master
1/*-
2 * SPDX-License-Identifier: BSD-2-Clause
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
29#ifndef _VMM_H_
30#define _VMM_H_
31
32#include <sys/cpuset.h>
33#include <sys/sdt.h>
34#include <x86/segments.h>
35
36struct vcpu;
37struct vm_snapshot_meta;
38
39#ifdef _KERNEL
40SDT_PROVIDER_DECLARE(vmm);
41#endif
42
43enum vm_suspend_how {
44 VM_SUSPEND_NONE,
45 VM_SUSPEND_RESET,
46 VM_SUSPEND_POWEROFF,
47 VM_SUSPEND_HALT,
48 VM_SUSPEND_TRIPLEFAULT,
49 VM_SUSPEND_LAST
50};
51
52/*
53 * Identifiers for architecturally defined registers.
54 */
55enum vm_reg_name {
56 VM_REG_GUEST_RAX,
57 VM_REG_GUEST_RBX,
58 VM_REG_GUEST_RCX,
59 VM_REG_GUEST_RDX,
60 VM_REG_GUEST_RSI,
61 VM_REG_GUEST_RDI,
62 VM_REG_GUEST_RBP,
63 VM_REG_GUEST_R8,
64 VM_REG_GUEST_R9,
65 VM_REG_GUEST_R10,
66 VM_REG_GUEST_R11,
67 VM_REG_GUEST_R12,
68 VM_REG_GUEST_R13,
69 VM_REG_GUEST_R14,
70 VM_REG_GUEST_R15,
71 VM_REG_GUEST_CR0,
72 VM_REG_GUEST_CR3,
73 VM_REG_GUEST_CR4,
74 VM_REG_GUEST_DR7,
75 VM_REG_GUEST_RSP,
76 VM_REG_GUEST_RIP,
77 VM_REG_GUEST_RFLAGS,
78 VM_REG_GUEST_ES,
79 VM_REG_GUEST_CS,
80 VM_REG_GUEST_SS,
81 VM_REG_GUEST_DS,
82 VM_REG_GUEST_FS,
83 VM_REG_GUEST_GS,
84 VM_REG_GUEST_LDTR,
85 VM_REG_GUEST_TR,
86 VM_REG_GUEST_IDTR,
87 VM_REG_GUEST_GDTR,
88 VM_REG_GUEST_EFER,
89 VM_REG_GUEST_CR2,
90 VM_REG_GUEST_PDPTE0,
91 VM_REG_GUEST_PDPTE1,
92 VM_REG_GUEST_PDPTE2,
93 VM_REG_GUEST_PDPTE3,
94 VM_REG_GUEST_INTR_SHADOW,
95 VM_REG_GUEST_DR0,
96 VM_REG_GUEST_DR1,
97 VM_REG_GUEST_DR2,
98 VM_REG_GUEST_DR3,
99 VM_REG_GUEST_DR6,
100 VM_REG_GUEST_ENTRY_INST_LENGTH,
101 VM_REG_GUEST_FS_BASE,
102 VM_REG_GUEST_GS_BASE,
103 VM_REG_GUEST_KGS_BASE,
104 VM_REG_GUEST_TPR,
105 VM_REG_LAST
106};
107
108enum x2apic_state {
109 X2APIC_DISABLED,
110 X2APIC_ENABLED,
111 X2APIC_STATE_LAST
112};
113
114#define VM_INTINFO_VECTOR(info) ((info) & 0xff)
115#define VM_INTINFO_DEL_ERRCODE 0x800
116#define VM_INTINFO_RSVD 0x7ffff000
117#define VM_INTINFO_VALID 0x80000000
118#define VM_INTINFO_TYPE 0x700
119#define VM_INTINFO_HWINTR (0 << 8)
120#define VM_INTINFO_NMI (2 << 8)
121#define VM_INTINFO_HWEXCEPTION (3 << 8)
122#define VM_INTINFO_SWINTR (4 << 8)
123
124/*
125 * The VM name has to fit into the pathname length constraints of devfs,
126 * governed primarily by SPECNAMELEN. The length is the total number of
127 * characters in the full path, relative to the mount point and not
128 * including any leading '/' characters.
129 * A prefix and a suffix are added to the name specified by the user.
130 * The prefix is usually "vmm/" or "vmm.io/", but can be a few characters
131 * longer for future use.
132 * The suffix is a string that identifies a bootrom image or some similar
133 * image that is attached to the VM. A separator character gets added to
134 * the suffix automatically when generating the full path, so it must be
135 * accounted for, reducing the effective length by 1.
136 * The effective length of a VM name is 229 bytes for FreeBSD 13 and 37
137 * bytes for FreeBSD 12. A minimum length is set for safety and supports
138 * a SPECNAMELEN as small as 32 on old systems.
139 */
140#define VM_MAX_PREFIXLEN 10
141#define VM_MAX_SUFFIXLEN 15
142#define VM_MIN_NAMELEN 6
143#define VM_MAX_NAMELEN \
144 (SPECNAMELEN - VM_MAX_PREFIXLEN - VM_MAX_SUFFIXLEN - 1)
145
146#ifdef _KERNEL
147#include <sys/kassert.h>
148
149CTASSERT(VM_MAX_NAMELEN >= VM_MIN_NAMELEN);
150
151struct vm;
152struct vm_exception;
153struct seg_desc;
154struct vm_exit;
155struct vm_run;
156struct vhpet;
157struct vioapic;
158struct vlapic;
159struct vmspace;
160struct vm_object;
161struct vm_guest_paging;
162struct pmap;
163enum snapshot_req;
164
165struct vm_eventinfo {
166 cpuset_t *rptr; /* rendezvous cookie */
167 int *sptr; /* suspend cookie */
168 int *iptr; /* reqidle cookie */
169};
170
171typedef int (*vmm_init_func_t)(int ipinum);
172typedef int (*vmm_cleanup_func_t)(void);
173typedef void (*vmm_resume_func_t)(void);
174typedef void * (*vmi_init_func_t)(struct vm *vm, struct pmap *pmap);
175typedef int (*vmi_run_func_t)(void *vcpui, register_t rip,
176 struct pmap *pmap, struct vm_eventinfo *info);
177typedef void (*vmi_cleanup_func_t)(void *vmi);
178typedef void * (*vmi_vcpu_init_func_t)(void *vmi, struct vcpu *vcpu,
179 int vcpu_id);
180typedef void (*vmi_vcpu_cleanup_func_t)(void *vcpui);
181typedef int (*vmi_get_register_t)(void *vcpui, int num, uint64_t *retval);
182typedef int (*vmi_set_register_t)(void *vcpui, int num, uint64_t val);
183typedef int (*vmi_get_desc_t)(void *vcpui, int num, struct seg_desc *desc);
184typedef int (*vmi_set_desc_t)(void *vcpui, int num, struct seg_desc *desc);
185typedef int (*vmi_get_cap_t)(void *vcpui, int num, int *retval);
186typedef int (*vmi_set_cap_t)(void *vcpui, int num, int val);
187typedef struct vmspace * (*vmi_vmspace_alloc)(vm_offset_t min, vm_offset_t max);
188typedef void (*vmi_vmspace_free)(struct vmspace *vmspace);
189typedef struct vlapic * (*vmi_vlapic_init)(void *vcpui);
190typedef void (*vmi_vlapic_cleanup)(struct vlapic *vlapic);
191typedef int (*vmi_snapshot_vcpu_t)(void *vcpui, struct vm_snapshot_meta *meta);
192typedef int (*vmi_restore_tsc_t)(void *vcpui, uint64_t now);
193
194struct vmm_ops {
195 vmm_init_func_t modinit; /* module wide initialization */
196 vmm_cleanup_func_t modcleanup;
197 vmm_resume_func_t modresume;
198
199 vmi_init_func_t init; /* vm-specific initialization */
200 vmi_run_func_t run;
201 vmi_cleanup_func_t cleanup;
202 vmi_vcpu_init_func_t vcpu_init;
203 vmi_vcpu_cleanup_func_t vcpu_cleanup;
204 vmi_get_register_t getreg;
205 vmi_set_register_t setreg;
206 vmi_get_desc_t getdesc;
207 vmi_set_desc_t setdesc;
208 vmi_get_cap_t getcap;
209 vmi_set_cap_t setcap;
210 vmi_vmspace_alloc vmspace_alloc;
211 vmi_vmspace_free vmspace_free;
212 vmi_vlapic_init vlapic_init;
213 vmi_vlapic_cleanup vlapic_cleanup;
214
215 /* checkpoint operations */
216 vmi_snapshot_vcpu_t vcpu_snapshot;
217 vmi_restore_tsc_t restore_tsc;
218};
219
220extern const struct vmm_ops vmm_ops_intel;
221extern const struct vmm_ops vmm_ops_amd;
222
223extern u_int vm_maxcpu; /* maximum virtual cpus */
224
225int vm_create(const char *name, struct vm **retvm);
226struct vcpu *vm_alloc_vcpu(struct vm *vm, int vcpuid);
227void vm_disable_vcpu_creation(struct vm *vm);
228void vm_slock_vcpus(struct vm *vm);
229void vm_unlock_vcpus(struct vm *vm);
230void vm_destroy(struct vm *vm);
231int vm_reinit(struct vm *vm);
232const char *vm_name(struct vm *vm);
233uint16_t vm_get_maxcpus(struct vm *vm);
234void vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores,
235 uint16_t *threads, uint16_t *maxcpus);
236int vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores,
237 uint16_t threads, uint16_t maxcpus);
238
239/*
240 * APIs that modify the guest memory map require all vcpus to be frozen.
241 */
242void vm_slock_memsegs(struct vm *vm);
243void vm_xlock_memsegs(struct vm *vm);
244void vm_unlock_memsegs(struct vm *vm);
245int vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t off,
246 size_t len, int prot, int flags);
247int vm_munmap_memseg(struct vm *vm, vm_paddr_t gpa, size_t len);
248int vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem);
249void vm_free_memseg(struct vm *vm, int ident);
250int vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa);
251int vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len);
252int vm_assign_pptdev(struct vm *vm, int bus, int slot, int func);
253int vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func);
254
255/*
256 * APIs that inspect the guest memory map require only a *single* vcpu to
257 * be frozen. This acts like a read lock on the guest memory map since any
258 * modification requires *all* vcpus to be frozen.
259 */
260int vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
261 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags);
262int vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
263 struct vm_object **objptr);
264vm_paddr_t vmm_sysmem_maxaddr(struct vm *vm);
265void *vm_gpa_hold(struct vcpu *vcpu, vm_paddr_t gpa, size_t len,
266 int prot, void **cookie);
267void *vm_gpa_hold_global(struct vm *vm, vm_paddr_t gpa, size_t len,
268 int prot, void **cookie);
269void vm_gpa_release(void *cookie);
270bool vm_mem_allocated(struct vcpu *vcpu, vm_paddr_t gpa);
271
272int vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval);
273int vm_set_register(struct vcpu *vcpu, int reg, uint64_t val);
274int vm_get_seg_desc(struct vcpu *vcpu, int reg,
275 struct seg_desc *ret_desc);
276int vm_set_seg_desc(struct vcpu *vcpu, int reg,
277 struct seg_desc *desc);
278int vm_run(struct vcpu *vcpu);
279int vm_suspend(struct vm *vm, enum vm_suspend_how how);
280int vm_inject_nmi(struct vcpu *vcpu);
281int vm_nmi_pending(struct vcpu *vcpu);
282void vm_nmi_clear(struct vcpu *vcpu);
283int vm_inject_extint(struct vcpu *vcpu);
284int vm_extint_pending(struct vcpu *vcpu);
285void vm_extint_clear(struct vcpu *vcpu);
