On 02/04/2015 01:29 PM, Benjamin Tissoires wrote:

On Feb 02 2015 or thereabouts, Benjamin Tissoires wrote:

quoted

On Mon, Feb 2, 2015 at 4:46 PM, Dmitry Torokhov
[off-list ref] wrote:

quoted

On Wed, Jan 28, 2015 at 03:10:05PM -0500, Benjamin Tissoires wrote:

quoted

The current code releases the extra buttons right after they are pressed.
As soon as a new serio report comes in, the hw state is reset to 0
and so the buttons are released.

Check for the report type before acting on the current extra buttons
state.

No:

"If Ext is 0 (if bit 1 of bytes 1 and 4 are the same) then there are no
external buttons being pressed (or all external buttons have been
released)." - Synaptics PS/2 TouchPad Interfacing Guide PN:
511-000275-01 Rev. B

Hmm, indeed, the current code follows the spec. Yeah! a new firmware bug!

So when I dumped the ps/2 reports, I clearly saw the release report
being sent with the Ext bit to 1, and other reports were in between
with no information (because no fingers were on the touchpad).

OK, so with the hardware, here are the results:

After each incoming event (being a button or touch), when there is no
more information to send (i.e. touch released or button pressed or
released), I receive 80 times the following buffer:
80 00 00 c0 00 00

The number 80 seems quite consistant, though I got once 96.
Not sure that this repeated buffer might be of interest however.

When a finger is touched on the sensor, I receive the following:

b0 ba 35 c0 8d ed <- first finger down on the sensor
....
b0 ba 3b c0 6f d0
b0 ba 3b c2 6d d0 <- button 1 pressed
b0 ba 3b c0 6f d0
....
b0 ba 3b c0 6f d0
b0 ba 3c c2 6c d0 <- button 1 released
b0 ba 3c c0 6f d0
....
90 ba 26 c0 6f d0
80 00 00 c0 00 00 <- first finger released from the sensor
80 00 00 c0 00 00 <- repeated 80 times

So here, either the spec is wrong, either the Synaptics with FW 8.1 in
the Lenovos are not following it. But I clearly see that the extended
buttons are reported only when the Ext bit is 1.

Andrew, could you help us determine which way to go?
Ideally, could you point out at a firmware version where we could be
sure that the spec has not been followed so we can add a quirk in the
driver?

I got confirmation that this is a firmware bug and that it has probably 
been there since 2011. This bug is definitely in 8.1 and probably 
several versions before it. But, we haven't shipped anything with 
extended buttons in a long time, so applying the quirk to firmware 
revision 8.1 should be adequate. When we fix the bug we will bump the 
minor version.

Andrew

Cheers,
Benjamin