Skip to content

RFC sched: Reduce runqueue lock contention -v3

Peter Zijlstra linux-arch linux-ide linux-scsi lkml 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 RFC Latest17 patches 31 repliesIn review

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/17sched: Always provide p->on_cpuinclude/linux/sched.h kernel/sched.c+30 -200 replies
2/17mutex: Use p->on_cpu for the adaptive spininclude/linux/mutex.h include/linux/sched.h kernel/mutex-debug.c +4 more+39 -560 replies
3/17sched: Change the ttwu success detailskernel/sched.c+8 -100 replies
4/17sched: Clean up ttwu statskernel/sched.c+32 -350 replies
5/17x86: Optimize arch_spin_unlock_wait()arch/x86/include/asm/spinlock.h+23 -312 replies
6/17sched: Provide p->on_rqinclude/linux/sched.h kernel/sched.c kernel/sched_debug.c +2 more+26 -251 reply
7/17sched: Serialize p->cpus_allowed and ttwu() using p->pi_lockkernel/sched.c+16 -212 replies
8/17sched: Drop the rq argument to sched_class::select_task_rq()include/linux/sched.h kernel/sched.c kernel/sched_fair.c +3 more+27 -3313 replies
9/17sched: Remove rq argument to sched_class::task_waking()include/linux/sched.h kernel/sched.c kernel/sched_fair.c+11 -50 replies
10/17sched: Add TASK_WAKING to task_rq_lockkernel/sched.c+7 -70 replies
11/17sched: Delay task_contributes_to_load()include/linux/sched.h kernel/sched.c+5 -120 replies
12/17sched: Also serialize ttwu_local() with p->pi_lockkernel/sched.c+17 -112 replies
13/17sched: Remove rq->lock from the first half of ttwu()kernel/sched.c kernel/sched_fair.c+30 -300 replies
14/17sched: Remove rq argument to ttwu_stat()kernel/sched.c+5 -32 replies
15/17sched: Rename ttwu_post_activationkernel/sched.c+6 -30 replies
16/17sched: Move the second half of ttwu() to the remote cpuarch/x86/kernel/smp.c include/linux/sched.h kernel/sched.c +1 more+114 -346 replies
17/17sched: Sort hotplug vs ttwu queueingkernel/sched.c+18 -22 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.

Read the v1 thread

Cover letter [RFC][PATCH 00/17] sched: Reduce runqueue lock contention -v3

This is the latest incarnation of my scary ttwu patch, its split up into tiny little pieces because at some point while working on the feedback from the last posting it all stopped working and tiny changes was the only way to find out wth I done wrong (I never did find out though, when I completely rewrote everything in tiny chunks it all magically worked again -- so I guess tglx is chuckling now).

Anyway, it still need a few more loose ends fixed, esp on 32bit there is the whole unlocked remote min_vruntime access where we can read partial updates and screw the timeline like problem. And I suppose there's a few more XXXs in there where I know things are needing TLC.

Also, I bet the eagle eyed lot Cc'ed will spot more holes, as all this is scary stuff :-)

Anyway, lots of thanks to Frank, Oleg, Ingo and others who commented on the last posting.

I guess most everybody is out having x-mas celebrations (and I will too not too long from now), so I expect replies to this won't happen until the new year, but who knows, maybe in between eating and drinking too much we all get an urge to stare at code.

Happy Holidays!