xref: /freebsd/sys/amd64/include/vmm.h (revision 0b32ef71f9f154f4da5037bfcbb4916960d38452)
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_GUEST_FS_BASE,
102 	VM_REG_GUEST_GS_BASE,
103 	VM_REG_GUEST_KGS_BASE,
104 	VM_REG_GUEST_TPR,
105 	VM_REG_LAST
106 };
107 
108 enum x2apic_state {
109 	X2APIC_DISABLED,
110 	X2APIC_ENABLED,
111 	X2APIC_STATE_LAST
112 };
113 
114 #define	VM_INTINFO_VECTOR(info)	((info) & 0xff)
115 #define	VM_INTINFO_DEL_ERRCODE	0x800
116 #define	VM_INTINFO_RSVD		0x7ffff000
117 #define	VM_INTINFO_VALID	0x80000000
118 #define	VM_INTINFO_TYPE		0x700
119 #define	VM_INTINFO_HWINTR	(0 << 8)
120 #define	VM_INTINFO_NMI		(2 << 8)
121 #define	VM_INTINFO_HWEXCEPTION	(3 << 8)
122 #define	VM_INTINFO_SWINTR	(4 << 8)
123 
124 /*
125  * The VM name has to fit into the pathname length constraints of devfs,
126  * governed primarily by SPECNAMELEN.  The length is the total number of
127  * characters in the full path, relative to the mount point and not
128  * including any leading '/' characters.
129  * A prefix and a suffix are added to the name specified by the user.
130  * The prefix is usually "vmm/" or "vmm.io/", but can be a few characters
131  * longer for future use.
132  * The suffix is a string that identifies a bootrom image or some similar
133  * image that is attached to the VM. A separator character gets added to
134  * the suffix automatically when generating the full path, so it must be
135  * accounted for, reducing the effective length by 1.
136  * The effective length of a VM name is 229 bytes for FreeBSD 13 and 37
137  * bytes for FreeBSD 12.  A minimum length is set for safety and supports
138  * a SPECNAMELEN as small as 32 on old systems.
139  */
140 #define VM_MAX_PREFIXLEN 10
141 #define VM_MAX_SUFFIXLEN 15
142 #define VM_MIN_NAMELEN   6
143 #define VM_MAX_NAMELEN \
144     (SPECNAMELEN - VM_MAX_PREFIXLEN - VM_MAX_SUFFIXLEN - 1)
145 
146 #ifdef _KERNEL
147 #include <sys/kassert.h>
148 
149 CTASSERT(VM_MAX_NAMELEN >= VM_MIN_NAMELEN);
150 
151 struct vm;
152 struct vm_exception;
153 struct seg_desc;
154 struct vm_exit;
155 struct vm_run;
156 struct vhpet;
157 struct vioapic;
158 struct vlapic;
159 struct vmspace;
160 struct vm_object;
161 struct vm_guest_paging;
162 struct pmap;
163 enum snapshot_req;
164 
165 struct vm_eventinfo {
166 	cpuset_t *rptr;		/* rendezvous cookie */
167 	int	*sptr;		/* suspend cookie */
168 	int	*iptr;		/* reqidle cookie */
169 };
170 
171 typedef int	(*vmm_init_func_t)(int ipinum);
172 typedef int	(*vmm_cleanup_func_t)(void);
173 typedef void	(*vmm_suspend_func_t)(void);
174 typedef void	(*vmm_resume_func_t)(void);
175 typedef void *	(*vmi_init_func_t)(struct vm *vm, struct pmap *pmap);
176 typedef int	(*vmi_run_func_t)(void *vcpui, register_t rip,
177 		    struct pmap *pmap, struct vm_eventinfo *info);
178 typedef void	(*vmi_cleanup_func_t)(void *vmi);
179 typedef void *	(*vmi_vcpu_init_func_t)(void *vmi, struct vcpu *vcpu,
180 		    int vcpu_id);
181 typedef void	(*vmi_vcpu_cleanup_func_t)(void *vcpui);
182 typedef int	(*vmi_get_register_t)(void *vcpui, int num, uint64_t *retval);
183 typedef int	(*vmi_set_register_t)(void *vcpui, int num, uint64_t val);
184 typedef int	(*vmi_get_desc_t)(void *vcpui, int num, struct seg_desc *desc);
