xref: /illumos-gate/usr/src/cmd/bhyve/mem.c (revision e0c0d44e)
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  * This file and its contents are supplied under the terms of the
32  * Common Development and Distribution License ("CDDL"), version 1.0.
33  * You may only use this file in accordance with the terms of version
34  * 1.0 of the CDDL.
35  *
36  * A full copy of the text of the CDDL should have accompanied this
37  * source.  A copy of the CDDL is also available via the Internet at
38  * http://www.illumos.org/license/CDDL.
39  *
40  * Copyright 2020 Oxide Computer Company
41  */
42 
43 /*
44  * Memory ranges are represented with an RB tree. On insertion, the range
45  * is checked for overlaps. On lookup, the key has the same base and limit
46  * so it can be searched within the range.
47  */
48 
49 #include <sys/cdefs.h>
50 __FBSDID("$FreeBSD$");
51 
52 #include <sys/types.h>
53 #include <sys/errno.h>
54 #include <sys/tree.h>
55 #include <machine/vmm.h>
56 
57 #include <assert.h>
58 #include <err.h>
59 #include <pthread.h>
60 #include <stdio.h>
61 #include <stdlib.h>
62 
63 #include "mem.h"
64 
65 struct mmio_rb_range {
66 	RB_ENTRY(mmio_rb_range)	mr_link;	/* RB tree links */
67 	struct mem_range	mr_param;
68 	uint64_t                mr_base;
69 	uint64_t                mr_end;
70 };
71 
72 struct mmio_rb_tree;
73 RB_PROTOTYPE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare);
74 
75 RB_HEAD(mmio_rb_tree, mmio_rb_range) mmio_rb_root, mmio_rb_fallback;
76 
77 /*
78  * Per-vCPU cache. Since most accesses from a vCPU will be to
79  * consecutive addresses in a range, it makes sense to cache the
80  * result of a lookup.
81  */
82 static struct mmio_rb_range	*mmio_hint[VM_MAXCPU];
83 
84 static pthread_rwlock_t mmio_rwlock;
85 
86 static int
mmio_rb_range_compare(struct mmio_rb_range * a,struct mmio_rb_range * b)87 mmio_rb_range_compare(struct mmio_rb_range *a, struct mmio_rb_range *b)
88 {
89 	if (a->mr_end < b->mr_base)
90 		return (-1);
91 	else if (a->mr_base > b->mr_end)
92 		return (1);
93 	return (0);
94 }
95 
96 static int
mmio_rb_lookup(struct mmio_rb_tree * rbt,uint64_t addr,struct mmio_rb_range ** entry)97 mmio_rb_lookup(struct mmio_rb_tree *rbt, uint64_t addr,
98     struct mmio_rb_range **entry)
99 {
100 	struct mmio_rb_range find, *res;
101 
102 	find.mr_base = find.mr_end = addr;
103 
104 	res = RB_FIND(mmio_rb_tree, rbt, &find);
105 
106 	if (res != NULL) {
107 		*entry = res;
108 		return (0);
109 	}
110 
111 	return (ENOENT);
112 }
113 
114 static int
mmio_rb_add(struct mmio_rb_tree * rbt,struct mmio_rb_range * new)115 mmio_rb_add(struct mmio_rb_tree *rbt, struct mmio_rb_range *new)
116 {
117 	struct mmio_rb_range *overlap;
118 
119 	overlap = RB_INSERT(mmio_rb_tree, rbt, new);
120 
121 	if (overlap != NULL) {
122 #ifdef RB_DEBUG
123 		printf("overlap detected: new %lx:%lx, tree %lx:%lx\n",
124 		       new->mr_base, new->mr_end,
125 		       overlap->mr_base, overlap->mr_end);
126 #endif
127 
128 		return (EEXIST);
129 	}
130 
131 	return (0);
132 }
133 
134 #if 0
135 static void
136 mmio_rb_dump(struct mmio_rb_tree *rbt)
137 {
138 	int perror;
139 	struct mmio_rb_range *np;
140 
141 	pthread_rwlock_rdlock(&mmio_rwlock);
142 	RB_FOREACH(np, mmio_rb_tree, rbt) {
143 		printf(" %lx:%lx, %s\n", np->mr_base, np->mr_end,
144 		       np->mr_param.name);
145 	}
146 	perror = pthread_rwlock_unlock(&mmio_rwlock);
147 	assert(perror == 0);
148 }
149 #endif
150 
