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 #ifndef QEMU_KVM_H 15 #define QEMU_KVM_H 16 17 #include "qemu/queue.h" 18 #include "hw/core/cpu.h" 19 #include "exec/memattrs.h" 20 #include "sysemu/accel.h" 21 #include "qom/object.h" 22 23 #ifdef NEED_CPU_H 24 # ifdef CONFIG_KVM 25 # include <linux/kvm.h> 26 # define CONFIG_KVM_IS_POSSIBLE 27 # endif 28 #else 29 # define CONFIG_KVM_IS_POSSIBLE 30 #endif 31 32 #ifdef CONFIG_KVM_IS_POSSIBLE 33 34 extern bool kvm_allowed; 35 extern bool kvm_kernel_irqchip; 36 extern bool kvm_split_irqchip; 37 extern bool kvm_async_interrupts_allowed; 38 extern bool kvm_halt_in_kernel_allowed; 39 extern bool kvm_eventfds_allowed; 40 extern bool kvm_irqfds_allowed; 41 extern bool kvm_resamplefds_allowed; 42 extern bool kvm_msi_via_irqfd_allowed; 43 extern bool kvm_gsi_routing_allowed; 44 extern bool kvm_gsi_direct_mapping; 45 extern bool kvm_readonly_mem_allowed; 46 extern bool kvm_direct_msi_allowed; 47 extern bool kvm_ioeventfd_any_length_allowed; 48 extern bool kvm_msi_use_devid; 49 50 #define kvm_enabled() (kvm_allowed) 51 /** 52 * kvm_irqchip_in_kernel: 53 * 54 * Returns: true if the user asked us to create an in-kernel 55 * irqchip via the "kernel_irqchip=on" machine option. 56 * What this actually means is architecture and machine model 57 * specific: on PC, for instance, it means that the LAPIC, 58 * IOAPIC and PIT are all in kernel. This function should never 59 * be used from generic target-independent code: use one of the 60 * following functions or some other specific check instead. 61 */ 62 #define kvm_irqchip_in_kernel() (kvm_kernel_irqchip) 63 64 /** 65 * kvm_irqchip_is_split: 66 * 67 * Returns: true if the user asked us to split the irqchip 68 * implementation between user and kernel space. The details are 69 * architecture and machine specific. On PC, it means that the PIC, 70 * IOAPIC, and PIT are in user space while the LAPIC is in the kernel. 71 */ 72 #define kvm_irqchip_is_split() (kvm_split_irqchip) 73 74 /** 75 * kvm_async_interrupts_enabled: 76 * 77 * Returns: true if we can deliver interrupts to KVM 78 * asynchronously (ie by ioctl from any thread at any time) 79 * rather than having to do interrupt delivery synchronously 80 * (where the vcpu must be stopped at a suitable point first). 81 */ 82 #define kvm_async_interrupts_enabled() (kvm_async_interrupts_allowed) 83 84 /** 85 * kvm_halt_in_kernel 86 * 87 * Returns: true if halted cpus should still get a KVM_RUN ioctl to run 88 * inside of kernel space. This only works if MP state is implemented. 89 */ 90 #define kvm_halt_in_kernel() (kvm_halt_in_kernel_allowed) 91 92 /** 93 * kvm_eventfds_enabled: 94 * 95 * Returns: true if we can use eventfds to receive notifications 96 * from a KVM CPU (ie the kernel supports eventds and we are running 97 * with a configuration where it is meaningful to use them). 98 */ 99 #define kvm_eventfds_enabled() (kvm_eventfds_allowed) 100 101 /** 102 * kvm_irqfds_enabled: 103 * 104 * Returns: true if we can use irqfds to inject interrupts into 105 * a KVM CPU (ie the kernel supports irqfds and we are running 106 * with a configuration where it is meaningful to use them). 107 */ 108 #define kvm_irqfds_enabled() (kvm_irqfds_allowed) 109 110 /** 111 * kvm_resamplefds_enabled: 112 * 113 * Returns: true if we can use resamplefds to inject interrupts into 114 * a KVM CPU (ie the kernel supports resamplefds and we are running 115 * with a configuration where it is meaningful to use them). 