xref: /freebsd/sys/riscv/vmm/vmm.c (revision bd27bd1f51d049538cc7a0053be9d99110a53ae1)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2015 Mihai Carabas <mihai.carabas@gmail.com>
5  * Copyright (c) 2024 Ruslan Bukin <br@bsdpad.com>
6  *
7  * This software was developed by the University of Cambridge Computer
8  * Laboratory (Department of Computer Science and Technology) under Innovate
9  * UK project 105694, "Digital Security by Design (DSbD) Technology Platform
10  * Prototype".
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  *
21  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/cpuset.h>
37 #include <sys/kernel.h>
38 #include <sys/linker.h>
39 #include <sys/lock.h>
40 #include <sys/malloc.h>
41 #include <sys/module.h>
42 #include <sys/mutex.h>
43 #include <sys/pcpu.h>
44 #include <sys/proc.h>
45 #include <sys/queue.h>
46 #include <sys/rwlock.h>
47 #include <sys/sched.h>
48 #include <sys/smp.h>
49 #include <sys/sysctl.h>
50 
51 #include <vm/vm.h>
52 #include <vm/vm_object.h>
53 #include <vm/vm_page.h>
54 #include <vm/pmap.h>
55 #include <vm/vm_map.h>
56 #include <vm/vm_extern.h>
57 #include <vm/vm_param.h>
58 
59 #include <machine/riscvreg.h>
60 #include <machine/cpu.h>
61 #include <machine/fpe.h>
62 #include <machine/machdep.h>
63 #include <machine/pcb.h>
64 #include <machine/smp.h>
65 #include <machine/vm.h>
66 #include <machine/vmparam.h>
67 #include <machine/vmm.h>
68 #include <machine/vmm_instruction_emul.h>
69 
70 #include <dev/pci/pcireg.h>
71 
72 #include <dev/vmm/vmm_dev.h>
73 #include <dev/vmm/vmm_ktr.h>
74 #include <dev/vmm/vmm_mem.h>
75 
76 #include "vmm_stat.h"
77 #include "riscv.h"
78 
79 #include "vmm_aplic.h"
80 
81 struct vcpu {
82 	int		flags;
83 	enum vcpu_state	state;
84 	struct mtx	mtx;
85 	int		hostcpu;	/* host cpuid this vcpu last ran on */
86 	int		vcpuid;
87 	void		*stats;
88 	struct vm_exit	exitinfo;
89 	uint64_t	nextpc;		/* (x) next instruction to execute */
90 	struct vm	*vm;		/* (o) */
91 	void		*cookie;	/* (i) cpu-specific data */
92 	struct fpreg	*guestfpu;	/* (a,i) guest fpu state */
93 };
94 
95 #define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
96 #define	vcpu_lock_destroy(v)	mtx_destroy(&((v)->mtx))
97 #define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
98 #define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
99 #define	vcpu_assert_locked(v)	mtx_assert(&((v)->mtx), MA_OWNED)
100 
101 struct vmm_mmio_region {
102 	uint64_t start;
103 	uint64_t end;
104 	mem_region_read_t read;
105 	mem_region_write_t write;
106 };
107 #define	VM_MAX_MMIO_REGIONS	4
108 
109 /*
110  * Initialization:
111  * (o) initialized the first time the VM is created
112  * (i) initialized when VM is created and when it is reinitialized
113  * (x) initialized before use
114  */
115 struct vm {
116 	void		*cookie;		/* (i) cpu-specific data */
117 	volatile cpuset_t active_cpus;		/* (i) active vcpus */
118 	volatile cpuset_t debug_cpus;		/* (i) vcpus stopped for debug*/
119 	int		suspend;		/* (i) stop VM execution */
120 	bool		dying;			/* (o) is dying */
121 	volatile cpuset_t suspended_cpus; 	/* (i) suspended vcpus */
122 	volatile cpuset_t halted_cpus;		/* (x) cpus in a hard halt */
123 	struct vm_mem	mem;			/* (i) [m+v] guest memory */
124 	char		name[VM_MAX_NAMELEN];	/* (o) virtual machine name */
125 	struct vcpu	**vcpu;			/* (i) guest vcpus */
126 	struct vmm_mmio_region mmio_region[VM_MAX_MMIO_REGIONS];
127 						/* (o) guest MMIO regions */
128 	/* The following describe the vm cpu topology */
129 	uint16_t	sockets;		/* (o) num of sockets */
130 	uint16_t	cores;			/* (o) num of cores/socket */
131 	uint16_t	threads;		/* (o) num of threads/core */
132 	uint16_t	maxcpus;		/* (o) max pluggable cpus */
133 	struct sx	vcpus_init_lock;	/* (o) */
134 };
135 
136 static bool vmm_initialized = false;
137 
138 static MALLOC_DEFINE(M_VMM, "vmm", "vmm");
139 
140 /* statistics */
141 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
142 
143 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
144 
145 static int vmm_ipinum;
146 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
147     "IPI vector used for vcpu notifications");
148 
149 u_int vm_maxcpu;
150 SYSCTL_UINT(_hw_vmm, OID_AUTO, maxcpu, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
151     &vm_maxcpu, 0, "Maximum number of vCPUs");
152 
153 static void vcpu_notify_event_locked(struct vcpu *vcpu);
154 
155 /* global statistics */
156 VMM_STAT(VMEXIT_COUNT, "total number of vm exits");
157 VMM_STAT(VMEXIT_IRQ, "number of vmexits for an irq");
158 VMM_STAT(VMEXIT_UNHANDLED, "number of vmexits for an unhandled exception");
159 
160 /*
161  * Upper limit on vm_maxcpu. We could increase this to 28 bits, but this
162  * is a safe value for now.
163  */
164 #define	VM_MAXCPU	MIN(0xffff - 1, CPU_SETSIZE)
165 
166 static void
167 vcpu_cleanup(struct vcpu *vcpu, bool destroy)
168 {
169 	vmmops_vcpu_cleanup(vcpu->cookie);
170 	vcpu->cookie = NULL;
171 	if (destroy) {
172 		vmm_stat_free(vcpu->stats);
173 		fpu_save_area_free(vcpu->guestfpu);
174 		vcpu_lock_destroy(vcpu);
175 		free(vcpu, M_VMM);
176 	}
177 }
178 
179 static struct vcpu *
180 vcpu_alloc(struct vm *vm, int vcpu_id)
181 {
182 	struct vcpu *vcpu;
183 
184 	KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus,
185 	    ("vcpu_alloc: invalid vcpu %d", vcpu_id));
186 
187 	vcpu = malloc(sizeof(*vcpu), M_VMM, M_WAITOK | M_ZERO);
188 	vcpu_lock_init(vcpu);
189 	vcpu->state = VCPU_IDLE;
190 	vcpu->hostcpu = NOCPU;
191 	vcpu->vcpuid = vcpu_id;
192 	vcpu->vm = vm;
193 	vcpu->guestfpu = fpu_save_area_alloc();
194 	vcpu->stats = vmm_stat_alloc();
195 	return (vcpu);
196 }
197 
198 static void
199 vcpu_init(struct vcpu *vcpu)
200 {
201 	vcpu->cookie = vmmops_vcpu_init(vcpu->vm->cookie, vcpu, vcpu->vcpuid);
202 	MPASS(vcpu->cookie != NULL);
203 	fpu_save_area_reset(vcpu->guestfpu);
204 	vmm_stat_init(vcpu->stats);
205 }
206 
207 struct vm_exit *
208 vm_exitinfo(struct vcpu *vcpu)
209 {
210 	return (&vcpu->exitinfo);
211 }
212 
213 static int
214 vmm_init(void)
215 {
216 
217 	vm_maxcpu = mp_ncpus;
218 
219 	TUNABLE_INT_FETCH("hw.vmm.maxcpu", &vm_maxcpu);
220 
221 	if (vm_maxcpu > VM_MAXCPU) {
222 		printf("vmm: vm_maxcpu clamped to %u\n", VM_MAXCPU);
223 		vm_maxcpu = VM_MAXCPU;
224 	}
225 
226 	if (vm_maxcpu == 0)
227 		vm_maxcpu = 1;
228 
229 	return (vmmops_modinit());
230 }
231 
232 static int
233 vmm_handler(module_t mod, int what, void *arg)
234 {
235 	int error;
236 
237 	switch (what) {
238 	case MOD_LOAD:
239 		error = vmmdev_init();
240 		if (error != 0)
241 			break;
242 		error = vmm_init();
243 		if (error == 0)
244 			vmm_initialized = true;
245 		else
246 			(void)vmmdev_cleanup();
247 		break;
248 	case MOD_UNLOAD:
249 		error = vmmdev_cleanup();
250 		if (error == 0 && vmm_initialized) {
251 			error = vmmops_modcleanup();
252 			if (error) {
253 				/*
254 				 * Something bad happened - prevent new
255 				 * VMs from being created
256 				 */
257 				vmm_initialized = false;
258 			}
259 		}
260 		break;
261 	default:
262 		error = 0;
263 		break;
264 	}
265 	return (error);
266 }
267 
268 static moduledata_t vmm_kmod = {
269 	"vmm",
270 	vmm_handler,
271 	NULL
272 };
273 
274 /*
275  * vmm initialization has the following dependencies:
276  *
277  * - vmm device initialization requires an initialized devfs.
278  */
279 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_DEVFS + 1, SI_ORDER_ANY);
280 MODULE_VERSION(vmm, 1);
281 
282 static void
283 vm_init(struct vm *vm, bool create)
284 {
285 	int i;
286 
287 	vm->cookie = vmmops_init(vm, vmspace_pmap(vm_vmspace(vm)));
288 	MPASS(vm->cookie != NULL);
289 
290 	CPU_ZERO(&vm->active_cpus);
291 	CPU_ZERO(&vm->debug_cpus);
292 
293 	vm->suspend = 0;
294 	CPU_ZERO(&vm->suspended_cpus);
295 
296 	memset(vm->mmio_region, 0, sizeof(vm->mmio_region));
297 
298 	if (!create) {
299 		for (i = 0; i < vm->maxcpus; i++) {
300 			if (vm->vcpu[i] != NULL)
301 				vcpu_init(vm->vcpu[i]);
302 		}
303 	}
304 }
305 
306 void
307 vm_disable_vcpu_creation(struct vm *vm)
308 {
309 	sx_xlock(&vm->vcpus_init_lock);
310 	vm->dying = true;
311 	sx_xunlock(&vm->vcpus_init_lock);
312 }
313 
314 struct vcpu *
315 vm_alloc_vcpu(struct vm *vm, int vcpuid)
316 {
317 	struct vcpu *vcpu;
318 
319 	if (vcpuid < 0 || vcpuid >= vm_get_maxcpus(vm))
320 		return (NULL);
321 
322 	vcpu = (struct vcpu *)
323 	    atomic_load_acq_ptr((uintptr_t *)&vm->vcpu[vcpuid]);
324 	if (__predict_true(vcpu != NULL))
325 		return (vcpu);
326 
327 	sx_xlock(&vm->vcpus_init_lock);
328 	vcpu = vm->vcpu[vcpuid];
329 	if (vcpu == NULL && !vm->dying) {
330 		vcpu = vcpu_alloc(vm, vcpuid);
331 		vcpu_init(vcpu);
332 
333 		/*
334 		 * Ensure vCPU is fully created before updating pointer
335 		 * to permit unlocked reads above.
336 		 */
337 		atomic_store_rel_ptr((uintptr_t *)&vm->vcpu[vcpuid],
338 		    (uintptr_t)vcpu);
339 	}
340 	sx_xunlock(&vm->vcpus_init_lock);
341 	return (vcpu);
342 }
343 
344 void
345 vm_lock_vcpus(struct vm *vm)
346 {
347 	sx_xlock(&vm->vcpus_init_lock);
348 }
349 
350 void
351 vm_unlock_vcpus(struct vm *vm)
352 {
353 	sx_unlock(&vm->vcpus_init_lock);
354 }
355 
356 int
357 vm_create(const char *name, struct vm **retvm)
358 {
359 	struct vm *vm;
360 	int error;
361 
362 	/*
363 	 * If vmm.ko could not be successfully initialized then don't attempt
364 	 * to create the virtual machine.
365 	 */
366 	if (!vmm_initialized)
367 		return (ENXIO);
368 
369 	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
370 		return (EINVAL);
371 
372 	vm = malloc(sizeof(struct vm), M_VMM, M_WAITOK | M_ZERO);
373 	error = vm_mem_init(&vm->mem, 0, 1ul << 39);
374 	if (error != 0) {
375 		free(vm, M_VMM);
376 		return (error);
377 	}
378 	strcpy(vm->name, name);
379 	sx_init(&vm->vcpus_init_lock, "vm vcpus");
380 
381 	vm->sockets = 1;
382 	vm->cores = 1;			/* XXX backwards compatibility */
383 	vm->threads = 1;		/* XXX backwards compatibility */
384 	vm->maxcpus = vm_maxcpu;
385 
386 	vm->vcpu = malloc(sizeof(*vm->vcpu) * vm->maxcpus, M_VMM,
387 	    M_WAITOK | M_ZERO);
388 
389 	vm_init(vm, true);
390 
391 	*retvm = vm;
392 	return (0);
393 }
394 
395 void
396 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores,
397     uint16_t *threads, uint16_t *maxcpus)
398 {
399 	*sockets = vm->sockets;
400 	*cores = vm->cores;
401 	*threads = vm->threads;
402 	*maxcpus = vm->maxcpus;
403 }
404 
405 uint16_t
406 vm_get_maxcpus(struct vm *vm)
407 {
408 	return (vm->maxcpus);
409 }
410 
411 int
412 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores,
413     uint16_t threads, uint16_t maxcpus)
414 {
415 	/* Ignore maxcpus. */
416 	if ((sockets * cores * threads) > vm->maxcpus)
417 		return (EINVAL);
418 	vm->sockets = sockets;
419 	vm->cores = cores;
420 	vm->threads = threads;
421 	return(0);
422 }
423 
424 static void
425 vm_cleanup(struct vm *vm, bool destroy)
426 {
427 	int i;
428 
429 	if (destroy)
430 		vm_xlock_memsegs(vm);
431 	else
432 		vm_assert_memseg_xlocked(vm);
433 
434 	aplic_detach_from_vm(vm->cookie);
435 
436 	for (i = 0; i < vm->maxcpus; i++) {
437 		if (vm->vcpu[i] != NULL)
438 			vcpu_cleanup(vm->vcpu[i], destroy);
439 	}
440 
441 	vmmops_cleanup(vm->cookie);
442 
443 	vm_mem_cleanup(vm);
444 	if (destroy) {
445 		vm_mem_destroy(vm);
446 
447 		free(vm->vcpu, M_VMM);
448 		sx_destroy(&vm->vcpus_init_lock);
449 	}
450 }
451 
452 void
453 vm_destroy(struct vm *vm)
454 {
455 
456 	vm_cleanup(vm, true);
457 
458 	free(vm, M_VMM);
459 }
460 
461 int
462 vm_reinit(struct vm *vm)
463 {
464 	int error;
465 
466 	/*
467 	 * A virtual machine can be reset only if all vcpus are suspended.
468 	 */
469 	if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
470 		vm_cleanup(vm, false);
471 		vm_init(vm, false);
472 		error = 0;
473 	} else {
474 		error = EBUSY;
475 	}
476 
477 	return (error);
478 }
479 
480 const char *
481 vm_name(struct vm *vm)
482 {
483 	return (vm->name);
484 }
485 
486 int
487 vm_gla2gpa_nofault(struct vcpu *vcpu, struct vm_guest_paging *paging,
488     uint64_t gla, int prot, uint64_t *gpa, int *is_fault)
489 {
490 	return (vmmops_gla2gpa(vcpu->cookie, paging, gla, prot, gpa, is_fault));
491 }
492 
493 void
494 vm_register_inst_handler(struct vm *vm, uint64_t start, uint64_t size,
495     mem_region_read_t mmio_read, mem_region_write_t mmio_write)
496 {
497 	int i;
498 
499 	for (i = 0; i < nitems(vm->mmio_region); i++) {
500 		if (vm->mmio_region[i].start == 0 &&
501 		    vm->mmio_region[i].end == 0) {
502 			vm->mmio_region[i].start = start;
503 			vm->mmio_region[i].end = start + size;
504 			vm->mmio_region[i].read = mmio_read;
505 			vm->mmio_region[i].write = mmio_write;
506 			return;
507 		}
508 	}
509 
510 	panic("%s: No free MMIO region", __func__);
511 }
512 
513 void
514 vm_deregister_inst_handler(struct vm *vm, uint64_t start, uint64_t size)
515 {
516 	int i;
517 
518 	for (i = 0; i < nitems(vm->mmio_region); i++) {
519 		if (vm->mmio_region[i].start == start &&
520 		    vm->mmio_region[i].end == start + size) {
521 			memset(&vm->mmio_region[i], 0,
522 			    sizeof(vm->mmio_region[i]));
523 			return;
524 		}
525 	}
526 
527 	panic("%s: Invalid MMIO region: %lx - %lx", __func__, start,
528 	    start + size);
529 }
530 
531 static int
532 vm_handle_inst_emul(struct vcpu *vcpu, bool *retu)
533 {
534 	struct vm *vm;
535 	struct vm_exit *vme;
536 	struct vie *vie;
537 	struct hyp *hyp;
538 	uint64_t fault_ipa;
539 	struct vm_guest_paging *paging;
540 	struct vmm_mmio_region *vmr;
541 	int error, i;
542 
543 	vm = vcpu->vm;
544 	hyp = vm->cookie;
545 	if (!hyp->aplic_attached)
546 		goto out_user;
547 
548 	vme = &vcpu->exitinfo;
549 	vie = &vme->u.inst_emul.vie;
550 	paging = &vme->u.inst_emul.paging;
551 
552 	fault_ipa = vme->u.inst_emul.gpa;
553 
554 	vmr = NULL;
555 	for (i = 0; i < nitems(vm->mmio_region); i++) {
556 		if (vm->mmio_region[i].start <= fault_ipa &&
557 		    vm->mmio_region[i].end > fault_ipa) {
558 			vmr = &vm->mmio_region[i];
559 			break;
560 		}
561 	}
562 	if (vmr == NULL)
563 		goto out_user;
564 
565 	error = vmm_emulate_instruction(vcpu, fault_ipa, vie, paging,
566 	    vmr->read, vmr->write, retu);
567 	return (error);
568 
569 out_user:
570 	*retu = true;
571 	return (0);
572 }
573 
574 int
575 vm_suspend(struct vm *vm, enum vm_suspend_how how)
576 {
577 	int i;
578 
579 	if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
580 		return (EINVAL);
581 
582 	if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
583 		VM_CTR2(vm, "virtual machine already suspended %d/%d",
584 		    vm->suspend, how);
585 		return (EALREADY);
586 	}
587 
588 	VM_CTR1(vm, "virtual machine successfully suspended %d", how);
589 
590 	/*
591 	 * Notify all active vcpus that they are now suspended.
