5d2adb9b31ea53794cb7644bb71a3506e874cf28
[dpdk.git] / lib / librte_eal / linuxapp / eal / eal.c
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5  *   Copyright(c) 2012-2014 6WIND S.A.
6  *   All rights reserved.
7  *
8  *   Redistribution and use in source and binary forms, with or without
9  *   modification, are permitted provided that the following conditions
10  *   are met:
11  *
12  *     * Redistributions of source code must retain the above copyright
13  *       notice, this list of conditions and the following disclaimer.
14  *     * Redistributions in binary form must reproduce the above copyright
15  *       notice, this list of conditions and the following disclaimer in
16  *       the documentation and/or other materials provided with the
17  *       distribution.
18  *     * Neither the name of Intel Corporation nor the names of its
19  *       contributors may be used to endorse or promote products derived
20  *       from this software without specific prior written permission.
21  *
22  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34
35 #include <stdio.h>
36 #include <stdlib.h>
37 #include <stdint.h>
38 #include <string.h>
39 #include <stdarg.h>
40 #include <unistd.h>
41 #include <pthread.h>
42 #include <syslog.h>
43 #include <getopt.h>
44 #include <sys/file.h>
45 #include <fcntl.h>
46 #include <dlfcn.h>
47 #include <stddef.h>
48 #include <errno.h>
49 #include <limits.h>
50 #include <errno.h>
51 #include <sys/mman.h>
52 #include <sys/queue.h>
53 #include <sys/io.h>
54
55 #include <rte_common.h>
56 #include <rte_debug.h>
57 #include <rte_memory.h>
58 #include <rte_memzone.h>
59 #include <rte_launch.h>
60 #include <rte_tailq.h>
61 #include <rte_eal.h>
62 #include <rte_eal_memconfig.h>
63 #include <rte_per_lcore.h>
64 #include <rte_lcore.h>
65 #include <rte_log.h>
66 #include <rte_random.h>
67 #include <rte_cycles.h>
68 #include <rte_string_fns.h>
69 #include <rte_cpuflags.h>
70 #include <rte_interrupts.h>
71 #include <rte_pci.h>
72 #include <rte_devargs.h>
73 #include <rte_common.h>
74 #include <rte_version.h>
75 #include <rte_atomic.h>
76 #include <malloc_heap.h>
77 #include <rte_eth_ring.h>
78 #include <rte_dev.h>
79
80 #include "eal_private.h"
81 #include "eal_thread.h"
82 #include "eal_internal_cfg.h"
83 #include "eal_filesystem.h"
84 #include "eal_hugepages.h"
85
86 #define OPT_HUGE_DIR    "huge-dir"
87 #define OPT_PROC_TYPE   "proc-type"
88 #define OPT_NO_SHCONF   "no-shconf"
89 #define OPT_NO_HPET     "no-hpet"
90 #define OPT_VMWARE_TSC_MAP   "vmware-tsc-map"
91 #define OPT_NO_PCI      "no-pci"
92 #define OPT_NO_HUGE     "no-huge"
93 #define OPT_FILE_PREFIX "file-prefix"
94 #define OPT_SOCKET_MEM  "socket-mem"
95 #define OPT_USE_DEVICE  "use-device"
96 #define OPT_PCI_WHITELIST "pci-whitelist"
97 #define OPT_PCI_BLACKLIST "pci-blacklist"
98 #define OPT_VDEV        "vdev"
99 #define OPT_SYSLOG      "syslog"
100 #define OPT_BASE_VIRTADDR   "base-virtaddr"
101 #define OPT_XEN_DOM0    "xen-dom0"
102 #define OPT_CREATE_UIO_DEV "create-uio-dev"
103 #define OPT_VFIO_INTR    "vfio-intr"
104
105 #define RTE_EAL_BLACKLIST_SIZE  0x100
106
107 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
108
109 #define SOCKET_MEM_STRLEN (RTE_MAX_NUMA_NODES * 10)
110
111 #define HIGHEST_RPL 3
112
113 #define BITS_PER_HEX 4
114
115 /* Allow the application to print its usage message too if set */
116 static rte_usage_hook_t rte_application_usage_hook = NULL;
117
118 TAILQ_HEAD(shared_driver_list, shared_driver);
119
120 /* Definition for shared object drivers. */
121 struct shared_driver {
122         TAILQ_ENTRY(shared_driver) next;
123
124         char    name[PATH_MAX];
125         void*   lib_handle;
126 };
127
128 /* List of external loadable drivers */
129 static struct shared_driver_list solib_list =
130 TAILQ_HEAD_INITIALIZER(solib_list);
131
132 /* early configuration structure, when memory config is not mmapped */
133 static struct rte_mem_config early_mem_config;
134
135 /* define fd variable here, because file needs to be kept open for the
136  * duration of the program, as we hold a write lock on it in the primary proc */
137 static int mem_cfg_fd = -1;
138
139 static struct flock wr_lock = {
140                 .l_type = F_WRLCK,
141                 .l_whence = SEEK_SET,
142                 .l_start = offsetof(struct rte_mem_config, memseg),
143                 .l_len = sizeof(early_mem_config.memseg),
144 };
145
146 /* Address of global and public configuration */
147 static struct rte_config rte_config = {
148                 .mem_config = &early_mem_config,
149 };
150
151 /* internal configuration (per-core) */
152 struct lcore_config lcore_config[RTE_MAX_LCORE];
153
154 /* internal configuration */
155 struct internal_config internal_config;
156
157 /* used by rte_rdtsc() */
158 int rte_cycles_vmware_tsc_map;
159
160 /* Return a pointer to the configuration structure */
161 struct rte_config *
162 rte_eal_get_configuration(void)
163 {
164         return &rte_config;
165 }
166
167 /* parse a sysfs (or other) file containing one integer value */
168 int
169 eal_parse_sysfs_value(const char *filename, unsigned long *val)
170 {
171         FILE *f;
172         char buf[BUFSIZ];
173         char *end = NULL;
174
175         if ((f = fopen(filename, "r")) == NULL) {
176                 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs value %s\n",
177                         __func__, filename);
178                 return -1;
179         }
180
181         if (fgets(buf, sizeof(buf), f) == NULL) {
182                 RTE_LOG(ERR, EAL, "%s(): cannot read sysfs value %s\n",
183                         __func__, filename);
184                 fclose(f);
185                 return -1;
186         }
187         *val = strtoul(buf, &end, 0);
188         if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
189                 RTE_LOG(ERR, EAL, "%s(): cannot parse sysfs value %s\n",
190                                 __func__, filename);
191                 fclose(f);
192                 return -1;
193         }
194         fclose(f);
195         return 0;
196 }
197
198
199 /* create memory configuration in shared/mmap memory. Take out
200  * a write lock on the memsegs, so we can auto-detect primary/secondary.
