Re: [Qemu-devel] Re: [6684] Fix "info registers" under kvm.

[Avi Kivity]

Anthony Liguori wrote:
> In the case of save/load registers, I'd prefer wrapper functions like:
>
> cpu_state_update(CPUState *env, int is_dirty)
>
> That could be hooked by something like Xen.  The implementation would be:
>
> static void cpu_state_update(CPUState *env, int is_dirty)
> {
>    if (kvm_enabled()) {
>        if (is_dirty)
>           kvm_arch_save_registers(env);
>        else
>           kvm_arch_load_registers(env);
>    }
> } 

That reduces readability IMO (boolean parameters do that, as well as a 
function where the direction of data movement isn't clear).

But we could keep dirty in the environment, and do everything automatically.

random qemu code:
 cpu_state_sync(env);
 // read registers
 // write registers
 cpu_state_dirty(env);

kvm arch code:
 if (env->registers_dirty) {
     copy registers to kernel
     env->registers_dirty = 0;
 }

The kernel code does similar things for registers which can be either in 
memory or in the vmcs; it uses accessors so cpu_state_sync() and 
cpu_state_dirty() aren't needed;

--
error compiling committee.c: too many arguments to function