xref: /freebsd/sys/amd64/include/vmm.h (revision e0c4386e7e71d93b0edc0c8fa156263fc4a8b0b6)
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 
36 struct vcpu;
37 struct vm_snapshot_meta;
38 
39 #ifdef _KERNEL
40 SDT_PROVIDER_DECLARE(vmm);
41 #endif
42 
43 enum 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  */
55 enum 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_LAST
102 };
103 
104 enum x2apic_state {
105 	X2APIC_DISABLED,
106 	X2APIC_ENABLED,
107 	X2APIC_STATE_LAST
108 };
109 
110 #define	VM_INTINFO_VECTOR(info)	((info) & 0xff)
111 #define	VM_INTINFO_DEL_ERRCODE	0x800
112 #define	VM_INTINFO_RSVD		0x7ffff000
113 #define	VM_INTINFO_VALID	0x80000000
114 #define	VM_INTINFO_TYPE		0x700
115 #define	VM_INTINFO_HWINTR	(0 << 8)
116 #define	VM_INTINFO_NMI		(2 << 8)
117 #define	VM_INTINFO_HWEXCEPTION	(3 << 8)
118 #define	VM_INTINFO_SWINTR	(4 << 8)
119 
120 /*
121  * The VM name has to fit into the pathname length constraints of devfs,
122  * governed primarily by SPECNAMELEN.  The length is the total number of
123  * characters in the full path, relative to the mount point and not
124  * including any leading '/' characters.
125  * A prefix and a suffix are added to the name specified by the user.
126  * The prefix is usually "vmm/" or "vmm.io/", but can be a few characters
127  * longer for future use.
128  * The suffix is a string that identifies a bootrom image or some similar
129  * image that is attached to the VM. A separator character gets added to
130  * the suffix automatically when generating the full path, so it must be
131  * accounted for, reducing the effective length by 1.
132  * The effective length of a VM name is 229 bytes for FreeBSD 13 and 37
133  * bytes for FreeBSD 12.  A minimum length is set for safety and supports
134  * a SPECNAMELEN as small as 32 on old systems.
135  */
136 #define VM_MAX_PREFIXLEN 10
137 #define VM_MAX_SUFFIXLEN 15
138 #define VM_MIN_NAMELEN   6
139 #define VM_MAX_NAMELEN \
140     (SPECNAMELEN - VM_MAX_PREFIXLEN - VM_MAX_SUFFIXLEN - 1)
141 
142 #ifdef _KERNEL
143 CTASSERT(VM_MAX_NAMELEN >= VM_MIN_NAMELEN);
144 
145 struct vm;
146 struct vm_exception;
147 struct seg_desc;
148 struct vm_exit;
149 struct vm_run;
150 struct vhpet;
151 struct vioapic;
152 struct vlapic;
153 struct vmspace;
154 struct vm_object;
155 struct vm_guest_paging;
156 struct pmap;
157 enum snapshot_req;
158 
159 struct vm_eventinfo {
160 	cpuset_t *rptr;		/* rendezvous cookie */
161 	int	*sptr;		/* suspend cookie */
162 	int	*iptr;		/* reqidle cookie */
163 };
164 
165 typedef int	(*vmm_init_func_t)(int ipinum);
166 typedef int	(*vmm_cleanup_func_t)(void);
167 typedef void	(*vmm_resume_func_t)(void);
168 typedef void *	(*vmi_init_func_t)(struct vm *vm, struct pmap *pmap);
169 typedef int	(*vmi_run_func_t)(void *vcpui, register_t rip,
