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; | |
1ed51011 | 74 | int do_smi; |
103a8fea LB |
75 | double bclk; |
76 | unsigned int show_pkg; | |
77 | unsigned int show_core; | |
78 | unsigned int show_cpu; | |
c98d5d94 LB |
79 | unsigned int show_pkg_only; |
80 | unsigned int show_core_only; | |
81 | char *output_buffer, *outp; | |
889facbe LB |
82 | unsigned int do_rapl; |
83 | unsigned int do_dts; | |
84 | unsigned int do_ptm; | |
85 | unsigned int tcc_activation_temp; | |
86 | unsigned int tcc_activation_temp_override; | |
40ee8e3b AS |
87 | double rapl_power_units, rapl_time_units; |
88 | double rapl_dram_energy_units, rapl_energy_units; | |
889facbe | 89 | double rapl_joule_counter_range; |
3a9a941d LB |
90 | unsigned int do_core_perf_limit_reasons; |
91 | unsigned int do_gfx_perf_limit_reasons; | |
92 | unsigned int do_ring_perf_limit_reasons; | |
8a5bdf41 LB |
93 | unsigned int crystal_hz; |
94 | unsigned long long tsc_hz; | |
7ce7d5de | 95 | int base_cpu; |
889facbe | 96 | |
e6f9bb3c LB |
97 | #define RAPL_PKG (1 << 0) |
98 | /* 0x610 MSR_PKG_POWER_LIMIT */ | |
99 | /* 0x611 MSR_PKG_ENERGY_STATUS */ | |
100 | #define RAPL_PKG_PERF_STATUS (1 << 1) | |
101 | /* 0x613 MSR_PKG_PERF_STATUS */ | |
102 | #define RAPL_PKG_POWER_INFO (1 << 2) | |
103 | /* 0x614 MSR_PKG_POWER_INFO */ | |
104 | ||
105 | #define RAPL_DRAM (1 << 3) | |
106 | /* 0x618 MSR_DRAM_POWER_LIMIT */ | |
107 | /* 0x619 MSR_DRAM_ENERGY_STATUS */ | |
e6f9bb3c LB |
108 | #define RAPL_DRAM_PERF_STATUS (1 << 4) |
109 | /* 0x61b MSR_DRAM_PERF_STATUS */ | |
0b2bb692 LB |
110 | #define RAPL_DRAM_POWER_INFO (1 << 5) |
111 | /* 0x61c MSR_DRAM_POWER_INFO */ | |
e6f9bb3c | 112 | |
0b2bb692 | 113 | #define RAPL_CORES (1 << 6) |
e6f9bb3c LB |
114 | /* 0x638 MSR_PP0_POWER_LIMIT */ |
115 | /* 0x639 MSR_PP0_ENERGY_STATUS */ | |
0b2bb692 | 116 | #define RAPL_CORE_POLICY (1 << 7) |
e6f9bb3c LB |
117 | /* 0x63a MSR_PP0_POLICY */ |
118 | ||
0b2bb692 | 119 | #define RAPL_GFX (1 << 8) |
e6f9bb3c LB |
120 | /* 0x640 MSR_PP1_POWER_LIMIT */ |
121 | /* 0x641 MSR_PP1_ENERGY_STATUS */ | |
122 | /* 0x642 MSR_PP1_POLICY */ | |
889facbe LB |
123 | #define TJMAX_DEFAULT 100 |
124 | ||
125 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
103a8fea LB |
126 | |
127 | int aperf_mperf_unstable; | |
128 | int backwards_count; | |
129 | char *progname; | |
103a8fea | 130 | |
c98d5d94 LB |
131 | cpu_set_t *cpu_present_set, *cpu_affinity_set; |
132 | size_t cpu_present_setsize, cpu_affinity_setsize; | |
133 | ||
134 | struct thread_data { | |
135 | unsigned long long tsc; | |
136 | unsigned long long aperf; | |
137 | unsigned long long mperf; | |
144b44b1 | 138 | unsigned long long c1; |
2f32edf1 | 139 | unsigned long long extra_msr64; |
8e180f3c LB |
140 | unsigned long long extra_delta64; |
141 | unsigned long long extra_msr32; | |
142 | unsigned long long extra_delta32; | |
1ed51011 | 143 | unsigned int smi_count; |
c98d5d94 LB |
144 | unsigned int cpu_id; |
145 | unsigned int flags; | |
146 | #define CPU_IS_FIRST_THREAD_IN_CORE 0x2 | |
147 | #define CPU_IS_FIRST_CORE_IN_PACKAGE 0x4 | |
148 | } *thread_even, *thread_odd; | |
149 | ||
150 | struct core_data { | |
151 | unsigned long long c3; | |
152 | unsigned long long c6; | |
153 | unsigned long long c7; | |
889facbe | 154 | unsigned int core_temp_c; |
c98d5d94 LB |
155 | unsigned int core_id; |
156 | } *core_even, *core_odd; | |
157 | ||
158 | struct pkg_data { | |
159 | unsigned long long pc2; | |
160 | unsigned long long pc3; | |
161 | unsigned long long pc6; | |
162 | unsigned long long pc7; | |
ca58710f KCA |
163 | unsigned long long pc8; |
164 | unsigned long long pc9; | |
165 | unsigned long long pc10; | |
0b2bb692 LB |
166 | unsigned long long pkg_wtd_core_c0; |
167 | unsigned long long pkg_any_core_c0; | |
168 | unsigned long long pkg_any_gfxe_c0; | |
169 | unsigned long long pkg_both_core_gfxe_c0; | |
c98d5d94 | 170 | unsigned int package_id; |
889facbe LB |
171 | unsigned int energy_pkg; /* MSR_PKG_ENERGY_STATUS */ |
172 | unsigned int energy_dram; /* MSR_DRAM_ENERGY_STATUS */ | |
173 | unsigned int energy_cores; /* MSR_PP0_ENERGY_STATUS */ | |
174 | unsigned int energy_gfx; /* MSR_PP1_ENERGY_STATUS */ | |
175 | unsigned int rapl_pkg_perf_status; /* MSR_PKG_PERF_STATUS */ | |
176 | unsigned int rapl_dram_perf_status; /* MSR_DRAM_PERF_STATUS */ | |
177 | unsigned int pkg_temp_c; | |
178 | ||
c98d5d94 LB |
179 | } *package_even, *package_odd; |
180 | ||
181 | #define ODD_COUNTERS thread_odd, core_odd, package_odd | |
182 | #define EVEN_COUNTERS thread_even, core_even, package_even | |
183 | ||
184 | #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \ | |
185 | (thread_base + (pkg_no) * topo.num_cores_per_pkg * \ | |
186 | topo.num_threads_per_core + \ | |
187 | (core_no) * topo.num_threads_per_core + (thread_no)) | |
188 | #define GET_CORE(core_base, core_no, pkg_no) \ | |
189 | (core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no)) | |
190 | #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no) | |
191 | ||
192 | struct system_summary { | |
193 | struct thread_data threads; | |
194 | struct core_data cores; | |
195 | struct pkg_data packages; | |
196 | } sum, average; | |
197 | ||
198 | ||
199 | struct topo_params { | |
200 | int num_packages; | |
201 | int num_cpus; | |
202 | int num_cores; | |
203 | int max_cpu_num; | |
204 | int num_cores_per_pkg; | |
205 | int num_threads_per_core; | |
206 | } topo; | |
207 | ||
208 | struct timeval tv_even, tv_odd, tv_delta; | |
209 | ||
210 | void setup_all_buffers(void); | |
211 | ||
212 | int cpu_is_not_present(int cpu) | |
d15cf7c1 | 213 | { |
c98d5d94 | 214 | return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set); |
d15cf7c1 | 215 | } |
88c3281f | 216 | /* |
c98d5d94 LB |
217 | * run func(thread, core, package) in topology order |
218 | * skip non-present cpus | |
88c3281f | 219 | */ |
c98d5d94 LB |
220 | |
221 | int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *), | |
222 | struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base) | |
88c3281f | 223 | { |
c98d5d94 | 224 | int retval, pkg_no, core_no, thread_no; |
d15cf7c1 | 225 | |
c98d5d94 LB |
226 | for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) { |
227 | for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) { | |
228 | for (thread_no = 0; thread_no < | |
229 | topo.num_threads_per_core; ++thread_no) { | |
230 | struct thread_data *t; | |
231 | struct core_data *c; | |
232 | struct pkg_data *p; | |
88c3281f | 233 | |
c98d5d94 LB |
234 | t = GET_THREAD(thread_base, thread_no, core_no, pkg_no); |
235 | ||
236 | if (cpu_is_not_present(t->cpu_id)) | |
237 | continue; | |
238 | ||
239 | c = GET_CORE(core_base, core_no, pkg_no); | |
240 | p = GET_PKG(pkg_base, pkg_no); | |
241 | ||
242 | retval = func(t, c, p); | |
243 | if (retval) | |
244 | return retval; | |
245 | } | |
246 | } | |
247 | } | |
248 | return 0; | |
88c3281f LB |
249 | } |
250 | ||
251 | int cpu_migrate(int cpu) | |
252 | { | |
c98d5d94 LB |
253 | CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set); |
254 | CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set); | |
255 | if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1) | |
88c3281f LB |
256 | return -1; |
257 | else | |
258 | return 0; | |
259 | } | |
260 | ||
15aaa346 | 261 | int get_msr(int cpu, off_t offset, unsigned long long *msr) |
103a8fea LB |
262 | { |
263 | ssize_t retval; | |
103a8fea LB |
264 | char pathname[32]; |
265 | int fd; | |
266 | ||
267 | sprintf(pathname, "/dev/cpu/%d/msr", cpu); | |
268 | fd = open(pathname, O_RDONLY); | |
15aaa346 | 269 | if (fd < 0) |
98481e79 | 270 | err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, or run as root", pathname); |
103a8fea | 271 | |
15aaa346 | 272 | retval = pread(fd, msr, sizeof *msr, offset); |
103a8fea | 273 | close(fd); |
15aaa346 | 274 | |
98481e79 LB |
275 | if (retval != sizeof *msr) |
276 | err(-1, "%s offset 0x%llx read failed", pathname, (unsigned long long)offset); | |
15aaa346 LB |
277 | |
278 | return 0; | |
103a8fea LB |
279 | } |
280 | ||
fc04cc67 LB |
281 | /* |
282 | * Example Format w/ field column widths: | |
283 | * | |
e7c95ff3 LB |
284 | * 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 |
285 | * 123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678 | |
fc04cc67 LB |
286 | */ |
287 | ||
a829eb4d | 288 | void print_header(void) |
103a8fea LB |
289 | { |
290 | if (show_pkg) | |
e7c95ff3 | 291 | outp += sprintf(outp, " Package"); |
103a8fea | 292 | if (show_core) |
e7c95ff3 | 293 | outp += sprintf(outp, " Core"); |
103a8fea | 294 | if (show_cpu) |
e7c95ff3 | 295 | outp += sprintf(outp, " CPU"); |
fc04cc67 | 296 | if (has_aperf) |
e7c95ff3 | 297 | outp += sprintf(outp, " Avg_MHz"); |
d7899447 | 298 | if (has_aperf) |
e7c95ff3 | 299 | outp += sprintf(outp, " %%Busy"); |
103a8fea | 300 | if (has_aperf) |
e7c95ff3 LB |
301 | outp += sprintf(outp, " Bzy_MHz"); |
302 | outp += sprintf(outp, " TSC_MHz"); | |
1cc21f7b | 303 | |
8e180f3c | 304 | if (extra_delta_offset32) |
e7c95ff3 | 305 | outp += sprintf(outp, " count 0x%03X", extra_delta_offset32); |
8e180f3c | 306 | if (extra_delta_offset64) |
e7c95ff3 | 307 | outp += sprintf(outp, " COUNT 0x%03X", extra_delta_offset64); |
2f32edf1 | 308 | if (extra_msr_offset32) |
e7c95ff3 | 309 | outp += sprintf(outp, " MSR 0x%03X", extra_msr_offset32); |
2f32edf1 | 310 | if (extra_msr_offset64) |
e7c95ff3 | 311 | outp += sprintf(outp, " MSR 0x%03X", extra_msr_offset64); |
1cc21f7b LB |
312 | |
313 | if (!debug) | |
314 | goto done; | |
315 | ||
316 | if (do_smi) | |
317 | outp += sprintf(outp, " SMI"); | |
318 | ||
103a8fea | 319 | if (do_nhm_cstates) |
e7c95ff3 | 320 | outp += sprintf(outp, " CPU%%c1"); |
fb5d4327 | 321 | if (do_nhm_cstates && !do_slm_cstates && !do_knl_cstates) |
e7c95ff3 | 322 | outp += sprintf(outp, " CPU%%c3"); |
103a8fea | 323 | if (do_nhm_cstates) |
e7c95ff3 | 324 | outp += sprintf(outp, " CPU%%c6"); |
103a8fea | 325 | if (do_snb_cstates) |
e7c95ff3 | 326 | outp += sprintf(outp, " CPU%%c7"); |
889facbe LB |
327 | |
328 | if (do_dts) | |
e7c95ff3 | 329 | outp += sprintf(outp, " CoreTmp"); |
889facbe | 330 | if (do_ptm) |
e7c95ff3 | 331 | outp += sprintf(outp, " PkgTmp"); |
889facbe | 332 | |
0b2bb692 LB |
333 | if (do_skl_residency) { |
334 | outp += sprintf(outp, " Totl%%C0"); | |
335 | outp += sprintf(outp, " Any%%C0"); | |
336 | outp += sprintf(outp, " GFX%%C0"); | |
337 | outp += sprintf(outp, " CPUGFX%%"); | |
338 | } | |
339 | ||
ee7e38e3 | 340 | if (do_pc2) |
e7c95ff3 | 341 | outp += sprintf(outp, " Pkg%%pc2"); |
ee7e38e3 | 342 | if (do_pc3) |
e7c95ff3 | 343 | outp += sprintf(outp, " Pkg%%pc3"); |
ee7e38e3 | 344 | if (do_pc6) |
e7c95ff3 | 345 | outp += sprintf(outp, " Pkg%%pc6"); |
ee7e38e3 | 346 | if (do_pc7) |
e7c95ff3 | 347 | outp += sprintf(outp, " Pkg%%pc7"); |
ca58710f | 348 | if (do_c8_c9_c10) { |
e7c95ff3 LB |
349 | outp += sprintf(outp, " Pkg%%pc8"); |
350 | outp += sprintf(outp, " Pkg%%pc9"); | |
351 | outp += sprintf(outp, " Pk%%pc10"); | |
ca58710f | 352 | } |
103a8fea | 353 | |
5c56be9a DB |
354 | if (do_rapl && !rapl_joules) { |
355 | if (do_rapl & RAPL_PKG) | |
e7c95ff3 | 356 | outp += sprintf(outp, " PkgWatt"); |
5c56be9a | 357 | if (do_rapl & RAPL_CORES) |
e7c95ff3 | 358 | outp += sprintf(outp, " CorWatt"); |
5c56be9a | 359 | if (do_rapl & RAPL_GFX) |
e7c95ff3 | 360 | outp += sprintf(outp, " GFXWatt"); |
5c56be9a | 361 | if (do_rapl & RAPL_DRAM) |
e7c95ff3 | 362 | outp += sprintf(outp, " RAMWatt"); |
5c56be9a | 363 | if (do_rapl & RAPL_PKG_PERF_STATUS) |
e7c95ff3 | 364 | outp += sprintf(outp, " PKG_%%"); |
5c56be9a | 365 | if (do_rapl & RAPL_DRAM_PERF_STATUS) |
e7c95ff3 | 366 | outp += sprintf(outp, " RAM_%%"); |
d7899447 | 367 | } else if (do_rapl && rapl_joules) { |
5c56be9a | 368 | if (do_rapl & RAPL_PKG) |
e7c95ff3 | 369 | outp += sprintf(outp, " Pkg_J"); |
5c56be9a | 370 | if (do_rapl & RAPL_CORES) |
e7c95ff3 | 371 | outp += sprintf(outp, " Cor_J"); |
5c56be9a | 372 | if (do_rapl & RAPL_GFX) |
e7c95ff3 | 373 | outp += sprintf(outp, " GFX_J"); |
5c56be9a | 374 | if (do_rapl & RAPL_DRAM) |
e7c95ff3 | 375 | outp += sprintf(outp, " RAM_W"); |
5c56be9a | 376 | if (do_rapl & RAPL_PKG_PERF_STATUS) |
e7c95ff3 | 377 | outp += sprintf(outp, " PKG_%%"); |
5c56be9a | 378 | if (do_rapl & RAPL_DRAM_PERF_STATUS) |
e7c95ff3 LB |
379 | outp += sprintf(outp, " RAM_%%"); |
380 | outp += sprintf(outp, " time"); | |
889facbe | 381 | |
5c56be9a | 382 | } |
1cc21f7b | 383 | done: |
c98d5d94 | 384 | outp += sprintf(outp, "\n"); |
103a8fea LB |
385 | } |
386 | ||
c98d5d94 LB |
387 | int dump_counters(struct thread_data *t, struct core_data *c, |
388 | struct pkg_data *p) | |
103a8fea | 389 | { |
3b4d5c7f | 390 | outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p); |
c98d5d94 LB |
391 | |
392 | if (t) { | |
3b4d5c7f AS |
393 | outp += sprintf(outp, "CPU: %d flags 0x%x\n", |
394 | t->cpu_id, t->flags); | |
395 | outp += sprintf(outp, "TSC: %016llX\n", t->tsc); | |
396 | outp += sprintf(outp, "aperf: %016llX\n", t->aperf); | |
397 | outp += sprintf(outp, "mperf: %016llX\n", t->mperf); | |
398 | outp += sprintf(outp, "c1: %016llX\n", t->c1); | |
399 | outp += sprintf(outp, "msr0x%x: %08llX\n", | |
8e180f3c | 400 | extra_delta_offset32, t->extra_delta32); |
3b4d5c7f | 401 | outp += sprintf(outp, "msr0x%x: %016llX\n", |
8e180f3c | 402 | extra_delta_offset64, t->extra_delta64); |
3b4d5c7f | 403 | outp += sprintf(outp, "msr0x%x: %08llX\n", |
2f32edf1 | 404 | extra_msr_offset32, t->extra_msr32); |
3b4d5c7f | 405 | outp += sprintf(outp, "msr0x%x: %016llX\n", |
2f32edf1 | 406 | extra_msr_offset64, t->extra_msr64); |
1ed51011 | 407 | if (do_smi) |
3b4d5c7f | 408 | outp += sprintf(outp, "SMI: %08X\n", t->smi_count); |
c98d5d94 | 409 | } |
103a8fea | 410 | |
c98d5d94 | 411 | if (c) { |
3b4d5c7f AS |
412 | outp += sprintf(outp, "core: %d\n", c->core_id); |
413 | outp += sprintf(outp, "c3: %016llX\n", c->c3); | |
414 | outp += sprintf(outp, "c6: %016llX\n", c->c6); | |
415 | outp += sprintf(outp, "c7: %016llX\n", c->c7); | |
416 | outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c); | |
c98d5d94 | 417 | } |
103a8fea | 418 | |
c98d5d94 | 419 | if (p) { |
3b4d5c7f | 420 | outp += sprintf(outp, "package: %d\n", p->package_id); |
0b2bb692 LB |
421 | |
422 | outp += sprintf(outp, "Weighted cores: %016llX\n", p->pkg_wtd_core_c0); | |
