Monday, 2017-08-21

*** svenkat has joined #openstack-powervm00:07
*** thorst has joined #openstack-powervm00:09
*** thorst has quit IRC00:13
*** thorst has joined #openstack-powervm00:44
*** thorst has quit IRC00:45
*** edmondsw has joined #openstack-powervm01:13
*** edmondsw has quit IRC01:18
*** thorst has joined #openstack-powervm01:46
*** thorst has quit IRC01:51
*** svenkat has quit IRC02:01
*** thorst has joined #openstack-powervm02:47
*** thorst has quit IRC02:51
*** edmondsw has joined #openstack-powervm03:02
*** edmondsw has quit IRC03:06
*** chhavi has joined #openstack-powervm03:48
*** thorst has joined #openstack-powervm03:48
*** thorst has quit IRC03:52
*** tonyb has quit IRC03:53
*** tonyb has joined #openstack-powervm04:00
*** edmondsw has joined #openstack-powervm04:50
*** edmondsw has quit IRC04:54
*** thorst has joined #openstack-powervm05:49
*** thorst has quit IRC05:54
*** k0da has joined #openstack-powervm06:03
*** thorst has joined #openstack-powervm06:51
*** mawi has joined #openstack-powervm06:53
*** thorst has quit IRC06:55
*** k0da has quit IRC07:06
*** thorst has joined #openstack-powervm07:51
*** thorst has quit IRC07:56
*** edmondsw has joined #openstack-powervm08:26
*** k0da has joined #openstack-powervm08:30
*** edmondsw has quit IRC08:30
*** thorst has joined #openstack-powervm08:52
*** thorst has quit IRC08:56
*** edmondsw has joined #openstack-powervm10:14
*** edmondsw has quit IRC10:19
*** thorst has joined #openstack-powervm10:54
*** thorst has quit IRC10:58
efriedmiltonm Yup, we've been tracking that.  esberglu and I are +1 on it - just waiting for devstack cores to decide it's good to go.11:12
efriedThink they may be a little gun-shy considering the snafu it caused last time :)11:12
efriedand/or busy with the release.11:13
efriedand/or everyone's on vacation last week and this.11:13
*** openstackgerrit has quit IRC11:17
*** chhavi has quit IRC11:45
*** edmondsw has joined #openstack-powervm11:55
*** thorst has joined #openstack-powervm11:57
*** svenkat has joined #openstack-powervm12:18
*** svenkat has quit IRC12:22
*** chhavi has joined #openstack-powervm12:23
*** svenkat has joined #openstack-powervm12:23
*** AndyWojo has quit IRC12:26
*** AndyWojo has joined #openstack-powervm12:27
*** edmondsw has quit IRC12:29
*** esberglu has joined #openstack-powervm13:15
*** apearson has joined #openstack-powervm13:28
*** edmondsw has joined #openstack-powervm13:29
*** apearson has quit IRC13:47
*** apearson has joined #openstack-powervm13:48
*** edmondsw has quit IRC16:05
*** edmondsw has joined #openstack-powervm16:07
*** edmondsw has quit IRC16:11
*** k0da has quit IRC16:20
*** miltonm has quit IRC16:21
*** miltonm has joined #openstack-powervm16:34
*** apearson_ has joined #openstack-powervm16:39
*** apearson has quit IRC16:40
*** edmondsw has joined #openstack-powervm17:39
*** edmondsw has quit IRC17:43
*** edmondsw has joined #openstack-powervm17:52
*** chhavi has quit IRC17:58
*** mawi has quit IRC18:13
edmondswefried what does a drc_name look like?19:09
efriededmondsw Like a location code.19:10
efriededmondsw U5791.001.CEC1202-P02-C027 kind of thing19:10
edmondswit's a unique id for a slot?19:10
efriedBut so is the DRC index.19:10
thorstefried: is DRC the PCI identifier for us?19:10
efriedthorst We're generating a pseudo PCI address from the DRC index.19:11
efriedWas just about to post up that change set.19:11
efriedYes, let's discuss.19:11
efriedCause there's not a good way to do this.19:11
efriedAnd this is the least awful way I've figured.19:11
thorstI'm just not a huge fan of DRC's19:14
thorstwe had that in the NL installer and it never worked out there19:14
thorstbut I think that was because pvmctl wasn't boot strapped yet19:14
efriedthorst edmondsw 575519:14
efriedThe only thing we need it for is to uniquely identify the slot in an environment (like nova) that's expecting a device to have a PCI address in the format of a PCI address.19:15
efriedSerendipitous that y'all were asking about this right as I was putting it together - where did the question come from?19:17
edmondswefried I'm looking at 574919:20
edmondswthe exception there says "slot at <drc_name>" and that just sounds odd to someone used to thinking of "slot <index>"19:21
efriededmondsw I think people are a lot more used to looking at DRC names (i.e. location codes) than anything else, when it comes to Power slots/cards.19:21
edmondswefried probably... POWER isn't my background19:22
edmondswhence the question19:23
edmondswleft you some more comments on 5749... not quite sold yet19:23
efriedThey're definitely not interested in looking at PCI addresses, cause there ain't no such thing.  And my hope is to keep these pseudo PCI addresses hidden from the user.19:24
efriedI debated whether to put the pseudo PCI address generator in nova instead.19:24
efriedWe should still discuss that.19:24
efriedThing that tips me over into having it in pypowervm is, I'm not sure we won't eventually need to use it in our neutron plugins as well.19:25
edmondswefried what would be the reason to put it in pypowervm? Do we think something besides nova would find that useful?19:25
efriedThat said, if we do, it may or may not be in the context of this SlotInfo thingy.19:26
efriedSo I just dunno.19:26
edmondswefried make it a utility method not part of SlotInfo?19:27
edmondswefried or do it on IOSlot...19:36
efriededmondsw Then I'd have to implement it twice, or make a base class.19:37
efriedWhich is part of the ew of this whole thing.19:37
*** k0da has joined #openstack-powervm19:41
edmondswesberglu ceilometer stable/pike is out19:50
esbergluOf course the day the workaround gets in...19:51
esbergluI thought that they were waiting to release with the official date?19:52
edmondswthey weren't supposed to be... I think they were confused19:52
edmondswesberglu seems like our emails on the ML helped move that along19:53
efriedI also pestered 'em in -release again earlier today.19:54
edmondswesberglu you might want to look at the last email in that ML thread for future reference when we're using releases19:54
efriedI had been tempted to propose the release change set myself over the weekend.19:54
efriededmondsw For 5755, I deliberately left the explanation out of the code comments.  I don't really want to expose that much of our dirty undercarriage to the world.  But I'm happy to explain it to you here.19:56
efried1) Why do we ignore the first byte?  seroyer led me to believe it's always the same (I've got test data that seems to belie that, but who knows where the test data came from?)19:57
efried2) The other pieces are being used as they are because they seem to map to right-sized pieces on the system (again according to seroyer).19:58
efriedA "domain" is two bytes and a "bus" is one; and the DRC index is apparently constructed as 0x21aabbbb.  So I used the two-byte part for the two-byte part, and the one-byte part for the one-byte part.19:58
efriedI didn't use the one-byte part for the slot because the slot of a PCI address is max 0x1F and that 'aa' part of the DRC index can apparently be more than that.19:59
edmondswefried ok. And why don't we have PCI addresses in POWER? Or maybe we do, but just buried at a layer we don't work at?20:00
edmondswI assume the latter20:00
efriedFinally, the 'func' (which has a max of 0x7) at least begins to relate to something I've dealt with before, which would have to be an SR-IOV physical function - for which we may be able to support passthrough at some point, whereupon we may use that field.20:01
efriededmondsw Right, because PHYP ain't Linux.20:01
efriedWe have a similar concept, which maps to physlocs (and presumably there's *some* correlation between physlocs and DRC indices).20:02
efriedPower has, like, CEC/Drawer/Slot/pport/vport or something.20:03
edmondswefried should we run this by someone on the PHYP team and see what they think of it?20:03
efriededmondsw Like seroyer?20:03
efriedCause that's where this came from.20:03
edmondswok, cool20:03
edmondswefried seems like if this is a place where nova is requiring a Linux-ism, we could go fix that in nova20:05
efriededmondsw Yes, I would like to do that eventually.20:06
edmondswefried cool... keeping a list somewhere?20:06
efriedGetting rid of it entirely is likely to be a fight.20:07
efriededmondsw At the moment, local chicken-scratch.  Was considering chucking it into an etherpad.  You interested in reading it?20:07
*** apearson_ has quit IRC20:27
*** apearson has joined #openstack-powervm20:30
*** edmondsw_ has joined #openstack-powervm20:47
*** apearson has quit IRC20:47
*** edmondsw has quit IRC20:49
*** apearson has joined #openstack-powervm21:00
*** thorst has quit IRC21:04
*** edmondsw_ has quit IRC21:04
*** k0da has quit IRC21:38
*** k0da has joined #openstack-powervm21:41
*** svenkat has quit IRC21:47
miltonmefried, part of the mismatch comes from linux only giving 16 bits to pci domain but phyp using 32 and 64 bit identifiers for their api22:02
miltonmthat, and phyp only presents an abstract machine to any partition22:03
*** apearson has quit IRC22:26
*** thorst has joined #openstack-powervm22:35
*** thorst has quit IRC22:53
efriedmiltonm Good to know.23:15
*** thorst has joined #openstack-powervm23:33
*** thorst has quit IRC23:33
*** k0da has quit IRC23:34
*** thorst has joined #openstack-powervm23:59

Generated by 2.15.3 by Marius Gedminas - find it at!