13:01:07 <baoli> #startmeeting PCI Passthrough 13:01:08 <openstack> Meeting started Mon Feb 17 13:01:07 2014 UTC and is due to finish in 60 minutes. The chair is baoli. Information about MeetBot at http://wiki.debian.org/MeetBot. 13:01:09 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 13:01:11 <openstack> The meeting name has been set to 'pci_passthrough' 13:01:18 <baoli> Hi 13:01:20 <beagles> good morning! 13:02:13 <irenab> hi 13:03:01 <baoli> Yongli will not join. Let's wait for rkukura 13:04:13 <irenab> I think he said that he may not join 13:05:04 <baoli> yea. He said he might miss it. 13:06:05 <irenab> baoli: I want to discuss the bug fix comment https://review.openstack.org/#/c/59093/ 13:06:38 <baoli> Irenab, sure. But first check this out: https://wiki.openstack.org/wiki/Meetings/Passthrough#BPs_.26_Patches and see I missed anything 13:07:46 <sadasu> baoli: thanks for putting links to all significant bps in one place 13:07:49 <sadasu> will take a look 13:08:24 <sadasu> irenab: regarding bug fix 13:08:46 <baoli> sadasu, my pleasure 13:09:00 <irenab> please take a look at my response on Robert's comment 13:09:43 <sadasu> irenab: didn't rkukura mention that due to multisegment network xtension, we cannot take segmentation id from network? 13:10:06 <sadasu> are we assuming that we are not dealing with multisegment networks in icehouse? 13:10:37 <baoli> I think initially we'll make it work with providernet. 13:10:47 <irenab> baoli: agree 13:10:55 <baoli> irenab, I saw your reponse. I'm ok if it's needed now 13:11:43 <irenab> baoli: thanks! We need it to work for current mellanox MD,before all the nova support 13:12:04 <irenab> baoli: If you can please remove -1 13:12:27 <sadasu> irenab: agreed: I think it would help if the BP description was edited to reflect this fact 13:12:42 <irenab> baoli: Did the part for getting physical_network seems suitable for what you planned? 13:12:52 <baoli> not mention that nova bps won't be approved and any change may only be in juno 13:13:36 <baoli> irenab, yes. I'll take a close look at it again and remove -1. Agree with sadasu to add some comments about the future plan, etc. 13:13:36 <irenab> sadasu: it is actually driven as bug fix, so will add more info to the bug and commit message 13:14:12 <irenab> makeing it bp will result into Juno ... 13:15:08 <irenab> baoli: so no hope for getting nova code into icehouse? 13:15:18 <sadasu> irenab: agreed...bug description...because this important detail about supporting only providernet extension is hidden inside the comments... 13:15:25 <sadasu> I guess it is a nit pick 13:16:24 <irenab> sadasu: Will update the bug description and mention future plans 13:16:30 <baoli> irenab, I am not sure. You have seen the emails from the two cores involved in this so far 13:17:01 <irenab> seen only from John 13:17:57 <irenab> regarding the listed bp there is also this one: https://blueprints.launchpad.net/nova/+spec/pci-extra-info-icehouse 13:18:07 <beagles> baoli, are there links to the aforementioned email threads? 13:19:02 <baoli> beagles, somehow that email was not on the ML 13:19:09 <beagles> ah okay 13:19:21 <baoli> Because the originator didn't sent it to the ML 13:19:48 <irenab> baoli: Are you talking about mail from Russel? 13:19:48 <baoli> irenab, that's the one Yunhong registered for the simplified plan. 13:19:57 <baoli> irenab, yes 13:20:26 <beagles> Where I am nova-focused, I would like to do what I can to mitigate/expedite the nova blueprint approval 13:21:00 <beagles> it would be cool to "get that out of the way" ASAP. 13:21:17 <baoli> beagles, that would be great. When is the deadline for BP approval? 13:21:40 <baoli> Irenab, let me add it to the wiki 13:22:05 <irenab> I guess if code if under review by tomorrow, its still maybe possible to make it to Icehouse 13:22:24 <irenab> baoli: great, thanks for putting it all together 13:22:27 <beagles> baoli, oops...I'm not certain, but knowing that detail would be a good start :) 13:24:18 <baoli> We need the nova change to make the whole thing work. I'm working on a patch that will make sure we can have SRIOV working. 13:25:12 <baoli> So if we get all the essential neutron patches, I probably can do some integration testing in a week 13:26:52 <irenab> baoli: since we closed the API's between neutron and nova, I think its possible to put WIP nova parts + unit tests 13:27:11 <irenab> from my experience unit testing takes more time than code itself ... 13:27:34 <baoli> Irenab, that's right. 