Anthony Liguori wrote:
> Jan Kiszka wrote:
>> Andrzej Zaborowski wrote:
>>  
>>> Revision: 6684
>>>           http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=6684
>>> Author:   balrog
>>> Date:     2009-03-04 21:00:07 +0000 (Wed, 04 Mar 2009)
>>> Log Message:
>>> -----------
>>> Fix "info registers" under kvm.
>>>
>>> Modified Paths:
>>> --------------
>>>     trunk/target-i386/helper.c
>>>
>>> Modified: trunk/target-i386/helper.c
>>> ===================================================================
>>> --- trunk/target-i386/helper.c    2009-03-04 19:25:22 UTC (rev 6683)
>>> +++ trunk/target-i386/helper.c    2009-03-04 21:00:07 UTC (rev 6684)
>>> @@ -578,6 +578,9 @@
>>>      char cc_op_name[32];
>>>      static const char *seg_name[6] = { "ES", "CS", "SS", "DS", "FS",
>>> "GS" };
>>>  
>>> +    if (kvm_enabled())
>>> +        kvm_arch_get_registers(env);
>>> +
>>>      eflags = env->eflags;
>>>  #ifdef TARGET_X86_64
>>>      if (env->hflags & HF_CS64_MASK) {
>>>
>>>     
>>
>> On the one hand, this patch also takes care of sync'ing with KVM in case
>> of cpu_dump_state on fatal exists. On the other hand, it only solves one
>> part of monitor issue. See [1] for a more complete sync.
>>
>> I'm just still waiting for a reply from Anthony on how to embed best all
>> the "if (kvm_enabled()) foo();" patterns [2]. That would also allow us
>> to merge gdbstub support for upstream kvm.
>>   
> 
> I really don't have a great suggestion.  I've been hoping we could come
> up with something better than if (kvm_enabled()) foo().  If we can't, we
> can't.

We could do

static inline void kvm_get_registers(CPUState *env)
{
	if (kvm_enabled())
		kvm_arch_get_registers(env);
}

in kvm.h (and kvm_put_registers analogously). That would still be
kvm-specific, though, but it would reduce the impact on the code that
has to be extended this way.

Generic hooks is something I would really like to postpone here until we
start working on some accelerator abstraction.

Jan