Documentation/arch/powerpc/kvm-nested.rst

   1 .. SPDX-License-Identifier: GPL-2.0
   2
   3 ====================================
   4 Nested KVM on POWER
   5 ====================================
   6
   7 Introduction
   8 ============
   9
  10 This document explains how a guest operating system can act as a
  11 hypervisor and run nested guests through the use of hypercalls, if the
  12 hypervisor has implemented them. The terms L0, L1, and L2 are used to
  13 refer to different software entities. L0 is the hypervisor mode entity
  14 that would normally be called the "host" or "hypervisor". L1 is a
  15 guest virtual machine that is directly run under L0 and is initiated
  16 and controlled by L0. L2 is a guest virtual machine that is initiated
  17 and controlled by L1 acting as a hypervisor.
  18
  19 Existing API
  20 ============
  21
  22 Linux/KVM has had support for Nesting as an L0 or L1 since 2018
  23
  24 The L0 code was added::
  25
  26    commit 8e3f5fc1045dc49fd175b978c5457f5f51e7a2ce
  27    Author: Paul Mackerras <paulus@ozlabs.org>
  28    Date:   Mon Oct 8 16:31:03 2018 +1100
  29    KVM: PPC: Book3S HV: Framework and hcall stubs for nested virtualization
  30
  31 The L1 code was added::
  32
  33    commit 360cae313702cdd0b90f82c261a8302fecef030a
  34    Author: Paul Mackerras <paulus@ozlabs.org>
  35    Date:   Mon Oct 8 16:31:04 2018 +1100
  36    KVM: PPC: Book3S HV: Nested guest entry via hypercall
  37
  38 This API works primarily using a single hcall h_enter_nested(). This
  39 call made by the L1 to tell the L0 to start an L2 vCPU with the given
  40 state. The L0 then starts this L2 and runs until an L2 exit condition
  41 is reached. Once the L2 exits, the state of the L2 is given back to
  42 the L1 by the L0. The full L2 vCPU state is always transferred from
  43 and to L1 when the L2 is run. The L0 doesn't keep any state on the L2
  44 vCPU (except in the short sequence in the L0 on L1 -> L2 entry and L2
  45 -> L1 exit).
  46
  47 The only state kept by the L0 is the partition table. The L1 registers
  48 it's partition table using the h_set_partition_table() hcall. All
  49 other state held by the L0 about the L2s is cached state (such as
  50 shadow page tables).
  51
  52 The L1 may run any L2 or vCPU without first informing the L0. It
  53 simply starts the vCPU using h_enter_nested(). The creation of L2s and
  54 vCPUs is done implicitly whenever h_enter_nested() is called.
  55
  56 In this document, we call this existing API the v1 API.
  57
  58 New PAPR API
  59 ===============
  60
  61 The new PAPR API changes from the v1 API such that the creating L2 and
  62 associated vCPUs is explicit. In this document, we call this the v2
  63 API.
  64
  65 h_enter_nested() is replaced with H_GUEST_VCPU_RUN().  Before this can
  66 be called the L1 must explicitly create the L2 using h_guest_create()
  67 and any associated vCPUs() created with h_guest_create_vCPU(). Getting
  68 and setting vCPU state can also be performed using h_guest_{g|s}et
  69 hcall.
  70
  71 The basic execution flow is for an L1 to create an L2, run it, and
  72 delete it is:
  73
  74 - L1 and L0 negotiate capabilities with H_GUEST_{G,S}ET_CAPABILITIES()
  75   (normally at L1 boot time).
  76
  77 - L1 requests the L0 create an L2 with H_GUEST_CREATE() and receives a token
  78
  79 - L1 requests the L0 create an L2 vCPU with H_GUEST_CREATE_VCPU()
  80
  81 - L1 and L0 communicate the vCPU state using the H_GUEST_{G,S}ET() hcall
  82
  83 - L1 requests the L0 runs the vCPU running H_GUEST_VCPU_RUN() hcall
  84
  85 - L1 deletes L2 with H_GUEST_DELETE()
  86
  87 More details of the individual hcalls follows:
  88
  89 HCALL Details
  90 =============
  91
  92 This documentation is provided to give an overall understating of the
  93 API. It doesn't aim to provide all the details required to implement
  94 an L1 or L0. Latest version of PAPR can be referred to for more details.
