[openib-general] Latest IPoIB Bringup Questions

[Hal Rosenstock]

On Thu, 2004-10-28 at 15:32, Roland Dreier wrote:
>     Hal> After ifconfig'ing an IPv4 address and netmask< I see the
>     Hal> that the QPN is filled in in the HWaddr (00-00-04-04). No
>     Hal> other bytes change. Is that correct ? Wouldn't the lower
>     Hal> bytes get filled in with some local GID ?
> 
> Actually with ifconfig I don't think you can see enough of the HW addr
> to see the GUID part of the GID.  On my box I have:
> 
>     # ifconfig ib0
>     ib0       Link encap:UNSPEC  HWaddr 00-00-04-04-FE-80-00-00-00-00-00-00-00-00-00-00
> 
> but
> 
>     # ip addr show dev ib0
>     6: ib0: <BROADCAST,MULTICAST,UP> mtu 2044 qdisc pfifo_fast qlen 128
>         link/[32] 00:00:04:04:fe:80:00:00:00:00:00:00:00:02:c9:01:07:8c:e4:61 brd 00:ff:ff:ff:ff:12:40:1b:ff:ff:00:00:00:00:00:00:ff:ff:ff:ff

Thanks. That looks better :-)

>     Hal> Some questions/comments on this sequence: 1. In general, how
>     Hal> long is the end node waiting before retransmitting a SA MC
>     Hal> request ?
> 
> It should wait 1 second and back off from there.  But I think I have a
> bug somewhere either in the IPoIB driver or the SA layer.
> 
>     Hal> 2. How are TIDs chosen ? It looks to me like there are 2
>     Hal> different MC requests potentially outstanding with the same
>     Hal> TID.
> 
> The SA layer uses idr.h to allocate the low TID -- so it shouldn't
> reuse a TID until the previous request times out or is completed.

This could be related to #1 above.

>     Hal> The last request is the insufficient component mask issue.
> 
> Right, that should be possible to fix now with the new SA API.
> 
>     Hal> One other minor comment: Should we teach ifconfig to display
>     Hal> Link Encap: INFINIBAND ?
> 
> Probably better to work on ip, since ifconfig has other issues (such
> as using an ioctl limited to 14 bytes to get the HW addr)

I have basic IP connectivity but tbe component mask issue is getting in
the way of some things in some cases I see a flood of join
requests/rejects until the application gives up. The "flood" may also be
due to the timing issues in #1 above.

-- Hal