201  * This means we never close the file while running (auto-close on exit).
202  * We also don't lock the whole file, so that in future we can use read-locks
203  * on other parts, e.g. memzones, to detect if there are running secondary
204  * processes. */
205 static void
206 rte_eal_config_create(void)
207 {
208         void *rte_mem_cfg_addr;
209         int retval;
210
211         const char *pathname = eal_runtime_config_path();
212
213         if (internal_config.no_shconf)
214                 return;
215
216         if (mem_cfg_fd < 0){
217                 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0660);
218                 if (mem_cfg_fd < 0)
219                         rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
220         }
221
222         retval = ftruncate(mem_cfg_fd, sizeof(*rte_config.mem_config));
223         if (retval < 0){
224                 close(mem_cfg_fd);
225                 rte_panic("Cannot resize '%s' for rte_mem_config\n", pathname);
226         }
227
228         retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
229         if (retval < 0){
230                 close(mem_cfg_fd);
231                 rte_exit(EXIT_FAILURE, "Cannot create lock on '%s'. Is another primary "
232                                 "process running?\n", pathname);
233         }
234
235         rte_mem_cfg_addr = mmap(NULL, sizeof(*rte_config.mem_config),
236                                 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
237
238         if (rte_mem_cfg_addr == MAP_FAILED){
239                 rte_panic("Cannot mmap memory for rte_config\n");
240         }
241         memcpy(rte_mem_cfg_addr, &early_mem_config, sizeof(early_mem_config));
242         rte_config.mem_config = (struct rte_mem_config *) rte_mem_cfg_addr;
243
244         /* store address of the config in the config itself so that secondary
245          * processes could later map the config into this exact location */
246         rte_config.mem_config->mem_cfg_addr = (uintptr_t) rte_mem_cfg_addr;
247
248 }
249
250 /* attach to an existing shared memory config */
251 static void
252 rte_eal_config_attach(void)
253 {
254         struct rte_mem_config *mem_config;
255
256         const char *pathname = eal_runtime_config_path();
257
258         if (internal_config.no_shconf)
259                 return;
260
261         if (mem_cfg_fd < 0){
262                 mem_cfg_fd = open(pathname, O_RDWR);
263                 if (mem_cfg_fd < 0)
264                         rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
265         }
266
267         /* map it as read-only first */
268         mem_config = (struct rte_mem_config *) mmap(NULL, sizeof(*mem_config),
269                         PROT_READ, MAP_SHARED, mem_cfg_fd, 0);
270         if (mem_config == MAP_FAILED)
271                 rte_panic("Cannot mmap memory for rte_config\n");
272
273         rte_config.mem_config = mem_config;
274 }
275
276 /* reattach the shared config at exact memory location primary process has it */
277 static void
278 rte_eal_config_reattach(void)
279 {
280         struct rte_mem_config *mem_config;
281         void *rte_mem_cfg_addr;
282
283         if (internal_config.no_shconf)
284                 return;
285
286         /* save the address primary process has mapped shared config to */
287         rte_mem_cfg_addr = (void *) (uintptr_t) rte_config.mem_config->mem_cfg_addr;
288
289         /* unmap original config */
290         munmap(rte_config.mem_config, sizeof(struct rte_mem_config));
291
292         /* remap the config at proper address */
293         mem_config = (struct rte_mem_config *) mmap(rte_mem_cfg_addr,
294                         sizeof(*mem_config), PROT_READ | PROT_WRITE, MAP_SHARED,
295                         mem_cfg_fd, 0);
296         close(mem_cfg_fd);
297         if (mem_config == MAP_FAILED || mem_config != rte_mem_cfg_addr)
298                 rte_panic("Cannot mmap memory for rte_config\n");
299
300         rte_config.mem_config = mem_config;
301 }
302
303 /* Detect if we are a primary or a secondary process */
304 static enum rte_proc_type_t
305 eal_proc_type_detect(void)
306 {
307         enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
308         const char *pathname = eal_runtime_config_path();
309
310         /* if we can open the file but not get a write-lock we are a secondary
311          * process. NOTE: if we get a file handle back, we keep that open
312          * and don't close it to prevent a race condition between multiple opens */
313         if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
314                         (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
315                 ptype = RTE_PROC_SECONDARY;
316
317         RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
318                         ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
319
320         return ptype;
321 }
322
323 /* Sets up rte_config structure with the pointer to shared memory config.*/
324 static void
325 rte_config_init(void)
326 {
327         rte_config.process_type = (internal_config.process_type == RTE_PROC_AUTO) ?
