include/sysemu/kvm.h

   1 /*
   2  * QEMU KVM support
   3  *
   4  * Copyright IBM, Corp. 2008
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  10  * See the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 /* header to be included in non-KVM-specific code */
  15
  16 #ifndef QEMU_KVM_H
  17 #define QEMU_KVM_H
  18
  19 #include "qemu/accel.h"
  20 #include "qom/object.h"
  21
  22 #ifdef NEED_CPU_H
  23 # ifdef CONFIG_KVM
  24 #  include <linux/kvm.h>
  25 #  define CONFIG_KVM_IS_POSSIBLE
  26 # endif
  27 #else
  28 # define CONFIG_KVM_IS_POSSIBLE
  29 #endif
  30
  31 #ifdef CONFIG_KVM_IS_POSSIBLE
  32
  33 extern bool kvm_allowed;
  34 extern bool kvm_kernel_irqchip;
  35 extern bool kvm_split_irqchip;
  36 extern bool kvm_async_interrupts_allowed;
  37 extern bool kvm_halt_in_kernel_allowed;
  38 extern bool kvm_eventfds_allowed;
  39 extern bool kvm_irqfds_allowed;
  40 extern bool kvm_resamplefds_allowed;
  41 extern bool kvm_msi_via_irqfd_allowed;
  42 extern bool kvm_gsi_routing_allowed;
  43 extern bool kvm_gsi_direct_mapping;
  44 extern bool kvm_readonly_mem_allowed;
  45 extern bool kvm_direct_msi_allowed;
  46 extern bool kvm_ioeventfd_any_length_allowed;
  47 extern bool kvm_msi_use_devid;
  48
  49 #define kvm_enabled()           (kvm_allowed)
  50 /**
  51  * kvm_irqchip_in_kernel:
  52  *
  53  * Returns: true if an in-kernel irqchip was created.
  54  * What this actually means is architecture and machine model
  55  * specific: on PC, for instance, it means that the LAPIC
  56  * is in kernel.  This function should never be used from generic
  57  * target-independent code: use one of the following functions or
  58  * some other specific check instead.
  59  */
  60 #define kvm_irqchip_in_kernel() (kvm_kernel_irqchip)
  61
  62 /**
  63  * kvm_irqchip_is_split:
  64  *
  65  * Returns: true if the irqchip implementation is split between
  66  * user and kernel space.  The details are architecture and
  67  * machine specific.  On PC, it means that the PIC, IOAPIC, and
  68  * PIT are in user space while the LAPIC is in the kernel.
  69  */
  70 #define kvm_irqchip_is_split() (kvm_split_irqchip)
  71
  72 /**
  73  * kvm_async_interrupts_enabled:
  74  *
  75  * Returns: true if we can deliver interrupts to KVM
  76  * asynchronously (ie by ioctl from any thread at any time)
  77  * rather than having to do interrupt delivery synchronously
  78  * (where the vcpu must be stopped at a suitable point first).
  79  */
  80 #define kvm_async_interrupts_enabled() (kvm_async_interrupts_allowed)
  81
  82 /**
  83  * kvm_halt_in_kernel
  84  *
  85  * Returns: true if halted cpus should still get a KVM_RUN ioctl to run
  86  * inside of kernel space. This only works if MP state is implemented.
  87  */
  88 #define kvm_halt_in_kernel() (kvm_halt_in_kernel_allowed)
  89
  90 /**
  91  * kvm_eventfds_enabled:
  92  *
  93  * Returns: true if we can use eventfds to receive notifications
  94  * from a KVM CPU (ie the kernel supports eventds and we are running
  95  * with a configuration where it is meaningful to use them).
  96  */
  97 #define kvm_eventfds_enabled() (kvm_eventfds_allowed)
  98
  99 /**
 100  * kvm_irqfds_enabled:
 101  *
 102  * Returns: true if we can use irqfds to inject interrupts into
 103  * a KVM CPU (ie the kernel supports irqfds and we are running
 104  * with a configuration where it is meaningful to use them).
 105  */
 106 #define kvm_irqfds_enabled() (kvm_irqfds_allowed)
 107
 108 /**
 109  * kvm_resamplefds_enabled:
 110  *
 111  * Returns: true if we can use resamplefds to inject interrupts into
 112  * a KVM CPU (ie the kernel supports resamplefds and we are running
 113  * with a configuration where it is meaningful to use them).
 114  */
 115 #define kvm_resamplefds_enabled() (kvm_resamplefds_allowed)
 116
 117 /**
 118  * kvm_msi_via_irqfd_enabled:
 119  *
 120  * Returns: true if we can route a PCI MSI (Message Signaled Interrupt)
 121  * to a KVM CPU via an irqfd. This requires that the kernel supports
 122  * this and that we're running in a configuration that permits it.
