2 * Copyright (c) 2010 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * - Monitor the cpu load and adjusts cpu and cpu power domain
38 * performance accordingly.
39 * - Monitor battery life. Alarm alerts and shutdown the machine
40 * if battery life goes low.
43 #define _KERNEL_STRUCTURES
44 #include <sys/types.h>
45 #include <sys/sysctl.h>
46 #include <sys/kinfo.h>
48 #include <sys/queue.h>
49 #include <sys/soundcard.h>
51 #include <machine/cpufunc.h>
61 #define MAXDOM MAXCPU /* worst case, 1 cpu per domain */
67 TAILQ_ENTRY(cpu_pwrdom) dom_link;
70 cpumask_t dom_cpumask;
75 double cpu_uavg; /* used for speeding up */
76 double cpu_davg; /* used for slowing down */
82 static void usage(void);
83 static void get_ncpus(void);
86 static void get_uschedcpus(void);
87 static void set_uschedcpus(void);
90 static int has_perfbias(void);
91 static void set_perfbias(int, int);
94 static void acpi_getcpufreq_str(int, int *, int *);
95 static int acpi_getcpufreq_bin(int, int *, int *);
96 static void acpi_get_cpufreq(int, int *, int *);
97 static void acpi_set_cpufreq(int, int);
98 static int acpi_get_cpupwrdom(void);
100 /* mwait C-state hint */
101 static int probe_cstate(void);
102 static void set_cstate(int, int);
104 /* Performance monitoring */
105 static void init_perf(void);
106 static void mon_perf(double);
107 static void adj_perf(cpumask_t, cpumask_t);
108 static void adj_cpu_pwrdom(int, int);
109 static void adj_cpu_perf(int, int);
110 static void get_cputime(double);
111 static int get_nstate(struct cpu_state *, double);
112 static void add_spare_cpus(const cpumask_t, int);
113 static void restore_perf(void);
115 /* Battery monitoring */
116 static int has_battery(void);
117 static int mon_battery(void);
118 static void low_battery_alert(int);
120 /* Runtime states for performance monitoring */
121 static int global_pcpu_limit;
122 static struct cpu_state pcpu_state[MAXCPU];
123 static struct cpu_state global_cpu_state;
124 static cpumask_t cpu_used; /* cpus w/ high perf */
125 static cpumask_t cpu_pwrdom_used; /* cpu power domains w/ high perf */
126 static cpumask_t usched_cpu_used; /* cpus for usched */
129 static cpumask_t cpu_pwrdom_mask; /* usable cpu power domains */
130 static int cpu2pwrdom[MAXCPU]; /* cpu to cpu power domain map */
131 static struct cpu_pwrdom *cpu_pwrdomain[MAXDOM];
132 static int NCpus; /* # of cpus */
133 static char orig_global_cx[CST_STRLEN];
134 static char cpu_perf_cx[CST_STRLEN];
135 static int cpu_perf_cxlen;
136 static char cpu_idle_cx[CST_STRLEN];
137 static int cpu_idle_cxlen;
140 static int TurboOpt = 1;
142 static int Hysteresis = 10; /* percentage */
143 static double TriggerUp = 0.25; /* single-cpu load to force max freq */
144 static double TriggerDown; /* load per cpu to force the min freq */
145 static int HasPerfbias = 0;
146 static int AdjustCpuFreq = 1;
147 static int AdjustCstate = 0;
148 static int HighestCpuFreq;
149 static int LowestCpuFreq;
151 static volatile int stopped;
153 /* Battery life monitoring */
154 static int BatLifeMin = 2; /* shutdown the box, if low on battery life */
155 static struct timespec BatLifePrevT;
156 static int BatLifePollIntvl = 5; /* unit: sec */
157 static struct timespec BatShutdownStartT;
158 static int BatShutdownLinger = -1;
159 static int BatShutdownLingerSet = 60; /* unit: sec */
160 static int BatShutdownLingerCnt;
161 static int BatShutdownAudioAlert = 1;
163 static void sigintr(int signo);
166 main(int ac, char **av)
174 srt = 8.0; /* time for samples - 8 seconds */
175 pollrate = 1.0; /* polling rate in seconds */
177 while ((ch = getopt(ac, av, "cdefh:l:p:r:tu:B:L:P:QT:")) != -1) {
192 HighestCpuFreq = strtol(optarg, NULL, 10);
195 LowestCpuFreq = strtol(optarg, NULL, 10);
198 Hysteresis = (int)strtol(optarg, NULL, 10);
201 pollrate = strtod(optarg, NULL);
207 TriggerUp = (double)strtol(optarg, NULL, 10) / 100;
210 BatLifeMin = strtol(optarg, NULL, 10);
213 BatShutdownLingerSet = strtol(optarg, NULL, 10);
214 if (BatShutdownLingerSet < 0)
215 BatShutdownLingerSet = 0;
218 BatLifePollIntvl = strtol(optarg, NULL, 10);
221 BatShutdownAudioAlert = 0;
224 srt = strtod(optarg, NULL);
236 /* Get number of cpus */
239 if (0 > Hysteresis || Hysteresis > 99) {
240 fprintf(stderr, "Invalid hysteresis value\n");
244 if (0 > TriggerUp || TriggerUp > 1) {
245 fprintf(stderr, "Invalid load limit value\n");
249 TriggerDown = TriggerUp - (TriggerUp * (double) Hysteresis / 100);
252 * Make sure powerd is not already running.
