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 21 #ifdef NEED_CPU_H 22 # ifdef CONFIG_KVM 23 # include <linux/kvm.h> 24 # define CONFIG_KVM_IS_POSSIBLE 25 # endif 26 #else 27 # define CONFIG_KVM_IS_POSSIBLE 28 #endif 29 30 #ifdef CONFIG_KVM_IS_POSSIBLE 31 32 extern bool kvm_allowed; 33 extern bool kvm_kernel_irqchip; 34 extern bool kvm_split_irqchip; 35 extern bool kvm_async_interrupts_allowed; 36 extern bool kvm_halt_in_kernel_allowed; 37 extern bool kvm_eventfds_allowed; 38 extern bool kvm_irqfds_allowed; 39 extern bool kvm_resamplefds_allowed; 40 extern bool kvm_msi_via_irqfd_allowed; 41 extern bool kvm_gsi_routing_allowed; 42 extern bool kvm_gsi_direct_mapping; 43 extern bool kvm_readonly_mem_allowed; 44 extern bool kvm_direct_msi_allowed; 45 extern bool kvm_ioeventfd_any_length_allowed; 46 extern bool kvm_msi_use_devid; 47 48 #define kvm_enabled() (kvm_allowed) 49 /** 50 * kvm_irqchip_in_kernel: 51 * 52 * Returns: true if the user asked us to create an in-kernel 53 * irqchip via the "kernel_irqchip=on" machine option. 54 * What this actually means is architecture and machine model 55 * specific: on PC, for instance, it means that the LAPIC, 56 * IOAPIC and PIT are all in kernel. This function should never 57 * be used from generic target-independent code: use one of the 58 * following functions or 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 user asked us to split the irqchip 66 * implementation between user and kernel space. The details are 67 * architecture and machine specific. On PC, it means that the PIC, 68 * IOAPIC, and 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 typedef struct KVMState KVMState; 203 extern KVMState *kvm_state; 204 typedef struct Notifier Notifier; 205 206 /* external API */ 207 208 bool kvm_has_free_slot(MachineState *ms); 209 bool kvm_has_sync_mmu(void); 210 int kvm_has_vcpu_events(void); 211 int kvm_has_robust_singlestep(void); 212 int kvm_has_debugregs(void); 213 int kvm_max_nested_state_length(void); 214 int kvm_has_pit_state2(void); 215 int kvm_has_many_ioeventfds(void); 216 int kvm_has_gsi_routing(void); 217 int kvm_has_intx_set_mask(void); 218 219 int kvm_init_vcpu(CPUState *cpu); 220 int kvm_cpu_exec(CPUState *cpu); 221 int kvm_destroy_vcpu(CPUState *cpu); 222 223 /** 224 * kvm_arm_supports_user_irq 225 * 226 * Not all KVM implementations support notifications for kernel generated 227 * interrupt events to user space. This function indicates whether the current 228 * KVM implementation does support them. 229 * 230 * Returns: true if KVM supports using kernel generated IRQs from user space 231 */ 232 bool kvm_arm_supports_user_irq(void); 233 234 /** 235 * kvm_memcrypt_enabled - return boolean indicating whether memory encryption 236 * is enabled 237 * Returns: 1 memory encryption is enabled 238 * 0 memory encryption is disabled 239 */ 240 bool kvm_memcrypt_enabled(void); 241 242 /** 243 * kvm_memcrypt_encrypt_data: encrypt the memory range 244 * 245 * Return: 1 failed to encrypt the range 246 * 0 succesfully encrypted memory region 247 */ 248 int kvm_memcrypt_encrypt_data(uint8_t *ptr, uint64_t len); 249 250 251 #ifdef NEED_CPU_H 252 #include "cpu.h" 253 254 void kvm_flush_coalesced_mmio_buffer(void); 255 256 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr, 257 target_ulong len, int type); 258 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr, 259 target_ulong len, int type); 260 void kvm_remove_all_breakpoints(CPUState *cpu); 261 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap); 262 263 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr); 264 int kvm_on_sigbus(int code, void *addr); 265 266 /* interface with exec.c */ 267 268 void phys_mem_set_alloc(void *(*alloc)(size_t, uint64_t *align, bool shared)); 269 270 /* internal API */ 271 272 int kvm_ioctl(KVMState *s, int type, ...); 273 274 int kvm_vm_ioctl(KVMState *s, int type, ...); 275 276 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...); 277 278 /** 279 * kvm_device_ioctl - call an ioctl on a kvm device 280 * @fd: The KVM device file descriptor as returned from KVM_CREATE_DEVICE 281 * @type: The device-ctrl ioctl number 282 * 283 * Returns: -errno on error, nonnegative on success 284 */ 285 int kvm_device_ioctl(int fd, int type, ...); 286 287 /** 288 * kvm_vm_check_attr - check for existence of a specific vm attribute 289 * @s: The KVMState pointer 290 * @group: the group 291 * @attr: the attribute of that group to query for 292 * 293 * Returns: 1 if the attribute exists 294 * 0 if the attribute either does not exist or if the vm device 295 * interface is unavailable 296 */ 297 int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr); 298 299 /** 300 * kvm_device_check_attr - check for existence of a specific device attribute 301 * @fd: The device file descriptor 302 * @group: the group 303 * @attr: the attribute of that group to query for 304 * 305 * Returns: 1 if the attribute exists 306 * 0 if the attribute either does not exist or if the vm device 307 * interface is unavailable 308 */ 309 int kvm_device_check_attr(int fd, uint32_t group, uint64_t attr); 310 311 /** 312 * kvm_device_access - set or get value of a specific device attribute 313 * @fd: The device file descriptor 314 * @group: the group 315 * @attr: the attribute of that group to set or get 316 * @val: pointer to a storage area for the value 317 * @write: true for set and false for get operation 318 * @errp: error object handle 319 * 320 * Returns: 0 on success 321 * < 0 on error 322 * Use kvm_device_check_attr() in order to check for the availability 323 * of optional attributes. 324 */ 325 int kvm_device_access(int fd, int group, uint64_t attr, 326 void *val, bool write, Error **errp); 327 328 /** 329 * kvm_create_device - create a KVM device for the device control API 330 * @KVMState: The KVMState pointer 331 * @type: The KVM device type (see Documentation/virtual/kvm/devices in the 332 * kernel source) 333 * @test: If true, only test if device can be created, but don't actually 334 * create the device. 335 * 336 * Returns: -errno on error, nonnegative on success: @test ? 0 : device fd; 337 */ 338 int kvm_create_device(KVMState *s, uint64_t type, bool test); 339 340 /** 341 * kvm_device_supported - probe whether KVM supports specific device 342 * 343 * @vmfd: The fd handler for VM 344 * @type: type of device 345 * 346 * @return: true if supported, otherwise false. 347 */ 348 bool kvm_device_supported(int vmfd, uint64_t type); 349 350 /* Arch specific hooks */ 351 352 extern const KVMCapabilityInfo kvm_arch_required_capabilities[]; 353 354 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run); 355 MemTxAttrs kvm_arch_post_run(CPUState *cpu, struct kvm_run *run); 356 357 int kvm_arch_handle_exit(CPUState *cpu, struct kvm_run *run); 358 359 int kvm_arch_process_async_events(CPUState *cpu); 360 361 int kvm_arch_get_registers(CPUState *cpu); 362 363 /* state subset only touched by the VCPU itself during runtime */ 364 #define KVM_PUT_RUNTIME_STATE 1 365 /* state subset modified during VCPU reset */ 366 #define KVM_PUT_RESET_STATE 2 367 /* full state set, modified during initialization or on vmload */ 368 #define KVM_PUT_FULL_STATE 3 369 370 int kvm_arch_put_registers(CPUState *cpu, int level); 371 372 int kvm_arch_init(MachineState *ms, KVMState *s); 373 374 int kvm_arch_init_vcpu(CPUState *cpu); 375 int kvm_arch_destroy_vcpu(CPUState *cpu); 376 377 bool kvm_vcpu_id_is_valid(int vcpu_id); 378 379 /* Returns VCPU ID to be used on KVM_CREATE_VCPU ioctl() */ 380 unsigned long kvm_arch_vcpu_id(CPUState *cpu); 381 382 #ifdef TARGET_I386 383 #define KVM_HAVE_MCE_INJECTION 1 384 void kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr); 385 #endif 386 387 void kvm_arch_init_irq_routing(KVMState *s); 388 389 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route, 390 uint64_t address, uint32_t data, PCIDevice *dev); 391 392 /* Notify arch about newly added MSI routes */ 393 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route, 394 int vector, PCIDevice *dev); 395 /* Notify arch about released MSI routes */ 396 int kvm_arch_release_virq_post(int virq); 397 398 int kvm_arch_msi_data_to_gsi(uint32_t data); 399 400 int kvm_set_irq(KVMState *s, int irq, int level); 401 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg); 402 403 void kvm_irqchip_add_irq_route(KVMState *s, int gsi, int irqchip, int pin); 404 405 void kvm_irqchip_add_change_notifier(Notifier *n); 406 void kvm_irqchip_remove_change_notifier(Notifier *n); 407 void kvm_irqchip_change_notify(void); 408 409 void kvm_get_apic_state(DeviceState *d, struct kvm_lapic_state *kapic); 410 411 struct kvm_guest_debug; 412 struct kvm_debug_exit_arch; 413 414 struct kvm_sw_breakpoint { 415 target_ulong pc; 416 target_ulong saved_insn; 417 int use_count; 418 QTAILQ_ENTRY(kvm_sw_breakpoint) entry; 419 }; 420 421 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu, 422 target_ulong pc); 423 424 int kvm_sw_breakpoints_active(CPUState *cpu); 425 426 int kvm_arch_insert_sw_breakpoint(CPUState *cpu, 427 struct kvm_sw_breakpoint *bp); 428 int kvm_arch_remove_sw_breakpoint(CPUState *cpu, 429 struct kvm_sw_breakpoint *bp); 430 int kvm_arch_insert_hw_breakpoint(target_ulong addr, 431 target_ulong len, int type); 432 int kvm_arch_remove_hw_breakpoint(target_ulong addr, 433 target_ulong len, int type); 434 void kvm_arch_remove_all_hw_breakpoints(void); 435 436 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg); 437 438 bool kvm_arch_stop_on_emulation_error(CPUState *cpu); 439 440 int kvm_check_extension(KVMState *s, unsigned int extension); 441 442 int kvm_vm_check_extension(KVMState *s, unsigned int extension); 443 444 #define kvm_vm_enable_cap(s, capability, cap_flags, ...) \ 445 ({ \ 446 struct kvm_enable_cap cap = { \ 447 .cap = capability, \ 448 .flags = cap_flags, \ 449 }; \ 450 uint64_t args_tmp[] = { __VA_ARGS__ }; \ 451 size_t n = MIN(ARRAY_SIZE(args_tmp), ARRAY_SIZE(cap.args)); \ 452 memcpy(cap.args, args_tmp, n * sizeof(cap.args[0])); \ 453 kvm_vm_ioctl(s, KVM_ENABLE_CAP, &cap); \ 454 }) 455 456 #define kvm_vcpu_enable_cap(cpu, capability, cap_flags, ...) \ 457 ({ \ 458 struct kvm_enable_cap cap = { \ 459 .cap = capability, \ 460 .flags = cap_flags, \ 461 }; \ 462 uint64_t args_tmp[] = { __VA_ARGS__ }; \ 463 size_t n = MIN(ARRAY_SIZE(args_tmp), ARRAY_SIZE(cap.args)); \ 464 memcpy(cap.args, args_tmp, n * sizeof(cap.args[0])); \ 465 kvm_vcpu_ioctl(cpu, KVM_ENABLE_CAP, &cap); \ 466 }) 467 468 uint32_t kvm_arch_get_supported_cpuid(KVMState *env, uint32_t function, 469 uint32_t index, int reg); 470 uint64_t kvm_arch_get_supported_msr_feature(KVMState *s, uint32_t index); 471 472 473 void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len); 474 475 #if !defined(CONFIG_USER_ONLY) 476 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram_addr, 477 hwaddr *phys_addr); 478 #endif 479 480 #endif /* NEED_CPU_H */ 481 482 void kvm_cpu_synchronize_state(CPUState *cpu); 483 void kvm_cpu_synchronize_post_reset(CPUState *cpu); 484 void kvm_cpu_synchronize_post_init(CPUState *cpu); 485 void kvm_cpu_synchronize_pre_loadvm(CPUState *cpu); 486 487 void kvm_init_cpu_signals(CPUState *cpu); 488 489 /** 490 * kvm_irqchip_add_msi_route - Add MSI route for specific vector 491 * @s: KVM state 492 * @vector: which vector to add. This can be either MSI/MSIX 493 * vector. The function will automatically detect whether 494 * MSI/MSIX is enabled, and fetch corresponding MSI 495 * message. 496 * @dev: Owner PCI device to add the route. If @dev is specified 497 * as @NULL, an empty MSI message will be inited. 498 * @return: virq (>=0) when success, errno (<0) when failed. 499 */ 500 int kvm_irqchip_add_msi_route(KVMState *s, int vector, PCIDevice *dev); 501 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg, 502 PCIDevice *dev); 503 void kvm_irqchip_commit_routes(KVMState *s); 504 void kvm_irqchip_release_virq(KVMState *s, int virq); 505 506 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter); 507 int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint); 508 509 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, 510 EventNotifier *rn, int virq); 511 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, 512 int virq); 513 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n, 514 EventNotifier *rn, qemu_irq irq); 515 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, 516 qemu_irq irq); 517 void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi); 518 void kvm_pc_gsi_handler(void *opaque, int n, int level); 519 void kvm_pc_setup_irq_routing(bool pci_enabled); 520 void kvm_init_irq_routing(KVMState *s); 521 522 /** 523 * kvm_arch_irqchip_create: 524 * @KVMState: The KVMState pointer 525 * @MachineState: The MachineState 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(MachineState *ms, 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 #endif 557