xref: /freebsd/sys/arm64/vmm/vmm.c (revision a97f683fe3c425b425cf8cc466319f54ea957c20)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (C) 2015 Mihai Carabas <mihai.carabas@gmail.com>
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 AUTHOR AND CONTRIBUTORS ``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 AUTHOR 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 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/cpuset.h>
32 #include <sys/kernel.h>
33 #include <sys/linker.h>
34 #include <sys/lock.h>
35 #include <sys/malloc.h>
36 #include <sys/module.h>
37 #include <sys/mutex.h>
38 #include <sys/pcpu.h>
39 #include <sys/proc.h>
40 #include <sys/queue.h>
41 #include <sys/rwlock.h>
42 #include <sys/sched.h>
43 #include <sys/smp.h>
44 #include <sys/sysctl.h>
45 
46 #include <vm/vm.h>
47 #include <vm/vm_object.h>
48 #include <vm/vm_page.h>
49 #include <vm/pmap.h>
50 #include <vm/vm_map.h>
51 #include <vm/vm_extern.h>
52 #include <vm/vm_param.h>
53 
54 #include <machine/armreg.h>
55 #include <machine/cpu.h>
56 #include <machine/fpu.h>
57 #include <machine/machdep.h>
58 #include <machine/pcb.h>
59 #include <machine/smp.h>
60 #include <machine/vm.h>
61 #include <machine/vmparam.h>
62 #include <machine/vmm.h>
63 #include <machine/vmm_instruction_emul.h>
64 
65 #include <dev/pci/pcireg.h>
66 #include <dev/vmm/vmm_dev.h>
67 #include <dev/vmm/vmm_ktr.h>
68 #include <dev/vmm/vmm_stat.h>
69 
70 #include "arm64.h"
71 #include "mmu.h"
72 
73 #include "io/vgic.h"
74 #include "io/vtimer.h"
75 
76 struct vcpu {
77 	int		flags;
78 	enum vcpu_state	state;
79 	struct mtx	mtx;
80 	int		hostcpu;	/* host cpuid this vcpu last ran on */
81 	int		vcpuid;
82 	void		*stats;
83 	struct vm_exit	exitinfo;
84 	uint64_t	nextpc;		/* (x) next instruction to execute */
85 	struct vm	*vm;		/* (o) */
86 	void		*cookie;	/* (i) cpu-specific data */
87 	struct vfpstate	*guestfpu;	/* (a,i) guest fpu state */
88 };
89 
90 #define	vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx))
91 #define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
92 #define	vcpu_lock_destroy(v)	mtx_destroy(&((v)->mtx))
93 #define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
94 #define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
95 #define	vcpu_assert_locked(v)	mtx_assert(&((v)->mtx), MA_OWNED)
96 
97 struct mem_seg {
98 	uint64_t	gpa;
99 	size_t		len;
100 	bool		wired;
101 	bool		sysmem;
102 	vm_object_t	object;
103 };
104 #define	VM_MAX_MEMSEGS	3
105 
106 struct mem_map {
107 	vm_paddr_t	gpa;
108 	size_t		len;
109 	vm_ooffset_t	segoff;
110 	int		segid;
111 	int		prot;
112 	int		flags;
113 };
114 #define	VM_MAX_MEMMAPS	4
115 
116 struct vmm_mmio_region {
117 	uint64_t start;
118 	uint64_t end;
119 	mem_region_read_t read;
120 	mem_region_write_t write;
121 };
122 #define	VM_MAX_MMIO_REGIONS	4
123 
124 struct vmm_special_reg {
125 	uint32_t	esr_iss;
126 	uint32_t	esr_mask;
127 	reg_read_t	reg_read;
128 	reg_write_t	reg_write;
129 	void		*arg;
130 };
131 #define	VM_MAX_SPECIAL_REGS	16
132 
133 /*
134  * Initialization:
135  * (o) initialized the first time the VM is created
136  * (i) initialized when VM is created and when it is reinitialized
137  * (x) initialized before use
138  */
139 struct vm {
140 	void		*cookie;		/* (i) cpu-specific data */
141 	volatile cpuset_t active_cpus;		/* (i) active vcpus */
142 	volatile cpuset_t debug_cpus;		/* (i) vcpus stopped for debug */
143 	int		suspend;		/* (i) stop VM execution */
144 	bool		dying;			/* (o) is dying */
145 	volatile cpuset_t suspended_cpus; 	/* (i) suspended vcpus */
146 	volatile cpuset_t halted_cpus;		/* (x) cpus in a hard halt */
147 	struct mem_map	mem_maps[VM_MAX_MEMMAPS]; /* (i) guest address space */
148 	struct mem_seg	mem_segs[VM_MAX_MEMSEGS]; /* (o) guest memory regions */
149 	struct vmspace	*vmspace;		/* (o) guest's address space */
150 	char		name[VM_MAX_NAMELEN];	/* (o) virtual machine name */
151 	struct vcpu	**vcpu;			/* (i) guest vcpus */
152 	struct vmm_mmio_region mmio_region[VM_MAX_MMIO_REGIONS];
153 						/* (o) guest MMIO regions */
154 	struct vmm_special_reg special_reg[VM_MAX_SPECIAL_REGS];
155 	/* The following describe the vm cpu topology */
156 	uint16_t	sockets;		/* (o) num of sockets */
157 	uint16_t	cores;			/* (o) num of cores/socket */
158 	uint16_t	threads;		/* (o) num of threads/core */
159 	uint16_t	maxcpus;		/* (o) max pluggable cpus */
160 	struct sx	mem_segs_lock;		/* (o) */
161 	struct sx	vcpus_init_lock;	/* (o) */
162 };
163 
164 static bool vmm_initialized = false;
165 
166 static int vm_handle_wfi(struct vcpu *vcpu,
167 			 struct vm_exit *vme, bool *retu);
168 
169 static MALLOC_DEFINE(M_VMM, "vmm", "vmm");
170 
171 /* statistics */
172 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
173 
174 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
175 
176 static int vmm_ipinum;
177 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
178     "IPI vector used for vcpu notifications");
179 
180 struct vmm_regs {
181 	uint64_t	id_aa64afr0;
182 	uint64_t	id_aa64afr1;
183 	uint64_t	id_aa64dfr0;
184 	uint64_t	id_aa64dfr1;
185 	uint64_t	id_aa64isar0;
186 	uint64_t	id_aa64isar1;
187 	uint64_t	id_aa64isar2;
188 	uint64_t	id_aa64mmfr0;
189 	uint64_t	id_aa64mmfr1;
190 	uint64_t	id_aa64mmfr2;
191 	uint64_t	id_aa64pfr0;
192 	uint64_t	id_aa64pfr1;
193 };
194 
195 static const struct vmm_regs vmm_arch_regs_masks = {
196 	.id_aa64dfr0 =
197 	    ID_AA64DFR0_CTX_CMPs_MASK |
198 	    ID_AA64DFR0_WRPs_MASK |
199 	    ID_AA64DFR0_BRPs_MASK |
200 	    ID_AA64DFR0_PMUVer_3 |
201 	    ID_AA64DFR0_DebugVer_8,
202 	.id_aa64isar0 =
203 	    ID_AA64ISAR0_TLB_TLBIOSR |
204 	    ID_AA64ISAR0_SHA3_IMPL |
205 	    ID_AA64ISAR0_RDM_IMPL |
206 	    ID_AA64ISAR0_Atomic_IMPL |
207 	    ID_AA64ISAR0_CRC32_BASE |
208 	    ID_AA64ISAR0_SHA2_512 |
209 	    ID_AA64ISAR0_SHA1_BASE |
210 	    ID_AA64ISAR0_AES_PMULL,
211 	.id_aa64mmfr0 =
212 	    ID_AA64MMFR0_TGran4_IMPL |
213 	    ID_AA64MMFR0_TGran64_IMPL |
214 	    ID_AA64MMFR0_TGran16_IMPL |
215 	    ID_AA64MMFR0_ASIDBits_16 |
216 	    ID_AA64MMFR0_PARange_4P,
217 	.id_aa64mmfr1 =
218 	    ID_AA64MMFR1_SpecSEI_IMPL |
219 	    ID_AA64MMFR1_PAN_ATS1E1 |
220 	    ID_AA64MMFR1_HAFDBS_AF,
221 	.id_aa64pfr0 =
222 	    ID_AA64PFR0_GIC_CPUIF_NONE |
223 	    ID_AA64PFR0_AdvSIMD_HP |
224 	    ID_AA64PFR0_FP_HP |
225 	    ID_AA64PFR0_EL3_64 |
226 	    ID_AA64PFR0_EL2_64 |
227 	    ID_AA64PFR0_EL1_64 |
228 	    ID_AA64PFR0_EL0_64,
229 };
230 
231 /* Host registers masked by vmm_arch_regs_masks. */
232 static struct vmm_regs vmm_arch_regs;
233 
234 u_int vm_maxcpu;
235 SYSCTL_UINT(_hw_vmm, OID_AUTO, maxcpu, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
236     &vm_maxcpu, 0, "Maximum number of vCPUs");
237 
238 static void vm_free_memmap(struct vm *vm, int ident);
239 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm);
240 static void vcpu_notify_event_locked(struct vcpu *vcpu);
241 
242 /* global statistics */
243 VMM_STAT(VMEXIT_COUNT, "total number of vm exits");
244 VMM_STAT(VMEXIT_UNKNOWN, "number of vmexits for the unknown exception");
245 VMM_STAT(VMEXIT_WFI, "number of times wfi was intercepted");
246 VMM_STAT(VMEXIT_WFE, "number of times wfe was intercepted");
247 VMM_STAT(VMEXIT_HVC, "number of times hvc was intercepted");
248 VMM_STAT(VMEXIT_MSR, "number of times msr/mrs was intercepted");
249 VMM_STAT(VMEXIT_DATA_ABORT, "number of vmexits for a data abort");
250 VMM_STAT(VMEXIT_INSN_ABORT, "number of vmexits for an instruction abort");
251 VMM_STAT(VMEXIT_UNHANDLED_SYNC, "number of vmexits for an unhandled synchronous exception");
252 VMM_STAT(VMEXIT_IRQ, "number of vmexits for an irq");
253 VMM_STAT(VMEXIT_FIQ, "number of vmexits for an interrupt");
254 VMM_STAT(VMEXIT_BRK, "number of vmexits for a breakpoint exception");
255 VMM_STAT(VMEXIT_SS, "number of vmexits for a single-step exception");
256 VMM_STAT(VMEXIT_UNHANDLED_EL2, "number of vmexits for an unhandled EL2 exception");
257 VMM_STAT(VMEXIT_UNHANDLED, "number of vmexits for an unhandled exception");
258 
259 /*
260  * Upper limit on vm_maxcpu. We could increase this to 28 bits, but this
261  * is a safe value for now.
262  */
263 #define	VM_MAXCPU	MIN(0xffff - 1, CPU_SETSIZE)
264 
265 static int
266 vmm_regs_init(struct vmm_regs *regs, const struct vmm_regs *masks)
267 {
268 #define	_FETCH_KERN_REG(reg, field) do {				\
269 	regs->field = vmm_arch_regs_masks.field;			\
270 	if (!get_kernel_reg_masked(reg, &regs->field, masks->field))	\
271 		regs->field = 0;					\
272 } while (0)
273 	_FETCH_KERN_REG(ID_AA64AFR0_EL1, id_aa64afr0);
274 	_FETCH_KERN_REG(ID_AA64AFR1_EL1, id_aa64afr1);
275 	_FETCH_KERN_REG(ID_AA64DFR0_EL1, id_aa64dfr0);
276 	_FETCH_KERN_REG(ID_AA64DFR1_EL1, id_aa64dfr1);
277 	_FETCH_KERN_REG(ID_AA64ISAR0_EL1, id_aa64isar0);
278 	_FETCH_KERN_REG(ID_AA64ISAR1_EL1, id_aa64isar1);
279 	_FETCH_KERN_REG(ID_AA64ISAR2_EL1, id_aa64isar2);
280 	_FETCH_KERN_REG(ID_AA64MMFR0_EL1, id_aa64mmfr0);
281 	_FETCH_KERN_REG(ID_AA64MMFR1_EL1, id_aa64mmfr1);
282 	_FETCH_KERN_REG(ID_AA64MMFR2_EL1, id_aa64mmfr2);
283 	_FETCH_KERN_REG(ID_AA64PFR0_EL1, id_aa64pfr0);
284 	_FETCH_KERN_REG(ID_AA64PFR1_EL1, id_aa64pfr1);
285 #undef _FETCH_KERN_REG
286 	return (0);
287 }
288 
289 static void
290 vcpu_cleanup(struct vcpu *vcpu, bool destroy)
291 {
292 	vmmops_vcpu_cleanup(vcpu->cookie);
293 	vcpu->cookie = NULL;
294 	if (destroy) {
295 		vmm_stat_free(vcpu->stats);
296 		fpu_save_area_free(vcpu->guestfpu);
297 		vcpu_lock_destroy(vcpu);
298 	}
299 }
300 
301 static struct vcpu *
302 vcpu_alloc(struct vm *vm, int vcpu_id)
303 {
304 	struct vcpu *vcpu;
305 
306 	KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus,
307 	    ("vcpu_alloc: invalid vcpu %d", vcpu_id));
308 
309 	vcpu = malloc(sizeof(*vcpu), M_VMM, M_WAITOK | M_ZERO);
310 	vcpu_lock_init(vcpu);
311 	vcpu->state = VCPU_IDLE;
312 	vcpu->hostcpu = NOCPU;
313 	vcpu->vcpuid = vcpu_id;
314 	vcpu->vm = vm;
315 	vcpu->guestfpu = fpu_save_area_alloc();
316 	vcpu->stats = vmm_stat_alloc();
317 	return (vcpu);
318 }
319 
320 static void
321 vcpu_init(struct vcpu *vcpu)
322 {
323 	vcpu->cookie = vmmops_vcpu_init(vcpu->vm->cookie, vcpu, vcpu->vcpuid);
324 	MPASS(vcpu->cookie != NULL);
325 	fpu_save_area_reset(vcpu->guestfpu);
326 	vmm_stat_init(vcpu->stats);
327 }
328 
329 struct vm_exit *
330 vm_exitinfo(struct vcpu *vcpu)
331 {
332 	return (&vcpu->exitinfo);
333 }
334 
335 static int
336 vmm_init(void)
337 {
338 	int error;
339 
340 	vm_maxcpu = mp_ncpus;
341 	TUNABLE_INT_FETCH("hw.vmm.maxcpu", &vm_maxcpu);
342 
343 	if (vm_maxcpu > VM_MAXCPU) {
344 		printf("vmm: vm_maxcpu clamped to %u\n", VM_MAXCPU);
345 		vm_maxcpu = VM_MAXCPU;
346 	}
347 	if (vm_maxcpu == 0)
348 		vm_maxcpu = 1;
349 
350 	error = vmm_regs_init(&vmm_arch_regs, &vmm_arch_regs_masks);
351 	if (error != 0)
352 		return (error);
353 
354 	return (vmmops_modinit(0));
355 }
356 
357 static int
358 vmm_handler(module_t mod, int what, void *arg)
359 {
360 	int error;
361 
362 	switch (what) {
363 	case MOD_LOAD:
364 		/* TODO: if (vmm_is_hw_supported()) { */
365 		error = vmmdev_init();
366 		if (error != 0)
367 			break;
368 		error = vmm_init();
369 		if (error == 0)
370 			vmm_initialized = true;
371 		break;
372 	case MOD_UNLOAD:
373 		/* TODO: if (vmm_is_hw_supported()) { */
374 		error = vmmdev_cleanup();
375 		if (error == 0 && vmm_initialized) {
376 			error = vmmops_modcleanup();
377 			if (error)
378 				vmm_initialized = false;
379 		}
380 		break;
381 	default:
382 		error = 0;
383 		break;
384 	}
385 	return (error);
386 }
387 
388 static moduledata_t vmm_kmod = {
389 	"vmm",
390 	vmm_handler,
391 	NULL
392 };
393 
394 /*
395  * vmm initialization has the following dependencies:
396  *
397  * - HYP initialization requires smp_rendezvous() and therefore must happen
398  *   after SMP is fully functional (after SI_SUB_SMP).
