From: Philip P. <phi...@re...> - 2010-12-12 19:12:44
|
On 12/7/10 12:37 PM, chas williams - CONTRACTOR wrote: > On Sun, 05 Dec 2010 12:39:56 -0800 > Philip Prindeville<phi...@re...> wrote: > >> Hi. >> >> I was thinking about how the kernel assigns IfIndex numbers for >> (amongst other things) SNMP to interfaces, and the fact that ATM >> doesn't use netdev's means that it can't be given one that's >> meaningful. >> >> It also can't be the source or sync for netlink messages >> (route/link.h). >> >> How could the ATM infrastructure be modified to better integrate with >> the kernel's notion of what an 'interface' is? > at one point there was talking about convering atm_device to use a > net_device instead. this would have fixed a couple of problems at once: > the device registration/allocation races, shutdown/removal races, > duplication of code, et al. the big drawback might be that raw atm > devices would appear if the output of ifconfig which might be confusing. > > atm isnt alone in this. infiniband maintains is own private device > structure and registration code. so conversion might not be necessary. > > as far as netlink, i think we just need to define a netlink family > for atm devices and them create the necessary routines to generate > and handle these messages. go through the archives. there was some > talk about this too at one point. What's the downside of using net_device's? -Philip |