aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--tools/power/x86/turbostat/Makefile1
-rw-r--r--tools/power/x86/turbostat/turbostat.877
-rw-r--r--tools/power/x86/turbostat/turbostat.c1329
3 files changed, 868 insertions, 539 deletions
diff --git a/tools/power/x86/turbostat/Makefile b/tools/power/x86/turbostat/Makefile
index fd8e1f1297aa..f85649554191 100644
--- a/tools/power/x86/turbostat/Makefile
+++ b/tools/power/x86/turbostat/Makefile
@@ -1,4 +1,5 @@
1turbostat : turbostat.c 1turbostat : turbostat.c
2CFLAGS += -Wall
2 3
3clean : 4clean :
4 rm -f turbostat 5 rm -f turbostat
diff --git a/tools/power/x86/turbostat/turbostat.8 b/tools/power/x86/turbostat/turbostat.8
index adf175f61496..74e44507dfe9 100644
--- a/tools/power/x86/turbostat/turbostat.8
+++ b/tools/power/x86/turbostat/turbostat.8
@@ -27,7 +27,11 @@ supports an "invariant" TSC, plus the APERF and MPERF MSRs.
27on processors that additionally support C-state residency counters. 27on processors that additionally support C-state residency counters.
28 28
29.SS Options 29.SS Options
30The \fB-s\fP option prints only a 1-line summary for each sample interval. 30The \fB-s\fP option limits output to a 1-line system summary for each interval.
31.PP
32The \fB-c\fP option limits output to the 1st thread in each core.
33.PP
34The \fB-p\fP option limits output to the 1st thread in each package.
31.PP 35.PP
32The \fB-v\fP option increases verbosity. 36The \fB-v\fP option increases verbosity.
33.PP 37.PP
@@ -65,19 +69,19 @@ Subsequent rows show per-CPU statistics.
65.nf 69.nf
66[root@x980]# ./turbostat 70[root@x980]# ./turbostat
67cor CPU %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6 71cor CPU %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6
68 0.60 1.63 3.38 2.91 0.00 96.49 0.00 76.64 72 0.09 1.62 3.38 1.83 0.32 97.76 1.26 83.61
69 0 0 0.59 1.62 3.38 4.51 0.00 94.90 0.00 76.64 73 0 0 0.15 1.62 3.38 10.23 0.05 89.56 1.26 83.61
70 0 6 1.13 1.64 3.38 3.97 0.00 94.90 0.00 76.64 74 0 6 0.05 1.62 3.38 10.34
71 1 2 0.08 1.62 3.38 0.07 0.00 99.85 0.00 76.64 75 1 2 0.03 1.62 3.38 0.07 0.05 99.86
72 1 8 0.03 1.62 3.38 0.12 0.00 99.85 0.00 76.64 76 1 8 0.03 1.62 3.38 0.06
73 2 4 0.01 1.62 3.38 0.06 0.00 99.93 0.00 76.64 77 2 4 0.21 1.62 3.38 0.10 1.49 98.21
74 2 10 0.04 1.62 3.38 0.02 0.00 99.93 0.00 76.64 78 2 10 0.02 1.62 3.38 0.29
75 8 1 2.85 1.62 3.38 11.71 0.00 85.44 0.00 76.64 79 8 1 0.04 1.62 3.38 0.04 0.08 99.84
76 8 7 1.98 1.62 3.38 12.58 0.00 85.44 0.00 76.64 80 8 7 0.01 1.62 3.38 0.06
77 9 3 0.36 1.62 3.38 0.71 0.00 98.93 0.00 76.64 81 9 3 0.53 1.62 3.38 0.10 0.20 99.17
78 9 9 0.09 1.62 3.38 0.98 0.00 98.93 0.00 76.64 82 9 9 0.02 1.62 3.38 0.60
79 10 5 0.03 1.62 3.38 0.09 0.00 99.87 0.00 76.64 83 10 5 0.01 1.62 3.38 0.02 0.04 99.92
80 10 11 0.07 1.62 3.38 0.06 0.00 99.87 0.00 76.64 84 10 11 0.02 1.62 3.38 0.02
81.fi 85.fi
82.SH SUMMARY EXAMPLE 86.SH SUMMARY EXAMPLE
83The "-s" option prints the column headers just once, 87The "-s" option prints the column headers just once,
@@ -86,9 +90,10 @@ and then the one line system summary for each sample interval.
86.nf 90.nf
87[root@x980]# ./turbostat -s 91[root@x980]# ./turbostat -s
88 %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6 92 %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6
89 0.61 1.89 3.38 5.95 0.00 93.44 0.00 66.33 93 0.23 1.67 3.38 2.00 0.30 97.47 1.07 82.12
90 0.52 1.62 3.38 6.83 0.00 92.65 0.00 61.11 94 0.10 1.62 3.38 1.87 2.25 95.77 12.02 72.60
91 0.62 1.92 3.38 5.47 0.00 93.91 0.00 67.31 95 0.20 1.64 3.38 1.98 0.11 97.72 0.30 83.36
96 0.11 1.70 3.38 1.86 1.81 96.22 9.71 74.90
92.fi 97.fi
93.SH VERBOSE EXAMPLE 98.SH VERBOSE EXAMPLE
94The "-v" option adds verbosity to the output: 99The "-v" option adds verbosity to the output:
@@ -120,30 +125,28 @@ until ^C while the other CPUs are mostly idle:
120[root@x980 lenb]# ./turbostat cat /dev/zero > /dev/null 125[root@x980 lenb]# ./turbostat cat /dev/zero > /dev/null
121^C 126^C
122cor CPU %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6 127cor CPU %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6
123 8.63 3.64 3.38 14.46 0.49 76.42 0.00 0.00 128 8.86 3.61 3.38 15.06 31.19 44.89 0.00 0.00
124 0 0 0.34 3.36 3.38 99.66 0.00 0.00 0.00 0.00 129 0 0 1.46 3.22 3.38 16.84 29.48 52.22 0.00 0.00
125 0 6 99.96 3.64 3.38 0.04 0.00 0.00 0.00 0.00 130 0 6 0.21 3.06 3.38 18.09
126 1 2 0.14 3.50 3.38 1.75 2.04 96.07 0.00 0.00 131 1 2 0.53 3.33 3.38 2.80 46.40 50.27
127 1 8 0.38 3.57 3.38 1.51 2.04 96.07 0.00 0.00 132 1 8 0.89 3.47 3.38 2.44
128 2 4 0.01 2.65 3.38 0.06 0.00 99.93 0.00 0.00 133 2 4 1.36 3.43 3.38 9.04 23.71 65.89
129 2 10 0.03 2.12 3.38 0.04 0.00 99.93 0.00 0.00 134 2 10 0.18 2.86 3.38 10.22
130 8 1 0.91 3.59 3.38 35.27 0.92 62.90 0.00 0.00 135 8 1 0.04 2.87 3.38 99.96 0.01 0.00
131 8 7 1.61 3.63 3.38 34.57 0.92 62.90 0.00 0.00 136 8 7 99.72 3.63 3.38 0.27
132 9 3 0.04 3.38 3.38 0.20 0.00 99.76 0.00 0.00 137 9 3 0.31 3.21 3.38 7.64 56.55 35.50
133 9 9 0.04 3.29 3.38 0.20 0.00 99.76 0.00 0.00 138 9 9 0.08 2.95 3.38 7.88
134 10 5 0.03 3.08 3.38 0.12 0.00 99.85 0.00 0.00 139 10 5 1.42 3.43 3.38 2.14 30.99 65.44
135 10 11 0.05 3.07 3.38 0.10 0.00 99.85 0.00 0.00 140 10 11 0.16 2.88 3.38 3.40
1364.907015 sec
137
138.fi 141.fi
139Above the cycle soaker drives cpu6 up 3.6 Ghz turbo limit 142Above the cycle soaker drives cpu7 up its 3.6 Ghz turbo limit
140while the other processors are generally in various states of idle. 143while the other processors are generally in various states of idle.
141 144
142Note that cpu0 is an HT sibling sharing core0 145Note that cpu1 and cpu7 are HT siblings within core8.
143with cpu6, and thus it is unable to get to an idle state 146As cpu7 is very busy, it prevents its sibling, cpu1,
144deeper than c1 while cpu6 is busy. 147from entering a c-state deeper than c1.
145 148
146Note that turbostat reports average GHz of 3.64, while 149Note that turbostat reports average GHz of 3.63, while
147the arithmetic average of the GHz column above is lower. 150the arithmetic average of the GHz column above is lower.
