On Tue, May 09 2023 at 13:07, Peter Zijlstra wrote:
> On Mon, May 08, 2023 at 09:43:55PM +0200, Thomas Gleixner wrote:
>
>> +static inline void cpuhp_ap_update_sync_state(enum cpuhp_sync_state state)
>> +{
>> + atomic_t *st = this_cpu_ptr(&cpuhp_state.ap_sync_state);
>> + int sync = atomic_read(st);
>> +
>> + while (!atomic_try_cmpxchg(st, &sync, state));
>> +}
>
> Why isn't:
>
> atomic_set(st, state);
>
> any good?
Good question. It's only the other side (control CPU) which needs to be
careful.
Thanks,
tglx
