4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
41 #include <sys/queue.h>
44 #include <rte_memory.h>
45 #include <rte_memzone.h>
46 #include <rte_tailq.h>
48 #include <rte_eal_memconfig.h>
49 #include <rte_per_lcore.h>
50 #include <rte_errno.h>
51 #include <rte_string_fns.h>
52 #include <rte_common.h>
54 #include "eal_private.h"
56 /* internal copy of free memory segments */
57 static struct rte_memseg *free_memseg = NULL;
59 static inline const struct rte_memzone *
60 memzone_lookup_thread_unsafe(const char *name)
62 const struct rte_mem_config *mcfg;
65 /* get pointer to global configuration */
66 mcfg = rte_eal_get_configuration()->mem_config;
69 * the algorithm is not optimal (linear), but there are few
70 * zones and this function should be called at init only
72 for (i = 0; i < RTE_MAX_MEMZONE && mcfg->memzone[i].addr != NULL; i++) {
73 if (!strncmp(name, mcfg->memzone[i].name, RTE_MEMZONE_NAMESIZE))
74 return &mcfg->memzone[i];
81 * Return a pointer to a correctly filled memzone descriptor. If the
82 * allocation cannot be done, return NULL.
84 const struct rte_memzone *
85 rte_memzone_reserve(const char *name, size_t len, int socket_id,
88 return rte_memzone_reserve_aligned(name,
89 len, socket_id, flags, CACHE_LINE_SIZE);
92 static const struct rte_memzone *
93 memzone_reserve_aligned_thread_unsafe(const char *name, size_t len,
94 int socket_id, unsigned flags, unsigned align)
96 struct rte_mem_config *mcfg;
100 size_t requested_len;
101 size_t memseg_len = 0;
102 phys_addr_t memseg_physaddr;
105 /* get pointer to global configuration */
106 mcfg = rte_eal_get_configuration()->mem_config;
108 /* no more room in config */
109 if (mcfg->memzone_idx >= RTE_MAX_MEMZONE) {
110 RTE_LOG(ERR, EAL, "%s(): No more room in config\n", __func__);
115 /* zone already exist */
116 if ((memzone_lookup_thread_unsafe(name)) != NULL) {
117 RTE_LOG(DEBUG, EAL, "%s(): memzone <%s> already exists\n",
123 /* align length on cache boundary. Check for overflow before doing so */
124 if (len > SIZE_MAX - CACHE_LINE_MASK) {
125 rte_errno = EINVAL; /* requested size too big */
128 len += CACHE_LINE_MASK;
129 len &= ~((size_t) CACHE_LINE_MASK);
131 /* save original length */
134 /* reserve extra space for future alignment */
138 /* save requested length */
141 /* reserve extra space for future alignment */
145 /* find the smallest segment matching requirements */
146 for (i = 0; i < RTE_MAX_MEMSEG; i++) {
148 if (free_memseg[i].addr == NULL)
151 /* empty segment, skip it */
152 if (free_memseg[i].len == 0)
156 if (socket_id != SOCKET_ID_ANY &&
157 socket_id != free_memseg[i].socket_id)
161 if (len != 0 && len > free_memseg[i].len)
164 /* check flags for hugepage sizes */
165 if ((flags & RTE_MEMZONE_2MB) &&
166 free_memseg[i].hugepage_sz == RTE_PGSIZE_1G )
168 if ((flags & RTE_MEMZONE_1GB) &&
169 free_memseg[i].hugepage_sz == RTE_PGSIZE_2M )
172 /* this segment is the best until now */
173 if (memseg_idx == -1) {
175 memseg_len = free_memseg[i].len;
177 /* find the biggest contiguous zone */
179 if (free_memseg[i].len > memseg_len) {
181 memseg_len = free_memseg[i].len;
185 * find the smallest (we already checked that current
186 * zone length is > len
188 else if (free_memseg[i].len < memseg_len) {
190 memseg_len = free_memseg[i].len;
194 /* no segment found */
195 if (memseg_idx == -1) {
197 * If RTE_MEMZONE_SIZE_HINT_ONLY flag is specified,
198 * try allocating again without the size parameter otherwise -fail.
200 if ((flags & RTE_MEMZONE_SIZE_HINT_ONLY) &&
201 ((flags & RTE_MEMZONE_1GB) || (flags & RTE_MEMZONE_2MB)))
202 return memzone_reserve_aligned_thread_unsafe(name, len - align,
203 socket_id, 0, align);
205 RTE_LOG(ERR, EAL, "%s(%s, %zu, %d): "
206 "No appropriate segment found\n",
207 __func__, name, requested_len, socket_id);
212 /* get offset needed to adjust alignment */
213 addr_offset = RTE_ALIGN_CEIL(free_memseg[memseg_idx].phys_addr, align) -
214 free_memseg[memseg_idx].phys_addr;
216 /* save aligned physical and virtual addresses */
217 memseg_physaddr = free_memseg[memseg_idx].phys_addr + addr_offset;
218 memseg_addr = RTE_PTR_ADD(free_memseg[memseg_idx].addr,
219 (uintptr_t) addr_offset);
221 /* if we are looking for a biggest memzone */
222 if (requested_len == 0)
223 requested_len = memseg_len - addr_offset;
225 /* set length to correct value */
226 len = (size_t)addr_offset + requested_len;
228 /* update our internal state */
229 free_memseg[memseg_idx].len -= len;
230 free_memseg[memseg_idx].phys_addr += len;
231 free_memseg[memseg_idx].addr =
232 (char *)free_memseg[memseg_idx].addr + len;
234 /* fill the zone in config */
235 struct rte_memzone *mz = &mcfg->memzone[mcfg->memzone_idx++];
236 rte_snprintf(mz->name, sizeof(mz->name), "%s", name);
237 mz->phys_addr = memseg_physaddr;
238 mz->addr = memseg_addr;
239 mz->len = requested_len;
240 mz->hugepage_sz = free_memseg[memseg_idx].hugepage_sz;
241 mz->socket_id = free_memseg[memseg_idx].socket_id;
248 * Return a pointer to a correctly filled memzone descriptor (with a
249 * specified alignment). If the allocation cannot be done, return NULL.
