Messages in this thread | | | Date | Tue, 5 Apr 2011 09:32:39 -0700 | From | Dmitry Torokhov <> | Subject | Re: [PATCH v2 4/4] input: evdev: Make device readable only when it contains a complete packet. |
| |
On Tue, Apr 05, 2011 at 02:03:27PM +0200, Henrik Rydberg wrote: > > >> Should use client->head here so that the SYN_DROPPED is readable. > > > > > > It is readable, but we do not want to signal on it. > > > > I think we do want to signal on it. We should signal whenever the > > device becomes readable. > > > > Signaling on dropped is useful in the case where a misbehaving device > > driver fails to ever call input_sync. If that happens, we might > > enqueue a dropped event and then never wake up the client which makes > > the issue hard to diagnose. > > A device that never wakes up the client seems like a detectable > symptom. I agree with Dmitry, the dropped event is more of a note in > passing, and as such can stay in the pipe until a real EV_SYN event > comes along.
Also we have evbug module to report raw event stream from theinput core without evdev involvement. It should show missing SYN_REPORT should such a driver appear.
> > > >> I don't think it's safe to modify last_syn outside of the spin lock. > > >> This should be done above. > > > > > > This is the only writer, plus we are running under event_lock with > > > interrupts off, so it is safe. > > > > The value will be read concurrently by evdev_fetch_next_event. So if > > this were safe, then we wouldn't need the spin lock at all. > > The spinlock ensures atomic read/write of the event buffer. The > position into the buffer does not need the lock. > > > At the very least for the sake of consistency, I think we should keep > > the buffer manipulations within the guarded region. > > Sounds reasonable.
OK, we can pull kill_fasync inside spin_lock/unlock pair, it should change nothing.
-- Dmitry
| |