170 		    struct pmap *pmap, struct vm_eventinfo *info);
171 typedef void	(*vmi_cleanup_func_t)(void *vmi);
172 typedef void *	(*vmi_vcpu_init_func_t)(void *vmi, struct vcpu *vcpu,
173 		    int vcpu_id);
174 typedef void	(*vmi_vcpu_cleanup_func_t)(void *vcpui);
175 typedef int	(*vmi_get_register_t)(void *vcpui, int num, uint64_t *retval);
176 typedef int	(*vmi_set_register_t)(void *vcpui, int num, uint64_t val);
177 typedef int	(*vmi_get_desc_t)(void *vcpui, int num, struct seg_desc *desc);
178 typedef int	(*vmi_set_desc_t)(void *vcpui, int num, struct seg_desc *desc);
179 typedef int	(*vmi_get_cap_t)(void *vcpui, int num, int *retval);
180 typedef int	(*vmi_set_cap_t)(void *vcpui, int num, int val);
181 typedef struct vmspace * (*vmi_vmspace_alloc)(vm_offset_t min, vm_offset_t max);
182 typedef void	(*vmi_vmspace_free)(struct vmspace *vmspace);
183 typedef struct vlapic * (*vmi_vlapic_init)(void *vcpui);
184 typedef void	(*vmi_vlapic_cleanup)(struct vlapic *vlapic);
185 typedef int	(*vmi_snapshot_vcpu_t)(void *vcpui, struct vm_snapshot_meta *meta);
186 typedef int	(*vmi_restore_tsc_t)(void *vcpui, uint64_t now);
187 
188 struct vmm_ops {
189 	vmm_init_func_t		modinit;	/* module wide initialization */
190 	vmm_cleanup_func_t	modcleanup;
191 	vmm_resume_func_t	modresume;
192 
193 	vmi_init_func_t		init;		/* vm-specific initialization */
194 	vmi_run_func_t		run;
195 	vmi_cleanup_func_t	cleanup;
196 	vmi_vcpu_init_func_t	vcpu_init;
197 	vmi_vcpu_cleanup_func_t	vcpu_cleanup;
198 	vmi_get_register_t	getreg;
199 	vmi_set_register_t	setreg;
200 	vmi_get_desc_t		getdesc;
201 	vmi_set_desc_t		setdesc;
202 	vmi_get_cap_t		getcap;
203 	vmi_set_cap_t		setcap;
204 	vmi_vmspace_alloc	vmspace_alloc;
205 	vmi_vmspace_free	vmspace_free;
206 	vmi_vlapic_init		vlapic_init;
207 	vmi_vlapic_cleanup	vlapic_cleanup;
208 
209 	/* checkpoint operations */
210 	vmi_snapshot_vcpu_t	vcpu_snapshot;
211 	vmi_restore_tsc_t	restore_tsc;
212 };
213 
214 extern const struct vmm_ops vmm_ops_intel;
215 extern const struct vmm_ops vmm_ops_amd;
216 
217 extern u_int vm_maxcpu;			/* maximum virtual cpus */
218 
219 int vm_create(const char *name, struct vm **retvm);
220 struct vcpu *vm_alloc_vcpu(struct vm *vm, int vcpuid);
221 void vm_disable_vcpu_creation(struct vm *vm);
222 void vm_slock_vcpus(struct vm *vm);
223 void vm_unlock_vcpus(struct vm *vm);
224 void vm_destroy(struct vm *vm);
225 int vm_reinit(struct vm *vm);
226 const char *vm_name(struct vm *vm);
227 uint16_t vm_get_maxcpus(struct vm *vm);
228 void vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores,
229     uint16_t *threads, uint16_t *maxcpus);
230 int vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores,
231     uint16_t threads, uint16_t maxcpus);
232 
233 /*
234  * APIs that modify the guest memory map require all vcpus to be frozen.
