[PATCH 5/9] Add i.MX5 framebuffer driver

[Sascha Hauer]

On Tue, Dec 14, 2010 at 02:40:09PM +0800, Liu Ying wrote:
> Hello, Sascha,
> 
> Please find my feedback with [LY] embedded.
> 
> And, a bug related with this patch:
> 1) Bootup the babbage board.
> 2) echo U:640x480p-60 > /sys/class/graphics/fb0/mode
> 3) cat /sys/class/graphics/fb0/virtual_size, it shows '640,480'.
> 4) echo 640,960 > /sys/class/graphics/fb0/virtual_size, change virtual
> yres to be 960.
> 5) cat /sys/class/graphics/fb0/virtual_size, it still shows '640,480'.
> I think this is caused by 'var->yres_virtual = var->yres;' in custom
> fb_set_par() function(Sorry, I cannot make the comment inline this
> time, as I don't find the original code within the mail I am
> replying). This makes fb0 use only one block of memory whose size is
> equal to screen size, so pan display can not really play her role and
> screen tearing issue may happen.
> 
> Best Regards,
> Liu Ying
> 
> 2010/12/13 Sascha Hauer <s.hauer at pengutronix.de>:
> > On Sun, Dec 12, 2010 at 02:13:40PM +0800, Liu Ying wrote:
> >> Hello, Sascha,
> >>
> >> I have following comments to this patch:
> >> 1) Please modify the commit message, as IPUv3 is not embedded in i.MX50 SoC.
> >
> > ok.
> >
> >> 2) ADC is not supported yet in the framebuffer driver, so please
> >> modify this comment:
> >>    > + * Framebuffer Framebuffer Driver for SDC and ADC.
> >
> > ok.
> >
> >> 3) 'ipu_dp_set_window_pos()' is called only once in
> >> imx_ipu_fb_set_par_overlay(). So, the framebuffer driver doesn't
> >> support to change the overlay framebuffer position. Need a
> >> mechanism/interface for users to change the overlay framebuffer
> >> position.
> >
> > Yes. The overlay support is currently only present in the driver because
> > I didn't want to break it in general. There is no generic overlay API in
> > the kernel and I didn't want to invent the n+1 API. Currently the
> > possible use cases of the overlay support are not clear to me. Number
> > one use case would probably be to display video output, but the
> > corresponding driver would be a v4l2 driver, so in this case we would
> > only need an in-kernel API.
> > Anyway, I have not the resources to implement complete overlay support.
> > So either we keep it like it is and it more has an example character
> > or we remove it completely from the driver for now.
> >
> >> 4) Need to make sure the framebuffer on DP-FG is blanked before the
> >> framebuffer on DP-BG is blanked. Meanwhile, the framebuffer on DP-FG
> >> should be unblanked after the framebuffer on DP-BG is unblanked
> >> 5) Need to check the framebuffer on DP-FG doesn't run out of the range
> >> of the framebuffer on DP-BG.
> >
> > I tend to remove the overlay support.
> >
> >> 6) I prefer to find the video mode in modedb first, and if we cannot
> >> find the video mode in common video mode data base, we can find a
> >> video mode in custom video mode data base which is defined in platform
> >> data. In this way, we don't need to export common modefb.
> >
> > I exported the modedb to be able to show the different modes under
> > /sys/class/graphics/fb0/modes and to set it with
> > /sys/class/graphics/fb0/mode. If you want this there is no way around
> > exporting it. The ioctl interface for mode setting is independent of the
> > specific modes anyway.
> [LY]  Just a concern. If a platform choose to add the generic video
> mode data base to IPUv3 framebuffer modelist, 'cat
> /sys/class/graphics/fb0/modes' will show all the video modes in the
> generic data base to users. I am not sure whether showing them to
> users means the IPUv3 framebuffer driver acknowledges to support all
> of them or not. I tried to do 'echo U:1800x1440p-70 >
> /sys/class/graphics/fb0/mode' with the kernel on ipuv3 branch, there
> will be a kernel dump(seems it can not allocate enough memory).

We'll have to increase the dma coherent size (and max zone order) for
these resolutions to work. I have patches for this in my local tree but
decided to wait until some contiguous memory allocator hits the tree.
Also, the fb driver should sanity check the generic modes before adding
them to the list. FOr example the IPU can't do 1920x1200 at 60. This code
is missing currently.

Sascha


-- 
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |