Turbostat: specify parse_int_file format
[collectd.git] / src / turbostat.c
1 /*
2  * turbostat -- Log CPU frequency and C-state residency
3  * on modern Intel turbo-capable processors for collectd.
4  *
5  * Based on the 'turbostat' tool of the Linux kernel, found at
6  * linux/tools/power/x86/turbostat/turbostat.c:
7  * ----
8  * Copyright (c) 2013 Intel Corporation.
9  * Len Brown <len.brown@intel.com>
10  *
11  * This program is free software; you can redistribute it and/or modify it
12  * under the terms and conditions of the GNU General Public License,
13  * version 2, as published by the Free Software Foundation.
14  *
15  * This program is distributed in the hope it will be useful, but WITHOUT
16  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
17  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
18  * more details.
19  *
20  * You should have received a copy of the GNU General Public License along with
21  * this program; if not, write to the Free Software Foundation, Inc.,
22  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
23  * ----
24  * Ported to collectd by Vincent Brillault <git@lerya.net>
25  */
26
27 #define _GNU_SOURCE
28 #include <asm/msr-index.h>
29 #include <stdarg.h>
30 #include <stdio.h>
31 #include <err.h>
32 #include <unistd.h>
33 #include <sys/types.h>
34 #include <sys/wait.h>
35 #include <sys/stat.h>
36 #include <sys/resource.h>
37 #include <fcntl.h>
38 #include <signal.h>
39 #include <sys/time.h>
40 #include <stdlib.h>
41 #include <dirent.h>
42 #include <string.h>
43 #include <ctype.h>
44 #include <sched.h>
45 #include <cpuid.h>
46
47 #include "collectd.h"
48 #include "common.h"
49 #include "plugin.h"
50
51 #define PLUGIN_NAME "turbostat"
52
53 static const char *proc_stat = "/proc/stat";
54 static unsigned int interval_sec = 5;   /* set with -i interval_sec */
55 static unsigned int skip_c0;
56 static unsigned int skip_c1;
57 static unsigned int do_nhm_cstates;
58 static unsigned int do_snb_cstates;
59 static unsigned int do_c8_c9_c10;
60 static unsigned int do_slm_cstates;
61 static unsigned int has_aperf;
62 static unsigned int has_epb;
63 static unsigned int units = 1000000000; /* Ghz etc */
64 static unsigned int genuine_intel;
65 static unsigned int has_invariant_tsc;
66 static unsigned int do_nehalem_platform_info;
67 static int do_smi;
68 static unsigned int show_pkg;
69 static unsigned int show_core;
70 static unsigned int show_cpu;
71 static unsigned int do_rapl;
72 static unsigned int do_dts;
73 static unsigned int do_ptm;
74 static unsigned int tcc_activation_temp;
75 static unsigned int tcc_activation_temp_override;
76 static double rapl_power_units, rapl_energy_units, rapl_time_units;
77 static double rapl_joule_counter_range;
78
79 #define RAPL_PKG                (1 << 0)
80                                         /* 0x610 MSR_PKG_POWER_LIMIT */
81                                         /* 0x611 MSR_PKG_ENERGY_STATUS */
82 #define RAPL_PKG_PERF_STATUS    (1 << 1)
83                                         /* 0x613 MSR_PKG_PERF_STATUS */
84 #define RAPL_PKG_POWER_INFO     (1 << 2)
85                                         /* 0x614 MSR_PKG_POWER_INFO */
86
87 #define RAPL_DRAM               (1 << 3)
88                                         /* 0x618 MSR_DRAM_POWER_LIMIT */
89                                         /* 0x619 MSR_DRAM_ENERGY_STATUS */
90                                         /* 0x61c MSR_DRAM_POWER_INFO */
91 #define RAPL_DRAM_PERF_STATUS   (1 << 4)
92                                         /* 0x61b MSR_DRAM_PERF_STATUS */
93
94 #define RAPL_CORES              (1 << 5)
95                                         /* 0x638 MSR_PP0_POWER_LIMIT */
96                                         /* 0x639 MSR_PP0_ENERGY_STATUS */
97 #define RAPL_CORE_POLICY        (1 << 6)
98                                         /* 0x63a MSR_PP0_POLICY */
99
100
101 #define RAPL_GFX                (1 << 7)
102                                         /* 0x640 MSR_PP1_POWER_LIMIT */
103                                         /* 0x641 MSR_PP1_ENERGY_STATUS */
104                                         /* 0x642 MSR_PP1_POLICY */
105 #define TJMAX_DEFAULT   100
106
107 int aperf_mperf_unstable;
108 int backwards_count;
109 char *progname;
110
111 cpu_set_t *cpu_present_set, *cpu_affinity_set;
112 size_t cpu_present_setsize, cpu_affinity_setsize;
113
114 struct thread_data {
115         unsigned long long tsc;
116         unsigned long long aperf;
117         unsigned long long mperf;
118         unsigned long long c1;
119         unsigned int smi_count;
120         unsigned int cpu_id;
121         unsigned int flags;
122 #define CPU_IS_FIRST_THREAD_IN_CORE     0x2
123 #define CPU_IS_FIRST_CORE_IN_PACKAGE    0x4
124 } *thread_even, *thread_odd;
125
126 struct core_data {
127         unsigned long long c3;
128         unsigned long long c6;
129         unsigned long long c7;
130         unsigned int core_temp_c;
131         unsigned int core_id;
132 } *core_even, *core_odd;
133
134 struct pkg_data {
135         unsigned long long pc2;
136         unsigned long long pc3;
137         unsigned long long pc6;
138         unsigned long long pc7;
139         unsigned long long pc8;
140         unsigned long long pc9;
141         unsigned long long pc10;
142         unsigned int package_id;
143         unsigned int energy_pkg;        /* MSR_PKG_ENERGY_STATUS */
144         unsigned int energy_dram;       /* MSR_DRAM_ENERGY_STATUS */
145         unsigned int energy_cores;      /* MSR_PP0_ENERGY_STATUS */
146         unsigned int energy_gfx;        /* MSR_PP1_ENERGY_STATUS */
147         unsigned int rapl_pkg_perf_status;      /* MSR_PKG_PERF_STATUS */
148         unsigned int rapl_dram_perf_status;     /* MSR_DRAM_PERF_STATUS */
149         unsigned int pkg_temp_c;
150
151 } *package_even, *package_odd;
152
153 #define ODD_COUNTERS thread_odd, core_odd, package_odd
154 #define EVEN_COUNTERS thread_even, core_even, package_even
155 static _Bool is_even = 1;
156
157 static _Bool allocated = 0;
158 static _Bool initialized = 0;
159
160 #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \
161         (thread_base + (pkg_no) * topo.num_cores_per_pkg * \
162                 topo.num_threads_per_core + \
163                 (core_no) * topo.num_threads_per_core + (thread_no))
164 #define GET_CORE(core_base, core_no, pkg_no) \
165         (core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no))
166 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
167
168 struct topo_params {
169         int num_packages;
170         int num_cpus;
171         int num_cores;
172         int max_cpu_num;
173         int num_cores_per_pkg;
174         int num_threads_per_core;
175 } topo;
176
177 struct timeval tv_even, tv_odd, tv_delta;
178
179 enum return_values {
180         OK = 0,
181         ERR_CPU_MIGRATE,
182         ERR_MSR_IA32_APERF,
183         ERR_MSR_IA32_MPERF,
184         ERR_MSR_SMI_COUNT,
185         ERR_MSR_CORE_C3_RESIDENCY,
186         ERR_MSR_CORE_C6_RESIDENCY,
187         ERR_MSR_CORE_C7_RESIDENCY,
188         ERR_MSR_IA32_THERM_STATUS,
189         ERR_MSR_PKG_C3_RESIDENCY,
190         ERR_MSR_PKG_C6_RESIDENCY,
191         ERR_MSR_PKG_C2_RESIDENCY,
192         ERR_MSR_PKG_C7_RESIDENCY,
193         ERR_MSR_PKG_C8_RESIDENCY,
194         ERR_MSR_PKG_C9_RESIDENCY,
195         ERR_MSR_PKG_C10_RESIDENCY,
196         ERR_MSR_PKG_ENERGY_STATUS,
197         ERR_MSR_PP0_ENERGY_STATUS,
198         ERR_MSR_DRAM_ENERGY_STATUS,
199         ERR_MSR_PP1_ENERGY_STATUS,
200         ERR_MSR_PKG_PERF_STATUS,
201         ERR_MSR_DRAM_PERF_STATUS,
202         ERR_MSR_IA32_PACKAGE_THERM_STATUS,
203         ERR_CPU_NOT_PRESENT,
204         ERR_NO_MSR,
205         ERR_CANT_OPEN_FILE,
206         ERR_CANT_READ_NUMBER,
207         ERR_CANT_READ_PROC_STAT,
208         ERR_NO_INVARIANT_TSC,
209         ERR_NO_APERF,
210         ERR_CALLOC,
211         ERR_CPU_ALLOC,
212         ERR_NOT_ROOT,
213 };
214
215 static int setup_all_buffers(void);
216
217 static int
218 cpu_is_not_present(int cpu)
219 {
220         return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
221 }
222 /*
223  * run func(thread, core, package) in topology order
224  * skip non-present cpus
225  */
226
227 static int __attribute__((warn_unused_result))
228 for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
229         struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
230 {
231         int retval, pkg_no, core_no, thread_no;
232
233         for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
234                 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
235                         for (thread_no = 0; thread_no <
236                                 topo.num_threads_per_core; ++thread_no) {
237                                 struct thread_data *t;
238                                 struct core_data *c;
239                                 struct pkg_data *p;
240
241                                 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
242
243                                 if (cpu_is_not_present(t->cpu_id))
244                                         continue;
245
246                                 c = GET_CORE(core_base, core_no, pkg_no);
247                                 p = GET_PKG(pkg_base, pkg_no);
248
249                                 retval = func(t, c, p);
250                                 if (retval)
251                                         return retval;
252                         }
253                 }
254         }
255         return 0;
256 }
257
258 static int __attribute__((warn_unused_result))
259 cpu_migrate(int cpu)
260 {
261         CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
262         CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
263         if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
264                 return -ERR_CPU_MIGRATE;
265         else
266                 return 0;
267 }
268
269 static int __attribute__((warn_unused_result))
270 get_msr(int cpu, off_t offset, unsigned long long *msr)
271 {
272         ssize_t retval;
273         char pathname[32];
274         int fd;
275
276         ssnprintf(pathname, 32, "/dev/cpu/%d/msr", cpu);
277         fd = open(pathname, O_RDONLY);
278         if (fd < 0)
279                 return -1;
280
281         retval = pread(fd, msr, sizeof *msr, offset);
282         close(fd);
283
284         if (retval != sizeof *msr) {
285                 ERROR ("%s offset 0x%llx read failed\n", pathname, (unsigned long long)offset);
286                 return -1;
287         }
288
289         return 0;
290 }
291
292 #define DELTA_WRAP32(new, old)                  \
293         if (new > old) {                        \
294                 old = new - old;                \
295         } else {                                \
296                 old = 0x100000000 + new - old;  \
297         }
298
299 static void
300 delta_package(struct pkg_data *new, struct pkg_data *old)
301 {
302         old->pc2 = new->pc2 - old->pc2;
303         old->pc3 = new->pc3 - old->pc3;
304         old->pc6 = new->pc6 - old->pc6;
305         old->pc7 = new->pc7 - old->pc7;
306         old->pc8 = new->pc8 - old->pc8;
307         old->pc9 = new->pc9 - old->pc9;
308         old->pc10 = new->pc10 - old->pc10;
309         old->pkg_temp_c = new->pkg_temp_c;
310
311         DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
312         DELTA_WRAP32(new->energy_cores, old->energy_cores);
313         DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
314         DELTA_WRAP32(new->energy_dram, old->energy_dram);
315         DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
316         DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
317 }
318
319 static void
320 delta_core(struct core_data *new, struct core_data *old)
321 {
322         old->c3 = new->c3 - old->c3;
323         old->c6 = new->c6 - old->c6;
324         old->c7 = new->c7 - old->c7;
325         old->core_temp_c = new->core_temp_c;
326 }
327
328 /*
329  * old = new - old
330  */
331 static int __attribute__((warn_unused_result))
332 delta_thread(struct thread_data *new, struct thread_data *old,
333         struct core_data *core_delta)
334 {
335         old->tsc = new->tsc - old->tsc;
336
337         /* check for TSC < 1 Mcycles over interval */
338         if (old->tsc < (1000 * 1000)) {
339                 WARNING("Insanely slow TSC rate, TSC stops in idle?\n"
340                         "You can disable all c-states by booting with \"idle=poll\"\n"
341                         "or just the deep ones with \"processor.max_cstate=1\"");
342                 return -1;
343         }
344
345         old->c1 = new->c1 - old->c1;
346
347         if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
348                 old->aperf = new->aperf - old->aperf;
349                 old->mperf = new->mperf - old->mperf;
350         } else {
351
352                 if (!aperf_mperf_unstable) {
353                         WARNING("%s: APERF or MPERF went backwards *\n", progname);
354                         WARNING("* Frequency results do not cover entire interval *\n");
355                         WARNING("* fix this by running Linux-2.6.30 or later *\n");
356
357                         aperf_mperf_unstable = 1;
358                 }
359                 /*
360                  * mperf delta is likely a huge "positive" number
361                  * can not use it for calculating c0 time
362                  */
363                 skip_c0 = 1;
364                 skip_c1 = 1;
365         }
366
367
368         /*
369          * As counter collection is not atomic,
370          * it is possible for mperf's non-halted cycles + idle states
371          * to exceed TSC's all cycles: show c1 = 0% in that case.
