[RPI BSP] resume, plan and questions for fb, fbcon gtk
pisa at cmp.felk.cvut.cz
Tue May 5 10:42:17 UTC 2015
Hello YANG Qiao,
On Tuesday 05 of May 2015 01:08:11 QIAO YANG wrote:
> Since I haven't yet received any respond of my last mail and I've now got
> the names of mentors who would instruct me, I create a new thread for a
> simple resume and post my questions .
> First of all, I've managed to use mailbox channel to get a framebuffer
> pointer and its informations. I can draw some simple images with the
> pointer and also the pointer can be retrieved correctly by ioctrl from
> applications, with the fb device interface. Painting works as well. I've
> already sent a patch with mailbox implementation, but I haven't yet
> confirmed if it's ready to be merged or any modification should be done.
> To complete fb implementation:
> 1. Some more functions need to be added : connected monitor EDID read and
> decoding for correct/optimal resolution selection. I don't know if we need
> a full VESA driver like what we've done on i386.
We do not need full VESA support but I suggest to look at EDID decoding
and related header file which is part of i386 VBE driver.
> 2. If we've enabled l2 cache, the memory mapping would be 0x40000000,
> while the mapping would be 0xC0000000 instead if l2 cache disabled.
> Normally it can be configured in the config.txt file.
I am not sure there, I would expect address remapping only in the case
that it is setup such way in the MMU. But MMU has to be enabled
to enable cache on ARM (at least on models which I have used).
> The file 'config.txt' is read by GPU before the ARM core is initialized.
> It's used as an BIOS configuration. It can be used to setup hdmi output
> mode, l2 cache, memory split etc. Such as memory split affect much the
> range of framebuffer address which should be covered in the mmu
> configuration table. What would be a better way to deal with it? Should I
> enable the memory pages that cover all possible range(This can vary depend
> on gpu_mem_split) ?
I you do not detect faramebuffer setup before MMU setup or use some RTEMS
functions to setup that frame-buffer region later in the code during
graphic driver startup then I suggest to setup (ideally 1 to 1) mapping
for all possible regions. There could be problem with number of available
TLBs but usually they are refilled from pagetable automatically so it
has only impact on speed. But I need to look more to that.
I have spent some time to read some more information about RPi VideoCore
communication channel and work on preparation support acceptable for
mainline Linux kernel. There are some links to share (I expect that you
have read some of them but some can be new for you, check please)
> To plan for fbcon implementation:
> 1. Find instructions for cursor. I've found that it's possible to get some
> infos from mailbox property channel.
> 2. Implementations refering to i386/pc386/console/outch.c
> 3. Implementations refering to i386/pc386/console/vgacons
Cursor is not a must for start. Microwindows provide soft cursor
for example. But if implemented then hardware cursor would
help with speed a lot.
> For GTK porting
Please, use RGTK for RTEMS Graphic Toolkit aka Microwindows or something
else than GTK because I have association to attempt to port
real GTK libraries set from the GTK name and taht would be instrresting
but really huge job for more/many developers.
> I tried to compile gtk for it. The libpng ... etc can be installed and I'm
> trying to make some samples for testing. Only the nxlib failed to build
> because we haven't yet the keyboard support (mouse support can be switched
> off, but I can't find out yet how to build the lib without keyboard).
It is possible to select keyboard driver named
By the way, I have setup Microwindows repository with most of my changes
and others RTEMS related changes.
When the school term/semester is over I hope to have some time for that
and prepare changes which should be safe for Greg Haer main Microwinodws
repository (now on GitHub)
More information about the devel