423 | outp += sprintf(outp, "Any cores: %016llX\n", p->pkg_any_core_c0); | |
424 | outp += sprintf(outp, "Any GFX: %016llX\n", p->pkg_any_gfxe_c0); | |
425 | outp += sprintf(outp, "CPU + GFX: %016llX\n", p->pkg_both_core_gfxe_c0); | |
426 | ||
3b4d5c7f | 427 | outp += sprintf(outp, "pc2: %016llX\n", p->pc2); |
ee7e38e3 LB |
428 | if (do_pc3) |
429 | outp += sprintf(outp, "pc3: %016llX\n", p->pc3); | |
430 | if (do_pc6) | |
431 | outp += sprintf(outp, "pc6: %016llX\n", p->pc6); | |
432 | if (do_pc7) | |
433 | outp += sprintf(outp, "pc7: %016llX\n", p->pc7); | |
3b4d5c7f AS |
434 | outp += sprintf(outp, "pc8: %016llX\n", p->pc8); |
435 | outp += sprintf(outp, "pc9: %016llX\n", p->pc9); | |
436 | outp += sprintf(outp, "pc10: %016llX\n", p->pc10); | |
437 | outp += sprintf(outp, "Joules PKG: %0X\n", p->energy_pkg); | |
438 | outp += sprintf(outp, "Joules COR: %0X\n", p->energy_cores); | |
439 | outp += sprintf(outp, "Joules GFX: %0X\n", p->energy_gfx); | |
440 | outp += sprintf(outp, "Joules RAM: %0X\n", p->energy_dram); | |
441 | outp += sprintf(outp, "Throttle PKG: %0X\n", | |
442 | p->rapl_pkg_perf_status); | |
443 | outp += sprintf(outp, "Throttle RAM: %0X\n", | |
444 | p->rapl_dram_perf_status); | |
445 | outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c); | |
c98d5d94 | 446 | } |
3b4d5c7f AS |
447 | |
448 | outp += sprintf(outp, "\n"); | |
449 | ||
c98d5d94 | 450 | return 0; |
103a8fea LB |
451 | } |
452 | ||
e23da037 LB |
453 | /* |
454 | * column formatting convention & formats | |
e23da037 | 455 | */ |
c98d5d94 LB |
456 | int format_counters(struct thread_data *t, struct core_data *c, |
457 | struct pkg_data *p) | |
103a8fea LB |
458 | { |
459 | double interval_float; | |
fc04cc67 | 460 | char *fmt8; |
103a8fea | 461 | |
c98d5d94 LB |
462 | /* if showing only 1st thread in core and this isn't one, bail out */ |
463 | if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
464 | return 0; | |
465 | ||
466 | /* if showing only 1st thread in pkg and this isn't one, bail out */ | |
467 | if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
468 | return 0; | |
469 | ||
103a8fea LB |
470 | interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0; |
471 | ||
c98d5d94 LB |
472 | /* topo columns, print blanks on 1st (average) line */ |
473 | if (t == &average.threads) { | |
103a8fea | 474 | if (show_pkg) |
fc04cc67 | 475 | outp += sprintf(outp, " -"); |
103a8fea | 476 | if (show_core) |
fc04cc67 | 477 | outp += sprintf(outp, " -"); |
103a8fea | 478 | if (show_cpu) |
fc04cc67 | 479 | outp += sprintf(outp, " -"); |
103a8fea | 480 | } else { |
c98d5d94 LB |
481 | if (show_pkg) { |
482 | if (p) | |
fc04cc67 | 483 | outp += sprintf(outp, "%8d", p->package_id); |
c98d5d94 | 484 | else |
fc04cc67 | 485 | outp += sprintf(outp, " -"); |
c98d5d94 | 486 | } |
c98d5d94 LB |
487 | if (show_core) { |
488 | if (c) | |
fc04cc67 | 489 | outp += sprintf(outp, "%8d", c->core_id); |
c98d5d94 | 490 | else |
fc04cc67 | 491 | outp += sprintf(outp, " -"); |
c98d5d94 | 492 | } |
103a8fea | 493 | if (show_cpu) |
fc04cc67 | 494 | outp += sprintf(outp, "%8d", t->cpu_id); |
103a8fea | 495 | } |
fc04cc67 | 496 | |
d7899447 | 497 | /* Avg_MHz */ |
fc04cc67 LB |
498 | if (has_aperf) |
499 | outp += sprintf(outp, "%8.0f", | |
500 | 1.0 / units * t->aperf / interval_float); | |
501 | ||
d7899447 LB |
502 | /* %Busy */ |
503 | if (has_aperf) { | |
103a8fea | 504 | if (!skip_c0) |
fc04cc67 | 505 | outp += sprintf(outp, "%8.2f", 100.0 * t->mperf/t->tsc); |
103a8fea | 506 | else |
fc04cc67 | 507 | outp += sprintf(outp, "********"); |
103a8fea LB |
508 | } |
509 | ||
d7899447 | 510 | /* Bzy_MHz */ |
fc04cc67 LB |
511 | if (has_aperf) |
512 | outp += sprintf(outp, "%8.0f", | |
513 | 1.0 * t->tsc / units * t->aperf / t->mperf / interval_float); | |
103a8fea | 514 | |
d7899447 | 515 | /* TSC_MHz */ |
fc04cc67 | 516 | outp += sprintf(outp, "%8.0f", 1.0 * t->tsc/units/interval_float); |
103a8fea | 517 | |
8e180f3c LB |
518 | /* delta */ |
519 | if (extra_delta_offset32) | |
520 | outp += sprintf(outp, " %11llu", t->extra_delta32); | |
521 | ||
522 | /* DELTA */ | |
523 | if (extra_delta_offset64) | |
524 | outp += sprintf(outp, " %11llu", t->extra_delta64); | |
2f32edf1 LB |
525 | /* msr */ |
526 | if (extra_msr_offset32) | |
8e180f3c | 527 | outp += sprintf(outp, " 0x%08llx", t->extra_msr32); |
2f32edf1 | 528 | |
130ff304 | 529 | /* MSR */ |
2f32edf1 LB |
530 | if (extra_msr_offset64) |
531 | outp += sprintf(outp, " 0x%016llx", t->extra_msr64); | |
130ff304 | 532 | |
1cc21f7b LB |
533 | if (!debug) |
534 | goto done; | |
535 | ||
536 | /* SMI */ | |
537 | if (do_smi) | |
538 | outp += sprintf(outp, "%8d", t->smi_count); | |
539 | ||
103a8fea LB |
540 | if (do_nhm_cstates) { |
541 | if (!skip_c1) | |
fc04cc67 | 542 | outp += sprintf(outp, "%8.2f", 100.0 * t->c1/t->tsc); |
103a8fea | 543 | else |
fc04cc67 | 544 | outp += sprintf(outp, "********"); |
103a8fea | 545 | } |
c98d5d94 LB |
546 | |
547 | /* print per-core data only for 1st thread in core */ | |
548 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
549 | goto done; | |
550 | ||
fb5d4327 | 551 | if (do_nhm_cstates && !do_slm_cstates && !do_knl_cstates) |
fc04cc67 | 552 | outp += sprintf(outp, "%8.2f", 100.0 * c->c3/t->tsc); |
103a8fea | 553 | if (do_nhm_cstates) |
fc04cc67 | 554 | outp += sprintf(outp, "%8.2f", 100.0 * c->c6/t->tsc); |
103a8fea | 555 | if (do_snb_cstates) |
fc04cc67 | 556 | outp += sprintf(outp, "%8.2f", 100.0 * c->c7/t->tsc); |
c98d5d94 | 557 | |
889facbe | 558 | if (do_dts) |
fc04cc67 | 559 | outp += sprintf(outp, "%8d", c->core_temp_c); |
889facbe | 560 | |
c98d5d94 LB |
561 | /* print per-package data only for 1st core in package */ |
562 | if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
563 | goto done; | |
564 | ||
0b2bb692 | 565 | /* PkgTmp */ |
889facbe | 566 | if (do_ptm) |
fc04cc67 | 567 | outp += sprintf(outp, "%8d", p->pkg_temp_c); |
889facbe | 568 | |
0b2bb692 LB |
569 | /* Totl%C0, Any%C0 GFX%C0 CPUGFX% */ |
570 | if (do_skl_residency) { | |
571 | outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_wtd_core_c0/t->tsc); | |
572 | outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_any_core_c0/t->tsc); | |
573 | outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_any_gfxe_c0/t->tsc); | |
574 | outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_both_core_gfxe_c0/t->tsc); | |
575 | } | |
576 | ||
ee7e38e3 | 577 | if (do_pc2) |
fc04cc67 | 578 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc2/t->tsc); |
ee7e38e3 | 579 | if (do_pc3) |
fc04cc67 | 580 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc3/t->tsc); |
ee7e38e3 | 581 | if (do_pc6) |
fc04cc67 | 582 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc6/t->tsc); |
ee7e38e3 | 583 | if (do_pc7) |
fc04cc67 | 584 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc7/t->tsc); |
ca58710f | 585 | if (do_c8_c9_c10) { |
fc04cc67 LB |
586 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc8/t->tsc); |
587 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc9/t->tsc); | |
588 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc10/t->tsc); | |
ca58710f | 589 | } |
889facbe LB |
590 | |
591 | /* | |
592 | * If measurement interval exceeds minimum RAPL Joule Counter range, | |
593 | * indicate that results are suspect by printing "**" in fraction place. | |
594 | */ | |
fc04cc67 LB |
595 | if (interval_float < rapl_joule_counter_range) |
596 | fmt8 = "%8.2f"; | |
597 | else | |
598 | fmt8 = " %6.0f**"; | |
889facbe | 599 | |
5c56be9a DB |
600 | if (do_rapl && !rapl_joules) { |
601 | if (do_rapl & RAPL_PKG) | |
fc04cc67 | 602 | outp += sprintf(outp, fmt8, p->energy_pkg * rapl_energy_units / interval_float); |
5c56be9a | 603 | if (do_rapl & RAPL_CORES) |
fc04cc67 | 604 | outp += sprintf(outp, fmt8, p->energy_cores * rapl_energy_units / interval_float); |
5c56be9a | 605 | if (do_rapl & RAPL_GFX) |
fc04cc67 | 606 | outp += sprintf(outp, fmt8, p->energy_gfx * rapl_energy_units / interval_float); |
5c56be9a | 607 | if (do_rapl & RAPL_DRAM) |
40ee8e3b | 608 | outp += sprintf(outp, fmt8, p->energy_dram * rapl_dram_energy_units / interval_float); |
5c56be9a | 609 | if (do_rapl & RAPL_PKG_PERF_STATUS) |
fc04cc67 | 610 | outp += sprintf(outp, fmt8, 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float); |
5c56be9a | 611 | if (do_rapl & RAPL_DRAM_PERF_STATUS) |
fc04cc67 | 612 | outp += sprintf(outp, fmt8, 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float); |
d7899447 | 613 | } else if (do_rapl && rapl_joules) { |
5c56be9a | 614 | if (do_rapl & RAPL_PKG) |
fc04cc67 | 615 | outp += sprintf(outp, fmt8, |
5c56be9a DB |
616 | p->energy_pkg * rapl_energy_units); |
617 | if (do_rapl & RAPL_CORES) | |
fc04cc67 | 618 | outp += sprintf(outp, fmt8, |
5c56be9a DB |
619 | p->energy_cores * rapl_energy_units); |
620 | if (do_rapl & RAPL_GFX) | |
fc04cc67 | 621 | outp += sprintf(outp, fmt8, |
5c56be9a DB |
622 | p->energy_gfx * rapl_energy_units); |
623 | if (do_rapl & RAPL_DRAM) | |
fc04cc67 | 624 | outp += sprintf(outp, fmt8, |
40ee8e3b | 625 | p->energy_dram * rapl_dram_energy_units); |
5c56be9a | 626 | if (do_rapl & RAPL_PKG_PERF_STATUS) |
fc04cc67 | 627 | outp += sprintf(outp, fmt8, 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float); |
5c56be9a | 628 | if (do_rapl & RAPL_DRAM_PERF_STATUS) |
fc04cc67 | 629 | outp += sprintf(outp, fmt8, 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float); |
889facbe | 630 | |
d7899447 | 631 | outp += sprintf(outp, fmt8, interval_float); |
5c56be9a | 632 | } |
c98d5d94 | 633 | done: |
c98d5d94 LB |
634 | outp += sprintf(outp, "\n"); |
635 | ||
636 | return 0; | |
103a8fea LB |
637 | } |
638 | ||
c98d5d94 LB |
639 | void flush_stdout() |
640 | { | |
641 | fputs(output_buffer, stdout); | |
ddac0d68 | 642 | fflush(stdout); |
c98d5d94 LB |
643 | outp = output_buffer; |
644 | } | |
645 | void flush_stderr() | |
646 | { | |
647 | fputs(output_buffer, stderr); | |
648 | outp = output_buffer; | |
649 | } | |
650 | void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
103a8fea | 651 | { |
e23da037 | 652 | static int printed; |
103a8fea | 653 | |
e23da037 LB |
654 | if (!printed || !summary_only) |
655 | print_header(); | |
103a8fea | 656 | |
c98d5d94 LB |
657 | if (topo.num_cpus > 1) |
658 | format_counters(&average.threads, &average.cores, | |
659 | &average.packages); | |
103a8fea | 660 | |
e23da037 LB |
661 | printed = 1; |
662 | ||
663 | if (summary_only) | |
664 | return; | |
665 | ||
c98d5d94 | 666 | for_all_cpus(format_counters, t, c, p); |
103a8fea LB |
667 | } |
668 | ||
889facbe LB |
669 | #define DELTA_WRAP32(new, old) \ |
670 | if (new > old) { \ | |
671 | old = new - old; \ | |
672 | } else { \ | |
673 | old = 0x100000000 + new - old; \ | |
674 | } | |
675 | ||
c98d5d94 LB |
676 | void |
677 | delta_package(struct pkg_data *new, struct pkg_data *old) | |
678 | { | |
0b2bb692 LB |
679 | |
680 | if (do_skl_residency) { | |
681 | old->pkg_wtd_core_c0 = new->pkg_wtd_core_c0 - old->pkg_wtd_core_c0; | |
682 | old->pkg_any_core_c0 = new->pkg_any_core_c0 - old->pkg_any_core_c0; | |
683 | old->pkg_any_gfxe_c0 = new->pkg_any_gfxe_c0 - old->pkg_any_gfxe_c0; | |
684 | old->pkg_both_core_gfxe_c0 = new->pkg_both_core_gfxe_c0 - old->pkg_both_core_gfxe_c0; | |
685 | } | |
c98d5d94 | 686 | old->pc2 = new->pc2 - old->pc2; |
ee7e38e3 LB |
687 | if (do_pc3) |
688 | old->pc3 = new->pc3 - old->pc3; | |
689 | if (do_pc6) | |
690 | old->pc6 = new->pc6 - old->pc6; | |
691 | if (do_pc7) | |
692 | old->pc7 = new->pc7 - old->pc7; | |
ca58710f KCA |
693 | old->pc8 = new->pc8 - old->pc8; |
694 | old->pc9 = new->pc9 - old->pc9; | |
695 | old->pc10 = new->pc10 - old->pc10; | |
889facbe LB |
696 | old->pkg_temp_c = new->pkg_temp_c; |
697 | ||
698 | DELTA_WRAP32(new->energy_pkg, old->energy_pkg); | |
699 | DELTA_WRAP32(new->energy_cores, old->energy_cores); | |
700 | DELTA_WRAP32(new->energy_gfx, old->energy_gfx); | |
701 | DELTA_WRAP32(new->energy_dram, old->energy_dram); | |
702 | DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status); | |
703 | DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status); | |
c98d5d94 | 704 | } |
103a8fea | 705 | |
c98d5d94 LB |
706 | void |
707 | delta_core(struct core_data *new, struct core_data *old) | |
103a8fea | 708 | { |
c98d5d94 LB |
709 | old->c3 = new->c3 - old->c3; |
710 | old->c6 = new->c6 - old->c6; | |
711 | old->c7 = new->c7 - old->c7; | |
889facbe | 712 | old->core_temp_c = new->core_temp_c; |
c98d5d94 | 713 | } |
103a8fea | 714 | |
c3ae331d LB |
715 | /* |
716 | * old = new - old | |
717 | */ | |
c98d5d94 LB |
718 | void |
719 | delta_thread(struct thread_data *new, struct thread_data *old, | |
720 | struct core_data *core_delta) | |
721 | { | |
722 | old->tsc = new->tsc - old->tsc; | |
723 | ||
724 | /* check for TSC < 1 Mcycles over interval */ | |
b2c95d90 JT |
725 | if (old->tsc < (1000 * 1000)) |
726 | errx(-3, "Insanely slow TSC rate, TSC stops in idle?\n" | |
727 | "You can disable all c-states by booting with \"idle=poll\"\n" | |
728 | "or just the deep ones with \"processor.max_cstate=1\""); | |
103a8fea | 729 | |
c98d5d94 | 730 | old->c1 = new->c1 - old->c1; |
103a8fea | 731 | |
a729617c LB |
732 | if (has_aperf) { |
733 | if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) { | |
734 | old->aperf = new->aperf - old->aperf; | |
735 | old->mperf = new->mperf - old->mperf; | |
736 | } else { | |
103a8fea | 737 | |
a729617c LB |
738 | if (!aperf_mperf_unstable) { |
739 | fprintf(stderr, "%s: APERF or MPERF went backwards *\n", progname); | |
740 | fprintf(stderr, "* Frequency results do not cover entire interval *\n"); | |
741 | fprintf(stderr, "* fix this by running Linux-2.6.30 or later *\n"); | |
103a8fea | 742 | |
a729617c LB |
743 | aperf_mperf_unstable = 1; |
744 | } | |
745 | /* | |
746 | * mperf delta is likely a huge "positive" number | |
747 | * can not use it for calculating c0 time | |
748 | */ | |
749 | skip_c0 = 1; | |
750 | skip_c1 = 1; | |
103a8fea | 751 | } |
c98d5d94 | 752 | } |
103a8fea | 753 | |
103a8fea | 754 | |
144b44b1 LB |
755 | if (use_c1_residency_msr) { |
756 | /* | |
757 | * Some models have a dedicated C1 residency MSR, | |
758 | * which should be more accurate than the derivation below. | |
759 | */ | |
760 | } else { | |
761 | /* | |
762 | * As counter collection is not atomic, | |
763 | * it is possible for mperf's non-halted cycles + idle states | |