372          */
373         if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > old->tsc)
374                 old->c1 = 0;
375         else {
376                 /* normal case, derive c1 */
377                 old->c1 = old->tsc - old->mperf - core_delta->c3
378                         - core_delta->c6 - core_delta->c7;
379         }
380
381         if (old->mperf == 0) {
382                 WARNING("cpu%d MPERF 0!\n", old->cpu_id);
383                 old->mperf = 1; /* divide by 0 protection */
384         }
385
386         if (do_smi)
387                 old->smi_count = new->smi_count - old->smi_count;
388
389         return 0;
390 }
391
392 static int __attribute__((warn_unused_result))
393 delta_cpu(struct thread_data *t, struct core_data *c,
394         struct pkg_data *p, struct thread_data *t2,
395         struct core_data *c2, struct pkg_data *p2)
396 {
397         int ret;
398
399         /* calculate core delta only for 1st thread in core */
400         if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
401                 delta_core(c, c2);
402
403         /* always calculate thread delta */
404         ret = delta_thread(t, t2, c2);  /* c2 is core delta */
405         if (ret != 0)
406                 return ret;
407
408         /* calculate package delta only for 1st core in package */
409         if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
410                 delta_package(p, p2);
411
412         return 0;
413 }
414
415 static unsigned long long
416 rdtsc(void)
417 {
418         unsigned int low, high;
419
420         asm volatile("rdtsc" : "=a" (low), "=d" (high));
421
422         return low | ((unsigned long long)high) << 32;
423 }
424
425
426 /*
427  * get_counters(...)
428  * migrate to cpu
429  * acquire and record local counters for that cpu
430  */
431 static int __attribute__((warn_unused_result))
432 get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
433 {
434         int cpu = t->cpu_id;
435         unsigned long long msr;
436
437         if (cpu_migrate(cpu)) {
438                 WARNING("Could not migrate to CPU %d\n", cpu);
439                 return -ERR_CPU_MIGRATE;
440         }
441
442         t->tsc = rdtsc();       /* we are running on local CPU of interest */
443
444         if (has_aperf) {
445                 if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
446                         return -ERR_MSR_IA32_APERF;
447                 if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
448                         return -ERR_MSR_IA32_MPERF;
449         }
450
451         if (do_smi) {
452                 if (get_msr(cpu, MSR_SMI_COUNT, &msr))
453                         return -ERR_MSR_SMI_COUNT;
454                 t->smi_count = msr & 0xFFFFFFFF;
455         }
456
457         /* collect core counters only for 1st thread in core */
458         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
459                 return 0;
460
461         if (do_nhm_cstates && !do_slm_cstates) {
462                 if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
463                         return -ERR_MSR_CORE_C3_RESIDENCY;
464         }
465
466         if (do_nhm_cstates) {
467                 if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
468                         return -ERR_MSR_CORE_C6_RESIDENCY;
469         }
470
471         if (do_snb_cstates)
472                 if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
473                         return -ERR_MSR_CORE_C7_RESIDENCY;
474
475         if (do_dts) {
476                 if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
477                         return -ERR_MSR_IA32_THERM_STATUS;
478                 c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
479         }
480
481
482         /* collect package counters only for 1st core in package */
483         if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
484                 return 0;
485
486         if (do_nhm_cstates && !do_slm_cstates) {
487                 if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
488                         return -ERR_MSR_PKG_C3_RESIDENCY;
489                 if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
490                         return -ERR_MSR_PKG_C6_RESIDENCY;
491         }
492         if (do_snb_cstates) {
493                 if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
494                         return -ERR_MSR_PKG_C2_RESIDENCY;
495                 if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
496                         return -ERR_MSR_PKG_C7_RESIDENCY;
497         }
498         if (do_c8_c9_c10) {
499                 if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
500                         return -ERR_MSR_PKG_C8_RESIDENCY;
501                 if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
502                         return -ERR_MSR_PKG_C9_RESIDENCY;
503                 if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
504                         return -ERR_MSR_PKG_C10_RESIDENCY;
505         }
506         if (do_rapl & RAPL_PKG) {
507                 if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
508                         return -ERR_MSR_PKG_ENERGY_STATUS;
509                 p->energy_pkg = msr & 0xFFFFFFFF;
510         }
511         if (do_rapl & RAPL_CORES) {
512                 if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
513                         return MSR_PP0_ENERGY_STATUS;
514                 p->energy_cores = msr & 0xFFFFFFFF;
515         }
516         if (do_rapl & RAPL_DRAM) {
517                 if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
518                         return -ERR_MSR_DRAM_ENERGY_STATUS;
519                 p->energy_dram = msr & 0xFFFFFFFF;
520         }
521         if (do_rapl & RAPL_GFX) {
522                 if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
523                         return -ERR_MSR_PP1_ENERGY_STATUS;
524                 p->energy_gfx = msr & 0xFFFFFFFF;
525         }
526         if (do_rapl & RAPL_PKG_PERF_STATUS) {
527                 if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
528                         return -ERR_MSR_PKG_PERF_STATUS;
529                 p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
530         }
531         if (do_rapl & RAPL_DRAM_PERF_STATUS) {
532                 if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
533                         return -ERR_MSR_DRAM_PERF_STATUS;
534                 p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
535         }
536         if (do_ptm) {
537                 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
538                         return -ERR_MSR_IA32_PACKAGE_THERM_STATUS;
539                 p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
540         }
541         return 0;
542 }
543
544 static void
545 free_all_buffers(void)
546 {
547         allocated = 0;
548         initialized = 0;
549
550         CPU_FREE(cpu_present_set);
551         cpu_present_set = NULL;
552         cpu_present_set = 0;
553
554         CPU_FREE(cpu_affinity_set);
555         cpu_affinity_set = NULL;
556         cpu_affinity_setsize = 0;
557
558         free(thread_even);
559         free(core_even);
560         free(package_even);
561
562         thread_even = NULL;
563         core_even = NULL;
564         package_even = NULL;
565
566         free(thread_odd);
567         free(core_odd);
568         free(package_odd);
569
570         thread_odd = NULL;
571         core_odd = NULL;
572         package_odd = NULL;
573 }
574
575 /*
576  * Parse a file containing a single int.
577  */
578 static int __attribute__ ((format(printf,1,2)))
579 parse_int_file(const char *fmt, ...)
580 {
581         va_list args;
582         char path[PATH_MAX];
583         FILE *filep;
584         int value;
585
586         va_start(args, fmt);
587         vsnprintf(path, sizeof(path), fmt, args);
588         va_end(args);
589         filep = fopen(path, "r");
590         if (!filep) {
591                 ERROR("%s: open failed", path);
592                 return -ERR_CANT_OPEN_FILE;
593         }
594         if (fscanf(filep, "%d", &value) != 1) {
595                 ERROR("%s: failed to parse number from file", path);
596                 return -ERR_CANT_READ_NUMBER;
597         }
598         fclose(filep);
599         return value;
600 }
601
602 /*
603  * cpu_is_first_sibling_in_core(cpu)
604  * return 1 if given CPU is 1st HT sibling in the core
605  */
606 static int
607 cpu_is_first_sibling_in_core(int cpu)
608 {
609         return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
610 }
611
612 /*
613  * cpu_is_first_core_in_package(cpu)
614  * return 1 if given CPU is 1st core in package
615  */
616 static int
617 cpu_is_first_core_in_package(int cpu)
618 {
619         return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
620 }
621
622 static int
623 get_physical_package_id(int cpu)
624 {
625         return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
626 }
627
628 static int
629 get_core_id(int cpu)
630 {
631         return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
632 }
633
634 static int
635 get_num_ht_siblings(int cpu)
636 {
637         char path[80];
638         FILE *filep;
639         int sib1, sib2;
640         int matches;
641         char character;
642
643         ssnprintf(path, 80, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
644         filep = fopen(path, "r");
645         if (!filep) {
646                 ERROR("%s: open failed", path);
647                 return -ERR_CANT_OPEN_FILE;
648         }
649         /*
650          * file format:
651          * if a pair of number with a character between: 2 siblings (eg. 1-2, or 1,4)
652          * otherwinse 1 sibling (self).
653          */
654         matches = fscanf(filep, "%d%c%d\n", &sib1, &character, &sib2);
655
656         fclose(filep);
657
658         if (matches == 3)
659                 return 2;
660         else
661                 return 1;
662 }
663
664 /*
665  * run func(thread, core, package) in topology order
666  * skip non-present cpus
667  */
668
669
670 static int __attribute__((warn_unused_result))
671 for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
672         struct pkg_data *, struct thread_data *, struct core_data *,
673         struct pkg_data *), struct thread_data *thread_base,
674         struct core_data *core_base, struct pkg_data *pkg_base,
675         struct thread_data *thread_base2, struct core_data *core_base2,
676         struct pkg_data *pkg_base2)
677 {
678         int retval, pkg_no, core_no, thread_no;
679
680         for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
681                 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
682                         for (thread_no = 0; thread_no <
683                                 topo.num_threads_per_core; ++thread_no) {
684                                 struct thread_data *t, *t2;
685                                 struct core_data *c, *c2;
686                                 struct pkg_data *p, *p2;
687
688                                 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
689
690                                 if (cpu_is_not_present(t->cpu_id))
691                                         continue;
692
693                                 t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no);
694
695                                 c = GET_CORE(core_base, core_no, pkg_no);
696                                 c2 = GET_CORE(core_base2, core_no, pkg_no);
697
698                                 p = GET_PKG(pkg_base, pkg_no);
699                                 p2 = GET_PKG(pkg_base2, pkg_no);
700
701                                 retval = func(t, c, p, t2, c2, p2);
702                                 if (retval)
703                                         return retval;
704                         }
705                 }
706         }
707         return 0;
708 }
709
710 /*
711  * run func(cpu) on every cpu in /proc/stat
712  * return max_cpu number
713  */
714 static int __attribute__((warn_unused_result))
715 for_all_proc_cpus(int (func)(int))
716 {
717         FILE *fp;
718         int cpu_num;
719         int retval;
720
721         fp = fopen(proc_stat, "r");
722         if (!fp) {
723                 ERROR("%s: open failed", proc_stat);
724                 return -ERR_CANT_OPEN_FILE;
725         }
726
727         retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
728         if (retval != 0) {
729                 ERROR("%s: failed to parse format", proc_stat);
730                 return -ERR_CANT_READ_PROC_STAT;
731         }
732
733         while (1) {
734                 retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
735                 if (retval != 1)
736                         break;
737
738                 retval = func(cpu_num);
739                 if (retval) {
740                         fclose(fp);
741                         return(retval);
742                 }
743         }
744         fclose(fp);
745         return 0;
746 }
747
748 /*
749  * count_cpus()
750  * remember the last one seen, it will be the max
751  */
752 static int
753 count_cpus(int cpu)
754 {
755         if (topo.max_cpu_num < cpu)
756                 topo.max_cpu_num = cpu;
757
758         topo.num_cpus += 1;
759         return 0;
760 }
761 static int
762 mark_cpu_present(int cpu)
763 {
764         CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
765         return 0;
766 }
767
768
769 static void
770 turbostat_submit (const char *plugin_instance,
771         const char *type, const char *type_instance,
772         gauge_t value)
773 {
774         value_list_t vl = VALUE_LIST_INIT;
775         value_t v;
776
777         v.gauge = value;
778         vl.values = &v;
779         vl.values_len = 1;
780         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
781         sstrncpy (vl.plugin, PLUGIN_NAME, sizeof (vl.plugin));
782         if (plugin_instance != NULL)
783                 sstrncpy (vl.plugin_instance, plugin_instance, sizeof (vl.plugin_instance));
784         sstrncpy (vl.type, type, sizeof (vl.type));
785         if (type_instance != NULL)
786                 sstrncpy (vl.type_instance, type_instance, sizeof (vl.type_instance));
787
788         plugin_dispatch_values (&vl);
789 }
790
791 /*
792  * column formatting convention & formats
793  * package: "pk" 2 columns %2d
794  * core: "cor" 3 columns %3d
795  * CPU: "CPU" 3 columns %3d
796  * Pkg_W: %6.2
797  * Cor_W: %6.2
798  * GFX_W: %5.2
799  * RAM_W: %5.2
800  * GHz: "GHz" 3 columns %3.2
801  * TSC: "TSC" 3 columns %3.2
802  * SMI: "SMI" 4 columns %4d
803  * percentage " %pc3" %6.2
804  * Perf Status percentage: %5.2
805  * "CTMP" 4 columns %4d
806  */
807 #define NAME_LEN 12
808 static int
809 submit_counters(struct thread_data *t, struct core_data *c,
810         struct pkg_data *p)
811 {
812         char name[NAME_LEN];
813         double interval_float;
814
815         interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
816
817         snprintf(name, NAME_LEN, "cpu%02d", t->cpu_id);
818
819         if (do_nhm_cstates) {
820                 if (!skip_c0)
821                         turbostat_submit(name, "percent", "c0", 100.0 * t->mperf/t->tsc);
822                 if (!skip_c1)
823                         turbostat_submit(name, "percent", "c1", 100.0 * t->c1/t->tsc);
824         }
825
826         /* GHz */
827         if (has_aperf && ((!aperf_mperf_unstable) || (!(t->aperf > t->tsc || t->mperf > t->tsc))))
828                 turbostat_submit(NULL, "frequency", name, 1.0 * t->tsc / units * t->aperf / t->mperf / interval_float);
829
830         /* SMI */
831         if (do_smi)
832                 turbostat_submit(NULL, "current", name, t->smi_count);
833
834         /* print per-core data only for 1st thread in core */
835         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
836                 goto done;
837
838         snprintf(name, NAME_LEN, "core%02d", c->core_id);
839
840         if (do_nhm_cstates && !do_slm_cstates)
841                 turbostat_submit(name, "percent", "c3", 100.0 * c->c3/t->tsc);
842         if (do_nhm_cstates)
843                 turbostat_submit(name, "percent", "c6", 100.0 * c->c6/t->tsc);
844         if (do_snb_cstates)
845                 turbostat_submit(name, "percent", "c7", 100.0 * c->c7/t->tsc);
846
847         if (do_dts)
848                 turbostat_submit(NULL, "temperature", name, c->core_temp_c);
849
850         /* print per-package data only for 1st core in package */
851         if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
852                 goto done;
853
854         snprintf(name, NAME_LEN, "pkg%02d", p->package_id);
855
856         if (do_ptm)
857                 turbostat_submit(NULL, "temperature", name, p->pkg_temp_c);
858
859         if (do_snb_cstates)
860                 turbostat_submit(name, "percent", "pc2", 100.0 * p->pc2/t->tsc);
861         if (do_nhm_cstates && !do_slm_cstates)
862                 turbostat_submit(name, "percent", "pc3", 100.0 * p->pc3/t->tsc);
863         if (do_nhm_cstates && !do_slm_cstates)
864                 turbostat_submit(name, "percent", "pc6", 100.0 * p->pc6/t->tsc);
865         if (do_snb_cstates)
866                 turbostat_submit(name, "percent", "pc7", 100.0 * p->pc7/t->tsc);
867         if (do_c8_c9_c10) {
868                 turbostat_submit(name, "percent", "pc8", 100.0 * p->pc8/t->tsc);
869                 turbostat_submit(name, "percent", "pc9", 100.0 * p->pc9/t->tsc);
870                 turbostat_submit(name, "percent", "pc10", 100.0 * p->pc10/t->tsc);
871         }
872
873         if (do_rapl) {
874                 if (do_rapl & RAPL_PKG)
875                         turbostat_submit(name, "power", "Pkg_W", p->energy_pkg * rapl_energy_units / interval_float);
876                 if (do_rapl & RAPL_CORES)
877                         turbostat_submit(name, "power", "Cor_W", p->energy_cores * rapl_energy_units / interval_float);
878                 if (do_rapl & RAPL_GFX)
879                         turbostat_submit(name, "power", "GFX_W", p->energy_gfx * rapl_energy_units / interval_float);
880                 if (do_rapl & RAPL_DRAM)
881                         turbostat_submit(name, "power", "RAM_W", p->energy_dram * rapl_energy_units / interval_float);
882         }
883 done:
884         return 0;
885 }
886
887 static int
888 turbostat_read(user_data_t * not_used)
889 {
890         int ret;
891
892         if (!allocated) {
893                 if ((ret = setup_all_buffers()) < 0)
894                         return ret;
895         }
896
897         if (for_all_proc_cpus(cpu_is_not_present)) {
898                 free_all_buffers();
899                 if ((ret = setup_all_buffers()) < 0)
900                         return ret;
901                 if (for_all_proc_cpus(cpu_is_not_present))
902                         return -ERR_CPU_NOT_PRESENT;
903         }
904
905         if (!initialized) {
906                 if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
907                         return ret;
908                 gettimeofday(&tv_even, (struct timezone *)NULL);
909                 is_even = 1;
910                 initialized = 1;
911                 return 0;
912         }
913
914         if (is_even) {
915                 if ((ret = for_all_cpus(get_counters, ODD_COUNTERS)) < 0)
916                         return ret;
917                 gettimeofday(&tv_odd, (struct timezone *)NULL);
918                 is_even = 0;