  95
  96 All these HCALLs are made by the L1 to the L0.
  97
  98 H_GUEST_GET_CAPABILITIES()
  99 --------------------------
 100
 101 This is called to get the capabilities of the L0 nested
 102 hypervisor. This includes capabilities such the CPU versions (eg
 103 POWER9, POWER10) that are supported as L2s::
 104
 105   H_GUEST_GET_CAPABILITIES(uint64 flags)
 106
 107   Parameters:
 108     Input:
 109       flags: Reserved
 110     Output:
 111       R3: Return code
 112       R4: Hypervisor Supported Capabilities bitmap 1
 113
 114 H_GUEST_SET_CAPABILITIES()
 115 --------------------------
 116
 117 This is called to inform the L0 of the capabilities of the L1
 118 hypervisor. The set of flags passed here are the same as
 119 H_GUEST_GET_CAPABILITIES()
 120
 121 Typically, GET will be called first and then SET will be called with a
 122 subset of the flags returned from GET. This process allows the L0 and
 123 L1 to negotiate an agreed set of capabilities::
 124
 125   H_GUEST_SET_CAPABILITIES(uint64 flags,
 126                            uint64 capabilitiesBitmap1)
 127   Parameters:
 128     Input:
 129       flags: Reserved
 130       capabilitiesBitmap1: Only capabilities advertised through
 131                            H_GUEST_GET_CAPABILITIES
 132     Output:
 133       R3: Return code
 134       R4: If R3 = H_P2: The number of invalid bitmaps
 135       R5: If R3 = H_P2: The index of first invalid bitmap
 136
 137 H_GUEST_CREATE()
 138 ----------------
 139
 140 This is called to create an L2. A unique ID of the L2 created
 141 (similar to an LPID) is returned, which can be used on subsequent HCALLs to
 142 identify the L2::
 143
 144   H_GUEST_CREATE(uint64 flags,
 145                  uint64 continueToken);
 146   Parameters:
 147     Input:
 148       flags: Reserved
 149       continueToken: Initial call set to -1. Subsequent calls,
 150                      after H_Busy or H_LongBusyOrder has been
 151                      returned, value that was returned in R4.
 152     Output:
 153       R3: Return code. Notable:
 154         H_Not_Enough_Resources: Unable to create Guest VCPU due to not
 155         enough Hypervisor memory. See H_GUEST_CREATE_GET_STATE(flags =
 156         takeOwnershipOfVcpuState)
 157       R4: If R3 = H_Busy or_H_LongBusyOrder -> continueToken
 158
 159 H_GUEST_CREATE_VCPU()
 160 ---------------------
 161
 162 This is called to create a vCPU associated with an L2. The L2 id
 163 (returned from H_GUEST_CREATE()) should be passed it. Also passed in
 164 is a unique (for this L2) vCPUid. This vCPUid is allocated by the
 165 L1::
 166
 167   H_GUEST_CREATE_VCPU(uint64 flags,
 168                       uint64 guestId,
 169                       uint64 vcpuId);
 170   Parameters:
 171     Input:
 172       flags: Reserved
 173       guestId: ID obtained from H_GUEST_CREATE
 174       vcpuId: ID of the vCPU to be created. This must be within the
 175               range of 0 to 2047
 176     Output:
 177       R3: Return code. Notable:
 178         H_Not_Enough_Resources: Unable to create Guest VCPU due to not
 179         enough Hypervisor memory. See H_GUEST_CREATE_GET_STATE(flags =
 180         takeOwnershipOfVcpuState)
 181
 182 H_GUEST_GET_STATE()
 183 -------------------
 184
 185 This is called to get state associated with an L2 (Guest-wide or vCPU specific).
 186 This info is passed via the Guest State Buffer (GSB), a standard format as
 187 explained later in this doc, necessary details below:
 188
 189 This can get either L2 wide or vcpu specific information. Examples of
 190 L2 wide is the timebase offset or process scoped page table
 191 info. Examples of vCPU specific are GPRs or VSRs. A bit in the flags
 192 parameter specifies if this call is L2 wide or vCPU specific and the
 193 IDs in the GSB must match this.
 194
 195 The L1 provides a pointer to the GSB as a parameter to this call. Also
 196 provided is the L2 and vCPU IDs associated with the state to set.
