On 07/11/2017 01:57 PM, Josh Boyer wrote: > The fact that i686 kernels continue to work in general is basically luck. I'm going to clarify this point and say that it's less luck and more inertia and if nobody cares enough sometimes it just stops. Some examples from recent history: i686 builds were broken for compilation in at least two builds around 4.12-rc6. At -rc6 things are supposed to be fairly stable in the kernel cycle. Breakages certainly happen from time to time but it's pretty telling for this to happen so late in the cycle. OOM Killing and memory management seems to need constant tuning for 32-bit machines. Most of the developers doing serious work on memory management are focused on 64-bit use cases and may not take into account 32-bit things such as highmem. The MM community is usually very responsive to bug reports and will give fixes for actual bugs but sometimes issues are more tuning problems. 32-bit came up in the recent stack clash discussion. 32-bit Java was doing some 'unique' handling of stack (https://marc.info/?l=linux-kernel&m=149925724902166) and there was a lot of discussion about how to work around this (https://marc.info/?l=linux-kernel&m=149970558924339). My overall concern with continuing to ship i686 kernels is any participants need to be active in advocating upstream. The only reason certain drivers or hardware are supported upstream is because of one or two people who stubbornly refuse to let it go. i686 is nowhere close to actually being deleted but fixing reported bugs in bugzilla is not a sustainable solution without at least some participation upstream to keep 32-bit healthy as Fedora relies very very heavily on upstream. Thanks, Laura _______________________________________________ devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx
