diff options
Diffstat (limited to 'tools/power/cpupower/utils/idle_monitor')
8 files changed, 414 insertions, 103 deletions
diff --git a/tools/power/cpupower/utils/idle_monitor/amd_fam14h_idle.c b/tools/power/cpupower/utils/idle_monitor/amd_fam14h_idle.c index 202e555988b..2116df9ad83 100644 --- a/tools/power/cpupower/utils/idle_monitor/amd_fam14h_idle.c +++ b/tools/power/cpupower/utils/idle_monitor/amd_fam14h_idle.c @@ -20,8 +20,6 @@ #include "idle_monitor/cpupower-monitor.h" #include "helpers/helpers.h" -/******** PCI parts could go into own file and get shared ***************/ - #define PCI_NON_PC0_OFFSET 0xb0 #define PCI_PC1_OFFSET 0xb4 #define PCI_PC6_OFFSET 0xb8 @@ -82,10 +80,7 @@ static cstate_t amd_fam14h_cstates[AMD_FAM14H_STATE_NUM] = { }; static struct pci_access *pci_acc; -static int pci_vendor_id = 0x1022; -static int pci_dev_ids[2] = {0x1716, 0}; static struct pci_dev *amd_fam14h_pci_dev; - static int nbp1_entered; struct timespec start_time; @@ -286,13 +281,13 @@ struct cpuidle_monitor *amd_fam14h_register(void) if (cpupower_cpu_info.vendor != X86_VENDOR_AMD) return NULL; - if (cpupower_cpu_info.family == 0x14) { - if (cpu_count <= 0 || cpu_count > 2) { - fprintf(stderr, "AMD fam14h: Invalid cpu count: %d\n", - cpu_count); - return NULL; - } - } else + if (cpupower_cpu_info.family == 0x14) + strncpy(amd_fam14h_monitor.name, "Fam_14h", + MONITOR_NAME_LEN - 1); + else if (cpupower_cpu_info.family == 0x12) + strncpy(amd_fam14h_monitor.name, "Fam_12h", + MONITOR_NAME_LEN - 1); + else return NULL; /* We do not alloc for nbp1 machine wide counter */ @@ -303,7 +298,9 @@ struct cpuidle_monitor *amd_fam14h_register(void) sizeof(unsigned long long)); } - amd_fam14h_pci_dev = pci_acc_init(&pci_acc, pci_vendor_id, pci_dev_ids); + /* We need PCI device: Slot 18, Func 6, compare with BKDG + for fam 12h/14h */ + amd_fam14h_pci_dev = pci_slot_func_init(&pci_acc, 0x18, 6); if (amd_fam14h_pci_dev == NULL || pci_acc == NULL) return NULL; @@ -325,7 +322,7 @@ static void amd_fam14h_unregister(void) } struct cpuidle_monitor amd_fam14h_monitor = { - .name = "Ontario", + .name = "", .hw_states = amd_fam14h_cstates, .hw_states_num = AMD_FAM14H_STATE_NUM, .start = amd_fam14h_start, diff --git a/tools/power/cpupower/utils/idle_monitor/cpuidle_sysfs.c b/tools/power/cpupower/utils/idle_monitor/cpuidle_sysfs.c index d048b96a615..bcd22a1a397 100644 --- a/tools/power/cpupower/utils/idle_monitor/cpuidle_sysfs.c +++ b/tools/power/cpupower/utils/idle_monitor/cpuidle_sysfs.c @@ -134,7 +134,7 @@ static struct cpuidle_monitor *cpuidle_register(void) /* Assume idle state count is the same for all CPUs */ cpuidle_sysfs_monitor.hw_states_num = sysfs_get_idlestate_count(0); - if (cpuidle_sysfs_monitor.hw_states_num == 0) + if (cpuidle_sysfs_monitor.hw_states_num <= 0) return NULL; for (num = 0; num < cpuidle_sysfs_monitor.hw_states_num; num++) { diff --git a/tools/power/cpupower/utils/idle_monitor/cpupower-monitor.c b/tools/power/cpupower/utils/idle_monitor/cpupower-monitor.c index ba4bf068380..c4bae9203a6 100644 --- a/tools/power/cpupower/utils/idle_monitor/cpupower-monitor.c +++ b/tools/power/cpupower/utils/idle_monitor/cpupower-monitor.c @@ -39,10 +39,17 @@ static int mode; static int interval = 1; static char *show_monitors_param; static struct cpupower_topology cpu_top; +static unsigned int wake_cpus; /* ToDo: Document this in the manpage */ static char range_abbr[RANGE_MAX] = { 'T', 'C', 'P', 'M', }; +static void print_wrong_arg_exit(void) +{ + printf(_("invalid or unknown argument\n")); + exit(EXIT_FAILURE); +} + long long timespec_diff_us(struct timespec start, struct timespec end) { struct timespec temp; @@ -56,21 +63,6 @@ long long timespec_diff_us(struct timespec start, struct timespec end) return (temp.tv_sec * 1000000) + (temp.tv_nsec / 1000); } -void monitor_help(void) -{ - printf(_("cpupower monitor: [-m <mon1>,[<mon2>],.. ] command\n")); - printf(_("cpupower monitor: [-m <mon1>,[<mon2>],.. ] [ -i interval_sec ]\n")); - printf(_("cpupower monitor: -l\n")); - printf(_("\t command: pass an arbitrary command to measure specific workload\n")); - printf(_("\t -i: time intervall to measure for in seconds (default 1)\n")); - printf(_("\t -l: list available CPU sleep monitors (for use with -m)\n")); - printf(_("\t -m: show specific CPU sleep monitors only (in same order)\n")); - printf(_("\t -h: print this help\n")); - printf("\n"); - printf(_("only one of: -l, -m are allowed\nIf none of them is passed,")); - printf(_(" all supported monitors are shown\n")); -} - void print_n_spaces(int n) { int x; @@ -93,7 +85,7 @@ int fill_string_with_spaces(char *s, int n) void print_header(int topology_depth) { int unsigned mon; - int state, need_len, pr_mon_len; + int state, need_len; cstate_t s; char buf[128] = ""; int percent_width = 4; @@ -102,7 +94,6 @@ void print_header(int topology_depth) printf("%s|", buf); for (mon = 0; mon < avail_monitors; mon++) { - pr_mon_len = 0; need_len = monitors[mon]->hw_states_num * (percent_width + 3) - 1; if (mon != 0) { @@ -149,6 +140,10 @@ void print_results(int topology_depth, int cpu) unsigned long long result; cstate_t s; + /* Be careful CPUs may got resorted for pkg value do not just use cpu */ + if (!bitmask_isbitset(cpus_chosen, cpu_top.core_info[cpu].cpu)) + return; + if (topology_depth > 2) printf("%4d|", cpu_top.core_info[cpu].pkg); if (topology_depth > 1) @@ -190,9 +185,13 @@ void print_results(int topology_depth, int cpu) } } } - /* cpu offline */ - if (cpu_top.core_info[cpu].pkg == -1 || - cpu_top.core_info[cpu].core == -1) { + /* + * The monitor could still provide useful data, for example + * AMD HW counters partly sit in PCI config space. + * It's up to the monitor plug-in to check .is_online, this one + * is just for additional info. + */ + if (!cpu_top.core_info[cpu].is_online) { printf(_(" *is offline\n")); return; } else @@ -238,7 +237,6 @@ static void parse_monitor_param(char *param) if (hits == 0) { printf(_("No matching monitor found in %s, " "try -l option\n"), param); - monitor_help(); exit(EXIT_FAILURE); } /* Override detected/registerd monitors array with requested one */ @@ -317,16 +315,28 @@ int fork_it(char **argv) int do_interval_measure(int i) { unsigned int num; + int cpu; + + if (wake_cpus) + for (cpu = 0; cpu < cpu_count; cpu++) + bind_cpu(cpu); for (num = 0; num < avail_monitors; num++) { dprint("HW C-state residency monitor: %s - States: %d\n", monitors[num]->name, monitors[num]->hw_states_num); monitors[num]->start(); } + sleep(i); + + if (wake_cpus) + for (cpu = 0; cpu < cpu_count; cpu++) + bind_cpu(cpu); + for (num = 0; num < avail_monitors; num++) monitors[num]->stop(); + return 0; } @@ -335,37 +345,30 @@ static void cmdline(int argc, char *argv[]) int opt; progname = basename(argv[0]); - while ((opt = getopt(argc, argv, "+hli:m:")) != -1) { + while ((opt = getopt(argc, argv, "+lci:m:")) != -1) { switch (opt) { - case 'h': - monitor_help(); - exit(EXIT_SUCCESS); case 'l': - if (mode) { - monitor_help(); - exit(EXIT_FAILURE); - } + if (mode) + print_wrong_arg_exit(); mode = list; break; case 'i': /* only allow -i with -m or no option */ - if (mode && mode != show) { - monitor_help(); - exit(EXIT_FAILURE); - } + if (mode && mode != show) + print_wrong_arg_exit(); interval = atoi(optarg); break; case 'm': - if (mode) { - monitor_help(); - exit(EXIT_FAILURE); - } + if (mode) + print_wrong_arg_exit(); mode = show; show_monitors_param = optarg; break; + case 'c': + wake_cpus = 1; + break; default: - monitor_help(); - exit(EXIT_FAILURE); + print_wrong_arg_exit(); } } if (!mode) @@ -385,6 +388,10 @@ int cmd_monitor(int argc, char **argv) return EXIT_FAILURE; } + /* Default is: monitor all CPUs */ + if (bitmask_isallclear(cpus_chosen)) + bitmask_setall(cpus_chosen); + dprint("System has up to %d CPU cores\n", cpu_count); for (num = 0; all_monitors[num]; num++) { diff --git a/tools/power/cpupower/utils/idle_monitor/cpupower-monitor.h b/tools/power/cpupower/utils/idle_monitor/cpupower-monitor.h index 9312ee1f2db..9e43f3371fb 100644 --- a/tools/power/cpupower/utils/idle_monitor/cpupower-monitor.h +++ b/tools/power/cpupower/utils/idle_monitor/cpupower-monitor.h @@ -65,4 +65,21 @@ extern long long timespec_diff_us(struct timespec start, struct timespec end); "could be inaccurate\n"), mes, ov); \ } + +/* Taken over from x86info project sources -> return 0 on success */ +#include <sched.h> +#include <sys/types.h> +#include <unistd.h> +static inline int bind_cpu(int cpu) +{ + cpu_set_t set; + + if (sched_getaffinity(getpid(), sizeof(set), &set) == 0) { + CPU_ZERO(&set); + CPU_SET(cpu, &set); + return sched_setaffinity(getpid(), sizeof(set), &set); + } + return 1; +} + #endif /* __CPUIDLE_INFO_HW__ */ diff --git a/tools/power/cpupower/utils/idle_monitor/hsw_ext_idle.c b/tools/power/cpupower/utils/idle_monitor/hsw_ext_idle.c new file mode 100644 index 00000000000..ebeaba6571a --- /dev/null +++ b/tools/power/cpupower/utils/idle_monitor/hsw_ext_idle.c @@ -0,0 +1,196 @@ +/* + * (C) 2010,2011 Thomas Renninger <trenn@suse.de>, Novell Inc. + * + * Licensed under the terms of the GNU GPL License version 2. + * + * Based on SandyBridge monitor. Implements the new package C-states + * (PC8, PC9, PC10) coming with a specific Haswell (family 0x45) CPU. + */ + +#if defined(__i386__) || defined(__x86_64__) + +#include <stdio.h> +#include <stdint.h> +#include <stdlib.h> +#include <string.h> + +#include "helpers/helpers.h" +#include "idle_monitor/cpupower-monitor.h" + +#define MSR_PKG_C8_RESIDENCY 0x00000630 +#define MSR_PKG_C9_RESIDENCY 0x00000631 +#define MSR_PKG_C10_RESIDENCY 0x00000632 + +#define MSR_TSC 0x10 + +enum intel_hsw_ext_id { PC8 = 0, PC9, PC10, HSW_EXT_CSTATE_COUNT, + TSC = 0xFFFF }; + +static int hsw_ext_get_count_percent(unsigned int self_id, double *percent, + unsigned int cpu); + +static cstate_t hsw_ext_cstates[HSW_EXT_CSTATE_COUNT] = { + { + .name = "PC8", + .desc = N_("Processor Package C8"), + .id = PC8, + .range = RANGE_PACKAGE, + .get_count_percent = hsw_ext_get_count_percent, + }, + { + .name = "PC9", + .desc = N_("Processor Package C9"), + .desc = N_("Processor Package C2"), + .id = PC9, + .range = RANGE_PACKAGE, + .get_count_percent = hsw_ext_get_count_percent, + }, + { + .name = "PC10", + .desc = N_("Processor Package C10"), + .id = PC10, + .range = RANGE_PACKAGE, + .get_count_percent = hsw_ext_get_count_percent, + }, +}; + +static unsigned long long tsc_at_measure_start; +static unsigned long long tsc_at_measure_end; +static unsigned long long *previous_count[HSW_EXT_CSTATE_COUNT]; +static unsigned long long *current_count[HSW_EXT_CSTATE_COUNT]; +/* valid flag for all CPUs. If a MSR read failed it will be zero */ +static int *is_valid; + +static int hsw_ext_get_count(enum intel_hsw_ext_id id, unsigned long long *val, + unsigned int cpu) +{ + int msr; + + switch (id) { + case PC8: + msr = MSR_PKG_C8_RESIDENCY; + break; + case PC9: + msr = MSR_PKG_C9_RESIDENCY; + break; + case PC10: + msr = MSR_PKG_C10_RESIDENCY; + break; + case TSC: + msr = MSR_TSC; + break; + default: + return -1; + }; + if (read_msr(cpu, msr, val)) + return -1; + return 0; +} + +static int hsw_ext_get_count_percent(unsigned int id, double *percent, + unsigned int cpu) +{ + *percent = 0.0; + + if (!is_valid[cpu]) + return -1; + + *percent = (100.0 * + (current_count[id][cpu] - previous_count[id][cpu])) / + (tsc_at_measure_end - tsc_at_measure_start); + + dprint("%s: previous: %llu - current: %llu - (%u)\n", + hsw_ext_cstates[id].name, previous_count[id][cpu], + current_count[id][cpu], cpu); + + dprint("%s: tsc_diff: %llu - count_diff: %llu - percent: %2.f (%u)\n", + hsw_ext_cstates[id].name, + (unsigned long long) tsc_at_measure_end - tsc_at_measure_start, + current_count[id][cpu] - previous_count[id][cpu], + *percent, cpu); + + return 0; +} + +static int hsw_ext_start(void) +{ + int num, cpu; + unsigned long long val; + + for (num = 0; num < HSW_EXT_CSTATE_COUNT; num++) { + for (cpu = 0; cpu < cpu_count; cpu++) { + hsw_ext_get_count(num, &val, cpu); + previous_count[num][cpu] = val; + } + } + hsw_ext_get_count(TSC, &tsc_at_measure_start, 0); + return 0; +} + +static int hsw_ext_stop(void) +{ + unsigned long long val; + int num, cpu; + + hsw_ext_get_count(TSC, &tsc_at_measure_end, 0); + + for (num = 0; num < HSW_EXT_CSTATE_COUNT; num++) { + for (cpu = 0; cpu < cpu_count; cpu++) { + is_valid[cpu] = !hsw_ext_get_count(num, &val, cpu); + current_count[num][cpu] = val; + } + } + return 0; +} + +struct cpuidle_monitor intel_hsw_ext_monitor; + +static struct cpuidle_monitor *hsw_ext_register(void) +{ + int num; + + if (cpupower_cpu_info.vendor != X86_VENDOR_INTEL + || cpupower_cpu_info.family != 6) + return NULL; + + switch (cpupower_cpu_info.model) { + case 0x45: /* HSW */ + break; + default: + return NULL; + } + + is_valid = calloc(cpu_count, sizeof(int)); + for (num = 0; num < HSW_EXT_CSTATE_COUNT; num++) { + previous_count[num] = calloc(cpu_count, + sizeof(unsigned long long)); + current_count[num] = calloc(cpu_count, + sizeof(unsigned long long)); + } + intel_hsw_ext_monitor.name_len = strlen(intel_hsw_ext_monitor.name); + return &intel_hsw_ext_monitor; +} + +void hsw_ext_unregister(void) +{ + int num; + free(is_valid); + for (num = 0; num < HSW_EXT_CSTATE_COUNT; num++) { + free(previous_count[num]); + free(current_count[num]); + } +} + +struct cpuidle_monitor intel_hsw_ext_monitor = { + .name = "HaswellExtended", + .hw_states = hsw_ext_cstates, + .hw_states_num = HSW_EXT_CSTATE_COUNT, + .start = hsw_ext_start, + .stop = hsw_ext_stop, + .do_register = hsw_ext_register, + .unregister = hsw_ext_unregister, + .needs_root = 1, + .overflow_s = 922000000 /* 922337203 