xref: /netbsd-src/sys/dev/nvmm/nvmm.c (revision 181254a7b1bdde6873432bffef2d2decc4b5c22f)
1 /*	$NetBSD: nvmm.c,v 1.39 2020/09/05 07:22:25 maxv Exp $	*/
2 
3 /*
4  * Copyright (c) 2018-2020 Maxime Villard, m00nbsd.net
5  * All rights reserved.
6  *
7  * This code is part of the NVMM hypervisor.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
23  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
24  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
25  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
26  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __KERNEL_RCSID(0, "$NetBSD: nvmm.c,v 1.39 2020/09/05 07:22:25 maxv Exp $");
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/kernel.h>
37 
38 #include <sys/atomic.h>
39 #include <sys/cpu.h>
40 #include <sys/conf.h>
41 #include <sys/kmem.h>
42 #include <sys/module.h>
43 #include <sys/proc.h>
44 #include <sys/mman.h>
45 #include <sys/file.h>
46 #include <sys/filedesc.h>
47 #include <sys/device.h>
48 
49 #include <uvm/uvm.h>
50 #include <uvm/uvm_page.h>
51 
52 #include "ioconf.h"
53 
54 #include <dev/nvmm/nvmm.h>
55 #include <dev/nvmm/nvmm_internal.h>
56 #include <dev/nvmm/nvmm_ioctl.h>
57 
58 static struct nvmm_machine machines[NVMM_MAX_MACHINES];
59 static volatile unsigned int nmachines __cacheline_aligned;
60 
61 static const struct nvmm_impl *nvmm_impl_list[] = {
62 #if defined(__x86_64__)
63 	&nvmm_x86_svm,	/* x86 AMD SVM */
64 	&nvmm_x86_vmx	/* x86 Intel VMX */
65 #endif
66 };
67 
68 static const struct nvmm_impl *nvmm_impl __read_mostly = NULL;
69 
70 static struct nvmm_owner root_owner;
71 
72 /* -------------------------------------------------------------------------- */
73 
74 static int
75 nvmm_machine_alloc(struct nvmm_machine **ret)
76 {
77 	struct nvmm_machine *mach;
78 	size_t i;
79 
80 	for (i = 0; i < NVMM_MAX_MACHINES; i++) {
81 		mach = &machines[i];
82 
83 		rw_enter(&mach->lock, RW_WRITER);
84 		if (mach->present) {
85 			rw_exit(&mach->lock);
86 			continue;
87 		}
88 
89 		mach->present = true;
90 		mach->time = time_second;
91 		*ret = mach;
92 		atomic_inc_uint(&nmachines);
93 		return 0;
94 	}
95 
96 	return ENOBUFS;
97 }
98 
99 static void
100 nvmm_machine_free(struct nvmm_machine *mach)
101 {
102 	KASSERT(rw_write_held(&mach->lock));
103 	KASSERT(mach->present);
104 	mach->present = false;
105 	atomic_dec_uint(&nmachines);
106 }
107 
108 static int
109 nvmm_machine_get(struct nvmm_owner *owner, nvmm_machid_t machid,
110     struct nvmm_machine **ret, bool writer)
111 {
112 	struct nvmm_machine *mach;
113 	krw_t op = writer ? RW_WRITER : RW_READER;
114 
115 	if (__predict_false(machid >= NVMM_MAX_MACHINES)) {
116 		return EINVAL;
117 	}
118 	mach = &machines[machid];
119 
120 	rw_enter(&mach->lock, op);
121 	if (__predict_false(!mach->present)) {
122 		rw_exit(&mach->lock);
123 		return ENOENT;
124 	}
125 	if (__predict_false(mach->owner != owner && owner != &root_owner)) {
126 		rw_exit(&mach->lock);
127 		return EPERM;
128 	}
129 	*ret = mach;
130 
131 	return 0;
132 }
133 
134 static void
135 nvmm_machine_put(struct nvmm_machine *mach)
136 {
137 	rw_exit(&mach->lock);
138 }
139 
140 /* -------------------------------------------------------------------------- */
141 
142 static int
143 nvmm_vcpu_alloc(struct nvmm_machine *mach, nvmm_cpuid_t cpuid,
144     struct nvmm_cpu **ret)
145 {
146 	struct nvmm_cpu *vcpu;
147 
148 	if (cpuid >= NVMM_MAX_VCPUS) {
149 		return EINVAL;
150 	}
151 	vcpu = &mach->cpus[cpuid];
152 
153 	mutex_enter(&vcpu->lock);
154 	if (vcpu->present) {
155 		mutex_exit(&vcpu->lock);
156 		return EBUSY;
157 	}
158 
159 	vcpu->present = true;
160 	vcpu->comm = NULL;
161 	vcpu->hcpu_last = -1;
162 	*ret = vcpu;
163 	return 0;
164 }
165 
166 static void
167 nvmm_vcpu_free(struct nvmm_machine *mach, struct nvmm_cpu *vcpu)
168 {
169 	KASSERT(mutex_owned(&vcpu->lock));
170 	vcpu->present = false;
171 	if (vcpu->comm != NULL) {
172 		uvm_deallocate(kernel_map, (vaddr_t)vcpu->comm, PAGE_SIZE);
173 	}
174 }
175 
176 static int
177 nvmm_vcpu_get(struct nvmm_machine *mach, nvmm_cpuid_t cpuid,
178     struct nvmm_cpu **ret)
