Re: [utah-glx-dev] dma, agp

[Robert L. <bo...@ya...>]

On Fri, 21 Feb 2003 05:27, Philip Brown wrote:
> On Thu, Feb 20, 2003 at 04:53:24PM +0200, Levent Yavas wrote:
> >...
> > is this means pc dma controller cannot  memcopy to video memory? i don't
> > know much about that.
> >
> >...
>
> See below
>
> > >If you do,
> > >perhaps you could modify them to copy from normal RAM space to video
> > >memory, instead of the current methods that only do from video mem to
> > > video mem.
> >
> > if you talk about screen to screen copy done in riva_ctx.c, i don't know
> > anything about that. it should be accelerated by video card (xaa?). if
> > you talk about texture uploading in riva_tex.c, it already copies texels
> > from system ram to video ram.
>
> Yes, I am talking about screen to screen copy.
>
> If you take the video cpu instructions that tell the GPU
> "copy THIS memory address (in video memory), to THAT memory address
>   (in video memory)", and adjust the first memory address to point to
> something in "normal" memory, you would then effectively have
> DMA transfer.
>
> This is "Dynamic Memory Access" by the GPU to main memory. DMA, of one
> form. (there are other forms)
>
> I hope you would like to experiment with this.
> If you want to try, I would suggest stealing the routines in the mga
> driver that find memory suitable for DMA.
> (and then see how it "locks" the user memory into physical address space)
>
> Once that is done, it should be reasonable for you to try the
>  "copy this address in main mem to that address in video mem" operation,
> and see what happens. I sugest copying the XAA CPU code, and
> making a new function riva_copy_mem_to_video() or something.
>
> I'd then suggest a simple test of just filling the userside DMA memory
> chunk with 'white', aka 0xffffffff, then forcing a large copy to the main
> video area somehow as an initial test.
>
>

Maybe it's time for me to rationalise the memory allocation as we discussed. 
Memory locking and virtual-physical adress tranaslation was to be built in to 
that allocator