xref: /freebsd/sys/amd64/include/vmm.h (revision 7cc42f6d25ef2e19059d088fa7d4853fe9afefb5)
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 #ifndef _VMM_H_
32 #define	_VMM_H_
33 
34 #include <sys/sdt.h>
35 #include <x86/segments.h>
36 
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 	void	*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 *vmi, int vcpu, register_t rip,
170 		    struct pmap *pmap, struct vm_eventinfo *info);
171 typedef void	(*vmi_cleanup_func_t)(void *vmi);
172 typedef int	(*vmi_get_register_t)(void *vmi, int vcpu, int num,
173 				      uint64_t *retval);
174 typedef int	(*vmi_set_register_t)(void *vmi, int vcpu, int num,
175 				      uint64_t val);
176 typedef int	(*vmi_get_desc_t)(void *vmi, int vcpu, int num,
177 				  struct seg_desc *desc);
178 typedef int	(*vmi_set_desc_t)(void *vmi, int vcpu, int num,
179 				  struct seg_desc *desc);
180 typedef int	(*vmi_get_cap_t)(void *vmi, int vcpu, int num, int *retval);
181 typedef int	(*vmi_set_cap_t)(void *vmi, int vcpu, int num, int val);
182 typedef struct vmspace * (*vmi_vmspace_alloc)(vm_offset_t min, vm_offset_t max);
183 typedef void	(*vmi_vmspace_free)(struct vmspace *vmspace);
184 typedef struct vlapic * (*vmi_vlapic_init)(void *vmi, int vcpu);
185 typedef void	(*vmi_vlapic_cleanup)(void *vmi, struct vlapic *vlapic);
186 typedef int	(*vmi_snapshot_t)(void *vmi, struct vm_snapshot_meta *meta);
187 typedef int	(*vmi_snapshot_vmcx_t)(void *vmi, struct vm_snapshot_meta *meta,
188 				       int vcpu);
189 typedef int	(*vmi_restore_tsc_t)(void *vmi, int vcpuid, uint64_t now);
190 
191 struct vmm_ops {
192 	vmm_init_func_t		init;		/* module wide initialization */
193 	vmm_cleanup_func_t	cleanup;
194 	vmm_resume_func_t	resume;
195 
196 	vmi_init_func_t		vminit;		/* vm-specific initialization */
197 	vmi_run_func_t		vmrun;
198 	vmi_cleanup_func_t	vmcleanup;
199 	vmi_get_register_t	vmgetreg;
200 	vmi_set_register_t	vmsetreg;
201 	vmi_get_desc_t		vmgetdesc;
202 	vmi_set_desc_t		vmsetdesc;
203 	vmi_get_cap_t		vmgetcap;
204 	vmi_set_cap_t		vmsetcap;
205 	vmi_vmspace_alloc	vmspace_alloc;
206 	vmi_vmspace_free	vmspace_free;
207 	vmi_vlapic_init		vlapic_init;
208 	vmi_vlapic_cleanup	vlapic_cleanup;
209 
210 	/* checkpoint operations */
211 	vmi_snapshot_t		vmsnapshot;
212 	vmi_snapshot_vmcx_t	vmcx_snapshot;
213 	vmi_restore_tsc_t	vm_restore_tsc;
214 };
215 
216 extern struct vmm_ops vmm_ops_intel;
217 extern struct vmm_ops vmm_ops_amd;
218 
219 int vm_create(const char *name, struct vm **retvm);
220 void vm_destroy(struct vm *vm);
221 int vm_reinit(struct vm *vm);
222 const char *vm_name(struct vm *vm);
223 uint16_t vm_get_maxcpus(struct vm *vm);
224 void vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores,
225     uint16_t *threads, uint16_t *maxcpus);
226 int vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores,
227     uint16_t threads, uint16_t maxcpus);
228 
229 /*
230  * APIs that modify the guest memory map require all vcpus to be frozen.
231  */
232 int vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t off,
233     size_t len, int prot, int flags);
234 int vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem);
235 void vm_free_memseg(struct vm *vm, int ident);
236 int vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa);
237 int vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len);
238 int vm_assign_pptdev(struct vm *vm, int bus, int slot, int func);
239 int vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func);
240 
241 /*
242  * APIs that inspect the guest memory map require only a *single* vcpu to
243  * be frozen. This acts like a read lock on the guest memory map since any
244  * modification requires *all* vcpus to be frozen.
