Hi,

as far as I can see, the design will already be at least as stable and 
secure as all consumer graphics cards on the market. Timothy has already 
explained in another mail in this thread that there will be some security 
feature for indirect buffers, so software filtering in the kernel will be a 
minimum.

Quoting Timothy:
> The only DMA memory mapped into the process space is assigned only to
> the process and executed only via indirect DMA.  To command the GPU,
> the process would generate command packets and then initiate an ioctl
> to tell the kernel to add a command to the ring buffer that would
> cause the process's commands to be fetched.

That's already a lot better than basically all DRI-supported cards out 
there. As far as I know, the only (DRI-supported) chip that has such a 
feature is Intel's i915 (and maybe other chips from that family, I'm not 
sure about that).

Quoting Daniel:
> We're still working on that.  I think we ought to make a valiant attempt 
> to solve the window security problem, so long as it doesn't need more 
> hardware support than the ownership test.  We will make some friends 
> that way, and hey, it might even be nice.

I definitely agree that window security is something that needs to be 
solved.

However, the problem lies a lot deeper than just the driver. Every X11 
client can arbitrarily create windows and draw basically anywhere, even 
*without* using nasty hardware-dependent hacks. Show me a setup that has 
feasible window security under certain limitations, such as no OpenGL 
support, and I will do what I can (in the form of driver fixes) to extend 
that security to OpenGL clients. But until that time, there are other 
things to worry about.

With that said, with a window ID buffer and all (which again means this 
design is potentially more secure than most - if not all - consumer cards 
out there), I believe the only real driver-related problem for supporting 
window security is video memory mmaps, and those can be avoided without 
additional hardware support if you really want to.

cu,
Nicolai

Attachment: pgp3jdpwOcMfG.pgp
Description: PGP signature

_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to