151 RB_GENERATE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare);
152 
153 typedef int (mem_cb_t)(struct vmctx *ctx, int vcpu, uint64_t gpa,
154     struct mem_range *mr, void *arg);
155 
156 static int
mem_read(void * ctx,int vcpu,uint64_t gpa,uint64_t * rval,int size,void * arg)157 mem_read(void *ctx, int vcpu, uint64_t gpa, uint64_t *rval, int size, void *arg)
158 {
159 	int error;
160 	struct mem_range *mr = arg;
161 
162 	error = (*mr->handler)(ctx, vcpu, MEM_F_READ, gpa, size,
163 			       rval, mr->arg1, mr->arg2);
164 	return (error);
165 }
166 
167 static int
mem_write(void * ctx,int vcpu,uint64_t gpa,uint64_t wval,int size,void * arg)168 mem_write(void *ctx, int vcpu, uint64_t gpa, uint64_t wval, int size, void *arg)
169 {
170 	int error;
171 	struct mem_range *mr = arg;
172 
173 	error = (*mr->handler)(ctx, vcpu, MEM_F_WRITE, gpa, size,
174 			       &wval, mr->arg1, mr->arg2);
175 	return (error);
176 }
177 
178 static int
access_memory(struct vmctx * ctx,int vcpu,uint64_t paddr,mem_cb_t * cb,void * arg)179 access_memory(struct vmctx *ctx, int vcpu, uint64_t paddr, mem_cb_t *cb,
180     void *arg)
181 {
182 	struct mmio_rb_range *entry;
183 	int err, perror, immutable;
184 
185 	pthread_rwlock_rdlock(&mmio_rwlock);
186 	/*
187 	 * First check the per-vCPU cache
188 	 */
189 	if (mmio_hint[vcpu] &&
190 	    paddr >= mmio_hint[vcpu]->mr_base &&
191 	    paddr <= mmio_hint[vcpu]->mr_end) {
192 		entry = mmio_hint[vcpu];
193 	} else
194 		entry = NULL;
195 
196 	if (entry == NULL) {
197 		if (mmio_rb_lookup(&mmio_rb_root, paddr, &entry) == 0) {
198 			/* Update the per-vCPU cache */
199 			mmio_hint[vcpu] = entry;
200 		} else if (mmio_rb_lookup(&mmio_rb_fallback, paddr, &entry)) {
201 			perror = pthread_rwlock_unlock(&mmio_rwlock);
202 			assert(perror == 0);
203 			return (ESRCH);
204 		}
205 	}
206 
207 	assert(entry != NULL);
208 
209 	/*
210 	 * An 'immutable' memory range is guaranteed to be never removed
211 	 * so there is no need to hold 'mmio_rwlock' while calling the
212 	 * handler.
213 	 *
214 	 * XXX writes to the PCIR_COMMAND register can cause register_mem()
215 	 * to be called. If the guest is using PCI extended config space
216 	 * to modify the PCIR_COMMAND register then register_mem() can
217 	 * deadlock on 'mmio_rwlock'. However by registering the extended
218 	 * config space window as 'immutable' the deadlock can be avoided.
219 	 */
220 	immutable = (entry->mr_param.flags & MEM_F_IMMUTABLE);
221 	if (immutable) {
222 		perror = pthread_rwlock_unlock(&mmio_rwlock);
223 		assert(perror == 0);
224 	}
225 
226 	err = cb(ctx, vcpu, paddr, &entry->mr_param, arg);
227 
228 	if (!immutable) {
229 		perror = pthread_rwlock_unlock(&mmio_rwlock);
230 		assert(perror == 0);
231 	}
232 
233 
234 	return (err);
235 }
236 
237 static int
emulate_mem_cb(struct vmctx * ctx,int vcpu,uint64_t paddr,struct mem_range * mr,void * arg)238 emulate_mem_cb(struct vmctx *ctx, int vcpu, uint64_t paddr, struct mem_range *mr,
239     void *arg)
240 {
241 	struct vm_mmio *mmio;
242 	int err = 0;
243 
244 	mmio = arg;
245 
246 	if (mmio->read != 0) {
247 		err = mem_read(ctx, vcpu, paddr, &mmio->data, mmio->bytes, mr);
248 	} else {
249 		err = mem_write(ctx, vcpu, paddr, mmio->data, mmio->bytes, mr);
250 	}
251 
252 	return (err);
253 }
254 
255 int
emulate_mem(struct vmctx * ctx,int vcpu,struct vm_mmio * mmio)256 emulate_mem(struct vmctx *ctx, int vcpu, struct vm_mmio *mmio)
257 {
