Intel and AMD are creating methods in which a virtual
Ring 0 (commonly called Ring -1) can be presented to the guest for direct
process execution through the VMM. ESX 3 cannot take advantage of this
functionality at this point. Instead, the VMM must send the instructions to the
VMkernel to execute the instruction and return it to the guest. The VMM virtualizes
this process to make the guest think that it is running in an actual Protection
Ring 0. Due to the fact that the instructions must be virtualized and
executed within the VMkernel there is virtualization overhead associated with
these calls. The amount of overhead varies depending on the type of instruction
and the overall workload being processed by the VMkernel. You can tell
how often a guest operating system needs to run in Virtualization Mode by capturing
the total percentage of Kernel or System Mode Process utilization.
50
Figure 2- 6: Processing Rings
Fortunately for us, the VMM responsible for running each virtual machine
knows when an instruction requires Virtualization Mode and has the capability
to properly move itself into the proper execution mode. When the requirement
for Virtualization Mode has been met, the VMM will move itself back into direct
execution mode.
Hardware Virtualization Enhancements
As mentioned earlier Intel and AMD have new processors on the market that
have virtualization enhancements integrated into the processors themselves. At
this point in time VMware ESX 3.0 does support some of the functionality, but
not all. As an example, in order to run a 64-bit guest operating system ESX
requires the use of a virtualization enhanced processor. This allows the
VMkernel, which is a 32-bit kernel to run 64-bit instructions for the 64-bit
guest. Alternatively, ESX 3 does not have the capability to present a virtual
Ring 0 to the guest operating system for Direct Execution Mode processing of
Kernel Mode Processes.
