el,
>                                              bool same_el, bool ea,
> @@ -122,6 +120,7 @@ void arm_cpu_do_unaligned_access(CPUState *cs, vaddr 
> vaddr,
>      arm_deliver_fault(cpu, vaddr, access_type, mmu_idx, &fi);
>  }
>
> +#ifndef CONFIG_USER_ONLY
>  /*
>   * arm_cpu_do_transaction_failed: handle a memory system error response
>   * (eg "no device/memory present at address") by raising an external abort
> --
> 2.17.1

Isn't this just enabling compilation of the softmmu
arm_cpu_do_unaligned_access() on linux-user? That codepath
does a lot of softmmu stuff including calling arm_deliver_fault()
(which implicitly does somewhat more looking at EL1 system
register state than we necessarily have set up correctly
for the user-mode code).

For arm_cpu_tlb_fill() which handles prefetch/data aborts
we just have a separate much simpler codepath for
CONFIG_USER_ONLY which doesn't call arm_deliver_fault().
I think being consistent here about how we handle the
CONFIG_USER_ONLY case would help avoid having a codepath
that isn't very well tested because it's only used in
the odd special case of unaligned-address exceptions.

thanks
-- PMM