seconds TSC overflow + at 20GHz */ +}; +#endif /* defined(__i386__) || defined(__x86_64__) */ diff --git a/tools/power/cpupower/utils/idle_monitor/idle_monitors.def b/tools/power/cpupower/utils/idle_monitor/idle_monitors.def index e3f8d9b2b18..0d6ba4dbb9c 100644 --- a/tools/power/cpupower/utils/idle_monitor/idle_monitors.def +++ b/tools/power/cpupower/utils/idle_monitor/idle_monitors.def @@ -2,6 +2,7 @@ DEF(amd_fam14h) DEF(intel_nhm) DEF(intel_snb) +DEF(intel_hsw_ext) DEF(mperf) #endif DEF(cpuidle_sysfs) diff --git a/tools/power/cpupower/utils/idle_monitor/mperf_monitor.c b/tools/power/cpupower/utils/idle_monitor/mperf_monitor.c index 63ca87a05e5..5650ab5a2c2 100644 --- a/tools/power/cpupower/utils/idle_monitor/mperf_monitor.c +++ b/tools/power/cpupower/utils/idle_monitor/mperf_monitor.c @@ -22,12 +22,15 @@ #define MSR_TSC 0x10 +#define MSR_AMD_HWCR 0xc0010015 + enum mperf_id { C0 = 0, Cx, AVG_FREQ, MPERF_CSTATE_COUNT }; static int mperf_get_count_percent(unsigned int self_id, double *percent, unsigned int cpu); static int mperf_get_count_freq(unsigned int id, unsigned long long *count, unsigned int cpu); +static struct timespec time_start, time_end; static cstate_t mperf_cstates[MPERF_CSTATE_COUNT] = { { @@ -54,19 +57,33 @@ static cstate_t mperf_cstates[MPERF_CSTATE_COUNT] = { }, }; +enum MAX_FREQ_MODE { MAX_FREQ_SYSFS, MAX_FREQ_TSC_REF }; +static int max_freq_mode; +/* + * The max frequency mperf is ticking at (in C0), either retrieved via: + * 1) calculated after measurements if we know TSC ticks at mperf/P0 frequency + * 2) cpufreq /sys/devices/.../cpu0/cpufreq/cpuinfo_max_freq at init time + * 1. Is preferred as it also works without cpufreq subsystem (e.g. on Xen) + */ +static unsigned long max_frequency; + static unsigned long long tsc_at_measure_start; static unsigned long long tsc_at_measure_end; -static unsigned long max_frequency; static unsigned long long *mperf_previous_count; static unsigned long long *aperf_previous_count; static unsigned long long *mperf_current_count; static unsigned long long *aperf_current_count; + /* valid flag for all CPUs. If a MSR read failed it will be zero */ static int *is_valid; static int mperf_get_tsc(unsigned long long *tsc) { - return read_msr(0, MSR_TSC, tsc); + int ret; + ret = read_msr(0, MSR_TSC, tsc); + if (ret) + dprint("Reading TSC MSR failed, returning %llu\n", *tsc); + return ret; } static int mperf_init_stats(unsigned int cpu) @@ -97,36 +114,11 @@ static int mperf_measure_stats(unsigned int cpu) return 0; } -/* - * get_average_perf() - * - * Returns the average performance (also considers boosted frequencies) - * - * Input: - * aperf_diff: Difference of the aperf register over a time period - * mperf_diff: Difference of the mperf register over the same time period - * max_freq: Maximum frequency (P0) - * - * Returns: - * Average performance over the time period - */ -static unsigned long get_average_perf(unsigned long long aperf_diff, - unsigned long long mperf_diff) -{ - unsigned int perf_percent = 0; - if (((unsigned long)(-1) / 100) < aperf_diff) { - int shift_count = 7; - aperf_diff >>= shift_count; - mperf_diff >>= shift_count; - } - perf_percent = (aperf_diff * 100) / mperf_diff; - return (max_frequency * perf_percent) / 100; -} - static int mperf_get_count_percent(unsigned int id, double *percent, unsigned int cpu) { unsigned long long aperf_diff, mperf_diff, tsc_diff; + unsigned