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
29/*
30 * Memory ranges are represented with an RB tree. On insertion, the range
31 * is checked for overlaps. On lookup, the key has the same base and limit
32 * so it can be searched within the range.
33 */
34
35#include <sys/types.h>
36#include <sys/errno.h>
37#include <sys/tree.h>
38#include <machine/vmm.h>
39#include <machine/vmm_instruction_emul.h>
40
41#include <assert.h>
42#include <err.h>
43#include <pthread.h>
44#include <stdio.h>
45#include <stdlib.h>
46#include <vmmapi.h>
47
48#include "mem.h"
49
50struct mmio_rb_range {
51	RB_ENTRY(mmio_rb_range)	mr_link;	/* RB tree links */
52	struct mem_range	mr_param;
53	uint64_t                mr_base;
54	uint64_t                mr_end;
55};
56
57struct mmio_rb_tree;
58RB_PROTOTYPE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare);
59
60static RB_HEAD(mmio_rb_tree, mmio_rb_range) mmio_rb_root, mmio_rb_fallback;
61
62/*
63 * Per-vCPU cache. Since most accesses from a vCPU will be to
64 * consecutive addresses in a range, it makes sense to cache the
65 * result of a lookup.
66 */
67static struct mmio_rb_range	**mmio_hint;
68static int mmio_ncpu;
69
70static pthread_rwlock_t mmio_rwlock;
71
72static int
73mmio_rb_range_compare(struct mmio_rb_range *a, struct mmio_rb_range *b)
74{
75	if (a->mr_end < b->mr_base)
76		return (-1);
77	else if (a->mr_base > b->mr_end)
78		return (1);
79	return (0);
80}
81
82static int
83mmio_rb_lookup(struct mmio_rb_tree *rbt, uint64_t addr,
84    struct mmio_rb_range **entry)
85{
86	struct mmio_rb_range find, *res;
87
88	find.mr_base = find.mr_end = addr;
89
90	res = RB_FIND(mmio_rb_tree, rbt, &find);
91
92	if (res != NULL) {
93		*entry = res;
94		return (0);
95	}
96
97	return (ENOENT);
98}
99
100static int
101mmio_rb_add(struct mmio_rb_tree *rbt, struct mmio_rb_range *new)
102{
103	struct mmio_rb_range *overlap;
104
105	overlap = RB_INSERT(mmio_rb_tree, rbt, new);
106
107	if (overlap != NULL) {
108#ifdef RB_DEBUG
109		printf("overlap detected: new %lx:%lx, tree %lx:%lx, '%s' "
110		       "claims region already claimed for '%s'\n",
111		       new->mr_base, new->mr_end,
112		       overlap->mr_base, overlap->mr_end,
113		       new->mr_param.name, overlap->mr_param.name);
114#endif
115
116		return (EEXIST);
117	}
118
119	return (0);
120}
121
122#if 0
123static void
124mmio_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
139RB_GENERATE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare);
140
141typedef int (mem_cb_t)(struct vcpu *vcpu, uint64_t gpa, struct mem_range *mr,
142    void *arg);
143
144static int
145mem_read(struct vcpu *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)(vcpu, MEM_F_READ, gpa, size, rval, mr->arg1,
151	    mr->arg2);
152	return (error);
153}
154
155static int
156mem_write(struct vcpu *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)(vcpu, MEM_F_WRITE, gpa, size, &wval, mr->arg1,
162	    mr->arg2);
163	return (error);
164}
165
166static int
167access_memory(struct vcpu *vcpu, uint64_t paddr, mem_cb_t *cb, void *arg)
168{
169	struct mmio_rb_range *entry;
170	int err, perror, immutable, vcpuid;
171
172	vcpuid = vcpu_id(vcpu);
173	pthread_rwlock_rdlock(&mmio_rwlock);
174	/*
175	 * First check the per-vCPU cache
176	 */
177	if (mmio_hint[vcpuid] &&
178	    paddr >= mmio_hint[vcpuid]->mr_base &&
179	    paddr <= mmio_hint[vcpuid]->mr_end) {
180		entry = mmio_hint[vcpuid];
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[vcpuid] = 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(vcpu, paddr, &entry->mr_param, arg);
215
216	if (!immutable) {
