@@ -3,6 +3,7 @@
#ifndef RESCTRL_H
#define RESCTRL_H
#include <stdio.h>
+#include <stdarg.h>
#include <errno.h>
#include <sched.h>
#include <stdlib.h>
@@ -28,10 +29,35 @@
exit(EXIT_FAILURE); \
} while (0)
+/*
+ * resctrl_val_param: resctrl test parameters
+ * @resctrl_val: Resctrl feature (Eg: mbm, mba.. etc)
+ * @ctrlgrp: Name of the control monitor group (con_mon grp)
+ * @mongrp: Name of the monitor group (mon grp)
+ * @cpu_no: CPU number to which the benchmark would be binded
+ * @span: Memory bytes accessed in each benchmark iteration
+ * @mum_resctrlfs: Should the resctrl FS be remounted?
+ * @filename: Name of file to which the o/p should be written
+ * @bw_report: Bandwidth report type (reads vs writes)
+ * @setup: Call back function to setup test environment
+ */
+struct resctrl_val_param {
+ char *resctrl_val;
+ char ctrlgrp[64];
+ char mongrp[64];
+ int cpu_no;
+ int span;
+ int mum_resctrlfs;
+ char filename[64];
+ char *bw_report;
+ int (*setup)(int num, ...);
+};
+
pid_t bm_pid, ppid;
int remount_resctrlfs(bool mum_resctrlfs);
int get_resource_id(int cpu_no, int *resource_id);
+int umount_resctrlfs(void);
int validate_bw_report_request(char *bw_report);
int validate_resctrl_feature_request(char *resctrl_val);
int taskset_benchmark(pid_t bm_pid, int cpu_no);
@@ -43,5 +69,6 @@ int write_bm_pid_to_resctrl(pid_t bm_pid, char *ctrlgrp, char *mongrp,
int perf_event_open(struct perf_event_attr *hw_event, pid_t pid, int cpu,
int group_fd, unsigned long flags);
int run_fill_buf(int span, int malloc_and_init_memory, int memflush, int op);
+int membw_val(char **benchmark_cmd, struct resctrl_val_param *param);
#endif /* RESCTRL_H */
@@ -11,6 +11,7 @@
*/
#include "resctrl.h"
+#define MB (1024 * 1024)
#define UNCORE_IMC "uncore_imc"
#define READ_FILE_NAME "events/cas_count_read"
#define WRITE_FILE_NAME "events/cas_count_write"
@@ -47,6 +48,9 @@ struct imc_counter_config {
struct membw_read_format return_value;
int fd;
};
+
+static char mbm_total_path[1024];
+static int imcs;
static struct imc_counter_config imc_counters_config[MAX_IMCS][2];
void membw_initialize_perf_event_attr(int i, int j)
@@ -116,3 +120,585 @@ void get_event_and_umask(char *cas_count_cfg, int count, bool op)
}
}
+static int open_perf_event(int i, int cpu_no, int j)
+{
+ imc_counters_config[i][j].fd =
+ perf_event_open(&imc_counters_config[i][j].pe, -1, cpu_no, -1,
+ PERF_FLAG_FD_CLOEXEC);
+
+ if (imc_counters_config[i][j].fd == -1) {
+ fprintf(stderr, "Error opening leader %llx\n",
+ imc_counters_config[i][j].pe.config);
+
+ return -1;
+ }
+
+ return 0;
+}
+
+/* Get type and config (read and write) of an iMC counter */
+static int read_from_imc_dir(char *imc_dir, int count)
+{
+ char cas_count_cfg[1024], imc_counter_cfg[1024], imc_counter_type[1024];
+ FILE *fp;
+
+ /* Get type of iMC counter */
+ sprintf(imc_counter_type, "%s%s", imc_dir, "type");
+ fp = fopen(imc_counter_type, "r");
+ if (!fp) {
+ perror("Failed to open imc counter type file");
+
+ return -1;
+ }
+ if (fscanf(fp, "%u", &imc_counters_config[count][READ].type) <= 0) {
+ perror("Could not get imc type");
+ fclose(fp);
+
+ return -1;
+ }
+ fclose(fp);
+
+ imc_counters_config[count][WRITE].type =
+ imc_counters_config[count][READ].type;
+
+ /* Get read config */
+ sprintf(imc_counter_cfg, "%s%s", imc_dir, READ_FILE_NAME);
+ fp = fopen(imc_counter_cfg, "r");
+ if (!fp) {
+ perror("Failed to open imc config file");
+
+ return -1;
+ }
+ if (fscanf(fp, "%s", cas_count_cfg) <= 0) {
+ perror("Could not get imc cas count read");
+ fclose(fp);
+
+ return -1;
+ }
+ fclose(fp);
+
+ get_event_and_umask(cas_count_cfg, count, READ);
+
+ /* Get write config */
+ sprintf(imc_counter_cfg, "%s%s", imc_dir, WRITE_FILE_NAME);
+ fp = fopen(imc_counter_cfg, "r");
+ if (!fp) {
+ perror("Failed to open imc config file");
+
+ return -1;
+ }
+ if (fscanf(fp, "%s", cas_count_cfg) <= 0) {
+ perror("Could not get imc cas count write");
+ fclose(fp);
+
+ return -1;
+ }
+ fclose(fp);
+
+ get_event_and_umask(cas_count_cfg, count, WRITE);
+
+ return 0;
+}
+
+/*
+ * A system can have 'n' number of iMC (Integrated Memory Controller)
+ * counters, get that 'n'. For each iMC counter get it's type and config.
