Commit | Line | Data |
---|---|---|
103a8fea LB |
1 | /* |
2 | * turbostat -- show CPU frequency and C-state residency | |
3 | * on modern Intel turbo-capable processors. | |
4 | * | |
144b44b1 | 5 | * Copyright (c) 2013 Intel Corporation. |
103a8fea LB |
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 | ||
88c3281f | 22 | #define _GNU_SOURCE |
b731f311 | 23 | #include MSRHEADER |
95aebc44 | 24 | #include <stdarg.h> |
103a8fea | 25 | #include <stdio.h> |
b2c95d90 | 26 | #include <err.h> |
103a8fea LB |
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> | |
d8af6f5f | 36 | #include <getopt.h> |
103a8fea LB |
37 | #include <dirent.h> |
38 | #include <string.h> | |
39 | #include <ctype.h> | |
88c3281f | 40 | #include <sched.h> |
2b92865e | 41 | #include <cpuid.h> |
98481e79 LB |
42 | #include <linux/capability.h> |
43 | #include <errno.h> | |
103a8fea | 44 | |
103a8fea | 45 | char *proc_stat = "/proc/stat"; |
d8af6f5f LB |
46 | unsigned int interval_sec = 5; |
47 | unsigned int debug; | |
48 | unsigned int rapl_joules; | |
49 | unsigned int summary_only; | |
50 | unsigned int dump_only; | |
103a8fea LB |
51 | unsigned int skip_c0; |
52 | unsigned int skip_c1; | |
53 | unsigned int do_nhm_cstates; | |
54 | unsigned int do_snb_cstates; | |
fb5d4327 | 55 | unsigned int do_knl_cstates; |
ee7e38e3 LB |
56 | unsigned int do_pc2; |
57 | unsigned int do_pc3; | |
58 | unsigned int do_pc6; | |
59 | unsigned int do_pc7; | |
ca58710f | 60 | unsigned int do_c8_c9_c10; |
0b2bb692 | 61 | unsigned int do_skl_residency; |
144b44b1 LB |
62 | unsigned int do_slm_cstates; |
63 | unsigned int use_c1_residency_msr; | |
103a8fea | 64 | unsigned int has_aperf; |
889facbe | 65 | unsigned int has_epb; |
fc04cc67 | 66 | unsigned int units = 1000000; /* MHz etc */ |
103a8fea LB |
67 | unsigned int genuine_intel; |
68 | unsigned int has_invariant_tsc; | |
d7899447 | 69 | unsigned int do_nhm_platform_info; |
2f32edf1 LB |
70 | unsigned int extra_msr_offset32; |
71 | unsigned int extra_msr_offset64; | |
8e180f3c LB |
72 | unsigned int extra_delta_offset32; |
73 | unsigned int extra_delta_offset64; | |
b2b34dfe | 74 | unsigned int aperf_mperf_multiplier = 1; |
1ed51011 | 75 | int do_smi; |
103a8fea | 76 | double bclk; |
a2b7b749 | 77 | double base_hz; |
21ed5574 | 78 | unsigned int has_base_hz; |
a2b7b749 | 79 | double tsc_tweak = 1.0; |
103a8fea LB |
80 | unsigned int show_pkg; |
81 | unsigned int show_core; | |
82 | unsigned int show_cpu; | |
c98d5d94 LB |
83 | unsigned int show_pkg_only; |
84 | unsigned int show_core_only; | |
85 | char *output_buffer, *outp; | |
889facbe LB |
86 | unsigned int do_rapl; |
87 | unsigned int do_dts; | |
88 | unsigned int do_ptm; | |
89 | unsigned int tcc_activation_temp; | |
90 | unsigned int tcc_activation_temp_override; | |
40ee8e3b AS |
91 | double rapl_power_units, rapl_time_units; |
92 | double rapl_dram_energy_units, rapl_energy_units; | |
889facbe | 93 | double rapl_joule_counter_range; |
3a9a941d LB |
94 | unsigned int do_core_perf_limit_reasons; |
95 | unsigned int do_gfx_perf_limit_reasons; | |
96 | unsigned int do_ring_perf_limit_reasons; | |
8a5bdf41 LB |
97 | unsigned int crystal_hz; |
98 | unsigned long long tsc_hz; | |
7ce7d5de | 99 | int base_cpu; |
21ed5574 | 100 | double discover_bclk(unsigned int family, unsigned int model); |
7f5c258e LB |
101 | unsigned int has_hwp; /* IA32_PM_ENABLE, IA32_HWP_CAPABILITIES */ |
102 | /* IA32_HWP_REQUEST, IA32_HWP_STATUS */ | |
103 | unsigned int has_hwp_notify; /* IA32_HWP_INTERRUPT */ | |
104 | unsigned int has_hwp_activity_window; /* IA32_HWP_REQUEST[bits 41:32] */ | |
105 | unsigned int has_hwp_epp; /* IA32_HWP_REQUEST[bits 31:24] */ | |
106 | unsigned int has_hwp_pkg; /* IA32_HWP_REQUEST_PKG */ | |
889facbe | 107 | |
e6f9bb3c LB |
108 | #define RAPL_PKG (1 << 0) |
109 | /* 0x610 MSR_PKG_POWER_LIMIT */ | |
110 | /* 0x611 MSR_PKG_ENERGY_STATUS */ | |
111 | #define RAPL_PKG_PERF_STATUS (1 << 1) | |
112 | /* 0x613 MSR_PKG_PERF_STATUS */ | |
113 | #define RAPL_PKG_POWER_INFO (1 << 2) | |
114 | /* 0x614 MSR_PKG_POWER_INFO */ | |
115 | ||
116 | #define RAPL_DRAM (1 << 3) | |
117 | /* 0x618 MSR_DRAM_POWER_LIMIT */ | |
118 | /* 0x619 MSR_DRAM_ENERGY_STATUS */ | |
e6f9bb3c LB |
119 | #define RAPL_DRAM_PERF_STATUS (1 << 4) |
120 | /* 0x61b MSR_DRAM_PERF_STATUS */ | |
0b2bb692 LB |
121 | #define RAPL_DRAM_POWER_INFO (1 << 5) |
122 | /* 0x61c MSR_DRAM_POWER_INFO */ | |
e6f9bb3c | 123 | |
0b2bb692 | 124 | #define RAPL_CORES (1 << 6) |
e6f9bb3c LB |
125 | /* 0x638 MSR_PP0_POWER_LIMIT */ |
126 | /* 0x639 MSR_PP0_ENERGY_STATUS */ | |
0b2bb692 | 127 | #define RAPL_CORE_POLICY (1 << 7) |
e6f9bb3c LB |
128 | /* 0x63a MSR_PP0_POLICY */ |
129 | ||
0b2bb692 | 130 | #define RAPL_GFX (1 << 8) |
e6f9bb3c LB |
131 | /* 0x640 MSR_PP1_POWER_LIMIT */ |
132 | /* 0x641 MSR_PP1_ENERGY_STATUS */ | |
133 | /* 0x642 MSR_PP1_POLICY */ | |
889facbe LB |
134 | #define TJMAX_DEFAULT 100 |
135 | ||
136 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
103a8fea LB |
137 | |
138 | int aperf_mperf_unstable; | |
139 | int backwards_count; | |
140 | char *progname; | |
103a8fea | 141 | |
c98d5d94 LB |
142 | cpu_set_t *cpu_present_set, *cpu_affinity_set; |
143 | size_t cpu_present_setsize, cpu_affinity_setsize; | |
144 | ||
145 | struct thread_data { | |
146 | unsigned long long tsc; | |
147 | unsigned long long aperf; | |
148 | unsigned long long mperf; | |
144b44b1 | 149 | unsigned long long c1; |
2f32edf1 | 150 | unsigned long long extra_msr64; |
8e180f3c LB |
151 | unsigned long long extra_delta64; |
152 | unsigned long long extra_msr32; | |
153 | unsigned long long extra_delta32; | |
1ed51011 | 154 | unsigned int smi_count; |
c98d5d94 LB |
155 | unsigned int cpu_id; |
156 | unsigned int flags; | |
157 | #define CPU_IS_FIRST_THREAD_IN_CORE 0x2 | |
158 | #define CPU_IS_FIRST_CORE_IN_PACKAGE 0x4 | |
159 | } *thread_even, *thread_odd; | |
160 | ||
161 | struct core_data { | |
162 | unsigned long long c3; | |
163 | unsigned long long c6; | |
164 | unsigned long long c7; | |
889facbe | 165 | unsigned int core_temp_c; |
c98d5d94 LB |
166 | unsigned int core_id; |
167 | } *core_even, *core_odd; | |
168 | ||
169 | struct pkg_data { | |
170 | unsigned long long pc2; | |
171 | unsigned long long pc3; | |
172 | unsigned long long pc6; | |
173 | unsigned long long pc7; | |
ca58710f KCA |
174 | unsigned long long pc8; |
175 | unsigned long long pc9; | |
176 | unsigned long long pc10; | |
0b2bb692 LB |
177 | unsigned long long pkg_wtd_core_c0; |
178 | unsigned long long pkg_any_core_c0; | |
179 | unsigned long long pkg_any_gfxe_c0; | |
180 | unsigned long long pkg_both_core_gfxe_c0; | |
c98d5d94 | 181 | unsigned int package_id; |
889facbe LB |
182 | unsigned int energy_pkg; /* MSR_PKG_ENERGY_STATUS */ |
183 | unsigned int energy_dram; /* MSR_DRAM_ENERGY_STATUS */ | |
184 | unsigned int energy_cores; /* MSR_PP0_ENERGY_STATUS */ | |
185 | unsigned int energy_gfx; /* MSR_PP1_ENERGY_STATUS */ | |
186 | unsigned int rapl_pkg_perf_status; /* MSR_PKG_PERF_STATUS */ | |
187 | unsigned int rapl_dram_perf_status; /* MSR_DRAM_PERF_STATUS */ | |
188 | unsigned int pkg_temp_c; | |
189 | ||
c98d5d94 LB |
190 | } *package_even, *package_odd; |
191 | ||
192 | #define ODD_COUNTERS thread_odd, core_odd, package_odd | |
193 | #define EVEN_COUNTERS thread_even, core_even, package_even | |
194 | ||
195 | #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \ | |
196 | (thread_base + (pkg_no) * topo.num_cores_per_pkg * \ | |
197 | topo.num_threads_per_core + \ | |
198 | (core_no) * topo.num_threads_per_core + (thread_no)) | |
199 | #define GET_CORE(core_base, core_no, pkg_no) \ | |
200 | (core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no)) | |
201 | #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no) | |
202 | ||
203 | struct system_summary { | |
204 | struct thread_data threads; | |
205 | struct core_data cores; | |
206 | struct pkg_data packages; | |
207 | } sum, average; | |
208 | ||
209 | ||
210 | struct topo_params { | |
211 | int num_packages; | |
212 | int num_cpus; | |
213 | int num_cores; | |
214 | int max_cpu_num; | |
215 | int num_cores_per_pkg; | |
216 | int num_threads_per_core; | |
217 | } topo; | |
218 | ||
219 | struct timeval tv_even, tv_odd, tv_delta; | |
220 | ||
221 | void setup_all_buffers(void); | |
222 | ||
223 | int cpu_is_not_present(int cpu) | |
d15cf7c1 | 224 | { |
c98d5d94 | 225 | return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set); |
d15cf7c1 | 226 | } |
88c3281f | 227 | /* |
c98d5d94 LB |
228 | * run func(thread, core, package) in topology order |
229 | * skip non-present cpus | |
88c3281f | 230 | */ |
c98d5d94 LB |
231 | |
232 | int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *), | |
233 | struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base) | |
88c3281f | 234 | { |
c98d5d94 | 235 | int retval, pkg_no, core_no, thread_no; |
d15cf7c1 | 236 | |
c98d5d94 LB |
237 | for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) { |
238 | for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) { | |
239 | for (thread_no = 0; thread_no < | |
240 | topo.num_threads_per_core; ++thread_no) { | |
241 | struct thread_data *t; | |
242 | struct core_data *c; | |
243 | struct pkg_data *p; | |
88c3281f | 244 | |
c98d5d94 LB |
245 | t = GET_THREAD(thread_base, thread_no, core_no, pkg_no); |
246 | ||
247 | if (cpu_is_not_present(t->cpu_id)) | |
248 | continue; | |
249 | ||
250 | c = GET_CORE(core_base, core_no, pkg_no); | |
251 | p = GET_PKG(pkg_base, pkg_no); | |
252 | ||
253 | retval = func(t, c, p); | |
254 | if (retval) | |
255 | return retval; | |
256 | } | |
257 | } | |
258 | } | |
259 | return 0; | |
88c3281f LB |
260 | } |
261 | ||
262 | int cpu_migrate(int cpu) | |
263 | { | |
c98d5d94 LB |
264 | CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set); |
265 | CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set); | |
266 | if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1) | |
88c3281f LB |
267 | return -1; |
268 | else | |
269 | return 0; | |
270 | } | |
271 | ||
15aaa346 | 272 | int get_msr(int cpu, off_t offset, unsigned long long *msr) |
103a8fea LB |
273 | { |
274 | ssize_t retval; | |
103a8fea LB |
275 | char pathname[32]; |
276 | int fd; | |
277 | ||
278 | sprintf(pathname, "/dev/cpu/%d/msr", cpu); | |
279 | fd = open(pathname, O_RDONLY); | |
15aaa346 | 280 | if (fd < 0) |
98481e79 | 281 | err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, or run as root", pathname); |
103a8fea | 282 | |
15aaa346 | 283 | retval = pread(fd, msr, sizeof *msr, offset); |
103a8fea | 284 | close(fd); |
15aaa346 | 285 | |
98481e79 LB |
286 | if (retval != sizeof *msr) |
287 | err(-1, "%s offset 0x%llx read failed", pathname, (unsigned long long)offset); | |
15aaa346 LB |
288 | |
289 | return 0; | |
103a8fea LB |
290 | } |
291 | ||
fc04cc67 LB |
292 | /* |
293 | * Example Format w/ field column widths: | |
294 | * | |
e7c95ff3 LB |
295 | * Package Core CPU Avg_MHz Bzy_MHz TSC_MHz SMI %Busy CPU_%c1 CPU_%c3 CPU_%c6 CPU_%c7 CoreTmp PkgTmp Pkg%pc2 Pkg%pc3 Pkg%pc6 Pkg%pc7 PkgWatt CorWatt GFXWatt |
296 | * 123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678 | |
fc04cc67 LB |
297 | */ |
298 | ||
a829eb4d | 299 | void print_header(void) |
103a8fea LB |
300 | { |
301 | if (show_pkg) | |
e7c95ff3 | 302 | outp += sprintf(outp, " Package"); |
103a8fea | 303 | if (show_core) |
e7c95ff3 | 304 | outp += sprintf(outp, " Core"); |
103a8fea | 305 | if (show_cpu) |
e7c95ff3 | 306 | outp += sprintf(outp, " CPU"); |
fc04cc67 | 307 | if (has_aperf) |
e7c95ff3 | 308 | outp += sprintf(outp, " Avg_MHz"); |
d7899447 | 309 | if (has_aperf) |
e7c95ff3 | 310 | outp += sprintf(outp, " %%Busy"); |
103a8fea | 311 | if (has_aperf) |
e7c95ff3 LB |
312 | outp += sprintf(outp, " Bzy_MHz"); |
313 | outp += sprintf(outp, " TSC_MHz"); | |
1cc21f7b | 314 | |
8e180f3c | 315 | if (extra_delta_offset32) |
e7c95ff3 | 316 | outp += sprintf(outp, " count 0x%03X", extra_delta_offset32); |
8e180f3c | 317 | if (extra_delta_offset64) |
e7c95ff3 | 318 | outp += sprintf(outp, " COUNT 0x%03X", extra_delta_offset64); |
2f32edf1 | 319 | if (extra_msr_offset32) |
e7c95ff3 | 320 | outp += sprintf(outp, " MSR 0x%03X", extra_msr_offset32); |
2f32edf1 | 321 | if (extra_msr_offset64) |
e7c95ff3 | 322 | outp += sprintf(outp, " MSR 0x%03X", extra_msr_offset64); |
1cc21f7b LB |
323 | |
324 | if (!debug) | |
325 | goto done; | |
326 | ||
327 | if (do_smi) | |
328 | outp += sprintf(outp, " SMI"); | |
329 | ||
103a8fea | 330 | if (do_nhm_cstates) |
e7c95ff3 | 331 | outp += sprintf(outp, " CPU%%c1"); |
fb5d4327 | 332 | if (do_nhm_cstates && !do_slm_cstates && !do_knl_cstates) |
e7c95ff3 | 333 | outp += sprintf(outp, " CPU%%c3"); |
103a8fea | 334 | if (do_nhm_cstates) |
e7c95ff3 | 335 | outp += sprintf(outp, " CPU%%c6"); |
103a8fea | 336 | if (do_snb_cstates) |
e7c95ff3 | 337 | outp += sprintf(outp, " CPU%%c7"); |
889facbe LB |
338 | |
339 | if (do_dts) | |
e7c95ff3 | 340 | outp += sprintf(outp, " CoreTmp"); |
889facbe | 341 | if (do_ptm) |
e7c95ff3 | 342 | outp += sprintf(outp, " PkgTmp"); |
889facbe | 343 | |
0b2bb692 LB |
344 | if (do_skl_residency) { |
345 | outp += sprintf(outp, " Totl%%C0"); | |
346 | outp += sprintf(outp, " Any%%C0"); | |
347 | outp += sprintf(outp, " GFX%%C0"); | |
348 | outp += sprintf(outp, " CPUGFX%%"); | |
349 | } | |
350 | ||
ee7e38e3 | 351 | if (do_pc2) |
e7c95ff3 | 352 | outp += sprintf(outp, " Pkg%%pc2"); |
ee7e38e3 | 353 | if (do_pc3) |
e7c95ff3 | 354 | outp += sprintf(outp, " Pkg%%pc3"); |
ee7e38e3 | 355 | if (do_pc6) |
e7c95ff3 | 356 | outp += sprintf(outp, " Pkg%%pc6"); |
ee7e38e3 | 357 | if (do_pc7) |
e7c95ff3 | 358 | outp += sprintf(outp, " Pkg%%pc7"); |
ca58710f | 359 | if (do_c8_c9_c10) { |
e7c95ff3 LB |
360 | outp += sprintf(outp, " Pkg%%pc8"); |
361 | outp += sprintf(outp, " Pkg%%pc9"); | |
362 | outp += sprintf(outp, " Pk%%pc10"); | |
ca58710f | 363 | } |
103a8fea | 364 | |
5c56be9a DB |
365 | if (do_rapl && !rapl_joules) { |
366 | if (do_rapl & RAPL_PKG) | |
e7c95ff3 | 367 | outp += sprintf(outp, " PkgWatt"); |
5c56be9a | 368 | if (do_rapl & RAPL_CORES) |
e7c95ff3 | 369 | outp += sprintf(outp, " CorWatt"); |
5c56be9a | 370 | if (do_rapl & RAPL_GFX) |
e7c95ff3 | 371 | outp += sprintf(outp, " GFXWatt"); |
5c56be9a | 372 | if (do_rapl & RAPL_DRAM) |
e7c95ff3 | 373 | outp += sprintf(outp, " RAMWatt"); |
5c56be9a | 374 | if (do_rapl & RAPL_PKG_PERF_STATUS) |
e7c95ff3 | 375 | outp += sprintf(outp, " PKG_%%"); |
5c56be9a | 376 | if (do_rapl & RAPL_DRAM_PERF_STATUS) |
e7c95ff3 | 377 | outp += sprintf(outp, " RAM_%%"); |
d7899447 | 378 | } else if (do_rapl && rapl_joules) { |
5c56be9a | 379 | if (do_rapl & RAPL_PKG) |
e7c95ff3 | 380 | outp += sprintf(outp, " Pkg_J"); |
5c56be9a | 381 | if (do_rapl & RAPL_CORES) |
e7c95ff3 | 382 | outp += sprintf(outp, " Cor_J"); |
5c56be9a | 383 | if (do_rapl & RAPL_GFX) |
e7c95ff3 | 384 | outp += sprintf(outp, " GFX_J"); |
5c56be9a | 385 | if (do_rapl & RAPL_DRAM) |
bd6906ed | 386 | outp += sprintf(outp, " RAM_J"); |
5c56be9a | 387 | if (do_rapl & RAPL_PKG_PERF_STATUS) |
e7c95ff3 | 388 | outp += sprintf(outp, " PKG_%%"); |
5c56be9a | 389 | if (do_rapl & RAPL_DRAM_PERF_STATUS) |
e7c95ff3 LB |
390 | outp += sprintf(outp, " RAM_%%"); |
391 | outp += sprintf(outp, " time"); | |
889facbe | 392 | |
5c56be9a | 393 | } |
1cc21f7b | 394 | done: |
c98d5d94 | 395 | outp += sprintf(outp, "\n"); |
103a8fea LB |
396 | } |
397 | ||
c98d5d94 LB |
398 | int dump_counters(struct thread_data *t, struct core_data *c, |
399 | struct pkg_data *p) | |
103a8fea | 400 | { |
3b4d5c7f | 401 | outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p); |
c98d5d94 LB |
402 | |
403 | if (t) { | |
3b4d5c7f AS |
404 | outp += sprintf(outp, "CPU: %d flags 0x%x\n", |
405 | t->cpu_id, t->flags); | |
406 | outp += sprintf(outp, "TSC: %016llX\n", t->tsc); | |
407 | outp += sprintf(outp, "aperf: %016llX\n", t->aperf); | |
408 | outp += sprintf(outp, "mperf: %016llX\n", t->mperf); | |
409 | outp += sprintf(outp, "c1: %016llX\n", t->c1); | |
410 | outp += sprintf(outp, "msr0x%x: %08llX\n", | |
8e180f3c | 411 | extra_delta_offset32, t->extra_delta32); |
3b4d5c7f | 412 | outp += sprintf(outp, "msr0x%x: %016llX\n", |
8e180f3c | 413 | extra_delta_offset64, t->extra_delta64); |
3b4d5c7f | 414 | outp += sprintf(outp, "msr0x%x: %08llX\n", |
2f32edf1 | 415 | extra_msr_offset32, t->extra_msr32); |
3b4d5c7f | 416 | outp += sprintf(outp, "msr0x%x: %016llX\n", |
2f32edf1 | 417 | extra_msr_offset64, t->extra_msr64); |
1ed51011 | 418 | if (do_smi) |
3b4d5c7f | 419 | outp += sprintf(outp, "SMI: %08X\n", t->smi_count); |
c98d5d94 | 420 | } |
103a8fea | 421 | |
c98d5d94 | 422 | if (c) { |
3b4d5c7f AS |
423 | outp += sprintf(outp, "core: %d\n", c->core_id); |
424 | outp += sprintf(outp, "c3: %016llX\n", c->c3); | |
425 | outp += sprintf(outp, "c6: %016llX\n", c->c6); | |
426 | outp += sprintf(outp, "c7: %016llX\n", c->c7); | |
427 | outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c); | |
c98d5d94 | 428 | } |
103a8fea | 429 | |
c98d5d94 | 430 | if (p) { |
3b4d5c7f | 431 | outp += sprintf(outp, "package: %d\n", p->package_id); |
0b2bb692 LB |
432 | |
433 | outp += sprintf(outp, "Weighted cores: %016llX\n", p->pkg_wtd_core_c0); | |
434 | outp += sprintf(outp, "Any cores: %016llX\n", p->pkg_any_core_c0); | |
435 | outp += sprintf(outp, "Any GFX: %016llX\n", p->pkg_any_gfxe_c0); | |
436 | outp += sprintf(outp, "CPU + GFX: %016llX\n", p->pkg_both_core_gfxe_c0); | |
437 | ||
3b4d5c7f | 438 | outp += sprintf(outp, "pc2: %016llX\n", p->pc2); |
ee7e38e3 LB |
439 | if (do_pc3) |
440 | outp += sprintf(outp, "pc3: %016llX\n", p->pc3); | |
441 | if (do_pc6) | |
442 | outp += sprintf(outp, "pc6: %016llX\n", p->pc6); | |
443 | if (do_pc7) | |
444 | outp += sprintf(outp, "pc7: %016llX\n", p->pc7); | |
3b4d5c7f AS |
445 | outp += sprintf(outp, "pc8: %016llX\n", p->pc8); |
446 | outp += sprintf(outp, "pc9: %016llX\n", p->pc9); | |
447 | outp += sprintf(outp, "pc10: %016llX\n", p->pc10); | |
448 | outp += sprintf(outp, "Joules PKG: %0X\n", p->energy_pkg); | |
449 | outp += sprintf(outp, "Joules COR: %0X\n", p->energy_cores); | |
450 | outp += sprintf(outp, "Joules GFX: %0X\n", p->energy_gfx); | |
451 | outp += sprintf(outp, "Joules RAM: %0X\n", p->energy_dram); | |
452 | outp += sprintf(outp, "Throttle PKG: %0X\n", | |
453 | p->rapl_pkg_perf_status); | |
454 | outp += sprintf(outp, "Throttle RAM: %0X\n", | |
455 | p->rapl_dram_perf_status); | |
456 | outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c); | |
c98d5d94 | 457 | } |
3b4d5c7f AS |
458 | |
459 | outp += sprintf(outp, "\n"); | |
460 | ||
c98d5d94 | 461 | return 0; |
103a8fea LB |
462 | } |
463 | ||
e23da037 LB |
464 | /* |
465 | * column formatting convention & formats | |
e23da037 | 466 | */ |
c98d5d94 LB |
467 | int format_counters(struct thread_data *t, struct core_data *c, |
468 | struct pkg_data *p) | |
103a8fea LB |
469 | { |
470 | double interval_float; | |
fc04cc67 | 471 | char *fmt8; |
103a8fea | 472 | |
c98d5d94 LB |
473 | /* if showing only 1st thread in core and this isn't one, bail out */ |
474 | if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
475 | return 0; | |
476 | ||
477 | /* if showing only 1st thread in pkg and this isn't one, bail out */ | |
478 | if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
479 | return 0; | |
480 | ||
103a8fea LB |
481 | interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0; |
482 | ||
c98d5d94 LB |
483 | /* topo columns, print blanks on 1st (average) line */ |
484 | if (t == &average.threads) { | |
103a8fea | 485 | if (show_pkg) |
fc04cc67 | 486 | outp += sprintf(outp, " -"); |
103a8fea | 487 | if (show_core) |
fc04cc67 | 488 | outp += sprintf(outp, " -"); |
103a8fea | 489 | if (show_cpu) |
fc04cc67 | 490 | outp += sprintf(outp, " -"); |
103a8fea | 491 | } else { |
c98d5d94 LB |
492 | if (show_pkg) { |
493 | if (p) | |
fc04cc67 | 494 | outp += sprintf(outp, "%8d", p->package_id); |
c98d5d94 | 495 | else |
fc04cc67 | 496 | outp += sprintf(outp, " -"); |
c98d5d94 | 497 | } |
c98d5d94 LB |
498 | if (show_core) { |
499 | if (c) | |
fc04cc67 | 500 | outp += sprintf(outp, "%8d", c->core_id); |
c98d5d94 | 501 | else |
fc04cc67 | 502 | outp += sprintf(outp, " -"); |
c98d5d94 | 503 | } |
103a8fea | 504 | if (show_cpu) |
fc04cc67 | 505 | outp += sprintf(outp, "%8d", t->cpu_id); |
103a8fea | 506 | } |
fc04cc67 | 507 | |
d7899447 | 508 | /* Avg_MHz */ |
fc04cc67 LB |
509 | if (has_aperf) |
510 | outp += sprintf(outp, "%8.0f", | |
511 | 1.0 / units * t->aperf / interval_float); | |
512 | ||
d7899447 LB |
513 | /* %Busy */ |
514 | if (has_aperf) { | |
103a8fea | 515 | if (!skip_c0) |
a2b7b749 | 516 | outp += sprintf(outp, "%8.2f", 100.0 * t->mperf/t->tsc/tsc_tweak); |
103a8fea | 517 | else |
fc04cc67 | 518 | outp += sprintf(outp, "********"); |
103a8fea LB |
519 | } |
520 | ||
d7899447 | 521 | /* Bzy_MHz */ |
21ed5574 LB |
522 | if (has_aperf) { |
523 | if (has_base_hz) | |
524 | outp += sprintf(outp, "%8.0f", base_hz / units * t->aperf / t->mperf); | |
525 | else | |
526 | outp += sprintf(outp, "%8.0f", | |
527 | 1.0 * t->tsc / units * t->aperf / t->mperf / interval_float); | |
528 | } | |
103a8fea | 529 | |
d7899447 | 530 | /* TSC_MHz */ |
fc04cc67 | 531 | outp += sprintf(outp, "%8.0f", 1.0 * t->tsc/units/interval_float); |
103a8fea | 532 | |
8e180f3c LB |
533 | /* delta */ |
534 | if (extra_delta_offset32) | |
535 | outp += sprintf(outp, " %11llu", t->extra_delta32); | |
536 | ||
537 | /* DELTA */ | |
538 | if (extra_delta_offset64) | |
539 | outp += sprintf(outp, " %11llu", t->extra_delta64); | |
2f32edf1 LB |
540 | /* msr */ |
541 | if (extra_msr_offset32) | |
8e180f3c | 542 | outp += sprintf(outp, " 0x%08llx", t->extra_msr32); |
2f32edf1 | 543 | |
130ff304 | 544 | /* MSR */ |
2f32edf1 LB |
545 | if (extra_msr_offset64) |
546 | outp += sprintf(outp, " 0x%016llx", t->extra_msr64); | |
130ff304 | 547 | |
1cc21f7b LB |
548 | if (!debug) |
549 | goto done; | |
550 | ||
551 | /* SMI */ | |
552 | if (do_smi) | |
553 | outp += sprintf(outp, "%8d", t->smi_count); | |
554 | ||
103a8fea LB |
555 | if (do_nhm_cstates) { |
556 | if (!skip_c1) | |
fc04cc67 | 557 | outp += sprintf(outp, "%8.2f", 100.0 * t->c1/t->tsc); |
103a8fea | 558 | else |
fc04cc67 | 559 | outp += sprintf(outp, "********"); |
103a8fea | 560 | } |
c98d5d94 LB |
561 | |
562 | /* print per-core data only for 1st thread in core */ | |
563 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
564 | goto done; | |
565 | ||
fb5d4327 | 566 | if (do_nhm_cstates && !do_slm_cstates && !do_knl_cstates) |
fc04cc67 | 567 | outp += sprintf(outp, "%8.2f", 100.0 * c->c3/t->tsc); |
103a8fea | 568 | if (do_nhm_cstates) |
fc04cc67 | 569 | outp += sprintf(outp, "%8.2f", 100.0 * c->c6/t->tsc); |
103a8fea | 570 | if (do_snb_cstates) |
fc04cc67 | 571 | outp += sprintf(outp, "%8.2f", 100.0 * c->c7/t->tsc); |
c98d5d94 | 572 | |
889facbe | 573 | if (do_dts) |
fc04cc67 | 574 | outp += sprintf(outp, "%8d", c->core_temp_c); |
889facbe | 575 | |
c98d5d94 LB |
576 | /* print per-package data only for 1st core in package */ |
577 | if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
578 | goto done; | |
579 | ||
0b2bb692 | 580 | /* PkgTmp */ |
889facbe | 581 | if (do_ptm) |
fc04cc67 | 582 | outp += sprintf(outp, "%8d", p->pkg_temp_c); |
889facbe | 583 | |
0b2bb692 LB |
584 | /* Totl%C0, Any%C0 GFX%C0 CPUGFX% */ |
585 | if (do_skl_residency) { | |
586 | outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_wtd_core_c0/t->tsc); | |
587 | outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_any_core_c0/t->tsc); | |
588 | outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_any_gfxe_c0/t->tsc); | |
589 | outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_both_core_gfxe_c0/t->tsc); | |
590 | } | |
591 | ||
ee7e38e3 | 592 | if (do_pc2) |
fc04cc67 | 593 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc2/t->tsc); |
ee7e38e3 | 594 | if (do_pc3) |
fc04cc67 | 595 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc3/t->tsc); |
ee7e38e3 | 596 | if (do_pc6) |
fc04cc67 | 597 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc6/t->tsc); |
ee7e38e3 | 598 | if (do_pc7) |
fc04cc67 | 599 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc7/t->tsc); |
ca58710f | 600 | if (do_c8_c9_c10) { |
fc04cc67 LB |
601 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc8/t->tsc); |
602 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc9/t->tsc); | |
603 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc10/t->tsc); | |
ca58710f | 604 | } |
889facbe LB |
605 | |
606 | /* | |
607 | * If measurement interval exceeds minimum RAPL Joule Counter range, | |
608 | * indicate that results are suspect by printing "**" in fraction place. | |
609 | */ | |
fc04cc67 LB |
610 | if (interval_float < rapl_joule_counter_range) |
611 | fmt8 = "%8.2f"; | |
612 | else | |
613 | fmt8 = " %6.0f**"; | |
889facbe | 614 | |
5c56be9a DB |
615 | if (do_rapl && !rapl_joules) { |
616 | if (do_rapl & RAPL_PKG) | |
fc04cc67 | 617 | outp += sprintf(outp, fmt8, p->energy_pkg * rapl_energy_units / interval_float); |
5c56be9a | 618 | if (do_rapl & RAPL_CORES) |
fc04cc67 | 619 | outp += sprintf(outp, fmt8, p->energy_cores * rapl_energy_units / interval_float); |
5c56be9a | 620 | if (do_rapl & RAPL_GFX) |
fc04cc67 | 621 | outp += sprintf(outp, fmt8, p->energy_gfx * rapl_energy_units / interval_float); |
5c56be9a | 622 | if (do_rapl & RAPL_DRAM) |
40ee8e3b | 623 | outp += sprintf(outp, fmt8, p->energy_dram * rapl_dram_energy_units / interval_float); |
5c56be9a | 624 | if (do_rapl & RAPL_PKG_PERF_STATUS) |
fc04cc67 | 625 | outp += sprintf(outp, fmt8, 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float); |
5c56be9a | 626 | if (do_rapl & RAPL_DRAM_PERF_STATUS) |
fc04cc67 | 627 | outp += sprintf(outp, fmt8, 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float); |
d7899447 | 628 | } else if (do_rapl && rapl_joules) { |
5c56be9a | 629 | if (do_rapl & RAPL_PKG) |
fc04cc67 | 630 | outp += sprintf(outp, fmt8, |
5c56be9a DB |
631 | p->energy_pkg * rapl_energy_units); |
632 | if (do_rapl & RAPL_CORES) | |
fc04cc67 | 633 | outp += sprintf(outp, fmt8, |
5c56be9a DB |
634 | p->energy_cores * rapl_energy_units); |
635 | if (do_rapl & RAPL_GFX) | |
fc04cc67 | 636 | outp += sprintf(outp, fmt8, |
5c56be9a DB |
637 | p->energy_gfx * rapl_energy_units); |
638 | if (do_rapl & RAPL_DRAM) | |
fc04cc67 | 639 | outp += sprintf(outp, fmt8, |
40ee8e3b | 640 | p->energy_dram * rapl_dram_energy_units); |
5c56be9a | 641 | if (do_rapl & RAPL_PKG_PERF_STATUS) |
fc04cc67 | 642 | outp += sprintf(outp, fmt8, 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float); |
5c56be9a | 643 | if (do_rapl & RAPL_DRAM_PERF_STATUS) |
fc04cc67 | 644 | outp += sprintf(outp, fmt8, 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float); |
889facbe | 645 | |
d7899447 | 646 | outp += sprintf(outp, fmt8, interval_float); |
5c56be9a | 647 | } |
c98d5d94 | 648 | done: |
c98d5d94 LB |
649 | outp += sprintf(outp, "\n"); |
650 | ||
651 | return 0; | |
103a8fea LB |
652 | } |
653 | ||
c98d5d94 LB |
654 | void flush_stdout() |
655 | { | |
656 | fputs(output_buffer, stdout); | |
ddac0d68 | 657 | fflush(stdout); |
c98d5d94 LB |
658 | outp = output_buffer; |
659 | } | |
660 | void flush_stderr() | |
661 | { | |
662 | fputs(output_buffer, stderr); | |
663 | outp = output_buffer; | |
664 | } | |
665 | void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
103a8fea | 666 | { |
e23da037 | 667 | static int printed; |
103a8fea | 668 | |
e23da037 LB |
669 | if (!printed || !summary_only) |
670 | print_header(); | |
103a8fea | 671 | |
c98d5d94 LB |
672 | if (topo.num_cpus > 1) |
673 | format_counters(&average.threads, &average.cores, | |
674 | &average.packages); | |
103a8fea | 675 | |
e23da037 LB |
676 | printed = 1; |
677 | ||
678 | if (summary_only) | |
679 | return; | |
680 | ||
c98d5d94 | 681 | for_all_cpus(format_counters, t, c, p); |
103a8fea LB |
682 | } |
683 | ||
889facbe LB |
684 | #define DELTA_WRAP32(new, old) \ |
685 | if (new > old) { \ | |
686 | old = new - old; \ | |
687 | } else { \ | |
688 | old = 0x100000000 + new - old; \ | |
689 | } | |
690 | ||
c98d5d94 LB |
691 | void |
692 | delta_package(struct pkg_data *new, struct pkg_data *old) | |
693 | { | |
0b2bb692 LB |
694 | |
695 | if (do_skl_residency) { | |
696 | old->pkg_wtd_core_c0 = new->pkg_wtd_core_c0 - old->pkg_wtd_core_c0; | |
697 | old->pkg_any_core_c0 = new->pkg_any_core_c0 - old->pkg_any_core_c0; | |
698 | old->pkg_any_gfxe_c0 = new->pkg_any_gfxe_c0 - old->pkg_any_gfxe_c0; | |
699 | old->pkg_both_core_gfxe_c0 = new->pkg_both_core_gfxe_c0 - old->pkg_both_core_gfxe_c0; | |
700 | } | |
c98d5d94 | 701 | old->pc2 = new->pc2 - old->pc2; |
ee7e38e3 LB |
702 | if (do_pc3) |
703 | old->pc3 = new->pc3 - old->pc3; | |
704 | if (do_pc6) | |
705 | old->pc6 = new->pc6 - old->pc6; | |
706 | if (do_pc7) | |
707 | old->pc7 = new->pc7 - old->pc7; | |
ca58710f KCA |
708 | old->pc8 = new->pc8 - old->pc8; |
709 | old->pc9 = new->pc9 - old->pc9; | |
710 | old->pc10 = new->pc10 - old->pc10; | |
889facbe LB |
711 | old->pkg_temp_c = new->pkg_temp_c; |
712 | ||
713 | DELTA_WRAP32(new->energy_pkg, old->energy_pkg); | |
714 | DELTA_WRAP32(new->energy_cores, old->energy_cores); | |
715 | DELTA_WRAP32(new->energy_gfx, old->energy_gfx); | |
716 | DELTA_WRAP32(new->energy_dram, old->energy_dram); | |
717 | DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status); | |
718 | DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status); | |
c98d5d94 | 719 | } |
103a8fea | 720 | |
c98d5d94 LB |
721 | void |
722 | delta_core(struct core_data *new, struct core_data *old) | |
103a8fea | 723 | { |
c98d5d94 LB |
724 | old->c3 = new->c3 - old->c3; |
725 | old->c6 = new->c6 - old->c6; | |
726 | old->c7 = new->c7 - old->c7; | |
889facbe | 727 | old->core_temp_c = new->core_temp_c; |
c98d5d94 | 728 | } |
103a8fea | 729 | |
c3ae331d LB |
730 | /* |
731 | * old = new - old | |
732 | */ | |
c98d5d94 LB |
733 | void |
734 | delta_thread(struct thread_data *new, struct thread_data *old, | |
735 | struct core_data *core_delta) | |
736 | { | |
737 | old->tsc = new->tsc - old->tsc; | |
738 | ||
739 | /* check for TSC < 1 Mcycles over interval */ | |
b2c95d90 JT |
740 | if (old->tsc < (1000 * 1000)) |
741 | errx(-3, "Insanely slow TSC rate, TSC stops in idle?\n" | |
742 | "You can disable all c-states by booting with \"idle=poll\"\n" | |
743 | "or just the deep ones with \"processor.max_cstate=1\""); | |
103a8fea | 744 | |
c98d5d94 | 745 | old->c1 = new->c1 - old->c1; |
103a8fea | 746 | |
a729617c LB |
747 | if (has_aperf) { |
748 | if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) { | |
749 | old->aperf = new->aperf - old->aperf; | |
750 | old->mperf = new->mperf - old->mperf; | |
751 | } else { | |
103a8fea | 752 | |
a729617c LB |
753 | if (!aperf_mperf_unstable) { |
754 | fprintf(stderr, "%s: APERF or MPERF went backwards *\n", progname); | |
755 | fprintf(stderr, "* Frequency results do not cover entire interval *\n"); | |
756 | fprintf(stderr, "* fix this by running Linux-2.6.30 or later *\n"); | |
103a8fea | 757 | |
a729617c LB |
758 | aperf_mperf_unstable = 1; |
759 | } | |
760 | /* | |
761 | * mperf delta is likely a huge "positive" number | |
762 | * can not use it for calculating c0 time | |
763 | */ | |
764 | skip_c0 = 1; | |
765 | skip_c1 = 1; | |
103a8fea | 766 | } |
c98d5d94 | 767 | } |
103a8fea | 768 | |
103a8fea | 769 | |
144b44b1 LB |
770 | if (use_c1_residency_msr) { |
771 | /* | |
772 | * Some models have a dedicated C1 residency MSR, | |
773 | * which should be more accurate than the derivation below. | |
774 | */ | |
775 | } else { | |
776 | /* | |
777 | * As counter collection is not atomic, | |
778 | * it is possible for mperf's non-halted cycles + idle states | |
779 | * to exceed TSC's all cycles: show c1 = 0% in that case. | |
780 | */ | |
781 | if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > old->tsc) | |
782 | old->c1 = 0; | |
783 | else { | |
784 | /* normal case, derive c1 */ | |
785 | old->c1 = old->tsc - old->mperf - core_delta->c3 | |
c98d5d94 | 786 | - core_delta->c6 - core_delta->c7; |
144b44b1 | 787 | } |
c98d5d94 | 788 | } |
c3ae331d | 789 | |
c98d5d94 | 790 | if (old->mperf == 0) { |
d8af6f5f | 791 | if (debug > 1) fprintf(stderr, "cpu%d MPERF 0!\n", old->cpu_id); |
c98d5d94 | 792 | old->mperf = 1; /* divide by 0 protection */ |
103a8fea | 793 | } |
c98d5d94 | 794 | |
8e180f3c LB |
795 | old->extra_delta32 = new->extra_delta32 - old->extra_delta32; |
796 | old->extra_delta32 &= 0xFFFFFFFF; | |
797 | ||
798 | old->extra_delta64 = new->extra_delta64 - old->extra_delta64; | |
799 | ||
c98d5d94 | 800 | /* |
8e180f3c | 801 | * Extra MSR is just a snapshot, simply copy latest w/o subtracting |
c98d5d94 | 802 | */ |
2f32edf1 LB |
803 | old->extra_msr32 = new->extra_msr32; |
804 | old->extra_msr64 = new->extra_msr64; | |
1ed51011 LB |
805 | |
806 | if (do_smi) | |
807 | old->smi_count = new->smi_count - old->smi_count; | |
c98d5d94 LB |
808 | } |
809 | ||
810 | int delta_cpu(struct thread_data *t, struct core_data *c, | |
811 | struct pkg_data *p, struct thread_data *t2, | |
812 | struct core_data *c2, struct pkg_data *p2) | |
813 | { | |
814 | /* calculate core delta only for 1st thread in core */ | |
815 | if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE) | |
816 | delta_core(c, c2); | |
817 | ||
818 | /* always calculate thread delta */ | |
819 | delta_thread(t, t2, c2); /* c2 is core delta */ | |
820 | ||
821 | /* calculate package delta only for 1st core in package */ | |
822 | if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE) | |
823 | delta_package(p, p2); | |
824 | ||
103a8fea LB |
825 | return 0; |
826 | } | |
827 | ||
c98d5d94 LB |
828 | void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) |
829 | { | |
830 | t->tsc = 0; | |
831 | t->aperf = 0; | |
832 | t->mperf = 0; | |
833 | t->c1 = 0; | |
834 | ||
1ed51011 | 835 | t->smi_count = 0; |
8e180f3c LB |
836 | t->extra_delta32 = 0; |
837 | t->extra_delta64 = 0; | |
838 | ||
c98d5d94 LB |
839 | /* tells format_counters to dump all fields from this set */ |
840 | t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE; | |
841 | ||
842 | c->c3 = 0; | |
843 | c->c6 = 0; | |
844 | c->c7 = 0; | |
889facbe | 845 | c->core_temp_c = 0; |
c98d5d94 | 846 | |
0b2bb692 LB |
847 | p->pkg_wtd_core_c0 = 0; |
848 | p->pkg_any_core_c0 = 0; | |
849 | p->pkg_any_gfxe_c0 = 0; | |
850 | p->pkg_both_core_gfxe_c0 = 0; | |
851 | ||
c98d5d94 | 852 | p->pc2 = 0; |
ee7e38e3 LB |
853 | if (do_pc3) |
854 | p->pc3 = 0; | |
855 | if (do_pc6) | |
856 | p->pc6 = 0; | |
857 | if (do_pc7) | |
858 | p->pc7 = 0; | |
ca58710f KCA |
859 | p->pc8 = 0; |
860 | p->pc9 = 0; | |
861 | p->pc10 = 0; | |
889facbe LB |
862 | |
863 | p->energy_pkg = 0; | |
864 | p->energy_dram = 0; | |
865 | p->energy_cores = 0; | |
866 | p->energy_gfx = 0; | |
867 | p->rapl_pkg_perf_status = 0; | |
868 | p->rapl_dram_perf_status = 0; | |
869 | p->pkg_temp_c = 0; | |
c98d5d94 LB |
870 | } |
871 | int sum_counters(struct thread_data *t, struct core_data *c, | |
872 | struct pkg_data *p) | |
103a8fea | 873 | { |
c98d5d94 LB |
874 | average.threads.tsc += t->tsc; |
875 | average.threads.aperf += t->aperf; | |
876 | average.threads.mperf += t->mperf; | |
877 | average.threads.c1 += t->c1; | |
103a8fea | 878 | |
8e180f3c LB |
879 | average.threads.extra_delta32 += t->extra_delta32; |
880 | average.threads.extra_delta64 += t->extra_delta64; | |
881 | ||
c98d5d94 LB |
882 | /* sum per-core values only for 1st thread in core */ |
883 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
884 | return 0; | |
103a8fea | 885 | |
c98d5d94 LB |
886 | average.cores.c3 += c->c3; |
887 | average.cores.c6 += c->c6; | |
888 | average.cores.c7 += c->c7; | |
889 | ||
889facbe LB |
890 | average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c); |
891 | ||
c98d5d94 LB |
892 | /* sum per-pkg values only for 1st core in pkg */ |
893 | if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
894 | return 0; | |
895 | ||
0b2bb692 LB |
896 | if (do_skl_residency) { |
897 | average.packages.pkg_wtd_core_c0 += p->pkg_wtd_core_c0; | |
898 | average.packages.pkg_any_core_c0 += p->pkg_any_core_c0; | |
899 | average.packages.pkg_any_gfxe_c0 += p->pkg_any_gfxe_c0; | |
900 | average.packages.pkg_both_core_gfxe_c0 += p->pkg_both_core_gfxe_c0; | |
901 | } | |
902 | ||
c98d5d94 | 903 | average.packages.pc2 += p->pc2; |
ee7e38e3 LB |
904 | if (do_pc3) |
905 | average.packages.pc3 += p->pc3; | |
906 | if (do_pc6) | |
907 | average.packages.pc6 += p->pc6; | |
908 | if (do_pc7) | |
909 | average.packages.pc7 += p->pc7; | |
ca58710f KCA |
910 | average.packages.pc8 += p->pc8; |
911 | average.packages.pc9 += p->pc9; | |
912 | average.packages.pc10 += p->pc10; | |
c98d5d94 | 913 | |
889facbe LB |
914 | average.packages.energy_pkg += p->energy_pkg; |
915 | average.packages.energy_dram += p->energy_dram; | |
916 | average.packages.energy_cores += p->energy_cores; | |
917 | average.packages.energy_gfx += p->energy_gfx; | |
918 | ||
919 | average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c); | |
920 | ||
921 | average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status; | |
922 | average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status; | |
c98d5d94 LB |
923 | return 0; |
924 | } | |
925 | /* | |
926 | * sum the counters for all cpus in the system | |
927 | * compute the weighted average | |
928 | */ | |
929 | void compute_average(struct thread_data *t, struct core_data *c, | |
930 | struct pkg_data *p) | |
931 | { | |
932 | clear_counters(&average.threads, &average.cores, &average.packages); | |
933 | ||
934 | for_all_cpus(sum_counters, t, c, p); | |
935 | ||
936 | average.threads.tsc /= topo.num_cpus; | |
937 | average.threads.aperf /= topo.num_cpus; | |
938 | average.threads.mperf /= topo.num_cpus; | |
939 | average.threads.c1 /= topo.num_cpus; | |
940 | ||
8e180f3c LB |
941 | average.threads.extra_delta32 /= topo.num_cpus; |
942 | average.threads.extra_delta32 &= 0xFFFFFFFF; | |
943 | ||
944 | average.threads.extra_delta64 /= topo.num_cpus; | |
945 | ||
c98d5d94 LB |
946 | average.cores.c3 /= topo.num_cores; |
947 | average.cores.c6 /= topo.num_cores; | |
948 | average.cores.c7 /= topo.num_cores; | |
949 | ||
0b2bb692 LB |
950 | if (do_skl_residency) { |
951 | average.packages.pkg_wtd_core_c0 /= topo.num_packages; | |
952 | average.packages.pkg_any_core_c0 /= topo.num_packages; | |
953 | average.packages.pkg_any_gfxe_c0 /= topo.num_packages; | |
954 | average.packages.pkg_both_core_gfxe_c0 /= topo.num_packages; | |
955 | } | |
956 | ||
c98d5d94 | 957 | average.packages.pc2 /= topo.num_packages; |
ee7e38e3 LB |
958 | if (do_pc3) |
959 | average.packages.pc3 /= topo.num_packages; | |
960 | if (do_pc6) | |
961 | average.packages.pc6 /= topo.num_packages; | |
962 | if (do_pc7) | |
963 | average.packages.pc7 /= topo.num_packages; | |
ca58710f KCA |
964 | |
965 | average.packages.pc8 /= topo.num_packages; | |
966 | average.packages.pc9 /= topo.num_packages; | |
967 | average.packages.pc10 /= topo.num_packages; | |
103a8fea LB |
968 | } |
969 | ||
c98d5d94 | 970 | static unsigned long long rdtsc(void) |
103a8fea | 971 | { |
c98d5d94 | 972 | unsigned int low, high; |
15aaa346 | 973 | |
c98d5d94 | 974 | asm volatile("rdtsc" : "=a" (low), "=d" (high)); |
15aaa346 | 975 | |
c98d5d94 LB |
976 | return low | ((unsigned long long)high) << 32; |
977 | } | |
15aaa346 | 978 | |
15aaa346 | 979 | |
c98d5d94 LB |
980 | /* |
981 | * get_counters(...) | |
982 | * migrate to cpu | |
983 | * acquire and record local counters for that cpu | |
984 | */ | |
985 | int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
986 | { | |
987 | int cpu = t->cpu_id; | |
889facbe | 988 | unsigned long long msr; |
88c3281f | 989 | |
e52966c0 LB |
990 | if (cpu_migrate(cpu)) { |
991 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
c98d5d94 | 992 | return -1; |
e52966c0 | 993 | } |
15aaa346 | 994 | |
c98d5d94 LB |
995 | t->tsc = rdtsc(); /* we are running on local CPU of interest */ |
996 | ||
997 | if (has_aperf) { | |
9c63a650 | 998 | if (get_msr(cpu, MSR_IA32_APERF, &t->aperf)) |
c98d5d94 | 999 | return -3; |
9c63a650 | 1000 | if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf)) |
c98d5d94 | 1001 | return -4; |
b2b34dfe HC |
1002 | t->aperf = t->aperf * aperf_mperf_multiplier; |
1003 | t->mperf = t->mperf * aperf_mperf_multiplier; | |
c98d5d94 LB |
1004 | } |
1005 | ||
1ed51011 LB |
1006 | if (do_smi) { |
1007 | if (get_msr(cpu, MSR_SMI_COUNT, &msr)) | |
1008 | return -5; | |
1009 | t->smi_count = msr & 0xFFFFFFFF; | |
1010 | } | |
8e180f3c | 1011 | if (extra_delta_offset32) { |
889facbe | 1012 | if (get_msr(cpu, extra_delta_offset32, &msr)) |
8e180f3c | 1013 | return -5; |
889facbe | 1014 | t->extra_delta32 = msr & 0xFFFFFFFF; |
8e180f3c LB |
1015 | } |
1016 | ||
1017 | if (extra_delta_offset64) | |
1018 | if (get_msr(cpu, extra_delta_offset64, &t->extra_delta64)) | |
2f32edf1 LB |
1019 | return -5; |
1020 | ||
8e180f3c | 1021 | if (extra_msr_offset32) { |
889facbe | 1022 | if (get_msr(cpu, extra_msr_offset32, &msr)) |
8e180f3c | 1023 | return -5; |
889facbe | 1024 | t->extra_msr32 = msr & 0xFFFFFFFF; |
8e180f3c LB |
1025 | } |
1026 | ||
2f32edf1 LB |
1027 | if (extra_msr_offset64) |
1028 | if (get_msr(cpu, extra_msr_offset64, &t->extra_msr64)) | |
c98d5d94 LB |
1029 | return -5; |
1030 | ||
144b44b1 LB |
1031 | if (use_c1_residency_msr) { |
1032 | if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1)) | |
1033 | return -6; | |
1034 | } | |
1035 | ||
c98d5d94 LB |
1036 | /* collect core counters only for 1st thread in core */ |
1037 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
1038 | return 0; | |
1039 | ||
fb5d4327 | 1040 | if (do_nhm_cstates && !do_slm_cstates && !do_knl_cstates) { |
c98d5d94 LB |
1041 | if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3)) |
1042 | return -6; | |
144b44b1 LB |
1043 | } |
1044 | ||
fb5d4327 | 1045 | if (do_nhm_cstates && !do_knl_cstates) { |
c98d5d94 LB |
1046 | if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6)) |
1047 | return -7; | |
fb5d4327 DC |
1048 | } else if (do_knl_cstates) { |
1049 | if (get_msr(cpu, MSR_KNL_CORE_C6_RESIDENCY, &c->c6)) | |
1050 | return -7; | |
c98d5d94 LB |
1051 | } |
1052 | ||
1053 | if (do_snb_cstates) | |
1054 | if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7)) | |
1055 | return -8; | |
1056 | ||
889facbe LB |
1057 | if (do_dts) { |
1058 | if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr)) | |
1059 | return -9; | |
1060 | c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F); | |
1061 | } | |
1062 | ||
1063 | ||
c98d5d94 LB |
1064 | /* collect package counters only for 1st core in package */ |
1065 | if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
1066 | return 0; | |
1067 | ||
0b2bb692 LB |
1068 | if (do_skl_residency) { |
1069 | if (get_msr(cpu, MSR_PKG_WEIGHTED_CORE_C0_RES, &p->pkg_wtd_core_c0)) | |
1070 | return -10; | |
1071 | if (get_msr(cpu, MSR_PKG_ANY_CORE_C0_RES, &p->pkg_any_core_c0)) | |
1072 | return -11; | |
1073 | if (get_msr(cpu, MSR_PKG_ANY_GFXE_C0_RES, &p->pkg_any_gfxe_c0)) | |
1074 | return -12; | |
1075 | if (get_msr(cpu, MSR_PKG_BOTH_CORE_GFXE_C0_RES, &p->pkg_both_core_gfxe_c0)) | |
1076 | return -13; | |
1077 | } | |
ee7e38e3 | 1078 | if (do_pc3) |
c98d5d94 LB |
1079 | if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3)) |
1080 | return -9; | |
ee7e38e3 | 1081 | if (do_pc6) |
c98d5d94 LB |
1082 | if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6)) |
1083 | return -10; | |
ee7e38e3 | 1084 | if (do_pc2) |
c98d5d94 LB |
1085 | if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2)) |
1086 | return -11; | |
ee7e38e3 | 1087 | if (do_pc7) |
c98d5d94 LB |
1088 | if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7)) |
1089 | return -12; | |
ca58710f KCA |
1090 | if (do_c8_c9_c10) { |
1091 | if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8)) | |
1092 | return -13; | |
1093 | if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9)) | |
1094 | return -13; | |
1095 | if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10)) | |
1096 | return -13; | |
1097 | } | |
889facbe LB |
1098 | if (do_rapl & RAPL_PKG) { |
1099 | if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr)) | |
1100 | return -13; | |
1101 | p->energy_pkg = msr & 0xFFFFFFFF; | |
1102 | } | |
1103 | if (do_rapl & RAPL_CORES) { | |
1104 | if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr)) | |
1105 | return -14; | |
1106 | p->energy_cores = msr & 0xFFFFFFFF; | |
1107 | } | |
1108 | if (do_rapl & RAPL_DRAM) { | |
1109 | if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr)) | |
1110 | return -15; | |
1111 | p->energy_dram = msr & 0xFFFFFFFF; | |
1112 | } | |
1113 | if (do_rapl & RAPL_GFX) { | |
1114 | if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr)) | |
1115 | return -16; | |
1116 | p->energy_gfx = msr & 0xFFFFFFFF; | |
1117 | } | |
1118 | if (do_rapl & RAPL_PKG_PERF_STATUS) { | |
1119 | if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr)) | |
1120 | return -16; | |
1121 | p->rapl_pkg_perf_status = msr & 0xFFFFFFFF; | |
1122 | } | |
1123 | if (do_rapl & RAPL_DRAM_PERF_STATUS) { | |
1124 | if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr)) | |
1125 | return -16; | |
1126 | p->rapl_dram_perf_status = msr & 0xFFFFFFFF; | |
1127 | } | |
1128 | if (do_ptm) { | |
1129 | if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr)) | |
1130 | return -17; | |
1131 | p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F); | |
1132 | } | |
15aaa346 | 1133 | return 0; |
103a8fea LB |
1134 | } |
1135 | ||
ee7e38e3 LB |
1136 | /* |
1137 | * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit: | |
1138 | * If you change the values, note they are used both in comparisons | |
1139 | * (>= PCL__7) and to index pkg_cstate_limit_strings[]. | |
1140 | */ | |
1141 | ||
1142 | #define PCLUKN 0 /* Unknown */ | |
1143 | #define PCLRSV 1 /* Reserved */ | |
1144 | #define PCL__0 2 /* PC0 */ | |
1145 | #define PCL__1 3 /* PC1 */ | |
1146 | #define PCL__2 4 /* PC2 */ | |
1147 | #define PCL__3 5 /* PC3 */ | |
1148 | #define PCL__4 6 /* PC4 */ | |
1149 | #define PCL__6 7 /* PC6 */ | |
1150 | #define PCL_6N 8 /* PC6 No Retention */ | |
1151 | #define PCL_6R 9 /* PC6 Retention */ | |
1152 | #define PCL__7 10 /* PC7 */ | |
1153 | #define PCL_7S 11 /* PC7 Shrink */ | |
0b2bb692 LB |
1154 | #define PCL__8 12 /* PC8 */ |
1155 | #define PCL__9 13 /* PC9 */ | |
1156 | #define PCLUNL 14 /* Unlimited */ | |
ee7e38e3 LB |
1157 | |
1158 | int pkg_cstate_limit = PCLUKN; | |
1159 | char *pkg_cstate_limit_strings[] = { "reserved", "unknown", "pc0", "pc1", "pc2", | |
0b2bb692 | 1160 | "pc3", "pc4", "pc6", "pc6n", "pc6r", "pc7", "pc7s", "pc8", "pc9", "unlimited"}; |
ee7e38e3 | 1161 | |
e9257f5f LB |
1162 | int nhm_pkg_cstate_limits[16] = {PCL__0, PCL__1, PCL__3, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV}; |
1163 | int snb_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCL_6N, PCL_6R, PCL__7, PCL_7S, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV}; | |
1164 | int hsw_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCL__3, PCL__6, PCL__7, PCL_7S, PCL__8, PCL__9, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV}; | |
1165 | int slv_pkg_cstate_limits[16] = {PCL__0, PCL__1, PCLRSV, PCLRSV, PCL__4, PCLRSV, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV}; | |
1166 | int amt_pkg_cstate_limits[16] = {PCL__0, PCL__1, PCL__2, PCLRSV, PCLRSV, PCLRSV, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV}; | |
1167 | int phi_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCL_6N, PCL_6R, PCLRSV, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV}; | |
ee7e38e3 | 1168 | |
a2b7b749 LB |
1169 | |
1170 | static void | |
1171 | calculate_tsc_tweak() | |
1172 | { | |
a2b7b749 LB |
1173 | tsc_tweak = base_hz / tsc_hz; |
1174 | } | |
1175 | ||
fcd17211 LB |
1176 | static void |
1177 | dump_nhm_platform_info(void) | |
103a8fea LB |
1178 | { |
1179 | unsigned long long msr; | |
1180 | unsigned int ratio; | |
1181 | ||
ec0adc53 | 1182 | get_msr(base_cpu, MSR_PLATFORM_INFO, &msr); |
103a8fea | 1183 | |
ec0adc53 | 1184 | fprintf(stderr, "cpu%d: MSR_PLATFORM_INFO: 0x%08llx\n", base_cpu, msr); |
6574a5d5 | 1185 | |
103a8fea | 1186 | ratio = (msr >> 40) & 0xFF; |
8f61f359 | 1187 | fprintf(stderr, "%d * %.0f = %.0f MHz max efficiency frequency\n", |