 123  */
 124 #define kvm_msi_via_irqfd_enabled() (kvm_msi_via_irqfd_allowed)
 125
 126 /**
 127  * kvm_gsi_routing_enabled:
 128  *
 129  * Returns: true if GSI routing is enabled (ie the kernel supports
 130  * it and we're running in a configuration that permits it).
 131  */
 132 #define kvm_gsi_routing_enabled() (kvm_gsi_routing_allowed)
 133
 134 /**
 135  * kvm_gsi_direct_mapping:
 136  *
 137  * Returns: true if GSI direct mapping is enabled.
 138  */
 139 #define kvm_gsi_direct_mapping() (kvm_gsi_direct_mapping)
 140
 141 /**
 142  * kvm_readonly_mem_enabled:
 143  *
 144  * Returns: true if KVM readonly memory is enabled (ie the kernel
 145  * supports it and we're running in a configuration that permits it).
 146  */
 147 #define kvm_readonly_mem_enabled() (kvm_readonly_mem_allowed)
 148
 149 /**
 150  * kvm_direct_msi_enabled:
 151  *
 152  * Returns: true if KVM allows direct MSI injection.
 153  */
 154 #define kvm_direct_msi_enabled() (kvm_direct_msi_allowed)
 155
 156 /**
 157  * kvm_ioeventfd_any_length_enabled:
 158  * Returns: true if KVM allows any length io eventfd.
 159  */
 160 #define kvm_ioeventfd_any_length_enabled() (kvm_ioeventfd_any_length_allowed)
 161
 162 /**
 163  * kvm_msi_devid_required:
 164  * Returns: true if KVM requires a device id to be provided while
 165  * defining an MSI routing entry.
 166  */
 167 #define kvm_msi_devid_required() (kvm_msi_use_devid)
 168
 169 #else
 170
 171 #define kvm_enabled()           (0)
 172 #define kvm_irqchip_in_kernel() (false)
 173 #define kvm_irqchip_is_split() (false)
 174 #define kvm_async_interrupts_enabled() (false)
 175 #define kvm_halt_in_kernel() (false)
 176 #define kvm_eventfds_enabled() (false)
 177 #define kvm_irqfds_enabled() (false)
 178 #define kvm_resamplefds_enabled() (false)
 179 #define kvm_msi_via_irqfd_enabled() (false)
 180 #define kvm_gsi_routing_allowed() (false)
 181 #define kvm_gsi_direct_mapping() (false)
 182 #define kvm_readonly_mem_enabled() (false)
 183 #define kvm_direct_msi_enabled() (false)
 184 #define kvm_ioeventfd_any_length_enabled() (false)
 185 #define kvm_msi_devid_required() (false)
 186
 187 #endif  /* CONFIG_KVM_IS_POSSIBLE */
 188
 189 struct kvm_run;
 190 struct kvm_lapic_state;
 191 struct kvm_irq_routing_entry;
 192
 193 typedef struct KVMCapabilityInfo {
 194     const char *name;
 195     int value;
 196 } KVMCapabilityInfo;
 197
 198 #define KVM_CAP_INFO(CAP) { "KVM_CAP_" stringify(CAP), KVM_CAP_##CAP }
 199 #define KVM_CAP_LAST_INFO { NULL, 0 }
 200
 201 struct KVMState;
 202
 203 #define TYPE_KVM_ACCEL ACCEL_CLASS_NAME("kvm")
 204 typedef struct KVMState KVMState;
 205 DECLARE_INSTANCE_CHECKER(KVMState, KVM_STATE,
 206                          TYPE_KVM_ACCEL)
 207
 208 extern KVMState *kvm_state;
 209 typedef struct Notifier Notifier;
 210
 211 typedef struct KVMRouteChange {
 212      KVMState *s;
 213      int changes;
 214 } KVMRouteChange;
 215
 216 /* external API */
 217
 218 bool kvm_has_free_slot(MachineState *ms);
 219 bool kvm_has_sync_mmu(void);
 220 int kvm_has_vcpu_events(void);
 221 int kvm_has_robust_singlestep(void);
 222 int kvm_has_debugregs(void);
 223 int kvm_max_nested_state_length(void);
 224 int kvm_has_pit_state2(void);
 225 int kvm_has_many_ioeventfds(void);
 226 int kvm_has_gsi_routing(void);
 227 int kvm_has_intx_set_mask(void);
 228
 229 /**
 230  * kvm_arm_supports_user_irq
 231  *
 232  * Not all KVM implementations support notifications for kernel generated
 233  * interrupt events to user space. This function indicates whether the current
 234  * KVM implementation does support them.
