page-demo.c

#ifndef __powerpc64__
#error "unsupported architecture"
#endif

#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include <assert.h>
#include <unistd.h>
#include <sys/mman.h>
#include <perfmon/pfmlib_perf_event.h>

static const char *evname[] = {
//	"page-faults",
//	"cpu-cycles",
//	"L1-dcache-loads",
	"L1-dcache-load-misses",
//	"L1-dcache-stores",
//	"dTLB-load-misses",
//	"dTLB-store-misses",
};

static inline
void perf_event_reset(int fd)
{
	assert(!ioctl(fd, PERF_EVENT_IOC_RESET, 0));
}

static inline
void perf_event_enable(int fd)
{
	assert(!ioctl(fd, PERF_EVENT_IOC_ENABLE, 0));
}

static inline
void perf_event_disable(int fd)
{
	assert(!ioctl(fd, PERF_EVENT_IOC_DISABLE, 0));
}

static inline
unsigned long perf_event_read(int fd)
{
	unsigned long count;
	assert(read(fd, &count, sizeof(unsigned long)) > 0);
	return count;
}

static
int perf_event_prepare(const char *name, int level, int group)
{
	struct perf_event_attr attr;
	int fd;

	memset(&attr, 0, sizeof(struct perf_event_attr));
	assert(pfm_get_perf_event_encoding(name, level, &attr,
					   NULL, NULL) == PFM_SUCCESS);
	attr.disabled = 1;
	assert((fd = perf_event_open(&attr, 0, -1, group, 0)) > 0);
	perf_event_reset(fd);

	return fd;
}

__attribute__((optimize("0")))
int main(int argc, char **argv)
{
	unsigned long pgsize, clsize, clmask;
	volatile unsigned int *addr;
	int *fd, evcount, idx, i;
	void *claddr;

	evcount = sizeof(evname) / sizeof(char *);

	pfm_initialize();
	pgsize = getpagesize();
	clsize = sysconf(_SC_LEVEL1_DCACHE_LINESIZE);
	clmask = ~(clsize - 1);
	printf("pid = %d, page size = %lu, "
	       "L1 dcache line size = %lu events = %d\n",
	       getpid(), pgsize, clsize, evcount);

	assert((fd = (int *) malloc(evcount * sizeof(int))));
	for (i = 0; i < evcount; i++)
		fd[i] = perf_event_prepare(evname[i], PFM_PLM3,
					   !i ? -1 : fd[0]);

	addr = (unsigned int *) mmap(NULL, pgsize, PROT_WRITE,
				     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
	perf_event_enable(fd[0]);

	for (i = 0; i < clsize; i++) {
		idx = i * clsize / sizeof(unsigned int);
		*(addr + idx) = 0xdeadbeef;
		claddr = (void *) (((unsigned long) (addr + idx)) & clmask);
		__builtin___clear_cache(claddr, claddr + clsize);
	}

//	assert(!mprotect((void *) addr, pgsize, PROT_READ | PROT_WRITE | PROT_EXEC));
//	__builtin___clear_cache(addr, addr + 1);
//	addr[0] = 0xdeadbeef;
//	__builtin___clear_cache(addr, addr + 1);

	perf_event_disable(fd[0]);

	for (i = 0; i < evcount; i++)
		printf("%s = %lu\n", evname[i], perf_event_read(fd[i]));

	munmap((void *) addr, pgsize);

	return EXIT_SUCCESS;
}