919                 timersub(&tv_odd, &tv_even, &tv_delta);
920                 if ((ret = for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) < 0)
921                         return ret;
922                 if ((ret = for_all_cpus(submit_counters, EVEN_COUNTERS)) < 0)
923                         return ret;
924         } else {
925                 if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
926                         return ret;
927                 gettimeofday(&tv_even, (struct timezone *)NULL);
928                 is_even = 1;
929                 timersub(&tv_even, &tv_odd, &tv_delta);
930                 if ((ret = for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS)) < 0)
931                         return ret;
932                 if ((ret = for_all_cpus(submit_counters, ODD_COUNTERS)) < 0)
933                         return ret;
934         }
935         return 0;
936 }
937
938 static int __attribute__((warn_unused_result))
939 check_dev_msr()
940 {
941         struct stat sb;
942
943         if (stat("/dev/cpu/0/msr", &sb)) {
944                 ERROR("no /dev/cpu/0/msr\n"
945                         "Try \"# modprobe msr\"");
946                 return -ERR_NO_MSR;
947         }
948         return 0;
949 }
950
951 static int __attribute__((warn_unused_result))
952 check_super_user()
953 {
954         if (getuid() != 0) {
955                 ERROR("must be root");
956                 return -ERR_NOT_ROOT;
957         }
958         return 0;
959 }
960
961
962 #define RAPL_POWER_GRANULARITY  0x7FFF  /* 15 bit power granularity */
963 #define RAPL_TIME_GRANULARITY   0x3F /* 6 bit time granularity */
964
965 static double
966 get_tdp(unsigned int model)
967 {
968         unsigned long long msr;
969
970         if (do_rapl & RAPL_PKG_POWER_INFO)
971                 if (!get_msr(0, MSR_PKG_POWER_INFO, &msr))
972                         return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
973
974         switch (model) {
975         case 0x37:
976         case 0x4D:
977                 return 30.0;
978         default:
979                 return 135.0;
980         }
981 }
982
983
984 /*
985  * rapl_probe()
986  *
987  * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
988  */
989 static void
990 rapl_probe(unsigned int family, unsigned int model)
991 {
992         unsigned long long msr;
993         unsigned int time_unit;
994         double tdp;
995
996         if (!genuine_intel)
997                 return;
998
999         if (family != 6)
1000                 return;
1001
1002         switch (model) {
1003         case 0x2A:
1004         case 0x3A:
1005         case 0x3C:      /* HSW */
1006         case 0x45:      /* HSW */
1007         case 0x46:      /* HSW */
1008                 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
1009                 break;
1010         case 0x3F:      /* HSX */
1011                 do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
1012                 break;
1013         case 0x2D:
1014         case 0x3E:
1015                 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_PKG_PERF_STATUS | RAPL_DRAM_PERF_STATUS | RAPL_PKG_POWER_INFO;
1016                 break;
1017         case 0x37:      /* BYT */
1018         case 0x4D:      /* AVN */
1019                 do_rapl = RAPL_PKG | RAPL_CORES ;
1020                 break;
1021         default:
1022                 return;
1023         }
1024
1025         /* units on package 0, verify later other packages match */
1026         if (get_msr(0, MSR_RAPL_POWER_UNIT, &msr))
1027                 return;
1028
1029         rapl_power_units = 1.0 / (1 << (msr & 0xF));
1030         if (model == 0x37)
1031                 rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
1032         else
1033                 rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
1034
1035         time_unit = msr >> 16 & 0xF;
1036         if (time_unit == 0)
1037                 time_unit = 0xA;
1038
1039         rapl_time_units = 1.0 / (1 << (time_unit));
1040
1041         tdp = get_tdp(model);
1042
1043         rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
1044 //      if (verbose)
1045 //              fprintf(stderr, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
1046
1047         return;
1048 }
1049
1050 static int
1051 is_snb(unsigned int family, unsigned int model)
1052 {
1053         if (!genuine_intel)
1054                 return 0;
1055
1056         switch (model) {
1057         case 0x2A:
1058         case 0x2D:
1059         case 0x3A:      /* IVB */
1060         case 0x3E:      /* IVB Xeon */
1061         case 0x3C:      /* HSW */
1062         case 0x3F:      /* HSW */
1063         case 0x45:      /* HSW */
1064         case 0x46:      /* HSW */
1065                 return 1;
1066         }
1067         return 0;
1068 }
1069
1070 static int
1071 has_c8_c9_c10(unsigned int family, unsigned int model)
1072 {
1073         if (!genuine_intel)
1074                 return 0;
1075
1076         switch (model) {
1077         case 0x45:
1078                 return 1;
1079         }
1080         return 0;
1081 }
1082
1083
1084 static int
1085 is_slm(unsigned int family, unsigned int model)
1086 {
1087         if (!genuine_intel)
1088                 return 0;
1089         switch (model) {
1090         case 0x37:      /* BYT */
1091         case 0x4D:      /* AVN */
1092                 return 1;
1093         }
1094         return 0;
1095 }
1096
1097 /*
1098  * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
1099  * the Thermal Control Circuit (TCC) activates.
1100  * This is usually equal to tjMax.
1101  *
1102  * Older processors do not have this MSR, so there we guess,
1103  * but also allow cmdline over-ride with -T.
1104  *
1105  * Several MSR temperature values are in units of degrees-C
1106  * below this value, including the Digital Thermal Sensor (DTS),
1107  * Package Thermal Management Sensor (PTM), and thermal event thresholds.