258 	return (access_memory(ctx, vcpu, mmio->gpa, emulate_mem_cb, mmio));
259 }
260 
261 struct rw_mem_args {
262 	uint64_t *val;
263 	int size;
264 	int operation;
265 };
266 
267 static int
rw_mem_cb(struct vmctx * ctx,int vcpu,uint64_t paddr,struct mem_range * mr,void * arg)268 rw_mem_cb(struct vmctx *ctx, int vcpu, uint64_t paddr, struct mem_range *mr,
269     void *arg)
270 {
271 	struct rw_mem_args *rma;
272 
273 	rma = arg;
274 	return (mr->handler(ctx, vcpu, rma->operation, paddr, rma->size,
275 	    rma->val, mr->arg1, mr->arg2));
276 }
277 
278 int
read_mem(struct vmctx * ctx,int vcpu,uint64_t gpa,uint64_t * rval,int size)279 read_mem(struct vmctx *ctx, int vcpu, uint64_t gpa, uint64_t *rval, int size)
280 {
281 	struct rw_mem_args rma;
282 
283 	rma.val = rval;
284 	rma.size = size;
285 	rma.operation = MEM_F_READ;
286 	return (access_memory(ctx, vcpu, gpa, rw_mem_cb, &rma));
287 }
288 
289 int
write_mem(struct vmctx * ctx,int vcpu,uint64_t gpa,uint64_t wval,int size)290 write_mem(struct vmctx *ctx, int vcpu, uint64_t gpa, uint64_t wval, int size)
291 {
292 	struct rw_mem_args rma;
293 
294 	rma.val = &wval;
295 	rma.size = size;
296 	rma.operation = MEM_F_WRITE;
297 	return (access_memory(ctx, vcpu, gpa, rw_mem_cb, &rma));
298 }
299 
300 static int
register_mem_int(struct mmio_rb_tree * rbt,struct mem_range * memp)301 register_mem_int(struct mmio_rb_tree *rbt, struct mem_range *memp)
302 {
303 	struct mmio_rb_range *entry, *mrp;
304 	int err, perror;
305 
306 	err = 0;
307 
308 	mrp = malloc(sizeof(struct mmio_rb_range));
309 	if (mrp == NULL) {
310 		warn("%s: couldn't allocate memory for mrp\n",
311 		     __func__);
312 		err = ENOMEM;
313 	} else {
314 		mrp->mr_param = *memp;
315 		mrp->mr_base = memp->base;
316 		mrp->mr_end = memp->base + memp->size - 1;
317 		pthread_rwlock_wrlock(&mmio_rwlock);
318 		if (mmio_rb_lookup(rbt, memp->base, &entry) != 0)
319 			err = mmio_rb_add(rbt, mrp);
320 		perror = pthread_rwlock_unlock(&mmio_rwlock);
321 		assert(perror == 0);
322 		if (err)
323 			free(mrp);
324 	}
325 
326 	return (err);
327 }
328 
329 int
register_mem(struct mem_range * memp)330 register_mem(struct mem_range *memp)
331 {
332 
333 	return (register_mem_int(&mmio_rb_root, memp));
334 }
335 
336 int
register_mem_fallback(struct mem_range * memp)337 register_mem_fallback(struct mem_range *memp)
338 {
339 
340 	return (register_mem_int(&mmio_rb_fallback, memp));
341 }
342 
343 int
unregister_mem(struct mem_range * memp)344 unregister_mem(struct mem_range *memp)
345 {
346 	struct mem_range *mr;
347 	struct mmio_rb_range *entry = NULL;
348 	int err, perror, i;
349 
350 	pthread_rwlock_wrlock(&mmio_rwlock);
351 	err = mmio_rb_lookup(&mmio_rb_root, memp->base, &entry);
352 	if (err == 0) {
353 		mr = &entry->mr_param;
354 		assert(mr->name == memp->name);
355 		assert(mr->base == memp->base && mr->size == memp->size);
356 		assert((mr->flags & MEM_F_IMMUTABLE) == 0);
357 		RB_REMOVE(mmio_rb_tree, &mmio_rb_root, entry);
358 
359 		/* flush Per-vCPU cache */
360 		for (i=0; i < VM_MAXCPU; i++) {
361 			if (mmio_hint[i] == entry)
362 				mmio_hint[i] = NULL;
363 		}
364 	}
365 	perror = pthread_rwlock_unlock(&mmio_rwlock);
366 	assert(perror == 0);
367 
368 	if (entry)
369 		free(entry);
370 
371 	return (err);
372 }
373 
374 void
init_mem(void)375 init_mem(void)
376 {
377 
378 	RB_INIT(&mmio_rb_root);
379 	RB_INIT(&mmio_rb_fallback);
380 	pthread_rwlock_init(&mmio_rwlock, NULL);
381 }
382