int create_perf_stat_counter(int pid, int system_wide)
{
struct perf_event_attr attr; //cache miss
memcpy(&attr, attrs, sizeof(struct perf_event_attr));
if (scale)
attr.read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
PERF_FORMAT_TOTAL_TIME_RUNNING;
if (system_wide) {
unsigned int cpu;
for (cpu = 0; cpu < nr_cpus; cpu++) {
fd[cpu] = sys_perf_event_open(&attr, -1, cpu, -1, 0);
}
return fd[cpu - 1];
} else {
attr.inherit = inherit;
attr.disabled = 0;
attr.enable_on_exec = 1;
return sys_perf_event_open(&attr, pid, -1, -1, 0);
}
}
int main(int argc, char *argv[])
{
(void) argc; (void) argv;
perf_event_attr attr, attr2;
// Execution sur un seul CPU
pid_t tid = gettid();
cpu_set_t mask;
CPU_ZERO(&mask);
CPU_SET(0, &mask);
qDebug() << sched_setaffinity(0, sizeof(mask), &mask);
//(perf_hw_cache_id) | (perf_hw_cache_op_id << 8) |
//(perf_hw_cache_op_result_id << 16)
// premiere structure d'attributs
memset(&attr, 0, sizeof(attr));
attr.size = sizeof(attr);
attr.type = PERF_TYPE_HW_CACHE;
attr.config = (PERF_COUNT_HW_CACHE_L1D) | (PERF_COUNT_HW_CACHE_OP_READ << 8) | (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16);
attr.inherit = 1;
attr.disabled = 0;
// Deuxieme structure
memset(&attr2, 0, sizeof(attr2));
attr2.size = sizeof(attr2);
attr2.type = PERF_TYPE_HW_CACHE;
attr2.config = (PERF_COUNT_HW_CACHE_L1D) | (PERF_COUNT_HW_CACHE_OP_READ << 8) | (PERF_COUNT_HW_CACHE_RESULT_MISS << 16);
attr2.inherit = 1;
attr2.disabled = 0;
//pid == 0 and cpu == -1
//This measures the calling process/thread on any CPU.
int fdRef = sys_perf_event_open(&attr, tid, -1, -1, 0);
int fdMiss = sys_perf_event_open(&attr2, tid, -1, -1, 0);
//int fdMiss = sys_perf_event_open(&attr2, tid, -1, fdRef, 0);
QMap<int, double> results;
for (qint64 len = 1; len < (1 << 26); len *= 2) {
double miss_rate = experiment(len, fdRef, fdMiss);
results.insert(len * sizeof(int), miss_rate);
}
// report
std::cout << "size,miss_rate" << std::endl;
for (int size : results.keys()) {
std::cout << size << "," << results[size] << std::endl;
}
close(fdRef);
close(fdMiss);
return 0;
}
static int
open_counter(size_t nr_switches)
{
struct perf_event_attr attr;
int fd;
struct f_owner_ex own;
memset(&attr, 0, sizeof(attr));
attr.size = sizeof(attr);
attr.type = PERF_TYPE_SOFTWARE;
attr.config = PERF_COUNT_SW_CONTEXT_SWITCHES;
attr.disabled = 1;
attr.sample_period = nr_switches;
fd = sys_perf_event_open(&attr, 0/*self*/, -1/*any cpu*/, -1, 0);
if (0 > fd)
err(1, "perf_event_open(cs, period=%u)", nr_switches);
if (fcntl(fd, F_SETFL, O_ASYNC))
err(1, "fcntl(O_ASYNC)");
own.type = F_OWNER_TID;
own.pid = sys_gettid();
if (fcntl(fd, F_SETOWN_EX, &own))
err(1, "fcntl(SETOWN_EX)");
if (fcntl(fd, F_SETSIG, SIGIO))
err(1, "fcntl(SETSIG, SIGIO)");
return fd;
}
static int perf_event_open_gen(struct perf_event_attr *attr,
int group_fd)
{
// Group leader: Request time values for reference (so we can
// detect multiplexing). Also ask for the whole group to return
// results at the same time.
