INFO: How VMs Can Communicate with VxDs

• Microsoft Windows Device Development Kit (DDK) for Windows versions 3.0 and 3.1

SUMMARY

Under the enhanced mode Windows environment, it is sometimes necessary or desirable to provide means by which processes (virtual machines) can communicate with virtual devices (VxDs). This article discusses the techniques by which this can be achieved.

MORE INFORMATION

Virtual machines (VMs) can communicate with virtual devices either explicitly (which means that the virtual machine is aware of running under control of enhanced mode Windows and submits explicit calls to the VxD) or implicitly (which means that the VxD traps VM requests to the operating system that would otherwise be serviced by the BIOS, TSRs, or the hardware directly). The latter notion is basically a subset of the idea of "virtualization," which means leaving the process under the impression that it runs on a dedicated machine instead of under a multitasking operating system.

When VM software wants to call into a VxD explicitly, it must first use the Interrupt 2fh call "get device api entry point" (ax=1684h) to receive the callback address to which the call into the VxD must be submitted. The functionality of the VxD when processing this call is up to the VxD; the software running in the virtual machine must know both the OEM VxD ID number that belongs to the VxD and the functionality of the VxD API function to utilize it. Because the VxD runs under a different protection level and in 32-bit code segments instead of 16-bit code segments, a VM cannot directly call into the VxD. The necessary ring transitions and mode switches are performed by the VMM.

Because VxDs may also hook software interrupts that are being submitted by a VM (as outlined below), a software interrupt may also be seen as a valid method for a VM to explicitly call into a VxD; however, prior to submitting the software interrupt, the VM should make sure that the VxD is installed, which is possible through the "get device api entry point" call mentioned above. If the software interrupt interface provided by the VxD includes an installation check call (such as the VDS implementation provided by a Windows virtual device), there is no need for an explicit VxD installation check. Note that VDS stands for "Virtual DMA Interface Specification," which is a software interface for virtual machines to access physical DMA (direct memory access). The specification is available in the Software/Data Library; just query for VDS.

The ways a VxD can cause a VM to trap into it implicitly are the following:

• By hooking software interrupts that will be submitted by the VM
• By trapping accesses from VMs to ports
• By inserting callback addresses in the VMs address space

1. Use Hook_V86_Int_Chain to register a callback function that is being called whenever a VM running in V86 mode submits the software interrupt

2. Use Get/Set V86_Int_Vector or Get/Set PM_Int_Vector to change the interrupt tables of the virtual machines

• Interrupt hooks established via Hook_V86_Int_Chain are not called by virtual machines running in protected mode unless the last handler in the VM protected-mode interrupt chain reflects the interrupt to V86 mode.
• Get/Set V86/PM_Int_Vector behaves just as if any real/protected mode software running in the VM had used the interrupt vectors; all code and data that will be touched by the interrupt handler must be visible to the VM in the corresponding modes and reside in 16-bit segments (unless it is being sent to a callback); the modified vectors can be manipulated (that is, looked at, chained to, replaced, and so forth) by the VM.
• The calling software in the VM will never be aware of a VxD V86 hook. That is, the hook chain can never be accessed or manipulated by a VM; the interrupt handler resides in a VxD (that is, in a 32-bit segment on ring 0). V86 hooks are generally used to virtualize software interrupt interfaces within VMs (such as providing the VDS or virtual NetBIOS interfaces to VMs).
• V86 hooks cannot be installed on a per-VM basis, whereas PM and V86-mode interrupt handlers can be.
• Scenario 2 is very rarely used because it requires that the VxD must provide code and data in an environment which is not easily accessible to it; in effect, by doing so it would extend the VM with software running in the VMs execution mode, whereas the technique outlined in scenario 1 contains code and data which is directly accessible to the VxD.

To use port trapping for communication, a VxD calls one of the services, Install_IO_Handler or Install_Mult_IO_Handler.

A callback, eventually, is an address within the visible range of a VM that will transfer control to a routine supplied in a VxD when being jumped to while the VM is running in its current mode. Callback functions can be combined with interrupt vectors in that an interrupt vector in a VM can be set to a callback function. Note, however, that for VMs running in V86 mode, hooking the interrupt is generally more efficient than setting the V86-mode interrupt vector to a callback because two ring transitions less are needed.

To allocate a callback, a VxD can call the service Allocate_V86_Call_back service for VMs running in V86 mode and Allocate_PM_Call_Back for VMs running in protected mode.

Calling VM Code from a VxD

When a VxD wants to transfer control to code residing in a VM, it will generally do this by scheduling an event for the VM that simulates a far call or an interrupt into the VM. Depending on whether the event has been scheduled using the CallxxxEvent or the SchedulexxxEvent service, and which VM was currently running when the event was scheduled, the call may or may not be executed immediately. In any case, a ring transition will occur that will execute the VM code in a less privileged level in 16-bit mode.

If a VxD wants to call software in a VM directly without scheduling an event, it should do so by using the Begin_Nest_Exec and End_Nest_Exec services. Because this mechanism uses the current virtual machine only, a VxD should use it primarily with global TSRs because global TSRs are guaranteed to be shared by all VMs; if a VxD wants to use nested execution to call code that is local to one VM, it must make sure to do this only if the VM hosting the code to call into is the current VM.

Nested calls and scheduled events can be used cooperatively. In particular, due to the non-reentrancy of the Windows enhanced mode operating system, certain routines (such as interrupt handlers) will need to schedule events that will eventually submit a nested execution call into a VM.