[lttng-dev] Question about URCU implementation: single global RCU "lock" VS per object RCU "locks"

[Eugene Ivanov]

Hi,

All RCU flavors seem to use single common global counter, thread list, etc
(to be precisely it seems that rcu_*lock is a single per-process "lock").
So if I use multiple data structures, i.e. rculfhashes, I protect them via
single shared RCU "lock". In case of QSBR it looks problematic: reclamation
of objects from object A depends on QS of threads which use another
objects. What is the reason of such design? Especially in comparison with
per-object RCU primitives. Any performance implications?

-- 
Eugene
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lttng.org/pipermail/lttng-dev/attachments/20141128/fe0f62bb/attachment.html>