103a8fea LB |
1188 | ratio, bclk, ratio * bclk); |
1189 | ||
1190 | ratio = (msr >> 8) & 0xFF; | |
8f61f359 | 1191 | fprintf(stderr, "%d * %.0f = %.0f MHz base frequency\n", |
103a8fea LB |
1192 | ratio, bclk, ratio * bclk); |
1193 | ||
7ce7d5de | 1194 | get_msr(base_cpu, MSR_IA32_POWER_CTL, &msr); |
bfae2052 LB |
1195 | fprintf(stderr, "cpu%d: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n", |
1196 | base_cpu, msr, msr & 0x2 ? "EN" : "DIS"); | |
67920418 | 1197 | |
fcd17211 LB |
1198 | return; |
1199 | } | |
1200 | ||
1201 | static void | |
1202 | dump_hsw_turbo_ratio_limits(void) | |
1203 | { | |
1204 | unsigned long long msr; | |
1205 | unsigned int ratio; | |
1206 | ||
7ce7d5de | 1207 | get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT2, &msr); |
fcd17211 | 1208 | |
bfae2052 | 1209 | fprintf(stderr, "cpu%d: MSR_TURBO_RATIO_LIMIT2: 0x%08llx\n", base_cpu, msr); |
fcd17211 LB |
1210 | |
1211 | ratio = (msr >> 8) & 0xFF; | |
1212 | if (ratio) | |
1213 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 18 active cores\n", | |
1214 | ratio, bclk, ratio * bclk); | |
1215 | ||
1216 | ratio = (msr >> 0) & 0xFF; | |
1217 | if (ratio) | |
1218 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 17 active cores\n", | |
1219 | ratio, bclk, ratio * bclk); | |
1220 | return; | |
1221 | } | |
1222 | ||
1223 | static void | |
1224 | dump_ivt_turbo_ratio_limits(void) | |
1225 | { | |
1226 | unsigned long long msr; | |
1227 | unsigned int ratio; | |
6574a5d5 | 1228 | |
7ce7d5de | 1229 | get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &msr); |
6574a5d5 | 1230 | |
bfae2052 | 1231 | fprintf(stderr, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, msr); |
6574a5d5 LB |
1232 | |
1233 | ratio = (msr >> 56) & 0xFF; | |
1234 | if (ratio) | |
1235 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 16 active cores\n", | |
1236 | ratio, bclk, ratio * bclk); | |
1237 | ||
1238 | ratio = (msr >> 48) & 0xFF; | |
1239 | if (ratio) | |
1240 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 15 active cores\n", | |
1241 | ratio, bclk, ratio * bclk); | |
1242 | ||
1243 | ratio = (msr >> 40) & 0xFF; | |
1244 | if (ratio) | |
1245 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 14 active cores\n", | |
1246 | ratio, bclk, ratio * bclk); | |
1247 | ||
1248 | ratio = (msr >> 32) & 0xFF; | |
1249 | if (ratio) | |
1250 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 13 active cores\n", | |
1251 | ratio, bclk, ratio * bclk); | |
1252 | ||
1253 | ratio = (msr >> 24) & 0xFF; | |
1254 | if (ratio) | |
1255 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 12 active cores\n", | |
1256 | ratio, bclk, ratio * bclk); | |
1257 | ||
1258 | ratio = (msr >> 16) & 0xFF; | |
1259 | if (ratio) | |
1260 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 11 active cores\n", | |
1261 | ratio, bclk, ratio * bclk); | |
1262 | ||
1263 | ratio = (msr >> 8) & 0xFF; | |
1264 | if (ratio) | |
1265 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 10 active cores\n", | |
1266 | ratio, bclk, ratio * bclk); | |
1267 | ||
1268 | ratio = (msr >> 0) & 0xFF; | |
1269 | if (ratio) | |
1270 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 9 active cores\n", | |
1271 | ratio, bclk, ratio * bclk); | |
fcd17211 LB |
1272 | return; |
1273 | } | |
6574a5d5 | 1274 | |
fcd17211 LB |
1275 | static void |
1276 | dump_nhm_turbo_ratio_limits(void) | |
1277 | { | |
1278 | unsigned long long msr; | |
1279 | unsigned int ratio; | |
103a8fea | 1280 | |
7ce7d5de | 1281 | get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr); |
103a8fea | 1282 | |
bfae2052 | 1283 | fprintf(stderr, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n", base_cpu, msr); |
6574a5d5 LB |
1284 | |
1285 | ratio = (msr >> 56) & 0xFF; | |
1286 | if (ratio) | |
1287 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 8 active cores\n", | |
1288 | ratio, bclk, ratio * bclk); | |
1289 | ||
1290 | ratio = (msr >> 48) & 0xFF; | |
1291 | if (ratio) | |
1292 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 7 active cores\n", | |
1293 | ratio, bclk, ratio * bclk); | |
1294 | ||
1295 | ratio = (msr >> 40) & 0xFF; | |
1296 | if (ratio) | |
1297 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 6 active cores\n", | |
1298 | ratio, bclk, ratio * bclk); | |
1299 | ||
1300 | ratio = (msr >> 32) & 0xFF; | |
1301 | if (ratio) | |
1302 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 5 active cores\n", | |
1303 | ratio, bclk, ratio * bclk); | |
1304 | ||
103a8fea LB |
1305 | ratio = (msr >> 24) & 0xFF; |
1306 | if (ratio) | |
1307 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 4 active cores\n", | |
1308 | ratio, bclk, ratio * bclk); | |
1309 | ||
1310 | ratio = (msr >> 16) & 0xFF; | |
1311 | if (ratio) | |
1312 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 3 active cores\n", | |
1313 | ratio, bclk, ratio * bclk); | |
1314 | ||
1315 | ratio = (msr >> 8) & 0xFF; | |
1316 | if (ratio) | |
1317 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 2 active cores\n", | |
1318 | ratio, bclk, ratio * bclk); | |
1319 | ||
1320 | ratio = (msr >> 0) & 0xFF; | |
1321 | if (ratio) | |
1322 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 1 active cores\n", | |
1323 | ratio, bclk, ratio * bclk); | |
fcd17211 LB |
1324 | return; |
1325 | } | |
3a9a941d | 1326 | |
fb5d4327 DC |
1327 | static void |
1328 | dump_knl_turbo_ratio_limits(void) | |
1329 | { | |
1330 | int cores; | |
1331 | unsigned int ratio; | |
1332 | unsigned long long msr; | |
1333 | int delta_cores; | |
1334 | int delta_ratio; | |
1335 | int i; | |
1336 | ||
7ce7d5de | 1337 | get_msr(base_cpu, MSR_NHM_TURBO_RATIO_LIMIT, &msr); |
fb5d4327 | 1338 | |
bfae2052 LB |
1339 | fprintf(stderr, "cpu%d: MSR_NHM_TURBO_RATIO_LIMIT: 0x%08llx\n", |
1340 | base_cpu, msr); | |
fb5d4327 DC |
1341 | |
1342 | /** | |
1343 | * Turbo encoding in KNL is as follows: | |
1344 | * [7:0] -- Base value of number of active cores of bucket 1. | |
1345 | * [15:8] -- Base value of freq ratio of bucket 1. | |
1346 | * [20:16] -- +ve delta of number of active cores of bucket 2. | |
1347 | * i.e. active cores of bucket 2 = | |
1348 | * active cores of bucket 1 + delta | |
1349 | * [23:21] -- Negative delta of freq ratio of bucket 2. | |
1350 | * i.e. freq ratio of bucket 2 = | |
1351 | * freq ratio of bucket 1 - delta | |
1352 | * [28:24]-- +ve delta of number of active cores of bucket 3. | |
1353 | * [31:29]-- -ve delta of freq ratio of bucket 3. | |
1354 | * [36:32]-- +ve delta of number of active cores of bucket 4. | |
1355 | * [39:37]-- -ve delta of freq ratio of bucket 4. | |
1356 | * [44:40]-- +ve delta of number of active cores of bucket 5. | |
1357 | * [47:45]-- -ve delta of freq ratio of bucket 5. | |
1358 | * [52:48]-- +ve delta of number of active cores of bucket 6. | |
1359 | * [55:53]-- -ve delta of freq ratio of bucket 6. | |
1360 | * [60:56]-- +ve delta of number of active cores of bucket 7. | |
1361 | * [63:61]-- -ve delta of freq ratio of bucket 7. | |
1362 | */ | |
1363 | cores = msr & 0xFF; | |
1364 | ratio = (msr >> 8) && 0xFF; | |
1365 | if (ratio > 0) | |
1366 | fprintf(stderr, | |
1367 | "%d * %.0f = %.0f MHz max turbo %d active cores\n", | |
1368 | ratio, bclk, ratio * bclk, cores); | |
1369 | ||
1370 | for (i = 16; i < 64; i = i + 8) { | |
1371 | delta_cores = (msr >> i) & 0x1F; | |
1372 | delta_ratio = (msr >> (i + 5)) && 0x7; | |
1373 | if (!delta_cores || !delta_ratio) | |
1374 | return; | |
1375 | cores = cores + delta_cores; | |
1376 | ratio = ratio - delta_ratio; | |
1377 | ||
1378 | /** -ve ratios will make successive ratio calculations | |
1379 | * negative. Hence return instead of carrying on. | |
1380 | */ | |
1381 | if (ratio > 0) | |
1382 | fprintf(stderr, | |
1383 | "%d * %.0f = %.0f MHz max turbo %d active cores\n", | |
1384 | ratio, bclk, ratio * bclk, cores); | |
1385 | } | |
1386 | } | |
1387 | ||
fcd17211 LB |
1388 | static void |
1389 | dump_nhm_cst_cfg(void) | |
1390 | { | |
1391 | unsigned long long msr; | |
1392 | ||
7ce7d5de | 1393 | get_msr(base_cpu, MSR_NHM_SNB_PKG_CST_CFG_CTL, &msr); |
fcd17211 LB |
1394 | |
1395 | #define SNB_C1_AUTO_UNDEMOTE (1UL << 27) | |
1396 | #define SNB_C3_AUTO_UNDEMOTE (1UL << 28) | |
1397 | ||
bfae2052 | 1398 | fprintf(stderr, "cpu%d: MSR_NHM_SNB_PKG_CST_CFG_CTL: 0x%08llx", base_cpu, msr); |
fcd17211 LB |
1399 | |
1400 | fprintf(stderr, " (%s%s%s%s%slocked: pkg-cstate-limit=%d: %s)\n", | |
1401 | (msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "", | |
1402 | (msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "", | |
1403 | (msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "", | |
1404 | (msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "", | |
1405 | (msr & (1 << 15)) ? "" : "UN", | |
1406 | (unsigned int)msr & 7, | |
1407 | pkg_cstate_limit_strings[pkg_cstate_limit]); | |
1408 | return; | |
103a8fea LB |
1409 | } |
1410 | ||
6fb3143b LB |
1411 | static void |
1412 | dump_config_tdp(void) | |
1413 | { | |
1414 | unsigned long long msr; | |
1415 | ||
1416 | get_msr(base_cpu, MSR_CONFIG_TDP_NOMINAL, &msr); | |
1417 | fprintf(stderr, "cpu%d: MSR_CONFIG_TDP_NOMINAL: 0x%08llx", base_cpu, msr); | |
1418 | fprintf(stderr, " (base_ratio=%d)\n", (unsigned int)msr & 0xEF); | |
1419 | ||
1420 | get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_1, &msr); | |
1421 | fprintf(stderr, "cpu%d: MSR_CONFIG_TDP_LEVEL_1: 0x%08llx (", base_cpu, msr); | |
1422 | if (msr) { | |
1423 | fprintf(stderr, "PKG_MIN_PWR_LVL1=%d ", (unsigned int)(msr >> 48) & 0xEFFF); | |
1424 | fprintf(stderr, "PKG_MAX_PWR_LVL1=%d ", (unsigned int)(msr >> 32) & 0xEFFF); | |
1425 | fprintf(stderr, "LVL1_RATIO=%d ", (unsigned int)(msr >> 16) & 0xEF); | |
1426 | fprintf(stderr, "PKG_TDP_LVL1=%d", (unsigned int)(msr) & 0xEFFF); | |
1427 | } | |
1428 | fprintf(stderr, ")\n"); | |
1429 | ||
1430 | get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_2, &msr); | |
1431 | fprintf(stderr, "cpu%d: MSR_CONFIG_TDP_LEVEL_2: 0x%08llx (", base_cpu, msr); | |
1432 | if (msr) { | |
1433 | fprintf(stderr, "PKG_MIN_PWR_LVL2=%d ", (unsigned int)(msr >> 48) & 0xEFFF); | |
1434 | fprintf(stderr, "PKG_MAX_PWR_LVL2=%d ", (unsigned int)(msr >> 32) & 0xEFFF); | |
1435 | fprintf(stderr, "LVL2_RATIO=%d ", (unsigned int)(msr >> 16) & 0xEF); | |
1436 | fprintf(stderr, "PKG_TDP_LVL2=%d", (unsigned int)(msr) & 0xEFFF); | |
1437 | } | |
1438 | fprintf(stderr, ")\n"); | |
1439 | ||
1440 | get_msr(base_cpu, MSR_CONFIG_TDP_CONTROL, &msr); | |
1441 | fprintf(stderr, "cpu%d: MSR_CONFIG_TDP_CONTROL: 0x%08llx (", base_cpu, msr); | |
1442 | if ((msr) & 0x3) | |
1443 | fprintf(stderr, "TDP_LEVEL=%d ", (unsigned int)(msr) & 0x3); | |
1444 | fprintf(stderr, " lock=%d", (unsigned int)(msr >> 31) & 1); | |
1445 | fprintf(stderr, ")\n"); | |
1446 | ||
1447 | get_msr(base_cpu, MSR_TURBO_ACTIVATION_RATIO, &msr); | |
1448 | fprintf(stderr, "cpu%d: MSR_TURBO_ACTIVATION_RATIO: 0x%08llx (", base_cpu, msr); | |
759d2a93 | 1449 | fprintf(stderr, "MAX_NON_TURBO_RATIO=%d", (unsigned int)(msr) & 0x7F); |
6fb3143b LB |
1450 | fprintf(stderr, " lock=%d", (unsigned int)(msr >> 31) & 1); |
1451 | fprintf(stderr, ")\n"); | |
1452 | } | |
1453 | ||
c98d5d94 | 1454 | void free_all_buffers(void) |
103a8fea | 1455 | { |
c98d5d94 LB |
1456 | CPU_FREE(cpu_present_set); |
1457 | cpu_present_set = NULL; | |
1458 | cpu_present_set = 0; | |
103a8fea | 1459 | |
c98d5d94 LB |
1460 | CPU_FREE(cpu_affinity_set); |
1461 | cpu_affinity_set = NULL; | |
1462 | cpu_affinity_setsize = 0; | |
103a8fea | 1463 | |
c98d5d94 LB |
1464 | free(thread_even); |
1465 | free(core_even); | |
1466 | free(package_even); | |
103a8fea | 1467 | |
c98d5d94 LB |
1468 | thread_even = NULL; |
1469 | core_even = NULL; | |
1470 | package_even = NULL; | |
103a8fea | 1471 | |
c98d5d94 LB |
1472 | free(thread_odd); |
1473 | free(core_odd); | |
1474 | free(package_odd); | |
103a8fea | 1475 | |
c98d5d94 LB |
1476 | thread_odd = NULL; |
1477 | core_odd = NULL; | |
1478 | package_odd = NULL; | |
103a8fea | 1479 | |
c98d5d94 LB |
1480 | free(output_buffer); |
1481 | output_buffer = NULL; | |
1482 | outp = NULL; | |
103a8fea LB |
1483 | } |
1484 | ||
57a42a34 JT |
1485 | /* |
1486 | * Open a file, and exit on failure | |
1487 | */ | |
1488 | FILE *fopen_or_die(const char *path, const char *mode) | |
1489 | { | |
1490 | FILE *filep = fopen(path, "r"); | |
b2c95d90 JT |
1491 | if (!filep) |
1492 | err(1, "%s: open failed", path); | |
57a42a34 JT |
1493 | return filep; |
1494 | } | |
1495 | ||
c98d5d94 | 1496 | /* |
95aebc44 | 1497 | * Parse a file containing a single int. |
c98d5d94 | 1498 | */ |
95aebc44 | 1499 | int parse_int_file(const char *fmt, ...) |
103a8fea | 1500 | { |
95aebc44 JT |
1501 | va_list args; |
1502 | char path[PATH_MAX]; | |
c98d5d94 | 1503 | FILE *filep; |
95aebc44 | 1504 | int value; |
103a8fea | 1505 | |
95aebc44 JT |
1506 | va_start(args, fmt); |
1507 | vsnprintf(path, sizeof(path), fmt, args); | |
1508 | va_end(args); | |
57a42a34 | 1509 | filep = fopen_or_die(path, "r"); |
b2c95d90 JT |
1510 | if (fscanf(filep, "%d", &value) != 1) |
1511 | err(1, "%s: failed to parse number from file", path); | |
c98d5d94 | 1512 | fclose(filep); |
95aebc44 JT |
1513 | return value; |
1514 | } | |
1515 | ||
1516 | /* | |
e275b388 DC |
1517 | * get_cpu_position_in_core(cpu) |
1518 | * return the position of the CPU among its HT siblings in the core | |
1519 | * return -1 if the sibling is not in list | |
95aebc44 | 1520 | */ |
e275b388 | 1521 | int get_cpu_position_in_core(int cpu) |
95aebc44 | 1522 | { |
e275b388 DC |
1523 | char path[64]; |
1524 | FILE *filep; | |
1525 | int this_cpu; | |
1526 | char character; | |
1527 | int i; | |
1528 | ||
1529 | sprintf(path, | |
1530 | "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", | |
1531 | cpu); | |
1532 | filep = fopen(path, "r"); | |
1533 | if (filep == NULL) { | |
1534 | perror(path); | |
1535 | exit(1); | |
1536 | } | |
1537 | ||
1538 | for (i = 0; i < topo.num_threads_per_core; i++) { | |
1539 | fscanf(filep, "%d", &this_cpu); | |
1540 | if (this_cpu == cpu) { | |
1541 | fclose(filep); | |
1542 | return i; | |
1543 | } | |
1544 | ||
1545 | /* Account for no separator after last thread*/ | |
1546 | if (i != (topo.num_threads_per_core - 1)) | |
1547 | fscanf(filep, "%c", &character); | |
1548 | } | |
1549 | ||
1550 | fclose(filep); | |
1551 | return -1; | |
103a8fea LB |
1552 | } |
1553 | ||
c98d5d94 LB |
1554 | /* |
1555 | * cpu_is_first_core_in_package(cpu) | |
1556 | * return 1 if given CPU is 1st core in package | |
1557 | */ | |
1558 | int cpu_is_first_core_in_package(int cpu) | |
103a8fea | 1559 | { |
95aebc44 | 1560 | return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu); |
103a8fea LB |
1561 | } |
1562 | ||
1563 | int get_physical_package_id(int cpu) | |
1564 | { | |
95aebc44 | 1565 | return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu); |
103a8fea LB |
1566 | } |
1567 | ||
1568 | int get_core_id(int cpu) | |
1569 | { | |
95aebc44 | 1570 | return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu); |
103a8fea LB |
1571 | } |
1572 | ||
c98d5d94 LB |
1573 | int get_num_ht_siblings(int cpu) |
1574 | { | |
1575 | char path[80]; | |
1576 | FILE *filep; | |
e275b388 DC |
1577 | int sib1; |
1578 | int matches = 0; | |
c98d5d94 | 1579 | char character; |
e275b388 DC |
1580 | char str[100]; |
1581 | char *ch; | |
c98d5d94 LB |
1582 | |
1583 | sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu); | |
57a42a34 | 1584 | filep = fopen_or_die(path, "r"); |
e275b388 | 1585 | |
c98d5d94 LB |
1586 | /* |
1587 | * file format: | |
e275b388 DC |
1588 | * A ',' separated or '-' separated set of numbers |
1589 | * (eg 1-2 or 1,3,4,5) | |
c98d5d94 | 1590 | */ |
e275b388 DC |
1591 | fscanf(filep, "%d%c\n", &sib1, &character); |
1592 | fseek(filep, 0, SEEK_SET); | |
1593 | fgets(str, 100, filep); | |
1594 | ch = strchr(str, character); | |
1595 | while (ch != NULL) { | |
1596 | matches++; | |
1597 | ch = strchr(ch+1, character); | |
1598 | } | |
c98d5d94 LB |
1599 | |
1600 | fclose(filep); | |
e275b388 | 1601 | return matches+1; |
c98d5d94 LB |
1602 | } |
1603 | ||
103a8fea | 1604 | /* |
