Hi Pavel, On Thu, Sep 10, 2015 at 11:11:41PM -0700, Pavel Roskin wrote: > Hi! Thanks for this detailed info. Adding Mika to CC who is an expert of this driver. Mika, maybe we should add you to MAINTAINERS, too? > I'm using Dell Inspiron 13, model P57G. It has a problem with the > touchpad under Linux. Approximately half of the time, the touchpad is > not working at all - the mouse cursor is not moving. In this case, > there are error messages written to the kernel log every second: > > [ 53.127339] i2c_designware INT3433:00: controller timed out > [ 54.219336] i2c_designware INT3433:00: controller timed out > [ 55.311346] i2c_designware INT3433:00: controller timed out > [ 56.403326] i2c_designware INT3433:00: controller timed out > > There is a simple fix - blacklist the driver > (i2c_designware_platform). In this case, the touchpad words as a mouse > and used IRQ 12. Googling for "blackist i2c_designware_platform" shows > that I'm not the one using that approach. > > I started looking at the driver in git. A patch by Romain Baeriswyl > applied on 2014-08-20 adds support for the "standard" mode with the > 100kHz clock, as opposed to the 400kHz "fast" mode. Unfortunately, > that patch only affects OpenFirmware systems, and I have ACPI. As soon > as I set the clock to 100kHz, the touchpad started working every time. > > One fix would be to have a module parameter to force slower clock. It > would still require users to deal with modprobe, so it's not optimal. > > I noticed that dev->get_clk_rate_khz(dev) returns 100000, but that > would be 100000kHz = 100MHz. Not sure is somebody confused the units > or it's the correct clock. > > Considering that I have no other systems to test, what would be the > best approach? Set the clock to 100kHz for INT3433? Recognize Dell > P57G system specifically? How? Preserve the original clock by reading > the DW_IC_CON register and keeping some bits? Note: The driver has been using 400kHz from its beginning. However, 100kHz is usually a better default for I2C busses. I wonder if we should take the risk of a performance regression in favour of a sane and working default? Regards, Wolfram
Attachment:
signature.asc
Description: Digital signature
