]> git.proxmox.com Git - ceph.git/blame - ceph/src/spdk/lib/env_dpdk/init.c
update sources to ceph Nautilus 14.2.1
[ceph.git] / ceph / src / spdk / lib / env_dpdk / init.c
CommitLineData
7c673cae
FG
1/*-
2 * BSD LICENSE
3 *
4 * Copyright (c) Intel Corporation.
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 *
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
16 * distribution.
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.
20 *
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.
32 */
33
11fdf7f2
TL
34#include "spdk/stdinc.h"
35
7c673cae
FG
36#include "env_internal.h"
37
11fdf7f2 38#include "spdk/version.h"
7c673cae
FG
39
40#include <rte_config.h>
41#include <rte_eal.h>
42
43#define SPDK_ENV_DPDK_DEFAULT_NAME "spdk"
44#define SPDK_ENV_DPDK_DEFAULT_SHM_ID -1
45#define SPDK_ENV_DPDK_DEFAULT_MEM_SIZE -1
46#define SPDK_ENV_DPDK_DEFAULT_MASTER_CORE -1
47#define SPDK_ENV_DPDK_DEFAULT_MEM_CHANNEL -1
48#define SPDK_ENV_DPDK_DEFAULT_CORE_MASK "0x1"
49
11fdf7f2
TL
50static char **eal_cmdline;
51static int eal_cmdline_argcount;
52
7c673cae
FG
53static char *
54_sprintf_alloc(const char *format, ...)
55{
56 va_list args;
57 va_list args_copy;
58 char *buf;
59 size_t bufsize;
60 int rc;
61
62 va_start(args, format);
63
64 /* Try with a small buffer first. */
65 bufsize = 32;
66
67 /* Limit maximum buffer size to something reasonable so we don't loop forever. */
68 while (bufsize <= 1024 * 1024) {
69 buf = malloc(bufsize);
70 if (buf == NULL) {
71 va_end(args);
72 return NULL;
73 }
74
75 va_copy(args_copy, args);
76 rc = vsnprintf(buf, bufsize, format, args_copy);
77 va_end(args_copy);
78
79 /*
80 * If vsnprintf() returned a count within our current buffer size, we are done.
81 * The count does not include the \0 terminator, so rc == bufsize is not OK.
82 */
83 if (rc >= 0 && (size_t)rc < bufsize) {
84 va_end(args);
85 return buf;
86 }
87
88 /*
89 * vsnprintf() should return the required space, but some libc versions do not
90 * implement this correctly, so just double the buffer size and try again.
91 *
92 * We don't need the data in buf, so rather than realloc(), use free() and malloc()
93 * again to avoid a copy.
94 */
95 free(buf);
96 bufsize *= 2;
97 }
98
99 va_end(args);
100 return NULL;
101}
102
11fdf7f2
TL
103static void
104spdk_env_unlink_shared_files(void)
105{
106 /* Starting with DPDK 18.05, there are more files with unpredictable paths
107 * and filenames. The --no-shconf option prevents from creating them, but
108 * only for DPDK 18.08+. For DPDK 18.05 we just leave them be.
109 */
110#if RTE_VERSION < RTE_VERSION_NUM(18, 05, 0, 0)
111 char buffer[PATH_MAX];
112
113 snprintf(buffer, PATH_MAX, "/var/run/.spdk_pid%d_hugepage_info", getpid());
114 if (unlink(buffer)) {
115 fprintf(stderr, "Unable to unlink shared memory file: %s. Error code: %d\n", buffer, errno);
116 }
117#endif
118}
119
7c673cae
FG
120void
121spdk_env_opts_init(struct spdk_env_opts *opts)
122{
123 if (!opts) {
124 return;
125 }
126
127 memset(opts, 0, sizeof(*opts));
128
129 opts->name = SPDK_ENV_DPDK_DEFAULT_NAME;
130 opts->core_mask = SPDK_ENV_DPDK_DEFAULT_CORE_MASK;
131 opts->shm_id = SPDK_ENV_DPDK_DEFAULT_SHM_ID;
11fdf7f2
TL
132 opts->mem_size = SPDK_ENV_DPDK_DEFAULT_MEM_SIZE;
133 opts->master_core = SPDK_ENV_DPDK_DEFAULT_MASTER_CORE;
134 opts->mem_channel = SPDK_ENV_DPDK_DEFAULT_MEM_CHANNEL;
7c673cae
FG
135}
136
137static void
138spdk_free_args(char **args, int argcount)
139{
140 int i;
141
7c673cae 142 for (i = 0; i < argcount; i++) {
7c673cae
FG
143 free(args[i]);
144 }
145
11fdf7f2
TL
146 if (argcount) {
147 free(args);
148 }
7c673cae
FG
149}
150
151static char **
152spdk_push_arg(char *args[], int *argcount, char *arg)
153{
154 char **tmp;
155
156 if (arg == NULL) {
11fdf7f2
TL
157 fprintf(stderr, "%s: NULL arg supplied\n", __func__);
158 spdk_free_args(args, *argcount);
7c673cae
FG
159 return NULL;
160 }
161
162 tmp = realloc(args, sizeof(char *) * (*argcount + 1));
163 if (tmp == NULL) {
164 spdk_free_args(args, *argcount);
165 return NULL;
166 }
167
168 tmp[*argcount] = arg;