c98d5d94 LB |
1605 | * run func(thread, core, package) in topology order |
1606 | * skip non-present cpus | |
103a8fea LB |
1607 | */ |
1608 | ||
c98d5d94 LB |
1609 | int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *, |
1610 | struct pkg_data *, struct thread_data *, struct core_data *, | |
1611 | struct pkg_data *), struct thread_data *thread_base, | |
1612 | struct core_data *core_base, struct pkg_data *pkg_base, | |
1613 | struct thread_data *thread_base2, struct core_data *core_base2, | |
1614 | struct pkg_data *pkg_base2) | |
1615 | { | |
1616 | int retval, pkg_no, core_no, thread_no; | |
1617 | ||
1618 | for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) { | |
1619 | for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) { | |
1620 | for (thread_no = 0; thread_no < | |
1621 | topo.num_threads_per_core; ++thread_no) { | |
1622 | struct thread_data *t, *t2; | |
1623 | struct core_data *c, *c2; | |
1624 | struct pkg_data *p, *p2; | |
1625 | ||
1626 | t = GET_THREAD(thread_base, thread_no, core_no, pkg_no); | |
1627 | ||
1628 | if (cpu_is_not_present(t->cpu_id)) | |
1629 | continue; | |
1630 | ||
1631 | t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no); | |
1632 | ||
1633 | c = GET_CORE(core_base, core_no, pkg_no); | |
1634 | c2 = GET_CORE(core_base2, core_no, pkg_no); | |
1635 | ||
1636 | p = GET_PKG(pkg_base, pkg_no); | |
1637 | p2 = GET_PKG(pkg_base2, pkg_no); | |
1638 | ||
1639 | retval = func(t, c, p, t2, c2, p2); | |
1640 | if (retval) | |
1641 | return retval; | |
1642 | } | |
1643 | } | |
1644 | } | |
1645 | return 0; | |
1646 | } | |
1647 | ||
1648 | /* | |
1649 | * run func(cpu) on every cpu in /proc/stat | |
1650 | * return max_cpu number | |
1651 | */ | |
1652 | int for_all_proc_cpus(int (func)(int)) | |
103a8fea LB |
1653 | { |
1654 | FILE *fp; | |
c98d5d94 | 1655 | int cpu_num; |
103a8fea LB |
1656 | int retval; |
1657 | ||
57a42a34 | 1658 | fp = fopen_or_die(proc_stat, "r"); |
103a8fea LB |
1659 | |
1660 | retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n"); | |
b2c95d90 JT |
1661 | if (retval != 0) |
1662 | err(1, "%s: failed to parse format", proc_stat); | |
103a8fea | 1663 | |
c98d5d94 LB |
1664 | while (1) { |
1665 | retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num); | |
103a8fea LB |
1666 | if (retval != 1) |
1667 | break; | |
1668 | ||
c98d5d94 LB |
1669 | retval = func(cpu_num); |
1670 | if (retval) { | |
1671 | fclose(fp); | |
1672 | return(retval); | |
1673 | } | |
103a8fea LB |
1674 | } |
1675 | fclose(fp); | |
c98d5d94 | 1676 | return 0; |
103a8fea LB |
1677 | } |
1678 | ||
1679 | void re_initialize(void) | |
1680 | { | |
c98d5d94 LB |
1681 | free_all_buffers(); |
1682 | setup_all_buffers(); | |
1683 | printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus); | |
103a8fea LB |
1684 | } |
1685 | ||
c98d5d94 | 1686 | |
103a8fea | 1687 | /* |
c98d5d94 LB |
1688 | * count_cpus() |
1689 | * remember the last one seen, it will be the max | |
103a8fea | 1690 | */ |
c98d5d94 | 1691 | int count_cpus(int cpu) |
103a8fea | 1692 | { |
c98d5d94 LB |
1693 | if (topo.max_cpu_num < cpu) |
1694 | topo.max_cpu_num = cpu; | |
103a8fea | 1695 | |
c98d5d94 LB |
1696 | topo.num_cpus += 1; |
1697 | return 0; | |
1698 | } | |
1699 | int mark_cpu_present(int cpu) | |
1700 | { | |
1701 | CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set); | |
15aaa346 | 1702 | return 0; |
103a8fea LB |
1703 | } |
1704 | ||
1705 | void turbostat_loop() | |
1706 | { | |
c98d5d94 | 1707 | int retval; |
e52966c0 | 1708 | int restarted = 0; |
c98d5d94 | 1709 | |
103a8fea | 1710 | restart: |
e52966c0 LB |
1711 | restarted++; |
1712 | ||
c98d5d94 | 1713 | retval = for_all_cpus(get_counters, EVEN_COUNTERS); |
d91bb17c LB |
1714 | if (retval < -1) { |
1715 | exit(retval); | |
1716 | } else if (retval == -1) { | |
e52966c0 LB |
1717 | if (restarted > 1) { |
1718 | exit(retval); | |
1719 | } | |
c98d5d94 LB |
1720 | re_initialize(); |
1721 | goto restart; | |
1722 | } | |
e52966c0 | 1723 | restarted = 0; |
103a8fea LB |
1724 | gettimeofday(&tv_even, (struct timezone *)NULL); |
1725 | ||
1726 | while (1) { | |
c98d5d94 | 1727 | if (for_all_proc_cpus(cpu_is_not_present)) { |
103a8fea LB |
1728 | re_initialize(); |
1729 | goto restart; | |
1730 | } | |
1731 | sleep(interval_sec); | |
c98d5d94 | 1732 | retval = for_all_cpus(get_counters, ODD_COUNTERS); |
d91bb17c LB |
1733 | if (retval < -1) { |
1734 | exit(retval); | |
1735 | } else if (retval == -1) { | |
15aaa346 LB |
1736 | re_initialize(); |
1737 | goto restart; | |
1738 | } | |
103a8fea | 1739 | gettimeofday(&tv_odd, (struct timezone *)NULL); |
103a8fea | 1740 | timersub(&tv_odd, &tv_even, &tv_delta); |
c98d5d94 LB |
1741 | for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS); |
1742 | compute_average(EVEN_COUNTERS); | |
1743 | format_all_counters(EVEN_COUNTERS); | |
1744 | flush_stdout(); | |
15aaa346 | 1745 | sleep(interval_sec); |
c98d5d94 | 1746 | retval = for_all_cpus(get_counters, EVEN_COUNTERS); |
d91bb17c LB |
1747 | if (retval < -1) { |
1748 | exit(retval); | |
1749 | } else if (retval == -1) { | |
103a8fea LB |
1750 | re_initialize(); |
1751 | goto restart; | |
1752 | } | |
103a8fea | 1753 | gettimeofday(&tv_even, (struct timezone *)NULL); |
103a8fea | 1754 | timersub(&tv_even, &tv_odd, &tv_delta); |
c98d5d94 LB |
1755 | for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS); |
1756 | compute_average(ODD_COUNTERS); | |
1757 | format_all_counters(ODD_COUNTERS); | |
1758 | flush_stdout(); | |
103a8fea LB |
1759 | } |
1760 | } | |
1761 | ||
1762 | void check_dev_msr() | |
1763 | { | |
1764 | struct stat sb; | |
7ce7d5de | 1765 | char pathname[32]; |
103a8fea | 1766 | |
7ce7d5de PB |
1767 | sprintf(pathname, "/dev/cpu/%d/msr", base_cpu); |
1768 | if (stat(pathname, &sb)) | |
a21d38c8 LB |
1769 | if (system("/sbin/modprobe msr > /dev/null 2>&1")) |
1770 | err(-5, "no /dev/cpu/0/msr, Try \"# modprobe msr\" "); | |
103a8fea LB |
1771 | } |
1772 | ||
98481e79 | 1773 | void check_permissions() |
103a8fea | 1774 | { |
98481e79 LB |
1775 | struct __user_cap_header_struct cap_header_data; |
1776 | cap_user_header_t cap_header = &cap_header_data; | |
1777 | struct __user_cap_data_struct cap_data_data; | |
1778 | cap_user_data_t cap_data = &cap_data_data; | |
1779 | extern int capget(cap_user_header_t hdrp, cap_user_data_t datap); | |
1780 | int do_exit = 0; | |
7ce7d5de | 1781 | char pathname[32]; |
98481e79 LB |
1782 | |
1783 | /* check for CAP_SYS_RAWIO */ | |
1784 | cap_header->pid = getpid(); | |
1785 | cap_header->version = _LINUX_CAPABILITY_VERSION; | |
1786 | if (capget(cap_header, cap_data) < 0) | |
1787 | err(-6, "capget(2) failed"); | |
1788 | ||
1789 | if ((cap_data->effective & (1 << CAP_SYS_RAWIO)) == 0) { | |
1790 | do_exit++; | |
1791 | warnx("capget(CAP_SYS_RAWIO) failed," | |
1792 | " try \"# setcap cap_sys_rawio=ep %s\"", progname); | |
1793 | } | |
1794 | ||
1795 | /* test file permissions */ | |
7ce7d5de PB |
1796 | sprintf(pathname, "/dev/cpu/%d/msr", base_cpu); |
1797 | if (euidaccess(pathname, R_OK)) { | |
98481e79 LB |
1798 | do_exit++; |
1799 | warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr"); | |
1800 | } | |
1801 | ||
1802 | /* if all else fails, thell them to be root */ | |
1803 | if (do_exit) | |
1804 | if (getuid() != 0) | |
d7899447 | 1805 | warnx("... or simply run as root"); |
98481e79 LB |
1806 | |
1807 | if (do_exit) | |
1808 | exit(-6); | |
103a8fea LB |
1809 | } |
1810 | ||
d7899447 LB |
1811 | /* |
1812 | * NHM adds support for additional MSRs: | |
1813 | * | |
1814 | * MSR_SMI_COUNT 0x00000034 | |
1815 | * | |
ec0adc53 | 1816 | * MSR_PLATFORM_INFO 0x000000ce |
d7899447 LB |
1817 | * MSR_NHM_SNB_PKG_CST_CFG_CTL 0x000000e2 |
1818 | * | |
1819 | * MSR_PKG_C3_RESIDENCY 0x000003f8 | |
1820 | * MSR_PKG_C6_RESIDENCY 0x000003f9 | |
1821 | * MSR_CORE_C3_RESIDENCY 0x000003fc | |
1822 | * MSR_CORE_C6_RESIDENCY 0x000003fd | |
1823 | * | |
ee7e38e3 LB |
1824 | * Side effect: |
1825 | * sets global pkg_cstate_limit to decode MSR_NHM_SNB_PKG_CST_CFG_CTL | |
d7899447 | 1826 | */ |
ee7e38e3 | 1827 | int probe_nhm_msrs(unsigned int family, unsigned int model) |
103a8fea | 1828 | { |
ee7e38e3 | 1829 | unsigned long long msr; |
21ed5574 | 1830 | unsigned int base_ratio; |
ee7e38e3 LB |
1831 | int *pkg_cstate_limits; |
1832 | ||
103a8fea LB |
1833 | if (!genuine_intel) |
1834 | return 0; | |
1835 | ||
1836 | if (family != 6) | |
1837 | return 0; | |
1838 | ||
21ed5574 LB |
1839 | bclk = discover_bclk(family, model); |
1840 | ||
103a8fea LB |
1841 | switch (model) { |
1842 | case 0x1A: /* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */ | |
1843 | case 0x1E: /* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */ | |
1844 | case 0x1F: /* Core i7 and i5 Processor - Nehalem */ | |
1845 | case 0x25: /* Westmere Client - Clarkdale, Arrandale */ | |
1846 | case 0x2C: /* Westmere EP - Gulftown */ | |
ee7e38e3 LB |
1847 | case 0x2E: /* Nehalem-EX Xeon - Beckton */ |
1848 | case 0x2F: /* Westmere-EX Xeon - Eagleton */ | |
1849 | pkg_cstate_limits = nhm_pkg_cstate_limits; | |
1850 | break; | |
103a8fea LB |
1851 | case 0x2A: /* SNB */ |
1852 | case 0x2D: /* SNB Xeon */ | |
553575f1 | 1853 | case 0x3A: /* IVB */ |
1300651b | 1854 | case 0x3E: /* IVB Xeon */ |
ee7e38e3 LB |
1855 | pkg_cstate_limits = snb_pkg_cstate_limits; |
1856 | break; | |
70b43400 | 1857 | case 0x3C: /* HSW */ |
e6f9bb3c | 1858 | case 0x3F: /* HSX */ |
70b43400 | 1859 | case 0x45: /* HSW */ |
149c2319 | 1860 | case 0x46: /* HSW */ |
4e8e863f | 1861 | case 0x3D: /* BDW */ |
48a0631c | 1862 | case 0x47: /* BDW */ |
4e8e863f LB |
1863 | case 0x4F: /* BDX */ |
1864 | case 0x56: /* BDX-DE */ | |
0b2bb692 LB |
1865 | case 0x4E: /* SKL */ |
1866 | case 0x5E: /* SKL */ | |
ee7e38e3 LB |
1867 | pkg_cstate_limits = hsw_pkg_cstate_limits; |
1868 | break; | |
1869 | case 0x37: /* BYT */ | |
1870 | case 0x4D: /* AVN */ | |
1871 | pkg_cstate_limits = slv_pkg_cstate_limits; | |
1872 | break; | |
1873 | case 0x4C: /* AMT */ | |
1874 | pkg_cstate_limits = amt_pkg_cstate_limits; | |
1875 | break; | |
1876 | case 0x57: /* PHI */ | |
1877 | pkg_cstate_limits = phi_pkg_cstate_limits; | |
1878 | break; | |
103a8fea LB |
1879 | default: |
1880 | return 0; | |
1881 | } | |
7ce7d5de | 1882 | get_msr(base_cpu, MSR_NHM_SNB_PKG_CST_CFG_CTL, &msr); |
e9257f5f | 1883 | pkg_cstate_limit = pkg_cstate_limits[msr & 0xF]; |
ee7e38e3 | 1884 | |
ec0adc53 | 1885 | get_msr(base_cpu, MSR_PLATFORM_INFO, &msr); |
21ed5574 LB |
1886 | base_ratio = (msr >> 8) & 0xFF; |
1887 | ||
1888 | base_hz = base_ratio * bclk * 1000000; | |
1889 | has_base_hz = 1; | |
ee7e38e3 | 1890 | return 1; |
103a8fea | 1891 | } |
d7899447 LB |
1892 | int has_nhm_turbo_ratio_limit(unsigned int family, unsigned int model) |
1893 | { | |
d7899447 LB |
1894 | switch (model) { |
1895 | /* Nehalem compatible, but do not include turbo-ratio limit support */ | |
1896 | case 0x2E: /* Nehalem-EX Xeon - Beckton */ | |
1897 | case 0x2F: /* Westmere-EX Xeon - Eagleton */ | |
1898 | return 0; | |
1899 | default: | |
1900 | return 1; | |
1901 | } | |
1902 | } | |
6574a5d5 LB |
1903 | int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model) |
1904 | { | |
1905 | if (!genuine_intel) | |
1906 | return 0; | |
1907 | ||
1908 | if (family != 6) | |
1909 | return 0; | |
1910 | ||
1911 | switch (model) { | |
1912 | case 0x3E: /* IVB Xeon */ | |
fcd17211 LB |
1913 | case 0x3F: /* HSW Xeon */ |
1914 | return 1; | |
1915 | default: | |
1916 | return 0; | |
1917 | } | |
1918 | } | |
1919 | int has_hsw_turbo_ratio_limit(unsigned int family, unsigned int model) | |
1920 | { | |
1921 | if (!genuine_intel) | |
1922 | return 0; | |
1923 | ||
1924 | if (family != 6) | |
1925 | return 0; | |
1926 | ||
1927 | switch (model) { | |
1928 | case 0x3F: /* HSW Xeon */ | |
6574a5d5 LB |
1929 | return 1; |
1930 | default: | |
1931 | return 0; | |
1932 | } | |
1933 | } | |
1934 | ||
fb5d4327 DC |
1935 | int has_knl_turbo_ratio_limit(unsigned int family, unsigned int model) |
1936 | { | |
1937 | if (!genuine_intel) | |
1938 | return 0; | |
1939 | ||
1940 | if (family != 6) | |
1941 | return 0; | |
1942 | ||
1943 | switch (model) { | |
1944 | case 0x57: /* Knights Landing */ | |
1945 | return 1; | |
1946 | default: | |
1947 | return 0; | |
1948 | } | |
1949 | } | |
6fb3143b LB |
1950 | int has_config_tdp(unsigned int family, unsigned int model) |
1951 | { | |
1952 | if (!genuine_intel) | |
1953 | return 0; | |
1954 | ||
1955 | if (family != 6) | |
1956 | return 0; | |
1957 | ||
1958 | switch (model) { | |
1959 | case 0x3A: /* IVB */ | |
6fb3143b LB |
1960 | case 0x3C: /* HSW */ |
1961 | case 0x3F: /* HSX */ | |
1962 | case 0x45: /* HSW */ | |
1963 | case 0x46: /* HSW */ | |
1964 | case 0x3D: /* BDW */ | |
1965 | case 0x47: /* BDW */ | |
1966 | case 0x4F: /* BDX */ | |
1967 | case 0x56: /* BDX-DE */ | |
1968 | case 0x4E: /* SKL */ | |
1969 | case 0x5E: /* SKL */ | |
1970 | ||
1971 | case 0x57: /* Knights Landing */ | |
1972 | return 1; | |
1973 | default: | |
1974 | return 0; | |
1975 | } | |
1976 | } | |
1977 | ||
fcd17211 LB |
1978 | static void |
1979 | dump_cstate_pstate_config_info(family, model) | |
1980 | { | |
1981 | if (!do_nhm_platform_info) | |
1982 | return; | |
1983 | ||
1984 | dump_nhm_platform_info(); | |
1985 | ||
1986 | if (has_hsw_turbo_ratio_limit(family, model)) | |
1987 | dump_hsw_turbo_ratio_limits(); | |
1988 | ||
1989 | if (has_ivt_turbo_ratio_limit(family, model)) | |
1990 | dump_ivt_turbo_ratio_limits(); | |
1991 | ||
1992 | if (has_nhm_turbo_ratio_limit(family, model)) | |
1993 | dump_nhm_turbo_ratio_limits(); | |
1994 | ||
fb5d4327 DC |
1995 | if (has_knl_turbo_ratio_limit(family, model)) |
1996 | dump_knl_turbo_ratio_limits(); | |
1997 | ||
6fb3143b LB |
1998 | if (has_config_tdp(family, model)) |
1999 | dump_config_tdp(); | |
2000 | ||
fcd17211 LB |
2001 | dump_nhm_cst_cfg(); |
2002 | } | |
2003 | ||
2004 | ||
889facbe LB |
2005 | /* |
2006 | * print_epb() | |
2007 | * Decode the ENERGY_PERF_BIAS MSR | |
2008 | */ | |
2009 | int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
2010 | { | |
2011 | unsigned long long msr; | |
2012 | char *epb_string; | |
2013 | int cpu; | |
2014 | ||
2015 | if (!has_epb) | |
2016 | return 0; | |
2017 | ||
2018 | cpu = t->cpu_id; | |
2019 | ||
2020 | /* EPB is per-package */ | |
2021 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
2022 | return 0; | |
2023 | ||
2024 | if (cpu_migrate(cpu)) { | |
2025 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
2026 | return -1; | |
2027 | } | |
2028 | ||
2029 | if (get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr)) | |
2030 | return 0; | |
2031 | ||
e9be7dd6 | 2032 | switch (msr & 0xF) { |
889facbe LB |
2033 | case ENERGY_PERF_BIAS_PERFORMANCE: |
2034 | epb_string = "performance"; | |
2035 | break; | |
2036 | case ENERGY_PERF_BIAS_NORMAL: | |
2037 | epb_string = "balanced"; | |
2038 | break; | |
2039 | case ENERGY_PERF_BIAS_POWERSAVE: | |
2040 | epb_string = "powersave"; | |
2041 | break; | |
2042 | default: | |
2043 | epb_string = "custom"; | |
2044 | break; | |
2045 | } | |
2046 | fprintf(stderr, "cpu%d: MSR_IA32_ENERGY_PERF_BIAS: 0x%08llx (%s)\n", cpu, msr, epb_string); | |
2047 | ||
2048 | return 0; | |
2049 | } | |
7f5c258e LB |
2050 | /* |
2051 | * print_hwp() | |
2052 | * Decode the MSR_HWP_CAPABILITIES | |
2053 | */ | |
2054 | int print_hwp(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
2055 | { | |
2056 | unsigned long long msr; | |
2057 | int cpu; | |
2058 | ||
2059 | if (!has_hwp) | |
2060 | return 0; | |
2061 | ||
2062 | cpu = t->cpu_id; | |
2063 | ||
2064 | /* MSR_HWP_CAPABILITIES is per-package */ | |
2065 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
2066 | return 0; | |
2067 | ||
2068 | if (cpu_migrate(cpu)) { | |
2069 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
2070 | return -1; | |
2071 | } | |
2072 | ||
2073 | if (get_msr(cpu, MSR_PM_ENABLE, &msr)) | |
2074 | return 0; | |
2075 | ||
2076 | fprintf(stderr, "cpu%d: MSR_PM_ENABLE: 0x%08llx (%sHWP)\n", | |
2077 | cpu, msr, (msr & (1 << 0)) ? "" : "No-"); | |
2078 | ||
2079 | /* MSR_PM_ENABLE[1] == 1 if HWP is enabled and MSRs visible */ | |