 235  *
 236  * Returns: true if KVM supports using kernel generated IRQs from user space
 237  */
 238 bool kvm_arm_supports_user_irq(void);
 239
 240
 241 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr);
 242 int kvm_on_sigbus(int code, void *addr);
 243
 244 #ifdef NEED_CPU_H
 245 #include "cpu.h"
 246
 247 void kvm_flush_coalesced_mmio_buffer(void);
 248
 249 /**
 250  * kvm_update_guest_debug(): ensure KVM debug structures updated
 251  * @cs: the CPUState for this cpu
 252  * @reinject_trap: KVM trap injection control
 253  *
 254  * There are usually per-arch specifics which will be handled by
 255  * calling down to kvm_arch_update_guest_debug after the generic
 256  * fields have been set.
 257  */
 258 #ifdef KVM_CAP_SET_GUEST_DEBUG
 259 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap);
 260 #else
 261 static inline int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
 262 {
 263     return -EINVAL;
 264 }
 265 #endif
 266
 267 /* internal API */
 268
 269 int kvm_ioctl(KVMState *s, int type, ...);
 270
 271 int kvm_vm_ioctl(KVMState *s, int type, ...);
 272
 273 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...);
 274
 275 /**
 276  * kvm_device_ioctl - call an ioctl on a kvm device
 277  * @fd: The KVM device file descriptor as returned from KVM_CREATE_DEVICE
 278  * @type: The device-ctrl ioctl number
 279  *
 280  * Returns: -errno on error, nonnegative on success
 281  */
 282 int kvm_device_ioctl(int fd, int type, ...);
 283
 284 /**
 285  * kvm_vm_check_attr - check for existence of a specific vm attribute
 286  * @s: The KVMState pointer
 287  * @group: the group
 288  * @attr: the attribute of that group to query for
 289  *
 290  * Returns: 1 if the attribute exists
 291  *          0 if the attribute either does not exist or if the vm device
 292  *            interface is unavailable
 293  */
 294 int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr);
 295
 296 /**
 297  * kvm_device_check_attr - check for existence of a specific device attribute
 298  * @fd: The device file descriptor
 299  * @group: the group
 300  * @attr: the attribute of that group to query for
 301  *
 302  * Returns: 1 if the attribute exists
 303  *          0 if the attribute either does not exist or if the vm device
 304  *            interface is unavailable
 305  */
 306 int kvm_device_check_attr(int fd, uint32_t group, uint64_t attr);
 307
 308 /**
 309  * kvm_device_access - set or get value of a specific device attribute
 310  * @fd: The device file descriptor
 311  * @group: the group
 312  * @attr: the attribute of that group to set or get
 313  * @val: pointer to a storage area for the value
 314  * @write: true for set and false for get operation
 315  * @errp: error object handle
 316  *
 317  * Returns: 0 on success
 318  *          < 0 on error
 319  * Use kvm_device_check_attr() in order to check for the availability
 320  * of optional attributes.
 321  */
 322 int kvm_device_access(int fd, int group, uint64_t attr,
 323                       void *val, bool write, Error **errp);
 324
 325 /**
 326  * kvm_create_device - create a KVM device for the device control API
 327  * @KVMState: The KVMState pointer
 328  * @type: The KVM device type (see Documentation/virtual/kvm/devices in the
 329  *        kernel source)
 330  * @test: If true, only test if device can be created, but don't actually
 331  *        create the device.
 332  *
 333  * Returns: -errno on error, nonnegative on success: @test ? 0 : device fd;
 334  */
 335 int kvm_create_device(KVMState *s, uint64_t type, bool test);
 336
 337 /**
 338  * kvm_device_supported - probe whether KVM supports specific device
 339  *
 340  * @vmfd: The fd handler for VM
 341  * @type: type of device
 342  *
 343  * @return: true if supported, otherwise false.
