src/cpu.c

   1 /**
   2  * collectd - src/cpu.c
   3  * Copyright (C) 2005,2006  Florian octo Forster
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License as published by the
   7  * Free Software Foundation; either version 2 of the License, or (at your
   8  * option) any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along
  16  * with this program; if not, write to the Free Software Foundation, Inc.,
  17  * 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
  18  *
  19  * Authors:
  20  *   Florian octo Forster <octo at verplant.org>
  21  **/
  22
  23 #include "collectd.h"
  24 #include "common.h"
  25 #include "plugin.h"
  26 #include "utils_debug.h"
  27
  28 #define MODULE_NAME "cpu"
  29
  30 #ifdef HAVE_MACH_KERN_RETURN_H
  31 # include <mach/kern_return.h>
  32 #endif
  33 #ifdef HAVE_MACH_MACH_INIT_H
  34 # include <mach/mach_init.h>
  35 #endif
  36 #ifdef HAVE_MACH_HOST_PRIV_H
  37 # include <mach/host_priv.h>
  38 #endif
  39 #if HAVE_MACH_MACH_ERROR_H
  40 #  include <mach/mach_error.h>
  41 #endif
  42 #ifdef HAVE_MACH_PROCESSOR_INFO_H
  43 # include <mach/processor_info.h>
  44 #endif
  45 #ifdef HAVE_MACH_PROCESSOR_H
  46 # include <mach/processor.h>
  47 #endif
  48 #ifdef HAVE_MACH_VM_MAP_H
  49 # include <mach/vm_map.h>
  50 #endif
  51
  52 #ifdef HAVE_LIBKSTAT
  53 # include <sys/sysinfo.h>
  54 #endif /* HAVE_LIBKSTAT */
  55
  56 #ifdef HAVE_SYSCTLBYNAME
  57 # ifdef HAVE_SYS_SYSCTL_H
  58 #  include <sys/sysctl.h>
  59 # endif
  60
  61 # ifdef HAVE_SYS_DKSTAT_H
  62 #  include <sys/dkstat.h>
  63 # endif
  64
  65 # if !defined(CP_USER) || !defined(CP_NICE) || !defined(CP_SYS) || !defined(CP_INTR) || !defined(CP_IDLE) || !defined(CPUSTATES)
  66 #  define CP_USER   0
  67 #  define CP_NICE   1
  68 #  define CP_SYS    2
  69 #  define CP_INTR   3
  70 #  define CP_IDLE   4
  71 #  define CPUSTATES 5
  72 # endif
  73 #endif /* HAVE_SYSCTLBYNAME */
  74
  75 #if defined(PROCESSOR_CPU_LOAD_INFO) || defined(KERNEL_LINUX) || defined(HAVE_LIBKSTAT) || defined(HAVE_SYSCTLBYNAME)
  76 # define CPU_HAVE_READ 1
  77 #else
  78 # define CPU_HAVE_READ 0
  79 #endif
  80
  81 #ifdef PROCESSOR_CPU_LOAD_INFO
  82 static mach_port_t port_host;
  83 static processor_port_array_t cpu_list;
  84 static mach_msg_type_number_t cpu_list_len;
  85
  86 #if PROCESSOR_TEMPERATURE
  87 static int cpu_temp_retry_counter = 0;
  88 static int cpu_temp_retry_step    = 1;
  89 static int cpu_temp_retry_max     = 1;
  90 #endif /* PROCESSOR_TEMPERATURE */
  91 /* #endif PROCESSOR_CPU_LOAD_INFO */
  92
  93 #elif defined(KERNEL_LINUX)
  94 /* no variables needed */
  95 /* #endif KERNEL_LINUX */
  96
  97 #elif defined(HAVE_LIBKSTAT)
  98 /* colleague tells me that Sun doesn't sell systems with more than 100 or so CPUs.. */
  99 # define MAX_NUMCPU 256
 100 extern kstat_ctl_t *kc;
 101 static kstat_t *ksp[MAX_NUMCPU];
 102 static int numcpu;
 103 /* #endif HAVE_LIBKSTAT */
 104
 105 #elif defined(HAVE_SYSCTLBYNAME)
 106 static int numcpu;
 107 #endif /* HAVE_SYSCTLBYNAME */
 108
 109 static char *cpu_filename = "cpu-%s.rrd";
 110
 111 static char *ds_def[] =
 112 {
 113         "DS:user:COUNTER:"COLLECTD_HEARTBEAT":0:U",
 114         "DS:nice:COUNTER:"COLLECTD_HEARTBEAT":0:U",
 115         "DS:syst:COUNTER:"COLLECTD_HEARTBEAT":0:U",
 116         "DS:idle:COUNTER:"COLLECTD_HEARTBEAT":0:U",
 117         "DS:wait:COUNTER:"COLLECTD_HEARTBEAT":0:U",
 118         NULL
 119 };
 120 static int ds_num = 5;
 121
 122 static void cpu_init (void)
 123 {
 124 #if PROCESSOR_CPU_LOAD_INFO || PROCESSOR_TEMPERATURE
 125         kern_return_t status;
 126         int collectd_step;
 127
 128         port_host = mach_host_self ();
 129
 130         /* FIXME: Free `cpu_list' if it's not NULL */
