xref: /netbsd-src/lib/libnvmm/libnvmm.c (revision 154bfe8e089c1a0a4e9ed8414f08d3da90949162) 
1 /*	$NetBSD: libnvmm.c,v 1.19 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 
33 #include <stdio.h>
34 #include <stdlib.h>
35 #include <string.h>
36 #include <unistd.h>
37 #include <fcntl.h>
38 #include <errno.h>
39 #include <sys/ioctl.h>
40 #include <sys/mman.h>
41 #include <sys/queue.h>
42 #include <machine/vmparam.h>
43 
44 #include "nvmm.h"
45 
46 static struct nvmm_capability __capability;
47 
48 #ifdef __x86_64__
49 #include "libnvmm_x86.c"
50 #endif
51 
52 typedef struct __area {
53 	LIST_ENTRY(__area) list;
54 	gpaddr_t gpa;
55 	uintptr_t hva;
56 	size_t size;
57 	nvmm_prot_t prot;
58 } area_t;
59 
60 typedef LIST_HEAD(, __area) area_list_t;
61 
62 static int nvmm_fd = -1;
63 
64 /* -------------------------------------------------------------------------- */
65 
66 static bool
67 __area_isvalid(struct nvmm_machine *mach, uintptr_t hva, gpaddr_t gpa,
68     size_t size)
69 {
70 	area_list_t *areas = mach->areas;
71 	area_t *ent;
72 
73 	LIST_FOREACH(ent, areas, list) {
74 		/* Collision on GPA */
75 		if (gpa >= ent->gpa && gpa < ent->gpa + ent->size) {
76 			return false;
77 		}
78 		if (gpa + size > ent->gpa &&
79 		    gpa + size <= ent->gpa + ent->size) {
80 			return false;
81 		}
82 		if (gpa <= ent->gpa && gpa + size >= ent->gpa + ent->size) {
83 			return false;
84 		}
85 	}
86 
87 	return true;
88 }
89 
90 static int
91 __area_add(struct nvmm_machine *mach, uintptr_t hva, gpaddr_t gpa, size_t size,
92     int prot)
93 {
94 	area_list_t *areas = mach->areas;
95 	nvmm_prot_t nprot;
96 	area_t *area;
97 
98 	nprot = 0;
99 	if (prot & PROT_READ)
100 		nprot |= NVMM_PROT_READ;
101 	if (prot & PROT_WRITE)
102 		nprot |= NVMM_PROT_WRITE;
103 	if (prot & PROT_EXEC)
104 		nprot |= NVMM_PROT_EXEC;
105 
106 	if (!__area_isvalid(mach, hva, gpa, size)) {
107 		errno = EINVAL;
108 		return -1;
109 	}
110 
111 	area = malloc(sizeof(*area));
112 	if (area == NULL)
113 		return -1;
114 	area->gpa = gpa;
115 	area->hva = hva;
116 	area->size = size;
117 	area->prot = nprot;
118 
119 	LIST_INSERT_HEAD(areas, area, list);
120 
121 	return 0;
122 }
123 
124 static int
125 __area_delete(struct nvmm_machine *mach, uintptr_t hva, gpaddr_t gpa,
126     size_t size)
127 {
128 	area_list_t *areas = mach->areas;
129 	area_t *ent, *nxt;
130 
131 	LIST_FOREACH_SAFE(ent, areas, list, nxt) {
132 		if (hva == ent->hva && gpa == ent->gpa && size == ent->size) {
133 			LIST_REMOVE(ent, list);
134 			free(ent);
135 			return 0;
136 		}
137 	}
138 
139 	return -1;
140 }
141 
142 static void
143 __area_remove_all(struct nvmm_machine *mach)
144 {
145 	area_list_t *areas = mach->areas;
146 	area_t *ent;
147 
148 	while ((ent = LIST_FIRST(areas)) != NULL) {
149 		LIST_REMOVE(ent, list);
150 		free(ent);
151 	}
152 
153 	free(areas);
154 }
155 
156 /* -------------------------------------------------------------------------- */
