Merge tag 'nfs-for-3.14-3' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
[platform/adaptation/renesas_rcar/renesas_kernel.git] / tools / power / x86 / turbostat / turbostat.c
1 /*
2  * turbostat -- show CPU frequency and C-state residency
3  * on modern Intel turbo-capable processors.
4  *
5  * Copyright (c) 2013 Intel Corporation.
6  * Len Brown <len.brown@intel.com>
7  *
8  * This program is free software; you can redistribute it and/or modify it
9  * under the terms and conditions of the GNU General Public License,
10  * version 2, as published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope it will be useful, but WITHOUT
13  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
15  * more details.
16  *
17  * You should have received a copy of the GNU General Public License along with
18  * this program; if not, write to the Free Software Foundation, Inc.,
19  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20  */
21
22 #define _GNU_SOURCE
23 #include MSRHEADER
24 #include <stdarg.h>
25 #include <stdio.h>
26 #include <err.h>
27 #include <unistd.h>
28 #include <sys/types.h>
29 #include <sys/wait.h>
30 #include <sys/stat.h>
31 #include <sys/resource.h>
32 #include <fcntl.h>
33 #include <signal.h>
34 #include <sys/time.h>
35 #include <stdlib.h>
36 #include <dirent.h>
37 #include <string.h>
38 #include <ctype.h>
39 #include <sched.h>
40 #include <cpuid.h>
41
42 char *proc_stat = "/proc/stat";
43 unsigned int interval_sec = 5;  /* set with -i interval_sec */
44 unsigned int verbose;           /* set with -v */
45 unsigned int rapl_verbose;      /* set with -R */
46 unsigned int rapl_joules;       /* set with -J */
47 unsigned int thermal_verbose;   /* set with -T */
48 unsigned int summary_only;      /* set with -S */
49 unsigned int dump_only;         /* set with -s */
50 unsigned int skip_c0;
51 unsigned int skip_c1;
52 unsigned int do_nhm_cstates;
53 unsigned int do_snb_cstates;
54 unsigned int do_c8_c9_c10;
55 unsigned int do_slm_cstates;
56 unsigned int use_c1_residency_msr;
57 unsigned int has_aperf;
58 unsigned int has_epb;
59 unsigned int units = 1000000000;        /* Ghz etc */
60 unsigned int genuine_intel;
61 unsigned int has_invariant_tsc;
62 unsigned int do_nehalem_platform_info;
63 unsigned int do_nehalem_turbo_ratio_limit;
64 unsigned int do_ivt_turbo_ratio_limit;
65 unsigned int extra_msr_offset32;
66 unsigned int extra_msr_offset64;
67 unsigned int extra_delta_offset32;
68 unsigned int extra_delta_offset64;
69 int do_smi;
70 double bclk;
71 unsigned int show_pkg;
72 unsigned int show_core;
73 unsigned int show_cpu;
74 unsigned int show_pkg_only;
75 unsigned int show_core_only;
76 char *output_buffer, *outp;
77 unsigned int do_rapl;
78 unsigned int do_dts;
79 unsigned int do_ptm;
80 unsigned int tcc_activation_temp;
81 unsigned int tcc_activation_temp_override;
82 double rapl_power_units, rapl_energy_units, rapl_time_units;
83 double rapl_joule_counter_range;
84
85 #define RAPL_PKG                (1 << 0)
86                                         /* 0x610 MSR_PKG_POWER_LIMIT */
87                                         /* 0x611 MSR_PKG_ENERGY_STATUS */
88 #define RAPL_PKG_PERF_STATUS    (1 << 1)
89                                         /* 0x613 MSR_PKG_PERF_STATUS */
90 #define RAPL_PKG_POWER_INFO     (1 << 2)
91                                         /* 0x614 MSR_PKG_POWER_INFO */
92
93 #define RAPL_DRAM               (1 << 3)
94                                         /* 0x618 MSR_DRAM_POWER_LIMIT */
95                                         /* 0x619 MSR_DRAM_ENERGY_STATUS */
96                                         /* 0x61c MSR_DRAM_POWER_INFO */
97 #define RAPL_DRAM_PERF_STATUS   (1 << 4)
98                                         /* 0x61b MSR_DRAM_PERF_STATUS */
99
100 #define RAPL_CORES              (1 << 5)
101                                         /* 0x638 MSR_PP0_POWER_LIMIT */
102                                         /* 0x639 MSR_PP0_ENERGY_STATUS */
103 #define RAPL_CORE_POLICY        (1 << 6)
104                                         /* 0x63a MSR_PP0_POLICY */
105
106
107 #define RAPL_GFX                (1 << 7)
108                                         /* 0x640 MSR_PP1_POWER_LIMIT */
109                                         /* 0x641 MSR_PP1_ENERGY_STATUS */
110                                         /* 0x642 MSR_PP1_POLICY */
111 #define TJMAX_DEFAULT   100
112
113 #define MAX(a, b) ((a) > (b) ? (a) : (b))
114
115 int aperf_mperf_unstable;
116 int backwards_count;
117 char *progname;
118
119 cpu_set_t *cpu_present_set, *cpu_affinity_set;
120 size_t cpu_present_setsize, cpu_affinity_setsize;
121
122 struct thread_data {
123         unsigned long long tsc;
124         unsigned long long aperf;
125         unsigned long long mperf;
126         unsigned long long c1;
127         unsigned long long extra_msr64;
128         unsigned long long extra_delta64;
129         unsigned long long extra_msr32;
130         unsigned long long extra_delta32;
131         unsigned int smi_count;
132         unsigned int cpu_id;
133         unsigned int flags;
134 #define CPU_IS_FIRST_THREAD_IN_CORE     0x2
135 #define CPU_IS_FIRST_CORE_IN_PACKAGE    0x4
136 } *thread_even, *thread_odd;
137
138 struct core_data {
139         unsigned long long c3;
140         unsigned long long c6;
141         unsigned long long c7;
142         unsigned int core_temp_c;
143         unsigned int core_id;
144 } *core_even, *core_odd;
145
146 struct pkg_data {
147         unsigned long long pc2;
148         unsigned long long pc3;
149         unsigned long long pc6;
150         unsigned long long pc7;
151         unsigned long long pc8;
152         unsigned long long pc9;
153         unsigned long long pc10;
154         unsigned int package_id;
155         unsigned int energy_pkg;        /* MSR_PKG_ENERGY_STATUS */
156         unsigned int energy_dram;       /* MSR_DRAM_ENERGY_STATUS */
157         unsigned int energy_cores;      /* MSR_PP0_ENERGY_STATUS */
158         unsigned int energy_gfx;        /* MSR_PP1_ENERGY_STATUS */
159         unsigned int rapl_pkg_perf_status;      /* MSR_PKG_PERF_STATUS */
160         unsigned int rapl_dram_perf_status;     /* MSR_DRAM_PERF_STATUS */
161         unsigned int pkg_temp_c;
162
163 } *package_even, *package_odd;
164
165 #define ODD_COUNTERS thread_odd, core_odd, package_odd
166 #define EVEN_COUNTERS thread_even, core_even, package_even
167
168 #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \
169         (thread_base + (pkg_no) * topo.num_cores_per_pkg * \
170                 topo.num_threads_per_core + \
171                 (core_no) * topo.num_threads_per_core + (thread_no))
172 #define GET_CORE(core_base, core_no, pkg_no) \
173         (core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no))
174 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
175
176 struct system_summary {
177         struct thread_data threads;
178         struct core_data cores;
179         struct pkg_data packages;
180 } sum, average;
181
182
183 struct topo_params {
184         int num_packages;
185         int num_cpus;
186         int num_cores;
187         int max_cpu_num;
188         int num_cores_per_pkg;
189         int num_threads_per_core;
190 } topo;
191
192 struct timeval tv_even, tv_odd, tv_delta;
193
194 void setup_all_buffers(void);
195
196 int cpu_is_not_present(int cpu)
197 {
198         return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
199 }
200 /*
201  * run func(thread, core, package) in topology order
202  * skip non-present cpus
203  */
204
205 int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
206         struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
207 {
208         int retval, pkg_no, core_no, thread_no;
209
210         for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
211                 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
212                         for (thread_no = 0; thread_no <
213                                 topo.num_threads_per_core; ++thread_no) {
214                                 struct thread_data *t;
215                                 struct core_data *c;
216                                 struct pkg_data *p;
217
218                                 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
219
220                                 if (cpu_is_not_present(t->cpu_id))
221                                         continue;
222
223                                 c = GET_CORE(core_base, core_no, pkg_no);
224                                 p = GET_PKG(pkg_base, pkg_no);
225
226                                 retval = func(t, c, p);
227                                 if (retval)
228                                         return retval;
229                         }
230                 }
231         }
232         return 0;
233 }
234
235 int cpu_migrate(int cpu)
236 {
237         CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
238         CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
239         if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
240                 return -1;
241         else
242                 return 0;
243 }
244
245 int get_msr(int cpu, off_t offset, unsigned long long *msr)
246 {
247         ssize_t retval;
248         char pathname[32];
249         int fd;
250
251         sprintf(pathname, "/dev/cpu/%d/msr", cpu);
252         fd = open(pathname, O_RDONLY);
253         if (fd < 0)
254                 return -1;
255
256         retval = pread(fd, msr, sizeof *msr, offset);
257         close(fd);
258
259         if (retval != sizeof *msr) {
260                 fprintf(stderr, "%s offset 0x%llx read failed\n", pathname, (unsigned long long)offset);
261                 return -1;
262         }
263
264         return 0;
265 }
266
267 void print_header(void)
268 {
269         if (show_pkg)
270                 outp += sprintf(outp, "pk");
271         if (show_pkg)
272                 outp += sprintf(outp, " ");
273         if (show_core)
274                 outp += sprintf(outp, "cor");
275         if (show_cpu)
276                 outp += sprintf(outp, " CPU");
277         if (show_pkg || show_core || show_cpu)
278                 outp += sprintf(outp, " ");
279         if (do_nhm_cstates)
280                 outp += sprintf(outp, "   %%c0");
281         if (has_aperf)
282                 outp += sprintf(outp, "  GHz");
283         outp += sprintf(outp, "  TSC");
284         if (do_smi)
285                 outp += sprintf(outp, " SMI");
286         if (extra_delta_offset32)
287                 outp += sprintf(outp, "  count 0x%03X", extra_delta_offset32);
288         if (extra_delta_offset64)
289                 outp += sprintf(outp, "  COUNT 0x%03X", extra_delta_offset64);
290         if (extra_msr_offset32)
291                 outp += sprintf(outp, "   MSR 0x%03X", extra_msr_offset32);
292         if (extra_msr_offset64)
293                 outp += sprintf(outp, "           MSR 0x%03X", extra_msr_offset64);
294         if (do_nhm_cstates)
295                 outp += sprintf(outp, "    %%c1");
296         if (do_nhm_cstates && !do_slm_cstates)
297                 outp += sprintf(outp, "    %%c3");
298         if (do_nhm_cstates)
299                 outp += sprintf(outp, "    %%c6");
300         if (do_snb_cstates)
301                 outp += sprintf(outp, "    %%c7");
302
303         if (do_dts)
304                 outp += sprintf(outp, " CTMP");
305         if (do_ptm)
306                 outp += sprintf(outp, " PTMP");
307
308         if (do_snb_cstates)
309                 outp += sprintf(outp, "   %%pc2");
310         if (do_nhm_cstates && !do_slm_cstates)
311                 outp += sprintf(outp, "   %%pc3");
312         if (do_nhm_cstates && !do_slm_cstates)
313                 outp += sprintf(outp, "   %%pc6");
314         if (do_snb_cstates)
315                 outp += sprintf(outp, "   %%pc7");
316         if (do_c8_c9_c10) {
317                 outp += sprintf(outp, "   %%pc8");
318                 outp += sprintf(outp, "   %%pc9");
319                 outp += sprintf(outp, "  %%pc10");
320         }
321
322         if (do_rapl && !rapl_joules) {
323                 if (do_rapl & RAPL_PKG)
324                         outp += sprintf(outp, "  Pkg_W");
325                 if (do_rapl & RAPL_CORES)
326                         outp += sprintf(outp, "  Cor_W");
327                 if (do_rapl & RAPL_GFX)
328                         outp += sprintf(outp, " GFX_W");
329                 if (do_rapl & RAPL_DRAM)
330                         outp += sprintf(outp, " RAM_W");
331                 if (do_rapl & RAPL_PKG_PERF_STATUS)
332                         outp += sprintf(outp, " PKG_%%");
333                 if (do_rapl & RAPL_DRAM_PERF_STATUS)
334                         outp += sprintf(outp, " RAM_%%");
335         } else {
336                 if (do_rapl & RAPL_PKG)
337                         outp += sprintf(outp, "  Pkg_J");
338                 if (do_rapl & RAPL_CORES)
339                         outp += sprintf(outp, "  Cor_J");
340                 if (do_rapl & RAPL_GFX)
341                         outp += sprintf(outp, " GFX_J");
342                 if (do_rapl & RAPL_DRAM)
343                         outp += sprintf(outp, " RAM_W");
344                 if (do_rapl & RAPL_PKG_PERF_STATUS)
345                         outp += sprintf(outp, " PKG_%%");
346                 if (do_rapl & RAPL_DRAM_PERF_STATUS)
347                         outp += sprintf(outp, " RAM_%%");
348                 outp += sprintf(outp, " time");
349
350         }
351         outp += sprintf(outp, "\n");
352 }
353
354 int dump_counters(struct thread_data *t, struct core_data *c,
355         struct pkg_data *p)
356 {
357         outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p);
358
359         if (t) {
360                 outp += sprintf(outp, "CPU: %d flags 0x%x\n",
361                         t->cpu_id, t->flags);
362                 outp += sprintf(outp, "TSC: %016llX\n", t->tsc);
363                 outp += sprintf(outp, "aperf: %016llX\n", t->aperf);
364                 outp += sprintf(outp, "mperf: %016llX\n", t->mperf);
365                 outp += sprintf(outp, "c1: %016llX\n", t->c1);
366                 outp += sprintf(outp, "msr0x%x: %08llX\n",
367                         extra_delta_offset32, t->extra_delta32);
368                 outp += sprintf(outp, "msr0x%x: %016llX\n",
369                         extra_delta_offset64, t->extra_delta64);
370                 outp += sprintf(outp, "msr0x%x: %08llX\n",
371                         extra_msr_offset32, t->extra_msr32);
372                 outp += sprintf(outp, "msr0x%x: %016llX\n",
373                         extra_msr_offset64, t->extra_msr64);
374                 if (do_smi)
375                         outp += sprintf(outp, "SMI: %08X\n", t->smi_count);
376         }
377
378         if (c) {
379                 outp += sprintf(outp, "core: %d\n", c->core_id);
380                 outp += sprintf(outp, "c3: %016llX\n", c->c3);
381                 outp += sprintf(outp, "c6: %016llX\n", c->c6);
382                 outp += sprintf(outp, "c7: %016llX\n", c->c7);
383                 outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c);
384         }
385
386         if (p) {
387                 outp += sprintf(outp, "package: %d\n", p->package_id);
388                 outp += sprintf(outp, "pc2: %016llX\n", p->pc2);
389                 outp += sprintf(outp, "pc3: %016llX\n", p->pc3);
390                 outp += sprintf(outp, "pc6: %016llX\n", p->pc6);
391                 outp += sprintf(outp, "pc7: %016llX\n", p->pc7);
392                 outp += sprintf(outp, "pc8: %016llX\n", p->pc8);
393                 outp += sprintf(outp, "pc9: %016llX\n", p->pc9);
394                 outp += sprintf(outp, "pc10: %016llX\n", p->pc10);
395                 outp += sprintf(outp, "Joules PKG: %0X\n", p->energy_pkg);
396                 outp += sprintf(outp, "Joules COR: %0X\n", p->energy_cores);
397                 outp += sprintf(outp, "Joules GFX: %0X\n", p->energy_gfx);
398                 outp += sprintf(outp, "Joules RAM: %0X\n", p->energy_dram);
399                 outp += sprintf(outp, "Throttle PKG: %0X\n",
400                         p->rapl_pkg_perf_status);
401                 outp += sprintf(outp, "Throttle RAM: %0X\n",
402                         p->rapl_dram_perf_status);
403                 outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c);
404         }
405
406         outp += sprintf(outp, "\n");
407
408         return 0;
409 }
410
411 /*
412  * column formatting convention & formats
413  * package: "pk" 2 columns %2d
414  * core: "cor" 3 columns %3d
415  * CPU: "CPU" 3 columns %3d
416  * Pkg_W: %6.2
417  * Cor_W: %6.2
418  * GFX_W: %5.2
419  * RAM_W: %5.2
420  * GHz: "GHz" 3 columns %3.2
421  * TSC: "TSC" 3 columns %3.2
422  * SMI: "SMI" 4 columns %4d
423  * percentage " %pc3" %6.2
424  * Perf Status percentage: %5.2
425  * "CTMP" 4 columns %4d
426  */
427 int format_counters(struct thread_data *t, struct core_data *c,
428         struct pkg_data *p)
429 {
430         double interval_float;
431         char *fmt5, *fmt6;
432
433          /* if showing only 1st thread in core and this isn't one, bail out */
434         if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
435                 return 0;
436
437          /* if showing only 1st thread in pkg and this isn't one, bail out */
438         if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
439                 return 0;
440
441         interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
442
443         /* topo columns, print blanks on 1st (average) line */
444         if (t == &average.threads) {
445                 if (show_pkg)
446                         outp += sprintf(outp, "  ");
447                 if (show_pkg && show_core)
448                         outp += sprintf(outp, " ");
449                 if (show_core)
450                         outp += sprintf(outp, "   ");
451                 if (show_cpu)
452                         outp += sprintf(outp, " " "   ");
453         } else {
454                 if (show_pkg) {
455                         if (p)
456                                 outp += sprintf(outp, "%2d", p->package_id);
457                         else
458                                 outp += sprintf(outp, "  ");
459                 }
460                 if (show_pkg && show_core)
461                         outp += sprintf(outp, " ");
462                 if (show_core) {
463                         if (c)
464                                 outp += sprintf(outp, "%3d", c->core_id);
465                         else
466                                 outp += sprintf(outp, "   ");
467                 }
468                 if (show_cpu)
469                         outp += sprintf(outp, " %3d", t->cpu_id);
470         }
471         /* %c0 */
472         if (do_nhm_cstates) {
473                 if (show_pkg || show_core || show_cpu)
474                         outp += sprintf(outp, " ");
475                 if (!skip_c0)
476                         outp += sprintf(outp, "%6.2f", 100.0 * t->mperf/t->tsc);
477                 else
478                         outp += sprintf(outp, "  ****");
479         }
480
481         /* GHz */
482         if (has_aperf) {
483                 if (!aperf_mperf_unstable) {
484                         outp += sprintf(outp, " %3.2f",
485                                 1.0 * t->tsc / units * t->aperf /
486                                 t->mperf / interval_float);
487                 } else {
488                         if (t->aperf > t->tsc || t->mperf > t->tsc) {
489                                 outp += sprintf(outp, " ***");
490                         } else {
491                                 outp += sprintf(outp, "%3.1f*",
492                                         1.0 * t->tsc /
493                                         units * t->aperf /
494                                         t->mperf / interval_float);
495                         }
496                 }
497         }
498
499         /* TSC */
500         outp += sprintf(outp, "%5.2f", 1.0 * t->tsc/units/interval_float);
501
502         /* SMI */
503         if (do_smi)
504                 outp += sprintf(outp, "%4d", t->smi_count);
505
506         /* delta */
507         if (extra_delta_offset32)
508                 outp += sprintf(outp, "  %11llu", t->extra_delta32);
509
510         /* DELTA */
511         if (extra_delta_offset64)
512                 outp += sprintf(outp, "  %11llu", t->extra_delta64);
513         /* msr */
514         if (extra_msr_offset32)
515                 outp += sprintf(outp, "  0x%08llx", t->extra_msr32);
516
517         /* MSR */
518         if (extra_msr_offset64)
519                 outp += sprintf(outp, "  0x%016llx", t->extra_msr64);
520
521         if (do_nhm_cstates) {
522                 if (!skip_c1)
523                         outp += sprintf(outp, " %6.2f", 100.0 * t->c1/t->tsc);
524                 else
525                         outp += sprintf(outp, "  ****");
526         }
527
528         /* print per-core data only for 1st thread in core */
529         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
530                 goto done;
531
532         if (do_nhm_cstates && !do_slm_cstates)
533                 outp += sprintf(outp, " %6.2f", 100.0 * c->c3/t->tsc);
534         if (do_nhm_cstates)
535                 outp += sprintf(outp, " %6.2f", 100.0 * c->c6/t->tsc);
536         if (do_snb_cstates)
537                 outp += sprintf(outp, " %6.2f", 100.0 * c->c7/t->tsc);
538
539         if (do_dts)
540                 outp += sprintf(outp, " %4d", c->core_temp_c);
541
542         /* print per-package data only for 1st core in package */
543         if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
544                 goto done;
545
546         if (do_ptm)
547                 outp += sprintf(outp, " %4d", p->pkg_temp_c);
548
549         if (do_snb_cstates)
550                 outp += sprintf(outp, " %6.2f", 100.0 * p->pc2/t->tsc);
551         if (do_nhm_cstates && !do_slm_cstates)
552                 outp += sprintf(outp, " %6.2f", 100.0 * p->pc3/t->tsc);
553         if (do_nhm_cstates && !do_slm_cstates)
554                 outp += sprintf(outp, " %6.2f", 100.0 * p->pc6/t->tsc);
555         if (do_snb_cstates)
556                 outp += sprintf(outp, " %6.2f", 100.0 * p->pc7/t->tsc);
557         if (do_c8_c9_c10) {
558                 outp += sprintf(outp, " %6.2f", 100.0 * p->pc8/t->tsc);
559                 outp += sprintf(outp, " %6.2f", 100.0 * p->pc9/t->tsc);
560                 outp += sprintf(outp, " %6.2f", 100.0 * p->pc10/t->tsc);
561         }
562
563         /*
564          * If measurement interval exceeds minimum RAPL Joule Counter range,
565          * indicate that results are suspect by printing "**" in fraction place.
566          */
567         if (interval_float < rapl_joule_counter_range) {
568                 fmt5 = " %5.2f";
569                 fmt6 = " %6.2f";
570         } else {
571                 fmt5 = " %3.0f**";
572                 fmt6 = " %4.0f**";
573         }
574
575         if (do_rapl && !rapl_joules) {
576                 if (do_rapl & RAPL_PKG)
577                         outp += sprintf(outp, fmt6, p->energy_pkg * rapl_energy_units / interval_float);
578                 if (do_rapl & RAPL_CORES)
579                         outp += sprintf(outp, fmt6, p->energy_cores * rapl_energy_units / interval_float);
580                 if (do_rapl & RAPL_GFX)
581                         outp += sprintf(outp, fmt5, p->energy_gfx * rapl_energy_units / interval_float);
582                 if (do_rapl & RAPL_DRAM)
583                         outp += sprintf(outp, fmt5, p->energy_dram * rapl_energy_units / interval_float);
584                 if (do_rapl & RAPL_PKG_PERF_STATUS)
585                         outp += sprintf(outp, fmt5, 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float);
586                 if (do_rapl & RAPL_DRAM_PERF_STATUS)
587                         outp += sprintf(outp, fmt5, 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float);
588         } else {
589                 if (do_rapl & RAPL_PKG)
590                         outp += sprintf(outp, fmt6,
591                                         p->energy_pkg * rapl_energy_units);
592                 if (do_rapl & RAPL_CORES)
593                         outp += sprintf(outp, fmt6,
594                                         p->energy_cores * rapl_energy_units);
595                 if (do_rapl & RAPL_GFX)
596                         outp += sprintf(outp, fmt5,
597                                         p->energy_gfx * rapl_energy_units);
598                 if (do_rapl & RAPL_DRAM)
599                         outp += sprintf(outp, fmt5,
600                                         p->energy_dram * rapl_energy_units);
601                 if (do_rapl & RAPL_PKG_PERF_STATUS)
602                         outp += sprintf(outp, fmt5, 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float);
603                 if (do_rapl & RAPL_DRAM_PERF_STATUS)
604                         outp += sprintf(outp, fmt5, 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float);
605         outp += sprintf(outp, fmt5, interval_float);
606
607         }
608 done:
609         outp += sprintf(outp, "\n");
610
611         return 0;
612 }
613
614 void flush_stdout()
615 {
616         fputs(output_buffer, stdout);
617         fflush(stdout);
618         outp = output_buffer;
619 }
620 void flush_stderr()
621 {
622         fputs(output_buffer, stderr);
623         outp = output_buffer;
624 }
625 void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
626 {
627         static int printed;
628
629         if (!printed || !summary_only)
630                 print_header();
631
632         if (topo.num_cpus > 1)
633                 format_counters(&average.threads, &average.cores,
634                         &average.packages);
635
636         printed = 1;
637
638         if (summary_only)
639                 return;
640
641         for_all_cpus(format_counters, t, c, p);
642 }
643
644 #define DELTA_WRAP32(new, old)                  \
645         if (new > old) {                        \
646                 old = new - old;                \
647         } else {                                \
648                 old = 0x100000000 + new - old;  \
649         }
650
651 void
652 delta_package(struct pkg_data *new, struct pkg_data *old)
653 {
654         old->pc2 = new->pc2 - old->pc2;
655         old->pc3 = new->pc3 - old->pc3;
656         old->pc6 = new->pc6 - old->pc6;
657         old->pc7 = new->pc7 - old->pc7;
658         old->pc8 = new->pc8 - old->pc8;
659         old->pc9 = new->pc9 - old->pc9;
660         old->pc10 = new->pc10 - old->pc10;
661         old->pkg_temp_c = new->pkg_temp_c;
662
663         DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
664         DELTA_WRAP32(new->energy_cores, old->energy_cores);
665         DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
666         DELTA_WRAP32(new->energy_dram, old->energy_dram);
667         DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
668         DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
669 }
670
671 void
672 delta_core(struct core_data *new, struct core_data *old)
673 {
674         old->c3 = new->c3 - old->c3;
675         old->c6 = new->c6 - old->c6;
676         old->c7 = new->c7 - old->c7;
677         old->core_temp_c = new->core_temp_c;
678 }
679
680 /*
681  * old = new - old
682  */
683 void
684 delta_thread(struct thread_data *new, struct thread_data *old,
685         struct core_data *core_delta)
686 {
687         old->tsc = new->tsc - old->tsc;
688
689         /* check for TSC < 1 Mcycles over interval */
690         if (old->tsc < (1000 * 1000))
691                 errx(-3, "Insanely slow TSC rate, TSC stops in idle?\n"
692                      "You can disable all c-states by booting with \"idle=poll\"\n"
693                      "or just the deep ones with \"processor.max_cstate=1\"");
694
695         old->c1 = new->c1 - old->c1;
696
697         if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
698                 old->aperf = new->aperf - old->aperf;
699                 old->mperf = new->mperf - old->mperf;
700         } else {
701
702                 if (!aperf_mperf_unstable) {
703                         fprintf(stderr, "%s: APERF or MPERF went backwards *\n", progname);
704                         fprintf(stderr, "* Frequency results do not cover entire interval *\n");
705                         fprintf(stderr, "* fix this by running Linux-2.6.30 or later *\n");
706
707                         aperf_mperf_unstable = 1;
708                 }
709                 /*
710                  * mperf delta is likely a huge "positive" number
711                  * can not use it for calculating c0 time
712                  */
713                 skip_c0 = 1;
714                 skip_c1 = 1;
715         }
716
717
718         if (use_c1_residency_msr) {
719                 /*
720                  * Some models have a dedicated C1 residency MSR,
721                  * which should be more accurate than the derivation below.
722                  */
723         } else {
724                 /*
725                  * As counter collection is not atomic,
726                  * it is possible for mperf's non-halted cycles + idle states
727                  * to exceed TSC's all cycles: show c1 = 0% in that case.
728                  */
729                 if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > old->tsc)
730                         old->c1 = 0;
731                 else {
732                         /* normal case, derive c1 */
733                         old->c1 = old->tsc - old->mperf - core_delta->c3
734                                 - core_delta->c6 - core_delta->c7;
735                 }
736         }
737
738         if (old->mperf == 0) {
739                 if (verbose > 1) fprintf(stderr, "cpu%d MPERF 0!\n", old->cpu_id);
740                 old->mperf = 1; /* divide by 0 protection */
741         }
742
743         old->extra_delta32 = new->extra_delta32 - old->extra_delta32;
744         old->extra_delta32 &= 0xFFFFFFFF;
745
746         old->extra_delta64 = new->extra_delta64 - old->extra_delta64;
747
748         /*
749          * Extra MSR is just a snapshot, simply copy latest w/o subtracting
750          */
751         old->extra_msr32 = new->extra_msr32;
752         old->extra_msr64 = new->extra_msr64;
753
754         if (do_smi)
755                 old->smi_count = new->smi_count - old->smi_count;
756 }
757
758 int delta_cpu(struct thread_data *t, struct core_data *c,
759         struct pkg_data *p, struct thread_data *t2,
760         struct core_data *c2, struct pkg_data *p2)
761 {
762         /* calculate core delta only for 1st thread in core */
763         if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
764                 delta_core(c, c2);
765
766         /* always calculate thread delta */
767         delta_thread(t, t2, c2);        /* c2 is core delta */
768
769         /* calculate package delta only for 1st core in package */
770         if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
771                 delta_package(p, p2);
772
773         return 0;
774 }
775
776 void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
777 {
778         t->tsc = 0;
779         t->aperf = 0;
780         t->mperf = 0;
781         t->c1 = 0;
782
783         t->smi_count = 0;
784         t->extra_delta32 = 0;
785         t->extra_delta64 = 0;
786
787         /* tells format_counters to dump all fields from this set */
788         t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE;
789
790         c->c3 = 0;
791         c->c6 = 0;
792         c->c7 = 0;
793         c->core_temp_c = 0;
794
795         p->pc2 = 0;
796         p->pc3 = 0;
797         p->pc6 = 0;
798         p->pc7 = 0;
799         p->pc8 = 0;
800         p->pc9 = 0;
801         p->pc10 = 0;
802
803         p->energy_pkg = 0;
804         p->energy_dram = 0;
805         p->energy_cores = 0;
806         p->energy_gfx = 0;
807         p->rapl_pkg_perf_status = 0;
808         p->rapl_dram_perf_status = 0;
809         p->pkg_temp_c = 0;
810 }
811 int sum_counters(struct thread_data *t, struct core_data *c,
812         struct pkg_data *p)
813 {
814         average.threads.tsc += t->tsc;
815         average.threads.aperf += t->aperf;
816         average.threads.mperf += t->mperf;
817         average.threads.c1 += t->c1;
818
819         average.threads.extra_delta32 += t->extra_delta32;
820         average.threads.extra_delta64 += t->extra_delta64;
821
822         /* sum per-core values only for 1st thread in core */
823         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
824                 return 0;
825
826         average.cores.c3 += c->c3;
827         average.cores.c6 += c->c6;
828         average.cores.c7 += c->c7;
829
830         average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c);
831
832         /* sum per-pkg values only for 1st core in pkg */
833         if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
834                 return 0;
835
836         average.packages.pc2 += p->pc2;
837         average.packages.pc3 += p->pc3;
838         average.packages.pc6 += p->pc6;
839         average.packages.pc7 += p->pc7;
840         average.packages.pc8 += p->pc8;
841         average.packages.pc9 += p->pc9;
842         average.packages.pc10 += p->pc10;
843
844         average.packages.energy_pkg += p->energy_pkg;
845         average.packages.energy_dram += p->energy_dram;
846         average.packages.energy_cores += p->energy_cores;
847         average.packages.energy_gfx += p->energy_gfx;
848
849         average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c);
850
851         average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status;
852         average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status;
853         return 0;
854 }
855 /*
856  * sum the counters for all cpus in the system
857  * compute the weighted average
858  */
859 void compute_average(struct thread_data *t, struct core_data *c,
860         struct pkg_data *p)
861 {
862         clear_counters(&average.threads, &average.cores, &average.packages);
863
864         for_all_cpus(sum_counters, t, c, p);
865
866         average.threads.tsc /= topo.num_cpus;
867         average.threads.aperf /= topo.num_cpus;
868         average.threads.mperf /= topo.num_cpus;
869         average.threads.c1 /= topo.num_cpus;
870
871         average.threads.extra_delta32 /= topo.num_cpus;
872         average.threads.extra_delta32 &= 0xFFFFFFFF;
873
874         average.threads.extra_delta64 /= topo.num_cpus;
875
876         average.cores.c3 /= topo.num_cores;
877         average.cores.c6 /= topo.num_cores;
878         average.cores.c7 /= topo.num_cores;
879
880         average.packages.pc2 /= topo.num_packages;
881         average.packages.pc3 /= topo.num_packages;
882         average.packages.pc6 /= topo.num_packages;
883         average.packages.pc7 /= topo.num_packages;
884
885         average.packages.pc8 /= topo.num_packages;
886         average.packages.pc9 /= topo.num_packages;
887         average.packages.pc10 /= topo.num_packages;
888 }
889
890 static unsigned long long rdtsc(void)
891 {
892         unsigned int low, high;
893
894         asm volatile("rdtsc" : "=a" (low), "=d" (high));
895
896         return low | ((unsigned long long)high) << 32;
897 }
898
899
900 /*
901  * get_counters(...)
902  * migrate to cpu
903  * acquire and record local counters for that cpu
904  */
905 int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
906 {
907         int cpu = t->cpu_id;
908         unsigned long long msr;
909
910         if (cpu_migrate(cpu)) {
911                 fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
912                 return -1;
913         }
914
915         t->tsc = rdtsc();       /* we are running on local CPU of interest */
916
917         if (has_aperf) {
918                 if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
919                         return -3;
920                 if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
921                         return -4;
922         }
923
924         if (do_smi) {
925                 if (get_msr(cpu, MSR_SMI_COUNT, &msr))
926                         return -5;
927                 t->smi_count = msr & 0xFFFFFFFF;
928         }
929         if (extra_delta_offset32) {
930                 if (get_msr(cpu, extra_delta_offset32, &msr))
931                         return -5;
932                 t->extra_delta32 = msr & 0xFFFFFFFF;
933         }
934
935         if (extra_delta_offset64)
936                 if (get_msr(cpu, extra_delta_offset64, &t->extra_delta64))
937                         return -5;
938
939         if (extra_msr_offset32) {
940                 if (get_msr(cpu, extra_msr_offset32, &msr))
941                         return -5;
942                 t->extra_msr32 = msr & 0xFFFFFFFF;
943         }
944
945         if (extra_msr_offset64)
946                 if (get_msr(cpu, extra_msr_offset64, &t->extra_msr64))
947                         return -5;
948
949         if (use_c1_residency_msr) {
950                 if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1))
951                         return -6;
952         }
953
954         /* collect core counters only for 1st thread in core */
955         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
956                 return 0;
957
958         if (do_nhm_cstates && !do_slm_cstates) {
959                 if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
960                         return -6;
961         }
962
963         if (do_nhm_cstates) {
964                 if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
965                         return -7;
966         }
967
968         if (do_snb_cstates)
969                 if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
970                         return -8;
971
972         if (do_dts) {
973                 if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
974                         return -9;
975                 c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
976         }
977
978
979         /* collect package counters only for 1st core in package */
980         if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
981                 return 0;
982
983         if (do_nhm_cstates && !do_slm_cstates) {
984                 if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
985                         return -9;
986                 if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
987                         return -10;
988         }
989         if (do_snb_cstates) {
990                 if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
991                         return -11;
992                 if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
993                         return -12;
994         }
995         if (do_c8_c9_c10) {
996                 if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
997                         return -13;
998                 if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
999                         return -13;
1000                 if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
1001                         return -13;
1002         }
1003         if (do_rapl & RAPL_PKG) {
1004                 if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
1005                         return -13;
1006                 p->energy_pkg = msr & 0xFFFFFFFF;
1007         }
1008         if (do_rapl & RAPL_CORES) {
1009                 if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
1010                         return -14;
1011                 p->energy_cores = msr & 0xFFFFFFFF;
1012         }
1013         if (do_rapl & RAPL_DRAM) {
1014                 if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
1015                         return -15;
1016                 p->energy_dram = msr & 0xFFFFFFFF;
1017         }
1018         if (do_rapl & RAPL_GFX) {
1019                 if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
1020                         return -16;
1021                 p->energy_gfx = msr & 0xFFFFFFFF;
1022         }
1023         if (do_rapl & RAPL_PKG_PERF_STATUS) {
1024                 if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
1025                         return -16;
1026                 p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
1027         }
1028         if (do_rapl & RAPL_DRAM_PERF_STATUS) {
1029                 if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
1030                         return -16;
1031                 p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
1032         }
1033         if (do_ptm) {
1034                 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
1035                         return -17;
1036                 p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
1037         }
1038         return 0;
1039 }
1040
1041 void print_verbose_header(void)
1042 {
1043         unsigned long long msr;
1044         unsigned int ratio;
1045
1046         if (!do_nehalem_platform_info)
1047                 return;
1048
1049         get_msr(0, MSR_NHM_PLATFORM_INFO, &msr);
1050
1051         fprintf(stderr, "cpu0: MSR_NHM_PLATFORM_INFO: 0x%08llx\n", msr);
1052
1053         ratio = (msr >> 40) & 0xFF;
1054         fprintf(stderr, "%d * %.0f = %.0f MHz max efficiency\n",
1055                 ratio, bclk, ratio * bclk);
1056
1057         ratio = (msr >> 8) & 0xFF;
1058         fprintf(stderr, "%d * %.0f = %.0f MHz TSC frequency\n",
1059                 ratio, bclk, ratio * bclk);
1060
1061         get_msr(0, MSR_IA32_POWER_CTL, &msr);
1062         fprintf(stderr, "cpu0: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n",
1063                 msr, msr & 0x2 ? "EN" : "DIS");
1064
1065         if (!do_ivt_turbo_ratio_limit)
1066                 goto print_nhm_turbo_ratio_limits;
1067
1068         get_msr(0, MSR_IVT_TURBO_RATIO_LIMIT, &msr);
1069
1070         fprintf(stderr, "cpu0: MSR_IVT_TURBO_RATIO_LIMIT: 0x%08llx\n", msr);
1071
1072         ratio = (msr >> 56) & 0xFF;
1073         if (ratio)
1074                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 16 active cores\n",
1075                         ratio, bclk, ratio * bclk);
1076
1077         ratio = (msr >> 48) & 0xFF;
1078         if (ratio)
1079                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 15 active cores\n",
1080                         ratio, bclk, ratio * bclk);
1081
1082         ratio = (msr >> 40) & 0xFF;
1083         if (ratio)
1084                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 14 active cores\n",
1085                         ratio, bclk, ratio * bclk);
1086
1087         ratio = (msr >> 32) & 0xFF;
1088         if (ratio)
1089                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 13 active cores\n",
1090                         ratio, bclk, ratio * bclk);
1091
1092         ratio = (msr >> 24) & 0xFF;
1093         if (ratio)
1094                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 12 active cores\n",
1095                         ratio, bclk, ratio * bclk);
1096
1097         ratio = (msr >> 16) & 0xFF;
1098         if (ratio)
1099                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 11 active cores\n",
1100                         ratio, bclk, ratio * bclk);
1101
1102         ratio = (msr >> 8) & 0xFF;
1103         if (ratio)
1104                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 10 active cores\n",
1105                         ratio, bclk, ratio * bclk);
1106
1107         ratio = (msr >> 0) & 0xFF;
1108         if (ratio)
1109                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 9 active cores\n",
1110                         ratio, bclk, ratio * bclk);
1111
1112 print_nhm_turbo_ratio_limits:
1113         get_msr(0, MSR_NHM_SNB_PKG_CST_CFG_CTL, &msr);
1114
1115 #define SNB_C1_AUTO_UNDEMOTE              (1UL << 27)
1116 #define SNB_C3_AUTO_UNDEMOTE              (1UL << 28)
1117
1118         fprintf(stderr, "cpu0: MSR_NHM_SNB_PKG_CST_CFG_CTL: 0x%08llx", msr);
1119
1120         fprintf(stderr, " (%s%s%s%s%slocked: pkg-cstate-limit=%d: ",
1121                 (msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "",
1122                 (msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "",
1123                 (msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "",
1124                 (msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "",
1125                 (msr & (1 << 15)) ? "" : "UN",
1126                 (unsigned int)msr & 7);
1127
1128
1129         switch(msr & 0x7) {
1130         case 0:
1131                 fprintf(stderr, do_slm_cstates ? "no pkg states" : "pc0");
1132                 break;
1133         case 1:
1134                 fprintf(stderr, do_slm_cstates ? "no pkg states" : do_snb_cstates ? "pc2" : "pc0");
1135                 break;
1136         case 2:
1137                 fprintf(stderr, do_slm_cstates ? "invalid" : do_snb_cstates ? "pc6-noret" : "pc3");
1138                 break;
1139         case 3:
1140                 fprintf(stderr, do_slm_cstates ? "invalid" : "pc6");
1141                 break;
1142         case 4:
1143                 fprintf(stderr, do_slm_cstates ? "pc4" : "pc7");
1144                 break;
1145         case 5:
1146                 fprintf(stderr, do_slm_cstates ? "invalid" : do_snb_cstates ? "pc7s" : "invalid");
1147                 break;
1148         case 6:
1149                 fprintf(stderr, do_slm_cstates ? "pc6" : "invalid");
1150                 break;
1151         case 7:
1152                 fprintf(stderr, do_slm_cstates ? "pc7" : "unlimited");
1153                 break;
1154         default:
1155                 fprintf(stderr, "invalid");
1156         }
1157         fprintf(stderr, ")\n");
1158
1159         if (!do_nehalem_turbo_ratio_limit)
1160                 return;
1161
1162         get_msr(0, MSR_NHM_TURBO_RATIO_LIMIT, &msr);
1163
1164         fprintf(stderr, "cpu0: MSR_NHM_TURBO_RATIO_LIMIT: 0x%08llx\n", msr);
1165
1166         ratio = (msr >> 56) & 0xFF;
1167         if (ratio)
1168                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 8 active cores\n",
1169                         ratio, bclk, ratio * bclk);
1170