254 PowerFd = open("/var/run/powerd.pid", O_CREAT|O_RDWR, 0644);
257 "Cannot create /var/run/powerd.pid, "
258 "continuing anyway\n");
260 if (flock(PowerFd, LOCK_EX|LOCK_NB) < 0) {
261 fprintf(stderr, "powerd is already running\n");
267 * Demonize and set pid
271 openlog("powerd", LOG_CONS | LOG_PID, LOG_DAEMON);
275 ftruncate(PowerFd, 0);
276 snprintf(buf, sizeof(buf), "%d\n", (int)getpid());
277 write(PowerFd, buf, strlen(buf));
280 /* Do we need to monitor battery life? */
281 if (BatLifePollIntvl <= 0)
284 monbat = has_battery();
286 /* Do we have perfbias(4)? */
288 HasPerfbias = has_perfbias();
290 /* Could we adjust C-state? */
292 AdjustCstate = probe_cstate();
295 * Wait hw.acpi.cpu.px_dom* sysctl to be created by kernel.
297 * Since hw.acpi.cpu.px_dom* creation is queued into ACPI
298 * taskqueue and ACPI taskqueue is shared across various
299 * ACPI modules, any delay in other modules may cause
300 * hw.acpi.cpu.px_dom* to be created at quite a later time
301 * (e.g. cmbat module's task could take quite a lot of time).
304 /* Prime delta cputime calculation. */
305 get_cputime(pollrate);
307 /* Wait for all cpus to appear */
308 if (acpi_get_cpupwrdom())
310 usleep((int)(pollrate * 1000000.0));
314 * Catch some signals so that max performance could be restored.
316 signal(SIGINT, sigintr);
317 signal(SIGTERM, sigintr);
319 /* Initialize performance states */
322 srt = srt / pollrate; /* convert to sample count */
324 printf("samples for downgrading: %5.2f\n", srt);
331 * Monitor performance
333 get_cputime(pollrate);
340 monbat = mon_battery();
342 usleep((int)(pollrate * 1000000.0));
346 * Set to maximum performance if killed.
348 syslog(LOG_INFO, "killed, setting max and exiting");
355 sigintr(int signo __unused)
361 * Figure out the cpu power domains.