179 {
180 	struct nvmm_cpu *vcpu;
181 
182 	if (__predict_false(cpuid >= NVMM_MAX_VCPUS)) {
183 		return EINVAL;
184 	}
185 	vcpu = &mach->cpus[cpuid];
186 
187 	mutex_enter(&vcpu->lock);
188 	if (__predict_false(!vcpu->present)) {
189 		mutex_exit(&vcpu->lock);
190 		return ENOENT;
191 	}
192 	*ret = vcpu;
193 
194 	return 0;
195 }
196 
197 static void
198 nvmm_vcpu_put(struct nvmm_cpu *vcpu)
199 {
200 	mutex_exit(&vcpu->lock);
201 }
202 
203 /* -------------------------------------------------------------------------- */
204 
205 static void
206 nvmm_kill_machines(struct nvmm_owner *owner)
207 {
208 	struct nvmm_machine *mach;
209 	struct nvmm_cpu *vcpu;
210 	size_t i, j;
211 	int error;
212 
213 	for (i = 0; i < NVMM_MAX_MACHINES; i++) {
214 		mach = &machines[i];
215 
216 		rw_enter(&mach->lock, RW_WRITER);
217 		if (!mach->present || mach->owner != owner) {
218 			rw_exit(&mach->lock);
219 			continue;
220 		}
221 
222 		/* Kill it. */
223 		for (j = 0; j < NVMM_MAX_VCPUS; j++) {
224 			error = nvmm_vcpu_get(mach, j, &vcpu);
225 			if (error)
226 				continue;
227 			(*nvmm_impl->vcpu_destroy)(mach, vcpu);
228 			nvmm_vcpu_free(mach, vcpu);
229 			nvmm_vcpu_put(vcpu);
230 			atomic_dec_uint(&mach->ncpus);
231 		}
232 		(*nvmm_impl->machine_destroy)(mach);
233 		uvmspace_free(mach->vm);
234 
235 		/* Drop the kernel UOBJ refs. */
236 		for (j = 0; j < NVMM_MAX_HMAPPINGS; j++) {
237 			if (!mach->hmap[j].present)
238 				continue;
239 			uao_detach(mach->hmap[j].uobj);
240 		}
241 
242 		nvmm_machine_free(mach);
243 
244 		rw_exit(&mach->lock);
245 	}
246 }
247 
248 /* -------------------------------------------------------------------------- */
249 
250 static int
251 nvmm_capability(struct nvmm_owner *owner, struct nvmm_ioc_capability *args)
252 {
253 	args->cap.version = NVMM_KERN_VERSION;
254 	args->cap.state_size = nvmm_impl->state_size;
255 	args->cap.max_machines = NVMM_MAX_MACHINES;
256 	args->cap.max_vcpus = NVMM_MAX_VCPUS;
257 	args->cap.max_ram = NVMM_MAX_RAM;
258 
259 	(*nvmm_impl->capability)(&args->cap);
260 
261 	return 0;
262 }
263 
264 static int
265 nvmm_machine_create(struct nvmm_owner *owner,
266     struct nvmm_ioc_machine_create *args)
267 {
268 	struct nvmm_machine *mach;
269 	int error;
270 
271 	error = nvmm_machine_alloc(&mach);
272 	if (error)
273 		return error;
274 
275 	/* Curproc owns the machine. */
276 	mach->owner = owner;
277 
278 	/* Zero out the host mappings. */
279 	memset(&mach->hmap, 0, sizeof(mach->hmap));
280 
281 	/* Create the machine vmspace. */
282 	mach->gpa_begin = 0;
283 	mach->gpa_end = NVMM_MAX_RAM;
284 	mach->vm = uvmspace_alloc(0, mach->gpa_end - mach->gpa_begin, false);
285 
286 	/* Create the comm uobj. */
287 	mach->commuobj = uao_create(NVMM_MAX_VCPUS * PAGE_SIZE, 0);
288 
289 	(*nvmm_impl->machine_create)(mach);
290 
291 	args->machid = mach->machid;
292 	nvmm_machine_put(mach);
293 
294 	return 0;
295 }
296 
297 static int
298 nvmm_machine_destroy(struct nvmm_owner *owner,
299     struct nvmm_ioc_machine_destroy *args)
300 {
301 	struct nvmm_machine *mach;
302 	struct nvmm_cpu *vcpu;
303 	int error;
304 	size_t i;
305 
306 	error = nvmm_machine_get(owner, args->machid, &mach, true);
307 	if (error)
308 		return error;
309 
310 	for (i = 0; i < NVMM_MAX_VCPUS; i++) {
311 		error = nvmm_vcpu_get(mach, i, &vcpu);
312 		if (error)
313 			continue;
314 
315 		(*nvmm_impl->vcpu_destroy)(mach, vcpu);
316 		nvmm_vcpu_free(mach, vcpu);
317 		nvmm_vcpu_put(vcpu);
318 		atomic_dec_uint(&mach->ncpus);
319 	}
320 
321 	(*nvmm_impl->machine_destroy)(mach);
322 
323 	/* Free the machine vmspace. */
324 	uvmspace_free(mach->vm);
325 
326 	/* Drop the kernel UOBJ refs. */
327 	for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
328 		if (!mach->hmap[i].present)
329 			continue;
330 		uao_detach(mach->hmap[i].uobj);
331 	}
332 
333 	nvmm_machine_free(mach);
334 	nvmm_machine_put(mach);
335 
336 	return 0;
337 }
338 
339 static int
340 nvmm_machine_configure(struct nvmm_owner *owner,
341     struct nvmm_ioc_machine_configure *args)
342 {
343 	struct nvmm_machine *mach;
