Re: [PATCH v4 1/3] time: add "NOW() good" indicator

On 28.07.2026 12:32, Roger Pau Monné wrote:
> On Tue, Jul 28, 2026 at 10:24:38AM +0200, Jan Beulich wrote:
>> On 28.07.2026 10:17, Roger Pau Monné wrote:
>>> On Tue, Jun 30, 2026 at 04:06:00PM +0200, Jan Beulich wrote:
>>>> --- a/xen/arch/x86/time.c
>>>> +++ b/xen/arch/x86/time.c
>>>> @@ -2660,6 +2660,7 @@ void __init early_time_init(void)
>>>>  
>>>>      set_time_scale(&t->tsc_scale, tmp);
>>>>      t->stamp.local_tsc = boot_tsc_stamp;
>>>> +    NOW_good = true;
>>>
>>> Would you need a barrier here to ensure compiler doesn't re-order the
>>> writes?  Maybe using ACCESS_ONCE(), or a smp_wmb() like it's used in
>>> other arches?
>>
>> Aiui it isn't needed here. The compiler can't re-order the two writes,
>> due to our use of -fno-strict-aliasing. The barrier is there on other
>> arch-es to avoid re-ordering in hardware.
> 
> You are the expert in compilers, but it was my understanding that
> `no-strict-aliasing` only affects the ordering of pointers accesses,
> but not plain variables.  IOW: NOW_good accesses could be reordered
> because it's not a pointer.  Does the compiler consider t can point to
> &NOW_good and hence it can't be reordered?

Yes, that's my understanding of how this work. (Yet no, "expert" surely
is going too far.) Things would be different if NOW_good was a function-
local variable, I think.

Jan