1171         ratio = (msr >> 48) & 0xFF;
1172         if (ratio)
1173                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 7 active cores\n",
1174                         ratio, bclk, ratio * bclk);
1175
1176         ratio = (msr >> 40) & 0xFF;
1177         if (ratio)
1178                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 6 active cores\n",
1179                         ratio, bclk, ratio * bclk);
1180
1181         ratio = (msr >> 32) & 0xFF;
1182         if (ratio)
1183                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 5 active cores\n",
1184                         ratio, bclk, ratio * bclk);
1185
1186         ratio = (msr >> 24) & 0xFF;
1187         if (ratio)
1188                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 4 active cores\n",
1189                         ratio, bclk, ratio * bclk);
1190
1191         ratio = (msr >> 16) & 0xFF;
1192         if (ratio)
1193                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 3 active cores\n",
1194                         ratio, bclk, ratio * bclk);
1195
1196         ratio = (msr >> 8) & 0xFF;
1197         if (ratio)
1198                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 2 active cores\n",
1199                         ratio, bclk, ratio * bclk);
1200
1201         ratio = (msr >> 0) & 0xFF;
1202         if (ratio)
1203                 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 1 active cores\n",
1204                         ratio, bclk, ratio * bclk);
1205 }
1206
1207 void free_all_buffers(void)
1208 {
1209         CPU_FREE(cpu_present_set);
1210         cpu_present_set = NULL;
1211         cpu_present_set = 0;
1212
1213         CPU_FREE(cpu_affinity_set);
1214         cpu_affinity_set = NULL;
1215         cpu_affinity_setsize = 0;
1216
1217         free(thread_even);
1218         free(core_even);
1219         free(package_even);
1220
1221         thread_even = NULL;
1222         core_even = NULL;
1223         package_even = NULL;
1224
1225         free(thread_odd);
1226         free(core_odd);
1227         free(package_odd);
1228
1229         thread_odd = NULL;
1230         core_odd = NULL;
1231         package_odd = NULL;
1232
1233         free(output_buffer);
1234         output_buffer = NULL;
1235         outp = NULL;
1236 }
1237
1238 /*
1239  * Open a file, and exit on failure
1240  */
1241 FILE *fopen_or_die(const char *path, const char *mode)
1242 {
1243         FILE *filep = fopen(path, "r");
1244         if (!filep)
1245                 err(1, "%s: open failed", path);
1246         return filep;
1247 }
1248
1249 /*
1250  * Parse a file containing a single int.
1251  */
1252 int parse_int_file(const char *fmt, ...)
1253 {
1254         va_list args;
1255         char path[PATH_MAX];
1256         FILE *filep;
1257         int value;
1258
1259         va_start(args, fmt);
1260         vsnprintf(path, sizeof(path), fmt, args);
1261         va_end(args);
1262         filep = fopen_or_die(path, "r");
1263         if (fscanf(filep, "%d", &value) != 1)
1264                 err(1, "%s: failed to parse number from file", path);
1265         fclose(filep);
1266         return value;
1267 }
1268
1269 /*
1270  * cpu_is_first_sibling_in_core(cpu)
1271  * return 1 if given CPU is 1st HT sibling in the core
1272  */
1273 int cpu_is_first_sibling_in_core(int cpu)
1274 {
1275         return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
1276 }
1277
1278 /*
1279  * cpu_is_first_core_in_package(cpu)
1280  * return 1 if given CPU is 1st core in package
1281  */
1282 int cpu_is_first_core_in_package(int cpu)
1283 {
1284         return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
1285 }
1286
1287 int get_physical_package_id(int cpu)
1288 {
1289         return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
1290 }
1291
1292 int get_core_id(int cpu)
1293 {
1294         return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
1295 }
1296
1297 int get_num_ht_siblings(int cpu)
1298 {
1299         char path[80];
1300         FILE *filep;
1301         int sib1, sib2;
1302         int matches;
1303         char character;
1304
1305         sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
1306         filep = fopen_or_die(path, "r");
1307         /*
1308          * file format:
1309          * if a pair of number with a character between: 2 siblings (eg. 1-2, or 1,4)
1310          * otherwinse 1 sibling (self).
1311          */
1312         matches = fscanf(filep, "%d%c%d\n", &sib1, &character, &sib2);
1313
1314         fclose(filep);
1315
1316         if (matches == 3)
1317                 return 2;
1318         else
1319                 return 1;
1320 }
1321
1322 /*
1323  * run func(thread, core, package) in topology order
1324  * skip non-present cpus
1325  */
1326
1327 int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
1328         struct pkg_data *, struct thread_data *, struct core_data *,
1329         struct pkg_data *), struct thread_data *thread_base,
1330         struct core_data *core_base, struct pkg_data *pkg_base,
1331         struct thread_data *thread_base2, struct core_data *core_base2,
1332         struct pkg_data *pkg_base2)
1333 {
1334         int retval, pkg_no, core_no, thread_no;
1335
1336         for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
1337                 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
1338                         for (thread_no = 0; thread_no <
1339                                 topo.num_threads_per_core; ++thread_no) {
1340                                 struct thread_data *t, *t2;
1341                                 struct core_data *c, *c2;
1342                                 struct pkg_data *p, *p2;
1343
1344                                 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
1345
1346                                 if (cpu_is_not_present(t->cpu_id))
1347                                         continue;
1348
1349                                 t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no);
1350
1351                                 c = GET_CORE(core_base, core_no, pkg_no);
1352                                 c2 = GET_CORE(core_base2, core_no, pkg_no);
1353
1354                                 p = GET_PKG(pkg_base, pkg_no);
1355                                 p2 = GET_PKG(pkg_base2, pkg_no);
1356
1357                                 retval = func(t, c, p, t2, c2, p2);
1358                                 if (retval)
1359                                         return retval;
1360                         }
1361                 }
1362         }
1363         return 0;
1364 }
1365
1366 /*
1367  * run func(cpu) on every cpu in /proc/stat
1368  * return max_cpu number
1369  */
1370 int for_all_proc_cpus(int (func)(int))
1371 {
1372         FILE *fp;
1373         int cpu_num;
1374         int retval;
1375
1376         fp = fopen_or_die(proc_stat, "r");
1377
1378         retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
1379         if (retval != 0)
1380                 err(1, "%s: failed to parse format", proc_stat);
1381
1382         while (1) {
1383                 retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
1384                 if (retval != 1)
1385                         break;
1386
1387                 retval = func(cpu_num);
1388                 if (retval) {
1389                         fclose(fp);
1390                         return(retval);
1391                 }
1392         }
1393         fclose(fp);
1394         return 0;
1395 }
1396
1397 void re_initialize(void)
1398 {
1399         free_all_buffers();
1400         setup_all_buffers();
1401         printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus);
1402 }
1403
1404
1405 /*
1406  * count_cpus()
1407  * remember the last one seen, it will be the max
1408  */
1409 int count_cpus(int cpu)
1410 {
1411         if (topo.max_cpu_num < cpu)
1412                 topo.max_cpu_num = cpu;
1413
1414         topo.num_cpus += 1;
1415         return 0;
1416 }
1417 int mark_cpu_present(int cpu)
1418 {
1419         CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
1420         return 0;
1421 }
1422
1423 void turbostat_loop()
1424 {
1425         int retval;
1426         int restarted = 0;
1427
1428 restart:
1429         restarted++;
1430
1431         retval = for_all_cpus(get_counters, EVEN_COUNTERS);
1432         if (retval < -1) {
1433                 exit(retval);
1434         } else if (retval == -1) {
1435                 if (restarted > 1) {
1436                         exit(retval);
1437                 }
1438                 re_initialize();
1439                 goto restart;
1440         }
1441         restarted = 0;
1442         gettimeofday(&tv_even, (struct timezone *)NULL);
1443
1444         while (1) {
1445                 if (for_all_proc_cpus(cpu_is_not_present)) {
1446                         re_initialize();
1447                         goto restart;
1448                 }
1449                 sleep(interval_sec);
1450                 retval = for_all_cpus(get_counters, ODD_COUNTERS);
1451                 if (retval < -1) {
1452                         exit(retval);
1453                 } else if (retval == -1) {
1454                         re_initialize();
1455                         goto restart;
1456                 }
1457                 gettimeofday(&tv_odd, (struct timezone *)NULL);
1458                 timersub(&tv_odd, &tv_even, &tv_delta);
1459                 for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS);
1460                 compute_average(EVEN_COUNTERS);
1461                 format_all_counters(EVEN_COUNTERS);
1462                 flush_stdout();
1463                 sleep(interval_sec);
1464                 retval = for_all_cpus(get_counters, EVEN_COUNTERS);
1465                 if (retval < -1) {
1466                         exit(retval);
1467                 } else if (retval == -1) {
1468                         re_initialize();
1469                         goto restart;
1470                 }
1471                 gettimeofday(&tv_even, (struct timezone *)NULL);
1472                 timersub(&tv_even, &tv_odd, &tv_delta);
1473                 for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS);
1474                 compute_average(ODD_COUNTERS);
1475                 format_all_counters(ODD_COUNTERS);
1476                 flush_stdout();
1477         }
1478 }
1479
1480 void check_dev_msr()
1481 {
1482         struct stat sb;
1483
1484         if (stat("/dev/cpu/0/msr", &sb))
1485                 err(-5, "no /dev/cpu/0/msr\n"
1486                     "Try \"# modprobe msr\"");
1487 }
1488
1489 void check_super_user()
1490 {
1491         if (getuid() != 0)
1492                 errx(-6, "must be root");
1493 }
1494
1495 int has_nehalem_turbo_ratio_limit(unsigned int family, unsigned int model)
1496 {
1497         if (!genuine_intel)
1498                 return 0;
1499
1500         if (family != 6)
1501                 return 0;