592 	 */
593 	for (i = 0; i < vm->maxcpus; i++) {
594 		if (CPU_ISSET(i, &vm->active_cpus))
595 			vcpu_notify_event(vm_vcpu(vm, i));
596 	}
597 
598 	return (0);
599 }
600 
601 void
602 vm_exit_suspended(struct vcpu *vcpu, uint64_t pc)
603 {
604 	struct vm *vm = vcpu->vm;
605 	struct vm_exit *vmexit;
606 
607 	KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
608 	    ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
609 
610 	vmexit = vm_exitinfo(vcpu);
611 	vmexit->pc = pc;
612 	vmexit->inst_length = 4;
613 	vmexit->exitcode = VM_EXITCODE_SUSPENDED;
614 	vmexit->u.suspended.how = vm->suspend;
615 }
616 
617 void
618 vm_exit_debug(struct vcpu *vcpu, uint64_t pc)
619 {
620 	struct vm_exit *vmexit;
621 
622 	vmexit = vm_exitinfo(vcpu);
623 	vmexit->pc = pc;
624 	vmexit->inst_length = 4;
625 	vmexit->exitcode = VM_EXITCODE_DEBUG;
626 }
627 
628 int
629 vm_activate_cpu(struct vcpu *vcpu)
630 {
631 	struct vm *vm = vcpu->vm;
632 
633 	if (CPU_ISSET(vcpu->vcpuid, &vm->active_cpus))
634 		return (EBUSY);
635 
636 	CPU_SET_ATOMIC(vcpu->vcpuid, &vm->active_cpus);
637 	return (0);
638 
639 }
640 
641 int
642 vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu)
643 {
644 	if (vcpu == NULL) {
645 		vm->debug_cpus = vm->active_cpus;
646 		for (int i = 0; i < vm->maxcpus; i++) {
647 			if (CPU_ISSET(i, &vm->active_cpus))
648 				vcpu_notify_event(vm_vcpu(vm, i));
649 		}
650 	} else {
651 		if (!CPU_ISSET(vcpu->vcpuid, &vm->active_cpus))
652 			return (EINVAL);
653 
654 		CPU_SET_ATOMIC(vcpu->vcpuid, &vm->debug_cpus);
655 		vcpu_notify_event(vcpu);
656 	}
657 	return (0);
658 }
659 
660 int
661 vm_resume_cpu(struct vm *vm, struct vcpu *vcpu)
662 {
663 
664 	if (vcpu == NULL) {
665 		CPU_ZERO(&vm->debug_cpus);
666 	} else {
667 		if (!CPU_ISSET(vcpu->vcpuid, &vm->debug_cpus))
668 			return (EINVAL);
669 
670 		CPU_CLR_ATOMIC(vcpu->vcpuid, &vm->debug_cpus);
671 	}
672 	return (0);
673 }
674 
675 int
676 vcpu_debugged(struct vcpu *vcpu)
677 {
678 
679 	return (CPU_ISSET(vcpu->vcpuid, &vcpu->vm->debug_cpus));
680 }
681 
682 cpuset_t
683 vm_active_cpus(struct vm *vm)
684 {
685 
686 	return (vm->active_cpus);
687 }
688 
689 cpuset_t
690 vm_debug_cpus(struct vm *vm)
691 {
692 
693 	return (vm->debug_cpus);
694 }
695 
696 cpuset_t
697 vm_suspended_cpus(struct vm *vm)
698 {
699 
700 	return (vm->suspended_cpus);
701 }
702 
703 
704 void *
705 vcpu_stats(struct vcpu *vcpu)
706 {
707 
708 	return (vcpu->stats);
709 }
710 
711 /*
712  * This function is called to ensure that a vcpu "sees" a pending event
713  * as soon as possible:
714  * - If the vcpu thread is sleeping then it is woken up.
715  * - If the vcpu is running on a different host_cpu then an IPI will be directed
716  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
717  */
718 static void
719 vcpu_notify_event_locked(struct vcpu *vcpu)
720 {
721 	int hostcpu;
722 
723 	hostcpu = vcpu->hostcpu;
724 	if (vcpu->state == VCPU_RUNNING) {
725 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
726 		if (hostcpu != curcpu) {
727 			ipi_cpu(hostcpu, vmm_ipinum);
728 		} else {
729 			/*
730 			 * If the 'vcpu' is running on 'curcpu' then it must
731 			 * be sending a notification to itself (e.g. SELF_IPI).
732 			 * The pending event will be picked up when the vcpu
733 			 * transitions back to guest context.
734 			 */
735 		}
736 	} else {
737 		KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
738 		    "with hostcpu %d", vcpu->state, hostcpu));
739 		if (vcpu->state == VCPU_SLEEPING)
740 			wakeup_one(vcpu);
741 	}
742 }
743 
744 void
745 vcpu_notify_event(struct vcpu *vcpu)
746 {
747 	vcpu_lock(vcpu);
748 	vcpu_notify_event_locked(vcpu);
749 	vcpu_unlock(vcpu);
750 }
751 
752 struct vm_mem *
753 vm_mem(struct vm *vm)
754 {
755 	return (&vm->mem);
756 }
757 
758 static void
759 restore_guest_fpustate(struct vcpu *vcpu)
760 {
761 
762 	/* Flush host state to the pcb. */
763 	fpe_state_save(curthread);
764 
765 	/* Ensure the VFP state will be re-loaded when exiting the guest. */
766 	PCPU_SET(fpcurthread, NULL);
767 
768 	/* restore guest FPU state */
769 	fpe_enable();
770 	fpe_restore(vcpu->guestfpu);
771 
772 	/*
773 	 * The FPU is now "dirty" with the guest's state so turn on emulation
774 	 * to trap any access to the FPU by the host.
775 	 */
776 	fpe_disable();
777 }
778 
779 static void
780 save_guest_fpustate(struct vcpu *vcpu)
781 {
782 
783 	/* Save guest FPE state. */
784 	fpe_enable();
785 	fpe_store(vcpu->guestfpu);
786 	fpe_disable();
787 
788 	KASSERT(PCPU_GET(fpcurthread) == NULL,
789 	    ("%s: fpcurthread set with guest registers", __func__));
790 }
791 
792 static int
793 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
794     bool from_idle)
795 {
796 	int error;
797 
798 	vcpu_assert_locked(vcpu);
799 
800 	/*
801 	 * State transitions from the vmmdev_ioctl() must always begin from
802 	 * the VCPU_IDLE state. This guarantees that there is only a single
803 	 * ioctl() operating on a vcpu at any point.
804 	 */
805 	if (from_idle) {
806 		while (vcpu->state != VCPU_IDLE) {
807 			vcpu_notify_event_locked(vcpu);
808 			msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
809 		}
810 	} else {
811 		KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
812 		    "vcpu idle state"));
813 	}
814 
815 	if (vcpu->state == VCPU_RUNNING) {
816 		KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
817 		    "mismatch for running vcpu", curcpu, vcpu->hostcpu));
818 	} else {
819 		KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
820 		    "vcpu that is not running", vcpu->hostcpu));
821 	}
822 
823 	/*
824 	 * The following state transitions are allowed:
825 	 * IDLE -> FROZEN -> IDLE
826 	 * FROZEN -> RUNNING -> FROZEN
827 	 * FROZEN -> SLEEPING -> FROZEN
828 	 */
829 	switch (vcpu->state) {
830 	case VCPU_IDLE:
831 	case VCPU_RUNNING:
832 	case VCPU_SLEEPING:
833 		error = (newstate != VCPU_FROZEN);
834 		break;
835 	case VCPU_FROZEN:
836 		error = (newstate == VCPU_FROZEN);
837 		break;
838 	default:
839 		error = 1;
840 		break;
841 	}
842 
843 	if (error)
844 		return (EBUSY);
845 
846 	vcpu->state = newstate;
847 	if (newstate == VCPU_RUNNING)
848 		vcpu->hostcpu = curcpu;
849 	else
850 		vcpu->hostcpu = NOCPU;
851 
852 	if (newstate == VCPU_IDLE)
853 		wakeup(&vcpu->state);
854 
855 	return (0);
856 }
857 
858 static void
859 vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate)
860 {
861 	int error;
862 
863 	if ((error = vcpu_set_state(vcpu, newstate, false)) != 0)
864 		panic("Error %d setting state to %d\n", error, newstate);
865 }
866 
867 static void
868 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
869 {
870 	int error;
871 
872 	if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0)
873 		panic("Error %d setting state to %d", error, newstate);
874 }
875 
876 int
877 vm_get_capability(struct vcpu *vcpu, int type, int *retval)
878 {
879 
880 	if (type < 0 || type >= VM_CAP_MAX)
881 		return (EINVAL);
882 
883 	return (vmmops_getcap(vcpu->cookie, type, retval));
884 }
885 
886 int
887 vm_set_capability(struct vcpu *vcpu, int type, int val)
888 {
889 
890 	if (type < 0 || type >= VM_CAP_MAX)
891 		return (EINVAL);
892 
893 	return (vmmops_setcap(vcpu->cookie, type, val));
894 }
895 
896 struct vm *
897 vcpu_vm(struct vcpu *vcpu)
898 {
899 
900 	return (vcpu->vm);
901 }
902 
903 int
904 vcpu_vcpuid(struct vcpu *vcpu)
905 {
906 
907 	return (vcpu->vcpuid);
908 }
909 
910 void *
911 vcpu_get_cookie(struct vcpu *vcpu)
912 {
913 
914 	return (vcpu->cookie);
915 }
916 
917 struct vcpu *
918 vm_vcpu(struct vm *vm, int vcpuid)
919 {
920 
921 	return (vm->vcpu[vcpuid]);
922 }
923 
924 int
925 vcpu_set_state(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle)
926 {
927 	int error;
928 
929 	vcpu_lock(vcpu);
930 	error = vcpu_set_state_locked(vcpu, newstate, from_idle);
931 	vcpu_unlock(vcpu);
932 
933 	return (error);
934 }
935 
936 enum vcpu_state
937 vcpu_get_state(struct vcpu *vcpu, int *hostcpu)
938 {
939 	enum vcpu_state state;
940 
941 	vcpu_lock(vcpu);
942 	state = vcpu->state;
943 	if (hostcpu != NULL)
944 		*hostcpu = vcpu->hostcpu;
945 	vcpu_unlock(vcpu);
946 
947 	return (state);
948 }
949 
950 int