286int vcpu_vcpuid(struct vcpu *vcpu);
287struct vm *vcpu_vm(struct vcpu *vcpu);
288struct vcpu *vm_vcpu(struct vm *vm, int cpu);
289struct vlapic *vm_lapic(struct vcpu *vcpu);
290struct vioapic *vm_ioapic(struct vm *vm);
291struct vhpet *vm_hpet(struct vm *vm);
292int vm_get_capability(struct vcpu *vcpu, int type, int *val);
293int vm_set_capability(struct vcpu *vcpu, int type, int val);
294int vm_get_x2apic_state(struct vcpu *vcpu, enum x2apic_state *state);
295int vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state);
296int vm_apicid2vcpuid(struct vm *vm, int apicid);
297int vm_activate_cpu(struct vcpu *vcpu);
298int vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu);
299int vm_resume_cpu(struct vm *vm, struct vcpu *vcpu);
300int vm_restart_instruction(struct vcpu *vcpu);
301struct vm_exit *vm_exitinfo(struct vcpu *vcpu);
302cpuset_t *vm_exitinfo_cpuset(struct vcpu *vcpu);
303void vm_exit_suspended(struct vcpu *vcpu, uint64_t rip);
304void vm_exit_debug(struct vcpu *vcpu, uint64_t rip);
305void vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip);
306void vm_exit_astpending(struct vcpu *vcpu, uint64_t rip);
307void vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip);
308int vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta);
309int vm_restore_time(struct vm *vm);
310
311#ifdef _SYS__CPUSET_H_
312/*
313 * Rendezvous all vcpus specified in 'dest' and execute 'func(arg)'.
314 * The rendezvous 'func(arg)' is not allowed to do anything that will
315 * cause the thread to be put to sleep.
316 *
317 * The caller cannot hold any locks when initiating the rendezvous.
318 *
319 * The implementation of this API may cause vcpus other than those specified
320 * by 'dest' to be stalled. The caller should not rely on any vcpus making
321 * forward progress when the rendezvous is in progress.
322 */
323typedef void (*vm_rendezvous_func_t)(struct vcpu *vcpu, void *arg);
324int vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest,
325 vm_rendezvous_func_t func, void *arg);
326
327cpuset_t vm_active_cpus(struct vm *vm);
328cpuset_t vm_debug_cpus(struct vm *vm);
329cpuset_t vm_suspended_cpus(struct vm *vm);
330cpuset_t vm_start_cpus(struct vm *vm, const cpuset_t *tostart);
331void vm_await_start(struct vm *vm, const cpuset_t *waiting);
332#endif /* _SYS__CPUSET_H_ */
333
334static __inline int
335vcpu_rendezvous_pending(struct vcpu *vcpu, struct vm_eventinfo *info)
336{
337 /*
338 * This check isn't done with atomic operations or under a lock because
339 * there's no need to. If the vcpuid bit is set, the vcpu is part of a
340 * rendezvous and the bit won't be cleared until the vcpu enters the
341 * rendezvous. On rendezvous exit, the cpuset is cleared and the vcpu
342 * will see an empty cpuset. So, the races are harmless.
343 */
344 return (CPU_ISSET(vcpu_vcpuid(vcpu), info->rptr));
345}
346
347static __inline int
348vcpu_suspended(struct vm_eventinfo *info)
349{
350
351 return (*info->sptr);
352}
353
354static __inline int
355vcpu_reqidle(struct vm_eventinfo *info)
356{
357
358 return (*info->iptr);
359}
360
361int vcpu_debugged(struct vcpu *vcpu);
362
363/*
364 * Return true if device indicated by bus/slot/func is supposed to be a
365 * pci passthrough device.
366 *
367 * Return false otherwise.
