perf_counter: fix the group mess

With these two patches the below proglet gives:

# ./test
EVNT: 0xffffffff811c0f4c scale: nan ID: 37 CNT: 1006180 ID: 38 CNT: 1010230 ID: 39 CNT: 1010332 ID: 40 CNT: 1010420

I can't seem to catch the SIGIO thing, but at least all the counters report more or less the same value. Also, that scale stuff doesn't seem to quite work yet.. I'll be poking at that too.

---
#include "perf.h"

#include <sys/types.h>
#include <sys/ioctl.h>
#include <sys/mman.h>
#include <stdlib.h>
#include <stdio.h>
#include <time.h>
#include <unistd.h>
#include <signal.h>
#include <fcntl.h>
#include <errno.h>
#include <string.h>

void work(void)

{
	int i;

	for (i = 0; i < 1000000000; i++) {
		asm("nop");
		asm("nop");
		asm("nop");
		asm("nop");
		asm("nop");
		asm("nop");
		asm("nop");
	}
}

unsigned long page_size;
int fd = -1, fd1 = 0;
pid_t me;
void *output;

void handle_sigio(int sig)

{
	printf("signal\n"); // not signal safe
	fflush(stdout);

	ioctl(fd, PERF_COUNTER_IOC_REFRESH, 1);
}

static unsigned long mmap_read_head(void)

{
	struct perf_counter_mmap_page *pc = output;
	long head;

	head = pc->data_head;
	rmb();

	return head;
}

static void *mmap_read_base(void)

{
	return output + page_size;
}

struct event {
	struct perf_event_header header;

	u64 ip;
	u64 nr;
	u64 time_enabled;
	u64 time_running;
	struct {
		u64 val;
		u64 id;
	} cnt[0];
};

int main(int argc, char **argv)

{
	struct perf_counter_attr attr;
	unsigned long offset = 0, head;
	int err, i;

	page_size = sysconf(_SC_PAGE_SIZE);
	me = getpid();

	memset(&attr, 0, sizeof(attr));

	attr.type = PERF_TYPE_HARDWARE;
	attr.config = PERF_COUNT_HW_CPU_CYCLES;
	attr.sample_period = 1000000;
	attr.sample_type = PERF_SAMPLE_IP |
			   PERF_SAMPLE_READ;
	attr.read_format = PERF_FORMAT_TOTAL_TIME_RUNNING |
			   PERF_FORMAT_TOTAL_TIME_ENABLED |
			   PERF_FORMAT_ID |
			   PERF_FORMAT_GROUP;
	attr.disabled = 1;
	attr.wakeup_events = 1;

	fd  = sys_perf_counter_open(&attr, me, -1, fd, 0);
	if (fd <= 0) {
		perror("FAIL fd: ");
		exit(-1);
	}

	attr.sample_period = 0;
	attr.disabled = 0;

	for (i = 0; i < 3; i++) {
		fd1 = sys_perf_counter_open(&attr, me, -1, fd, 0);
		if (fd1 <= 0) {
			perror("FAIL fd1: ");
			exit(-1);
		}
	}

	signal(SIGIO, handle_sigio);
	err = fcntl(fd, F_SETOWN, me);
	if (err == -1) {
		perror("FAIL fcntl: ");
		exit(-1);
	}

	output = mmap(NULL, page_size * 3, PROT_READ, MAP_SHARED, fd, 0);
	if (output == ((void *)-1)) {
		perror("FAIL mmap:");
		exit(-1);
	}

	ioctl(fd, PERF_COUNTER_IOC_REFRESH, 1);

	work();

	ioctl(fd, PERF_COUNTER_IOC_DISABLE, 0);

	head = mmap_read_head();

	for (; offset < head; ) {
		struct event *evnt = mmap_read_base() + offset;

		offset += evnt->header.size;

		printf("EVNT: %p scale: %f ", (void *)evnt->ip,
				((double)evnt->time_running)/evnt->time_enabled
				);
		for (i = 0; i < evnt->nr; i++) {
			printf("ID: %Lu CNT: %Lu ", 
					evnt->cnt[i].id, evnt->cnt[i].val);
		}
		printf("\n");
	}

	return 0;
}