185 typedef int	(*vmi_set_desc_t)(void *vcpui, int num, struct seg_desc *desc);
186 typedef int	(*vmi_get_cap_t)(void *vcpui, int num, int *retval);
187 typedef int	(*vmi_set_cap_t)(void *vcpui, int num, int val);
188 typedef struct vmspace * (*vmi_vmspace_alloc)(vm_offset_t min, vm_offset_t max);
189 typedef void	(*vmi_vmspace_free)(struct vmspace *vmspace);
190 typedef struct vlapic * (*vmi_vlapic_init)(void *vcpui);
191 typedef void	(*vmi_vlapic_cleanup)(struct vlapic *vlapic);
192 typedef int	(*vmi_snapshot_vcpu_t)(void *vcpui, struct vm_snapshot_meta *meta);
193 typedef int	(*vmi_restore_tsc_t)(void *vcpui, uint64_t now);
194 
195 struct vmm_ops {
196 	vmm_init_func_t		modinit;	/* module wide initialization */
197 	vmm_cleanup_func_t	modcleanup;
198 	vmm_resume_func_t	modsuspend;
199 	vmm_resume_func_t	modresume;
200 
201 	vmi_init_func_t		init;		/* vm-specific initialization */
202 	vmi_run_func_t		run;
203 	vmi_cleanup_func_t	cleanup;
204 	vmi_vcpu_init_func_t	vcpu_init;
205 	vmi_vcpu_cleanup_func_t	vcpu_cleanup;
206 	vmi_get_register_t	getreg;
207 	vmi_set_register_t	setreg;
208 	vmi_get_desc_t		getdesc;
209 	vmi_set_desc_t		setdesc;
210 	vmi_get_cap_t		getcap;
211 	vmi_set_cap_t		setcap;
212 	vmi_vmspace_alloc	vmspace_alloc;
213 	vmi_vmspace_free	vmspace_free;
214 	vmi_vlapic_init		vlapic_init;
215 	vmi_vlapic_cleanup	vlapic_cleanup;
216 
217 	/* checkpoint operations */
218 	vmi_snapshot_vcpu_t	vcpu_snapshot;
219 	vmi_restore_tsc_t	restore_tsc;
220 };
221 
222 extern const struct vmm_ops vmm_ops_intel;
223 extern const struct vmm_ops vmm_ops_amd;
224 
225 extern u_int vm_maxcpu;			/* maximum virtual cpus */
226 
227 int vm_create(const char *name, struct vm **retvm);
228 struct vcpu *vm_alloc_vcpu(struct vm *vm, int vcpuid);
229 void vm_disable_vcpu_creation(struct vm *vm);
230 void vm_slock_vcpus(struct vm *vm);
231 void vm_unlock_vcpus(struct vm *vm);
232 void vm_destroy(struct vm *vm);
233 int vm_reinit(struct vm *vm);
234 const char *vm_name(struct vm *vm);
235 uint16_t vm_get_maxcpus(struct vm *vm);
236 void vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores,
237     uint16_t *threads, uint16_t *maxcpus);
238 int vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores,
239     uint16_t threads, uint16_t maxcpus);
240 
241 /*
242  * APIs that modify the guest memory map require all vcpus to be frozen.
243  */
244 void vm_slock_memsegs(struct vm *vm);
245 void vm_xlock_memsegs(struct vm *vm);
246 void vm_unlock_memsegs(struct vm *vm);
247 int vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t off,
248     size_t len, int prot, int flags);
249 int vm_munmap_memseg(struct vm *vm, vm_paddr_t gpa, size_t len);
250 int vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem);
251 void vm_free_memseg(struct vm *vm, int ident);
252 int vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa);
253 int vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len);
254 int vm_assign_pptdev(struct vm *vm, int bus, int slot, int func);
255 int vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func);
256 
257 /*
258  * APIs that inspect the guest memory map require only a *single* vcpu to
259  * be frozen. This acts like a read lock on the guest memory map since any
260  * modification requires *all* vcpus to be frozen.
261  */
262 int vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
263     vm_ooffset_t *segoff, size_t *len, int *prot, int *flags);
264 int vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
265     struct vm_object **objptr);
266 vm_paddr_t vmm_sysmem_maxaddr(struct vm *vm);
267 void *vm_gpa_hold(struct vcpu *vcpu, vm_paddr_t gpa, size_t len,
268     int prot, void **cookie);
269 void *vm_gpa_hold_global(struct vm *vm, vm_paddr_t gpa, size_t len,
270     int prot, void **cookie);
271 void vm_gpa_release(void *cookie);
272 bool vm_mem_allocated(struct vcpu *vcpu, vm_paddr_t gpa);
273 
274 int vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval);
275 int vm_set_register(struct vcpu *vcpu, int reg, uint64_t val);
276 int vm_get_seg_desc(struct vcpu *vcpu, int reg,
277 		    struct seg_desc *ret_desc);
278 int vm_set_seg_desc(struct vcpu *vcpu, int reg,
279 		    struct seg_desc *desc);
280 int vm_run(struct vcpu *vcpu);
281 int vm_suspend(struct vm *vm, enum vm_suspend_how how);
282 int vm_inject_nmi(struct vcpu *vcpu);
283 int vm_nmi_pending(struct vcpu *vcpu);
284 void vm_nmi_clear(struct vcpu *vcpu);
285 int vm_inject_extint(struct vcpu *vcpu);
286 int vm_extint_pending(struct vcpu *vcpu);
287 void vm_extint_clear(struct vcpu *vcpu);
288 int vcpu_vcpuid(struct vcpu *vcpu);
289 struct vm *vcpu_vm(struct vcpu *vcpu);
290 struct vcpu *vm_vcpu(struct vm *vm, int cpu);
291 struct vlapic *vm_lapic(struct vcpu *vcpu);
292 struct vioapic *vm_ioapic(struct vm *vm);
293 struct vhpet *vm_hpet(struct vm *vm);
294 int vm_get_capability(struct vcpu *vcpu, int type, int *val);
295 int vm_set_capability(struct vcpu *vcpu, int type, int val);
296 int vm_get_x2apic_state(struct vcpu *vcpu, enum x2apic_state *state);
297 int vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state);
298 int vm_apicid2vcpuid(struct vm *vm, int apicid);
299 int vm_activate_cpu(struct vcpu *vcpu);
300 int vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu);
301 int vm_resume_cpu(struct vm *vm, struct vcpu *vcpu);
302 int vm_restart_instruction(struct vcpu *vcpu);
303 struct vm_exit *vm_exitinfo(struct vcpu *vcpu);
304 cpuset_t *vm_exitinfo_cpuset(struct vcpu *vcpu);
305 void vm_exit_suspended(struct vcpu *vcpu, uint64_t rip);
306 void vm_exit_debug(struct vcpu *vcpu, uint64_t rip);
307 void vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip);
308 void vm_exit_astpending(struct vcpu *vcpu, uint64_t rip);
309 void vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip);
310 int vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta);
311 int vm_restore_time(struct vm *vm);
312 
313 #ifdef _SYS__CPUSET_H_
314 /*
315  * Rendezvous all vcpus specified in 'dest' and execute 'func(arg)'.
316  * The rendezvous 'func(arg)' is not allowed to do anything that will
317  * cause the thread to be put to sleep.
318  *
319  * The caller cannot hold any locks when initiating the rendezvous.
320  *
321  * The implementation of this API may cause vcpus other than those specified
322  * by 'dest' to be stalled. The caller should not rely on any vcpus making
323  * forward progress when the rendezvous is in progress.
