So far, get_device_system_crosststamp() unconditionally passes
system_counterval.cycles to timekeeping_cycles_to_ns(). But when
interpolating system time (do_interp == true), system_counterval.cycles is
before tkr_mono.cycle_last, contrary to the timekeeping_cycles_to_ns()
expectations.
On x86, CONFIG_CLOCKSOURCE_VALIDATE_LAST_CYCLE will mitigate on
interpolating, setting delta to 0. With delta == 0, xtstamp->sys_monoraw
and xtstamp->sys_realtime are then set to the last update time, as
implicitly expected by adjust_historical_crosststamp(). On other
architectures, the resulting nonsense xtstamp->sys_monoraw and
xtstamp->sys_realtime corrupt the xtstamp (ts) adjustment in
adjust_historical_crosststamp().
Fix this by deriving xtstamp->sys_monoraw and xtstamp->sys_realtime from
the last update time when interpolating, by using the local variable
"cycles". The local variable already has the right value when
interpolating, unlike system_counterval.cycles.
Fixes: 2c756feb18d9 ("time: Add history to cross timestamp interface supporting slower devices")
Signed-off-by: Peter Hilber <[email protected]>
---
Notes:
v2:
- simplify fix (John Stultz)
kernel/time/timekeeping.c | 6 ++----
1 file changed, 2 insertions(+), 4 deletions(-)
diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
index 70ecd44fdd9e..c145601ea062 100644
--- a/kernel/time/timekeeping.c
+++ b/kernel/time/timekeeping.c
@@ -1261,10 +1261,8 @@ int get_device_system_crosststamp(int (*get_time_fn)
tk_core.timekeeper.offs_real);
base_raw = tk->tkr_raw.base;
- nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono,
- system_counterval.cycles);
- nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw,
- system_counterval.cycles);
+ nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, cycles);
+ nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw, cycles);
} while (read_seqcount_retry(&tk_core.seq, seq));
xtstamp->sys_realtime = ktime_add_ns(base_real, nsec_real);
--
2.39.2
On Fri, Aug 25, 2023 6:04 John Stultz <[email protected]> wrote:
> On Thu, Aug 17, 2023 at 6:20 PM Peter Hilber
> <[email protected]> wrote:
>>
>> So far, get_device_system_crosststamp() unconditionally passes
>> system_counterval.cycles to timekeeping_cycles_to_ns(). But when
>> interpolating system time (do_interp == true), system_counterval.cycles is
>> before tkr_mono.cycle_last, contrary to the timekeeping_cycles_to_ns()
>> expectations.
>>
>> On x86, CONFIG_CLOCKSOURCE_VALIDATE_LAST_CYCLE will mitigate on
>> interpolating, setting delta to 0. With delta == 0, xtstamp->sys_monoraw
>> and xtstamp->sys_realtime are then set to the last update time, as
>> implicitly expected by adjust_historical_crosststamp(). On other
>> architectures, the resulting nonsense xtstamp->sys_monoraw and
>> xtstamp->sys_realtime corrupt the xtstamp (ts) adjustment in
>> adjust_historical_crosststamp().
>>
>> Fix this by deriving xtstamp->sys_monoraw and xtstamp->sys_realtime from
>> the last update time when interpolating, by using the local variable
>> "cycles". The local variable already has the right value when
>> interpolating, unlike system_counterval.cycles.
>>
>> Fixes: 2c756feb18d9 ("time: Add history to cross timestamp interface supporting slower devices")
>> Signed-off-by: Peter Hilber <[email protected]>
>
> Thanks again for iterating on this. This looks much better!
>
> Now, I've never had an environment that used this logic, so I'm
> trusting you've tested it well?
>
> Assuming so:
> Acked-by: John Stultz <[email protected]>
Thanks for re-reviewing!
I did automated tests with various chrony [1] configurations. The tests
check that all PTP_SYS_OFFSET_PRECISE2 ioctls issued by chrony are
successful for a combined test time of many hours, and that the
cross-timestamps look plausible.
I will add a description of the relevant tests when changing the series to
non-RFC.
[1] https://chrony-project.org/