364 acpi_get_cpupwrdom(void)
366 struct cpu_pwrdom *dom;
367 cpumask_t pwrdom_mask;
372 int n, i, ncpu = 0, dom_id;
374 memset(cpu2pwrdom, 0, sizeof(cpu2pwrdom));
375 memset(cpu_pwrdomain, 0, sizeof(cpu_pwrdomain));
376 CPUMASK_ASSZERO(cpu_pwrdom_mask);
378 for (i = 0; i < MAXDOM; ++i) {
379 snprintf(buf, sizeof(buf),
380 "hw.acpi.cpu.px_dom%d.available", i);
381 if (sysctlbyname(buf, NULL, NULL, NULL, 0) < 0)
384 dom = calloc(1, sizeof(*dom));
387 if (cpu_pwrdomain[i] != NULL) {
388 fprintf(stderr, "cpu power domain %d exists\n", i);
391 cpu_pwrdomain[i] = dom;
392 CPUMASK_ORBIT(cpu_pwrdom_mask, i);
394 pwrdom_mask = cpu_pwrdom_mask;
396 while (CPUMASK_TESTNZERO(pwrdom_mask)) {
397 dom_id = BSFCPUMASK(pwrdom_mask);
398 CPUMASK_NANDBIT(pwrdom_mask, dom_id);
399 dom = cpu_pwrdomain[dom_id];
401 CPUMASK_ASSZERO(dom->dom_cpumask);
403 snprintf(buf, sizeof(buf),
404 "hw.acpi.cpu.px_dom%d.members", dom->dom_id);
405 msize = sizeof(members);
406 if (sysctlbyname(buf, members, &msize, NULL, 0) < 0) {
407 cpu_pwrdomain[dom_id] = NULL;
413 for (str = strtok(members, " "); str; str = strtok(NULL, " ")) {
415 sscanf(str, "cpu%d", &n);
419 CPUMASK_ORBIT(dom->dom_cpumask, n);
420 cpu2pwrdom[n] = dom->dom_id;
423 if (dom->dom_ncpus == 0) {
424 cpu_pwrdomain[dom_id] = NULL;
429 printf("dom%d cpumask: ", dom->dom_id);
430 for (i = 0; i < (int)NELEM(dom->dom_cpumask.ary); ++i) {
432 (uintmax_t)dom->dom_cpumask.ary[i]);
440 printf("Found %d cpus, expecting %d\n", ncpu, NCpus);
442 pwrdom_mask = cpu_pwrdom_mask;
443 while (CPUMASK_TESTNZERO(pwrdom_mask)) {
444 dom_id = BSFCPUMASK(pwrdom_mask);
445 CPUMASK_NANDBIT(pwrdom_mask, dom_id);
446 dom = cpu_pwrdomain[dom_id];
456 * Save per-cpu load and sum of per-cpu load.
459 get_cputime(double pollrate)
461 static struct kinfo_cputime ocpu_time[MAXCPU];
462 static struct kinfo_cputime ncpu_time[MAXCPU];
468 bcopy(ncpu_time, ocpu_time, sizeof(struct kinfo_cputime) * NCpus);
470 slen = sizeof(ncpu_time);
471 if (sysctlbyname("kern.cputime", &ncpu_time, &slen, NULL, 0) < 0) {
472 fprintf(stderr, "kern.cputime sysctl not available\n");
475 ncpu = slen / sizeof(ncpu_time[0]);
478 for (cpu = 0; cpu < ncpu; ++cpu) {
481 d = (ncpu_time[cpu].cp_user + ncpu_time[cpu].cp_sys +
482 ncpu_time[cpu].cp_nice + ncpu_time[cpu].cp_intr) -
483 (ocpu_time[cpu].cp_user + ocpu_time[cpu].cp_sys +
484 ocpu_time[cpu].cp_nice + ocpu_time[cpu].cp_intr);
485 pcpu_state[cpu].cpu_qavg = (double)d / (pollrate * 1000000.0);
489 global_cpu_state.cpu_qavg = (double)delta / (pollrate * 1000000.0);
493 acpi_getcpufreq_str(int dom_id, int *highest0, int *lowest0)
495 char buf[256], sysid[64];
498 int v, highest, lowest;
501 * Retrieve availability list
503 snprintf(sysid, sizeof(sysid), "hw.acpi.cpu.px_dom%d.available",
505 buflen = sizeof(buf) - 1;
506 if (sysctlbyname(sysid, buf, &buflen, NULL, 0) < 0)
511 * Parse out the highest and lowest cpu frequencies
514 highest = lowest = 0;
515 while (ptr && (v = strtol(ptr, &ptr, 10)) > 0) {
516 if ((lowest == 0 || lowest > v) &&
517 (LowestCpuFreq <= 0 || v >= LowestCpuFreq))
519 if ((highest == 0 || highest < v) &&
520 (HighestCpuFreq <= 0 || v <= HighestCpuFreq))
525 if (!TurboOpt && highest - v == 1)
534 acpi_getcpufreq_bin(int dom_id, int *highest0, int *lowest0)
542 * Retrieve availability list
544 snprintf(sysid, sizeof(sysid), "hw.acpi.cpu.px_dom%d.avail", dom_id);
545 freqlen = sizeof(freq);