235  */
236 void vm_slock_memsegs(struct vm *vm);
237 void vm_xlock_memsegs(struct vm *vm);
238 void vm_unlock_memsegs(struct vm *vm);
239 int vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t off,
240     size_t len, int prot, int flags);
241 int vm_munmap_memseg(struct vm *vm, vm_paddr_t gpa, size_t len);
242 int vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem);
243 void vm_free_memseg(struct vm *vm, int ident);
244 int vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa);
245 int vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len);
246 int vm_assign_pptdev(struct vm *vm, int bus, int slot, int func);
247 int vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func);
248 
249 /*
250  * APIs that inspect the guest memory map require only a *single* vcpu to
251  * be frozen. This acts like a read lock on the guest memory map since any
252  * modification requires *all* vcpus to be frozen.
253  */
254 int vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
255     vm_ooffset_t *segoff, size_t *len, int *prot, int *flags);
256 int vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
257     struct vm_object **objptr);
258 vm_paddr_t vmm_sysmem_maxaddr(struct vm *vm);
259 void *vm_gpa_hold(struct vcpu *vcpu, vm_paddr_t gpa, size_t len,
260     int prot, void **cookie);
261 void *vm_gpa_hold_global(struct vm *vm, vm_paddr_t gpa, size_t len,
262     int prot, void **cookie);
263 void vm_gpa_release(void *cookie);
264 bool vm_mem_allocated(struct vcpu *vcpu, vm_paddr_t gpa);
265 
266 int vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval);
267 int vm_set_register(struct vcpu *vcpu, int reg, uint64_t val);
268 int vm_get_seg_desc(struct vcpu *vcpu, int reg,
269 		    struct seg_desc *ret_desc);
270 int vm_set_seg_desc(struct vcpu *vcpu, int reg,
271 		    struct seg_desc *desc);
272 int vm_run(struct vcpu *vcpu);
273 int vm_suspend(struct vm *vm, enum vm_suspend_how how);
274 int vm_inject_nmi(struct vcpu *vcpu);
275 int vm_nmi_pending(struct vcpu *vcpu);
276 void vm_nmi_clear(struct vcpu *vcpu);
277 int vm_inject_extint(struct vcpu *vcpu);
278 int vm_extint_pending(struct vcpu *vcpu);
279 void vm_extint_clear(struct vcpu *vcpu);
280 int vcpu_vcpuid(struct vcpu *vcpu);
281 struct vm *vcpu_vm(struct vcpu *vcpu);
282 struct vcpu *vm_vcpu(struct vm *vm, int cpu);
283 struct vlapic *vm_lapic(struct vcpu *vcpu);
284 struct vioapic *vm_ioapic(struct vm *vm);
285 struct vhpet *vm_hpet(struct vm *vm);
286 int vm_get_capability(struct vcpu *vcpu, int type, int *val);
287 int vm_set_capability(struct vcpu *vcpu, int type, int val);
288 int vm_get_x2apic_state(struct vcpu *vcpu, enum x2apic_state *state);
289 int vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state);
290 int vm_apicid2vcpuid(struct vm *vm, int apicid);
291 int vm_activate_cpu(struct vcpu *vcpu);
292 int vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu);
293 int vm_resume_cpu(struct vm *vm, struct vcpu *vcpu);
294 int vm_restart_instruction(struct vcpu *vcpu);
295 struct vm_exit *vm_exitinfo(struct vcpu *vcpu);
296 cpuset_t *vm_exitinfo_cpuset(struct vcpu *vcpu);
297 void vm_exit_suspended(struct vcpu *vcpu, uint64_t rip);
298 void vm_exit_debug(struct vcpu *vcpu, uint64_t rip);
299 void vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip);
300 void vm_exit_astpending(struct vcpu *vcpu, uint64_t rip);
301 void vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip);
302 int vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta);
303 int vm_restore_time(struct vm *vm);
304 
305 #ifdef _SYS__CPUSET_H_
306 /*
307  * Rendezvous all vcpus specified in 'dest' and execute 'func(arg)'.
308  * The rendezvous 'func(arg)' is not allowed to do anything that will
309  * cause the thread to be put to sleep.
310  *
311  * The caller cannot hold any locks when initiating the rendezvous.
312  *
313  * The implementation of this API may cause vcpus other than those specified
314  * by 'dest' to be stalled. The caller should not rely on any vcpus making
315  * forward progress when the rendezvous is in progress.
316  */
317 typedef void (*vm_rendezvous_func_t)(struct vcpu *vcpu, void *arg);
318 int vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest,
319     vm_rendezvous_func_t func, void *arg);
320 
321 cpuset_t vm_active_cpus(struct vm *vm);
322 cpuset_t vm_debug_cpus(struct vm *vm);
323 cpuset_t vm_suspended_cpus(struct vm *vm);
324 cpuset_t vm_start_cpus(struct vm *vm, const cpuset_t *tostart);
325 void vm_await_start(struct vm *vm, const cpuset_t *waiting);
326 #endif	/* _SYS__CPUSET_H_ */
327 
328 static __inline int
329 vcpu_rendezvous_pending(struct vcpu *vcpu, struct vm_eventinfo *info)
330 {
331 	/*
332 	 * This check isn't done with atomic operations or under a lock because
333 	 * there's no need to. If the vcpuid bit is set, the vcpu is part of a
334 	 * rendezvous and the bit won't be cleared until the vcpu enters the
335 	 * rendezvous. On rendezvous exit, the cpuset is cleared and the vcpu
336 	 * will see an empty cpuset. So, the races are harmless.
337 	 */
338 	return (CPU_ISSET(vcpu_vcpuid(vcpu), info->rptr));
339 }
340 
341 static __inline int
342 vcpu_suspended(struct vm_eventinfo *info)
343 {
344 
345 	return (*info->sptr);
346 }
347 
348 static __inline int
349 vcpu_reqidle(struct vm_eventinfo *info)
350 {
351 
352 	return (*info->iptr);
353 }
354 
355 int vcpu_debugged(struct vcpu *vcpu);
356 
357 /*
358  * Return true if device indicated by bus/slot/func is supposed to be a
359  * pci passthrough device.
360  *
361  * Return false otherwise.
362  */
363 bool vmm_is_pptdev(int bus, int slot, int func);
364 
365 void *vm_iommu_domain(struct vm *vm);
366 
367 enum vcpu_state {
368 	VCPU_IDLE,
369 	VCPU_FROZEN,
370 	VCPU_RUNNING,
371 	VCPU_SLEEPING,
372 };
373 
374 int vcpu_set_state(struct vcpu *vcpu, enum vcpu_state state, bool from_idle);
375 enum vcpu_state vcpu_get_state(struct vcpu *vcpu, int *hostcpu);
376 
377 static int __inline
378 vcpu_is_running(struct vcpu *vcpu, int *hostcpu)
379 {
380 	return (vcpu_get_state(vcpu, hostcpu) == VCPU_RUNNING);
381 }
382 
383 #ifdef _SYS_PROC_H_
384 static int __inline
385 vcpu_should_yield(struct vcpu *vcpu)
386 {
387 	struct thread *td;
388 
389 	td = curthread;
390 	return (td->td_ast != 0 || td->td_owepreempt != 0);
391 }
392 #endif
393 
394 void *vcpu_stats(struct vcpu *vcpu);
395 void vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr);
396 struct vmspace *vm_get_vmspace(struct vm *vm);
397 struct vatpic *vm_atpic(struct vm *vm);
398 struct vatpit *vm_atpit(struct vm *vm);
399 struct vpmtmr *vm_pmtmr(struct vm *vm);
400 struct vrtc *vm_rtc(struct vm *vm);
401 
402 /*
403  * Inject exception 'vector' into the guest vcpu. This function returns 0 on
404  * success and non-zero on failure.
405  *
406  * Wrapper functions like 'vm_inject_gp()' should be preferred to calling
407  * this function directly because they enforce the trap-like or fault-like
408  * behavior of an exception.
409  *
410  * This function should only be called in the context of the thread that is
411  * executing this vcpu.
412  */
413 int vm_inject_exception(struct vcpu *vcpu, int vector, int err_valid,
414     uint32_t errcode, int restart_instruction);
415 
416 /*
417  * This function is called after a VM-exit that occurred during exception or
418  * interrupt delivery through the IDT. The format of 'intinfo' is described
419  * in Figure 15-1, "EXITINTINFO for All Intercepts", APM, Vol 2.
420  *
421  * If a VM-exit handler completes the event delivery successfully then it
422  * should call vm_exit_intinfo() to extinguish the pending event. For e.g.,
423  * if the task switch emulation is triggered via a task gate then it should
424  * call this function with 'intinfo=0' to indicate that the external event
425  * is not pending anymore.
426  *
427  * Return value is 0 on success and non-zero on failure.
428  */
429 int vm_exit_intinfo(struct vcpu *vcpu, uint64_t intinfo);
430 
431 /*
432  * This function is called before every VM-entry to retrieve a pending
433  * event that should be injected into the guest. This function combines
434  * nested events into a double or triple fault.
435  *
436  * Returns 0 if there are no events that need to be injected into the guest
437  * and non-zero otherwise.
438  */
439 int vm_entry_intinfo(struct vcpu *vcpu, uint64_t *info);
440 
441 int vm_get_intinfo(struct vcpu *vcpu, uint64_t *info1, uint64_t *info2);
442 
443 /*
444  * Function used to keep track of the guest's TSC offset. The
445  * offset is used by the virutalization extensions to provide a consistent
446  * value for the Time Stamp Counter to the guest.
447  */
448 void vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset);
449 
450 enum vm_reg_name vm_segment_name(int seg_encoding);
451 
452 struct vm_copyinfo {
453 	uint64_t	gpa;
454 	size_t		len;
455 	void		*hva;
456 	void		*cookie;
457 };
458 
459 /*
460  * Set up 'copyinfo[]' to copy to/from guest linear address space starting
461  * at 'gla' and 'len' bytes long. The 'prot' should be set to PROT_READ for
462  * a copyin or PROT_WRITE for a copyout.
463  *
464  * retval	is_fault	Interpretation
465  *   0		   0		Success
466  *   0		   1		An exception was injected into the guest
467  * EFAULT	  N/A		Unrecoverable error
468  *
469  * The 'copyinfo[]' can be passed to 'vm_copyin()' or 'vm_copyout()' only if
470  * the return value is 0. The 'copyinfo[]' resources should be freed by calling
471  * 'vm_copy_teardown()' after the copy is done.
472  */
473 int vm_copy_setup(struct vcpu *vcpu, struct vm_guest_paging *paging,
474     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
475     int num_copyinfo, int *is_fault);
476 void vm_copy_teardown(struct vm_copyinfo *copyinfo, int num_copyinfo);
477 void vm_copyin(struct vm_copyinfo *copyinfo, void *kaddr, size_t len);
478 void vm_copyout(const void *kaddr, struct vm_copyinfo *copyinfo, size_t len);
479 
480 int vcpu_trace_exceptions(struct vcpu *vcpu);
481 int vcpu_trap_wbinvd(struct vcpu *vcpu);
482 #endif	/* KERNEL */
483 
484 /*
485  * Identifiers for optional vmm capabilities
486  */
487 enum vm_cap_type {
488 	VM_CAP_HALT_EXIT,
489 	VM_CAP_MTRAP_EXIT,
490 	VM_CAP_PAUSE_EXIT,
491 	VM_CAP_UNRESTRICTED_GUEST,
492 	VM_CAP_ENABLE_INVPCID,
493 	VM_CAP_BPT_EXIT,
494 	VM_CAP_RDPID,
495 	VM_CAP_RDTSCP,
496 	VM_CAP_IPI_EXIT,
497 	VM_CAP_MASK_HWINTR,
498 	VM_CAP_RFLAGS_TF,
499 	VM_CAP_MAX
500 };
501 
502 enum vm_intr_trigger {
503 	EDGE_TRIGGER,
504 	LEVEL_TRIGGER
505 };
506 
507 /*
508  * The 'access' field has the format specified in Table 21-2 of the Intel
509  * Architecture Manual vol 3b.
510  *
511  * XXX The contents of the 'access' field are architecturally defined except
512  * bit 16 - Segment Unusable.
513  */
514 struct seg_desc {
515 	uint64_t	base;
516 	uint32_t	limit;
517 	uint32_t	access;
518 };
519 #define	SEG_DESC_TYPE(access)		((access) & 0x001f)
520 #define	SEG_DESC_DPL(access)		(((access) >> 5) & 0x3)
521 #define	SEG_DESC_PRESENT(access)	(((access) & 0x0080) ? 1 : 0)
522 #define	SEG_DESC_DEF32(access)		(((access) & 0x4000) ? 1 : 0)
523 #define	SEG_DESC_GRANULARITY(access)	(((access) & 0x8000) ? 1 : 0)
524 #define	SEG_DESC_UNUSABLE(access)	(((access) & 0x10000) ? 1 : 0)
525 
526 enum vm_cpu_mode {
527 	CPU_MODE_REAL,
528 	CPU_MODE_PROTECTED,
529 	CPU_MODE_COMPATIBILITY,		/* IA-32E mode (CS.L = 0) */
530 	CPU_MODE_64BIT,			/* IA-32E mode (CS.L = 1) */
531 };
532 
533 enum vm_paging_mode {
534 	PAGING_MODE_FLAT,
535 	PAGING_MODE_32,
536 	PAGING_MODE_PAE,
537 	PAGING_MODE_64,
538 	PAGING_MODE_64_LA57,
539 };
540 
541 struct vm_guest_paging {
542 	uint64_t	cr3;
543 	int		cpl;
544 	enum vm_cpu_mode cpu_mode;
545 	enum vm_paging_mode paging_mode;
546 };
547 
548 /*
549  * The data structures 'vie' and 'vie_op' are meant to be opaque to the
550  * consumers of instruction decoding. The only reason why their contents
551  * need to be exposed is because they are part of the 'vm_exit' structure.
552  */
553 struct vie_op {
554 	uint8_t		op_byte;	/* actual opcode byte */
555 	uint8_t		op_type;	/* type of operation (e.g. MOV) */
556 	uint16_t	op_flags;
557 };
558 _Static_assert(sizeof(struct vie_op) == 4, "ABI");
559 _Static_assert(_Alignof(struct vie_op) == 2, "ABI");
560 
561 #define	VIE_INST_SIZE	15
562 struct vie {
563 	uint8_t		inst[VIE_INST_SIZE];	/* instruction bytes */
564 	uint8_t		num_valid;		/* size of the instruction */
565 
566 /* The following fields are all zeroed upon restart. */
567 #define	vie_startzero	num_processed
568 	uint8_t		num_processed;
569 
570 	uint8_t		addrsize:4, opsize:4;	/* address and operand sizes */
571 	uint8_t		rex_w:1,		/* REX prefix */
572 			rex_r:1,
573 			rex_x:1,
574 			rex_b:1,
575 			rex_present:1,
576 			repz_present:1,		/* REP/REPE/REPZ prefix */
577 			repnz_present:1,	/* REPNE/REPNZ prefix */
578 			opsize_override:1,	/* Operand size override */
579 			addrsize_override:1,	/* Address size override */
580 			segment_override:1;	/* Segment override */
581 
582 	uint8_t		mod:2,			/* ModRM byte */
583 			reg:4,
584 			rm:4;
585 
586 	uint8_t		ss:2,			/* SIB byte */
587 			vex_present:1,		/* VEX prefixed */
588 			vex_l:1,		/* L bit */
589 			index:4,		/* SIB byte */
590 			base:4;			/* SIB byte */
591 
592 	uint8_t		disp_bytes;
593 	uint8_t		imm_bytes;
594 
595 	uint8_t		scale;
596 
597 	uint8_t		vex_reg:4,		/* vvvv: first source register specifier */
598 			vex_pp:2,		/* pp */
599 			_sparebits:2;
600 
601 	uint8_t		_sparebytes[2];
602 
603 	int		base_register;		/* VM_REG_GUEST_xyz */
604 	int		index_register;		/* VM_REG_GUEST_xyz */
605 	int		segment_register;	/* VM_REG_GUEST_xyz */
606 
607 	int64_t		displacement;		/* optional addr displacement */
608 	int64_t		immediate;		/* optional immediate operand */
609 
610 	uint8_t		decoded;	/* set to 1 if successfully decoded */
611 
612 	uint8_t		_sparebyte;
613 
614 	struct vie_op	op;			/* opcode description */
615 };
616 _Static_assert(sizeof(struct vie) == 64, "ABI");
617 _Static_assert(__offsetof(struct vie, disp_bytes) == 22, "ABI");
618 _Static_assert(__offsetof(struct vie, scale) == 24, "ABI");
619 _Static_assert(__offsetof(struct vie, base_register) == 28, "ABI");
620 
621 enum vm_exitcode {
622 	VM_EXITCODE_INOUT,
623 	VM_EXITCODE_VMX,
624 	VM_EXITCODE_BOGUS,
625 	VM_EXITCODE_RDMSR,
626 	VM_EXITCODE_WRMSR,
627 	VM_EXITCODE_HLT,
628 	VM_EXITCODE_MTRAP,
629 	VM_EXITCODE_PAUSE,
630 	VM_EXITCODE_PAGING,
631 	VM_EXITCODE_INST_EMUL,
632 	VM_EXITCODE_SPINUP_AP,
633 	VM_EXITCODE_DEPRECATED1,	/* used to be SPINDOWN_CPU */
634 	VM_EXITCODE_RENDEZVOUS,
635 	VM_EXITCODE_IOAPIC_EOI,
636 	VM_EXITCODE_SUSPENDED,
637 	VM_EXITCODE_INOUT_STR,
638 	VM_EXITCODE_TASK_SWITCH,
639 	VM_EXITCODE_MONITOR,
640 	VM_EXITCODE_MWAIT,
641 	VM_EXITCODE_SVM,
642 	VM_EXITCODE_REQIDLE,
643 	VM_EXITCODE_DEBUG,
644 	VM_EXITCODE_VMINSN,
645 	VM_EXITCODE_BPT,
646 	VM_EXITCODE_IPI,
647 	VM_EXITCODE_DB,
648 	VM_EXITCODE_MAX
649 };
650 
651 struct vm_inout {
652 	uint16_t	bytes:3;	/* 1 or 2 or 4 */
653 	uint16_t	in:1;
654 	uint16_t	string:1;
655 	uint16_t	rep:1;
656 	uint16_t	port;
657 	uint32_t	eax;		/* valid for out */
658 };
659 
660 struct vm_inout_str {
661 	struct vm_inout	inout;		/* must be the first element */
662 	struct vm_guest_paging paging;
663 	uint64_t	rflags;
664 	uint64_t	cr0;
665 	uint64_t	index;
666 	uint64_t	count;		/* rep=1 (%rcx), rep=0 (1) */
667 	int		addrsize;
668 	enum vm_reg_name seg_name;
669 	struct seg_desc seg_desc;
670 };
671 
672 enum task_switch_reason {
673 	TSR_CALL,
674 	TSR_IRET,
675 	TSR_JMP,
676 	TSR_IDT_GATE,	/* task gate in IDT */
677 };
678 
679 struct vm_task_switch {
680 	uint16_t	tsssel;		/* new TSS selector */
681 	int		ext;		/* task switch due to external event */
682 	uint32_t	errcode;
683 	int		errcode_valid;	/* push 'errcode' on the new stack */
684 	enum task_switch_reason reason;
685 	struct vm_guest_paging paging;
686 };
687 
688 struct vm_exit {
689 	enum vm_exitcode	exitcode;
690 	int			inst_length;	/* 0 means unknown */
691 	uint64_t		rip;
692 	union {
693 		struct vm_inout	inout;
694 		struct vm_inout_str inout_str;
695 		struct {
696 			uint64_t	gpa;
697 			int		fault_type;
698 		} paging;
699 		struct {
700 			uint64_t	gpa;
701 			uint64_t	gla;
702 			uint64_t	cs_base;
703 			int		cs_d;		/* CS.D */
704 			struct vm_guest_paging paging;
705 			struct vie	vie;
706 		} inst_emul;
707 		/*
708 		 * VMX specific payload. Used when there is no "better"
709 		 * exitcode to represent the VM-exit.