764 | * to exceed TSC's all cycles: show c1 = 0% in that case. | |
765 | */ | |
766 | if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > old->tsc) | |
767 | old->c1 = 0; | |
768 | else { | |
769 | /* normal case, derive c1 */ | |
770 | old->c1 = old->tsc - old->mperf - core_delta->c3 | |
c98d5d94 | 771 | - core_delta->c6 - core_delta->c7; |
144b44b1 | 772 | } |
c98d5d94 | 773 | } |
c3ae331d | 774 | |
c98d5d94 | 775 | if (old->mperf == 0) { |
d8af6f5f | 776 | if (debug > 1) fprintf(stderr, "cpu%d MPERF 0!\n", old->cpu_id); |
c98d5d94 | 777 | old->mperf = 1; /* divide by 0 protection */ |
103a8fea | 778 | } |
c98d5d94 | 779 | |
8e180f3c LB |
780 | old->extra_delta32 = new->extra_delta32 - old->extra_delta32; |
781 | old->extra_delta32 &= 0xFFFFFFFF; | |
782 | ||
783 | old->extra_delta64 = new->extra_delta64 - old->extra_delta64; | |
784 | ||
c98d5d94 | 785 | /* |
8e180f3c | 786 | * Extra MSR is just a snapshot, simply copy latest w/o subtracting |
c98d5d94 | 787 | */ |
2f32edf1 LB |
788 | old->extra_msr32 = new->extra_msr32; |
789 | old->extra_msr64 = new->extra_msr64; | |
1ed51011 LB |
790 | |
791 | if (do_smi) | |
792 | old->smi_count = new->smi_count - old->smi_count; | |
c98d5d94 LB |
793 | } |
794 | ||
795 | int delta_cpu(struct thread_data *t, struct core_data *c, | |
796 | struct pkg_data *p, struct thread_data *t2, | |
797 | struct core_data *c2, struct pkg_data *p2) | |
798 | { | |
799 | /* calculate core delta only for 1st thread in core */ | |
800 | if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE) | |
801 | delta_core(c, c2); | |
802 | ||
803 | /* always calculate thread delta */ | |
804 | delta_thread(t, t2, c2); /* c2 is core delta */ | |
805 | ||
806 | /* calculate package delta only for 1st core in package */ | |
807 | if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE) | |
808 | delta_package(p, p2); | |
809 | ||
103a8fea LB |
810 | return 0; |
811 | } | |
812 | ||
c98d5d94 LB |
813 | void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) |
814 | { | |
815 | t->tsc = 0; | |
816 | t->aperf = 0; | |
817 | t->mperf = 0; | |
818 | t->c1 = 0; | |
819 | ||
1ed51011 | 820 | t->smi_count = 0; |
8e180f3c LB |
821 | t->extra_delta32 = 0; |
822 | t->extra_delta64 = 0; | |
823 | ||
c98d5d94 LB |
824 | /* tells format_counters to dump all fields from this set */ |
825 | t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE; | |
826 | ||
827 | c->c3 = 0; | |
828 | c->c6 = 0; | |
829 | c->c7 = 0; | |
889facbe | 830 | c->core_temp_c = 0; |
c98d5d94 | 831 | |
0b2bb692 LB |
832 | p->pkg_wtd_core_c0 = 0; |
833 | p->pkg_any_core_c0 = 0; | |
834 | p->pkg_any_gfxe_c0 = 0; | |
835 | p->pkg_both_core_gfxe_c0 = 0; | |
836 | ||
c98d5d94 | 837 | p->pc2 = 0; |
ee7e38e3 LB |
838 | if (do_pc3) |
839 | p->pc3 = 0; | |
840 | if (do_pc6) | |
841 | p->pc6 = 0; | |
842 | if (do_pc7) | |
843 | p->pc7 = 0; | |
ca58710f KCA |
844 | p->pc8 = 0; |
845 | p->pc9 = 0; | |
846 | p->pc10 = 0; | |
889facbe LB |
847 | |
848 | p->energy_pkg = 0; | |
849 | p->energy_dram = 0; | |
850 | p->energy_cores = 0; | |
851 | p->energy_gfx = 0; | |
852 | p->rapl_pkg_perf_status = 0; | |
853 | p->rapl_dram_perf_status = 0; | |
854 | p->pkg_temp_c = 0; | |
c98d5d94 LB |
855 | } |
856 | int sum_counters(struct thread_data *t, struct core_data *c, | |
857 | struct pkg_data *p) | |
103a8fea | 858 | { |
c98d5d94 LB |
859 | average.threads.tsc += t->tsc; |
860 | average.threads.aperf += t->aperf; | |
861 | average.threads.mperf += t->mperf; | |
862 | average.threads.c1 += t->c1; | |
103a8fea | 863 | |
8e180f3c LB |
864 | average.threads.extra_delta32 += t->extra_delta32; |
865 | average.threads.extra_delta64 += t->extra_delta64; | |
866 | ||
c98d5d94 LB |
867 | /* sum per-core values only for 1st thread in core */ |
868 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
869 | return 0; | |
103a8fea | 870 | |
c98d5d94 LB |
871 | average.cores.c3 += c->c3; |
872 | average.cores.c6 += c->c6; | |
873 | average.cores.c7 += c->c7; | |
874 | ||
889facbe LB |
875 | average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c); |
876 | ||
c98d5d94 LB |
877 | /* sum per-pkg values only for 1st core in pkg */ |
878 | if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
879 | return 0; | |
880 | ||
0b2bb692 LB |
881 | if (do_skl_residency) { |
882 | average.packages.pkg_wtd_core_c0 += p->pkg_wtd_core_c0; | |
883 | average.packages.pkg_any_core_c0 += p->pkg_any_core_c0; | |
884 | average.packages.pkg_any_gfxe_c0 += p->pkg_any_gfxe_c0; | |
885 | average.packages.pkg_both_core_gfxe_c0 += p->pkg_both_core_gfxe_c0; | |
886 | } | |
887 | ||
c98d5d94 | 888 | average.packages.pc2 += p->pc2; |
ee7e38e3 LB |
889 | if (do_pc3) |
890 | average.packages.pc3 += p->pc3; | |
891 | if (do_pc6) | |
892 | average.packages.pc6 += p->pc6; | |
893 | if (do_pc7) | |
894 | average.packages.pc7 += p->pc7; | |
ca58710f KCA |
895 | average.packages.pc8 += p->pc8; |
896 | average.packages.pc9 += p->pc9; | |
897 | average.packages.pc10 += p->pc10; | |
c98d5d94 | 898 | |
889facbe LB |
899 | average.packages.energy_pkg += p->energy_pkg; |
900 | average.packages.energy_dram += p->energy_dram; | |
901 | average.packages.energy_cores += p->energy_cores; | |
902 | average.packages.energy_gfx += p->energy_gfx; | |
903 | ||
904 | average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c); | |
905 | ||
906 | average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status; | |
907 | average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status; | |
c98d5d94 LB |
908 | return 0; |
909 | } | |
910 | /* | |
911 | * sum the counters for all cpus in the system | |
912 | * compute the weighted average | |
913 | */ | |
914 | void compute_average(struct thread_data *t, struct core_data *c, | |
915 | struct pkg_data *p) | |
916 | { | |
917 | clear_counters(&average.threads, &average.cores, &average.packages); | |
918 | ||
919 | for_all_cpus(sum_counters, t, c, p); | |
920 | ||
921 | average.threads.tsc /= topo.num_cpus; | |
922 | average.threads.aperf /= topo.num_cpus; | |
923 | average.threads.mperf /= topo.num_cpus; | |
924 | average.threads.c1 /= topo.num_cpus; | |
925 | ||
8e180f3c LB |
926 | average.threads.extra_delta32 /= topo.num_cpus; |
927 | average.threads.extra_delta32 &= 0xFFFFFFFF; | |
928 | ||
929 | average.threads.extra_delta64 /= topo.num_cpus; | |
930 | ||
c98d5d94 LB |
931 | average.cores.c3 /= topo.num_cores; |
932 | average.cores.c6 /= topo.num_cores; | |
933 | average.cores.c7 /= topo.num_cores; | |
934 | ||
0b2bb692 LB |
935 | if (do_skl_residency) { |
936 | average.packages.pkg_wtd_core_c0 /= topo.num_packages; | |
937 | average.packages.pkg_any_core_c0 /= topo.num_packages; | |
938 | average.packages.pkg_any_gfxe_c0 /= topo.num_packages; | |
939 | average.packages.pkg_both_core_gfxe_c0 /= topo.num_packages; | |
940 | } | |
941 | ||
c98d5d94 | 942 | average.packages.pc2 /= topo.num_packages; |
ee7e38e3 LB |
943 | if (do_pc3) |
944 | average.packages.pc3 /= topo.num_packages; | |
945 | if (do_pc6) | |
946 | average.packages.pc6 /= topo.num_packages; | |
947 | if (do_pc7) | |
948 | average.packages.pc7 /= topo.num_packages; | |
ca58710f KCA |
949 | |
950 | average.packages.pc8 /= topo.num_packages; | |
951 | average.packages.pc9 /= topo.num_packages; | |
952 | average.packages.pc10 /= topo.num_packages; | |
103a8fea LB |
953 | } |
954 | ||
c98d5d94 | 955 | static unsigned long long rdtsc(void) |
103a8fea | 956 | { |
c98d5d94 | 957 | unsigned int low, high; |
15aaa346 | 958 | |
c98d5d94 | 959 | asm volatile("rdtsc" : "=a" (low), "=d" (high)); |
15aaa346 | 960 | |
c98d5d94 LB |
961 | return low | ((unsigned long long)high) << 32; |
962 | } | |
15aaa346 | 963 | |
15aaa346 | 964 | |
c98d5d94 LB |
965 | /* |
966 | * get_counters(...) | |
967 | * migrate to cpu | |
968 | * acquire and record local counters for that cpu | |
969 | */ | |
970 | int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
971 | { | |
972 | int cpu = t->cpu_id; | |
889facbe | 973 | unsigned long long msr; |
88c3281f | 974 | |
e52966c0 LB |
975 | if (cpu_migrate(cpu)) { |
976 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
c98d5d94 | 977 | return -1; |
e52966c0 | 978 | } |
15aaa346 | 979 | |
c98d5d94 LB |
980 | t->tsc = rdtsc(); /* we are running on local CPU of interest */ |
981 | ||
982 | if (has_aperf) { | |
9c63a650 | 983 | if (get_msr(cpu, MSR_IA32_APERF, &t->aperf)) |
c98d5d94 | 984 | return -3; |
9c63a650 | 985 | if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf)) |
c98d5d94 LB |
986 | return -4; |
987 | } | |
988 | ||
1ed51011 LB |
989 | if (do_smi) { |
990 | if (get_msr(cpu, MSR_SMI_COUNT, &msr)) | |
991 | return -5; | |
992 | t->smi_count = msr & 0xFFFFFFFF; | |
993 | } | |
8e180f3c | 994 | if (extra_delta_offset32) { |
889facbe | 995 | if (get_msr(cpu, extra_delta_offset32, &msr)) |
8e180f3c | 996 | return -5; |
889facbe | 997 | t->extra_delta32 = msr & 0xFFFFFFFF; |
8e180f3c LB |
998 | } |
999 | ||
1000 | if (extra_delta_offset64) | |
1001 | if (get_msr(cpu, extra_delta_offset64, &t->extra_delta64)) | |
2f32edf1 LB |
1002 | return -5; |
1003 | ||
8e180f3c | 1004 | if (extra_msr_offset32) { |
889facbe | 1005 | if (get_msr(cpu, extra_msr_offset32, &msr)) |
8e180f3c | 1006 | return -5; |
889facbe | 1007 | t->extra_msr32 = msr & 0xFFFFFFFF; |
8e180f3c LB |
1008 | } |
1009 | ||
2f32edf1 LB |
1010 | if (extra_msr_offset64) |
1011 | if (get_msr(cpu, extra_msr_offset64, &t->extra_msr64)) | |
c98d5d94 LB |
1012 | return -5; |
1013 | ||
144b44b1 LB |
1014 | if (use_c1_residency_msr) { |
1015 | if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1)) | |
1016 | return -6; | |
1017 | } | |
1018 | ||
c98d5d94 LB |
1019 | /* collect core counters only for 1st thread in core */ |
1020 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
1021 | return 0; | |
1022 | ||
fb5d4327 | 1023 | if (do_nhm_cstates && !do_slm_cstates && !do_knl_cstates) { |
c98d5d94 LB |
1024 | if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3)) |
1025 | return -6; | |
144b44b1 LB |
1026 | } |
1027 | ||
fb5d4327 | 1028 | if (do_nhm_cstates && !do_knl_cstates) { |
c98d5d94 LB |
1029 | if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6)) |
1030 | return -7; | |
fb5d4327 DC |
1031 | } else if (do_knl_cstates) { |
1032 | if (get_msr(cpu, MSR_KNL_CORE_C6_RESIDENCY, &c->c6)) | |
1033 | return -7; | |
c98d5d94 LB |
1034 | } |
1035 | ||
1036 | if (do_snb_cstates) | |
1037 | if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7)) | |
1038 | return -8; | |
1039 | ||
889facbe LB |
1040 | if (do_dts) { |
1041 | if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr)) | |
1042 | return -9; | |
1043 | c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F); | |
1044 | } | |
1045 | ||
1046 | ||
c98d5d94 LB |
1047 | /* collect package counters only for 1st core in package */ |
1048 | if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
1049 | return 0; | |
1050 | ||
0b2bb692 LB |
1051 | if (do_skl_residency) { |
1052 | if (get_msr(cpu, MSR_PKG_WEIGHTED_CORE_C0_RES, &p->pkg_wtd_core_c0)) | |
1053 | return -10; | |
1054 | if (get_msr(cpu, MSR_PKG_ANY_CORE_C0_RES, &p->pkg_any_core_c0)) | |
1055 | return -11; | |
1056 | if (get_msr(cpu, MSR_PKG_ANY_GFXE_C0_RES, &p->pkg_any_gfxe_c0)) | |
1057 | return -12; | |
1058 | if (get_msr(cpu, MSR_PKG_BOTH_CORE_GFXE_C0_RES, &p->pkg_both_core_gfxe_c0)) | |
1059 | return -13; | |
1060 | } | |
ee7e38e3 | 1061 | if (do_pc3) |
c98d5d94 LB |
1062 | if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3)) |
1063 | return -9; | |
ee7e38e3 | 1064 | if (do_pc6) |
c98d5d94 LB |
1065 | if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6)) |
1066 | return -10; | |
ee7e38e3 | 1067 | if (do_pc2) |
c98d5d94 LB |
1068 | if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2)) |
1069 | return -11; | |
ee7e38e3 | 1070 | if (do_pc7) |
c98d5d94 LB |
1071 | if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7)) |
1072 | return -12; | |
ca58710f KCA |
1073 | if (do_c8_c9_c10) { |
1074 | if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8)) | |
1075 | return -13; | |
1076 | if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9)) | |
1077 | return -13; | |
1078 | if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10)) | |
1079 | return -13; | |
1080 | } | |
889facbe LB |
1081 | if (do_rapl & RAPL_PKG) { |
1082 | if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr)) | |
1083 | return -13; | |
1084 | p->energy_pkg = msr & 0xFFFFFFFF; | |
1085 | } | |
1086 | if (do_rapl & RAPL_CORES) { | |
1087 | if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr)) | |
1088 | return -14; | |
1089 | p->energy_cores = msr & 0xFFFFFFFF; | |
1090 | } | |
1091 | if (do_rapl & RAPL_DRAM) { | |
1092 | if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr)) | |
1093 | return -15; | |
1094 | p->energy_dram = msr & 0xFFFFFFFF; | |
1095 | } | |
1096 | if (do_rapl & RAPL_GFX) { | |
1097 | if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr)) | |
1098 | return -16; | |
1099 | p->energy_gfx = msr & 0xFFFFFFFF; | |
1100 | } | |
1101 | if (do_rapl & RAPL_PKG_PERF_STATUS) { | |
1102 | if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr)) | |
1103 | return -16; | |
1104 | p->rapl_pkg_perf_status = msr & 0xFFFFFFFF; | |
1105 | } | |
1106 | if (do_rapl & RAPL_DRAM_PERF_STATUS) { | |
1107 | if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr)) | |
1108 | return -16; | |
1109 | p->rapl_dram_perf_status = msr & 0xFFFFFFFF; | |
1110 | } | |
1111 | if (do_ptm) { | |
1112 | if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr)) | |
1113 | return -17; | |
1114 | p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F); | |
1115 | } | |
15aaa346 | 1116 | return 0; |
103a8fea LB |
1117 | } |
1118 | ||
ee7e38e3 LB |
1119 | /* |
1120 | * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit: | |
1121 | * If you change the values, note they are used both in comparisons | |
1122 | * (>= PCL__7) and to index pkg_cstate_limit_strings[]. | |
1123 | */ | |
1124 | ||
1125 | #define PCLUKN 0 /* Unknown */ | |
1126 | #define PCLRSV 1 /* Reserved */ | |
1127 | #define PCL__0 2 /* PC0 */ | |
1128 | #define PCL__1 3 /* PC1 */ | |
1129 | #define PCL__2 4 /* PC2 */ | |
1130 | #define PCL__3 5 /* PC3 */ | |
1131 | #define PCL__4 6 /* PC4 */ | |
1132 | #define PCL__6 7 /* PC6 */ | |
1133 | #define PCL_6N 8 /* PC6 No Retention */ | |