 197
 198 The L1 writes only the IDs and sizes in the GSB.  L0 writes the
 199 associated values for each ID in the GSB::
 200
 201   H_GUEST_GET_STATE(uint64 flags,
 202                            uint64 guestId,
 203                            uint64 vcpuId,
 204                            uint64 dataBuffer,
 205                            uint64 dataBufferSizeInBytes);
 206   Parameters:
 207     Input:
 208       flags:
 209          Bit 0: getGuestWideState: Request state of the Guest instead
 210            of an individual VCPU.
 211          Bit 1: takeOwnershipOfVcpuState Indicate the L1 is taking
 212            over ownership of the VCPU state and that the L0 can free
 213            the storage holding the state. The VCPU state will need to
 214            be returned to the Hypervisor via H_GUEST_SET_STATE prior
 215            to H_GUEST_RUN_VCPU being called for this VCPU. The data
 216            returned in the dataBuffer is in a Hypervisor internal
 217            format.
 218          Bits 2-63: Reserved
 219       guestId: ID obtained from H_GUEST_CREATE
 220       vcpuId: ID of the vCPU pass to H_GUEST_CREATE_VCPU
 221       dataBuffer: A L1 real address of the GSB.
 222         If takeOwnershipOfVcpuState, size must be at least the size
 223         returned by ID=0x0001
 224       dataBufferSizeInBytes: Size of dataBuffer
 225     Output:
 226       R3: Return code
 227       R4: If R3 = H_Invalid_Element_Id: The array index of the bad
 228             element ID.
 229           If R3 = H_Invalid_Element_Size: The array index of the bad
 230              element size.
 231           If R3 = H_Invalid_Element_Value: The array index of the bad
 232              element value.
 233
 234 H_GUEST_SET_STATE()
 235 -------------------
 236
 237 This is called to set L2 wide or vCPU specific L2 state. This info is
 238 passed via the Guest State Buffer (GSB), necessary details below:
 239
 240 This can set either L2 wide or vcpu specific information. Examples of
 241 L2 wide is the timebase offset or process scoped page table
 242 info. Examples of vCPU specific are GPRs or VSRs. A bit in the flags
 243 parameter specifies if this call is L2 wide or vCPU specific and the
 244 IDs in the GSB must match this.
 245
 246 The L1 provides a pointer to the GSB as a parameter to this call. Also
 247 provided is the L2 and vCPU IDs associated with the state to set.
 248
 249 The L1 writes all values in the GSB and the L0 only reads the GSB for
 250 this call::
 251
 252   H_GUEST_SET_STATE(uint64 flags,
 253                     uint64 guestId,
 254                     uint64 vcpuId,
 255                     uint64 dataBuffer,
 256                     uint64 dataBufferSizeInBytes);
 257   Parameters:
 258     Input:
 259       flags:
 260          Bit 0: getGuestWideState: Request state of the Guest instead
 261            of an individual VCPU.
 262          Bit 1: returnOwnershipOfVcpuState Return Guest VCPU state. See
 263            GET_STATE takeOwnershipOfVcpuState
 264          Bits 2-63: Reserved
 265       guestId: ID obtained from H_GUEST_CREATE
 266       vcpuId: ID of the vCPU pass to H_GUEST_CREATE_VCPU
 267       dataBuffer: A L1 real address of the GSB.
 268         If takeOwnershipOfVcpuState, size must be at least the size
 269         returned by ID=0x0001
 270       dataBufferSizeInBytes: Size of dataBuffer
 271     Output:
 272       R3: Return code
 273       R4: If R3 = H_Invalid_Element_Id: The array index of the bad
 274             element ID.
 275           If R3 = H_Invalid_Element_Size: The array index of the bad
 276              element size.
