1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2011 NetApp, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 * 28 * $FreeBSD$ 29 */ 30 /* 31 * This file and its contents are supplied under the terms of the 32 * Common Development and Distribution License ("CDDL"), version 1.0. 33 * You may only use this file in accordance with the terms of version 34 * 1.0 of the CDDL. 35 * 36 * A full copy of the text of the CDDL should have accompanied this 37 * source. A copy of the CDDL is also available via the Internet at 38 * http://www.illumos.org/license/CDDL. 39 * 40 * Copyright 2015 Pluribus Networks Inc. 41 * Copyright 2019 Joyent, Inc. 42 * Copyright 2022 Oxide Computer Company 43 * Copyright 2021 OmniOS Community Edition (OmniOSce) Association. 44 */ 45 46 #ifndef _VMM_KERNEL_H_ 47 #define _VMM_KERNEL_H_ 48 49 #include <sys/sdt.h> 50 #include <x86/segments.h> 51 #include <sys/vmm.h> 52 #include <sys/vmm_data.h> 53 #include <sys/linker_set.h> 54 55 SDT_PROVIDER_DECLARE(vmm); 56 57 struct vm; 58 struct vm_exception; 59 struct seg_desc; 60 struct vm_exit; 61 struct vie; 62 struct vm_run; 63 struct vhpet; 64 struct vioapic; 65 struct vlapic; 66 struct vmspace; 67 struct vm_client; 68 struct vm_object; 69 struct vm_guest_paging; 70 struct vmm_data_req; 71 72 typedef int (*vmm_init_func_t)(void); 73 typedef int (*vmm_cleanup_func_t)(void); 74 typedef void (*vmm_resume_func_t)(void); 75 typedef void * (*vmi_init_func_t)(struct vm *vm); 76 typedef int (*vmi_run_func_t)(void *vmi, int vcpu, uint64_t rip); 77 typedef void (*vmi_cleanup_func_t)(void *vmi); 78 typedef int (*vmi_get_register_t)(void *vmi, int vcpu, int num, 79 uint64_t *retval); 80 typedef int (*vmi_set_register_t)(void *vmi, int vcpu, int num, 81 uint64_t val); 82 typedef int (*vmi_get_desc_t)(void *vmi, int vcpu, int num, 83 struct seg_desc *desc); 84 typedef int (*vmi_set_desc_t)(void *vmi, int vcpu, int num, 85 const struct seg_desc *desc); 86 typedef int (*vmi_get_cap_t)(void *vmi, int vcpu, int num, int *retval); 87 typedef int (*vmi_set_cap_t)(void *vmi, int vcpu, int num, int val); 88 typedef struct vlapic *(*vmi_vlapic_init)(void *vmi, int vcpu); 89 typedef void (*vmi_vlapic_cleanup)(void *vmi, struct vlapic *vlapic); 90 typedef void (*vmi_savectx)(void *vmi, int vcpu); 91 typedef void (*vmi_restorectx)(void *vmi, int vcpu); 92 93 typedef int (*vmi_get_msr_t)(void *vmi, int vcpu, uint32_t msr, 94 uint64_t *valp); 95 typedef int (*vmi_set_msr_t)(void *vmi, int vcpu, uint32_t msr, 96 uint64_t val); 97 98 struct vmm_ops { 99 vmm_init_func_t init; /* module wide initialization */ 100 vmm_cleanup_func_t cleanup; 101 vmm_resume_func_t resume; 102 103 vmi_init_func_t vminit; /* vm-specific initialization */ 104 vmi_run_func_t vmrun; 105 vmi_cleanup_func_t vmcleanup; 106 vmi_get_register_t vmgetreg; 107 vmi_set_register_t vmsetreg; 108 vmi_get_desc_t vmgetdesc; 109 vmi_set_desc_t vmsetdesc; 110 vmi_get_cap_t vmgetcap; 111 vmi_set_cap_t vmsetcap; 112 vmi_vlapic_init vlapic_init; 113 vmi_vlapic_cleanup vlapic_cleanup; 114 115 vmi_savectx vmsavectx; 116 vmi_restorectx vmrestorectx; 117 118 vmi_get_msr_t vmgetmsr; 119 vmi_set_msr_t vmsetmsr; 120 }; 121 122 extern struct vmm_ops vmm_ops_intel; 123 extern struct vmm_ops vmm_ops_amd; 124 125 int vm_create(uint64_t flags, struct vm **retvm); 126 void vm_destroy(struct vm *vm); 127 int vm_reinit(struct vm *vm, uint64_t); 128 uint16_t vm_get_maxcpus(struct vm *vm); 129 void vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores, 130 uint16_t *threads, uint16_t *maxcpus); 131 int vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores, 132 uint16_t threads, uint16_t maxcpus); 133 134 /* 135 * APIs that race against hardware. 