Over the course of a long job with lots of forks like build.sh, small but
frequent contention on fstrans_lock can start to add up. This cures it by
maintaining the global list of fstrans_lwp_info in a pool cache constructor
and destructor, much the same pattern as is done for struct file and struct
vm_amap.
fstrans_mount_lock is taken in the same path but less work is done with that
lock held, so it doesn't really show up in the profile for me.
http://www.netbsd.org/~ad/2020/fstrans_lock.diff
Any comments?
Thanks,
Andrew