 131         if ((status = host_processors (port_host, &cpu_list, &cpu_list_len)) != KERN_SUCCESS)
 132         {
 133                 syslog (LOG_ERR, "cpu-plugin: host_processors returned %i\n", (int) status);
 134                 cpu_list_len = 0;
 135                 return;
 136         }
 137
 138         DBG ("host_processors returned %i %s", (int) cpu_list_len, cpu_list_len == 1 ? "processor" : "processors");
 139         syslog (LOG_INFO, "cpu-plugin: Found %i processor%s.", (int) cpu_list_len, cpu_list_len == 1 ? "" : "s");
 140
 141         collectd_step = atoi (COLLECTD_STEP);
 142         if ((collectd_step > 0) && (collectd_step <= 86400))
 143                 cpu_temp_retry_max = 86400 / collectd_step;
 144
 145 /* #endif PROCESSOR_CPU_LOAD_INFO */
 146
 147 #elif defined(HAVE_LIBKSTAT)
 148         kstat_t *ksp_chain;
 149
 150         numcpu = 0;
 151
 152         if (kc == NULL)
 153                 return;
 154
 155         /* Solaris doesn't count linear.. *sigh* */
 156         for (numcpu = 0, ksp_chain = kc->kc_chain;
 157                         (numcpu < MAX_NUMCPU) && (ksp_chain != NULL);
 158                         ksp_chain = ksp_chain->ks_next)
 159                 if (strncmp (ksp_chain->ks_module, "cpu_stat", 8) == 0)
 160                         ksp[numcpu++] = ksp_chain;
 161 /* #endif HAVE_LIBKSTAT */
 162
 163 #elif defined (HAVE_SYSCTLBYNAME)
 164         size_t numcpu_size;
 165
 166         numcpu_size = sizeof (numcpu);
 167
 168         if (sysctlbyname ("hw.ncpu", &numcpu, &numcpu_size, NULL, 0) < 0)
 169         {
 170                 syslog (LOG_WARNING, "cpu: sysctlbyname: %s", strerror (errno));
 171                 return;
 172         }
 173
 174         if (numcpu != 1)
 175                 syslog (LOG_NOTICE, "cpu: Only one processor supported when using `sysctlbyname' (found %i)", numcpu);
 176 #endif
 177
 178         return;
 179 }
 180
 181 static void cpu_write (char *host, char *inst, char *val)
 182 {
 183         char file[512];
 184         int status;
 185
 186         status = snprintf (file, 512, cpu_filename, inst);
 187         if (status < 1)
 188                 return;
 189         else if (status >= 512)
 190                 return;
 191
 192         rrd_update_file (host, file, val, ds_def, ds_num);
 193 }
 194
 195 #if CPU_HAVE_READ
 196 #define BUFSIZE 512
 197 static void cpu_submit (int cpu_num, unsigned long long user,
 198                 unsigned long long nice, unsigned long long syst,
 199                 unsigned long long idle, unsigned long long wait)
 200 {
 201         char buf[BUFSIZE];
 202         char cpu[16];
 203
 204         if (snprintf (buf, BUFSIZE, "%u:%llu:%llu:%llu:%llu:%llu", (unsigned int) curtime,
 205                                 user, nice, syst, idle, wait) >= BUFSIZE)
 206                 return;
 207         snprintf (cpu, 16, "%i", cpu_num);
 208
 209         plugin_submit (MODULE_NAME, cpu, buf);
 210 }
 211 #undef BUFSIZE
 212
 213 static void cpu_read (void)
 214 {
 215 #if PROCESSOR_CPU_LOAD_INFO || PROCESSOR_TEMPERATURE
 216         int cpu;
 217
 218         kern_return_t status;
 219
 220 #if PROCESSOR_CPU_LOAD_INFO
 221         processor_cpu_load_info_data_t cpu_info;
 222         mach_msg_type_number_t         cpu_info_len;
 223 #endif
 224 #if PROCESSOR_TEMPERATURE
 225         processor_info_data_t          cpu_temp;
 226         mach_msg_type_number_t         cpu_temp_len;
 227 #endif
 228
 229         host_t cpu_host;
 230
 231         for (cpu = 0; cpu < cpu_list_len; cpu++)
 232         {
 233 #if PROCESSOR_CPU_LOAD_INFO
 234                 cpu_host = 0;
 235                 cpu_info_len = PROCESSOR_BASIC_INFO_COUNT;
