ATI R600/700 Command Checker Published

[phoronix]

Phoronix: ATI R600/700 Command Checker Published

Jerome Glisse has sent a new patch to the other DRI developers that adds a command checker for the ATI R600/700 series graphics processors. The command stream (CS) checker ensures that no commands are attempting to access the GPU in a way that it shouldn't, such as accessing memory out of its resource size, which without the safety mechanism could be exploited by those with malicious intentions...

http://www.phoronix.com/vr.php?view=NzkzOA

[bugmenot]

Nice and important work, thank you.
But does this influence the performance?

[kernelOfTruth]

Nice and important work, thank you.
But does this influence the performance?

why is everyone so eager to get performance out of every patch ?

to answer your question:

it rather helps the devs to create a stable and feature-rich and high-performance driver instead of a (probably) slightly faster BUT rather unstable and hardlocking driver

[droidhacker]

Hardlock is the slowest it can ever be....

[drag]

It would probably help development. If a person is playing around with driver code and makes a mistake; this should catch it instead of just locking up the machine or whatever.

That should help speed up development, at least.

[glisse]

Command checking is slowing down rendering as we have to perform analysis of what the userspace sends us. That being said the overhead of this process seems very small as of today, ie our bottleneck is elsewhere.

Anyway, it's a feature opensource driver needs to have because you don't want some one to use your GPU to crash your kernel or do anythings malicious. Closed source drivers (what ever their platform is) likely doesn't do any kind of command checking, thus i believe some one very clever and with a very very deep understanding of GPU inner working can use those to gain access to all system memory. But hey this is theory and you know how hard, sometimes, it's to put theory into practise

[bridgman]

I guess command checking should only slow things down noticably if the system is running CPU-limited, ie the command queue is often empty and the limiting factor is the ability of app plus driver to push commands through the driver stack into the GPU.

I don't think that's going to be a problem on most systems - maybe the "small CPU plus big honkin GPU" systems will get there sooner

[Yfrwlf]

I guess command checking should only slow things down noticably if the system is running CPU-limited, ie the command queue is often empty and the limiting factor is the ability of app plus driver to push commands through the driver stack into the GPU.

I don't think that's going to be a problem on most systems - maybe the "small CPU plus big honkin GPU" systems will get there sooner

I knew I shouldn't have bought that 5970 for my 8086. :P

[droidhacker]

I knew I shouldn't have bought that 5970 for my 8086. :P

As a side note, I'm having a hard time getting it to plug in. How did you get yours to hook up?

[rohcQaH]

As a side note, I'm having a hard time getting it to plug in. How did you get yours to hook up?

duct tape.


I noticed that bridgman said in that fun and productive "70% performance" discussion that there are no linux apps to utilize most recent GPUs anyway (i.e. everything is CPU bound), but now it seems to have switched around


Of course a command checker is needed, so the (small) performance penalty cannot be avoided.