13:00:20 <baoli> #startmeeting PCI passthrough 13:00:21 <openstack> Meeting started Tue May 20 13:00:20 2014 UTC and is due to finish in 60 minutes. The chair is baoli. Information about MeetBot at http://wiki.debian.org/MeetBot. 13:00:22 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 13:00:24 <openstack> The meeting name has been set to 'pci_passthrough' 13:00:37 <baoli> Hi everyone. 13:00:52 <yjiang51> hi 13:00:54 <baoli> Hope everyone had a good trip back home 13:01:01 <irenab> hi 13:01:03 <beagles> hi 13:01:04 <BrianB_> Hi 13:01:25 <irenab> I came in 2 days delay... 13:01:50 <baoli> irenab, why was your flight getting cancelled? 13:02:05 <yjiang51> irenab: 2 days delay? are you also american airline? 13:02:16 <irenab> yes :-) 13:02:28 <irenab> never again,,, 13:03:08 <irenab> it was bead weather in Philadephia and weather radar not working on the plain 13:03:24 <beagles> wow 13:03:52 <irenab> But at least finally I am back, a little sick though 13:04:50 <irenab> sorry for not having progress on the spec, hope to catch up tomorrow and push for review 13:05:03 <baoli> Irenab, wish you a quick recovery. 13:05:16 <heyongli> hello, sorry late 13:05:16 <irenab> baoli: thanks 13:05:17 <baoli> let's get started with the nova-spec 13:05:28 <baoli> #topic sriov nova-spec 13:05:40 <baoli> irenab, do you need me to edit the spec? 13:06:08 <baoli> heyongli, hi 13:06:11 <irenab> baoli: if you have time today, please do and let me know. I hope can get to it tomorrow 13:06:12 <heyongli> hi 13:06:26 <baoli> irenab, sure, let me start it today, then 13:06:52 <irenab> baoli: fine, let me know by the end of the day if I can assist further 13:07:49 <baoli> irenab, I think initially I will just use the etherpad's contents. the current spec has much more. 13:08:46 <baoli> so the goal is to make it contain what we have discussed to deliver and hopefully to get it approved soon 13:09:17 <beagles> +1 13:09:27 <heyongli> sure 13:09:27 <irenab> baoli: agree, please us the spec we already have, just remove all unneeded content for now 13:09:44 <baoli> irenab, will do. 13:11:33 <baoli> Please take a look at the update once it's pushed up. And if everyone is ok with it, we should contact the cores for approval. 13:11:53 <heyongli> ok 13:11:56 <irenab> baoli: sure 13:12:04 <sadasu> sure 13:12:10 <baoli> Cool 13:12:14 <irenab> any other topics we want to cover today? 13:12:33 <baoli> irenab, yes. let's talk about the real work 13:12:45 <baoli> #topic nova patches 13:13:42 <BrianB_> Can we talk about baoli patch 13:14:35 <irenab> BrianB_: sure 13:14:36 <sadasu> +1 for cleaning up baoli's patch and posting it for review 13:14:36 <BrianB_> irenab, sadasu how has your testing been 13:14:51 <irenab> sadasu: +1 13:15:18 <irenab> BrianB_: faking neutron agent, it did the expected job 13:15:24 <sadasu> I have been working with baoli's patches in my lab setup and having success with it 13:15:59 <irenab> so seems that neutron "consumers" are quite satisfied :-) 13:16:27 <BrianB_> great so baoli can this be pushed up 13:16:43 <BrianB_> to get basic funionality in 13:16:59 <baoli> BrianB_: yes we can 13:17:17 <BrianB_> thanks 13:17:32 <irenab> baoli, heyongli: need to resolve the white_list/pci_stats modifications 13:17:43 <heyongli> i like to take it 13:18:55 <baoli> irenab, that's what i had wanted to talk about during the summit 13:19:13 <yjiang51> baoli: will you split your patch, or simply push it as a big one? 13:19:40 <heyongli> split might better 13:19:49 <irenab> I think we need clarly identify the tasks in the spec, so each one can be picked and pushed as separate patch. It is not mandatory to be done by different people 13:20:13 <yjiang51> irenab: +1 13:20:24 <baoli> irenab, +1 13:20:37 <heyongli> sure 13:20:50 <baoli> yjiang51, heyongli, we should talk about the changes with the changes in the whitelist 13:21:32 <heyongli> what's is it? 13:21:33 <beagles> I haven't gone through all of my email yet (yesterday was a holiday here) - so I'm assuming nobody has done this already, but it just occurred to me that we should've planned on summarizing the outcome and discussions of the summit for those who weren't there. 