Re: [PATCH 2/2] viafb: Add OLPC XO-1.5 port configs

[Florian Tobias Schandinat <FlorianSchandinat@gmx.de>]

Daniel Drake schrieb:
> On 21 September 2010 18:54, Florian Tobias Schandinat
> <FlorianSchandinat@gmx.de> wrote:
>> *after a deeper look at the patch*
>> Hm, one of my patches enables 0x26 also as I2C. So the only difference
>> between the current config and your config is whether 0x2c is GPIO or I2C
>> where your version make a lot more sense.
>> Jon, do you agree that this should become the default config?
>> Handling a GPIO device as such and not as an I2C device sounds right to me,
>> although we have to take care about the old code which is doing so.
> 
> Jon, ping :)
> 
> Something possibly of relevance:
> In addition to enabling the camera and DCON, changing the port config
> of 0x2c from VIA_MODE_I2C to VIA_MODE_GPIO makes a fairly substantial
> difference in the viafb init paths for XO-1.5.
> 
> On linus master, before making the change in my patch, viafb init
> takes a long time, maybe 10 seconds. With linux-next 20100924, it
> takes 2.5 seconds. After the patch, it takes 0.07 seconds.
> 
> Timings from linux-next:
> 
> [    1.021495] VIA Graphics Intergration Chipset framebuffer 2.4 initializing
> [    1.029309] viafb 0000:00:01.0: PCI INT A -> GSI 16 (level, low) -> IRQ 16
> [    2.970023] viafb_init_dvi_size: DVI panel size undetected!
> [    3.475728] Console: switching to colour frame buffer device 100x40
> 
> And from the same kernel with the port config adjusted:
> 
> [    1.025436] VIA Graphics Intergration Chipset framebuffer 2.4 initializing
> [    1.033234] viafb 0000:00:01.0: PCI INT A -> GSI 16 (level, low) -> IRQ 16
> [    1.057799] viafb_init_dvi_size: DVI panel size undetected!
> [    1.090548] Console: switching to colour frame buffer device 100x40
> 
> So, changing this value makes a pretty significant difference. If you
> changed it globally I'm wondering if you'd be breaking current non-XO
> systems, for example you'd be presumably affecting the behaviour of
> viafb_lvds_trasmitter_identify()

You're right, we may not only change the entry. The device code also needs to be 
changed. The situation is like this:
0x2c are essentially two GPIO ports but those can be used to emulate one I2C 
port. At the moment it is only used as I2C port with a little hack in the I2C 
code. I believe it would be good to change it to always be a GPIO port (as that 
is what it really is, at least for all documented IGPs) and emulate I2C on top 
of that if needed. Sadly VIAs emulation (using 2 pins) differs from the generic 
one in i2c-gpio.c (using 1 pin), so we can't reuse that but the emulation code 
shouldn't be much code anyway.
I hope that this change is simple enough to get it right without testing as I 
don't have any device with a I2C device on port 0x2c.
But I hope we can get a bit faster than 2.5 seconds.


Regards,

Florian Tobias Schandinat