2080 | if ((msr & (1 << 0)) == 0) | |
2081 | return 0; | |
2082 | ||
2083 | if (get_msr(cpu, MSR_HWP_CAPABILITIES, &msr)) | |
2084 | return 0; | |
2085 | ||
2086 | fprintf(stderr, "cpu%d: MSR_HWP_CAPABILITIES: 0x%08llx " | |
2087 | "(high 0x%x guar 0x%x eff 0x%x low 0x%x)\n", | |
2088 | cpu, msr, | |
2089 | (unsigned int)HWP_HIGHEST_PERF(msr), | |
2090 | (unsigned int)HWP_GUARANTEED_PERF(msr), | |
2091 | (unsigned int)HWP_MOSTEFFICIENT_PERF(msr), | |
2092 | (unsigned int)HWP_LOWEST_PERF(msr)); | |
2093 | ||
2094 | if (get_msr(cpu, MSR_HWP_REQUEST, &msr)) | |
2095 | return 0; | |
2096 | ||
2097 | fprintf(stderr, "cpu%d: MSR_HWP_REQUEST: 0x%08llx " | |
2098 | "(min 0x%x max 0x%x des 0x%x epp 0x%x window 0x%x pkg 0x%x)\n", | |
2099 | cpu, msr, | |
2100 | (unsigned int)(((msr) >> 0) & 0xff), | |
2101 | (unsigned int)(((msr) >> 8) & 0xff), | |
2102 | (unsigned int)(((msr) >> 16) & 0xff), | |
2103 | (unsigned int)(((msr) >> 24) & 0xff), | |
2104 | (unsigned int)(((msr) >> 32) & 0xff3), | |
2105 | (unsigned int)(((msr) >> 42) & 0x1)); | |
2106 | ||
2107 | if (has_hwp_pkg) { | |
2108 | if (get_msr(cpu, MSR_HWP_REQUEST_PKG, &msr)) | |
2109 | return 0; | |
2110 | ||
2111 | fprintf(stderr, "cpu%d: MSR_HWP_REQUEST_PKG: 0x%08llx " | |
2112 | "(min 0x%x max 0x%x des 0x%x epp 0x%x window 0x%x)\n", | |
2113 | cpu, msr, | |
2114 | (unsigned int)(((msr) >> 0) & 0xff), | |
2115 | (unsigned int)(((msr) >> 8) & 0xff), | |
2116 | (unsigned int)(((msr) >> 16) & 0xff), | |
2117 | (unsigned int)(((msr) >> 24) & 0xff), | |
2118 | (unsigned int)(((msr) >> 32) & 0xff3)); | |
2119 | } | |
2120 | if (has_hwp_notify) { | |
2121 | if (get_msr(cpu, MSR_HWP_INTERRUPT, &msr)) | |
2122 | return 0; | |
2123 | ||
2124 | fprintf(stderr, "cpu%d: MSR_HWP_INTERRUPT: 0x%08llx " | |
2125 | "(%s_Guaranteed_Perf_Change, %s_Excursion_Min)\n", | |
2126 | cpu, msr, | |
2127 | ((msr) & 0x1) ? "EN" : "Dis", | |
2128 | ((msr) & 0x2) ? "EN" : "Dis"); | |
2129 | } | |
2130 | if (get_msr(cpu, MSR_HWP_STATUS, &msr)) | |
2131 | return 0; | |
2132 | ||
2133 | fprintf(stderr, "cpu%d: MSR_HWP_STATUS: 0x%08llx " | |
2134 | "(%sGuaranteed_Perf_Change, %sExcursion_Min)\n", | |
2135 | cpu, msr, | |
2136 | ((msr) & 0x1) ? "" : "No-", | |
2137 | ((msr) & 0x2) ? "" : "No-"); | |
2138 | ||
2139 | return 0; | |
2140 | } | |
889facbe | 2141 | |
3a9a941d LB |
2142 | /* |
2143 | * print_perf_limit() | |
2144 | */ | |
2145 | int print_perf_limit(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
2146 | { | |
2147 | unsigned long long msr; | |
2148 | int cpu; | |
2149 | ||
2150 | cpu = t->cpu_id; | |
2151 | ||
2152 | /* per-package */ | |
2153 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
2154 | return 0; | |
2155 | ||
2156 | if (cpu_migrate(cpu)) { | |
2157 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
2158 | return -1; | |
2159 | } | |
2160 | ||
2161 | if (do_core_perf_limit_reasons) { | |
2162 | get_msr(cpu, MSR_CORE_PERF_LIMIT_REASONS, &msr); | |
2163 | fprintf(stderr, "cpu%d: MSR_CORE_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr); | |
2164 | fprintf(stderr, " (Active: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)", | |
e33cbe85 | 2165 | (msr & 1 << 15) ? "bit15, " : "", |
3a9a941d | 2166 | (msr & 1 << 14) ? "bit14, " : "", |
e33cbe85 LB |
2167 | (msr & 1 << 13) ? "Transitions, " : "", |
2168 | (msr & 1 << 12) ? "MultiCoreTurbo, " : "", | |
2169 | (msr & 1 << 11) ? "PkgPwrL2, " : "", | |
2170 | (msr & 1 << 10) ? "PkgPwrL1, " : "", | |
2171 | (msr & 1 << 9) ? "CorePwr, " : "", | |
2172 | (msr & 1 << 8) ? "Amps, " : "", | |
2173 | (msr & 1 << 6) ? "VR-Therm, " : "", | |
2174 | (msr & 1 << 5) ? "Auto-HWP, " : "", | |
2175 | (msr & 1 << 4) ? "Graphics, " : "", | |
2176 | (msr & 1 << 2) ? "bit2, " : "", | |
2177 | (msr & 1 << 1) ? "ThermStatus, " : "", | |
2178 | (msr & 1 << 0) ? "PROCHOT, " : ""); | |
3a9a941d | 2179 | fprintf(stderr, " (Logged: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)\n", |
e33cbe85 | 2180 | (msr & 1 << 31) ? "bit31, " : "", |
3a9a941d | 2181 | (msr & 1 << 30) ? "bit30, " : "", |
e33cbe85 LB |
2182 | (msr & 1 << 29) ? "Transitions, " : "", |
2183 | (msr & 1 << 28) ? "MultiCoreTurbo, " : "", | |
2184 | (msr & 1 << 27) ? "PkgPwrL2, " : "", | |
2185 | (msr & 1 << 26) ? "PkgPwrL1, " : "", | |
2186 | (msr & 1 << 25) ? "CorePwr, " : "", | |
2187 | (msr & 1 << 24) ? "Amps, " : "", | |
2188 | (msr & 1 << 22) ? "VR-Therm, " : "", | |
2189 | (msr & 1 << 21) ? "Auto-HWP, " : "", | |
2190 | (msr & 1 << 20) ? "Graphics, " : "", | |
2191 | (msr & 1 << 18) ? "bit18, " : "", | |
2192 | (msr & 1 << 17) ? "ThermStatus, " : "", | |
2193 | (msr & 1 << 16) ? "PROCHOT, " : ""); | |
3a9a941d LB |
2194 | |
2195 | } | |
2196 | if (do_gfx_perf_limit_reasons) { | |
2197 | get_msr(cpu, MSR_GFX_PERF_LIMIT_REASONS, &msr); | |
2198 | fprintf(stderr, "cpu%d: MSR_GFX_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr); | |
2199 | fprintf(stderr, " (Active: %s%s%s%s%s%s%s%s)", | |
2200 | (msr & 1 << 0) ? "PROCHOT, " : "", | |
2201 | (msr & 1 << 1) ? "ThermStatus, " : "", | |
2202 | (msr & 1 << 4) ? "Graphics, " : "", | |
2203 | (msr & 1 << 6) ? "VR-Therm, " : "", | |
2204 | (msr & 1 << 8) ? "Amps, " : "", | |
2205 | (msr & 1 << 9) ? "GFXPwr, " : "", | |
2206 | (msr & 1 << 10) ? "PkgPwrL1, " : "", | |
2207 | (msr & 1 << 11) ? "PkgPwrL2, " : ""); | |
2208 | fprintf(stderr, " (Logged: %s%s%s%s%s%s%s%s)\n", | |
2209 | (msr & 1 << 16) ? "PROCHOT, " : "", | |
2210 | (msr & 1 << 17) ? "ThermStatus, " : "", | |
2211 | (msr & 1 << 20) ? "Graphics, " : "", | |
2212 | (msr & 1 << 22) ? "VR-Therm, " : "", | |
2213 | (msr & 1 << 24) ? "Amps, " : "", | |
2214 | (msr & 1 << 25) ? "GFXPwr, " : "", | |
2215 | (msr & 1 << 26) ? "PkgPwrL1, " : "", | |
2216 | (msr & 1 << 27) ? "PkgPwrL2, " : ""); | |
2217 | } | |
2218 | if (do_ring_perf_limit_reasons) { | |
2219 | get_msr(cpu, MSR_RING_PERF_LIMIT_REASONS, &msr); | |
2220 | fprintf(stderr, "cpu%d: MSR_RING_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr); | |
2221 | fprintf(stderr, " (Active: %s%s%s%s%s%s)", | |
2222 | (msr & 1 << 0) ? "PROCHOT, " : "", | |
2223 | (msr & 1 << 1) ? "ThermStatus, " : "", | |
2224 | (msr & 1 << 6) ? "VR-Therm, " : "", | |
2225 | (msr & 1 << 8) ? "Amps, " : "", | |
2226 | (msr & 1 << 10) ? "PkgPwrL1, " : "", | |
2227 | (msr & 1 << 11) ? "PkgPwrL2, " : ""); | |
2228 | fprintf(stderr, " (Logged: %s%s%s%s%s%s)\n", | |
2229 | (msr & 1 << 16) ? "PROCHOT, " : "", | |
2230 | (msr & 1 << 17) ? "ThermStatus, " : "", | |
2231 | (msr & 1 << 22) ? "VR-Therm, " : "", | |
2232 | (msr & 1 << 24) ? "Amps, " : "", | |
2233 | (msr & 1 << 26) ? "PkgPwrL1, " : "", | |
2234 | (msr & 1 << 27) ? "PkgPwrL2, " : ""); | |
2235 | } | |
2236 | return 0; | |
2237 | } | |
2238 | ||
889facbe LB |
2239 | #define RAPL_POWER_GRANULARITY 0x7FFF /* 15 bit power granularity */ |
2240 | #define RAPL_TIME_GRANULARITY 0x3F /* 6 bit time granularity */ | |
2241 | ||
144b44b1 LB |
2242 | double get_tdp(model) |
2243 | { | |
2244 | unsigned long long msr; | |
2245 | ||
2246 | if (do_rapl & RAPL_PKG_POWER_INFO) | |
7ce7d5de | 2247 | if (!get_msr(base_cpu, MSR_PKG_POWER_INFO, &msr)) |
144b44b1 LB |
2248 | return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units; |
2249 | ||
2250 | switch (model) { | |
2251 | case 0x37: | |
2252 | case 0x4D: | |
2253 | return 30.0; | |
2254 | default: | |
2255 | return 135.0; | |
2256 | } | |
2257 | } | |
2258 | ||
40ee8e3b AS |
2259 | /* |
2260 | * rapl_dram_energy_units_probe() | |
2261 | * Energy units are either hard-coded, or come from RAPL Energy Unit MSR. | |
2262 | */ | |
2263 | static double | |
2264 | rapl_dram_energy_units_probe(int model, double rapl_energy_units) | |
2265 | { | |
2266 | /* only called for genuine_intel, family 6 */ | |
2267 | ||
2268 | switch (model) { | |
2269 | case 0x3F: /* HSX */ | |
2270 | case 0x4F: /* BDX */ | |
2271 | case 0x56: /* BDX-DE */ | |
fb5d4327 | 2272 | case 0x57: /* KNL */ |
40ee8e3b AS |
2273 | return (rapl_dram_energy_units = 15.3 / 1000000); |
2274 | default: | |
2275 | return (rapl_energy_units); | |
2276 | } | |
2277 | } | |
2278 | ||
144b44b1 | 2279 | |
889facbe LB |
2280 | /* |
2281 | * rapl_probe() | |
2282 | * | |
144b44b1 | 2283 | * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units |
889facbe LB |
2284 | */ |
2285 | void rapl_probe(unsigned int family, unsigned int model) | |
2286 | { | |
2287 | unsigned long long msr; | |
144b44b1 | 2288 | unsigned int time_unit; |
889facbe LB |
2289 | double tdp; |
2290 | ||
2291 | if (!genuine_intel) | |
2292 | return; | |
2293 | ||
2294 | if (family != 6) | |
2295 | return; | |
2296 | ||
2297 | switch (model) { | |
2298 | case 0x2A: | |
2299 | case 0x3A: | |
70b43400 | 2300 | case 0x3C: /* HSW */ |
70b43400 | 2301 | case 0x45: /* HSW */ |
149c2319 | 2302 | case 0x46: /* HSW */ |
4e8e863f | 2303 | case 0x3D: /* BDW */ |
48a0631c | 2304 | case 0x47: /* BDW */ |
144b44b1 | 2305 | do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO; |
889facbe | 2306 | break; |
0b2bb692 LB |
2307 | case 0x4E: /* SKL */ |
2308 | case 0x5E: /* SKL */ | |
2309 | do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO; | |
2310 | break; | |
e6f9bb3c | 2311 | case 0x3F: /* HSX */ |
4e8e863f LB |
2312 | case 0x4F: /* BDX */ |
2313 | case 0x56: /* BDX-DE */ | |
fb5d4327 | 2314 | case 0x57: /* KNL */ |
0b2bb692 | 2315 | do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO; |
e6f9bb3c | 2316 | break; |
889facbe LB |
2317 | case 0x2D: |
2318 | case 0x3E: | |
0b2bb692 | 2319 | do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_PKG_PERF_STATUS | RAPL_DRAM_PERF_STATUS | RAPL_PKG_POWER_INFO; |
144b44b1 LB |
2320 | break; |
2321 | case 0x37: /* BYT */ | |
2322 | case 0x4D: /* AVN */ | |
2323 | do_rapl = RAPL_PKG | RAPL_CORES ; | |
889facbe LB |
2324 | break; |
2325 | default: | |
2326 | return; | |
2327 | } | |
2328 | ||
2329 | /* units on package 0, verify later other packages match */ | |
7ce7d5de | 2330 | if (get_msr(base_cpu, MSR_RAPL_POWER_UNIT, &msr)) |
889facbe LB |
2331 | return; |
2332 | ||
2333 | rapl_power_units = 1.0 / (1 << (msr & 0xF)); | |
144b44b1 LB |
2334 | if (model == 0x37) |
2335 | rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000; | |
2336 | else | |
2337 | rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F)); | |
889facbe | 2338 | |
40ee8e3b AS |
2339 | rapl_dram_energy_units = rapl_dram_energy_units_probe(model, rapl_energy_units); |
2340 | ||
144b44b1 LB |
2341 | time_unit = msr >> 16 & 0xF; |
2342 | if (time_unit == 0) | |
2343 | time_unit = 0xA; | |
889facbe | 2344 | |
144b44b1 | 2345 | rapl_time_units = 1.0 / (1 << (time_unit)); |
889facbe | 2346 | |
144b44b1 | 2347 | tdp = get_tdp(model); |
889facbe | 2348 | |
144b44b1 | 2349 | rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp; |
d8af6f5f | 2350 | if (debug) |
144b44b1 | 2351 | fprintf(stderr, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp); |
889facbe LB |
2352 | |
2353 | return; | |
2354 | } | |
2355 | ||
3a9a941d LB |
2356 | void perf_limit_reasons_probe(family, model) |
2357 | { | |
2358 | if (!genuine_intel) | |
2359 | return; | |
2360 | ||
2361 | if (family != 6) | |
2362 | return; | |
2363 | ||
2364 | switch (model) { | |
2365 | case 0x3C: /* HSW */ | |
2366 | case 0x45: /* HSW */ | |
2367 | case 0x46: /* HSW */ | |
2368 | do_gfx_perf_limit_reasons = 1; | |
2369 | case 0x3F: /* HSX */ | |
2370 | do_core_perf_limit_reasons = 1; | |
2371 | do_ring_perf_limit_reasons = 1; | |
2372 | default: | |
2373 | return; | |
2374 | } | |
2375 | } | |
2376 | ||
889facbe LB |
2377 | int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p) |
2378 | { | |
2379 | unsigned long long msr; | |
2380 | unsigned int dts; | |
2381 | int cpu; | |
2382 | ||
2383 | if (!(do_dts || do_ptm)) | |
2384 | return 0; | |
2385 | ||
2386 | cpu = t->cpu_id; | |
2387 | ||
2388 | /* DTS is per-core, no need to print for each thread */ | |
2389 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
2390 | return 0; | |
2391 | ||
2392 | if (cpu_migrate(cpu)) { | |
2393 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
2394 | return -1; | |
2395 | } | |
2396 | ||
2397 | if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) { | |
2398 | if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr)) | |
2399 | return 0; | |
2400 | ||
2401 | dts = (msr >> 16) & 0x7F; | |
2402 | fprintf(stderr, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n", | |
2403 | cpu, msr, tcc_activation_temp - dts); | |
2404 | ||
2405 | #ifdef THERM_DEBUG | |
2406 | if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr)) | |
2407 | return 0; | |
2408 | ||
2409 | dts = (msr >> 16) & 0x7F; | |
2410 | dts2 = (msr >> 8) & 0x7F; | |
2411 | fprintf(stderr, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n", | |
2412 | cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2); | |
2413 | #endif | |
2414 | } | |
2415 | ||
2416 | ||
2417 | if (do_dts) { | |
2418 | unsigned int resolution; | |
2419 | ||
2420 | if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr)) | |
2421 | return 0; | |
2422 | ||
2423 | dts = (msr >> 16) & 0x7F; | |
2424 | resolution = (msr >> 27) & 0xF; | |
2425 | fprintf(stderr, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n", | |
2426 | cpu, msr, tcc_activation_temp - dts, resolution); | |
2427 | ||
2428 | #ifdef THERM_DEBUG | |
2429 | if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr)) | |
2430 | return 0; | |
2431 | ||
2432 | dts = (msr >> 16) & 0x7F; | |
2433 | dts2 = (msr >> 8) & 0x7F; | |
2434 | fprintf(stderr, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n", | |
2435 | cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2); | |
2436 | #endif | |
2437 | } | |
2438 | ||
2439 | return 0; | |
2440 | } | |
2441 | ||
2442 | void print_power_limit_msr(int cpu, unsigned long long msr, char *label) | |
2443 | { | |
2444 | fprintf(stderr, "cpu%d: %s: %sabled (%f Watts, %f sec, clamp %sabled)\n", | |
2445 | cpu, label, | |
2446 | ((msr >> 15) & 1) ? "EN" : "DIS", | |
2447 | ((msr >> 0) & 0x7FFF) * rapl_power_units, | |
2448 | (1.0 + (((msr >> 22) & 0x3)/4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units, | |
2449 | (((msr >> 16) & 1) ? "EN" : "DIS")); | |
2450 | ||
2451 | return; | |
2452 | } | |
2453 | ||
2454 | int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
2455 | { | |
2456 | unsigned long long msr; | |
2457 | int cpu; | |
889facbe LB |
2458 | |
2459 | if (!do_rapl) | |
2460 | return 0; | |
2461 | ||
2462 | /* RAPL counters are per package, so print only for 1st thread/package */ | |
2463 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
2464 | return 0; | |
2465 | ||
2466 | cpu = t->cpu_id; | |
2467 | if (cpu_migrate(cpu)) { | |
2468 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
2469 | return -1; | |
2470 | } | |
2471 | ||
2472 | if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr)) | |
2473 | return -1; | |
2474 | ||
d8af6f5f | 2475 | if (debug) { |
889facbe LB |
2476 | fprintf(stderr, "cpu%d: MSR_RAPL_POWER_UNIT: 0x%08llx " |
2477 | "(%f Watts, %f Joules, %f sec.)\n", cpu, msr, | |
144b44b1 | 2478 | rapl_power_units, rapl_energy_units, rapl_time_units); |
889facbe | 2479 | } |
144b44b1 LB |
2480 | if (do_rapl & RAPL_PKG_POWER_INFO) { |
2481 | ||
889facbe LB |
2482 | if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr)) |
2483 | return -5; | |
2484 | ||
2485 | ||
2486 | fprintf(stderr, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n", | |
2487 | cpu, msr, | |
2488 | ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2489 | ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2490 | ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2491 | ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units); | |
2492 | ||
144b44b1 LB |
2493 | } |
2494 | if (do_rapl & RAPL_PKG) { | |
2495 | ||
889facbe LB |
2496 | if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr)) |
2497 | return -9; | |
2498 | ||
2499 | fprintf(stderr, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n", | |
2500 | cpu, msr, (msr >> 63) & 1 ? "": "UN"); | |
2501 | ||
2502 | print_power_limit_msr(cpu, msr, "PKG Limit #1"); | |