344 	size_t allocsz;
345 	uint64_t op;
346 	void *data;
347 	int error;
348 
349 	op = NVMM_MACH_CONF_MD(args->op);
350 	if (__predict_false(op >= nvmm_impl->mach_conf_max)) {
351 		return EINVAL;
352 	}
353 
354 	allocsz = nvmm_impl->mach_conf_sizes[op];
355 	data = kmem_alloc(allocsz, KM_SLEEP);
356 
357 	error = nvmm_machine_get(owner, args->machid, &mach, true);
358 	if (error) {
359 		kmem_free(data, allocsz);
360 		return error;
361 	}
362 
363 	error = copyin(args->conf, data, allocsz);
364 	if (error) {
365 		goto out;
366 	}
367 
368 	error = (*nvmm_impl->machine_configure)(mach, op, data);
369 
370 out:
371 	nvmm_machine_put(mach);
372 	kmem_free(data, allocsz);
373 	return error;
374 }
375 
376 static int
377 nvmm_vcpu_create(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_create *args)
378 {
379 	struct nvmm_machine *mach;
380 	struct nvmm_cpu *vcpu;
381 	int error;
382 
383 	error = nvmm_machine_get(owner, args->machid, &mach, false);
384 	if (error)
385 		return error;
386 
387 	error = nvmm_vcpu_alloc(mach, args->cpuid, &vcpu);
388 	if (error)
389 		goto out;
390 
391 	/* Allocate the comm page. */
392 	uao_reference(mach->commuobj);
393 	error = uvm_map(kernel_map, (vaddr_t *)&vcpu->comm, PAGE_SIZE,
394 	    mach->commuobj, args->cpuid * PAGE_SIZE, 0, UVM_MAPFLAG(UVM_PROT_RW,
395 	    UVM_PROT_RW, UVM_INH_SHARE, UVM_ADV_RANDOM, 0));
396 	if (error) {
397 		uao_detach(mach->commuobj);
398 		nvmm_vcpu_free(mach, vcpu);
399 		nvmm_vcpu_put(vcpu);
400 		goto out;
401 	}
402 	error = uvm_map_pageable(kernel_map, (vaddr_t)vcpu->comm,
403 	    (vaddr_t)vcpu->comm + PAGE_SIZE, false, 0);
404 	if (error) {
405 		nvmm_vcpu_free(mach, vcpu);
406 		nvmm_vcpu_put(vcpu);
407 		goto out;
408 	}
409 	memset(vcpu->comm, 0, PAGE_SIZE);
410 
411 	error = (*nvmm_impl->vcpu_create)(mach, vcpu);
412 	if (error) {
413 		nvmm_vcpu_free(mach, vcpu);
414 		nvmm_vcpu_put(vcpu);
415 		goto out;
416 	}
417 
418 	nvmm_vcpu_put(vcpu);
419 	atomic_inc_uint(&mach->ncpus);
420 
421 out:
422 	nvmm_machine_put(mach);
423 	return error;
424 }
425 
426 static int
427 nvmm_vcpu_destroy(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_destroy *args)
428 {
429 	struct nvmm_machine *mach;
430 	struct nvmm_cpu *vcpu;
431 	int error;
432 
433 	error = nvmm_machine_get(owner, args->machid, &mach, false);
434 	if (error)
435 		return error;
436 
437 	error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
438 	if (error)
439 		goto out;
440 
441 	(*nvmm_impl->vcpu_destroy)(mach, vcpu);
442 	nvmm_vcpu_free(mach, vcpu);
443 	nvmm_vcpu_put(vcpu);
444 	atomic_dec_uint(&mach->ncpus);
445 
446 out:
447 	nvmm_machine_put(mach);
448 	return error;
449 }
450 
451 static int
452 nvmm_vcpu_configure(struct nvmm_owner *owner,
453     struct nvmm_ioc_vcpu_configure *args)
454 {
455 	struct nvmm_machine *mach;
456 	struct nvmm_cpu *vcpu;
457 	size_t allocsz;
458 	uint64_t op;
459 	void *data;
460 	int error;
461 
462 	op = NVMM_VCPU_CONF_MD(args->op);
463 	if (__predict_false(op >= nvmm_impl->vcpu_conf_max))
464 		return EINVAL;
465 
466 	allocsz = nvmm_impl->vcpu_conf_sizes[op];
467 	data = kmem_alloc(allocsz, KM_SLEEP);
468 
469 	error = nvmm_machine_get(owner, args->machid, &mach, false);
470 	if (error) {
471 		kmem_free(data, allocsz);
472 		return error;
473 	}
474 
475 	error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
476 	if (error) {
477 		nvmm_machine_put(mach);
478 		kmem_free(data, allocsz);
479 		return error;
480 	}
481 
482 	error = copyin(args->conf, data, allocsz);
483 	if (error) {
484 		goto out;
485 	}
486 
487 	error = (*nvmm_impl->vcpu_configure)(vcpu, op, data);
488 
489 out:
490 	nvmm_vcpu_put(vcpu);
491 	nvmm_machine_put(mach);
492 	kmem_free(data, allocsz);
493 	return error;
494 }
495 
496 static int
497 nvmm_vcpu_setstate(struct nvmm_owner *owner,
498     struct nvmm_ioc_vcpu_setstate *args)
499 {
500 	struct nvmm_machine *mach;
501 	struct nvmm_cpu *vcpu;
502 	int error;
503 
504 	error = nvmm_machine_get(owner, args->machid, &mach, false);
505 	if (error)
506 		return error;
507 
