Conversation

Thorsten Leemhuis (acct. 1/4)

The the "Lazy preemption" model (CONFIG_PREEMPT_LAZY=y) was merged for 6.13.

It optimizes fair-class preemption by delaying preemption requests to the tick boundary, while working as full preemption for RR/FIFO/DEADLINE classes.

In the end it hopefully will thus allow distributions to ship one äkernel that can serve both regular needs as well as RT needs.

https://git.kernel.org/torvalds/c/3f020399e4f1c690ce87b4c472f75b1fc89e07d5

For details, see also https://lore.kernel.org/all/20241007074609.447006177@infradead.org/t/#u and https://lwn.net/Articles/945422/

1
0
1
@kernellogger if you mean RT as in PREEMPT_RT then I doubt that, that's way more fundamentally different for this mechanism.
2
0
0
@vbabka @kernellogger I do RT_LAZY, that is I lazily retoot stuff all the time, if that helps
0
1
2

@vbabka yes, sorry, you are right, removed that, mixed that up with something else; thx for pointing this out!

0
0
1