546 if (sysctlbyname(sysid, freq, &freqlen, NULL, 0) < 0)
549 freqcnt = freqlen / sizeof(freq[0]);
553 for (i = freqcnt - 1; i >= 0; --i) {
555 if (LowestCpuFreq <= 0 || *lowest0 >= LowestCpuFreq)
561 if (!TurboOpt && freqcnt > 1 && freq[0] - freq[1] == 1) {
565 for (; i < freqcnt; ++i) {
566 if (HighestCpuFreq <= 0 || *highest0 <= HighestCpuFreq)
574 acpi_get_cpufreq(int dom_id, int *highest, int *lowest)
579 if (acpi_getcpufreq_bin(dom_id, highest, lowest))
581 acpi_getcpufreq_str(dom_id, highest, lowest);
588 fprintf(stderr, "usage: powerd [-cdeftQ] [-p hysteresis] "
589 "[-h highest_freq] [-l lowest_freq] "
590 "[-r poll_interval] [-u trigger_up] "
591 "[-B min_battery_life] [-L low_battery_linger] "
592 "[-P battery_poll_interval] [-T sample_interval]\n");
597 #define timespecsub(vvp, uvp) \
599 (vvp)->tv_sec -= (uvp)->tv_sec; \
600 (vvp)->tv_nsec -= (uvp)->tv_nsec; \
601 if ((vvp)->tv_nsec < 0) { \
603 (vvp)->tv_nsec += 1000000000; \
608 #define BAT_SYSCTL_TIME_MAX 50000000 /* unit: nanosecond */
613 struct timespec s, e;
617 clock_gettime(CLOCK_MONOTONIC_FAST, &s);
621 if (sysctlbyname("hw.acpi.acline", &val, &len, NULL, 0) < 0) {
622 /* No AC line information */
625 clock_gettime(CLOCK_MONOTONIC_FAST, &e);
628 if (e.tv_sec > 0 || e.tv_nsec > BAT_SYSCTL_TIME_MAX) {
629 /* hw.acpi.acline takes to long to be useful */
630 syslog(LOG_NOTICE, "hw.acpi.acline takes too long");
634 clock_gettime(CLOCK_MONOTONIC_FAST, &s);
636 if (sysctlbyname("hw.acpi.battery.life", &val, &len, NULL, 0) < 0) {
637 /* No battery life */
640 clock_gettime(CLOCK_MONOTONIC_FAST, &e);
643 if (e.tv_sec > 0 || e.tv_nsec > BAT_SYSCTL_TIME_MAX) {
644 /* hw.acpi.battery.life takes to long to be useful */
645 syslog(LOG_NOTICE, "hw.acpi.battery.life takes too long");
652 low_battery_alert(int life)
654 int fmt, stereo, freq;
657 syslog(LOG_ALERT, "low battery life %d%%, please plugin AC line, #%d",
658 life, BatShutdownLingerCnt);
659 ++BatShutdownLingerCnt;
661 if (!BatShutdownAudioAlert)
664 fd = open("/dev/dsp", O_WRONLY);
669 if (ioctl(fd, SNDCTL_DSP_SETFMT, &fmt, sizeof(fmt)) < 0)
673 if (ioctl(fd, SNDCTL_DSP_STEREO, &stereo, sizeof(stereo)) < 0)
677 if (ioctl(fd, SNDCTL_DSP_SPEED, &freq, sizeof(freq)) < 0)
680 write(fd, alert1, sizeof(alert1));
681 write(fd, alert1, sizeof(alert1));
690 struct timespec cur, ts;
694 clock_gettime(CLOCK_MONOTONIC_FAST, &cur);
696 timespecsub(&ts, &BatLifePrevT);
697 if (ts.tv_sec < BatLifePollIntvl)
701 len = sizeof(acline);
702 if (sysctlbyname("hw.acpi.acline", &acline, &len, NULL, 0) < 0)
705 BatShutdownLinger = -1;
706 BatShutdownLingerCnt = 0;
711 if (sysctlbyname("hw.acpi.battery.life", &life, &len, NULL, 0) < 0)
714 if (BatShutdownLinger > 0) {
716 timespecsub(&ts, &BatShutdownStartT);
717 if (ts.tv_sec > BatShutdownLinger)
718 BatShutdownLinger = 0;
721 if (life <= BatLifeMin) {
722 if (BatShutdownLinger == 0 || BatShutdownLingerSet == 0) {
723 syslog(LOG_ALERT, "low battery life %d%%, "
724 "shutting down", life);
726 execlp("poweroff", "poweroff", NULL);
728 } else if (BatShutdownLinger < 0) {
729 BatShutdownLinger = BatShutdownLingerSet;
730 BatShutdownStartT = cur;
732 low_battery_alert(life);
742 slen = sizeof(NCpus);
743 if (sysctlbyname("hw.ncpu", &NCpus, &slen, NULL, 0) < 0)
744 err(1, "sysctlbyname hw.ncpu failed");
746 printf("hw.ncpu %d\n", NCpus);
754 slen = sizeof(usched_cpu_used);
