Re: [PATCH 5/8] x86/time: correctly honor late clearing of TSC related feature flags

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 04/07/16 16:53, Jan Beulich wrote:
>>>> On 04.07.16 at 17:39, <andrew.cooper3@citrix.com> wrote:
>> On 20/06/16 16:20, Jan Beulich wrote:
>>>>>> On 20.06.16 at 16:32, <andrew.cooper3@citrix.com> wrote:
>>>> On 15/06/16 11:28, Jan Beulich wrote:
>>>>> --- a/xen/arch/x86/time.c
>>>>> +++ b/xen/arch/x86/time.c
>>>>> @@ -1358,6 +1358,24 @@ static void time_calibration(void *unuse
>>>>>                       &r, 1);
>>>>>  }
>>>>>  
>>>>> +void __init clear_tsc_cap(unsigned int feature)
>>>>> +{
>>>>> +    void (*rendezvous_fn)(void *) = time_calibration_std_rendezvous;
>>>> This should read time_calibration_rendezvous_fn rather than assuming
>>>> time_calibration_std_rendezvous.
>>>>
>>>> Otherwise, there is a risk that it could be reset back to
>>>> time_calibration_std_rendezvous.
>>> But that's the purpose: We may need to switch back.
>> Under what circumstances could we ever move from re-syncing back to not
>> re-syncing?
> verify_tsc_reliability() may result in X86_FEATURE_TSC_RELIABLE
> getting cleared. That's an initcall, which means it runs after
> init_xen_time(), and hence after the rendezvous function got
> established initially.

Right, but that isn't important.

There will never be a case where, once TSC_RELIABLE is cleared, it is
safe to revert back to std_rendezvous, even if TSC_RELIABLE is
subsequently re-set.

Therefore, this function must never accidentally revert
time_calibration_rendezvous_fn from time_calibration_tsc_rendezvous back
to time_calibration_std_rendezvous.

~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel