xref: /freebsd/sys/amd64/include/vmm.h (revision 716fd348e01c5f2ba125f878a634a753436c2994)
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		modinit;	/* module wide initialization */
193 	vmm_cleanup_func_t	modcleanup;
194 	vmm_resume_func_t	modresume;
195 
196 	vmi_init_func_t		init;		/* vm-specific initialization */
197 	vmi_run_func_t		run;
198 	vmi_cleanup_func_t	cleanup;
199 	vmi_get_register_t	getreg;
200 	vmi_set_register_t	setreg;
201 	vmi_get_desc_t		getdesc;
202 	vmi_set_desc_t		setdesc;
203 	vmi_get_cap_t		getcap;
204 	vmi_set_cap_t		setcap;
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		snapshot;
212 	vmi_snapshot_vmcx_t	vmcx_snapshot;
213 	vmi_restore_tsc_t	restore_tsc;
214 };
215 
216 extern const struct vmm_ops vmm_ops_intel;
217 extern const 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_munmap_memseg(struct vm *vm, vm_paddr_t gpa, size_t len);
235 int vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem);
236 void vm_free_memseg(struct vm *vm, int ident);
237 int vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa);
238 int vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len);
239 int vm_assign_pptdev(struct vm *vm, int bus, int slot, int func);
240 int vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func);
241 
242 /*
243  * APIs that inspect the guest memory map require only a *single* vcpu to
244  * be frozen. This acts like a read lock on the guest memory map since any
245  * modification requires *all* vcpus to be frozen.
246  */
247 int vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
248     vm_ooffset_t *segoff, size_t *len, int *prot, int *flags);
249 int vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
250     struct vm_object **objptr);
251 vm_paddr_t vmm_sysmem_maxaddr(struct vm *vm);
252 void *vm_gpa_hold(struct vm *, int vcpuid, vm_paddr_t gpa, size_t len,
253     int prot, void **cookie);
254 void vm_gpa_release(void *cookie);
255 bool vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa);
256 
257 int vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval);
258 int vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val);
259 int vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
260 		    struct seg_desc *ret_desc);
261 int vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
262 		    struct seg_desc *desc);
263 int vm_run(struct vm *vm, struct vm_run *vmrun);
264 int vm_suspend(struct vm *vm, enum vm_suspend_how how);
265 int vm_inject_nmi(struct vm *vm, int vcpu);
266 int vm_nmi_pending(struct vm *vm, int vcpuid);
267 void vm_nmi_clear(struct vm *vm, int vcpuid);
268 int vm_inject_extint(struct vm *vm, int vcpu);
269 int vm_extint_pending(struct vm *vm, int vcpuid);
270 void vm_extint_clear(struct vm *vm, int vcpuid);
271 struct vlapic *vm_lapic(struct vm *vm, int cpu);
272 struct vioapic *vm_ioapic(struct vm *vm);
273 struct vhpet *vm_hpet(struct vm *vm);
274 int vm_get_capability(struct vm *vm, int vcpu, int type, int *val);
275 int vm_set_capability(struct vm *vm, int vcpu, int type, int val);
276 int vm_get_x2apic_state(struct vm *vm, int vcpu, enum x2apic_state *state);
277 int vm_set_x2apic_state(struct vm *vm, int vcpu, enum x2apic_state state);
278 int vm_apicid2vcpuid(struct vm *vm, int apicid);
279 int vm_activate_cpu(struct vm *vm, int vcpu);
280 int vm_suspend_cpu(struct vm *vm, int vcpu);
281 int vm_resume_cpu(struct vm *vm, int vcpu);
282 struct vm_exit *vm_exitinfo(struct vm *vm, int vcpuid);
283 void vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip);
284 void vm_exit_debug(struct vm *vm, int vcpuid, uint64_t rip);
285 void vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip);
286 void vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip);
287 void vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip);
288 int vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta);
289 int vm_restore_time(struct vm *vm);
290 
291 #ifdef _SYS__CPUSET_H_
292 /*
293  * Rendezvous all vcpus specified in 'dest' and execute 'func(arg)'.
294  * The rendezvous 'func(arg)' is not allowed to do anything that will
295  * cause the thread to be put to sleep.
296  *
297  * If the rendezvous is being initiated from a vcpu context then the
298  * 'vcpuid' must refer to that vcpu, otherwise it should be set to -1.
299  *
300  * The caller cannot hold any locks when initiating the rendezvous.
301  *
302  * The implementation of this API may cause vcpus other than those specified
303  * by 'dest' to be stalled. The caller should not rely on any vcpus making
304  * forward progress when the rendezvous is in progress.
305  */
306 typedef void (*vm_rendezvous_func_t)(struct vm *vm, int vcpuid, void *arg);
307 int vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
308     vm_rendezvous_func_t func, void *arg);
309 cpuset_t vm_active_cpus(struct vm *vm);
310 cpuset_t vm_debug_cpus(struct vm *vm);
311 cpuset_t vm_suspended_cpus(struct vm *vm);
312 #endif	/* _SYS__CPUSET_H_ */
313 
314 static __inline int
315 vcpu_rendezvous_pending(struct vm_eventinfo *info)
316 {
317 
318 	return (*((uintptr_t *)(info->rptr)) != 0);
319 }
320 
321 static __inline int
322 vcpu_suspended(struct vm_eventinfo *info)
323 {
324 
325 	return (*info->sptr);
326 }
327 
328 static __inline int
329 vcpu_reqidle(struct vm_eventinfo *info)
330 {
331 
332 	return (*info->iptr);
333 }
334 
335 int vcpu_debugged(struct vm *vm, int vcpuid);
336 
337 /*
338  * Return true if device indicated by bus/slot/func is supposed to be a
339  * pci passthrough device.
340  *
341  * Return false otherwise.
342  */
343 bool vmm_is_pptdev(int bus, int slot, int func);
344 
345 void *vm_iommu_domain(struct vm *vm);
346 
347 enum vcpu_state {
348 	VCPU_IDLE,
349 	VCPU_FROZEN,
350 	VCPU_RUNNING,
351 	VCPU_SLEEPING,
352 };
353 
354 int vcpu_set_state(struct vm *vm, int vcpu, enum vcpu_state state,
355     bool from_idle);
356 enum vcpu_state vcpu_get_state(struct vm *vm, int vcpu, int *hostcpu);
357 
358 static int __inline
359 vcpu_is_running(struct vm *vm, int vcpu, int *hostcpu)
360 {
361 	return (vcpu_get_state(vm, vcpu, hostcpu) == VCPU_RUNNING);
362 }
363 
364 #ifdef _SYS_PROC_H_
365 static int __inline
366 vcpu_should_yield(struct vm *vm, int vcpu)
367 {
368 
369 	if (curthread->td_flags & (TDF_ASTPENDING | TDF_NEEDRESCHED))
370 		return (1);
371 	else if (curthread->td_owepreempt)
372 		return (1);
373 	else
374 		return (0);
375 }
376 #endif
377 
378 void *vcpu_stats(struct vm *vm, int vcpu);
379 void vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr);
380 struct vmspace *vm_get_vmspace(struct vm *vm);
381 struct vatpic *vm_atpic(struct vm *vm);
382 struct vatpit *vm_atpit(struct vm *vm);
383 struct vpmtmr *vm_pmtmr(struct vm *vm);
384 struct vrtc *vm_rtc(struct vm *vm);
385 
386 /*
387  * Inject exception 'vector' into the guest vcpu. This function returns 0 on
388  * success and non-zero on failure.
389  *
390  * Wrapper functions like 'vm_inject_gp()' should be preferred to calling
391  * this function directly because they enforce the trap-like or fault-like
392  * behavior of an exception.
393  *
394  * This function should only be called in the context of the thread that is
395  * executing this vcpu.
396  */
397 int vm_inject_exception(struct vm *vm, int vcpuid, int vector, int err_valid,
398     uint32_t errcode, int restart_instruction);
399 
400 /*
401  * This function is called after a VM-exit that occurred during exception or
402  * interrupt delivery through the IDT. The format of 'intinfo' is described
403  * in Figure 15-1, "EXITINTINFO for All Intercepts", APM, Vol 2.
404  *
405  * If a VM-exit handler completes the event delivery successfully then it
406  * should call vm_exit_intinfo() to extinguish the pending event. For e.g.,
407  * if the task switch emulation is triggered via a task gate then it should
408  * call this function with 'intinfo=0' to indicate that the external event
409  * is not pending anymore.
410  *
411  * Return value is 0 on success and non-zero on failure.
412  */
413 int vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t intinfo);
414 
415 /*
416  * This function is called before every VM-entry to retrieve a pending
417  * event that should be injected into the guest. This function combines
418  * nested events into a double or triple fault.
419  *
420  * Returns 0 if there are no events that need to be injected into the guest
421  * and non-zero otherwise.
422  */
423 int vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *info);
424 
425 int vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2);
426 
427 /*
428  * Function used to keep track of the guest's TSC offset. The
429  * offset is used by the virutalization extensions to provide a consistent
430  * value for the Time Stamp Counter to the guest.
431  *
432  * Return value is 0 on success and non-zero on failure.
433  */
434 int vm_set_tsc_offset(struct vm *vm, int vcpu_id, uint64_t offset);
435 
436 enum vm_reg_name vm_segment_name(int seg_encoding);
437 
438 struct vm_copyinfo {
439 	uint64_t	gpa;
440 	size_t		len;
441 	void		*hva;
442 	void		*cookie;
443 };
444 
445 /*
446  * Set up 'copyinfo[]' to copy to/from guest linear address space starting
447  * at 'gla' and 'len' bytes long. The 'prot' should be set to PROT_READ for
448  * a copyin or PROT_WRITE for a copyout.
449  *
450  * retval	is_fault	Interpretation
451  *   0		   0		Success
452  *   0		   1		An exception was injected into the guest
453  * EFAULT	  N/A		Unrecoverable error
454  *
455  * The 'copyinfo[]' can be passed to 'vm_copyin()' or 'vm_copyout()' only if
456  * the return value is 0. The 'copyinfo[]' resources should be freed by calling
457  * 'vm_copy_teardown()' after the copy is done.
458  */
459 int vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
460     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
461     int num_copyinfo, int *is_fault);
462 void vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
463     int num_copyinfo);
464 void vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
465     void *kaddr, size_t len);
466 void vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
467     struct vm_copyinfo *copyinfo, size_t len);
468 
469 int vcpu_trace_exceptions(struct vm *vm, int vcpuid);
470 #endif	/* KERNEL */
471 
472 #ifdef _KERNEL
473 #define	VM_MAXCPU	16			/* maximum virtual cpus */
474 #endif
475 
476 /*
477  * Identifiers for optional vmm capabilities
478  */
479 enum vm_cap_type {
480 	VM_CAP_HALT_EXIT,
481 	VM_CAP_MTRAP_EXIT,
482 	VM_CAP_PAUSE_EXIT,
483 	VM_CAP_UNRESTRICTED_GUEST,
484 	VM_CAP_ENABLE_INVPCID,
485 	VM_CAP_BPT_EXIT,
486 	VM_CAP_RDPID,
487 	VM_CAP_RDTSCP,
488 	VM_CAP_MAX
489 };
490 
491 enum vm_intr_trigger {
492 	EDGE_TRIGGER,
493 	LEVEL_TRIGGER
494 };
495 
496 /*
497  * The 'access' field has the format specified in Table 21-2 of the Intel
498  * Architecture Manual vol 3b.
499  *
500  * XXX The contents of the 'access' field are architecturally defined except
501  * bit 16 - Segment Unusable.
502  */
503 struct seg_desc {
504 	uint64_t	base;
505 	uint32_t	limit;
506 	uint32_t	access;
507 };
508 #define	SEG_DESC_TYPE(access)		((access) & 0x001f)
509 #define	SEG_DESC_DPL(access)		(((access) >> 5) & 0x3)
510 #define	SEG_DESC_PRESENT(access)	(((access) & 0x0080) ? 1 : 0)
511 #define	SEG_DESC_DEF32(access)		(((access) & 0x4000) ? 1 : 0)
512 #define	SEG_DESC_GRANULARITY(access)	(((access) & 0x8000) ? 1 : 0)
513 #define	SEG_DESC_UNUSABLE(access)	(((access) & 0x10000) ? 1 : 0)
514 
515 enum vm_cpu_mode {
516 	CPU_MODE_REAL,
517 	CPU_MODE_PROTECTED,
518 	CPU_MODE_COMPATIBILITY,		/* IA-32E mode (CS.L = 0) */
519 	CPU_MODE_64BIT,			/* IA-32E mode (CS.L = 1) */
520 };
521 
522 enum vm_paging_mode {
523 	PAGING_MODE_FLAT,
524 	PAGING_MODE_32,
525 	PAGING_MODE_PAE,
526 	PAGING_MODE_64,
527 	PAGING_MODE_64_LA57,
528 };
529 
530 struct vm_guest_paging {
531 	uint64_t	cr3;
532 	int		cpl;
533 	enum vm_cpu_mode cpu_mode;
534 	enum vm_paging_mode paging_mode;
535 };
536 
537 /*
538  * The data structures 'vie' and 'vie_op' are meant to be opaque to the
539  * consumers of instruction decoding. The only reason why their contents
540  * need to be exposed is because they are part of the 'vm_exit' structure.
541  */
542 struct vie_op {
543 	uint8_t		op_byte;	/* actual opcode byte */
544 	uint8_t		op_type;	/* type of operation (e.g. MOV) */
545 	uint16_t	op_flags;
546 };
547 _Static_assert(sizeof(struct vie_op) == 4, "ABI");
548 _Static_assert(_Alignof(struct vie_op) == 2, "ABI");
549 
550 #define	VIE_INST_SIZE	15
551 struct vie {
552 	uint8_t		inst[VIE_INST_SIZE];	/* instruction bytes */
553 	uint8_t		num_valid;		/* size of the instruction */
554 
555 /* The following fields are all zeroed upon restart. */
556 #define	vie_startzero	num_processed
557 	uint8_t		num_processed;
558 
559 	uint8_t		addrsize:4, opsize:4;	/* address and operand sizes */
560 	uint8_t		rex_w:1,		/* REX prefix */
561 			rex_r:1,
562 			rex_x:1,
563 			rex_b:1,
564 			rex_present:1,
565 			repz_present:1,		/* REP/REPE/REPZ prefix */
566 			repnz_present:1,	/* REPNE/REPNZ prefix */
567 			opsize_override:1,	/* Operand size override */
568 			addrsize_override:1,	/* Address size override */
569 			segment_override:1;	/* Segment override */
570 
571 	uint8_t		mod:2,			/* ModRM byte */
572 			reg:4,
573 			rm:4;
574 
575 	uint8_t		ss:2,			/* SIB byte */
576 			vex_present:1,		/* VEX prefixed */
577 			vex_l:1,		/* L bit */
578 			index:4,		/* SIB byte */
579 			base:4;			/* SIB byte */
580 
581 	uint8_t		disp_bytes;
582 	uint8_t		imm_bytes;
583 
584 	uint8_t		scale;
585 
586 	uint8_t		vex_reg:4,		/* vvvv: first source register specifier */
587 			vex_pp:2,		/* pp */
588 			_sparebits:2;
589 
590 	uint8_t		_sparebytes[2];
591 
592 	int		base_register;		/* VM_REG_GUEST_xyz */
593 	int		index_register;		/* VM_REG_GUEST_xyz */
594 	int		segment_register;	/* VM_REG_GUEST_xyz */
595 
596 	int64_t		displacement;		/* optional addr displacement */
597 	int64_t		immediate;		/* optional immediate operand */
598 
599 	uint8_t		decoded;	/* set to 1 if successfully decoded */
600 
601 	uint8_t		_sparebyte;
602 
603 	struct vie_op	op;			/* opcode description */
604 };
605 _Static_assert(sizeof(struct vie) == 64, "ABI");
606 _Static_assert(__offsetof(struct vie, disp_bytes) == 22, "ABI");
607 _Static_assert(__offsetof(struct vie, scale) == 24, "ABI");
608 _Static_assert(__offsetof(struct vie, base_register) == 28, "ABI");
609 
610 enum vm_exitcode {
611 	VM_EXITCODE_INOUT,
612 	VM_EXITCODE_VMX,
613 	VM_EXITCODE_BOGUS,
614 	VM_EXITCODE_RDMSR,
615 	VM_EXITCODE_WRMSR,
616 	VM_EXITCODE_HLT,
617 	VM_EXITCODE_MTRAP,
618 	VM_EXITCODE_PAUSE,
619 	VM_EXITCODE_PAGING,
620 	VM_EXITCODE_INST_EMUL,
621 	VM_EXITCODE_SPINUP_AP,
622 	VM_EXITCODE_DEPRECATED1,	/* used to be SPINDOWN_CPU */
623 	VM_EXITCODE_RENDEZVOUS,
624 	VM_EXITCODE_IOAPIC_EOI,
625 	VM_EXITCODE_SUSPENDED,
626 	VM_EXITCODE_INOUT_STR,
627 	VM_EXITCODE_TASK_SWITCH,
628 	VM_EXITCODE_MONITOR,
629 	VM_EXITCODE_MWAIT,
630 	VM_EXITCODE_SVM,
631 	VM_EXITCODE_REQIDLE,
632 	VM_EXITCODE_DEBUG,
633 	VM_EXITCODE_VMINSN,
634 	VM_EXITCODE_BPT,
635 	VM_EXITCODE_MAX
636 };
637 
638 struct vm_inout {
639 	uint16_t	bytes:3;	/* 1 or 2 or 4 */
640 	uint16_t	in:1;
641 	uint16_t	string:1;
642 	uint16_t	rep:1;
643 	uint16_t	port;
644 	uint32_t	eax;		/* valid for out */
645 };
646 
647 struct vm_inout_str {
648 	struct vm_inout	inout;		/* must be the first element */
649 	struct vm_guest_paging paging;
650 	uint64_t	rflags;
651 	uint64_t	cr0;
652 	uint64_t	index;
653 	uint64_t	count;		/* rep=1 (%rcx), rep=0 (1) */
654 	int		addrsize;
655 	enum vm_reg_name seg_name;
656 	struct seg_desc seg_desc;
657 };
658 
659 enum task_switch_reason {
660 	TSR_CALL,
661 	TSR_IRET,
662 	TSR_JMP,
663 	TSR_IDT_GATE,	/* task gate in IDT */
664 };
665 
666 struct vm_task_switch {
667 	uint16_t	tsssel;		/* new TSS selector */
668 	int		ext;		/* task switch due to external event */
669 	uint32_t	errcode;
670 	int		errcode_valid;	/* push 'errcode' on the new stack */
671 	enum task_switch_reason reason;
672 	struct vm_guest_paging paging;
673 };
674 
675 struct vm_exit {
676 	enum vm_exitcode	exitcode;
677 	int			inst_length;	/* 0 means unknown */
678 	uint64_t		rip;
679 	union {
680 		struct vm_inout	inout;
681 		struct vm_inout_str inout_str;
682 		struct {
683 			uint64_t	gpa;
684 			int		fault_type;
685 		} paging;
686 		struct {
687 			uint64_t	gpa;
688 			uint64_t	gla;
689 			uint64_t	cs_base;
690 			int		cs_d;		/* CS.D */
691 			struct vm_guest_paging paging;
692 			struct vie	vie;
693 		} inst_emul;
694 		/*
695 		 * VMX specific payload. Used when there is no "better"
696 		 * exitcode to represent the VM-exit.
697 		 */
698 		struct {
699 			int		status;		/* vmx inst status */
700 			/*
701 			 * 'exit_reason' and 'exit_qualification' are valid
702 			 * only if 'status' is zero.
703 			 */
704 			uint32_t	exit_reason;
705 			uint64_t	exit_qualification;
706 			/*
707 			 * 'inst_error' and 'inst_type' are valid
708 			 * only if 'status' is non-zero.
709 			 */
710 			int		inst_type;
711 			int		inst_error;
712 		} vmx;
713 		/*
714 		 * SVM specific payload.
715 		 */
716 		struct {
717 			uint64_t	exitcode;
718 			uint64_t	exitinfo1;
719 			uint64_t	exitinfo2;
720 		} svm;
721 		struct {
722 			int		inst_length;
723 		} bpt;
724 		struct {
725 			uint32_t	code;		/* ecx value */
726 			uint64_t	wval;
727 		} msr;
728 		struct {
729 			int		vcpu;
730 			uint64_t	rip;
731 		} spinup_ap;
732 		struct {
733 			uint64_t	rflags;
734 			uint64_t	intr_status;
735 		} hlt;
736 		struct {
737 			int		vector;
738 		} ioapic_eoi;
739 		struct {
740 			enum vm_suspend_how how;
741 		} suspended;
742 		struct vm_task_switch task_switch;
743 	} u;
744 };
745 
746 /* APIs to inject faults into the guest */
747 void vm_inject_fault(void *vm, int vcpuid, int vector, int errcode_valid,
748     int errcode);
749 
750 static __inline void
751 vm_inject_ud(void *vm, int vcpuid)
752 {
753 	vm_inject_fault(vm, vcpuid, IDT_UD, 0, 0);
754 }
755 
756 static __inline void
757 vm_inject_gp(void *vm, int vcpuid)
758 {
759 	vm_inject_fault(vm, vcpuid, IDT_GP, 1, 0);
760 }
761 
762 static __inline void
763 vm_inject_ac(void *vm, int vcpuid, int errcode)
764 {
765 	vm_inject_fault(vm, vcpuid, IDT_AC, 1, errcode);
766 }
767 
768 static __inline void
769 vm_inject_ss(void *vm, int vcpuid, int errcode)
770 {
771 	vm_inject_fault(vm, vcpuid, IDT_SS, 1, errcode);
772 }
773 
774 void vm_inject_pf(void *vm, int vcpuid, int error_code, uint64_t cr2);
775 
776 int vm_restart_instruction(void *vm, int vcpuid);
777 
778 #endif	/* _VMM_H_ */
779