On Sat, Sep 01, 2012 at 09:47:00PM +0200, Henrik Rydberg wrote: > On heavy event loads, such as a multitouch driver, the irqsoff latency > can be as high as 250 us. By accumulating a frame worth of data > before passing it on, the latency can be dramatically reduced. As a > side effect, the special EV_SYN handling can be removed, since the > frame is now atomic. > > This patch adds the events() handler callback and uses it if it > exists. The latency is improved by 50 us even without the callback. > > Signed-off-by: Henrik Rydberg <[email protected]> > --- > drivers/input/input.c | 176 > +++++++++++++++++++++++++++++++++++--------------- > include/linux/input.h | 24 ++++++- > 2 files changed, 144 insertions(+), 56 deletions(-) > > diff --git a/drivers/input/input.c b/drivers/input/input.c > index a57c4a5..5b66b2f 100644 > --- a/drivers/input/input.c > +++ b/drivers/input/input.c > @@ -47,6 +47,8 @@ static DEFINE_MUTEX(input_mutex); > > static struct input_handler *input_table[8]; > > +static const struct input_value input_value_sync = { EV_SYN, SYN_REPORT, 1 }; > + > static inline int is_event_supported(unsigned int code, > unsigned long *bm, unsigned int max) > { > @@ -90,46 +92,81 @@ static void input_stop_autorepeat(struct input_dev *dev) > * filtered out, through all open handles. This function is called with > * dev->event_lock held and interrupts disabled. > */ > -static void input_pass_event(struct input_dev *dev, > - unsigned int type, unsigned int code, int value) > +static unsigned int input_to_handler(struct input_handle *handle, > + struct input_value *vals, unsigned int count) > { > - struct input_handler *handler; > - struct input_handle *handle; > + struct input_handler *handler = handle->handler; > + struct input_value *end = vals; > + struct input_value *v; > > - rcu_read_lock(); > + for (v = vals; v != vals + count; v++) { > + if (handler->filter && > + handler->filter(handle, v->type, v->code, v->value)) > + continue; > + if (end != v) > + *end = *v; > + end++; > + } > > - handle = rcu_dereference(dev->grab); > - if (handle) > - handle->handler->event(handle, type, code, value); > - else { > - bool filtered = false; > + count = end - vals; > + if (!count) > + return 0; > > - list_for_each_entry_rcu(handle, &dev->h_list, d_node) { > - if (!handle->open) > - continue; > + if (handler->events) > + handler->events(handle, vals, count); > + else > + for (v = vals; v != end; v++) > + handler->event(handle, v->type, v->code, v->value);
Filter handlers do not typically have ->event() methods, that's the reason for reported panic. Thanks. -- Dmitry -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