 344  */
 345 bool kvm_device_supported(int vmfd, uint64_t type);
 346
 347 /* Arch specific hooks */
 348
 349 extern const KVMCapabilityInfo kvm_arch_required_capabilities[];
 350
 351 void kvm_arch_accel_class_init(ObjectClass *oc);
 352
 353 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run);
 354 MemTxAttrs kvm_arch_post_run(CPUState *cpu, struct kvm_run *run);
 355
 356 int kvm_arch_handle_exit(CPUState *cpu, struct kvm_run *run);
 357
 358 int kvm_arch_process_async_events(CPUState *cpu);
 359
 360 int kvm_arch_get_registers(CPUState *cpu);
 361
 362 /* state subset only touched by the VCPU itself during runtime */
 363 #define KVM_PUT_RUNTIME_STATE   1
 364 /* state subset modified during VCPU reset */
 365 #define KVM_PUT_RESET_STATE     2
 366 /* full state set, modified during initialization or on vmload */
 367 #define KVM_PUT_FULL_STATE      3
 368
 369 int kvm_arch_put_registers(CPUState *cpu, int level);
 370
 371 int kvm_arch_init(MachineState *ms, KVMState *s);
 372
 373 int kvm_arch_init_vcpu(CPUState *cpu);
 374 int kvm_arch_destroy_vcpu(CPUState *cpu);
 375
 376 bool kvm_vcpu_id_is_valid(int vcpu_id);
 377
 378 /* Returns VCPU ID to be used on KVM_CREATE_VCPU ioctl() */
 379 unsigned long kvm_arch_vcpu_id(CPUState *cpu);
 380
 381 #ifdef KVM_HAVE_MCE_INJECTION
 382 void kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr);
 383 #endif
 384
 385 void kvm_arch_init_irq_routing(KVMState *s);
 386
 387 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
 388                              uint64_t address, uint32_t data, PCIDevice *dev);
 389
 390 /* Notify arch about newly added MSI routes */
 391 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
 392                                 int vector, PCIDevice *dev);
 393 /* Notify arch about released MSI routes */
 394 int kvm_arch_release_virq_post(int virq);
 395
 396 int kvm_arch_msi_data_to_gsi(uint32_t data);
 397
 398 int kvm_set_irq(KVMState *s, int irq, int level);
 399 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg);
 400
 401 void kvm_irqchip_add_irq_route(KVMState *s, int gsi, int irqchip, int pin);
 402
 403 void kvm_irqchip_add_change_notifier(Notifier *n);
 404 void kvm_irqchip_remove_change_notifier(Notifier *n);
 405 void kvm_irqchip_change_notify(void);
 406
 407 void kvm_get_apic_state(DeviceState *d, struct kvm_lapic_state *kapic);
 408
 409 struct kvm_guest_debug;
 410 struct kvm_debug_exit_arch;
 411
 412 struct kvm_sw_breakpoint {
 413     target_ulong pc;
 414     target_ulong saved_insn;
 415     int use_count;
 416     QTAILQ_ENTRY(kvm_sw_breakpoint) entry;
 417 };
 418
 419 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
 420                                                  target_ulong pc);
 421
 422 int kvm_sw_breakpoints_active(CPUState *cpu);
 423
 424 int kvm_arch_insert_sw_breakpoint(CPUState *cpu,
 425                                   struct kvm_sw_breakpoint *bp);
 426 int kvm_arch_remove_sw_breakpoint(CPUState *cpu,
 427                                   struct kvm_sw_breakpoint *bp);
 428 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
 429                                   target_ulong len, int type);
 430 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
 431                                   target_ulong len, int type);
 432 void kvm_arch_remove_all_hw_breakpoints(void);
 433
 434 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg);
 435
 436 bool kvm_arch_stop_on_emulation_error(CPUState *cpu);
 437
 438 int kvm_check_extension(KVMState *s, unsigned int extension);
 439
 440 int kvm_vm_check_extension(KVMState *s, unsigned int extension);
 441
 442 #define kvm_vm_enable_cap(s, capability, cap_flags, ...)             \
 443     ({                                                               \
 444         struct kvm_enable_cap cap = {                                \
 445             .cap = capability,                                       \
 446             .flags = cap_flags,                                      \
 447         };                                                           \
 448         uint64_t args_tmp[] = { __VA_ARGS__ };                       \
 449         size_t n = MIN(ARRAY_SIZE(args_tmp), ARRAY_SIZE(cap.args));  \
 450         memcpy(cap.args, args_tmp, n * sizeof(cap.args[0]));         \
 451         kvm_vm_ioctl(s, KVM_ENABLE_CAP, &cap);                       \
 452     })
 453
 454 #define kvm_vcpu_enable_cap(cpu, capability, cap_flags, ...)         \
 455     ({                                                               \
 456         struct kvm_enable_cap cap = {                                \
 457             .cap = capability,                                       \
 458             .flags = cap_flags,                                      \
 459         };                                                           \
 460         uint64_t args_tmp[] = { __VA_ARGS__ };                       \
 461         size_t n = MIN(ARRAY_SIZE(args_tmp), ARRAY_SIZE(cap.args));  \
 462         memcpy(cap.args, args_tmp, n * sizeof(cap.args[0]));         \
 463         kvm_vcpu_ioctl(cpu, KVM_ENABLE_CAP, &cap);                   \
 464     })
 465
 466 uint32_t kvm_arch_get_supported_cpuid(KVMState *env, uint32_t function,
 467                                       uint32_t index, int reg);
 468 uint64_t kvm_arch_get_supported_msr_feature(KVMState *s, uint32_t index);
 469
 470
 471 void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len);
 472
 473 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram_addr,
 474                                        hwaddr *phys_addr);
 475
 476 #endif /* NEED_CPU_H */
 477
 478 void kvm_cpu_synchronize_state(CPUState *cpu);
 479
 480 void kvm_init_cpu_signals(CPUState *cpu);
 481
 482 /**
 483  * kvm_irqchip_add_msi_route - Add MSI route for specific vector
 484  * @c:      KVMRouteChange instance.
