Hi Linus and All,

This patch breaks down the three global IPC locks into one lock per IPC
ID. 

In current implementation, the operations on any IPC semaphores are
synchronized by one single IPC semaphore lock.  Changing the IPC locks
from one lock per IPC resource type into one lock per IPC ID makes sense
to me.  By doing so could reduce the possible lock contention in some
applications where the IPC resources are heavily used.

Test results from the LMbench Pipe and IPC latency test shows this patch 
improves the performance of those functions from 1% to 9%.  

Patch applies to 2.5.30 kernel. Please consider it.

--
Mingming Cao