Return to enetOld.questions CVS log

Up to [Apple Darwin 0.x] / driverkit / notes

Darwin 0.2 Driver Kit

Ethernet questions

1) current driver - how is enprobe() called? It's in bus_driver endriver, but
   I don't see where it's referred to...
   > in ioconf.c - endriver inferred (during config) from en0 in MASTER.next?
   
2) netif stuff: In Unix server? In kernel? In separate task containing all of
   the network drivers and protocols etc.?
   
   In Unix server:
     -- easy to implement "Network Object" which provides methods analogous 
        to init_func, input_func, etc. (args to if_attach()). 
	
     -- very dirty to implement all of the functions in netif (e.g.,
        if_oerrors(), if_oerrors_set(), etc.). Need objc interface to
	all of these (in Unix server, accessible from network driver tasks)?
	Or RPCs to Unix server? Who would do all of this?
	
	-- a lot of these (like if_oerrors(), if_oerrors_set()) are pretty
	   dumb and shouldn't require RPCs. Just maintain this stuff
	   locally; these functions could be provided by NetworkDevice
	   superclass.
	   
	-- Necessary netif functions (called by driver, in netif.c):
	
		if_attach
		if_down
		if_handle_input (rld'd into driver task)?

     -- rld things like venip_input (the if_input function, in
        net/if_venip.c) into Enet task? What happens to the current 
	ifnet.if_input?
	
     -- What about ifnet.if_getbuf()? Seems way too expensive to do an RPC
        every time you do need to get an output buffer (one RPC to get
	the buffer, one to send to the driver). Also, mbuf chains ain't
	gonna work across RPCs. Need to unroll mbuf chains to get one
	continguous piece of vm to send to driver task...

     -- besides, this doesn't really seem right...IP is not Unix!
     
     -- How are we going to provide 3rd party network protocol support? 
        Rld into Unix server? Huh-uh. 
	
   In kernel: 
     -- could implement the netif functions needed by driver task as
        traps...seems kind of kludgy.
     
   One big "network" task, rld'd into at run time for 3rd party or optional
   protocols:
      -- rld inherently evil?
      -- use existing netif, or redo it in ObjC?
     
   Separate task for each driver and protocol:
      -- seems cleanest from one point of view (no rld, one module can't 
         crash another)
      -- Redo entire netif in ObjC? What's the interface to the rest of the 
         world (i.e., Unix and the kernel)? IODevice? NetworkDevice?

This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.