xref: /freebsd/sys/amd64/include/vmm.h (revision 79959b84754f856487ecaba16916a8fb62d743cb)
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 #ifdef _KERNEL
38 SDT_PROVIDER_DECLARE(vmm);
39 #endif
40 
41 enum vm_suspend_how {
42 	VM_SUSPEND_NONE,
43 	VM_SUSPEND_RESET,
44 	VM_SUSPEND_POWEROFF,
45 	VM_SUSPEND_HALT,
46 	VM_SUSPEND_TRIPLEFAULT,
47 	VM_SUSPEND_LAST
48 };
49 
50 /*
51  * Identifiers for architecturally defined registers.
52  */
53 enum vm_reg_name {
54 	VM_REG_GUEST_RAX,
55 	VM_REG_GUEST_RBX,
56 	VM_REG_GUEST_RCX,
57 	VM_REG_GUEST_RDX,
58 	VM_REG_GUEST_RSI,
59 	VM_REG_GUEST_RDI,
60 	VM_REG_GUEST_RBP,
61 	VM_REG_GUEST_R8,
62 	VM_REG_GUEST_R9,
63 	VM_REG_GUEST_R10,
64 	VM_REG_GUEST_R11,
65 	VM_REG_GUEST_R12,
66 	VM_REG_GUEST_R13,
67 	VM_REG_GUEST_R14,
68 	VM_REG_GUEST_R15,
69 	VM_REG_GUEST_CR0,
70 	VM_REG_GUEST_CR3,
71 	VM_REG_GUEST_CR4,
72 	VM_REG_GUEST_DR7,
73 	VM_REG_GUEST_RSP,
74 	VM_REG_GUEST_RIP,
75 	VM_REG_GUEST_RFLAGS,
76 	VM_REG_GUEST_ES,
77 	VM_REG_GUEST_CS,
78 	VM_REG_GUEST_SS,
79 	VM_REG_GUEST_DS,
80 	VM_REG_GUEST_FS,
81 	VM_REG_GUEST_GS,
82 	VM_REG_GUEST_LDTR,
83 	VM_REG_GUEST_TR,
84 	VM_REG_GUEST_IDTR,
85 	VM_REG_GUEST_GDTR,
86 	VM_REG_GUEST_EFER,
87 	VM_REG_GUEST_CR2,
88 	VM_REG_GUEST_PDPTE0,
89 	VM_REG_GUEST_PDPTE1,
90 	VM_REG_GUEST_PDPTE2,
91 	VM_REG_GUEST_PDPTE3,
92 	VM_REG_GUEST_INTR_SHADOW,
93 	VM_REG_GUEST_DR0,
94 	VM_REG_GUEST_DR1,
95 	VM_REG_GUEST_DR2,
96 	VM_REG_GUEST_DR3,
97 	VM_REG_GUEST_DR6,
98 	VM_REG_GUEST_ENTRY_INST_LENGTH,
99 	VM_REG_LAST
100 };
101 
102 enum x2apic_state {
103 	X2APIC_DISABLED,
104 	X2APIC_ENABLED,
105 	X2APIC_STATE_LAST
106 };
107 
108 #define	VM_INTINFO_VECTOR(info)	((info) & 0xff)
109 #define	VM_INTINFO_DEL_ERRCODE	0x800
110 #define	VM_INTINFO_RSVD		0x7ffff000
111 #define	VM_INTINFO_VALID	0x80000000
112 #define	VM_INTINFO_TYPE		0x700
113 #define	VM_INTINFO_HWINTR	(0 << 8)
114 #define	VM_INTINFO_NMI		(2 << 8)
115 #define	VM_INTINFO_HWEXCEPTION	(3 << 8)
116 #define	VM_INTINFO_SWINTR	(4 << 8)
117 
118 /*
119  * The VM name has to fit into the pathname length constraints of devfs,
120  * governed primarily by SPECNAMELEN.  The length is the total number of
121  * characters in the full path, relative to the mount point and not
122  * including any leading '/' characters.
123  * A prefix and a suffix are added to the name specified by the user.
124  * The prefix is usually "vmm/" or "vmm.io/", but can be a few characters
125  * longer for future use.