245  */
246 int vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
247     vm_ooffset_t *segoff, size_t *len, int *prot, int *flags);
248 int vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
249     struct vm_object **objptr);
250 vm_paddr_t vmm_sysmem_maxaddr(struct vm *vm);
251 void *vm_gpa_hold(struct vm *, int vcpuid, vm_paddr_t gpa, size_t len,
252     int prot, void **cookie);
253 void vm_gpa_release(void *cookie);
254 bool vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa);
255 
256 int vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval);
257 int vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val);
258 int vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
259 		    struct seg_desc *ret_desc);
260 int vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
261 		    struct seg_desc *desc);
262 int vm_run(struct vm *vm, struct vm_run *vmrun);
263 int vm_suspend(struct vm *vm, enum vm_suspend_how how);
264 int vm_inject_nmi(struct vm *vm, int vcpu);
265 int vm_nmi_pending(struct vm *vm, int vcpuid);
266 void vm_nmi_clear(struct vm *vm, int vcpuid);
267 int vm_inject_extint(struct vm *vm, int vcpu);
268 int vm_extint_pending(struct vm *vm, int vcpuid);
269 void vm_extint_clear(struct vm *vm, int vcpuid);
270 struct vlapic *vm_lapic(struct vm *vm, int cpu);
271 struct vioapic *vm_ioapic(struct vm *vm);
272 struct vhpet *vm_hpet(struct vm *vm);
273 int vm_get_capability(struct vm *vm, int vcpu, int type, int *val);
274 int vm_set_capability(struct vm *vm, int vcpu, int type, int val);
275 int vm_get_x2apic_state(struct vm *vm, int vcpu, enum x2apic_state *state);
276 int vm_set_x2apic_state(struct vm *vm, int vcpu, enum x2apic_state state);
277 int vm_apicid2vcpuid(struct vm *vm, int apicid);
278 int vm_activate_cpu(struct vm *vm, int vcpu);
279 int vm_suspend_cpu(struct vm *vm, int vcpu);
280 int vm_resume_cpu(struct vm *vm, int vcpu);
281 struct vm_exit *vm_exitinfo(struct vm *vm, int vcpuid);
282 void vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip);
283 void vm_exit_debug(struct vm *vm, int vcpuid, uint64_t rip);
284 void vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip);
285 void vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip);
286 void vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip);
287 int vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta);
288 int vm_restore_time(struct vm *vm);
289 
290 #ifdef _SYS__CPUSET_H_
291 /*
292  * Rendezvous all vcpus specified in 'dest' and execute 'func(arg)'.
293  * The rendezvous 'func(arg)' is not allowed to do anything that will
294  * cause the thread to be put to sleep.
295  *
296  * If the rendezvous is being initiated from a vcpu context then the
297  * 'vcpuid' must refer to that vcpu, otherwise it should be set to -1.
298  *
299  * The caller cannot hold any locks when initiating the rendezvous.
300  *
301  * The implementation of this API may cause vcpus other than those specified
302  * by 'dest' to be stalled. The caller should not rely on any vcpus making
303  * forward progress when the rendezvous is in progress.
304  */
305 typedef void (*vm_rendezvous_func_t)(struct vm *vm, int vcpuid, void *arg);
306 int vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
307     vm_rendezvous_func_t func, void *arg);
308 cpuset_t vm_active_cpus(struct vm *vm);
309 cpuset_t vm_debug_cpus(struct vm *vm);
310 cpuset_t vm_suspended_cpus(struct vm *vm);
311 #endif	/* _SYS__CPUSET_H_ */
312 
313 static __inline int
314 vcpu_rendezvous_pending(struct vm_eventinfo *info)
315 {
316 
317 	return (*((uintptr_t *)(info->rptr)) != 0);
318 }
319 
320 static __inline int
321 vcpu_suspended(struct vm_eventinfo *info)
322 {
323 
324 	return (*info->sptr);
325 }
326 
327 static __inline int
328 vcpu_reqidle(struct vm_eventinfo *info)
329 {
330 
331 	return (*info->iptr);
332 }
333 
334 int vcpu_debugged(struct vm *vm, int vcpuid);
335 
336 /*
337  * Return true if device indicated by bus/slot/func is supposed to be a
338  * pci passthrough device.