368 */
369bool vmm_is_pptdev(int bus, int slot, int func);
370
371void *vm_iommu_domain(struct vm *vm);
372
373enum vcpu_state {
374 VCPU_IDLE,
375 VCPU_FROZEN,
376 VCPU_RUNNING,
377 VCPU_SLEEPING,
378};
379
380int vcpu_set_state(struct vcpu *vcpu, enum vcpu_state state, bool from_idle);
381enum vcpu_state vcpu_get_state(struct vcpu *vcpu, int *hostcpu);
382
383static int __inline
384vcpu_is_running(struct vcpu *vcpu, int *hostcpu)
385{
386 return (vcpu_get_state(vcpu, hostcpu) == VCPU_RUNNING);
387}
388
389#ifdef _SYS_PROC_H_
390static int __inline
391vcpu_should_yield(struct vcpu *vcpu)
392{
393 struct thread *td;
394
395 td = curthread;
396 return (td->td_ast != 0 || td->td_owepreempt != 0);
397}
398#endif
399
400void *vcpu_stats(struct vcpu *vcpu);
401void vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr);
402struct vmspace *vm_get_vmspace(struct vm *vm);
403struct vatpic *vm_atpic(struct vm *vm);
404struct vatpit *vm_atpit(struct vm *vm);
405struct vpmtmr *vm_pmtmr(struct vm *vm);
406struct vrtc *vm_rtc(struct vm *vm);
407
408/*
409 * Inject exception 'vector' into the guest vcpu. This function returns 0 on
410 * success and non-zero on failure.
411 *
412 * Wrapper functions like 'vm_inject_gp()' should be preferred to calling
413 * this function directly because they enforce the trap-like or fault-like
414 * behavior of an exception.
415 *
416 * This function should only be called in the context of the thread that is
417 * executing this vcpu.
418 */
419int vm_inject_exception(struct vcpu *vcpu, int vector, int err_valid,
420 uint32_t errcode, int restart_instruction);
421
422/*
423 * This function is called after a VM-exit that occurred during exception or
424 * interrupt delivery through the IDT. The format of 'intinfo' is described
425 * in Figure 15-1, "EXITINTINFO for All Intercepts", APM, Vol 2.
426 *
427 * If a VM-exit handler completes the event delivery successfully then it
428 * should call vm_exit_intinfo() to extinguish the pending event. For e.g.,
429 * if the task switch emulation is triggered via a task gate then it should
430 * call this function with 'intinfo=0' to indicate that the external event
431 * is not pending anymore.
432 *
433 * Return value is 0 on success and non-zero on failure.
434 */
435int vm_exit_intinfo(struct vcpu *vcpu, uint64_t intinfo);
436
437/*
438 * This function is called before every VM-entry to retrieve a pending
439 * event that should be injected into the guest. This function combines
440 * nested events into a double or triple fault.
441 *
442 * Returns 0 if there are no events that need to be injected into the guest
443 * and non-zero otherwise.
444 */
445int vm_entry_intinfo(struct vcpu *vcpu, uint64_t *info);
446
447int vm_get_intinfo(struct vcpu *vcpu, uint64_t *info1, uint64_t *info2);
448
449/*
450 * Function used to keep track of the guest's TSC offset. The
451 * offset is used by the virutalization extensions to provide a consistent
452 * value for the Time Stamp Counter to the guest.
453 */
454void vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset);
455
456enum vm_reg_name vm_segment_name(int seg_encoding);
457
458struct vm_copyinfo {
459 uint64_t gpa;
460 size_t len;
461 void *hva;
462 void *cookie;
463};
464
465/*
466 * Set up 'copyinfo[]' to copy to/from guest linear address space starting
467 * at 'gla' and 'len' bytes long. The 'prot' should be set to PROT_READ for
468 * a copyin or PROT_WRITE for a copyout.
469 *
470 * retval is_fault Interpretation
471 * 0 0 Success
472 * 0 1 An exception was injected into the guest
473 * EFAULT N/A Unrecoverable error
474 *
475 * The 'copyinfo[]' can be passed to 'vm_copyin()' or 'vm_copyout()' only if
476 * the return value is 0. The 'copyinfo[]' resources should be freed by calling
477 * 'vm_copy_teardown()' after the copy is done.