324  */
325 typedef void (*vm_rendezvous_func_t)(struct vcpu *vcpu, void *arg);
326 int vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest,
327     vm_rendezvous_func_t func, void *arg);
328 
329 cpuset_t vm_active_cpus(struct vm *vm);
330 cpuset_t vm_debug_cpus(struct vm *vm);
331 cpuset_t vm_suspended_cpus(struct vm *vm);
332 cpuset_t vm_start_cpus(struct vm *vm, const cpuset_t *tostart);
333 void vm_await_start(struct vm *vm, const cpuset_t *waiting);
334 #endif	/* _SYS__CPUSET_H_ */
335 
336 static __inline int
vcpu_rendezvous_pending(struct vcpu * vcpu,struct vm_eventinfo * info)337 vcpu_rendezvous_pending(struct vcpu *vcpu, struct vm_eventinfo *info)
338 {
339 	/*
340 	 * This check isn't done with atomic operations or under a lock because
341 	 * there's no need to. If the vcpuid bit is set, the vcpu is part of a
342 	 * rendezvous and the bit won't be cleared until the vcpu enters the
343 	 * rendezvous. On rendezvous exit, the cpuset is cleared and the vcpu
344 	 * will see an empty cpuset. So, the races are harmless.
345 	 */
346 	return (CPU_ISSET(vcpu_vcpuid(vcpu), info->rptr));
347 }
348 
349 static __inline int
vcpu_suspended(struct vm_eventinfo * info)350 vcpu_suspended(struct vm_eventinfo *info)
351 {
352 
353 	return (*info->sptr);
354 }
355 
356 static __inline int
vcpu_reqidle(struct vm_eventinfo * info)357 vcpu_reqidle(struct vm_eventinfo *info)
358 {
359 
360 	return (*info->iptr);
361 }
362 
363 int vcpu_debugged(struct vcpu *vcpu);
364 
365 /*
366  * Return true if device indicated by bus/slot/func is supposed to be a
367  * pci passthrough device.
368  *
369  * Return false otherwise.
370  */
371 bool vmm_is_pptdev(int bus, int slot, int func);
372 
373 void *vm_iommu_domain(struct vm *vm);
374 
375 enum vcpu_state {
376 	VCPU_IDLE,
377 	VCPU_FROZEN,
378 	VCPU_RUNNING,
379 	VCPU_SLEEPING,
380 };
381 
382 int vcpu_set_state(struct vcpu *vcpu, enum vcpu_state state, bool from_idle);
383 enum vcpu_state vcpu_get_state(struct vcpu *vcpu, int *hostcpu);
384 
385 static int __inline
vcpu_is_running(struct vcpu * vcpu,int * hostcpu)386 vcpu_is_running(struct vcpu *vcpu, int *hostcpu)
387 {
388 	return (vcpu_get_state(vcpu, hostcpu) == VCPU_RUNNING);
389 }
390 
391 #ifdef _SYS_PROC_H_
392 static int __inline
vcpu_should_yield(struct vcpu * vcpu)393 vcpu_should_yield(struct vcpu *vcpu)
394 {
395 	struct thread *td;
396 
397 	td = curthread;
398 	return (td->td_ast != 0 || td->td_owepreempt != 0);
399 }
400 #endif
401 
402 void *vcpu_stats(struct vcpu *vcpu);
403 void vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr);
404 struct vmspace *vm_get_vmspace(struct vm *vm);
405 struct vatpic *vm_atpic(struct vm *vm);
406 struct vatpit *vm_atpit(struct vm *vm);
407 struct vpmtmr *vm_pmtmr(struct vm *vm);
408 struct vrtc *vm_rtc(struct vm *vm);
409 
410 /*
411  * Inject exception 'vector' into the guest vcpu. This function returns 0 on
412  * success and non-zero on failure.
413  *
414  * Wrapper functions like 'vm_inject_gp()' should be preferred to calling
415  * this function directly because they enforce the trap-like or fault-like
416  * behavior of an exception.
417  *
418  * This function should only be called in the context of the thread that is
419  * executing this vcpu.
420  */
421 int vm_inject_exception(struct vcpu *vcpu, int vector, int err_valid,
422     uint32_t errcode, int restart_instruction);
423 
424 /*
425  * This function is called after a VM-exit that occurred during exception or
426  * interrupt delivery through the IDT. The format of 'intinfo' is described
427  * in Figure 15-1, "EXITINTINFO for All Intercepts", APM, Vol 2.
428  *
429  * If a VM-exit handler completes the event delivery successfully then it
430  * should call vm_exit_intinfo() to extinguish the pending event. For e.g.,
431  * if the task switch emulation is triggered via a task gate then it should
432  * call this function with 'intinfo=0' to indicate that the external event
433  * is not pending anymore.
434  *
435  * Return value is 0 on success and non-zero on failure.
436  */
437 int vm_exit_intinfo(struct vcpu *vcpu, uint64_t intinfo);
438 
439 /*
440  * This function is called before every VM-entry to retrieve a pending
441  * event that should be injected into the guest. This function combines
442  * nested events into a double or triple fault.
443  *
444  * Returns 0 if there are no events that need to be injected into the guest
445  * and non-zero otherwise.
446  */
447 int vm_entry_intinfo(struct vcpu *vcpu, uint64_t *info);
448 
449 int vm_get_intinfo(struct vcpu *vcpu, uint64_t *info1, uint64_t *info2);
450 
451 /*
452  * Function used to keep track of the guest's TSC offset. The
453  * offset is used by the virtualization extensions to provide a consistent
454  * value for the Time Stamp Counter to the guest.
455  */
456 void vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset);
457 
458 enum vm_reg_name vm_segment_name(int seg_encoding);
459 
460 struct vm_copyinfo {
461 	uint64_t	gpa;
462 	size_t		len;
463 	void		*hva;
464 	void		*cookie;
465 };
466 
467 /*
468  * Set up 'copyinfo[]' to copy to/from guest linear address space starting
469  * at 'gla' and 'len' bytes long. The 'prot' should be set to PROT_READ for
470  * a copyin or PROT_WRITE for a copyout.
471  *
472  * retval	is_fault	Interpretation
473  *   0		   0		Success
474  *   0		   1		An exception was injected into the guest
475  * EFAULT	  N/A		Unrecoverable error
476  *
477  * The 'copyinfo[]' can be passed to 'vm_copyin()' or 'vm_copyout()' only if
478  * the return value is 0. The 'copyinfo[]' resources should be freed by calling
479  * 'vm_copy_teardown()' after the copy is done.
480  */
481 int vm_copy_setup(struct vcpu *vcpu, struct vm_guest_paging *paging,
482     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
483     int num_copyinfo, int *is_fault);
484 void vm_copy_teardown(struct vm_copyinfo *copyinfo, int num_copyinfo);
485 void vm_copyin(struct vm_copyinfo *copyinfo, void *kaddr, size_t len);
486 void vm_copyout(const void *kaddr, struct vm_copyinfo *copyinfo, size_t len);
487 
488 int vcpu_trace_exceptions(struct vcpu *vcpu);
489 int vcpu_trap_wbinvd(struct vcpu *vcpu);
490 #endif	/* KERNEL */
491 
492 /*
493  * Identifiers for optional vmm capabilities
494  */
495 enum vm_cap_type {
496 	VM_CAP_HALT_EXIT,
497 	VM_CAP_MTRAP_EXIT,
498 	VM_CAP_PAUSE_EXIT,
499 	VM_CAP_UNRESTRICTED_GUEST,
500 	VM_CAP_ENABLE_INVPCID,
501 	VM_CAP_BPT_EXIT,
502 	VM_CAP_RDPID,
503 	VM_CAP_RDTSCP,
504 	VM_CAP_IPI_EXIT,
505 	VM_CAP_MASK_HWINTR,
506 	VM_CAP_RFLAGS_TF,
507 	VM_CAP_MAX
508 };
509 
510 enum vm_intr_trigger {
511 	EDGE_TRIGGER,
512 	LEVEL_TRIGGER
513 };
514 
515 /*
516  * The 'access' field has the format specified in Table 21-2 of the Intel
517  * Architecture Manual vol 3b.
