Re: of DMA,TDFX and Alpha (was:[Dri-devel] On Mice and X. Ignore me)

[Dave G. <gil...@tr...>]

On Mon, 3 Jan 2000, Daryll Strauss wrote:

> On Mon, Jan 03, 2000 at 09:57:48AM -0600, rsd wrote:
> > tends to aggrivate this. (sound cards, IDE drivers)  This happens on all
> > alpha's from the cabrolet to the newest DS20.  Is there a hardware bug here?
> > No other acceleration suffers from this.

Its possible but don't bet on it!

> > The bug (for me) misplaces the target.  It always expands the data correctly,
> > but - for instance, can shift the text to the right, or
> > repeatedly overwrite itself.  Others see no text at all.  But there is
> > something wrong!

Yeh, I get very little text displayed; typically nothing at all.

> I'm not much of a hardware guy myself. Can anyone guess why should DMA
> transfers be causing problems? All we're doing is writing to addresses
> on the board via a memory mapped region. The board itself isn't doing
> any DMA. I would think the hardware should handle the bus access
> transparently.

Are you sure? It isn't pulling the data out of memory by itself?
These type of things can go wrong for lots of reasons:

   1) If there is actually some DMA going on has the data got to the RAM
yet? i.e. if processor does a RAM write is it caught up in some cache
instead?

   2) Perhaps there is a hardware bug which requires a certain time
between some operations, or a certain type of PCI access to be used.

   3) Does the Alpha do any form of write combining or any other magic
with the PCI accesses that you perform?

Dave

-- 
 ---------------- Have a happy GNU millennium! ----------------------   
/ Dr. David Alan Gilbert      | Running GNU/Linux on       |  Happy  \ 
\   gro.gilbert @ treblig.org |  Alpha, x86, ARM and SPARC |  In Hex /
 ____________________________|___ http://www.treblig.clara.net   __/