399  */
400 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
401 MODULE_VERSION(vmm, 1);
402 
403 static void
404 vm_init(struct vm *vm, bool create)
405 {
406 	int i;
407 
408 	vm->cookie = vmmops_init(vm, vmspace_pmap(vm->vmspace));
409 	MPASS(vm->cookie != NULL);
410 
411 	CPU_ZERO(&vm->active_cpus);
412 	CPU_ZERO(&vm->debug_cpus);
413 
414 	vm->suspend = 0;
415 	CPU_ZERO(&vm->suspended_cpus);
416 
417 	memset(vm->mmio_region, 0, sizeof(vm->mmio_region));
418 	memset(vm->special_reg, 0, sizeof(vm->special_reg));
419 
420 	if (!create) {
421 		for (i = 0; i < vm->maxcpus; i++) {
422 			if (vm->vcpu[i] != NULL)
423 				vcpu_init(vm->vcpu[i]);
424 		}
425 	}
426 }
427 
428 void
429 vm_disable_vcpu_creation(struct vm *vm)
430 {
431 	sx_xlock(&vm->vcpus_init_lock);
432 	vm->dying = true;
433 	sx_xunlock(&vm->vcpus_init_lock);
434 }
435 
436 struct vcpu *
437 vm_alloc_vcpu(struct vm *vm, int vcpuid)
438 {
439 	struct vcpu *vcpu;
440 
441 	if (vcpuid < 0 || vcpuid >= vm_get_maxcpus(vm))
442 		return (NULL);
443 
444 	/* Some interrupt controllers may have a CPU limit */
445 	if (vcpuid >= vgic_max_cpu_count(vm->cookie))
446 		return (NULL);
447 
448 	vcpu = (struct vcpu *)
449 	    atomic_load_acq_ptr((uintptr_t *)&vm->vcpu[vcpuid]);
450 	if (__predict_true(vcpu != NULL))
451 		return (vcpu);
452 
453 	sx_xlock(&vm->vcpus_init_lock);
454 	vcpu = vm->vcpu[vcpuid];
455 	if (vcpu == NULL && !vm->dying) {
456 		vcpu = vcpu_alloc(vm, vcpuid);
457 		vcpu_init(vcpu);
458 
459 		/*
460 		 * Ensure vCPU is fully created before updating pointer
461 		 * to permit unlocked reads above.
462 		 */
463 		atomic_store_rel_ptr((uintptr_t *)&vm->vcpu[vcpuid],
464 		    (uintptr_t)vcpu);
465 	}
466 	sx_xunlock(&vm->vcpus_init_lock);
467 	return (vcpu);
468 }
469 
470 void
471 vm_slock_vcpus(struct vm *vm)
472 {
473 	sx_slock(&vm->vcpus_init_lock);
474 }
475 
476 void
477 vm_unlock_vcpus(struct vm *vm)
478 {
479 	sx_unlock(&vm->vcpus_init_lock);
480 }
481 
482 int
483 vm_create(const char *name, struct vm **retvm)
484 {
485 	struct vm *vm;
486 	struct vmspace *vmspace;
487 
488 	/*
489 	 * If vmm.ko could not be successfully initialized then don't attempt
490 	 * to create the virtual machine.
491 	 */
492 	if (!vmm_initialized)
493 		return (ENXIO);
494 
495 	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
496 		return (EINVAL);
497 
498 	vmspace = vmmops_vmspace_alloc(0, 1ul << 39);
499 	if (vmspace == NULL)
500 		return (ENOMEM);
501 
502 	vm = malloc(sizeof(struct vm), M_VMM, M_WAITOK | M_ZERO);
503 	strcpy(vm->name, name);
504 	vm->vmspace = vmspace;
505 	sx_init(&vm->mem_segs_lock, "vm mem_segs");
506 	sx_init(&vm->vcpus_init_lock, "vm vcpus");
507 
508 	vm->sockets = 1;
509 	vm->cores = 1;			/* XXX backwards compatibility */
510 	vm->threads = 1;		/* XXX backwards compatibility */
511 	vm->maxcpus = vm_maxcpu;
512 
513 	vm->vcpu = malloc(sizeof(*vm->vcpu) * vm->maxcpus, M_VMM,
514 	    M_WAITOK | M_ZERO);
515 
516 	vm_init(vm, true);
517 
518 	*retvm = vm;
519 	return (0);
520 }
521 
522 void
523 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores,
524     uint16_t *threads, uint16_t *maxcpus)
525 {
526 	*sockets = vm->sockets;
527 	*cores = vm->cores;
528 	*threads = vm->threads;
529 	*maxcpus = vm->maxcpus;
530 }
531 
532 uint16_t
533 vm_get_maxcpus(struct vm *vm)
534 {
535 	return (vm->maxcpus);
536 }
537 
538 int
539 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores,
540     uint16_t threads, uint16_t maxcpus)
541 {
542 	/* Ignore maxcpus. */
543 	if ((sockets * cores * threads) > vm->maxcpus)
544 		return (EINVAL);
545 	vm->sockets = sockets;
546 	vm->cores = cores;
547 	vm->threads = threads;
548 	return(0);
549 }
550 
551 static void
552 vm_cleanup(struct vm *vm, bool destroy)
553 {
554 	struct mem_map *mm;
555 	pmap_t pmap __diagused;
556 	int i;
557 
558 	if (destroy) {
559 		pmap = vmspace_pmap(vm->vmspace);
560 		sched_pin();
561 		PCPU_SET(curvmpmap, NULL);
562 		sched_unpin();
563 		CPU_FOREACH(i) {
564 			MPASS(cpuid_to_pcpu[i]->pc_curvmpmap != pmap);
565 		}
566 	}
567 
568 	vgic_detach_from_vm(vm->cookie);
569 
570 	for (i = 0; i < vm->maxcpus; i++) {
571 		if (vm->vcpu[i] != NULL)
572 			vcpu_cleanup(vm->vcpu[i], destroy);
573 	}
574 
575 	vmmops_cleanup(vm->cookie);
576 
577 	/*
578 	 * System memory is removed from the guest address space only when
579 	 * the VM is destroyed. This is because the mapping remains the same
580 	 * across VM reset.
581 	 *
582 	 * Device memory can be relocated by the guest (e.g. using PCI BARs)
583 	 * so those mappings are removed on a VM reset.
584 	 */
585 	if (!destroy) {
586 		for (i = 0; i < VM_MAX_MEMMAPS; i++) {
587 			mm = &vm->mem_maps[i];
588 			if (destroy || !sysmem_mapping(vm, mm))
589 				vm_free_memmap(vm, i);
590 		}
591 	}
592 
593 	if (destroy) {
594 		for (i = 0; i < VM_MAX_MEMSEGS; i++)
595 			vm_free_memseg(vm, i);
596 
597 		vmmops_vmspace_free(vm->vmspace);
598 		vm->vmspace = NULL;
599 
600 		for (i = 0; i < vm->maxcpus; i++)
601 			free(vm->vcpu[i], M_VMM);
602 		free(vm->vcpu, M_VMM);
603 		sx_destroy(&vm->vcpus_init_lock);
604 		sx_destroy(&vm->mem_segs_lock);
605 	}
606 }
607 
608 void
609 vm_destroy(struct vm *vm)
610 {
611 	vm_cleanup(vm, true);
612 	free(vm, M_VMM);
613 }
614 
615 int
616 vm_reinit(struct vm *vm)
617 {
618 	int error;
619 
620 	/*
621 	 * A virtual machine can be reset only if all vcpus are suspended.
622 	 */
623 	if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
624 		vm_cleanup(vm, false);
625 		vm_init(vm, false);
626 		error = 0;
627 	} else {
628 		error = EBUSY;
629 	}
630 
631 	return (error);
632 }
633 
634 const char *
635 vm_name(struct vm *vm)
636 {
637 	return (vm->name);
638 }
639 
640 void
641 vm_slock_memsegs(struct vm *vm)
642 {
643 	sx_slock(&vm->mem_segs_lock);
644 }
645 
646 void
647 vm_xlock_memsegs(struct vm *vm)
648 {
649 	sx_xlock(&vm->mem_segs_lock);
650 }
651 
652 void
653 vm_unlock_memsegs(struct vm *vm)
654 {
655 	sx_unlock(&vm->mem_segs_lock);
656 }
657 
658 /*
659  * Return 'true' if 'gpa' is allocated in the guest address space.
660  *
661  * This function is called in the context of a running vcpu which acts as
662  * an implicit lock on 'vm->mem_maps[]'.
663  */
664 bool
665 vm_mem_allocated(struct vcpu *vcpu, vm_paddr_t gpa)
666 {
667 	struct vm *vm = vcpu->vm;
668 	struct mem_map *mm;
669 	int i;
670 
671 #ifdef INVARIANTS
672 	int hostcpu, state;
673 	state = vcpu_get_state(vcpu, &hostcpu);
674 	KASSERT(state == VCPU_RUNNING && hostcpu == curcpu,
675 	    ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu));
676 #endif
677 
678 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
679 		mm = &vm->mem_maps[i];
680 		if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len)
681 			return (true);		/* 'gpa' is sysmem or devmem */
682 	}
683 
684 	return (false);
685 }
686 
687 int
688 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem)
689 {
690 	struct mem_seg *seg;
691 	vm_object_t obj;
692 
693 	sx_assert(&vm->mem_segs_lock, SX_XLOCKED);
694 
695 	if (ident < 0 || ident >= VM_MAX_MEMSEGS)
696 		return (EINVAL);
697 
698 	if (len == 0 || (len & PAGE_MASK))
699 		return (EINVAL);
700 
701 	seg = &vm->mem_segs[ident];
702 	if (seg->object != NULL) {
703 		if (seg->len == len && seg->sysmem == sysmem)
704 			return (EEXIST);
705 		else
706 			return (EINVAL);
707 	}
708 
709 	obj = vm_object_allocate(OBJT_DEFAULT, len >> PAGE_SHIFT);
710 	if (obj == NULL)
711 		return (ENOMEM);
712 
713 	seg->len = len;
714 	seg->object = obj;
715 	seg->sysmem = sysmem;
716 	return (0);
717 }
718 
719 int
720 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
721     vm_object_t *objptr)
722 {
723 	struct mem_seg *seg;
724 
725 	sx_assert(&vm->mem_segs_lock, SX_LOCKED);
726 
727 	if (ident < 0 || ident >= VM_MAX_MEMSEGS)
728 		return (EINVAL);
729 
730 	seg = &vm->mem_segs[ident];
731 	if (len)
732 		*len = seg->len;
733 	if (sysmem)
734 		*sysmem = seg->sysmem;
735 	if (objptr)
736 		*objptr = seg->object;
737 	return (0);
738 }
739 
740 void
741 vm_free_memseg(struct vm *vm, int ident)
742 {
743 	struct mem_seg *seg;
744 
745 	KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS,
746 	    ("%s: invalid memseg ident %d", __func__, ident));
747 
748 	seg = &vm->mem_segs[ident];
749 	if (seg->object != NULL) {
750 		vm_object_deallocate(seg->object);
751 		bzero(seg, sizeof(struct mem_seg));
752 	}
753 }
754 
755 int
756 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first,
757     size_t len, int prot, int flags)
758 {
759 	struct mem_seg *seg;
760 	struct mem_map *m, *map;
761 	vm_ooffset_t last;
762 	int i, error;
763 
764 	if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0)
765 		return (EINVAL);
766 
767 	if (flags & ~VM_MEMMAP_F_WIRED)
768 		return (EINVAL);
769 
770 	if (segid < 0 || segid >= VM_MAX_MEMSEGS)
771 		return (EINVAL);
772 
773 	seg = &vm->mem_segs[segid];
774 	if (seg->object == NULL)
775 		return (EINVAL);
776 
777 	last = first + len;
778 	if (first < 0 || first >= last || last > seg->len)
779 		return (EINVAL);
780 
781 	if ((gpa | first | last) & PAGE_MASK)
782 		return (EINVAL);
783 
784 	map = NULL;
785 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
786 		m = &vm->mem_maps[i];
787 		if (m->len == 0) {
788 			map = m;
789 			break;
790 		}
791 	}
792 
793 	if (map == NULL)
794 		return (ENOSPC);
795 
796 	error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa,
797 	    len, 0, VMFS_NO_SPACE, prot, prot, 0);
798 	if (error != KERN_SUCCESS)
799 		return (EFAULT);
800 
801 	vm_object_reference(seg->object);
802 
803 	if (flags & VM_MEMMAP_F_WIRED) {
804 		error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len,
805 		    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
806 		if (error != KERN_SUCCESS) {
807 			vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len);
808 			return (error == KERN_RESOURCE_SHORTAGE ? ENOMEM :
809 			    EFAULT);
810 		}
811 	}
812 
813 	map->gpa = gpa;
814 	map->len = len;
815 	map->segoff = first;
816 	map->segid = segid;
817 	map->prot = prot;
818 	map->flags = flags;
819 	return (0);
820 }
821 
822 int
823 vm_munmap_memseg(struct vm *vm, vm_paddr_t gpa, size_t len)
824 {
825 	struct mem_map *m;
826 	int i;
827 
828 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
829 		m = &vm->mem_maps[i];
830 		if (m->gpa == gpa && m->len == len) {
831 			vm_free_memmap(vm, i);
832 			return (0);
833 		}
834 	}
835 
836 	return (EINVAL);
837 }
838 
839 int
840 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
841     vm_ooffset_t *segoff, size_t *len, int *prot, int *flags)
842 {
843 	struct mem_map *mm, *mmnext;
844 	int i;
845 
846 	mmnext = NULL;
847 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
848 		mm = &vm->mem_maps[i];
849 		if (mm->len == 0 || mm->gpa < *gpa)
850 			continue;
851 		if (mmnext == NULL || mm->gpa < mmnext->gpa)
852 			mmnext = mm;
853 	}
854 
855 	if (mmnext != NULL) {
856 		*gpa = mmnext->gpa;
857 		if (segid)
858 			*segid = mmnext->segid;
859 		if (segoff)
860 			*segoff = mmnext->segoff;
861 		if (len)
862 			*len = mmnext->len;
863 		if (prot)
864 			*prot = mmnext->prot;
865 		if (flags)
866 			*flags = mmnext->flags;
867 		return (0);
868 	} else {
869 		return (ENOENT);
870 	}
871 }
872 
873 static void
874 vm_free_memmap(struct vm *vm, int ident)
875 {
876 	struct mem_map *mm;
877 	int error __diagused;
878 
879 	mm = &vm->mem_maps[ident];
880 	if (mm->len) {
881 		error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa,
882 		    mm->gpa + mm->len);
883 		KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d",
884 		    __func__, error));
885 		bzero(mm, sizeof(struct mem_map));
886 	}
887 }
888 
889 static __inline bool
890 sysmem_mapping(struct vm *vm, struct mem_map *mm)
891 {
892 
893 	if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem)
894 		return (true);
895 	else
896 		return (false);
897 }
898 
899 vm_paddr_t
900 vmm_sysmem_maxaddr(struct vm *vm)
901 {
902 	struct mem_map *mm;
903 	vm_paddr_t maxaddr;
904 	int i;
905 
906 	maxaddr = 0;
907 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
908 		mm = &vm->mem_maps[i];
909 		if (sysmem_mapping(vm, mm)) {
910 			if (maxaddr < mm->gpa + mm->len)
911 				maxaddr = mm->gpa + mm->len;
912 		}
913 	}
914 	return (maxaddr);
915 }
916 
917 int
918 vm_gla2gpa_nofault(struct vcpu *vcpu, struct vm_guest_paging *paging,
919     uint64_t gla, int prot, uint64_t *gpa, int *is_fault)
920 {
921 
922 	vmmops_gla2gpa(vcpu->cookie, paging, gla, prot, gpa, is_fault);
923 	return (0);
924 }
925 
926 static int
927 vmm_reg_raz(struct vcpu *vcpu, uint64_t *rval, void *arg)
928 {
929 	*rval = 0;
930 	return (0);
931 }
932 
933 static int
934 vmm_reg_read_arg(struct vcpu *vcpu, uint64_t *rval, void *arg)
935 {
936 	*rval = *(uint64_t *)arg;
937 	return (0);
938 }
939 
940 static int
941 vmm_reg_wi(struct vcpu *vcpu, uint64_t wval, void *arg)
942 {
943 	return (0);
944 }
945 
946 static const struct vmm_special_reg vmm_special_regs[] = {
947 #define	SPECIAL_REG(_reg, _read, _write)				\
948 	{								\
949 		.esr_iss = ((_reg ## _op0) << ISS_MSR_OP0_SHIFT) |	\
950 		    ((_reg ## _op1) << ISS_MSR_OP1_SHIFT) |		\
951 		    ((_reg ## _CRn) << ISS_MSR_CRn_SHIFT) |		\
952 		    ((_reg ## _CRm) << ISS_MSR_CRm_SHIFT) |		\
953 		    ((_reg ## _op2) << ISS_MSR_OP2_SHIFT),		\
954 		.esr_mask = ISS_MSR_REG_MASK,				\
955 		.reg_read = (_read),					\
956 		.reg_write = (_write),					\
957 		.arg = NULL,						\
958 	}
959 #define	ID_SPECIAL_REG(_reg, _name)					\
960 	{								\
961 		.esr_iss = ((_reg ## _op0) << ISS_MSR_OP0_SHIFT) |	\
962 		    ((_reg ## _op1) << ISS_MSR_OP1_SHIFT) |		\
963 		    ((_reg ## _CRn) << ISS_MSR_CRn_SHIFT) |		\
964 		    ((_reg ## _CRm) << ISS_MSR_CRm_SHIFT) |		\
965 		    ((_reg ## _op2) << ISS_MSR_OP2_SHIFT),		\
966 		.esr_mask = ISS_MSR_REG_MASK,				\
967 		.reg_read = vmm_reg_read_arg,				\
968 		.reg_write = vmm_reg_wi,				\
969 		.arg = &(vmm_arch_regs._name),				\
970 	}
971 
972 	/* ID registers */
973 	ID_SPECIAL_REG(ID_AA64PFR0_EL1, id_aa64pfr0),
974 	ID_SPECIAL_REG(ID_AA64DFR0_EL1, id_aa64dfr0),
975 	ID_SPECIAL_REG(ID_AA64ISAR0_EL1, id_aa64isar0),
976 	ID_SPECIAL_REG(ID_AA64MMFR0_EL1, id_aa64mmfr0),
977 	ID_SPECIAL_REG(ID_AA64MMFR1_EL1, id_aa64mmfr1),
978 
979 	/*
980 	 * All other ID registers are read as zero.
981 	 * They are all in the op0=3, op1=0, CRn=0, CRm={0..7} space.
982 	 */
983 	{
984 		.esr_iss = (3 << ISS_MSR_OP0_SHIFT) |
985 		    (0 << ISS_MSR_OP1_SHIFT) |
986 		    (0 << ISS_MSR_CRn_SHIFT) |
987 		    (0 << ISS_MSR_CRm_SHIFT),
988 		.esr_mask = ISS_MSR_OP0_MASK | ISS_MSR_OP1_MASK |
989 		    ISS_MSR_CRn_MASK | (0x8 << ISS_MSR_CRm_SHIFT),
990 		.reg_read = vmm_reg_raz,
991 		.reg_write = vmm_reg_wi,
992 		.arg = NULL,
993 	},
994 
995 	/* Counter physical registers */
996 	SPECIAL_REG(CNTP_CTL_EL0, vtimer_phys_ctl_read, vtimer_phys_ctl_write),
997 	SPECIAL_REG(CNTP_CVAL_EL0, vtimer_phys_cval_read,
998 	    vtimer_phys_cval_write),
999 	SPECIAL_REG(CNTP_TVAL_EL0, vtimer_phys_tval_read,
1000 	    vtimer_phys_tval_write),
1001 	SPECIAL_REG(CNTPCT_EL0, vtimer_phys_cnt_read, vtimer_phys_cnt_write),
1002 #undef SPECIAL_REG
1003 };
1004 
1005 void
1006 vm_register_reg_handler(struct vm *vm, uint64_t iss, uint64_t mask,
1007     reg_read_t reg_read, reg_write_t reg_write, void *arg)
1008 {
1009 	int i;
1010 
1011 	for (i = 0; i < nitems(vm->special_reg); i++) {
1012 		if (vm->special_reg[i].esr_iss == 0 &&
1013 		    vm->special_reg[i].esr_mask == 0) {
1014 			vm->special_reg[i].esr_iss = iss;
1015 			vm->special_reg[i].esr_mask = mask;
1016 			vm->special_reg[i].reg_read = reg_read;
1017 			vm->special_reg[i].reg_write = reg_write;
1018 			vm->special_reg[i].arg = arg;
1019 			return;
1020 		}
1021 	}
1022 
1023 	panic("%s: No free special register slot", __func__);
1024 }
1025 
1026 void
1027 vm_deregister_reg_handler(struct vm *vm, uint64_t iss, uint64_t mask)
1028 {
1029 	int i;
1030 
1031 	for (i = 0; i < nitems(vm->special_reg); i++) {
1032 		if (vm->special_reg[i].esr_iss == iss &&
1033 		    vm->special_reg[i].esr_mask == mask) {
1034 			memset(&vm->special_reg[i], 0,
1035 			    sizeof(vm->special_reg[i]));
1036 			return;
1037 		}
1038 	}
1039 
1040 	panic("%s: Invalid special register: iss %lx mask %lx", __func__, iss,
1041 	    mask);
1042 }
1043 
1044 static int
1045 vm_handle_reg_emul(struct vcpu *vcpu, bool *retu)
1046 {
1047 	struct vm *vm;
1048 	struct vm_exit *vme;
1049 	struct vre *vre;
1050 	int i, rv;
1051 
1052 	vm = vcpu->vm;
1053 	vme = &vcpu->exitinfo;
1054 	vre = &vme->u.reg_emul.vre;
1055 
1056 	for (i = 0; i < nitems(vm->special_reg); i++) {
1057 		if (vm->special_reg[i].esr_iss == 0 &&
1058 		    vm->special_reg[i].esr_mask == 0)
1059 			continue;
1060 
1061 		if ((vre->inst_syndrome & vm->special_reg[i].esr_mask) ==
1062 		    vm->special_reg[i].esr_iss) {
1063 			rv = vmm_emulate_register(vcpu, vre,
1064 			    vm->special_reg[i].reg_read,
1065 			    vm->special_reg[i].reg_write,
1066 			    vm->special_reg[i].arg);
1067 			if (rv == 0) {
1068 				*retu = false;
1069 			}
1070 			return (rv);
1071 		}
1072 	}
1073 	for (i = 0; i < nitems(vmm_special_regs); i++) {
1074 		if ((vre->inst_syndrome & vmm_special_regs[i].esr_mask) ==
1075 		    vmm_special_regs[i].esr_iss) {
1076 			rv = vmm_emulate_register(vcpu, vre,
1077 			    vmm_special_regs[i].reg_read,
1078 			    vmm_special_regs[i].reg_write,
1079 			    vmm_special_regs[i].arg);
1080 			if (rv == 0) {
1081 				*retu = false;
1082 			}
1083 			return (rv);
1084 		}
1085 	}
1086 
1087 
1088 	*retu = true;
1089 	return (0);
1090 }
1091 
1092 void
1093 vm_register_inst_handler(struct vm *vm, uint64_t start, uint64_t size,
1094     mem_region_read_t mmio_read, mem_region_write_t mmio_write)
1095 {
1096 	int i;
1097 
1098 	for (i = 0; i < nitems(vm->mmio_region); i++) {
1099 		if (vm->mmio_region[i].start == 0 &&
1100 		    vm->mmio_region[i].end == 0) {
1101 			vm->mmio_region[i].start = start;
1102 			vm->mmio_region[i].end = start + size;
1103 			vm->mmio_region[i].read = mmio_read;
1104 			vm->mmio_region[i].write = mmio_write;
1105 			return;
1106 		}
1107 	}
1108 
1109 	panic("%s: No free MMIO region", __func__);
1110 }
1111 
1112 void
1113 vm_deregister_inst_handler(struct vm *vm, uint64_t start, uint64_t size)
1114 {
1115 	int i;
1116 
1117 	for (i = 0; i < nitems(vm->mmio_region); i++) {
1118 		if (vm->mmio_region[i].start == start &&
1119 		    vm->mmio_region[i].end == start + size) {
1120 			memset(&vm->mmio_region[i], 0,
1121 			    sizeof(vm->mmio_region[i]));
1122 			return;
1123 		}
1124 	}
1125 
1126 	panic("%s: Invalid MMIO region: %lx - %lx", __func__, start,
1127 	    start + size);
1128 }
1129 
1130 static int
1131 vm_handle_inst_emul(struct vcpu *vcpu, bool *retu)
1132 {
1133 	struct vm *vm;
1134 	struct vm_exit *vme;
1135 	struct vie *vie;
1136 	struct hyp *hyp;
1137 	uint64_t fault_ipa;
1138 	struct vm_guest_paging *paging;
1139 	struct vmm_mmio_region *vmr;
1140 	int error, i;
1141 
1142 	vm = vcpu->vm;
1143 	hyp = vm->cookie;
1144 	if (!hyp->vgic_attached)
1145 		goto out_user;
1146 
1147 	vme = &vcpu->exitinfo;
1148 	vie = &vme->u.inst_emul.vie;
1149 	paging = &vme->u.inst_emul.paging;
1150 
1151 	fault_ipa = vme->u.inst_emul.gpa;
1152 
1153 	vmr = NULL;
1154 	for (i = 0; i < nitems(vm->mmio_region); i++) {
1155 		if (vm->mmio_region[i].start <= fault_ipa &&
1156 		    vm->mmio_region[i].end > fault_ipa) {
1157 			vmr = &vm->mmio_region[i];
1158 			break;
1159 		}
1160 	}
1161 	if (vmr == NULL)
1162 		goto out_user;
1163 
1164 	error = vmm_emulate_instruction(vcpu, fault_ipa, vie, paging,
1165 	    vmr->read, vmr->write, retu);
1166 	return (error);
1167 
1168 out_user:
1169 	*retu = true;
1170 	return (0);
1171 }
1172 
1173 int
1174 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1175 {
1176 	int i;
1177 
1178 	if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1179 		return (EINVAL);
1180 
1181 	if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1182 		VM_CTR2(vm, "virtual machine already suspended %d/%d",
1183 		    vm->suspend, how);
1184 		return (EALREADY);
1185 	}
1186 
1187 	VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1188 
1189 	/*
1190 	 * Notify all active vcpus that they are now suspended.
1191 	 */
1192 	for (i = 0; i < vm->maxcpus; i++) {
1193 		if (CPU_ISSET(i, &vm->active_cpus))
1194 			vcpu_notify_event(vm_vcpu(vm, i));
1195 	}
1196 
1197 	return (0);
1198 }
1199 
1200 void
1201 vm_exit_suspended(struct vcpu *vcpu, uint64_t pc)
1202 {
1203 	struct vm *vm = vcpu->vm;
1204 	struct vm_exit *vmexit;
1205 
1206 	KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1207 	    ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1208 
1209 	vmexit = vm_exitinfo(vcpu);
1210 	vmexit->pc = pc;
1211 	vmexit->inst_length = 4;
1212 	vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1213 	vmexit->u.suspended.how = vm->suspend;
1214 }
1215 
1216 void
1217 vm_exit_debug(struct vcpu *vcpu, uint64_t pc)
1218 {
1219 	struct vm_exit *vmexit;
1220 
1221 	vmexit = vm_exitinfo(vcpu);
1222 	vmexit->pc = pc;
1223 	vmexit->inst_length = 4;
1224 	vmexit->exitcode = VM_EXITCODE_DEBUG;
1225 }
1226 
1227 int
1228 vm_activate_cpu(struct vcpu *vcpu)
1229 {
1230 	struct vm *vm = vcpu->vm;
1231 
1232 	if (CPU_ISSET(vcpu->vcpuid, &vm->active_cpus))
1233 		return (EBUSY);
1234 
1235 	CPU_SET_ATOMIC(vcpu->vcpuid, &vm->active_cpus);
1236 	return (0);
1237 
1238 }
1239 
1240 int
1241 vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu)
1242 {
1243 	if (vcpu == NULL) {
1244 		vm->debug_cpus = vm->active_cpus;
1245 		for (int i = 0; i < vm->maxcpus; i++) {
1246 			if (CPU_ISSET(i, &vm->active_cpus))
1247 				vcpu_notify_event(vm_vcpu(vm, i));
1248 		}
1249 	} else {
1250 		if (!CPU_ISSET(vcpu->vcpuid, &vm->active_cpus))
1251 			return (EINVAL);
1252 
1253 		CPU_SET_ATOMIC(vcpu->vcpuid, &vm->debug_cpus);
1254 		vcpu_notify_event(vcpu);
1255 	}
1256 	return (0);
1257 }
1258 
1259 int
1260 vm_resume_cpu(struct vm *vm, struct vcpu *vcpu)
1261 {
1262 
1263 	if (vcpu == NULL) {
1264 		CPU_ZERO(&vm->debug_cpus);
1265 	} else {
1266 		if (!CPU_ISSET(vcpu->vcpuid, &vm->debug_cpus))
1267 			return (EINVAL);
1268 
1269 		CPU_CLR_ATOMIC(vcpu->vcpuid, &vm->debug_cpus);
1270 	}
1271 	return (0);
1272 }
1273 
1274 int
1275 vcpu_debugged(struct vcpu *vcpu)
1276 {
1277 
1278 	return (CPU_ISSET(vcpu->vcpuid, &vcpu->vm->debug_cpus));
1279 }
1280 
1281 cpuset_t
1282 vm_active_cpus(struct vm *vm)
1283 {
1284 
1285 	return (vm->active_cpus);
1286 }
1287 
1288 cpuset_t
1289 vm_debug_cpus(struct vm *vm)
1290 {
1291 
1292 	return (vm->debug_cpus);
1293 }
1294 
1295 cpuset_t
1296 vm_suspended_cpus(struct vm *vm)
1297 {
1298 
1299 	return (vm->suspended_cpus);
1300 }
1301 
1302 
1303 void *
1304 vcpu_stats(struct vcpu *vcpu)
1305 {
1306 
1307 	return (vcpu->stats);
1308 }
1309 
1310 /*
1311  * This function is called to ensure that a vcpu "sees" a pending event
1312  * as soon as possible:
1313  * - If the vcpu thread is sleeping then it is woken up.
1314  * - If the vcpu is running on a different host_cpu then an IPI will be directed
1315  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
1316  */
1317 static void
1318 vcpu_notify_event_locked(struct vcpu *vcpu)
1319 {
1320 	int hostcpu;
1321 
1322 	hostcpu = vcpu->hostcpu;
1323 	if (vcpu->state == VCPU_RUNNING) {
1324 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
1325 		if (hostcpu != curcpu) {
1326 			ipi_cpu(hostcpu, vmm_ipinum);
1327 		} else {
1328 			/*
1329 			 * If the 'vcpu' is running on 'curcpu' then it must
1330 			 * be sending a notification to itself (e.g. SELF_IPI).
1331 			 * The pending event will be picked up when the vcpu
1332 			 * transitions back to guest context.
1333 			 */
1334 		}
1335 	} else {
1336 		KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
1337 		    "with hostcpu %d", vcpu->state, hostcpu));
1338 		if (vcpu->state == VCPU_SLEEPING)
1339 			wakeup_one(vcpu);
1340 	}
1341 }
1342 
1343 void
1344 vcpu_notify_event(struct vcpu *vcpu)
1345 {
1346 	vcpu_lock(vcpu);
1347 	vcpu_notify_event_locked(vcpu);
1348 	vcpu_unlock(vcpu);
1349 }
1350 
1351 static void
1352 restore_guest_fpustate(struct vcpu *vcpu)
1353 {
1354 
1355 	/* flush host state to the pcb */
1356 	vfp_save_state(curthread, curthread->td_pcb);
1357 	/* Ensure the VFP state will be re-loaded when exiting the guest */
1358 	PCPU_SET(fpcurthread, NULL);
1359 
1360 	/* restore guest FPU state */
1361 	vfp_enable();
1362 	vfp_restore(vcpu->guestfpu);
1363 
1364 	/*
1365 	 * The FPU is now "dirty" with the guest's state so turn on emulation
1366 	 * to trap any access to the FPU by the host.
1367 	 */
1368 	vfp_disable();
1369 }
1370 
1371 static void
1372 save_guest_fpustate(struct vcpu *vcpu)
1373 {
1374 	if ((READ_SPECIALREG(cpacr_el1) & CPACR_FPEN_MASK) !=
1375 	    CPACR_FPEN_TRAP_ALL1)
1376 		panic("VFP not enabled in host!");
1377 
1378 	/* save guest FPU state */
1379 	vfp_enable();
1380 	vfp_store(vcpu->guestfpu);
1381 	vfp_disable();
1382 
1383 	KASSERT(PCPU_GET(fpcurthread) == NULL,
1384 	    ("%s: fpcurthread set with guest registers", __func__));
1385 }
1386 static int
1387 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
1388     bool from_idle)
1389 {
1390 	int error;
1391 
1392 	vcpu_assert_locked(vcpu);
1393 
1394 	/*
1395 	 * State transitions from the vmmdev_ioctl() must always begin from
1396 	 * the VCPU_IDLE state. This guarantees that there is only a single
1397 	 * ioctl() operating on a vcpu at any point.
1398 	 */
1399 	if (from_idle) {
1400 		while (vcpu->state != VCPU_IDLE) {
1401 			vcpu_notify_event_locked(vcpu);
1402 			msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
1403 		}
1404 	} else {
1405 		KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
1406 		    "vcpu idle state"));
1407 	}
1408 
1409 	if (vcpu->state == VCPU_RUNNING) {
1410 		KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
1411 		    "mismatch for running vcpu", curcpu, vcpu->hostcpu));
1412 	} else {
1413 		KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
1414 		    "vcpu that is not running", vcpu->hostcpu));
1415 	}
1416 
1417 	/*
1418 	 * The following state transitions are allowed:
1419 	 * IDLE -> FROZEN -> IDLE
1420 	 * FROZEN -> RUNNING -> FROZEN
1421 	 * FROZEN -> SLEEPING -> FROZEN
1422 	 */
1423 	switch (vcpu->state) {
1424 	case VCPU_IDLE:
1425 	case VCPU_RUNNING:
1426 	case VCPU_SLEEPING:
1427 		error = (newstate != VCPU_FROZEN);
1428 		break;
1429 	case VCPU_FROZEN:
1430 		error = (newstate == VCPU_FROZEN);
1431 		break;
1432 	default:
1433 		error = 1;
1434 		break;
1435 	}
1436 
1437 	if (error)
1438 		return (EBUSY);
1439 
1440 	vcpu->state = newstate;
1441 	if (newstate == VCPU_RUNNING)
1442 		vcpu->hostcpu = curcpu;
1443 	else
1444 		vcpu->hostcpu = NOCPU;
1445 
1446 	if (newstate == VCPU_IDLE)
1447 		wakeup(&vcpu->state);
1448 
1449 	return (0);
1450 }
1451 
1452 static void
1453 vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate)
1454 {
1455 	int error;
1456 
1457 	if ((error = vcpu_set_state(vcpu, newstate, false)) != 0)
1458 		panic("Error %d setting state to %d\n", error, newstate);
1459 }
1460 
1461 static void
1462 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
1463 {
1464 	int error;
1465 
1466 	if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0)
1467 		panic("Error %d setting state to %d", error, newstate);
1468 }
1469 
1470 int
1471 vm_get_capability(struct vcpu *vcpu, int type, int *retval)
1472 {
1473 	if (type < 0 || type >= VM_CAP_MAX)
1474 		return (EINVAL);
1475 
1476 	return (vmmops_getcap(vcpu->cookie, type, retval));
1477 }
1478 
1479 int
1480 vm_set_capability(struct vcpu *vcpu, int type, int val)
1481 {
1482 	if (type < 0 || type >= VM_CAP_MAX)
1483 		return (EINVAL);
1484 
1485 	return (vmmops_setcap(vcpu->cookie, type, val));
1486 }
1487 
1488 struct vm *
1489 vcpu_vm(struct vcpu *vcpu)
1490 {
1491 	return (vcpu->vm);
1492 }
1493 
1494 int
1495 vcpu_vcpuid(struct vcpu *vcpu)
1496 {
1497 	return (vcpu->vcpuid);
1498 }
1499 
1500 void *
1501 vcpu_get_cookie(struct vcpu *vcpu)
1502 {
1503 	return (vcpu->cookie);
1504 }
1505 
1506 struct vcpu *
1507 vm_vcpu(struct vm *vm, int vcpuid)
1508 {
1509 	return (vm->vcpu[vcpuid]);
1510 }
1511 
1512 int
1513 vcpu_set_state(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle)
1514 {
1515 	int error;
1516 
1517 	vcpu_lock(vcpu);
1518 	error = vcpu_set_state_locked(vcpu, newstate, from_idle);
1519 	vcpu_unlock(vcpu);
1520 
1521 	return (error);
1522 }
1523 
1524 enum vcpu_state
1525 vcpu_get_state(struct vcpu *vcpu, int *hostcpu)
1526 {
1527 	enum vcpu_state state;
1528 
1529 	vcpu_lock(vcpu);
1530 	state = vcpu->state;
1531 	if (hostcpu != NULL)
1532 		*hostcpu = vcpu->hostcpu;
1533 	vcpu_unlock(vcpu);
1534 
1535 	return (state);
1536 }
1537 
1538 static void *
1539 _vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
1540     void **cookie)
1541 {
1542 	int i, count, pageoff;
1543 	struct mem_map *mm;
1544 	vm_page_t m;
1545 
1546 	pageoff = gpa & PAGE_MASK;
1547 	if (len > PAGE_SIZE - pageoff)
1548 		panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
1549 
1550 	count = 0;
1551 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
1552 		mm = &vm->mem_maps[i];
1553 		if (sysmem_mapping(vm, mm) && gpa >= mm->gpa &&
1554 		    gpa < mm->gpa + mm->len) {
1555 			count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
1556 			    trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
1557 			break;
1558 		}
1559 	}
1560 
1561 	if (count == 1) {
1562 		*cookie = m;
1563 		return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
1564 	} else {
1565 		*cookie = NULL;
1566 		return (NULL);
1567 	}
1568 }
1569 
1570 void *
1571 vm_gpa_hold(struct vcpu *vcpu, vm_paddr_t gpa, size_t len, int reqprot,
1572 	    void **cookie)
1573 {
1574 #ifdef INVARIANTS
1575 	/*
1576 	 * The current vcpu should be frozen to ensure 'vm_memmap[]'
1577 	 * stability.
1578 	 */
1579 	int state = vcpu_get_state(vcpu, NULL);
1580 	KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d",
1581 	    __func__, state));
1582 #endif
1583 	return (_vm_gpa_hold(vcpu->vm, gpa, len, reqprot, cookie));
1584 }
1585 
1586 void *
1587 vm_gpa_hold_global(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
1588     void **cookie)
1589 {
1590 	sx_assert(&vm->mem_segs_lock, SX_LOCKED);
1591 	return (_vm_gpa_hold(vm, gpa, len, reqprot, cookie));
1592 }
1593 
1594 void
1595 vm_gpa_release(void *cookie)
1596 {
1597 	vm_page_t m = cookie;
1598 
1599 	vm_page_unwire(m, PQ_ACTIVE);
1600 }
1601 
1602 int
1603 vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval)
1604 {
1605 
1606 	if (reg >= VM_REG_LAST)
1607 		return (EINVAL);
1608 
1609 	return (vmmops_getreg(vcpu->cookie, reg, retval));
1610 }
1611 
1612 int
1613 vm_set_register(struct vcpu *vcpu, int reg, uint64_t val)
1614 {
1615 	int error;
1616 
1617 	if (reg >= VM_REG_LAST)
1618 		return (EINVAL);
1619 	error = vmmops_setreg(vcpu->cookie, reg, val);
1620 	if (error || reg != VM_REG_GUEST_PC)
1621 		return (error);
1622 
1623 	vcpu->nextpc = val;
1624 
1625 	return (0);
1626 }
1627 
1628 void *
1629 vm_get_cookie(struct vm *vm)
1630 {
1631 	return (vm->cookie);
1632 }
1633 
1634 int
1635 vm_inject_exception(struct vcpu *vcpu, uint64_t esr, uint64_t far)
1636 {
1637 	return (vmmops_exception(vcpu->cookie, esr, far));
1638 }
1639 
1640 int
1641 vm_attach_vgic(struct vm *vm, struct vm_vgic_descr *descr)
1642 {
1643 	return (vgic_attach_to_vm(vm->cookie, descr));
1644 }
1645 
1646 int
1647 vm_assert_irq(struct vm *vm, uint32_t irq)
1648 {
1649 	return (vgic_inject_irq(vm->cookie, -1, irq, true));
1650 }
1651 
1652 int
1653 vm_deassert_irq(struct vm *vm, uint32_t irq)
1654 {
1655 	return (vgic_inject_irq(vm->cookie, -1, irq, false));
1656 }
1657 
1658 int
1659 vm_raise_msi(struct vm *vm, uint64_t msg, uint64_t addr, int bus, int slot,
1660     int func)
1661 {
1662 	/* TODO: Should we raise an SError? */
1663 	return (vgic_inject_msi(vm->cookie, msg, addr));
1664 }
1665 
1666 static int
1667 vm_handle_smccc_call(struct vcpu *vcpu, struct vm_exit *vme, bool *retu)
1668 {
1669 	struct hypctx *hypctx;
1670 	int i;
1671 
1672 	hypctx = vcpu_get_cookie(vcpu);
1673 
1674 	if ((hypctx->tf.tf_esr & ESR_ELx_ISS_MASK) != 0)
1675 		return (1);
1676 
1677 	vme->exitcode = VM_EXITCODE_SMCCC;
1678 	vme->u.smccc_call.func_id = hypctx->tf.tf_x[0];
1679 	for (i = 0; i < nitems(vme->u.smccc_call.args); i++)
1680 		vme->u.smccc_call.args[i] = hypctx->tf.tf_x[i + 1];
1681 
1682 	*retu = true;
1683 	return (0);
1684 }
1685 
1686 static int
1687 vm_handle_wfi(struct vcpu *vcpu, struct vm_exit *vme, bool *retu)
1688 {
1689 	vcpu_lock(vcpu);
1690 	while (1) {
1691 		if (vgic_has_pending_irq(vcpu->cookie))
1692 			break;
1693 
1694 		if (vcpu_should_yield(vcpu))
1695 			break;
1696 
1697 		vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1698 		/*
1699 		 * XXX msleep_spin() cannot be interrupted by signals so
1700 		 * wake up periodically to check pending signals.