136 */ 137 void vm_track_dirty_pages(struct vm *, uint64_t, size_t, uint8_t *); 138 139 /* 140 * APIs that modify the guest memory map require all vcpus to be frozen. 141 */ 142 int vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t off, 143 size_t len, int prot, int flags); 144 int vm_munmap_memseg(struct vm *vm, vm_paddr_t gpa, size_t len); 145 int vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem); 146 void vm_free_memseg(struct vm *vm, int ident); 147 int vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa); 148 int vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len); 149 int vm_assign_pptdev(struct vm *vm, int pptfd); 150 int vm_unassign_pptdev(struct vm *vm, int pptfd); 151 152 /* 153 * APIs that inspect the guest memory map require only a *single* vcpu to 154 * be frozen. This acts like a read lock on the guest memory map since any 155 * modification requires *all* vcpus to be frozen. 156 */ 157 int vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid, 158 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags); 159 int vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem, 160 struct vm_object **objptr); 161 vm_paddr_t vmm_sysmem_maxaddr(struct vm *vm); 162 bool vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa); 163 164 int vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval); 165 int vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val); 166 int vm_get_seg_desc(struct vm *vm, int vcpu, int reg, 167 struct seg_desc *ret_desc); 168 int vm_set_seg_desc(struct vm *vm, int vcpu, int reg, 169 const struct seg_desc *desc); 170 int vm_get_run_state(struct vm *vm, int vcpuid, uint32_t *state, 171 uint8_t *sipi_vec); 172 int vm_set_run_state(struct vm *vm, int vcpuid, uint32_t state, 173 uint8_t sipi_vec); 174 int vm_get_fpu(struct vm *vm, int vcpuid, void *buf, size_t len); 175 int vm_set_fpu(struct vm *vm, int vcpuid, void *buf, size_t len); 176 int vm_run(struct vm *vm, int vcpuid, const struct vm_entry *); 177 int vm_suspend(struct vm *vm, enum vm_suspend_how how); 178 int vm_inject_nmi(struct vm *vm, int vcpu); 179 bool vm_nmi_pending(struct vm *vm, int vcpuid); 180 void vm_nmi_clear(struct vm *vm, int vcpuid); 181 int vm_inject_extint(struct vm *vm, int vcpu); 182 bool vm_extint_pending(struct vm *vm, int vcpuid); 183 void vm_extint_clear(struct vm *vm, int vcpuid); 184 int vm_inject_init(struct vm *vm, int vcpuid); 185 int vm_inject_sipi(struct vm *vm, int vcpuid, uint8_t vec); 186 struct vlapic *vm_lapic(struct vm *vm, int cpu); 187 struct vioapic *vm_ioapic(struct vm *vm); 188 struct vhpet *vm_hpet(struct vm *vm); 189 int vm_get_capability(struct vm *vm, int vcpu, int type, int *val); 190 int vm_set_capability(struct vm *vm, int vcpu, int type, int val); 191 int vm_get_x2apic_state(struct vm *vm, int vcpu, enum x2apic_state *state); 192 int vm_set_x2apic_state(struct vm *vm, int vcpu, enum x2apic_state state); 193 int vm_apicid2vcpuid(struct vm *vm, int apicid); 194 int vm_activate_cpu(struct vm *vm, int vcpu); 195 int vm_suspend_cpu(struct vm *vm, int vcpu); 196 int vm_resume_cpu(struct vm *vm, int vcpu); 197 struct vm_exit *vm_exitinfo(struct vm *vm, int vcpuid); 198 struct vie *vm_vie_ctx(struct vm *vm, int vcpuid); 199 void vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip); 200 void vm_exit_debug(struct vm *vm, int vcpuid, uint64_t rip); 201 void vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip); 202 void vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip); 203 void vm_exit_run_state(struct vm *vm, int vcpuid, uint64_t rip); 204 int vm_service_mmio_read(struct vm *vm, int cpuid, uint64_t gpa, uint64_t *rval, 205 int rsize); 206 int vm_service_mmio_write(struct vm *vm, int cpuid, uint64_t gpa, uint64_t wval, 207 int wsize); 208 209 #ifdef _SYS__CPUSET_H_ 210 cpuset_t vm_active_cpus(struct vm *vm); 211 cpuset_t vm_debug_cpus(struct vm *vm); 212 cpuset_t vm_suspended_cpus(struct vm *vm); 213 #endif /* _SYS__CPUSET_H_ */ 214 215 bool vcpu_entry_bailout_checks(struct vm *vm, int vcpuid, uint64_t rip); 216 bool vcpu_run_state_pending(struct vm *vm, int vcpuid); 217 int vcpu_arch_reset(struct vm *vm, int vcpuid, bool init_only); 218 219 /* 220 * Return true if device indicated by bus/slot/func is supposed to be a 221 * pci passthrough device. 222 * 223 * Return false otherwise. 224 */ 225 bool vmm_is_pptdev(int bus, int slot, int func); 226 227 void *vm_iommu_domain(struct vm *vm); 228 229 enum vcpu_state { 230 VCPU_IDLE, 231 VCPU_FROZEN, 232 VCPU_RUNNING, 233 VCPU_SLEEPING, 234 }; 235 236 int vcpu_set_state(struct vm *vm, int vcpu, enum vcpu_state state, 237 bool from_idle); 238 enum vcpu_state vcpu_get_state(struct vm *vm, int vcpu, int *hostcpu); 239 void vcpu_block_run(struct vm *, int); 240 void vcpu_unblock_run(struct vm *, int); 241 242 uint64_t vcpu_tsc_offset(struct vm *vm, int vcpuid, bool phys_adj); 243 hrtime_t vm_normalize_hrtime(struct vm *, hrtime_t); 244 hrtime_t vm_denormalize_hrtime(struct vm *, hrtime_t); 245 246 static __inline bool 247 vcpu_is_running(struct vm *vm, int vcpu, int *hostcpu) 248 { 249 return (vcpu_get_state(vm, vcpu, hostcpu) == VCPU_RUNNING); 250 } 251 252 #ifdef _SYS_THREAD_H 253 static __inline int 254 vcpu_should_yield(struct vm *vm, int vcpu) 255 { 256 257 if (curthread->t_astflag) 258 return (1); 259 else if (CPU->cpu_runrun) 260 return (1); 261 else 262 return (0); 263 } 264 #endif /* _SYS_THREAD_H */ 265 266 typedef enum vcpu_notify { 267 VCPU_NOTIFY_NONE, 268 VCPU_NOTIFY_APIC, /* Posted intr notification (if possible) */ 269 VCPU_NOTIFY_EXIT, /* IPI to cause VM exit */ 270 } vcpu_notify_t; 271 272 void *vcpu_stats(struct vm *vm, int vcpu); 273 void vcpu_notify_event(struct vm *vm, int vcpuid); 274 void vcpu_notify_event_type(struct vm *vm, int vcpuid, vcpu_notify_t); 275 struct vmspace *vm_get_vmspace(struct vm *vm); 276 struct vm_client *vm_get_vmclient(struct vm *vm, int vcpuid); 277 struct vatpic *vm_atpic(struct vm *vm); 278 struct vatpit *vm_atpit(struct vm *vm); 279 struct vpmtmr *vm_pmtmr(struct vm *vm); 280 struct vrtc *vm_rtc(struct vm *vm); 281 282 /* 283 * Inject exception 'vector' into the guest vcpu. This function returns 0 on 284 * success and non-zero on failure. 285 * 286 * Wrapper functions like 'vm_inject_gp()' should be preferred to calling 287 * this function directly because they enforce the trap-like or fault-like 288 * behavior of an exception. 289 * 290 * This function should only be called in the context of the thread that is 291 * executing this vcpu. 292 */ 293 int vm_inject_exception(struct vm *vm, int vcpuid, uint8_t vector, 294 bool err_valid, uint32_t errcode, bool restart_instruction); 295 296 /* 297 * This function is called after a VM-exit that occurred during exception or 298 * interrupt delivery through the IDT. The format of 'intinfo' is described 299 * in Figure 15-1, "EXITINTINFO for All Intercepts", APM, Vol 2. 