148This is a weighted average, where the weight is %c0. ie. it is the total number of 151This is a weighted average, where the weight is %c0. ie. it is the total number of
149un-halted cycles elapsed per time divided by the number of CPUs. 152un-halted cycles elapsed per time divided by the number of CPUs.
diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c
index 16de7ad4850f..b815a12159b2 100644
--- a/tools/power/x86/turbostat/turbostat.c
+++ b/tools/power/x86/turbostat/turbostat.c
@@ -67,92 +67,119 @@ double bclk;
67unsigned int show_pkg; 67unsigned int show_pkg;
68unsigned int show_core; 68unsigned int show_core;
69unsigned int show_cpu; 69unsigned int show_cpu;
70unsigned int show_pkg_only;
71unsigned int show_core_only;
72char *output_buffer, *outp;
70 73
71int aperf_mperf_unstable; 74int aperf_mperf_unstable;
72int backwards_count; 75int backwards_count;
73char *progname; 76char *progname;
74 77
75int num_cpus; 78cpu_set_t *cpu_present_set, *cpu_affinity_set;
76cpu_set_t *cpu_present_set, *cpu_mask; 79size_t cpu_present_setsize, cpu_affinity_setsize;
77size_t cpu_present_setsize, cpu_mask_size; 80
78 81struct thread_data {
79struct counters { 82 unsigned long long tsc;
80 unsigned long long tsc; /* per thread */ 83 unsigned long long aperf;
81 unsigned long long aperf; /* per thread */ 84 unsigned long long mperf;
82 unsigned long long mperf; /* per thread */ 85 unsigned long long c1; /* derived */
83 unsigned long long c1; /* per thread (calculated) */ 86 unsigned long long extra_msr;
84 unsigned long long c3; /* per core */ 87 unsigned int cpu_id;
85 unsigned long long c6; /* per core */ 88 unsigned int flags;
86 unsigned long long c7; /* per core */ 89#define CPU_IS_FIRST_THREAD_IN_CORE 0x2
87 unsigned long long pc2; /* per package */ 90#define CPU_IS_FIRST_CORE_IN_PACKAGE 0x4
88 unsigned long long pc3; /* per package */ 91} *thread_even, *thread_odd;
89 unsigned long long pc6; /* per package */ 92
90 unsigned long long pc7; /* per package */ 93struct core_data {
91 unsigned long long extra_msr; /* per thread */ 94 unsigned long long c3;
92 int pkg; 95 unsigned long long c6;
93 int core; 96 unsigned long long c7;
94 int cpu; 97 unsigned int core_id;
95 struct counters *next; 98} *core_even, *core_odd;
96}; 99
97 100struct pkg_data {
98struct counters *cnt_even; 101 unsigned long long pc2;
99struct counters *cnt_odd; 102 unsigned long long pc3;
100struct counters *cnt_delta; 103 unsigned long long pc6;
101struct counters *cnt_average; 104 unsigned long long pc7;
102struct timeval tv_even; 105 unsigned int package_id;
103struct timeval tv_odd; 106} *package_even, *package_odd;
104struct timeval tv_delta; 107
105 108#define ODD_COUNTERS thread_odd, core_odd, package_odd
106int mark_cpu_present(int pkg, int core, int cpu) 109#define EVEN_COUNTERS thread_even, core_even, package_even
110
111#define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \
112 (thread_base + (pkg_no) * topo.num_cores_per_pkg * \
113 topo.num_threads_per_core + \
114 (core_no) * topo.num_threads_per_core + (thread_no))
115#define GET_CORE(core_base, core_no, pkg_no) \
116 (core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no))
117#define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
118
119struct system_summary {
120 struct thread_data threads;
121 struct core_data cores;
122 struct pkg_data packages;
123} sum, average;
124
125
126struct topo_params {
127 int num_packages;
128 int num_cpus;
129 int num_cores;
130 int max_cpu_num;
131 int num_cores_per_pkg;
132 int num_threads_per_core;
133} topo;
134
135struct timeval tv_even, tv_odd, tv_delta;
136
137void setup_all_buffers(void);
138
139int cpu_is_not_present(int cpu)
107{ 140{
108 CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set); 141 return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
109 return 0;
110} 142}
111
112/* 143/*
113 * cpu_mask_init(ncpus) 144 * run func(thread, core, package) in topology order
114 * 145 * skip non-present cpus
115 * allocate and clear cpu_mask
116 * set cpu_mask_size
117 */ 146 */
118void cpu_mask_init(int ncpus) 147
148int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
149 struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
119{ 150{
120 cpu_mask = CPU_ALLOC(ncpus); 151 int retval, pkg_no, core_no, thread_no;
121 if (cpu_mask == NULL) {
122 perror("CPU_ALLOC");
123 exit(3);
124 }
125 cpu_mask_size = CPU_ALLOC_SIZE(ncpus);
126 CPU_ZERO_S(cpu_mask_size, cpu_mask);
127 152
128 /* 153 for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
129 * Allocate and initialize cpu_present_set 154 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
130 */ 155 for (thread_no = 0; thread_no <
131 cpu_present_set = CPU_ALLOC(ncpus); 156 topo.num_threads_per_core; ++thread_no) {
132 if (cpu_present_set == NULL) { 157 struct thread_data *t;
133 perror("CPU_ALLOC"); 158 struct core_data *c;
134 exit(3); 159 struct pkg_data *p;
135 }
136 cpu_present_setsize = CPU_ALLOC_SIZE(ncpus);
137 CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
138 for_all_cpus(mark_cpu_present);
139}
140 160
141void cpu_mask_uninit() 161 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
142{ 162
143 CPU_FREE(cpu_mask); 163 if (cpu_is_not_present(t->cpu_id))
144 cpu_mask = NULL; 164 continue;
145 cpu_mask_size = 0; 165
146 CPU_FREE(cpu_present_set); 166 c = GET_CORE(core_base, core_no, pkg_no);
147 cpu_present_set = NULL; 167 p = GET_PKG(pkg_base, pkg_no);
148 cpu_present_setsize = 0; 168
169 retval = func(t, c, p);
170 if (retval)
171 return retval;
172 }
173 }
174 }
175 return 0;
149} 176}
150 177
151int cpu_migrate(int cpu) 178int cpu_migrate(int cpu)
152{ 179{
153 CPU_ZERO_S(cpu_mask_size, cpu_mask); 180 CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
154 CPU_SET_S(cpu, cpu_mask_size, cpu_mask); 181 CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
155 if (sched_setaffinity(0, cpu_mask_size, cpu_mask) == -1) 182 if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
156 return -1; 183 return -1;
157 else 184 else
158 return 0; 185 return 0;
@@ -181,67 +208,72 @@ int get_msr(int cpu, off_t offset, unsigned long long *msr)
181void print_header(void) 208void print_header(void)
182{ 209{
183 if (show_pkg) 210 if (show_pkg)
184 fprintf(stderr, "pk"); 211 outp += sprintf(outp, "pk");
185 if (show_pkg) 212 if (show_pkg)
186 fprintf(stderr, " "); 213 outp += sprintf(outp, " ");
187 if (show_core) 214 if (show_core)
188 fprintf(stderr, "cor"); 215 outp += sprintf(outp, "cor");
189 if (show_cpu) 216 if (show_cpu)
190 fprintf(stderr, " CPU"); 217 outp += sprintf(outp, " CPU");
191 if (show_pkg || show_core || show_cpu) 218 if (show_pkg || show_core || show_cpu)
192 fprintf(stderr, " "); 219 outp += sprintf(outp, " ");
193 if (do_nhm_cstates) 220 if (do_nhm_cstates)
194 fprintf(stderr, " %%c0"); 221 outp += sprintf(outp, " %%c0");
195 if (has_aperf) 222 if (has_aperf)
196 fprintf(stderr, " GHz"); 223 outp += sprintf(outp, " GHz");
197 fprintf(stderr, " TSC"); 224 outp += sprintf(outp, " TSC");
198 if (do_nhm_cstates) 225 if (do_nhm_cstates)
199 fprintf(stderr, " %%c1"); 226 outp += sprintf(outp, " %%c1");
200 if (do_nhm_cstates) 227 if (do_nhm_cstates)
201 fprintf(stderr, " %%c3"); 228 outp += sprintf(outp, " %%c3");