755 if (sysctlbyname("kern.usched_global_cpumask", &usched_cpu_used, &slen,
757 err(1, "sysctlbyname kern.usched_global_cpumask failed");
761 printf("usched cpumask was: ");
762 for (i = 0; i < (int)NELEM(usched_cpu_used.ary); ++i)
763 printf("%jx ", (uintmax_t)usched_cpu_used.ary[i]);
774 printf("usched cpumask: ");
775 for (i = 0; i < (int)NELEM(usched_cpu_used.ary); ++i) {
777 (uintmax_t)usched_cpu_used.ary[i]);
781 sysctlbyname("kern.usched_global_cpumask", NULL, 0,
782 &usched_cpu_used, sizeof(usched_cpu_used));
792 if (sysctlbyname("machdep.perfbias0.hint", &hint, &len, NULL, 0) < 0)
798 set_perfbias(int cpu, int inc)
800 int hint = inc ? 0 : 15;
804 printf("cpu%d set perfbias hint %d\n", cpu, hint);
805 snprintf(sysid, sizeof(sysid), "machdep.perfbias%d.hint", cpu);
806 sysctlbyname(sysid, NULL, NULL, &hint, sizeof(hint));
812 struct cpu_state *state;
815 /* Get usched cpumask */
819 * Assume everything are used and are maxed out, before we
823 CPUMASK_ASSBMASK(cpu_used, NCpus);
824 cpu_pwrdom_used = cpu_pwrdom_mask;
825 global_pcpu_limit = NCpus;
827 for (cpu = 0; cpu < NCpus; ++cpu) {
828 state = &pcpu_state[cpu];
830 state->cpu_uavg = 0.0;
831 state->cpu_davg = 0.0;
832 state->cpu_limit = 1;
833 state->cpu_count = 1;
834 snprintf(state->cpu_name, sizeof(state->cpu_name), "cpu%d",
838 state = &global_cpu_state;
839 state->cpu_uavg = 0.0;
840 state->cpu_davg = 0.0;
841 state->cpu_limit = NCpus;
842 state->cpu_count = NCpus;
843 strlcpy(state->cpu_name, "global", sizeof(state->cpu_name));
847 get_nstate(struct cpu_state *state, double srt)
849 int ustate, dstate, nstate;
852 state->cpu_uavg = (state->cpu_uavg * 2.0 + state->cpu_qavg) / 3.0;
854 state->cpu_davg = (state->cpu_davg * srt + state->cpu_qavg) / (srt + 1);
855 if (state->cpu_davg < state->cpu_uavg)
856 state->cpu_davg = state->cpu_uavg;
858 ustate = state->cpu_uavg / TriggerUp;
859 if (ustate < state->cpu_limit)
860 ustate = state->cpu_uavg / TriggerDown;
861 dstate = state->cpu_davg / TriggerUp;
862 if (dstate < state->cpu_limit)
863 dstate = state->cpu_davg / TriggerDown;
865 nstate = (ustate > dstate) ? ustate : dstate;
866 if (nstate > state->cpu_count)
867 nstate = state->cpu_count;
870 printf("%s qavg=%5.2f uavg=%5.2f davg=%5.2f "
871 "%2d ncpus=%d\n", state->cpu_name,
872 state->cpu_qavg, state->cpu_uavg, state->cpu_davg,
873 state->cpu_limit, nstate);
881 cpumask_t ocpu_used, ocpu_pwrdom_used;
882 int pnstate = 0, nstate;
886 * Find cpus requiring performance and their cooresponding power
887 * domains. Save the number of cpus requiring performance in
890 ocpu_used = cpu_used;
891 ocpu_pwrdom_used = cpu_pwrdom_used;
893 CPUMASK_ASSZERO(cpu_used);
894 CPUMASK_ASSZERO(cpu_pwrdom_used);
896 for (cpu = 0; cpu < NCpus; ++cpu) {
897 struct cpu_state *state = &pcpu_state[cpu];
900 s = get_nstate(state, srt);
902 CPUMASK_ORBIT(cpu_used, cpu);
903 CPUMASK_ORBIT(cpu_pwrdom_used, cpu2pwrdom[cpu]);
907 state->cpu_limit = s;
911 * Calculate nstate, the number of cpus we wish to run at max
914 nstate = get_nstate(&global_cpu_state, srt);
916 if (nstate == global_cpu_state.cpu_limit &&
917 (pnstate == global_pcpu_limit || nstate > pnstate)) {
918 /* Nothing changed; keep the sets */
919 cpu_used = ocpu_used;
920 cpu_pwrdom_used = ocpu_pwrdom_used;
922 global_pcpu_limit = pnstate;
925 global_pcpu_limit = pnstate;
927 if (nstate > pnstate) {
929 * Add spare cpus to meet global performance requirement.
