On 18 April 2017 at 20:19, Pranith Kumar <bobby.pr...@gmail.com> wrote:
> On Tue, Apr 18, 2017 at 5:56 AM, Paolo Bonzini <pbonz...@redhat.com> wrote:
>>
>>
>> On 17/04/2017 20:55, Pranith Kumar wrote:
>>>>> +/* ARM does not have a user-space readble cycle counter available.
>>>>> + * This is a compromise to get monotonically increasing time.
>>>>> + */
>>>>> +static inline int64_t cpu_get_host_ticks(void)
>>>>> +{
>>>>> + return get_clock();
>>>>> +}
>>>> This doesn't look like it should be ARM-specific. Is it
>>>> better than the current default implementation? If so,
>>>> why not make this the default implementation?
>>>
>>> I think we can do that...
>>
>> Yes, it is always better for emulation accuracy. It may be much slower,
>> depending on your OS (especially if get_clock requires a
>> user->kernel->user transition), but the current code is quite broken.
>>
>
> OK, I sent an updated patch using get_clock() for all other cases.
Thanks. As it happens I just checked against what configure
supports for host CPU architectures, and ARM is the only one
without a special-purpose cpu_get_host_ticks() (except perhaps
elderly MIPS chips).
thanks
-- PMM
