xref: /freebsd/usr.sbin/bhyve/mem.c (revision 488ab515d6cc02f6f743f0badfc8e94eb553cd30)
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 	struct mmio_rb_range *np;
127 
128 	pthread_rwlock_rdlock(&mmio_rwlock);
129 	RB_FOREACH(np, mmio_rb_tree, rbt) {
130 		printf(" %lx:%lx, %s\n", np->mr_base, np->mr_end,
131 		       np->mr_param.name);
132 	}
133 	pthread_rwlock_unlock(&mmio_rwlock);
134 }
135 #endif
136 
137 RB_GENERATE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare);
138 
139 typedef int (mem_cb_t)(struct vmctx *ctx, int vcpu, uint64_t gpa,
140     struct mem_range *mr, void *arg);
141 
142 static int
143 mem_read(void *ctx, int vcpu, uint64_t gpa, uint64_t *rval, int size, void *arg)
144 {
145 	int error;
146 	struct mem_range *mr = arg;
147 
148 	error = (*mr->handler)(ctx, vcpu, MEM_F_READ, gpa, size,
149 			       rval, mr->arg1, mr->arg2);
150 	return (error);
151 }
152 
153 static int
154 mem_write(void *ctx, int vcpu, uint64_t gpa, uint64_t wval, int size, void *arg)
155 {
156 	int error;
157 	struct mem_range *mr = arg;
158 
159 	error = (*mr->handler)(ctx, vcpu, MEM_F_WRITE, gpa, size,
160 			       &wval, mr->arg1, mr->arg2);
161 	return (error);
162 }
163 
164 static int
165 access_memory(struct vmctx *ctx, int vcpu, uint64_t paddr, mem_cb_t *cb,
166     void *arg)
167 {
168 	struct mmio_rb_range *entry;
169 	int err, immutable;
170 
171 	pthread_rwlock_rdlock(&mmio_rwlock);
172 	/*
173 	 * First check the per-vCPU cache
174 	 */
175 	if (mmio_hint[vcpu] &&
176 	    paddr >= mmio_hint[vcpu]->mr_base &&
177 	    paddr <= mmio_hint[vcpu]->mr_end) {
178 		entry = mmio_hint[vcpu];
179 	} else
180 		entry = NULL;
181 
182 	if (entry == NULL) {
183 		if (mmio_rb_lookup(&mmio_rb_root, paddr, &entry) == 0) {
184 			/* Update the per-vCPU cache */
185 			mmio_hint[vcpu] = entry;
186 		} else if (mmio_rb_lookup(&mmio_rb_fallback, paddr, &entry)) {
187 			pthread_rwlock_unlock(&mmio_rwlock);
188 			return (ESRCH);
189 		}
190 	}
191 
192 	assert(entry != NULL);
193 
194 	/*
195 	 * An 'immutable' memory range is guaranteed to be never removed
196 	 * so there is no need to hold 'mmio_rwlock' while calling the
197 	 * handler.
198 	 *
199 	 * XXX writes to the PCIR_COMMAND register can cause register_mem()
200 	 * to be called. If the guest is using PCI extended config space
201 	 * to modify the PCIR_COMMAND register then register_mem() can
202 	 * deadlock on 'mmio_rwlock'. However by registering the extended
203 	 * config space window as 'immutable' the deadlock can be avoided.