508 	error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
509 	if (error)
510 		goto out;
511 
512 	(*nvmm_impl->vcpu_setstate)(vcpu);
513 	nvmm_vcpu_put(vcpu);
514 
515 out:
516 	nvmm_machine_put(mach);
517 	return error;
518 }
519 
520 static int
521 nvmm_vcpu_getstate(struct nvmm_owner *owner,
522     struct nvmm_ioc_vcpu_getstate *args)
523 {
524 	struct nvmm_machine *mach;
525 	struct nvmm_cpu *vcpu;
526 	int error;
527 
528 	error = nvmm_machine_get(owner, args->machid, &mach, false);
529 	if (error)
530 		return error;
531 
532 	error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
533 	if (error)
534 		goto out;
535 
536 	(*nvmm_impl->vcpu_getstate)(vcpu);
537 	nvmm_vcpu_put(vcpu);
538 
539 out:
540 	nvmm_machine_put(mach);
541 	return error;
542 }
543 
544 static int
545 nvmm_vcpu_inject(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_inject *args)
546 {
547 	struct nvmm_machine *mach;
548 	struct nvmm_cpu *vcpu;
549 	int error;
550 
551 	error = nvmm_machine_get(owner, args->machid, &mach, false);
552 	if (error)
553 		return error;
554 
555 	error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
556 	if (error)
557 		goto out;
558 
559 	error = (*nvmm_impl->vcpu_inject)(vcpu);
560 	nvmm_vcpu_put(vcpu);
561 
562 out:
563 	nvmm_machine_put(mach);
564 	return error;
565 }
566 
567 static int
568 nvmm_do_vcpu_run(struct nvmm_machine *mach, struct nvmm_cpu *vcpu,
569     struct nvmm_vcpu_exit *exit)
570 {
571 	struct vmspace *vm = mach->vm;
572 	int ret;
573 
574 	while (1) {
575 		/* Got a signal? Or pending resched? Leave. */
576 		if (__predict_false(nvmm_return_needed())) {
577 			exit->reason = NVMM_VCPU_EXIT_NONE;
578 			return 0;
579 		}
580 
581 		/* Run the VCPU. */
582 		ret = (*nvmm_impl->vcpu_run)(mach, vcpu, exit);
583 		if (__predict_false(ret != 0)) {
584 			return ret;
585 		}
586 
587 		/* Process nested page faults. */
588 		if (__predict_true(exit->reason != NVMM_VCPU_EXIT_MEMORY)) {
589 			break;
590 		}
591 		if (exit->u.mem.gpa >= mach->gpa_end) {
592 			break;
593 		}
594 		if (uvm_fault(&vm->vm_map, exit->u.mem.gpa, exit->u.mem.prot)) {
595 			break;
596 		}
597 	}
598 
599 	return 0;
600 }
601 
602 static int
603 nvmm_vcpu_run(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_run *args)
604 {
605 	struct nvmm_machine *mach;
606 	struct nvmm_cpu *vcpu;
607 	int error;
608 
609 	error = nvmm_machine_get(owner, args->machid, &mach, false);
610 	if (error)
611 		return error;
612 
613 	error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
614 	if (error)
615 		goto out;
616 
617 	error = nvmm_do_vcpu_run(mach, vcpu, &args->exit);
618 	nvmm_vcpu_put(vcpu);
619 
620 out:
621 	nvmm_machine_put(mach);
622 	return error;
623 }
624 
625 /* -------------------------------------------------------------------------- */
626 
627 static struct uvm_object *
628 nvmm_hmapping_getuobj(struct nvmm_machine *mach, uintptr_t hva, size_t size,
629    size_t *off)
630 {
631 	struct nvmm_hmapping *hmapping;
632 	size_t i;
633 
634 	for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
635 		hmapping = &mach->hmap[i];
636 		if (!hmapping->present) {
637 			continue;
638 		}
639 		if (hva >= hmapping->hva &&
640 		    hva + size <= hmapping->hva + hmapping->size) {
641 			*off = hva - hmapping->hva;
642 			return hmapping->uobj;
643 		}
644 	}
645 
646 	return NULL;
647 }
648 
649 static int
650 nvmm_hmapping_validate(struct nvmm_machine *mach, uintptr_t hva, size_t size)
651 {
652 	struct nvmm_hmapping *hmapping;
653 	size_t i;
654 
655 	if ((hva % PAGE_SIZE) != 0 || (size % PAGE_SIZE) != 0) {
656 		return EINVAL;
657 	}
658 	if (hva == 0) {
659 		return EINVAL;
660 	}
661 
662 	for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
663 		hmapping = &mach->hmap[i];
664 		if (!hmapping->present) {
665 			continue;
666 		}
667 
668 		if (hva >= hmapping->hva &&
669 		    hva + size <= hmapping->hva + hmapping->size) {
670 			break;
671 		}
672 
673 		if (hva >= hmapping->hva &&
674 		    hva < hmapping->hva + hmapping->size) {
675 			return EEXIST;
676 		}
677 		if (hva + size > hmapping->hva &&