202 if (do_nhm_cstates) 229 if (do_nhm_cstates)
203 fprintf(stderr, " %%c6"); 230 outp += sprintf(outp, " %%c6");
204 if (do_snb_cstates) 231 if (do_snb_cstates)
205 fprintf(stderr, " %%c7"); 232 outp += sprintf(outp, " %%c7");
206 if (do_snb_cstates) 233 if (do_snb_cstates)
207 fprintf(stderr, " %%pc2"); 234 outp += sprintf(outp, " %%pc2");
208 if (do_nhm_cstates) 235 if (do_nhm_cstates)
209 fprintf(stderr, " %%pc3"); 236 outp += sprintf(outp, " %%pc3");
210 if (do_nhm_cstates) 237 if (do_nhm_cstates)
211 fprintf(stderr, " %%pc6"); 238 outp += sprintf(outp, " %%pc6");
212 if (do_snb_cstates) 239 if (do_snb_cstates)
213 fprintf(stderr, " %%pc7"); 240 outp += sprintf(outp, " %%pc7");
214 if (extra_msr_offset) 241 if (extra_msr_offset)
215 fprintf(stderr, " MSR 0x%x ", extra_msr_offset); 242 outp += sprintf(outp, " MSR 0x%x ", extra_msr_offset);
216 243
217 putc('\n', stderr); 244 outp += sprintf(outp, "\n");
218} 245}
219 246
220void dump_cnt(struct counters *cnt) 247int dump_counters(struct thread_data *t, struct core_data *c,
248 struct pkg_data *p)
221{ 249{
222 if (!cnt) 250 fprintf(stderr, "t %p, c %p, p %p\n", t, c, p);
223 return; 251
224 if (cnt->pkg) fprintf(stderr, "package: %d ", cnt->pkg); 252 if (t) {
225 if (cnt->core) fprintf(stderr, "core:: %d ", cnt->core); 253 fprintf(stderr, "CPU: %d flags 0x%x\n", t->cpu_id, t->flags);
226 if (cnt->cpu) fprintf(stderr, "CPU: %d ", cnt->cpu); 254 fprintf(stderr, "TSC: %016llX\n", t->tsc);
227 if (cnt->tsc) fprintf(stderr, "TSC: %016llX\n", cnt->tsc); 255 fprintf(stderr, "aperf: %016llX\n", t->aperf);
228 if (cnt->c3) fprintf(stderr, "c3: %016llX\n", cnt->c3); 256 fprintf(stderr, "mperf: %016llX\n", t->mperf);
229 if (cnt->c6) fprintf(stderr, "c6: %016llX\n", cnt->c6); 257 fprintf(stderr, "c1: %016llX\n", t->c1);
230 if (cnt->c7) fprintf(stderr, "c7: %016llX\n", cnt->c7); 258 fprintf(stderr, "msr0x%x: %016llX\n",
231 if (cnt->aperf) fprintf(stderr, "aperf: %016llX\n", cnt->aperf); 259 extra_msr_offset, t->extra_msr);
232 if (cnt->pc2) fprintf(stderr, "pc2: %016llX\n", cnt->pc2); 260 }
233 if (cnt->pc3) fprintf(stderr, "pc3: %016llX\n", cnt->pc3);
234 if (cnt->pc6) fprintf(stderr, "pc6: %016llX\n", cnt->pc6);
235 if (cnt->pc7) fprintf(stderr, "pc7: %016llX\n", cnt->pc7);
236 if (cnt->extra_msr) fprintf(stderr, "msr0x%x: %016llX\n", extra_msr_offset, cnt->extra_msr);
237}
238 261
239void dump_list(struct counters *cnt) 262 if (c) {
240{ 263 fprintf(stderr, "core: %d\n", c->core_id);
241 printf("dump_list 0x%p\n", cnt); 264 fprintf(stderr, "c3: %016llX\n", c->c3);
265 fprintf(stderr, "c6: %016llX\n", c->c6);
266 fprintf(stderr, "c7: %016llX\n", c->c7);
267 }
242 268
243 for (; cnt; cnt = cnt->next) 269 if (p) {
244 dump_cnt(cnt); 270 fprintf(stderr, "package: %d\n", p->package_id);
271 fprintf(stderr, "pc2: %016llX\n", p->pc2);
272 fprintf(stderr, "pc3: %016llX\n", p->pc3);
273 fprintf(stderr, "pc6: %016llX\n", p->pc6);
274 fprintf(stderr, "pc7: %016llX\n", p->pc7);
275 }
276 return 0;
245} 277}
246 278
247/* 279/*
@@ -253,321 +285,385 @@ void dump_list(struct counters *cnt)
253 * TSC: "TSC" 3 columns %3.2 285 * TSC: "TSC" 3 columns %3.2
254 * percentage " %pc3" %6.2 286 * percentage " %pc3" %6.2
255 */ 287 */
256void print_cnt(struct counters *p) 288int format_counters(struct thread_data *t, struct core_data *c,
289 struct pkg_data *p)
257{ 290{
258 double interval_float; 291 double interval_float;
259 292
293 /* if showing only 1st thread in core and this isn't one, bail out */
294 if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
295 return 0;
296
297 /* if showing only 1st thread in pkg and this isn't one, bail out */
298 if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
299 return 0;
300
260 interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0; 301 interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
261 302
262 /* topology columns, print blanks on 1st (average) line */ 303 /* topo columns, print blanks on 1st (average) line */
263 if (p == cnt_average) { 304 if (t == &average.threads) {
264 if (show_pkg) 305 if (show_pkg)
265 fprintf(stderr, " "); 306 outp += sprintf(outp, " ");
266 if (show_pkg && show_core) 307 if (show_pkg && show_core)
267 fprintf(stderr, " "); 308 outp += sprintf(outp, " ");
268 if (show_core) 309 if (show_core)
269 fprintf(stderr, " "); 310 outp += sprintf(outp, " ");
270 if (show_cpu) 311 if (show_cpu)
271 fprintf(stderr, " " " "); 312 outp += sprintf(outp, " " " ");
272 } else { 313 } else {
273 if (show_pkg) 314 if (show_pkg) {
274 fprintf(stderr, "%2d", p->pkg); 315 if (p)
316 outp += sprintf(outp, "%2d", p->package_id);
317 else
318 outp += sprintf(outp, " ");
319 }
275 if (show_pkg && show_core) 320 if (show_pkg && show_core)
276 fprintf(stderr, " "); 321 outp += sprintf(outp, " ");
277 if (show_core) 322 if (show_core) {
278 fprintf(stderr, "%3d", p->core); 323 if (c)
324 outp += sprintf(outp, "%3d", c->core_id);
325 else
326 outp += sprintf(outp, " ");
327 }
279 if (show_cpu) 328 if (show_cpu)
280 fprintf(stderr, " %3d", p->cpu); 329 outp += sprintf(outp, " %3d", t->cpu_id);
281 } 330 }
282 331
283 /* %c0 */ 332 /* %c0 */
284 if (do_nhm_cstates) { 333 if (do_nhm_cstates) {
285 if (show_pkg || show_core || show_cpu) 334 if (show_pkg || show_core || show_cpu)
286 fprintf(stderr, " "); 335 outp += sprintf(outp, " ");
287 if (!skip_c0) 336 if (!skip_c0)
288 fprintf(stderr, "%6.2f", 100.0 * p->mperf/p->tsc); 337 outp += sprintf(outp, "%6.2f", 100.0 * t->mperf/t->tsc);
289 else 338 else
290 fprintf(stderr, " ****"); 339 outp += sprintf(outp, " ****");
291 } 340 }
292 341
293 /* GHz */ 342 /* GHz */
294 if (has_aperf) { 343 if (has_aperf) {
295 if (!aperf_mperf_unstable) { 344 if (!aperf_mperf_unstable) {
296 fprintf(stderr, " %3.2f", 345 outp += sprintf(outp, " %3.2f",
297 1.0 * p->tsc / units * p->aperf / 346 1.0 * t->tsc / units * t->aperf /
298 p->mperf / interval_float); 347 t->mperf / interval_float);
299 } else { 348 } else {
300 if (p->aperf > p->tsc || p->mperf > p->tsc) { 349 if (t->aperf > t->tsc || t->mperf > t->tsc) {
301 fprintf(stderr, " ***"); 350 outp += sprintf(outp, " ***");
302 } else { 351 } else {
303 fprintf(stderr, "%3.1f*", 352 outp += sprintf(outp, "%3.1f*",
304 1.0 * p->tsc / 353 1.0 * t->tsc /
305 units * p->aperf / 354 units * t->aperf /
306 p->mperf / interval_float); 355 t->mperf / interval_float);
307 } 356 }
308 } 357 }
309 } 358 }
310 359
311 /* TSC */ 360 /* TSC */
312 fprintf(stderr, "%5.2f", 1.0 * p->tsc/units/interval_float); 361 outp += sprintf(outp, "%5.2f", 1.0 * t->tsc/units/interval_float);
313 362
314 if (do_nhm_cstates) { 363 if (do_nhm_cstates) {
315 if (!skip_c1) 364 if (!skip_c1)
316 fprintf(stderr, " %6.2f", 100.0 * p->c1/p->tsc); 365 outp += sprintf(outp, " %6.2f", 100.0 * t->c1/t->tsc);
317 else 366 else
318 fprintf(stderr, " ****"); 367 outp += sprintf(outp, " ****");
319 } 368 }
369
370 /* print per-core data only for 1st thread in core */
371 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
372 goto done;
373
320 if (do_nhm_cstates) 374 if (do_nhm_cstates)
321 fprintf(stderr, " %6.2f", 100.0 * p->c3/p->tsc); 375 outp += sprintf(outp, " %6.2f", 100.0 * c->c3/t->tsc);
322 if (do_nhm_cstates) 376 if (do_nhm_cstates)
