| Message ID | 20240426135814.141837-1-inv.git-commit@tdk.com (mailing list archive) |
|---|---|
| State | Accepted |
| Headers | show |
| Series | iio: invensense: fix interrupt timestamp alignment | expand |
On Fri, 26 Apr 2024 13:58:14 +0000 inv.git-commit@tdk.com wrote: > From: Jean-Baptiste Maneyrol <jean-baptiste.maneyrol@tdk.com> > > Restrict interrupt timestamp alignment for not overflowing max/min > period thresholds. > > Fixes: 0ecc363ccea7 ("iio: make invensense timestamp module generic") > Cc: stable@vger.kernel.org > Signed-off-by: Jean-Baptiste Maneyrol <jean-baptiste.maneyrol@tdk.com> Hi Jean-Baptiste, I'll pick this up, but for future similar cases, please make a clear statement in the patch description on whether this is a theoretical problem, one found by some tooling, or (the most important bit) something that actually happens in real usage! That info helps people decided on how aggressively to backport that change. Applied to the fixes-togreg branch of iio.git. Given that has a link tag to this thread, replying here with the above will make that info somewhat available. We are late in cycle, so I may just move this to the final pull request for the merge window if I don't have many other fixes queued up. Thanks, Jonathan > --- > drivers/iio/common/inv_sensors/inv_sensors_timestamp.c | 9 +++++++-- > 1 file changed, 7 insertions(+), 2 deletions(-) > > diff --git a/drivers/iio/common/inv_sensors/inv_sensors_timestamp.c b/drivers/iio/common/inv_sensors/inv_sensors_timestamp.c > index 3b0f9598a7c7..4b8ec16240b5 100644 > --- a/drivers/iio/common/inv_sensors/inv_sensors_timestamp.c > +++ b/drivers/iio/common/inv_sensors/inv_sensors_timestamp.c > @@ -101,6 +101,9 @@ static bool inv_update_chip_period(struct inv_sensors_timestamp *ts, > > static void inv_align_timestamp_it(struct inv_sensors_timestamp *ts) > { > + const int64_t period_min = ts->min_period * ts->mult; > + const int64_t period_max = ts->max_period * ts->mult; > + int64_t add_max, sub_max; > int64_t delta, jitter; > int64_t adjust; > > @@ -108,11 +111,13 @@ static void inv_align_timestamp_it(struct inv_sensors_timestamp *ts) > delta = ts->it.lo - ts->timestamp; > > /* adjust timestamp while respecting jitter */ > + add_max = period_max - (int64_t)ts->period; > + sub_max = period_min - (int64_t)ts->period; > jitter = INV_SENSORS_TIMESTAMP_JITTER((int64_t)ts->period, ts->chip.jitter); > if (delta > jitter) > - adjust = jitter; > + adjust = add_max; > else if (delta < -jitter) > - adjust = -jitter; > + adjust = sub_max; > else > adjust = 0; > > -- > 2.34.1 >
Hello Jonathan, this is a bug that happens in real usage, especially when using high frequencies or low-end system. The delta timestamps between 2 samples periodically exceeds the configured jitter when doing interrupt timestamp alignment, while this timestamping method should assure it never happens. This is one of the goals of this timestamping module. I will make sure to mention that in future patches. Thanks for the guidance, JB
diff --git a/drivers/iio/common/inv_sensors/inv_sensors_timestamp.c b/drivers/iio/common/inv_sensors/inv_sensors_timestamp.c index 3b0f9598a7c7..4b8ec16240b5 100644 --- a/drivers/iio/common/inv_sensors/inv_sensors_timestamp.c +++ b/drivers/iio/common/inv_sensors/inv_sensors_timestamp.c @@ -101,6 +101,9 @@ static bool inv_update_chip_period(struct inv_sensors_timestamp *ts, static void inv_align_timestamp_it(struct inv_sensors_timestamp *ts) { + const int64_t period_min = ts->min_period * ts->mult; + const int64_t period_max = ts->max_period * ts->mult; + int64_t add_max, sub_max; int64_t delta, jitter; int64_t adjust; @@ -108,11 +111,13 @@ static void inv_align_timestamp_it(struct inv_sensors_timestamp *ts) delta = ts->it.lo - ts->timestamp; /* adjust timestamp while respecting jitter */ + add_max = period_max - (int64_t)ts->period; + sub_max = period_min - (int64_t)ts->period; jitter = INV_SENSORS_TIMESTAMP_JITTER((int64_t)ts->period, ts->chip.jitter); if (delta > jitter) - adjust = jitter; + adjust = add_max; else if (delta < -jitter) - adjust = -jitter; + adjust = sub_max; else adjust = 0;