157 
158 int
159 nvmm_init(void)
160 {
161 	if (nvmm_fd != -1)
162 		return 0;
163 	nvmm_fd = open("/dev/nvmm", O_RDONLY | O_CLOEXEC);
164 	if (nvmm_fd == -1)
165 		return -1;
166 	if (nvmm_capability(&__capability) == -1) {
167 		close(nvmm_fd);
168 		nvmm_fd = -1;
169 		return -1;
170 	}
171 	if (__capability.version != NVMM_KERN_VERSION) {
172 		close(nvmm_fd);
173 		nvmm_fd = -1;
174 		errno = EPROGMISMATCH;
175 		return -1;
176 	}
177 
178 	return 0;
179 }
180 
181 int
182 nvmm_root_init(void)
183 {
184 	if (nvmm_fd != -1)
185 		return 0;
186 	nvmm_fd = open("/dev/nvmm", O_WRONLY | O_CLOEXEC);
187 	if (nvmm_fd == -1)
188 		return -1;
189 	if (nvmm_capability(&__capability) == -1) {
190 		close(nvmm_fd);
191 		nvmm_fd = -1;
192 		return -1;
193 	}
194 	if (__capability.version != NVMM_KERN_VERSION) {
195 		close(nvmm_fd);
196 		nvmm_fd = -1;
197 		errno = EPROGMISMATCH;
198 		return -1;
199 	}
200 
201 	return 0;
202 }
203 
204 int
205 nvmm_capability(struct nvmm_capability *cap)
206 {
207 	struct nvmm_ioc_capability args;
208 	int ret;
209 
210 	ret = ioctl(nvmm_fd, NVMM_IOC_CAPABILITY, &args);
211 	if (ret == -1)
212 		return -1;
213 
214 	memcpy(cap, &args.cap, sizeof(args.cap));
215 
216 	return 0;
217 }
218 
219 int
220 nvmm_machine_create(struct nvmm_machine *mach)
221 {
222 	struct nvmm_ioc_machine_create args;
223 	struct nvmm_comm_page **pages;
224 	area_list_t *areas;
225 	int ret;
226 
227 	areas = calloc(1, sizeof(*areas));
228 	if (areas == NULL)
229 		return -1;
230 
231 	pages = calloc(__capability.max_vcpus, sizeof(*pages));
232 	if (pages == NULL) {
233 		free(areas);
234 		return -1;
235 	}
236 
237 	ret = ioctl(nvmm_fd, NVMM_IOC_MACHINE_CREATE, &args);
238 	if (ret == -1) {
239 		free(areas);
240 		return -1;
241 	}
242 
243 	LIST_INIT(areas);
244 
245 	memset(mach, 0, sizeof(*mach));
246 	mach->machid = args.machid;
247 	mach->pages = pages;
248 	mach->areas = areas;
249 
250 	return 0;
251 }
252 
253 int
254 nvmm_machine_destroy(struct nvmm_machine *mach)
255 {
256 	struct nvmm_ioc_machine_destroy args;
257 	int ret;
258 
259 	args.machid = mach->machid;
260 
261 	ret = ioctl(nvmm_fd, NVMM_IOC_MACHINE_DESTROY, &args);
262 	if (ret == -1)
263 		return -1;
264 
265 	__area_remove_all(mach);
266 	free(mach->pages);
267 
268 	return 0;
269 }
270 
271 int
272 nvmm_machine_configure(struct nvmm_machine *mach, uint64_t op, void *conf)
273 {
274 	struct nvmm_ioc_machine_configure args;
275 	int ret;
276 
277 	args.machid = mach->machid;
278 	args.op = op;
279 	args.conf = conf;
280 
281 	ret = ioctl(nvmm_fd, NVMM_IOC_MACHINE_CONFIGURE, &args);
282 	if (ret == -1)
283 		return -1;
284 
285 	return 0;
286 }
287 
288 int
289 nvmm_vcpu_create(struct nvmm_machine *mach, nvmm_cpuid_t cpuid,
290     struct nvmm_vcpu *vcpu)
291 {
292 	struct nvmm_ioc_vcpu_create args;
293 	struct nvmm_comm_page *comm;
294 	int ret;
295 
296 	args.machid = mach->machid;
297 	args.cpuid = cpuid;
298 
299 	ret = ioctl(nvmm_fd, NVMM_IOC_VCPU_CREATE, &args);
300 	if (ret == -1)
301 		return -1;
302 
303 	comm = mmap(NULL, PAGE_SIZE, PROT_READ|PROT_WRITE, MAP_SHARED|MAP_FILE,
304 	    nvmm_fd, NVMM_COMM_OFF(mach->machid, cpuid));
305 	if (comm == MAP_FAILED)
306 		return -1;
307 
308 	mach->pages[cpuid] = comm;
309 
310 	vcpu->cpuid = cpuid;
311 	vcpu->state = &comm->state;
312 	vcpu->event = &comm->event;
313 	vcpu->exit = malloc(sizeof(*vcpu->exit));
314 
315 	return 0;
316 }
317 
318 int
319 nvmm_vcpu_destroy(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
320 {
321 	struct nvmm_ioc_vcpu_destroy args;
322 	struct nvmm_comm_page *comm;
323 	int ret;
324 
325 	args.machid = mach->machid;
326 	args.cpuid = vcpu->cpuid;
327 
328 	ret = ioctl(nvmm_fd, NVMM_IOC_VCPU_DESTROY, &args);
329 	if (ret == -1)
330 		return -1;
331 
332 	comm = mach->pages[vcpu->cpuid];
333 	munmap(comm, PAGE_SIZE);
334 	free(vcpu->exit);
335 
336 	return 0;
337 }
338 
339 int
340 nvmm_vcpu_configure(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu,
341     uint64_t op, void *conf)
342 {
343 	struct nvmm_ioc_vcpu_configure args;
344 	int ret;
345 
346 	switch (op) {
347 	case NVMM_VCPU_CONF_CALLBACKS:
348 		memcpy(&vcpu->cbs, conf, sizeof(vcpu->cbs));
349 		return 0;
350 	}
351 
352 	args.machid = mach->machid;
353 	args.cpuid = vcpu->cpuid;
354 	args.op = op;
355 	args.conf = conf;
356 
357 	ret = ioctl(nvmm_fd, NVMM_IOC_VCPU_CONFIGURE, &args);
358 	if (ret == -1)
359 		return -1;
360 
361 	return 0;
362 }
363 
364 int
365 nvmm_vcpu_setstate(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu,
366     uint64_t flags)
367 {
368 	struct nvmm_comm_page *comm;
369 
370 	comm = mach->pages[vcpu->cpuid];
371 	comm->state_commit |= flags;
372 	comm->state_cached |= flags;
373 
374 	return 0;
375 }
376 
377 int
378 nvmm_vcpu_getstate(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu,
379     uint64_t flags)
380 {
381 	struct nvmm_ioc_vcpu_getstate args;
382 	struct nvmm_comm_page *comm;
383 	int ret;
384 
385 	comm = mach->pages[vcpu->cpuid];
386 
387 	if (__predict_true((flags & ~comm->state_cached) == 0)) {
388 		return 0;
389 	}
390 	comm->state_wanted = flags & ~comm->state_cached;
391 
392 	args.machid = mach->machid;
393 	args.cpuid = vcpu->cpuid;
394 
395 	ret = ioctl(nvmm_fd, NVMM_IOC_VCPU_GETSTATE, &args);
396 	if (ret == -1)
397 		return -1;
398 
399 	return 0;
400 }
401 
402 int
403 nvmm_vcpu_inject(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
404 {
405 	struct nvmm_comm_page *comm;
406 
407 	comm = mach->pages[vcpu->cpuid];
408 	comm->event_commit = true;
409 
410 	return 0;
411 }
412 
413 int
414 nvmm_vcpu_run(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
415 {
416 	struct nvmm_ioc_vcpu_run args;
417 	int ret;
418 
419 	args.machid = mach->machid;
420 	args.cpuid = vcpu->cpuid;