2503 | fprintf(stderr, "cpu%d: PKG Limit #2: %sabled (%f Watts, %f* sec, clamp %sabled)\n", | |
2504 | cpu, | |
2505 | ((msr >> 47) & 1) ? "EN" : "DIS", | |
2506 | ((msr >> 32) & 0x7FFF) * rapl_power_units, | |
2507 | (1.0 + (((msr >> 54) & 0x3)/4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units, | |
2508 | ((msr >> 48) & 1) ? "EN" : "DIS"); | |
2509 | } | |
2510 | ||
0b2bb692 | 2511 | if (do_rapl & RAPL_DRAM_POWER_INFO) { |
889facbe LB |
2512 | if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr)) |
2513 | return -6; | |
2514 | ||
889facbe LB |
2515 | fprintf(stderr, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n", |
2516 | cpu, msr, | |
2517 | ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2518 | ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2519 | ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2520 | ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units); | |
0b2bb692 LB |
2521 | } |
2522 | if (do_rapl & RAPL_DRAM) { | |
889facbe LB |
2523 | if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr)) |
2524 | return -9; | |
2525 | fprintf(stderr, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n", | |
2526 | cpu, msr, (msr >> 31) & 1 ? "": "UN"); | |
2527 | ||
2528 | print_power_limit_msr(cpu, msr, "DRAM Limit"); | |
2529 | } | |
144b44b1 | 2530 | if (do_rapl & RAPL_CORE_POLICY) { |
d8af6f5f | 2531 | if (debug) { |
889facbe LB |
2532 | if (get_msr(cpu, MSR_PP0_POLICY, &msr)) |
2533 | return -7; | |
2534 | ||
2535 | fprintf(stderr, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF); | |
144b44b1 LB |
2536 | } |
2537 | } | |
2538 | if (do_rapl & RAPL_CORES) { | |
d8af6f5f | 2539 | if (debug) { |
889facbe LB |
2540 | |
2541 | if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr)) | |
2542 | return -9; | |
2543 | fprintf(stderr, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n", | |
2544 | cpu, msr, (msr >> 31) & 1 ? "": "UN"); | |
2545 | print_power_limit_msr(cpu, msr, "Cores Limit"); | |
2546 | } | |
2547 | } | |
2548 | if (do_rapl & RAPL_GFX) { | |
d8af6f5f | 2549 | if (debug) { |
889facbe LB |
2550 | if (get_msr(cpu, MSR_PP1_POLICY, &msr)) |
2551 | return -8; | |
2552 | ||
2553 | fprintf(stderr, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF); | |
2554 | ||
2555 | if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr)) | |
2556 | return -9; | |
2557 | fprintf(stderr, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n", | |
2558 | cpu, msr, (msr >> 31) & 1 ? "": "UN"); | |
2559 | print_power_limit_msr(cpu, msr, "GFX Limit"); | |
2560 | } | |
2561 | } | |
2562 | return 0; | |
2563 | } | |
2564 | ||
d7899447 LB |
2565 | /* |
2566 | * SNB adds support for additional MSRs: | |
2567 | * | |
2568 | * MSR_PKG_C7_RESIDENCY 0x000003fa | |
2569 | * MSR_CORE_C7_RESIDENCY 0x000003fe | |
2570 | * MSR_PKG_C2_RESIDENCY 0x0000060d | |
2571 | */ | |
103a8fea | 2572 | |
d7899447 | 2573 | int has_snb_msrs(unsigned int family, unsigned int model) |
103a8fea LB |
2574 | { |
2575 | if (!genuine_intel) | |
2576 | return 0; | |
2577 | ||
2578 | switch (model) { | |
2579 | case 0x2A: | |
2580 | case 0x2D: | |
650a37f3 | 2581 | case 0x3A: /* IVB */ |
1300651b | 2582 | case 0x3E: /* IVB Xeon */ |
70b43400 LB |
2583 | case 0x3C: /* HSW */ |
2584 | case 0x3F: /* HSW */ | |
2585 | case 0x45: /* HSW */ | |
149c2319 | 2586 | case 0x46: /* HSW */ |
4e8e863f | 2587 | case 0x3D: /* BDW */ |
48a0631c | 2588 | case 0x47: /* BDW */ |
4e8e863f LB |
2589 | case 0x4F: /* BDX */ |
2590 | case 0x56: /* BDX-DE */ | |
0b2bb692 LB |
2591 | case 0x4E: /* SKL */ |
2592 | case 0x5E: /* SKL */ | |
103a8fea LB |
2593 | return 1; |
2594 | } | |
2595 | return 0; | |
2596 | } | |
2597 | ||
d7899447 LB |
2598 | /* |
2599 | * HSW adds support for additional MSRs: | |
2600 | * | |
2601 | * MSR_PKG_C8_RESIDENCY 0x00000630 | |
2602 | * MSR_PKG_C9_RESIDENCY 0x00000631 | |
2603 | * MSR_PKG_C10_RESIDENCY 0x00000632 | |
2604 | */ | |
2605 | int has_hsw_msrs(unsigned int family, unsigned int model) | |
ca58710f KCA |
2606 | { |
2607 | if (!genuine_intel) | |
2608 | return 0; | |
2609 | ||
2610 | switch (model) { | |
4e8e863f LB |
2611 | case 0x45: /* HSW */ |
2612 | case 0x3D: /* BDW */ | |
0b2bb692 LB |
2613 | case 0x4E: /* SKL */ |
2614 | case 0x5E: /* SKL */ | |
2615 | return 1; | |
2616 | } | |
2617 | return 0; | |
2618 | } | |
2619 | ||
2620 | /* | |
2621 | * SKL adds support for additional MSRS: | |
2622 | * | |
2623 | * MSR_PKG_WEIGHTED_CORE_C0_RES 0x00000658 | |
2624 | * MSR_PKG_ANY_CORE_C0_RES 0x00000659 | |
2625 | * MSR_PKG_ANY_GFXE_C0_RES 0x0000065A | |
2626 | * MSR_PKG_BOTH_CORE_GFXE_C0_RES 0x0000065B | |
2627 | */ | |
2628 | int has_skl_msrs(unsigned int family, unsigned int model) | |
2629 | { | |
2630 | if (!genuine_intel) | |
2631 | return 0; | |
2632 | ||
2633 | switch (model) { | |
2634 | case 0x4E: /* SKL */ | |
2635 | case 0x5E: /* SKL */ | |
ca58710f KCA |
2636 | return 1; |
2637 | } | |
2638 | return 0; | |
2639 | } | |
2640 | ||
2641 | ||
0b2bb692 | 2642 | |
144b44b1 LB |
2643 | int is_slm(unsigned int family, unsigned int model) |
2644 | { | |
2645 | if (!genuine_intel) | |
2646 | return 0; | |
2647 | switch (model) { | |
2648 | case 0x37: /* BYT */ | |
2649 | case 0x4D: /* AVN */ | |
2650 | return 1; | |
2651 | } | |
2652 | return 0; | |
2653 | } | |
2654 | ||
fb5d4327 DC |
2655 | int is_knl(unsigned int family, unsigned int model) |
2656 | { | |
2657 | if (!genuine_intel) | |
2658 | return 0; | |
2659 | switch (model) { | |
2660 | case 0x57: /* KNL */ | |
2661 | return 1; | |
2662 | } | |
2663 | return 0; | |
2664 | } | |
2665 | ||
b2b34dfe HC |
2666 | unsigned int get_aperf_mperf_multiplier(unsigned int family, unsigned int model) |
2667 | { | |
2668 | if (is_knl(family, model)) | |
2669 | return 1024; | |
2670 | return 1; | |
2671 | } | |
2672 | ||
144b44b1 LB |
2673 | #define SLM_BCLK_FREQS 5 |
2674 | double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0}; | |
2675 | ||
2676 | double slm_bclk(void) | |
2677 | { | |
2678 | unsigned long long msr = 3; | |
2679 | unsigned int i; | |
2680 | double freq; | |
2681 | ||
7ce7d5de | 2682 | if (get_msr(base_cpu, MSR_FSB_FREQ, &msr)) |
144b44b1 LB |
2683 | fprintf(stderr, "SLM BCLK: unknown\n"); |
2684 | ||
2685 | i = msr & 0xf; | |
2686 | if (i >= SLM_BCLK_FREQS) { | |
2687 | fprintf(stderr, "SLM BCLK[%d] invalid\n", i); | |
2688 | msr = 3; | |
2689 | } | |
2690 | freq = slm_freq_table[i]; | |
2691 | ||
2692 | fprintf(stderr, "SLM BCLK: %.1f Mhz\n", freq); | |
2693 | ||
2694 | return freq; | |
2695 | } | |
2696 | ||
103a8fea LB |
2697 | double discover_bclk(unsigned int family, unsigned int model) |
2698 | { | |
d7899447 | 2699 | if (has_snb_msrs(family, model)) |
103a8fea | 2700 | return 100.00; |
144b44b1 LB |
2701 | else if (is_slm(family, model)) |
2702 | return slm_bclk(); | |
103a8fea LB |
2703 | else |
2704 | return 133.33; | |
2705 | } | |
2706 | ||
889facbe LB |
2707 | /* |
2708 | * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where | |
2709 | * the Thermal Control Circuit (TCC) activates. | |
2710 | * This is usually equal to tjMax. | |
2711 | * | |
2712 | * Older processors do not have this MSR, so there we guess, | |
2713 | * but also allow cmdline over-ride with -T. | |
2714 | * | |
2715 | * Several MSR temperature values are in units of degrees-C | |
2716 | * below this value, including the Digital Thermal Sensor (DTS), | |
2717 | * Package Thermal Management Sensor (PTM), and thermal event thresholds. | |
2718 | */ | |
2719 | int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
2720 | { | |
2721 | unsigned long long msr; | |
2722 | unsigned int target_c_local; | |
2723 | int cpu; | |
2724 | ||
2725 | /* tcc_activation_temp is used only for dts or ptm */ | |
2726 | if (!(do_dts || do_ptm)) | |
2727 | return 0; | |
2728 | ||
2729 | /* this is a per-package concept */ | |
2730 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
2731 | return 0; | |
2732 | ||
2733 | cpu = t->cpu_id; | |
2734 | if (cpu_migrate(cpu)) { | |
2735 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
2736 | return -1; | |
2737 | } | |
2738 | ||
2739 | if (tcc_activation_temp_override != 0) { | |
2740 | tcc_activation_temp = tcc_activation_temp_override; | |
2741 | fprintf(stderr, "cpu%d: Using cmdline TCC Target (%d C)\n", | |
2742 | cpu, tcc_activation_temp); | |
2743 | return 0; | |
2744 | } | |
2745 | ||
2746 | /* Temperature Target MSR is Nehalem and newer only */ | |
d7899447 | 2747 | if (!do_nhm_platform_info) |
889facbe LB |
2748 | goto guess; |
2749 | ||
7ce7d5de | 2750 | if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr)) |
889facbe LB |
2751 | goto guess; |
2752 | ||
3482124a | 2753 | target_c_local = (msr >> 16) & 0xFF; |
889facbe | 2754 | |
d8af6f5f | 2755 | if (debug) |
889facbe LB |
2756 | fprintf(stderr, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n", |
2757 | cpu, msr, target_c_local); | |
2758 | ||
3482124a | 2759 | if (!target_c_local) |
889facbe LB |
2760 | goto guess; |
2761 | ||
2762 | tcc_activation_temp = target_c_local; | |
2763 | ||
2764 | return 0; | |
2765 | ||
2766 | guess: | |
2767 | tcc_activation_temp = TJMAX_DEFAULT; | |
2768 | fprintf(stderr, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n", | |
2769 | cpu, tcc_activation_temp); | |
2770 | ||
2771 | return 0; | |
2772 | } | |
69807a63 LB |
2773 | |
2774 | void decode_misc_enable_msr(void) | |
2775 | { | |
2776 | unsigned long long msr; | |
2777 | ||
2778 | if (!get_msr(base_cpu, MSR_IA32_MISC_ENABLE, &msr)) | |
2779 | fprintf(stderr, "cpu%d: MSR_IA32_MISC_ENABLE: 0x%08llx (%s %s %s)\n", | |
2780 | base_cpu, msr, | |
2781 | msr & (1 << 3) ? "TCC" : "", | |
2782 | msr & (1 << 16) ? "EIST" : "", | |
2783 | msr & (1 << 18) ? "MONITOR" : ""); | |
2784 | } | |
2785 | ||
f0057310 LB |
2786 | /* |
2787 | * Decode MSR_MISC_PWR_MGMT | |
2788 | * | |
2789 | * Decode the bits according to the Nehalem documentation | |
2790 | * bit[0] seems to continue to have same meaning going forward | |
2791 | * bit[1] less so... | |
2792 | */ | |
2793 | void decode_misc_pwr_mgmt_msr(void) | |
2794 | { | |
2795 | unsigned long long msr; | |
2796 | ||
2797 | if (!do_nhm_platform_info) | |
2798 | return; | |
2799 | ||
2800 | if (!get_msr(base_cpu, MSR_MISC_PWR_MGMT, &msr)) | |
2801 | fprintf(stderr, "cpu%d: MSR_MISC_PWR_MGMT: 0x%08llx (%sable-EIST_Coordination %sable-EPB)\n", | |
2802 | base_cpu, msr, | |
2803 | msr & (1 << 0) ? "DIS" : "EN", | |
2804 | msr & (1 << 1) ? "EN" : "DIS"); | |
2805 | } | |
7f5c258e | 2806 | |
fcd17211 | 2807 | void process_cpuid() |
103a8fea | 2808 | { |
61a87ba7 | 2809 | unsigned int eax, ebx, ecx, edx, max_level, max_extended_level; |
103a8fea LB |
2810 | unsigned int fms, family, model, stepping; |
2811 | ||
2812 | eax = ebx = ecx = edx = 0; | |
2813 | ||
2b92865e | 2814 | __get_cpuid(0, &max_level, &ebx, &ecx, &edx); |
103a8fea LB |
2815 | |
2816 | if (ebx == 0x756e6547 && edx == 0x49656e69 && ecx == 0x6c65746e) | |
2817 | genuine_intel = 1; | |
2818 | ||
d8af6f5f | 2819 | if (debug) |
889facbe | 2820 | fprintf(stderr, "CPUID(0): %.4s%.4s%.4s ", |
103a8fea LB |
2821 | (char *)&ebx, (char *)&edx, (char *)&ecx); |
2822 | ||
2b92865e | 2823 | __get_cpuid(1, &fms, &ebx, &ecx, &edx); |
103a8fea LB |
2824 | family = (fms >> 8) & 0xf; |
2825 | model = (fms >> 4) & 0xf; | |
2826 | stepping = fms & 0xf; | |
2827 | if (family == 6 || family == 0xf) | |
2828 | model += ((fms >> 16) & 0xf) << 4; | |
2829 | ||
69807a63 | 2830 | if (debug) { |
103a8fea LB |
2831 | fprintf(stderr, "%d CPUID levels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n", |
2832 | max_level, family, model, stepping, family, model, stepping); | |
69807a63 LB |
2833 | fprintf(stderr, "CPUID(1): %s %s %s %s %s %s %s %s\n", |
2834 | ecx & (1 << 0) ? "SSE3" : "-", | |
2835 | ecx & (1 << 3) ? "MONITOR" : "-", | |
2836 | ecx & (1 << 7) ? "EIST" : "-", | |
2837 | ecx & (1 << 8) ? "TM2" : "-", | |
2838 | edx & (1 << 4) ? "TSC" : "-", | |
2839 | edx & (1 << 5) ? "MSR" : "-", | |
2840 | edx & (1 << 22) ? "ACPI-TM" : "-", | |
2841 | edx & (1 << 29) ? "TM" : "-"); | |
2842 | } | |
103a8fea | 2843 | |
b2c95d90 JT |
2844 | if (!(edx & (1 << 5))) |
2845 | errx(1, "CPUID: no MSR"); | |
103a8fea LB |
2846 | |
2847 | /* | |
2848 | * check max extended function levels of CPUID. | |
2849 | * This is needed to check for invariant TSC. | |
2850 | * This check is valid for both Intel and AMD. | |
2851 | */ | |
2852 | ebx = ecx = edx = 0; | |
61a87ba7 | 2853 | __get_cpuid(0x80000000, &max_extended_level, &ebx, &ecx, &edx); |
103a8fea | 2854 | |
61a87ba7 | 2855 | if (max_extended_level >= 0x80000007) { |
103a8fea | 2856 | |
d7899447 LB |
2857 | /* |
2858 | * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8 | |
2859 | * this check is valid for both Intel and AMD | |
2860 | */ | |
2861 | __get_cpuid(0x80000007, &eax, &ebx, &ecx, &edx); | |
2862 | has_invariant_tsc = edx & (1 << 8); | |
2863 | } | |
103a8fea LB |
2864 | |
2865 | /* | |
2866 | * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0 | |
2867 | * this check is valid for both Intel and AMD | |
2868 | */ | |
2869 | ||
2b92865e | 2870 | __get_cpuid(0x6, &eax, &ebx, &ecx, &edx); |
8209e054 | 2871 | has_aperf = ecx & (1 << 0); |
889facbe LB |
2872 | do_dts = eax & (1 << 0); |
2873 | do_ptm = eax & (1 << 6); | |
7f5c258e LB |
2874 | has_hwp = eax & (1 << 7); |
2875 | has_hwp_notify = eax & (1 << 8); | |
2876 | has_hwp_activity_window = eax & (1 << 9); | |
2877 | has_hwp_epp = eax & (1 << 10); | |
2878 | has_hwp_pkg = eax & (1 << 11); | |
889facbe LB |
2879 | has_epb = ecx & (1 << 3); |
2880 | ||
d8af6f5f | 2881 | if (debug) |
7f5c258e LB |
2882 | fprintf(stderr, "CPUID(6): %sAPERF, %sDTS, %sPTM, %sHWP, " |
2883 | "%sHWPnotify, %sHWPwindow, %sHWPepp, %sHWPpkg, %sEPB\n", | |
2884 | has_aperf ? "" : "No-", | |
2885 | do_dts ? "" : "No-", | |
2886 | do_ptm ? "" : "No-", | |
2887 | has_hwp ? "" : "No-", | |
2888 | has_hwp_notify ? "" : "No-", | |
2889 | has_hwp_activity_window ? "" : "No-", | |
2890 | has_hwp_epp ? "" : "No-", | |
2891 | has_hwp_pkg ? "" : "No-", | |
2892 | has_epb ? "" : "No-"); | |
103a8fea | 2893 | |
69807a63 LB |
2894 | if (debug) |
2895 | decode_misc_enable_msr(); | |
2896 | ||
61a87ba7 | 2897 | if (max_level >= 0x15) { |
8a5bdf41 LB |
2898 | unsigned int eax_crystal; |
2899 | unsigned int ebx_tsc; | |
2900 | ||
2901 | /* | |
2902 | * CPUID 15H TSC/Crystal ratio, possibly Crystal Hz | |
2903 | */ | |
2904 | eax_crystal = ebx_tsc = crystal_hz = edx = 0; | |
2905 | __get_cpuid(0x15, &eax_crystal, &ebx_tsc, &crystal_hz, &edx); | |
2906 | ||
2907 | if (ebx_tsc != 0) { | |
2908 | ||
2909 | if (debug && (ebx != 0)) | |
2910 | fprintf(stderr, "CPUID(0x15): eax_crystal: %d ebx_tsc: %d ecx_crystal_hz: %d\n", | |
2911 | eax_crystal, ebx_tsc, crystal_hz); | |
2912 | ||
2913 | if (crystal_hz == 0) | |
2914 | switch(model) { | |
2915 | case 0x4E: /* SKL */ | |
2916 | case 0x5E: /* SKL */ | |
2917 | crystal_hz = 24000000; /* 24 MHz */ | |
2918 | break; | |
2919 | default: | |
2920 | crystal_hz = 0; | |
2921 | } | |
2922 | ||
2923 | if (crystal_hz) { | |
2924 | tsc_hz = (unsigned long long) crystal_hz * ebx_tsc / eax_crystal; | |
2925 | if (debug) | |
2926 | fprintf(stderr, "TSC: %lld MHz (%d Hz * %d / %d / 1000000)\n", | |
2927 | tsc_hz / 1000000, crystal_hz, ebx_tsc, eax_crystal); | |
2928 | } | |
2929 | } | |
2930 | } | |
61a87ba7 LB |
2931 | if (max_level >= 0x16) { |
2932 | unsigned int base_mhz, max_mhz, bus_mhz, edx; | |
2933 | ||
2934 | /* | |
2935 | * CPUID 16H Base MHz, Max MHz, Bus MHz | |
2936 | */ | |
2937 | base_mhz = max_mhz = bus_mhz = edx = 0; | |
2938 | ||
2939 | __get_cpuid(0x16, &base_mhz, &max_mhz, &bus_mhz, &edx); | |
2940 | if (debug) | |
2941 | fprintf(stderr, "CPUID(0x16): base_mhz: %d max_mhz: %d bus_mhz: %d\n", | |
2942 | base_mhz, max_mhz, bus_mhz); | |
2943 | } | |
8a5bdf41 | 2944 | |
b2b34dfe HC |
2945 | if (has_aperf) |
2946 | aperf_mperf_multiplier = get_aperf_mperf_multiplier(family, model); | |
2947 | ||
ee7e38e3 | 2948 | do_nhm_platform_info = do_nhm_cstates = do_smi = probe_nhm_msrs(family, model); |
d7899447 | 2949 | do_snb_cstates = has_snb_msrs(family, model); |
ee7e38e3 LB |
