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 "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_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 #ifdef NEED_CPU_H 238 #include "cpu.h" 239 240 void kvm_flush_coalesced_mmio_buffer(void); 241 242 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr, 243 target_ulong len, int type); 244 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr, 245 target_ulong len, int type); 246 void kvm_remove_all_breakpoints(CPUState *cpu); 247 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap); 248 249 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr); 250 int kvm_on_sigbus(int code, void *addr); 251 252 /* interface with exec.c */ 253 254 void phys_mem_set_alloc(void *(*alloc)(size_t, uint64_t *align, bool shared)); 255 256 /* internal API */ 257 258 int kvm_ioctl(KVMState *s, int type, ...); 259 260 int kvm_vm_ioctl(KVMState *s, int type, ...); 261 262 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...); 263 264 /** 265 * kvm_device_ioctl - call an ioctl on a kvm device 266 * @fd: The KVM device file descriptor as returned from KVM_CREATE_DEVICE 267 * @type: The device-ctrl ioctl number 268 * 269 * Returns: -errno on error, nonnegative on success 270 */ 271 int kvm_device_ioctl(int fd, int type, ...); 272 273 /** 274 * kvm_vm_check_attr - check for existence of a specific vm attribute 275 * @s: The KVMState pointer 276 * @group: the group 277 * @attr: the attribute of that group to query for 278 * 279 * Returns: 1 if the attribute exists 280 * 0 if the attribute either does not exist or if the vm device 281 * interface is unavailable 282 */ 283 int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr); 284 285 /** 286 * kvm_device_check_attr - check for existence of a specific device attribute 287 * @fd: The device file descriptor 288 * @group: the group 289 * @attr: the attribute of that group to query for 290 * 291 * Returns: 1 if the attribute exists 292 * 0 if the attribute either does not exist or if the vm device 293 * interface is unavailable 294 */ 295 int kvm_device_check_attr(int fd, uint32_t group, uint64_t attr); 296 297 /** 298 * kvm_device_access - set or get value of a specific device attribute 299 * @fd: The device file descriptor 300 * @group: the group 301 * @attr: the attribute of that group to set or get 302 * @val: pointer to a storage area for the value 303 * @write: true for set and false for get operation 304 * @errp: error object handle 305 * 306 * Returns: 0 on success 307 * < 0 on error 308 * Use kvm_device_check_attr() in order to check for the availability 309 * of optional attributes. 310 */ 311 int kvm_device_access(int fd, int group, uint64_t attr, 312 void *val, bool write, Error **errp); 313 314 /** 315 * kvm_create_device - create a KVM device for the device control API 316 * @KVMState: The KVMState pointer 317 * @type: The KVM device type (see Documentation/virtual/kvm/devices in the 318 * kernel source) 319 * @test: If true, only test if device can be created, but don't actually 320 * create the device. 321 * 322 * Returns: -errno on error, nonnegative on success: @test ? 0 : device fd; 323 */ 324 int kvm_create_device(KVMState *s, uint64_t type, bool test); 325 326 /** 327 * kvm_device_supported - probe whether KVM supports specific device 328 * 329 * @vmfd: The fd handler for VM 330 * @type: type of device 331 * 332 * @return: true if supported, otherwise false. 