 277           If R3 = H_Invalid_Element_Value: The array index of the bad
 278              element value.
 279
 280 H_GUEST_RUN_VCPU()
 281 ------------------
 282
 283 This is called to run an L2 vCPU. The L2 and vCPU IDs are passed in as
 284 parameters. The vCPU runs with the state set previously using
 285 H_GUEST_SET_STATE(). When the L2 exits, the L1 will resume from this
 286 hcall.
 287
 288 This hcall also has associated input and output GSBs. Unlike
 289 H_GUEST_{S,G}ET_STATE(), these GSB pointers are not passed in as
 290 parameters to the hcall (This was done in the interest of
 291 performance). The locations of these GSBs must be preregistered using
 292 the H_GUEST_SET_STATE() call with ID 0x0c00 and 0x0c01 (see table
 293 below).
 294
 295 The input GSB may contain only VCPU specific elements to be set. This
 296 GSB may also contain zero elements (ie 0 in the first 4 bytes of the
 297 GSB) if nothing needs to be set.
 298
 299 On exit from the hcall, the output buffer is filled with elements
 300 determined by the L0. The reason for the exit is contained in GPR4 (ie
 301 NIP is put in GPR4).  The elements returned depend on the exit
 302 type. For example, if the exit reason is the L2 doing a hcall (GPR4 =
 303 0xc00), then GPR3-12 are provided in the output GSB as this is the
 304 state likely needed to service the hcall. If additional state is
 305 needed, H_GUEST_GET_STATE() may be called by the L1.
 306
 307 To synthesize interrupts in the L2, when calling H_GUEST_RUN_VCPU()
 308 the L1 may set a flag (as a hcall parameter) and the L0 will
 309 synthesize the interrupt in the L2. Alternatively, the L1 may
 310 synthesize the interrupt itself using H_GUEST_SET_STATE() or the
 311 H_GUEST_RUN_VCPU() input GSB to set the state appropriately::
 312
 313   H_GUEST_RUN_VCPU(uint64 flags,
 314                    uint64 guestId,
 315                    uint64 vcpuId,
 316                    uint64 dataBuffer,
 317                    uint64 dataBufferSizeInBytes);
 318   Parameters:
 319     Input:
 320       flags:
 321          Bit 0: generateExternalInterrupt: Generate an external interrupt
 322          Bit 1: generatePrivilegedDoorbell: Generate a Privileged Doorbell
 323          Bit 2: sendToSystemReset”: Generate a System Reset Interrupt
 324          Bits 3-63: Reserved
 325       guestId: ID obtained from H_GUEST_CREATE
 326       vcpuId: ID of the vCPU pass to H_GUEST_CREATE_VCPU
 327     Output:
 328       R3: Return code
 329       R4: If R3 = H_Success: The reason L1 VCPU exited (ie. NIA)
 330             0x000: The VCPU stopped running for an unspecified reason. An
 331               example of this is the Hypervisor stopping a VCPU running
 332               due to an outstanding interrupt for the Host Partition.
 333             0x980: HDEC
 334             0xC00: HCALL
 335             0xE00: HDSI
 336             0xE20: HISI
 337             0xE40: HEA
 338             0xF80: HV Fac Unavail
 339           If R3 = H_Invalid_Element_Id, H_Invalid_Element_Size, or
 340             H_Invalid_Element_Value: R4 is offset of the invalid element
 341             in the input buffer.
 342
 343 H_GUEST_DELETE()
 344 ----------------
 345
 346 This is called to delete an L2. All associated vCPUs are also
 347 deleted. No specific vCPU delete call is provided.