#ifdef USE_PERF_FORMAT_GROUP
if (group_fd == -1) {
#endif
attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED
| PERF_FORMAT_TOTAL_TIME_RUNNING
#ifdef USE_PERF_FORMAT_GROUP
| PERF_FORMAT_GROUP
#endif
;
#ifdef USE_PERF_FORMAT_GROUP
}
#endif
// Exclude kernel and hypervisor, start disabled
attr->exclude_kernel = 1;
attr->exclude_hv = 1;
attr->disabled = 1;
#ifndef USE_PERF_FORMAT_GROUP
attr->inherit = 1;
#endif
// Create the performance counter
return sys_perf_event_open(attr, 0, -1, group_fd, 0);
}
int perf_open(int inherit, counter_config config)
{
perf_event_attr attr;
memset(&attr,0,sizeof(attr));
attr.size = sizeof(attr);
attr.exclude_kernel = 1;
attr.type = PERF_TYPE_HARDWARE;
switch(config){
case INST :
attr.config = PERF_COUNT_HW_INSTRUCTIONS;
break;
case HW_CYCLES :
attr.config = PERF_COUNT_HW_CPU_CYCLES;
break;
default :
fprintf(stderr, "Invalid counter configuration.\n");
exit(EXIT_FAILURE);
break;
}
attr.inherit = inherit;
attr.disabled = 0;
return sys_perf_event_open(&attr, 0, -1, -1, 0);
}
/*
* perf_open()
* open perf, get leader and perf fd's
*/
int perf_open(stress_perf_t *sp)
{
size_t i;
if (!sp)
return -1;
if (g_shared->perf.no_perf)
return -1;
(void)memset(sp, 0, sizeof(*sp));
sp->perf_opened = 0;
for (i = 0; i < STRESS_PERF_MAX; i++) {
sp->perf_stat[i].fd = -1;
sp->perf_stat[i].counter = 0;
}
for (i = 0; i < STRESS_PERF_MAX && perf_info[i].label; i++) {
if (perf_info[i].config != UNRESOLVED) {
struct perf_event_attr attr;
(void)memset(&attr, 0, sizeof(attr));
attr.type = perf_info[i].type;
attr.config = perf_info[i].config;
attr.disabled = 1;
attr.inherit = 1;
attr.read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
PERF_FORMAT_TOTAL_TIME_RUNNING;
attr.size = sizeof(attr);
sp->perf_stat[i].fd =
sys_perf_event_open(&attr, 0, -1, -1, 0);
if (sp->perf_stat[i].fd > -1)
sp->perf_opened++;
}
}
if (!sp->perf_opened) {
int ret;
ret = shim_pthread_spin_lock(&g_shared->perf.lock);
if (!ret) {
pr_dbg("perf: spin lock on perf.lock failed: %d (%s)\n",
ret, strerror(ret));
return -1;
}
if (!g_shared->perf.no_perf) {
pr_dbg("perf: perf_event_open failed, no "
"perf events [%u]\n", getpid());
g_shared->perf.no_perf = true;
}
ret = shim_pthread_spin_unlock(&g_shared->perf.lock);
if (!ret) {
pr_dbg("perf: spin unlock on perf.lock failed: %d (%s)\n",
ret, strerror(ret));
return -1;
}
return -1;
}
return 0;
}
static int sampling_start(int *pmu_fd, int freq)
{
int i;
struct perf_event_attr pe_sample_attr = {
.type = PERF_TYPE_SOFTWARE,
.freq = 1,
.sample_period = freq,
.config = PERF_COUNT_SW_CPU_CLOCK,
.inherit = 1,
};
for (i = 0; i < nr_cpus; i++) {
pmu_fd[i] = sys_perf_event_open(&pe_sample_attr, -1 /* pid */, i,
-1 /* group_fd */, 0 /* flags */);
if (pmu_fd[i] < 0) {
fprintf(stderr, "ERROR: Initializing perf sampling\n");
return 1;
}
assert(ioctl(pmu_fd[i], PERF_EVENT_IOC_SET_BPF,
prog_fd[0]) == 0);
assert(ioctl(pmu_fd[i], PERF_EVENT_IOC_ENABLE, 0) == 0);
}
return 0;
}
static void sampling_end(int *pmu_fd)
{
int i;
for (i = 0; i < nr_cpus; i++)
close(pmu_fd[i]);
}
static int __event(bool is_x, void *addr, struct perf_event_attr *attr)
{
int fd;
memset(attr, 0, sizeof(struct perf_event_attr));
attr->type = PERF_TYPE_BREAKPOINT;
attr->size = sizeof(struct perf_event_attr);
attr->config = 0;
attr->bp_type = is_x ? HW_BREAKPOINT_X : HW_BREAKPOINT_W;
attr->bp_addr = (unsigned long) addr;
attr->bp_len = sizeof(long);
attr->sample_period = 1;
attr->sample_type = PERF_SAMPLE_IP;
attr->exclude_kernel = 1;
attr->exclude_hv = 1;
fd = sys_perf_event_open(attr, -1, 0, -1,
perf_event_open_cloexec_flag());
if (fd < 0) {
pr_debug("failed opening event %llx\n", attr->config);
return TEST_FAIL;
}
return fd;
}
static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
struct thread_map *threads)
{
int cpu, thread;
if (evsel->fd == NULL &&
perf_evsel__alloc_fd(evsel, cpus->nr, threads->nr) < 0)
return -1;
for (cpu = 0; cpu < cpus->nr; cpu++) {
for (thread = 0; thread < threads->nr; thread++) {
FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr,
threads->map[thread],
cpus->map[cpu], -1, 0);
if (FD(evsel, cpu, thread) < 0)
goto out_close;
}
}
return 0;
out_close:
do {
while (--thread >= 0) {
close(FD(evsel, cpu, thread));
FD(evsel, cpu, thread) = -1;
}
thread = threads->nr;
} while (--cpu >= 0);
return -1;
}
bool PerfGroup::prepareCPU(const int cpu) {
logg->logMessage("%s(%s:%i): Onlining cpu %i", __FUNCTION__, __FILE__, __LINE__, cpu);
for (int i = 0; i < ARRAY_LENGTH(mKeys); ++i) {
if (mKeys[i] < 0) {
continue;
}
const int offset = i * gSessionData->mCores;
if (mFds[cpu + offset] >= 0) {
logg->logMessage("%s(%s:%i): cpu already online or not correctly cleaned up", __FUNCTION__, __FILE__, __LINE__);
return false;
}
logg->logMessage("%s(%s:%i): perf_event_open cpu: %i type: %lli config: %lli sample: %lli sample_type: %lli", __FUNCTION__, __FILE__, __LINE__, cpu, (long long)mAttrs[i].type, (long long)mAttrs[i].config, (long long)mAttrs[i].sample_period, (long long)mAttrs[i].sample_type);
mFds[cpu + offset] = sys_perf_event_open(&mAttrs[i], -1, cpu, i == 0 ? -1 : mFds[cpu], i == 0 ? 0 : PERF_FLAG_FD_OUTPUT);
if (mFds[cpu + offset] < 0) {
logg->logMessage("%s(%s:%i): failed %s", __FUNCTION__, __FILE__, __LINE__, strerror(errno));
continue;
}
if (!mPb->useFd(cpu, mFds[cpu + offset], mFds[cpu])) {
logg->logMessage("%s(%s:%i): PerfBuffer::useFd failed", __FUNCTION__, __FILE__, __LINE__);
return false;
}
}
return true;
}
bool
PerfMeasurement::canMeasureSomething()
{
// Find out if the kernel implements the performance measurement
// API. If it doesn't, syscall(__NR_perf_event_open, ...) is
// guaranteed to return -1 and set errno to ENOSYS.
//
// We set up input parameters that should provoke an EINVAL error
// from a kernel that does implement perf_event_open, but we can't
// be sure it will (newer kernels might add more event types), so
// we have to take care to close any valid fd it might return.
struct perf_event_attr attr;
memset(&attr, 0, sizeof(attr));
attr.size = sizeof(attr);
attr.type = PERF_TYPE_MAX;
int fd = sys_perf_event_open(&attr, 0, -1, -1, 0);
if (fd >= 0) {
close(fd);
return true;
} else {
return errno != ENOSYS;
}
}
static bool perfPoll(struct perf_event_attr *const pea) {
int fd = sys_perf_event_open(pea, -1, 0, -1, 0);
if (fd < 0) {
return false;
}
close(fd);
return true;
}
int PerfGroup::prepareCPU(const int cpu, Monitor *const monitor) {
logg.logMessage("Onlining cpu %i", cpu);
for (int i = 0; i < ARRAY_LENGTH(mKeys); ++i) {
if (mKeys[i] < 0) {
continue;
}
if ((cpu != 0) && !(mFlags[i] & PERF_GROUP_PER_CPU)) {
continue;
}
if ((mFlags[i] & PERF_GROUP_PER_CPU) && !(mFlags[i] & PERF_GROUP_ALL_CLUSTERS) && gSessionData.mSharedData->mClusters[gSessionData.mSharedData->mClusterIds[cpu]] != mClusters[i]) {
continue;
}
const int offset = i * gSessionData.mCores + cpu;
if (mFds[offset] >= 0) {
logg.logMessage("cpu already online or not correctly cleaned up");
return PG_FAILURE;
}
logg.logMessage("perf_event_open cpu: %i type: %i config: %lli sample: %lli sample_type: 0x%llx pinned: %lli mmap: %lli comm: %lli freq: %lli task: %lli sample_id_all: %lli", cpu, mAttrs[i].type, mAttrs[i].config, mAttrs[i].sample_period, mAttrs[i].sample_type, mAttrs[i].pinned, mAttrs[i].mmap, mAttrs[i].comm, mAttrs[i].freq, mAttrs[i].task, mAttrs[i].sample_id_all);
mFds[offset] = sys_perf_event_open(&mAttrs[i], -1, cpu, mAttrs[i].pinned ? -1 : mFds[mLeaders[getEffectiveType(mAttrs[i].type, mFlags[i])] * gSessionData.mCores + cpu], mAttrs[i].pinned ? 0 : PERF_FLAG_FD_OUTPUT);
if (mFds[offset] < 0) {
logg.logMessage("failed %s", strerror(errno));
if (errno == ENODEV) {
// The core is offline
return PG_CPU_OFFLINE;
}
if (errno == ENOENT) {
// This event doesn't apply to this CPU but should apply to a different one, ex bL
continue;
}
logg.logMessage("perf_event_open failed");
return PG_FAILURE;
}
if (!mPb->useFd(cpu, mFds[offset])) {
logg.logMessage("PerfBuffer::useFd failed");
return PG_FAILURE;
}
if (!monitor->add(mFds[offset])) {
logg.logMessage("Monitor::add failed");
return PG_FAILURE;
}
}
return PG_SUCCESS;
}
static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
struct thread_map *threads, bool group,
struct xyarray *group_fds)
{
int cpu, thread;
unsigned long flags = 0;
int pid = -1, err;
if (evsel->fd == NULL &&
perf_evsel__alloc_fd(evsel, cpus->nr, threads->nr) < 0)
return -ENOMEM;
if (evsel->cgrp) {
flags = PERF_FLAG_PID_CGROUP;
pid = evsel->cgrp->fd;
}
for (cpu = 0; cpu < cpus->nr; cpu++) {
int group_fd = group_fds ? GROUP_FD(group_fds, cpu) : -1;
for (thread = 0; thread < threads->nr; thread++) {
if (!evsel->cgrp)
pid = threads->map[thread];
FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr,
pid,
cpus->map[cpu],
group_fd, flags);
if (FD(evsel, cpu, thread) < 0) {
err = -errno;
goto out_close;
}
if (group && group_fd == -1)
group_fd = FD(evsel, cpu, thread);
}
}
return 0;
out_close:
do {
while (--thread >= 0) {
close(FD(evsel, cpu, thread));
FD(evsel, cpu, thread) = -1;
}
thread = threads->nr;
} while (--cpu >= 0);
return err;
}
static int perf_flag_probe(void)
{
/* use 'safest' configuration as used in perf_evsel__fallback() */
struct perf_event_attr attr = {
.type = PERF_COUNT_SW_CPU_CLOCK,
.config = PERF_COUNT_SW_CPU_CLOCK,
};
int fd;
int err;
/* check cloexec flag */
fd = sys_perf_event_open(&attr, 0, -1, -1,
PERF_FLAG_FD_CLOEXEC);
err = errno;
if (fd >= 0) {
close(fd);
return 1;
}
WARN_ONCE(err != EINVAL,
"perf_event_open(..., PERF_FLAG_FD_CLOEXEC) failed with unexpected error %d (%s)\n",
err, strerror(err));
/* not supported, confirm error related to PERF_FLAG_FD_CLOEXEC */
fd = sys_perf_event_open(&attr, 0, -1, -1, 0);
err = errno;
if (WARN_ONCE(fd < 0,
"perf_event_open(..., 0) failed unexpectedly with error %d (%s)\n",
err, strerror(err)))
return -1;
close(fd);
return 0;
}
static int sampling_do_open(PyObject *obj)
{
int tid;
int flags;
PyPerfEvent *ev;
struct f_owner_ex fown;
struct perf_event_attr attr;
if (!PyPerfEvent_CheckExact(obj))
return -1;
ev = (PyPerfEvent *)obj;
tid = gettid();
/* Open the perf event */
attr = ev->attr;
attr.disabled = 0; /* do not start the event yet */
attr.watermark = 1;
attr.wakeup_events = 1;
ev->fd = sys_perf_event_open(&attr, tid, -1, -1, 0);
if (ev->fd < 0) {
ev->status = EVENT_STATUS_FAILED;
return -1;
}
/* Configure fasync */
fown.type = F_OWNER_TID;
fown.pid = tid;
if (fcntl(ev->fd, F_SETOWN_EX, &fown) < 0)
goto fail;
flags = fcntl(ev->fd, F_GETFL);
if (fcntl(ev->fd, F_SETFL, flags | FASYNC | O_ASYNC) < 0)
goto fail;
ev->status = EVENT_STATUS_OPENED;
xev = ev;
__sync_synchronize();
ioctl(xev->fd, PERF_EVENT_IOC_REFRESH, 1);
ioctl(xev->fd, PERF_EVENT_IOC_ENABLE, 0);
return 0;
fail:
close(ev->fd);
ev->fd = -1;
ev->status = EVENT_STATUS_FAILED;
PyErr_SetString(PyExc_RuntimeError, "fcntl() failed");
return -1;
}
static void test_perf_event_task(struct perf_event_attr *attr)
{
int pmu_fd;
/* open task bound event */
pmu_fd = sys_perf_event_open(attr, 0, -1, -1, 0);
if (pmu_fd < 0) {
printf("sys_perf_event_open failed\n");
return;
}
assert(ioctl(pmu_fd, PERF_EVENT_IOC_SET_BPF, prog_fd[0]) == 0);
assert(ioctl(pmu_fd, PERF_EVENT_IOC_ENABLE, 0) == 0);
system("dd if=/dev/zero of=/dev/null count=5000k");
print_stacks();
close(pmu_fd);
}
static int self_open_counters(void)
{
struct perf_event_attr attr;
int fd;
memset(&attr, 0, sizeof(attr));
attr.type = PERF_TYPE_SOFTWARE;
attr.config = PERF_COUNT_SW_TASK_CLOCK;
fd = sys_perf_event_open(&attr, 0, -1, -1, 0);
if (fd < 0)
pr_err("Error: sys_perf_event_open() syscall returned "
"with %d (%s)\n", fd, strerror(errno));
return fd;
}
static int create_perf_counter(struct perf_setup_s *p)
{
struct cpu_map *cpus;
int cpu;
cpus = cpu_map__new(NULL);
if (p == NULL)
return PERF_PERIODIC_ERROR;
for (cpu = 0; cpu < cpus->nr; cpu++) {
if (((1 << cpu) & cpumask) == 0)
continue;
p->perf_fd[cpu] = sys_perf_event_open(p->attr, target_pid, cpu,
-1, 0);
if (p->perf_fd[cpu] < 0)
return PERF_PERIODIC_ERROR;
}
return 0;
}
static int self_open_counters(struct perf_sched *sched, unsigned long cur_task)
{
struct perf_event_attr attr;
char sbuf[STRERR_BUFSIZE], info[STRERR_BUFSIZE];
int fd;
struct rlimit limit;
bool need_privilege = false;
memset(&attr, 0, sizeof(attr));
attr.type = PERF_TYPE_SOFTWARE;
attr.config = PERF_COUNT_SW_TASK_CLOCK;
force_again:
fd = sys_perf_event_open(&attr, 0, -1, -1,
perf_event_open_cloexec_flag());
if (fd < 0) {
if (errno == EMFILE) {
if (sched->force) {
BUG_ON(getrlimit(RLIMIT_NOFILE, &limit) == -1);
limit.rlim_cur += sched->nr_tasks - cur_task;
if (limit.rlim_cur > limit.rlim_max) {
limit.rlim_max = limit.rlim_cur;
need_privilege = true;
}
if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
if (need_privilege && errno == EPERM)
strcpy(info, "Need privilege\n");
} else
goto force_again;
} else
strcpy(info, "Have a try with -f option\n");
}
pr_err("Error: sys_perf_event_open() syscall returned "
"with %d (%s)\n%s", fd,
strerror_r(errno, sbuf, sizeof(sbuf)), info);
exit(EXIT_FAILURE);
}
return fd;
}
static void test_perf_event_all_cpu(struct perf_event_attr *attr)
{
int nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
int *pmu_fd = malloc(nr_cpus * sizeof(int));
int i;
/* open perf_event on all cpus */
for (i = 0; i < nr_cpus; i++) {
pmu_fd[i] = sys_perf_event_open(attr, -1, i, -1, 0);
if (pmu_fd[i] < 0) {
printf("sys_perf_event_open failed\n");
goto all_cpu_err;
}
assert(ioctl(pmu_fd[i], PERF_EVENT_IOC_SET_BPF, prog_fd[0]) == 0);
assert(ioctl(pmu_fd[i], PERF_EVENT_IOC_ENABLE, 0) == 0);
}
system("dd if=/dev/zero of=/dev/null count=5000k");
print_stacks();
all_cpu_err:
for (i--; i >= 0; i--)
close(pmu_fd[i]);
free(pmu_fd);
}
static int load_and_attach(const char *event, struct bpf_insn *prog, int size)
{
bool is_socket = strncmp(event, "socket", 6) == 0;