1502
1503         switch (model) {
1504         case 0x1A:      /* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
1505         case 0x1E:      /* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
1506         case 0x1F:      /* Core i7 and i5 Processor - Nehalem */
1507         case 0x25:      /* Westmere Client - Clarkdale, Arrandale */
1508         case 0x2C:      /* Westmere EP - Gulftown */
1509         case 0x2A:      /* SNB */
1510         case 0x2D:      /* SNB Xeon */
1511         case 0x3A:      /* IVB */
1512         case 0x3E:      /* IVB Xeon */
1513         case 0x3C:      /* HSW */
1514         case 0x3F:      /* HSX */
1515         case 0x45:      /* HSW */
1516         case 0x46:      /* HSW */
1517         case 0x37:      /* BYT */
1518         case 0x4D:      /* AVN */
1519                 return 1;
1520         case 0x2E:      /* Nehalem-EX Xeon - Beckton */
1521         case 0x2F:      /* Westmere-EX Xeon - Eagleton */
1522         default:
1523                 return 0;
1524         }
1525 }
1526 int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model)
1527 {
1528         if (!genuine_intel)
1529                 return 0;
1530
1531         if (family != 6)
1532                 return 0;
1533
1534         switch (model) {
1535         case 0x3E:      /* IVB Xeon */
1536                 return 1;
1537         default:
1538                 return 0;
1539         }
1540 }
1541
1542 /*
1543  * print_epb()
1544  * Decode the ENERGY_PERF_BIAS MSR
1545  */
1546 int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1547 {
1548         unsigned long long msr;
1549         char *epb_string;
1550         int cpu;
1551
1552         if (!has_epb)
1553                 return 0;
1554
1555         cpu = t->cpu_id;
1556
1557         /* EPB is per-package */
1558         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1559                 return 0;
1560
1561         if (cpu_migrate(cpu)) {
1562                 fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
1563                 return -1;
1564         }
1565
1566         if (get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr))
1567                 return 0;
1568
1569         switch (msr & 0x7) {
1570         case ENERGY_PERF_BIAS_PERFORMANCE:
1571                 epb_string = "performance";
1572                 break;
1573         case ENERGY_PERF_BIAS_NORMAL:
1574                 epb_string = "balanced";
1575                 break;
1576         case ENERGY_PERF_BIAS_POWERSAVE:
1577                 epb_string = "powersave";
1578                 break;
1579         default:
1580                 epb_string = "custom";
1581                 break;
1582         }
1583         fprintf(stderr, "cpu%d: MSR_IA32_ENERGY_PERF_BIAS: 0x%08llx (%s)\n", cpu, msr, epb_string);
1584
1585         return 0;
1586 }
1587
1588 #define RAPL_POWER_GRANULARITY  0x7FFF  /* 15 bit power granularity */
1589 #define RAPL_TIME_GRANULARITY   0x3F /* 6 bit time granularity */
1590
1591 double get_tdp(model)
1592 {
1593         unsigned long long msr;
1594
1595         if (do_rapl & RAPL_PKG_POWER_INFO)
1596                 if (!get_msr(0, MSR_PKG_POWER_INFO, &msr))
1597                         return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
1598
1599         switch (model) {
1600         case 0x37:
1601         case 0x4D:
1602                 return 30.0;
1603         default:
1604                 return 135.0;
1605         }
1606 }
1607
1608
1609 /*
1610  * rapl_probe()
1611  *
1612  * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
1613  */
1614 void rapl_probe(unsigned int family, unsigned int model)
1615 {
1616         unsigned long long msr;
1617         unsigned int time_unit;
1618         double tdp;
1619
1620         if (!genuine_intel)
1621                 return;
1622
1623         if (family != 6)
1624                 return;
1625
1626         switch (model) {
1627         case 0x2A:
1628         case 0x3A:
1629         case 0x3C:      /* HSW */
1630         case 0x45:      /* HSW */
1631         case 0x46:      /* HSW */
1632                 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
1633                 break;
1634         case 0x3F:      /* HSX */
1635                 do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
1636                 break;
1637         case 0x2D:
1638         case 0x3E:
1639                 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_PKG_PERF_STATUS | RAPL_DRAM_PERF_STATUS | RAPL_PKG_POWER_INFO;
1640                 break;
1641         case 0x37:      /* BYT */
1642         case 0x4D:      /* AVN */
1643                 do_rapl = RAPL_PKG | RAPL_CORES ;
1644                 break;
1645         default:
1646                 return;
1647         }
1648
1649         /* units on package 0, verify later other packages match */
1650         if (get_msr(0, MSR_RAPL_POWER_UNIT, &msr))
1651                 return;
1652
1653         rapl_power_units = 1.0 / (1 << (msr & 0xF));
1654         if (model == 0x37)
1655                 rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
1656         else
1657                 rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
1658
1659         time_unit = msr >> 16 & 0xF;
1660         if (time_unit == 0)
1661                 time_unit = 0xA;
1662
1663         rapl_time_units = 1.0 / (1 << (time_unit));
1664
1665         tdp = get_tdp(model);
1666
1667         rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
1668         if (verbose)
1669                 fprintf(stderr, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
1670
1671         return;
1672 }
1673
1674 int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1675 {
1676         unsigned long long msr;
1677         unsigned int dts;
1678         int cpu;
1679
1680         if (!(do_dts || do_ptm))
1681                 return 0;
1682
1683         cpu = t->cpu_id;
1684
1685         /* DTS is per-core, no need to print for each thread */
1686         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) 
1687                 return 0;
1688
1689         if (cpu_migrate(cpu)) {
1690                 fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
1691                 return -1;
1692         }
1693
1694         if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
1695                 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
1696                         return 0;
1697
1698                 dts = (msr >> 16) & 0x7F;
1699                 fprintf(stderr, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n",
1700                         cpu, msr, tcc_activation_temp - dts);
1701
1702 #ifdef  THERM_DEBUG
1703                 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr))
1704                         return 0;
1705
1706                 dts = (msr >> 16) & 0x7F;
1707                 dts2 = (msr >> 8) & 0x7F;
1708                 fprintf(stderr, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
1709                         cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
1710 #endif
1711         }
1712
1713
1714         if (do_dts) {
1715                 unsigned int resolution;
1716
1717                 if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
1718                         return 0;
1719
1720                 dts = (msr >> 16) & 0x7F;
1721                 resolution = (msr >> 27) & 0xF;
1722                 fprintf(stderr, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n",
1723                         cpu, msr, tcc_activation_temp - dts, resolution);
1724
1725 #ifdef THERM_DEBUG
1726                 if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr))
1727                         return 0;
1728
1729                 dts = (msr >> 16) & 0x7F;
1730                 dts2 = (msr >> 8) & 0x7F;
1731                 fprintf(stderr, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
1732                         cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
1733 #endif
1734         }
1735
1736         return 0;
1737 }
1738         
1739 void print_power_limit_msr(int cpu, unsigned long long msr, char *label)
1740 {
1741         fprintf(stderr, "cpu%d: %s: %sabled (%f Watts, %f sec, clamp %sabled)\n",
1742                 cpu, label,
1743                 ((msr >> 15) & 1) ? "EN" : "DIS",
1744                 ((msr >> 0) & 0x7FFF) * rapl_power_units,
1745                 (1.0 + (((msr >> 22) & 0x3)/4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units,
1746                 (((msr >> 16) & 1) ? "EN" : "DIS"));
1747
1748         return;
1749 }
1750
1751 int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1752 {
1753         unsigned long long msr;
1754         int cpu;
1755
1756         if (!do_rapl)
1757                 return 0;
1758
1759         /* RAPL counters are per package, so print only for 1st thread/package */
1760         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1761                 return 0;
1762
1763         cpu = t->cpu_id;
1764         if (cpu_migrate(cpu)) {
1765                 fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
1766                 return -1;
1767         }
1768
1769         if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr))
1770                 return -1;
1771
1772         if (verbose) {
1773                 fprintf(stderr, "cpu%d: MSR_RAPL_POWER_UNIT: 0x%08llx "
1774                         "(%f Watts, %f Joules, %f sec.)\n", cpu, msr,
1775                         rapl_power_units, rapl_energy_units, rapl_time_units);
1776         }
1777         if (do_rapl & RAPL_PKG_POWER_INFO) {
1778
1779                 if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr))
1780                         return -5;
1781
1782
1783                 fprintf(stderr, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
1784                         cpu, msr,
1785                         ((msr >>  0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1786                         ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1787                         ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1788                         ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
1789
1790         }
1791         if (do_rapl & RAPL_PKG) {
1792
1793                 if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr))
1794                         return -9;
1795
1796                 fprintf(stderr, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n",
1797                         cpu, msr, (msr >> 63) & 1 ? "": "UN");
1798
1799                 print_power_limit_msr(cpu, msr, "PKG Limit #1");
1800                 fprintf(stderr, "cpu%d: PKG Limit #2: %sabled (%f Watts, %f* sec, clamp %sabled)\n",
1801                         cpu,
1802                         ((msr >> 47) & 1) ? "EN" : "DIS",
1803                         ((msr >> 32) & 0x7FFF) * rapl_power_units,
1804                         (1.0 + (((msr >> 54) & 0x3)/4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units,
