KVM: nVMX: Documentation
authorNadav Har'El <nyh@il.ibm.com>
Wed, 25 May 2011 20:17:11 +0000 (23:17 +0300)
committerAvi Kivity <avi@redhat.com>
Tue, 12 Jul 2011 10:15:22 +0000 (13:15 +0300)
This patch includes a brief introduction to the nested vmx feature in the
Documentation/kvm directory. The document also includes a copy of the
vmcs12 structure, as requested by Avi Kivity.

[marcelo: move to Documentation/virtual/kvm]

Signed-off-by: Nadav Har'El <nyh@il.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Documentation/virtual/kvm/nested-vmx.txt [new file with mode: 0644]

diff --git a/Documentation/virtual/kvm/nested-vmx.txt b/Documentation/virtual/kvm/nested-vmx.txt
new file mode 100644 (file)
index 0000000..8ed937d
--- /dev/null
@@ -0,0 +1,251 @@
+Nested VMX
+==========
+
+Overview
+---------
+
+On Intel processors, KVM uses Intel's VMX (Virtual-Machine eXtensions)
+to easily and efficiently run guest operating systems. Normally, these guests
+*cannot* themselves be hypervisors running their own guests, because in VMX,
+guests cannot use VMX instructions.
+
+The "Nested VMX" feature adds this missing capability - of running guest
+hypervisors (which use VMX) with their own nested guests. It does so by
+allowing a guest to use VMX instructions, and correctly and efficiently
+emulating them using the single level of VMX available in the hardware.
+
+We describe in much greater detail the theory behind the nested VMX feature,
+its implementation and its performance characteristics, in the OSDI 2010 paper
+"The Turtles Project: Design and Implementation of Nested Virtualization",
+available at:
+
+       http://www.usenix.org/events/osdi10/tech/full_papers/Ben-Yehuda.pdf
+
+
+Terminology
+-----------
+
+Single-level virtualization has two levels - the host (KVM) and the guests.
+In nested virtualization, we have three levels: The host (KVM), which we call
+L0, the guest hypervisor, which we call L1, and its nested guest, which we
+call L2.
+
+
+Known limitations
+-----------------
+
+The current code supports running Linux guests under KVM guests.
+Only 64-bit guest hypervisors are supported.
+
+Additional patches for running Windows under guest KVM, and Linux under
+guest VMware server, and support for nested EPT, are currently running in
+the lab, and will be sent as follow-on patchsets.
+
+
+Running nested VMX
+------------------
+
+The nested VMX feature is disabled by default. It can be enabled by giving
+the "nested=1" option to the kvm-intel module.
+
+No modifications are required to user space (qemu). However, qemu's default
+emulated CPU type (qemu64) does not list the "VMX" CPU feature, so it must be
+explicitly enabled, by giving qemu one of the following options:
+
+     -cpu host              (emulated CPU has all features of the real CPU)
+
+     -cpu qemu64,+vmx       (add just the vmx feature to a named CPU type)
+
+
+ABIs
+----
+
+Nested VMX aims to present a standard and (eventually) fully-functional VMX
+implementation for the a guest hypervisor to use. As such, the official
+specification of the ABI that it provides is Intel's VMX specification,
+namely volume 3B of their "Intel 64 and IA-32 Architectures Software
+Developer's Manual". Not all of VMX's features are currently fully supported,
+but the goal is to eventually support them all, starting with the VMX features
+which are used in practice by popular hypervisors (KVM and others).
+
+As a VMX implementation, nested VMX presents a VMCS structure to L1.
+As mandated by the spec, other than the two fields revision_id and abort,
+this structure is *opaque* to its user, who is not supposed to know or care
+about its internal structure. Rather, the structure is accessed through the
+VMREAD and VMWRITE instructions.
+Still, for debugging purposes, KVM developers might be interested to know the
+internals of this structure; This is struct vmcs12 from arch/x86/kvm/vmx.c.
+
+The name "vmcs12" refers to the VMCS that L1 builds for L2. In the code we
+also have "vmcs01", the VMCS that L0 built for L1, and "vmcs02" is the VMCS
+which L0 builds to actually run L2 - how this is done is explained in the
+aforementioned paper.
+
+For convenience, we repeat the content of struct vmcs12 here. If the internals
+of this structure changes, this can break live migration across KVM versions.
+VMCS12_REVISION (from vmx.c) should be changed if struct vmcs12 or its inner
+struct shadow_vmcs is ever changed.
+
+       typedef u64 natural_width;
+       struct __packed vmcs12 {
+               /* According to the Intel spec, a VMCS region must start with
+                * these two user-visible fields */
+               u32 revision_id;
+               u32 abort;
+
+               u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
+               u32 padding[7]; /* room for future expansion */
+
+               u64 io_bitmap_a;
+               u64 io_bitmap_b;
+               u64 msr_bitmap;
+               u64 vm_exit_msr_store_addr;
+               u64 vm_exit_msr_load_addr;
+               u64 vm_entry_msr_load_addr;
+               u64 tsc_offset;
+               u64 virtual_apic_page_addr;
+               u64 apic_access_addr;
+               u64 ept_pointer;
+               u64 guest_physical_address;
+               u64 vmcs_link_pointer;
+               u64 guest_ia32_debugctl;
+               u64 guest_ia32_pat;
+               u64 guest_ia32_efer;
+               u64 guest_pdptr0;
+               u64 guest_pdptr1;