1134 | #define PCL_6R 9 /* PC6 Retention */ | |
1135 | #define PCL__7 10 /* PC7 */ | |
1136 | #define PCL_7S 11 /* PC7 Shrink */ | |
0b2bb692 LB |
1137 | #define PCL__8 12 /* PC8 */ |
1138 | #define PCL__9 13 /* PC9 */ | |
1139 | #define PCLUNL 14 /* Unlimited */ | |
ee7e38e3 LB |
1140 | |
1141 | int pkg_cstate_limit = PCLUKN; | |
1142 | char *pkg_cstate_limit_strings[] = { "reserved", "unknown", "pc0", "pc1", "pc2", | |
0b2bb692 | 1143 | "pc3", "pc4", "pc6", "pc6n", "pc6r", "pc7", "pc7s", "pc8", "pc9", "unlimited"}; |
ee7e38e3 | 1144 | |
e9257f5f LB |
1145 | 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}; |
1146 | 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}; | |
1147 | 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}; | |
1148 | 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}; | |
1149 | 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}; | |
1150 | 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 | 1151 | |
fcd17211 LB |
1152 | static void |
1153 | dump_nhm_platform_info(void) | |
103a8fea LB |
1154 | { |
1155 | unsigned long long msr; | |
1156 | unsigned int ratio; | |
1157 | ||
7ce7d5de | 1158 | get_msr(base_cpu, MSR_NHM_PLATFORM_INFO, &msr); |
103a8fea | 1159 | |
67920418 | 1160 | fprintf(stderr, "cpu0: MSR_NHM_PLATFORM_INFO: 0x%08llx\n", msr); |
6574a5d5 | 1161 | |
103a8fea | 1162 | ratio = (msr >> 40) & 0xFF; |
8f61f359 | 1163 | fprintf(stderr, "%d * %.0f = %.0f MHz max efficiency frequency\n", |
103a8fea LB |
1164 | ratio, bclk, ratio * bclk); |
1165 | ||
1166 | ratio = (msr >> 8) & 0xFF; | |
8f61f359 | 1167 | fprintf(stderr, "%d * %.0f = %.0f MHz base frequency\n", |
103a8fea LB |
1168 | ratio, bclk, ratio * bclk); |
1169 | ||
7ce7d5de | 1170 | get_msr(base_cpu, MSR_IA32_POWER_CTL, &msr); |
144b44b1 | 1171 | fprintf(stderr, "cpu0: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n", |
67920418 LB |
1172 | msr, msr & 0x2 ? "EN" : "DIS"); |
1173 | ||
fcd17211 LB |
1174 | return; |
1175 | } | |
1176 | ||
1177 | static void | |
1178 | dump_hsw_turbo_ratio_limits(void) | |
1179 | { | |
1180 | unsigned long long msr; | |
1181 | unsigned int ratio; | |
1182 | ||
7ce7d5de | 1183 | get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT2, &msr); |
fcd17211 LB |
1184 | |
1185 | fprintf(stderr, "cpu0: MSR_TURBO_RATIO_LIMIT2: 0x%08llx\n", msr); | |
1186 | ||
1187 | ratio = (msr >> 8) & 0xFF; | |
1188 | if (ratio) | |
1189 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 18 active cores\n", | |
1190 | ratio, bclk, ratio * bclk); | |
1191 | ||
1192 | ratio = (msr >> 0) & 0xFF; | |
1193 | if (ratio) | |
1194 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 17 active cores\n", | |
1195 | ratio, bclk, ratio * bclk); | |
1196 | return; | |
1197 | } | |
1198 | ||
1199 | static void | |
1200 | dump_ivt_turbo_ratio_limits(void) | |
1201 | { | |
1202 | unsigned long long msr; | |
1203 | unsigned int ratio; | |
6574a5d5 | 1204 | |
7ce7d5de | 1205 | get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &msr); |
6574a5d5 | 1206 | |
12bb43c6 | 1207 | fprintf(stderr, "cpu0: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", msr); |
6574a5d5 LB |
1208 | |
1209 | ratio = (msr >> 56) & 0xFF; | |
1210 | if (ratio) | |
1211 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 16 active cores\n", | |
1212 | ratio, bclk, ratio * bclk); | |
1213 | ||
1214 | ratio = (msr >> 48) & 0xFF; | |
1215 | if (ratio) | |
1216 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 15 active cores\n", | |
1217 | ratio, bclk, ratio * bclk); | |
1218 | ||
1219 | ratio = (msr >> 40) & 0xFF; | |
1220 | if (ratio) | |
1221 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 14 active cores\n", | |
1222 | ratio, bclk, ratio * bclk); | |
1223 | ||
1224 | ratio = (msr >> 32) & 0xFF; | |
1225 | if (ratio) | |
1226 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 13 active cores\n", | |
1227 | ratio, bclk, ratio * bclk); | |
1228 | ||
1229 | ratio = (msr >> 24) & 0xFF; | |
1230 | if (ratio) | |
1231 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 12 active cores\n", | |
1232 | ratio, bclk, ratio * bclk); | |
1233 | ||
1234 | ratio = (msr >> 16) & 0xFF; | |
1235 | if (ratio) | |
1236 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 11 active cores\n", | |
1237 | ratio, bclk, ratio * bclk); | |
1238 | ||
1239 | ratio = (msr >> 8) & 0xFF; | |
1240 | if (ratio) | |
1241 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 10 active cores\n", | |
1242 | ratio, bclk, ratio * bclk); | |
1243 | ||
1244 | ratio = (msr >> 0) & 0xFF; | |
1245 | if (ratio) | |
1246 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 9 active cores\n", | |
1247 | ratio, bclk, ratio * bclk); | |
fcd17211 LB |
1248 | return; |
1249 | } | |
6574a5d5 | 1250 | |
fcd17211 LB |
1251 | static void |
1252 | dump_nhm_turbo_ratio_limits(void) | |
1253 | { | |
1254 | unsigned long long msr; | |
1255 | unsigned int ratio; | |
103a8fea | 1256 | |
7ce7d5de | 1257 | get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr); |
103a8fea | 1258 | |
12bb43c6 | 1259 | fprintf(stderr, "cpu0: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n", msr); |
6574a5d5 LB |
1260 | |
1261 | ratio = (msr >> 56) & 0xFF; | |
1262 | if (ratio) | |
1263 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 8 active cores\n", | |
1264 | ratio, bclk, ratio * bclk); | |
1265 | ||
1266 | ratio = (msr >> 48) & 0xFF; | |
1267 | if (ratio) | |
1268 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 7 active cores\n", | |
1269 | ratio, bclk, ratio * bclk); | |
1270 | ||
1271 | ratio = (msr >> 40) & 0xFF; | |
1272 | if (ratio) | |
1273 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 6 active cores\n", | |
1274 | ratio, bclk, ratio * bclk); | |
1275 | ||
1276 | ratio = (msr >> 32) & 0xFF; | |
1277 | if (ratio) | |
1278 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 5 active cores\n", | |
1279 | ratio, bclk, ratio * bclk); | |
1280 | ||
103a8fea LB |
1281 | ratio = (msr >> 24) & 0xFF; |
1282 | if (ratio) | |
1283 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 4 active cores\n", | |
1284 | ratio, bclk, ratio * bclk); | |
1285 | ||
1286 | ratio = (msr >> 16) & 0xFF; | |
1287 | if (ratio) | |
1288 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 3 active cores\n", | |
1289 | ratio, bclk, ratio * bclk); | |
1290 | ||
1291 | ratio = (msr >> 8) & 0xFF; | |
1292 | if (ratio) | |
1293 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 2 active cores\n", | |
1294 | ratio, bclk, ratio * bclk); | |
1295 | ||
1296 | ratio = (msr >> 0) & 0xFF; | |
1297 | if (ratio) | |
1298 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 1 active cores\n", | |
1299 | ratio, bclk, ratio * bclk); | |
fcd17211 LB |
1300 | return; |
1301 | } | |
3a9a941d | 1302 | |
fb5d4327 DC |
1303 | static void |
1304 | dump_knl_turbo_ratio_limits(void) | |
1305 | { | |
1306 | int cores; | |
1307 | unsigned int ratio; | |
1308 | unsigned long long msr; | |
1309 | int delta_cores; | |
1310 | int delta_ratio; | |
1311 | int i; | |
1312 | ||
7ce7d5de | 1313 | get_msr(base_cpu, MSR_NHM_TURBO_RATIO_LIMIT, &msr); |
fb5d4327 DC |
1314 | |
1315 | fprintf(stderr, "cpu0: MSR_NHM_TURBO_RATIO_LIMIT: 0x%08llx\n", | |
1316 | msr); | |
1317 | ||
1318 | /** | |
1319 | * Turbo encoding in KNL is as follows: | |
1320 | * [7:0] -- Base value of number of active cores of bucket 1. | |
1321 | * [15:8] -- Base value of freq ratio of bucket 1. | |
1322 | * [20:16] -- +ve delta of number of active cores of bucket 2. | |
1323 | * i.e. active cores of bucket 2 = | |
1324 | * active cores of bucket 1 + delta | |
1325 | * [23:21] -- Negative delta of freq ratio of bucket 2. | |
1326 | * i.e. freq ratio of bucket 2 = | |
1327 | * freq ratio of bucket 1 - delta | |
1328 | * [28:24]-- +ve delta of number of active cores of bucket 3. | |
1329 | * [31:29]-- -ve delta of freq ratio of bucket 3. | |
1330 | * [36:32]-- +ve delta of number of active cores of bucket 4. | |
1331 | * [39:37]-- -ve delta of freq ratio of bucket 4. | |
1332 | * [44:40]-- +ve delta of number of active cores of bucket 5. | |
1333 | * [47:45]-- -ve delta of freq ratio of bucket 5. | |
1334 | * [52:48]-- +ve delta of number of active cores of bucket 6. | |
1335 | * [55:53]-- -ve delta of freq ratio of bucket 6. | |
1336 | * [60:56]-- +ve delta of number of active cores of bucket 7. | |
1337 | * [63:61]-- -ve delta of freq ratio of bucket 7. | |
1338 | */ | |
1339 | cores = msr & 0xFF; | |
1340 | ratio = (msr >> 8) && 0xFF; | |
1341 | if (ratio > 0) | |
1342 | fprintf(stderr, | |
1343 | "%d * %.0f = %.0f MHz max turbo %d active cores\n", | |
1344 | ratio, bclk, ratio * bclk, cores); | |
1345 | ||
1346 | for (i = 16; i < 64; i = i + 8) { | |
1347 | delta_cores = (msr >> i) & 0x1F; | |
1348 | delta_ratio = (msr >> (i + 5)) && 0x7; | |
1349 | if (!delta_cores || !delta_ratio) | |
1350 | return; | |
1351 | cores = cores + delta_cores; | |
1352 | ratio = ratio - delta_ratio; | |
1353 | ||
1354 | /** -ve ratios will make successive ratio calculations | |
1355 | * negative. Hence return instead of carrying on. | |
1356 | */ | |
1357 | if (ratio > 0) | |
1358 | fprintf(stderr, | |
1359 | "%d * %.0f = %.0f MHz max turbo %d active cores\n", | |
1360 | ratio, bclk, ratio * bclk, cores); | |
1361 | } | |
1362 | } | |
1363 | ||
fcd17211 LB |
1364 | static void |
1365 | dump_nhm_cst_cfg(void) | |
1366 | { | |
1367 | unsigned long long msr; | |
1368 | ||
7ce7d5de | 1369 | get_msr(base_cpu, MSR_NHM_SNB_PKG_CST_CFG_CTL, &msr); |
fcd17211 LB |
1370 | |
1371 | #define SNB_C1_AUTO_UNDEMOTE (1UL << 27) | |
1372 | #define SNB_C3_AUTO_UNDEMOTE (1UL << 28) | |
1373 | ||
1374 | fprintf(stderr, "cpu0: MSR_NHM_SNB_PKG_CST_CFG_CTL: 0x%08llx", msr); | |
1375 | ||
1376 | fprintf(stderr, " (%s%s%s%s%slocked: pkg-cstate-limit=%d: %s)\n", | |
1377 | (msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "", | |
1378 | (msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "", | |
1379 | (msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "", | |
1380 | (msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "", | |
1381 | (msr & (1 << 15)) ? "" : "UN", | |
1382 | (unsigned int)msr & 7, | |
1383 | pkg_cstate_limit_strings[pkg_cstate_limit]); | |
1384 | return; | |
103a8fea LB |
1385 | } |
1386 | ||
c98d5d94 | 1387 | void free_all_buffers(void) |
103a8fea | 1388 | { |
c98d5d94 LB |
1389 | CPU_FREE(cpu_present_set); |
1390 | cpu_present_set = NULL; | |
1391 | cpu_present_set = 0; | |
103a8fea | 1392 | |
c98d5d94 LB |
1393 | CPU_FREE(cpu_affinity_set); |
1394 | cpu_affinity_set = NULL; | |
1395 | cpu_affinity_setsize = 0; | |
103a8fea | 1396 | |
c98d5d94 LB |
1397 | free(thread_even); |
1398 | free(core_even); | |
1399 | free(package_even); | |
103a8fea | 1400 | |
c98d5d94 LB |
1401 | thread_even = NULL; |
1402 | core_even = NULL; | |
1403 | package_even = NULL; | |
103a8fea | 1404 | |
c98d5d94 LB |
1405 | free(thread_odd); |
1406 | free(core_odd); | |
1407 | free(package_odd); | |
103a8fea | 1408 | |
c98d5d94 LB |
1409 | thread_odd = NULL; |
1410 | core_odd = NULL; | |
1411 | package_odd = NULL; | |
103a8fea | 1412 | |
c98d5d94 LB |
1413 | free(output_buffer); |
1414 | output_buffer = NULL; | |
1415 | outp = NULL; | |
103a8fea LB |
1416 | } |
1417 | ||
57a42a34 JT |
1418 | /* |
1419 | * Open a file, and exit on failure | |
1420 | */ | |
1421 | FILE *fopen_or_die(const char *path, const char *mode) | |
1422 | { | |
1423 | FILE *filep = fopen(path, "r"); | |
b2c95d90 JT |
1424 | if (!filep) |
1425 | err(1, "%s: open failed", path); | |
57a42a34 JT |
1426 | return filep; |
1427 | } | |
1428 | ||
c98d5d94 | 1429 | /* |
95aebc44 | 1430 | * Parse a file containing a single int. |
c98d5d94 | 1431 | */ |
95aebc44 | 1432 | int parse_int_file(const char *fmt, ...) |
103a8fea | 1433 | { |
95aebc44 JT |
1434 | va_list args; |
1435 | char path[PATH_MAX]; | |
c98d5d94 | 1436 | FILE *filep; |
95aebc44 | 1437 | int value; |
103a8fea | 1438 | |
95aebc44 JT |
1439 | va_start(args, fmt); |
1440 | vsnprintf(path, sizeof(path), fmt, args); | |
1441 | va_end(args); | |
57a42a34 | 1442 | filep = fopen_or_die(path, "r"); |
b2c95d90 JT |
1443 | if (fscanf(filep, "%d", &value) != 1) |
1444 | err(1, "%s: failed to parse number from file", path); | |
c98d5d94 | 1445 | fclose(filep); |
95aebc44 JT |
1446 | return value; |
1447 | } | |
1448 | ||
1449 | /* | |
e275b388 DC |
1450 | * get_cpu_position_in_core(cpu) |
1451 | * return the position of the CPU among its HT siblings in the core | |
1452 | * return -1 if the sibling is not in list | |
95aebc44 | 1453 | */ |
e275b388 | 1454 | int get_cpu_position_in_core(int cpu) |
95aebc44 | 1455 | { |
e275b388 DC |
1456 | char path[64]; |
1457 | FILE *filep; | |
1458 | int this_cpu; | |
1459 | char character; | |
1460 | int i; | |
1461 | ||
1462 | sprintf(path, | |
1463 | "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", | |
1464 | cpu); | |
1465 | filep = fopen(path, "r"); | |
1466 | if (filep == NULL) { | |
1467 | perror(path); | |
1468 | exit(1); | |
1469 | } | |
1470 | ||
1471 | for (i = 0; i < topo.num_threads_per_core; i++) { | |
1472 | fscanf(filep, "%d", &this_cpu); | |
1473 | if (this_cpu == cpu) { | |
1474 | fclose(filep); | |
1475 | return i; | |
1476 | } | |
1477 | ||
1478 | /* Account for no separator after last thread*/ | |
1479 | if (i != (topo.num_threads_per_core - 1)) | |
1480 | fscanf(filep, "%c", &character); | |
1481 | } | |
1482 | ||
1483 | fclose(filep); | |
1484 | return -1; | |
103a8fea LB |
1485 | } |
1486 | ||
c98d5d94 LB |
1487 | /* |
1488 | * cpu_is_first_core_in_package(cpu) | |
1489 | * return 1 if given CPU is 1st core in package | |
1490 | */ | |
1491 | int cpu_is_first_core_in_package(int cpu) | |
103a8fea | 1492 | { |
95aebc44 | 1493 | return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu); |
103a8fea LB |
1494 | } |
1495 | ||
1496 | int get_physical_package_id(int cpu) | |
1497 | { | |
95aebc44 | 1498 | return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu); |
103a8fea LB |
1499 | } |
1500 | ||
1501 | int get_core_id(int cpu) | |
1502 | { | |
95aebc44 | 1503 | return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu); |
103a8fea LB |
1504 | } |
1505 | ||
c98d5d94 LB |
1506 | int get_num_ht_siblings(int cpu) |
1507 | { | |
1508 | char path[80]; | |
1509 | FILE *filep; | |
e275b388 DC |
1510 | int sib1; |
1511 | int matches = 0; | |
c98d5d94 | 1512 | char character; |
e275b388 DC |
1513 | char str[100]; |
1514 | char *ch; | |
c98d5d94 LB |
1515 | |
1516 | sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu); | |
57a42a34 | 1517 | filep = fopen_or_die(path, "r"); |
e275b388 | 1518 | |
c98d5d94 LB |
1519 | /* |
1520 | * file format: | |
e275b388 DC |
1521 | * A ',' separated or '-' separated set of numbers |
1522 | * (eg 1-2 or 1,3,4,5) | |
c98d5d94 | 1523 | */ |
e275b388 DC |
1524 | fscanf(filep, "%d%c\n", &sib1, &character); |
1525 | fseek(filep, 0, SEEK_SET); | |
1526 | fgets(str, 100, filep); | |
1527 | ch = strchr(str, character); | |
1528 | while (ch != NULL) { | |
1529 | matches++; | |
1530 | ch = strchr(ch+1, character); | |
1531 | } | |
c98d5d94 LB |
1532 | |
1533 | fclose(filep); | |
e275b388 | 1534 | return matches+1; |
c98d5d94 LB |
1535 | } |
1536 | ||
103a8fea | 1537 | /* |
c98d5d94 LB |
1538 | * run func(thread, core, package) in topology order |
1539 | * skip non-present cpus | |
103a8fea LB |
1540 | */ |
1541 | ||
c98d5d94 LB |
1542 | int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *, |
1543 | struct pkg_data *, struct thread_data *, struct core_data *, | |
1544 | struct pkg_data *), struct thread_data *thread_base, | |
1545 | struct core_data *core_base, struct pkg_data *pkg_base, | |
1546 | struct thread_data *thread_base2, struct core_data *core_base2, | |
1547 | struct pkg_data *pkg_base2) | |
1548 | { | |
1549 | int retval, pkg_no, core_no, thread_no; | |
1550 | ||
1551 | for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) { | |
1552 | for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) { | |
1553 | for (thread_no = 0; thread_no < | |
1554 | topo.num_threads_per_core; ++thread_no) { | |
1555 | struct thread_data *t, *t2; | |
1556 | struct core_data *c, *c2; | |
1557 | struct pkg_data *p, *p2; | |
1558 | ||
1559 | t = GET_THREAD(thread_base, thread_no, core_no, pkg_no); | |
1560 | ||
1561 | if (cpu_is_not_present(t->cpu_id)) | |
1562 | continue; | |
1563 | ||
1564 | t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no); | |
1565 | ||
1566 | c = GET_CORE(core_base, core_no, pkg_no); | |
1567 | c2 = GET_CORE(core_base2, core_no, pkg_no); | |
1568 | ||
1569 | p = GET_PKG(pkg_base, pkg_no); | |
1570 | p2 = GET_PKG(pkg_base2, pkg_no); | |
1571 | ||
1572 | retval = func(t, c, p, t2, c2, p2); | |
1573 | if (retval) | |
1574 | return retval; | |
1575 | } | |
1576 | } | |
1577 | } | |
1578 | return 0; | |
1579 | } | |
1580 | ||
1581 | /* | |
1582 | * run func(cpu) on every cpu in /proc/stat | |
1583 | * return max_cpu number | |
1584 | */ | |
1585 | int for_all_proc_cpus(int (func)(int)) | |
103a8fea LB |
1586 | { |
1587 | FILE *fp; | |
c98d5d94 | 1588 | int cpu_num; |
103a8fea LB |
1589 | int retval; |
1590 | ||
57a42a34 | 1591 | fp = fopen_or_die(proc_stat, "r"); |
103a8fea LB |
1592 | |
1593 | retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n"); | |
b2c95d90 JT |
1594 | if (retval != 0) |
1595 | err(1, "%s: failed to parse format", proc_stat); | |
103a8fea | 1596 | |
c98d5d94 LB |
1597 | while (1) { |
1598 | retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num); | |
103a8fea LB |
1599 | if (retval != 1) |
1600 | break; | |
1601 | ||
c98d5d94 LB |
1602 | retval = func(cpu_num); |
1603 | if (retval) { | |
1604 | fclose(fp); | |
1605 | return(retval); | |
1606 | } | |
103a8fea LB |
1607 | } |
1608 | fclose(fp); | |
c98d5d94 | 1609 | return 0; |
103a8fea LB |
1610 | } |
1611 | ||
1612 | void re_initialize(void) | |
1613 | { | |
c98d5d94 LB |
1614 | free_all_buffers(); |
1615 | setup_all_buffers(); | |
1616 | printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus); | |
103a8fea LB |
1617 | } |
1618 | ||
c98d5d94 | 1619 | |
103a8fea | 1620 | /* |
c98d5d94 LB |
1621 | * count_cpus() |
1622 | * remember the last one seen, it will be the max | |
103a8fea | 1623 | */ |
c98d5d94 | 1624 | int count_cpus(int cpu) |
103a8fea | 1625 | { |
c98d5d94 LB |
1626 | if (topo.max_cpu_num < cpu) |
1627 | topo.max_cpu_num = cpu; | |
103a8fea | 1628 | |
c98d5d94 LB |
1629 | topo.num_cpus += 1; |
1630 | return 0; | |
1631 | } | |
1632 | int mark_cpu_present(int cpu) | |
1633 | { | |
1634 | CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set); | |
15aaa346 | 1635 | return 0; |
103a8fea LB |
1636 | } |
1637 | ||
1638 | void turbostat_loop() | |
1639 | { | |
c98d5d94 | 1640 | int retval; |
e52966c0 | 1641 | int restarted = 0; |
c98d5d94 | 1642 | |
103a8fea | 1643 | restart: |
e52966c0 LB |
1644 | restarted++; |
1645 | ||
c98d5d94 | 1646 | retval = for_all_cpus(get_counters, EVEN_COUNTERS); |
d91bb17c LB |
1647 | if (retval < -1) { |
1648 | exit(retval); | |
1649 | } else if (retval == -1) { | |
e52966c0 LB |
1650 | if (restarted > 1) { |
1651 | exit(retval); | |
1652 | } | |
c98d5d94 LB |
1653 | re_initialize(); |
1654 | goto restart; | |
1655 | } | |
e52966c0 | 1656 | restarted = 0; |
103a8fea LB |
1657 | gettimeofday(&tv_even, (struct timezone *)NULL); |
1658 | ||
1659 | while (1) { | |
c98d5d94 | 1660 | if (for_all_proc_cpus(cpu_is_not_present)) { |
103a8fea LB |
1661 | re_initialize(); |
1662 | goto restart; | |
1663 | } | |
1664 | sleep(interval_sec); | |
c98d5d94 | 1665 | retval = for_all_cpus(get_counters, ODD_COUNTERS); |
d91bb17c LB |
1666 | if (retval < -1) { |
1667 | exit(retval); | |
1668 | } else if (retval == -1) { | |
15aaa346 LB |
1669 | re_initialize(); |
1670 | goto restart; | |
1671 | } | |
103a8fea | 1672 | gettimeofday(&tv_odd, (struct timezone *)NULL); |
103a8fea | 1673 | timersub(&tv_odd, &tv_even, &tv_delta); |
c98d5d94 LB |
1674 | for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS); |
1675 | compute_average(EVEN_COUNTERS); | |
1676 | format_all_counters(EVEN_COUNTERS); | |
1677 | flush_stdout(); | |
15aaa346 | 1678 | sleep(interval_sec); |
c98d5d94 | 1679 | retval = for_all_cpus(get_counters, EVEN_COUNTERS); |
d91bb17c LB |
1680 | if (retval < -1) { |
1681 | exit(retval); | |
1682 | } else if (retval == -1) { | |
103a8fea LB |
1683 | re_initialize(); |
1684 | goto restart; | |
1685 | } | |
103a8fea | 1686 | gettimeofday(&tv_even, (struct timezone *)NULL); |
103a8fea | 1687 | timersub(&tv_even, &tv_odd, &tv_delta); |
c98d5d94 LB |
1688 | for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS); |
1689 | compute_average(ODD_COUNTERS); | |
1690 | format_all_counters(ODD_COUNTERS); | |
1691 | flush_stdout(); | |
103a8fea LB |
1692 | } |
1693 | } | |
1694 | ||
1695 | void check_dev_msr() | |
1696 | { | |
1697 | struct stat sb; | |
7ce7d5de | 1698 | char pathname[32]; |
103a8fea | 1699 | |
7ce7d5de PB |
1700 | sprintf(pathname, "/dev/cpu/%d/msr", base_cpu); |
1701 | if (stat(pathname, &sb)) | |
a21d38c8 LB |
1702 | if (system("/sbin/modprobe msr > /dev/null 2>&1")) |
1703 | err(-5, "no /dev/cpu/0/msr, Try \"# modprobe msr\" "); | |
103a8fea LB |
1704 | } |
1705 | ||
98481e79 | 1706 | void check_permissions() |
103a8fea | 1707 | { |
98481e79 LB |
1708 | struct __user_cap_header_struct cap_header_data; |
1709 | cap_user_header_t cap_header = &cap_header_data; | |
1710 | struct __user_cap_data_struct cap_data_data; | |
1711 | cap_user_data_t cap_data = &cap_data_data; | |
1712 | extern int capget(cap_user_header_t hdrp, cap_user_data_t datap); | |
1713 | int do_exit = 0; | |
7ce7d5de | 1714 | char pathname[32]; |
98481e79 LB |
1715 | |
1716 | /* check for CAP_SYS_RAWIO */ | |
1717 | cap_header->pid = getpid(); | |
1718 | cap_header->version = _LINUX_CAPABILITY_VERSION; | |
1719 | if (capget(cap_header, cap_data) < 0) | |
1720 | err(-6, "capget(2) failed"); | |
1721 | ||
1722 | if ((cap_data->effective & (1 << CAP_SYS_RAWIO)) == 0) { | |
1723 | do_exit++; | |
1724 | warnx("capget(CAP_SYS_RAWIO) failed," | |
1725 | " try \"# setcap cap_sys_rawio=ep %s\"", progname); | |
1726 | } | |
1727 | ||
1728 | /* test file permissions */ | |
7ce7d5de PB |
1729 | sprintf(pathname, "/dev/cpu/%d/msr", base_cpu); |
1730 | if (euidaccess(pathname, R_OK)) { | |
98481e79 LB |
1731 | do_exit++; |
1732 | warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr"); | |
1733 | } | |
1734 | ||
1735 | /* if all else fails, thell them to be root */ | |
1736 | if (do_exit) | |
1737 | if (getuid() != 0) | |
d7899447 | 1738 | warnx("... or simply run as root"); |
98481e79 LB |
1739 | |
1740 | if (do_exit) | |
1741 | exit(-6); | |
103a8fea LB |
1742 | } |
1743 | ||
d7899447 LB |
1744 | /* |
1745 | * NHM adds support for additional MSRs: | |
1746 | * | |
1747 | * MSR_SMI_COUNT 0x00000034 | |
1748 | * | |
1749 | * MSR_NHM_PLATFORM_INFO 0x000000ce | |
1750 | * MSR_NHM_SNB_PKG_CST_CFG_CTL 0x000000e2 | |
1751 | * | |
1752 | * MSR_PKG_C3_RESIDENCY 0x000003f8 | |
1753 | * MSR_PKG_C6_RESIDENCY 0x000003f9 | |
1754 | * MSR_CORE_C3_RESIDENCY 0x000003fc | |
1755 | * MSR_CORE_C6_RESIDENCY 0x000003fd | |
1756 | * | |
ee7e38e3 LB |
1757 | * Side effect: |
1758 | * sets global pkg_cstate_limit to decode MSR_NHM_SNB_PKG_CST_CFG_CTL | |
d7899447 | 1759 | */ |
ee7e38e3 | 1760 | int probe_nhm_msrs(unsigned int family, unsigned int model) |
103a8fea | 1761 | { |
ee7e38e3 LB |
1762 | unsigned long long msr; |
1763 | int *pkg_cstate_limits; | |
1764 | ||
103a8fea LB |
1765 | if (!genuine_intel) |
1766 | return 0; | |
1767 | ||
1768 | if (family != 6) | |
1769 | return 0; | |
1770 | ||
1771 | switch (model) { | |
1772 | case 0x1A: /* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */ | |
1773 | case 0x1E: /* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */ | |
1774 | case 0x1F: /* Core i7 and i5 Processor - Nehalem */ | |
1775 | case 0x25: /* Westmere Client - Clarkdale, Arrandale */ | |
1776 | case 0x2C: /* Westmere EP - Gulftown */ | |
ee7e38e3 LB |
1777 | case 0x2E: /* Nehalem-EX Xeon - Beckton */ |
1778 | case 0x2F: /* Westmere-EX Xeon - Eagleton */ | |
1779 | pkg_cstate_limits = nhm_pkg_cstate_limits; | |
1780 | break; | |
103a8fea LB |
1781 | case 0x2A: /* SNB */ |
1782 | case 0x2D: /* SNB Xeon */ | |
553575f1 | 1783 | case 0x3A: /* IVB */ |
1300651b | 1784 | case 0x3E: /* IVB Xeon */ |
ee7e38e3 LB |
1785 | pkg_cstate_limits = snb_pkg_cstate_limits; |
1786 | break; | |
70b43400 | 1787 | case 0x3C: /* HSW */ |
e6f9bb3c | 1788 | case 0x3F: /* HSX */ |
70b43400 | 1789 | case 0x45: /* HSW */ |
149c2319 | 1790 | case 0x46: /* HSW */ |
4e8e863f | 1791 | case 0x3D: /* BDW */ |
48a0631c | 1792 | case 0x47: /* BDW */ |
4e8e863f LB |
1793 | case 0x4F: /* BDX */ |
1794 | case 0x56: /* BDX-DE */ | |
0b2bb692 LB |
1795 | case 0x4E: /* SKL */ |
1796 | case 0x5E: /* SKL */ | |
ee7e38e3 LB |
1797 | pkg_cstate_limits = hsw_pkg_cstate_limits; |
1798 | break; | |
1799 | case 0x37: /* BYT */ | |
1800 | case 0x4D: /* AVN */ | |
1801 | pkg_cstate_limits = slv_pkg_cstate_limits; | |
1802 | break; | |
1803 | case 0x4C: /* AMT */ | |
1804 | pkg_cstate_limits = amt_pkg_cstate_limits; | |
1805 | break; | |
1806 | case 0x57: /* PHI */ | |
1807 | pkg_cstate_limits = phi_pkg_cstate_limits; | |
1808 | break; | |
103a8fea LB |
1809 | default: |
1810 | return 0; | |
1811 | } | |
7ce7d5de | 1812 | get_msr(base_cpu, MSR_NHM_SNB_PKG_CST_CFG_CTL, &msr); |
ee7e38e3 | 1813 | |
e9257f5f | 1814 | pkg_cstate_limit = pkg_cstate_limits[msr & 0xF]; |
ee7e38e3 LB |
1815 | |
1816 | return 1; | |
103a8fea | 1817 | } |
d7899447 LB |
1818 | int has_nhm_turbo_ratio_limit(unsigned int family, unsigned int model) |
1819 | { | |
d7899447 LB |
1820 | switch (model) { |
1821 | /* Nehalem compatible, but do not include turbo-ratio limit support */ | |
1822 | case 0x2E: /* Nehalem-EX Xeon - Beckton */ | |
1823 | case 0x2F: /* Westmere-EX Xeon - Eagleton */ | |
1824 | return 0; | |
1825 | default: | |
1826 | return 1; | |
1827 | } | |
1828 | } | |
6574a5d5 LB |
1829 | int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model) |
1830 | { | |
1831 | if (!genuine_intel) | |
1832 | return 0; | |
1833 | ||
1834 | if (family != 6) | |
1835 | return 0; | |
1836 | ||
1837 | switch (model) { | |
1838 | case 0x3E: /* IVB Xeon */ | |
fcd17211 LB |
1839 | case 0x3F: /* HSW Xeon */ |
1840 | return 1; | |
1841 | default: | |
1842 | return 0; | |
1843 | } | |
1844 | } | |
1845 | int has_hsw_turbo_ratio_limit(unsigned int family, unsigned int model) | |
1846 | { | |
1847 | if (!genuine_intel) | |
1848 | return 0; | |
1849 | ||
1850 | if (family != 6) | |
1851 | return 0; | |
1852 | ||
1853 | switch (model) { | |
1854 | case 0x3F: /* HSW Xeon */ | |
6574a5d5 LB |
1855 | return 1; |
1856 | default: | |
1857 | return 0; | |
1858 | } | |
1859 | } | |
1860 | ||
fb5d4327 DC |
1861 | int has_knl_turbo_ratio_limit(unsigned int family, unsigned int model) |
1862 | { | |
1863 | if (!genuine_intel) | |
1864 | return 0; | |
1865 | ||
1866 | if (family != 6) | |
1867 | return 0; | |
1868 | ||
1869 | switch (model) { | |
1870 | case 0x57: /* Knights Landing */ | |
1871 | return 1; | |
1872 | default: | |
1873 | return 0; | |
1874 | } | |
1875 | } | |
fcd17211 LB |
1876 | static void |
1877 | dump_cstate_pstate_config_info(family, model) | |
1878 | { | |
1879 | if (!do_nhm_platform_info) | |
1880 | return; | |
1881 | ||
1882 | dump_nhm_platform_info(); | |
1883 | ||
1884 | if (has_hsw_turbo_ratio_limit(family, model)) | |
1885 | dump_hsw_turbo_ratio_limits(); | |
1886 | ||
1887 | if (has_ivt_turbo_ratio_limit(family, model)) | |
1888 | dump_ivt_turbo_ratio_limits(); | |
1889 | ||
1890 | if (has_nhm_turbo_ratio_limit(family, model)) | |
1891 | dump_nhm_turbo_ratio_limits(); | |
1892 | ||
fb5d4327 DC |
1893 | if (has_knl_turbo_ratio_limit(family, model)) |
1894 | dump_knl_turbo_ratio_limits(); | |
1895 | ||
fcd17211 LB |
1896 | dump_nhm_cst_cfg(); |
1897 | } | |
1898 | ||
1899 | ||
889facbe LB |
1900 | /* |
1901 | * print_epb() | |
1902 | * Decode the ENERGY_PERF_BIAS MSR | |
1903 | */ | |
1904 | int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
1905 | { | |
1906 | unsigned long long msr; | |
1907 | char *epb_string; | |
1908 | int cpu; | |
1909 | ||
1910 | if (!has_epb) | |
1911 | return 0; | |
1912 | ||
1913 | cpu = t->cpu_id; | |
1914 | ||
1915 | /* EPB is per-package */ | |
1916 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
1917 | return 0; | |
1918 | ||
1919 | if (cpu_migrate(cpu)) { | |
1920 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
1921 | return -1; | |
1922 | } | |
1923 | ||
1924 | if (get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr)) | |
1925 | return 0; | |
1926 | ||
e9be7dd6 | 1927 | switch (msr & 0xF) { |
889facbe LB |
1928 | case ENERGY_PERF_BIAS_PERFORMANCE: |