931 add_spare_cpus(ocpu_used, nstate - pnstate);
934 global_cpu_state.cpu_limit = nstate;
937 * Adjust cpu and cpu power domain performance
939 adj_perf(ocpu_used, ocpu_pwrdom_used);
943 add_spare_cpus(const cpumask_t ocpu_used, int ncpu)
945 cpumask_t saved_pwrdom, xcpu_used;
949 * Find more cpus in the previous cpu set.
951 xcpu_used = cpu_used;
952 CPUMASK_XORMASK(xcpu_used, ocpu_used);
953 while (CPUMASK_TESTNZERO(xcpu_used)) {
954 cpu = BSFCPUMASK(xcpu_used);
955 CPUMASK_NANDBIT(xcpu_used, cpu);
957 if (CPUMASK_TESTBIT(ocpu_used, cpu)) {
958 CPUMASK_ORBIT(cpu_pwrdom_used, cpu2pwrdom[cpu]);
959 CPUMASK_ORBIT(cpu_used, cpu);
967 * Find more cpus in the used cpu power domains.
969 saved_pwrdom = cpu_pwrdom_used;
971 while (CPUMASK_TESTNZERO(saved_pwrdom)) {
972 cpumask_t unused_cpumask;
975 dom = BSFCPUMASK(saved_pwrdom);
976 CPUMASK_NANDBIT(saved_pwrdom, dom);
978 unused_cpumask = cpu_pwrdomain[dom]->dom_cpumask;
979 CPUMASK_NANDMASK(unused_cpumask, cpu_used);
981 while (CPUMASK_TESTNZERO(unused_cpumask)) {
982 cpu = BSFCPUMASK(unused_cpumask);
983 CPUMASK_NANDBIT(unused_cpumask, cpu);
985 CPUMASK_ORBIT(cpu_pwrdom_used, dom);
986 CPUMASK_ORBIT(cpu_used, cpu);
995 * Find more cpus in unused cpu power domains
997 saved_pwrdom = cpu_pwrdom_mask;
998 CPUMASK_NANDMASK(saved_pwrdom, cpu_pwrdom_used);
1002 printf("%d cpus not found\n", ncpu);
1006 acpi_set_cpufreq(int dom, int inc)
1008 int lowest, highest, desired;
1011 acpi_get_cpufreq(dom, &highest, &lowest);
1012 if (highest == 0 || lowest == 0)
1014 desired = inc ? highest : lowest;
1017 printf("dom%d set frequency %d\n", dom, desired);
1018 snprintf(sysid, sizeof(sysid), "hw.acpi.cpu.px_dom%d.select", dom);
1019 sysctlbyname(sysid, NULL, NULL, &desired, sizeof(desired));
1023 adj_cpu_pwrdom(int dom, int inc)
1026 acpi_set_cpufreq(dom, inc);
1030 adj_cpu_perf(int cpu, int inc)
1034 printf("cpu%d increase perf\n", cpu);
1036 printf("cpu%d decrease perf\n", cpu);
1040 set_perfbias(cpu, inc);
1042 set_cstate(cpu, inc);
1046 adj_perf(cpumask_t xcpu_used, cpumask_t xcpu_pwrdom_used)
1048 cpumask_t old_usched_used;
1052 * Set cpus requiring performance to the userland process
1053 * scheduler. Leave the rest of cpus unmapped.
1055 old_usched_used = usched_cpu_used;
1056 usched_cpu_used = cpu_used;
1057 if (CPUMASK_TESTZERO(usched_cpu_used))
1058 CPUMASK_ORBIT(usched_cpu_used, 0);
1059 if (CPUMASK_CMPMASKNEQ(usched_cpu_used, old_usched_used))
1063 * Adjust per-cpu performance.