116 */ 117 #define kvm_resamplefds_enabled() (kvm_resamplefds_allowed) 118 119 /** 120 * kvm_msi_via_irqfd_enabled: 121 * 122 * Returns: true if we can route a PCI MSI (Message Signaled Interrupt) 123 * to a KVM CPU via an irqfd. This requires that the kernel supports 124 * this and that we're running in a configuration that permits it. 125 */ 126 #define kvm_msi_via_irqfd_enabled() (kvm_msi_via_irqfd_allowed) 127 128 /** 129 * kvm_gsi_routing_enabled: 130 * 131 * Returns: true if GSI routing is enabled (ie the kernel supports 132 * it and we're running in a configuration that permits it). 133 */ 134 #define kvm_gsi_routing_enabled() (kvm_gsi_routing_allowed) 135 136 /** 137 * kvm_gsi_direct_mapping: 138 * 139 * Returns: true if GSI direct mapping is enabled. 140 */ 141 #define kvm_gsi_direct_mapping() (kvm_gsi_direct_mapping) 142 143 /** 144 * kvm_readonly_mem_enabled: 145 * 146 * Returns: true if KVM readonly memory is enabled (ie the kernel 147 * supports it and we're running in a configuration that permits it). 148 */ 149 #define kvm_readonly_mem_enabled() (kvm_readonly_mem_allowed) 150 151 /** 152 * kvm_direct_msi_enabled: 153 * 154 * Returns: true if KVM allows direct MSI injection. 155 */ 156 #define kvm_direct_msi_enabled() (kvm_direct_msi_allowed) 157 158 /** 159 * kvm_ioeventfd_any_length_enabled: 160 * Returns: true if KVM allows any length io eventfd. 161 */ 162 #define kvm_ioeventfd_any_length_enabled() (kvm_ioeventfd_any_length_allowed) 163 164 /** 165 * kvm_msi_devid_required: 166 * Returns: true if KVM requires a device id to be provided while 167 * defining an MSI routing entry. 168 */ 169 #define kvm_msi_devid_required() (kvm_msi_use_devid) 170 171 #else 172 173 #define kvm_enabled() (0) 174 #define kvm_irqchip_in_kernel() (false) 175 #define kvm_irqchip_is_split() (false) 176 #define kvm_async_interrupts_enabled() (false) 177 #define kvm_halt_in_kernel() (false) 178 #define kvm_eventfds_enabled() (false) 179 #define kvm_irqfds_enabled() (false) 180 #define kvm_resamplefds_enabled() (false) 181 #define kvm_msi_via_irqfd_enabled() (false) 182 #define kvm_gsi_routing_allowed() (false) 183 #define kvm_gsi_direct_mapping() (false) 184 #define kvm_readonly_mem_enabled() (false) 185 #define kvm_direct_msi_enabled() (false) 186 #define kvm_ioeventfd_any_length_enabled() (false) 187 #define kvm_msi_devid_required() (false) 188 189 #endif /* CONFIG_KVM_IS_POSSIBLE */ 190 191 struct kvm_run; 192 struct kvm_lapic_state; 193 struct kvm_irq_routing_entry; 194 195 typedef struct KVMCapabilityInfo { 196 const char *name; 197 int value; 198 } KVMCapabilityInfo; 199 200 #define KVM_CAP_INFO(CAP) { "KVM_CAP_" stringify(CAP), KVM_CAP_##CAP } 201 #define KVM_CAP_LAST_INFO { NULL, 0 } 202 203 struct KVMState; 204 205 #define TYPE_KVM_ACCEL ACCEL_CLASS_NAME("kvm") 206 typedef struct KVMState KVMState; 207 DECLARE_INSTANCE_CHECKER(KVMState, KVM_STATE, 208 TYPE_KVM_ACCEL) 209 210 extern KVMState *kvm_state; 211 typedef struct Notifier Notifier; 212 213 /* external API */ 214 215 bool kvm_has_free_slot(MachineState *ms); 216 bool kvm_has_sync_mmu(void); 217 int kvm_has_vcpu_events(void); 218 int kvm_has_robust_singlestep(void); 219 int kvm_has_debugregs(void); 220 int kvm_max_nested_state_length(void); 221 int kvm_has_pit_state2(void); 222 int kvm_has_many_ioeventfds(void); 223 int kvm_has_gsi_routing(void); 224 int kvm_has_intx_set_mask(void); 225 226 /** 227 * kvm_arm_supports_user_irq 228 * 229 * Not all KVM implementations support notifications for kernel generated 230 * interrupt events to user space. This function indicates whether the current 231 * KVM implementation does support them. 232 * 233 * Returns: true if KVM supports using kernel generated IRQs from user space 234 */ 235 bool kvm_arm_supports_user_irq(void); 236 237 /** 238 * kvm_memcrypt_enabled - return boolean indicating whether memory encryption 239 * is enabled 240 * Returns: 1 memory encryption is enabled 241 * 0 memory encryption is disabled 242 */ 243 bool kvm_memcrypt_enabled(void); 244 245 /** 246 * kvm_memcrypt_encrypt_data: encrypt the memory range 247 * 248 * Return: 1 failed to encrypt the range 249 * 0 succesfully encrypted memory region 250 */ 251 int kvm_memcrypt_encrypt_data(uint8_t *ptr, uint64_t len); 252 253 254 #ifdef NEED_CPU_H 255 #include "cpu.h" 256 257 void kvm_flush_coalesced_mmio_buffer(void); 258 259 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr, 260 target_ulong len, int type); 261 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr, 262 target_ulong len, int type); 263 void kvm_remove_all_breakpoints(CPUState *cpu); 264 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap); 265 266 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr); 267 int kvm_on_sigbus(int code, void *addr); 268 269 /* interface with exec.c */ 270 271 void phys_mem_set_alloc(void *(*alloc)(size_t, uint64_t *align, bool shared)); 272 273 /* internal API */ 274 275 int kvm_ioctl(KVMState *s, int type, ...); 276 277 int kvm_vm_ioctl(KVMState *s, int type, ...); 278 279 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...); 280 281 /** 282 * kvm_device_ioctl - call an ioctl on a kvm device 283 * @fd: The KVM device file descriptor as returned from KVM_CREATE_DEVICE 284 * @type: The device-ctrl ioctl number 285 * 286 * Returns: -errno on error, nonnegative on success 287 */ 288 int kvm_device_ioctl(int fd, int type, ...); 289 290 /** 291 * kvm_vm_check_attr - check for existence of a specific vm attribute 292 * @s: The KVMState pointer 293 * @group: the group 294 * @attr: the attribute of that group to query for 295 * 296 * Returns: 1 if the attribute exists 297 * 0 if the attribute either does not exist or if the vm device 298 * interface is unavailable 299 */ 300 int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr); 301 302 /** 303 * kvm_device_check_attr - check for existence of a specific device attribute 304 * @fd: The device file descriptor 305 * @group: the group 306 * @attr: the attribute of that group to query for 307 * 308 * Returns: 1 if the attribute exists 309 * 0 if the attribute either does not exist or if the vm device 310 * interface is unavailable 311 */ 312 int kvm_device_check_attr(int fd, uint32_t group, uint64_t attr); 313 314 /** 315 * kvm_device_access - set or get value of a specific device attribute 316 * @fd: The device file descriptor 317 * @group: the group 318 * @attr: the attribute of that group to set or get 319 * @val: pointer to a storage area for the value 320 * @write: true for set and false for get operation 321 * @errp: error object handle 322 * 323 * Returns: 0 on success 324 * < 0 on error 325 * Use kvm_device_check_attr() in order to check for the availability 326 * of optional attributes. 327 */ 328 int kvm_device_access(int fd, int group, uint64_t attr, 329 void *val, bool write, Error **errp); 330 331 /** 332 * kvm_create_device - create a KVM device for the device control API 333 * @KVMState: The KVMState pointer 334 * @type: The KVM device type (see Documentation/virtual/kvm/devices in the 335 * kernel source) 336 * @test: If true, only test if device can be created, but don't actually 337 * create the device. 338 * 339 * Returns: -errno on error, nonnegative on success: @test ? 0 : device fd; 340 */ 341 int kvm_create_device(KVMState *s, uint64_t type, bool test); 342 343 /** 344 * kvm_device_supported - probe whether KVM supports specific device 345 * 346 * @vmfd: The fd handler for VM 347 * @type: type of device 348 * 349 * @return: true if supported, otherwise false. 