char netcpu_kstat10_id[]="\ @(#)netcpu_kstat10.c (c) Copyright 2005-2012, Hewlett-Packard Company Version 2.6.0"; #if HAVE_CONFIG_H # include #endif #include #if HAVE_INTTYPES_H # include #else # if HAVE_STDINT_H # include # endif #endif #if HAVE_UNISTD_H # include #endif #if HAVE_STRINGS_H # include #endif #if STDC_HEADERS # include # include #else # if HAVE_STDLIB_H # include # endif #endif #include #include #include #include "netsh.h" #include "netlib.h" static kstat_ctl_t *kc = NULL; static kid_t kcid = 0; typedef struct cpu_time_counters { uint64_t idle; uint64_t user; uint64_t kernel; uint64_t interrupt; } cpu_time_counters_t; static cpu_time_counters_t starting_cpu_counters[MAXCPUS]; static cpu_time_counters_t ending_cpu_counters[MAXCPUS]; static cpu_time_counters_t delta_cpu_counters[MAXCPUS]; static cpu_time_counters_t corrected_cpu_counters[MAXCPUS]; static void print_cpu_time_counters(char *name, int instance, cpu_time_counters_t *counters) { fprintf(where, "%s[%d]:\n" "\t idle %llu\n" "\t user %llu\n" "\t kernel %llu\n" "\t interrupt %llu\n", name,instance, counters[instance].idle, counters[instance].user, counters[instance].kernel, counters[instance].interrupt); } void cpu_util_init(void) { kstat_t *ksp; int i; kc = kstat_open(); if (kc == NULL) { fprintf(where, "cpu_util_init: kstat_open: errno %d %s\n", errno, strerror(errno)); fflush(where); exit(-1); } /* lets flesh-out a CPU instance number map since it seems that some systems, not even those which are partitioned, can have non-contiguous CPU numbers. discovered "the hard way" on a T5220. raj 20080804 */ i = 0; for (ksp = kc->kc_chain, i = 0; (ksp != NULL) && (i < MAXCPUS); ksp = ksp->ks_next) { if ((strcmp(ksp->ks_module,"cpu") == 0) && (strcmp(ksp->ks_name,"sys") == 0)) { if (debug) { fprintf(where,"Mapping CPU instance %d to entry %d\n", ksp->ks_instance,i); fflush(where); } lib_cpu_map[i++] = ksp->ks_instance; } } if (MAXCPUS == i) { fprintf(where, "Sorry, this system has more CPUs (%d) than netperf can handle (%d).\n" "Please alter MAXCPUS in netlib.h and recompile.\n", i, MAXCPUS); fflush(where); exit(1); } return; } void cpu_util_terminate(void) { kstat_close(kc); return; } int get_cpu_method(void) { return KSTAT_10; } static void print_unexpected_statistic_warning(char *who, char *what, char *why) { if (why) { fprintf(where, "WARNING! WARNING! WARNING! WARNING!\n" "%s found an unexpected %s statistic %.16s\n", who, why, what); } else { fprintf(where, "%s is ignoring statistic %.16s\n", who, what); } } static void get_cpu_counters(int cpu_num, cpu_time_counters_t *counters) { kstat_t *ksp; int found=0; kid_t nkcid; kstat_named_t *knp; int i; ksp = kstat_lookup(kc, "cpu", lib_cpu_map[cpu_num], "sys"); if ((ksp) && (ksp->ks_type == KSTAT_TYPE_NAMED)) { /* happiness and joy, keep going */ nkcid = kstat_read(kc, ksp, NULL); if (nkcid != -1) { /* happiness and joy, keep going. we could consider adding a "found < 3" to the end conditions, but then we wouldn't search to the end and find that Sun added some nsec. we probably want to see if they add an nsec. raj 2005-01-28 */ for (i = ksp->ks_ndata, knp = ksp->ks_data; i > 0; knp++,i--) { /* we would be hosed if the same name could appear twice */ if (!strcmp("cpu_nsec_idle",knp->name)) { found++; counters[cpu_num].idle = knp->value.ui64; } else if (!strcmp("cpu_nsec_user",knp->name)) { found++; counters[cpu_num].user = knp->value.ui64; } else if (!strcmp("cpu_nsec_kernel",knp->name)) { found++; counters[cpu_num].kernel = knp->value.ui64; } else if (!strcmp("cpu_nsec_intr",knp->name)) { if (debug >= 2) { fprintf(where, "Found a cpu_nsec_intr but it doesn't do what we want\n"); fflush(where); } } else if (strstr(knp->name,"nsec")) { /* finding another nsec here means Sun have changed something and we need to warn the user. raj 2005-01-28 */ print_unexpected_statistic_warning("get_cpu_counters", knp->name, "nsec"); } else if (debug >=2) { /* might want to tell people about what we are skipping. however, only display other names debug >=2. raj 2005-01-28 */ print_unexpected_statistic_warning("get_cpu_counters", knp->name, NULL); } } if (3 == found) { /* happiness and joy */ return; } else { fprintf(where, "get_cpu_counters could not find one or more of the expected counters!\n"); fflush(where); exit(-1); } } else { /* the kstat_read returned an error or the chain changed */ fprintf(where, "get_cpu_counters: kstat_read failed or chain id changed %d %s\n", errno, strerror(errno)); fflush(where); exit(-1); } } else { /* the lookup failed or found the wrong type */ fprintf(where, "get_cpu_counters: kstat_lookup failed for module 'cpu' number %d instance %d name 'sys' and KSTAT_TYPE_NAMED: errno %d %s\n", cpu_num, lib_cpu_map[cpu_num], errno, strerror(errno)); fflush(where); exit(-1); } } static void get_interrupt_counters(int cpu_num, cpu_time_counters_t *counters) { kstat_t *ksp; int found=0; kid_t nkcid; kstat_named_t *knp; int i; ksp = kstat_lookup(kc, "cpu", lib_cpu_map[cpu_num], "intrstat"); counters[cpu_num].interrupt = 0; if ((ksp) && (ksp->ks_type == KSTAT_TYPE_NAMED)) { /* happiness and joy, keep going */ nkcid = kstat_read(kc, ksp, NULL); if (nkcid != -1) { /* happiness and joy, keep going. we could consider adding a "found < 15" to the end conditions, but then we wouldn't search to the end and find that Sun added some "time." we probably want to see if they add a "nsec." raj 2005-01-28 */ for (i = ksp->ks_ndata, knp = ksp->ks_data; i > 0; knp++,i--) { if (strstr(knp->name,"time")) { found++; counters[cpu_num].interrupt += knp->value.ui64; } else if (debug >=2) { /* might want to tell people about what we are skipping. however, only display other names debug >=2. raj 2005-01-28 */ print_unexpected_statistic_warning("get_cpu_counters", knp->name, NULL); } } if (15 == found) { /* happiness and joy */ return; } else { fprintf(where, "get_cpu_counters could not find one or more of the expected counters!\n"); fflush(where); exit(-1); } } else { /* the kstat_read returned an error or the chain changed */ fprintf(where, "get_cpu_counters: kstat_read failed or chain id changed %d %s\n", errno, strerror(errno)); fflush(where); exit(-1); } } else { /* the lookup failed or found the wrong type */ fprintf(where, "get_cpu_counters: kstat_lookup failed for module 'cpu' %d instance %d class 'intrstat' and KSTAT_TYPE_NAMED: errno %d %s\n", cpu_num, lib_cpu_map[cpu_num], errno, strerror(errno)); fflush(where); exit(-1); } } static void get_cpu_time_counters(cpu_time_counters_t *counters) { int i; for (i = 0; i < lib_num_loc_cpus; i++){ get_cpu_counters(i, counters); get_interrupt_counters(i, counters); } return; } /* the kstat10 mechanism, since it is based on actual nanosecond counters is not going to use a comparison to an idle rate. so, the calibrate_idle_rate routine will be rather simple :) raj 2005-01-28 */ float calibrate_idle_rate(int iterations, int interval) { return 0.0; } float calc_cpu_util_internal(float elapsed_time) { int i; float correction_factor; float actual_rate; uint64_t total_cpu_nsec; /* multiply by 100 and divide by total and you get whole percentages. multiply by 1000 and divide by total and you get tenths of percentages. multiply by 10000 and divide by total and you get hundredths of percentages. etc etc etc raj 2005-01-28 */ #define CALC_PERCENT 100 #define CALC_TENTH_PERCENT 