333 */ 334 bool kvm_device_supported(int vmfd, uint64_t type); 335 336 /* Arch specific hooks */ 337 338 extern const KVMCapabilityInfo kvm_arch_required_capabilities[]; 339 340 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run); 341 MemTxAttrs kvm_arch_post_run(CPUState *cpu, struct kvm_run *run); 342 343 int kvm_arch_handle_exit(CPUState *cpu, struct kvm_run *run); 344 345 int kvm_arch_process_async_events(CPUState *cpu); 346 347 int kvm_arch_get_registers(CPUState *cpu); 348 349 /* state subset only touched by the VCPU itself during runtime */ 350 #define KVM_PUT_RUNTIME_STATE 1 351 /* state subset modified during VCPU reset */ 352 #define KVM_PUT_RESET_STATE 2 353 /* full state set, modified during initialization or on vmload */ 354 #define KVM_PUT_FULL_STATE 3 355 356 int kvm_arch_put_registers(CPUState *cpu, int level); 357 358 int kvm_arch_init(MachineState *ms, KVMState *s); 359 360 int kvm_arch_init_vcpu(CPUState *cpu); 361 int kvm_arch_destroy_vcpu(CPUState *cpu); 362 363 bool kvm_vcpu_id_is_valid(int vcpu_id); 364 365 /* Returns VCPU ID to be used on KVM_CREATE_VCPU ioctl() */ 366 unsigned long kvm_arch_vcpu_id(CPUState *cpu); 367 368 #ifdef KVM_HAVE_MCE_INJECTION 369 void kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr); 370 #endif 371 372 void kvm_arch_init_irq_routing(KVMState *s); 373 374 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route, 375 uint64_t address, uint32_t data, PCIDevice *dev); 376 377 /* Notify arch about newly added MSI routes */ 378 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route, 379 int vector, PCIDevice *dev); 380 /* Notify arch about released MSI routes */ 381 int kvm_arch_release_virq_post(int virq); 382 383 int kvm_arch_msi_data_to_gsi(uint32_t data); 384 385 int kvm_set_irq(KVMState *s, int irq, int level); 386 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg); 387 388 void kvm_irqchip_add_irq_route(KVMState *s, int gsi, int irqchip, int pin); 389 390 void kvm_irqchip_add_change_notifier(Notifier *n); 391 void kvm_irqchip_remove_change_notifier(Notifier *n); 392 void kvm_irqchip_change_notify(void); 393 394 void kvm_get_apic_state(DeviceState *d, struct kvm_lapic_state *kapic); 395 396 struct kvm_guest_debug; 397 struct kvm_debug_exit_arch; 398 399 struct kvm_sw_breakpoint { 400 target_ulong pc; 401 target_ulong saved_insn; 402 int use_count; 403 QTAILQ_ENTRY(kvm_sw_breakpoint) entry; 404 }; 405 406 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu, 407 target_ulong pc); 408 409 int kvm_sw_breakpoints_active(CPUState *cpu); 410 411 int kvm_arch_insert_sw_breakpoint(CPUState *cpu, 412 struct kvm_sw_breakpoint *bp); 413 int kvm_arch_remove_sw_breakpoint(CPUState *cpu, 414 struct kvm_sw_breakpoint *bp); 415 int kvm_arch_insert_hw_breakpoint(target_ulong addr, 416 target_ulong len, int type); 417 int kvm_arch_remove_hw_breakpoint(target_ulong addr, 418 target_ulong len, int type); 419 void kvm_arch_remove_all_hw_breakpoints(void); 420 421 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg); 422 423 bool kvm_arch_stop_on_emulation_error(CPUState *cpu); 424 425 int kvm_check_extension(KVMState *s, unsigned int extension); 426 427 int kvm_vm_check_extension(KVMState *s, unsigned int extension); 428 429 #define kvm_vm_enable_cap(s, capability, cap_flags, ...) \ 430 ({ \ 431 struct kvm_enable_cap cap = { \ 432 .cap = capability, \ 433 .flags = cap_flags, \ 434 }; \ 435 uint64_t args_tmp[] = { __VA_ARGS__ }; \ 436 size_t n = MIN(ARRAY_SIZE(args_tmp), ARRAY_SIZE(cap.args)); \ 437 memcpy(cap.args, args_tmp, n * sizeof(cap.args[0])); \ 438 kvm_vm_ioctl(s, KVM_ENABLE_CAP, &cap); \ 439 }) 440 441 #define kvm_vcpu_enable_cap(cpu, capability, cap_flags, ...) \ 442 ({ \ 443 struct kvm_enable_cap cap = { \ 444 .cap = capability, \ 445 .flags = cap_flags, \ 446 }; \ 447 uint64_t