[syslinux] Re: Q: passing DHCP options to the booted linux-kernel?

[H. Peter Anvin]

Ph. Marek wrote:
>>>That would be perfect, especially, if the kernel understood this value
>>>too :-)
>>
>>Why should the kernel care about it?
> 
> At least in 2.4.x the option "kernel IP autoconfiguration" uses the "ip=" 
> parameter, which specifies the ip- and gateway-address, netmask and 
> dhcp-server.
> This configuration will be valid ONLY on the network interface which got this 
> information; therefore, only this interface can be configured.
> So there needs to be a way to tell the kernel about that.
> 

No, you're not getting my point.

Don't do that.

> 
>>>Thanks a lot, I'll be hoping for that.
>>>If you've got a patch for something around 2.4.20 I'm ready to test.
>>>(I'll need a newer pxelinux too, I suppose).
>>
>>The ipconfig/nfsroot stuff in the kernel is going away for 2.5, so I'm
>>not particularly interested in hacking on it since it's a dead end
>>anyway.  It's all moving to userspace in 2.5, and you can do it in
>>userspace in 2.4 as well by using an initrd and pivot_root.
> 
> I know. I currently use a self-written program which checks the line-status of 
> the interfaces (via MII-registers), and if only 1 cable is plugged into the 
> cards, this card is configured.
> 
> My fear is mostly that using userspace for a lot of things tends to need more 
> and more libraries, which I have to present on a boot floppy for 
> not-PXE-enabled machines - there space is very valuable.

Use a PXE bootfloppy instead.  Much easier.

> And no, I can't use NFSroot- I have to use some network drivers which can't be 
> compiled into the kernel but are modules, so I need userspace to load them.
> 
> So in the long term you're right - just passing an argument specifying the 
> hardware address from pxelinux to the kernel should be sufficient for most 
> cases (except when two cards share the hardware address *). As 2.5 will get a 
> klibc (although I have to admit that I don't really know what that means - is 
> that a minimal substitute for libc?) I hope that my needed userspace 
> libraries set will get smaller.
> 

Yes, that's what it is.  Note that there is nothing that keeps you from 
using klibc on an initrd in 2.4 -- it's just userspace.

> 
> *) No, it's not a 1/2^48 chance. I know some machines which record a UAA 
> across reboots, and if these are set for ease of network management ... 
> conflicts happen.
> 

If that's how you manage your network, then you deserve what you get. 
Really.

	-hpa