On Thu, 31 Aug 2023 at 14:22, Andreas Hasenack <2028...@bugs.launchpad.net> wrote: > > I suppose it's possible that the particular dkms module that failed to > build is essential for booting the system, so the new approach that > hides this build failure could render the system unbootable without the > user realizing it. A bit less catastrophic, but still very serious, > would be for the machine to lose network access (wifi driver), which > could pose a chicken an egg problem (need network to fix the problem).
I'm not sure that is possible. Because it means it was not possible to boot the installer, or install the system. Because for example when one assumes secureboot by default it is not possible to enroll a Mok key without a reboot, and meaning having a successful install that at least can boot (but might not have a weird wifi). We haven't yet had any cases of drivers being dropped by a kernel, that transition to be a 3rd party dkms. Or ability to complete install with dkms module that is essential for boot (i.e. disk driver for disk install, or network driver for network boot install). To the point of, I am happy to assert that if dkms is required for boot, the image for such a system was pre-built externally. And in such cases, it should be possible to pre-built again. > > So what we have here is a balance between a system with an interrupted > release-upgrade, requiring a lot of apt-foo to fix, or a system where > dkms rebuilds might have failed, but otherwise upgraded successfully. > Both cases could result in an unbootable system, but I tend to agree Note quite. The completed upgrade system with missing dkms is bootable with new kernel. Or also bootable with the old kernel (and old dkms module) those are not purged and remain on disk. So one can do successful boot using old kernel (and thus still have access to weird wlan) to figure out how to update their 3rd party dkms to be compatible with a new kernel. So it is a massive improvement on the situation. > that the first one (interrupted release-upgrade) is more likely to be > much harder to recover (been there). > This is very true and the core point of this SRU. > I would like some sort of commitment on the follow-up work for the > release upgrader to check the status of dkms builds after the upgrade. > Can we get an LP bug for it please? > That's this bug report that was opened in May and still being discussed https://bugs.launchpad.net/ubuntu/+source/ubuntu-release-upgrader/+bug/2020406 -- okurrr, Dimitri -- You received this bug notification because you are a member of Kernel Packages, which is subscribed to dkms in Ubuntu. https://bugs.launchpad.net/bugs/2028366 Title: Kernel header installation fails for incompatible DKMS modules Status in dkms package in Ubuntu: Fix Released Status in dkms source package in Jammy: Confirmed Status in dkms source package in Lunar: Fix Released Status in dkms source package in Mantic: Fix Released Bug description: [ Impact ] If a new kernel is installed, all installed DKMS modules are built for that new kernel. There might be incompatible modules that won't compile for the new kernel which results in a kernel header package installation failure. That's bad and not really correct, the incompatible DKMS module is the problem and not the new kernel. In this case, DKMS module build failures should be ignored so that the kernel installation completes. This is especially acute during release-upgrades, as dkms modules are upgraded out of order, and major kernel version are upgraded out of order. Majority of the time there is a new dkms available, which should attempt build & load. However, many modules are often remain broken, no longer needed, or need user to fetch updated versions themselves. [ Test Plan ] * Install jammy * Add module that support v5.15 kernel, but fails to compile with any newer kernels (one can find examples of such dkms modules in the archive, or out of the archive) * Perform release upgrade with patched dkms pre-installed * Release upgrade should succeed, despite unable to compile all dkms modules [ Where problems could occur ] * This is an improvement to the current situation of aborting release upgrade half way through. It doesn't quite resolve the UX to notify the user which dkms modules did not manage to compile, or to ask user to uninstall or to update them. Further UX / hooks might be needed in the release upgrade to complete the story of asking the user what they want to do with regressed dkms modules. [ Other Info ] * See lots and lots of upgrade bugs, failing on dkms module installation To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/dkms/+bug/2028366/+subscriptions -- Mailing list: https://launchpad.net/~kernel-packages Post to : kernel-packages@lists.launchpad.net Unsubscribe : https://launchpad.net/~kernel-packages More help : https://help.launchpad.net/ListHelp