339  *
340  * Return false otherwise.
341  */
342 bool vmm_is_pptdev(int bus, int slot, int func);
343 
344 void *vm_iommu_domain(struct vm *vm);
345 
346 enum vcpu_state {
347 	VCPU_IDLE,
348 	VCPU_FROZEN,
349 	VCPU_RUNNING,
350 	VCPU_SLEEPING,
351 };
352 
353 int vcpu_set_state(struct vm *vm, int vcpu, enum vcpu_state state,
354     bool from_idle);
355 enum vcpu_state vcpu_get_state(struct vm *vm, int vcpu, int *hostcpu);
356 
357 static int __inline
358 vcpu_is_running(struct vm *vm, int vcpu, int *hostcpu)
359 {
360 	return (vcpu_get_state(vm, vcpu, hostcpu) == VCPU_RUNNING);
361 }
362 
363 #ifdef _SYS_PROC_H_
364 static int __inline
365 vcpu_should_yield(struct vm *vm, int vcpu)
366 {
367 
368 	if (curthread->td_flags & (TDF_ASTPENDING | TDF_NEEDRESCHED))
369 		return (1);
370 	else if (curthread->td_owepreempt)
371 		return (1);
372 	else
373 		return (0);
374 }
375 #endif
376 
377 void *vcpu_stats(struct vm *vm, int vcpu);
378 void vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr);
379 struct vmspace *vm_get_vmspace(struct vm *vm);
380 struct vatpic *vm_atpic(struct vm *vm);
381 struct vatpit *vm_atpit(struct vm *vm);
382 struct vpmtmr *vm_pmtmr(struct vm *vm);
383 struct vrtc *vm_rtc(struct vm *vm);
384 
385 /*
386  * Inject exception 'vector' into the guest vcpu. This function returns 0 on
387  * success and non-zero on failure.
388  *
389  * Wrapper functions like 'vm_inject_gp()' should be preferred to calling
390  * this function directly because they enforce the trap-like or fault-like
391  * behavior of an exception.
392  *
393  * This function should only be called in the context of the thread that is
394  * executing this vcpu.
395  */
396 int vm_inject_exception(struct vm *vm, int vcpuid, int vector, int err_valid,
397     uint32_t errcode, int restart_instruction);
398 
399 /*
400  * This function is called after a VM-exit that occurred during exception or
401  * interrupt delivery through the IDT. The format of 'intinfo' is described
402  * in Figure 15-1, "EXITINTINFO for All Intercepts", APM, Vol 2.
403  *
404  * If a VM-exit handler completes the event delivery successfully then it
405  * should call vm_exit_intinfo() to extinguish the pending event. For e.g.,
406  * if the task switch emulation is triggered via a task gate then it should
407  * call this function with 'intinfo=0' to indicate that the external event
408  * is not pending anymore.
409  *
410  * Return value is 0 on success and non-zero on failure.
411  */
412 int vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t intinfo);
413 
414 /*
415  * This function is called before every VM-entry to retrieve a pending
416  * event that should be injected into the guest. This function combines
417  * nested events into a double or triple fault.
418  *
419  * Returns 0 if there are no events that need to be injected into the guest
420  * and non-zero otherwise.
421  */
422 int vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *info);
423 
424 int vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2);
425 
426 /*
427  * Function used to keep track of the guest's TSC offset. The
428  * offset is used by the virutalization extensions to provide a consistent
429  * value for the Time Stamp Counter to the guest.
430  *
431  * Return value is 0 on success and non-zero on failure.
432  */
433 int vm_set_tsc_offset(struct vm *vm, int vcpu_id, uint64_t offset);
434 
435 enum vm_reg_name vm_segment_name(int seg_encoding);
436 
437 struct vm_copyinfo {
438 	uint64_t	gpa;
439 	size_t		len;
440 	void		*hva;
441 	void		*cookie;
442 };
443 
444 /*
445  * Set up 'copyinfo[]' to copy to/from guest linear address space starting
446  * at 'gla' and 'len' bytes long. The 'prot' should be set to PROT_READ for
447  * a copyin or PROT_WRITE for a copyout.
