On 7/7/20 6:58 PM, Paolo Bonzini wrote:
> On 07/07/20 15:58, Claudio Fontana wrote:
>> +
>> +CpusAccel tcg_cpus = {
>> + .create_vcpu_thread = tcg_start_vcpu_thread,
>> + .kick_vcpu_thread = tcg_kick_vcpu_thread,
>> +
>> + .synchronize_post_reset = tcg_cpu_synchronize_noop,
>> + .synchronize_post_init = tcg_cpu_synchronize_noop,
>> + .synchronize_state = tcg_cpu_synchronize_noop,
>> + .synchronize_pre_loadvm = tcg_cpu_synchronize_noop,
>> +};
>
> Could this struct reside in AccelClass instead, so that there's no need
> to register the operations explicitly?
This I tried a few times before, you can dig the history a bit of the comments.
The result of my attempts is that you end up replacing an "explicit
registration", with an explicit internal copy from a structure defined in
kvm-cpus.c
(for example) to the accelerator state, so you just switch one registration for
another.
The result is a pile of boilerplate of no use that is just confusing for the
reader.
> We could still cache it in a
> global variable in accel_init_machine, in order to avoid pointer chasing.
Indeed, the "cache" though ends up being the only useful thing,
and it can be even more confusing for the reader to see the whole machinery in
AccelClass.
>
> Thanks,
>
> Paolo
>
Still if you are interested to see what the result ends up being I can share
it, will need some fiddling..
Ciao,
Claudio
