Skip to content

rcu inline, expedited, ->completed cleanups

Paul E. McKenney lkml 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 Latest4 patches 16 repliesIn review

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/4rcu: remove inline from forward-referenced functionskernel/rcutree.h kernel/rcutree_plugin.h+3 -313 replies
2/4rcu: enable synchronize_sched_expedited() fastpathkernel/sched.c+1 -01 reply
3/4rcu: rename dynticks_completed to completed_fqskernel/rcutree.c kernel/rcutree.h+4 -41 reply
4/4rcu: simplify association of quiescent states with grace periodskernel/rcutree.c kernel/rcutree_plugin.h+3 -31 reply

Reviews trailers per patch

1 of 4 patches has a Reviewed-by or Acked-by.

PatchJosh Triplett
1/4A-b
2/4
3/4
4/4

Trailers given in replies are filled; those carried in the patches from earlier versions are outlined with dashes.

Read the v1 thread

Cover letter [PATCH tip/core/rcu 0/4] rcu inline, expedited, ->completed cleanups

This patchset contains a few fixes from testing and review of treercu. The review was spurred by the earlier bugs.

1. Some compilers don't like forward references to inline functions.
Make the compiler happy by removing "inline".

2. Although synchronize_sched_expedited() was implemented with a
fastpath to efficiently handle multiple concurrent calls, the counter increment that enabled this fastpath was omitted. Add it back in. (And yes, this does make a big difference in performance and scalability for multiple concurrent calls to synchronize_sched_expedited(), not that such calls should happen all that often. Still, no point in bringing the machine to its knees unnecessarily.)

3. The field rsp->dynticks_completed is now used whether or not

	dynticks is enabled.  Rename it to rsp->completed_fqs to better
	document its new role.

4. Much of the confusion leading to the bugs provoked by testing
with long RCU grace periods was due to the method used to determine which grace period a given context-switch quiescent state is to be applied to. This change provides a much more straightforward detection method for these quiescent states. The quiescent states induced by force_quiescent_state() still use the old method, and a separate patch for them is in the works.