"giving up on N buffers" revisited

[Don Lewis]

I think there is another cause for the "giving up on N buffers" problems
besides the fsck'ing the root partition in single user mode problem that
I fixed and the mounted ext2fs file system problem that has not been
fixed.

I occasionally run into this problem if I reboot too quickly after doing
	make installworld
	mergemaster

My file systems are mounted with soft-updates enabled.

My suspicion is that the syncer thread is getting disabled before its
worklist is empty and the remaining dirty bufs can't be immediately
flushed in the kernel shutdown code due to dependencies.  Mergemaster
would have a tendency to trigger this because it nukes its temproot tree
just before it exits.  Waiting a little while before rebooting gives the
syncer time to do some cleanup before it gets shut down.