| |||||||||||||
Big reader locksBig reader locksPosted Mar 19, 2010 16:53 UTC (Fri) by dvhart (guest, #19636)Parent article: Big reader locks
What is it about this usage case that makes the brlock a more appropriate solution than RCU? They both cater to read-mostly datastructures, and both rely on some sort of global quiescent state (the brlock version being when the write acquires al NR_CPUS locks).
(Log in to post comments)
Big reader locks Posted Mar 23, 2010 18:47 UTC (Tue) by jzbiciak (✭ supporter ✭, #5246) [Link]
I guess the major difference is that this one disallows any concurrency between readers and writers. In RCU, my understanding is that readers always get a "consistent" structure for as long as they need. A writer can start preparing a new copy while readers are still active.
In this, it sounds like a writer must wait until all readers are truly finished. When mounting/unmounting, that sounds like a useful, stronger semantic.
Or am I missing something?
Big reader locks Posted Mar 25, 2010 15:36 UTC (Thu) by dvhart (guest, #19636) [Link]
Good point, I think you nailed it.
|
|||||||||||||