| [13:28] <sforshee> LocutusOfBorg: reverted the patch, thanks for the heads up | |
| [13:54] <LocutusOfBorg> thanks sforshee | |
| [13:57] <LocutusOfBorg> sforshee, just to understand, in case it fails, will some automatic test notify me? | |
| [13:57] <LocutusOfBorg> I double checked and virtualbox 6.2.14 in groovy release doesn't call the two functions anymore | |
| [13:57] <LocutusOfBorg> but I don't want to break the world :D | |
| [13:58] <LocutusOfBorg> (Debian never applied the patch and sid has 5.8.7 or whatever) | |
| [14:14] <sforshee> LocutusOfBorg: the automated testing will check that the dkms packages still build, I don't think you will be automatically notifiied but it will block the kernel from promoting, so we will see it and let you know | |
| [15:14] <LocutusOfBorg> thanks! I'm pretty sure it will work | |
| [15:14] <LocutusOfBorg> but better safe than sorry | |
| [18:20] <tyhicks> hello! I'm thinking back, trying to remember the packages that typically needed to be updated when a new enablement kernel is introduced into a given Ubuntu release (or a new upstream kernel is introduced into the dev release) | |
| [18:21] <tyhicks> the DKMS modules obviously are the majority of the work and, for my purposes, I'm just lumping them all into one bucket | |
| [18:21] <tyhicks> there's the stuff in linux-tools-* that gets updated lockstep with the kernel (such as perf, usbip, etc.) | |
| [18:22] <tyhicks> what I'm trying to remember are the additional packages spread thruoghout the archive | |
| [18:23] <tyhicks> I came up with crash, kexec-tools, makedumpfile, ktap, and kpatch | |
| [18:23] <tyhicks> are there any others that commonly need to be updated when bumping from 5.x to 5.y? | |
| [18:25] <tyhicks> sforshee, cascardo, mhcerri: ^ you folks would have more recent memories of this work than I do | |
| [18:25] <sforshee> tyhicks: the big one the comes to mind is zfs | |
| [18:26] <tyhicks> I had forgotten about that one | |
| [18:26] <tyhicks> I'm primarily trying to inventory userspace code that needs to move forward with new kernel releases | |
| [18:27] <sforshee> it is mostly dkms stuff | |
| [18:27] <tyhicks> sforshee: can you think of anything that broke that's outside of the DKMS module and kernel tool categories? | |
| [18:28] <sforshee> tyhicks: I was about to say iproute2, usually not because it broke but to support new things | |
| [18:28] <sforshee> otherwise I can't think of anything which we consistently need to update | |
| [18:28] <tyhicks> sforshee: I was curious about that one, too. I'll go through the iproute2 changelog. | |
| [18:28] <sforshee> just things which show up with test failures from time to time | |
| [18:29] <tyhicks> sforshee: yeah, those are the ones that I don't know how to identify very well | |
| [18:29] <sforshee> they are hard to identify, and very often if they break it's because the kernel regressed somehow | |
| [18:30] <sforshee> you can look at the adt matrix to see the ones we are testing against - https://people.canonical.com/~kernel/status/adt-matrix/groovy-linux-meta.html | |
| [18:30] <sashal> sforshee: could you guesstimate how many of these break in your average kernel upgrade? | |
| [18:32] <sforshee> sashal: we probably see 10 or so dkms packages which need updated each time, zfs and nvidia pretty much always needs updates. Aside from those it's usually just a few | |
| [18:32] <sforshee> most commonly glibc and systemd | |
| [18:32] <sforshee> oh and trace things like lttng and sysdig, though that's ususally the dkms parts | |
| [18:37] <sarnold> tyhicks: seccomp syscall rules? | |
| [18:38] <tyhicks> sarnold: aha! that's right, we've had to bump/patch libseccomp to filter new syscalls | |
| [18:39] <sforshee> sarnold: that's not currently one of the testing requirements for our kernels. Should it be? | |
| [18:39] <sarnold> sforshee: jdstrand would be more familiar, but it probably should be -- seccomp is an important part of snapd story | |
| [18:40] <sarnold> tyhicks: I believe a cronjob recently emailed us to let us know about new syscalls in a kernel :) hooray automation | |
| [18:40] <sforshee> sarnold: perhaps it is already covered by qa-regression-tests or snapd testing | |
| [18:41] <sforshee> I don't recall ever seeing any kind of test failures related to it though | |
| [18:41] <tyhicks> sforshee: it is covered by both, I believe | |
| [18:41] <sforshee> ok, good | |
| [18:42] <sarnold> ah good | |
| [18:42] <tyhicks> sforshee: it wouldn't have been a failure - it would have been a gap in being able to write a seccomp filter that specifies one of the new syscalls | |
| [18:43] <sforshee> tyhicks: so if I'm testing e.g. 5.9 right now there wouldn't be anything in the test results that would prompt me to know that we needed a libseccomp update | |
| [18:45] <tyhicks> sforshee: I don't think so but let me take a quick look at the autopkgtests because I know Jamie added a few in there but I don't remember if they detect new syscalls | |
| [18:46] <sforshee> tyhicks: well libseccomp dep8 tests don't block our kernels, unless they are getting run as part of q-r-t | |
| [18:47] <sforshee> we do have some sort of seccomp test in our autotests stuff, I'm not sure what it runs though | |
| [18:47] <tyhicks> sforshee: ah, then they probably should block the kernels. the tests are quick and simple. | |
| [18:48] <tyhicks> sforshee: I don't see anything in there that would fail on new syscalls | |
| [18:49] <sforshee> tyhicks: our seccomp test clones the upstream libseccomp source and runs 'make check', essentially | |
| [18:49] <sforshee> doesn't really tell us about the state of the ubuntu package | |
| [18:50] <tyhicks> sforshee: I'm not sure if that would build a local (to the git repo) libseccomp and test with that or if it would use the system libseccomp - I'd think the former | |
| [18:50] <tyhicks> (fwiw, the upstream libseccomp tests are pretty good0 | |
| [18:50] <sforshee> or both :-) | |
| [18:52] <sforshee> tyhicks: I mean it might be nice to test both, but yeah I don't know which ends up being tested either | |
| [22:50] <Kon> Did something change with the mainline kernel PPA with the July 15 update? I have a utility to check for updates that stopped returning results after that time | |
| [22:50] <Kon> Or rather, it still finds all kernels uploaded prior to July 15, but not after | |
| [22:57] <Kon> I could just download the debs, but I am curious | |
| [23:04] <Kon> Oh, the patches are no longer included in the repo comments | |
| [23:04] <Kon> Nor the BUILD.LOG | |