[00:56:21] *** Quits: ziyeyang_ (~ziyeyang@192.55.54.45) (Remote host closed the connection) [04:02:46] *** Quits: drv (daniel@oak.drv.nu) (Ping timeout: 264 seconds) [04:02:52] *** Quits: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) (Ping timeout: 240 seconds) [04:06:06] *** Joins: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) [04:10:13] *** Joins: drv (daniel@oak.drv.nu) [04:10:13] *** ChanServ sets mode: +o drv [06:43:39] *** Joins: lhodev (~Adium@inet-hqmc01-o.oracle.com) [10:38:31] *** Joins: konv81 (cf8c2b51@gateway/web/freenode/ip.207.140.43.81) [10:40:32] Hi guys, I was wondering if you have experience running with vmware nvme controller emulation [10:40:40] I'm running into an issue with queue allocation [10:41:21] hmm, we've got it working with virtualbox and qemu emulation but I'm not aware of any effort on vmware [10:41:27] what's the error you're seeing? [10:41:42] during set feature for number of queues [10:41:49] vmware returns cdw0 of completion as 0 [10:41:58] I believe it's a bug on their side [10:42:03] everyone gets that wrong [10:42:10] but get features return the correct parameter [10:42:10] in emulation, in real firmware, etc [10:44:20] we can probably provide a workaround - what's the real value? [10:44:34] 15 or 7 i believe [10:44:38] 0's based [10:45:07] i'm thinking of sending a get features after set features, in nvme_ctrlr_set_num_qpairs [10:45:56] what's the PCI vendor id, device id [10:48:24] one sec, let me check [10:52:56] konv81: it's probably best to change SPDK's start up to send a get features, then set features, and if set features returns 0 just assume we got the min(what we requested, what get features said was max) [10:53:36] if it always returns 0 - i assume they don't even set it [10:53:41] the min will always be the 0 [10:53:45] I was thinking about adding a quirk just for VMWare, but drv just talked me out of it - firmware implementors get this wrong way too often so we should just handle it [10:54:00] I mean the min of the value we requested in the set features command [10:54:06] and the value that get features returned as the max [10:54:14] that will work [10:55:58] I think we can just do Set Features, ignore the return value, Get Features, and use the Get response as the result [10:56:18] I thought get features returned the number of support queues, but it doesn't. It returns the number allocated [10:56:20] shouldn't hurt on controllers that work correctly, and it'll work around ones with broken Set Features [10:56:32] (it would still be good to get VMware to fix their implementation as well) [10:56:51] konv81: do you want to take a shot at the patch? [10:57:38] sure [11:00:00] http://www.spdk.io/development/#contributing if you aren't set up yet [11:00:38] I'll need a little time to get it setup, should have something by end of day [11:00:44] great! [11:00:46] I'm at work - have to deal with other issues as well [11:00:59] we understand all too well [12:42:35] *** Quits: lhodev (~Adium@inet-hqmc01-o.oracle.com) (Remote host closed the connection) [12:44:22] *** Joins: lhodev (~Adium@66-90-218-190.dyn.grandenetworks.net) [12:44:44] *** Quits: lhodev (~Adium@66-90-218-190.dyn.grandenetworks.net) (Client Quit) [12:46:11] *** Joins: lhodev (~Adium@inet-hqmc01-o.oracle.com) [14:53:09] bwalker: could you review the readv/writev patch and the "defer destroying bs_dev" patch right after it? [14:53:11] https://review.gerrithub.io/#/c/371885/ [14:54:28] sure [14:54:51] Hi guys, I just sent my changes for review at https://review.gerrithub.io/#/c/375234/ [15:00:01] it's running through the tests now, but it looks great [15:00:04] thanks! [15:00:20] you can see the test pool status (and your patch in the queue) at ci.spdk.io [15:00:59] Ohh thats cool! [15:01:41] sethhowe, drv: do we have a link to ci.spdk.io anywhere on the spdk.io website? I can't seem to find it [15:01:58] it's buried somewhere in http://spdk.io/development [15:02:33] I thought I pushed a patch to add it to the links at the top, but I guess we never merged that [15:02:41] let me revisit that [15:05:14] https://review.gerrithub.io/#/c/375236/ [15:05:40] oh, I think bwalker's comment last time was that the navbar was getting crowded, which is fair enough [15:07:20] maybe put it prominently somewhere on spdk.io/development [15:07:22] toward the top [15:07:43] we should also have more prominent links on that page to GitHub and GerritHub [15:07:55] instead of having them embedded in the text [15:24:16] konv81: after your patch, does everything work OK in vmware? [15:25:01] yes tried in vmware and a samsung drive [15:25:12] cool, looks good then [15:26:45] it's pretty straight forward, took a while to understand how to go about unit tests [16:31:04] *** Quits: patrickmacarthur (~smuxi@2606:4100:3880:1240:9922:701e:33a4:d270) (Ping timeout: 276 seconds) [16:33:18] *** Joins: patrickmacarthur (~smuxi@2606:4100:3880:1240:85a6:a4ef:6ebd:4ce4) [18:30:28] *** Quits: lhodev (~Adium@inet-hqmc01-o.oracle.com) (Quit: Leaving.) [19:20:49] konv81, well done!