在2024年5月28日五月 下午8:17，Thomas Gleixner写道：
> On Tue, May 28 2024 at 15:31, Jiaxun Yang wrote:
>> 在2024年5月27日五月 下午5:32，Jiaxun Yang写道：
>>> 在2024年5月27日五月 下午4:51，Thomas Gleixner写道：
>>>> Jiaxun!
>>>>
>>> Hi Thomas,
>>> [...]
>>>>
>>>> What's this indirection for? Why dont you update
>>> [...]
>>>>
>>>>> +static struct clocksource nodecnt_clocksource = {
>>>>> + .name = "nodecnt",
>>>>> + .read = nodecnt_clocksource_read,
>>>>
>>>> the read function pointer here and spare the indirection?
>>> Smart! sched_clock takes slightly different function pointer argument type
>>> but as we don't use the argument anyway, it should be safe to relax this
>>> indirection.
>>
>> Sadly, I'm unable to remove it with force type catsing :-/
>>
>> drivers/clocksource/loongson-nodecnt.c: In function ‘nodecnt_clocksource_init’:
>> drivers/clocksource/loongson-nodecnt.c:89:36: warning: cast between incompatible function types from ‘u64 (*)(void)’ {aka ‘long long unsigned int (*)(void)’} to ‘u64 (*)(struct clocksource *)’ {aka ‘long long unsigned int (*)(struct clocksource *)’} [-Wcast-function-type]
>> 89 | nodecnt_clocksource.read = (u64 (*)(struct clocksource *))nodecnt_read_fn;
>>
>
> What about making the actual read functions have the required function
> signature? The clocksource argument is not used in those real functions.
Sadly sched_clock requires another signature :-/
I can see timer-clint hit into the same pitfall.
Thanks
>
> Thanks,
>
> tglx
--
- Jiaxun