bool is_kprobe = strncmp(event, "kprobe/", 7) == 0;
bool is_kretprobe = strncmp(event, "kretprobe/", 10) == 0;
bool is_tracepoint = strncmp(event, "tracepoint/", 11) == 0;
bool is_xdp = strncmp(event, "xdp", 3) == 0;
bool is_perf_event = strncmp(event, "perf_event", 10) == 0;
bool is_cgroup_skb = strncmp(event, "cgroup/skb", 10) == 0;
bool is_cgroup_sk = strncmp(event, "cgroup/sock", 11) == 0;
size_t insns_cnt = size / sizeof(struct bpf_insn);
enum bpf_prog_type prog_type;
char buf[256];
int fd, efd, err, id;
struct perf_event_attr attr = {};
attr.type = PERF_TYPE_TRACEPOINT;
attr.sample_type = PERF_SAMPLE_RAW;
attr.sample_period = 1;
attr.wakeup_events = 1;
if (is_socket) {
prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
} else if (is_kprobe || is_kretprobe) {
prog_type = BPF_PROG_TYPE_KPROBE;
} else if (is_tracepoint) {
prog_type = BPF_PROG_TYPE_TRACEPOINT;
} else if (is_xdp) {
prog_type = BPF_PROG_TYPE_XDP;
} else if (is_perf_event) {
prog_type = BPF_PROG_TYPE_PERF_EVENT;
} else if (is_cgroup_skb) {
prog_type = BPF_PROG_TYPE_CGROUP_SKB;
} else if (is_cgroup_sk) {
prog_type = BPF_PROG_TYPE_CGROUP_SOCK;
} else {
printf("Unknown event '%s'\n", event);
return -1;
}
fd = bpf_load_program(prog_type, prog, insns_cnt, license, kern_version,
bpf_log_buf, BPF_LOG_BUF_SIZE);
if (fd < 0) {
printf("bpf_load_program() err=%d\n%s", errno, bpf_log_buf);
return -1;
}
prog_fd[prog_cnt++] = fd;
if (is_xdp || is_perf_event || is_cgroup_skb || is_cgroup_sk)
return 0;
if (is_socket) {
event += 6;
if (*event != '/')
return 0;
event++;
if (!isdigit(*event)) {
printf("invalid prog number\n");
return -1;
}
return populate_prog_array(event, fd);
}
if (is_kprobe || is_kretprobe) {
if (is_kprobe)
event += 7;
else
event += 10;
if (*event == 0) {
printf("event name cannot be empty\n");
return -1;
}
if (isdigit(*event))
return populate_prog_array(event, fd);
snprintf(buf, sizeof(buf),
"echo '%c:%s %s' >> /sys/kernel/debug/tracing/kprobe_events",
is_kprobe ? 'p' : 'r', event, event);
err = system(buf);
if (err < 0) {
printf("failed to create kprobe '%s' error '%s'\n",
event, strerror(errno));
return -1;
}
strcpy(buf, DEBUGFS);
strcat(buf, "events/kprobes/");
strcat(buf, event);
strcat(buf, "/id");
} else if (is_tracepoint) {
event += 11;
if (*event == 0) {
printf("event name cannot be empty\n");
return -1;
}
strcpy(buf, DEBUGFS);
strcat(buf, "events/");
strcat(buf, event);
strcat(buf, "/id");
}
efd = open(buf, O_RDONLY, 0);
if (efd < 0) {
printf("failed to open event %s\n", event);
return -1;
}
err = read(efd, buf, sizeof(buf));
if (err < 0 || err >= sizeof(buf)) {
printf("read from '%s' failed '%s'\n", event, strerror(errno));
return -1;
}
close(efd);
buf[err] = 0;
id = atoi(buf);
attr.config = id;
efd = sys_perf_event_open(&attr, -1/*pid*/, 0/*cpu*/, -1/*group_fd*/, 0);
if (efd < 0) {
printf("event %d fd %d err %s\n", id, efd, strerror(errno));
return -1;
}
event_fd[prog_cnt - 1] = efd;
ioctl(efd, PERF_EVENT_IOC_ENABLE, 0);
ioctl(efd, PERF_EVENT_IOC_SET_BPF, fd);
return 0;
}
static int perf_flag_probe(void)
{
/* use 'safest' configuration as used in perf_evsel__fallback() */
struct perf_event_attr attr = {
.type = PERF_TYPE_SOFTWARE,
.config = PERF_COUNT_SW_CPU_CLOCK,
.exclude_kernel = 1,
};
int fd;
int err;
int cpu;
pid_t pid = -1;
char sbuf[STRERR_BUFSIZE];
cpu = sched_getcpu();
if (cpu < 0)
cpu = 0;
/*
* Using -1 for the pid is a workaround to avoid gratuitous jump label
* changes.
