There's one thing to add, though: By running device drivers in dedicated user-level processes, a crashing driver in the common case only takes down its own process, but the rest of the system continues to run. That's a major advantage, because you can have a monitoring process that then restarts the driver [1] this way.
However, even a user space driver can still crash the system if it goes rampant by misprogramming the DMA engine and you don't use an IOMMU.
Bjoern
Are there plans to provide IOMMU support in Fiasco.OC?
On Thu Jun 21, 2012 at 15:49:28 -0500, Robert Campbell wrote:
There's one thing to add, though: By running device drivers in dedicated user-level processes, a crashing driver in the common case only takes down its own process, but the rest of the system continues to run. That's a major advantage, because you can have a monitoring process that then restarts the driver [1] this way.
However, even a user space driver can still crash the system if it goes rampant by misprogramming the DMA engine and you don't use an IOMMU.
Are there plans to provide IOMMU support in Fiasco.OC?
Yes.
Adam
l4-hackers@os.inf.tu-dresden.de