1000 #define CALC_HUNDREDTH_PERCENT 10000 #define CALC_THOUSANDTH_PERCENT 100000 #define CALC_ACCURACY CALC_THOUSANDTH_PERCENT uint64_t fraction_idle; uint64_t fraction_user; uint64_t fraction_kernel; uint64_t fraction_interrupt; uint64_t interrupt_idle; uint64_t interrupt_user; uint64_t interrupt_kernel; memset(&lib_local_cpu_stats, 0, sizeof(lib_local_cpu_stats)); /* It is possible that the library measured a time other than the one that the user want for the cpu utilization calculations - for example, tests that were ended by watchdog timers such as the udp stream test. We let these tests tell up what the elapsed time should be. */ if (elapsed_time != 0.0) { correction_factor = (float) 1.0 + ((lib_elapsed - elapsed_time) / elapsed_time); } else { correction_factor = (float) 1.0; } for (i = 0; i < lib_num_loc_cpus; i++) { /* this is now the fun part. we have the nanoseconds _allegedly_ spent in user, idle and kernel. We also have nanoseconds spent servicing interrupts. Sadly, in the developer's finite wisdom, the interrupt time accounting is in parallel with the other accounting. this means that time accounted in user, kernel or idle will also include time spent in interrupt. for netperf's porpoises we do not really care about that for user and kernel, but we certainly do care for idle. the $64B question becomes - how to "correct" for this? we could just subtract interrupt time from idle. that has the virtue of simplicity and also "punishes" Sun for doing something that seems to be so stupid. however, we probably have to be "fair" even to the allegedly stupid so the other mechanism, suggested by a Sun engineer is to subtract interrupt time from each of user, kernel and idle in proportion to their numbers. then we sum the corrected user, kernel and idle along with the interrupt time and use that to calculate a new idle percentage and thus a CPU util percentage. that is what we will attempt to do here. raj 2005-01-28 of course, we also have to wonder what we should do if there is more interrupt time than the sum of user, kernel and idle. that is a theoretical possibility I suppose, but for the time-being, one that we will blythly ignore, except perhaps for a quick check. raj 2005-01-31 */ /* we ass-u-me that these counters will never wrap during a netperf run. this may not be a particularly safe thing to do. raj 2005-01-28 */ delta_cpu_counters[i].idle = ending_cpu_counters[i].idle - starting_cpu_counters[i].idle; delta_cpu_counters[i].user = ending_cpu_counters[i].user - starting_cpu_counters[i].user; delta_cpu_counters[i].kernel = ending_cpu_counters[i].kernel - starting_cpu_counters[i].kernel; delta_cpu_counters[i].interrupt = ending_cpu_counters[i].interrupt - starting_cpu_counters[i].interrupt; if (debug) { print_cpu_time_counters("delta_cpu_counters",i,delta_cpu_counters); } /* for this summation, we do not include interrupt time */ total_cpu_nsec = delta_cpu_counters[i].idle + delta_cpu_counters[i].user + delta_cpu_counters[i].kernel; if (debug) { fprintf(where,"total_cpu_nsec %llu\n",total_cpu_nsec); } if (delta_cpu_counters[i].interrupt > total_cpu_nsec) { /* we are not in Kansas any more Toto, and I am not quite sure the best way to get our tails out of here so let us just punt. raj 2005-01-31 */ fprintf(where, "WARNING! WARNING! WARNING! WARNING! WARNING! \n" "calc_cpu_util_internal: more interrupt time than others combined!