126  * The suffix is a string that identifies a bootrom image or some similar
127  * image that is attached to the VM. A separator character gets added to
128  * the suffix automatically when generating the full path, so it must be
129  * accounted for, reducing the effective length by 1.
130  * The effective length of a VM name is 229 bytes for FreeBSD 13 and 37
131  * bytes for FreeBSD 12.  A minimum length is set for safety and supports
132  * a SPECNAMELEN as small as 32 on old systems.
133  */
134 #define VM_MAX_PREFIXLEN 10
135 #define VM_MAX_SUFFIXLEN 15
136 #define VM_MIN_NAMELEN   6
137 #define VM_MAX_NAMELEN \
138     (SPECNAMELEN - VM_MAX_PREFIXLEN - VM_MAX_SUFFIXLEN - 1)
139 
140 #ifdef _KERNEL
141 CTASSERT(VM_MAX_NAMELEN >= VM_MIN_NAMELEN);
142 
143 struct vm;
144 struct vm_exception;
145 struct seg_desc;
146 struct vm_exit;
147 struct vm_run;
148 struct vhpet;
149 struct vioapic;
150 struct vlapic;
151 struct vmspace;
152 struct vm_object;
153 struct vm_guest_paging;
154 struct pmap;
155 
156 struct vm_eventinfo {
157 	void	*rptr;		/* rendezvous cookie */
158 	int	*sptr;		/* suspend cookie */
159 	int	*iptr;		/* reqidle cookie */
160 };
161 
162 typedef int	(*vmm_init_func_t)(int ipinum);
163 typedef int	(*vmm_cleanup_func_t)(void);
164 typedef void	(*vmm_resume_func_t)(void);
165 typedef void *	(*vmi_init_func_t)(struct vm *vm, struct pmap *pmap);
166 typedef int	(*vmi_run_func_t)(void *vmi, int vcpu, register_t rip,
167 		    struct pmap *pmap, struct vm_eventinfo *info);
168 typedef void	(*vmi_cleanup_func_t)(void *vmi);
169 typedef int	(*vmi_get_register_t)(void *vmi, int vcpu, int num,
170 				      uint64_t *retval);
171 typedef int	(*vmi_set_register_t)(void *vmi, int vcpu, int num,
172 				      uint64_t val);
173 typedef int	(*vmi_get_desc_t)(void *vmi, int vcpu, int num,
174 				  struct seg_desc *desc);
175 typedef int	(*vmi_set_desc_t)(void *vmi, int vcpu, int num,
176 				  struct seg_desc *desc);
177 typedef int	(*vmi_get_cap_t)(void *vmi, int vcpu, int num, int *retval);
178 typedef int	(*vmi_set_cap_t)(void *vmi, int vcpu, int num, int val);
179 typedef struct vmspace * (*vmi_vmspace_alloc)(vm_offset_t min, vm_offset_t max);
180 typedef void	(*vmi_vmspace_free)(struct vmspace *vmspace);
181 typedef struct vlapic * (*vmi_vlapic_init)(void *vmi, int vcpu);
182 typedef void	(*vmi_vlapic_cleanup)(void *vmi, struct vlapic *vlapic);
183 
184 struct vmm_ops {
185 	vmm_init_func_t		init;		/* module wide initialization */
186 	vmm_cleanup_func_t	cleanup;
187 	vmm_resume_func_t	resume;
188 
189 	vmi_init_func_t		vminit;		/* vm-specific initialization */
190 	vmi_run_func_t		vmrun;
191 	vmi_cleanup_func_t	vmcleanup;
192 	vmi_get_register_t	vmgetreg;
193 	vmi_set_register_t	vmsetreg;
194 	vmi_get_desc_t		vmgetdesc;
195 	vmi_set_desc_t		vmsetdesc;
196 	vmi_get_cap_t		vmgetcap;
197 	vmi_set_cap_t		vmsetcap;
198 	vmi_vmspace_alloc	vmspace_alloc;
199 	vmi_vmspace_free	vmspace_free;
200 	vmi_vlapic_init		vlapic_init;
201 	vmi_vlapic_cleanup	vlapic_cleanup;
202 };
203 
204 extern struct vmm_ops vmm_ops_intel;
205 extern struct vmm_ops vmm_ops_amd;
206 
207 int vm_create(const char *name, struct vm **retvm);
208 void vm_destroy(struct vm *vm);
209 int vm_reinit(struct vm *vm);
210 const char *vm_name(struct vm *vm);
211 uint16_t vm_get_maxcpus(struct vm *vm);
212 void vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores,
213     uint16_t *threads, uint16_t *maxcpus);
214 int vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores,
215     uint16_t threads, uint16_t maxcpus);
216 
217 /*
218  * APIs that modify the guest memory map require all vcpus to be frozen.