1929 | epb_string = "performance"; | |
1930 | break; | |
1931 | case ENERGY_PERF_BIAS_NORMAL: | |
1932 | epb_string = "balanced"; | |
1933 | break; | |
1934 | case ENERGY_PERF_BIAS_POWERSAVE: | |
1935 | epb_string = "powersave"; | |
1936 | break; | |
1937 | default: | |
1938 | epb_string = "custom"; | |
1939 | break; | |
1940 | } | |
1941 | fprintf(stderr, "cpu%d: MSR_IA32_ENERGY_PERF_BIAS: 0x%08llx (%s)\n", cpu, msr, epb_string); | |
1942 | ||
1943 | return 0; | |
1944 | } | |
1945 | ||
3a9a941d LB |
1946 | /* |
1947 | * print_perf_limit() | |
1948 | */ | |
1949 | int print_perf_limit(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
1950 | { | |
1951 | unsigned long long msr; | |
1952 | int cpu; | |
1953 | ||
1954 | cpu = t->cpu_id; | |
1955 | ||
1956 | /* per-package */ | |
1957 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
1958 | return 0; | |
1959 | ||
1960 | if (cpu_migrate(cpu)) { | |
1961 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
1962 | return -1; | |
1963 | } | |
1964 | ||
1965 | if (do_core_perf_limit_reasons) { | |
1966 | get_msr(cpu, MSR_CORE_PERF_LIMIT_REASONS, &msr); | |
1967 | fprintf(stderr, "cpu%d: MSR_CORE_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr); | |
1968 | fprintf(stderr, " (Active: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)", | |
e33cbe85 | 1969 | (msr & 1 << 15) ? "bit15, " : "", |
3a9a941d | 1970 | (msr & 1 << 14) ? "bit14, " : "", |
e33cbe85 LB |
1971 | (msr & 1 << 13) ? "Transitions, " : "", |
1972 | (msr & 1 << 12) ? "MultiCoreTurbo, " : "", | |
1973 | (msr & 1 << 11) ? "PkgPwrL2, " : "", | |
1974 | (msr & 1 << 10) ? "PkgPwrL1, " : "", | |
1975 | (msr & 1 << 9) ? "CorePwr, " : "", | |
1976 | (msr & 1 << 8) ? "Amps, " : "", | |
1977 | (msr & 1 << 6) ? "VR-Therm, " : "", | |
1978 | (msr & 1 << 5) ? "Auto-HWP, " : "", | |
1979 | (msr & 1 << 4) ? "Graphics, " : "", | |
1980 | (msr & 1 << 2) ? "bit2, " : "", | |
1981 | (msr & 1 << 1) ? "ThermStatus, " : "", | |
1982 | (msr & 1 << 0) ? "PROCHOT, " : ""); | |
3a9a941d | 1983 | fprintf(stderr, " (Logged: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)\n", |
e33cbe85 | 1984 | (msr & 1 << 31) ? "bit31, " : "", |
3a9a941d | 1985 | (msr & 1 << 30) ? "bit30, " : "", |
e33cbe85 LB |
1986 | (msr & 1 << 29) ? "Transitions, " : "", |
1987 | (msr & 1 << 28) ? "MultiCoreTurbo, " : "", | |
1988 | (msr & 1 << 27) ? "PkgPwrL2, " : "", | |
1989 | (msr & 1 << 26) ? "PkgPwrL1, " : "", | |
1990 | (msr & 1 << 25) ? "CorePwr, " : "", | |
1991 | (msr & 1 << 24) ? "Amps, " : "", | |
1992 | (msr & 1 << 22) ? "VR-Therm, " : "", | |
1993 | (msr & 1 << 21) ? "Auto-HWP, " : "", | |
1994 | (msr & 1 << 20) ? "Graphics, " : "", | |
1995 | (msr & 1 << 18) ? "bit18, " : "", | |
1996 | (msr & 1 << 17) ? "ThermStatus, " : "", | |
1997 | (msr & 1 << 16) ? "PROCHOT, " : ""); | |
3a9a941d LB |
1998 | |
1999 | } | |
2000 | if (do_gfx_perf_limit_reasons) { | |
2001 | get_msr(cpu, MSR_GFX_PERF_LIMIT_REASONS, &msr); | |
2002 | fprintf(stderr, "cpu%d: MSR_GFX_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr); | |
2003 | fprintf(stderr, " (Active: %s%s%s%s%s%s%s%s)", | |
2004 | (msr & 1 << 0) ? "PROCHOT, " : "", | |
2005 | (msr & 1 << 1) ? "ThermStatus, " : "", | |
2006 | (msr & 1 << 4) ? "Graphics, " : "", | |
2007 | (msr & 1 << 6) ? "VR-Therm, " : "", | |
2008 | (msr & 1 << 8) ? "Amps, " : "", | |
2009 | (msr & 1 << 9) ? "GFXPwr, " : "", | |
2010 | (msr & 1 << 10) ? "PkgPwrL1, " : "", | |
2011 | (msr & 1 << 11) ? "PkgPwrL2, " : ""); | |
2012 | fprintf(stderr, " (Logged: %s%s%s%s%s%s%s%s)\n", | |
2013 | (msr & 1 << 16) ? "PROCHOT, " : "", | |
2014 | (msr & 1 << 17) ? "ThermStatus, " : "", | |
2015 | (msr & 1 << 20) ? "Graphics, " : "", | |
2016 | (msr & 1 << 22) ? "VR-Therm, " : "", | |
2017 | (msr & 1 << 24) ? "Amps, " : "", | |
2018 | (msr & 1 << 25) ? "GFXPwr, " : "", | |
2019 | (msr & 1 << 26) ? "PkgPwrL1, " : "", | |
2020 | (msr & 1 << 27) ? "PkgPwrL2, " : ""); | |
2021 | } | |
2022 | if (do_ring_perf_limit_reasons) { | |
2023 | get_msr(cpu, MSR_RING_PERF_LIMIT_REASONS, &msr); | |
2024 | fprintf(stderr, "cpu%d: MSR_RING_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr); | |
2025 | fprintf(stderr, " (Active: %s%s%s%s%s%s)", | |
2026 | (msr & 1 << 0) ? "PROCHOT, " : "", | |
2027 | (msr & 1 << 1) ? "ThermStatus, " : "", | |
2028 | (msr & 1 << 6) ? "VR-Therm, " : "", | |
2029 | (msr & 1 << 8) ? "Amps, " : "", | |
2030 | (msr & 1 << 10) ? "PkgPwrL1, " : "", | |
2031 | (msr & 1 << 11) ? "PkgPwrL2, " : ""); | |
2032 | fprintf(stderr, " (Logged: %s%s%s%s%s%s)\n", | |
2033 | (msr & 1 << 16) ? "PROCHOT, " : "", | |
2034 | (msr & 1 << 17) ? "ThermStatus, " : "", | |
2035 | (msr & 1 << 22) ? "VR-Therm, " : "", | |
2036 | (msr & 1 << 24) ? "Amps, " : "", | |
2037 | (msr & 1 << 26) ? "PkgPwrL1, " : "", | |
2038 | (msr & 1 << 27) ? "PkgPwrL2, " : ""); | |
2039 | } | |
2040 | return 0; | |
2041 | } | |
2042 | ||
889facbe LB |
2043 | #define RAPL_POWER_GRANULARITY 0x7FFF /* 15 bit power granularity */ |
2044 | #define RAPL_TIME_GRANULARITY 0x3F /* 6 bit time granularity */ | |
2045 | ||
144b44b1 LB |
2046 | double get_tdp(model) |
2047 | { | |
2048 | unsigned long long msr; | |
2049 | ||
2050 | if (do_rapl & RAPL_PKG_POWER_INFO) | |
7ce7d5de | 2051 | if (!get_msr(base_cpu, MSR_PKG_POWER_INFO, &msr)) |
144b44b1 LB |
2052 | return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units; |
2053 | ||
2054 | switch (model) { | |
2055 | case 0x37: | |
2056 | case 0x4D: | |
2057 | return 30.0; | |
2058 | default: | |
2059 | return 135.0; | |
2060 | } | |
2061 | } | |
2062 | ||
40ee8e3b AS |
2063 | /* |
2064 | * rapl_dram_energy_units_probe() | |
2065 | * Energy units are either hard-coded, or come from RAPL Energy Unit MSR. | |
2066 | */ | |
2067 | static double | |
2068 | rapl_dram_energy_units_probe(int model, double rapl_energy_units) | |
2069 | { | |
2070 | /* only called for genuine_intel, family 6 */ | |
2071 | ||
2072 | switch (model) { | |
2073 | case 0x3F: /* HSX */ | |
2074 | case 0x4F: /* BDX */ | |
2075 | case 0x56: /* BDX-DE */ | |
fb5d4327 | 2076 | case 0x57: /* KNL */ |
40ee8e3b AS |
2077 | return (rapl_dram_energy_units = 15.3 / 1000000); |
2078 | default: | |
2079 | return (rapl_energy_units); | |
2080 | } | |
2081 | } | |
2082 | ||
144b44b1 | 2083 | |
889facbe LB |
2084 | /* |
2085 | * rapl_probe() | |
2086 | * | |
144b44b1 | 2087 | * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units |
889facbe LB |
2088 | */ |
2089 | void rapl_probe(unsigned int family, unsigned int model) | |
2090 | { | |
2091 | unsigned long long msr; | |
144b44b1 | 2092 | unsigned int time_unit; |
889facbe LB |
2093 | double tdp; |
2094 | ||
2095 | if (!genuine_intel) | |
2096 | return; | |
2097 | ||
2098 | if (family != 6) | |
2099 | return; | |
2100 | ||
2101 | switch (model) { | |
2102 | case 0x2A: | |
2103 | case 0x3A: | |
70b43400 | 2104 | case 0x3C: /* HSW */ |
70b43400 | 2105 | case 0x45: /* HSW */ |
149c2319 | 2106 | case 0x46: /* HSW */ |
4e8e863f | 2107 | case 0x3D: /* BDW */ |
48a0631c | 2108 | case 0x47: /* BDW */ |
144b44b1 | 2109 | do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO; |
889facbe | 2110 | break; |
0b2bb692 LB |
2111 | case 0x4E: /* SKL */ |
2112 | case 0x5E: /* SKL */ | |
2113 | do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO; | |
2114 | break; | |
e6f9bb3c | 2115 | case 0x3F: /* HSX */ |
4e8e863f LB |
2116 | case 0x4F: /* BDX */ |
2117 | case 0x56: /* BDX-DE */ | |
fb5d4327 | 2118 | case 0x57: /* KNL */ |
0b2bb692 | 2119 | do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO; |
e6f9bb3c | 2120 | break; |
889facbe LB |
2121 | case 0x2D: |
2122 | case 0x3E: | |
0b2bb692 | 2123 | 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 |
2124 | break; |
2125 | case 0x37: /* BYT */ | |
2126 | case 0x4D: /* AVN */ | |
2127 | do_rapl = RAPL_PKG | RAPL_CORES ; | |
889facbe LB |
2128 | break; |
2129 | default: | |
2130 | return; | |
2131 | } | |
2132 | ||
2133 | /* units on package 0, verify later other packages match */ | |
7ce7d5de | 2134 | if (get_msr(base_cpu, MSR_RAPL_POWER_UNIT, &msr)) |
889facbe LB |
2135 | return; |
2136 | ||
2137 | rapl_power_units = 1.0 / (1 << (msr & 0xF)); | |
144b44b1 LB |
2138 | if (model == 0x37) |
2139 | rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000; | |
2140 | else | |
2141 | rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F)); | |
889facbe | 2142 | |
40ee8e3b AS |
2143 | rapl_dram_energy_units = rapl_dram_energy_units_probe(model, rapl_energy_units); |
2144 | ||
144b44b1 LB |
2145 | time_unit = msr >> 16 & 0xF; |
2146 | if (time_unit == 0) | |
2147 | time_unit = 0xA; | |
889facbe | 2148 | |
144b44b1 | 2149 | rapl_time_units = 1.0 / (1 << (time_unit)); |
889facbe | 2150 | |
144b44b1 | 2151 | tdp = get_tdp(model); |
889facbe | 2152 | |
144b44b1 | 2153 | rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp; |
d8af6f5f | 2154 | if (debug) |
144b44b1 | 2155 | fprintf(stderr, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp); |
889facbe LB |
2156 | |
2157 | return; | |
2158 | } | |
2159 | ||
3a9a941d LB |
2160 | void perf_limit_reasons_probe(family, model) |
2161 | { | |
2162 | if (!genuine_intel) | |
2163 | return; | |
2164 | ||
2165 | if (family != 6) | |
2166 | return; | |
2167 | ||
2168 | switch (model) { | |
2169 | case 0x3C: /* HSW */ | |
2170 | case 0x45: /* HSW */ | |
2171 | case 0x46: /* HSW */ | |
2172 | do_gfx_perf_limit_reasons = 1; | |
2173 | case 0x3F: /* HSX */ | |
2174 | do_core_perf_limit_reasons = 1; | |
2175 | do_ring_perf_limit_reasons = 1; | |
2176 | default: | |
2177 | return; | |
2178 | } | |
2179 | } | |
2180 | ||
889facbe LB |
2181 | int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p) |
2182 | { | |
2183 | unsigned long long msr; | |
2184 | unsigned int dts; | |
2185 | int cpu; | |
2186 | ||
2187 | if (!(do_dts || do_ptm)) | |
2188 | return 0; | |
2189 | ||
2190 | cpu = t->cpu_id; | |
2191 | ||
2192 | /* DTS is per-core, no need to print for each thread */ | |
2193 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
2194 | return 0; | |
2195 | ||
2196 | if (cpu_migrate(cpu)) { | |
2197 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
2198 | return -1; | |
2199 | } | |
2200 | ||
2201 | if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) { | |
2202 | if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr)) | |
2203 | return 0; | |
2204 | ||
2205 | dts = (msr >> 16) & 0x7F; | |
2206 | fprintf(stderr, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n", | |
2207 | cpu, msr, tcc_activation_temp - dts); | |
2208 | ||
2209 | #ifdef THERM_DEBUG | |
2210 | if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr)) | |
2211 | return 0; | |
2212 | ||
2213 | dts = (msr >> 16) & 0x7F; | |
2214 | dts2 = (msr >> 8) & 0x7F; | |
2215 | fprintf(stderr, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n", | |
2216 | cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2); | |
2217 | #endif | |
2218 | } | |
2219 | ||
2220 | ||
2221 | if (do_dts) { | |
2222 | unsigned int resolution; | |
2223 | ||
2224 | if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr)) | |
2225 | return 0; | |
2226 | ||
2227 | dts = (msr >> 16) & 0x7F; | |
2228 | resolution = (msr >> 27) & 0xF; | |
2229 | fprintf(stderr, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n", | |
2230 | cpu, msr, tcc_activation_temp - dts, resolution); | |
2231 | ||
2232 | #ifdef THERM_DEBUG | |
2233 | if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr)) | |
2234 | return 0; | |
2235 | ||
2236 | dts = (msr >> 16) & 0x7F; | |
2237 | dts2 = (msr >> 8) & 0x7F; | |
2238 | fprintf(stderr, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n", | |
2239 | cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2); | |
2240 | #endif | |
2241 | } | |
2242 | ||
2243 | return 0; | |
2244 | } | |
2245 | ||
2246 | void print_power_limit_msr(int cpu, unsigned long long msr, char *label) | |
2247 | { | |
2248 | fprintf(stderr, "cpu%d: %s: %sabled (%f Watts, %f sec, clamp %sabled)\n", | |
2249 | cpu, label, | |
2250 | ((msr >> 15) & 1) ? "EN" : "DIS", | |
2251 | ((msr >> 0) & 0x7FFF) * rapl_power_units, | |
2252 | (1.0 + (((msr >> 22) & 0x3)/4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units, | |
2253 | (((msr >> 16) & 1) ? "EN" : "DIS")); | |
2254 | ||
2255 | return; | |
2256 | } | |
2257 | ||
2258 | int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
2259 | { | |
2260 | unsigned long long msr; | |
2261 | int cpu; | |
889facbe LB |
2262 | |
2263 | if (!do_rapl) | |
2264 | return 0; | |
2265 | ||
2266 | /* RAPL counters are per package, so print only for 1st thread/package */ | |
2267 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
2268 | return 0; | |
2269 | ||
2270 | cpu = t->cpu_id; | |
2271 | if (cpu_migrate(cpu)) { | |
2272 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
2273 | return -1; | |
2274 | } | |
2275 | ||
2276 | if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr)) | |
2277 | return -1; | |
2278 | ||
d8af6f5f | 2279 | if (debug) { |
889facbe LB |
2280 | fprintf(stderr, "cpu%d: MSR_RAPL_POWER_UNIT: 0x%08llx " |
2281 | "(%f Watts, %f Joules, %f sec.)\n", cpu, msr, | |
144b44b1 | 2282 | rapl_power_units, rapl_energy_units, rapl_time_units); |
889facbe | 2283 | } |
144b44b1 LB |
2284 | if (do_rapl & RAPL_PKG_POWER_INFO) { |
2285 | ||
889facbe LB |
2286 | if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr)) |
2287 | return -5; | |
2288 | ||
2289 | ||
2290 | fprintf(stderr, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n", | |
2291 | cpu, msr, | |
2292 | ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2293 | ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2294 | ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2295 | ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units); | |
2296 | ||
144b44b1 LB |
2297 | } |
2298 | if (do_rapl & RAPL_PKG) { | |
2299 | ||
889facbe LB |
2300 | if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr)) |
2301 | return -9; | |
2302 | ||
2303 | fprintf(stderr, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n", | |
2304 | cpu, msr, (msr >> 63) & 1 ? "": "UN"); | |
2305 | ||
2306 | print_power_limit_msr(cpu, msr, "PKG Limit #1"); | |
2307 | fprintf(stderr, "cpu%d: PKG Limit #2: %sabled (%f Watts, %f* sec, clamp %sabled)\n", | |
2308 | cpu, | |
2309 | ((msr >> 47) & 1) ? "EN" : "DIS", | |
2310 | ((msr >> 32) & 0x7FFF) * rapl_power_units, | |