328                         eal_proc_type_detect() : /* for auto, detect the type */
329                         internal_config.process_type; /* otherwise use what's already set */
330
331         switch (rte_config.process_type){
332         case RTE_PROC_PRIMARY:
333                 rte_eal_config_create();
334                 break;
335         case RTE_PROC_SECONDARY:
336                 rte_eal_config_attach();
337                 rte_eal_mcfg_wait_complete(rte_config.mem_config);
338                 rte_eal_config_reattach();
339                 break;
340         case RTE_PROC_AUTO:
341         case RTE_PROC_INVALID:
342                 rte_panic("Invalid process type\n");
343         }
344 }
345
346 /* Unlocks hugepage directories that were locked by eal_hugepage_info_init */
347 static void
348 eal_hugedirs_unlock(void)
349 {
350         int i;
351
352         for (i = 0; i < MAX_HUGEPAGE_SIZES; i++)
353         {
354                 /* skip uninitialized */
355                 if (internal_config.hugepage_info[i].lock_descriptor < 0)
356                         continue;
357                 /* unlock hugepage file */
358                 flock(internal_config.hugepage_info[i].lock_descriptor, LOCK_UN);
359                 close(internal_config.hugepage_info[i].lock_descriptor);
360                 /* reset the field */
361                 internal_config.hugepage_info[i].lock_descriptor = -1;
362         }
363 }
364
365 /* display usage */
366 static void
367 eal_usage(const char *prgname)
368 {
369         printf("\nUsage: %s -c COREMASK -n NUM [-m NB] [-r NUM] [-b <domain:bus:devid.func>]"
370                "[--proc-type primary|secondary|auto] \n\n"
371                "EAL options:\n"
372                "  -c COREMASK  : A hexadecimal bitmask of cores to run on\n"
373                "  -n NUM       : Number of memory channels\n"
374                    "  -v           : Display version information on startup\n"
375                "  -d LIB.so    : add driver (can be used multiple times)\n"
376                "  -m MB        : memory to allocate (see also --"OPT_SOCKET_MEM")\n"
377                "  -r NUM       : force number of memory ranks (don't detect)\n"
378                "  --"OPT_XEN_DOM0" : support application running on Xen Domain0 "
379                            "without hugetlbfs\n"
380                "  --"OPT_SYSLOG"     : set syslog facility\n"
381                "  --"OPT_SOCKET_MEM" : memory to allocate on specific \n"
382                    "                 sockets (use comma separated values)\n"
383                "  --"OPT_HUGE_DIR"   : directory where hugetlbfs is mounted\n"
384                "  --"OPT_PROC_TYPE"  : type of this process\n"
385                "  --"OPT_FILE_PREFIX": prefix for hugepage filenames\n"
386                "  --"OPT_PCI_BLACKLIST", -b: add a PCI device in black list.\n"
387                "               Prevent EAL from using this PCI device. The argument\n"
388                "               format is <domain:bus:devid.func>.\n"
389                "  --"OPT_PCI_WHITELIST", -w: add a PCI device in white list.\n"
390                "               Only use the specified PCI devices. The argument format\n"
391                "               is <[domain:]bus:devid.func>. This option can be present\n"
392                "               several times (once per device).\n"
393                "               [NOTE: PCI whitelist cannot be used with -b option]\n"
394                "  --"OPT_VDEV": add a virtual device.\n"
395                "               The argument format is <driver><id>[,key=val,...]\n"
396                "               (ex: --vdev=eth_pcap0,iface=eth2).\n"
397                "  --"OPT_VMWARE_TSC_MAP": use VMware TSC map instead of native RDTSC\n"
398                "  --"OPT_BASE_VIRTADDR": specify base virtual address\n"
399                "  --"OPT_VFIO_INTR": specify desired interrupt mode for VFIO "
400                            "(legacy|msi|msix)\n"
401                "  --"OPT_CREATE_UIO_DEV": create /dev/uioX (usually done by hotplug)\n"
402                "\nEAL options for DEBUG use only:\n"
403                "  --"OPT_NO_HUGE"  : use malloc instead of hugetlbfs\n"
404                "  --"OPT_NO_PCI"   : disable pci\n"
405                "  --"OPT_NO_HPET"  : disable hpet\n"
406                "  --"OPT_NO_SHCONF": no shared config (mmap'd files)\n"
407                "\n",
408                prgname);
409         /* Allow the application to print its usage message too if hook is set */
410         if ( rte_application_usage_hook ) {
411                 printf("===== Application Usage =====\n\n");
412                 rte_application_usage_hook(prgname);
413         }
414 }
415
416 /* Set a per-application usage message */
417 rte_usage_hook_t
418 rte_set_application_usage_hook( rte_usage_hook_t usage_func )
419 {
420         rte_usage_hook_t        old_func;
421
422         /* Will be NULL on the first call to denote the last usage routine. */
423         old_func                                        = rte_application_usage_hook;
424         rte_application_usage_hook      = usage_func;
425
426         return old_func;
427 }
428
429 /*
430  * Parse the coremask given as argument (hexadecimal string) and fill
431  * the global configuration (core role and core count) with the parsed
432  * value.
433  */
434 static int xdigit2val(unsigned char c)
435 {
436         int val;
437         if(isdigit(c))
438                 val = c - '0';
439         else if(isupper(c))
440                 val = c - 'A' + 10;
441         else
442                 val = c - 'a' + 10;
443         return val;
444 }
445 static int
446 eal_parse_coremask(const char *coremask)
447 {
448         struct rte_config *cfg = rte_eal_get_configuration();
449         int i, j, idx = 0 ;
450         unsigned count = 0;
451         char c;
452         int val;
453
454         if (coremask == NULL)
455                 return -1;
456         /* Remove all blank characters ahead and after .
457          * Remove 0x/0X if exists.
458          */
459         while (isblank(*coremask))
460                 coremask++;
461         if (coremask[0] == '0' && ((coremask[1] == 'x')
462                 ||  (coremask[1] == 'X')) )
463                 coremask += 2;
464         i = strnlen(coremask, PATH_MAX);
465         while ((i > 0) && isblank(coremask[i - 1]))
466                 i--;
467         if (i == 0)
468                 return -1;
469
470         for (i = i - 1; i >= 0 && idx < RTE_MAX_LCORE; i--) {
471                 c = coremask[i];
472                 if (isxdigit(c) == 0) {
473                         /* invalid characters */
474                         return (-1);
475                 }
476                 val = xdigit2val(c);
477                 for(j = 0; j < BITS_PER_HEX && idx < RTE_MAX_LCORE; j++, idx++) {
478                         if((1 << j) & val) {
479                                 if (!lcore_config[idx].detected) {
480                                         RTE_LOG(ERR, EAL, "lcore %u "
481                                                 "unavailable\n", idx);
482                                         return -1;
483                                 }
484                                 cfg->lcore_role[idx] = ROLE_RTE;
485                                 if(count == 0)
486                                         cfg->master_lcore = idx;
487                                 count++;
488                         } else  {
489                                 cfg->lcore_role[idx] = ROLE_OFF;
490                         }
491                 }
492         }
493         for(; i >= 0; i--)
494                 if(coremask[i] != '0')
495                         return -1;
496         for(; idx < RTE_MAX_LCORE; idx++)
497                 cfg->lcore_role[idx] = ROLE_OFF;
498         if(count == 0)
499                 return -1;
500         /* Update the count of enabled logical cores of the EAL configuration */
501         cfg->lcore_count = count;
502         return 0;
503 }
504
505 static int
506 eal_parse_syslog(const char *facility)
507 {
508         int i;
509         static struct {
510                 const char *name;
511                 int value;
512         } map[] = {
513                 { "auth", LOG_AUTH },
514                 { "cron", LOG_CRON },
515                 { "daemon", LOG_DAEMON },
516                 { "ftp", LOG_FTP },
517                 { "kern", LOG_KERN },
518                 { "lpr", LOG_LPR },
519                 { "mail", LOG_MAIL },
520                 { "news", LOG_NEWS },
521                 { "syslog", LOG_SYSLOG },
522                 { "user", LOG_USER },
523                 { "uucp", LOG_UUCP },
524                 { "local0", LOG_LOCAL0 },
525                 { "local1", LOG_LOCAL1 },
526                 { "local2", LOG_LOCAL2 },
527                 { "local3", LOG_LOCAL3 },
528                 { "local4", LOG_LOCAL4 },
529                 { "local5", LOG_LOCAL5 },
530                 { "local6", LOG_LOCAL6 },
531                 { "local7", LOG_LOCAL7 },
532                 { NULL, 0 }
533         };
534
535         for (i = 0; map[i].name; i++) {
536                 if (!strcmp(facility, map[i].name)) {
537                         internal_config.syslog_facility = map[i].value;
538                         return 0;
539                 }
540         }
541         return -1;
542 }
543
544 static int
545 eal_parse_socket_mem(char *socket_mem)
546 {
547         char * arg[RTE_MAX_NUMA_NODES];
548         char *end;
549         int arg_num, i, len;
550         uint64_t total_mem = 0;
551
552         len = strnlen(socket_mem, SOCKET_MEM_STRLEN);
553         if (len == SOCKET_MEM_STRLEN) {
554                 RTE_LOG(ERR, EAL, "--socket-mem is too long\n");
555                 return -1;
556         }
557
558         /* all other error cases will be caught later */
559         if (!isdigit(socket_mem[len-1]))
560                 return -1;
561
562         /* split the optarg into separate socket values */
563         arg_num = rte_strsplit(socket_mem, len,
564                         arg, RTE_MAX_NUMA_NODES, ',');
565
566         /* if split failed, or 0 arguments */
567         if (arg_num <= 0)
568                 return -1;
569
570         internal_config.force_sockets = 1;
571
572         /* parse each defined socket option */
573         errno = 0;
574         for (i = 0; i < arg_num; i++) {
575                 end = NULL;
576                 internal_config.socket_mem[i] = strtoull(arg[i], &end, 10);
577
578                 /* check for invalid input */
579                 if ((errno != 0)  ||
580                                 (arg[i][0] == '\0') || (end == NULL) || (*end != '\0'))
581                         return -1;
582                 internal_config.socket_mem[i] *= 1024ULL;
583                 internal_config.socket_mem[i] *= 1024ULL;
584                 total_mem += internal_config.socket_mem[i];
585         }
586
587         /* check if we have a positive amount of total memory */
588         if (total_mem == 0)
589                 return -1;
590
591         return 0;
592 }
593
594 static int
595 eal_parse_base_virtaddr(const char *arg)
596 {
597         char *end;
598         uint64_t addr;
599
600         errno = 0;
601         addr = strtoull(arg, &end, 16);
602
603         /* check for errors */
604         if ((errno != 0) || (arg[0] == '\0') || end == NULL || (*end != '\0'))
605                 return -1;
606
607         /* make sure we don't exceed 32-bit boundary on 32-bit target */
608 #ifndef RTE_ARCH_X86_64
609         if (addr >= UINTPTR_MAX)
610                 return -1;
611 #endif
612
613         /* align the addr on 2M boundary */
614         internal_config.base_virtaddr = RTE_PTR_ALIGN_CEIL((uintptr_t)addr,
615                                                            RTE_PGSIZE_2M);
616
617         return 0;
618 }
619
620 static int
621 eal_parse_vfio_intr(const char *mode)
622 {
623         unsigned i;
624         static struct {
625                 const char *name;
626                 enum rte_intr_mode value;
627         } map[] = {
628                 { "legacy", RTE_INTR_MODE_LEGACY },
629                 { "msi", RTE_INTR_MODE_MSI },
