1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation
3 */
4
5 #include <stdlib.h>
6 #include <stdio.h>
7 #include <stdint.h>
8 #include <stdarg.h>
9 #include <inttypes.h>
10 #include <string.h>
11 #include <errno.h>
12 #include <sys/queue.h>
13
14 #include <rte_log.h>
15 #include <rte_memory.h>
16 #include <rte_memzone.h>
17 #include <rte_eal.h>
18 #include <rte_per_lcore.h>
19 #include <rte_errno.h>
20 #include <rte_string_fns.h>
21 #include <rte_common.h>
22 #include <rte_eal_trace.h>
23
24 #include "malloc_heap.h"
25 #include "malloc_elem.h"
26 #include "eal_private.h"
27 #include "eal_memcfg.h"
28
29 static inline const struct rte_memzone *
memzone_lookup_thread_unsafe(const char * name)30 memzone_lookup_thread_unsafe(const char *name)
31 {
32 struct rte_mem_config *mcfg;
33 struct rte_fbarray *arr;
34 const struct rte_memzone *mz;
35 int i = 0;
36
37 /* get pointer to global configuration */
38 mcfg = rte_eal_get_configuration()->mem_config;
39 arr = &mcfg->memzones;
40
41 /*
42 * the algorithm is not optimal (linear), but there are few
43 * zones and this function should be called at init only
44 */
45 i = rte_fbarray_find_next_used(arr, 0);
46 while (i >= 0) {
47 mz = rte_fbarray_get(arr, i);
48 if (mz->addr != NULL &&
49 !strncmp(name, mz->name, RTE_MEMZONE_NAMESIZE))
50 return mz;
51 i = rte_fbarray_find_next_used(arr, i + 1);
52 }
53 return NULL;
54 }
55
56 static const struct rte_memzone *
memzone_reserve_aligned_thread_unsafe(const char * name,size_t len,int socket_id,unsigned int flags,unsigned int align,unsigned int bound)57 memzone_reserve_aligned_thread_unsafe(const char *name, size_t len,
58 int socket_id, unsigned int flags, unsigned int align,
59 unsigned int bound)
60 {
61 struct rte_memzone *mz;
62 struct rte_mem_config *mcfg;
63 struct rte_fbarray *arr;
64 void *mz_addr;
65 size_t requested_len;
66 int mz_idx;
67 bool contig;
68
69 /* get pointer to global configuration */
70 mcfg = rte_eal_get_configuration()->mem_config;
71 arr = &mcfg->memzones;
72
73 /* no more room in config */
74 if (arr->count >= arr->len) {
75 RTE_LOG(ERR, EAL,
76 "%s(): Number of requested memzone segments exceeds RTE_MAX_MEMZONE\n",
77 __func__);
78 rte_errno = ENOSPC;
79 return NULL;
80 }
81
82 if (strlen(name) > sizeof(mz->name) - 1) {
83 RTE_LOG(DEBUG, EAL, "%s(): memzone <%s>: name too long\n",
84 __func__, name);
85 rte_errno = ENAMETOOLONG;
86 return NULL;
87 }
88
89 /* zone already exist */
90 if ((memzone_lookup_thread_unsafe(name)) != NULL) {
91 RTE_LOG(DEBUG, EAL, "%s(): memzone <%s> already exists\n",
92 __func__, name);
93 rte_errno = EEXIST;
94 return NULL;
95 }
96
97 /* if alignment is not a power of two */
98 if (align && !rte_is_power_of_2(align)) {
99 RTE_LOG(ERR, EAL, "%s(): Invalid alignment: %u\n", __func__,
100 align);
101 rte_errno = EINVAL;
102 return NULL;
103 }
104
105 /* alignment less than cache size is not allowed */
106 if (align < RTE_CACHE_LINE_SIZE)
107 align = RTE_CACHE_LINE_SIZE;
108
109 /* align length on cache boundary. Check for overflow before doing so */
110 if (len > SIZE_MAX - RTE_CACHE_LINE_MASK) {
111 rte_errno = EINVAL; /* requested size too big */
112 return NULL;
113 }
114
115 len = RTE_ALIGN_CEIL(len, RTE_CACHE_LINE_SIZE);
116
117 /* save minimal requested length */
118 requested_len = RTE_MAX((size_t)RTE_CACHE_LINE_SIZE, len);
119
120 /* check that boundary condition is valid */
121 if (bound != 0 && (requested_len > bound || !rte_is_power_of_2(bound))) {
122 rte_errno = EINVAL;
123 return NULL;
124 }
125
126 if ((socket_id != SOCKET_ID_ANY) && socket_id < 0) {
127 rte_errno = EINVAL;
128 return NULL;
129 }
130
131 /* only set socket to SOCKET_ID_ANY if we aren't allocating for an
132 * external heap.
