Documentation/virt/kvm/nested-vmx.rst

   1 .. SPDX-License-Identifier: GPL-2.0
   2
   3 ==========
   4 Nested VMX
   5 ==========
   6
   7 Overview
   8 ---------
   9
  10 On Intel processors, KVM uses Intel's VMX (Virtual-Machine eXtensions)
  11 to easily and efficiently run guest operating systems. Normally, these guests
  12 *cannot* themselves be hypervisors running their own guests, because in VMX,
  13 guests cannot use VMX instructions.
  14
  15 The "Nested VMX" feature adds this missing capability - of running guest
  16 hypervisors (which use VMX) with their own nested guests. It does so by
  17 allowing a guest to use VMX instructions, and correctly and efficiently
  18 emulating them using the single level of VMX available in the hardware.
  19
  20 We describe in much greater detail the theory behind the nested VMX feature,
  21 its implementation and its performance characteristics, in the OSDI 2010 paper
  22 "The Turtles Project: Design and Implementation of Nested Virtualization",
  23 available at:
  24
  25         https://www.usenix.org/events/osdi10/tech/full_papers/Ben-Yehuda.pdf
  26
  27
  28 Terminology
  29 -----------
  30
  31 Single-level virtualization has two levels - the host (KVM) and the guests.
  32 In nested virtualization, we have three levels: The host (KVM), which we call
  33 L0, the guest hypervisor, which we call L1, and its nested guest, which we
  34 call L2.
  35
  36
  37 Running nested VMX
  38 ------------------
  39
  40 The nested VMX feature is disabled by default. It can be enabled by giving
  41 the "nested=1" option to the kvm-intel module.
  42
  43 No modifications are required to user space (qemu). However, qemu's default
  44 emulated CPU type (qemu64) does not list the "VMX" CPU feature, so it must be
  45 explicitly enabled, by giving qemu one of the following options:
  46
  47      - cpu host              (emulated CPU has all features of the real CPU)
  48
  49      - cpu qemu64,+vmx       (add just the vmx feature to a named CPU type)
  50
  51
  52 ABIs
  53 ----
  54
  55 Nested VMX aims to present a standard and (eventually) fully-functional VMX
  56 implementation for the a guest hypervisor to use. As such, the official
  57 specification of the ABI that it provides is Intel's VMX specification,
  58 namely volume 3B of their "Intel 64 and IA-32 Architectures Software
  59 Developer's Manual". Not all of VMX's features are currently fully supported,
  60 but the goal is to eventually support them all, starting with the VMX features
  61 which are used in practice by popular hypervisors (KVM and others).
  62
  63 As a VMX implementation, nested VMX presents a VMCS structure to L1.
  64 As mandated by the spec, other than the two fields revision_id and abort,
  65 this structure is *opaque* to its user, who is not supposed to know or care
  66 about its internal structure. Rather, the structure is accessed through the
  67 VMREAD and VMWRITE instructions.
  68 Still, for debugging purposes, KVM developers might be interested to know the
  69 internals of this structure; This is struct vmcs12 from arch/x86/kvm/vmx.c.
  70
  71 The name "vmcs12" refers to the VMCS that L1 builds for L2. In the code we
  72 also have "vmcs01", the VMCS that L0 built for L1, and "vmcs02" is the VMCS
  73 which L0 builds to actually run L2 - how this is done is explained in the
  74 aforementioned paper.
  75
  76 For convenience, we repeat the content of struct vmcs12 here. If the internals
  77 of this structure changes, this can break live migration across KVM versions.
  78 VMCS12_REVISION (from vmx.c) should be changed if struct vmcs12 or its inner
  79 struct shadow_vmcs is ever changed.
  80
  81 ::
  82
  83         typedef u64 natural_width;
  84         struct __packed vmcs12 {
  85                 /* According to the Intel spec, a VMCS region must start with
  86                  * these two user-visible fields */
  87                 u32 revision_id;
  88                 u32 abort;
  89
  90                 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
  91                 u32 padding[7]; /* room for future expansion */
  92
  93                 u64 io_bitmap_a;
  94                 u64 io_bitmap_b;
  95                 u64 msr_bitmap;
  96                 u64 vm_exit_msr_store_addr;
  97                 u64 vm_exit_msr_load_addr;
  98                 u64 vm_entry_msr_load_addr;
  99                 u64 tsc_offset;
 100                 u64 virtual_apic_page_addr;
 101                 u64 apic_access_addr;
 102                 u64 ept_pointer;
 103                 u64 guest_physical_address;
 104                 u64 vmcs_link_pointer;
 105                 u64 guest_ia32_debugctl;
 106                 u64 guest_ia32_pat;
 107                 u64 guest_ia32_efer;
 108                 u64 guest_pdptr0;
 109                 u64 guest_pdptr1;
 110                 u64 guest_pdptr2;
 111                 u64 guest_pdptr3;
 112                 u64 host_ia32_pat;
 113                 u64 host_ia32_efer;
 114                 u64 padding64[8]; /* room for future expansion */
 115                 natural_width cr0_guest_host_mask;
 116                 natural_width cr4_guest_host_mask;
 117                 natural_width cr0_read_shadow;
 118                 natural_width cr4_read_shadow;
 119                 natural_width dead_space[4]; /* Last remnants of cr3_target_value[0-3]. */