678 		    hva + size <= hmapping->hva + hmapping->size) {
679 			return EEXIST;
680 		}
681 		if (hva <= hmapping->hva &&
682 		    hva + size >= hmapping->hva + hmapping->size) {
683 			return EEXIST;
684 		}
685 	}
686 
687 	return 0;
688 }
689 
690 static struct nvmm_hmapping *
691 nvmm_hmapping_alloc(struct nvmm_machine *mach)
692 {
693 	struct nvmm_hmapping *hmapping;
694 	size_t i;
695 
696 	for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
697 		hmapping = &mach->hmap[i];
698 		if (!hmapping->present) {
699 			hmapping->present = true;
700 			return hmapping;
701 		}
702 	}
703 
704 	return NULL;
705 }
706 
707 static int
708 nvmm_hmapping_free(struct nvmm_machine *mach, uintptr_t hva, size_t size)
709 {
710 	struct vmspace *vmspace = curproc->p_vmspace;
711 	struct nvmm_hmapping *hmapping;
712 	size_t i;
713 
714 	for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
715 		hmapping = &mach->hmap[i];
716 		if (!hmapping->present || hmapping->hva != hva ||
717 		    hmapping->size != size) {
718 			continue;
719 		}
720 
721 		uvm_unmap(&vmspace->vm_map, hmapping->hva,
722 		    hmapping->hva + hmapping->size);
723 		uao_detach(hmapping->uobj);
724 
725 		hmapping->uobj = NULL;
726 		hmapping->present = false;
727 
728 		return 0;
729 	}
730 
731 	return ENOENT;
732 }
733 
734 static int
735 nvmm_hva_map(struct nvmm_owner *owner, struct nvmm_ioc_hva_map *args)
736 {
737 	struct vmspace *vmspace = curproc->p_vmspace;
738 	struct nvmm_machine *mach;
739 	struct nvmm_hmapping *hmapping;
740 	vaddr_t uva;
741 	int error;
742 
743 	error = nvmm_machine_get(owner, args->machid, &mach, true);
744 	if (error)
745 		return error;
746 
747 	error = nvmm_hmapping_validate(mach, args->hva, args->size);
748 	if (error)
749 		goto out;
750 
751 	hmapping = nvmm_hmapping_alloc(mach);
752 	if (hmapping == NULL) {
753 		error = ENOBUFS;
754 		goto out;
755 	}
756 
757 	hmapping->hva = args->hva;
758 	hmapping->size = args->size;
759 	hmapping->uobj = uao_create(hmapping->size, 0);
760 	uva = hmapping->hva;
761 
762 	/* Take a reference for the user. */
763 	uao_reference(hmapping->uobj);
764 
765 	/* Map the uobj into the user address space, as pageable. */
766 	error = uvm_map(&vmspace->vm_map, &uva, hmapping->size, hmapping->uobj,
767 	    0, 0, UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW, UVM_INH_SHARE,
768 	    UVM_ADV_RANDOM, UVM_FLAG_FIXED|UVM_FLAG_UNMAP));
769 	if (error) {
770 		uao_detach(hmapping->uobj);
771 	}
772 
773 out:
774 	nvmm_machine_put(mach);
775 	return error;
776 }
777 
778 static int
779 nvmm_hva_unmap(struct nvmm_owner *owner, struct nvmm_ioc_hva_unmap *args)
780 {
781 	struct nvmm_machine *mach;
782 	int error;
783 
784 	error = nvmm_machine_get(owner, args->machid, &mach, true);
785 	if (error)
786 		return error;
787 
788 	error = nvmm_hmapping_free(mach, args->hva, args->size);
789 
790 	nvmm_machine_put(mach);
791 	return error;
792 }
793 
794 /* -------------------------------------------------------------------------- */
795 
796 static int
797 nvmm_gpa_map(struct nvmm_owner *owner, struct nvmm_ioc_gpa_map *args)
798 {
799 	struct nvmm_machine *mach;
800 	struct uvm_object *uobj;
801 	gpaddr_t gpa;
802 	size_t off;
803 	int error;
804 
805 	error = nvmm_machine_get(owner, args->machid, &mach, false);
806 	if (error)
807 		return error;
808 
809 	if ((args->prot & ~(PROT_READ|PROT_WRITE|PROT_EXEC)) != 0) {
810 		error = EINVAL;
811 		goto out;
812 	}
813 
814 	if ((args->gpa % PAGE_SIZE) != 0 || (args->size % PAGE_SIZE) != 0 ||
815 	    (args->hva % PAGE_SIZE) != 0) {
816 		error = EINVAL;
817 		goto out;
818 	}
819 	if (args->hva == 0) {
820 		error = EINVAL;
821 		goto out;
822 	}
823 	if (args->gpa < mach->gpa_begin || args->gpa >= mach->gpa_end) {
824 		error = EINVAL;
825 		goto out;
826 	}
827 	if (args->gpa + args->size <= args->gpa) {
828 		error = EINVAL;
829 		goto out;
830 	}
831 	if (args->gpa + args->size > mach->gpa_end) {
832 		error = EINVAL;
833 		goto out;
834 	}
835 	gpa = args->gpa;
836 
837 	uobj = nvmm_hmapping_getuobj(mach, args->hva, args->size, &off);
