xref: /freebsd/usr.sbin/bhyve/mem.c (revision e08e9e999091f86081377b7cedc3fd2fe2ab70fc)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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/tree.h>
42 #include <sys/errno.h>
43 #include <machine/vmm.h>
44 #include <machine/vmm_instruction_emul.h>
45 
46 #include <stdio.h>
47 #include <stdlib.h>
48 #include <assert.h>
49 #include <pthread.h>
50 
51 #include "mem.h"
52 
53 struct mmio_rb_range {
54 	RB_ENTRY(mmio_rb_range)	mr_link;	/* RB tree links */
55 	struct mem_range	mr_param;
56 	uint64_t                mr_base;
57 	uint64_t                mr_end;
58 };
59 
60 struct mmio_rb_tree;
61 RB_PROTOTYPE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare);
62 
63 RB_HEAD(mmio_rb_tree, mmio_rb_range) mmio_rb_root, mmio_rb_fallback;
64 
65 /*
66  * Per-vCPU cache. Since most accesses from a vCPU will be to
67  * consecutive addresses in a range, it makes sense to cache the
68  * result of a lookup.
69  */
70 static struct mmio_rb_range	*mmio_hint[VM_MAXCPU];
71 
72 static pthread_rwlock_t mmio_rwlock;
73 
74 static int
75 mmio_rb_range_compare(struct mmio_rb_range *a, struct mmio_rb_range *b)
76 {
77 	if (a->mr_end < b->mr_base)
78 		return (-1);
79 	else if (a->mr_base > b->mr_end)
80 		return (1);
81 	return (0);
82 }
83 
84 static int
85 mmio_rb_lookup(struct mmio_rb_tree *rbt, uint64_t addr,
86     struct mmio_rb_range **entry)
87 {
88 	struct mmio_rb_range find, *res;
89 
90 	find.mr_base = find.mr_end = addr;
91 
92 	res = RB_FIND(mmio_rb_tree, rbt, &find);
93 
94 	if (res != NULL) {
95 		*entry = res;
96 		return (0);
97 	}
98 
99 	return (ENOENT);
100 }
101 
102 static int
103 mmio_rb_add(struct mmio_rb_tree *rbt, struct mmio_rb_range *new)
104 {
105 	struct mmio_rb_range *overlap;
106 
107 	overlap = RB_INSERT(mmio_rb_tree, rbt, new);
108 
109 	if (overlap != NULL) {
110 #ifdef RB_DEBUG
111 		printf("overlap detected: new %lx:%lx, tree %lx:%lx\n",
112 		       new->mr_base, new->mr_end,
113 		       overlap->mr_base, overlap->mr_end);
114 #endif
115 
116 		return (EEXIST);
117 	}
118 
119 	return (0);
120 }
121 
122 #if 0
123 static void
124 mmio_rb_dump(struct mmio_rb_tree *rbt)
125 {
126 	int perror;
127 	struct mmio_rb_range *np;
128 
129 	pthread_rwlock_rdlock(&mmio_rwlock);
130 	RB_FOREACH(np, mmio_rb_tree, rbt) {
131 		printf(" %lx:%lx, %s\n", np->mr_base, np->mr_end,
132 		       np->mr_param.name);
133 	}
134 	perror = pthread_rwlock_unlock(&mmio_rwlock);
135 	assert(perror == 0);
136 }
137 #endif
138 
139 RB_GENERATE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare);
140 
141 typedef int (mem_cb_t)(struct vmctx *ctx, int vcpu, uint64_t gpa,
142     struct mem_range *mr, void *arg);
143 
144 static int
145 mem_read(void *ctx, int vcpu, uint64_t gpa, uint64_t *rval, int size, void *arg)
146 {
147 	int error;
148 	struct mem_range *mr = arg;
149 
150 	error = (*mr->handler)(ctx, vcpu, MEM_F_READ, gpa, size,
151 			       rval, mr->arg1, mr->arg2);
152 	return (error);
153 }
154 
155 static int
156 mem_write(void *ctx, int vcpu, uint64_t gpa, uint64_t wval, int size, void *arg)
157 {
158 	int error;
159 	struct mem_range *mr = arg;
160 
161 	error = (*mr->handler)(ctx, vcpu, MEM_F_WRITE, gpa, size,
162 			       &wval, mr->arg1, mr->arg2);
163 	return (error);
164 }
165 
166 static int
167 access_memory(struct vmctx *ctx, int vcpu, uint64_t paddr, mem_cb_t *cb,
168     void *arg)
169 {
170 	struct mmio_rb_range *entry;
171 	int err, perror, immutable;
172 
173 	pthread_rwlock_rdlock(&mmio_rwlock);
174 	/*
175 	 * First check the per-vCPU cache
176 	 */
177 	if (mmio_hint[vcpu] &&
178 	    paddr >= mmio_hint[vcpu]->mr_base &&
179 	    paddr <= mmio_hint[vcpu]->mr_end) {
180 		entry = mmio_hint[vcpu];
181 	} else
182 		entry = NULL;
183 
184 	if (entry == NULL) {
185 		if (mmio_rb_lookup(&mmio_rb_root, paddr, &entry) == 0) {
186 			/* Update the per-vCPU cache */
187 			mmio_hint[vcpu] = entry;
188 		} else if (mmio_rb_lookup(&mmio_rb_fallback, paddr, &entry)) {
189 			perror = pthread_rwlock_unlock(&mmio_rwlock);
190 			assert(perror == 0);
191 			return (ESRCH);
192 		}
193 	}
194 
195 	assert(entry != NULL);
196 
197 	/*
198 	 * An 'immutable' memory range is guaranteed to be never removed
199 	 * so there is no need to hold 'mmio_rwlock' while calling the
200 	 * handler.