219  */
220 int vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t off,
221     size_t len, int prot, int flags);
222 int vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem);
223 void vm_free_memseg(struct vm *vm, int ident);
224 int vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa);
225 int vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len);
226 int vm_assign_pptdev(struct vm *vm, int bus, int slot, int func);
227 int vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func);
228 
229 /*
230  * APIs that inspect the guest memory map require only a *single* vcpu to
231  * be frozen. This acts like a read lock on the guest memory map since any
232  * modification requires *all* vcpus to be frozen.
233  */
234 int vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
235     vm_ooffset_t *segoff, size_t *len, int *prot, int *flags);
236 int vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
237     struct vm_object **objptr);
238 vm_paddr_t vmm_sysmem_maxaddr(struct vm *vm);
239 void *vm_gpa_hold(struct vm *, int vcpuid, vm_paddr_t gpa, size_t len,
240     int prot, void **cookie);
241 void vm_gpa_release(void *cookie);
242 bool vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa);
243 
244 int vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval);
245 int vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val);
246 int vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
247 		    struct seg_desc *ret_desc);
248 int vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
249 		    struct seg_desc *desc);
250 int vm_run(struct vm *vm, struct vm_run *vmrun);
251 int vm_suspend(struct vm *vm, enum vm_suspend_how how);
252 int vm_inject_nmi(struct vm *vm, int vcpu);
253 int vm_nmi_pending(struct vm *vm, int vcpuid);
254 void vm_nmi_clear(struct vm *vm, int vcpuid);
255 int vm_inject_extint(struct vm *vm, int vcpu);
256 int vm_extint_pending(struct vm *vm, int vcpuid);
257 void vm_extint_clear(struct vm *vm, int vcpuid);
258 struct vlapic *vm_lapic(struct vm *vm, int cpu);
259 struct vioapic *vm_ioapic(struct vm *vm);
260 struct vhpet *vm_hpet(struct vm *vm);
261 int vm_get_capability(struct vm *vm, int vcpu, int type, int *val);
262 int vm_set_capability(struct vm *vm, int vcpu, int type, int val);
263 int vm_get_x2apic_state(struct vm *vm, int vcpu, enum x2apic_state *state);
264 int vm_set_x2apic_state(struct vm *vm, int vcpu, enum x2apic_state state);
265 int vm_apicid2vcpuid(struct vm *vm, int apicid);
266 int vm_activate_cpu(struct vm *vm, int vcpu);
267 int vm_suspend_cpu(struct vm *vm, int vcpu);
268 int vm_resume_cpu(struct vm *vm, int vcpu);
269 struct vm_exit *vm_exitinfo(struct vm *vm, int vcpuid);
270 void vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip);
271 void vm_exit_debug(struct vm *vm, int vcpuid, uint64_t rip);
272 void vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip);
273 void vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip);
274 void vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip);
275 
276 #ifdef _SYS__CPUSET_H_
277 /*
278  * Rendezvous all vcpus specified in 'dest' and execute 'func(arg)'.
279  * The rendezvous 'func(arg)' is not allowed to do anything that will
280  * cause the thread to be put to sleep.
281  *
282  * If the rendezvous is being initiated from a vcpu context then the
283  * 'vcpuid' must refer to that vcpu, otherwise it should be set to -1.
284  *
285  * The caller cannot hold any locks when initiating the rendezvous.
286  *
287  * The implementation of this API may cause vcpus other than those specified
288  * by 'dest' to be stalled. The caller should not rely on any vcpus making
289  * forward progress when the rendezvous is in progress.
