Changes to Xen in support of Intel(R) Vanderpool Technology
-------------------------------------------------------------

Our VT extensions to the Xen hypervisor provide full platform
virtualization, including CPU(s), memory, and I/O infrastructure. The
generic code in Xen handles and schedules those virtual machines as it
does for the existing para-virtualized domains.

Full virtualization required by the OS guests requires full device
virtualization as well. The device models in BOCHS
(http://bochs.sourceforge.net/) were decoupled from the CPU
virtualization, and are used to virtualize the legacy devices (such as
keyboard, mouse, VGA, IDE) in the PC platform. At this point, the
device models run in user mode on domain 0, not in the Xen hypervisor.

We would like to thank Ian Pratt and Keir Fraser for reviewing our
design and code intensively, and for providing numerous useful
suggestions to improve the architecture and code. 

We have a list of Intel team members who take credit for making this
release happen: Yunhong Jiang, Nitin Kamble, Chengyuan Li, Xin Li,
Xiaofeng Ling, Benjamin Liu, Asit Mallick, Jun Nakajima, Sunil Saxena,
Arun Sharma, Edwin Zhai, Jeff Zheng, and Louis Zhuang. We'll continue
to add more features to complete full virtualization in Xen using VT.

The notes document the changes to the Xen hypervisor in order to add
VT support. The changes to other areas, such as Control Panel will be
added as we deliver the code.

Summary of changes for the first release
----------------------------------------
December 15, 2004

    * VT specific event handling and domain management were added. 

    * Shadow mode was extended to support full 32-bit guests
    
    * Domain switching code was extended to support VT domain
    
    * I/O request handling was added to communicate with the device model

    * Domain builder was extended to provide the environment when the
      guest enters the protected mode, including E820 memory and VGA
      info, typically obtained by BIOS calls.

New code:
---------
    VT (Vanderpool Technology) is based on the new VMX (Virtual
    Machine Extensions) architecture. The current release of the
    software supports 32-bit only.

    * arch/x86/vmx.[ch] and arch/x86/vmx_*.[ch]: created to handle
      VMX-specific events in order to provide virtual machine.

    * arch/x86/x86_32/entry.S: new code path was added to have the
      first-level handler from VM exits. The first-level handler calls
      the second-level handler in arch/x86/vmx.c.

    * arch/x86/setup.c: new function start_vmx() to init_intel() to
      enable VMX mode.

    * include/asm-x86/config.h: #ifdef CONFIG_VMX was added.

    * arch/x86/domain.c: new code patch was added to create a VMX
      domain given the flag from the control panel.

    * include/public/io/ioreq.h: A new data structure was added to
      define the I/O requests between the Xen hypervisor and the
      device models.

Changes to the existing code:
-----------------------------

    * arch/x86/shadow.[ch]: new mode SHM_full_32 was added to support
      full virtualization. The current Xen code assumes that the guest
      page directory and tables have _machine_ (or host) physical page
      frame numbers, and the new code allows to support _guest_
      physical page frame numbers

    * include/asm-x86/processor.h: struct arch_vmx_struct arch_vmx has
      been added to the thread_struct data structure. The arch_vmx has
      the addtional VMX-related CPU context.

    * arch/x86/io_apic.c: reverse mapping between vector and irq has
      been added. We will revisit this code when considering MSI
      support.

--- Jun