 120                 natural_width exit_qualification;
 121                 natural_width guest_linear_address;
 122                 natural_width guest_cr0;
 123                 natural_width guest_cr3;
 124                 natural_width guest_cr4;
 125                 natural_width guest_es_base;
 126                 natural_width guest_cs_base;
 127                 natural_width guest_ss_base;
 128                 natural_width guest_ds_base;
 129                 natural_width guest_fs_base;
 130                 natural_width guest_gs_base;
 131                 natural_width guest_ldtr_base;
 132                 natural_width guest_tr_base;
 133                 natural_width guest_gdtr_base;
 134                 natural_width guest_idtr_base;
 135                 natural_width guest_dr7;
 136                 natural_width guest_rsp;
 137                 natural_width guest_rip;
 138                 natural_width guest_rflags;
 139                 natural_width guest_pending_dbg_exceptions;
 140                 natural_width guest_sysenter_esp;
 141                 natural_width guest_sysenter_eip;
 142                 natural_width host_cr0;
 143                 natural_width host_cr3;
 144                 natural_width host_cr4;
 145                 natural_width host_fs_base;
 146                 natural_width host_gs_base;
 147                 natural_width host_tr_base;
 148                 natural_width host_gdtr_base;
 149                 natural_width host_idtr_base;
 150                 natural_width host_ia32_sysenter_esp;
 151                 natural_width host_ia32_sysenter_eip;
 152                 natural_width host_rsp;
 153                 natural_width host_rip;
 154                 natural_width paddingl[8]; /* room for future expansion */
 155                 u32 pin_based_vm_exec_control;
 156                 u32 cpu_based_vm_exec_control;
 157                 u32 exception_bitmap;
 158                 u32 page_fault_error_code_mask;
 159                 u32 page_fault_error_code_match;
 160                 u32 cr3_target_count;
 161                 u32 vm_exit_controls;
 162                 u32 vm_exit_msr_store_count;
 163                 u32 vm_exit_msr_load_count;
 164                 u32 vm_entry_controls;
 165                 u32 vm_entry_msr_load_count;
 166                 u32 vm_entry_intr_info_field;
 167                 u32 vm_entry_exception_error_code;
 168                 u32 vm_entry_instruction_len;
 169                 u32 tpr_threshold;
 170                 u32 secondary_vm_exec_control;
 171                 u32 vm_instruction_error;
 172                 u32 vm_exit_reason;
 173                 u32 vm_exit_intr_info;
 174                 u32 vm_exit_intr_error_code;
 175                 u32 idt_vectoring_info_field;
 176                 u32 idt_vectoring_error_code;
 177                 u32 vm_exit_instruction_len;
 178                 u32 vmx_instruction_info;
 179                 u32 guest_es_limit;
 180                 u32 guest_cs_limit;
 181                 u32 guest_ss_limit;
 182                 u32 guest_ds_limit;
 183                 u32 guest_fs_limit;
 184                 u32 guest_gs_limit;
 185                 u32 guest_ldtr_limit;
 186                 u32 guest_tr_limit;
 187                 u32 guest_gdtr_limit;
 188                 u32 guest_idtr_limit;
 189                 u32 guest_es_ar_bytes;
 190                 u32 guest_cs_ar_bytes;
 191                 u32 guest_ss_ar_bytes;
 192                 u32 guest_ds_ar_bytes;
 193                 u32 guest_fs_ar_bytes;
 194                 u32 guest_gs_ar_bytes;
 195                 u32 guest_ldtr_ar_bytes;
 196                 u32 guest_tr_ar_bytes;
 197                 u32 guest_interruptibility_info;
 198                 u32 guest_activity_state;
 199                 u32 guest_sysenter_cs;
 200                 u32 host_ia32_sysenter_cs;
 201                 u32 padding32[8]; /* room for future expansion */
 202                 u16 virtual_processor_id;
 203                 u16 guest_es_selector;
 204                 u16 guest_cs_selector;
 205                 u16 guest_ss_selector;
 206                 u16 guest_ds_selector;
 207                 u16 guest_fs_selector;
 208                 u16 guest_gs_selector;
 209                 u16 guest_ldtr_selector;
 210                 u16 guest_tr_selector;
 211                 u16 host_es_selector;
 212                 u16 host_cs_selector;
 213                 u16 host_ss_selector;
 214                 u16 host_ds_selector;
 215                 u16 host_fs_selector;
 216                 u16 host_gs_selector;
 217                 u16 host_tr_selector;
 218         };
 219
 220
 221 Authors
 222 -------
 223
 224 These patches were written by:
 225     - Abel Gordon, abelg <at> il.ibm.com
 226     - Nadav Har'El, nyh <at> il.ibm.com
 227     - Orit Wasserman, oritw <at> il.ibm.com
 228     - Ben-Ami Yassor, benami <at> il.ibm.com
 229     - Muli Ben-Yehuda, muli <at> il.ibm.com
 230
 231 With contributions by:
 232     - Anthony Liguori, aliguori <at> us.ibm.com
 233     - Mike Day, mdday <at> us.ibm.com
 234     - Michael Factor, factor <at> il.ibm.com
 235     - Zvi Dubitzky, dubi <at> il.ibm.com
 236
 237 And valuable reviews by:
 238     - Avi Kivity, avi <at> redhat.com
 239     - Gleb Natapov, gleb <at> redhat.com
 240     - Marcelo Tosatti, mtosatti <at> redhat.com
 241     - Kevin Tian, kevin.tian <at> intel.com
 242     - and others.