448  *
449  * retval	is_fault	Interpretation
450  *   0		   0		Success
451  *   0		   1		An exception was injected into the guest
452  * EFAULT	  N/A		Unrecoverable error
453  *
454  * The 'copyinfo[]' can be passed to 'vm_copyin()' or 'vm_copyout()' only if
455  * the return value is 0. The 'copyinfo[]' resources should be freed by calling
456  * 'vm_copy_teardown()' after the copy is done.
457  */
458 int vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
459     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
460     int num_copyinfo, int *is_fault);
461 void vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
462     int num_copyinfo);
463 void vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
464     void *kaddr, size_t len);
465 void vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
466     struct vm_copyinfo *copyinfo, size_t len);
467 
468 int vcpu_trace_exceptions(struct vm *vm, int vcpuid);
469 #endif	/* KERNEL */
470 
471 #define	VM_MAXCPU	16			/* maximum virtual cpus */
472 
473 /*
474  * Identifiers for optional vmm capabilities
475  */
476 enum vm_cap_type {
477 	VM_CAP_HALT_EXIT,
478 	VM_CAP_MTRAP_EXIT,
479 	VM_CAP_PAUSE_EXIT,
480 	VM_CAP_UNRESTRICTED_GUEST,
481 	VM_CAP_ENABLE_INVPCID,
482 	VM_CAP_BPT_EXIT,
483 	VM_CAP_RDPID,
484 	VM_CAP_RDTSCP,
485 	VM_CAP_MAX
486 };
487 
488 enum vm_intr_trigger {
489 	EDGE_TRIGGER,
490 	LEVEL_TRIGGER
491 };
492 
493 /*
494  * The 'access' field has the format specified in Table 21-2 of the Intel
495  * Architecture Manual vol 3b.
496  *
497  * XXX The contents of the 'access' field are architecturally defined except
498  * bit 16 - Segment Unusable.
499  */
500 struct seg_desc {
501 	uint64_t	base;
502 	uint32_t	limit;
503 	uint32_t	access;
504 };
505 #define	SEG_DESC_TYPE(access)		((access) & 0x001f)
506 #define	SEG_DESC_DPL(access)		(((access) >> 5) & 0x3)
507 #define	SEG_DESC_PRESENT(access)	(((access) & 0x0080) ? 1 : 0)
508 #define	SEG_DESC_DEF32(access)		(((access) & 0x4000) ? 1 : 0)
509 #define	SEG_DESC_GRANULARITY(access)	(((access) & 0x8000) ? 1 : 0)
510 #define	SEG_DESC_UNUSABLE(access)	(((access) & 0x10000) ? 1 : 0)
511 
512 enum vm_cpu_mode {
513 	CPU_MODE_REAL,
514 	CPU_MODE_PROTECTED,
515 	CPU_MODE_COMPATIBILITY,		/* IA-32E mode (CS.L = 0) */
516 	CPU_MODE_64BIT,			/* IA-32E mode (CS.L = 1) */
517 };
518 
519 enum vm_paging_mode {
520 	PAGING_MODE_FLAT,
521 	PAGING_MODE_32,
522 	PAGING_MODE_PAE,
523 	PAGING_MODE_64,
524 	PAGING_MODE_64_LA57,
525 };
526 
527 struct vm_guest_paging {
528 	uint64_t	cr3;
529 	int		cpl;
530 	enum vm_cpu_mode cpu_mode;
531 	enum vm_paging_mode paging_mode;
532 };
533 
534 /*
535  * The data structures 'vie' and 'vie_op' are meant to be opaque to the
536  * consumers of instruction decoding. The only reason why their contents
537  * need to be exposed is because they are part of the 'vm_exit' structure.