290  */
291 typedef void (*vm_rendezvous_func_t)(struct vm *vm, int vcpuid, void *arg);
292 int vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
293     vm_rendezvous_func_t func, void *arg);
294 cpuset_t vm_active_cpus(struct vm *vm);
295 cpuset_t vm_debug_cpus(struct vm *vm);
296 cpuset_t vm_suspended_cpus(struct vm *vm);
297 #endif	/* _SYS__CPUSET_H_ */
298 
299 static __inline int
300 vcpu_rendezvous_pending(struct vm_eventinfo *info)
301 {
302 
303 	return (*((uintptr_t *)(info->rptr)) != 0);
304 }
305 
306 static __inline int
307 vcpu_suspended(struct vm_eventinfo *info)
308 {
309 
310 	return (*info->sptr);
311 }
312 
313 static __inline int
314 vcpu_reqidle(struct vm_eventinfo *info)
315 {
316 
317 	return (*info->iptr);
318 }
319 
320 int vcpu_debugged(struct vm *vm, int vcpuid);
321 
322 /*
323  * Return true if device indicated by bus/slot/func is supposed to be a
324  * pci passthrough device.
325  *
326  * Return false otherwise.
327  */
328 bool vmm_is_pptdev(int bus, int slot, int func);
329 
330 void *vm_iommu_domain(struct vm *vm);
331 
332 enum vcpu_state {
333 	VCPU_IDLE,
334 	VCPU_FROZEN,
335 	VCPU_RUNNING,
336 	VCPU_SLEEPING,
337 };
338 
339 int vcpu_set_state(struct vm *vm, int vcpu, enum vcpu_state state,
340     bool from_idle);
341 enum vcpu_state vcpu_get_state(struct vm *vm, int vcpu, int *hostcpu);
342 
343 static int __inline
344 vcpu_is_running(struct vm *vm, int vcpu, int *hostcpu)
345 {
346 	return (vcpu_get_state(vm, vcpu, hostcpu) == VCPU_RUNNING);
347 }
348 
349 #ifdef _SYS_PROC_H_
350 static int __inline
351 vcpu_should_yield(struct vm *vm, int vcpu)
352 {
353 
354 	if (curthread->td_flags & (TDF_ASTPENDING | TDF_NEEDRESCHED))
355 		return (1);
356 	else if (curthread->td_owepreempt)
357 		return (1);
358 	else
359 		return (0);
360 }
361 #endif
362 
363 void *vcpu_stats(struct vm *vm, int vcpu);
364 void vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr);
365 struct vmspace *vm_get_vmspace(struct vm *vm);
366 struct vatpic *vm_atpic(struct vm *vm);
367 struct vatpit *vm_atpit(struct vm *vm);
368 struct vpmtmr *vm_pmtmr(struct vm *vm);
369 struct vrtc *vm_rtc(struct vm *vm);
370 
371 /*
372  * Inject exception 'vector' into the guest vcpu. This function returns 0 on
373  * success and non-zero on failure.
374  *
375  * Wrapper functions like 'vm_inject_gp()' should be preferred to calling
376  * this function directly because they enforce the trap-like or fault-like
377  * behavior of an exception.
378  *
379  * This function should only be called in the context of the thread that is
380  * executing this vcpu.
381  */
382 int vm_inject_exception(struct vm *vm, int vcpuid, int vector, int err_valid,
383     uint32_t errcode, int restart_instruction);
384 
385 /*
386  * This function is called after a VM-exit that occurred during exception or
387  * interrupt delivery through the IDT. The format of 'intinfo' is described
388  * in Figure 15-1, "EXITINTINFO for All Intercepts", APM, Vol 2.
389  *
390  * If a VM-exit handler completes the event delivery successfully then it
391  * should call vm_exit_intinfo() to extinguish the pending event. For e.g.,
392  * if the task switch emulation is triggered via a task gate then it should
393  * call this function with 'intinfo=0' to indicate that the external event
394  * is not pending anymore.
395  *
396  * Return value is 0 on success and non-zero on failure.
397  */
398 int vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t intinfo);
399 
400 /*
401  * This function is called before every VM-entry to retrieve a pending
402  * event that should be injected into the guest. This function combines
403  * nested events into a double or triple fault.
404  *
405  * Returns 0 if there are no events that need to be injected into the guest
406  * and non-zero otherwise.
