Re: [Qemu-devel] [PATCH 6/7] monitor: "i": Add ARM specifics

[Peter Crosthwaite]

On Tue, May 5, 2015 at 10:43 AM, Richard Henderson <address@hidden> wrote:
> On 05/05/2015 10:19 AM, Peter Maydell wrote:
>> On 5 May 2015 at 05:45, Peter Crosthwaite <address@hidden> wrote:
>>> Add the ARM specific disassembly flags setup, so ARM can be correctly
>>> disassembled from the monitor.
>>>
>>> Signed-off-by: Peter Crosthwaite <address@hidden>
>>> ---
>>>  monitor.c | 11 +++++++++++
>>>  1 file changed, 11 insertions(+)
>>>
>>> diff --git a/monitor.c b/monitor.c
>>> index d831d98..9d9f1e2 100644
>>> --- a/monitor.c
>>> +++ b/monitor.c
>>> @@ -1217,6 +1217,17 @@ static void memory_dump(Monitor *mon, int count, int 
>>> format, int wsize,
>>>          int flags;
>>>          flags = 0;
>>>          env = mon_get_cpu();
>>> +#ifdef TARGET_ARM
>>> +        if (env->thumb) {
>>> +            flags |= 1;
>>> +        }
>>> +        if (env->bswap_code) {
>>> +            flags |= 2;
>>> +        }
>>> +        if (env->aarch64) {
>>> +            flags |= 4;
>>> +        }
>>> +#endif
>>
>> monitor.c has no business poking around in the CPU state
>> internals like this... You probably want a CPU method
>> get_disas_flags() or something.
>>
>> -- PMM
>>
>
> While this patch set does improve the current dismal state of affairs, I think
> the ideal solution is a cpu method that takes care of all the disassembly info
> setup.
>
> Indeed, the flags setup becomes less obscure when, instead of
>
> #ifdef TARGET_I386
>         if (wsize == 2) {
>             flags = 1;
>         } else if (wsize == 4) {
>             flags = 0;
>         } else {
>             /* as default we use the current CS size */
>             flags = 0;
>             if (env) {
> #ifdef TARGET_X86_64
>                 if ((env->efer & MSR_EFER_LMA) &&
>                     (env->segs[R_CS].flags & DESC_L_MASK))
>                     flags = 2;
>                 else
> #endif
>                 if (!(env->segs[R_CS].flags & DESC_B_MASK))
>                     flags = 1;
>             }
>         }
>
> in one place and
>
> #if defined(TARGET_I386)
>     if (flags == 2) {
>         s.info.mach = bfd_mach_x86_64;
>     } else if (flags == 1) {
>         s.info.mach = bfd_mach_i386_i8086;
>     } else {
>         s.info.mach = bfd_mach_i386_i386;
>     }
>     print_insn = print_insn_i386;
>
> in another, we merge the two so that we get
>
>     s.info.mach = bfd_mach_i386_i8086;
>     if (env->hflags & (1U << HF_CS32_SHIFT)) {
>         s.info.mach = bfd_mach_i386_i386;
>     }
> #ifdef TARGET_X86_64
>     if (env->hflags & (1U << HF_CS64_SHIFT)) {
>         s.info.mach = bfd_mach_x86_64;
>     }
> #endif
>
>
> I'm not sure what the right interface for this is, exactly.  But I'd think 
> that
> the CPUDebug structure would be initialized in the caller -- target or monitor
> -- while the mach and whatnot get filled in by the cpu hook.  Maybe even have
> the cpu hook return the print_insn function to use.
>

So something else I just thought of and started worrying about, is the
the target_disas path is driving some of the flags from the disas
context, whereas a hook will only have access to the CPUState/env. Can
we rely on the env/CPUState always being up to date during
target_disas (which happens at translate time?) or will we need to go
field by field to make sure any env updates explicitly occur before
target_disas?

Regards,
Peter

>
>
> r~
>