Jonathan Stone <jonathan@DSG.Stanford.EDU> wrote:

   I think the person starting the thread acuatlly *did* want to move
   compression and data-transfer to userspace.  Which you're also
   against, yes?

Yes. e.g. compression: the performance gain of in-kernel
(de)compression, compared to in-modem compression, comes from the
reduced interupt and context switch rate. I think its wise to keep the
core of it there. Somebody already cited the reasons.

   >The only stuff you need in-kernel (or lkm'd, eventually) are the
   >compression protocols and the data transfer protocols used (but only
   >data compression part, not the negotiation part/state machine).

   Actually, according to Paul Mackerras, a small part of the CCP state
   *is* in the kernel, and gets handled there, because receiving  such
   a packet should affects  how the next output packet is handled.

Hm... this might well be, I don't know the details of that stuff.

Regards,
	Ignatios Souvatzis