On 6/19/23 12:30 PM, Greg Kroah-Hartman wrote:
From: Stephen Hemminger stephen@networkplumber.org
commit 1202cdd665315c525b5237e96e0bedc76d7e754f upstream.
DECnet is an obsolete network protocol that receives more attention from kernel janitors than users. It belongs in computer protocol history museum not in Linux kernel.
[...]
May I ask, how and why this patch made it into the stable kernels?
Did this patch "fix a real bug that bothers people?", is Documentation/process/stable-kernel-rules.rs obsolete or something else? I apologize, if I overlooked something obvious.
Background:
No, we don't use DECNET since 25 years or so. But still any change of kconfig patterns bothers us.
We automatically build each released kernel and our config evolves automatically following a `cp config-mpi .config && make olddefconfig && make savedefconfig && cp defconfig config-mpi && git commit -m"Update for new kernel version" config-mpi` pattern.
Historically, the changes of savedefconfig in a stable series were zero most of the time. Only when we start a new series with a -rc1 kernel, we had to fix things because, for example, we lost one config because of reorganization of dependencies or a module went from tristate to binary as CONFIG_UNIX did or an option was renamed as CONFIG_AUTOFS4_FS was. But we only needed to care for that seldomly and in expected moments, when we actually want to move our fleet to a new series and manually compare and test the config.
These changes complicate continuous upgrades and add distractions during reviews and investigations of unintended config changes. And of course, they increase the risk that you lose a feature you depend on during a upgrade from one release of a stable series to another. That's fine, if the changes are really justified by real world problems and maybe these are. But I failed to find the explicit reasoning here.
Another example of such a change in stable for 5.15 might be the addition of CONFIG_MICROCODE_LATE_LOADING with default=y [1] which removed the feature by default, which had to be detected and undone.
I just wanted to note, that backporting "Cleanup"-Patches to stable has cost for the users, too, even if it was true, that nobody needs the removed feature and that there are no unforeseen side effects or other bugs in the patches. It produces manual workload. So I hope, the "only real bug fixes"-policy is not generally obsoleted?
Best
Donald
[1]: https://lore.kernel.org/stable/20230307165910.899232186@linuxfoundation.org/