[maemo-developers] ARM MMU questions

[From: Frantisek Dufka]

Simon Pickering wrote:

> If I try to name my memory section FRAMEBUFFER (same code, just changing
> the name in the command file), I'm back to the original problematic sort
> of output:
> 
> dsp_dld output:
> 
> dsp_dld: event detected.
> dsp_dld: event detected.
> device sharedmem is requesting for TADD.
> starting TADD process for device sharedmem.
> loading /lib/dsp/modules/sharedmem.o.
> exporting framebuffer to 0x300000
> FBEXPORT failed
> adding _task_sharedmem(@145120) into system.

Well it makes sense this is really a special case since (at least on 
770) you don't want to map just any random memory, you want to map 
_framebuffer_ memory (i.e. specific memory preallocated in bootloader 
and used by omapfb driver). It doesn't explain why the mapping should be 
done differently on DSP side but it is definitely handled as a special 
case in kernel so it needs special keyword. With different name you get 
some random memory not related to video (thus you need to do additional 
copy to real video memory on ARM side).


> Another thing I need to look at is whether I can share some large shared
> memory regions on the 770 at higher addresses in the DSP memory map (to
> be used by mplayer to avoid memcpy operations on frames).

Well that would be nice but most probably you need continuous physical 
(not virtual) memory for this so the frame can be used as real video 
memory = transferable via DMA to external video chip. Such regions have 
to be preallocated when kernel boots since at later time physical memory 
becomes fragmented.

Not sure how good is 2.6.21 (used future N8XX firmwares) but in older 
kernels such fragmentation is not easy thing to solve. In recent kernels 
there is some ongoing work related to defragmenting or preventing 
physcal memory fragmentation.

Also in N800 (and up?) some video RAM is in SRAM so there is definitely 
not space for more frames there. Even if using main SDRAM would be 
possible (i.e. there is no specific SRAM related setting somewhere in 
various pieces of omap framebuffer driver - disp.c omapfb_main.c 
blizzard.c rfbi.c - each hadling different piece of video hardware 
puzzle) we may suffer some performance hit for such frames.

Frantisek