[01:00:27] *** Quits: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) (Read error: Connection reset by peer) [01:14:29] *** Joins: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) [01:31:32] *** Quits: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) (Read error: Connection reset by peer) [01:36:24] *** Joins: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) [04:38:21] *** Quits: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) (Read error: Connection reset by peer) [04:42:52] *** Joins: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) [06:14:43] *** Quits: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) (Quit: My Mac Pro has gone to sleep. ZZZzzz…) [06:32:12] *** Joins: lhodev (~Adium@inet-hqmc01-o.oracle.com) [08:06:01] bwalker: this patch looks good but I posted a comment on it: https://review.gerrithub.io/#/c/375204/ [09:20:17] jimharris: is it possible to add a simple smoke test for the virtio bdev before we submit Dariusz's patches? [09:21:13] we can ask him to prioritize that [09:21:54] it's probably going to be easiest to do a smoke test with vhost-user mode (instead of PCI mode in a guest VM) [09:22:06] we'll add tests for both eventually [09:22:19] but we'll need at least one test system set up with 1GB hugepages to make that happen [09:28:03] hmm, ok [09:28:30] do we need 1 GB hugepages for both vhost-user and PCI? [09:28:43] no - just vhost-user [09:29:11] for PCI, QEMU allocates the memory for the guest, and won't do the DPDK thing where it mmaps each 2MB huge page individually [09:38:02] jimharris: I'll add in some stronger pool checks for sure - just as soon as I stop moving the pool code around [09:38:24] *** Parts: lhodev (~Adium@inet-hqmc01-o.oracle.com) () [09:42:49] *** Joins: dstojacx (~dstojacx@192.55.54.36) [09:43:48] *** Joins: klateck (~klateck@192.55.54.36) [09:44:54] *** Joins: lgalkax (lgalkax@nat/intel/x-pzkaqzihqkpyaszo) [09:45:19] *** Joins: pniedzwx (pniedzwx@nat/intel/x-fzffxfkqtmomoifd) [09:50:18] cool [09:50:32] *** ChanServ sets mode: +o bwalker [10:00:12] drv: can you take another look at hailiang's pkgdep patch? is there any reason we can't just get this script to use /bin/sh on both linux and freebsd? [10:01:11] i'm also wondering if we just skip the makefile altogether and just call the script directly - i think you had already suggested this previously [10:04:06] I think having a script named "install_deps.sh" in scripts is reasonable [10:04:35] I'm not opposed to the makefile target in addition to that either, but I think the README should tell users to run the script directly [10:04:45] because you need root to install packages, but not for the rest of the build [10:05:29] I don't think having a separate makefile just for this in the scripts directory is the right way to go [10:05:40] i don't see any reason why that target couldn't go in the top level Makefile [10:05:50] oh a makefile in scripts - definitely not [10:05:56] just add a new target to the top level makefile [10:15:06] as far as making it compatible with /bin/sh rather than using bash, I'm all for it [10:15:11] (we just need to test it against something that isn't bash) [10:15:39] and I'd prefer to just have it as a separate script rather than a Makefile target unless there's some standard name for this [10:27:40] jimharris: please take a look at my comment on https://review.gerrithub.io/#/c/374399/ - I'm not sure if this is in the hot path for I/O [10:28:22] oh - yes, you're right [13:39:01] bwalker: the last patch in your vtophys series failed due to trying to include linux/version.h on freebsd [13:40:13] jimharris: do you remember why we had the rte_mem_virt2phy() check first in the vtophys lookup? [13:40:33] I seem to remember that it was required for vhost, but it passes anyway when bwalker changes it in https://review.gerrithub.io/#/c/375208/ [13:41:05] yeah I'm fixing it now [13:41:24] as far as I can tell, what I changed it to is the right thing to do [13:41:44] I agree, but I thought vhost had some kind of problem when we did it that way [13:42:20] after we get this one in I'm going to do one more patch [13:42:32] that handles the case where we can't get physical addresses because we're not running as root [13:42:53] then we should be able to run vhost as an unprivileged user, assuming we grant the user permission to the domain sockets and such [13:44:19] I don't remember offhand - I'm going back through the history on that file now [13:46:16] 7336c00 put that code snippet before the memseg loop, but I don't see any reason nor explanation in the commit message why it couldn't have been put afterwards [13:46:53] anyway, if it works, I'm happy to merge it this way [13:47:23] would there be any cases where a vhost guest VM page is in the DPDK memseg table? [13:47:29] no [13:48:38] not sure how critical it is to get these patches in but might be worth asking pawel to review them (i think he's on vaca until monday though) [13:48:58] these aren't for anything - I was just reviewing vhost stuff and trying to comprehend what was going on here [13:49:03] and saw an opportunity to simplify [13:49:16] and to also make it work as an unprivileged user [13:49:19] so they can wait [13:49:28] drv wants me to fix some comment format stuff anyway [14:33:45] CI system has been busy today [17:12:12] having visions of consolidating most of the split, gpt and error bdev modules [17:12:55] most of the logic in these modules are identical - and for these generic bdev changes to better handle resets, I'm having to do the same surgery on each module [17:15:50] hmm, I'm mostly OK with merging split and GPT (or just removing split), but error seems different enough that we probably shouldn't mix it in with GPT [17:16:11] well they will still be three distinct modules [17:16:37] at the beginning (current state) there's not a lot of overlap between error and split/gpt [17:17:32] but in the new code, resubmitting an I/O requires discrete channels in these modules (which aren't there today), a switch statement to do things like translate bdev_io type to spdk_bdev_readv/writev/etc. [17:22:44] ah, yeah, that's probably fine [18:58:12] *** Joins: Ray__ (2a784a5d@gateway/web/freenode/ip.42.120.74.93) [18:59:46] *** Quits: Ray__ (2a784a5d@gateway/web/freenode/ip.42.120.74.93) (Client Quit)