Re: USB NICs

[kimo_sabe@atdot.org]

On Mon, Mar 13, 2000 at 07:58:15AM -0500, Marty Connor wrote:
> 
> On 3/12/2000 11:01 PM Ken Yap ken@nlc.net.au wrote:
> >Apparently they are starting to be supported in Linux kernel 2.3.
> >Anybody played with these? I just wonder what would be needed to support
> >network booting with USB NICs.
  More core type(non-NIC) code.  There are two USB controler standards, both
should be supported, ideally.  Then there are the NICs them selves. There
are two drivers for USB NICs in the current devel kernels, and rumors of a
third being worked on by a manufacturer(Belkin I think).  One of them is
even a 10/100, on 12Mbits of bandwidth minus lots of protocol overhead. :)

> I have not played with any USB NICs yet, but I would imagine that if we 
> consider the Universal Serial Bus to be just another kind of bus, with 
> some controller chip, and some method to probe it, some method to address 
> devices on it, and some method to transfer data to and from devices 
> connected to it, one might imagine that support for USB (just like ISA 
> and PCI) would have to be added to Etherboot.  The of course begs the 
> questions, where would the boot media be for such a device?  How big 
> would the code be to support USB?
  boot media: system bios eeprom, or flash(Thinking of the Iopener, but that
has enough room to boot the kernel and have a filesystem itself) 

> I imagine looking at the Linux sources would be instructive;  I would 
> suspect we would have a "usb.c" and "usb.h" files, and we would scan the 
> bus somewhat like the pci code and isa code do.  If we ignore the 
> possibility of someone trying to "hot swap" the usb adapter during 
> etherboot, it could be possible.  The question I have is what the code 
> size for this might be.
  The three USB host controler drivers in the Linux kernel are 12.6k, 15.5k,
and 16.8k(according to size on the .o's).  The NIC drivers are 5k for the
10/100 chip, and 5.5k for the other.  That looks like it might be a little
tight, but do able.

> Next, there is the question of the kernel detecting and using a USB nic.
  As that's probably where the code will be stol, err, borrowed from, I
don't think that will be a problem. 

> Let's imagine for a moment we have a laptop.  We use a floppy, because we 
> don't have a place for a boot prom.  Suddenly, this looks a lot like the 
> PCMCIA problem.  And didn't I hear that PCMCIA code is coming inside the 
> kernel in 2.4 as well?   Could Firewire be too far off?
  Yup, it's there. I'm not sure if it can do rarp/bootp for NFSroot, but I
can test it, later today probably.
 
> In the USB case we have an ethernet card attached on a bus, and we have 
> to find it, initialize it, and do IO with it long enough to get a kernel 
> (and possibly RAM disk) loaded.  The we assume the kernel can handle what 
> must be done.
> 
> OK, there's some thinking out loud to "stir the pot".  I'm hoping others 
> with more experience will now jump in and comment :-)
  Thanks. It gave me some questions from the etherboot perspective to
attack.

  - Nick Lopez
    kimo_sabe@atdot.org
===========================================================================
This Mail was sent to netboot mailing list by:
kimo_sabe@atdot.org
To get help about this list, send a mail with 'help' as the only string in
it's body to majordomo@baghira.han.de. If you have problems with this list,
send a mail to netboot-owner@baghira.han.de.