13:28:36 <baoli> From neutron side, we need the vnic_type, binding:profile and binding:vif_details ASAP. 13:29:13 <baoli> we have patches for vnic_type and binding:vif_details 13:29:43 <irenab> there is also for binding:profile 13:30:07 <sadasu> regarding binding:profile 13:30:07 <irenab> https://review.openstack.org/#/c/73500/ 13:30:15 <baoli> Yea, just saw it. 13:30:29 <baoli> I'll take a look, and hopefully I can get something working this week 13:30:36 <sadasu> irenab: do you need to differentiate between a PF and VF? 13:30:51 <irenab> I am dealing with VF only 13:31:12 <irenab> assuming that nova will allocate VF 13:31:38 <sadasu> irenab: ok 13:31:50 <baoli> sadasu, do you need PF? 13:32:14 <irenab> sadasu: Do you need additional property on the pci_info or pci_slot? 13:33:39 <sadasu> the NICs (cisco & intel) currently support allocation of the PF also for SR-IOV 13:34:13 <irenab> sadsu: does it matter on neutron side? 13:34:44 <sadasu> but Nova would only allocate VFs, we can assume in Neutron that it is always going to be a VF 13:35:14 <sadasu> just thinking if there might be a case in the future if we might have to support allocation of a PF 13:35:46 <baoli> sadasu, do you mean to say that a VM can use the entire PF for passthrough? Does libvirt support this in its XML? 13:35:56 <sadasu> in that case, do we want to pre-provision for this field in binding:profile 13:36:12 <sadasu> baoli: apparently yes 13:36:26 <irenab> sadasu: Do you see any different handling on neutron side? 13:36:42 <irenab> Or its just for nova VIF Driver? 13:36:54 <sadasu> irenab: not from my side...thats why I was checking with you 13:37:24 <irenab> sadasu: I think the same for me. 13:38:21 <baoli> sadasu, that will have a little implication on the nova side. 13:38:26 <sadasu> ok..so the current approach is: Nova will allocate only VFs 13:38:41 <baoli> sadasu, that's right 13:38:53 <irenab> sadsu: I guess it depends on white list definition 13:39:13 <sadasu> baoli: yes, Nova would be impacted...if we needed to PF support 13:39:24 <sadasu> continuing... 13:39:43 <sadasu> Neutron will assume only VFs are being allocated 13:39:59 <sadasu> binding:profile will not contain PF/VF info 13:40:07 <irenab> so it maybe the problem, getting the wrong device 13:40:10 <baoli> sadasu, any quick pointer to an example of the libvirt xml with PF? 13:40:29 <irenab> and nova vifdriver trying to plug it 13:41:06 <irenab> somewhere should be sanity check that device is suitable for vendor networking case 13:41:55 <sadasu> ok...will look for example and update 13:42:32 <irenab> so adding vf/pf to pci_info or pci_slot or nova will allow to plug only VFs? 13:43:09 <baoli> I think that we can add PF support in the next release. But let's find more information about it first 13:43:24 <irenab> baoli: agree 13:43:41 <irenab> I sent few questions regarding vnic_type implementation earlier today. Shall we discuss it now or over email? 13:43:58 <baoli> irenab, let's see what Bob has to say. 13:44:25 <baoli> I don't think that we need VNIC_AGENT. 13:44:30 <irenab> baoli: fine. please also review the patch 13:44:58 <baoli> vnic_type update before it's bound may be allowed. 13:45:04 <irenab> baoli: it just seems stange that in port-show it shows virtio 13:45:57 <irenab> what about changing vnic_type? 13:47:03 <irenab> baoli: just saw your answer. thanks 13:47:27 <irenab> it was my thoughts too 13:47:34 <baoli> irenab, cool 13:48:22 <irenab> baoli: any more for discussion today? 13:48:53 <baoli> irenab, I don't have anything left to discuss today. 13:49:15 <irenab> next sync on Wed? 13:49:43 <baoli> Maybe we should switch back to weekly? 13:50:10 <baoli> irenab, if needed, we can have a sync-up on WEd 13:50:17 <irenab> we missed Heyongli and rkukura today, maybe still do one more time this week? 13:50:48 <baoli> irenab, sure. let me send an invite after this meeting 13:50:56 <irenab> baoli: thanks 13:51:52 <irenab> baoli: please keep in sync regarding nova progress and if there is some WIP that can be taken for testing 13:52:23 <baoli> Irenab, sure. 13:52:26 <baoli> If we switch back to weekly, which date do you guys prefer? 13:52:43 <sadasu> yeah, sometime mid week for the weekly meeting would be good. 13:52:54 <irenab> Tue? 13:53:04 <sadasu> Tues/Wed 13:53:13 <baoli> then, let's do Tue 13:53:51 <irenab> great, so next week switching to weekly? 13:54:19 <sadasu> I think we are meeting only tomorrow for this week too 13:54:55 <irenab> Wed or tomorrow? 13:54:55 <baoli> Unless somebody propose a sync-up, let's do weekly from next week. we'll do one on Wed this week. 13:55:10 <irenab> baoli: great 13:55:18 <sadasu> baoli: sounds good 13:55:19 * beagles nods 13:55:26 <baoli> Cool. 13:55:32 <baoli> Thanks everyone 13:55:37 <sadasu> thanks 13:55:40 <irenab> Thanks 13:55:51 <baoli> #endmeeting