On Tue, Dec 19, 2017 at 6:34 AM, Greentime Hu <green.hu@xxxxxxxxx> wrote:
> Hi, Arnd:
>
> 2017-12-18 19:13 GMT+08:00 Arnd Bergmann <arnd@xxxxxxxx>:
>> On Mon, Dec 18, 2017 at 7:46 AM, Greentime Hu <green.hu@xxxxxxxxx> wrote:
>>> From: Greentime Hu <greentime@xxxxxxxxxxxxx>
>>>
>>> This patch introduces some miscellaneous header files.
>>
>>> +static inline void __delay(unsigned long loops)
>>> +{
>>> + __asm__ __volatile__(".align 2\n"
>>> + "1:\n"
>>> + "\taddi\t%0, %0, -1\n"
>>> + "\tbgtz\t%0, 1b\n"
>>> + :"=r"(loops)
>>> + :"0"(loops));
>>> +}
>>> +
>>> +static inline void __udelay(unsigned long usecs, unsigned long lpj)
>>> +{
>>> + usecs *= (unsigned long)(((0x8000000000000000ULL / (500000 / HZ)) +
>>> + 0x80000000ULL) >> 32);
>>> + usecs = (unsigned long)(((unsigned long long)usecs * lpj) >> 32);
>>> + __delay(usecs);
>>> +}
>>
>> Do you have a reliable clocksource that you can read here instead of doing the
>> loop? It's generally preferred to have an accurate delay if at all possible, the
>> delay loop calibration is only for those architectures that don't have any
>> way to observe how much time has passed accurately.
>>
>
> We currently only have atcpit100 as clocksource but it is an IP of SoC.
> These delay API will be unavailable if we changed to another SoC
> unless all these timer driver provided the same APIs.
> It may suffer our customers if they forget to port these APIs in their
> timer drivers when they try to use nds32 in the first beginning.
Ok, thanks for the clarification.
> Or maybe I can use a CONFIG_USE_ACCURATE_DELAY to keep these 2
> implementions for these purposes?
I'd just add a one-line comment in delay.h to explain that there is no
cycle counter in the CPU.
Arnd