300 * 301 * If a VM-exit handler completes the event delivery successfully then it 302 * should call vm_exit_intinfo() to extinguish the pending event. For e.g., 303 * if the task switch emulation is triggered via a task gate then it should 304 * call this function with 'intinfo=0' to indicate that the external event 305 * is not pending anymore. 306 * 307 * Return value is 0 on success and non-zero on failure. 308 */ 309 int vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t intinfo); 310 311 /* 312 * This function is called before every VM-entry to retrieve a pending 313 * event that should be injected into the guest. This function combines 314 * nested events into a double or triple fault. 315 * 316 * Returns false if there are no events that need to be injected into the guest. 317 */ 318 bool vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *info); 319 320 int vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2); 321 322 enum vm_reg_name vm_segment_name(int seg_encoding); 323 324 struct vm_copyinfo { 325 uint64_t gpa; 326 size_t len; 327 int prot; 328 void *hva; 329 void *cookie; 330 }; 331 332 /* 333 * Set up 'copyinfo[]' to copy to/from guest linear address space starting 334 * at 'gla' and 'len' bytes long. The 'prot' should be set to PROT_READ for 335 * a copyin or PROT_WRITE for a copyout. 336 * 337 * retval is_fault Interpretation 338 * 0 0 Success 339 * 0 1 An exception was injected into the guest 340 * EFAULT N/A Unrecoverable error 341 * 342 * The 'copyinfo[]' can be passed to 'vm_copyin()' or 'vm_copyout()' only if 343 * the return value is 0. The 'copyinfo[]' resources should be freed by calling 344 * 'vm_copy_teardown()' after the copy is done. 345 */ 346 int vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging, 347 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 348 uint_t num_copyinfo, int *is_fault); 349 void vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, 350 uint_t num_copyinfo); 351 void vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, 352 void *kaddr, size_t len); 353 void vm_copyout(struct vm *vm, int vcpuid, const void *kaddr, 354 struct vm_copyinfo *copyinfo, size_t len); 355 356 int vcpu_trace_exceptions(struct vm *vm, int vcpuid); 357 358 void vm_inject_ud(struct vm *vm, int vcpuid); 359 void vm_inject_gp(struct vm *vm, int vcpuid); 360 void vm_inject_ac(struct vm *vm, int vcpuid, uint32_t errcode); 361 void vm_inject_ss(struct vm *vm, int vcpuid, uint32_t errcode); 362 void vm_inject_pf(struct vm *vm, int vcpuid, uint32_t errcode, uint64_t cr2); 363 364 /* 365 * Both SVM and VMX have complex logic for injecting events such as exceptions 366 * or interrupts into the guest. Within those two backends, the progress of 367 * event injection is tracked by event_inject_state, hopefully making it easier 368 * to reason about. 