tmpfs: avoid the vm object lock in tmpfs_pager_getvp in the common case

[mjg]

Summary

This saves about 8% of wait time on vm object lock when doing -j 104 incremental kernel-only tinderbox run.

[kib]

sys/fs/tmpfs/tmpfs_subr.c

212

What for is this acquire fence?

247

Either this check is needed in the tmpfs_pager_getvp_smp(), or it is not needed there.

In fact, I am not sure that we clean up enough of the un_pager to guarantee that swp_tmpfs is always NULL for newly allocated object, so I think that OBJT_SWAP check is still needed.