 236
 237                 if ((status = processor_info (cpu_list[cpu],
 238                                                 PROCESSOR_CPU_LOAD_INFO, &cpu_host,
 239                                                 (processor_info_t) &cpu_info, &cpu_info_len)) != KERN_SUCCESS)
 240                 {
 241                         syslog (LOG_ERR, "cpu-plugin: processor_info failed with status %i\n", (int) status);
 242                         continue;
 243                 }
 244
 245                 if (cpu_info_len < CPU_STATE_MAX)
 246                 {
 247                         syslog (LOG_ERR, "cpu-plugin: processor_info returned only %i elements..\n", cpu_info_len);
 248                         continue;
 249                 }
 250
 251                 cpu_submit (cpu, cpu_info.cpu_ticks[CPU_STATE_USER],
 252                                 cpu_info.cpu_ticks[CPU_STATE_NICE],
 253                                 cpu_info.cpu_ticks[CPU_STATE_SYSTEM],
 254                                 cpu_info.cpu_ticks[CPU_STATE_IDLE],
 255                                 0ULL);
 256 #endif /* PROCESSOR_CPU_LOAD_INFO */
 257 #if PROCESSOR_TEMPERATURE
 258                 /*
 259                  * Not all Apple computers do have this ability. To minimize
 260                  * the messages sent to the syslog we do an exponential
 261                  * stepback if `processor_info' fails. We still try ~once a day
 262                  * though..
 263                  */
 264                 if (cpu_temp_retry_counter > 0)
 265                 {
 266                         cpu_temp_retry_counter--;
 267                         continue;
 268                 }
 269
 270                 cpu_temp_len = PROCESSOR_INFO_MAX;
 271
 272                 status = processor_info (cpu_list[cpu],
 273                                 PROCESSOR_TEMPERATURE,
 274                                 &cpu_host,
 275                                 cpu_temp, &cpu_temp_len);
 276                 if (status != KERN_SUCCESS)
 277                 {
 278                         syslog (LOG_ERR, "cpu-plugin: processor_info failed: %s",
 279                                         mach_error_string (status));
 280
 281                         cpu_temp_retry_counter = cpu_temp_retry_step;
 282                         cpu_temp_retry_step *= 2;
 283                         if (cpu_temp_retry_step > cpu_temp_retry_max)
 284                                 cpu_temp_retry_step = cpu_temp_retry_max;
 285
 286                         continue;
 287                 }
 288
 289                 if (cpu_temp_len != 1)
 290                 {
 291                         DBG ("processor_info (PROCESSOR_TEMPERATURE) returned %i elements..?",
 292                                         (int) cpu_temp_len);
 293                         continue;
 294                 }
 295
 296                 cpu_temp_retry_counter = 0;
 297                 cpu_temp_retry_step    = 1;
 298
 299                 DBG ("cpu_temp = %i", (int) cpu_temp);
 300 #endif /* PROCESSOR_TEMPERATURE */
 301         }
 302 /* #endif PROCESSOR_CPU_LOAD_INFO */
 303
 304 #elif defined(KERNEL_LINUX)
 305 # define BUFSIZE 1024
 306         int cpu;
 307         unsigned long long user, nice, syst, idle;
 308         unsigned long long wait, intr, sitr; /* sitr == soft interrupt */
 309         FILE *fh;
 310         char buf[BUFSIZE];
 311
 312         char *fields[9];
 313         int numfields;
 314
 315         static complain_t complain_obj;
 316
 317         if ((fh = fopen ("/proc/stat", "r")) == NULL)
 318         {
 319                 plugin_complain (LOG_ERR, &complain_obj, "cpu plugin: "
 320                                 "fopen (/proc/stat) failed: %s",
