Linux Scheduler Build Improvements From "Fast Kernel Headers" Queued, FKH v3 Posted
Published at the start of the new year was 2.3k patches providing "fast kernel headers" as a major speed-up to Linux kernel build times and addressing the dependency hell among all the header files in the Linux kernel source tree. It will likely take some time for that massive patch series to work its way to mainline in full, but at least for Linux 5.18 already the patches touching the kernel's scheduler area are ready to land.
Thanks to Linux 5.17 getting pushed back a week and therefore allowing extra time for material to queue in "-next" branches ahead of the Linux 5.18 merge window, the fast kernel header patches affecting the scheduler area were queued up.
Merged to sched/core this morning were the "fast header" patches affecting it. Several hundred lines of code were reworked just for the "sched/" code as part of this massive rework to the Linux kernel's header file structuring.
As for the time needed to build just the scheduler portion of the kernel, these patches dropped the CPU time used by 60.9%. The wall clock time dropped by 3.9%. Though with the scheduler code being a small fraction of the overall kernel size, this is just a tiny fraction of the overall build time. Once the fast header series is mainlined in larger portion is when users and CI/CD farms will likely begin noticing the fruits of this long in-development work.
Meanwhile, minutes ago, Ingo Molnar has just published his v3 patches of the "fast kernel headers". With this series is still over 2,300 patches and affecting 180k lines of new code and removing 74.5k lines of code.
Thanks to Linux 5.17 getting pushed back a week and therefore allowing extra time for material to queue in "-next" branches ahead of the Linux 5.18 merge window, the fast kernel header patches affecting the scheduler area were queued up.
Merged to sched/core this morning were the "fast header" patches affecting it. Several hundred lines of code were reworked just for the "sched/" code as part of this massive rework to the Linux kernel's header file structuring.
As for the time needed to build just the scheduler portion of the kernel, these patches dropped the CPU time used by 60.9%. The wall clock time dropped by 3.9%. Though with the scheduler code being a small fraction of the overall kernel size, this is just a tiny fraction of the overall build time. Once the fast header series is mainlined in larger portion is when users and CI/CD farms will likely begin noticing the fruits of this long in-development work.
Meanwhile, minutes ago, Ingo Molnar has just published his v3 patches of the "fast kernel headers". With this series is still over 2,300 patches and affecting 180k lines of new code and removing 74.5k lines of code.
7 Comments