538  */
539 struct vie_op {
540 	uint8_t		op_byte;	/* actual opcode byte */
541 	uint8_t		op_type;	/* type of operation (e.g. MOV) */
542 	uint16_t	op_flags;
543 };
544 _Static_assert(sizeof(struct vie_op) == 4, "ABI");
545 _Static_assert(_Alignof(struct vie_op) == 2, "ABI");
546 
547 #define	VIE_INST_SIZE	15
548 struct vie {
549 	uint8_t		inst[VIE_INST_SIZE];	/* instruction bytes */
550 	uint8_t		num_valid;		/* size of the instruction */
551 
552 /* The following fields are all zeroed upon restart. */
553 #define	vie_startzero	num_processed
554 	uint8_t		num_processed;
555 
556 	uint8_t		addrsize:4, opsize:4;	/* address and operand sizes */
557 	uint8_t		rex_w:1,		/* REX prefix */
558 			rex_r:1,
559 			rex_x:1,
560 			rex_b:1,
561 			rex_present:1,
562 			repz_present:1,		/* REP/REPE/REPZ prefix */
563 			repnz_present:1,	/* REPNE/REPNZ prefix */
564 			opsize_override:1,	/* Operand size override */
565 			addrsize_override:1,	/* Address size override */
566 			segment_override:1;	/* Segment override */
567 
568 	uint8_t		mod:2,			/* ModRM byte */
569 			reg:4,
570 			rm:4;
571 
572 	uint8_t		ss:2,			/* SIB byte */
573 			vex_present:1,		/* VEX prefixed */
574 			vex_l:1,		/* L bit */
575 			index:4,		/* SIB byte */
576 			base:4;			/* SIB byte */
577 
578 	uint8_t		disp_bytes;
579 	uint8_t		imm_bytes;
580 
581 	uint8_t		scale;
582 
583 	uint8_t		vex_reg:4,		/* vvvv: first source register specifier */
584 			vex_pp:2,		/* pp */
585 			_sparebits:2;
586 
587 	uint8_t		_sparebytes[2];
588 
589 	int		base_register;		/* VM_REG_GUEST_xyz */
590 	int		index_register;		/* VM_REG_GUEST_xyz */
591 	int		segment_register;	/* VM_REG_GUEST_xyz */
592 
593 	int64_t		displacement;		/* optional addr displacement */
594 	int64_t		immediate;		/* optional immediate operand */
595 
596 	uint8_t		decoded;	/* set to 1 if successfully decoded */
597 
598 	uint8_t		_sparebyte;
599 
600 	struct vie_op	op;			/* opcode description */
601 };
602 _Static_assert(sizeof(struct vie) == 64, "ABI");
603 _Static_assert(__offsetof(struct vie, disp_bytes) == 22, "ABI");
604 _Static_assert(__offsetof(struct vie, scale) == 24, "ABI");
605 _Static_assert(__offsetof(struct vie, base_register) == 28, "ABI");
606 
607 enum vm_exitcode {
608 	VM_EXITCODE_INOUT,
609 	VM_EXITCODE_VMX,
610 	VM_EXITCODE_BOGUS,
611 	VM_EXITCODE_RDMSR,
612 	VM_EXITCODE_WRMSR,
613 	VM_EXITCODE_HLT,
614 	VM_EXITCODE_MTRAP,
615 	VM_EXITCODE_PAUSE,
616 	VM_EXITCODE_PAGING,
617 	VM_EXITCODE_INST_EMUL,
618 	VM_EXITCODE_SPINUP_AP,
619 	VM_EXITCODE_DEPRECATED1,	/* used to be SPINDOWN_CPU */
620 	VM_EXITCODE_RENDEZVOUS,
621 	VM_EXITCODE_IOAPIC_EOI,
622 	VM_EXITCODE_SUSPENDED,
623 	VM_EXITCODE_INOUT_STR,
624 	VM_EXITCODE_TASK_SWITCH,
625 	VM_EXITCODE_MONITOR,
626 	VM_EXITCODE_MWAIT,
627 	VM_EXITCODE_SVM,
628 	VM_EXITCODE_REQIDLE,
629 	VM_EXITCODE_DEBUG,
630 	VM_EXITCODE_VMINSN,
631 	VM_EXITCODE_BPT,
632 	VM_EXITCODE_MAX
633 };
634 
635 struct vm_inout {
636 	uint16_t	bytes:3;	/* 1 or 2 or 4 */
637 	uint16_t	in:1;
638 	uint16_t	string:1;
639 	uint16_t	rep:1;
640 	uint16_t	port;
641 	uint32_t	eax;		/* valid for out */
642 };
643 
644 struct vm_inout_str {
645 	struct vm_inout	inout;		/* must be the first element */