 348
 349 A flag may be provided to delete all guests. This is used to reset the
 350 L0 in the case of kdump/kexec::
 351
 352   H_GUEST_DELETE(uint64 flags,
 353                  uint64 guestId)
 354   Parameters:
 355     Input:
 356       flags:
 357          Bit 0: deleteAllGuests: deletes all guests
 358          Bits 1-63: Reserved
 359       guestId: ID obtained from H_GUEST_CREATE
 360     Output:
 361       R3: Return code
 362
 363 Guest State Buffer
 364 ==================
 365
 366 The Guest State Buffer (GSB) is the main method of communicating state
 367 about the L2 between the L1 and L0 via H_GUEST_{G,S}ET() and
 368 H_GUEST_VCPU_RUN() calls.
 369
 370 State may be associated with a whole L2 (eg timebase offset) or a
 371 specific L2 vCPU (eg. GPR state). Only L2 VCPU state maybe be set by
 372 H_GUEST_VCPU_RUN().
 373
 374 All data in the GSB is big endian (as is standard in PAPR)
 375
 376 The Guest state buffer has a header which gives the number of
 377 elements, followed by the GSB elements themselves.
 378
 379 GSB header:
 380
 381 +----------+----------+-------------------------------------------+
 382 |  Offset  |  Size    |  Purpose                                  |
 383 |  Bytes   |  Bytes   |                                           |
 384 +==========+==========+===========================================+
 385 |    0     |    4     |  Number of elements                       |
 386 +----------+----------+-------------------------------------------+
 387 |    4     |          |  Guest state buffer elements              |
 388 +----------+----------+-------------------------------------------+
 389
 390 GSB element:
 391
 392 +----------+----------+-------------------------------------------+
 393 |  Offset  |  Size    |  Purpose                                  |
 394 |  Bytes   |  Bytes   |                                           |
 395 +==========+==========+===========================================+
 396 |    0     |    2     |  ID                                       |
 397 +----------+----------+-------------------------------------------+
 398 |    2     |    2     |  Size of Value                            |
 399 +----------+----------+-------------------------------------------+
 400 |    4     | As above |  Value                                    |
 401 +----------+----------+-------------------------------------------+
 402
 403 The ID in the GSB element specifies what is to be set. This includes
 404 archtected state like GPRs, VSRs, SPRs, plus also some meta data about
 405 the partition like the timebase offset and partition scoped page
 406 table information.
 407
 408 +--------+-------+----+--------+----------------------------------+
 409 |   ID   | Size  | RW | Thread | Details                          |
 410 |        | Bytes |    | Guest  |                                  |
 411 |        |       |    | Scope  |                                  |
 412 +========+=======+====+========+==================================+
 413 | 0x0000 |       | RW |   TG   | NOP element                      |
 414 +--------+-------+----+--------+----------------------------------+
 415 | 0x0001 | 0x08  | R  |   G    | Size of L0 vCPU state. See:      |
 416 |        |       |    |        | H_GUEST_GET_STATE:               |
 417 |        |       |    |        | flags = takeOwnershipOfVcpuState |
 418 +--------+-------+----+--------+----------------------------------+
 419 | 0x0002 | 0x08  | R  |   G    | Size Run vCPU out buffer         |
 420 +--------+-------+----+--------+----------------------------------+
 421 | 0x0003 | 0x04  | RW |   G    | Logical PVR                      |
 422 +--------+-------+----+--------+----------------------------------+
 423 | 0x0004 | 0x08  | RW |   G    | TB Offset (L1 relative)          |
 424 +--------+-------+----+--------+----------------------------------+
 425 | 0x0005 | 0x18  | RW |   G    |Partition scoped page tbl info:   |
 426 |        |       |    |        |                                  |
 427 |        |       |    |        |- 0x00 Addr part scope table      |
 428 |        |       |    |        |- 0x08 Num addr bits              |
 429 |        |       |    |        |- 0x10 Size root dir              |
 430 +--------+-------+----+--------+----------------------------------+
 431 | 0x0006 | 0x10  | RW |   G    |Process Table Information:        |
 432 |        |       |    |        |                                  |