630                 { "msix", RTE_INTR_MODE_MSIX },
631         };
632
633         for (i = 0; i < RTE_DIM(map); i++) {
634                 if (!strcmp(mode, map[i].name)) {
635                         internal_config.vfio_intr_mode = map[i].value;
636                         return 0;
637                 }
638         }
639         return -1;
640 }
641
642 static inline size_t
643 eal_get_hugepage_mem_size(void)
644 {
645         uint64_t size = 0;
646         unsigned i, j;
647
648         for (i = 0; i < internal_config.num_hugepage_sizes; i++) {
649                 struct hugepage_info *hpi = &internal_config.hugepage_info[i];
650                 if (hpi->hugedir != NULL) {
651                         for (j = 0; j < RTE_MAX_NUMA_NODES; j++) {
652                                 size += hpi->hugepage_sz * hpi->num_pages[j];
653                         }
654                 }
655         }
656
657         return (size < SIZE_MAX) ? (size_t)(size) : SIZE_MAX;
658 }
659
660 static enum rte_proc_type_t
661 eal_parse_proc_type(const char *arg)
662 {
663         if (strncasecmp(arg, "primary", sizeof("primary")) == 0)
664                 return RTE_PROC_PRIMARY;
665         if (strncasecmp(arg, "secondary", sizeof("secondary")) == 0)
666                 return RTE_PROC_SECONDARY;
667         if (strncasecmp(arg, "auto", sizeof("auto")) == 0)
668                 return RTE_PROC_AUTO;
669
670         return RTE_PROC_INVALID;
671 }
672
673 /* Parse the argument given in the command line of the application */
674 static int
675 eal_parse_args(int argc, char **argv)
676 {
677         int opt, ret, i;
678         char **argvopt;
679         int option_index;
680         int coremask_ok = 0;
681         char *prgname = argv[0];
682         static struct option lgopts[] = {
683                 {OPT_NO_HUGE, 0, 0, 0},
684                 {OPT_NO_PCI, 0, 0, 0},
685                 {OPT_NO_HPET, 0, 0, 0},
686                 {OPT_VMWARE_TSC_MAP, 0, 0, 0},
687                 {OPT_HUGE_DIR, 1, 0, 0},
688                 {OPT_NO_SHCONF, 0, 0, 0},
689                 {OPT_PROC_TYPE, 1, 0, 0},
690                 {OPT_FILE_PREFIX, 1, 0, 0},
691                 {OPT_SOCKET_MEM, 1, 0, 0},
692                 {OPT_PCI_WHITELIST, 1, 0, 0},
693                 {OPT_PCI_BLACKLIST, 1, 0, 0},
694                 {OPT_VDEV, 1, 0, 0},
695                 {OPT_SYSLOG, 1, NULL, 0},
696                 {OPT_VFIO_INTR, 1, NULL, 0},
697                 {OPT_BASE_VIRTADDR, 1, 0, 0},
698                 {OPT_XEN_DOM0, 0, 0, 0},
699                 {OPT_CREATE_UIO_DEV, 1, NULL, 0},
700                 {0, 0, 0, 0}
701         };
702         struct shared_driver *solib;
703
704         argvopt = argv;
705
706         internal_config.memory = 0;
707         internal_config.force_nrank = 0;
708         internal_config.force_nchannel = 0;
709         internal_config.hugefile_prefix = HUGEFILE_PREFIX_DEFAULT;
710         internal_config.hugepage_dir = NULL;
711         internal_config.force_sockets = 0;
712         internal_config.syslog_facility = LOG_DAEMON;
713         internal_config.xen_dom0_support = 0;
714         /* if set to NONE, interrupt mode is determined automatically */
715         internal_config.vfio_intr_mode = RTE_INTR_MODE_NONE;
716 #ifdef RTE_LIBEAL_USE_HPET
717         internal_config.no_hpet = 0;
718 #else
719         internal_config.no_hpet = 1;
720 #endif
721         /* zero out the NUMA config */
722         for (i = 0; i < RTE_MAX_NUMA_NODES; i++)
723                 internal_config.socket_mem[i] = 0;
724
725         /* zero out hugedir descriptors */
726         for (i = 0; i < MAX_HUGEPAGE_SIZES; i++)
727                 internal_config.hugepage_info[i].lock_descriptor = -1;
728
729         internal_config.vmware_tsc_map = 0;
730         internal_config.base_virtaddr = 0;
731
732         while ((opt = getopt_long(argc, argvopt, "b:w:c:d:m:n:r:v",
733                                   lgopts, &option_index)) != EOF) {
734
735                 switch (opt) {
736                 /* blacklist */
737                 case 'b':
738                         if (rte_eal_devargs_add(RTE_DEVTYPE_BLACKLISTED_PCI,
739                                         optarg) < 0) {
740                                 eal_usage(prgname);
741                                 return (-1);
742                         }
743                         break;
744                 /* whitelist */
745                 case 'w':
746                         if (rte_eal_devargs_add(RTE_DEVTYPE_WHITELISTED_PCI,
747                                         optarg) < 0) {
748                                 eal_usage(prgname);
749                                 return -1;
750                         }
751                         break;
752                 /* coremask */
753                 case 'c':
754                         if (eal_parse_coremask(optarg) < 0) {
755                                 RTE_LOG(ERR, EAL, "invalid coremask\n");
756                                 eal_usage(prgname);
757                                 return -1;
758                         }
759                         coremask_ok = 1;
760                         break;
761                 /* force loading of external driver */
762                 case 'd':
763                         solib = malloc(sizeof(*solib));
764                         if (solib == NULL) {
765                                 RTE_LOG(ERR, EAL, "malloc(solib) failed\n");
766                                 return -1;
767                         }
768                         memset(solib, 0, sizeof(*solib));
769                         strncpy(solib->name, optarg, PATH_MAX-1);
770                         solib->name[PATH_MAX-1] = 0;
771                         TAILQ_INSERT_TAIL(&solib_list, solib, next);
772                         break;
773                 /* size of memory */
774                 case 'm':
775                         internal_config.memory = atoi(optarg);
776                         internal_config.memory *= 1024ULL;
777                         internal_config.memory *= 1024ULL;
778                         break;
779                 /* force number of channels */
