[ofw] RE: [PATCH] Kill ATS registration in IPoIB
stan.smith at intel.com
Tue Jul 8 15:36:42 PDT 2008
Fab Tillier wrote:
>> From: Smith, Stan [mailto:stan.smith at intel.com]
>> Sent: Monday, July 07, 2008 10:29 AM
>> If the ipoib patch could be reworked such that the ipoib interface
>> IPv6 address GID pair is still registered with the SA under
>> ATS_NAME, then DAPL will continue to work correctly.
> IPoIB doesn't register the IPv6 addresses with the SA. In fact, I'm
> not sure it supports IPv6 properly - shouldn't there be some packet
> filtering/snooping to get the MAC to GID mappings from the neighbor
> discovery protocol?
IPoIB uses an IPv6 structure when registering an IPv4 address; sorry for
uDAPL/2 providers use an IPv6 address structure to retrieve the IPv4
address from the SA.
> Is there a dependency on ATS for IPv4 in uDAPL on Windows? If not,
> this is probably not an issue for DAPL at this point (or at least it
> doesn't break anything that isn't already broken.)
Yes, Windows uDAPL/2 have an explicit dependency on ATS. From looking at
the ibat interface, it may be less of a job to rewire DAPL/2 providers
for ibat than anticipated, testing will consume the bulk of the time....
Now that Fab has joined the party, if he were to work on supporting
multiple add_device() calls in ibbus as a filter driver, that would
allow me to work on wiring ibat into uDAPL/2, thus allowing the Kill ATS
patch to proceed forward..... After all Fab, it was your idea. A
Win-Win situation if you will? :-)
More information about the ofw