1805                         ((msr >> 48) & 1) ? "EN" : "DIS");
1806         }
1807
1808         if (do_rapl & RAPL_DRAM) {
1809                 if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr))
1810                         return -6;
1811
1812
1813                 fprintf(stderr, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
1814                         cpu, msr,
1815                         ((msr >>  0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1816                         ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1817                         ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1818                         ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
1819
1820
1821                 if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr))
1822                         return -9;
1823                 fprintf(stderr, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n",
1824                                 cpu, msr, (msr >> 31) & 1 ? "": "UN");
1825
1826                 print_power_limit_msr(cpu, msr, "DRAM Limit");
1827         }
1828         if (do_rapl & RAPL_CORE_POLICY) {
1829                 if (verbose) {
1830                         if (get_msr(cpu, MSR_PP0_POLICY, &msr))
1831                                 return -7;
1832
1833                         fprintf(stderr, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF);
1834                 }
1835         }
1836         if (do_rapl & RAPL_CORES) {
1837                 if (verbose) {
1838
1839                         if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr))
1840                                 return -9;
1841                         fprintf(stderr, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n",
1842                                         cpu, msr, (msr >> 31) & 1 ? "": "UN");
1843                         print_power_limit_msr(cpu, msr, "Cores Limit");
1844                 }
1845         }
1846         if (do_rapl & RAPL_GFX) {
1847                 if (verbose) {
1848                         if (get_msr(cpu, MSR_PP1_POLICY, &msr))
1849                                 return -8;
1850
1851                         fprintf(stderr, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF);
1852
1853                         if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr))
1854                                 return -9;
1855                         fprintf(stderr, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n",
1856                                         cpu, msr, (msr >> 31) & 1 ? "": "UN");
1857                         print_power_limit_msr(cpu, msr, "GFX Limit");
1858                 }
1859         }
1860         return 0;
1861 }
1862
1863
1864 int is_snb(unsigned int family, unsigned int model)
1865 {
1866         if (!genuine_intel)
1867                 return 0;
1868
1869         switch (model) {
1870         case 0x2A:
1871         case 0x2D:
1872         case 0x3A:      /* IVB */
1873         case 0x3E:      /* IVB Xeon */
1874         case 0x3C:      /* HSW */
1875         case 0x3F:      /* HSW */
1876         case 0x45:      /* HSW */
1877         case 0x46:      /* HSW */
1878                 return 1;
1879         }
1880         return 0;
1881 }
1882
1883 int has_c8_c9_c10(unsigned int family, unsigned int model)
1884 {
1885         if (!genuine_intel)
1886                 return 0;
1887
1888         switch (model) {
1889         case 0x45:
1890                 return 1;
1891         }
1892         return 0;
1893 }
1894
1895
1896 int is_slm(unsigned int family, unsigned int model)
1897 {
1898         if (!genuine_intel)
1899                 return 0;
1900         switch (model) {
1901         case 0x37:      /* BYT */
1902         case 0x4D:      /* AVN */
1903                 return 1;
1904         }
1905         return 0;
1906 }
1907
1908 #define SLM_BCLK_FREQS 5
1909 double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0};
1910
1911 double slm_bclk(void)
1912 {
1913         unsigned long long msr = 3;
1914         unsigned int i;
1915         double freq;
1916
1917         if (get_msr(0, MSR_FSB_FREQ, &msr))
1918                 fprintf(stderr, "SLM BCLK: unknown\n");
1919
1920         i = msr & 0xf;
1921         if (i >= SLM_BCLK_FREQS) {
1922                 fprintf(stderr, "SLM BCLK[%d] invalid\n", i);
1923                 msr = 3;
1924         }
1925         freq = slm_freq_table[i];
1926
1927         fprintf(stderr, "SLM BCLK: %.1f Mhz\n", freq);
1928
1929         return freq;
1930 }
1931
1932 double discover_bclk(unsigned int family, unsigned int model)
1933 {
1934         if (is_snb(family, model))
1935                 return 100.00;
1936         else if (is_slm(family, model))
1937                 return slm_bclk();
1938         else
1939                 return 133.33;
1940 }
1941
1942 /*
1943  * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
1944  * the Thermal Control Circuit (TCC) activates.
1945  * This is usually equal to tjMax.
1946  *
1947  * Older processors do not have this MSR, so there we guess,
1948  * but also allow cmdline over-ride with -T.
1949  *
1950  * Several MSR temperature values are in units of degrees-C
1951  * below this value, including the Digital Thermal Sensor (DTS),
1952  * Package Thermal Management Sensor (PTM), and thermal event thresholds.
1953  */
1954 int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1955 {
1956         unsigned long long msr;
1957         unsigned int target_c_local;
1958         int cpu;
1959
1960         /* tcc_activation_temp is used only for dts or ptm */
1961         if (!(do_dts || do_ptm))
1962                 return 0;
1963
1964         /* this is a per-package concept */
1965         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1966                 return 0;
1967
1968         cpu = t->cpu_id;
1969         if (cpu_migrate(cpu)) {
1970                 fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
1971                 return -1;
1972         }
1973
1974         if (tcc_activation_temp_override != 0) {
1975                 tcc_activation_temp = tcc_activation_temp_override;
1976                 fprintf(stderr, "cpu%d: Using cmdline TCC Target (%d C)\n",
1977                         cpu, tcc_activation_temp);
1978                 return 0;
1979         }
1980
1981         /* Temperature Target MSR is Nehalem and newer only */
1982         if (!do_nehalem_platform_info)
1983                 goto guess;
1984
1985         if (get_msr(0, MSR_IA32_TEMPERATURE_TARGET, &msr))
1986                 goto guess;
1987
1988         target_c_local = (msr >> 16) & 0x7F;
1989
1990         if (verbose)
1991                 fprintf(stderr, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n",
1992                         cpu, msr, target_c_local);
1993
1994         if (target_c_local < 85 || target_c_local > 127)
1995                 goto guess;
1996
1997         tcc_activation_temp = target_c_local;
1998
1999         return 0;
2000
2001 guess:
2002         tcc_activation_temp = TJMAX_DEFAULT;
2003         fprintf(stderr, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
2004                 cpu, tcc_activation_temp);
2005
2006         return 0;
2007 }
2008 void check_cpuid()
2009 {
2010         unsigned int eax, ebx, ecx, edx, max_level;
2011         unsigned int fms, family, model, stepping;
2012
2013         eax = ebx = ecx = edx = 0;
2014
2015         __get_cpuid(0, &max_level, &ebx, &ecx, &edx);
2016
2017         if (ebx == 0x756e6547 && edx == 0x49656e69 && ecx == 0x6c65746e)
2018                 genuine_intel = 1;
2019
2020         if (verbose)
2021                 fprintf(stderr, "CPUID(0): %.4s%.4s%.4s ",
2022                         (char *)&ebx, (char *)&edx, (char *)&ecx);
2023
2024         __get_cpuid(1, &fms, &ebx, &ecx, &edx);
2025         family = (fms >> 8) & 0xf;
2026         model = (fms >> 4) & 0xf;
2027         stepping = fms & 0xf;
2028         if (family == 6 || family == 0xf)
2029                 model += ((fms >> 16) & 0xf) << 4;
2030
2031         if (verbose)
2032                 fprintf(stderr, "%d CPUID levels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n",
2033                         max_level, family, model, stepping, family, model, stepping);
2034
2035         if (!(edx & (1 << 5)))
2036                 errx(1, "CPUID: no MSR");
2037
2038         /*
2039          * check max extended function levels of CPUID.
2040          * This is needed to check for invariant TSC.
2041          * This check is valid for both Intel and AMD.
2042          */
2043         ebx = ecx = edx = 0;
2044         __get_cpuid(0x80000000, &max_level, &ebx, &ecx, &edx);
2045
2046         if (max_level < 0x80000007)
2047                 errx(1, "CPUID: no invariant TSC (max_level 0x%x)", max_level);
2048
2049         /*
2050          * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
2051          * this check is valid for both Intel and AMD
2052          */
2053         __get_cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
2054         has_invariant_tsc = edx & (1 << 8);
2055
2056         if (!has_invariant_tsc)
2057                 errx(1, "No invariant TSC");
2058
2059         /*
2060          * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
2061          * this check is valid for both Intel and AMD
2062          */
2063
2064         __get_cpuid(0x6, &eax, &ebx, &ecx, &edx);
2065         has_aperf = ecx & (1 << 0);
2066         do_dts = eax & (1 << 0);
2067         do_ptm = eax & (1 << 6);
2068         has_epb = ecx & (1 << 3);
2069
2070         if (verbose)
2071                 fprintf(stderr, "CPUID(6): %s%s%s%s\n",
2072                         has_aperf ? "APERF" : "No APERF!",
2073                         do_dts ? ", DTS" : "",
2074                         do_ptm ? ", PTM": "",
2075                         has_epb ? ", EPB": "");
2076
2077         if (!has_aperf)
2078                 errx(-1, "No APERF");
2079
2080         do_nehalem_platform_info = genuine_intel && has_invariant_tsc;
2081         do_nhm_cstates = genuine_intel; /* all Intel w/ non-stop TSC have NHM counters */
2082         do_smi = do_nhm_cstates;
2083         do_snb_cstates = is_snb(family, model);
2084         do_c8_c9_c10 = has_c8_c9_c10(family, model);
2085         do_slm_cstates = is_slm(family, model);
2086         bclk = discover_bclk(family, model);
2087
2088         do_nehalem_turbo_ratio_limit = has_nehalem_turbo_ratio_limit(family, model);
2089         do_ivt_turbo_ratio_limit = has_ivt_turbo_ratio_limit(family, model);
2090         rapl_probe(family, model);
2091
2092         return;
2093 }
2094
2095
2096 void usage()
2097 {
2098         errx(1, "%s: [-v][-R][-T][-p|-P|-S][-c MSR#][-C MSR#][-m MSR#][-M MSR#][-i interval_sec | command ...]\n",
2099              progname);
2100 }
2101
2102
2103 /*
2104  * in /dev/cpu/ return success for names that are numbers
2105  * ie. filter out ".", "..", "microcode".