780                 case 'n':
781                         internal_config.force_nchannel = atoi(optarg);
782                         if (internal_config.force_nchannel == 0 ||
783                             internal_config.force_nchannel > 4) {
784                                 RTE_LOG(ERR, EAL, "invalid channel number\n");
785                                 eal_usage(prgname);
786                                 return -1;
787                         }
788                         break;
789                 /* force number of ranks */
790                 case 'r':
791                         internal_config.force_nrank = atoi(optarg);
792                         if (internal_config.force_nrank == 0 ||
793                             internal_config.force_nrank > 16) {
794                                 RTE_LOG(ERR, EAL, "invalid rank number\n");
795                                 eal_usage(prgname);
796                                 return -1;
797                         }
798                         break;
799                 case 'v':
800                         /* since message is explicitly requested by user, we
801                          * write message at highest log level so it can always be seen
802                          * even if info or warning messages are disabled */
803                         RTE_LOG(CRIT, EAL, "RTE Version: '%s'\n", rte_version());
804                         break;
805
806                 /* long options */
807                 case 0:
808                         if (!strcmp(lgopts[option_index].name, OPT_NO_HUGE)) {
809                                 internal_config.no_hugetlbfs = 1;
810                         }
811                         if (!strcmp(lgopts[option_index].name, OPT_XEN_DOM0)) {
812                 #ifdef RTE_LIBRTE_XEN_DOM0
813                                 internal_config.xen_dom0_support = 1;
814                 #else
815                                 RTE_LOG(ERR, EAL, "Can't support DPDK app "
816                                         "running on Dom0, please configure"
817                                         " RTE_LIBRTE_XEN_DOM0=y\n");
818                                 return -1;
819                 #endif
820                         }
821                         else if (!strcmp(lgopts[option_index].name, OPT_NO_PCI)) {
822                                 internal_config.no_pci = 1;
823                         }
824                         else if (!strcmp(lgopts[option_index].name, OPT_NO_HPET)) {
825                                 internal_config.no_hpet = 1;
826                         }
827                         else if (!strcmp(lgopts[option_index].name, OPT_VMWARE_TSC_MAP)) {
828                                 internal_config.vmware_tsc_map = 1;
829                         }
830                         else if (!strcmp(lgopts[option_index].name, OPT_NO_SHCONF)) {
831                                 internal_config.no_shconf = 1;
832                         }
833                         else if (!strcmp(lgopts[option_index].name, OPT_HUGE_DIR)) {
834                                 internal_config.hugepage_dir = optarg;
835                         }
836                         else if (!strcmp(lgopts[option_index].name, OPT_PROC_TYPE)) {
837                                 internal_config.process_type = eal_parse_proc_type(optarg);
838                         }
839                         else if (!strcmp(lgopts[option_index].name, OPT_FILE_PREFIX)) {
840                                 internal_config.hugefile_prefix = optarg;
841                         }
842                         else if (!strcmp(lgopts[option_index].name, OPT_SOCKET_MEM)) {
843                                 if (eal_parse_socket_mem(optarg) < 0) {
844                                         RTE_LOG(ERR, EAL, "invalid parameters for --"
845                                                         OPT_SOCKET_MEM "\n");
846                                         eal_usage(prgname);
847                                         return -1;
848                                 }
849                         }
850                         else if (!strcmp(lgopts[option_index].name, OPT_USE_DEVICE)) {
851                                 printf("The --use-device option is deprecated, please use\n"
852                                         "--whitelist or --vdev instead.\n");
853                                 eal_usage(prgname);
854                                 return -1;
855                         }
856                         else if (!strcmp(lgopts[option_index].name, OPT_PCI_BLACKLIST)) {
857                                 if (rte_eal_devargs_add(RTE_DEVTYPE_BLACKLISTED_PCI,
858                                                 optarg) < 0) {
859                                         eal_usage(prgname);
860                                         return -1;
861                                 }
862                         }
863                         else if (!strcmp(lgopts[option_index].name, OPT_PCI_WHITELIST)) {
864                                 if (rte_eal_devargs_add(RTE_DEVTYPE_WHITELISTED_PCI,
865                                                 optarg) < 0) {
866                                         eal_usage(prgname);
867                                         return -1;
868                                 }
869                         }
870                         else if (!strcmp(lgopts[option_index].name, OPT_VDEV)) {
871                                 if (rte_eal_devargs_add(RTE_DEVTYPE_VIRTUAL,
872                                                 optarg) < 0) {
873                                         eal_usage(prgname);
874                                         return -1;
875                                 }
876                         }
877                         else if (!strcmp(lgopts[option_index].name, OPT_SYSLOG)) {
878                                 if (eal_parse_syslog(optarg) < 0) {
879                                         RTE_LOG(ERR, EAL, "invalid parameters for --"
880                                                         OPT_SYSLOG "\n");
881                                         eal_usage(prgname);
