xref: /freebsd/usr.sbin/bhyve/mem.c (revision 7ef62cebc2f965b0f640263e179276928885e33d)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2012 NetApp, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  * $FreeBSD$
29  */
30 
31 /*
32  * Memory ranges are represented with an RB tree. On insertion, the range
33  * is checked for overlaps. On lookup, the key has the same base and limit
34  * so it can be searched within the range.
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include <sys/types.h>
41 #include <sys/errno.h>
42 #include <sys/tree.h>
43 #include <machine/vmm.h>
44 #include <machine/vmm_instruction_emul.h>
45 
46 #include <assert.h>
47 #include <err.h>
48 #include <pthread.h>
49 #include <stdio.h>
50 #include <stdlib.h>
51 #include <vmmapi.h>
52 
53 #include "mem.h"
54 
55 struct mmio_rb_range {
56 	RB_ENTRY(mmio_rb_range)	mr_link;	/* RB tree links */
57 	struct mem_range	mr_param;
58 	uint64_t                mr_base;
59 	uint64_t                mr_end;
60 };
61 
62 struct mmio_rb_tree;
63 RB_PROTOTYPE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare);
64 
65 static RB_HEAD(mmio_rb_tree, mmio_rb_range) mmio_rb_root, mmio_rb_fallback;
66 
67 /*
68  * Per-vCPU cache. Since most accesses from a vCPU will be to
69  * consecutive addresses in a range, it makes sense to cache the
70  * result of a lookup.
71  */
72 static struct mmio_rb_range	**mmio_hint;
73 static int mmio_ncpu;
74 
75 static pthread_rwlock_t mmio_rwlock;
76 
77 static int
78 mmio_rb_range_compare(struct mmio_rb_range *a, struct mmio_rb_range *b)
79 {
80 	if (a->mr_end < b->mr_base)
81 		return (-1);
82 	else if (a->mr_base > b->mr_end)
83 		return (1);
84 	return (0);
85 }
86 
87 static int
88 mmio_rb_lookup(struct mmio_rb_tree *rbt, uint64_t addr,
89     struct mmio_rb_range **entry)
90 {
91 	struct mmio_rb_range find, *res;
92 
93 	find.mr_base = find.mr_end = addr;
94 
95 	res = RB_FIND(mmio_rb_tree, rbt, &find);
96 
97 	if (res != NULL) {
98 		*entry = res;
99 		return (0);
100 	}
101 
102 	return (ENOENT);
103 }
104 
105 static int
106 mmio_rb_add(struct mmio_rb_tree *rbt, struct mmio_rb_range *new)
107 {
108 	struct mmio_rb_range *overlap;
109 
110 	overlap = RB_INSERT(mmio_rb_tree, rbt, new);
111 
112 	if (overlap != NULL) {
113 #ifdef RB_DEBUG
114 		printf("overlap detected: new %lx:%lx, tree %lx:%lx, '%s' "
115 		       "claims region already claimed for '%s'\n",
116 		       new->mr_base, new->mr_end,
117 		       overlap->mr_base, overlap->mr_end,
118 		       new->mr_param.name, overlap->mr_param.name);
119 #endif
120 
121 		return (EEXIST);
122 	}
123 
124 	return (0);
125 }
126 
127 #if 0
128 static void
129 mmio_rb_dump(struct mmio_rb_tree *rbt)
130 {
131 	int perror;
132 	struct mmio_rb_range *np;
133 
134 	pthread_rwlock_rdlock(&mmio_rwlock);
135 	RB_FOREACH(np, mmio_rb_tree, rbt) {
136 		printf(" %lx:%lx, %s\n", np->mr_base, np->mr_end,
137 		       np->mr_param.name);
138 	}
139 	perror = pthread_rwlock_unlock(&mmio_rwlock);
140 	assert(perror == 0);
141 }
142 #endif
143 
144 RB_GENERATE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare);
145 
146 typedef int (mem_cb_t)(struct vcpu *vcpu, uint64_t gpa, struct mem_range *mr,
147     void *arg);
148 
149 static int
150 mem_read(struct vcpu *vcpu, uint64_t gpa, uint64_t *rval, int size, void *arg)
151 {
152 	int error;
153 	struct mem_range *mr = arg;
154 
155 	error = (*mr->handler)(vcpu, MEM_F_READ, gpa, size, rval, mr->arg1,
156 	    mr->arg2);
157 	return (error);
158 }
159 
160 static int
161 mem_write(struct vcpu *vcpu, uint64_t gpa, uint64_t wval, int size, void *arg)
162 {
163 	int error;
164 	struct mem_range *mr = arg;
165 
166 	error = (*mr->handler)(vcpu, MEM_F_WRITE, gpa, size, &wval, mr->arg1,
167 	    mr->arg2);
168 	return (error);
169 }
170 
171 static int
172 access_memory(struct vcpu *vcpu, uint64_t paddr, mem_cb_t *cb, void *arg)
173 {
174 	struct mmio_rb_range *entry;
175 	int err, perror, immutable, vcpuid;
176 
177 	vcpuid = vcpu_id(vcpu);
178 	pthread_rwlock_rdlock(&mmio_rwlock);
179 	/*
180 	 * First check the per-vCPU cache
181 	 */
182 	if (mmio_hint[vcpuid] &&
183 	    paddr >= mmio_hint[vcpuid]->mr_base &&
184 	    paddr <= mmio_hint[vcpuid]->mr_end) {
185 		entry = mmio_hint[vcpuid];
186 	} else
187 		entry = NULL;
188 
189 	if (entry == NULL) {
190 		if (mmio_rb_lookup(&mmio_rb_root, paddr, &entry) == 0) {
191 			/* Update the per-vCPU cache */
192 			mmio_hint[vcpuid] = entry;
193 		} else if (mmio_rb_lookup(&mmio_rb_fallback, paddr, &entry)) {
194 			perror = pthread_rwlock_unlock(&mmio_rwlock);
195 			assert(perror == 0);
196 			return (ESRCH);
197 		}
198 	}
199 
200 	assert(entry != NULL);
201 
202 	/*
203 	 * An 'immutable' memory range is guaranteed to be never removed
204 	 * so there is no need to hold 'mmio_rwlock' while calling the
205 	 * handler.
206 	 *
207 	 * XXX writes to the PCIR_COMMAND register can cause register_mem()
208 	 * to be called. If the guest is using PCI extended config space
209 	 * to modify the PCIR_COMMAND register then register_mem() can
210 	 * deadlock on 'mmio_rwlock'. However by registering the extended
211 	 * config space window as 'immutable' the deadlock can be avoided.
212 	 */
213 	immutable = (entry->mr_param.flags & MEM_F_IMMUTABLE);
214 	if (immutable) {
215 		perror = pthread_rwlock_unlock(&mmio_rwlock);
216 		assert(perror == 0);
217 	}
218 
219 	err = cb(vcpu, paddr, &entry->mr_param, arg);
220 
221 	if (!immutable) {
222 		perror = pthread_rwlock_unlock(&mmio_rwlock);
223 		assert(perror == 0);
224 	}
225 
226 	return (err);
227 }
228 
229 struct emulate_mem_args {
230 	struct vie *vie;
231 	struct vm_guest_paging *paging;
232 };
233 
234 static int
235 emulate_mem_cb(struct vcpu *vcpu, uint64_t paddr, struct mem_range *mr,
236     void *arg)
237 {
238 	struct emulate_mem_args *ema;
239 
240 	ema = arg;
241 	return (vmm_emulate_instruction(vcpu, paddr, ema->vie, ema->paging,
242 	    mem_read, mem_write, mr));
243 }
244 
245 int
246 emulate_mem(struct vcpu *vcpu, uint64_t paddr, struct vie *vie,
247     struct vm_guest_paging *paging)
248 {
249 	struct emulate_mem_args ema;
250 
251 	ema.vie = vie;