 433 |        |       |    |        |- 0x0 Addr proc scope table       |
 434 |        |       |    |        |- 0x8 Table size.                 |
 435 +--------+-------+----+--------+----------------------------------+
 436 | 0x0007-|       |    |        | Reserved                         |
 437 | 0x0BFF |       |    |        |                                  |
 438 +--------+-------+----+--------+----------------------------------+
 439 | 0x0C00 | 0x10  | RW |   T    |Run vCPU Input Buffer:            |
 440 |        |       |    |        |                                  |
 441 |        |       |    |        |- 0x0 Addr of buffer              |
 442 |        |       |    |        |- 0x8 Buffer Size.                |
 443 +--------+-------+----+--------+----------------------------------+
 444 | 0x0C01 | 0x10  | RW |   T    |Run vCPU Output Buffer:           |
 445 |        |       |    |        |                                  |
 446 |        |       |    |        |- 0x0 Addr of buffer              |
 447 |        |       |    |        |- 0x8 Buffer Size.                |
 448 +--------+-------+----+--------+----------------------------------+
 449 | 0x0C02 | 0x08  | RW |   T    | vCPU VPA Address                 |
 450 +--------+-------+----+--------+----------------------------------+
 451 | 0x0C03-|       |    |        | Reserved                         |
 452 | 0x0FFF |       |    |        |                                  |
 453 +--------+-------+----+--------+----------------------------------+
 454 | 0x1000-| 0x08  | RW |   T    | GPR 0-31                         |
 455 | 0x101F |       |    |        |                                  |
 456 +--------+-------+----+--------+----------------------------------+
 457 | 0x1020 |  0x08 | T  |   T    | HDEC expiry TB                   |
 458 +--------+-------+----+--------+----------------------------------+
 459 | 0x1021 | 0x08  | RW |   T    | NIA                              |
 460 +--------+-------+----+--------+----------------------------------+
 461 | 0x1022 | 0x08  | RW |   T    | MSR                              |
 462 +--------+-------+----+--------+----------------------------------+
 463 | 0x1023 | 0x08  | RW |   T    | LR                               |
 464 +--------+-------+----+--------+----------------------------------+
 465 | 0x1024 | 0x08  | RW |   T    | XER                              |
 466 +--------+-------+----+--------+----------------------------------+
 467 | 0x1025 | 0x08  | RW |   T    | CTR                              |
 468 +--------+-------+----+--------+----------------------------------+
 469 | 0x1026 | 0x08  | RW |   T    | CFAR                             |
 470 +--------+-------+----+--------+----------------------------------+
 471 | 0x1027 | 0x08  | RW |   T    | SRR0                             |
 472 +--------+-------+----+--------+----------------------------------+
 473 | 0x1028 | 0x08  | RW |   T    | SRR1                             |
 474 +--------+-------+----+--------+----------------------------------+
 475 | 0x1029 | 0x08  | RW |   T    | DAR                              |
 476 +--------+-------+----+--------+----------------------------------+
 477 | 0x102A | 0x08  | RW |   T    | DEC expiry TB                    |
 478 +--------+-------+----+--------+----------------------------------+
 479 | 0x102B | 0x08  | RW |   T    | VTB                              |
 480 +--------+-------+----+--------+----------------------------------+
 481 | 0x102C | 0x08  | RW |   T    | LPCR                             |
 482 +--------+-------+----+--------+----------------------------------+
 483 | 0x102D | 0x08  | RW |   T    | HFSCR                            |
 484 +--------+-------+----+--------+----------------------------------+
 485 | 0x102E | 0x08  | RW |   T    | FSCR                             |
 486 +--------+-------+----+--------+----------------------------------+
 487 | 0x102F | 0x08  | RW |   T    | FPSCR                            |
 488 +--------+-------+----+--------+----------------------------------+
 489 | 0x1030 | 0x08  | RW |   T    | DAWR0                            |
 490 +--------+-------+----+--------+----------------------------------+
 491 | 0x1031 | 0x08  | RW |   T    | DAWR1                            |
 492 +--------+-------+----+--------+----------------------------------+
 493 | 0x1032 | 0x08  | RW |   T    | CIABR                            |
 494 +--------+-------+----+--------+----------------------------------+