323 fprintf(stderr, " %6.2f", 100.0 * p->c6/p->tsc); 377 outp += sprintf(outp, " %6.2f", 100.0 * c->c6/t->tsc);
324 if (do_snb_cstates) 378 if (do_snb_cstates)
325 fprintf(stderr, " %6.2f", 100.0 * p->c7/p->tsc); 379 outp += sprintf(outp, " %6.2f", 100.0 * c->c7/t->tsc);
380
381 /* print per-package data only for 1st core in package */
382 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
383 goto done;
384
326 if (do_snb_cstates) 385 if (do_snb_cstates)
327 fprintf(stderr, " %6.2f", 100.0 * p->pc2/p->tsc); 386 outp += sprintf(outp, " %6.2f", 100.0 * p->pc2/t->tsc);
328 if (do_nhm_cstates) 387 if (do_nhm_cstates)
329 fprintf(stderr, " %6.2f", 100.0 * p->pc3/p->tsc); 388 outp += sprintf(outp, " %6.2f", 100.0 * p->pc3/t->tsc);
330 if (do_nhm_cstates) 389 if (do_nhm_cstates)
331 fprintf(stderr, " %6.2f", 100.0 * p->pc6/p->tsc); 390 outp += sprintf(outp, " %6.2f", 100.0 * p->pc6/t->tsc);
332 if (do_snb_cstates) 391 if (do_snb_cstates)
333 fprintf(stderr, " %6.2f", 100.0 * p->pc7/p->tsc); 392 outp += sprintf(outp, " %6.2f", 100.0 * p->pc7/t->tsc);
393done:
334 if (extra_msr_offset) 394 if (extra_msr_offset)
335 fprintf(stderr, " 0x%016llx", p->extra_msr); 395 outp += sprintf(outp, " 0x%016llx", t->extra_msr);
336 putc('\n', stderr); 396 outp += sprintf(outp, "\n");
397
398 return 0;
337} 399}
338 400
339void print_counters(struct counters *counters) 401void flush_stdout()
402{
403 fputs(output_buffer, stdout);
404 outp = output_buffer;
405}
406void flush_stderr()
407{
408 fputs(output_buffer, stderr);
409 outp = output_buffer;
410}
411void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
340{ 412{
341 struct counters *cnt;
342 static int printed; 413 static int printed;
343 414
344
345 if (!printed || !summary_only) 415 if (!printed || !summary_only)
346 print_header(); 416 print_header();
347 417
348 if (num_cpus > 1) 418 if (topo.num_cpus > 1)
349 print_cnt(cnt_average); 419 format_counters(&average.threads, &average.cores,
420 &average.packages);
350 421
351 printed = 1; 422 printed = 1;
352 423
353 if (summary_only) 424 if (summary_only)
354 return; 425 return;
355 426
356 for (cnt = counters; cnt != NULL; cnt = cnt->next) 427 for_all_cpus(format_counters, t, c, p);
357 print_cnt(cnt);
358
359} 428}
360 429
361#define SUBTRACT_COUNTER(after, before, delta) (delta = (after - before), (before > after)) 430void
431delta_package(struct pkg_data *new, struct pkg_data *old)
432{
433 old->pc2 = new->pc2 - old->pc2;
434 old->pc3 = new->pc3 - old->pc3;
435 old->pc6 = new->pc6 - old->pc6;
436 old->pc7 = new->pc7 - old->pc7;
437}
362 438
363int compute_delta(struct counters *after, 439void
364 struct counters *before, struct counters *delta) 440delta_core(struct core_data *new, struct core_data *old)
365{ 441{
366 int errors = 0; 442 old->c3 = new->c3 - old->c3;
367 int perf_err = 0; 443 old->c6 = new->c6 - old->c6;
444 old->c7 = new->c7 - old->c7;
445}
368 446
369 skip_c0 = skip_c1 = 0; 447void
448delta_thread(struct thread_data *new, struct thread_data *old,
449 struct core_data *core_delta)
450{
451 old->tsc = new->tsc - old->tsc;
452
453 /* check for TSC < 1 Mcycles over interval */
454 if (old->tsc < (1000 * 1000)) {
455 fprintf(stderr, "Insanely slow TSC rate, TSC stops in idle?\n");
456 fprintf(stderr, "You can disable all c-states by booting with \"idle=poll\"\n");
457 fprintf(stderr, "or just the deep ones with \"processor.max_cstate=1\"\n");
458 exit(-3);
459 }
370 460
371 for ( ; after && before && delta; 461 old->c1 = new->c1 - old->c1;
372 after = after->next, before = before->next, delta = delta->next) {
373 if (before->cpu != after->cpu) {
374 printf("cpu configuration changed: %d != %d\n",
375 before->cpu, after->cpu);
376 return -1;
377 }
378 462
379 if (SUBTRACT_COUNTER(after->tsc, before->tsc, delta->tsc)) { 463 if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
380 fprintf(stderr, "cpu%d TSC went backwards %llX to %llX\n", 464 old->aperf = new->aperf - old->aperf;
381 before->cpu, before->tsc, after->tsc); 465 old->mperf = new->mperf - old->mperf;
382 errors++; 466 } else {
383 }
384 /* check for TSC < 1 Mcycles over interval */
385 if (delta->tsc < (1000 * 1000)) {
386 fprintf(stderr, "Insanely slow TSC rate,"
387 " TSC stops in idle?\n");
388 fprintf(stderr, "You can disable all c-states"
389 " by booting with \"idle=poll\"\n");
390 fprintf(stderr, "or just the deep ones with"
391 " \"processor.max_cstate=1\"\n");
392 exit(-3);
393 }
394 if (SUBTRACT_COUNTER(after->c3, before->c3, delta->c3)) {
395 fprintf(stderr, "cpu%d c3 counter went backwards %llX to %llX\n",
396 before->cpu, before->c3, after->c3);
397 errors++;
398 }
399 if (SUBTRACT_COUNTER(after->c6, before->c6, delta->c6)) {
400 fprintf(stderr, "cpu%d c6 counter went backwards %llX to %llX\n",
401 before->cpu, before->c6, after->c6);
402 errors++;
403 }
404 if (SUBTRACT_COUNTER(after->c7, before->c7, delta->c7)) {
405 fprintf(stderr, "cpu%d c7 counter went backwards %llX to %llX\n",
406 before->cpu, before->c7, after->c7);
407 errors++;
408 }
409 if (SUBTRACT_COUNTER(after->pc2, before->pc2, delta->pc2)) {
410 fprintf(stderr, "cpu%d pc2 counter went backwards %llX to %llX\n",
411 before->cpu, before->pc2, after->pc2);
412 errors++;
413 }
414 if (SUBTRACT_COUNTER(after->pc3, before->pc3, delta->pc3)) {
415 fprintf(stderr, "cpu%d pc3 counter went backwards %llX to %llX\n",
416 before->cpu, before->pc3, after->pc3);
417 errors++;
418 }
419 if (SUBTRACT_COUNTER(after->pc6, before->pc6, delta->pc6)) {
420 fprintf(stderr, "cpu%d pc6 counter went backwards %llX to %llX\n",
421 before->cpu, before->pc6, after->pc6);
422 errors++;
423 }
424 if (SUBTRACT_COUNTER(after->pc7, before->pc7, delta->pc7)) {
425 fprintf(stderr, "cpu%d pc7 counter went backwards %llX to %llX\n",
426 before->cpu, before->pc7, after->pc7);
427 errors++;
428 }
429 467
430 perf_err = SUBTRACT_COUNTER(after->aperf, before->aperf, delta->aperf); 468 if (!aperf_mperf_unstable) {
431 if (perf_err) { 469 fprintf(stderr, "%s: APERF or MPERF went backwards *\n", progname);
432 fprintf(stderr, "cpu%d aperf counter went backwards %llX to %llX\n", 470 fprintf(stderr, "* Frequency results do not cover entire interval *\n");
433 before->cpu, before->aperf, after->aperf); 471 fprintf(stderr, "* fix this by running Linux-2.6.30 or later *\n");
434 }
435 perf_err |= SUBTRACT_COUNTER(after->mperf, before->mperf, delta->mperf);
436 if (perf_err) {
437 fprintf(stderr, "cpu%d mperf counter went backwards %llX to %llX\n",
438 before->cpu, before->mperf, after->mperf);
439 }
440 if (perf_err) {
441 if (!aperf_mperf_unstable) {
442 fprintf(stderr, "%s: APERF or MPERF went backwards *\n", progname);
443 fprintf(stderr, "* Frequency results do not cover entire interval *\n");
444 fprintf(stderr, "* fix this by running Linux-2.6.30 or later *\n");
445 472
446 aperf_mperf_unstable = 1; 473 aperf_mperf_unstable = 1;
447 }
448 /*
449 * mperf delta is likely a huge "positive" number
450 * can not use it for calculating c0 time
451 */
452 skip_c0 = 1;
453 skip_c1 = 1;
454 } 474 }
455
456 /* 475 /*
457 * As mperf and tsc collection are not atomic, 476 * mperf delta is likely a huge "positive" number
458 * it is possible for mperf's non-halted cycles 477 * can not use it for calculating c0 time
459 * to exceed TSC's all cycles: show c1 = 0% in that case.