252 	ema.paging = paging;
253 	return (access_memory(vcpu, paddr, emulate_mem_cb, &ema));
254 }
255 
256 struct rw_mem_args {
257 	uint64_t *val;
258 	int size;
259 	int operation;
260 };
261 
262 static int
263 rw_mem_cb(struct vcpu *vcpu, uint64_t paddr, struct mem_range *mr, void *arg)
264 {
265 	struct rw_mem_args *rma;
266 
267 	rma = arg;
268 	return (mr->handler(vcpu, rma->operation, paddr, rma->size,
269 	    rma->val, mr->arg1, mr->arg2));
270 }
271 
272 int
273 read_mem(struct vcpu *vcpu, uint64_t gpa, uint64_t *rval, int size)
274 {
275 	struct rw_mem_args rma;
276 
277 	rma.val = rval;
278 	rma.size = size;
279 	rma.operation = MEM_F_READ;
280 	return (access_memory(vcpu, gpa, rw_mem_cb, &rma));
281 }
282 
283 int
284 write_mem(struct vcpu *vcpu, uint64_t gpa, uint64_t wval, int size)
285 {
286 	struct rw_mem_args rma;
287 
288 	rma.val = &wval;
289 	rma.size = size;
290 	rma.operation = MEM_F_WRITE;
291 	return (access_memory(vcpu, gpa, rw_mem_cb, &rma));
292 }
293 
294 static int
295 register_mem_int(struct mmio_rb_tree *rbt, struct mem_range *memp)
296 {
297 	struct mmio_rb_range *entry, *mrp;
298 	int err, perror;
299 
300 	err = 0;
301 
302 	mrp = malloc(sizeof(struct mmio_rb_range));
303 	if (mrp == NULL) {
304 		warn("%s: couldn't allocate memory for mrp\n",
305 		     __func__);
306 		err = ENOMEM;
307 	} else {
308 		mrp->mr_param = *memp;
309 		mrp->mr_base = memp->base;
310 		mrp->mr_end = memp->base + memp->size - 1;
311 		pthread_rwlock_wrlock(&mmio_rwlock);
312 		if (mmio_rb_lookup(rbt, memp->base, &entry) != 0)
313 			err = mmio_rb_add(rbt, mrp);
314 		perror = pthread_rwlock_unlock(&mmio_rwlock);
315 		assert(perror == 0);
316 		if (err)
317 			free(mrp);
318 	}
319 
320 	return (err);
321 }
322 
323 int
324 register_mem(struct mem_range *memp)
325 {
326 
327 	return (register_mem_int(&mmio_rb_root, memp));
328 }
329 
330 int
331 register_mem_fallback(struct mem_range *memp)
332 {
333 
334 	return (register_mem_int(&mmio_rb_fallback, memp));
335 }
336 
337 int
338 unregister_mem(struct mem_range *memp)
339 {
340 	struct mem_range *mr;
341 	struct mmio_rb_range *entry = NULL;
342 	int err, perror, i;
343 
344 	pthread_rwlock_wrlock(&mmio_rwlock);
345 	err = mmio_rb_lookup(&mmio_rb_root, memp->base, &entry);
346 	if (err == 0) {
347 		mr = &entry->mr_param;
348 		assert(mr->name == memp->name);
349 		assert(mr->base == memp->base && mr->size == memp->size);
350 		assert((mr->flags & MEM_F_IMMUTABLE) == 0);
351 		RB_REMOVE(mmio_rb_tree, &mmio_rb_root, entry);
352 
353 		/* flush Per-vCPU cache */
354 		for (i = 0; i < mmio_ncpu; i++) {
355 			if (mmio_hint[i] == entry)
356 				mmio_hint[i] = NULL;
357 		}
358 	}
359 	perror = pthread_rwlock_unlock(&mmio_rwlock);
360 	assert(perror == 0);
361 
362 	if (entry)
363 		free(entry);
364 
365 	return (err);
366 }
367 
368 void
369 init_mem(int ncpu)
370 {
371 
372 	mmio_ncpu = ncpu;
373 	mmio_hint = calloc(ncpu, sizeof(*mmio_hint));
374 	RB_INIT(&mmio_rb_root);
375 	RB_INIT(&mmio_rb_fallback);
376 	pthread_rwlock_init(&mmio_rwlock, NULL);
377 }
378