13:21:43 <beagles> Might reduce confusion and miscommunication 13:22:30 <yjiang51> heyongli: I think baoli is talking about the pci config white list change on your patch and his patch. 13:23:01 <heyongli> ok, what's the concern, baoli? 13:23:11 <baoli> beagles, I assumed that yjiang51 had talked with heyongli about the summit. If not, we can brief heyongli now 13:23:25 <heyongli> yes, we did 13:23:34 <baoli> ok, cool 13:23:44 <yjiang51> heyongli: the signle/multiple tag as discussed yesterday. 13:24:07 <heyongli> ok, for now only phy netowrk, right? 13:24:15 <yjiang51> heyongli: yes 13:24:19 <baoli> heyongli, it's just the changes we made in that area were quite different. 13:24:40 <baoli> heyongli, I gave comments to your patches 13:24:57 <heyongli> recently? 13:27:28 <yjiang51> heyongli: yes, for now only phy network. 13:27:38 <heyongli> i'm fine with it 13:29:25 <irenab> heyongli: can you please briefly describe what is proposed to cover phy net support? 13:29:56 <heyongli> add tag to whitelist, preset it to pci stats 13:30:24 <heyongli> do you expect this? 13:31:22 <irenab> is it same as baoli's suggestion? 13:31:59 <heyongli> no real conflict actually 13:32:43 <heyongli> make sure pci stats contain this tags information, and other information already exist 13:33:03 <irenab> baoli: Are you ok to go with this? 13:34:21 <baoli> irenab, for networking devices, I don't think that we want the vendor_id and product_id in the stats key, if that's what you guys are talking about 13:35:28 <baoli> if the device is tagged, then the tag is used for stats keys 13:35:44 <heyongli> does it crash the sriov? 13:36:10 <baoli> heyongli, can you rephrase? 13:36:45 <yjiang51> baoli: as discussed in summit, the pci stats should not block future potential extension, because we need consider upgrade. so we should keep the pci_stats more generic one. I think the first thing for pci_stats is to make it an object, because the compute manager objectify is waiting for it. 13:36:55 <heyongli> had that information is not stop sriov, and provide extra info for later enhancement maybe 13:38:04 <baoli> yjiang51, objectifying the pci_stats may be a new BP for it's own. 13:38:21 <baoli> The change allows multiple tags 13:38:42 <baoli> So it doesn't prevent future enhancement such as pci flavor, etc 13:38:57 <heyongli> image will need that 13:39:11 <yjiang51> baoli: Its covered by an existed BP already. If the changes allows multiple tags, it's ok and we can discuss on the review? 13:39:31 <baoli> yjiang51, which review? 13:40:25 <baoli> heyongli, other existing scheudler filters should cover the image requirement you brought up 13:40:58 <yjiang51> baoli: I assume you will split the patch into multiple one, and then one of them will be changes to pci_stats, right? We can review that patch then. 13:41:03 <heyongli> other filter never access pci information now 13:42:13 <baoli> heyongli, so you want to include vid, pid in the stats key as mandatory? 13:42:22 <yjiang51> baoli: let's put multiple info implementation aside, and only make sure no blocking for future changes. And also, we should split the patch and push them in parrellel. So that if there are disagreement on one patch, others will continue move forward. 