Re: [PATCH 02/19] Input: Improve the events-per-packet estimate

On Tue, Aug 14, 2012 at 01:50:38PM -0700, Ping Cheng wrote:

On Tue, Aug 14, 2012 at 12:53 PM, Dmitry Torokhov
[off-list ref] wrote:

quoted

On Tuesday, August 14, 2012 12:32:21 PM Ping Cheng wrote:

quoted

On Sun, Aug 12, 2012 at 2:42 PM, Henrik Rydberg [off-list ref] wrote:

quoted

Many MT devices send a number of keys along with the mt information.
This patch makes sure that there is room for them in the packet
buffer.

Signed-off-by: Henrik Rydberg <redacted>
---

 drivers/input/input.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/drivers/input/input.c b/drivers/input/input.c
index 6e90705..8ebf116 100644
--- a/drivers/input/input.c
+++ b/drivers/input/input.c

@@ -1777,6 +1777,9 @@ static unsigned int

input_estimate_events_per_packet(struct input_dev *dev)>
                if (test_bit(i, dev->relbit))

                        events++;

+       /* Make room for KEY and MSC events */
+       events += 7;

Hi Henrik,

It is nice to get rid of the redundant pieces and to incorporate
common functions. Thank you.

I have a question about the code above though.  Why do we use 7
instead of going through the keys like:

      for (i = 0; i < KEY_MAX; i++)
              if (test_bit(i, dev->keybit))
                      events++;

Because that would result in gross over-estimation for many devices -
my keyboard has 100+ keys but it never sends all of them in one event
frame, not even if I can get a cat to lay on it ;)

Thanks for the prompt reply. I thought you were on vacation ;-).

No, just generally busy ;(

So, what device are we talking about here? I thought it is a touch
device with a few extra buttons, which are reported as key events. Am
I missing something?

I was talking about a bog-standard computer keyboard here.

If it is a touch device, we won't have too many buttons. So,
test_bit(i, dev->keybit) won't be true for more than the number of
buttons that declared by __set_bit().

input_estimate_events_per_packet() is a generic routine that is used for
all devices, not only [multi]touch.

I would think we could play a keyboard (this keyboard does not have
letters on it ;-) with ten fingers.

But even that keyboard would have more than 10 keys, right? So even
though max_events should be 10 + 10 + 1 (10 keys, 10 msc, syn) your loop
would produce what 88 + 88 + 1 for full size music keyboard?

Thanks.

-- 
Dmitry