\n" "\tso CPU util cannot be estimated\n" "\t delta[%d].interrupt %llu\n" "\t delta[%d].idle %llu\n" "\t delta[%d].user %llu\n" "\t delta[%d].kernel %llu\n", i,delta_cpu_counters[i].interrupt, i,delta_cpu_counters[i].idle, i,delta_cpu_counters[i].user, i,delta_cpu_counters[i].kernel); fflush(where); lib_local_cpu_stats.cpu_util = -1.0; lib_local_per_cpu_util[i] = -1.0; return -1.0; } /* and now some fun with integer math. i initially tried to promote things to long doubled but that didn't seem to result in happiness and joy. raj 2005-01-28 */ fraction_idle = (delta_cpu_counters[i].idle * CALC_ACCURACY) / total_cpu_nsec; fraction_user = (delta_cpu_counters[i].user * CALC_ACCURACY) / total_cpu_nsec; fraction_kernel = (delta_cpu_counters[i].kernel * CALC_ACCURACY) / total_cpu_nsec; /* ok, we have our fractions, now we want to take that fraction of the interrupt time and subtract that from the bucket. */ interrupt_idle = ((delta_cpu_counters[i].interrupt * fraction_idle) / CALC_ACCURACY); interrupt_user = ((delta_cpu_counters[i].interrupt * fraction_user) / CALC_ACCURACY); interrupt_kernel = ((delta_cpu_counters[i].interrupt * fraction_kernel) / CALC_ACCURACY); if (debug) { fprintf(where, "\tfraction_idle %llu interrupt_idle %llu\n" "\tfraction_user %llu interrupt_user %llu\n" "\tfraction_kernel %llu interrupt_kernel %llu\n", fraction_idle, interrupt_idle, fraction_user, interrupt_user, fraction_kernel, interrupt_kernel); } corrected_cpu_counters[i].idle = delta_cpu_counters[i].idle - interrupt_idle; corrected_cpu_counters[i].user = delta_cpu_counters[i].user - interrupt_user; corrected_cpu_counters[i].kernel = delta_cpu_counters[i].kernel - interrupt_kernel; corrected_cpu_counters[i].interrupt = delta_cpu_counters[i].interrupt; if (debug) { print_cpu_time_counters("corrected_cpu_counters", i, corrected_cpu_counters); } /* I was going to check for going less than zero, but since all the calculations are in unsigned quantities that would seem to be a triffle silly... raj 2005-01-28 */ /* ok, now we sum the numbers again, this time including interrupt */ total_cpu_nsec = corrected_cpu_counters[i].idle + corrected_cpu_counters[i].user + corrected_cpu_counters[i].kernel + corrected_cpu_counters[i].interrupt; /* and recalculate our fractions we are really only going to use fraction_idle, but lets calculate the rest just for the heck of it. one day we may want to display them. raj 2005-01-28 */ /* multiply by 100 and divide by total and you get whole percentages. multiply by 1000 and divide by total and you get tenths of percentages. multiply by 10000 and divide by total and you get hundredths of percentages. etc etc etc raj 2005-01-28 */ fraction_idle = (corrected_cpu_counters[i].idle * CALC_ACCURACY) / total_cpu_nsec; fraction_user = (corrected_cpu_counters[i].user * CALC_ACCURACY) / total_cpu_nsec; fraction_kernel = (corrected_cpu_counters[i].kernel * CALC_ACCURACY) / total_cpu_nsec; fraction_interrupt = (corrected_cpu_counters[i].interrupt * CALC_ACCURACY) / total_cpu_nsec; if (debug) { fprintf(where,"\tfraction_idle %lu\n",fraction_idle); fprintf(where,"\tfraction_user %lu\n",fraction_user); fprintf(where,"\tfraction_kernel %lu\n",fraction_kernel); fprintf(where,"\tfraction_interrupt %lu\n",fraction_interrupt); } /* and finally, what is our CPU utilization? */ lib_local_per_cpu_util[i] = 100.0 - (((float)fraction_idle / (float)CALC_ACCURACY) * 100.0); lib_local_per_cpu_util[i] *= correction_factor; if (debug) { fprintf(where, "lib_local_per_cpu_util[%d] %g cf %f\n", i, lib_local_per_cpu_util[i], correction_factor); } lib_local_cpu_stats.cpu_util += lib_local_per_cpu_util[i]; } /* we want the average across all n processors */ lib_local_cpu_stats.cpu_util /= (float)lib_num_loc_cpus; return lib_local_cpu_stats.cpu_util; } void cpu_start_internal(void) { get_cpu_time_counters(starting_cpu_counters); return; } void cpu_stop_internal(void) { get_cpu_time_counters(ending_cpu_counters); }