478 */
479int vm_copy_setup(struct vcpu *vcpu, struct vm_guest_paging *paging,
480 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
481 int num_copyinfo, int *is_fault);
482void vm_copy_teardown(struct vm_copyinfo *copyinfo, int num_copyinfo);
483void vm_copyin(struct vm_copyinfo *copyinfo, void *kaddr, size_t len);
484void vm_copyout(const void *kaddr, struct vm_copyinfo *copyinfo, size_t len);
485
486int vcpu_trace_exceptions(struct vcpu *vcpu);
487int vcpu_trap_wbinvd(struct vcpu *vcpu);
488#endif /* KERNEL */
489
490/*
491 * Identifiers for optional vmm capabilities
492 */
493enum vm_cap_type {
494 VM_CAP_HALT_EXIT,
495 VM_CAP_MTRAP_EXIT,
496 VM_CAP_PAUSE_EXIT,
497 VM_CAP_UNRESTRICTED_GUEST,
498 VM_CAP_ENABLE_INVPCID,
499 VM_CAP_BPT_EXIT,
500 VM_CAP_RDPID,
501 VM_CAP_RDTSCP,
502 VM_CAP_IPI_EXIT,
503 VM_CAP_MASK_HWINTR,
504 VM_CAP_RFLAGS_TF,
505 VM_CAP_MAX
506};
507
508enum vm_intr_trigger {
509 EDGE_TRIGGER,
510 LEVEL_TRIGGER
511};
512
513/*
514 * The 'access' field has the format specified in Table 21-2 of the Intel
515 * Architecture Manual vol 3b.
516 *
517 * XXX The contents of the 'access' field are architecturally defined except
518 * bit 16 - Segment Unusable.
519 */
520struct seg_desc {
521 uint64_t base;
522 uint32_t limit;
523 uint32_t access;
524};
525#define SEG_DESC_TYPE(access) ((access) & 0x001f)
526#define SEG_DESC_DPL(access) (((access) >> 5) & 0x3)
527#define SEG_DESC_PRESENT(access) (((access) & 0x0080) ? 1 : 0)
528#define SEG_DESC_DEF32(access) (((access) & 0x4000) ? 1 : 0)
529#define SEG_DESC_GRANULARITY(access) (((access) & 0x8000) ? 1 : 0)
530#define SEG_DESC_UNUSABLE(access) (((access) & 0x10000) ? 1 : 0)
531
532enum vm_cpu_mode {
533 CPU_MODE_REAL,
534 CPU_MODE_PROTECTED,
535 CPU_MODE_COMPATIBILITY, /* IA-32E mode (CS.L = 0) */
536 CPU_MODE_64BIT, /* IA-32E mode (CS.L = 1) */
537};
538
539enum vm_paging_mode {
540 PAGING_MODE_FLAT,
541 PAGING_MODE_32,
542 PAGING_MODE_PAE,
543 PAGING_MODE_64,
544 PAGING_MODE_64_LA57,
545};
546
547struct vm_guest_paging {
548 uint64_t cr3;
549 int cpl;
550 enum vm_cpu_mode cpu_mode;
551 enum vm_paging_mode paging_mode;
552};
553
554/*
555 * The data structures 'vie' and 'vie_op' are meant to be opaque to the
556 * consumers of instruction decoding. The only reason why their contents
557 * need to be exposed is because they are part of the 'vm_exit' structure.
558 */
559struct vie_op {
560 uint8_t op_byte; /* actual opcode byte */
561 uint8_t op_type; /* type of operation (e.g. MOV) */
562 uint16_t op_flags;
563};
564_Static_assert(sizeof(struct vie_op) == 4, "ABI");
565_Static_assert(_Alignof(struct vie_op) == 2, "ABI");
566
567#define VIE_INST_SIZE 15
568struct vie {
569 uint8_t inst[VIE_INST_SIZE]; /* instruction bytes */
570 uint8_t num_valid; /* size of the instruction */
571
572/* The following fields are all zeroed upon restart. */
573#define vie_startzero num_processed
574 uint8_t num_processed;
575
576 uint8_t addrsize:4, opsize:4; /* address and operand sizes */
577 uint8_t rex_w:1, /* REX prefix */
578 rex_r:1,
579 rex_x:1,
580 rex_b:1,
581 rex_present:1,
582 repz_present:1, /* REP/REPE/REPZ prefix */
583 repnz_present:1, /* REPNE/REPNZ prefix */
584 opsize_override:1, /* Operand size override */
585 addrsize_override:1, /* Address size override */
586 segment_override:1; /* Segment override */
587
588 uint8_t mod:2, /* ModRM byte */
589 reg:4,
590 rm:4;
591
592 uint8_t ss:2, /* SIB byte */
593 vex_present:1, /* VEX prefixed */
594 vex_l:1, /* L bit */
595 index:4, /* SIB byte */
596 base:4; /* SIB byte */
597
598 uint8_t disp_bytes;
599 uint8_t imm_bytes;
600
601 uint8_t scale;
602
603 uint8_t vex_reg:4, /* vvvv: first source register specifier */
604 vex_pp:2, /* pp */
605 _sparebits:2;
606
607 uint8_t _sparebytes[2];
608
609 int base_register; /* VM_REG_GUEST_xyz */
610 int index_register; /* VM_REG_GUEST_xyz */
611 int segment_register; /* VM_REG_GUEST_xyz */
612
613 int64_t displacement; /* optional addr displacement */
614 int64_t immediate; /* optional immediate operand */
615
616 uint8_t decoded; /* set to 1 if successfully decoded */
617
618 uint8_t _sparebyte;
619
620 struct vie_op op; /* opcode description */
621};
622_Static_assert(sizeof(struct vie) == 64, "ABI");
623_Static_assert(__offsetof(struct vie, disp_bytes) == 22, "ABI");
624_Static_assert(__offsetof(struct vie, scale) == 24, "ABI");
625_Static_assert(__offsetof(struct vie, base_register) == 28, "ABI");
626
627enum vm_exitcode {
628 VM_EXITCODE_INOUT,
629 VM_EXITCODE_VMX,
630 VM_EXITCODE_BOGUS,
631 VM_EXITCODE_RDMSR,
632 VM_EXITCODE_WRMSR,
633 VM_EXITCODE_HLT,
634 VM_EXITCODE_MTRAP,
635 VM_EXITCODE_PAUSE,
636 VM_EXITCODE_PAGING,
637 VM_EXITCODE_INST_EMUL,
638 VM_EXITCODE_SPINUP_AP,
639 VM_EXITCODE_DEPRECATED1, /* used to be SPINDOWN_CPU */
640 VM_EXITCODE_RENDEZVOUS,
641 VM_EXITCODE_IOAPIC_EOI,
642 VM_EXITCODE_SUSPENDED,
643 VM_EXITCODE_INOUT_STR,
644 VM_EXITCODE_TASK_SWITCH,
645 VM_EXITCODE_MONITOR,
646 VM_EXITCODE_MWAIT,
647 VM_EXITCODE_SVM,
648 VM_EXITCODE_REQIDLE,
649 VM_EXITCODE_DEBUG,
650 VM_EXITCODE_VMINSN,
651 VM_EXITCODE_BPT,
652 VM_EXITCODE_IPI,
653 VM_EXITCODE_DB,
654 VM_EXITCODE_MAX
655};
656
657struct vm_inout {
658 uint16_t bytes:3; /* 1 or 2 or 4 */
659 uint16_t in:1;
660 uint16_t string:1;
661 uint16_t rep:1;
662 uint16_t port;
663 uint32_t eax; /* valid for out */
664};
665
666struct vm_inout_str {
667 struct vm_inout inout; /* must be the first element */
668 struct vm_guest_paging paging;
669 uint64_t rflags;
