Introduction

kMemvisor is a hypervisor providing software controlled memory mirroring based on hardware virtualization and static binary translation. Specifically, kMemvisor first creates a backup memory space with the same size of a specified memory for applications or virtual machines.

MemV6 is a kMemvisor-like implementation for xv6. By modifying memory management module in the OS, MemV6 can achieve high-available memory access in xv6.

The above picture shows the workflow of kMemvisor's strategy

1. A block of physical memory will be reserved when a VM startups.
2. When native PTEs are updated, the related mirror PTEs will also be created.
3. Native instructions and mirror instructions will write the same data in different address.
4. If a page is corrupted, a new page will be allocated to the VM.
5. Map the virtual address to the new page.
6. Copy data from the mirror address.

Because of simplicity, we only implement memory redundant in MemV6.
MemV6

Memv6 will create the redundant page table while the native page table is created. When an application that has been binary translated runs, the duplicated memory write won't cause a page fualt. And the memory layout of xv6 is analyzed carefully so that there is no overlap between native address and redundant address.

Changed Files

• memlayout.h
• vm.c
• umalloc.c
• sysproc.c
• proc.c

Binary Translate

We use static binary translate to duplicate memory-write instructions so that every memory write will commit in mirrored memory either.
The following picture shows the overall architecture of binary translation.

Evaluation

We evaluated the performance of usertest(xv6's benchmark) between native xv6 and MemV6. This usertest contains several memory tests and the result shows that the largest overhead is still less than 60% and the average overhead is only 43%.

Reference

kMemvisor: flexible system wide memory mirroring in virtual environments.
Memvisor: Application Level Memory Mirroring via Binary Translation