 495 | 0x1033 | 0x08  | RW |   T    | PURR                             |
 496 +--------+-------+----+--------+----------------------------------+
 497 | 0x1034 | 0x08  | RW |   T    | SPURR                            |
 498 +--------+-------+----+--------+----------------------------------+
 499 | 0x1035 | 0x08  | RW |   T    | IC                               |
 500 +--------+-------+----+--------+----------------------------------+
 501 | 0x1036-| 0x08  | RW |   T    | SPRG 0-3                         |
 502 | 0x1039 |       |    |        |                                  |
 503 +--------+-------+----+--------+----------------------------------+
 504 | 0x103A | 0x08  | W  |   T    | PPR                              |
 505 +--------+-------+----+--------+----------------------------------+
 506 | 0x103B | 0x08  | RW |   T    | MMCR 0-3                         |
 507 | 0x103E |       |    |        |                                  |
 508 +--------+-------+----+--------+----------------------------------+
 509 | 0x103F | 0x08  | RW |   T    | MMCRA                            |
 510 +--------+-------+----+--------+----------------------------------+
 511 | 0x1040 | 0x08  | RW |   T    | SIER                             |
 512 +--------+-------+----+--------+----------------------------------+
 513 | 0x1041 | 0x08  | RW |   T    | SIER 2                           |
 514 +--------+-------+----+--------+----------------------------------+
 515 | 0x1042 | 0x08  | RW |   T    | SIER 3                           |
 516 +--------+-------+----+--------+----------------------------------+
 517 | 0x1043 | 0x08  | RW |   T    | BESCR                            |
 518 +--------+-------+----+--------+----------------------------------+
 519 | 0x1044 | 0x08  | RW |   T    | EBBHR                            |
 520 +--------+-------+----+--------+----------------------------------+
 521 | 0x1045 | 0x08  | RW |   T    | EBBRR                            |
 522 +--------+-------+----+--------+----------------------------------+
 523 | 0x1046 | 0x08  | RW |   T    | AMR                              |
 524 +--------+-------+----+--------+----------------------------------+
 525 | 0x1047 | 0x08  | RW |   T    | IAMR                             |
 526 +--------+-------+----+--------+----------------------------------+
 527 | 0x1048 | 0x08  | RW |   T    | AMOR                             |
 528 +--------+-------+----+--------+----------------------------------+
 529 | 0x1049 | 0x08  | RW |   T    | UAMOR                            |
 530 +--------+-------+----+--------+----------------------------------+
 531 | 0x104A | 0x08  | RW |   T    | SDAR                             |
 532 +--------+-------+----+--------+----------------------------------+
 533 | 0x104B | 0x08  | RW |   T    | SIAR                             |
 534 +--------+-------+----+--------+----------------------------------+
 535 | 0x104C | 0x08  | RW |   T    | DSCR                             |
 536 +--------+-------+----+--------+----------------------------------+
 537 | 0x104D | 0x08  | RW |   T    | TAR                              |
 538 +--------+-------+----+--------+----------------------------------+
 539 | 0x104E | 0x08  | RW |   T    | DEXCR                            |
 540 +--------+-------+----+--------+----------------------------------+
 541 | 0x104F | 0x08  | RW |   T    | HDEXCR                           |
 542 +--------+-------+----+--------+----------------------------------+
 543 | 0x1050 | 0x08  | RW |   T    | HASHKEYR                         |
 544 +--------+-------+----+--------+----------------------------------+
 545 | 0x1051 | 0x08  | RW |   T    | HASHPKEYR                        |
 546 +--------+-------+----+--------+----------------------------------+
 547 | 0x1052 | 0x08  | RW |   T    | CTRL                             |
 548 +--------+-------+----+--------+----------------------------------+
 549 | 0x1053-|       |    |        | Reserved                         |
 550 | 0x1FFF |       |    |        |                                  |
 551 +--------+-------+----+--------+----------------------------------+
 552 | 0x2000 | 0x04  | RW |   T    | CR                               |
 553 +--------+-------+----+--------+----------------------------------+
 554 | 0x2001 | 0x04  | RW |   T    | PIDR                             |
 555 +--------+-------+----+--------+----------------------------------+
 556 | 0x2002 | 0x04  | RW |   T    | DSISR                            |