882                                         return -1;
883                                 }
884                         }
885                         else if (!strcmp(lgopts[option_index].name, OPT_BASE_VIRTADDR)) {
886                                 if (eal_parse_base_virtaddr(optarg) < 0) {
887                                         RTE_LOG(ERR, EAL, "invalid parameter for --"
888                                                         OPT_BASE_VIRTADDR "\n");
889                                         eal_usage(prgname);
890                                         return -1;
891                                 }
892                         }
893                         else if (!strcmp(lgopts[option_index].name, OPT_VFIO_INTR)) {
894                                 if (eal_parse_vfio_intr(optarg) < 0) {
895                                         RTE_LOG(ERR, EAL, "invalid parameters for --"
896                                                         OPT_VFIO_INTR "\n");
897                                         eal_usage(prgname);
898                                         return -1;
899                                 }
900                         }
901                         else if (!strcmp(lgopts[option_index].name, OPT_CREATE_UIO_DEV)) {
902                                 internal_config.create_uio_dev = 1;
903                         }
904                         break;
905
906                 default:
907                         eal_usage(prgname);
908                         return -1;
909                 }
910         }
911
912         /* sanity checks */
913         if (!coremask_ok) {
914                 RTE_LOG(ERR, EAL, "coremask not specified\n");
915                 eal_usage(prgname);
916                 return -1;
917         }
918         if (internal_config.process_type == RTE_PROC_AUTO){
919                 internal_config.process_type = eal_proc_type_detect();
920         }
921         if (internal_config.process_type == RTE_PROC_INVALID){
922                 RTE_LOG(ERR, EAL, "Invalid process type specified\n");
923                 eal_usage(prgname);
924                 return -1;
925         }
926         if (internal_config.process_type == RTE_PROC_PRIMARY &&
927                         internal_config.force_nchannel == 0) {
928                 RTE_LOG(ERR, EAL, "Number of memory channels (-n) not specified\n");
929                 eal_usage(prgname);
930                 return -1;
931         }
932         if (index(internal_config.hugefile_prefix,'%') != NULL){
933                 RTE_LOG(ERR, EAL, "Invalid char, '%%', in '"OPT_FILE_PREFIX"' option\n");
934                 eal_usage(prgname);
935                 return -1;
936         }
937         if (internal_config.memory > 0 && internal_config.force_sockets == 1) {
938                 RTE_LOG(ERR, EAL, "Options -m and --socket-mem cannot be specified "
939                                 "at the same time\n");
940                 eal_usage(prgname);
941                 return -1;
942         }
943         /* --no-huge doesn't make sense with either -m or --socket-mem */
944         if (internal_config.no_hugetlbfs &&
945                         (internal_config.memory > 0 ||
946                                         internal_config.force_sockets == 1)) {
947                 RTE_LOG(ERR, EAL, "Options -m or --socket-mem cannot be specified "
948                                 "together with --no-huge!\n");
949                 eal_usage(prgname);
950                 return -1;
951         }
952         /* --xen-dom0 doesn't make sense with --socket-mem */
953         if (internal_config.xen_dom0_support && internal_config.force_sockets == 1) {
954                 RTE_LOG(ERR, EAL, "Options --socket-mem cannot be specified "
955                                         "together with --xen_dom0!\n");
956                 eal_usage(prgname);
957                 return -1;
958         }
959
960         if (rte_eal_devargs_type_count(RTE_DEVTYPE_WHITELISTED_PCI) != 0 &&
961                 rte_eal_devargs_type_count(RTE_DEVTYPE_BLACKLISTED_PCI) != 0) {
962                 RTE_LOG(ERR, EAL, "Error: blacklist [-b] and whitelist "
963                         "[-w] options cannot be used at the same time\n");
964                 eal_usage(prgname);
965                 return -1;
966         }
967
968         if (optind >= 0)
969                 argv[optind-1] = prgname;
970
971         /* if no memory amounts were requested, this will result in 0 and
972          * will be overriden later, right after eal_hugepage_info_init() */
973         for (i = 0; i < RTE_MAX_NUMA_NODES; i++)
974                 internal_config.memory += internal_config.socket_mem[i];
975
976         ret = optind-1;
977         optind = 0; /* reset getopt lib */
978         return ret;
979 }
980
981 static void
982 eal_check_mem_on_local_socket(void)
983 {
984         const struct rte_memseg *ms;
985         int i, socket_id;
986
987         socket_id = rte_lcore_to_socket_id(rte_config.master_lcore);
988
989         ms = rte_eal_get_physmem_layout();
990
991         for (i = 0; i < RTE_MAX_MEMSEG; i++)
992                 if (ms[i].socket_id == socket_id &&
993                                 ms[i].len > 0)
994                         return;
995
996         RTE_LOG(WARNING, EAL, "WARNING: Master core has no "
997                         "memory on local socket!\n");
998 }
999
1000 static int
1001 sync_func(__attribute__((unused)) void *arg)
1002 {
1003         return 0;
1004 }
1005
1006 inline static void
1007 rte_eal_mcfg_complete(void)
1008 {
1009         /* ALL shared mem_config related INIT DONE */
1010         if (rte_config.process_type == RTE_PROC_PRIMARY)
1011                 rte_config.mem_config->magic = RTE_MAGIC;
1012 }
1013
1014 /*
1015  * Request iopl privilege for all RPL, returns 0 on success
1016  */
1017 static int
1018 rte_eal_iopl_init(void)
1019 {
1020         return iopl(HIGHEST_RPL);
1021 }
1022
1023 /* Launch threads, called at application init(). */
1024 int
1025 rte_eal_init(int argc, char **argv)
1026 {
1027         int i, fctret, ret;
1028         pthread_t thread_id;