838 	if (uobj == NULL) {
839 		error = EINVAL;
840 		goto out;
841 	}
842 
843 	/* Take a reference for the machine. */
844 	uao_reference(uobj);
845 
846 	/* Map the uobj into the machine address space, as pageable. */
847 	error = uvm_map(&mach->vm->vm_map, &gpa, args->size, uobj, off, 0,
848 	    UVM_MAPFLAG(args->prot, UVM_PROT_RWX, UVM_INH_NONE,
849 	    UVM_ADV_RANDOM, UVM_FLAG_FIXED|UVM_FLAG_UNMAP));
850 	if (error) {
851 		uao_detach(uobj);
852 		goto out;
853 	}
854 	if (gpa != args->gpa) {
855 		uao_detach(uobj);
856 		printf("[!] uvm_map problem\n");
857 		error = EINVAL;
858 		goto out;
859 	}
860 
861 out:
862 	nvmm_machine_put(mach);
863 	return error;
864 }
865 
866 static int
867 nvmm_gpa_unmap(struct nvmm_owner *owner, struct nvmm_ioc_gpa_unmap *args)
868 {
869 	struct nvmm_machine *mach;
870 	gpaddr_t gpa;
871 	int error;
872 
873 	error = nvmm_machine_get(owner, args->machid, &mach, false);
874 	if (error)
875 		return error;
876 
877 	if ((args->gpa % PAGE_SIZE) != 0 || (args->size % PAGE_SIZE) != 0) {
878 		error = EINVAL;
879 		goto out;
880 	}
881 	if (args->gpa < mach->gpa_begin || args->gpa >= mach->gpa_end) {
882 		error = EINVAL;
883 		goto out;
884 	}
885 	if (args->gpa + args->size <= args->gpa) {
886 		error = EINVAL;
887 		goto out;
888 	}
889 	if (args->gpa + args->size >= mach->gpa_end) {
890 		error = EINVAL;
891 		goto out;
892 	}
893 	gpa = args->gpa;
894 
895 	/* Unmap the memory from the machine. */
896 	uvm_unmap(&mach->vm->vm_map, gpa, gpa + args->size);
897 
898 out:
899 	nvmm_machine_put(mach);
900 	return error;
901 }
902 
903 /* -------------------------------------------------------------------------- */
904 
905 static int
906 nvmm_ctl_mach_info(struct nvmm_owner *owner, struct nvmm_ioc_ctl *args)
907 {
908 	struct nvmm_ctl_mach_info ctl;
909 	struct nvmm_machine *mach;
910 	int error;
911 	size_t i;
912 
913 	if (args->size != sizeof(ctl))
914 		return EINVAL;
915 	error = copyin(args->data, &ctl, sizeof(ctl));
916 	if (error)
917 		return error;
918 
919 	error = nvmm_machine_get(owner, ctl.machid, &mach, true);
920 	if (error)
921 		return error;
922 
923 	ctl.nvcpus = mach->ncpus;
924 
925 	ctl.nram = 0;
926 	for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
927 		if (!mach->hmap[i].present)
928 			continue;
929 		ctl.nram += mach->hmap[i].size;
930 	}
931 
932 	ctl.pid = mach->owner->pid;
933 	ctl.time = mach->time;
934 
935 	nvmm_machine_put(mach);
936 
937 	error = copyout(&ctl, args->data, sizeof(ctl));
938 	if (error)
939 		return error;
940 
941 	return 0;
942 }
943 
944 static int
945 nvmm_ctl(struct nvmm_owner *owner, struct nvmm_ioc_ctl *args)
946 {
947 	switch (args->op) {
948 	case NVMM_CTL_MACH_INFO:
949 		return nvmm_ctl_mach_info(owner, args);
950 	default:
951 		return EINVAL;
952 	}
953 }
954 
955 /* -------------------------------------------------------------------------- */
956 
957 static const struct nvmm_impl *
958 nvmm_ident(void)
959 {
960 	size_t i;
961 
962 	for (i = 0; i < __arraycount(nvmm_impl_list); i++) {
963 		if ((*nvmm_impl_list[i]->ident)())
964 			return nvmm_impl_list[i];
965 	}
966 
967 	return NULL;
968 }
969 
970 static int
971 nvmm_init(void)
972 {
973 	size_t i, n;
974 
975 	nvmm_impl = nvmm_ident();
976 	if (nvmm_impl == NULL)
977 		return ENOTSUP;
978 
979 	for (i = 0; i < NVMM_MAX_MACHINES; i++) {
980 		machines[i].machid = i;
981 		rw_init(&machines[i].lock);
982 		for (n = 0; n < NVMM_MAX_VCPUS; n++) {
983 			machines[i].cpus[n].present = false;
984 			machines[i].cpus[n].cpuid = n;
985 			mutex_init(&machines[i].cpus[n].lock, MUTEX_DEFAULT,
986 			    IPL_NONE);
987 		}
988 	}
989 
990 	(*nvmm_impl->init)();
991 
992 	return 0;
993 }
994 
995 static void
996 nvmm_fini(void)
997 {
998 	size_t i, n;
999 
1000 	for (i = 0; i < NVMM_MAX_MACHINES; i++) {
1001 		rw_destroy(&machines[i].lock);
1002 		for (n = 0; n < NVMM_MAX_VCPUS; n++) {
1003 			mutex_destroy(&machines[i].cpus[n].lock);
1004 		}
1005 	}
1006 
1007 	(*nvmm_impl->fini)();
1008 	nvmm_impl = NULL;
1009 }
1010 