169 (*argcount)++;
170
171 return tmp;
172}
173
11fdf7f2
TL
174static void
175spdk_destruct_eal_cmdline(void)
176{
177 spdk_free_args(eal_cmdline, eal_cmdline_argcount);
178}
179
180
7c673cae 181static int
11fdf7f2 182spdk_build_eal_cmdline(const struct spdk_env_opts *opts)
7c673cae
FG
183{
184 int argcount = 0;
185 char **args;
186
7c673cae
FG
187 args = NULL;
188
189 /* set the program name */
190 args = spdk_push_arg(args, &argcount, _sprintf_alloc("%s", opts->name));
191 if (args == NULL) {
192 return -1;
193 }
194
11fdf7f2
TL
195 /* disable shared configuration files when in single process mode. This allows for cleaner shutdown */
196 if (opts->shm_id < 0) {
197 args = spdk_push_arg(args, &argcount, _sprintf_alloc("%s", "--no-shconf"));
198 if (args == NULL) {
199 return -1;
200 }
201 }
202
7c673cae 203 /* set the coremask */
11fdf7f2
TL
204 /* NOTE: If coremask starts with '[' and ends with ']' it is a core list
205 */
206 if (opts->core_mask[0] == '[') {
207 char *l_arg = _sprintf_alloc("-l %s", opts->core_mask + 1);
208 int len = strlen(l_arg);
209 if (l_arg[len - 1] == ']') {
210 l_arg[len - 1] = '\0';
211 }
212 args = spdk_push_arg(args, &argcount, l_arg);
213 } else {
214 args = spdk_push_arg(args, &argcount, _sprintf_alloc("-c %s", opts->core_mask));
215 }
216
7c673cae
FG
217 if (args == NULL) {
218 return -1;
219 }
220
221 /* set the memory channel number */
11fdf7f2
TL
222 if (opts->mem_channel > 0) {
223 args = spdk_push_arg(args, &argcount, _sprintf_alloc("-n %d", opts->mem_channel));
7c673cae
FG
224 if (args == NULL) {
225 return -1;
226 }
227 }
228
229 /* set the memory size */
11fdf7f2
TL
230 if (opts->mem_size >= 0) {
231 args = spdk_push_arg(args, &argcount, _sprintf_alloc("-m %d", opts->mem_size));
7c673cae
FG
232 if (args == NULL) {
233 return -1;
234 }
235 }
236
237 /* set the master core */
11fdf7f2 238 if (opts->master_core > 0) {
7c673cae 239 args = spdk_push_arg(args, &argcount, _sprintf_alloc("--master-lcore=%d",
11fdf7f2
TL
240 opts->master_core));
241 if (args == NULL) {
242 return -1;
243 }
244 }
245
246 /* set no pci if enabled */
247 if (opts->no_pci) {
248 args = spdk_push_arg(args, &argcount, _sprintf_alloc("--no-pci"));
249 if (args == NULL) {
250 return -1;
251 }
252 }
253
254 /* create just one hugetlbfs file */
255 if (opts->hugepage_single_segments) {
256 args = spdk_push_arg(args, &argcount, _sprintf_alloc("--single-file-segments"));
257 if (args == NULL) {
258 return -1;
259 }
260 }
261
262 /* unlink hugepages after initialization */
263 if (opts->unlink_hugepage) {
264 args = spdk_push_arg(args, &argcount, _sprintf_alloc("--huge-unlink"));
7c673cae
FG
265 if (args == NULL) {
266 return -1;
267 }
268 }
269
11fdf7f2
TL
270#if RTE_VERSION >= RTE_VERSION_NUM(18, 05, 0, 0) && RTE_VERSION < RTE_VERSION_NUM(18, 5, 1, 0)
271 /* Dynamic memory management is buggy in DPDK 18.05.0. Don't use it. */
272 args = spdk_push_arg(args, &argcount, _sprintf_alloc("--legacy-mem"));
273 if (args == NULL) {
274 return -1;
275 }
276#endif
277
278 if (opts->num_pci_addr) {
279 size_t i;
280 char bdf[32];
281 struct spdk_pci_addr *pci_addr =
282 opts->pci_blacklist ? opts->pci_blacklist : opts->pci_whitelist;
283
284 for (i = 0; i < opts->num_pci_addr; i++) {
285 spdk_pci_addr_fmt(bdf, 32, &pci_addr[i]);
286 args = spdk_push_arg(args, &argcount, _sprintf_alloc("%s=%s",
287 (opts->pci_blacklist ? "--pci-blacklist" : "--pci-whitelist"),
288 bdf));
289 if (args == NULL) {
290 return -1;
291 }
292 }
293 }
294
7c673cae
FG
295#ifdef __linux__
296 if (opts->shm_id < 0) {
297 args = spdk_push_arg(args, &argcount, _sprintf_alloc("--file-prefix=spdk_pid%d",
298 getpid()));
299 if (args == NULL) {
300 return -1;
301 }
302 } else {
303 args = spdk_push_arg(args, &argcount, _sprintf_alloc("--file-prefix=spdk%d",
304 opts->shm_id));
305 if (args == NULL) {
306 return -1;
307 }
308
11fdf7f2
TL
309 /* Set the base virtual address - it must be an address that is not in the
310 * ASAN shadow region, otherwise ASAN-enabled builds will ignore the
311 * mmap hint.
