Ahead of the planned Linux 6.19 stable kernel release tomorrow, there have been some last-minute fixes submitted for the scheduler code, including for performance regressions.
Linux 6.19 has been bumpy this cycle on the scheduler front. At the start of the Linux 6.19 cycle I pointed out some Linux 6.19 performance regressions bisected back to the scheduler code. In taking care at least some of those, in late January the NEXT_BUDDY feature was disabled due to performance regressions in Linux 6.19. Now ahead of tomorrow’s v6.19 stable tag are some additional scheduler regression work.
The fixes sent out today for Linux 6.19 pertain to the scheduler mm/cid code rewrite in 6.19:
“Miscellaneous MMCID fixes to address bugs and performance regressions in the recent rewrite of the SCHED_MM_CID management code.”
The patch to optimize transitional CIDs when scheduling out is a notable one on its performance impact:
“With a thread pool benchmark which is configured to cross the mode switch boundaries frequently this reduces the number of bitmap operations by about 30% and increases the fastpath utilization in the low single digit percentage range.”
This patch to protect transitions on weakly ordered systems fixes hard lock-ups.
This patch fixes a live lock caught by the BPF CI.
These MM_CID regression fixes were sent out this morning as fixes for Linux 6.19 final. There’s a chance Linus Torvalds could delay them since they are rather substantive a day before the stable release, but we’ll see.