1011 /* -------------------------------------------------------------------------- */
1012 
1013 static dev_type_open(nvmm_open);
1014 
1015 const struct cdevsw nvmm_cdevsw = {
1016 	.d_open = nvmm_open,
1017 	.d_close = noclose,
1018 	.d_read = noread,
1019 	.d_write = nowrite,
1020 	.d_ioctl = noioctl,
1021 	.d_stop = nostop,
1022 	.d_tty = notty,
1023 	.d_poll = nopoll,
1024 	.d_mmap = nommap,
1025 	.d_kqfilter = nokqfilter,
1026 	.d_discard = nodiscard,
1027 	.d_flag = D_OTHER | D_MPSAFE
1028 };
1029 
1030 static int nvmm_ioctl(file_t *, u_long, void *);
1031 static int nvmm_close(file_t *);
1032 static int nvmm_mmap(file_t *, off_t *, size_t, int, int *, int *,
1033     struct uvm_object **, int *);
1034 
1035 static const struct fileops nvmm_fileops = {
1036 	.fo_read = fbadop_read,
1037 	.fo_write = fbadop_write,
1038 	.fo_ioctl = nvmm_ioctl,
1039 	.fo_fcntl = fnullop_fcntl,
1040 	.fo_poll = fnullop_poll,
1041 	.fo_stat = fbadop_stat,
1042 	.fo_close = nvmm_close,
1043 	.fo_kqfilter = fnullop_kqfilter,
1044 	.fo_restart = fnullop_restart,
1045 	.fo_mmap = nvmm_mmap,
1046 };
1047 
1048 static int
1049 nvmm_open(dev_t dev, int flags, int type, struct lwp *l)
1050 {
1051 	struct nvmm_owner *owner;
1052 	struct file *fp;
1053 	int error, fd;
1054 
1055 	if (__predict_false(nvmm_impl == NULL))
1056 		return ENXIO;
1057 	if (minor(dev) != 0)
1058 		return EXDEV;
1059 	if (!(flags & O_CLOEXEC))
1060 		return EINVAL;
1061 	error = fd_allocfile(&fp, &fd);
1062 	if (error)
1063 		return error;
1064 
1065 	if (OFLAGS(flags) & O_WRONLY) {
1066 		owner = &root_owner;
1067 	} else {
1068 		owner = kmem_alloc(sizeof(*owner), KM_SLEEP);
1069 		owner->pid = l->l_proc->p_pid;
1070 	}
1071 
1072 	return fd_clone(fp, fd, flags, &nvmm_fileops, owner);
1073 }
1074 
1075 static int
1076 nvmm_close(file_t *fp)
1077 {
1078 	struct nvmm_owner *owner = fp->f_data;
1079 
1080 	KASSERT(owner != NULL);
1081 	nvmm_kill_machines(owner);
1082 	if (owner != &root_owner) {
1083 		kmem_free(owner, sizeof(*owner));
1084 	}
1085 	fp->f_data = NULL;
1086 
1087    	return 0;
1088 }
1089 
1090 static int
1091 nvmm_mmap(file_t *fp, off_t *offp, size_t size, int prot, int *flagsp,
1092     int *advicep, struct uvm_object **uobjp, int *maxprotp)
1093 {
1094 	struct nvmm_owner *owner = fp->f_data;
1095 	struct nvmm_machine *mach;
1096 	nvmm_machid_t machid;
1097 	nvmm_cpuid_t cpuid;
1098 	int error;
1099 
1100 	if (prot & PROT_EXEC)
1101 		return EACCES;
1102 	if (size != PAGE_SIZE)
1103 		return EINVAL;
1104 
1105 	cpuid = NVMM_COMM_CPUID(*offp);
1106 	if (__predict_false(cpuid >= NVMM_MAX_VCPUS))
1107 		return EINVAL;
1108 
1109 	machid = NVMM_COMM_MACHID(*offp);
1110 	error = nvmm_machine_get(owner, machid, &mach, false);
1111 	if (error)
1112 		return error;
1113 
1114 	uao_reference(mach->commuobj);
1115 	*uobjp = mach->commuobj;
1116 	*offp = cpuid * PAGE_SIZE;
1117 	*maxprotp = prot;
1118 	*advicep = UVM_ADV_RANDOM;
1119 
1120 	nvmm_machine_put(mach);
1121 	return 0;
1122 }
1123 
1124 static int
1125 nvmm_ioctl(file_t *fp, u_long cmd, void *data)
1126 {
1127 	struct nvmm_owner *owner = fp->f_data;
1128 
1129 	KASSERT(owner != NULL);
1130 
1131 	switch (cmd) {
1132 	case NVMM_IOC_CAPABILITY:
1133 		return nvmm_capability(owner, data);
1134 	case NVMM_IOC_MACHINE_CREATE:
1135 		return nvmm_machine_create(owner, data);
1136 	case NVMM_IOC_MACHINE_DESTROY:
1137 		return nvmm_machine_destroy(owner, data);
1138 	case NVMM_IOC_MACHINE_CONFIGURE:
1139 		return nvmm_machine_configure(owner, data);
1140 	case NVMM_IOC_VCPU_CREATE:
1141 		return nvmm_vcpu_create(owner, data);
1142 	case NVMM_IOC_VCPU_DESTROY:
1143 		return nvmm_vcpu_destroy(owner, data);
1144 	case NVMM_IOC_VCPU_CONFIGURE:
1145 		return nvmm_vcpu_configure(owner, data);
1146 	case NVMM_IOC_VCPU_SETSTATE:
1147 		return nvmm_vcpu_setstate(owner, data);
1148 	case NVMM_IOC_VCPU_GETSTATE:
1149 		return nvmm_vcpu_getstate(owner, data);
1150 	case NVMM_IOC_VCPU_INJECT:
1151 		return nvmm_vcpu_inject(owner, data);
1152 	case NVMM_IOC_VCPU_RUN:
1153 		return nvmm_vcpu_run(owner, data);
1154 	case NVMM_IOC_GPA_MAP:
1155 		return nvmm_gpa_map(owner, data);