133 */
134 if (!rte_eal_has_hugepages() && socket_id < RTE_MAX_NUMA_NODES)
135 socket_id = SOCKET_ID_ANY;
136
137 contig = (flags & RTE_MEMZONE_IOVA_CONTIG) != 0;
138 /* malloc only cares about size flags, remove contig flag from flags */
139 flags &= ~RTE_MEMZONE_IOVA_CONTIG;
140
141 if (len == 0 && bound == 0) {
142 /* no size constraints were placed, so use malloc elem len */
143 requested_len = 0;
144 mz_addr = malloc_heap_alloc_biggest(NULL, socket_id, flags,
145 align, contig);
146 } else {
147 if (len == 0)
148 requested_len = bound;
149 /* allocate memory on heap */
150 mz_addr = malloc_heap_alloc(NULL, requested_len, socket_id,
151 flags, align, bound, contig);
152 }
153 if (mz_addr == NULL) {
154 rte_errno = ENOMEM;
155 return NULL;
156 }
157
158 struct malloc_elem *elem = malloc_elem_from_data(mz_addr);
159
160 /* fill the zone in config */
161 mz_idx = rte_fbarray_find_next_free(arr, 0);
162
163 if (mz_idx < 0) {
164 mz = NULL;
165 } else {
166 rte_fbarray_set_used(arr, mz_idx);
167 mz = rte_fbarray_get(arr, mz_idx);
168 }
169
170 if (mz == NULL) {
171 RTE_LOG(ERR, EAL, "%s(): Cannot find free memzone\n", __func__);
172 malloc_heap_free(elem);
173 rte_errno = ENOSPC;
174 return NULL;
175 }
176
177 strlcpy(mz->name, name, sizeof(mz->name));
178 mz->iova = rte_malloc_virt2iova(mz_addr);
179 mz->addr = mz_addr;
180 mz->len = requested_len == 0 ?
181 elem->size - elem->pad - MALLOC_ELEM_OVERHEAD :
182 requested_len;
183 mz->hugepage_sz = elem->msl->page_sz;
184 mz->socket_id = elem->msl->socket_id;
185 mz->flags = 0;
186
187 return mz;
188 }
189
190 static const struct rte_memzone *
rte_memzone_reserve_thread_safe(const char * name,size_t len,int socket_id,unsigned int flags,unsigned int align,unsigned int bound)191 rte_memzone_reserve_thread_safe(const char *name, size_t len, int socket_id,
192 unsigned int flags, unsigned int align, unsigned int bound)
193 {
194 struct rte_mem_config *mcfg;
195 const struct rte_memzone *mz = NULL;
196
197 /* get pointer to global configuration */
198 mcfg = rte_eal_get_configuration()->mem_config;
199
200 rte_rwlock_write_lock(&mcfg->mlock);
201
202 mz = memzone_reserve_aligned_thread_unsafe(
203 name, len, socket_id, flags, align, bound);
204
205 rte_eal_trace_memzone_reserve(name, len, socket_id, flags, align,
206 bound, mz);
207
208 rte_rwlock_write_unlock(&mcfg->mlock);
209
210 return mz;
211 }
212
213 /*
214 * Return a pointer to a correctly filled memzone descriptor (with a
215 * specified alignment and boundary). If the allocation cannot be done,
216 * return NULL.