+ * Also, each counter has two configs, one for read and the other for write.
+ * A config again has two parts, event and umask.
+ * Enumerate all these details into an array of structures.
+ *
+ * Return: >= 0 on success. < 0 on failure.
+ */
+static int num_of_imcs(void)
+{
+ unsigned int count = 0;
+ char imc_dir[1024];
+ struct dirent *ep;
+ int ret;
+ DIR *dp;
+
+ dp = opendir(DYN_PMU_PATH);
+ if (dp) {
+ while ((ep = readdir(dp))) {
+ if (strstr(ep->d_name, UNCORE_IMC)) {
+ sprintf(imc_dir, "%s/%s/", DYN_PMU_PATH,
+ ep->d_name);
+ ret = read_from_imc_dir(imc_dir, count);
+ if (ret) {
+ closedir(dp);
+
+ return ret;
+ }
+ count++;
+ }
+ }
+ closedir(dp);
+ if (count == 0) {
+ perror("Unable find iMC counters!\n");
+
+ return -1;
+ }
+ } else {
+ perror("Unable to open PMU directory!\n");
+
+ return -1;
+ }
+
+ return count;
+}
+
+static int initialize_mem_bw_imc(void)
+{
+ int imc, j;
+
+ imcs = num_of_imcs();
+ if (imcs <= 0)
+ return imcs;
+
+ /* Initialize perf_event_attr structures for all iMC's */
+ for (imc = 0; imc < imcs; imc++) {
+ for (j = 0; j < 2; j++)
+ membw_initialize_perf_event_attr(imc, j);
+ }
+
+ return 0;
+}
+
+/*
+ * get_mem_bw_imc: Memory band width as reported by iMC counters
+ * @cpu_no: CPU number that the benchmark PID is binded to
+ * @bw_report: Bandwidth report type (reads, writes)
+ *
+ * Memory B/W utilized by a process on a socket can be calculated using
+ * iMC counters. Perf events are used to read these counters.
+ *
+ * Return: >= 0 on success. < 0 on failure.