369 */ 370 enum event_inject_state { 371 EIS_CAN_INJECT = 0, /* exception/interrupt can be injected */ 372 EIS_EV_EXISTING = 1, /* blocked by existing event */ 373 EIS_EV_INJECTED = 2, /* blocked by injected event */ 374 EIS_GI_BLOCK = 3, /* blocked by guest interruptability */ 375 376 /* 377 * Flag to request an immediate exit from VM context after event 378 * injection in order to perform more processing 379 */ 380 EIS_REQ_EXIT = (1 << 15), 381 }; 382 383 /* Possible result codes for MSR access emulation */ 384 typedef enum vm_msr_result { 385 VMR_OK = 0, /* succesfully emulated */ 386 VMR_GP = 1, /* #GP should be injected */ 387 VMR_UNHANLDED = 2, /* handle in userspace, kernel cannot emulate */ 388 } vm_msr_result_t; 389 390 enum vm_cpuid_capability { 391 VCC_NONE, 392 VCC_NO_EXECUTE, 393 VCC_FFXSR, 394 VCC_TCE, 395 VCC_LAST 396 }; 397 398 int x86_emulate_cpuid(struct vm *, int, uint64_t *, uint64_t *, uint64_t *, 399 uint64_t *); 400 bool vm_cpuid_capability(struct vm *, int, enum vm_cpuid_capability); 401 bool validate_guest_xcr0(uint64_t, uint64_t); 402 403 void vmm_sol_glue_init(void); 404 void vmm_sol_glue_cleanup(void); 405 406 void *vmm_contig_alloc(size_t); 407 void vmm_contig_free(void *, size_t); 408 409 int vmm_mod_load(void); 410 int vmm_mod_unload(void); 411 412 bool vmm_check_iommu(void); 413 414 void vmm_call_trap(uint64_t); 415 416 /* 417 * Because of tangled headers, this is not exposed directly via the vmm_drv 418 * interface, but rather mirrored as vmm_drv_iop_cb_t in vmm_drv.h. 419 */ 420 typedef int (*ioport_handler_t)(void *, bool, uint16_t, uint8_t, uint32_t *); 421 422 int vm_ioport_access(struct vm *vm, int vcpuid, bool in, uint16_t port, 423 uint8_t bytes, uint32_t *val); 424 425 int vm_ioport_attach(struct vm *vm, uint16_t port, ioport_handler_t func, 426 void *arg, void **cookie); 427 int vm_ioport_detach(struct vm *vm, void **cookie, ioport_handler_t *old_func, 428 void **old_arg); 429 430 int vm_ioport_hook(struct vm *, uint16_t, ioport_handler_t, void *, void **); 431 void vm_ioport_unhook(struct vm *, void **); 432 433 enum vcpu_ustate { 434 VU_INIT = 0, /* initialized but has not yet attempted to run */ 435 VU_RUN, /* running in guest context */ 436 VU_IDLE, /* idle (HLTed, wait-for-SIPI, etc) */ 437 VU_EMU_KERN, /* emulation performed in-kernel */ 438 VU_EMU_USER, /* emulation performed in userspace */ 439 VU_SCHED, /* off-cpu for interrupt, preempt, lock contention */ 440 VU_MAX 441 }; 442 443 void vcpu_ustate_change(struct vm *, int, enum vcpu_ustate); 444 445 typedef struct vmm_kstats { 446 kstat_named_t vk_name; 447 } vmm_kstats_t; 448 449 typedef struct vmm_vcpu_kstats { 450 kstat_named_t vvk_vcpu; 451 kstat_named_t vvk_time_init; 452 kstat_named_t vvk_time_run; 453 kstat_named_t vvk_time_idle; 454 kstat_named_t vvk_time_emu_kern; 455 kstat_named_t vvk_time_emu_user; 456 kstat_named_t vvk_time_sched; 457 } vmm_vcpu_kstats_t; 458 459 #define VMM_KSTAT_CLASS "misc" 460 461 int vmm_kstat_update_vcpu(struct kstat *, int); 462 463 typedef struct vmm_data_req { 464 uint16_t vdr_class; 465 uint16_t vdr_version; 466 uint32_t vdr_flags; 467 uint32_t vdr_len; 468 void *vdr_data; 469 uint32_t *vdr_result_len; 470 } vmm_data_req_t; 471 typedef struct vmm_data_req vmm_data_req_t; 472 473 typedef int (*vmm_data_writef_t)(void *, const vmm_data_req_t *); 474 typedef int (*vmm_data_readf_t)(void *, const vmm_data_req_t *); 475 476 typedef struct vmm_data_version_entry { 477 uint16_t vdve_class; 478 uint16_t vdve_version; 479 uint16_t vdve_len_expect; 480 uint16_t vdve_len_per_item; 481 vmm_data_readf_t vdve_readf; 482 vmm_data_writef_t vdve_writef; 483 } vmm_data_version_entry_t; 484 485 #define VMM_DATA_VERSION(sym) SET_ENTRY(vmm_data_version_entries, sym) 486 487 int vmm_data_read(struct vm *, int, const vmm_data_req_t *); 488 int vmm_data_write(struct vm *, int, const vmm_data_req_t *); 489 490 #endif /* _VMM_KERNEL_H_ */ 491