1029         static rte_atomic32_t run_once = RTE_ATOMIC32_INIT(0);
1030         struct shared_driver *solib = NULL;
1031         const char *logid;
1032
1033         if (!rte_atomic32_test_and_set(&run_once))
1034                 return -1;
1035
1036         logid = strrchr(argv[0], '/');
1037         logid = strdup(logid ? logid + 1: argv[0]);
1038
1039         thread_id = pthread_self();
1040
1041         if (rte_eal_log_early_init() < 0)
1042                 rte_panic("Cannot init early logs\n");
1043
1044         if (rte_eal_cpu_init() < 0)
1045                 rte_panic("Cannot detect lcores\n");
1046
1047         fctret = eal_parse_args(argc, argv);
1048         if (fctret < 0)
1049                 exit(1);
1050
1051         if (internal_config.no_hugetlbfs == 0 &&
1052                         internal_config.process_type != RTE_PROC_SECONDARY &&
1053                         internal_config.xen_dom0_support == 0 &&
1054                         eal_hugepage_info_init() < 0)
1055                 rte_panic("Cannot get hugepage information\n");
1056
1057         if (internal_config.memory == 0 && internal_config.force_sockets == 0) {
1058                 if (internal_config.no_hugetlbfs)
1059                         internal_config.memory = MEMSIZE_IF_NO_HUGE_PAGE;
1060                 else
1061                         internal_config.memory = eal_get_hugepage_mem_size();
1062         }
1063
1064         if (internal_config.vmware_tsc_map == 1) {
1065 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
1066                 rte_cycles_vmware_tsc_map = 1;
1067                 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
1068                                 "you must have monitor_control.pseudo_perfctr = TRUE\n");
1069 #else
1070                 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
1071                                 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
1072 #endif
1073         }
1074
1075         rte_srand(rte_rdtsc());
1076
1077         rte_config_init();
1078
1079         if (rte_eal_iopl_init() == 0)
1080                 rte_config.flags |= EAL_FLG_HIGH_IOPL;
1081
1082         if (rte_eal_pci_init() < 0)
1083                 rte_panic("Cannot init PCI\n");
1084
1085 #ifdef RTE_LIBRTE_IVSHMEM
1086         if (rte_eal_ivshmem_init() < 0)
1087                 rte_panic("Cannot init IVSHMEM\n");
1088 #endif
1089
1090         if (rte_eal_memory_init() < 0)
1091                 rte_panic("Cannot init memory\n");
1092
1093         /* the directories are locked during eal_hugepage_info_init */
1094         eal_hugedirs_unlock();
1095
1096         if (rte_eal_memzone_init() < 0)
1097                 rte_panic("Cannot init memzone\n");
1098
1099         if (rte_eal_tailqs_init() < 0)
1100                 rte_panic("Cannot init tail queues for objects\n");
1101
1102 #ifdef RTE_LIBRTE_IVSHMEM
1103         if (rte_eal_ivshmem_obj_init() < 0)
1104                 rte_panic("Cannot init IVSHMEM objects\n");
1105 #endif
1106
1107         if (rte_eal_log_init(logid, internal_config.syslog_facility) < 0)
1108                 rte_panic("Cannot init logs\n");
1109
1110         if (rte_eal_alarm_init() < 0)
1111                 rte_panic("Cannot init interrupt-handling thread\n");
1112
1113         if (rte_eal_intr_init() < 0)
1114                 rte_panic("Cannot init interrupt-handling thread\n");
1115
1116         if (rte_eal_timer_init() < 0)
1117                 rte_panic("Cannot init HPET or TSC timers\n");
1118
1119         eal_check_mem_on_local_socket();
1120
1121         rte_eal_mcfg_complete();
1122
1123         TAILQ_FOREACH(solib, &solib_list, next) {
1124                 RTE_LOG(INFO, EAL, "open shared lib %s\n", solib->name);
1125                 solib->lib_handle = dlopen(solib->name, RTLD_NOW);
1126                 if (solib->lib_handle == NULL)
1127                         RTE_LOG(WARNING, EAL, "%s\n", dlerror());
1128         }
1129
1130         eal_thread_init_master(rte_config.master_lcore);
1131
1132         RTE_LOG(DEBUG, EAL, "Master core %u is ready (tid=%x)\n",
1133                 rte_config.master_lcore, (int)thread_id);
1134
1135         if (rte_eal_dev_init(PMD_INIT_PRE_PCI_PROBE) < 0)
1136                 rte_panic("Cannot init pmd devices\n");
1137
1138         RTE_LCORE_FOREACH_SLAVE(i) {
1139
1140                 /*
1141                  * create communication pipes between master thread
1142                  * and children
1143                  */
1144                 if (pipe(lcore_config[i].pipe_master2slave) < 0)
1145                         rte_panic("Cannot create pipe\n");
1146                 if (pipe(lcore_config[i].pipe_slave2master) < 0)
1147                         rte_panic("Cannot create pipe\n");
1148
1149                 lcore_config[i].state = WAIT;
1150
1151                 /* create a thread for each lcore */
1152                 ret = pthread_create(&lcore_config[i].thread_id, NULL,
1153                                      eal_thread_loop, NULL);
1154                 if (ret != 0)
1155                         rte_panic("Cannot create thread\n");
1156         }
1157
1158         /*
1159          * Launch a dummy function on all slave lcores, so that master lcore
1160          * knows they are all ready when this function returns.
1161          */
1162         rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MASTER);
1163         rte_eal_mp_wait_lcore();
1164
1165         /* Probe & Initialize PCI devices */
1166         if (rte_eal_pci_probe())
1167                         rte_panic("Cannot probe PCI\n");
1168
1169         /* Initialize any outstanding devices */
1170         if (rte_eal_dev_init(PMD_INIT_POST_PCI_PROBE) < 0)
1171                 rte_panic("Cannot init pmd devices\n");
1172
1173         return fctret;
1174 }
1175
1176 /* get core role */
1177 enum rte_lcore_role_t
1178 rte_eal_lcore_role(unsigned lcore_id)
1179 {
1180         return (rte_config.lcore_role[lcore_id]);
1181 }
1182
1183 enum rte_proc_type_t
1184 rte_eal_process_type(void)
1185 {
1186         return (rte_config.process_type);
1187 }
1188
1189 int rte_eal_has_hugepages(void)
1190 {
1191         return ! internal_config.no_hugetlbfs;
1192 }