407  */
408 int vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *info);
409 
410 int vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2);
411 
412 enum vm_reg_name vm_segment_name(int seg_encoding);
413 
414 struct vm_copyinfo {
415 	uint64_t	gpa;
416 	size_t		len;
417 	void		*hva;
418 	void		*cookie;
419 };
420 
421 /*
422  * Set up 'copyinfo[]' to copy to/from guest linear address space starting
423  * at 'gla' and 'len' bytes long. The 'prot' should be set to PROT_READ for
424  * a copyin or PROT_WRITE for a copyout.
425  *
426  * retval	is_fault	Interpretation
427  *   0		   0		Success
428  *   0		   1		An exception was injected into the guest
429  * EFAULT	  N/A		Unrecoverable error
430  *
431  * The 'copyinfo[]' can be passed to 'vm_copyin()' or 'vm_copyout()' only if
432  * the return value is 0. The 'copyinfo[]' resources should be freed by calling
433  * 'vm_copy_teardown()' after the copy is done.
434  */
435 int vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
436     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
437     int num_copyinfo, int *is_fault);
438 void vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
439     int num_copyinfo);
440 void vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
441     void *kaddr, size_t len);
442 void vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
443     struct vm_copyinfo *copyinfo, size_t len);
444 
445 int vcpu_trace_exceptions(struct vm *vm, int vcpuid);
446 #endif	/* KERNEL */
447 
448 #define	VM_MAXCPU	16			/* maximum virtual cpus */
449 
450 /*
451  * Identifiers for optional vmm capabilities
452  */
453 enum vm_cap_type {
454 	VM_CAP_HALT_EXIT,
455 	VM_CAP_MTRAP_EXIT,
456 	VM_CAP_PAUSE_EXIT,
457 	VM_CAP_UNRESTRICTED_GUEST,
458 	VM_CAP_ENABLE_INVPCID,
459 	VM_CAP_BPT_EXIT,
460 	VM_CAP_MAX
461 };
462 
463 enum vm_intr_trigger {
464 	EDGE_TRIGGER,
465 	LEVEL_TRIGGER
466 };
467 
468 /*
469  * The 'access' field has the format specified in Table 21-2 of the Intel
470  * Architecture Manual vol 3b.
471  *
472  * XXX The contents of the 'access' field are architecturally defined except
473  * bit 16 - Segment Unusable.
474  */
475 struct seg_desc {
476 	uint64_t	base;
477 	uint32_t	limit;
478 	uint32_t	access;
479 };
480 #define	SEG_DESC_TYPE(access)		((access) & 0x001f)
481 #define	SEG_DESC_DPL(access)		(((access) >> 5) & 0x3)
482 #define	SEG_DESC_PRESENT(access)	(((access) & 0x0080) ? 1 : 0)
483 #define	SEG_DESC_DEF32(access)		(((access) & 0x4000) ? 1 : 0)
484 #define	SEG_DESC_GRANULARITY(access)	(((access) & 0x8000) ? 1 : 0)
485 #define	SEG_DESC_UNUSABLE(access)	(((access) & 0x10000) ? 1 : 0)
486 
487 enum vm_cpu_mode {
488 	CPU_MODE_REAL,
489 	CPU_MODE_PROTECTED,
490 	CPU_MODE_COMPATIBILITY,		/* IA-32E mode (CS.L = 0) */
491 	CPU_MODE_64BIT,			/* IA-32E mode (CS.L = 1) */
492 };
493 
494 enum vm_paging_mode {
495 	PAGING_MODE_FLAT,
496 	PAGING_MODE_32,
497 	PAGING_MODE_PAE,
498 	PAGING_MODE_64,
499 };
500 
501 struct vm_guest_paging {
502 	uint64_t	cr3;
503 	int		cpl;
504 	enum vm_cpu_mode cpu_mode;
505 	enum vm_paging_mode paging_mode;
506 };
507 
508 /*
509  * The data structures 'vie' and 'vie_op' are meant to be opaque to the
510  * consumers of instruction decoding. The only reason why their contents
511  * need to be exposed is because they are part of the 'vm_exit' structure.