201 	 *
202 	 * XXX writes to the PCIR_COMMAND register can cause register_mem()
203 	 * to be called. If the guest is using PCI extended config space
204 	 * to modify the PCIR_COMMAND register then register_mem() can
205 	 * deadlock on 'mmio_rwlock'. However by registering the extended
206 	 * config space window as 'immutable' the deadlock can be avoided.
207 	 */
208 	immutable = (entry->mr_param.flags & MEM_F_IMMUTABLE);
209 	if (immutable) {
210 		perror = pthread_rwlock_unlock(&mmio_rwlock);
211 		assert(perror == 0);
212 	}
213 
214 	err = cb(ctx, vcpu, paddr, &entry->mr_param, arg);
215 
216 	if (!immutable) {
217 		perror = pthread_rwlock_unlock(&mmio_rwlock);
218 		assert(perror == 0);
219 	}
220 
221 
222 	return (err);
223 }
224 
225 struct emulate_mem_args {
226 	struct vie *vie;
227 	struct vm_guest_paging *paging;
228 };
229 
230 static int
231 emulate_mem_cb(struct vmctx *ctx, int vcpu, uint64_t paddr, struct mem_range *mr,
232     void *arg)
233 {
234 	struct emulate_mem_args *ema;
235 
236 	ema = arg;
237 	return (vmm_emulate_instruction(ctx, vcpu, paddr, ema->vie, ema->paging,
238 	    mem_read, mem_write, mr));
239 }
240 
241 int
242 emulate_mem(struct vmctx *ctx, int vcpu, uint64_t paddr, struct vie *vie,
243     struct vm_guest_paging *paging)
244 
245 {
246 	struct emulate_mem_args ema;
247 
248 	ema.vie = vie;
249 	ema.paging = paging;
250 	return (access_memory(ctx, vcpu, paddr, emulate_mem_cb, &ema));
251 }
252 
253 struct read_mem_args {
254 	uint64_t *rval;
255 	int size;
256 };
257 
258 static int
259 read_mem_cb(struct vmctx *ctx, int vcpu, uint64_t paddr, struct mem_range *mr,
260     void *arg)
261 {
262 	struct read_mem_args *rma;
263 
264 	rma = arg;
265 	return (mr->handler(ctx, vcpu, MEM_F_READ, paddr, rma->size,
266 	    rma->rval, mr->arg1, mr->arg2));
267 }
268 
269 int
270 read_mem(struct vmctx *ctx, int vcpu, uint64_t gpa, uint64_t *rval, int size)
271 {
272 	struct read_mem_args rma;
273 
274 	rma.rval = rval;
275 	rma.size = size;
276 	return (access_memory(ctx, vcpu, gpa, read_mem_cb, &rma));
277 }
278 
279 static int
280 register_mem_int(struct mmio_rb_tree *rbt, struct mem_range *memp)
281 {
282 	struct mmio_rb_range *entry, *mrp;
283 	int err, perror;
284 
285 	err = 0;
286 
287 	mrp = malloc(sizeof(struct mmio_rb_range));
288 
289 	if (mrp != NULL) {
290 		mrp->mr_param = *memp;
291 		mrp->mr_base = memp->base;
292 		mrp->mr_end = memp->base + memp->size - 1;
293 		pthread_rwlock_wrlock(&mmio_rwlock);
294 		if (mmio_rb_lookup(rbt, memp->base, &entry) != 0)
295 			err = mmio_rb_add(rbt, mrp);
296 		perror = pthread_rwlock_unlock(&mmio_rwlock);
297 		assert(perror == 0);
298 		if (err)
299 			free(mrp);
300 	} else
301 		err = ENOMEM;
302 
303 	return (err);
304 }
305 
306 int
307 register_mem(struct mem_range *memp)
308 {
309 
310 	return (register_mem_int(&mmio_rb_root, memp));
311 }
312 
313 int
314 register_mem_fallback(struct mem_range *memp)
315 {
316 
317 	return (register_mem_int(&mmio_rb_fallback, memp));
318 }
319 
320 int
321 unregister_mem(struct mem_range *memp)
322 {
323 	struct mem_range *mr;
324 	struct mmio_rb_range *entry = NULL;
325 	int err, perror, i;
326 
327 	pthread_rwlock_wrlock(&mmio_rwlock);
328 	err = mmio_rb_lookup(&mmio_rb_root, memp->base, &entry);
329 	if (err == 0) {
330 		mr = &entry->mr_param;
331 		assert(mr->name == memp->name);
332 		assert(mr->base == memp->base && mr->size == memp->size);
333 		assert((mr->flags & MEM_F_IMMUTABLE) == 0);
334 		RB_REMOVE(mmio_rb_tree, &mmio_rb_root, entry);
335 
336 		/* flush Per-vCPU cache */
337 		for (i=0; i < VM_MAXCPU; i++) {
338 			if (mmio_hint[i] == entry)
339 				mmio_hint[i] = NULL;
340 		}
341 	}
342 	perror = pthread_rwlock_unlock(&mmio_rwlock);
343 	assert(perror == 0);
344 
345 	if (entry)
346 		free(entry);
347 
348 	return (err);
349 }
350 
351 void
352 init_mem(void)
353 {
354 
355 	RB_INIT(&mmio_rb_root);
356 	RB_INIT(&mmio_rb_fallback);
357 	pthread_rwlock_init(&mmio_rwlock, NULL);
358 }
359