*/
while (1) {
/* check cloexec flag */
fd = sys_perf_event_open(&attr, pid, cpu, -1,
PERF_FLAG_FD_CLOEXEC);
if (fd < 0 && pid == -1 && errno == EACCES) {
pid = 0;
continue;
}
break;
}
err = errno;
if (fd >= 0) {
close(fd);
return 1;
}
WARN_ONCE(err != EINVAL && err != EBUSY,
"perf_event_open(..., PERF_FLAG_FD_CLOEXEC) failed with unexpected error %d (%s)\n",
err, strerror_r(err, sbuf, sizeof(sbuf)));
/* not supported, confirm error related to PERF_FLAG_FD_CLOEXEC */
while (1) {
fd = sys_perf_event_open(&attr, pid, cpu, -1, 0);
if (fd < 0 && pid == -1 && errno == EACCES) {
pid = 0;
continue;
}
break;
}
err = errno;
if (fd >= 0)
close(fd);
if (WARN_ONCE(fd < 0 && err != EBUSY,
"perf_event_open(..., 0) failed unexpectedly with error %d (%s)\n",
err, strerror_r(err, sbuf, sizeof(sbuf))))
return -1;
return 0;
}
unsigned long perf_event_open_cloexec_flag(void)
{
static bool probed;
if (!probed) {
if (perf_flag_probe() <= 0)
flag = 0;
probed = true;
}
return flag;
}
static void create_perf_event(void)
{
struct perf_event_attr attr;
int ret;
struct {
__u64 count;
__u64 time_enabled;
__u64 time_running;
__u64 id;
} read_data;
memset(&attr, 0, sizeof(attr));
attr.read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
PERF_FORMAT_TOTAL_TIME_RUNNING |
PERF_FORMAT_ID;
attr.sample_freq = 0;
attr.sample_period = 1;
attr.sample_type |= PERF_SAMPLE_RAW;
attr.mmap = 1;
attr.comm = 1;
attr.inherit = 0;
attr.disabled = 1;
attr.type = PERF_TYPE_TRACEPOINT;
attr.config = get_trace_type();
if (attr.config <= 0)
return;
perf_fd = sys_perf_event_open(&attr, -1, 0, -1, 0);
if (perf_fd < 0) {
fprintf(stderr, "Perf syscall failed: %i / %i (%s)\n", perf_fd, errno, strerror(errno));
return;
}
if (read(perf_fd, &read_data, sizeof(read_data)) == -1) {
perror("Unable to read perf file descriptor\n");
exit(-1);
}
fcntl(perf_fd, F_SETFL, O_NONBLOCK);
perf_mmap = mmap(NULL, (128+1)*getpagesize(),
PROT_READ | PROT_WRITE, MAP_SHARED, perf_fd, 0);
if (perf_mmap == MAP_FAILED) {
fprintf(stderr, "failed to mmap with %d (%s)\n", errno, strerror(errno));
return;
}
ret = ioctl(perf_fd, PERF_EVENT_IOC_ENABLE);
if (ret < 0)
fprintf(stderr, "failed to enable perf \n");
pc = perf_mmap;
data_mmap = perf_mmap + getpagesize();
}
HRESULT CaPerfEvent::startEvent(bool enable)
{
PerfPmuTarget& pmuTgt = m_profConfig->getPMUTarget();
const CAThreadVec& threadVec = m_profConfig->getThreadVec();
size_t nbr = m_nbrfds;
OS_OUTPUT_FORMAT_DEBUG_LOG(OS_DEBUG_LOG_DEBUG, L"nbr(%d) in startEvent", nbr);
pid_t pid = -1;
int group_fd = -1;
int ret;
int cpu;
int fd;
// Set perf_event_attr::disabled field
m_attr.disabled = ~enable;
if (pmuTgt.isSWP())
{
for (size_t i = 0; i < nbr; i++)
{
cpu = pmuTgt.m_pCpus[i];
fd = sys_perf_event_open(&m_attr, pid, cpu, group_fd, 0);
int err = errno;
if (fd < 0)
{
OS_OUTPUT_DEBUG_LOG(L"sys_perf_event_open failed..", OS_DEBUG_LOG_ERROR);
if ((EACCES == err) || (EPERM == err))
{
OS_OUTPUT_DEBUG_LOG(L"sys_perf_event_open failed due to Access Permissions...", OS_DEBUG_LOG_EXTENSIVE);
return E_ACCESSDENIED;
}
else if (ENODEV == err)
{
// invalid cpu list
OS_OUTPUT_DEBUG_LOG(L"sys_perf_event_open failed - ENODEV...", OS_DEBUG_LOG_EXTENSIVE);
return E_INVALIDARG;
}
else if (ENOENT == err)
{
// invalid cpu list
OS_OUTPUT_DEBUG_LOG(L"sys_perf_event_open failed - ENOENT...", OS_DEBUG_LOG_EXTENSIVE);
return E_INVALIDARG;
}
else if (EINVAL == err)
{
// invalid cpu list
OS_OUTPUT_DEBUG_LOG(L"sys_perf_event_open failed - EINVAL...", OS_DEBUG_LOG_EXTENSIVE);
return E_INVALIDARG;
}
else
{
OS_OUTPUT_FORMAT_DEBUG_LOG(OS_DEBUG_LOG_EXTENSIVE, L"sys_perf_event_open failed - errno(%d)", err);
return E_FAIL;
}
}
else
{