512  */
513 struct vie_op {
514 	uint8_t		op_byte;	/* actual opcode byte */
515 	uint8_t		op_type;	/* type of operation (e.g. MOV) */
516 	uint16_t	op_flags;
517 };
518 _Static_assert(sizeof(struct vie_op) == 4, "ABI");
519 _Static_assert(_Alignof(struct vie_op) == 2, "ABI");
520 
521 #define	VIE_INST_SIZE	15
522 struct vie {
523 	uint8_t		inst[VIE_INST_SIZE];	/* instruction bytes */
524 	uint8_t		num_valid;		/* size of the instruction */
525 	uint8_t		num_processed;
526 
527 	uint8_t		addrsize:4, opsize:4;	/* address and operand sizes */
528 	uint8_t		rex_w:1,		/* REX prefix */
529 			rex_r:1,
530 			rex_x:1,
531 			rex_b:1,
532 			rex_present:1,
533 			repz_present:1,		/* REP/REPE/REPZ prefix */
534 			repnz_present:1,	/* REPNE/REPNZ prefix */
535 			opsize_override:1,	/* Operand size override */
536 			addrsize_override:1,	/* Address size override */
537 			segment_override:1;	/* Segment override */
538 
539 	uint8_t		mod:2,			/* ModRM byte */
540 			reg:4,
541 			rm:4;
542 
543 	uint8_t		ss:2,			/* SIB byte */
544 			vex_present:1,		/* VEX prefixed */
545 			vex_l:1,		/* L bit */
546 			index:4,		/* SIB byte */
547 			base:4;			/* SIB byte */
548 
549 	uint8_t		disp_bytes;
550 	uint8_t		imm_bytes;
551 
552 	uint8_t		scale;
553 
554 	uint8_t		vex_reg:4,		/* vvvv: first source register specifier */
555 			vex_pp:2,		/* pp */
556 			_sparebits:2;
557 
558 	uint8_t		_sparebytes[2];
559 
560 	int		base_register;		/* VM_REG_GUEST_xyz */
561 	int		index_register;		/* VM_REG_GUEST_xyz */
562 	int		segment_register;	/* VM_REG_GUEST_xyz */
563 
564 	int64_t		displacement;		/* optional addr displacement */
565 	int64_t		immediate;		/* optional immediate operand */
566 
567 	uint8_t		decoded;	/* set to 1 if successfully decoded */
568 
569 	uint8_t		_sparebyte;
570 
571 	struct vie_op	op;			/* opcode description */
572 };
573 _Static_assert(sizeof(struct vie) == 64, "ABI");
574 _Static_assert(__offsetof(struct vie, disp_bytes) == 22, "ABI");
575 _Static_assert(__offsetof(struct vie, scale) == 24, "ABI");
576 _Static_assert(__offsetof(struct vie, base_register) == 28, "ABI");
577 
578 enum vm_exitcode {
579 	VM_EXITCODE_INOUT,
580 	VM_EXITCODE_VMX,
581 	VM_EXITCODE_BOGUS,
582 	VM_EXITCODE_RDMSR,
583 	VM_EXITCODE_WRMSR,
584 	VM_EXITCODE_HLT,
585 	VM_EXITCODE_MTRAP,
586 	VM_EXITCODE_PAUSE,
587 	VM_EXITCODE_PAGING,
588 	VM_EXITCODE_INST_EMUL,
589 	VM_EXITCODE_SPINUP_AP,
590 	VM_EXITCODE_DEPRECATED1,	/* used to be SPINDOWN_CPU */
591 	VM_EXITCODE_RENDEZVOUS,
592 	VM_EXITCODE_IOAPIC_EOI,
593 	VM_EXITCODE_SUSPENDED,
594 	VM_EXITCODE_INOUT_STR,
595 	VM_EXITCODE_TASK_SWITCH,
596 	VM_EXITCODE_MONITOR,
597 	VM_EXITCODE_MWAIT,
598 	VM_EXITCODE_SVM,
599 	VM_EXITCODE_REQIDLE,
600 	VM_EXITCODE_DEBUG,
601 	VM_EXITCODE_VMINSN,
602 	VM_EXITCODE_BPT,
603 	VM_EXITCODE_MAX
604 };
605 
606 struct vm_inout {
607 	uint16_t	bytes:3;	/* 1 or 2 or 4 */
608 	uint16_t	in:1;
609 	uint16_t	string:1;
610 	uint16_t	rep:1;
611 	uint16_t	port;
612 	uint32_t	eax;		/* valid for out */
613 };
614 
615 struct vm_inout_str {
616 	struct vm_inout	inout;		/* must be the first element */
617 	struct vm_guest_paging paging;
