PS3 Open Platform - some big news

I think their current worry is the fact that RSX can read and write XDR and they are afraid people might use it to reverse engineer their security mechanism/hypervisor/etc...
So can Cell! Why's that a worry? Is it not possible to lock out RSX from certain areas of RAM via the Hypervisor? I'd have thought there'd be address limiting somehow on RSX for both security and system robustness.
 
So can Cell! Why's that a worry? Is it not possible to lock out RSX from certain areas of RAM via the Hypervisor? I'd have thought there'd be address limiting somehow on RSX for both security and system robustness.

Yes, CELL can, but evidently it is much easier to block from CELL software compiled to run on CELL which is actually running on CELL rather than Shader programs starting Loads and Writes from RSX's Memory Controller.

Can the Hypervisor monitor Loads/Writes initiated by RSX's Memory Controller or is that regulated by hardware mechanism that are not software controlled ? Even if it can be done through the Hypervisor, it might not be as trivial as we might think (I am being overly cautious because I am fearful of making wrong assumptions on the Hypervisor and its ability to control PLAYSTATION 3's mist "intimate" details so I might have my brain in sleep mode and some very stupid things might be coming out...).

I do think it should be possible to limit/regulate RSX's XDR memory accesses, but there are some things to be said:

1.) it was not needed in developers' officially licensed kits.

2.) it is something SCEI would do JUST for consumers' PLAYSTATION 3 units running Linux/Other OS and it is not something that they wanted/felt need to dedicate a lot of man hours to address.

So, if SCE will ever allow some sort of 3D acceleration through RSX (even at high level) it will be when they will have developed a way to monitor all kinds of RSX behaviors from the Hypervisor.

I would not mind loosing 10-15% of RSX performance due to Hypervisor security checks, it would still infinitely better than not having any access to it.
 
So can Cell! Why's that a worry? Is it not possible to lock out RSX from certain areas of RAM via the Hypervisor? I'd have thought there'd be address limiting somehow on RSX for both security and system robustness.
Hypervisor is a thing running on Cell, and GPUs traditionally dont care for virtual memory or protection (My blind guess is that RSX aint different). I think Hypervisor could block System-XDR-RAM, but not System-GDDR-RAM. (How `bout ripping Blu-Ray Movies directly from framebuffer ?)
Further, switching OSes might be the weak spot, as you have to store the state with a possibly still working and fetching/writing GPU. You`d need to make sure RSX can get into a "switchable-state", as it possible (likely) cant save all task it has active (unlike DX10 GPUs) and has to finish all started polygons or even the most recent library-call as whole. Theres no easy way to make an "atomic" switch between processes in other words (unlike Cell) and if shaders cant be stalled,saved and continued then the switching cant be done in predictable time (unlike Cell).
I still hope we get HW-accellerated XWindows extensions atleast.. and maybe a fixed function OpenGL-Driver.
 
I don't know if this will work on PS3, but there is MythTV PVR http://www.mythtv.org/
Apparently this guy could use it on PS3 with some hacks
http://www.keshi.org/moin/moin.cgi/PS3/GO7007
http://www.keshi.org/blog/2007/01/mythtv_on_ps3_linux.html (Japanese)
As it has no acceleration by SPE/RSX right now it doesn't run very well for real use, nonetheless the future is bright.

BTW this guy wrote a Wiimote driver for PS3 Linux too
http://www.keshi.org/moin/moin.cgi/PS3/WiiRemote
 
Back
Top