 557 +--------+-------+----+--------+----------------------------------+
 558 | 0x2003 | 0x04  | RW |   T    | VSCR                             |
 559 +--------+-------+----+--------+----------------------------------+
 560 | 0x2004 | 0x04  | RW |   T    | VRSAVE                           |
 561 +--------+-------+----+--------+----------------------------------+
 562 | 0x2005 | 0x04  | RW |   T    | DAWRX0                           |
 563 +--------+-------+----+--------+----------------------------------+
 564 | 0x2006 | 0x04  | RW |   T    | DAWRX1                           |
 565 +--------+-------+----+--------+----------------------------------+
 566 | 0x2007-| 0x04  | RW |   T    | PMC 1-6                          |
 567 | 0x200c |       |    |        |                                  |
 568 +--------+-------+----+--------+----------------------------------+
 569 | 0x200D | 0x04  | RW |   T    | WORT                             |
 570 +--------+-------+----+--------+----------------------------------+
 571 | 0x200E | 0x04  | RW |   T    | PSPB                             |
 572 +--------+-------+----+--------+----------------------------------+
 573 | 0x200F-|       |    |        | Reserved                         |
 574 | 0x2FFF |       |    |        |                                  |
 575 +--------+-------+----+--------+----------------------------------+
 576 | 0x3000-| 0x10  | RW |   T    | VSR 0-63                         |
 577 | 0x303F |       |    |        |                                  |
 578 +--------+-------+----+--------+----------------------------------+
 579 | 0x3040-|       |    |        | Reserved                         |
 580 | 0xEFFF |       |    |        |                                  |
 581 +--------+-------+----+--------+----------------------------------+
 582 | 0xF000 | 0x08  | R  |   T    | HDAR                             |
 583 +--------+-------+----+--------+----------------------------------+
 584 | 0xF001 | 0x04  | R  |   T    | HDSISR                           |
 585 +--------+-------+----+--------+----------------------------------+
 586 | 0xF002 | 0x04  | R  |   T    | HEIR                             |
 587 +--------+-------+----+--------+----------------------------------+
 588 | 0xF003 | 0x08  | R  |   T    | ASDR                             |
 589 +--------+-------+----+--------+----------------------------------+
 590
 591
 592 Miscellaneous info
 593 ==================
 594
 595 State not in ptregs/hvregs
 596 --------------------------
 597
 598 In the v1 API, some state is not in the ptregs/hvstate. This includes
 599 the vector register and some SPRs. For the L1 to set this state for
 600 the L2, the L1 loads up these hardware registers before the
 601 h_enter_nested() call and the L0 ensures they end up as the L2 state
 602 (by not touching them).
 603
 604 The v2 API removes this and explicitly sets this state via the GSB.
 605
 606 L1 Implementation details: Caching state
 607 ----------------------------------------
 608
 609 In the v1 API, all state is sent from the L1 to the L0 and vice versa
 610 on every h_enter_nested() hcall. If the L0 is not currently running
 611 any L2s, the L0 has no state information about them. The only
 612 exception to this is the location of the partition table, registered
 613 via h_set_partition_table().
 614
 615 The v2 API changes this so that the L0 retains the L2 state even when
 616 it's vCPUs are no longer running. This means that the L1 only needs to
 617 communicate with the L0 about L2 state when it needs to modify the L2
 618 state, or when it's value is out of date. This provides an opportunity
 619 for performance optimisation.
 620
 621 When a vCPU exits from a H_GUEST_RUN_VCPU() call, the L1 internally
 622 marks all L2 state as invalid. This means that if the L1 wants to know
 623 the L2 state (say via a kvm_get_one_reg() call), it needs call
 624 H_GUEST_GET_STATE() to get that state. Once it's read, it's marked as
 625 valid in L1 until the L2 is run again.
 626
 627 Also, when an L1 modifies L2 vcpu state, it doesn't need to write it
 628 to the L0 until that L2 vcpu runs again. Hence when the L1 updates
 629 state (say via a kvm_set_one_reg() call), it writes to an internal L1
 630 copy and only flushes this copy to the L0 when the L2 runs again via
 631 the H_GUEST_VCPU_RUN() input buffer.
 632
 633 This lazy updating of state by the L1 avoids unnecessary
 634 H_GUEST_{G|S}ET_STATE() calls.