![[tokkee]](http://tokkee.org/images/avatar.png){kind=avatar}
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 */
27 /*
28 * _GNU_SOURCE is required because of the following functions:
29 * - CPU_ISSET_S
30 * - CPU_ZERO_S
31 * - CPU_SET_S
32 * - CPU_FREE
33 * - CPU_ALLOC
34 * - CPU_ALLOC_SIZE
35 */
36 #define _GNU_SOURCE
38 #include "collectd.h"
39 #include "common.h"
40 #include "plugin.h"
41 #include "utils_time.h"
43 #include <asm/msr-index.h>
44 #include <cpuid.h>
45 #ifdef HAVE_SYS_CAPABILITY_H
46 #include <sys/capability.h>
47 #endif /* HAVE_SYS_CAPABILITY_H */
49 #define PLUGIN_NAME "turbostat"
51 /*
52 * This tool uses the Model-Specific Registers (MSRs) present on Intel processors.
53 * The general description each of these registers, depending on the architecture,
54 * can be found in the Intel® 64 and IA-32 Architectures Software Developer Manual,
55 * Volume 3 Chapter 35.
56 */
58 /*
59 * If set, aperf_mperf_unstable disables a/mperf based stats.
60 * This includes: C0 & C1 states, frequency
61 *
62 * This value is automatically set if mperf or aperf go backward
63 */
64 static _Bool aperf_mperf_unstable;
66 /*
67 * Bitmask of the list of core C states supported by the processor.
68 * Currently supported C-states (by this plugin): 3, 6, 7
69 */
70 static unsigned int do_core_cstate;
71 static unsigned int config_core_cstate;
72 static _Bool apply_config_core_cstate;
74 /*
75 * Bitmask of the list of pacages C states supported by the processor.
76 * Currently supported C-states (by this plugin): 2, 3, 6, 7, 8, 9, 10
77 */
78 static unsigned int do_pkg_cstate;
79 static unsigned int config_pkg_cstate;
80 static _Bool apply_config_pkg_cstate;
82 /*
83 * Boolean indicating if the processor supports 'I/O System-Management Interrupt counter'
84 */
85 static _Bool do_smi;
86 static _Bool config_smi;
87 static _Bool apply_config_smi;
89 /*
90 * Boolean indicating if the processor supports 'Digital temperature sensor'
91 * This feature enables the monitoring of the temperature of each core
92 *
93 * This feature has two limitations:
94 * - if MSR_IA32_TEMPERATURE_TARGET is not supported, the absolute temperature might be wrong
95 * - Temperatures above the tcc_activation_temp are not recorded
96 */
97 static _Bool do_dts;
98 static _Bool config_dts;
99 static _Bool apply_config_dts;
101 /*
102 * Boolean indicating if the processor supports 'Package thermal management'
103 * This feature allows the monitoring of the temperature of each package
104 *
105 * This feature has two limitations:
106 * - if MSR_IA32_TEMPERATURE_TARGET is not supported, the absolute temperature might be wrong
107 * - Temperatures above the tcc_activation_temp are not recorded
108 */
109 static _Bool do_ptm;
110 static _Bool config_ptm;
111 static _Bool apply_config_ptm;
113 /*
114 * Thermal Control Circuit Activation Temperature as configured by the user.
115 * This override the automated detection via MSR_IA32_TEMPERATURE_TARGET
116 * and should only be used if the automated detection fails.
117 */
118 static unsigned int tcc_activation_temp;
120 static unsigned int do_rapl;
121 static unsigned int config_rapl;
122 static _Bool apply_config_rapl;
123 static double rapl_energy_units;
125 #define RAPL_PKG (1 << 0)
126 /* 0x610 MSR_PKG_POWER_LIMIT */
127 /* 0x611 MSR_PKG_ENERGY_STATUS */
128 #define RAPL_DRAM (1 << 1)
129 /* 0x618 MSR_DRAM_POWER_LIMIT */
130 /* 0x619 MSR_DRAM_ENERGY_STATUS */
131 /* 0x61c MSR_DRAM_POWER_INFO */
132 #define RAPL_CORES (1 << 2)
133 /* 0x638 MSR_PP0_POWER_LIMIT */
134 /* 0x639 MSR_PP0_ENERGY_STATUS */
136 #define RAPL_GFX (1 << 3)
137 /* 0x640 MSR_PP1_POWER_LIMIT */
138 /* 0x641 MSR_PP1_ENERGY_STATUS */
139 /* 0x642 MSR_PP1_POLICY */
140 #define TJMAX_DEFAULT 100
142 static cpu_set_t *cpu_present_set, *cpu_affinity_set, *cpu_saved_affinity_set;
143 static size_t cpu_present_setsize, cpu_affinity_setsize, cpu_saved_affinity_setsize;
145 static struct thread_data {
146 unsigned long long tsc;
147 unsigned long long aperf;
148 unsigned long long mperf;
149 unsigned long long c1;
150 unsigned int smi_count;
151 unsigned int cpu_id;
152 unsigned int flags;
153 #define CPU_IS_FIRST_THREAD_IN_CORE 0x2
154 #define CPU_IS_FIRST_CORE_IN_PACKAGE 0x4
155 } *thread_delta, *thread_even, *thread_odd;
157 static struct core_data {
158 unsigned long long c3;
159 unsigned long long c6;
160 unsigned long long c7;
161 unsigned int core_temp_c;
162 unsigned int core_id;
163 } *core_delta, *core_even, *core_odd;
165 static struct pkg_data {
166 unsigned long long pc2;
167 unsigned long long pc3;
168 unsigned long long pc6;
169 unsigned long long pc7;
170 unsigned long long pc8;
171 unsigned long long pc9;
172 unsigned long long pc10;
173 unsigned int package_id;
174 uint32_t energy_pkg; /* MSR_PKG_ENERGY_STATUS */
175 uint32_t energy_dram; /* MSR_DRAM_ENERGY_STATUS */
176 uint32_t energy_cores; /* MSR_PP0_ENERGY_STATUS */
177 uint32_t energy_gfx; /* MSR_PP1_ENERGY_STATUS */
178 unsigned int tcc_activation_temp;
179 unsigned int pkg_temp_c;
180 } *package_delta, *package_even, *package_odd;
182 #define DELTA_COUNTERS thread_delta, core_delta, package_delta
183 #define ODD_COUNTERS thread_odd, core_odd, package_odd
184 #define EVEN_COUNTERS thread_even, core_even, package_even
185 static _Bool is_even = 1;
187 static _Bool allocated = 0;
188 static _Bool initialized = 0;
190 #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \
191 (thread_base + \
192 (pkg_no) * topology.num_cores * topology.num_threads + \
193 (core_no) * topology.num_threads + \
194 (thread_no))
195 #define GET_CORE(core_base, core_no, pkg_no) \
196 (core_base + \
197 (pkg_no) * topology.num_cores + \
198 (core_no))
199 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
201 struct cpu_topology {
202 unsigned int package_id;
203 unsigned int core_id;
204 _Bool first_core_in_package;
205 _Bool first_thread_in_core;
206 };
208 static struct topology {
209 unsigned int max_cpu_id;
210 unsigned int num_packages;
211 unsigned int num_cores;
212 unsigned int num_threads;
213 struct cpu_topology *cpus;
214 } topology;
216 static cdtime_t time_even, time_odd, time_delta;
218 static const char *config_keys[] =
219 {
220 "CoreCstates",
221 "PackageCstates",
222 "SystemManagementInterrupt",
223 "DigitalTemperatureSensor",
224 "PackageThermalManagement",
225 "TCCActivationTemp",
226 "RunningAveragePowerLimit",
227 };
228 static const int config_keys_num = STATIC_ARRAY_SIZE (config_keys);
230 /*****************************
231 * MSR Manipulation helpers *
232 *****************************/
234 /*
235 * Open a MSR device for reading
236 * Can change the scheduling affinity of the current process if multiple_read is 1
237 */
238 static int __attribute__((warn_unused_result))
239 open_msr(unsigned int cpu, _Bool multiple_read)
240 {
241 char pathname[32];
242 int fd;
244 /*
245 * If we need to do multiple read, let's migrate to the CPU
246 * Otherwise, we would lose time calling functions on another CPU
247 *
248 * If we are not yet initialized (cpu_affinity_setsize = 0),
249 * we need to skip this optimisation.
250 */
251 if (multiple_read && cpu_affinity_setsize) {
252 CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
253 CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
254 if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1) {
255 ERROR("turbostat plugin: Could not migrate to CPU %d", cpu);
256 return -1;
257 }
258 }
260 ssnprintf(pathname, sizeof(pathname), "/dev/cpu/%d/msr", cpu);
261 fd = open(pathname, O_RDONLY);
262 if (fd < 0) {
263 ERROR("turbostat plugin: failed to open %s", pathname);
264 return -1;
265 }
266 return fd;
267 }
269 /*
270 * Read a single MSR from an open file descriptor
271 */
272 static int __attribute__((warn_unused_result))
273 read_msr(int fd, off_t offset, unsigned long long *msr)
274 {
275 ssize_t retval;
277 retval = pread(fd, msr, sizeof *msr, offset);
279 if (retval != sizeof *msr) {
280 ERROR("turbostat plugin: MSR offset 0x%llx read failed",
281 (unsigned long long)offset);
282 return -1;
283 }
284 return 0;
285 }
287 /*
288 * Open a MSR device for reading, read the value asked for and close it.
289 * This call will not affect the scheduling affinity of this thread.
290 */
291 static ssize_t __attribute__((warn_unused_result))
292 get_msr(unsigned int cpu, off_t offset, unsigned long long *msr)
293 {
294 ssize_t retval;
295 int fd;
297 fd = open_msr(cpu, 0);
298 if (fd < 0)
299 return fd;
300 retval = read_msr(fd, offset, msr);
301 close(fd);
302 return retval;
303 }
306 /********************************
307 * Raw data acquisition (1 CPU) *
308 ********************************/
310 /*
311 * Read every data avalaible for a single CPU
312 *
313 * Core data is shared for all threads in one core: extracted only for the first thread
314 * Package data is shared for all core in one package: extracted only for the first thread of the first core
315 *
316 * Side effect: migrates to the targeted CPU
317 */
318 static int __attribute__((warn_unused_result))
319 get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
320 {
321 unsigned int cpu = t->cpu_id;
322 unsigned long long msr;
323 int msr_fd;
324 int retval = 0;
326 msr_fd = open_msr(cpu, 1);
327 if (msr_fd < 0)
328 return msr_fd;
330 #define READ_MSR(msr, dst) \
331 do { \
332 if (read_msr(msr_fd, msr, dst)) { \
333 ERROR("turbostat plugin: Unable to read " #msr); \
334 retval = -1; \
335 goto out; \
336 } \
337 } while (0)
339 READ_MSR(MSR_IA32_TSC, &t->tsc);
341 READ_MSR(MSR_IA32_APERF, &t->aperf);
342 READ_MSR(MSR_IA32_MPERF, &t->mperf);
344 if (do_smi) {
345 READ_MSR(MSR_SMI_COUNT, &msr);
346 t->smi_count = msr & 0xFFFFFFFF;
347 }
349 /* collect core counters only for 1st thread in core */
350 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) {
351 retval = 0;
352 goto out;
353 }
355 if (do_core_cstate & (1 << 3))
356 READ_MSR(MSR_CORE_C3_RESIDENCY, &c->c3);
357 if (do_core_cstate & (1 << 6))
358 READ_MSR(MSR_CORE_C6_RESIDENCY, &c->c6);
359 if (do_core_cstate & (1 << 7))
360 READ_MSR(MSR_CORE_C7_RESIDENCY, &c->c7);
362 if (do_dts) {
363 READ_MSR(MSR_IA32_THERM_STATUS, &msr);
364 c->core_temp_c = p->tcc_activation_temp - ((msr >> 16) & 0x7F);
365 }
367 /* collect package counters only for 1st core in package */
368 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
369 retval = 0;
370 goto out;
371 }
373 if (do_pkg_cstate & (1 << 2))
374 READ_MSR(MSR_PKG_C2_RESIDENCY, &p->pc2);
375 if (do_pkg_cstate & (1 << 3))
376 READ_MSR(MSR_PKG_C3_RESIDENCY, &p->pc3);
377 if (do_pkg_cstate & (1 << 6))
378 READ_MSR(MSR_PKG_C6_RESIDENCY, &p->pc6);
379 if (do_pkg_cstate & (1 << 7))
380 READ_MSR(MSR_PKG_C7_RESIDENCY, &p->pc7);
381 if (do_pkg_cstate & (1 << 8))
382 READ_MSR(MSR_PKG_C8_RESIDENCY, &p->pc8);
383 if (do_pkg_cstate & (1 << 9))
384 READ_MSR(MSR_PKG_C9_RESIDENCY, &p->pc9);
385 if (do_pkg_cstate & (1 << 10))
386 READ_MSR(MSR_PKG_C10_RESIDENCY, &p->pc10);
388 if (do_rapl & RAPL_PKG) {
389 READ_MSR(MSR_PKG_ENERGY_STATUS, &msr);
390 p->energy_pkg = msr & 0xFFFFFFFF;
391 }
392 if (do_rapl & RAPL_CORES) {
393 READ_MSR(MSR_PP0_ENERGY_STATUS, &msr);
394 p->energy_cores = msr & 0xFFFFFFFF;
395 }
396 if (do_rapl & RAPL_DRAM) {
397 READ_MSR(MSR_DRAM_ENERGY_STATUS, &msr);
398 p->energy_dram = msr & 0xFFFFFFFF;
399 }
400 if (do_rapl & RAPL_GFX) {
401 READ_MSR(MSR_PP1_ENERGY_STATUS, &msr);
402 p->energy_gfx = msr & 0xFFFFFFFF;
403 }
404 if (do_ptm) {
405 READ_MSR(MSR_IA32_PACKAGE_THERM_STATUS, &msr);
406 p->pkg_temp_c = p->tcc_activation_temp - ((msr >> 16) & 0x7F);
407 }
409 out:
410 close(msr_fd);
411 return retval;
412 }
415 /**********************************
416 * Evaluating the changes (1 CPU) *
417 **********************************/
419 /*
420 * Extract the evolution old->new in delta at a package level
421 * (some are not new-delta, e.g. temperature)
422 */
423 static inline void
424 delta_package(struct pkg_data *delta, const struct pkg_data *new, const struct pkg_data *old)
425 {
426 delta->pc2 = new->pc2 - old->pc2;
427 delta->pc3 = new->pc3 - old->pc3;
428 delta->pc6 = new->pc6 - old->pc6;
429 delta->pc7 = new->pc7 - old->pc7;
430 delta->pc8 = new->pc8 - old->pc8;
431 delta->pc9 = new->pc9 - old->pc9;
432 delta->pc10 = new->pc10 - old->pc10;
433 delta->pkg_temp_c = new->pkg_temp_c;
435 delta->energy_pkg = new->energy_pkg - old->energy_pkg;
436 delta->energy_cores = new->energy_cores - old->energy_cores;
437 delta->energy_gfx = new->energy_gfx - old->energy_gfx;
438 delta->energy_dram = new->energy_dram - old->energy_dram;
439 }
441 /*
442 * Extract the evolution old->new in delta at a core level
443 * (some are not new-delta, e.g. temperature)
444 */
445 static inline void
446 delta_core(struct core_data *delta, const struct core_data *new, const struct core_data *old)
447 {
448 delta->c3 = new->c3 - old->c3;
449 delta->c6 = new->c6 - old->c6;
450 delta->c7 = new->c7 - old->c7;
451 delta->core_temp_c = new->core_temp_c;
452 }
454 /*
455 * Extract the evolution old->new in delta at a package level
456 * core_delta is required for c1 estimation (tsc - c0 - all core cstates)
457 */
458 static inline int __attribute__((warn_unused_result))
459 delta_thread(struct thread_data *delta, const struct thread_data *new, const struct thread_data *old,
460 const struct core_data *cdelta)
461 {
462 delta->tsc = new->tsc - old->tsc;
464 /* check for TSC < 1 Mcycles over interval */
465 if (delta->tsc < (1000 * 1000)) {
466 WARNING("turbostat plugin: Insanely slow TSC rate, TSC stops "
467 "in idle? You can disable all c-states by booting with"
468 " 'idle=poll' or just the deep ones with"
469 " 'processor.max_cstate=1'");
470 return -1;
471 }
473 delta->c1 = new->c1 - old->c1;
475 if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
476 delta->aperf = new->aperf - old->aperf;
477 delta->mperf = new->mperf - old->mperf;
478 } else {
479 if (!aperf_mperf_unstable) {
480 WARNING("turbostat plugin: APERF or MPERF went "
481 "backwards. Frequency results do not cover "
482 "the entire interval. Fix this by running "
483 "Linux-2.6.30 or later.");
485 aperf_mperf_unstable = 1;
486 }
487 }
489 /*
490 * As counter collection is not atomic,
491 * it is possible for mperf's non-halted cycles + idle states
492 * to exceed TSC's all cycles: show c1 = 0% in that case.
493 */
494 if ((delta->mperf + cdelta->c3 + cdelta->c6 + cdelta->c7) > delta->tsc)
495 delta->c1 = 0;
496 else {
497 /* normal case, derive c1 */
498 delta->c1 = delta->tsc - delta->mperf - cdelta->c3
499 - cdelta->c6 - cdelta->c7;
500 }
502 if (delta->mperf == 0) {
503 WARNING("turbostat plugin: cpu%d MPERF 0!", old->cpu_id);
504 delta->mperf = 1; /* divide by 0 protection */
505 }
507 if (do_smi)
508 delta->smi_count = new->smi_count - old->smi_count;
510 return 0;
511 }
513 /**********************************
514 * Submitting the results (1 CPU) *
515 **********************************/
517 /*
518 * Submit one gauge value
519 */
520 static void
521 turbostat_submit (const char *plugin_instance,
522 const char *type, const char *type_instance,
523 gauge_t value)
524 {
525 value_list_t vl = VALUE_LIST_INIT;
526 value_t v;
528 v.gauge = value;
529 vl.values = &v;
530 vl.values_len = 1;
531 sstrncpy (vl.host, hostname_g, sizeof (vl.host));
532 sstrncpy (vl.plugin, PLUGIN_NAME, sizeof (vl.plugin));
533 if (plugin_instance != NULL)
534 sstrncpy (vl.plugin_instance, plugin_instance, sizeof (vl.plugin_instance));
535 sstrncpy (vl.type, type, sizeof (vl.type));
536 if (type_instance != NULL)
537 sstrncpy (vl.type_instance, type_instance, sizeof (vl.type_instance));
539 plugin_dispatch_values (&vl);
540 }
542 /*
543 * Submit every data for a single CPU
544 *
545 * Core data is shared for all threads in one core: submitted only for the first thread
546 * Package data is shared for all core in one package: submitted only for the first thread of the first core
547 */
548 static int
549 submit_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
550 {
551 char name[DATA_MAX_NAME_LEN];
552 double interval_float;
554 interval_float = CDTIME_T_TO_DOUBLE(time_delta);
556 ssnprintf(name, sizeof(name), "cpu%02d", t->cpu_id);
558 if (!aperf_mperf_unstable)
559 turbostat_submit(name, "percent", "c0", 100.0 * t->mperf/t->tsc);
560 if (!aperf_mperf_unstable)
561 turbostat_submit(name, "percent", "c1", 100.0 * t->c1/t->tsc);
563 turbostat_submit(name, "frequency", "average", 1.0 / 1000000 * t->aperf / interval_float);
565 if ((!aperf_mperf_unstable) || (!(t->aperf > t->tsc || t->mperf > t->tsc)))
566 turbostat_submit(name, "frequency", "busy", 1.0 * t->tsc / 1000000 * t->aperf / t->mperf / interval_float);
568 /* Sanity check (should stay stable) */
569 turbostat_submit(name, "gauge", "TSC", 1.0 * t->tsc / 1000000 / interval_float);
571 /* SMI */
572 if (do_smi)
573 turbostat_submit(name, "count", NULL, t->smi_count);
575 /* submit per-core data only for 1st thread in core */
576 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
577 goto done;
579 ssnprintf(name, sizeof(name), "core%02d", c->core_id);
581 if (do_core_cstate & (1 << 3))
582 turbostat_submit(name, "percent", "c3", 100.0 * c->c3/t->tsc);
583 if (do_core_cstate & (1 << 6))
584 turbostat_submit(name, "percent", "c6", 100.0 * c->c6/t->tsc);
585 if (do_core_cstate & (1 << 7))
586 turbostat_submit(name, "percent", "c7", 100.0 * c->c7/t->tsc);
588 if (do_dts)
589 turbostat_submit(name, "temperature", NULL, c->core_temp_c);
591 /* submit per-package data only for 1st core in package */
592 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
593 goto done;
595 ssnprintf(name, sizeof(name), "pkg%02d", p->package_id);
597 if (do_ptm)
598 turbostat_submit(name, "temperature", NULL, p->pkg_temp_c);
600 if (do_pkg_cstate & (1 << 2))
601 turbostat_submit(name, "percent", "pc2", 100.0 * p->pc2/t->tsc);
602 if (do_pkg_cstate & (1 << 3))
603 turbostat_submit(name, "percent", "pc3", 100.0 * p->pc3/t->tsc);
604 if (do_pkg_cstate & (1 << 6))
605 turbostat_submit(name, "percent", "pc6", 100.0 * p->pc6/t->tsc);
606 if (do_pkg_cstate & (1 << 7))
607 turbostat_submit(name, "percent", "pc7", 100.0 * p->pc7/t->tsc);
608 if (do_pkg_cstate & (1 << 8))
609 turbostat_submit(name, "percent", "pc8", 100.0 * p->pc8/t->tsc);
610 if (do_pkg_cstate & (1 << 9))
611 turbostat_submit(name, "percent", "pc9", 100.0 * p->pc9/t->tsc);
612 if (do_pkg_cstate & (1 << 10))
613 turbostat_submit(name, "percent", "pc10", 100.0 * p->pc10/t->tsc);
615 if (do_rapl) {
616 if (do_rapl & RAPL_PKG)
617 turbostat_submit(name, "power", "pkg", p->energy_pkg * rapl_energy_units / interval_float);
618 if (do_rapl & RAPL_CORES)
619 turbostat_submit(name, "power", "cores", p->energy_cores * rapl_energy_units / interval_float);
620 if (do_rapl & RAPL_GFX)
621 turbostat_submit(name, "power", "GFX", p->energy_gfx * rapl_energy_units / interval_float);
622 if (do_rapl & RAPL_DRAM)
623 turbostat_submit(name, "power", "DRAM", p->energy_dram * rapl_energy_units / interval_float);
624 }
625 done:
626 return 0;
627 }
630 /**********************************
631 * Looping function over all CPUs *
632 **********************************/
634 /*
635 * Check if a given cpu id is in our compiled list of existing CPUs
636 */
637 static int
638 cpu_is_not_present(unsigned int cpu)
639 {
640 return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
641 }
643 /*
644 * Loop on all CPUs in topological order
645 *
646 * Skip non-present cpus
647 * Return the error code at the first error or 0
648 */
649 static int __attribute__((warn_unused_result))
650 for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
651 struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
652 {
653 int retval;
654 unsigned int pkg_no, core_no, thread_no;
656 for (pkg_no = 0; pkg_no < topology.num_packages; ++pkg_no) {
657 for (core_no = 0; core_no < topology.num_cores; ++core_no) {
658 for (thread_no = 0; thread_no < topology.num_threads; ++thread_no) {
659 struct thread_data *t;
660 struct core_data *c;
661 struct pkg_data *p;
663 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
665 if (cpu_is_not_present(t->cpu_id))
666 continue;
668 c = GET_CORE(core_base, core_no, pkg_no);
669 p = GET_PKG(pkg_base, pkg_no);
671 retval = func(t, c, p);
672 if (retval)
673 return retval;
674 }
675 }
676 }
677 return 0;
678 }
680 /*
681 * Dedicated loop: Extract every data evolution for all CPU
682 *
683 * Skip non-present cpus
684 * Return the error code at the first error or 0
685 *
686 * Core data is shared for all threads in one core: extracted only for the first thread
687 * Package data is shared for all core in one package: extracted only for the first thread of the first core
688 */
689 static int __attribute__((warn_unused_result))
690 for_all_cpus_delta(const struct thread_data *thread_new_base, const struct core_data *core_new_base, const struct pkg_data *pkg_new_base,
691 const struct thread_data *thread_old_base, const struct core_data *core_old_base, const struct pkg_data *pkg_old_base)
692 {
693 int retval;
694 unsigned int pkg_no, core_no, thread_no;
696 for (pkg_no = 0; pkg_no < topology.num_packages; ++pkg_no) {
697 for (core_no = 0; core_no < topology.num_cores; ++core_no) {
698 for (thread_no = 0; thread_no < topology.num_threads; ++thread_no) {
699 struct thread_data *t_delta;
700 const struct thread_data *t_old, *t_new;
701 struct core_data *c_delta;
703 /* Get correct pointers for threads */
704 t_delta = GET_THREAD(thread_delta, thread_no, core_no, pkg_no);
705 t_new = GET_THREAD(thread_new_base, thread_no, core_no, pkg_no);
706 t_old = GET_THREAD(thread_old_base, thread_no, core_no, pkg_no);
708 /* Skip threads that disappeared */
709 if (cpu_is_not_present(t_delta->cpu_id))
710 continue;
712 /* c_delta is always required for delta_thread */
713 c_delta = GET_CORE(core_delta, core_no, pkg_no);
715 /* calculate core delta only for 1st thread in core */
716 if (t_new->flags & CPU_IS_FIRST_THREAD_IN_CORE) {
717 const struct core_data *c_old, *c_new;
719 c_new = GET_CORE(core_new_base, core_no, pkg_no);
720 c_old = GET_CORE(core_old_base, core_no, pkg_no);
722 delta_core(c_delta, c_new, c_old);
723 }
725 /* Always calculate thread delta */
726 retval = delta_thread(t_delta, t_new, t_old, c_delta);
727 if (retval)
728 return retval;
730 /* calculate package delta only for 1st core in package */
731 if (t_new->flags & CPU_IS_FIRST_CORE_IN_PACKAGE) {
732 struct pkg_data *p_delta;
733 const struct pkg_data *p_old, *p_new;
735 p_delta = GET_PKG(package_delta, pkg_no);
736 p_new = GET_PKG(pkg_new_base, pkg_no);
737 p_old = GET_PKG(pkg_old_base, pkg_no);
739 delta_package(p_delta, p_new, p_old);
740 }
741 }
742 }
743 }
744 return 0;
745 }
748 /***************
749 * CPU Probing *
750 ***************/
752 /*
753 * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
754 * the Thermal Control Circuit (TCC) activates.
755 * This is usually equal to tjMax.
756 *
757 * Older processors do not have this MSR, so there we guess,
758 * but also allow conficuration over-ride with "TCCActivationTemp".
759 *
760 * Several MSR temperature values are in units of degrees-C
761 * below this value, including the Digital Thermal Sensor (DTS),
762 * Package Thermal Management Sensor (PTM), and thermal event thresholds.