+ */
+static float get_mem_bw_imc(int cpu_no, char *bw_report)
+{
+ float reads, writes, of_mul_read, of_mul_write;
+ int imc, j, ret;
+
+ /* Start all iMC counters to log values (both read and write) */
+ reads = 0, writes = 0, of_mul_read = 1, of_mul_write = 1;
+ for (imc = 0; imc < imcs; imc++) {
+ for (j = 0; j < 2; j++) {
+ ret = open_perf_event(imc, cpu_no, j);
+ if (ret)
+ return -1;
+ }
+ for (j = 0; j < 2; j++)
+ membw_ioctl_perf_event_ioc_reset_enable(imc, j);
+ }
+
+ sleep(1);
+
+ /* Stop counters after a second to get results (both read and write) */
+ for (imc = 0; imc < imcs; imc++) {
+ for (j = 0; j < 2; j++)
+ membw_ioctl_perf_event_ioc_disable(imc, j);
+ }
+
+ /*
+ * Get results which are stored in struct type imc_counter_config
+ * Take over flow into consideration before calculating total b/w
+ */
+ for (imc = 0; imc < imcs; imc++) {
+ struct imc_counter_config *r =
+ &imc_counters_config[imc][READ];
+ struct imc_counter_config *w =
+ &imc_counters_config[imc][WRITE];
+
+ if (read(r->fd, &r->return_value,
+ sizeof(struct membw_read_format)) == -1) {
+ perror("Couldn't get read b/w through iMC");
+
+ return -1;
+ }
+
+ if (read(w->fd, &w->return_value,
+ sizeof(struct membw_read_format)) == -1) {
+ perror("Couldn't get write bw through iMC");
+
+ return -1;
+ }
+
+ __u64 r_time_enabled = r->return_value.time_enabled;
+ __u64 r_time_running = r->return_value.time_running;
+
+ if (r_time_enabled != r_time_running)
+ of_mul_read = (float)r_time_enabled /
+ (float)r_time_running;
+
+ __u64 w_time_enabled = w->return_value.time_enabled;
+ __u64 w_time_running = w->return_value.time_running;
+
+ if (w_time_enabled != w_time_running)
+ of_mul_write = (float)w_time_enabled /
+ (float)w_time_running;
+ reads += r->return_value.value * of_mul_read * SCALE;
+ writes += w->return_value.value * of_mul_write * SCALE;
+ }
+
+ for (imc = 0; imc < imcs; imc++) {
+ close(imc_counters_config[imc][READ].fd);
+ close(imc_counters_config[imc][WRITE].fd);
+ }
+
+ if (strcmp(bw_report, "reads") == 0)
+ return reads;
+
+ if (strcmp(bw_report, "writes") == 0)
+ return writes;
+
+ return (reads + writes);
+}
+
+void set_mbm_path(const char *ctrlgrp, const char *mongrp, int resource_id)
+{
+ if (ctrlgrp && mongrp)
+ sprintf(mbm_total_path, CON_MON_MBM_LOCAL_BYTES_PATH,
+ RESCTRL_PATH, ctrlgrp, mongrp, resource_id);
+ else if (!ctrlgrp && mongrp)
+ sprintf(mbm_total_path, MON_MBM_LOCAL_BYTES_PATH, RESCTRL_PATH,
+ mongrp, resource_id);
+ else if (ctrlgrp && !mongrp)
+ sprintf(mbm_total_path, CON_MBM_LOCAL_BYTES_PATH, RESCTRL_PATH,
+ ctrlgrp, resource_id);
+ else if (!ctrlgrp && !mongrp)
+ sprintf(mbm_total_path, MBM_LOCAL_BYTES_PATH, RESCTRL_PATH,
+ resource_id);
+}
+
+/*
+ * initialize_mem_bw_resctrl: Appropriately populate "mbm_total_path"
+ * @ctrlgrp: Name of the control monitor group (con_mon grp)
+ * @mongrp: Name of the monitor group (mon grp)
+ * @cpu_no: CPU number that the benchmark PID is binded to
+ * @resctrl_val: Resctrl feature (Eg: mbm, mba.. etc)
+ */
+static void initialize_mem_bw_resctrl(const char *ctrlgrp, const char *mongrp,
+ int cpu_no, char *resctrl_val)
+{
+ int resource_id;
+
+ if (get_resource_id(cpu_no, &resource_id) < 0) {
+ perror("Could not get resource_id");
+ return;
+ }
+
+ if (strcmp(resctrl_val, "mbm") == 0)
+ set_mbm_path(ctrlgrp, mongrp, resource_id);
+
+ if ((strcmp(resctrl_val, "mba") == 0)) {
+ if (ctrlgrp)
+ sprintf(mbm_total_path, CON_MBM_LOCAL_BYTES_PATH,
+ RESCTRL_PATH, ctrlgrp, resource_id);
+ else
+ sprintf(mbm_total_path, MBM_LOCAL_BYTES_PATH,
+ RESCTRL_PATH, resource_id);
+ }
+}
+
+/*
+ * Get MBM Local bytes as reported by resctrl FS
+ * For MBM,
+ * 1. If con_mon grp and mon grp are given, then read from con_mon grp's mon grp
+ * 2. If only con_mon grp is given, then read from con_mon grp
+ * 3. If both are not given, then read from root con_mon grp
+ * For MBA,
+ * 1. If con_mon grp is given, then read from it
+ * 2. If con_mon grp is not given, then read from root con_mon grp
+ */
+static unsigned long get_mem_bw_resctrl(void)
+{
+ unsigned long mbm_total = 0;
+ FILE *fp;
+
+ fp = fopen(mbm_total_path, "r");
+ if (!fp) {
+ perror("Failed to open total bw file");
+
+ return -1;
+ }
+ if (fscanf(fp, "%lu", &mbm_total) <= 0) {
+ perror("Could not get mbm local bytes");
+ fclose(fp);
+
+ return -1;
+ }
+ fclose(fp);
+
+ return mbm_total;
+}
+
+pid_t bm_pid, ppid;
+
+static void ctrlc_handler(int signum, siginfo_t *info, void *ptr)
+{
+ kill(bm_pid, SIGKILL);
+ printf("Ending\n\n");
+
+ exit(EXIT_SUCCESS);
+}
+
+/*
+ * print_results_bw: the memory bandwidth results are stored in a file
+ * @filename: file that stores the results
+ * @bm_pid: child pid that runs benchmark
+ * @bw_imc: perf imc counter value
+ * @bw_resc: memory bandwidth value
+ *
+ * Return: 0 on success. non-zero on failure.