460 */ 478 */
461 if (delta->mperf > delta->tsc) 479 skip_c0 = 1;
462 delta->c1 = 0; 480 skip_c1 = 1;
463 else /* normal case, derive c1 */ 481 }
464 delta->c1 = delta->tsc - delta->mperf
465 - delta->c3 - delta->c6 - delta->c7;
466 482
467 if (delta->mperf == 0)
468 delta->mperf = 1; /* divide by 0 protection */
469 483
470 /* 484 /*
471 * for "extra msr", just copy the latest w/o subtracting 485 * As mperf and tsc collection are not atomic,
472 */ 486 * it is possible for mperf's non-halted cycles
473 delta->extra_msr = after->extra_msr; 487 * to exceed TSC's all cycles: show c1 = 0% in that case.
474 if (errors) { 488 */
475 fprintf(stderr, "ERROR cpu%d before:\n", before->cpu); 489 if (old->mperf > old->tsc)
476 dump_cnt(before); 490 old->c1 = 0;
477 fprintf(stderr, "ERROR cpu%d after:\n", before->cpu); 491 else {
478 dump_cnt(after); 492 /* normal case, derive c1 */
479 errors = 0; 493 old->c1 = old->tsc - old->mperf - core_delta->c3
480 } 494 - core_delta->c6 - core_delta->c7;
495 }
496 if (old->mperf == 0) {
497 if (verbose) fprintf(stderr, "cpu%d MPERF 0!\n", old->cpu_id);
498 old->mperf = 1; /* divide by 0 protection */
481 } 499 }
500
501 /*
502 * for "extra msr", just copy the latest w/o subtracting
503 */
504 old->extra_msr = new->extra_msr;
505}
506
507int delta_cpu(struct thread_data *t, struct core_data *c,
508 struct pkg_data *p, struct thread_data *t2,
509 struct core_data *c2, struct pkg_data *p2)
510{
511 /* calculate core delta only for 1st thread in core */
512 if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
513 delta_core(c, c2);
514
515 /* always calculate thread delta */
516 delta_thread(t, t2, c2); /* c2 is core delta */
517
518 /* calculate package delta only for 1st core in package */
519 if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
520 delta_package(p, p2);
521
482 return 0; 522 return 0;
483} 523}
484 524
485void compute_average(struct counters *delta, struct counters *avg) 525void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
526{
527 t->tsc = 0;
528 t->aperf = 0;
529 t->mperf = 0;
530 t->c1 = 0;
531
532 /* tells format_counters to dump all fields from this set */
533 t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE;
534
535 c->c3 = 0;
536 c->c6 = 0;
537 c->c7 = 0;
538
539 p->pc2 = 0;
540 p->pc3 = 0;
541 p->pc6 = 0;
542 p->pc7 = 0;
543}
544int sum_counters(struct thread_data *t, struct core_data *c,
545 struct pkg_data *p)
486{ 546{
487 struct counters *sum; 547 average.threads.tsc += t->tsc;
548 average.threads.aperf += t->aperf;
549 average.threads.mperf += t->mperf;
550 average.threads.c1 += t->c1;
488 551
489 sum = calloc(1, sizeof(struct counters)); 552 /* sum per-core values only for 1st thread in core */
490 if (sum == NULL) { 553 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
491 perror("calloc sum"); 554 return 0;
492 exit(1);
493 }
494 555
495 for (; delta; delta = delta->next) { 556 average.cores.c3 += c->c3;
496 sum->tsc += delta->tsc; 557 average.cores.c6 += c->c6;
497 sum->c1 += delta->c1; 558 average.cores.c7 += c->c7;
498 sum->c3 += delta->c3; 559
499 sum->c6 += delta->c6; 560 /* sum per-pkg values only for 1st core in pkg */
500 sum->c7 += delta->c7; 561 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
501 sum->aperf += delta->aperf; 562 return 0;
502 sum->mperf += delta->mperf; 563
503 sum->pc2 += delta->pc2; 564 average.packages.pc2 += p->pc2;
504 sum->pc3 += delta->pc3; 565 average.packages.pc3 += p->pc3;
505 sum->pc6 += delta->pc6; 566 average.packages.pc6 += p->pc6;
506 sum->pc7 += delta->pc7; 567 average.packages.pc7 += p->pc7;
507 } 568
508 avg->tsc = sum->tsc/num_cpus; 569 return 0;
509 avg->c1 = sum->c1/num_cpus; 570}
510 avg->c3 = sum->c3/num_cpus; 571/*
511 avg->c6 = sum->c6/num_cpus; 572 * sum the counters for all cpus in the system
512 avg->c7 = sum->c7/num_cpus; 573 * compute the weighted average
513 avg->aperf = sum->aperf/num_cpus; 574 */
514 avg->mperf = sum->mperf/num_cpus; 575void compute_average(struct thread_data *t, struct core_data *c,
515 avg->pc2 = sum->pc2/num_cpus; 576 struct pkg_data *p)
516 avg->pc3 = sum->pc3/num_cpus; 577{
517 avg->pc6 = sum->pc6/num_cpus; 578 clear_counters(&average.threads, &average.cores, &average.packages);
518 avg->pc7 = sum->pc7/num_cpus; 579
519 580 for_all_cpus(sum_counters, t, c, p);
520 free(sum); 581
582 average.threads.tsc /= topo.num_cpus;
583 average.threads.aperf /= topo.num_cpus;
584 average.threads.mperf /= topo.num_cpus;
585 average.threads.c1 /= topo.num_cpus;
586
587 average.cores.c3 /= topo.num_cores;
588 average.cores.c6 /= topo.num_cores;
589 average.cores.c7 /= topo.num_cores;
590
591 average.packages.pc2 /= topo.num_packages;
592 average.packages.pc3 /= topo.num_packages;
593 average.packages.pc6 /= topo.num_packages;
594 average.packages.pc7 /= topo.num_packages;
521} 595}
522 596
523int get_counters(struct counters *cnt) 597static unsigned long long rdtsc(void)
524{ 598{
525 for ( ; cnt; cnt = cnt->next) { 599 unsigned int low, high;
526 600
527 if (cpu_migrate(cnt->cpu)) 601 asm volatile("rdtsc" : "=a" (low), "=d" (high));
528 return -1;
529 602
530 if (get_msr(cnt->cpu, MSR_TSC, &cnt->tsc)) 603 return low | ((unsigned long long)high) << 32;
531 return -1; 604}
532 605
533 if (has_aperf) {
534 if (get_msr(cnt->cpu, MSR_APERF, &cnt->aperf))
535 return -1;
536 if (get_msr(cnt->cpu, MSR_MPERF, &cnt->mperf))
537 return -1;
538 }
539 606
540 if (do_nhm_cstates) { 607/*
541 if (get_msr(cnt->cpu, MSR_CORE_C3_RESIDENCY, &cnt->c3)) 608 * get_counters(...)
542 return -1; 609 * migrate to cpu
543 if (get_msr(cnt->cpu, MSR_CORE_C6_RESIDENCY, &cnt->c6)) 610 * acquire and record local counters for that cpu
544 return -1; 611 */
545 } 612int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
613{
614 int cpu = t->cpu_id;
546 615
547 if (do_snb_cstates) 616 if (cpu_migrate(cpu))
548 if (get_msr(cnt->cpu, MSR_CORE_C7_RESIDENCY, &cnt->c7)) 617 return -1;
549 return -1;
550 618
551 if (do_nhm_cstates) { 619 t->tsc = rdtsc(); /* we are running on local CPU of interest */
552 if (get_msr(cnt->cpu, MSR_PKG_C3_RESIDENCY, &cnt->pc3)) 620
553 return -1; 621 if (has_aperf) {
554 if (get_msr(cnt->cpu, MSR_PKG_C6_RESIDENCY, &cnt->pc6)) 622 if (get_msr(cpu, MSR_APERF, &t->aperf))
555 return -1; 623 return -3;
556 } 624 if (get_msr(cpu, MSR_MPERF, &t->mperf))
557 if (do_snb_cstates) { 625 return -4;
558 if (get_msr(cnt->cpu, MSR_PKG_C2_RESIDENCY, &cnt->pc2)) 626 }
559 return -1; 627
560 if (get_msr(cnt->cpu, MSR_PKG_C7_RESIDENCY, &cnt->pc7)) 628 if (extra_msr_offset)
561 return -1; 629 if (get_msr(cpu, extra_msr_offset, &t->extra_msr))
562 } 630 return -5;
563 if (extra_msr_offset) 631
564 if (get_msr(cnt->cpu, extra_msr_offset, &cnt->extra_msr)) 632 /* collect core counters only for 1st thread in core */
565 return -1; 633 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
634 return 0;
635
636 if (do_nhm_cstates) {
637 if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
638 return -6;
639 if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
640 return -7;
641 }
642
643 if (do_snb_cstates)
644 if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
645 return -8;
646
647 /* collect package counters only for 1st core in package */
648 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
649 return 0;
650
651 if (do_nhm_cstates) {
652 if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
653 return -9;
654 if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
655 return -10;
656 }
657 if (do_snb_cstates) {
658 if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
659 return -11;
660 if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
661 return -12;
566 } 662 }
567 return 0; 663 return 0;
568} 664}
569 665
570void print_nehalem_info(void) 666void print_verbose_header(void)
571{ 667{
572 unsigned long long msr; 668 unsigned long long msr;
573 unsigned int ratio; 669 unsigned int ratio;
@@ -615,143 +711,82 @@ void print_nehalem_info(void)
615 711
616} 712}
617 713
618void free_counter_list(struct counters *list) 714void free_all_buffers(void)
619{ 715{
620 struct counters *p; 716 CPU_FREE(cpu_present_set);
717 cpu_present_set = NULL;
718 cpu_present_set = 0;
621 719
622 for (p = list; p; ) { 720 CPU_FREE(cpu_affinity_set);
623 struct counters *free_me; 721 cpu_affinity_set = NULL;
722 cpu_affinity_setsize = 0;
624 723
625 free_me = p; 724 free(thread_even);
626 p = p->next; 725 free(core_even);
627 free(free_me); 726 free(package_even);
628 }
629}
630 727
631void free_all_counters(void) 728 thread_even = NULL;
632{ 729 core_even = NULL;
633 free_counter_list(cnt_even); 730 package_even = NULL;
634 cnt_even = NULL;
635 731
636 free_counter_list(cnt_odd); 732 free(thread_odd);
637 cnt_odd = NULL; 733 free(core_odd);
734 free(package_odd);
638 735
639 free_counter_list(cnt_delta); 736 thread_odd = NULL;
640 cnt_delta = NULL; 737 core_odd = NULL;
738 package_odd = NULL;
641 739
642 free_counter_list(cnt_average); 740 free(output_buffer);
643 cnt_average = NULL; 741 output_buffer = NULL;
742 outp = NULL;
644} 743}
645 744
646void insert_counters(struct counters **list, 745/*
647 struct counters *new) 746 * cpu_is_first_sibling_in_core(cpu)
747 * return 1 if given CPU is 1st HT sibling in the core
748 */
749int cpu_is_first_sibling_in_core(int cpu)
648{ 750{
649 struct counters *prev; 751 char path[64];
650 752 FILE *filep;
651 /* 753 int first_cpu;
652 * list was empty
653 */
654 if (*list == NULL) {
655 new->next = *list;
656 *list = new;
657 return;
658 }
659
660 if (!summary_only)
661 show_cpu = 1; /* there is more than one CPU */
662
663 /*
664 * insert on front of list.
665 * It is sorted by ascending package#, core#, cpu#
666 */
667 if (((*list)->pkg > new->pkg) ||
668 (((*list)->pkg == new->pkg) && ((*list)->core > new->core)) ||
669 (((*list)->pkg == new->pkg) && ((*list)->core == new->core) && ((*list)->cpu > new->cpu))) {
670 new->next = *list;
671 *list = new;
672 return;
673 }
674
675 prev = *list;
676
677 while (prev->next && (prev->next->pkg < new->pkg)) {
678 prev = prev->next;
679 if (!summary_only)
680 show_pkg = 1; /* there is more than 1 package */
681 }
682
683 while (prev->next && (prev->next->pkg == new->pkg)
684 && (prev->next->core < new->core)) {
685 prev = prev->next;
686 if (!summary_only)
687 show_core = 1; /* there is more than 1 core */
688 }
689 754
690 while (prev->next && (prev->next->pkg == new->pkg) 755 sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
691 && (prev->next->core == new->core) 756 filep = fopen(path, "r");
692 && (prev->next->cpu < new->cpu)) { 757 if (filep == NULL) {
693 prev = prev->next; 758 perror(path);
759 exit(1);
694 } 760 }
695 761 fscanf(filep, "%d", &first_cpu);
696 /* 762 fclose(filep);
697 * insert after "prev" 763 return (cpu == first_cpu);
698 */
699 new->next = prev->next;
700 prev->next = new;
701} 764}
702 765
703void alloc_new_counters(int pkg, int core, int cpu) 766/*
767 * cpu_is_first_core_in_package(cpu)
768 * return 1 if given CPU is 1st core in package
769 */
770int cpu_is_first_core_in_package(int cpu)
704{ 771{
705 struct counters *new; 772 char path[64];
706 773 FILE *filep;
707 if (verbose > 1) 774 int first_cpu;
708 printf("pkg%d core%d, cpu%d\n", pkg, core, cpu);
709
710 new = (struct counters *)calloc(1, sizeof(struct counters));
711 if (new == NULL) {
712 perror("calloc");
713 exit(1);
714 }
715 new->pkg = pkg;
716 new->core = core;
717 new->cpu = cpu;
718 insert_counters(&cnt_odd, new);
719
720 new = (struct counters *)calloc(1,
721 sizeof(struct counters));
722 if (new == NULL) {
723 perror("calloc");
724 exit(1);
725 }
726 new->pkg = pkg;
727 new->core = core;
728 new->cpu = cpu;
729 insert_counters(&cnt_even, new);
730
731 new = (struct counters *)calloc(1, sizeof(struct counters));
732 if (new == NULL) {
733 perror("calloc");
734 exit(1);
735 }
736 new->pkg = pkg;
737 new->core = core;
738 new->cpu = cpu;
739 insert_counters(&cnt_delta, new);
740 775
741 new = (struct counters *)calloc(1, sizeof(struct counters)); 776 sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
742 if (new == NULL) { 777 filep = fopen(path, "r");
743 perror("calloc"); 778 if (filep == NULL) {
779 perror(path);
744 exit(1); 780 exit(1);
745 } 781 }
746 new->pkg = pkg; 782 fscanf(filep, "%d", &first_cpu);
747 new->core = core; 783 fclose(filep);
748 new->cpu = cpu; 784 return (cpu == first_cpu);
749 cnt_average = new;
750} 785}
751 786
752int get_physical_package_id(int cpu) 787int get_physical_package_id(int cpu)
753{ 788{
754 char path[64]; 789 char path[80];
755 FILE *filep; 790 FILE *filep;
756 int pkg; 791 int pkg;
757 792
@@ -768,7 +803,7 @@ int get_physical_package_id(int cpu)
768 803
769int get_core_id(int cpu) 804int get_core_id(int cpu)
770{ 805{
771 char path[64]; 806 char path[80];
772 FILE *filep; 807 FILE *filep;
773 int core; 808 int core;
774 809
@@ -783,14 +818,87 @@ int get_core_id(int cpu)
783 return core; 818 return core;
784} 819}
785 820
821int get_num_ht_siblings(int cpu)
822{
823 char path[80];
824 FILE *filep;
825 int sib1, sib2;
826 int matches;
827 char character;
828
829 sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
830 filep = fopen(path, "r");
831 if (filep == NULL) {
832 perror(path);
833 exit(1);
834 }
835 /*
836 * file format:
837 * if a pair of number with a character between: 2 siblings (eg. 1-2, or 1,4)
838 * otherwinse 1 sibling (self).