670 uint64_t cr0;
671 uint64_t index;
672 uint64_t count; /* rep=1 (%rcx), rep=0 (1) */
673 int addrsize;
674 enum vm_reg_name seg_name;
675 struct seg_desc seg_desc;
676};
677
678enum task_switch_reason {
679 TSR_CALL,
680 TSR_IRET,
681 TSR_JMP,
682 TSR_IDT_GATE, /* task gate in IDT */
683};
684
685struct vm_task_switch {
686 uint16_t tsssel; /* new TSS selector */
687 int ext; /* task switch due to external event */
688 uint32_t errcode;
689 int errcode_valid; /* push 'errcode' on the new stack */
690 enum task_switch_reason reason;
691 struct vm_guest_paging paging;
692};
693
694struct vm_exit {
695 enum vm_exitcode exitcode;
696 int inst_length; /* 0 means unknown */
697 uint64_t rip;
698 union {
699 struct vm_inout inout;
700 struct vm_inout_str inout_str;
701 struct {
702 uint64_t gpa;
703 int fault_type;
704 } paging;
705 struct {
706 uint64_t gpa;
707 uint64_t gla;
708 uint64_t cs_base;
709 int cs_d; /* CS.D */
710 struct vm_guest_paging paging;
711 struct vie vie;
712 } inst_emul;
713 /*
714 * VMX specific payload. Used when there is no "better"
715 * exitcode to represent the VM-exit.
716 */
717 struct {
718 int status; /* vmx inst status */
719 /*
720 * 'exit_reason' and 'exit_qualification' are valid
721 * only if 'status' is zero.
722 */
723 uint32_t exit_reason;
724 uint64_t exit_qualification;
725 /*
726 * 'inst_error' and 'inst_type' are valid
727 * only if 'status' is non-zero.
728 */
729 int inst_type;
730 int inst_error;
731 } vmx;
732 /*
733 * SVM specific payload.
734 */
735 struct {
736 uint64_t exitcode;
737 uint64_t exitinfo1;
738 uint64_t exitinfo2;
739 } svm;
740 struct {
741 int inst_length;
742 } bpt;
743 struct {
744 int trace_trap;
745 int pushf_intercept;
746 int tf_shadow_val;
747 struct vm_guest_paging paging;
748 } dbg;
749 struct {
750 uint32_t code; /* ecx value */
751 uint64_t wval;
752 } msr;
753 struct {
754 int vcpu;
755 uint64_t rip;
756 } spinup_ap;
757 struct {
758 uint64_t rflags;
759 uint64_t intr_status;
760 } hlt;
761 struct {
762 int vector;
763 } ioapic_eoi;
764 struct {
765 enum vm_suspend_how how;
766 } suspended;
767 struct {
768 /*
769 * The destination vCPU mask is saved in vcpu->cpuset
770 * and is copied out to userspace separately to avoid
771 * ABI concerns.
772 */
773 uint32_t mode;
774 uint8_t vector;
775 } ipi;
776 struct vm_task_switch task_switch;
777 } u;
778};
779
780/* APIs to inject faults into the guest */
781void vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid,
782 int errcode);
783
784static __inline void
785vm_inject_ud(struct vcpu *vcpu)
786{
787 vm_inject_fault(vcpu, IDT_UD, 0, 0);
788}
789
790static __inline void
791vm_inject_gp(struct vcpu *vcpu)
792{
793 vm_inject_fault(vcpu, IDT_GP, 1, 0);
794}
795
796static __inline void
797vm_inject_ac(struct vcpu *vcpu, int errcode)
798{
799 vm_inject_fault(vcpu, IDT_AC, 1, errcode);
800}
801
802static __inline void
803vm_inject_ss(struct vcpu *vcpu, int errcode)
804{
805 vm_inject_fault(vcpu, IDT_SS, 1, errcode);
806}
807
808void vm_inject_pf(struct vcpu *vcpu, int error_code, uint64_t cr2);
809
810#endif /* _VMM_H_ */