* [RFC] 0/5 rcu lock update
@ 2004-05-25 5:35 Manfred Spraul
0 siblings, 0 replies; only message in thread
From: Manfred Spraul @ 2004-05-25 5:35 UTC (permalink / raw)
To: linux-kernel, lse-tech
Hi,
I tried to tackle the rcu scalability problem that Jack Steiner
noticed:
I found three changes that reduce the cache line trashing with
rcu_ctrlblk:
- Use generation numbers instead of the rcu_cpu_mask bitmap for
polling in rcu_pending. (patch 1)
- Use a sequence lock to check if a new batch must be started. Right
now each cpu acquires the global spinlock and usually notices that
the batch is already running. (patch 2)
- Hide implementation details - only fields that are accessed by
rcu_pending need to be visible in <linux/rcupdate.h> (patch 3)
- Preparation for hierarchical bitmap : hide locking (patch 4)
- Add a hierarchical bitmap (patch 5)
The first three patches are IMHO steps in the right direction,
suitable for inclusion. The last two patches are proof of concept
changes.
--
Manfred
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2004-05-25 5:41 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2004-05-25 5:35 [RFC] 0/5 rcu lock update Manfred Spraul
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).