312 *
313 * Ref: https://github.com/google/sanitizers/wiki/AddressSanitizerAlgorithm
314 */
315 args = spdk_push_arg(args, &argcount, _sprintf_alloc("--base-virtaddr=0x200000000000"));
7c673cae
FG
316 if (args == NULL) {
317 return -1;
318 }
319
320 /* set the process type */
321 args = spdk_push_arg(args, &argcount, _sprintf_alloc("--proc-type=auto"));
322 if (args == NULL) {
323 return -1;
324 }
325 }
326#endif
327
11fdf7f2
TL
328 eal_cmdline = args;
329 eal_cmdline_argcount = argcount;
330 if (atexit(spdk_destruct_eal_cmdline) != 0) {
331 fprintf(stderr, "Failed to register cleanup handler\n");
332 }
7c673cae
FG
333
334 return argcount;
335}
336
11fdf7f2 337int spdk_env_init(const struct spdk_env_opts *opts)
7c673cae 338{
7c673cae 339 char **dpdk_args = NULL;
11fdf7f2
TL
340 int i, rc;
341 int orig_optind;
7c673cae 342
11fdf7f2
TL
343 rc = spdk_build_eal_cmdline(opts);
344 if (rc < 0) {
7c673cae 345 fprintf(stderr, "Invalid arguments to initialize DPDK\n");
11fdf7f2 346 return -1;
7c673cae
FG
347 }
348
11fdf7f2 349 printf("Starting %s / %s initialization...\n", SPDK_VERSION_STRING, rte_version());
7c673cae 350 printf("[ DPDK EAL parameters: ");
11fdf7f2
TL
351 for (i = 0; i < eal_cmdline_argcount; i++) {
352 printf("%s ", eal_cmdline[i]);
7c673cae
FG
353 }
354 printf("]\n");
355
356 /* DPDK rearranges the array we pass to it, so make a copy
357 * before passing so we can still free the individual strings
358 * correctly.
359 */
11fdf7f2 360 dpdk_args = calloc(eal_cmdline_argcount, sizeof(char *));
7c673cae
FG
361 if (dpdk_args == NULL) {
362 fprintf(stderr, "Failed to allocate dpdk_args\n");
11fdf7f2 363 return -1;
7c673cae 364 }
11fdf7f2 365 memcpy(dpdk_args, eal_cmdline, sizeof(char *) * eal_cmdline_argcount);
7c673cae
FG
366
367 fflush(stdout);
11fdf7f2
TL
368 orig_optind = optind;
369 optind = 1;
370 rc = rte_eal_init(eal_cmdline_argcount, dpdk_args);
371 optind = orig_optind;
7c673cae 372
7c673cae
FG
373 free(dpdk_args);
374
375 if (rc < 0) {
376 fprintf(stderr, "Failed to initialize DPDK\n");
11fdf7f2
TL
377 return -1;
378 }
379
380 if (opts->shm_id < 0 && !opts->hugepage_single_segments) {
381 /*
382 * Unlink hugepage and config info files after init. This will ensure they get
383 * deleted on app exit, even if the app crashes and does not exit normally.
384 * Only do this when not in multi-process mode, since for multi-process other
385 * apps will need to open these files. These files are not created for
386 * "single file segments".
387 */
388 spdk_env_unlink_shared_files();
389 }
390
391 if (spdk_mem_map_init() < 0) {
392 fprintf(stderr, "Failed to allocate mem_map\n");
393 return -1;
394 }
395 if (spdk_vtophys_init() < 0) {
396 fprintf(stderr, "Failed to initialize vtophys\n");
397 return -1;
7c673cae
FG
398 }
399
11fdf7f2 400 return 0;
7c673cae 401}