1701 		 */
1702 		msleep_spin(vcpu, &vcpu->mtx, "vmidle", hz);
1703 		vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1704 	}
1705 	vcpu_unlock(vcpu);
1706 
1707 	*retu = false;
1708 	return (0);
1709 }
1710 
1711 static int
1712 vm_handle_paging(struct vcpu *vcpu, bool *retu)
1713 {
1714 	struct vm *vm = vcpu->vm;
1715 	struct vm_exit *vme;
1716 	struct vm_map *map;
1717 	uint64_t addr, esr;
1718 	pmap_t pmap;
1719 	int ftype, rv;
1720 
1721 	vme = &vcpu->exitinfo;
1722 
1723 	pmap = vmspace_pmap(vcpu->vm->vmspace);
1724 	addr = vme->u.paging.gpa;
1725 	esr = vme->u.paging.esr;
1726 
1727 	/* The page exists, but the page table needs to be updated. */
1728 	if (pmap_fault(pmap, esr, addr) == KERN_SUCCESS)
1729 		return (0);
1730 
1731 	switch (ESR_ELx_EXCEPTION(esr)) {
1732 	case EXCP_INSN_ABORT_L:
1733 	case EXCP_DATA_ABORT_L:
1734 		ftype = VM_PROT_EXECUTE | VM_PROT_READ | VM_PROT_WRITE;
1735 		break;
1736 	default:
1737 		panic("%s: Invalid exception (esr = %lx)", __func__, esr);
1738 	}
1739 
1740 	map = &vm->vmspace->vm_map;
1741 	rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL);
1742 	if (rv != KERN_SUCCESS)
1743 		return (EFAULT);
1744 
1745 	return (0);
1746 }
1747 
1748 static int
1749 vm_handle_suspend(struct vcpu *vcpu, bool *retu)
1750 {
1751 	struct vm *vm = vcpu->vm;
1752 	int error, i;
1753 	struct thread *td;
1754 
1755 	error = 0;
1756 	td = curthread;
1757 
1758 	CPU_SET_ATOMIC(vcpu->vcpuid, &vm->suspended_cpus);
1759 
1760 	/*
1761 	 * Wait until all 'active_cpus' have suspended themselves.
1762 	 *
1763 	 * Since a VM may be suspended at any time including when one or
1764 	 * more vcpus are doing a rendezvous we need to call the rendezvous
1765 	 * handler while we are waiting to prevent a deadlock.
1766 	 */
1767 	vcpu_lock(vcpu);
1768 	while (error == 0) {
1769 		if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0)
1770 			break;
1771 
1772 		vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1773 		msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1774 		vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1775 		if (td_ast_pending(td, TDA_SUSPEND)) {
1776 			vcpu_unlock(vcpu);
1777 			error = thread_check_susp(td, false);
1778 			vcpu_lock(vcpu);
1779 		}
1780 	}
1781 	vcpu_unlock(vcpu);
1782 
1783 	/*
1784 	 * Wakeup the other sleeping vcpus and return to userspace.
1785 	 */
1786 	for (i = 0; i < vm->maxcpus; i++) {
1787 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
1788 			vcpu_notify_event(vm_vcpu(vm, i));
1789 		}
1790 	}
1791 
1792 	*retu = true;
1793 	return (error);
1794 }
1795 
1796 int
1797 vm_run(struct vcpu *vcpu)
1798 {
1799 	struct vm *vm = vcpu->vm;
1800 	struct vm_eventinfo evinfo;
1801 	int error, vcpuid;
1802 	struct vm_exit *vme;
1803 	bool retu;
1804 	pmap_t pmap;
1805 
1806 	vcpuid = vcpu->vcpuid;
1807 
1808 	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1809 		return (EINVAL);
1810 
1811 	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1812 		return (EINVAL);
1813 
1814 	pmap = vmspace_pmap(vm->vmspace);
1815 	vme = &vcpu->exitinfo;
1816 	evinfo.rptr = NULL;
1817 	evinfo.sptr = &vm->suspend;
1818 	evinfo.iptr = NULL;
1819 restart:
1820 	critical_enter();
1821 
1822 	restore_guest_fpustate(vcpu);
1823 
1824 	vcpu_require_state(vcpu, VCPU_RUNNING);
1825 	error = vmmops_run(vcpu->cookie, vcpu->nextpc, pmap, &evinfo);
1826 	vcpu_require_state(vcpu, VCPU_FROZEN);
1827 
1828 	save_guest_fpustate(vcpu);
1829 
1830 	critical_exit();
1831 
1832 	if (error == 0) {
1833 		retu = false;
1834 		switch (vme->exitcode) {
1835 		case VM_EXITCODE_INST_EMUL:
1836 			vcpu->nextpc = vme->pc + vme->inst_length;
1837 			error = vm_handle_inst_emul(vcpu, &retu);
1838 			break;
1839 
1840 		case VM_EXITCODE_REG_EMUL:
1841 			vcpu->nextpc = vme->pc + vme->inst_length;
1842 			error = vm_handle_reg_emul(vcpu, &retu);
1843 			break;
1844 
1845 		case VM_EXITCODE_HVC:
1846 			/*
1847 			 * The HVC instruction saves the address for the
1848 			 * next instruction as the return address.
1849 			 */
1850 			vcpu->nextpc = vme->pc;
1851 			/*
1852 			 * The PSCI call can change the exit information in the
1853 			 * case of suspend/reset/poweroff/cpu off/cpu on.
1854 			 */
1855 			error = vm_handle_smccc_call(vcpu, vme, &retu);
1856 			break;
1857 
1858 		case VM_EXITCODE_WFI:
1859 			vcpu->nextpc = vme->pc + vme->inst_length;
1860 			error = vm_handle_wfi(vcpu, vme, &retu);
1861 			break;
1862 
1863 		case VM_EXITCODE_PAGING:
1864 			vcpu->nextpc = vme->pc;
1865 			error = vm_handle_paging(vcpu, &retu);
1866 			break;
1867 
1868 		case VM_EXITCODE_SUSPENDED:
1869 			vcpu->nextpc = vme->pc;
1870 			error = vm_handle_suspend(vcpu, &retu);
1871 			break;
1872 
1873 		default:
1874 			/* Handle in userland */
1875 			vcpu->nextpc = vme->pc;
1876 			retu = true;
1877 			break;
1878 		}
1879 	}
1880 
1881 	if (error == 0 && retu == false)
1882 		goto restart;
1883 
1884 	return (error);
1885 }
1886