217		perror = pthread_rwlock_unlock(&mmio_rwlock);
218		assert(perror == 0);
219	}
220
221	return (err);
222}
223
224struct emulate_mem_args {
225	struct vie *vie;
226	struct vm_guest_paging *paging;
227};
228
229static int
230emulate_mem_cb(struct vcpu *vcpu, uint64_t paddr, struct mem_range *mr,
231    void *arg)
232{
233	struct emulate_mem_args *ema;
234
235	ema = arg;
236	return (vmm_emulate_instruction(vcpu, paddr, ema->vie, ema->paging,
237	    mem_read, mem_write, mr));
238}
239
240int
241emulate_mem(struct vcpu *vcpu, uint64_t paddr, struct vie *vie,
242    struct vm_guest_paging *paging)
243{
244	struct emulate_mem_args ema;
245
246	ema.vie = vie;
247	ema.paging = paging;
248	return (access_memory(vcpu, paddr, emulate_mem_cb, &ema));
249}
250
251struct rw_mem_args {
252	uint64_t *val;
253	int size;
254	int operation;
255};
256
257static int
258rw_mem_cb(struct vcpu *vcpu, uint64_t paddr, struct mem_range *mr, void *arg)
259{
260	struct rw_mem_args *rma;
261
262	rma = arg;
263	return (mr->handler(vcpu, rma->operation, paddr, rma->size,
264	    rma->val, mr->arg1, mr->arg2));
265}
266
267int
268read_mem(struct vcpu *vcpu, uint64_t gpa, uint64_t *rval, int size)
269{
270	struct rw_mem_args rma;
271
272	rma.val = rval;
273	rma.size = size;
274	rma.operation = MEM_F_READ;
275	return (access_memory(vcpu, gpa, rw_mem_cb, &rma));
276}
277
278int
279write_mem(struct vcpu *vcpu, uint64_t gpa, uint64_t wval, int size)
280{
281	struct rw_mem_args rma;
282
283	rma.val = &wval;
284	rma.size = size;
285	rma.operation = MEM_F_WRITE;
286	return (access_memory(vcpu, gpa, rw_mem_cb, &rma));
287}
288
289static int
290register_mem_int(struct mmio_rb_tree *rbt, struct mem_range *memp)
291{
292	struct mmio_rb_range *entry, *mrp;
293	int err, perror;
294
295	err = 0;
296
297	mrp = malloc(sizeof(struct mmio_rb_range));
298	if (mrp == NULL) {
299		warn("%s: couldn't allocate memory for mrp\n",
300		     __func__);
301		err = ENOMEM;
302	} else {
303		mrp->mr_param = *memp;
304		mrp->mr_base = memp->base;
305		mrp->mr_end = memp->base + memp->size - 1;
306		pthread_rwlock_wrlock(&mmio_rwlock);
307		if (mmio_rb_lookup(rbt, memp->base, &entry) != 0)
308			err = mmio_rb_add(rbt, mrp);
309		perror = pthread_rwlock_unlock(&mmio_rwlock);
310		assert(perror == 0);
311		if (err)
312			free(mrp);
313	}
314
315	return (err);
316}
317
318int
319register_mem(struct mem_range *memp)
320{
321
322	return (register_mem_int(&mmio_rb_root, memp));
323}
324
325int
326register_mem_fallback(struct mem_range *memp)
327{
328
329	return (register_mem_int(&mmio_rb_fallback, memp));
330}
331
332int
333unregister_mem(struct mem_range *memp)
334{
335	struct mem_range *mr;
336	struct mmio_rb_range *entry = NULL;
337	int err, perror, i;
338
339	pthread_rwlock_wrlock(&mmio_rwlock);
340	err = mmio_rb_lookup(&mmio_rb_root, memp->base, &entry);
341	if (err == 0) {
342		mr = &entry->mr_param;
343		assert(mr->name == memp->name);
344		assert(mr->base == memp->base && mr->size == memp->size);
345		assert((mr->flags & MEM_F_IMMUTABLE) == 0);
346		RB_REMOVE(mmio_rb_tree, &mmio_rb_root, entry);
347
348		/* flush Per-vCPU cache */
349		for (i = 0; i < mmio_ncpu; i++) {
350			if (mmio_hint[i] == entry)
351				mmio_hint[i] = NULL;
352		}
353	}
354	perror = pthread_rwlock_unlock(&mmio_rwlock);
355	assert(perror == 0);
356
357	if (entry)
358		free(entry);
359
360	return (err);
361}
362
363void
364init_mem(int ncpu)
365{
366
367	mmio_ncpu = ncpu;
368	mmio_hint = calloc(ncpu, sizeof(*mmio_hint));
369	RB_INIT(&mmio_rb_root);
370	RB_INIT(&mmio_rb_fallback);
371	pthread_rwlock_init(&mmio_rwlock, NULL);
372}
373