217 */
218 const struct rte_memzone *
rte_memzone_reserve_bounded(const char * name,size_t len,int socket_id,unsigned flags,unsigned align,unsigned bound)219 rte_memzone_reserve_bounded(const char *name, size_t len, int socket_id,
220 unsigned flags, unsigned align, unsigned bound)
221 {
222 return rte_memzone_reserve_thread_safe(name, len, socket_id, flags,
223 align, bound);
224 }
225
226 /*
227 * Return a pointer to a correctly filled memzone descriptor (with a
228 * specified alignment). If the allocation cannot be done, return NULL.
229 */
230 const struct rte_memzone *
rte_memzone_reserve_aligned(const char * name,size_t len,int socket_id,unsigned flags,unsigned align)231 rte_memzone_reserve_aligned(const char *name, size_t len, int socket_id,
232 unsigned flags, unsigned align)
233 {
234 return rte_memzone_reserve_thread_safe(name, len, socket_id, flags,
235 align, 0);
236 }
237
238 /*
239 * Return a pointer to a correctly filled memzone descriptor. If the
240 * allocation cannot be done, return NULL.
241 */
242 const struct rte_memzone *
rte_memzone_reserve(const char * name,size_t len,int socket_id,unsigned flags)243 rte_memzone_reserve(const char *name, size_t len, int socket_id,
244 unsigned flags)
245 {
246 return rte_memzone_reserve_thread_safe(name, len, socket_id,
247 flags, RTE_CACHE_LINE_SIZE, 0);
248 }
249
250 int
rte_memzone_free(const struct rte_memzone * mz)251 rte_memzone_free(const struct rte_memzone *mz)
252 {
253 char name[RTE_MEMZONE_NAMESIZE];
254 struct rte_mem_config *mcfg;
255 struct rte_fbarray *arr;
256 struct rte_memzone *found_mz;
257 int ret = 0;
258 void *addr = NULL;
259 unsigned idx;
260
261 if (mz == NULL)
262 return -EINVAL;
263
264 rte_strlcpy(name, mz->name, RTE_MEMZONE_NAMESIZE);
265 mcfg = rte_eal_get_configuration()->mem_config;
266 arr = &mcfg->memzones;
267
268 rte_rwlock_write_lock(&mcfg->mlock);
269
270 idx = rte_fbarray_find_idx(arr, mz);
271 found_mz = rte_fbarray_get(arr, idx);
272
273 if (found_mz == NULL) {
274 ret = -EINVAL;
275 } else if (found_mz->addr == NULL) {
276 RTE_LOG(ERR, EAL, "Memzone is not allocated\n");
277 ret = -EINVAL;
278 } else {
279 addr = found_mz->addr;
280 memset(found_mz, 0, sizeof(*found_mz));
281 rte_fbarray_set_free(arr, idx);
282 }
283
284 rte_rwlock_write_unlock(&mcfg->mlock);
285
286 if (addr != NULL)
287 rte_free(addr);
288
289 rte_eal_trace_memzone_free(name, addr, ret);
290 return ret;
291 }
292
293 /*
294 * Lookup for the memzone identified by the given name
295 */
296 const struct rte_memzone *
rte_memzone_lookup(const char * name)297 rte_memzone_lookup(const char *name)
298 {
299 struct rte_mem_config *mcfg;
300 const struct rte_memzone *memzone = NULL;
301
302 mcfg = rte_eal_get_configuration()->mem_config;
303
304 rte_rwlock_read_lock(&mcfg->mlock);
305
306 memzone = memzone_lookup_thread_unsafe(name);
307
308 rte_rwlock_read_unlock(&mcfg->mlock);
309
310 rte_eal_trace_memzone_lookup(name, memzone);
311 return memzone;
312 }
313
314 static void
