Kernel 2.6.30.4 XFS(..?) regression (happened again)

[Christoph Hellwig]

All your traces seem to have in common that you're block on locks hold
by processed doing I/O.  Especially a lot of page locks which are
controlled by the VM and not the filesystem.  Can you try with commit
8aa7e847d834ed937a9ad37a0f2ad5b8584c1ab0 from 2.6.31-rc applied, or
better directly 2.6.3-rc6?

I would be surprised if this is something inside XFS, but running with
CONFIG_XFS_DEBUG never hurts.