Re: [MirageOS-devel] Passing tap devices to the main module

[Anil Madhavapeddy <anil@xxxxxxxxxx>]

On 5 Oct 2014, at 22:03, Mindy <mindy@xxxxxxxxxxxxxxxxxxx> wrote:
> 
> On 10/05/2014 03:03 PM, Mindy wrote:
>> 
>> On 10/05/2014 11:57 AM, Anil Madhavapeddy wrote:
>>> I was wondering how to solve this more generally (for instance, via an 
>>> ARM-style device tree).  Does the solution below of picking monotonic 
>>> network device numbers work on cloud providers?
>>> 
>>> Amazon still seems to only provide one VIF, so maybe this isn't a real 
>>> issue.  It'll come up quite quickly on ARM when we are bridging 
>>> internal/external networks though, so it would be good to figure out a 
>>> solution..
>> Excellent question.  It apparently doesn't work on EC2, which does allow you 
>> to give an instance >1 VIF if you futz around in the console a bit.  Even 
>> though the second interface is listed as "device index" 1, versus "device 
>> index" 0 for the interface created automatically for the interface, 
>> attempting to get information for the second netif fails silently and 
>> information for the first is used instead (not great).
> 
> All that stuff I just said is only true if both VIFs aren't attached (i.e. 
> assigned to the interface in the Amazon console) at boot time; otherwise it 
> appears to work OK (the VM boots and does DHCP for two distinct and correct 
> MAC addresses representing the VIFs Amazon provides).

Thanks for checking!  That's encouraging -- I would only expect them to work if 
explicitly attached.

Block devices have another Xenstore numbering scheme that seems far more 
Linux-specific (and afaict, one that only Dave and Mort understand :-)

-anil
_______________________________________________
MirageOS-devel mailing list
MirageOS-devel@xxxxxxxxxxxxxxxxxxxx
http://lists.xenproject.org/cgi-bin/mailman/listinfo/mirageos-devel