1108  */
1109 static int __attribute__((warn_unused_result))
1110 set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1111 {
1112         unsigned long long msr;
1113         unsigned int target_c_local;
1114         int cpu;
1115
1116         /* tcc_activation_temp is used only for dts or ptm */
1117         if (!(do_dts || do_ptm))
1118                 return 0;
1119
1120         /* this is a per-package concept */
1121         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1122                 return 0;
1123
1124         cpu = t->cpu_id;
1125         if (cpu_migrate(cpu)) {
1126                 ERROR("Could not migrate to CPU %d\n", cpu);
1127                 return -ERR_CPU_MIGRATE;
1128         }
1129
1130         if (tcc_activation_temp_override != 0) {
1131                 tcc_activation_temp = tcc_activation_temp_override;
1132                 ERROR("cpu%d: Using cmdline TCC Target (%d C)\n",
1133                         cpu, tcc_activation_temp);
1134                 return 0;
1135         }
1136
1137         /* Temperature Target MSR is Nehalem and newer only */
1138         if (!do_nehalem_platform_info)
1139                 goto guess;
1140
1141         if (get_msr(0, MSR_IA32_TEMPERATURE_TARGET, &msr))
1142                 goto guess;
1143
1144         target_c_local = (msr >> 16) & 0x7F;
1145
1146         if (target_c_local < 85 || target_c_local > 127)
1147                 goto guess;
1148
1149         tcc_activation_temp = target_c_local;
1150
1151         return 0;
1152
1153 guess:
1154         tcc_activation_temp = TJMAX_DEFAULT;
1155         WARNING("cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
1156                 cpu, tcc_activation_temp);
1157
1158         return 0;
1159 }
1160
1161 static int __attribute__((warn_unused_result))
1162 check_cpuid()
1163 {
1164         unsigned int eax, ebx, ecx, edx, max_level;
1165         unsigned int fms, family, model;
1166
1167         eax = ebx = ecx = edx = 0;
1168
1169         __get_cpuid(0, &max_level, &ebx, &ecx, &edx);
1170
1171         if (ebx == 0x756e6547 && edx == 0x49656e69 && ecx == 0x6c65746e)
1172                 genuine_intel = 1;
1173
1174         fms = 0;
1175         __get_cpuid(1, &fms, &ebx, &ecx, &edx);
1176         family = (fms >> 8) & 0xf;
1177         model = (fms >> 4) & 0xf;
1178         if (family == 6 || family == 0xf)
1179                 model += ((fms >> 16) & 0xf) << 4;
1180
1181         if (!(edx & (1 << 5))) {
1182                 ERROR("CPUID: no MSR");
1183                 return -ERR_NO_MSR;
1184         }
1185
1186         /*
1187          * check max extended function levels of CPUID.
1188          * This is needed to check for invariant TSC.
1189          * This check is valid for both Intel and AMD.
1190          */
1191         ebx = ecx = edx = 0;
1192         __get_cpuid(0x80000000, &max_level, &ebx, &ecx, &edx);
1193
1194         if (max_level < 0x80000007) {
1195                 ERROR("CPUID: no invariant TSC (max_level 0x%x)", max_level);
1196                 return -ERR_NO_INVARIANT_TSC;
1197         }
1198
1199         /*
1200          * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
1201          * this check is valid for both Intel and AMD
1202          */
1203         __get_cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
1204         has_invariant_tsc = edx & (1 << 8);
1205
1206         if (!has_invariant_tsc) {
1207                 ERROR("No invariant TSC");
1208                 return -ERR_NO_INVARIANT_TSC;
1209         }
1210
1211         /*
1212          * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
1213          * this check is valid for both Intel and AMD
1214          */
1215
1216         __get_cpuid(0x6, &eax, &ebx, &ecx, &edx);
1217         has_aperf = ecx & (1 << 0);
1218         do_dts = eax & (1 << 0);
1219         do_ptm = eax & (1 << 6);
1220         has_epb = ecx & (1 << 3);
1221
1222         if (!has_aperf) {
1223                 ERROR("No APERF");
1224                 return -ERR_NO_APERF;
1225         }
1226
1227         do_nehalem_platform_info = genuine_intel && has_invariant_tsc;
1228         do_nhm_cstates = genuine_intel; /* all Intel w/ non-stop TSC have NHM counters */
1229         do_smi = do_nhm_cstates;
1230         do_snb_cstates = is_snb(family, model);
1231         do_c8_c9_c10 = has_c8_c9_c10(family, model);
1232         do_slm_cstates = is_slm(family, model);
1233
1234         rapl_probe(family, model);
1235
1236         return 0;
1237 }
1238
1239
1240
1241 static int __attribute__((warn_unused_result))
1242 topology_probe()
1243 {
1244         int i;
1245         int ret;
1246         int max_core_id = 0;
1247         int max_package_id = 0;
1248         int max_siblings = 0;
1249         struct cpu_topology {
1250                 int core_id;
1251                 int physical_package_id;
1252         } *cpus;
1253
1254         /* Initialize num_cpus, max_cpu_num */
1255         topo.num_cpus = 0;
1256         topo.max_cpu_num = 0;
1257         ret = for_all_proc_cpus(count_cpus);
1258         if (ret < 0)
1259                 return ret;
1260         if (topo.num_cpus > 1)
1261                 show_cpu = 1;
1262
1263         DEBUG("num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
1264
1265         cpus = calloc(1, (topo.max_cpu_num  + 1) * sizeof(struct cpu_topology));
1266         if (cpus == NULL) {
1267                 ERROR("calloc cpus");
1268                 return -ERR_CALLOC;
1269         }
1270
1271         /*
1272          * Allocate and initialize cpu_present_set
1273          */
1274         cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
1275         if (cpu_present_set == NULL) {
1276                 free(cpus);
1277                 ERROR("CPU_ALLOC");
1278                 return -ERR_CPU_ALLOC;
1279         }
1280         cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
1281         CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
1282         ret = for_all_proc_cpus(mark_cpu_present);
1283         if (ret < 0) {
1284                 free(cpus);
1285                 return ret;
1286         }
1287
1288         /*
1289          * Allocate and initialize cpu_affinity_set
1290          */
1291         cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
1292         if (cpu_affinity_set == NULL) {
1293                 free(cpus);
1294                 ERROR("CPU_ALLOC");
1295                 return -ERR_CPU_ALLOC;
1296         }
1297         cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
1298         CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
1299
1300
1301         /*
1302          * For online cpus
1303          * find max_core_id, max_package_id
1304          */
1305         for (i = 0; i <= topo.max_cpu_num; ++i) {
1306                 int siblings;
1307
1308                 if (cpu_is_not_present(i)) {
1309                         //if (verbose > 1)
1310                                 fprintf(stderr, "cpu%d NOT PRESENT\n", i);
1311                         continue;
1312                 }
1313                 cpus[i].core_id = get_core_id(i);