839 */
840 matches = fscanf(filep, "%d%c%d\n", &sib1, &character, &sib2);
841
842 fclose(filep);
843
844 if (matches == 3)
845 return 2;
846 else
847 return 1;
848}
849
786/* 850/*
787 * run func(pkg, core, cpu) on every cpu in /proc/stat 851 * run func(thread, core, package) in topology order
852 * skip non-present cpus
788 */ 853 */
789 854
790int for_all_cpus(void (func)(int, int, int)) 855int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
856 struct pkg_data *, struct thread_data *, struct core_data *,
857 struct pkg_data *), struct thread_data *thread_base,
858 struct core_data *core_base, struct pkg_data *pkg_base,
859 struct thread_data *thread_base2, struct core_data *core_base2,
860 struct pkg_data *pkg_base2)
861{
862 int retval, pkg_no, core_no, thread_no;
863
864 for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
865 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
866 for (thread_no = 0; thread_no <
867 topo.num_threads_per_core; ++thread_no) {
868 struct thread_data *t, *t2;
869 struct core_data *c, *c2;
870 struct pkg_data *p, *p2;
871
872 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
873
874 if (cpu_is_not_present(t->cpu_id))
875 continue;
876
877 t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no);
878
879 c = GET_CORE(core_base, core_no, pkg_no);
880 c2 = GET_CORE(core_base2, core_no, pkg_no);
881
882 p = GET_PKG(pkg_base, pkg_no);
883 p2 = GET_PKG(pkg_base2, pkg_no);
884
885 retval = func(t, c, p, t2, c2, p2);
886 if (retval)
887 return retval;
888 }
889 }
890 }
891 return 0;
892}
893
894/*
895 * run func(cpu) on every cpu in /proc/stat
896 * return max_cpu number
897 */
898int for_all_proc_cpus(int (func)(int))
791{ 899{
792 FILE *fp; 900 FILE *fp;
793 int cpu_count; 901 int cpu_num;
794 int retval; 902 int retval;
795 903
796 fp = fopen(proc_stat, "r"); 904 fp = fopen(proc_stat, "r");
@@ -805,78 +913,88 @@ int for_all_cpus(void (func)(int, int, int))
805 exit(1); 913 exit(1);
806 } 914 }
807 915
808 for (cpu_count = 0; ; cpu_count++) { 916 while (1) {
809 int cpu; 917 retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
810
811 retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu);
812 if (retval != 1) 918 if (retval != 1)
813 break; 919 break;
814 920
815 func(get_physical_package_id(cpu), get_core_id(cpu), cpu); 921 retval = func(cpu_num);
922 if (retval) {
923 fclose(fp);
924 return(retval);
925 }
816 } 926 }
817 fclose(fp); 927 fclose(fp);
818 return cpu_count; 928 return 0;
819} 929}
820 930
821void re_initialize(void) 931void re_initialize(void)
822{ 932{
823 free_all_counters(); 933 free_all_buffers();
824 num_cpus = for_all_cpus(alloc_new_counters); 934 setup_all_buffers();
825 cpu_mask_uninit(); 935 printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus);
826 cpu_mask_init(num_cpus);
827 printf("turbostat: re-initialized with num_cpus %d\n", num_cpus);
828} 936}
829 937
830void dummy(int pkg, int core, int cpu) { return; } 938
831/* 939/*
832 * check to see if a cpu came on-line 940 * count_cpus()
941 * remember the last one seen, it will be the max
833 */ 942 */
834int verify_num_cpus(void) 943int count_cpus(int cpu)
835{ 944{
836 int new_num_cpus; 945 if (topo.max_cpu_num < cpu)
946 topo.max_cpu_num = cpu;
837 947
838 new_num_cpus = for_all_cpus(dummy); 948 topo.num_cpus += 1;
839 949 return 0;
840 if (new_num_cpus != num_cpus) { 950}
841 if (verbose) 951int mark_cpu_present(int cpu)
842 printf("num_cpus was %d, is now %d\n", 952{
843 num_cpus, new_num_cpus); 953 CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
844 return -1;
845 }
846 return 0; 954 return 0;
847} 955}
848 956
849void turbostat_loop() 957void turbostat_loop()
850{ 958{
959 int retval;
960
851restart: 961restart:
852 get_counters(cnt_even); 962 retval = for_all_cpus(get_counters, EVEN_COUNTERS);
963 if (retval) {
964 re_initialize();
965 goto restart;
966 }
853 gettimeofday(&tv_even, (struct timezone *)NULL); 967 gettimeofday(&tv_even, (struct timezone *)NULL);
854 968
855 while (1) { 969 while (1) {
856 if (verify_num_cpus()) { 970 if (for_all_proc_cpus(cpu_is_not_present)) {
857 re_initialize(); 971 re_initialize();
858 goto restart; 972 goto restart;
859 } 973 }
860 sleep(interval_sec); 974 sleep(interval_sec);
861 if (get_counters(cnt_odd)) { 975 retval = for_all_cpus(get_counters, ODD_COUNTERS);
976 if (retval) {
862 re_initialize(); 977 re_initialize();
863 goto restart; 978 goto restart;
864 } 979 }
865 gettimeofday(&tv_odd, (struct timezone *)NULL); 980 gettimeofday(&tv_odd, (struct timezone *)NULL);
866 compute_delta(cnt_odd, cnt_even, cnt_delta);
867 timersub(&tv_odd, &tv_even, &tv_delta); 981 timersub(&tv_odd, &tv_even, &tv_delta);
868 compute_average(cnt_delta, cnt_average); 982 for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS);
869 print_counters(cnt_delta); 983 compute_average(EVEN_COUNTERS);
984 format_all_counters(EVEN_COUNTERS);
985 flush_stdout();
870 sleep(interval_sec); 986 sleep(interval_sec);
871 if (get_counters(cnt_even)) { 987 retval = for_all_cpus(get_counters, EVEN_COUNTERS);
988 if (retval) {
872 re_initialize(); 989 re_initialize();
873 goto restart; 990 goto restart;
874 } 991 }
875 gettimeofday(&tv_even, (struct timezone *)NULL); 992 gettimeofday(&tv_even, (struct timezone *)NULL);
876 compute_delta(cnt_even, cnt_odd, cnt_delta);
877 timersub(&tv_even, &tv_odd, &tv_delta); 993 timersub(&tv_even, &tv_odd, &tv_delta);
878 compute_average(cnt_delta, cnt_average); 994 for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS);
879 print_counters(cnt_delta); 995 compute_average(ODD_COUNTERS);
996 format_all_counters(ODD_COUNTERS);
997 flush_stdout();
880 } 998 }
881} 999}
882 1000
@@ -1051,6 +1169,208 @@ int open_dev_cpu_msr(int dummy1)
1051 return 0; 1169 return 0;
1052} 1170}
1053 1171
1172void topology_probe()
1173{
1174 int i;
1175 int max_core_id = 0;
1176 int max_package_id = 0;
1177 int max_siblings = 0;
1178 struct cpu_topology {
1179 int core_id;
1180 int physical_package_id;
1181 } *cpus;
1182
1183 /* Initialize num_cpus, max_cpu_num */
1184 topo.num_cpus = 0;
1185 topo.max_cpu_num = 0;
1186 for_all_proc_cpus(count_cpus);
1187 if (!summary_only && topo.num_cpus > 1)
1188 show_cpu = 1;
1189
1190 if (verbose > 1)
1191 fprintf(stderr, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
1192
1193 cpus = calloc(1, (topo.max_cpu_num + 1) * sizeof(struct cpu_topology));
1194 if (cpus == NULL) {
1195 perror("calloc cpus");
1196 exit(1);
1197 }
1198
1199 /*
1200 * Allocate and initialize cpu_present_set
1201 */
1202 cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
1203 if (cpu_present_set == NULL) {
1204 perror("CPU_ALLOC");
1205 exit(3);
1206 }
1207 cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
1208 CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
1209 for_all_proc_cpus(mark_cpu_present);
1210
1211 /*
1212 * Allocate and initialize cpu_affinity_set
1213 */
1214 cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
1215 if (cpu_affinity_set == NULL) {
1216 perror("CPU_ALLOC");
1217 exit(3);
1218 }
1219 cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
1220 CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
1221
1222
1223 /*
1224 * For online cpus
1225 * find max_core_id, max_package_id
1226 */
1227 for (i = 0; i <= topo.max_cpu_num; ++i) {
1228 int siblings;