518  *
519  * XXX The contents of the 'access' field are architecturally defined except
520  * bit 16 - Segment Unusable.
521  */
522 struct seg_desc {
523 	uint64_t	base;
524 	uint32_t	limit;
525 	uint32_t	access;
526 };
527 #define	SEG_DESC_TYPE(access)		((access) & 0x001f)
528 #define	SEG_DESC_DPL(access)		(((access) >> 5) & 0x3)
529 #define	SEG_DESC_PRESENT(access)	(((access) & 0x0080) ? 1 : 0)
530 #define	SEG_DESC_DEF32(access)		(((access) & 0x4000) ? 1 : 0)
531 #define	SEG_DESC_GRANULARITY(access)	(((access) & 0x8000) ? 1 : 0)
532 #define	SEG_DESC_UNUSABLE(access)	(((access) & 0x10000) ? 1 : 0)
533 
534 enum vm_cpu_mode {
535 	CPU_MODE_REAL,
536 	CPU_MODE_PROTECTED,
537 	CPU_MODE_COMPATIBILITY,		/* IA-32E mode (CS.L = 0) */
538 	CPU_MODE_64BIT,			/* IA-32E mode (CS.L = 1) */
539 };
540 
541 enum vm_paging_mode {
542 	PAGING_MODE_FLAT,
543 	PAGING_MODE_32,
544 	PAGING_MODE_PAE,
545 	PAGING_MODE_64,
546 	PAGING_MODE_64_LA57,
547 };
548 
549 struct vm_guest_paging {
550 	uint64_t	cr3;
551 	int		cpl;
552 	enum vm_cpu_mode cpu_mode;
553 	enum vm_paging_mode paging_mode;
554 };
555 
556 /*
557  * The data structures 'vie' and 'vie_op' are meant to be opaque to the
558  * consumers of instruction decoding. The only reason why their contents
559  * need to be exposed is because they are part of the 'vm_exit' structure.
560  */
561 struct vie_op {
562 	uint8_t		op_byte;	/* actual opcode byte */
563 	uint8_t		op_type;	/* type of operation (e.g. MOV) */
564 	uint16_t	op_flags;
565 };
566 _Static_assert(sizeof(struct vie_op) == 4, "ABI");
567 _Static_assert(_Alignof(struct vie_op) == 2, "ABI");
568 
569 #define	VIE_INST_SIZE	15
570 struct vie {
571 	uint8_t		inst[VIE_INST_SIZE];	/* instruction bytes */
572 	uint8_t		num_valid;		/* size of the instruction */
573 
574 /* The following fields are all zeroed upon restart. */
575 #define	vie_startzero	num_processed
576 	uint8_t		num_processed;
577 
578 	uint8_t		addrsize:4, opsize:4;	/* address and operand sizes */
579 	uint8_t		rex_w:1,		/* REX prefix */
580 			rex_r:1,
581 			rex_x:1,
582 			rex_b:1,
583 			rex_present:1,
584 			repz_present:1,		/* REP/REPE/REPZ prefix */
585 			repnz_present:1,	/* REPNE/REPNZ prefix */
586 			opsize_override:1,	/* Operand size override */
587 			addrsize_override:1,	/* Address size override */
588 			segment_override:1;	/* Segment override */
589 
590 	uint8_t		mod:2,			/* ModRM byte */
591 			reg:4,
592 			rm:4;
593 
594 	uint8_t		ss:2,			/* SIB byte */
595 			vex_present:1,		/* VEX prefixed */
596 			vex_l:1,		/* L bit */
597 			index:4,		/* SIB byte */
598 			base:4;			/* SIB byte */
599 
600 	uint8_t		disp_bytes;
601 	uint8_t		imm_bytes;
602 
603 	uint8_t		scale;
604 
605 	uint8_t		vex_reg:4,		/* vvvv: first source register specifier */
606 			vex_pp:2,		/* pp */
607 			_sparebits:2;
608 
609 	uint8_t		_sparebytes[2];
610 
611 	int		base_register;		/* VM_REG_GUEST_xyz */