2311 | (1.0 + (((msr >> 54) & 0x3)/4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units, | |
2312 | ((msr >> 48) & 1) ? "EN" : "DIS"); | |
2313 | } | |
2314 | ||
0b2bb692 | 2315 | if (do_rapl & RAPL_DRAM_POWER_INFO) { |
889facbe LB |
2316 | if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr)) |
2317 | return -6; | |
2318 | ||
889facbe LB |
2319 | fprintf(stderr, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n", |
2320 | cpu, msr, | |
2321 | ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2322 | ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2323 | ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2324 | ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units); | |
0b2bb692 LB |
2325 | } |
2326 | if (do_rapl & RAPL_DRAM) { | |
889facbe LB |
2327 | if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr)) |
2328 | return -9; | |
2329 | fprintf(stderr, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n", | |
2330 | cpu, msr, (msr >> 31) & 1 ? "": "UN"); | |
2331 | ||
2332 | print_power_limit_msr(cpu, msr, "DRAM Limit"); | |
2333 | } | |
144b44b1 | 2334 | if (do_rapl & RAPL_CORE_POLICY) { |
d8af6f5f | 2335 | if (debug) { |
889facbe LB |
2336 | if (get_msr(cpu, MSR_PP0_POLICY, &msr)) |
2337 | return -7; | |
2338 | ||
2339 | fprintf(stderr, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF); | |
144b44b1 LB |
2340 | } |
2341 | } | |
2342 | if (do_rapl & RAPL_CORES) { | |
d8af6f5f | 2343 | if (debug) { |
889facbe LB |
2344 | |
2345 | if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr)) | |
2346 | return -9; | |
2347 | fprintf(stderr, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n", | |
2348 | cpu, msr, (msr >> 31) & 1 ? "": "UN"); | |
2349 | print_power_limit_msr(cpu, msr, "Cores Limit"); | |
2350 | } | |
2351 | } | |
2352 | if (do_rapl & RAPL_GFX) { | |
d8af6f5f | 2353 | if (debug) { |
889facbe LB |
2354 | if (get_msr(cpu, MSR_PP1_POLICY, &msr)) |
2355 | return -8; | |
2356 | ||
2357 | fprintf(stderr, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF); | |
2358 | ||
2359 | if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr)) | |
2360 | return -9; | |
2361 | fprintf(stderr, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n", | |
2362 | cpu, msr, (msr >> 31) & 1 ? "": "UN"); | |
2363 | print_power_limit_msr(cpu, msr, "GFX Limit"); | |
2364 | } | |
2365 | } | |
2366 | return 0; | |
2367 | } | |
2368 | ||
d7899447 LB |
2369 | /* |
2370 | * SNB adds support for additional MSRs: | |
2371 | * | |
2372 | * MSR_PKG_C7_RESIDENCY 0x000003fa | |
2373 | * MSR_CORE_C7_RESIDENCY 0x000003fe | |
2374 | * MSR_PKG_C2_RESIDENCY 0x0000060d | |
2375 | */ | |
103a8fea | 2376 | |
d7899447 | 2377 | int has_snb_msrs(unsigned int family, unsigned int model) |
103a8fea LB |
2378 | { |
2379 | if (!genuine_intel) | |
2380 | return 0; | |
2381 | ||
2382 | switch (model) { | |
2383 | case 0x2A: | |
2384 | case 0x2D: | |
650a37f3 | 2385 | case 0x3A: /* IVB */ |
1300651b | 2386 | case 0x3E: /* IVB Xeon */ |
70b43400 LB |
2387 | case 0x3C: /* HSW */ |
2388 | case 0x3F: /* HSW */ | |
2389 | case 0x45: /* HSW */ | |
149c2319 | 2390 | case 0x46: /* HSW */ |
4e8e863f | 2391 | case 0x3D: /* BDW */ |
48a0631c | 2392 | case 0x47: /* BDW */ |
4e8e863f LB |
2393 | case 0x4F: /* BDX */ |
2394 | case 0x56: /* BDX-DE */ | |
0b2bb692 LB |
2395 | case 0x4E: /* SKL */ |
2396 | case 0x5E: /* SKL */ | |
103a8fea LB |
2397 | return 1; |
2398 | } | |
2399 | return 0; | |
2400 | } | |
2401 | ||
d7899447 LB |
2402 | /* |
2403 | * HSW adds support for additional MSRs: | |
2404 | * | |
2405 | * MSR_PKG_C8_RESIDENCY 0x00000630 | |
2406 | * MSR_PKG_C9_RESIDENCY 0x00000631 | |
2407 | * MSR_PKG_C10_RESIDENCY 0x00000632 | |
2408 | */ | |
2409 | int has_hsw_msrs(unsigned int family, unsigned int model) | |
ca58710f KCA |
2410 | { |
2411 | if (!genuine_intel) | |
2412 | return 0; | |
2413 | ||
2414 | switch (model) { | |
4e8e863f LB |
2415 | case 0x45: /* HSW */ |
2416 | case 0x3D: /* BDW */ | |
0b2bb692 LB |
2417 | case 0x4E: /* SKL */ |
2418 | case 0x5E: /* SKL */ | |
2419 | return 1; | |
2420 | } | |
2421 | return 0; | |
2422 | } | |
2423 | ||
2424 | /* | |
2425 | * SKL adds support for additional MSRS: | |
2426 | * | |
2427 | * MSR_PKG_WEIGHTED_CORE_C0_RES 0x00000658 | |
2428 | * MSR_PKG_ANY_CORE_C0_RES 0x00000659 | |
2429 | * MSR_PKG_ANY_GFXE_C0_RES 0x0000065A | |
2430 | * MSR_PKG_BOTH_CORE_GFXE_C0_RES 0x0000065B | |
2431 | */ | |
2432 | int has_skl_msrs(unsigned int family, unsigned int model) | |
2433 | { | |
2434 | if (!genuine_intel) | |
2435 | return 0; | |
2436 | ||
2437 | switch (model) { | |
2438 | case 0x4E: /* SKL */ | |
2439 | case 0x5E: /* SKL */ | |
ca58710f KCA |
2440 | return 1; |
2441 | } | |
2442 | return 0; | |
2443 | } | |
2444 | ||
2445 | ||
0b2bb692 | 2446 | |
144b44b1 LB |
2447 | int is_slm(unsigned int family, unsigned int model) |
2448 | { | |
2449 | if (!genuine_intel) | |
2450 | return 0; | |
2451 | switch (model) { | |
2452 | case 0x37: /* BYT */ | |
2453 | case 0x4D: /* AVN */ | |
2454 | return 1; | |
2455 | } | |
2456 | return 0; | |
2457 | } | |
2458 | ||
fb5d4327 DC |
2459 | int is_knl(unsigned int family, unsigned int model) |
2460 | { | |
2461 | if (!genuine_intel) | |
2462 | return 0; | |
2463 | switch (model) { | |
2464 | case 0x57: /* KNL */ | |
2465 | return 1; | |
2466 | } | |
2467 | return 0; | |
2468 | } | |
2469 | ||
144b44b1 LB |
2470 | #define SLM_BCLK_FREQS 5 |
2471 | double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0}; | |
2472 | ||
2473 | double slm_bclk(void) | |
2474 | { | |
2475 | unsigned long long msr = 3; | |
2476 | unsigned int i; | |
2477 | double freq; | |
2478 | ||
7ce7d5de | 2479 | if (get_msr(base_cpu, MSR_FSB_FREQ, &msr)) |
144b44b1 LB |
2480 | fprintf(stderr, "SLM BCLK: unknown\n"); |
2481 | ||
2482 | i = msr & 0xf; | |
2483 | if (i >= SLM_BCLK_FREQS) { | |
2484 | fprintf(stderr, "SLM BCLK[%d] invalid\n", i); | |
2485 | msr = 3; | |
2486 | } | |
2487 | freq = slm_freq_table[i]; | |
2488 | ||
2489 | fprintf(stderr, "SLM BCLK: %.1f Mhz\n", freq); | |
2490 | ||
2491 | return freq; | |
2492 | } | |
2493 | ||
103a8fea LB |
2494 | double discover_bclk(unsigned int family, unsigned int model) |
2495 | { | |
d7899447 | 2496 | if (has_snb_msrs(family, model)) |
103a8fea | 2497 | return 100.00; |
144b44b1 LB |
2498 | else if (is_slm(family, model)) |
2499 | return slm_bclk(); | |
103a8fea LB |
2500 | else |
2501 | return 133.33; | |
2502 | } | |
2503 | ||
889facbe LB |
2504 | /* |
2505 | * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where | |
2506 | * the Thermal Control Circuit (TCC) activates. | |
2507 | * This is usually equal to tjMax. | |
2508 | * | |
2509 | * Older processors do not have this MSR, so there we guess, | |
2510 | * but also allow cmdline over-ride with -T. | |
2511 | * | |
2512 | * Several MSR temperature values are in units of degrees-C | |
2513 | * below this value, including the Digital Thermal Sensor (DTS), | |
2514 | * Package Thermal Management Sensor (PTM), and thermal event thresholds. | |
2515 | */ | |
2516 | int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
2517 | { | |
2518 | unsigned long long msr; | |
2519 | unsigned int target_c_local; | |
2520 | int cpu; | |
2521 | ||
2522 | /* tcc_activation_temp is used only for dts or ptm */ | |
2523 | if (!(do_dts || do_ptm)) | |
2524 | return 0; | |
2525 | ||
2526 | /* this is a per-package concept */ | |
2527 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
2528 | return 0; | |
2529 | ||
2530 | cpu = t->cpu_id; | |
2531 | if (cpu_migrate(cpu)) { | |
2532 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
2533 | return -1; | |
2534 | } | |
2535 | ||
2536 | if (tcc_activation_temp_override != 0) { | |
2537 | tcc_activation_temp = tcc_activation_temp_override; | |
2538 | fprintf(stderr, "cpu%d: Using cmdline TCC Target (%d C)\n", | |
2539 | cpu, tcc_activation_temp); | |
2540 | return 0; | |
2541 | } | |
2542 | ||
2543 | /* Temperature Target MSR is Nehalem and newer only */ | |
d7899447 | 2544 | if (!do_nhm_platform_info) |
889facbe LB |
2545 | goto guess; |
2546 | ||
7ce7d5de | 2547 | if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr)) |
889facbe LB |
2548 | goto guess; |
2549 | ||
3482124a | 2550 | target_c_local = (msr >> 16) & 0xFF; |
889facbe | 2551 | |
d8af6f5f | 2552 | if (debug) |
889facbe LB |
2553 | fprintf(stderr, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n", |
2554 | cpu, msr, target_c_local); | |
2555 | ||
3482124a | 2556 | if (!target_c_local) |
889facbe LB |
2557 | goto guess; |
2558 | ||
2559 | tcc_activation_temp = target_c_local; | |
2560 | ||
2561 | return 0; | |
2562 | ||
2563 | guess: | |
2564 | tcc_activation_temp = TJMAX_DEFAULT; | |
2565 | fprintf(stderr, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n", | |
2566 | cpu, tcc_activation_temp); | |
2567 | ||
2568 | return 0; | |
2569 | } | |
fcd17211 | 2570 | void process_cpuid() |
103a8fea LB |
2571 | { |
2572 | unsigned int eax, ebx, ecx, edx, max_level; | |
2573 | unsigned int fms, family, model, stepping; | |
2574 | ||
2575 | eax = ebx = ecx = edx = 0; | |
2576 | ||
2b92865e | 2577 | __get_cpuid(0, &max_level, &ebx, &ecx, &edx); |
103a8fea LB |
2578 | |
2579 | if (ebx == 0x756e6547 && edx == 0x49656e69 && ecx == 0x6c65746e) | |
2580 | genuine_intel = 1; | |
2581 | ||
d8af6f5f | 2582 | if (debug) |
889facbe | 2583 | fprintf(stderr, "CPUID(0): %.4s%.4s%.4s ", |
103a8fea LB |
2584 | (char *)&ebx, (char *)&edx, (char *)&ecx); |
2585 | ||
2b92865e | 2586 | __get_cpuid(1, &fms, &ebx, &ecx, &edx); |
103a8fea LB |
2587 | family = (fms >> 8) & 0xf; |
2588 | model = (fms >> 4) & 0xf; | |
2589 | stepping = fms & 0xf; | |
2590 | if (family == 6 || family == 0xf) | |
2591 | model += ((fms >> 16) & 0xf) << 4; | |
2592 | ||
d8af6f5f | 2593 | if (debug) |
103a8fea LB |
2594 | fprintf(stderr, "%d CPUID levels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n", |
2595 | max_level, family, model, stepping, family, model, stepping); | |
2596 | ||
b2c95d90 JT |
2597 | if (!(edx & (1 << 5))) |
2598 | errx(1, "CPUID: no MSR"); | |
103a8fea LB |
2599 | |
2600 | /* | |
2601 | * check max extended function levels of CPUID. | |
2602 | * This is needed to check for invariant TSC. | |
2603 | * This check is valid for both Intel and AMD. | |
2604 | */ | |
2605 | ebx = ecx = edx = 0; | |
2b92865e | 2606 | __get_cpuid(0x80000000, &max_level, &ebx, &ecx, &edx); |
103a8fea | 2607 | |
d7899447 | 2608 | if (max_level >= 0x80000007) { |
103a8fea | 2609 | |
d7899447 LB |
2610 | /* |
2611 | * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8 | |
2612 | * this check is valid for both Intel and AMD | |
2613 | */ | |
2614 | __get_cpuid(0x80000007, &eax, &ebx, &ecx, &edx); | |
2615 | has_invariant_tsc = edx & (1 << 8); | |
2616 | } | |
103a8fea LB |
2617 | |
2618 | /* | |
2619 | * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0 | |
2620 | * this check is valid for both Intel and AMD | |
2621 | */ | |
2622 | ||
2b92865e | 2623 | __get_cpuid(0x6, &eax, &ebx, &ecx, &edx); |
8209e054 | 2624 | has_aperf = ecx & (1 << 0); |
889facbe LB |
2625 | do_dts = eax & (1 << 0); |
2626 | do_ptm = eax & (1 << 6); | |
2627 | has_epb = ecx & (1 << 3); | |
2628 | ||
d8af6f5f | 2629 | if (debug) |
a729617c LB |
2630 | fprintf(stderr, "CPUID(6): %sAPERF, %sDTS, %sPTM, %sEPB\n", |
2631 | has_aperf ? "" : "No ", | |
2632 | do_dts ? "" : "No ", | |
2633 | do_ptm ? "" : "No ", | |
2634 | has_epb ? "" : "No "); | |
103a8fea | 2635 | |
8a5bdf41 LB |
2636 | if (max_level > 0x15) { |
2637 | unsigned int eax_crystal; | |
2638 | unsigned int ebx_tsc; | |
2639 | ||
2640 | /* | |
2641 | * CPUID 15H TSC/Crystal ratio, possibly Crystal Hz | |
2642 | */ | |
2643 | eax_crystal = ebx_tsc = crystal_hz = edx = 0; | |
2644 | __get_cpuid(0x15, &eax_crystal, &ebx_tsc, &crystal_hz, &edx); | |
2645 | ||
2646 | if (ebx_tsc != 0) { | |
2647 | ||
2648 | if (debug && (ebx != 0)) | |
2649 | fprintf(stderr, "CPUID(0x15): eax_crystal: %d ebx_tsc: %d ecx_crystal_hz: %d\n", | |
2650 | eax_crystal, ebx_tsc, crystal_hz); | |
2651 | ||
2652 | if (crystal_hz == 0) | |
2653 | switch(model) { | |
2654 | case 0x4E: /* SKL */ | |
2655 | case 0x5E: /* SKL */ | |
2656 | crystal_hz = 24000000; /* 24 MHz */ | |
2657 | break; | |
2658 | default: | |
2659 | crystal_hz = 0; | |
2660 | } | |
2661 | ||
2662 | if (crystal_hz) { | |
2663 | tsc_hz = (unsigned long long) crystal_hz * ebx_tsc / eax_crystal; | |
2664 | if (debug) | |
2665 | fprintf(stderr, "TSC: %lld MHz (%d Hz * %d / %d / 1000000)\n", | |
2666 | tsc_hz / 1000000, crystal_hz, ebx_tsc, eax_crystal); | |
2667 | } | |
2668 | } | |
2669 | } | |
2670 | ||
ee7e38e3 | 2671 | do_nhm_platform_info = do_nhm_cstates = do_smi = probe_nhm_msrs(family, model); |
d7899447 | 2672 | do_snb_cstates = has_snb_msrs(family, model); |
ee7e38e3 LB |
2673 | do_pc2 = do_snb_cstates && (pkg_cstate_limit >= PCL__2); |
2674 | do_pc3 = (pkg_cstate_limit >= PCL__3); | |
2675 | do_pc6 = (pkg_cstate_limit >= PCL__6); | |
2676 | do_pc7 = do_snb_cstates && (pkg_cstate_limit >= PCL__7); | |
d7899447 | 2677 | do_c8_c9_c10 = has_hsw_msrs(family, model); |
0b2bb692 | 2678 | do_skl_residency = has_skl_msrs(family, model); |
144b44b1 | 2679 | do_slm_cstates = is_slm(family, model); |
fb5d4327 | 2680 | do_knl_cstates = is_knl(family, model); |
103a8fea LB |
2681 | bclk = discover_bclk(family, model); |
2682 | ||
889facbe | 2683 | rapl_probe(family, model); |
3a9a941d | 2684 | perf_limit_reasons_probe(family, model); |
889facbe | 2685 | |
fcd17211 LB |
2686 | if (debug) |
2687 | dump_cstate_pstate_config_info(); | |
2688 | ||
889facbe | 2689 | return; |
103a8fea LB |
2690 | } |
2691 | ||
d8af6f5f | 2692 | void help() |
103a8fea | 2693 | { |
d8af6f5f LB |
2694 | fprintf(stderr, |
2695 | "Usage: turbostat [OPTIONS][(--interval seconds) | COMMAND ...]\n" | |
2696 | "\n" | |
2697 | "Turbostat forks the specified COMMAND and prints statistics\n" | |
2698 | "when COMMAND completes.\n" | |
2699 | "If no COMMAND is specified, turbostat wakes every 5-seconds\n" | |
2700 | "to print statistics, until interrupted.\n" | |
2701 | "--debug run in \"debug\" mode\n" | |
2702 | "--interval sec Override default 5-second measurement interval\n" | |
2703 | "--help print this help message\n" | |
2704 | "--counter msr print 32-bit counter at address \"msr\"\n" | |
2705 | "--Counter msr print 64-bit Counter at address \"msr\"\n" | |