951 vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval)
952 {
953 	if (reg < 0 || reg >= VM_REG_LAST)
954 		return (EINVAL);
955 
956 	return (vmmops_getreg(vcpu->cookie, reg, retval));
957 }
958 
959 int
960 vm_set_register(struct vcpu *vcpu, int reg, uint64_t val)
961 {
962 	int error;
963 
964 	if (reg < 0 || reg >= VM_REG_LAST)
965 		return (EINVAL);
966 	error = vmmops_setreg(vcpu->cookie, reg, val);
967 	if (error || reg != VM_REG_GUEST_SEPC)
968 		return (error);
969 
970 	vcpu->nextpc = val;
971 
972 	return (0);
973 }
974 
975 void *
976 vm_get_cookie(struct vm *vm)
977 {
978 
979 	return (vm->cookie);
980 }
981 
982 int
983 vm_inject_exception(struct vcpu *vcpu, uint64_t scause)
984 {
985 
986 	return (vmmops_exception(vcpu->cookie, scause));
987 }
988 
989 int
990 vm_attach_aplic(struct vm *vm, struct vm_aplic_descr *descr)
991 {
992 
993 	return (aplic_attach_to_vm(vm->cookie, descr));
994 }
995 
996 int
997 vm_assert_irq(struct vm *vm, uint32_t irq)
998 {
999 
1000 	return (aplic_inject_irq(vm->cookie, -1, irq, true));
1001 }
1002 
1003 int
1004 vm_deassert_irq(struct vm *vm, uint32_t irq)
1005 {
1006 
1007 	return (aplic_inject_irq(vm->cookie, -1, irq, false));
1008 }
1009 
1010 int
1011 vm_raise_msi(struct vm *vm, uint64_t msg, uint64_t addr, int bus, int slot,
1012     int func)
1013 {
1014 
1015 	return (aplic_inject_msi(vm->cookie, msg, addr));
1016 }
1017 
1018 static int
1019 vm_handle_wfi(struct vcpu *vcpu, struct vm_exit *vme, bool *retu)
1020 {
1021 	struct vm *vm;
1022 
1023 	vm = vcpu->vm;
1024 	vcpu_lock(vcpu);
1025 	while (1) {
1026 		if (vm->suspend)
1027 			break;
1028 
1029 		if (aplic_check_pending(vcpu->cookie))
1030 			break;
1031 
1032 		if (riscv_check_ipi(vcpu->cookie, false))
1033 			break;
1034 
1035 		if (riscv_check_interrupts_pending(vcpu->cookie))
1036 			break;
1037 
1038 		if (vcpu_should_yield(vcpu))
1039 			break;
1040 
1041 		vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1042 		/*
1043 		 * XXX msleep_spin() cannot be interrupted by signals so
1044 		 * wake up periodically to check pending signals.
1045 		 */
1046 		msleep_spin(vcpu, &vcpu->mtx, "vmidle", hz);
1047 		vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1048 	}
1049 	vcpu_unlock(vcpu);
1050 
1051 	*retu = false;
1052 
1053 	return (0);
1054 }
1055 
1056 static int
1057 vm_handle_paging(struct vcpu *vcpu, bool *retu)
1058 {
1059 	struct vm *vm;
1060 	struct vm_exit *vme;
1061 	struct vm_map *map;
1062 	uint64_t addr;
1063 	pmap_t pmap;
1064 	int ftype, rv;
1065 
1066 	vm = vcpu->vm;
1067 	vme = &vcpu->exitinfo;
1068 
1069 	pmap = vmspace_pmap(vm_vmspace(vm));
1070 	addr = (vme->htval << 2) & ~(PAGE_SIZE - 1);
1071 
1072 	dprintf("%s: %lx\n", __func__, addr);
1073 
1074 	switch (vme->scause) {
1075 	case SCAUSE_STORE_GUEST_PAGE_FAULT:
1076 		ftype = VM_PROT_WRITE;
1077 		break;
1078 	case SCAUSE_FETCH_GUEST_PAGE_FAULT:
1079 		ftype = VM_PROT_EXECUTE;
1080 		break;
1081 	case SCAUSE_LOAD_GUEST_PAGE_FAULT:
1082 		ftype = VM_PROT_READ;
1083 		break;
1084 	default:
1085 		panic("unknown page trap: %lu", vme->scause);
1086 	}
1087 
1088 	/* The page exists, but the page table needs to be updated. */
1089 	if (pmap_fault(pmap, addr, ftype))
1090 		return (0);
1091 
1092 	map = &vm_vmspace(vm)->vm_map;
1093 	rv = vm_fault(map, addr, ftype, VM_FAULT_NORMAL, NULL);
1094 	if (rv != KERN_SUCCESS) {
1095 		printf("%s: vm_fault failed, addr %lx, ftype %d, err %d\n",
1096 		    __func__, addr, ftype, rv);
1097 		return (EFAULT);
1098 	}
1099 
1100 	return (0);
1101 }
1102 
1103 static int
1104 vm_handle_suspend(struct vcpu *vcpu, bool *retu)
1105 {
1106 	struct vm *vm = vcpu->vm;
1107 	int error, i;
1108 	struct thread *td;
1109 
1110 	error = 0;
1111 	td = curthread;
1112 
1113 	CPU_SET_ATOMIC(vcpu->vcpuid, &vm->suspended_cpus);
1114 
1115 	/*
1116 	 * Wait until all 'active_cpus' have suspended themselves.