dump_memzone(const struct rte_memzone * mz,void * arg)315 dump_memzone(const struct rte_memzone *mz, void *arg)
316 {
317 struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
318 struct rte_memseg_list *msl = NULL;
319 void *cur_addr, *mz_end;
320 struct rte_memseg *ms;
321 int mz_idx, ms_idx;
322 size_t page_sz;
323 FILE *f = arg;
324
325 mz_idx = rte_fbarray_find_idx(&mcfg->memzones, mz);
326
327 fprintf(f, "Zone %u: name:<%s>, len:0x%zx, virt:%p, "
328 "socket_id:%"PRId32", flags:%"PRIx32"\n",
329 mz_idx,
330 mz->name,
331 mz->len,
332 mz->addr,
333 mz->socket_id,
334 mz->flags);
335
336 /* go through each page occupied by this memzone */
337 msl = rte_mem_virt2memseg_list(mz->addr);
338 if (!msl) {
339 RTE_LOG(DEBUG, EAL, "Skipping bad memzone\n");
340 return;
341 }
342 page_sz = (size_t)mz->hugepage_sz;
343 cur_addr = RTE_PTR_ALIGN_FLOOR(mz->addr, page_sz);
344 mz_end = RTE_PTR_ADD(cur_addr, mz->len);
345
346 fprintf(f, "physical segments used:\n");
347 ms_idx = RTE_PTR_DIFF(mz->addr, msl->base_va) / page_sz;
348 ms = rte_fbarray_get(&msl->memseg_arr, ms_idx);
349
350 do {
351 fprintf(f, " addr: %p iova: 0x%" PRIx64 " "
352 "len: 0x%zx "
353 "pagesz: 0x%zx\n",
354 cur_addr, ms->iova, ms->len, page_sz);
355
356 /* advance VA to next page */
357 cur_addr = RTE_PTR_ADD(cur_addr, page_sz);
358
359 /* memzones occupy contiguous segments */
360 ++ms;
361 } while (cur_addr < mz_end);
362 }
363
364 /* Dump all reserved memory zones on console */
365 void
rte_memzone_dump(FILE * f)366 rte_memzone_dump(FILE *f)
367 {
368 rte_memzone_walk(dump_memzone, f);
369 }
370
371 /*
372 * Init the memzone subsystem
373 */
374 int
rte_eal_memzone_init(void)375 rte_eal_memzone_init(void)
376 {
377 struct rte_mem_config *mcfg;
378 int ret = 0;
379
380 /* get pointer to global configuration */
381 mcfg = rte_eal_get_configuration()->mem_config;
382
383 rte_rwlock_write_lock(&mcfg->mlock);
384
385 if (rte_eal_process_type() == RTE_PROC_PRIMARY &&
386 rte_fbarray_init(&mcfg->memzones, "memzone",
387 RTE_MAX_MEMZONE, sizeof(struct rte_memzone))) {
388 RTE_LOG(ERR, EAL, "Cannot allocate memzone list\n");
389 ret = -1;
390 } else if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
391 rte_fbarray_attach(&mcfg->memzones)) {
392 RTE_LOG(ERR, EAL, "Cannot attach to memzone list\n");
393 ret = -1;
394 }
395
396 rte_rwlock_write_unlock(&mcfg->mlock);
397
398 return ret;
399 }
400
401 /* Walk all reserved memory zones */
rte_memzone_walk(void (* func)(const struct rte_memzone *,void *),void * arg)402 void rte_memzone_walk(void (*func)(const struct rte_memzone *, void *),
403 void *arg)
404 {
405 struct rte_mem_config *mcfg;
406 struct rte_fbarray *arr;
407 int i;
408
409 mcfg = rte_eal_get_configuration()->mem_config;
410 arr = &mcfg->memzones;
411
412 rte_rwlock_read_lock(&mcfg->mlock);
413 i = rte_fbarray_find_next_used(arr, 0);
414 while (i >= 0) {
415 struct rte_memzone *mz = rte_fbarray_get(arr, i);
416 (*func)(mz, arg);
417 i = rte_fbarray_find_next_used(arr, i + 1);
418 }
419 rte_rwlock_read_unlock(&mcfg->mlock);
420 }
421