350 */ 351 bool kvm_device_supported(int vmfd, uint64_t type); 352 353 /* Arch specific hooks */ 354 355 extern const KVMCapabilityInfo kvm_arch_required_capabilities[]; 356 357 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run); 358 MemTxAttrs kvm_arch_post_run(CPUState *cpu, struct kvm_run *run); 359 360 int kvm_arch_handle_exit(CPUState *cpu, struct kvm_run *run); 361 362 int kvm_arch_process_async_events(CPUState *cpu); 363 364 int kvm_arch_get_registers(CPUState *cpu); 365 366 /* state subset only touched by the VCPU itself during runtime */ 367 #define KVM_PUT_RUNTIME_STATE 1 368 /* state subset modified during VCPU reset */ 369 #define KVM_PUT_RESET_STATE 2 370 /* full state set, modified during initialization or on vmload */ 371 #define KVM_PUT_FULL_STATE 3 372 373 int kvm_arch_put_registers(CPUState *cpu, int level); 374 375 int kvm_arch_init(MachineState *ms, KVMState *s); 376 377 int kvm_arch_init_vcpu(CPUState *cpu); 378 int kvm_arch_destroy_vcpu(CPUState *cpu); 379 380 bool kvm_vcpu_id_is_valid(int vcpu_id); 381 382 /* Returns VCPU ID to be used on KVM_CREATE_VCPU ioctl() */ 383 unsigned long kvm_arch_vcpu_id(CPUState *cpu); 384 385 #ifdef KVM_HAVE_MCE_INJECTION 386 void kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr); 387 #endif 388 389 void kvm_arch_init_irq_routing(KVMState *s); 390 391 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route, 392 uint64_t address, uint32_t data, PCIDevice *dev); 393 394 /* Notify arch about newly added MSI routes */ 395 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route, 396 int vector, PCIDevice *dev); 397 /* Notify arch about released MSI routes */ 398 int kvm_arch_release_virq_post(int virq); 399 400 int kvm_arch_msi_data_to_gsi(uint32_t data); 401 402 int kvm_set_irq(KVMState *s, int irq, int level); 403 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg); 404 405 void kvm_irqchip_add_irq_route(KVMState *s, int gsi, int irqchip, int pin); 406 407 void kvm_irqchip_add_change_notifier(Notifier *n); 408 void kvm_irqchip_remove_change_notifier(Notifier *n); 409 void kvm_irqchip_change_notify(void); 410 411 void kvm_get_apic_state(DeviceState *d, struct kvm_lapic_state *kapic); 412 413 struct kvm_guest_debug; 414 struct kvm_debug_exit_arch; 415 416 struct kvm_sw_breakpoint { 417 target_ulong pc; 418 target_ulong saved_insn; 419 int use_count; 420 QTAILQ_ENTRY(kvm_sw_breakpoint) entry; 421 }; 422 423 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu, 424 target_ulong pc); 425 426 int kvm_sw_breakpoints_active(CPUState *cpu); 427 428 int kvm_arch_insert_sw_breakpoint(CPUState *cpu, 429 struct kvm_sw_breakpoint *bp); 430 int kvm_arch_remove_sw_breakpoint(CPUState *cpu, 431 struct kvm_sw_breakpoint *bp); 432 int kvm_arch_insert_hw_breakpoint(target_ulong addr, 433 target_ulong len, int type); 434 int kvm_arch_remove_hw_breakpoint(target_ulong addr, 435 target_ulong len, int type); 436 void kvm_arch_remove_all_hw_breakpoints(void); 437 438 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg); 439 440 bool kvm_arch_stop_on_emulation_error(CPUState *cpu); 441 442 int kvm_check_extension(KVMState *s, unsigned int extension); 443 444 int kvm_vm_check_extension(KVMState *s, unsigned int extension); 445 446 #define kvm_vm_enable_cap(s, capability, cap_flags, ...) \ 447 ({ \ 448 struct kvm_enable_cap cap = { \ 449 .cap = capability, \ 450 .flags = cap_flags, \ 451 }; \ 452 uint64_t args_tmp[] = { __VA_ARGS__ }; \ 453 size_t n = MIN(ARRAY_SIZE(args_tmp), ARRAY_SIZE(cap.args)); \ 454 memcpy(cap.args, args_tmp, n * sizeof(cap.args[0])); \ 455 kvm_vm_ioctl(s, KVM_ENABLE_CAP, &cap); \ 456 }) 457 458 #define kvm_vcpu_enable_cap(cpu, capability, cap_flags, ...) \ 459 ({ \ 460 struct kvm_enable_cap cap = { \ 461 .cap = capability, \ 462 .flags = cap_flags, \ 463 }; \ 464 uint64_t args_tmp[] = { __VA_ARGS__ }; \ 465 size_t n = MIN(ARRAY_SIZE(args_tmp), ARRAY_SIZE(cap.args)); \ 466 memcpy(cap.args, args_tmp, n * sizeof(cap.args[0])); \ 467 kvm_vcpu_ioctl(cpu, KVM_ENABLE_CAP, &cap); \ 468 }) 469 470 uint32_t kvm_arch_get_supported_cpuid(KVMState *env, uint32_t function, 471 uint32_t index, int reg); 472 uint64_t kvm_arch_get_supported_msr_feature(KVMState *s, uint32_t index); 473 474 475 void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len); 476 477 #if !defined(CONFIG_USER_ONLY) 478 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram_addr, 479 hwaddr *phys_addr); 480 #endif 481 482 #endif /* NEED_CPU_H */ 483 484 void kvm_cpu_synchronize_state(CPUState *cpu); 485 486 void kvm_init_cpu_signals(CPUState *cpu); 487 488 /** 489 * kvm_irqchip_add_msi_route - Add MSI route for specific vector 490 * @s: KVM state 491 * @vector: which vector to add. This can be either MSI/MSIX 492 * vector. The function will automatically detect whether 493 * MSI/MSIX is enabled, and fetch corresponding MSI 494 * message. 495 * @dev: Owner PCI device to add the route. If @dev is specified 496 * as @NULL, an empty MSI message will be inited. 497 * @return: virq (>=0) when success, errno (<0) when failed. 498 */ 499 int kvm_irqchip_add_msi_route(KVMState *s, int vector, PCIDevice *dev); 500 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg, 501 PCIDevice *dev); 502 void kvm_irqchip_commit_routes(KVMState *s); 503 void kvm_irqchip_release_virq(KVMState *s, int virq); 504 505 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter); 506 int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint); 507 508 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, 509 EventNotifier *rn, int virq); 510 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, 511 int virq); 512 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n, 513 EventNotifier *rn, qemu_irq irq); 514 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, 515 qemu_irq irq); 516 void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi); 517 void kvm_pc_setup_irq_routing(bool pci_enabled); 518 void kvm_init_irq_routing(KVMState *s); 519 520 bool kvm_kernel_irqchip_allowed(void); 521 bool kvm_kernel_irqchip_required(void); 522 bool kvm_kernel_irqchip_split(void); 523 524 /** 525 * kvm_arch_irqchip_create: 526 * @KVMState: The KVMState pointer 527 * 528 * Allow architectures to create an in-kernel irq chip themselves. 529 * 530 * Returns: < 0: error 531 * 0: irq chip was not created 532 * > 0: irq chip was created 533 */ 534 int kvm_arch_irqchip_create(KVMState *s); 535 536 /** 537 * kvm_set_one_reg - set a register value in KVM via KVM_SET_ONE_REG ioctl 538 * @id: The register ID 539 * @source: The pointer to the value to be set. It must point to a variable 540 * of the correct type/size for the register being accessed. 541 * 542 * Returns: 0 on success, or a negative errno on failure. 543 */ 544 int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source); 545 546 /** 547 * kvm_get_one_reg - get a register value from KVM via KVM_GET_ONE_REG ioctl 548 * @id: The register ID 549 * @target: The pointer where the value is to be stored. It must point to a 550 * variable of the correct type/size for the register being accessed. 551 * 552 * Returns: 0 on success, or a negative errno on failure. 553 */ 554 int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target); 555 struct ppc_radix_page_info *kvm_get_radix_page_info(void); 556 int kvm_get_max_memslots(void); 557 558 /* Notify resamplefd for EOI of specific interrupts. */ 559 void kvm_resample_fd_notify(int gsi); 560 561 #endif 562