2706 | "--msr msr print 32-bit value at address \"msr\"\n" | |
2707 | "--MSR msr print 64-bit Value at address \"msr\"\n" | |
2708 | "--version print version information\n" | |
2709 | "\n" | |
2710 | "For more help, run \"man turbostat\"\n"); | |
103a8fea LB |
2711 | } |
2712 | ||
2713 | ||
2714 | /* | |
2715 | * in /dev/cpu/ return success for names that are numbers | |
2716 | * ie. filter out ".", "..", "microcode". | |
2717 | */ | |
2718 | int dir_filter(const struct dirent *dirp) | |
2719 | { | |
2720 | if (isdigit(dirp->d_name[0])) | |
2721 | return 1; | |
2722 | else | |
2723 | return 0; | |
2724 | } | |
2725 | ||
2726 | int open_dev_cpu_msr(int dummy1) | |
2727 | { | |
2728 | return 0; | |
2729 | } | |
2730 | ||
c98d5d94 LB |
2731 | void topology_probe() |
2732 | { | |
2733 | int i; | |
2734 | int max_core_id = 0; | |
2735 | int max_package_id = 0; | |
2736 | int max_siblings = 0; | |
2737 | struct cpu_topology { | |
2738 | int core_id; | |
2739 | int physical_package_id; | |
2740 | } *cpus; | |
2741 | ||
2742 | /* Initialize num_cpus, max_cpu_num */ | |
2743 | topo.num_cpus = 0; | |
2744 | topo.max_cpu_num = 0; | |
2745 | for_all_proc_cpus(count_cpus); | |
2746 | if (!summary_only && topo.num_cpus > 1) | |
2747 | show_cpu = 1; | |
2748 | ||
d8af6f5f | 2749 | if (debug > 1) |
c98d5d94 LB |
2750 | fprintf(stderr, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num); |
2751 | ||
2752 | cpus = calloc(1, (topo.max_cpu_num + 1) * sizeof(struct cpu_topology)); | |
b2c95d90 JT |
2753 | if (cpus == NULL) |
2754 | err(1, "calloc cpus"); | |
c98d5d94 LB |
2755 | |
2756 | /* | |
2757 | * Allocate and initialize cpu_present_set | |
2758 | */ | |
2759 | cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1)); | |
b2c95d90 JT |
2760 | if (cpu_present_set == NULL) |
2761 | err(3, "CPU_ALLOC"); | |
c98d5d94 LB |
2762 | cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1)); |
2763 | CPU_ZERO_S(cpu_present_setsize, cpu_present_set); | |
2764 | for_all_proc_cpus(mark_cpu_present); | |
2765 | ||
2766 | /* | |
2767 | * Allocate and initialize cpu_affinity_set | |
2768 | */ | |
2769 | cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1)); | |
b2c95d90 JT |
2770 | if (cpu_affinity_set == NULL) |
2771 | err(3, "CPU_ALLOC"); | |
c98d5d94 LB |
2772 | cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1)); |
2773 | CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set); | |
2774 | ||
2775 | ||
2776 | /* | |
2777 | * For online cpus | |
2778 | * find max_core_id, max_package_id | |
2779 | */ | |
2780 | for (i = 0; i <= topo.max_cpu_num; ++i) { | |
2781 | int siblings; | |
2782 | ||
2783 | if (cpu_is_not_present(i)) { | |
d8af6f5f | 2784 | if (debug > 1) |
c98d5d94 LB |
2785 | fprintf(stderr, "cpu%d NOT PRESENT\n", i); |
2786 | continue; | |
2787 | } | |
2788 | cpus[i].core_id = get_core_id(i); | |
2789 | if (cpus[i].core_id > max_core_id) | |
2790 | max_core_id = cpus[i].core_id; | |
2791 | ||
2792 | cpus[i].physical_package_id = get_physical_package_id(i); | |
2793 | if (cpus[i].physical_package_id > max_package_id) | |
2794 | max_package_id = cpus[i].physical_package_id; | |
2795 | ||
2796 | siblings = get_num_ht_siblings(i); | |
2797 | if (siblings > max_siblings) | |
2798 | max_siblings = siblings; | |
d8af6f5f | 2799 | if (debug > 1) |
c98d5d94 LB |
2800 | fprintf(stderr, "cpu %d pkg %d core %d\n", |
2801 | i, cpus[i].physical_package_id, cpus[i].core_id); | |
2802 | } | |
2803 | topo.num_cores_per_pkg = max_core_id + 1; | |
d8af6f5f | 2804 | if (debug > 1) |
c98d5d94 LB |
2805 | fprintf(stderr, "max_core_id %d, sizing for %d cores per package\n", |
2806 | max_core_id, topo.num_cores_per_pkg); | |
1cc21f7b | 2807 | if (debug && !summary_only && topo.num_cores_per_pkg > 1) |
c98d5d94 LB |
2808 | show_core = 1; |
2809 | ||
2810 | topo.num_packages = max_package_id + 1; | |
d8af6f5f | 2811 | if (debug > 1) |
c98d5d94 LB |
2812 | fprintf(stderr, "max_package_id %d, sizing for %d packages\n", |
2813 | max_package_id, topo.num_packages); | |
1cc21f7b | 2814 | if (debug && !summary_only && topo.num_packages > 1) |
c98d5d94 LB |
2815 | show_pkg = 1; |
2816 | ||
2817 | topo.num_threads_per_core = max_siblings; | |
d8af6f5f | 2818 | if (debug > 1) |
c98d5d94 LB |
2819 | fprintf(stderr, "max_siblings %d\n", max_siblings); |
2820 | ||
2821 | free(cpus); | |
2822 | } | |
2823 | ||
2824 | void | |
2825 | allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p) | |
2826 | { | |
2827 | int i; | |
2828 | ||
2829 | *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg * | |
2830 | topo.num_packages, sizeof(struct thread_data)); | |
2831 | if (*t == NULL) | |
2832 | goto error; | |
2833 | ||
2834 | for (i = 0; i < topo.num_threads_per_core * | |
2835 | topo.num_cores_per_pkg * topo.num_packages; i++) | |
2836 | (*t)[i].cpu_id = -1; | |
2837 | ||
2838 | *c = calloc(topo.num_cores_per_pkg * topo.num_packages, | |
2839 | sizeof(struct core_data)); | |
2840 | if (*c == NULL) | |
2841 | goto error; | |
2842 | ||
2843 | for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++) | |
2844 | (*c)[i].core_id = -1; | |
2845 | ||
2846 | *p = calloc(topo.num_packages, sizeof(struct pkg_data)); | |
2847 | if (*p == NULL) | |
2848 | goto error; | |
2849 | ||
2850 | for (i = 0; i < topo.num_packages; i++) | |
2851 | (*p)[i].package_id = i; | |
2852 | ||
2853 | return; | |
2854 | error: | |
b2c95d90 | 2855 | err(1, "calloc counters"); |
c98d5d94 LB |
2856 | } |
2857 | /* | |
2858 | * init_counter() | |
2859 | * | |
2860 | * set cpu_id, core_num, pkg_num | |
2861 | * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE | |
2862 | * | |
2863 | * increment topo.num_cores when 1st core in pkg seen | |
2864 | */ | |
2865 | void init_counter(struct thread_data *thread_base, struct core_data *core_base, | |
2866 | struct pkg_data *pkg_base, int thread_num, int core_num, | |
2867 | int pkg_num, int cpu_id) | |
2868 | { | |
2869 | struct thread_data *t; | |
2870 | struct core_data *c; | |
2871 | struct pkg_data *p; | |
2872 | ||
2873 | t = GET_THREAD(thread_base, thread_num, core_num, pkg_num); | |
2874 | c = GET_CORE(core_base, core_num, pkg_num); | |
2875 | p = GET_PKG(pkg_base, pkg_num); | |
2876 | ||
2877 | t->cpu_id = cpu_id; | |
2878 | if (thread_num == 0) { | |
2879 | t->flags |= CPU_IS_FIRST_THREAD_IN_CORE; | |
2880 | if (cpu_is_first_core_in_package(cpu_id)) | |
2881 | t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE; | |
2882 | } | |
2883 | ||
2884 | c->core_id = core_num; | |
2885 | p->package_id = pkg_num; | |
2886 | } | |
2887 | ||
2888 | ||
2889 | int initialize_counters(int cpu_id) | |
2890 | { | |
2891 | int my_thread_id, my_core_id, my_package_id; | |
2892 | ||
2893 | my_package_id = get_physical_package_id(cpu_id); | |
2894 | my_core_id = get_core_id(cpu_id); | |
e275b388 DC |
2895 | my_thread_id = get_cpu_position_in_core(cpu_id); |
2896 | if (!my_thread_id) | |
c98d5d94 | 2897 | topo.num_cores++; |
c98d5d94 LB |
2898 | |
2899 | init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id); | |
2900 | init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id); | |
2901 | return 0; | |
2902 | } | |
2903 | ||
2904 | void allocate_output_buffer() | |
2905 | { | |
3b4d5c7f | 2906 | output_buffer = calloc(1, (1 + topo.num_cpus) * 1024); |
c98d5d94 | 2907 | outp = output_buffer; |
b2c95d90 JT |
2908 | if (outp == NULL) |
2909 | err(-1, "calloc output buffer"); | |
c98d5d94 LB |
2910 | } |
2911 | ||
2912 | void setup_all_buffers(void) | |
2913 | { | |
2914 | topology_probe(); | |
2915 | allocate_counters(&thread_even, &core_even, &package_even); | |
2916 | allocate_counters(&thread_odd, &core_odd, &package_odd); | |
2917 | allocate_output_buffer(); | |
2918 | for_all_proc_cpus(initialize_counters); | |
2919 | } | |
3b4d5c7f | 2920 | |
7ce7d5de PB |
2921 | void set_base_cpu(void) |
2922 | { | |
2923 | base_cpu = sched_getcpu(); | |
2924 | if (base_cpu < 0) | |
2925 | err(-ENODEV, "No valid cpus found"); | |
2926 | ||
2927 | if (debug > 1) | |
2928 | fprintf(stderr, "base_cpu = %d\n", base_cpu); | |
2929 | } | |
2930 | ||
103a8fea LB |
2931 | void turbostat_init() |
2932 | { | |
7ce7d5de PB |
2933 | setup_all_buffers(); |
2934 | set_base_cpu(); | |
103a8fea | 2935 | check_dev_msr(); |
98481e79 | 2936 | check_permissions(); |
fcd17211 | 2937 | process_cpuid(); |
103a8fea | 2938 | |
103a8fea | 2939 | |
d8af6f5f | 2940 | if (debug) |
889facbe LB |
2941 | for_all_cpus(print_epb, ODD_COUNTERS); |
2942 | ||
d8af6f5f | 2943 | if (debug) |
3a9a941d LB |
2944 | for_all_cpus(print_perf_limit, ODD_COUNTERS); |
2945 | ||
d8af6f5f | 2946 | if (debug) |
889facbe LB |
2947 | for_all_cpus(print_rapl, ODD_COUNTERS); |
2948 | ||
2949 | for_all_cpus(set_temperature_target, ODD_COUNTERS); | |
2950 | ||
d8af6f5f | 2951 | if (debug) |
889facbe | 2952 | for_all_cpus(print_thermal, ODD_COUNTERS); |
103a8fea LB |
2953 | } |
2954 | ||
2955 | int fork_it(char **argv) | |
2956 | { | |
103a8fea | 2957 | pid_t child_pid; |
d91bb17c | 2958 | int status; |
d15cf7c1 | 2959 | |
d91bb17c LB |
2960 | status = for_all_cpus(get_counters, EVEN_COUNTERS); |
2961 | if (status) | |
2962 | exit(status); | |
c98d5d94 LB |
2963 | /* clear affinity side-effect of get_counters() */ |
2964 | sched_setaffinity(0, cpu_present_setsize, cpu_present_set); | |
103a8fea LB |
2965 | gettimeofday(&tv_even, (struct timezone *)NULL); |
2966 | ||
2967 | child_pid = fork(); | |
2968 | if (!child_pid) { | |
2969 | /* child */ | |
2970 | execvp(argv[0], argv); | |
2971 | } else { | |
103a8fea LB |
2972 | |
2973 | /* parent */ | |
b2c95d90 JT |
2974 | if (child_pid == -1) |
2975 | err(1, "fork"); | |
103a8fea LB |
2976 | |
2977 | signal(SIGINT, SIG_IGN); | |
2978 | signal(SIGQUIT, SIG_IGN); | |
b2c95d90 JT |
2979 | if (waitpid(child_pid, &status, 0) == -1) |
2980 | err(status, "waitpid"); | |
103a8fea | 2981 | } |
c98d5d94 LB |
2982 | /* |
2983 | * n.b. fork_it() does not check for errors from for_all_cpus() | |
2984 | * because re-starting is problematic when forking | |
2985 | */ | |
2986 | for_all_cpus(get_counters, ODD_COUNTERS); | |
103a8fea | 2987 | gettimeofday(&tv_odd, (struct timezone *)NULL); |
103a8fea | 2988 | timersub(&tv_odd, &tv_even, &tv_delta); |
c98d5d94 LB |
2989 | for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS); |
2990 | compute_average(EVEN_COUNTERS); | |
2991 | format_all_counters(EVEN_COUNTERS); | |
2992 | flush_stderr(); | |
103a8fea | 2993 | |
6eab04a8 | 2994 | fprintf(stderr, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0); |
103a8fea | 2995 | |
d91bb17c | 2996 | return status; |
103a8fea LB |
2997 | } |
2998 | ||
3b4d5c7f AS |
2999 | int get_and_dump_counters(void) |
3000 | { | |
3001 | int status; | |
3002 | ||
3003 | status = for_all_cpus(get_counters, ODD_COUNTERS); | |
3004 | if (status) | |
3005 | return status; | |
3006 | ||
3007 | status = for_all_cpus(dump_counters, ODD_COUNTERS); | |
3008 | if (status) | |
3009 | return status; | |
3010 | ||
3011 | flush_stdout(); | |
3012 | ||
3013 | return status; | |
3014 | } | |
3015 | ||
d8af6f5f | 3016 | void print_version() { |
a68c7c3f | 3017 | fprintf(stderr, "turbostat version 4.7 27-May, 2015" |
d8af6f5f LB |
3018 | " - Len Brown <lenb@kernel.org>\n"); |
3019 | } | |
3020 | ||
103a8fea LB |
3021 | void cmdline(int argc, char **argv) |
3022 | { | |
3023 | int opt; | |
d8af6f5f LB |
3024 | int option_index = 0; |
3025 | static struct option long_options[] = { | |
3026 | {"Counter", required_argument, 0, 'C'}, | |
3027 | {"counter", required_argument, 0, 'c'}, | |
3028 | {"Dump", no_argument, 0, 'D'}, | |
3029 | {"debug", no_argument, 0, 'd'}, | |
3030 | {"interval", required_argument, 0, 'i'}, | |
3031 | {"help", no_argument, 0, 'h'}, | |
3032 | {"Joules", no_argument, 0, 'J'}, | |
3033 | {"MSR", required_argument, 0, 'M'}, | |
3034 | {"msr", required_argument, 0, 'm'}, | |
3035 | {"Package", no_argument, 0, 'p'}, | |
3036 | {"processor", no_argument, 0, 'p'}, | |
3037 | {"Summary", no_argument, 0, 'S'}, | |
3038 | {"TCC", required_argument, 0, 'T'}, | |
3039 | {"version", no_argument, 0, 'v' }, | |
3040 | {0, 0, 0, 0 } | |
3041 | }; | |
103a8fea LB |
3042 | |
3043 | progname = argv[0]; | |
3044 | ||
d8af6f5f LB |
3045 | while ((opt = getopt_long_only(argc, argv, "C:c:Ddhi:JM:m:PpST:v", |
3046 | long_options, &option_index)) != -1) { | |
103a8fea | 3047 | switch (opt) { |
d8af6f5f LB |
3048 | case 'C': |
3049 | sscanf(optarg, "%x", &extra_delta_offset64); | |
c98d5d94 | 3050 | break; |
d8af6f5f LB |
3051 | case 'c': |
3052 | sscanf(optarg, "%x", &extra_delta_offset32); | |
c98d5d94 | 3053 | break; |
d8af6f5f | 3054 | case 'D': |
3b4d5c7f AS |
3055 | dump_only++; |
3056 | break; | |
d8af6f5f LB |
3057 | case 'd': |
3058 | debug++; | |
103a8fea | 3059 | break; |
d8af6f5f LB |
3060 | case 'h': |
3061 | default: | |
3062 | help(); | |
3063 | exit(1); | |
103a8fea LB |
3064 | case 'i': |
3065 | interval_sec = atoi(optarg); | |
3066 | break; | |
d8af6f5f LB |
3067 | case 'J': |
3068 | rapl_joules++; | |
8e180f3c | 3069 | break; |
d8af6f5f LB |
3070 | case 'M': |
3071 | sscanf(optarg, "%x", &extra_msr_offset64); | |
8e180f3c | 3072 | break; |
2f32edf1 LB |
3073 | case 'm': |
3074 | sscanf(optarg, "%x", &extra_msr_offset32); | |
2f32edf1 | 3075 | break; |
d8af6f5f LB |
3076 | case 'P': |
3077 | show_pkg_only++; | |
3078 | break; | |
3079 | case 'p': | |
3080 | show_core_only++; | |
103a8fea | 3081 | break; |
d8af6f5f LB |
3082 | case 'S': |
3083 | summary_only++; | |
889facbe LB |
3084 | break; |
3085 | case 'T': | |
3086 | tcc_activation_temp_override = atoi(optarg); | |
3087 | break; | |
d8af6f5f LB |
3088 | case 'v': |
3089 | print_version(); | |
3090 | exit(0); | |
5c56be9a | 3091 | break; |
103a8fea LB |
3092 | } |
3093 | } | |
3094 | } | |
3095 | ||
3096 | int main(int argc, char **argv) | |
3097 | { | |
3098 | cmdline(argc, argv); | |
3099 | ||
d8af6f5f LB |
3100 | if (debug) |
3101 | print_version(); | |
103a8fea LB |
3102 | |
3103 | turbostat_init(); | |
3104 | ||
3b4d5c7f AS |
3105 | /* dump counters and exit */ |
3106 | if (dump_only) | |
3107 | return get_and_dump_counters(); | |
3108 | ||
103a8fea LB |
3109 | /* |
3110 | * if any params left, it must be a command to fork | |
3111 | */ | |
3112 | if (argc - optind) | |
3113 | return fork_it(argv + optind); | |
3114 | else | |
3115 | turbostat_loop(); | |
3116 | ||
3117 | return 0; | |
3118 | } |