On 20.01.2026 09:50, Jan Beulich wrote:
> On 15.01.2026 09:00, Jan Beulich wrote:
>> On 14.01.2026 18:49, Roger Pau Monné wrote:
>>> On Tue, Jan 06, 2026 at 02:58:11PM +0100, Jan Beulich wrote:
>>>> stime2tsc() guards against negative deltas by using 0 instead; I'm not
>>>> quite sure that's correct either.
>>>
>>> Hm, we should likely do the same for stime2tsc() that you do for
>>> get_s_time_fixed(). Given the current callers I think we might be
>>> safe, but it's a risk.
>>
>> Will do then.
>
> While doing so, I came to wonder if there isn't a reason for this "capping".
> In local_time_calibration() we also have
>
> /* Local time warps forward if it lags behind master time. */
> if ( curr.local_stime < curr.master_stime )
> curr.local_stime = curr.master_stime;
>
> Which for the use of stime2tsc() in cstate_restore_tsc() might mean that
> indeed there is a worry of the delta being negative, and the desire to
> "warp forward" in that case.
Proposed new function implementation (easier to look at than the diff):
uint64_t stime2tsc(s_time_t stime)
{
const struct cpu_time *t = &this_cpu(cpu_time);
s_time_t stime_delta = stime - t->stamp.local_stime;
int64_t delta = 0;
/*
* While for reprogram_timer() the capping at 0 isn't relevant (the returned
* value is likely in the past anyway then, by the time it is used), for
* cstate_restore_tsc() it is meaningful: We need to avoid moving the TSC
* backwards (relative to when it may last have been read).
*/
if ( stime_delta > 0 )
{
struct time_scale sys_to_tsc = scale_reciprocal(t->tsc_scale);
delta = scale_delta(stime_delta, &sys_to_tsc);
}
return t->stamp.local_tsc + delta;
}
Jan
