2023-06-30 17:34:55

by Peter Hilber

Subject: [RFC PATCH 3/7] timekeeping: Fix cross-timestamp interpolation for non-x86

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 always setting the delta to 0 when interpolating.

Fixes: 2c756feb18d9 ("time: Add history to cross timestamp interface supporting slower devices")
Signed-off-by: Peter Hilber <[email protected]>
---
kernel/time/timekeeping.c | 13 +++++++++----
1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
index 7e86d5cd784d..7ccc2377c319 100644
--- a/kernel/time/timekeeping.c
+++ b/kernel/time/timekeeping.c
@@ -1259,10 +1259,15 @@ 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);
+ if (do_interp) {
+ nsec_real = timekeeping_delta_to_ns(&tk->tkr_mono, 0);
+ nsec_raw = timekeeping_delta_to_ns(&tk->tkr_raw, 0);
+ } else {
+ nsec_real = timekeeping_cycles_to_ns(
+ &tk->tkr_mono, system_counterval.cycles);
+ nsec_raw = timekeeping_cycles_to_ns(
+ &tk->tkr_raw, system_counterval.cycles);
+ }
} while (read_seqcount_retry(&tk_core.seq, seq));

xtstamp->sys_realtime = ktime_add_ns(base_real, nsec_real);
--
2.39.2

2023-07-08 00:48:12

by John Stultz

Subject: Re: [RFC PATCH 3/7] timekeeping: Fix cross-timestamp interpolation for non-x86

On Fri, Jun 30, 2023 at 10:12 AM 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 always setting the delta to 0 when interpolating.
>
> Fixes: 2c756feb18d9 ("time: Add history to cross timestamp interface supporting slower devices")
> Signed-off-by: Peter Hilber <[email protected]>
> ---
> kernel/time/timekeeping.c | 13 +++++++++----
> 1 file changed, 9 insertions(+), 4 deletions(-)
>
> diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
> index 7e86d5cd784d..7ccc2377c319 100644
> --- a/kernel/time/timekeeping.c
> +++ b/kernel/time/timekeeping.c
> @@ -1259,10 +1259,15 @@ 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);
> + if (do_interp) {
> + nsec_real = timekeeping_delta_to_ns(&tk->tkr_mono, 0);
> + nsec_raw = timekeeping_delta_to_ns(&tk->tkr_raw, 0);
> + } else {
> + nsec_real = timekeeping_cycles_to_ns(
> + &tk->tkr_mono, system_counterval.cycles);
> + nsec_raw = timekeeping_cycles_to_ns(
> + &tk->tkr_raw, system_counterval.cycles);
> + }

Rather than adding another conditional branch here to go through, why
not just use "cycles" instead of system_counterval.cycles as it seems
to be set properly already?

thanks
-john

2023-07-27 10:40:35

by Peter Hilber

Subject: Re: [RFC PATCH 3/7] timekeeping: Fix cross-timestamp interpolation for non-x86

On 08.07.23 01:31, John Stultz wrote:
> On Fri, Jun 30, 2023 at 10:12 AM 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 always setting the delta to 0 when interpolating.
>>
>> Fixes: 2c756feb18d9 ("time: Add history to cross timestamp interface supporting slower devices")
>> Signed-off-by: Peter Hilber <[email protected]>
>> ---
>> kernel/time/timekeeping.c | 13 +++++++++----
>> 1 file changed, 9 insertions(+), 4 deletions(-)
>>
>> diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
>> index 7e86d5cd784d..7ccc2377c319 100644
>> --- a/kernel/time/timekeeping.c
>> +++ b/kernel/time/timekeeping.c
>> @@ -1259,10 +1259,15 @@ 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);
>> + if (do_interp) {
>> + nsec_real = timekeeping_delta_to_ns(&tk->tkr_mono, 0);
>> + nsec_raw = timekeeping_delta_to_ns(&tk->tkr_raw, 0);
>> + } else {
>> + nsec_real = timekeeping_cycles_to_ns(
>> + &tk->tkr_mono, system_counterval.cycles);
>> + nsec_raw = timekeeping_cycles_to_ns(
>> + &tk->tkr_raw, system_counterval.cycles);
>> + }
>
> Rather than adding another conditional branch here to go through, why
> not just use "cycles" instead of system_counterval.cycles as it seems
> to be set properly already?

OK. Thanks for the review and suggestion!