+ */
+static int print_results_bw(char *filename, int bm_pid, float bw_imc,
+ unsigned long bw_resc)
+{
+ unsigned long diff = labs(bw_imc - bw_resc);
+ FILE *fp;
+
+ if (strcmp(filename, "stdio") == 0 || strcmp(filename, "stderr") == 0) {
+ printf("Pid: %d \t Mem_BW_iMC: %f \t ", bm_pid, bw_imc);
+ printf("Mem_BW_resc: %lu \t Difference: %lu\n", bw_resc, diff);
+ } else {
+ fp = fopen(filename, "a");
+ if (!fp) {
+ perror("Cannot open results file");
+
+ return errno;
+ }
+ if (fprintf(fp, "Pid: %d \t Mem_BW_iMC: %f \t Mem_BW_resc: %lu \t Difference: %lu\n",
+ bm_pid, bw_imc, bw_resc, diff) <= 0) {
+ fclose(fp);
+ perror("Could not log results.");
+
+ return errno;
+ }
+ fclose(fp);
+ }
+
+ return 0;
+}
+
+static int
+measure_vals(struct resctrl_val_param *param, unsigned long *bw_resc_start)
+{
+ unsigned long bw_imc, bw_resc, bw_resc_end;
+ int ret;
+
+ /*
+ * Measure memory bandwidth from resctrl and from
+ * another source which is perf imc value or could
+ * be something else if perf imc event is not available.
+ * Compare the two values to validate resctrl value.
+ * It takes 1sec to measure the data.
+ */
+ bw_imc = get_mem_bw_imc(param->cpu_no, param->bw_report);
+ if (bw_imc <= 0)
+ return bw_imc;
+
+ bw_resc_end = get_mem_bw_resctrl();
+ if (bw_resc_end <= 0)
+ return bw_resc_end;
+
+ bw_resc = (bw_resc_end - *bw_resc_start) / MB;
+ ret = print_results_bw(param->filename, bm_pid, bw_imc, bw_resc);
+ if (ret)
+ return ret;
+
+ *bw_resc_start = bw_resc_end;
+
+ return 0;
+}
+
+/*
+ * membw_val: execute benchmark and measure memory bandwidth on
+ * the benchmark
+ * @benchmark_cmd: benchmark command and its arguments
+ * @param: parameters passed to membw_val()
+ *
+ * Return: 0 on success. non-zero on failure.