2950 | do_pc2 = do_snb_cstates && (pkg_cstate_limit >= PCL__2); |
2951 | do_pc3 = (pkg_cstate_limit >= PCL__3); | |
2952 | do_pc6 = (pkg_cstate_limit >= PCL__6); | |
2953 | do_pc7 = do_snb_cstates && (pkg_cstate_limit >= PCL__7); | |
d7899447 | 2954 | do_c8_c9_c10 = has_hsw_msrs(family, model); |
0b2bb692 | 2955 | do_skl_residency = has_skl_msrs(family, model); |
144b44b1 | 2956 | do_slm_cstates = is_slm(family, model); |
fb5d4327 | 2957 | do_knl_cstates = is_knl(family, model); |
103a8fea | 2958 | |
f0057310 LB |
2959 | if (debug) |
2960 | decode_misc_pwr_mgmt_msr(); | |
2961 | ||
889facbe | 2962 | rapl_probe(family, model); |
3a9a941d | 2963 | perf_limit_reasons_probe(family, model); |
889facbe | 2964 | |
fcd17211 LB |
2965 | if (debug) |
2966 | dump_cstate_pstate_config_info(); | |
2967 | ||
a2b7b749 LB |
2968 | if (has_skl_msrs(family, model)) |
2969 | calculate_tsc_tweak(); | |
2970 | ||
889facbe | 2971 | return; |
103a8fea LB |
2972 | } |
2973 | ||
d8af6f5f | 2974 | void help() |
103a8fea | 2975 | { |
d8af6f5f LB |
2976 | fprintf(stderr, |
2977 | "Usage: turbostat [OPTIONS][(--interval seconds) | COMMAND ...]\n" | |
2978 | "\n" | |
2979 | "Turbostat forks the specified COMMAND and prints statistics\n" | |
2980 | "when COMMAND completes.\n" | |
2981 | "If no COMMAND is specified, turbostat wakes every 5-seconds\n" | |
2982 | "to print statistics, until interrupted.\n" | |
2983 | "--debug run in \"debug\" mode\n" | |
2984 | "--interval sec Override default 5-second measurement interval\n" | |
2985 | "--help print this help message\n" | |
2986 | "--counter msr print 32-bit counter at address \"msr\"\n" | |
2987 | "--Counter msr print 64-bit Counter at address \"msr\"\n" | |
2988 | "--msr msr print 32-bit value at address \"msr\"\n" | |
2989 | "--MSR msr print 64-bit Value at address \"msr\"\n" | |
2990 | "--version print version information\n" | |
2991 | "\n" | |
2992 | "For more help, run \"man turbostat\"\n"); | |
103a8fea LB |
2993 | } |
2994 | ||
2995 | ||
2996 | /* | |
2997 | * in /dev/cpu/ return success for names that are numbers | |
2998 | * ie. filter out ".", "..", "microcode". | |
2999 | */ | |
3000 | int dir_filter(const struct dirent *dirp) | |
3001 | { | |
3002 | if (isdigit(dirp->d_name[0])) | |
3003 | return 1; | |
3004 | else | |
3005 | return 0; | |
3006 | } | |
3007 | ||
3008 | int open_dev_cpu_msr(int dummy1) | |
3009 | { | |
3010 | return 0; | |
3011 | } | |
3012 | ||
c98d5d94 LB |
3013 | void topology_probe() |
3014 | { | |
3015 | int i; | |
3016 | int max_core_id = 0; | |
3017 | int max_package_id = 0; | |
3018 | int max_siblings = 0; | |
3019 | struct cpu_topology { | |
3020 | int core_id; | |
3021 | int physical_package_id; | |
3022 | } *cpus; | |
3023 | ||
3024 | /* Initialize num_cpus, max_cpu_num */ | |
3025 | topo.num_cpus = 0; | |
3026 | topo.max_cpu_num = 0; | |
3027 | for_all_proc_cpus(count_cpus); | |
3028 | if (!summary_only && topo.num_cpus > 1) | |
3029 | show_cpu = 1; | |
3030 | ||
d8af6f5f | 3031 | if (debug > 1) |
c98d5d94 LB |
3032 | fprintf(stderr, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num); |
3033 | ||
3034 | cpus = calloc(1, (topo.max_cpu_num + 1) * sizeof(struct cpu_topology)); | |
b2c95d90 JT |
3035 | if (cpus == NULL) |
3036 | err(1, "calloc cpus"); | |
c98d5d94 LB |
3037 | |
3038 | /* | |
3039 | * Allocate and initialize cpu_present_set | |
3040 | */ | |
3041 | cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1)); | |
b2c95d90 JT |
3042 | if (cpu_present_set == NULL) |
3043 | err(3, "CPU_ALLOC"); | |
c98d5d94 LB |
3044 | cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1)); |
3045 | CPU_ZERO_S(cpu_present_setsize, cpu_present_set); | |
3046 | for_all_proc_cpus(mark_cpu_present); | |
3047 | ||
3048 | /* | |
3049 | * Allocate and initialize cpu_affinity_set | |
3050 | */ | |
3051 | cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1)); | |
b2c95d90 JT |
3052 | if (cpu_affinity_set == NULL) |
3053 | err(3, "CPU_ALLOC"); | |
c98d5d94 LB |
3054 | cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1)); |
3055 | CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set); | |
3056 | ||
3057 | ||
3058 | /* | |
3059 | * For online cpus | |
3060 | * find max_core_id, max_package_id | |
3061 | */ | |
3062 | for (i = 0; i <= topo.max_cpu_num; ++i) { | |
3063 | int siblings; | |
3064 | ||
3065 | if (cpu_is_not_present(i)) { | |
d8af6f5f | 3066 | if (debug > 1) |
c98d5d94 LB |
3067 | fprintf(stderr, "cpu%d NOT PRESENT\n", i); |
3068 | continue; | |
3069 | } | |
3070 | cpus[i].core_id = get_core_id(i); | |
3071 | if (cpus[i].core_id > max_core_id) | |
3072 | max_core_id = cpus[i].core_id; | |
3073 | ||
3074 | cpus[i].physical_package_id = get_physical_package_id(i); | |
3075 | if (cpus[i].physical_package_id > max_package_id) | |
3076 | max_package_id = cpus[i].physical_package_id; | |
3077 | ||
3078 | siblings = get_num_ht_siblings(i); | |
3079 | if (siblings > max_siblings) | |
3080 | max_siblings = siblings; | |
d8af6f5f | 3081 | if (debug > 1) |
c98d5d94 LB |
3082 | fprintf(stderr, "cpu %d pkg %d core %d\n", |
3083 | i, cpus[i].physical_package_id, cpus[i].core_id); | |
3084 | } | |
3085 | topo.num_cores_per_pkg = max_core_id + 1; | |
d8af6f5f | 3086 | if (debug > 1) |
c98d5d94 LB |
3087 | fprintf(stderr, "max_core_id %d, sizing for %d cores per package\n", |
3088 | max_core_id, topo.num_cores_per_pkg); | |
1cc21f7b | 3089 | if (debug && !summary_only && topo.num_cores_per_pkg > 1) |
c98d5d94 LB |
3090 | show_core = 1; |
3091 | ||
3092 | topo.num_packages = max_package_id + 1; | |
d8af6f5f | 3093 | if (debug > 1) |
c98d5d94 LB |
3094 | fprintf(stderr, "max_package_id %d, sizing for %d packages\n", |
3095 | max_package_id, topo.num_packages); | |
1cc21f7b | 3096 | if (debug && !summary_only && topo.num_packages > 1) |
c98d5d94 LB |
3097 | show_pkg = 1; |
3098 | ||
3099 | topo.num_threads_per_core = max_siblings; | |
d8af6f5f | 3100 | if (debug > 1) |
c98d5d94 LB |
3101 | fprintf(stderr, "max_siblings %d\n", max_siblings); |
3102 | ||
3103 | free(cpus); | |
3104 | } | |
3105 | ||
3106 | void | |
3107 | allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p) | |
3108 | { | |
3109 | int i; | |
3110 | ||
3111 | *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg * | |
3112 | topo.num_packages, sizeof(struct thread_data)); | |
3113 | if (*t == NULL) | |
3114 | goto error; | |
3115 | ||
3116 | for (i = 0; i < topo.num_threads_per_core * | |
3117 | topo.num_cores_per_pkg * topo.num_packages; i++) | |
3118 | (*t)[i].cpu_id = -1; | |
3119 | ||
3120 | *c = calloc(topo.num_cores_per_pkg * topo.num_packages, | |
3121 | sizeof(struct core_data)); | |
3122 | if (*c == NULL) | |
3123 | goto error; | |
3124 | ||
3125 | for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++) | |
3126 | (*c)[i].core_id = -1; | |
3127 | ||
3128 | *p = calloc(topo.num_packages, sizeof(struct pkg_data)); | |
3129 | if (*p == NULL) | |
3130 | goto error; | |
3131 | ||
3132 | for (i = 0; i < topo.num_packages; i++) | |
3133 | (*p)[i].package_id = i; | |
3134 | ||
3135 | return; | |
3136 | error: | |
b2c95d90 | 3137 | err(1, "calloc counters"); |
c98d5d94 LB |
3138 | } |
3139 | /* | |
3140 | * init_counter() | |
3141 | * | |
3142 | * set cpu_id, core_num, pkg_num | |
3143 | * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE | |
3144 | * | |
3145 | * increment topo.num_cores when 1st core in pkg seen | |
3146 | */ | |
3147 | void init_counter(struct thread_data *thread_base, struct core_data *core_base, | |
3148 | struct pkg_data *pkg_base, int thread_num, int core_num, | |
3149 | int pkg_num, int cpu_id) | |
3150 | { | |
3151 | struct thread_data *t; | |
3152 | struct core_data *c; | |
3153 | struct pkg_data *p; | |
3154 | ||
3155 | t = GET_THREAD(thread_base, thread_num, core_num, pkg_num); | |
3156 | c = GET_CORE(core_base, core_num, pkg_num); | |
3157 | p = GET_PKG(pkg_base, pkg_num); | |
3158 | ||
3159 | t->cpu_id = cpu_id; | |
3160 | if (thread_num == 0) { | |
3161 | t->flags |= CPU_IS_FIRST_THREAD_IN_CORE; | |
3162 | if (cpu_is_first_core_in_package(cpu_id)) | |
3163 | t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE; | |
3164 | } | |
3165 | ||
3166 | c->core_id = core_num; | |
3167 | p->package_id = pkg_num; | |
3168 | } | |
3169 | ||
3170 | ||
3171 | int initialize_counters(int cpu_id) | |
3172 | { | |
3173 | int my_thread_id, my_core_id, my_package_id; | |
3174 | ||
3175 | my_package_id = get_physical_package_id(cpu_id); | |
3176 | my_core_id = get_core_id(cpu_id); | |
e275b388 DC |
3177 | my_thread_id = get_cpu_position_in_core(cpu_id); |
3178 | if (!my_thread_id) | |
c98d5d94 | 3179 | topo.num_cores++; |
c98d5d94 LB |
3180 | |
3181 | init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id); | |
3182 | init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id); | |
3183 | return 0; | |
3184 | } | |
3185 | ||
3186 | void allocate_output_buffer() | |
3187 | { | |
3b4d5c7f | 3188 | output_buffer = calloc(1, (1 + topo.num_cpus) * 1024); |
c98d5d94 | 3189 | outp = output_buffer; |
b2c95d90 JT |
3190 | if (outp == NULL) |
3191 | err(-1, "calloc output buffer"); | |
c98d5d94 LB |
3192 | } |
3193 | ||
3194 | void setup_all_buffers(void) | |
3195 | { | |
3196 | topology_probe(); | |
3197 | allocate_counters(&thread_even, &core_even, &package_even); | |
3198 | allocate_counters(&thread_odd, &core_odd, &package_odd); | |
3199 | allocate_output_buffer(); | |
3200 | for_all_proc_cpus(initialize_counters); | |
3201 | } | |
3b4d5c7f | 3202 | |
7ce7d5de PB |
3203 | void set_base_cpu(void) |
3204 | { | |
3205 | base_cpu = sched_getcpu(); | |
3206 | if (base_cpu < 0) | |
3207 | err(-ENODEV, "No valid cpus found"); | |
3208 | ||
3209 | if (debug > 1) | |
3210 | fprintf(stderr, "base_cpu = %d\n", base_cpu); | |
3211 | } | |
3212 | ||
103a8fea LB |
3213 | void turbostat_init() |
3214 | { | |
7ce7d5de PB |
3215 | setup_all_buffers(); |
3216 | set_base_cpu(); | |
103a8fea | 3217 | check_dev_msr(); |
98481e79 | 3218 | check_permissions(); |
fcd17211 | 3219 | process_cpuid(); |
103a8fea | 3220 | |
103a8fea | 3221 | |
7f5c258e LB |
3222 | if (debug) |
3223 | for_all_cpus(print_hwp, ODD_COUNTERS); | |
3224 | ||
d8af6f5f | 3225 | if (debug) |
889facbe LB |
3226 | for_all_cpus(print_epb, ODD_COUNTERS); |
3227 | ||
d8af6f5f | 3228 | if (debug) |
3a9a941d LB |
3229 | for_all_cpus(print_perf_limit, ODD_COUNTERS); |
3230 | ||
d8af6f5f | 3231 | if (debug) |
889facbe LB |
3232 | for_all_cpus(print_rapl, ODD_COUNTERS); |
3233 | ||
3234 | for_all_cpus(set_temperature_target, ODD_COUNTERS); | |
3235 | ||
d8af6f5f | 3236 | if (debug) |
889facbe | 3237 | for_all_cpus(print_thermal, ODD_COUNTERS); |
103a8fea LB |
3238 | } |
3239 | ||
3240 | int fork_it(char **argv) | |
3241 | { | |
103a8fea | 3242 | pid_t child_pid; |
d91bb17c | 3243 | int status; |
d15cf7c1 | 3244 | |
d91bb17c LB |
3245 | status = for_all_cpus(get_counters, EVEN_COUNTERS); |
3246 | if (status) | |
3247 | exit(status); | |
c98d5d94 LB |
3248 | /* clear affinity side-effect of get_counters() */ |
3249 | sched_setaffinity(0, cpu_present_setsize, cpu_present_set); | |
103a8fea LB |
3250 | gettimeofday(&tv_even, (struct timezone *)NULL); |
3251 | ||
3252 | child_pid = fork(); | |
3253 | if (!child_pid) { | |
3254 | /* child */ | |
3255 | execvp(argv[0], argv); | |
3256 | } else { | |
103a8fea LB |
3257 | |
3258 | /* parent */ | |
b2c95d90 JT |
3259 | if (child_pid == -1) |
3260 | err(1, "fork"); | |
103a8fea LB |
3261 | |
3262 | signal(SIGINT, SIG_IGN); | |
3263 | signal(SIGQUIT, SIG_IGN); | |
b2c95d90 JT |
3264 | if (waitpid(child_pid, &status, 0) == -1) |
3265 | err(status, "waitpid"); | |
103a8fea | 3266 | } |
c98d5d94 LB |
3267 | /* |
3268 | * n.b. fork_it() does not check for errors from for_all_cpus() | |
3269 | * because re-starting is problematic when forking | |
3270 | */ | |
3271 | for_all_cpus(get_counters, ODD_COUNTERS); | |
103a8fea | 3272 | gettimeofday(&tv_odd, (struct timezone *)NULL); |
103a8fea | 3273 | timersub(&tv_odd, &tv_even, &tv_delta); |
c98d5d94 LB |
3274 | for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS); |
3275 | compute_average(EVEN_COUNTERS); | |
3276 | format_all_counters(EVEN_COUNTERS); | |
3277 | flush_stderr(); | |
103a8fea | 3278 | |
6eab04a8 | 3279 | fprintf(stderr, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0); |
103a8fea | 3280 | |
d91bb17c | 3281 | return status; |
103a8fea LB |
3282 | } |
3283 | ||
3b4d5c7f AS |
3284 | int get_and_dump_counters(void) |
3285 | { | |
3286 | int status; | |
3287 | ||
3288 | status = for_all_cpus(get_counters, ODD_COUNTERS); | |
3289 | if (status) | |
3290 | return status; | |
3291 | ||
3292 | status = for_all_cpus(dump_counters, ODD_COUNTERS); | |
3293 | if (status) | |
3294 | return status; | |
3295 | ||
3296 | flush_stdout(); | |
3297 | ||
3298 | return status; | |
3299 | } | |
3300 | ||
d8af6f5f | 3301 | void print_version() { |
7f5c258e | 3302 | fprintf(stderr, "turbostat version 4.10 10 Dec, 2015" |
d8af6f5f LB |
3303 | " - Len Brown <lenb@kernel.org>\n"); |
3304 | } | |
3305 | ||
103a8fea LB |
3306 | void cmdline(int argc, char **argv) |
3307 | { | |
3308 | int opt; | |
d8af6f5f LB |
3309 | int option_index = 0; |
3310 | static struct option long_options[] = { | |
3311 | {"Counter", required_argument, 0, 'C'}, | |
3312 | {"counter", required_argument, 0, 'c'}, | |
3313 | {"Dump", no_argument, 0, 'D'}, | |
3314 | {"debug", no_argument, 0, 'd'}, | |
3315 | {"interval", required_argument, 0, 'i'}, | |
3316 | {"help", no_argument, 0, 'h'}, | |
3317 | {"Joules", no_argument, 0, 'J'}, | |
3318 | {"MSR", required_argument, 0, 'M'}, | |
3319 | {"msr", required_argument, 0, 'm'}, | |
3320 | {"Package", no_argument, 0, 'p'}, | |
3321 | {"processor", no_argument, 0, 'p'}, | |
3322 | {"Summary", no_argument, 0, 'S'}, | |
3323 | {"TCC", required_argument, 0, 'T'}, | |
3324 | {"version", no_argument, 0, 'v' }, | |
3325 | {0, 0, 0, 0 } | |
3326 | }; | |
103a8fea LB |
3327 | |
3328 | progname = argv[0]; | |
3329 | ||
a01e72fb | 3330 | while ((opt = getopt_long_only(argc, argv, "+C:c:Ddhi:JM:m:PpST:v", |
d8af6f5f | 3331 | long_options, &option_index)) != -1) { |
103a8fea | 3332 | switch (opt) { |
d8af6f5f LB |
3333 | case 'C': |
3334 | sscanf(optarg, "%x", &extra_delta_offset64); | |
c98d5d94 | 3335 | break; |
d8af6f5f LB |
3336 | case 'c': |
3337 | sscanf(optarg, "%x", &extra_delta_offset32); | |
c98d5d94 | 3338 | break; |
d8af6f5f | 3339 | case 'D': |
3b4d5c7f AS |
3340 | dump_only++; |
3341 | break; | |
d8af6f5f LB |
3342 | case 'd': |
3343 | debug++; | |
103a8fea | 3344 | break; |
d8af6f5f LB |
3345 | case 'h': |
3346 | default: | |
3347 | help(); | |
3348 | exit(1); | |
103a8fea LB |
3349 | case 'i': |
3350 | interval_sec = atoi(optarg); | |
3351 | break; | |
d8af6f5f LB |
3352 | case 'J': |
3353 | rapl_joules++; | |
8e180f3c | 3354 | break; |
d8af6f5f LB |
3355 | case 'M': |
3356 | sscanf(optarg, "%x", &extra_msr_offset64); | |
8e180f3c | 3357 | break; |
2f32edf1 LB |
3358 | case 'm': |
3359 | sscanf(optarg, "%x", &extra_msr_offset32); | |
2f32edf1 | 3360 | break; |
d8af6f5f LB |
3361 | case 'P': |
3362 | show_pkg_only++; | |
3363 | break; | |
3364 | case 'p': | |
3365 | show_core_only++; | |
103a8fea | 3366 | break; |
d8af6f5f LB |
3367 | case 'S': |
3368 | summary_only++; | |
889facbe LB |
3369 | break; |
3370 | case 'T': | |
3371 | tcc_activation_temp_override = atoi(optarg); | |
3372 | break; | |
d8af6f5f LB |
3373 | case 'v': |
3374 | print_version(); | |
3375 | exit(0); | |
5c56be9a | 3376 | break; |
103a8fea LB |
3377 | } |
3378 | } | |
3379 | } | |
3380 | ||
3381 | int main(int argc, char **argv) | |
3382 | { | |
3383 | cmdline(argc, argv); | |
3384 | ||
d8af6f5f LB |
3385 | if (debug) |
3386 | print_version(); | |
103a8fea LB |
3387 | |
3388 | turbostat_init(); | |
3389 | ||
3b4d5c7f AS |
3390 | /* dump counters and exit */ |
3391 | if (dump_only) | |
3392 | return get_and_dump_counters(); | |
3393 | ||
103a8fea LB |
3394 | /* |
3395 | * if any params left, it must be a command to fork | |
3396 | */ | |
3397 | if (argc - optind) | |
3398 | return fork_it(argv + optind); | |
3399 | else | |
3400 | turbostat_loop(); | |
3401 | ||
3402 | return 0; | |
3403 | } |