On 8/14/20 10:39 PM, Richard Henderson wrote:
> On 8/12/20 11:32 AM, Claudio Fontana wrote:
>> +static int64_t tcg_get_virtual_clock(void)
>> +{
>> + if (icount_enabled()) {
>> + return icount_get();
>> + }
>> + return cpu_get_clock();
>> +}
>> +
>> +static int64_t tcg_get_elapsed_ticks(void)
>> +{
>> + if (icount_enabled()) {
>> + return icount_get();
>> + }
>> + return cpu_get_ticks();
>> +}
>> +
>> +CpusAccel tcg_cpus = {
>> + .create_vcpu_thread = tcg_start_vcpu_thread,
>> + .kick_vcpu_thread = tcg_kick_vcpu_thread,
>> + .get_virtual_clock = tcg_get_virtual_clock,
>> + .get_elapsed_ticks = tcg_get_elapsed_ticks,
>> +};
>
> I think this variable should be const. Which of course means that the
> previous
> patch needs to add const annotations.
Will add.
>
> I think you should actually have multiple dispatch variables: with icount and
> without, with mttcg and without. That way these methods don't have to check
> icount_enabled() or qemu_tcg_mttcg_enabled() at runtime, only at startup.
I gave it a try, it would work, however it is basically a refactoring of the
existing qemu_tcg_init_vcpu ,
I'd do it in a separate series if possible (with some help from Alex?)
Claudio
>
>
> r~
>
