The thread_get(Readable|Writeable)Pointer interfaces were designed so that the implementations could potentially be zero-copy, but right now they both copy data from the plugin into a Shadow-allocated buffer. Worse, this copy involves syscalls, and for the threadpreload involves context switches between Shadow and the plugin.

In most or all cases, we should be able to arrange for plugin memory that Shadow needs to access to be directly mmap'd in shadow, such that these interfaces should only need to do some arithmetic to translate a plugin-pointer to a shadow-pointer.