2106  */
2107 int dir_filter(const struct dirent *dirp)
2108 {
2109         if (isdigit(dirp->d_name[0]))
2110                 return 1;
2111         else
2112                 return 0;
2113 }
2114
2115 int open_dev_cpu_msr(int dummy1)
2116 {
2117         return 0;
2118 }
2119
2120 void topology_probe()
2121 {
2122         int i;
2123         int max_core_id = 0;
2124         int max_package_id = 0;
2125         int max_siblings = 0;
2126         struct cpu_topology {
2127                 int core_id;
2128                 int physical_package_id;
2129         } *cpus;
2130
2131         /* Initialize num_cpus, max_cpu_num */
2132         topo.num_cpus = 0;
2133         topo.max_cpu_num = 0;
2134         for_all_proc_cpus(count_cpus);
2135         if (!summary_only && topo.num_cpus > 1)
2136                 show_cpu = 1;
2137
2138         if (verbose > 1)
2139                 fprintf(stderr, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
2140
2141         cpus = calloc(1, (topo.max_cpu_num  + 1) * sizeof(struct cpu_topology));
2142         if (cpus == NULL)
2143                 err(1, "calloc cpus");
2144
2145         /*
2146          * Allocate and initialize cpu_present_set
2147          */
2148         cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
2149         if (cpu_present_set == NULL)
2150                 err(3, "CPU_ALLOC");
2151         cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
2152         CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
2153         for_all_proc_cpus(mark_cpu_present);
2154
2155         /*
2156          * Allocate and initialize cpu_affinity_set
2157          */
2158         cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
2159         if (cpu_affinity_set == NULL)
2160                 err(3, "CPU_ALLOC");
2161         cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
2162         CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
2163
2164
2165         /*
2166          * For online cpus
2167          * find max_core_id, max_package_id
2168          */
2169         for (i = 0; i <= topo.max_cpu_num; ++i) {
2170                 int siblings;
2171
2172                 if (cpu_is_not_present(i)) {
2173                         if (verbose > 1)
2174                                 fprintf(stderr, "cpu%d NOT PRESENT\n", i);
2175                         continue;
2176                 }
2177                 cpus[i].core_id = get_core_id(i);
2178                 if (cpus[i].core_id > max_core_id)
2179                         max_core_id = cpus[i].core_id;
2180
2181                 cpus[i].physical_package_id = get_physical_package_id(i);
2182                 if (cpus[i].physical_package_id > max_package_id)
2183                         max_package_id = cpus[i].physical_package_id;
2184
2185                 siblings = get_num_ht_siblings(i);
2186                 if (siblings > max_siblings)
2187                         max_siblings = siblings;
2188                 if (verbose > 1)
2189                         fprintf(stderr, "cpu %d pkg %d core %d\n",
2190                                 i, cpus[i].physical_package_id, cpus[i].core_id);
2191         }
2192         topo.num_cores_per_pkg = max_core_id + 1;
2193         if (verbose > 1)
2194                 fprintf(stderr, "max_core_id %d, sizing for %d cores per package\n",
2195                         max_core_id, topo.num_cores_per_pkg);
2196         if (!summary_only && topo.num_cores_per_pkg > 1)
2197                 show_core = 1;
2198
2199         topo.num_packages = max_package_id + 1;
2200         if (verbose > 1)
2201                 fprintf(stderr, "max_package_id %d, sizing for %d packages\n",
2202                         max_package_id, topo.num_packages);
2203         if (!summary_only && topo.num_packages > 1)
2204                 show_pkg = 1;
2205
2206         topo.num_threads_per_core = max_siblings;
2207         if (verbose > 1)
2208                 fprintf(stderr, "max_siblings %d\n", max_siblings);
2209
2210         free(cpus);
2211 }
2212
2213 void
2214 allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p)
2215 {
2216         int i;
2217
2218         *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg *
2219                 topo.num_packages, sizeof(struct thread_data));
2220         if (*t == NULL)
2221                 goto error;
2222
2223         for (i = 0; i < topo.num_threads_per_core *
2224                 topo.num_cores_per_pkg * topo.num_packages; i++)
2225                 (*t)[i].cpu_id = -1;
2226
2227         *c = calloc(topo.num_cores_per_pkg * topo.num_packages,
2228                 sizeof(struct core_data));
2229         if (*c == NULL)
2230                 goto error;
2231
2232         for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++)
2233                 (*c)[i].core_id = -1;
2234
2235         *p = calloc(topo.num_packages, sizeof(struct pkg_data));
2236         if (*p == NULL)
2237                 goto error;
2238
2239         for (i = 0; i < topo.num_packages; i++)
2240                 (*p)[i].package_id = i;
2241
2242         return;
2243 error:
2244         err(1, "calloc counters");
2245 }
2246 /*
2247  * init_counter()
2248  *
2249  * set cpu_id, core_num, pkg_num
2250  * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
2251  *
2252  * increment topo.num_cores when 1st core in pkg seen
2253  */
2254 void init_counter(struct thread_data *thread_base, struct core_data *core_base,
2255         struct pkg_data *pkg_base, int thread_num, int core_num,
2256         int pkg_num, int cpu_id)
2257 {
2258         struct thread_data *t;
2259         struct core_data *c;
2260         struct pkg_data *p;
2261
2262         t = GET_THREAD(thread_base, thread_num, core_num, pkg_num);
2263         c = GET_CORE(core_base, core_num, pkg_num);
2264         p = GET_PKG(pkg_base, pkg_num);
2265
2266         t->cpu_id = cpu_id;
2267         if (thread_num == 0) {
2268                 t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
2269                 if (cpu_is_first_core_in_package(cpu_id))
2270                         t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
2271         }
2272
2273         c->core_id = core_num;
2274         p->package_id = pkg_num;
2275 }
2276
2277
2278 int initialize_counters(int cpu_id)
2279 {
2280         int my_thread_id, my_core_id, my_package_id;
2281
2282         my_package_id = get_physical_package_id(cpu_id);
2283         my_core_id = get_core_id(cpu_id);
2284
2285         if (cpu_is_first_sibling_in_core(cpu_id)) {
2286                 my_thread_id = 0;
2287                 topo.num_cores++;
2288         } else {
2289                 my_thread_id = 1;
2290         }
2291
2292         init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
2293         init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
2294         return 0;
2295 }
2296
2297 void allocate_output_buffer()
2298 {
2299         output_buffer = calloc(1, (1 + topo.num_cpus) * 1024);
2300         outp = output_buffer;
2301         if (outp == NULL)
2302                 err(-1, "calloc output buffer");
2303 }
2304
2305 void setup_all_buffers(void)
2306 {
2307         topology_probe();
2308         allocate_counters(&thread_even, &core_even, &package_even);
2309         allocate_counters(&thread_odd, &core_odd, &package_odd);
2310         allocate_output_buffer();
2311         for_all_proc_cpus(initialize_counters);
2312 }
2313
2314 void turbostat_init()
2315 {
2316         check_cpuid();
2317
2318         check_dev_msr();
2319         check_super_user();
2320
2321         setup_all_buffers();
2322
2323         if (verbose)
2324                 print_verbose_header();
2325
2326         if (verbose)
2327                 for_all_cpus(print_epb, ODD_COUNTERS);
2328
2329         if (verbose)
2330                 for_all_cpus(print_rapl, ODD_COUNTERS);
2331
2332         for_all_cpus(set_temperature_target, ODD_COUNTERS);
2333
2334         if (verbose)
2335                 for_all_cpus(print_thermal, ODD_COUNTERS);
2336 }
2337
2338 int fork_it(char **argv)
2339 {
2340         pid_t child_pid;
2341         int status;
2342
2343         status = for_all_cpus(get_counters, EVEN_COUNTERS);
2344         if (status)
2345                 exit(status);
2346         /* clear affinity side-effect of get_counters() */
2347         sched_setaffinity(0, cpu_present_setsize, cpu_present_set);
2348         gettimeofday(&tv_even, (struct timezone *)NULL);
2349
2350         child_pid = fork();
2351         if (!child_pid) {
2352                 /* child */
2353                 execvp(argv[0], argv);
2354         } else {
2355
2356                 /* parent */
2357                 if (child_pid == -1)
2358                         err(1, "fork");
2359
2360                 signal(SIGINT, SIG_IGN);
2361                 signal(SIGQUIT, SIG_IGN);
2362                 if (waitpid(child_pid, &status, 0) == -1)
2363                         err(status, "waitpid");
2364         }
2365         /*
2366          * n.b. fork_it() does not check for errors from for_all_cpus()
2367          * because re-starting is problematic when forking
2368          */
2369         for_all_cpus(get_counters, ODD_COUNTERS);
2370         gettimeofday(&tv_odd, (struct timezone *)NULL);
2371         timersub(&tv_odd, &tv_even, &tv_delta);
2372         for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS);
2373         compute_average(EVEN_COUNTERS);
2374         format_all_counters(EVEN_COUNTERS);
2375         flush_stderr();
2376
2377         fprintf(stderr, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);
2378
2379         return status;
2380 }
2381
2382 int get_and_dump_counters(void)
2383 {
2384         int status;
2385
2386         status = for_all_cpus(get_counters, ODD_COUNTERS);
2387         if (status)
2388                 return status;
2389
2390         status = for_all_cpus(dump_counters, ODD_COUNTERS);
2391         if (status)
2392                 return status;
2393
2394         flush_stdout();
2395
2396         return status;
2397 }
2398
2399 void cmdline(int argc, char **argv)
2400 {
2401         int opt;
2402
2403         progname = argv[0];
2404
2405         while ((opt = getopt(argc, argv, "+pPsSvi:c:C:m:M:RJT:")) != -1) {
2406                 switch (opt) {
2407                 case 'p':
2408                         show_core_only++;
2409                         break;
2410                 case 'P':
2411                         show_pkg_only++;
2412                         break;
2413                 case 's':
2414                         dump_only++;
2415                         break;
2416                 case 'S':
2417                         summary_only++;
2418                         break;
2419                 case 'v':
2420                         verbose++;
2421                         break;
2422                 case 'i':
2423                         interval_sec = atoi(optarg);
2424                         break;
2425                 case 'c':
2426                         sscanf(optarg, "%x", &extra_delta_offset32);
2427                         break;
2428                 case 'C':
2429                         sscanf(optarg, "%x", &extra_delta_offset64);
2430                         break;
2431                 case 'm':
2432                         sscanf(optarg, "%x", &extra_msr_offset32);
2433                         break;
2434                 case 'M':
2435                         sscanf(optarg, "%x", &extra_msr_offset64);
2436                         break;
2437                 case 'R':
2438                         rapl_verbose++;
2439                         break;
2440                 case 'T':
2441                         tcc_activation_temp_override = atoi(optarg);
2442                         break;
2443                 case 'J':
2444                         rapl_joules++;
2445                         break;
2446
2447                 default:
2448                         usage();
2449                 }
2450         }
2451 }
2452
2453 int main(int argc, char **argv)
2454 {
2455         cmdline(argc, argv);
2456
2457         if (verbose)
2458                 fprintf(stderr, "turbostat v3.6 Dec 2, 2013"
2459                         " - Len Brown <lenb@kernel.org>\n");
2460
2461         turbostat_init();
2462
2463         /* dump counters and exit */
2464         if (dump_only)
2465                 return get_and_dump_counters();
2466
2467         /*
2468          * if any params left, it must be a command to fork
2469          */
2470         if (argc - optind)
2471                 return fork_it(argv + optind);
2472         else
2473                 turbostat_loop();
2474
2475         return 0;
2476 }