251 const struct rte_memzone *
252 rte_memzone_reserve_aligned(const char *name, size_t len,
253 int socket_id, unsigned flags, unsigned align)
255 struct rte_mem_config *mcfg;
256 const struct rte_memzone *mz = NULL;
258 /* both sizes cannot be explicitly called for */
259 if ((flags & RTE_MEMZONE_1GB) && (flags & RTE_MEMZONE_2MB)) {
264 /* if alignment is not a power of two */
265 if (!rte_is_power_of_2(align)) {
266 RTE_LOG(ERR, EAL, "%s(): Invalid alignment: %u\n", __func__,
272 /* alignment less than cache size is not allowed */
273 if (align < CACHE_LINE_SIZE)
274 align = CACHE_LINE_SIZE;
276 /* get pointer to global configuration */
277 mcfg = rte_eal_get_configuration()->mem_config;
279 rte_rwlock_write_lock(&mcfg->mlock);
281 mz = memzone_reserve_aligned_thread_unsafe(
282 name, len, socket_id, flags, align);
284 rte_rwlock_write_unlock(&mcfg->mlock);
290 * Lookup for the memzone identified by the given name
292 const struct rte_memzone *
293 rte_memzone_lookup(const char *name)
295 struct rte_mem_config *mcfg;
296 const struct rte_memzone *memzone = NULL;
298 mcfg = rte_eal_get_configuration()->mem_config;
300 rte_rwlock_read_lock(&mcfg->mlock);
302 memzone = memzone_lookup_thread_unsafe(name);
304 rte_rwlock_read_unlock(&mcfg->mlock);
309 /* Dump all reserved memory zones on console */
311 rte_memzone_dump(void)
313 struct rte_mem_config *mcfg;
316 /* get pointer to global configuration */
317 mcfg = rte_eal_get_configuration()->mem_config;
319 rte_rwlock_read_lock(&mcfg->mlock);
321 for (i=0; i<RTE_MAX_MEMZONE; i++) {
322 if (mcfg->memzone[i].addr == NULL)
324 printf("Zone %u: name:<%s>, phys:0x%"PRIx64", len:0x%zx"
325 ", virt:%p, socket_id:%"PRId32", flags:%"PRIx32"\n", i,
326 mcfg->memzone[i].name,
327 mcfg->memzone[i].phys_addr,
328 mcfg->memzone[i].len,
329 mcfg->memzone[i].addr,
330 mcfg->memzone[i].socket_id,
331 mcfg->memzone[i].flags);
333 rte_rwlock_read_unlock(&mcfg->mlock);
337 * called by init: modify the free memseg list to have cache-aligned
338 * addresses and cache-aligned lengths
341 memseg_sanitize(struct rte_memseg *memseg)
347 phys_align = memseg->phys_addr & CACHE_LINE_MASK;
348 virt_align = (unsigned long)memseg->addr & CACHE_LINE_MASK;
351 * sanity check: phys_addr and addr must have the same
354 if (phys_align != virt_align)
357 /* memseg is really too small, don't bother with it */
358 if (memseg->len < (2 * CACHE_LINE_SIZE)) {
363 /* align start address */
364 off = (CACHE_LINE_SIZE - phys_align) & CACHE_LINE_MASK;
365 memseg->phys_addr += off;
366 memseg->addr = (char *)memseg->addr + off;
369 /* align end address */
370 memseg->len &= ~((uint64_t)CACHE_LINE_MASK);
376 * Init the memzone subsystem
379 rte_eal_memzone_init(void)
381 struct rte_mem_config *mcfg;
382 const struct rte_memseg *memseg;
385 /* get pointer to global configuration */
386 mcfg = rte_eal_get_configuration()->mem_config;
388 /* mirror the runtime memsegs from config */
389 free_memseg = mcfg->free_memseg;
391 /* secondary processes don't need to initialise anything */
392 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
395 memseg = rte_eal_get_physmem_layout();
396 if (memseg == NULL) {
397 RTE_LOG(ERR, EAL, "%s(): Cannot get physical layout\n", __func__);
401 rte_rwlock_write_lock(&mcfg->mlock);
403 /* duplicate the memsegs from config */
404 memcpy(free_memseg, memseg, sizeof(struct rte_memseg) * RTE_MAX_MEMSEG);
406 /* make all zones cache-aligned */
407 for (i=0; i<RTE_MAX_MEMSEG; i++) {
408 if (free_memseg[i].addr == NULL)
410 if (memseg_sanitize(&free_memseg[i]) < 0) {
411 RTE_LOG(ERR, EAL, "%s(): Sanity check failed\n", __func__);
412 rte_rwlock_write_unlock(&mcfg->mlock);
417 /* delete all zones */
418 mcfg->memzone_idx = 0;
419 memset(mcfg->memzone, 0, sizeof(mcfg->memzone));
421 rte_rwlock_write_unlock(&mcfg->mlock);