long long timediff; if (!is_valid[cpu]) return -1; @@ -136,11 +128,19 @@ static int mperf_get_count_percent(unsigned int id, double *percent, mperf_diff = mperf_current_count[cpu] - mperf_previous_count[cpu]; aperf_diff = aperf_current_count[cpu] - aperf_previous_count[cpu]; - tsc_diff = tsc_at_measure_end - tsc_at_measure_start; - *percent = 100.0 * mperf_diff / tsc_diff; - dprint("%s: mperf_diff: %llu, tsc_diff: %llu\n", - mperf_cstates[id].name, mperf_diff, tsc_diff); + if (max_freq_mode == MAX_FREQ_TSC_REF) { + tsc_diff = tsc_at_measure_end - tsc_at_measure_start; + *percent = 100.0 * mperf_diff / tsc_diff; + dprint("%s: TSC Ref - mperf_diff: %llu, tsc_diff: %llu\n", + mperf_cstates[id].name, mperf_diff, tsc_diff); + } else if (max_freq_mode == MAX_FREQ_SYSFS) { + timediff = timespec_diff_us(time_start, time_end); + *percent = 100.0 * mperf_diff / timediff; + dprint("%s: MAXFREQ - mperf_diff: %llu, time_diff: %llu\n", + mperf_cstates[id].name, mperf_diff, timediff); + } else + return -1; if (id == Cx) *percent = 100.0 - *percent; @@ -154,7 +154,7 @@ static int mperf_get_count_percent(unsigned int id, double *percent, static int mperf_get_count_freq(unsigned int id, unsigned long long *count, unsigned int cpu) { - unsigned long long aperf_diff, mperf_diff; + unsigned long long aperf_diff, mperf_diff, time_diff, tsc_diff; if (id != AVG_FREQ) return 1; @@ -165,11 +165,21 @@ static int mperf_get_count_freq(unsigned int id, unsigned long long *count, mperf_diff = mperf_current_count[cpu] - mperf_previous_count[cpu]; aperf_diff = aperf_current_count[cpu] - aperf_previous_count[cpu]; - /* Return MHz for now, might want to return KHz if column width is more - generic */ - *count = get_average_perf(aperf_diff, mperf_diff) / 1000; - dprint("%s: %llu\n", mperf_cstates[id].name, *count); + if (max_freq_mode == MAX_FREQ_TSC_REF) { + /* Calculate max_freq from TSC count */ + tsc_diff = tsc_at_measure_end - tsc_at_measure_start; + time_diff = timespec_diff_us(time_start, time_end); + max_frequency = tsc_diff / time_diff; + } + *count = max_frequency * ((double)aperf_diff / mperf_diff); + dprint("%s: Average freq based on %s maximum frequency:\n", + mperf_cstates[id].name, + (max_freq_mode == MAX_FREQ_TSC_REF) ? "TSC calculated" : "sysfs read"); + dprint("%max_frequency: %lu", max_frequency); + dprint("aperf_diff: %llu\n", aperf_diff); + dprint("mperf_diff: %llu\n", mperf_diff); + dprint("avg freq: %llu\n", *count); return 0; } @@ -178,6 +188,7 @@ static int mperf_start(void) int cpu; unsigned long long dbg; + clock_gettime(CLOCK_REALTIME, &time_start); mperf_get_tsc(&tsc_at_measure_start); for (cpu = 0; cpu < cpu_count; cpu++) @@ -193,32 +204,104 @@ static int mperf_stop(void) unsigned long long dbg; int cpu; - mperf_get_tsc(&tsc_at_measure_end); - for (cpu = 0; cpu < cpu_count; cpu++) mperf_measure_stats(cpu); + mperf_get_tsc(&tsc_at_measure_end); + clock_gettime(CLOCK_REALTIME, &time_end); + mperf_get_tsc(&dbg); dprint("TSC diff: %llu\n", dbg - tsc_at_measure_end); return 0; } -struct cpuidle_monitor mperf_monitor; - -struct cpuidle_monitor *mperf_register(void) +/* + * Mperf register is defined to tick at P0 (maximum) frequency + * + * Instead of reading out P0 which can be tricky to read out from HW, + * we use TSC counter if it reliably ticks at P0/mperf frequency. + * + * Still try to fall back to: + * /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_max_freq + * on older Intel HW without invariant TSC feature. + * Or on AMD machines where TSC does not tick at P0 (do not exist yet, but + * it's still double checked (MSR_AMD_HWCR)). + * + * On these machines the user would still get useful mperf + * stats when acpi-cpufreq driver is loaded. + */ +static int init_maxfreq_mode(void) { + int ret; + unsigned long long hwcr; unsigned long min; - if (!