In my P4 branch: //depot/user/attilio/attilio_smpng/... you can find a sx locks rewriting using the optimized semantic of rwlocks; in the end this might result in a valuable performance improvement. Some hints about it: - new sx locks alredy support inlined s*lock operations and try* operations; (they have a fully functional support) - new sx locks doesn't have support for adaptive spinning yet; this is beacause the code is under revision even for mutex/rwlock. - we could allow a sharers tracking in debugging mode, at least, in order to detect eventual recursion in slock operation - currently, sx locks mantain the exclusive holder tracking even if this is not really necessary (we can get rid of it). In this moment a strong phase of test is *very* welcome, so please, for every people having a p4 account, dowload the kernel and try to put it under stress (at this purpose I cc'ed, in particular, kris_at_, pho_at_ and jhb_at_ in order to have tests, revisions, etc. etc.) Note1: the branch contains other diffs respect CURRENT. Note2: due to same strange struct sx members accesses, I had to disable XFS locking primitives in order to not hurt the current work. This is possibly buggy... Thanks, Attilio -- Peace can only be achieved by understanding - A. EinsteinReceived on Wed Oct 18 2006 - 20:18:30 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:01 UTC