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