710 		 */
711 		struct {
712 			int		status;		/* vmx inst status */
713 			/*
714 			 * 'exit_reason' and 'exit_qualification' are valid
715 			 * only if 'status' is zero.
716 			 */
717 			uint32_t	exit_reason;
718 			uint64_t	exit_qualification;
719 			/*
720 			 * 'inst_error' and 'inst_type' are valid
721 			 * only if 'status' is non-zero.
722 			 */
723 			int		inst_type;
724 			int		inst_error;
725 		} vmx;
726 		/*
727 		 * SVM specific payload.
728 		 */
729 		struct {
730 			uint64_t	exitcode;
731 			uint64_t	exitinfo1;
732 			uint64_t	exitinfo2;
733 		} svm;
734 		struct {
735 			int		inst_length;
736 		} bpt;
737 		struct {
738 			int		trace_trap;
739 			int		pushf_intercept;
740 			int		tf_shadow_val;
741 			struct		vm_guest_paging paging;
742 		} dbg;
743 		struct {
744 			uint32_t	code;		/* ecx value */
745 			uint64_t	wval;
746 		} msr;
747 		struct {
748 			int		vcpu;
749 			uint64_t	rip;
750 		} spinup_ap;
751 		struct {
752 			uint64_t	rflags;
753 			uint64_t	intr_status;
754 		} hlt;
755 		struct {
756 			int		vector;
757 		} ioapic_eoi;
758 		struct {
759 			enum vm_suspend_how how;
760 		} suspended;
761 		struct {
762 			/*
763 			 * The destination vCPU mask is saved in vcpu->cpuset
764 			 * and is copied out to userspace separately to avoid
765 			 * ABI concerns.
766 			 */
767 			uint32_t mode;
768 			uint8_t vector;
769 		} ipi;
770 		struct vm_task_switch task_switch;
771 	} u;
772 };
773 
774 /* APIs to inject faults into the guest */
775 void vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid,
776     int errcode);
777 
778 static __inline void
779 vm_inject_ud(struct vcpu *vcpu)
780 {
781 	vm_inject_fault(vcpu, IDT_UD, 0, 0);
782 }
783 
784 static __inline void
785 vm_inject_gp(struct vcpu *vcpu)
786 {
787 	vm_inject_fault(vcpu, IDT_GP, 1, 0);
788 }
789 
790 static __inline void
791 vm_inject_ac(struct vcpu *vcpu, int errcode)
792 {
793 	vm_inject_fault(vcpu, IDT_AC, 1, errcode);
794 }
795 
796 static __inline void
797 vm_inject_ss(struct vcpu *vcpu, int errcode)
798 {
799 	vm_inject_fault(vcpu, IDT_SS, 1, errcode);
800 }
801 
802 void vm_inject_pf(struct vcpu *vcpu, int error_code, uint64_t cr2);
803 
804 #endif	/* _VMM_H_ */
805