646 	struct vm_guest_paging paging;
647 	uint64_t	rflags;
648 	uint64_t	cr0;
649 	uint64_t	index;
650 	uint64_t	count;		/* rep=1 (%rcx), rep=0 (1) */
651 	int		addrsize;
652 	enum vm_reg_name seg_name;
653 	struct seg_desc seg_desc;
654 };
655 
656 enum task_switch_reason {
657 	TSR_CALL,
658 	TSR_IRET,
659 	TSR_JMP,
660 	TSR_IDT_GATE,	/* task gate in IDT */
661 };
662 
663 struct vm_task_switch {
664 	uint16_t	tsssel;		/* new TSS selector */
665 	int		ext;		/* task switch due to external event */
666 	uint32_t	errcode;
667 	int		errcode_valid;	/* push 'errcode' on the new stack */
668 	enum task_switch_reason reason;
669 	struct vm_guest_paging paging;
670 };
671 
672 struct vm_exit {
673 	enum vm_exitcode	exitcode;
674 	int			inst_length;	/* 0 means unknown */
675 	uint64_t		rip;
676 	union {
677 		struct vm_inout	inout;
678 		struct vm_inout_str inout_str;
679 		struct {
680 			uint64_t	gpa;
681 			int		fault_type;
682 		} paging;
683 		struct {
684 			uint64_t	gpa;
685 			uint64_t	gla;
686 			uint64_t	cs_base;
687 			int		cs_d;		/* CS.D */
688 			struct vm_guest_paging paging;
689 			struct vie	vie;
690 		} inst_emul;
691 		/*
692 		 * VMX specific payload. Used when there is no "better"
693 		 * exitcode to represent the VM-exit.
694 		 */
695 		struct {
696 			int		status;		/* vmx inst status */
697 			/*
698 			 * 'exit_reason' and 'exit_qualification' are valid
699 			 * only if 'status' is zero.
700 			 */
701 			uint32_t	exit_reason;
702 			uint64_t	exit_qualification;
703 			/*
704 			 * 'inst_error' and 'inst_type' are valid
705 			 * only if 'status' is non-zero.
706 			 */
707 			int		inst_type;
708 			int		inst_error;
709 		} vmx;
710 		/*
711 		 * SVM specific payload.
712 		 */
713 		struct {
714 			uint64_t	exitcode;
715 			uint64_t	exitinfo1;
716 			uint64_t	exitinfo2;
717 		} svm;
718 		struct {
719 			int		inst_length;
720 		} bpt;
721 		struct {
722 			uint32_t	code;		/* ecx value */
723 			uint64_t	wval;
724 		} msr;
725 		struct {
726 			int		vcpu;
727 			uint64_t	rip;
728 		} spinup_ap;
729 		struct {
730 			uint64_t	rflags;
731 			uint64_t	intr_status;
732 		} hlt;
733 		struct {
734 			int		vector;
735 		} ioapic_eoi;
736 		struct {
737 			enum vm_suspend_how how;
738 		} suspended;
739 		struct vm_task_switch task_switch;
740 	} u;
741 };
742 
743 /* APIs to inject faults into the guest */
744 void vm_inject_fault(void *vm, int vcpuid, int vector, int errcode_valid,
745     int errcode);
746 
747 static __inline void
748 vm_inject_ud(void *vm, int vcpuid)
749 {
750 	vm_inject_fault(vm, vcpuid, IDT_UD, 0, 0);
751 }
752 
753 static __inline void
754 vm_inject_gp(void *vm, int vcpuid)
755 {
756 	vm_inject_fault(vm, vcpuid, IDT_GP, 1, 0);
757 }
758 
759 static __inline void
760 vm_inject_ac(void *vm, int vcpuid, int errcode)
761 {
762 	vm_inject_fault(vm, vcpuid, IDT_AC, 1, errcode);
763 }
764 
765 static __inline void
766 vm_inject_ss(void *vm, int vcpuid, int errcode)
767 {
768 	vm_inject_fault(vm, vcpuid, IDT_SS, 1, errcode);
769 }
770 
771 void vm_inject_pf(void *vm, int vcpuid, int error_code, uint64_t cr2);
772 
773 int vm_restart_instruction(void *vm, int vcpuid);
774 
775 #endif	/* _VMM_H_ */
776