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