Skip to contentSkip to conversation

Small PCI core patch

RFC Robert Hancock 3 participants 3 messages Nov 25, 2005 5 messages not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

#5 lore raw
Quoted 3 lines · Jeff Garzik

One sticking point is validation: ensuring userspace cannot cause invalid GPU microcode to be generated. [I can just hear Al Viro swearing, just thinking about creating secure compilers...]

I suspect the amount of data going through is large enough that this wouldn't really be practical. I think you'd have to deal with the code generating GPU instructions having to be trusted and have the device interface require root privileges..

Signature · 3 lines
Robert Hancock      Saskatoon, SK, Canada
To email, remove "nospam" from [email protected]
Home Page: http://www.roberthancock.com/
#6 lore raw
Quoted 8 lines · Robert Hancock
Quoted 3 lines · Jeff Garzik

One sticking point is validation: ensuring userspace cannot cause invalid GPU microcode to be generated. [I can just hear Al Viro swearing, just thinking about creating secure compilers...]

I suspect the amount of data going through is large enough that this wouldn't really be practical. I think you'd have to deal with the code generating GPU instructions having to be trusted and have the device interface require root privileges..

All I said was "ensuring userspace cannot cause invalid GPU microcode to be generated."

No matter what runs with root priveleges, other graphics processes do not, and one must ensure that app clients cannot generate sequences which cause the hardware to fail. Which is a lot more difficult, when the unpriveleged app clients are submitting GLSL.

Jeff

#7 lore raw
In reply to Robert Hancock
Quoted 4 lines · Robert Hancock

I suspect the amount of data going through is large enough that this wouldn't really be practical. I think you'd have to deal with the code generating GPU instructions having to be trusted and have the device interface require root privileges..

You may wish to compare your suspicion with the DRI driver for via GPUs which does exactly this without problem. The majority of video traffic is not GPU instructions but textures which go via a different route. Also the GPU streams are generally very clean to parse.

Alan