204 	 */
205 	immutable = (entry->mr_param.flags & MEM_F_IMMUTABLE);
206 	if (immutable)
207 		pthread_rwlock_unlock(&mmio_rwlock);
208 
209 	err = cb(ctx, vcpu, paddr, &entry->mr_param, arg);
210 
211 	if (!immutable)
212 		pthread_rwlock_unlock(&mmio_rwlock);
213 
214 	return (err);
215 }
216 
217 struct emulate_mem_args {
218 	struct vie *vie;
219 	struct vm_guest_paging *paging;
220 };
221 
222 static int
223 emulate_mem_cb(struct vmctx *ctx, int vcpu, uint64_t paddr, struct mem_range *mr,
224     void *arg)
225 {
226 	struct emulate_mem_args *ema;
227 
228 	ema = arg;
229 	return (vmm_emulate_instruction(ctx, vcpu, paddr, ema->vie, ema->paging,
230 	    mem_read, mem_write, mr));
231 }
232 
233 int
234 emulate_mem(struct vmctx *ctx, int vcpu, uint64_t paddr, struct vie *vie,
235     struct vm_guest_paging *paging)
236 
237 {
238 	struct emulate_mem_args ema;
239 
240 	ema.vie = vie;
241 	ema.paging = paging;
242 	return (access_memory(ctx, vcpu, paddr, emulate_mem_cb, &ema));
243 }
244 
245 struct read_mem_args {
246 	uint64_t *rval;
247 	int size;
248 };
249 
250 static int
251 read_mem_cb(struct vmctx *ctx, int vcpu, uint64_t paddr, struct mem_range *mr,
252     void *arg)
253 {
254 	struct read_mem_args *rma;
255 
256 	rma = arg;
257 	return (mr->handler(ctx, vcpu, MEM_F_READ, paddr, rma->size,
258 	    rma->rval, mr->arg1, mr->arg2));
259 }
260 
261 int
262 read_mem(struct vmctx *ctx, int vcpu, uint64_t gpa, uint64_t *rval, int size)
263 {
264 	struct read_mem_args rma;
265 
266 	rma.rval = rval;
267 	rma.size = size;
268 	return (access_memory(ctx, vcpu, gpa, read_mem_cb, &rma));
269 }
270 
271 static int
272 register_mem_int(struct mmio_rb_tree *rbt, struct mem_range *memp)
273 {
274 	struct mmio_rb_range *entry, *mrp;
275 	int		err;
276 
277 	err = 0;
278 
279 	mrp = malloc(sizeof(struct mmio_rb_range));
280 
281 	if (mrp != NULL) {
282 		mrp->mr_param = *memp;
283 		mrp->mr_base = memp->base;
284 		mrp->mr_end = memp->base + memp->size - 1;
285 		pthread_rwlock_wrlock(&mmio_rwlock);
286 		if (mmio_rb_lookup(rbt, memp->base, &entry) != 0)
287 			err = mmio_rb_add(rbt, mrp);
288 		pthread_rwlock_unlock(&mmio_rwlock);
289 		if (err)
290 			free(mrp);
291 	} else
292 		err = ENOMEM;
293 
294 	return (err);
295 }
296 
297 int
298 register_mem(struct mem_range *memp)
299 {
300 
301 	return (register_mem_int(&mmio_rb_root, memp));
302 }
303 
304 int
305 register_mem_fallback(struct mem_range *memp)
306 {
307 
308 	return (register_mem_int(&mmio_rb_fallback, memp));
309 }
310 
311 int
312 unregister_mem(struct mem_range *memp)
313 {
314 	struct mem_range *mr;
315 	struct mmio_rb_range *entry = NULL;
316 	int err, i;
317 
318 	pthread_rwlock_wrlock(&mmio_rwlock);
319 	err = mmio_rb_lookup(&mmio_rb_root, memp->base, &entry);
320 	if (err == 0) {
321 		mr = &entry->mr_param;
322 		assert(mr->name == memp->name);
323 		assert(mr->base == memp->base && mr->size == memp->size);
324 		assert((mr->flags & MEM_F_IMMUTABLE) == 0);
325 		RB_REMOVE(mmio_rb_tree, &mmio_rb_root, entry);
326 
327 		/* flush Per-vCPU cache */
328 		for (i=0; i < VM_MAXCPU; i++) {
329 			if (mmio_hint[i] == entry)
330 				mmio_hint[i] = NULL;
331 		}
332 	}
333 	pthread_rwlock_unlock(&mmio_rwlock);
334 
335 	if (entry)
336 		free(entry);
337 
338 	return (err);
339 }
340 
341 void
342 init_mem(void)
343 {
344 
345 	RB_INIT(&mmio_rb_root);
346 	RB_INIT(&mmio_rb_fallback);
347 	pthread_rwlock_init(&mmio_rwlock, NULL);
348 }
349