Does it really stabilise the system?

Posted Mar 17, 2005 13:47 UTC (Thu) by nix (subscriber, #2304)
In reply to: Does it really stabilise the system? by NAR
Parent article: Handling interrupts in user space

Ideally, the X server would not need to use iopl() or run suid root: this is a step in that direction. In that situation, the most it could do without a kernel bug is lock up the console (maybe not even that).

to post comments

Does it really stabilise the system?

Posted Mar 17, 2005 15:20 UTC (Thu) by gnb (subscriber, #5132) [Link] (1 responses)

> In that situation, the most it could do without a kernel bug is lock up
>the console (maybe not even that).
That's optimistic. Any driver for a DMA-capable device can pretty much
obliterate the system. Moving stuff into user-space should eliminate
some classes of crash, but it's not the answer to everything.

Does it really stabilise the system?

Posted Mar 18, 2005 23:06 UTC (Fri) by giraffedata (guest, #1954) [Link]

Moving stuff into user-space should eliminate some classes of crash, but it's not the answer to everything.

Indeed. But the point of this proposal is that it eliminates not only some classes of crash, but the vast majority of the crashes. If it were only a few classes, it wouldn't be worth it.

A few examples of common coding errors that would obliterate the system in a real interrupt handler, but have relatively minor impact in a user interrupt handler process: infinite loop, wild pointer to nonexistent address, store into an arbitrary address.

The SGI ULI thing, though, is a horse of a different color. This eliminates relatively few risks and adds others.