args_tmp[] = { __VA_ARGS__ }; \ 448 size_t n = MIN(ARRAY_SIZE(args_tmp), ARRAY_SIZE(cap.args)); \ 449 memcpy(cap.args, args_tmp, n * sizeof(cap.args[0])); \ 450 kvm_vcpu_ioctl(cpu, KVM_ENABLE_CAP, &cap); \ 451 }) 452 453 uint32_t kvm_arch_get_supported_cpuid(KVMState *env, uint32_t function, 454 uint32_t index, int reg); 455 uint64_t kvm_arch_get_supported_msr_feature(KVMState *s, uint32_t index); 456 457 458 void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len); 459 460 #if !defined(CONFIG_USER_ONLY) 461 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram_addr, 462 hwaddr *phys_addr); 463 #endif 464 465 #endif /* NEED_CPU_H */ 466 467 void kvm_cpu_synchronize_state(CPUState *cpu); 468 469 void kvm_init_cpu_signals(CPUState *cpu); 470 471 /** 472 * kvm_irqchip_add_msi_route - Add MSI route for specific vector 473 * @s: KVM state 474 * @vector: which vector to add. This can be either MSI/MSIX 475 * vector. The function will automatically detect whether 476 * MSI/MSIX is enabled, and fetch corresponding MSI 477 * message. 478 * @dev: Owner PCI device to add the route. If @dev is specified 479 * as @NULL, an empty MSI message will be inited. 480 * @return: virq (>=0) when success, errno (<0) when failed. 481 */ 482 int kvm_irqchip_add_msi_route(KVMState *s, int vector, PCIDevice *dev); 483 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg, 484 PCIDevice *dev); 485 void kvm_irqchip_commit_routes(KVMState *s); 486 void kvm_irqchip_release_virq(KVMState *s, int virq); 487 488 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter); 489 int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint); 490 491 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, 492 EventNotifier *rn, int virq); 493 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, 494 int virq); 495 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n, 496 EventNotifier *rn, qemu_irq irq); 497 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, 498 qemu_irq irq); 499 void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi); 500 void kvm_pc_setup_irq_routing(bool pci_enabled); 501 void kvm_init_irq_routing(KVMState *s); 502 503 bool kvm_kernel_irqchip_allowed(void); 504 bool kvm_kernel_irqchip_required(void); 505 bool kvm_kernel_irqchip_split(void); 506 507 /** 508 * kvm_arch_irqchip_create: 509 * @KVMState: The KVMState pointer 510 * 511 * Allow architectures to create an in-kernel irq chip themselves. 512 * 513 * Returns: < 0: error 514 * 0: irq chip was not created 515 * > 0: irq chip was created 516 */ 517 int kvm_arch_irqchip_create(KVMState *s); 518 519 /** 520 * kvm_set_one_reg - set a register value in KVM via KVM_SET_ONE_REG ioctl 521 * @id: The register ID 522 * @source: The pointer to the value to be set. It must point to a variable 523 * of the correct type/size for the register being accessed. 524 * 525 * Returns: 0 on success, or a negative errno on failure. 526 */ 527 int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source); 528 529 /** 530 * kvm_get_one_reg - get a register value from KVM via KVM_GET_ONE_REG ioctl 531 * @id: The register ID 532 * @target: The pointer where the value is to be stored. It must point to a 533 * variable of the correct type/size for the register being accessed. 534 * 535 * Returns: 0 on success, or a negative errno on failure. 536 */ 537 int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target); 538 struct ppc_radix_page_info *kvm_get_radix_page_info(void); 539 int kvm_get_max_memslots(void); 540 541 /* Notify resamplefd for EOI of specific interrupts. */ 542 void kvm_resample_fd_notify(int gsi); 543 544 #endif 545