1314                 if (cpus[i].core_id < 0)
1315                         return cpus[i].core_id;
1316                 if (cpus[i].core_id > max_core_id)
1317                         max_core_id = cpus[i].core_id;
1318
1319                 cpus[i].physical_package_id = get_physical_package_id(i);
1320                 if (cpus[i].physical_package_id < 0)
1321                         return cpus[i].physical_package_id;
1322                 if (cpus[i].physical_package_id > max_package_id)
1323                         max_package_id = cpus[i].physical_package_id;
1324
1325                 siblings = get_num_ht_siblings(i);
1326                 if (siblings < 0)
1327                         return siblings;
1328                 if (siblings > max_siblings)
1329                         max_siblings = siblings;
1330                 DEBUG("cpu %d pkg %d core %d\n",
1331                         i, cpus[i].physical_package_id, cpus[i].core_id);
1332         }
1333         topo.num_cores_per_pkg = max_core_id + 1;
1334         DEBUG("max_core_id %d, sizing for %d cores per package\n",
1335                 max_core_id, topo.num_cores_per_pkg);
1336         if (topo.num_cores_per_pkg > 1)
1337                 show_core = 1;
1338
1339         topo.num_packages = max_package_id + 1;
1340         DEBUG("max_package_id %d, sizing for %d packages\n",
1341                 max_package_id, topo.num_packages);
1342         if (topo.num_packages > 1)
1343                 show_pkg = 1;
1344
1345         topo.num_threads_per_core = max_siblings;
1346         DEBUG("max_siblings %d\n", max_siblings);
1347
1348         free(cpus);
1349         return 0;
1350 }
1351
1352 static int
1353 allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p)
1354 {
1355         int i;
1356
1357         *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg *
1358                 topo.num_packages, sizeof(struct thread_data));
1359         if (*t == NULL)
1360                 goto error;
1361
1362         for (i = 0; i < topo.num_threads_per_core *
1363                 topo.num_cores_per_pkg * topo.num_packages; i++)
1364                 (*t)[i].cpu_id = -1;
1365
1366         *c = calloc(topo.num_cores_per_pkg * topo.num_packages,
1367                 sizeof(struct core_data));
1368         if (*c == NULL)
1369                 goto error;
1370
1371         for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++)
1372                 (*c)[i].core_id = -1;
1373
1374         *p = calloc(topo.num_packages, sizeof(struct pkg_data));
1375         if (*p == NULL)
1376                 goto error;
1377
1378         for (i = 0; i < topo.num_packages; i++)
1379                 (*p)[i].package_id = i;
1380
1381         return 0;
1382 error:
1383         ERROR("calloc counters");
1384         return -ERR_CALLOC;
1385 }
1386 /*
1387  * init_counter()
1388  *
1389  * set cpu_id, core_num, pkg_num
1390  * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
1391  *
1392  * increment topo.num_cores when 1st core in pkg seen
1393  */
1394 static int
1395 init_counter(struct thread_data *thread_base, struct core_data *core_base,
1396         struct pkg_data *pkg_base, int thread_num, int core_num,
1397         int pkg_num, int cpu_id)
1398 {
1399         int ret;
1400         struct thread_data *t;
1401         struct core_data *c;
1402         struct pkg_data *p;
1403
1404         t = GET_THREAD(thread_base, thread_num, core_num, pkg_num);
1405         c = GET_CORE(core_base, core_num, pkg_num);
1406         p = GET_PKG(pkg_base, pkg_num);
1407
1408         t->cpu_id = cpu_id;
1409         if (thread_num == 0) {
1410                 t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
1411                 if ((ret = cpu_is_first_core_in_package(cpu_id)) < 0) {
1412                         return ret;
1413                 } else if (ret != 0) {
1414                         t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
1415                 }
1416         }
1417
1418         c->core_id = core_num;
1419         p->package_id = pkg_num;
1420
1421         return 0;
1422 }
1423
1424
1425 static int
1426 initialize_counters(int cpu_id)
1427 {
1428         int my_thread_id, my_core_id, my_package_id;
1429         int ret;
1430
1431         my_package_id = get_physical_package_id(cpu_id);
1432         if (my_package_id < 0)
1433                 return my_package_id;
1434         my_core_id = get_core_id(cpu_id);
1435         if (my_core_id < 0)
1436                 return my_core_id;
1437
1438         if ((ret = cpu_is_first_sibling_in_core(cpu_id)) < 0) {
1439                 return ret;
1440         } else if (ret != 0) {
1441                 my_thread_id = 0;
1442                 topo.num_cores++;
1443         } else {
1444                 my_thread_id = 1;
1445         }
1446
1447         ret = init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
1448         if (ret < 0)
1449                 return ret;
1450         ret = init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
1451         if (ret < 0)
1452                 return ret;
1453         return 0;
1454 }
1455
1456 #define DO_OR_GOTO_ERR(something) \
1457 do {                         \
1458         ret = (something);     \
1459         if (ret < 0)         \
1460                 goto err;    \
1461 } while (0)
1462
1463 static int setup_all_buffers(void)
1464 {
1465         int ret;
1466
1467         DO_OR_GOTO_ERR(topology_probe());
1468         DO_OR_GOTO_ERR(allocate_counters(&thread_even, &core_even, &package_even));
1469         DO_OR_GOTO_ERR(allocate_counters(&thread_odd, &core_odd, &package_odd));
1470         DO_OR_GOTO_ERR(for_all_proc_cpus(initialize_counters));
1471
1472         allocated = 1;
1473         return 0;
1474 err:
1475         free_all_buffers();
1476         return ret;
1477 }
1478
1479 static int
1480 turbostat_init(void)
1481 {
1482         int ret;
1483         struct timespec ts;
1484
1485         DO_OR_GOTO_ERR(check_cpuid());
1486         DO_OR_GOTO_ERR(check_dev_msr());
1487         DO_OR_GOTO_ERR(check_super_user());
1488         DO_OR_GOTO_ERR(setup_all_buffers());
1489         DO_OR_GOTO_ERR(for_all_cpus(set_temperature_target, EVEN_COUNTERS));
1490
1491         ts.tv_sec = interval_sec;
1492         ts.tv_nsec = 0;
1493
1494         plugin_register_complex_read(NULL, PLUGIN_NAME, turbostat_read, &ts, NULL);
1495
1496         return 0;
1497 err:
1498         free_all_buffers();
1499         return ret;
1500 }
1501
1502 static const char *config_keys[] =
1503 {
1504         "Interval",
1505 };
1506 static int config_keys_num = STATIC_ARRAY_SIZE (config_keys);
1507
1508 static int
1509 turbostat_config(const char *key, const char *value)
1510 {
1511         if (strcasecmp("Interval", key) == 0)
1512                 interval_sec = atoi(value);
1513         else
1514                 return -1;
1515         return 0;
1516 }
1517
1518 void module_register(void);
1519 void module_register(void)
1520 {
1521         plugin_register_init(PLUGIN_NAME, turbostat_init);
1522         plugin_register_config(PLUGIN_NAME, turbostat_config, config_keys, config_keys_num);
1523 }