Date:   Mon, 12 Nov 2018 21:15:15 +0000
From:   "Moger, Babu" <Babu.Moger@....com>
To:     Fenghua Yu <fenghua.yu@...el.com>,
        Thomas Gleixner <tglx@...utronix.de>,
        Ingo Molnar <mingo@...hat.com>, H Peter Anvin <hpa@...or.com>,
        Tony Luck <tony.luck@...el.com>,
        Peter Zijlstra <peterz@...radead.org>,
        Reinette Chatre <reinette.chatre@...el.com>,
        James Morse <james.morse@....com>,
        Ravi V Shankar <ravi.v.shankar@...el.com>,
        Sai Praneeth Prakhya <sai.praneeth.prakhya@...el.com>,
        Arshiya Hayatkhan Pathan <arshiya.hayatkhan.pathan@...el.com>
CC:     linux-kernel <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v3 3/8] selftests/resctrl: Read memory bandwidth from perf
 IMC counter and from resctrl file system



On 10/31/18 4:02 PM, Fenghua Yu wrote:
> From: Sai Praneeth Prakhya <sai.praneeth.prakhya@...el.com>
> 
> Total memory bandwidth can be monitored from perf IMC counter and from
> resctrl file system. Later the two will be compared to verify the total
> memory bandwidth read from resctrl is correct.
> 
> Signed-off-by: Sai Praneeth Prakhya <sai.praneeth.prakhya@...el.com>
> Signed-off-by: Arshiya Hayatkhan Pathan <arshiya.hayatkhan.pathan@...el.com>
> Signed-off-by: Fenghua Yu <fenghua.yu@...el.com>
> ---
>  tools/testing/selftests/resctrl/membw.c | 431 ++++++++++++++++++++++++++++++++
>  1 file changed, 431 insertions(+)
>  create mode 100644 tools/testing/selftests/resctrl/membw.c
> 
> diff --git a/tools/testing/selftests/resctrl/membw.c b/tools/testing/selftests/resctrl/membw.c
> new file mode 100644
> index 000000000000..3146cf8b7468
> --- /dev/null
> +++ b/tools/testing/selftests/resctrl/membw.c
> @@ -0,0 +1,431 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Memory bandwidth monitoring and allocation library
> + *
> + * Copyright (C) 2018 Intel Corporation
> + *
> + * Authors:
> + *    Arshiya Hayatkhan Pathan <arshiya.hayatkhan.pathan@...el.com>
> + *    Sai Praneeth Prakhya <sai.praneeth.prakhya@...el.com>,
> + *    Fenghua Yu <fenghua.yu@...el.com>
> + */
> +#include "resctrl.h"
> +
> +#define UNCORE_IMC		"uncore_imc"
> +#define READ_FILE_NAME		"events/cas_count_read"
> +#define WRITE_FILE_NAME		"events/cas_count_write"
> +#define DYN_PMU_PATH		"/sys/bus/event_source/devices"
> +#define SCALE			0.00006103515625
> +#define MAX_IMCS		20
> +#define MAX_TOKENS		5
> +#define READ			0
> +#define WRITE			1
> +#define CON_MON_MBM_LOCAL_BYTES_PATH				\
> +	"%s/%s/mon_groups/%s/mon_data/mon_L3_0%c/mbm_local_bytes"
> +
> +#define CON_MBM_LOCAL_BYTES_PATH		\
> +	"%s/%s/mon_data/mon_L3_0%c/mbm_local_bytes"
> +
> +#define MON_MBM_LOCAL_BYTES_PATH		\
> +	"%s/mon_groups/%s/mon_data/mon_L3_0%c/mbm_local_bytes"
> +
> +#define MBM_LOCAL_BYTES_PATH			\
> +	"%s/mon_data/mon_L3_0%c/mbm_local_bytes"
> +
> +struct membw_read_format {
> +	__u64 value;         /* The value of the event */
> +	__u64 time_enabled;  /* if PERF_FORMAT_TOTAL_TIME_ENABLED */
> +	__u64 time_running;  /* if PERF_FORMAT_TOTAL_TIME_RUNNING */
> +	__u64 id;            /* if PERF_FORMAT_ID */
> +};
> +
> +struct imc_counter_config {
> +	__u32 type;
> +	__u64 event;
> +	__u64 umask;
> +	struct perf_event_attr pe;
> +	struct membw_read_format return_value;
> +	int fd;
> +};
> +
> +static struct imc_counter_config imc_counters_config[MAX_IMCS][2];
> +static char mbm_total_path[1024];
> +static int imcs;
> +
> +void membw_initialize_perf_event_attr(int i, int j)
> +{
> +	memset(&imc_counters_config[i][j].pe, 0,
> +	       sizeof(struct perf_event_attr));
> +	imc_counters_config[i][j].pe.type = imc_counters_config[i][j].type;
> +	imc_counters_config[i][j].pe.size = sizeof(struct perf_event_attr);
> +	imc_counters_config[i][j].pe.disabled = 1;
> +	imc_counters_config[i][j].pe.inherit = 1;
> +	imc_counters_config[i][j].pe.exclude_guest = 1;
> +	imc_counters_config[i][j].pe.config =
> +		imc_counters_config[i][j].umask << 8 |
> +		imc_counters_config[i][j].event;
> +	imc_counters_config[i][j].pe.sample_type = PERF_SAMPLE_IDENTIFIER;
> +	imc_counters_config[i][j].pe.read_format =
> +		PERF_FORMAT_TOTAL_TIME_ENABLED | PERF_FORMAT_TOTAL_TIME_RUNNING;
> +}
> +
> +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;
> +}
> +
> +void membw_ioctl_perf_event_ioc_reset_enable(int i, int j)
> +{
> +	ioctl(imc_counters_config[i][j].fd, PERF_EVENT_IOC_RESET, 0);
> +	ioctl(imc_counters_config[i][j].fd, PERF_EVENT_IOC_ENABLE, 0);
> +}
> +
> +void membw_ioctl_perf_event_ioc_disable(int i, int j)
> +{
> +	ioctl(imc_counters_config[i][j].fd, PERF_EVENT_IOC_DISABLE, 0);
> +}
> +
> +/*
> + * get_event_and_umask:	Parse config into event and umask
> + * @cas_count_cfg:	Config
> + * @count:		iMC number
> + * @op:			Operation (read/write)
> + */
> +void get_event_and_umask(char *cas_count_cfg, int count, bool op)
> +{
> +	char *token[MAX_TOKENS];
> +	int i = 0;
> +
> +	strcat(cas_count_cfg, ",");
> +	token[0] = strtok(cas_count_cfg, "=,");
> +
> +	for (i = 1; i < MAX_TOKENS; i++)
> +		token[i] = strtok(NULL, "=,");
> +
> +	for (i = 0; i < MAX_TOKENS; i++) {
> +		if (!token[i])
> +			break;
> +		if (strcmp(token[i], "event") == 0) {
> +			if (op == READ)
> +				imc_counters_config[count][READ].event =
> +				strtol(token[i + 1], NULL, 16);
> +			else
> +				imc_counters_config[count][WRITE].event =
> +				strtol(token[i + 1], NULL, 16);
> +		}
> +		if (strcmp(token[i], "umask") == 0) {
> +			if (op == READ)
> +				imc_counters_config[count][READ].umask =
> +				strtol(token[i + 1], NULL, 16);
> +			else
> +				imc_counters_config[count][WRITE].umask =
> +				strtol(token[i + 1], NULL, 16);
> +		}
> +	}
> +}
> +
> +/* 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)