1229
1230 if (cpu_is_not_present(i)) {
1231 if (verbose > 1)
1232 fprintf(stderr, "cpu%d NOT PRESENT\n", i);
1233 continue;
1234 }
1235 cpus[i].core_id = get_core_id(i);
1236 if (cpus[i].core_id > max_core_id)
1237 max_core_id = cpus[i].core_id;
1238
1239 cpus[i].physical_package_id = get_physical_package_id(i);
1240 if (cpus[i].physical_package_id > max_package_id)
1241 max_package_id = cpus[i].physical_package_id;
1242
1243 siblings = get_num_ht_siblings(i);
1244 if (siblings > max_siblings)
1245 max_siblings = siblings;
1246 if (verbose > 1)
1247 fprintf(stderr, "cpu %d pkg %d core %d\n",
1248 i, cpus[i].physical_package_id, cpus[i].core_id);
1249 }
1250 topo.num_cores_per_pkg = max_core_id + 1;
1251 if (verbose > 1)
1252 fprintf(stderr, "max_core_id %d, sizing for %d cores per package\n",
1253 max_core_id, topo.num_cores_per_pkg);
1254 if (!summary_only && topo.num_cores_per_pkg > 1)
1255 show_core = 1;
1256
1257 topo.num_packages = max_package_id + 1;
1258 if (verbose > 1)
1259 fprintf(stderr, "max_package_id %d, sizing for %d packages\n",
1260 max_package_id, topo.num_packages);
1261 if (!summary_only && topo.num_packages > 1)
1262 show_pkg = 1;
1263
1264 topo.num_threads_per_core = max_siblings;
1265 if (verbose > 1)
1266 fprintf(stderr, "max_siblings %d\n", max_siblings);
1267
1268 free(cpus);
1269}
1270
1271void
1272allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p)
1273{
1274 int i;
1275
1276 *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg *
1277 topo.num_packages, sizeof(struct thread_data));
1278 if (*t == NULL)
1279 goto error;
1280
1281 for (i = 0; i < topo.num_threads_per_core *
1282 topo.num_cores_per_pkg * topo.num_packages; i++)
1283 (*t)[i].cpu_id = -1;
1284
1285 *c = calloc(topo.num_cores_per_pkg * topo.num_packages,
1286 sizeof(struct core_data));
1287 if (*c == NULL)
1288 goto error;
1289
1290 for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++)
1291 (*c)[i].core_id = -1;
1292
1293 *p = calloc(topo.num_packages, sizeof(struct pkg_data));
1294 if (*p == NULL)
1295 goto error;
1296
1297 for (i = 0; i < topo.num_packages; i++)
1298 (*p)[i].package_id = i;
1299
1300 return;
1301error:
1302 perror("calloc counters");
1303 exit(1);
1304}
1305/*
1306 * init_counter()
1307 *
1308 * set cpu_id, core_num, pkg_num
1309 * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
1310 *
1311 * increment topo.num_cores when 1st core in pkg seen
1312 */
1313void init_counter(struct thread_data *thread_base, struct core_data *core_base,
1314 struct pkg_data *pkg_base, int thread_num, int core_num,
1315 int pkg_num, int cpu_id)
1316{
1317 struct thread_data *t;
1318 struct core_data *c;
1319 struct pkg_data *p;
1320
1321 t = GET_THREAD(thread_base, thread_num, core_num, pkg_num);
1322 c = GET_CORE(core_base, core_num, pkg_num);
1323 p = GET_PKG(pkg_base, pkg_num);
1324
1325 t->cpu_id = cpu_id;
1326 if (thread_num == 0) {
1327 t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
1328 if (cpu_is_first_core_in_package(cpu_id))
1329 t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
1330 }
1331
1332 c->core_id = core_num;
1333 p->package_id = pkg_num;
1334}
1335
1336
1337int initialize_counters(int cpu_id)
1338{
1339 int my_thread_id, my_core_id, my_package_id;
1340
1341 my_package_id = get_physical_package_id(cpu_id);
1342 my_core_id = get_core_id(cpu_id);
1343
1344 if (cpu_is_first_sibling_in_core(cpu_id)) {
1345 my_thread_id = 0;
1346 topo.num_cores++;
1347 } else {
1348 my_thread_id = 1;
1349 }
1350
1351 init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
1352 init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
1353 return 0;
1354}
1355
1356void allocate_output_buffer()
1357{
1358 output_buffer = calloc(1, (1 + topo.num_cpus) * 128);
1359 outp = output_buffer;
1360 if (outp == NULL) {
1361 perror("calloc");
1362 exit(-1);
1363 }
1364}
1365
1366void setup_all_buffers(void)
1367{
1368 topology_probe();
1369 allocate_counters(&thread_even, &core_even, &package_even);
1370 allocate_counters(&thread_odd, &core_odd, &package_odd);
1371 allocate_output_buffer();
1372 for_all_proc_cpus(initialize_counters);
1373}
1054void turbostat_init() 1374void turbostat_init()
1055{ 1375{
1056 check_cpuid(); 1376 check_cpuid();
@@ -1058,21 +1378,19 @@ void turbostat_init()
1058 check_dev_msr(); 1378 check_dev_msr();
1059 check_super_user(); 1379 check_super_user();
1060 1380
1061 num_cpus = for_all_cpus(alloc_new_counters); 1381 setup_all_buffers();
1062 cpu_mask_init(num_cpus);
1063 1382
1064 if (verbose) 1383 if (verbose)
1065 print_nehalem_info(); 1384 print_verbose_header();
1066} 1385}
1067 1386
1068int fork_it(char **argv) 1387int fork_it(char **argv)
1069{ 1388{
1070 int retval;
1071 pid_t child_pid; 1389 pid_t child_pid;
1072 get_counters(cnt_even);
1073 1390
1074 /* clear affinity side-effect of get_counters() */ 1391 for_all_cpus(get_counters, EVEN_COUNTERS);
1075 sched_setaffinity(0, cpu_present_setsize, cpu_present_set); 1392 /* clear affinity side-effect of get_counters() */
1393 sched_setaffinity(0, cpu_present_setsize, cpu_present_set);
1076 gettimeofday(&tv_even, (struct timezone *)NULL); 1394 gettimeofday(&tv_even, (struct timezone *)NULL);
1077 1395
1078 child_pid = fork(); 1396 child_pid = fork();
@@ -1095,14 +1413,17 @@ int fork_it(char **argv)
1095 exit(1); 1413 exit(1);
1096 } 1414 }
1097 } 1415 }
1098 get_counters(cnt_odd); 1416 /*
1417 * n.b. fork_it() does not check for errors from for_all_cpus()
1418 * because re-starting is problematic when forking
1419 */
1420 for_all_cpus(get_counters, ODD_COUNTERS);
1099 gettimeofday(&tv_odd, (struct timezone *)NULL); 1421 gettimeofday(&tv_odd, (struct timezone *)NULL);
1100 retval = compute_delta(cnt_odd, cnt_even, cnt_delta);
1101
1102 timersub(&tv_odd, &tv_even, &tv_delta); 1422 timersub(&tv_odd, &tv_even, &tv_delta);
1103 compute_average(cnt_delta, cnt_average); 1423 for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS);
1104 if (!retval) 1424 compute_average(EVEN_COUNTERS);
1105 print_counters(cnt_delta); 1425 format_all_counters(EVEN_COUNTERS);
1426 flush_stderr();
1106 1427
1107 fprintf(stderr, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0); 1428 fprintf(stderr, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);
1108 1429
@@ -1115,8 +1436,14 @@ void cmdline(int argc, char **argv)
1115 1436
1116 progname = argv[0]; 1437 progname = argv[0];
1117 1438
1118 while ((opt = getopt(argc, argv, "+svi:M:")) != -1) { 1439 while ((opt = getopt(argc, argv, "+cpsvi:M:")) != -1) {
1119 switch (opt) { 1440 switch (opt) {
1441 case 'c':
1442 show_core_only++;
1443 break;
1444 case 'p':
1445 show_pkg_only++;
1446 break;
1120 case 's': 1447 case 's':
1121 summary_only++; 1448 summary_only++;
1122 break; 1449 break;
@@ -1142,10 +1469,8 @@ int main(int argc, char **argv)
1142 cmdline(argc, argv); 1469 cmdline(argc, argv);
1143 1470
1144 if (verbose > 1) 1471 if (verbose > 1)
1145 fprintf(stderr, "turbostat Dec 6, 2010" 1472 fprintf(stderr, "turbostat v2.0 May 16, 2012"
1146 " - Len Brown <lenb@kernel.org>\n"); 1473 " - Len Brown <lenb@kernel.org>\n");
1147 if (verbose > 1)
1148 fprintf(stderr, "http://userweb.kernel.org/~lenb/acpi/utils/pmtools/turbostat/\n");
1149 1474
1150 turbostat_init(); 1475 turbostat_init();
1151 1476
generated by cgit v1.2.2 (git 2.25.0) at 2025-12-30 17:05:11 -0500