763 */
764 static int __attribute__((warn_unused_result))
765 set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
766 {
767 unsigned long long msr;
768 unsigned int target_c_local;
770 /* tcc_activation_temp is used only for dts or ptm */
771 if (!(do_dts || do_ptm))
772 return 0;
774 /* this is a per-package concept */
775 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
776 return 0;
778 if (tcc_activation_temp != 0) {
779 p->tcc_activation_temp = tcc_activation_temp;
780 return 0;
781 }
783 if (get_msr(t->cpu_id, MSR_IA32_TEMPERATURE_TARGET, &msr))
784 goto guess;
786 target_c_local = (msr >> 16) & 0xFF;
788 if (!target_c_local)
789 goto guess;
791 p->tcc_activation_temp = target_c_local;
793 return 0;
795 guess:
796 p->tcc_activation_temp = TJMAX_DEFAULT;
797 WARNING("turbostat plugin: cpu%d: Guessing tjMax %d C,"
798 " Please use TCCActivationTemp to specify it.",
799 t->cpu_id, p->tcc_activation_temp);
801 return 0;
802 }
804 /*
805 * Identify the functionality of the CPU
806 */
807 static int __attribute__((warn_unused_result))
808 probe_cpu(void)
809 {
810 unsigned int eax, ebx, ecx, edx, max_level;
811 unsigned int fms, family, model;
813 /* CPUID(0):
814 * - EAX: Maximum Input Value for Basic CPUID Information
815 * - EBX: "Genu" (0x756e6547)
816 * - EDX: "ineI" (0x49656e69)
817 * - ECX: "ntel" (0x6c65746e)
818 */
819 max_level = ebx = ecx = edx = 0;
820 __get_cpuid(0, &max_level, &ebx, &ecx, &edx);
821 if (ebx != 0x756e6547 && edx != 0x49656e69 && ecx != 0x6c65746e) {
822 ERROR("turbostat plugin: Unsupported CPU (not Intel)");
823 return -1;
824 }
826 /* CPUID(1):
827 * - EAX: Version Information: Type, Family, Model, and Stepping ID
828 * + 4-7: Model ID
829 * + 8-11: Family ID
830 * + 12-13: Processor type
831 * + 16-19: Extended Model ID
832 * + 20-27: Extended Family ID
833 * - EDX: Feature Information:
834 * + 5: Support for MSR read/write operations
835 */
836 fms = ebx = ecx = edx = 0;
837 __get_cpuid(1, &fms, &ebx, &ecx, &edx);
838 family = (fms >> 8) & 0xf;
839 model = (fms >> 4) & 0xf;
840 if (family == 0xf)
841 family += (fms >> 20) & 0xf;
842 if (family == 6 || family == 0xf)
843 model += ((fms >> 16) & 0xf) << 4;
844 if (!(edx & (1 << 5))) {
845 ERROR("turbostat plugin: Unsupported CPU (no MSR support)");
846 return -1;
847 }
849 /*
850 * CPUID(6):
851 * - EAX:
852 * + 0: Digital temperature sensor is supported if set
853 * + 6: Package thermal management is supported if set
854 * - ECX:
855 * + 0: Hardware Coordination Feedback Capability (Presence of IA32_MPERF and IA32_APERF).
856 * + 3: The processor supports performance-energy bias preference if set.
857 * It also implies the presence of a new architectural MSR called IA32_ENERGY_PERF_BIAS
858 *
859 * This check is valid for both Intel and AMD
860 */
861 eax = ebx = ecx = edx = 0;
862 __get_cpuid(0x6, &eax, &ebx, &ecx, &edx);
863 do_dts = eax & (1 << 0);
864 do_ptm = eax & (1 << 6);
865 if (!(ecx & (1 << 0))) {
866 ERROR("turbostat plugin: Unsupported CPU (No APERF)");
867 return -1;
868 }
870 /*
871 * Enable or disable C states depending on the model and family
872 */
873 if (family == 6) {
874 switch (model) {
875 /* Atom (partial) */
876 case 0x27:
877 do_smi = 0;
878 do_core_cstate = 0;
879 do_pkg_cstate = (1 << 2) | (1 << 4) | (1 << 6);
880 break;
881 /* Silvermont */
882 case 0x37: /* BYT */
883 case 0x4D: /* AVN */
884 do_smi = 1;
885 do_core_cstate = (1 << 1) | (1 << 6);
886 do_pkg_cstate = (1 << 6);
887 break;
888 /* Nehalem */
889 case 0x1A: /* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
890 case 0x1E: /* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
891 case 0x1F: /* Core i7 and i5 Processor - Nehalem */
892 case 0x2E: /* Nehalem-EX Xeon - Beckton */
893 do_smi = 1;
894 do_core_cstate = (1 << 3) | (1 << 6);
895 do_pkg_cstate = (1 << 3) | (1 << 6) | (1 << 7);
896 break;
897 /* Westmere */
898 case 0x25: /* Westmere Client - Clarkdale, Arrandale */
899 case 0x2C: /* Westmere EP - Gulftown */
900 case 0x2F: /* Westmere-EX Xeon - Eagleton */
901 do_smi = 1;
902 do_core_cstate = (1 << 3) | (1 << 6);
903 do_pkg_cstate = (1 << 3) | (1 << 6) | (1 << 7);
904 break;
905 /* Sandy Bridge */
906 case 0x2A: /* SNB */
907 case 0x2D: /* SNB Xeon */
908 do_smi = 1;
909 do_core_cstate = (1 << 3) | (1 << 6) | (1 << 7);
910 do_pkg_cstate = (1 << 2) | (1 << 3) | (1 << 6) | (1 << 7);
911 break;
912 /* Ivy Bridge */
913 case 0x3A: /* IVB */
914 case 0x3E: /* IVB Xeon */
915 do_smi = 1;
916 do_core_cstate = (1 << 3) | (1 << 6) | (1 << 7);
917 do_pkg_cstate = (1 << 2) | (1 << 3) | (1 << 6) | (1 << 7);
918 break;
919 /* Haswell Bridge */
920 case 0x3C: /* HSW */
921 case 0x3F: /* HSW */
922 case 0x46: /* HSW */
923 do_smi = 1;
924 do_core_cstate = (1 << 3) | (1 << 6) | (1 << 7);
925 do_pkg_cstate = (1 << 2) | (1 << 3) | (1 << 6) | (1 << 7);
926 break;
927 case 0x45: /* HSW */
928 do_smi = 1;
929 do_core_cstate = (1 << 3) | (1 << 6) | (1 << 7);
930 do_pkg_cstate = (1 << 2) | (1 << 3) | (1 << 6) | (1 << 7) | (1 << 8) | (1 << 9) | (1 << 10);
931 break;
932 /* Broadwel */
933 case 0x4F: /* BDW */
934 case 0x56: /* BDX-DE */
935 do_smi = 1;
936 do_core_cstate = (1 << 3) | (1 << 6) | (1 << 7);
937 do_pkg_cstate = (1 << 2) | (1 << 3) | (1 << 6) | (1 << 7);
938 break;
939 case 0x3D: /* BDW */
940 do_smi = 1;
941 do_core_cstate = (1 << 3) | (1 << 6) | (1 << 7);
942 do_pkg_cstate = (1 << 2) | (1 << 3) | (1 << 6) | (1 << 7) | (1 << 8) | (1 << 9) | (1 << 10);
943 break;
944 default:
945 do_smi = 0;
946 do_core_cstate = 0;
947 do_pkg_cstate = 0;
948 break;
949 }
950 switch (model) {
951 case 0x2A: /* SNB */
952 case 0x3A: /* IVB */
953 case 0x3C: /* HSW */
954 case 0x45: /* HSW */
955 case 0x46: /* HSW */
956 case 0x3D: /* BDW */
957 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_GFX;
958 break;
959 case 0x3F: /* HSX */
960 case 0x4F: /* BDX */
961 case 0x56: /* BDX-DE */
962 do_rapl = RAPL_PKG | RAPL_DRAM ;
963 break;
964 case 0x2D: /* SNB Xeon */
965 case 0x3E: /* IVB Xeon */
966 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_DRAM;
967 break;
968 case 0x37: /* BYT */
969 case 0x4D: /* AVN */
970 do_rapl = RAPL_PKG | RAPL_CORES;
971 break;
972 default:
973 do_rapl = 0;
974 }
975 } else {
976 ERROR("turbostat plugin: Unsupported CPU (family: %#x, "
977 "model: %#x)", family, model);
978 return -1;
979 }
981 /* Override detected values with configuration */
982 if (apply_config_core_cstate)
983 do_core_cstate = config_core_cstate;
984 if (apply_config_pkg_cstate)
985 do_pkg_cstate = config_pkg_cstate;
986 if (apply_config_smi)
987 do_smi = config_smi;
988 if (apply_config_dts)
989 do_dts = config_dts;
990 if (apply_config_ptm)
991 do_ptm = config_ptm;
992 if (apply_config_rapl)
993 do_rapl = config_rapl;
995 if (do_rapl) {
996 unsigned long long msr;
997 if (get_msr(0, MSR_RAPL_POWER_UNIT, &msr))
998 return 0;
1000 if (model == 0x37)
1001 rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
1002 else
1003 rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
1004 }
1006 return 0;
1007 }
1010 /********************
1011 * Topology Probing *
1012 ********************/
1014 /*
1015 * Read a single int from a file.
1016 */
1017 static int __attribute__ ((format(printf,1,2)))
1018 parse_int_file(const char *fmt, ...)
1019 {
1020 va_list args;
1021 char path[PATH_MAX];
1022 FILE *filep;
1023 int len, value;
1025 va_start(args, fmt);
1026 len = vsnprintf(path, sizeof(path), fmt, args);
1027 va_end(args);
1028 if (len < 0 || len >= PATH_MAX) {
1029 ERROR("turbostat plugin: path truncated: '%s'", path);
1030 return -1;
1031 }
1033 filep = fopen(path, "r");
1034 if (!filep) {
1035 ERROR("turbostat plugin: Failed to open '%s'", path);
1036 return -1;
1037 }
1038 if (fscanf(filep, "%d", &value) != 1) {
1039 ERROR("turbostat plugin: Failed to parse number from '%s'", path);
1040 fclose(filep);
1041 return -1;
1042 }
1043 fclose(filep);
1044 return value;
1045 }
1047 static int
1048 get_threads_on_core(unsigned int cpu)
1049 {
1050 char path[80];
1051 FILE *filep;
1052 int sib1, sib2;
1053 int matches;
1054 char character;
1056 ssnprintf(path, sizeof(path), "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
1057 filep = fopen(path, "r");
1058 if (!filep) {
1059 ERROR("turbostat plugin: Failed to open '%s'", path);
1060 return -1;
1061 }
1062 /*
1063 * file format:
1064 * if a pair of number with a character between: 2 siblings (eg. 1-2, or 1,4)
1065 * otherwinse 1 sibling (self).
1066 */
1067 matches = fscanf(filep, "%d%c%d\n", &sib1, &character, &sib2);
1069 fclose(filep);
1071 if (matches == 3)
1072 return 2;
1073 else
1074 return 1;
1075 }
1077 /*
1078 * run func(cpu) on every cpu in /proc/stat
1079 * return max_cpu number
1080 */
1081 static int __attribute__((warn_unused_result))
1082 for_all_proc_cpus(int (func)(unsigned int))
1083 {
1084 FILE *fp;
1085 unsigned int cpu_num;
1086 int retval;
1088 fp = fopen("/proc/stat", "r");
1089 if (!fp) {
1090 ERROR("turbostat plugin: Failed to open /proc/stat");
1091 return -1;
1092 }
1094 retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
1095 if (retval != 0) {
1096 ERROR("turbostat plugin: Failed to parse /proc/stat");
1097 fclose(fp);
1098 return -1;
1099 }
1101 while (1) {
1102 retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
1103 if (retval != 1)
1104 break;
1106 retval = func(cpu_num);
1107 if (retval) {
1108 fclose(fp);
1109 return(retval);
1110 }
1111 }
1112 fclose(fp);
1113 return 0;
1114 }
1116 /*
1117 * Update the stored topology.max_cpu_id
1118 */
1119 static int
1120 update_max_cpu_id(unsigned int cpu)
1121 {
1122 if (topology.max_cpu_id < cpu)
1123 topology.max_cpu_id = cpu;
1124 return 0;
1125 }
1127 static int
1128 mark_cpu_present(unsigned int cpu)
1129 {
1130 CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
1131 return 0;
1132 }
1134 static int __attribute__((warn_unused_result))
1135 allocate_cpu_set(cpu_set_t ** set, size_t * size) {
1136 *set = CPU_ALLOC(topology.max_cpu_id + 1);
1137 if (*set == NULL) {
1138 ERROR("turbostat plugin: Unable to allocate CPU state");
1139 return -1;
1140 }
1141 *size = CPU_ALLOC_SIZE(topology.max_cpu_id + 1);
1142 CPU_ZERO_S(*size, *set);
1143 return 0;
1144 }
1146 /*
1147 * Build a local representation of the cpu distribution
1148 */
1149 static int __attribute__((warn_unused_result))
1150 topology_probe(void)
1151 {
1152 unsigned int i;
1153 int ret;
1154 unsigned int max_package_id, max_core_id, max_threads;
1155 max_package_id = max_core_id = max_threads = 0;
1157 /* Clean topology */
1158 free(topology.cpus);
1159 memset(&topology, 0, sizeof(topology));
1161 ret = for_all_proc_cpus(update_max_cpu_id);
1162 if (ret != 0)
1163 goto err;
1165 topology.cpus = calloc(1, (topology.max_cpu_id + 1) * sizeof(struct cpu_topology));
1166 if (topology.cpus == NULL) {
1167 ERROR("turbostat plugin: Unable to allocate memory for CPU topology");
1168 return -1;
1169 }
1171 ret = allocate_cpu_set(&cpu_present_set, &cpu_present_setsize);
1172 if (ret != 0)
1173 goto err;
1174 ret = allocate_cpu_set(&cpu_affinity_set, &cpu_affinity_setsize);
1175 if (ret != 0)
1176 goto err;
1177 ret = allocate_cpu_set(&cpu_saved_affinity_set, &cpu_saved_affinity_setsize);
1178 if (ret != 0)
1179 goto err;
1181 ret = for_all_proc_cpus(mark_cpu_present);
1182 if (ret != 0)
1183 goto err;
1185 /*
1186 * For online cpus
1187 * find max_core_id, max_package_id
1188 */
1189 for (i = 0; i <= topology.max_cpu_id; ++i) {
1190 unsigned int num_threads;
1191 struct cpu_topology *cpu = &topology.cpus[i];
1193 if (cpu_is_not_present(i)) {
1194 WARNING("turbostat plugin: cpu%d NOT PRESENT", i);
1195 continue;
1196 }
1198 ret = parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", i);
1199 if (ret < 0)
1200 goto err;
1201 else
1202 cpu->package_id = (unsigned int) ret;
1203 if (cpu->package_id > max_package_id)
1204 max_package_id = cpu->package_id;
1206 ret = parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", i);
1207 if (ret < 0)
1208 goto err;
1209 else
1210 cpu->core_id = (unsigned int) ret;
1211 if (cpu->core_id > max_core_id)
1212 max_core_id = cpu->core_id;
1213 ret = parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", i);
1214 if (ret < 0)
1215 goto err;
1216 else if ((unsigned int) ret == i)
1217 cpu->first_core_in_package = 1;
1219 ret = get_threads_on_core(i);
1220 if (ret < 0)
1221 goto err;
1222 else
1223 num_threads = (unsigned int) ret;
1224 if (num_threads > max_threads)
1225 max_threads = num_threads;
1226 ret = parse_int_file("/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", i);
1227 if (ret < 0)
1228 goto err;
1229 else if ((unsigned int) ret == i)
1230 cpu->first_thread_in_core = 1;
1232 DEBUG("turbostat plugin: cpu %d pkg %d core %d\n",
1233 i, cpu->package_id, cpu->core_id);
1234 }
1235 /* Num is max + 1 (need to count 0) */
1236 topology.num_packages = max_package_id + 1;
1237 topology.num_cores = max_core_id + 1;
1238 topology.num_threads = max_threads;
1240 return 0;
1241 err:
1242 free(topology.cpus);
1243 return ret;
1244 }
1247 /************************
1248 * Main alloc/init/free *
