Hi all,
We are seeing an issue with gpio_keys where the first event after
a SYN_DROPPED has the same timestamp as the SYN_DROPPED event itself.
After some investigation it looks like this is an issue with evdev
and not specific to gpio_keys.
The issue was originally introduced in commit 3b51c44 ("Input: allow
drivers specify timestamp for input events").
This commit introduced input_set_timestamp and input_get_timestamp.
The latter (despite the name) actually generates and stores a timestamp
in dev->timestamp if the driver did not set one itself. This timestamp
needs to be reset when events are flushed; otherwise the next event
will use a stale timestamp. A partial fix was implemented in 4370b23
("Input: reset device timestamp on sync"), but this does not cover the
case of SYN_DROPPED.
If a SYN_DROPPED is generated (currently only done by evdev), the
following happens:
- evdev calls input_get_timestamp to generate a timestamp for the
SYN_DROPPED event.
- input_get_timestamp generates a timestamp and stores it in
dev->timestamp
- When the next real event is reported (in evdev_events), evdev
calls input_get_timestamp, which uses the timestamp already
stored in dev->timestamp, which corresponds to the SYN_DROPPED event
How to fix:
- When a SYN_DROPPED is generated, after calling input_get_timestamp,
the timestamp stored in dev->timestamp should be reset (same as is
currently done in input_handle_event). The attached patch does that.
(Perhaps the underlying problem is that it is not expected that a
function called input_get_timestamp will have side effects. The
commit history of input.c shows that this has actually caused a
few issues since 3b51c44.)
Guillermo Rodríguez (1):
Input: evdev - fix wrong timestamp after SYN_DROPPED event
drivers/input/evdev.c | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
base-commit: e70140ba0d2b1a30467d4af6bcfe761327b9ec95
2.25.1