+ */
+int membw_val(char **benchmark_cmd, struct resctrl_val_param *param)
+{
+ int ret = 0, pipefd[2], pipe_message = 0;
+ char *resctrl_val = param->resctrl_val;
+ unsigned long bw_resc_start = 0;
+ struct sigaction sigact;
+ union sigval value;
+ FILE *fp;
+
+ if (strcmp(param->filename, "") == 0)
+ sprintf(param->filename, "stdio");
+
+ if (strcmp(param->bw_report, "") == 0)
+ param->bw_report = "total";
+
+ ret = validate_resctrl_feature_request(resctrl_val);
+ if (ret)
+ return ret;
+
+ if ((strcmp(resctrl_val, "mba")) == 0 ||
+ (strcmp(resctrl_val, "mbm")) == 0) {
+ ret = validate_bw_report_request(param->bw_report);
+ if (ret)
+ return ret;
+ }
+
+ ret = remount_resctrlfs(param->mum_resctrlfs);
+ if (ret)
+ return ret;
+
+ /*
+ * If benchmark wasn't successfully started by child, then child should
+ * kill parent, so save parent's pid
+ */
+ ppid = getpid();
+
+ /* File based synchronization between parent and child */
+ fp = fopen("sig", "w");
+ if (!fp) {
+ perror("Failed to open sig file");
+
+ return -1;
+ }
+ if (fprintf(fp, "%d\n", 0) <= 0) {
+ perror("Unable to establish sync bw parent & child");
+ fclose(fp);
+
+ return -1;
+ }
+ fclose(fp);
+
+ if (pipe(pipefd)) {
+ perror("Unable to create pipe");
+
+ return -1;
+ }
+
+ /*
+ * Fork to start benchmark, save child's pid so that it can be killed
+ * when needed
+ */
+ bm_pid = fork();
+ if (bm_pid == -1) {
+ perror("Unable to fork");
+
+ return -1;
+ }
+
+ if (bm_pid == 0) {
+ /*
+ * Mask all signals except SIGUSR1, parent uses SIGUSR1 to
+ * start benchmark
+ */
+ sigfillset(&sigact.sa_mask);
+ sigdelset(&sigact.sa_mask, SIGUSR1);
+
+ sigact.sa_sigaction = run_benchmark;
+ sigact.sa_flags = SA_SIGINFO;
+
+ /* Register for "SIGUSR1" signal from parent */
+ if (sigaction(SIGUSR1, &sigact, NULL))
+ PARENT_EXIT("Can't register child for signal");
+
+ /* Tell parent that child is ready */
+ close(pipefd[0]);
+ pipe_message = 1;
+ write(pipefd[1], &pipe_message, sizeof(pipe_message));
+ close(pipefd[1]);
+
+ /* Suspend child until delivery of "SIGUSR1" from parent */
+ sigsuspend(&sigact.sa_mask);
+
+ PARENT_EXIT("Child is done");
+ }
+
+ printf("Benchmark PID: %d\n", bm_pid);
+
+ /*
+ * Register CTRL-C handler for parent, as it has to kill benchmark
+ * before exiting
+ */
+ sigact.sa_sigaction = ctrlc_handler;
+ sigemptyset(&sigact.sa_mask);
+ sigact.sa_flags = SA_SIGINFO;
+ if (sigaction(SIGINT, &sigact, NULL) ||
+ sigaction(SIGHUP, &sigact, NULL)) {
+ perror("Can't register parent for CTRL-C handler");
+ ret = errno;
+ goto out;
+ }
+
+ value.sival_ptr = benchmark_cmd;
+
+ /* Taskset benchmark to specified cpu */
+ ret = taskset_benchmark(bm_pid, param->cpu_no);
+ if (ret)
+ goto out;
+
+ /* Write benchmark to specified control&monitoring grp in resctrl FS */
+ ret = write_bm_pid_to_resctrl(bm_pid, param->ctrlgrp, param->mongrp,
+ resctrl_val);
+ if (ret)
+ goto out;
+
+ if ((strcmp(resctrl_val, "mbm") == 0) ||
+ (strcmp(resctrl_val, "mba") == 0)) {
+ ret = initialize_mem_bw_imc();
+ if (ret)
+ goto out;
+
+ initialize_mem_bw_resctrl(param->ctrlgrp, param->mongrp,
+ param->cpu_no, resctrl_val);
+ }
+
+ /* Parent waits for child to be ready. */
+ close(pipefd[1]);
+ while (pipe_message != 1)
+ read(pipefd[0], &pipe_message, sizeof(pipe_message));
+ close(pipefd[0]);
+
+ /* Signal child to start benchmark */
+ if (sigqueue(bm_pid, SIGUSR1, value) == -1) {
+ perror("Unable to signal child to start execution");
+ ret = errno;
+ goto out;
+ }
+
+ /* Give benchmark enough time to fully run */
+ sleep(1);
+
+ /* Test runs until the callback setup() tells the test to stop. */
+ while (1) {
+ if (strcmp(resctrl_val, "mbm") == 0) {
+ ret = param->setup(1, param);
+ if (ret) {
+ ret = 0;
+ break;
+ }
+
+ ret = measure_vals(param, &bw_resc_start);
+ if (ret)
+ break;
+ } else if ((strcmp(resctrl_val, "mba") == 0)) {
+ ret = param->setup(1, param);
+ if (ret) {
+ ret = 0;
+ break;
+ }
+
+ ret = measure_vals(param, &bw_resc_start);
+ if (ret)
+ break;
+ } else {
+ break;
+ }
+ }
+
+out:
+ kill(bm_pid, SIGKILL);
+ umount_resctrlfs();
+
+ return ret;
+}