or does the new inode.c stuff solve this problem?
plus:
pc5829:~/trace/ktrace/samples$ grep get_empty_filp execve.out
Trace: c0127da8 <get_empty_filp+10/e4> (50.33)
Trace: c0127da8 <get_empty_filp+10/e4> (41.72)
thats 100 usecs more overhead per exec() ... get_empty_filp() sucks hard.
-- mingo