1156 	case NVMM_IOC_GPA_UNMAP:
1157 		return nvmm_gpa_unmap(owner, data);
1158 	case NVMM_IOC_HVA_MAP:
1159 		return nvmm_hva_map(owner, data);
1160 	case NVMM_IOC_HVA_UNMAP:
1161 		return nvmm_hva_unmap(owner, data);
1162 	case NVMM_IOC_CTL:
1163 		return nvmm_ctl(owner, data);
1164 	default:
1165 		return EINVAL;
1166 	}
1167 }
1168 
1169 /* -------------------------------------------------------------------------- */
1170 
1171 static int nvmm_match(device_t, cfdata_t, void *);
1172 static void nvmm_attach(device_t, device_t, void *);
1173 static int nvmm_detach(device_t, int);
1174 
1175 extern struct cfdriver nvmm_cd;
1176 
1177 CFATTACH_DECL_NEW(nvmm, 0, nvmm_match, nvmm_attach, nvmm_detach, NULL);
1178 
1179 static struct cfdata nvmm_cfdata[] = {
1180 	{
1181 		.cf_name = "nvmm",
1182 		.cf_atname = "nvmm",
1183 		.cf_unit = 0,
1184 		.cf_fstate = FSTATE_STAR,
1185 		.cf_loc = NULL,
1186 		.cf_flags = 0,
1187 		.cf_pspec = NULL,
1188 	},
1189 	{ NULL, NULL, 0, FSTATE_NOTFOUND, NULL, 0, NULL }
1190 };
1191 
1192 static int
1193 nvmm_match(device_t self, cfdata_t cfdata, void *arg)
1194 {
1195 	return 1;
1196 }
1197 
1198 static void
1199 nvmm_attach(device_t parent, device_t self, void *aux)
1200 {
1201 	int error;
1202 
1203 	error = nvmm_init();
1204 	if (error)
1205 		panic("%s: impossible", __func__);
1206 	aprint_normal_dev(self, "attached, using backend %s\n",
1207 	    nvmm_impl->name);
1208 }
1209 
1210 static int
1211 nvmm_detach(device_t self, int flags)
1212 {
1213 	if (atomic_load_relaxed(&nmachines) > 0)
1214 		return EBUSY;
1215 	nvmm_fini();
1216 	return 0;
1217 }
1218 
1219 void
1220 nvmmattach(int nunits)
1221 {
1222 	/* nothing */
1223 }
1224 
1225 MODULE(MODULE_CLASS_MISC, nvmm, NULL);
1226 
1227 #if defined(_MODULE)
1228 CFDRIVER_DECL(nvmm, DV_VIRTUAL, NULL);
1229 #endif
1230 
1231 static int
1232 nvmm_modcmd(modcmd_t cmd, void *arg)
1233 {
1234 #if defined(_MODULE)
1235 	devmajor_t bmajor = NODEVMAJOR;
1236 	devmajor_t cmajor = 345;
1237 #endif
1238 	int error;
1239 
1240 	switch (cmd) {
1241 	case MODULE_CMD_INIT:
1242 		if (nvmm_ident() == NULL) {
1243 			aprint_error("%s: cpu not supported\n",
1244 			    nvmm_cd.cd_name);
1245 			return ENOTSUP;
1246 		}
1247 #if defined(_MODULE)
1248 		error = config_cfdriver_attach(&nvmm_cd);
1249 		if (error)
1250 			return error;
1251 #endif
1252 		error = config_cfattach_attach(nvmm_cd.cd_name, &nvmm_ca);
1253 		if (error) {
1254 			config_cfdriver_detach(&nvmm_cd);
1255 			aprint_error("%s: config_cfattach_attach failed\n",
1256 			    nvmm_cd.cd_name);
1257 			return error;
1258 		}
1259 
1260 		error = config_cfdata_attach(nvmm_cfdata, 1);
1261 		if (error) {
1262 			config_cfattach_detach(nvmm_cd.cd_name, &nvmm_ca);
1263 			config_cfdriver_detach(&nvmm_cd);
1264 			aprint_error("%s: unable to register cfdata\n",
1265 			    nvmm_cd.cd_name);
1266 			return error;
1267 		}
1268 
1269 		if (config_attach_pseudo(nvmm_cfdata) == NULL) {
1270 			aprint_error("%s: config_attach_pseudo failed\n",
1271 			    nvmm_cd.cd_name);
1272 			config_cfattach_detach(nvmm_cd.cd_name, &nvmm_ca);
1273 			config_cfdriver_detach(&nvmm_cd);
1274 			return ENXIO;
1275 		}
1276 
1277 #if defined(_MODULE)
1278 		/* mknod /dev/nvmm c 345 0 */
1279 		error = devsw_attach(nvmm_cd.cd_name, NULL, &bmajor,
1280 			&nvmm_cdevsw, &cmajor);
1281 		if (error) {
1282 			aprint_error("%s: unable to register devsw\n",
1283 			    nvmm_cd.cd_name);
1284 			config_cfattach_detach(nvmm_cd.cd_name, &nvmm_ca);
1285 			config_cfdriver_detach(&nvmm_cd);
1286 			return error;
1287 		}
1288 #endif
1289 		return 0;
1290 	case MODULE_CMD_FINI:
1291 		error = config_cfdata_detach(nvmm_cfdata);
1292 		if (error)
1293 			return error;
1294 		error = config_cfattach_detach(nvmm_cd.cd_name, &nvmm_ca);
1295 		if (error)
1296 			return error;
1297 #if defined(_MODULE)
1298 		config_cfdriver_detach(&nvmm_cd);
1299 		devsw_detach(NULL, &nvmm_cdevsw);
1300 #endif
1301 		return 0;
1302 	case MODULE_CMD_AUTOUNLOAD:
1303 		return EBUSY;
1304 	default:
1305 		return ENOTTY;
1306 	}
1307 }
1308