612 	int		index_register;		/* VM_REG_GUEST_xyz */
613 	int		segment_register;	/* VM_REG_GUEST_xyz */
614 
615 	int64_t		displacement;		/* optional addr displacement */
616 	int64_t		immediate;		/* optional immediate operand */
617 
618 	uint8_t		decoded;	/* set to 1 if successfully decoded */
619 
620 	uint8_t		_sparebyte;
621 
622 	struct vie_op	op;			/* opcode description */
623 };
624 _Static_assert(sizeof(struct vie) == 64, "ABI");
625 _Static_assert(__offsetof(struct vie, disp_bytes) == 22, "ABI");
626 _Static_assert(__offsetof(struct vie, scale) == 24, "ABI");
627 _Static_assert(__offsetof(struct vie, base_register) == 28, "ABI");
628 
629 enum vm_exitcode {
630 	VM_EXITCODE_INOUT,
631 	VM_EXITCODE_VMX,
632 	VM_EXITCODE_BOGUS,
633 	VM_EXITCODE_RDMSR,
634 	VM_EXITCODE_WRMSR,
635 	VM_EXITCODE_HLT,
636 	VM_EXITCODE_MTRAP,
637 	VM_EXITCODE_PAUSE,
638 	VM_EXITCODE_PAGING,
639 	VM_EXITCODE_INST_EMUL,
640 	VM_EXITCODE_SPINUP_AP,
641 	VM_EXITCODE_DEPRECATED1,	/* used to be SPINDOWN_CPU */
642 	VM_EXITCODE_RENDEZVOUS,
643 	VM_EXITCODE_IOAPIC_EOI,
644 	VM_EXITCODE_SUSPENDED,
645 	VM_EXITCODE_INOUT_STR,
646 	VM_EXITCODE_TASK_SWITCH,
647 	VM_EXITCODE_MONITOR,
648 	VM_EXITCODE_MWAIT,
649 	VM_EXITCODE_SVM,
650 	VM_EXITCODE_REQIDLE,
651 	VM_EXITCODE_DEBUG,
652 	VM_EXITCODE_VMINSN,
653 	VM_EXITCODE_BPT,
654 	VM_EXITCODE_IPI,
655 	VM_EXITCODE_DB,
656 	VM_EXITCODE_MAX
657 };
658 
659 struct vm_inout {
660 	uint16_t	bytes:3;	/* 1 or 2 or 4 */
661 	uint16_t	in:1;
662 	uint16_t	string:1;
663 	uint16_t	rep:1;
664 	uint16_t	port;
665 	uint32_t	eax;		/* valid for out */
666 };
667 
668 struct vm_inout_str {
669 	struct vm_inout	inout;		/* must be the first element */
670 	struct vm_guest_paging paging;
671 	uint64_t	rflags;
672 	uint64_t	cr0;
673 	uint64_t	index;
674 	uint64_t	count;		/* rep=1 (%rcx), rep=0 (1) */
675 	int		addrsize;
676 	enum vm_reg_name seg_name;
677 	struct seg_desc seg_desc;
678 };
679 
680 enum task_switch_reason {
681 	TSR_CALL,
682 	TSR_IRET,
683 	TSR_JMP,
684 	TSR_IDT_GATE,	/* task gate in IDT */
685 };
686 
687 struct vm_task_switch {
688 	uint16_t	tsssel;		/* new TSS selector */
689 	int		ext;		/* task switch due to external event */
690 	uint32_t	errcode;
691 	int		errcode_valid;	/* push 'errcode' on the new stack */
692 	enum task_switch_reason reason;
693 	struct vm_guest_paging paging;
694 };
695 
696 struct vm_exit {
697 	enum vm_exitcode	exitcode;
698 	int			inst_length;	/* 0 means unknown */
699 	uint64_t		rip;
700 	union {
701 		struct vm_inout	inout;
702 		struct vm_inout_str inout_str;
703 		struct {
704 			uint64_t	gpa;
705 			int		fault_type;
706 		} paging;
707 		struct {
708 			uint64_t	gpa;
709 			uint64_t	gla;
710 			uint64_t	cs_base;
711 			int		cs_d;		/* CS.D */
712 			struct vm_guest_paging paging;
713 			struct vie	vie;
714 		} inst_emul;
715 		/*
716 		 * VMX specific payload. Used when there is no "better"
717 		 * exitcode to represent the VM-exit.