1249 ************************/
1251 static int
1252 allocate_counters(struct thread_data **threads, struct core_data **cores, struct pkg_data **packages)
1253 {
1254 unsigned int i;
1255 unsigned int total_threads, total_cores;
1257 if ((topology.num_threads == 0)
1258 || (topology.num_cores == 0)
1259 || (topology.num_packages == 0))
1260 {
1261 ERROR ("turbostat plugin: Invalid topology: %u threads, %u cores, %u packages",
1262 topology.num_threads, topology.num_cores, topology.num_packages);
1263 return -1;
1264 }
1266 total_threads = topology.num_threads * topology.num_cores * topology.num_packages;
1267 *threads = calloc(total_threads, sizeof(struct thread_data));
1268 if (*threads == NULL)
1269 {
1270 ERROR ("turbostat plugin: calloc failed");
1271 return -1;
1272 }
1274 for (i = 0; i < total_threads; ++i)
1275 (*threads)[i].cpu_id = topology.max_cpu_id + 1;
1277 total_cores = topology.num_cores * topology.num_packages;
1278 *cores = calloc(total_cores, sizeof(struct core_data));
1279 if (*cores == NULL)
1280 {
1281 ERROR ("turbostat plugin: calloc failed");
1282 sfree (threads);
1283 return -1;
1284 }
1286 *packages = calloc(topology.num_packages, sizeof(struct pkg_data));
1287 if (*packages == NULL)
1288 {
1289 ERROR ("turbostat plugin: calloc failed");
1290 sfree (cores);
1291 sfree (threads);
1292 return -1;
1293 }
1295 return 0;
1296 }
1298 static void
1299 init_counter(struct thread_data *thread_base, struct core_data *core_base,
1300 struct pkg_data *pkg_base, unsigned int cpu_id)
1301 {
1302 struct thread_data *t;
1303 struct core_data *c;
1304 struct pkg_data *p;
1305 struct cpu_topology *cpu = &topology.cpus[cpu_id];
1307 t = GET_THREAD(thread_base, !(cpu->first_thread_in_core), cpu->core_id, cpu->package_id);
1308 c = GET_CORE(core_base, cpu->core_id, cpu->package_id);
1309 p = GET_PKG(pkg_base, cpu->package_id);
1311 t->cpu_id = cpu_id;
1312 if (cpu->first_thread_in_core)
1313 t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
1314 if (cpu->first_core_in_package)
1315 t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
1317 c->core_id = cpu->core_id;
1318 p->package_id = cpu->package_id;
1319 }
1321 static void
1322 initialize_counters(void)
1323 {
1324 unsigned int cpu_id;
1326 for (cpu_id = 0; cpu_id <= topology.max_cpu_id; ++cpu_id) {
1327 if (cpu_is_not_present(cpu_id))
1328 continue;
1329 init_counter(EVEN_COUNTERS, cpu_id);
1330 init_counter(ODD_COUNTERS, cpu_id);
1331 init_counter(DELTA_COUNTERS, cpu_id);
1332 }
1333 }
1337 static void
1338 free_all_buffers(void)
1339 {
1340 allocated = 0;
1341 initialized = 0;
1343 CPU_FREE(cpu_present_set);
1344 cpu_present_set = NULL;
1345 cpu_present_setsize = 0;
1347 CPU_FREE(cpu_affinity_set);
1348 cpu_affinity_set = NULL;
1349 cpu_affinity_setsize = 0;
1351 CPU_FREE(cpu_saved_affinity_set);
1352 cpu_saved_affinity_set = NULL;
1353 cpu_saved_affinity_setsize = 0;
1355 free(thread_even);
1356 free(core_even);
1357 free(package_even);
1359 thread_even = NULL;
1360 core_even = NULL;
1361 package_even = NULL;
1363 free(thread_odd);
1364 free(core_odd);
1365 free(package_odd);
1367 thread_odd = NULL;
1368 core_odd = NULL;
1369 package_odd = NULL;
1371 free(thread_delta);
1372 free(core_delta);
1373 free(package_delta);
1375 thread_delta = NULL;
1376 core_delta = NULL;
1377 package_delta = NULL;
1378 }
1381 /**********************
1382 * Collectd functions *
1383 **********************/
1385 #define DO_OR_GOTO_ERR(something) \
1386 do { \
1387 ret = (something); \
1388 if (ret < 0) \
1389 goto err; \
1390 } while (0)
1392 static int setup_all_buffers(void)
1393 {
1394 int ret;
1396 DO_OR_GOTO_ERR(topology_probe());
1397 DO_OR_GOTO_ERR(allocate_counters(&thread_even, &core_even, &package_even));
1398 DO_OR_GOTO_ERR(allocate_counters(&thread_odd, &core_odd, &package_odd));
1399 DO_OR_GOTO_ERR(allocate_counters(&thread_delta, &core_delta, &package_delta));
1400 initialize_counters();
1401 DO_OR_GOTO_ERR(for_all_cpus(set_temperature_target, EVEN_COUNTERS));
1402 DO_OR_GOTO_ERR(for_all_cpus(set_temperature_target, ODD_COUNTERS));
1404 allocated = 1;
1405 return 0;
1406 err:
1407 free_all_buffers();
1408 return ret;
1409 }
1411 static int
1412 turbostat_read(void)
1413 {
1414 int ret;
1416 if (!allocated) {
1417 if ((ret = setup_all_buffers()) < 0)
1418 return ret;
1419 }
1421 if (for_all_proc_cpus(cpu_is_not_present)) {
1422 free_all_buffers();
1423 if ((ret = setup_all_buffers()) < 0)
1424 return ret;
1425 if (for_all_proc_cpus(cpu_is_not_present)) {
1426 ERROR("turbostat plugin: CPU appeared just after "
1427 "initialization");
1428 return -1;
1429 }
1430 }
1432 /* Saving the scheduling affinity, as it will be modified by get_counters */
1433 if (sched_getaffinity(0, cpu_saved_affinity_setsize, cpu_saved_affinity_set) != 0) {
1434 ERROR("turbostat plugin: Unable to save the CPU affinity");
1435 return -1;
1436 }
1438 if (!initialized) {
1439 if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
1440 goto out;
1441 time_even = cdtime();
1442 is_even = 1;
1443 initialized = 1;
1444 ret = 0;
1445 goto out;
1446 }
1448 if (is_even) {
1449 if ((ret = for_all_cpus(get_counters, ODD_COUNTERS)) < 0)
1450 goto out;
1451 time_odd = cdtime();
1452 is_even = 0;
1453 time_delta = time_odd - time_even;
1454 if ((ret = for_all_cpus_delta(ODD_COUNTERS, EVEN_COUNTERS)) < 0)
1455 goto out;
1456 if ((ret = for_all_cpus(submit_counters, DELTA_COUNTERS)) < 0)
1457 goto out;
1458 } else {
1459 if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
1460 goto out;
1461 time_even = cdtime();
1462 is_even = 1;
1463 time_delta = time_even - time_odd;
1464 if ((ret = for_all_cpus_delta(EVEN_COUNTERS, ODD_COUNTERS)) < 0)
1465 goto out;
1466 if ((ret = for_all_cpus(submit_counters, DELTA_COUNTERS)) < 0)
1467 goto out;
1468 }
1469 ret = 0;
1470 out:
1471 /*
1472 * Let's restore the affinity
1473 * This might fail if the number of CPU changed, but we can't do anything in that case..
