LVDS and DP through i2c

[Alex Deucher]

On Fri, Dec 24, 2010 at 7:14 AM, Kai-Uwe Behrmann <ku.b at gmx.de> wrote:
> Am 23.12.10, 11:09 -0500 schrieb Alex Deucher:
>>
>> On Thu, Dec 23, 2010 at 2:25 AM, Kai-Uwe Behrmann <ku.b at gmx.de> wrote:
>>>
>>> Hello,
>>>
>>> I try to talk to LVDS and DP connected displays over the i2c bus. What
>>> works
>>> nicely for external DVI displays appears silent for the internal one and
>>> DP.
>>> Any pointer what could be used or what shall be done to get a
>>> communication?
>>>
>>> example code for EDID fetching:
>>> /* 1. select i2c-dev node */
>>> const char * device_name = "/dev/i2c-0";
>>> /* 2. open device node */
>>> int fd = open( device_name, O_RDWR );
>>> /* 3. set adress */
>>> ioctl(fd, I2C_SLAVE, 0x50)
>>>
>>> /* 4. request desired block */
>>> usleep(50000);
>>> char bits[128] = {0};
>>> int ret = write( fd, &bits, 1 );
>>> if(ret == 1)
>>> {
>>>  /* 5. read data from node */
>>>  char data[128];
>>>  usleep(50000);
>>>  ret = read( fd, &data, 128 );
>>> }
>>>
>>> Already point 4 fails )-1) with a LVDS i2c node.
>>> The DP port returns arbitrary data in point 5.
>>>
>>> The DP is on a nvidia card, LVDS seems good on nvidia but not working on
>>> a
>>> evergreen.
>>>
>>
>> For radeon, make sure you are using at least 2.6.34, as previous
>> kernels did not expose i2c buses properly due to missing pre/post_xfer
>> i2c bit algo hooks.  Next, make sure you are opening the correct i2c
>
> 2.6.35 worked nicely. Thanks for the tip.
>
>> bus.   On radeons, we expose all of the on board i2c buses.  In most
>> cases, each connector uses a different i2c bus and there are
>> additional buses for thermal chips and fan controllers, etc.  Finally,
>
> I check the return values in the code above and the received EDID header.
>
>> not all oems implement an EDID for the laptop panels, or in some cases
>> the EDID image is stored in the vbios.  If you just want the EDID, you
>> can query that using randr or via sysfs.
>
> For xrandr people seem to have to wait until the KMS patch is not rejecting
> for wrong checksum. sysfs looks interessting too.

Yeah, that should be changed or at least an option provided to not reject them.

Alex

>
>> Alex
>
>
> kind regards
> Kai-Uwe Behrmann
> --
> developing for colour management www.behrmann.name + www.oyranos.org
>