 321                                 strerror (errno));
 322                 return;
 323         }
 324
 325         plugin_relief (LOG_NOTICE, &complain_obj, "cpu plugin: "
 326                         "fopen (/proc/stat) succeeded.");
 327
 328         while (fgets (buf, BUFSIZE, fh) != NULL)
 329         {
 330                 if (strncmp (buf, "cpu", 3))
 331                         continue;
 332                 if ((buf[3] < '0') || (buf[3] > '9'))
 333                         continue;
 334
 335                 numfields = strsplit (buf, fields, 9);
 336                 if (numfields < 5)
 337                         continue;
 338
 339                 cpu = atoi (fields[0] + 3);
 340                 user = atoll (fields[1]);
 341                 nice = atoll (fields[2]);
 342                 syst = atoll (fields[3]);
 343                 idle = atoll (fields[4]);
 344
 345                 if (numfields >= 8)
 346                 {
 347                         wait = atoll (fields[5]);
 348                         intr = atoll (fields[6]);
 349                         sitr = atoll (fields[7]);
 350
 351                         /* I doubt anyone cares about the time spent in
 352                          * interrupt handlers.. */
 353                         syst += intr + sitr;
 354                 }
 355                 else
 356                 {
 357                         wait = 0LL;
 358                 }
 359
 360                 cpu_submit (cpu, user, nice, syst, idle, wait);
 361         }
 362
 363         fclose (fh);
 364 #undef BUFSIZE
 365 /* #endif defined(KERNEL_LINUX) */
 366
 367 #elif defined(HAVE_LIBKSTAT)
 368         int cpu;
 369         unsigned long long user, syst, idle, wait;
 370         static cpu_stat_t cs;
 371
 372         if (kc == NULL)
 373                 return;
 374
 375         for (cpu = 0; cpu < numcpu; cpu++)
 376         {
 377                 if (kstat_read (kc, ksp[cpu], &cs) == -1)
 378                         continue; /* error message? */
 379
 380                 idle = (unsigned long long) cs.cpu_sysinfo.cpu[CPU_IDLE];
 381                 user = (unsigned long long) cs.cpu_sysinfo.cpu[CPU_USER];
 382                 syst = (unsigned long long) cs.cpu_sysinfo.cpu[CPU_KERNEL];
 383                 wait = (unsigned long long) cs.cpu_sysinfo.cpu[CPU_WAIT];
 384
 385                 cpu_submit (ksp[cpu]->ks_instance,
 386                                 user, 0LL, syst, idle, wait);
 387         }
 388 /* #endif defined(HAVE_LIBKSTAT) */
 389
 390 #elif defined(HAVE_SYSCTLBYNAME)
 391         long cpuinfo[CPUSTATES];
 392         size_t cpuinfo_size;
 393
 394         static complain_t complain_obj;
 395
 396         cpuinfo_size = sizeof (cpuinfo);
 397
 398         if (sysctlbyname("kern.cp_time", &cpuinfo, &cpuinfo_size, NULL, 0) < 0)
 399         {
 400                 plugin_complain (LOG_ERR, &complain_obj, "cpu plugin: "
 401                                 "sysctlbyname failed: %s.",
 402                                 strerror (errno));
 403                 return;
 404         }
 405
 406         plugin_relief (LOG_NOTICE, &complain_obj, "cpu plugin: "
 407                         "sysctlbyname succeeded.");
 408
 409         cpuinfo[CP_SYS] += cpuinfo[CP_INTR];
 410
 411         /* FIXME: Instance is always `0' */
 412         cpu_submit (0, cpuinfo[CP_USER], cpuinfo[CP_NICE], cpuinfo[CP_SYS], cpuinfo[CP_IDLE], 0LL);
 413 #endif
 414
 415         return;
 416 }
 417 #else
 418 # define cpu_read NULL
 419 #endif /* CPU_HAVE_READ */
 420
 421 void module_register (void)
 422 {
 423         plugin_register (MODULE_NAME, cpu_init, cpu_read, cpu_write);
 424 }
 425
 426 #undef MODULE_NAME