1065 CPUMASK_XORMASK(xcpu_used, cpu_used);
1066 while (CPUMASK_TESTNZERO(xcpu_used)) {
1067 cpu = BSFCPUMASK(xcpu_used);
1068 CPUMASK_NANDBIT(xcpu_used, cpu);
1070 if (CPUMASK_TESTBIT(cpu_used, cpu)) {
1071 /* Increase cpu performance */
1074 /* Decrease cpu performance */
1077 adj_cpu_perf(cpu, inc);
1081 * Adjust cpu power domain performance. This could affect
1084 CPUMASK_XORMASK(xcpu_pwrdom_used, cpu_pwrdom_used);
1085 while (CPUMASK_TESTNZERO(xcpu_pwrdom_used)) {
1088 dom = BSFCPUMASK(xcpu_pwrdom_used);
1089 CPUMASK_NANDBIT(xcpu_pwrdom_used, dom);
1091 if (CPUMASK_TESTBIT(cpu_pwrdom_used, dom)) {
1092 /* Increase cpu power domain performance */
1095 /* Decrease cpu power domain performance */
1098 adj_cpu_pwrdom(dom, inc);
1105 cpumask_t ocpu_used, ocpu_pwrdom_used;
1107 /* Remove highest cpu frequency limitation */
1110 ocpu_used = cpu_used;
1111 ocpu_pwrdom_used = cpu_pwrdom_used;
1113 /* Max out all cpus and cpu power domains performance */
1114 CPUMASK_ASSBMASK(cpu_used, NCpus);
1115 cpu_pwrdom_used = cpu_pwrdom_mask;
1117 adj_perf(ocpu_used, ocpu_pwrdom_used);
1121 * Restore the original mwait C-state
1124 printf("global set cstate %s\n", orig_global_cx);
1125 sysctlbyname("machdep.mwait.CX.idle", NULL, NULL,
1126 orig_global_cx, strlen(orig_global_cx) + 1);
1133 char cx_supported[1024];
1136 int idle_hlt, deep = 1;
1139 len = sizeof(idle_hlt);
1140 if (sysctlbyname("machdep.cpu_idle_hlt", &idle_hlt, &len, NULL, 0) < 0)
1145 len = sizeof(cx_supported);
1146 if (sysctlbyname("machdep.mwait.CX.supported", cx_supported, &len,
1150 len = sizeof(orig_global_cx);
1151 if (sysctlbyname("machdep.mwait.CX.idle", orig_global_cx, &len,
1155 strlcpy(cpu_perf_cx, "AUTODEEP", sizeof(cpu_perf_cx));
1156 cpu_perf_cxlen = strlen(cpu_perf_cx) + 1;
1157 if (sysctlbyname("machdep.mwait.CX.idle", NULL, NULL,
1158 cpu_perf_cx, cpu_perf_cxlen) < 0) {
1159 /* AUTODEEP is not supported; try AUTO */
1161 strlcpy(cpu_perf_cx, "AUTO", sizeof(cpu_perf_cx));
1162 cpu_perf_cxlen = strlen(cpu_perf_cx) + 1;
1163 if (sysctlbyname("machdep.mwait.CX.idle", NULL, NULL,
1164 cpu_perf_cx, cpu_perf_cxlen) < 0)
1172 for (ptr = strtok(cx_supported, " "); ptr != NULL;
1173 ptr = strtok(NULL, " ")) {
1174 if (target == NULL ||
1175 (target != NULL && strcmp(ptr, target) == 0)) {
1176 strlcpy(cpu_idle_cx, ptr, sizeof(cpu_idle_cx));
1177 cpu_idle_cxlen = strlen(cpu_idle_cx) + 1;
1182 if (cpu_idle_cxlen == 0)
1186 printf("cstate orig %s, perf %s, idle %s\n",
1187 orig_global_cx, cpu_perf_cx, cpu_idle_cx);
1193 set_cstate(int cpu, int inc)
1201 len = cpu_perf_cxlen;
1204 len = cpu_idle_cxlen;
1208 printf("cpu%d set cstate %s\n", cpu, cst);
1209 snprintf(sysid, sizeof(sysid), "machdep.mwait.CX.idle%d", cpu);
1210 sysctlbyname(sysid, NULL, NULL, cst, len);