I think this condition should be "<=". Pls check.
> +		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, char sock_num)
> +{
> +	if (ctrlgrp && mongrp)
> +		sprintf(mbm_total_path, CON_MON_MBM_LOCAL_BYTES_PATH,
> +			RESCTRL_PATH, ctrlgrp, mongrp, sock_num);
> +	else if (!ctrlgrp && mongrp)
> +		sprintf(mbm_total_path, MON_MBM_LOCAL_BYTES_PATH, RESCTRL_PATH,
> +			mongrp, sock_num);
> +	else if (ctrlgrp && !mongrp)
> +		sprintf(mbm_total_path, CON_MBM_LOCAL_BYTES_PATH, RESCTRL_PATH,
> +			ctrlgrp, sock_num);
> +	else if (!ctrlgrp && !mongrp)
> +		sprintf(mbm_total_path, MBM_LOCAL_BYTES_PATH, RESCTRL_PATH,
> +			sock_num);
> +}
> +
> +/*
> + * 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)
> +{
> +	char sock_num;
> +
> +	sock_num = get_sock_num(cpu_no);
> +	if (sock_num < 0)
> +		return;
> +
> +	if (strcmp(resctrl_val, "mbm") == 0)
> +		set_mbm_path(ctrlgrp, mongrp, sock_num);
> +
> +	if ((strcmp(resctrl_val, "mba") == 0)) {
> +		if (ctrlgrp)
> +			sprintf(mbm_total_path, CON_MBM_LOCAL_BYTES_PATH,
> +				RESCTRL_PATH, ctrlgrp, sock_num);
> +		else
> +			sprintf(mbm_total_path, MBM_LOCAL_BYTES_PATH,
> +				RESCTRL_PATH, sock_num);
> +	}
> +}
> +
> +/*
> + * 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;
> +}
> 

Powered by blists - more mailing lists