Exokernel / Device drivers

William DUCK guillaume.fortaine at wanadoo.fr
Fri Oct 20 00:20:04 CEST 2006


The more I go forward, the more questions come to me :


the exokernel paper says somewhere that all the exokernel does is to
securely multiplex the cpu (that's time) and memory (that's space)
ressources.  that's what you said you want from a good os in your `why ...'
paper.  and some months ago, i observed that the os may be viewed as a
device driver for these two because that's what device drivers should do:
securely multiplex some hardware ressource


"In contract to our approach [L4], [Exokernel] is based on the
  philosophy that a kernel should /not/ provide abstractions but only
  a minimal set of primitives.  Consequently, the Exokernel interface
  is architecture dependent . . .  We believe that dropping the
  abstractional [sic] approach could only be justified by substantial
  performance gains. . . . It might turn out that the right
  abstractions are even more efficient than securely multiplexing
  hardware primitives or, on the other hand, that abstractions are too


So would it be possible to build a general purpose OS without a kernel ?

Isn't just a matter of language *and/or* hardware architecure ?

Could we imagine a full application-level OS and link the apps directly to the 
drivers => so the "kernel" would be the device drivers framework ?

Or at least the ressource management system ...

Best Regards,


More information about the l4-hackers mailing list