On 2023-07-05 17:21:04 +0200, Andreas Beckmann wrote: > On 05/07/2023 16.17, Vincent Lefevre wrote: > > I have autoinstall_all_kernels="1" in /etc/dkms/framework.conf > > so that all kernel modules are rebuilt when some dependency > > changes. > > Does that at least cause a rebuild if the linux-headers-* package gets > updated? IIRC as long as KVERS does not change, dkms does not rebuild > modules on header change. > > More general: if linux-headers-* gets updated but the ABI does not change > (and the ABI is not 0), does dkms have to rebuild the modules? Why?
I'm quoting what you said in bug 1040178: | This needs to be fixed before linux 6.3.0-2-* can migrate to testing, | otherwise it will break dkms module building for everyone still having | linux-headers-6.3.0-1-* installed (which is probably for the currently | running kernel). Though I track unstable instead of testing, the issue is basically the same (I'm concerned now instead of later, if nothing is changed). What you said is that the upgrade to 6.3.0-2-* is expected to break dkms module building for everyone still having linux-headers-6.3.0-1-* installed (which is exactly my case), and 6.3.0-1 was indeed the currently running kernel in my case. This means that you expect the upgrade to 6.3.0-2-* to rebuild the kernel modules for 6.3.0-1 (otherwise, i.e. if the 6.3.0-1 modules are not rebuilt, there are no bugs at all). But this is not what happened. And it doesn't seem that autoinstall_all_kernels actually matters. > from dkms(8): > > .B $autoinstall_all_kernels > > Used by the common postinst for DKMS modules. It controls if the build > > should be done for all installed kernels or only for the current and > > latest installed kernel. It has no command > > line equivalent. > > So that flag is not doing what you expected. And I think on Debian it is > ignored and we always build for all kernels. But only on new header > installation or on dkms module update. Not on header update. Yes, it is probably a different matter. I now remember that autoinstall_all_kernels is not sufficient as the ABI may change without being announced. See https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=856355#54 This is crap (I had written a workaround in /etc/kernel/postinst.d, and I've just noticed that it got broken after an upgrade 3 years ago), but this is unrelated to the above issue. -- Vincent Lefèvre <vinc...@vinc17.net> - Web: <https://www.vinc17.net/> 100% accessible validated (X)HTML - Blog: <https://www.vinc17.net/blog/> Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)