+               u64 guest_pdptr2;
+               u64 guest_pdptr3;
+               u64 host_ia32_pat;
+               u64 host_ia32_efer;
+               u64 padding64[8]; /* room for future expansion */
+               natural_width cr0_guest_host_mask;
+               natural_width cr4_guest_host_mask;
+               natural_width cr0_read_shadow;
+               natural_width cr4_read_shadow;
+               natural_width cr3_target_value0;
+               natural_width cr3_target_value1;
+               natural_width cr3_target_value2;
+               natural_width cr3_target_value3;
+               natural_width exit_qualification;
+               natural_width guest_linear_address;
+               natural_width guest_cr0;
+               natural_width guest_cr3;
+               natural_width guest_cr4;
+               natural_width guest_es_base;
+               natural_width guest_cs_base;
+               natural_width guest_ss_base;
+               natural_width guest_ds_base;
+               natural_width guest_fs_base;
+               natural_width guest_gs_base;
+               natural_width guest_ldtr_base;
+               natural_width guest_tr_base;
+               natural_width guest_gdtr_base;
+               natural_width guest_idtr_base;
+               natural_width guest_dr7;
+               natural_width guest_rsp;
+               natural_width guest_rip;
+               natural_width guest_rflags;
+               natural_width guest_pending_dbg_exceptions;
+               natural_width guest_sysenter_esp;
+               natural_width guest_sysenter_eip;
+               natural_width host_cr0;
+               natural_width host_cr3;
+               natural_width host_cr4;
+               natural_width host_fs_base;
+               natural_width host_gs_base;
+               natural_width host_tr_base;
+               natural_width host_gdtr_base;
+               natural_width host_idtr_base;
+               natural_width host_ia32_sysenter_esp;
+               natural_width host_ia32_sysenter_eip;
+               natural_width host_rsp;
+               natural_width host_rip;
+               natural_width paddingl[8]; /* room for future expansion */
+               u32 pin_based_vm_exec_control;
+               u32 cpu_based_vm_exec_control;
+               u32 exception_bitmap;
+               u32 page_fault_error_code_mask;
+               u32 page_fault_error_code_match;
+               u32 cr3_target_count;
+               u32 vm_exit_controls;
+               u32 vm_exit_msr_store_count;
+               u32 vm_exit_msr_load_count;
+               u32 vm_entry_controls;
+               u32 vm_entry_msr_load_count;
+               u32 vm_entry_intr_info_field;
+               u32 vm_entry_exception_error_code;
+               u32 vm_entry_instruction_len;
+               u32 tpr_threshold;
+               u32 secondary_vm_exec_control;
+               u32 vm_instruction_error;
+               u32 vm_exit_reason;
+               u32 vm_exit_intr_info;
+               u32 vm_exit_intr_error_code;
+               u32 idt_vectoring_info_field;
+               u32 idt_vectoring_error_code;
+               u32 vm_exit_instruction_len;
+               u32 vmx_instruction_info;
+               u32 guest_es_limit;
+               u32 guest_cs_limit;
+               u32 guest_ss_limit;
+               u32 guest_ds_limit;
+               u32 guest_fs_limit;
+               u32 guest_gs_limit;
+               u32 guest_ldtr_limit;
+               u32 guest_tr_limit;
+               u32 guest_gdtr_limit;
+               u32 guest_idtr_limit;
+               u32 guest_es_ar_bytes;
+               u32 guest_cs_ar_bytes;
+               u32 guest_ss_ar_bytes;
+               u32 guest_ds_ar_bytes;
+               u32 guest_fs_ar_bytes;
+               u32 guest_gs_ar_bytes;
+               u32 guest_ldtr_ar_bytes;
+               u32 guest_tr_ar_bytes;
+               u32 guest_interruptibility_info;
+               u32 guest_activity_state;
+               u32 guest_sysenter_cs;
+               u32 host_ia32_sysenter_cs;
+               u32 padding32[8]; /* room for future expansion */
+               u16 virtual_processor_id;
+               u16 guest_es_selector;
+               u16 guest_cs_selector;
+               u16 guest_ss_selector;
+               u16 guest_ds_selector;
+               u16 guest_fs_selector;
+               u16 guest_gs_selector;
+               u16 guest_ldtr_selector;
+               u16 guest_tr_selector;
+               u16 host_es_selector;
+               u16 host_cs_selector;
+               u16 host_ss_selector;
+               u16 host_ds_selector;
+               u16 host_fs_selector;
+               u16 host_gs_selector;
+               u16 host_tr_selector;
+       };
+
+
+Authors
+-------
+
+These patches were written by:
+     Abel Gordon, abelg <at> il.ibm.com
+     Nadav Har'El, nyh <at> il.ibm.com
+     Orit Wasserman, oritw <at> il.ibm.com
+     Ben-Ami Yassor, benami <at> il.ibm.com
+     Muli Ben-Yehuda, muli <at> il.ibm.com
+
+With contributions by:
+     Anthony Liguori, aliguori <at> us.ibm.com
+     Mike Day, mdday <at> us.ibm.com
+     Michael Factor, factor <at> il.ibm.com
+     Zvi Dubitzky, dubi <at> il.ibm.com
+
+And valuable reviews by:
+     Avi Kivity, avi <at> redhat.com
+     Gleb Natapov, gleb <at> redhat.com
+     Marcelo Tosatti, mtosatti <at> redhat.com
+     Kevin Tian, kevin.tian <at> intel.com
+     and others.