421 	memset(&args.exit, 0, sizeof(args.exit));
422 
423 	ret = ioctl(nvmm_fd, NVMM_IOC_VCPU_RUN, &args);
424 	if (ret == -1)
425 		return -1;
426 
427 	/* No comm support yet, just copy. */
428 	memcpy(vcpu->exit, &args.exit, sizeof(args.exit));
429 
430 	return 0;
431 }
432 
433 int
434 nvmm_gpa_map(struct nvmm_machine *mach, uintptr_t hva, gpaddr_t gpa,
435     size_t size, int prot)
436 {
437 	struct nvmm_ioc_gpa_map args;
438 	int ret;
439 
440 	ret = __area_add(mach, hva, gpa, size, prot);
441 	if (ret == -1)
442 		return -1;
443 
444 	args.machid = mach->machid;
445 	args.hva = hva;
446 	args.gpa = gpa;
447 	args.size = size;
448 	args.prot = prot;
449 
450 	ret = ioctl(nvmm_fd, NVMM_IOC_GPA_MAP, &args);
451 	if (ret == -1) {
452 		/* Can't recover. */
453 		abort();
454 	}
455 
456 	return 0;
457 }
458 
459 int
460 nvmm_gpa_unmap(struct nvmm_machine *mach, uintptr_t hva, gpaddr_t gpa,
461     size_t size)
462 {
463 	struct nvmm_ioc_gpa_unmap args;
464 	int ret;
465 
466 	ret = __area_delete(mach, hva, gpa, size);
467 	if (ret == -1)
468 		return -1;
469 
470 	args.machid = mach->machid;
471 	args.gpa = gpa;
472 	args.size = size;
473 
474 	ret = ioctl(nvmm_fd, NVMM_IOC_GPA_UNMAP, &args);
475 	if (ret == -1) {
476 		/* Can't recover. */
477 		abort();
478 	}
479 
480 	return 0;
481 }
482 
483 int
484 nvmm_hva_map(struct nvmm_machine *mach, uintptr_t hva, size_t size)
485 {
486 	struct nvmm_ioc_hva_map args;
487 	int ret;
488 
489 	args.machid = mach->machid;
490 	args.hva = hva;
491 	args.size = size;
492 
493 	ret = ioctl(nvmm_fd, NVMM_IOC_HVA_MAP, &args);
494 	if (ret == -1)
495 		return -1;
496 
497 	return 0;
498 }
499 
500 int
501 nvmm_hva_unmap(struct nvmm_machine *mach, uintptr_t hva, size_t size)
502 {
503 	struct nvmm_ioc_hva_unmap args;
504 	int ret;
505 
506 	args.machid = mach->machid;
507 	args.hva = hva;
508 	args.size = size;
509 
510 	ret = ioctl(nvmm_fd, NVMM_IOC_HVA_UNMAP, &args);
511 	if (ret == -1)
512 		return -1;
513 
514 	return 0;
515 }
516 
517 /*
518  * nvmm_gva_to_gpa(): architecture-specific.
519  */
520 
521 int
522 nvmm_gpa_to_hva(struct nvmm_machine *mach, gpaddr_t gpa, uintptr_t *hva,
523     nvmm_prot_t *prot)
524 {
525 	area_list_t *areas = mach->areas;
526 	area_t *ent;
527 
528 	LIST_FOREACH(ent, areas, list) {
529 		if (gpa >= ent->gpa && gpa < ent->gpa + ent->size) {
530 			*hva = ent->hva + (gpa - ent->gpa);
531 			*prot = ent->prot;
532 			return 0;
533 		}
534 	}
535 
536 	errno = ENOENT;
537 	return -1;
538 }
539 
540 /*
541  * nvmm_assist_io(): architecture-specific.
542  */
543 
544 /*
545  * nvmm_assist_mem(): architecture-specific.
546  */
547 
548 int
549 nvmm_ctl(int op, void *data, size_t size)
550 {
551 	struct nvmm_ioc_ctl args;
552 	int ret;
553 
554 	args.op = op;
555 	args.data = data;
556 	args.size = size;
557 
558 	ret = ioctl(nvmm_fd, NVMM_IOC_CTL, &args);
559 	if (ret == -1)
560 		return -1;
561 
562 	return 0;
563 }
564