1117 	 *
1118 	 * Since a VM may be suspended at any time including when one or
1119 	 * more vcpus are doing a rendezvous we need to call the rendezvous
1120 	 * handler while we are waiting to prevent a deadlock.
1121 	 */
1122 	vcpu_lock(vcpu);
1123 	while (error == 0) {
1124 		if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0)
1125 			break;
1126 
1127 		vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1128 		msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1129 		vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1130 		if (td_ast_pending(td, TDA_SUSPEND)) {
1131 			vcpu_unlock(vcpu);
1132 			error = thread_check_susp(td, false);
1133 			vcpu_lock(vcpu);
1134 		}
1135 	}
1136 	vcpu_unlock(vcpu);
1137 
1138 	/*
1139 	 * Wakeup the other sleeping vcpus and return to userspace.
1140 	 */
1141 	for (i = 0; i < vm->maxcpus; i++) {
1142 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
1143 			vcpu_notify_event(vm_vcpu(vm, i));
1144 		}
1145 	}
1146 
1147 	*retu = true;
1148 	return (error);
1149 }
1150 
1151 int
1152 vm_run(struct vcpu *vcpu)
1153 {
1154 	struct vm_eventinfo evinfo;
1155 	struct vm_exit *vme;
1156 	struct vm *vm;
1157 	pmap_t pmap;
1158 	int error;
1159 	int vcpuid;
1160 	bool retu;
1161 
1162 	vm = vcpu->vm;
1163 
1164 	dprintf("%s\n", __func__);
1165 
1166 	vcpuid = vcpu->vcpuid;
1167 
1168 	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1169 		return (EINVAL);
1170 
1171 	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1172 		return (EINVAL);
1173 
1174 	pmap = vmspace_pmap(vm_vmspace(vm));
1175 	vme = &vcpu->exitinfo;
1176 	evinfo.rptr = NULL;
1177 	evinfo.sptr = &vm->suspend;
1178 	evinfo.iptr = NULL;
1179 restart:
1180 	critical_enter();
1181 
1182 	restore_guest_fpustate(vcpu);
1183 
1184 	vcpu_require_state(vcpu, VCPU_RUNNING);
1185 	error = vmmops_run(vcpu->cookie, vcpu->nextpc, pmap, &evinfo);
1186 	vcpu_require_state(vcpu, VCPU_FROZEN);
1187 
1188 	save_guest_fpustate(vcpu);
1189 
1190 	critical_exit();
1191 
1192 	if (error == 0) {
1193 		retu = false;
1194 		switch (vme->exitcode) {
1195 		case VM_EXITCODE_INST_EMUL:
1196 			vcpu->nextpc = vme->pc + vme->inst_length;
1197 			error = vm_handle_inst_emul(vcpu, &retu);
1198 			break;
1199 		case VM_EXITCODE_WFI:
1200 			vcpu->nextpc = vme->pc + vme->inst_length;
1201 			error = vm_handle_wfi(vcpu, vme, &retu);
1202 			break;
1203 		case VM_EXITCODE_ECALL:
1204 			/* Handle in userland. */
1205 			vcpu->nextpc = vme->pc + vme->inst_length;
1206 			retu = true;
1207 			break;
1208 		case VM_EXITCODE_PAGING:
1209 			vcpu->nextpc = vme->pc;
1210 			error = vm_handle_paging(vcpu, &retu);
1211 			break;
1212 		case VM_EXITCODE_BOGUS:
1213 			vcpu->nextpc = vme->pc;
1214 			retu = false;
1215 			error = 0;
1216 			break;
1217 		case VM_EXITCODE_SUSPENDED:
1218 			vcpu->nextpc = vme->pc;
1219 			error = vm_handle_suspend(vcpu, &retu);
1220 			break;
1221 		default:
1222 			/* Handle in userland. */
1223 			vcpu->nextpc = vme->pc;
1224 			retu = true;
1225 			break;
1226 		}
1227 	}
1228 
1229 	if (error == 0 && retu == false)
1230 		goto restart;
1231 
1232 	return (error);
1233 }
1234