1474 */
1475 (void)sched_setaffinity(0, cpu_saved_affinity_setsize, cpu_saved_affinity_set);
1476 return ret;
1477 }
1479 static int
1480 check_permissions(void)
1481 {
1482 #ifdef HAVE_SYS_CAPABILITY_H
1483 struct __user_cap_header_struct cap_header_data;
1484 cap_user_header_t cap_header = &cap_header_data;
1485 struct __user_cap_data_struct cap_data_data;
1486 cap_user_data_t cap_data = &cap_data_data;
1487 int ret = 0;
1488 #endif /* HAVE_SYS_CAPABILITY_H */
1490 if (getuid() == 0) {
1491 /* We have everything we need */
1492 return 0;
1493 #ifndef HAVE_SYS_CAPABILITY_H
1494 } else {
1495 ERROR("turbostat plugin: Initialization failed: this plugin "
1496 "requires collectd to run as root");
1497 return -1;
1498 }
1499 #else /* HAVE_SYS_CAPABILITY_H */
1500 }
1502 /* check for CAP_SYS_RAWIO */
1503 cap_header->pid = getpid();
1504 cap_header->version = _LINUX_CAPABILITY_VERSION;
1505 if (capget(cap_header, cap_data) < 0) {
1506 ERROR("turbostat plugin: capget failed");
1507 return -1;
1508 }
1510 if ((cap_data->effective & (1 << CAP_SYS_RAWIO)) == 0) {
1511 WARNING("turbostat plugin: Collectd doesn't have the "
1512 "CAP_SYS_RAWIO capability. If you don't want to run "
1513 "collectd as root, try running \"setcap "
1514 "cap_sys_rawio=ep\" on collectd binary");
1515 ret = -1;
1516 }
1518 if (euidaccess("/dev/cpu/0/msr", R_OK)) {
1519 WARNING("turbostat plugin: Collectd cannot open "
1520 "/dev/cpu/0/msr. If you don't want to run collectd as "
1521 "root, you need to change the ownership (chown) and "
1522 "permissions on /dev/cpu/*/msr to allow such access");
1523 ret = -1;
1524 }
1526 if (ret != 0)
1527 ERROR("turbostat plugin: Initialization failed: this plugin "
1528 "requires collectd to either to run as root or give "
1529 "collectd a special capability (CAP_SYS_RAWIO) and read "
1530 "access to /dev/cpu/*/msr (see previous warnings)");
1531 return ret;
1532 #endif /* HAVE_SYS_CAPABILITY_H */
1533 }
1535 static int
1536 turbostat_init(void)
1537 {
1538 struct stat sb;
1539 int ret;
1541 if (stat("/dev/cpu/0/msr", &sb)) {
1542 ERROR("turbostat plugin: Initialization failed: /dev/cpu/0/msr "
1543 "does not exist while the CPU supports MSR. You may be "
1544 "missing the corresponding kernel module, please try '# "
1545 "modprobe msr'");
1546 return -1;
1547 }
1549 DO_OR_GOTO_ERR(check_permissions());
1551 DO_OR_GOTO_ERR(probe_cpu());
1553 DO_OR_GOTO_ERR(setup_all_buffers());
1555 plugin_register_read(PLUGIN_NAME, turbostat_read);
1557 return 0;
1558 err:
1559 free_all_buffers();
1560 return ret;
1561 }
1563 static int
1564 turbostat_config(const char *key, const char *value)
1565 {
1566 long unsigned int tmp_val;
1567 char *end;
1569 if (strcasecmp("CoreCstates", key) == 0) {
1570 tmp_val = strtoul(value, &end, 0);
1571 if (*end != '\0' || tmp_val > UINT_MAX) {
1572 ERROR("turbostat plugin: Invalid CoreCstates '%s'",
1573 value);
1574 return -1;
1575 }
1576 config_core_cstate = (unsigned int) tmp_val;
1577 apply_config_core_cstate = 1;
1578 } else if (strcasecmp("PackageCstates", key) == 0) {
1579 tmp_val = strtoul(value, &end, 0);
1580 if (*end != '\0' || tmp_val > UINT_MAX) {
1581 ERROR("turbostat plugin: Invalid PackageCstates '%s'",
1582 value);
1583 return -1;
1584 }
1585 config_pkg_cstate = (unsigned int) tmp_val;
1586 apply_config_pkg_cstate = 1;
1587 } else if (strcasecmp("SystemManagementInterrupt", key) == 0) {
1588 config_smi = IS_TRUE(value);
1589 apply_config_smi = 1;
1590 } else if (strcasecmp("DigitalTemperatureSensor", key) == 0) {
1591 config_dts = IS_TRUE(value);
1592 apply_config_dts = 1;
1593 } else if (strcasecmp("PackageThermalManagement", key) == 0) {
1594 config_ptm = IS_TRUE(value);
1595 apply_config_ptm = 1;
1596 } else if (strcasecmp("RunningAveragePowerLimit", key) == 0) {
1597 tmp_val = strtoul(value, &end, 0);
1598 if (*end != '\0' || tmp_val > UINT_MAX) {
1599 ERROR("turbostat plugin: Invalid RunningAveragePowerLimit '%s'",
1600 value);
1601 return -1;
1602 }
1603 config_rapl = (unsigned int) tmp_val;
1604 apply_config_rapl = 1;
1605 } else if (strcasecmp("TCCActivationTemp", key) == 0) {
1606 tmp_val = strtoul(value, &end, 0);
1607 if (*end != '\0' || tmp_val > UINT_MAX) {
1608 ERROR("turbostat plugin: Invalid TCCActivationTemp '%s'",
1609 value);
1610 return -1;
1611 }
1612 tcc_activation_temp = (unsigned int) tmp_val;
1613 } else {
1614 ERROR("turbostat plugin: Invalid configuration option '%s'",
1615 key);
1616 return -1;
1617 }
1618 return 0;
1619 }
1621 void module_register(void)
1622 {
1623 plugin_register_init(PLUGIN_NAME, turbostat_init);
1624 plugin_register_config(PLUGIN_NAME, turbostat_config, config_keys, config_keys_num);
1625 }