13:42:44 <heyongli> +1 13:42:49 <baoli> yjiang51, +1 13:43:27 <irenab> shall we discuss alternative time for this meeting? Everyone can make current time slot? 13:43:31 <baoli> So we'll split my POC up and push them individually? 13:44:32 <yjiang51> baoli: I don't think we can resolve the pci_stats now, but at least we agree wih the white list, and it's a progress now. I had a quick check on the pci_stats part, frankly not quite sure on the changes. If it's splitted, it will be much better. 13:44:53 <irenab> baoli: I think it may help us to move faster 13:45:35 <baoli> cool 13:45:35 <irenab> baoli: you current POC still good to have as is to progress with neutron side implemention 13:46:50 <baoli> So we'll split it up in smaller patches 13:47:04 <baoli> And we will review them individually 13:48:54 <irenab> so this week, let's focus on the spec 13:49:28 <baoli> irenab, yes. let's push it for approval 13:49:55 <baoli> Anything else we want to talk about? 13:50:04 <yjiang51> baoli: heyongli, for the white list spec, do we still support address regexp? 13:50:41 <baoli> yjiang51, I'd like to use the lspci format that is known to people 13:51:17 <yjiang51> baoli: I remember someone strongly suggested something like 0000.*.3.*, but forgot who give the request. Does your patch support it? 13:51:42 <baoli> yjiang51, yes, and it's using the same format as lspci 13:52:40 <yjiang51> baoli: lspci has no regex, right? And what's the exact difference of lspci format and current implementation? I think currently it's also domain:b:d.f in the code. 13:53:17 <yjiang51> baoli: asking because seems your patch has no support for widecard etc. Of coruse, we can discuss in the patch review. 13:53:39 <baoli> yjiang51, take a look at the lspci man page. basically, it allows the use of . and * in individual fields. 13:54:25 <baoli> yjiang51, we don't have to reinvent it 13:54:40 <yjiang51> baoli: cool, it's ok only if it's supported. 13:54:47 <irenab> yjiang51, baoli: if lspci format provides required flexibility, I think its better to follow it 13:55:11 <baoli> irenab, that's the idea 13:55:28 <irenab> good 13:56:06 <irenab> So to summarize, baoli is planning to work on spec today and push if possible, right? 13:56:33 <baoli> irenab, yes 13:56:35 <irenab> irenab to join the effort on spec tomorrow, all to review one pushed. Agree? 13:56:39 <yjiang51> irenab: what's next meeting? 13:56:59 <irenab> next week, smae time? 13:56:59 <heyongli> irenab, fine 13:57:04 <baoli> #action baoli pushing the sriov nova-spec update 13:57:50 <irenab> yjiang51: can you make this time slotor want alternative time? 13:58:01 <irenab> ^time slot 13:58:45 <yjiang51> irenab: I thought either yongli or me will be in. 6:00 am is still a bit early to me. 13:59:07 <baoli> in that case, would 10:00pm too late for yongli? 13:59:15 <yjiang51> irenab: I will try, but you know, sometimes alerm does not work as expected. I had 3 alarm this morning. 13:59:32 <baoli> it is 9pm for him now 13:59:42 <irenab> yjiang51: :-) 13:59:55 <yjiang51> irenab: let' keep this time, and I will try to attend. I hope NFV meeting will replace this one sooon :) 14:00:16 <yjiang51> baoli: it's 10:00 pm for yongli already. 14:00:37 <heyongli> yeah, now it's 10:00pm 14:00:37 <baoli> yjiang51, I'm talking about the meeting start time 14:00:43 <baoli> sorry for the confusion 14:00:46 <yjiang51> baoli: :) 14:01:01 <irenab> ok, so see you next week same time 14:01:01 <baoli> so, same time next week? 14:01:07 <yjiang51> baoli: yes. 14:01:21 <baoli> thanks everyone. We're making a lot of progress 14:01:24 <baoli> today 14:01:29 <baoli> See you next week 14:01:31 <irenab> thanks! 14:01:35 <baoli> #endmeeting