718 		 */
719 		struct {
720 			int		status;		/* vmx inst status */
721 			/*
722 			 * 'exit_reason' and 'exit_qualification' are valid
723 			 * only if 'status' is zero.
724 			 */
725 			uint32_t	exit_reason;
726 			uint64_t	exit_qualification;
727 			/*
728 			 * 'inst_error' and 'inst_type' are valid
729 			 * only if 'status' is non-zero.
730 			 */
731 			int		inst_type;
732 			int		inst_error;
733 		} vmx;
734 		/*
735 		 * SVM specific payload.
736 		 */
737 		struct {
738 			uint64_t	exitcode;
739 			uint64_t	exitinfo1;
740 			uint64_t	exitinfo2;
741 		} svm;
742 		struct {
743 			int		inst_length;
744 		} bpt;
745 		struct {
746 			int		trace_trap;
747 			int		pushf_intercept;
748 			int		tf_shadow_val;
749 			struct		vm_guest_paging paging;
750 		} dbg;
751 		struct {
752 			uint32_t	code;		/* ecx value */
753 			uint64_t	wval;
754 		} msr;
755 		struct {
756 			int		vcpu;
757 			uint64_t	rip;
758 		} spinup_ap;
759 		struct {
760 			uint64_t	rflags;
761 			uint64_t	intr_status;
762 		} hlt;
763 		struct {
764 			int		vector;
765 		} ioapic_eoi;
766 		struct {
767 			enum vm_suspend_how how;
768 		} suspended;
769 		struct {
770 			/*
771 			 * The destination vCPU mask is saved in vcpu->cpuset
772 			 * and is copied out to userspace separately to avoid
773 			 * ABI concerns.
774 			 */
775 			uint32_t mode;
776 			uint8_t vector;
777 		} ipi;
778 		struct vm_task_switch task_switch;
779 	} u;
780 };
781 
782 /* APIs to inject faults into the guest */
783 void vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid,
784     int errcode);
785 
786 static __inline void
vm_inject_ud(struct vcpu * vcpu)787 vm_inject_ud(struct vcpu *vcpu)
788 {
789 	vm_inject_fault(vcpu, IDT_UD, 0, 0);
790 }
791 
792 static __inline void
vm_inject_gp(struct vcpu * vcpu)793 vm_inject_gp(struct vcpu *vcpu)
794 {
795 	vm_inject_fault(vcpu, IDT_GP, 1, 0);
796 }
797 
798 static __inline void
vm_inject_ac(struct vcpu * vcpu,int errcode)799 vm_inject_ac(struct vcpu *vcpu, int errcode)
800 {
801 	vm_inject_fault(vcpu, IDT_AC, 1, errcode);
802 }
803 
804 static __inline void
vm_inject_ss(struct vcpu * vcpu,int errcode)805 vm_inject_ss(struct vcpu *vcpu, int errcode)
806 {
807 	vm_inject_fault(vcpu, IDT_SS, 1, errcode);
808 }
809 
810 void vm_inject_pf(struct vcpu *vcpu, int error_code, uint64_t cr2);
811 
812 #endif	/* _VMM_H_ */
813