(cpupower_cpu_info.caps & CPUPOWER_CAP_APERF)) - return NULL; - - /* Assume min/max all the same on all cores */ + if (!cpupower_cpu_info.caps & CPUPOWER_CAP_INV_TSC) + goto use_sysfs; + + if (cpupower_cpu_info.vendor == X86_VENDOR_AMD) { + /* MSR_AMD_HWCR tells us whether TSC runs at P0/mperf + * freq. + * A test whether hwcr is accessable/available would be: + * (cpupower_cpu_info.family > 0x10 || + * cpupower_cpu_info.family == 0x10 && + * cpupower_cpu_info.model >= 0x2)) + * This should be the case for all aperf/mperf + * capable AMD machines and is therefore safe to test here. + * Compare with Linus kernel git commit: acf01734b1747b1ec4 + */ + ret = read_msr(0, MSR_AMD_HWCR, &hwcr); + /* + * If the MSR read failed, assume a Xen system that did + * not explicitly provide access to it and assume TSC works + */ + if (ret != 0) { + dprint("TSC read 0x%x failed - assume TSC working\n", + MSR_AMD_HWCR); + return 0; + } else if (1 & (hwcr >> 24)) { + max_freq_mode = MAX_FREQ_TSC_REF; + return 0; + } else { /* Use sysfs max frequency if available */ } + } else if (cpupower_cpu_info.vendor == X86_VENDOR_INTEL) { + /* + * On Intel we assume mperf (in C0) is ticking at same + * rate than TSC + */ + max_freq_mode = MAX_FREQ_TSC_REF; + return 0; + } +use_sysfs: if (cpufreq_get_hardware_limits(0, &min, &max_frequency)) { dprint("Cannot retrieve max freq from cpufreq kernel " "subsystem\n"); - return NULL; + return -1; } + max_freq_mode = MAX_FREQ_SYSFS; + return 0; +} + +/* + * This monitor provides: + * + * 1) Average frequency a CPU resided in + * This always works if the CPU has aperf/mperf capabilities + * + * 2) C0 and Cx (any sleep state) time a CPU resided in + * Works if mperf timer stops ticking in sleep states which + * seem to be the case on all current HW. + * Both is directly retrieved from HW registers and is independent + * from kernel statistics. + */ +struct cpuidle_monitor mperf_monitor; +struct cpuidle_monitor *mperf_register(void) +{ + if (!(cpupower_cpu_info.caps & CPUPOWER_CAP_APERF)) + return NULL; + + if (init_maxfreq_mode()) + return NULL; /* Free this at program termination */ is_valid = calloc(cpu_count, sizeof(int)); diff --git a/tools/power/cpupower/utils/idle_monitor/snb_idle.c b/tools/power/cpupower/utils/idle_monitor/snb_idle.c index a1bc07cd53e..efc8a69c9ab 100644 --- a/tools/power/cpupower/utils/idle_monitor/snb_idle.c +++ b/tools/power/cpupower/utils/idle_monitor/snb_idle.c @@ -150,9 +150,19 @@ static struct cpuidle_monitor *snb_register(void) || cpupower_cpu_info.family != 6) return NULL; - if (cpupower_cpu_info.model != 0x2A - && cpupower_cpu_info.model != 0x2D) + switch (cpupower_cpu_info.model) { + case 0x2A: /* SNB */ + case 0x2D: /* SNB Xeon */ + case 0x3A: /* IVB */ + case 0x3E: /* IVB Xeon */ + case 0x3C: /* HSW */ + case 0x3F: /* HSW */ + case 0x45: /* HSW */ + case 0x46: /* HSW */ + break; + default: return NULL; + } is_valid = calloc(cpu_count, sizeof(int)); for (num = 0; num < SNB_CSTATE_COUNT; num++) { |