 485  * @vector: which vector to add. This can be either MSI/MSIX
 486  *          vector. The function will automatically detect whether
 487  *          MSI/MSIX is enabled, and fetch corresponding MSI
 488  *          message.
 489  * @dev:    Owner PCI device to add the route. If @dev is specified
 490  *          as @NULL, an empty MSI message will be inited.
 491  * @return: virq (>=0) when success, errno (<0) when failed.
 492  */
 493 int kvm_irqchip_add_msi_route(KVMRouteChange *c, int vector, PCIDevice *dev);
 494 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg,
 495                                  PCIDevice *dev);
 496 void kvm_irqchip_commit_routes(KVMState *s);
 497
 498 static inline KVMRouteChange kvm_irqchip_begin_route_changes(KVMState *s)
 499 {
 500     return (KVMRouteChange) { .s = s, .changes = 0 };
 501 }
 502
 503 static inline void kvm_irqchip_commit_route_changes(KVMRouteChange *c)
 504 {
 505     if (c->changes) {
 506         kvm_irqchip_commit_routes(c->s);
 507         c->changes = 0;
 508     }
 509 }
 510
 511 void kvm_irqchip_release_virq(KVMState *s, int virq);
 512
 513 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter);
 514 int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint);
 515
 516 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
 517                                        EventNotifier *rn, int virq);
 518 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
 519                                           int virq);
 520 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
 521                                    EventNotifier *rn, qemu_irq irq);
 522 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n,
 523                                       qemu_irq irq);
 524 void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi);
 525 void kvm_pc_setup_irq_routing(bool pci_enabled);
 526 void kvm_init_irq_routing(KVMState *s);
 527
 528 bool kvm_kernel_irqchip_allowed(void);
 529 bool kvm_kernel_irqchip_required(void);
 530 bool kvm_kernel_irqchip_split(void);
 531
 532 /**
 533  * kvm_arch_irqchip_create:
 534  * @KVMState: The KVMState pointer
 535  *
 536  * Allow architectures to create an in-kernel irq chip themselves.
 537  *
 538  * Returns: < 0: error
 539  *            0: irq chip was not created
 540  *          > 0: irq chip was created
 541  */
 542 int kvm_arch_irqchip_create(KVMState *s);
 543
 544 /**
 545  * kvm_set_one_reg - set a register value in KVM via KVM_SET_ONE_REG ioctl
 546  * @id: The register ID
 547  * @source: The pointer to the value to be set. It must point to a variable
 548  *          of the correct type/size for the register being accessed.
 549  *
 550  * Returns: 0 on success, or a negative errno on failure.
 551  */
 552 int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source);
 553
 554 /**
 555  * kvm_get_one_reg - get a register value from KVM via KVM_GET_ONE_REG ioctl
 556  * @id: The register ID
 557  * @target: The pointer where the value is to be stored. It must point to a
 558  *          variable of the correct type/size for the register being accessed.
 559  *
 560  * Returns: 0 on success, or a negative errno on failure.
 561  */
 562 int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target);
 563 struct ppc_radix_page_info *kvm_get_radix_page_info(void);
 564 int kvm_get_max_memslots(void);
 565
 566 /* Notify resamplefd for EOI of specific interrupts. */
 567 void kvm_resample_fd_notify(int gsi);
 568
 569 /**
 570  * kvm_cpu_check_are_resettable - return whether CPUs can be reset
 571  *
 572  * Returns: true: CPUs are resettable
 573  *          false: CPUs are not resettable
 574  */
 575 bool kvm_cpu_check_are_resettable(void);
 576
 577 bool kvm_arch_cpu_check_are_resettable(void);
 578
 579 bool kvm_dirty_ring_enabled(void);
 580
 581 uint32_t kvm_dirty_ring_size(void);
 582 #endif