Memory Architecture

Each Denali VM is given its own (virtualized) physical 32-bit address space. A VM may only access a subset of this 32-bit address space, the size and range of which is chosen by the isolation kernel when the VM is instantiated. The kernel itself is mapped into a portion of the address space that the VM cannot access; because of this, we can avoid physical TLB flushes on VM/VMM crossings.

By default, a VM cannot virtualize its own (virtualized) physical address space: in other words, by default, there is no virtual MMU. In this configuration, a VM's OS shares its address space with applications, similar to a libOS in Exokernel [20]. Exposing a single address space to each VM improves performance, by avoiding TLB flushes during context switches between applications in the same VM [31].

We have recently added support for an optional, virtual MMU to Denali; this virtual MMU looks nothing like the underlying x86-based physical MMU, but instead is modeled after a simple software-loaded TLB, similar to those of modern RISC architectures. A software-loaded TLB has the advantage that the VM itself gets to define its own page-table structure, and the software TLB interface between the VMM and the VM is substantially simpler than the more complex page table interface mandated by the x86 hardware-loaded TLB architecture.