618 	uint64_t	rflags;
619 	uint64_t	cr0;
620 	uint64_t	index;
621 	uint64_t	count;		/* rep=1 (%rcx), rep=0 (1) */
622 	int		addrsize;
623 	enum vm_reg_name seg_name;
624 	struct seg_desc seg_desc;
625 };
626 
627 enum task_switch_reason {
628 	TSR_CALL,
629 	TSR_IRET,
630 	TSR_JMP,
631 	TSR_IDT_GATE,	/* task gate in IDT */
632 };
633 
634 struct vm_task_switch {
635 	uint16_t	tsssel;		/* new TSS selector */
636 	int		ext;		/* task switch due to external event */
637 	uint32_t	errcode;
638 	int		errcode_valid;	/* push 'errcode' on the new stack */
639 	enum task_switch_reason reason;
640 	struct vm_guest_paging paging;
641 };
642 
643 struct vm_exit {
644 	enum vm_exitcode	exitcode;
645 	int			inst_length;	/* 0 means unknown */
646 	uint64_t		rip;
647 	union {
648 		struct vm_inout	inout;
649 		struct vm_inout_str inout_str;
650 		struct {
651 			uint64_t	gpa;
652 			int		fault_type;
653 		} paging;
654 		struct {
655 			uint64_t	gpa;
656 			uint64_t	gla;
657 			uint64_t	cs_base;
658 			int		cs_d;		/* CS.D */
659 			struct vm_guest_paging paging;
660 			struct vie	vie;
661 		} inst_emul;
662 		/*
663 		 * VMX specific payload. Used when there is no "better"
664 		 * exitcode to represent the VM-exit.
665 		 */
666 		struct {
667 			int		status;		/* vmx inst status */
668 			/*
669 			 * 'exit_reason' and 'exit_qualification' are valid
670 			 * only if 'status' is zero.
671 			 */
672 			uint32_t	exit_reason;
673 			uint64_t	exit_qualification;
674 			/*
675 			 * 'inst_error' and 'inst_type' are valid
676 			 * only if 'status' is non-zero.
677 			 */
678 			int		inst_type;
679 			int		inst_error;
680 		} vmx;
681 		/*
682 		 * SVM specific payload.
683 		 */
684 		struct {
685 			uint64_t	exitcode;
686 			uint64_t	exitinfo1;
687 			uint64_t	exitinfo2;
688 		} svm;
689 		struct {
690 			int		inst_length;
691 		} bpt;
692 		struct {
693 			uint32_t	code;		/* ecx value */
694 			uint64_t	wval;
695 		} msr;
696 		struct {
697 			int		vcpu;
698 			uint64_t	rip;
699 		} spinup_ap;
700 		struct {
701 			uint64_t	rflags;
702 			uint64_t	intr_status;
703 		} hlt;
704 		struct {
705 			int		vector;
706 		} ioapic_eoi;
707 		struct {
708 			enum vm_suspend_how how;
709 		} suspended;
710 		struct vm_task_switch task_switch;
711 	} u;
712 };
713 
714 /* APIs to inject faults into the guest */
715 void vm_inject_fault(void *vm, int vcpuid, int vector, int errcode_valid,
716     int errcode);
717 
718 static __inline void
719 vm_inject_ud(void *vm, int vcpuid)
720 {
721 	vm_inject_fault(vm, vcpuid, IDT_UD, 0, 0);
722 }
723 
724 static __inline void
725 vm_inject_gp(void *vm, int vcpuid)
726 {
727 	vm_inject_fault(vm, vcpuid, IDT_GP, 1, 0);
728 }
729 
730 static __inline void
731 vm_inject_ac(void *vm, int vcpuid, int errcode)
732 {
733 	vm_inject_fault(vm, vcpuid, IDT_AC, 1, errcode);
734 }
735 
736 static __inline void
737 vm_inject_ss(void *vm, int vcpuid, int errcode)
738 {
739 	vm_inject_fault(vm, vcpuid, IDT_SS, 1, errcode);
740 }
741 
742 void vm_inject_pf(void *vm, int vcpuid, int error_code, uint64_t cr2);
743 
744 int vm_restart_instruction(void *vm, int vcpuid);
745 
746 #endif	/* _VMM_H_ */
747