OS_OUTPUT_FORMAT_DEBUG_LOG(OS_DEBUG_LOG_EXTENSIVE, L"for cpu(%d) fd(%d)", cpu, fd);
}
m_pCountData[i].m_fd = fd;
// If the event is sampling event, get the sample-id
// We set the attr::read_format to PERF_FORMAT_TOTAL_TIME_ENABLED | PERF_TOTAL_TIME_RUNNING
// | PERF_FORMAT_ID;
// for this the format of the data returned by read() on the fd is:
// struct read_format {
// u64 value;
// u64 time_enabled;
// u64 time_running;
// u64 id;
// }
//
if (m_samplingEvent)
{
ret = readn(fd, m_pCountData[i].m_values, (PERF_MAX_NBR_VALUES * sizeof(uint64_t)));
// on error, return
if (-1 == ret)
{
OS_OUTPUT_FORMAT_DEBUG_LOG(OS_DEBUG_LOG_ERROR, L"error in reading samplie-id for event(%d, %lx), fd(%d) errno(%d)\n",
m_attr.type, m_attr.config, fd, errno);
return E_FAIL;
}
m_pCountData[i].m_sampleId = m_pCountData[i].m_values[3];
// Now push this into m_eventDataList
// TODO: Eventually we will discard pCountData stuff
OS_OUTPUT_FORMAT_DEBUG_LOG(OS_DEBUG_LOG_EXTENSIVE, L"push elemnt in m_eventDataList i(%d)\n", i);
m_eventDataList.push_back(m_pCountData[i]);
}
}
}
else
{
OS_OUTPUT_FORMAT_DEBUG_LOG(OS_DEBUG_LOG_DEBUG, L"nbrPids(%d) nbrThreads(%d).\n", pmuTgt.m_nbrPids, threadVec.size());
int cnt = 0;
for (size_t i = 0; i < pmuTgt.m_nbrCpus; i++)
{
cpu = pmuTgt.m_pCpus[i];
for (size_t j = 0; j < threadVec.size(); j++, cnt++)
{
pid = threadVec[j];
fd = sys_perf_event_open(&m_attr, pid, cpu, group_fd, 0);
if (-1 == fd)
{
OS_OUTPUT_FORMAT_DEBUG_LOG(OS_DEBUG_LOG_ERROR, L"sys_perf_event_open failed cpu(%d), pid(%d), cnt(%d)", cpu, pid, cnt);
}
else
{
OS_OUTPUT_FORMAT_DEBUG_LOG(OS_DEBUG_LOG_EXTENSIVE, L"for cpu(%d), pid(%d), cnt(%d), fd(%d)", cpu, pid, cnt, fd);
}
m_pCountData[cnt].m_fd = fd;
// If the event is sampling event, get the sample-id
if (m_samplingEvent)
{
ret = readn(fd, m_pCountData[cnt].m_values, (PERF_MAX_NBR_VALUES * sizeof(uint64_t)));
// on error, return
if (-1 == ret)
{
OS_OUTPUT_FORMAT_DEBUG_LOG(OS_DEBUG_LOG_ERROR, L"error in reading samplie-id for event(%d, %lx), fd(%d) errno(%d)",
m_attr.type, m_attr.config, fd, errno);
return E_FAIL;
}
// If the event is sampling event, get the sample-id
// We set the attr::read_format to PERF_FORMAT_TOTAL_TIME_ENABLED | PERF_TOTAL_TIME_RUNNING
// | PERF_FORMAT_ID;
// for this the format of the data returned by read() on the fd is:
// struct read_format {
// u64 value;
// u64 time_enabled;
// u64 time_running;
// u64 id;
// }
//
m_pCountData[cnt].m_sampleId = m_pCountData[cnt].m_values[3];
// Now push this into m_eventDataList
// TODO: Eventually we will discard pCountData stuff
OS_OUTPUT_FORMAT_DEBUG_LOG(OS_DEBUG_LOG_EXTENSIVE, L"push elemnt in m_eventDataList cnt(%d)\n", cnt);
m_eventDataList.push_back(m_pCountData[i]);
}
} // threads
} // cpus
} // per process
OS_OUTPUT_FORMAT_DEBUG_LOG(OS_DEBUG_LOG_DEBUG, L"size of m_eventDataList (%d)\n", m_eventDataList.size());
return S_OK;
}
static int bpf_perf_event_open(int map_fd, int key, int cpu)
{
struct perf_event_attr attr = {
.sample_type = PERF_SAMPLE_RAW | PERF_SAMPLE_TIME,
.type = PERF_TYPE_SOFTWARE,
.config = PERF_COUNT_SW_BPF_OUTPUT,
};
int pmu_fd;
pmu_fd = sys_perf_event_open(&attr, -1, cpu, -1, 0);
if (pmu_fd < 0) {
p_err("failed to open perf event %d for CPU %d", key, cpu);
return -1;
}
if (bpf_map_update_elem(map_fd, &key, &pmu_fd, BPF_ANY)) {
p_err("failed to update map for event %d for CPU %d", key, cpu);
goto err_close;
}
if (ioctl(pmu_fd, PERF_EVENT_IOC_ENABLE, 0)) {
p_err("failed to enable event %d for CPU %d", key, cpu);
goto err_close;
}
return pmu_fd;
err_close:
close(pmu_fd);
return -1;
}
int do_event_pipe(int argc, char **argv)
{
int i, nfds, map_fd, index = -1, cpu = -1;
struct bpf_map_info map_info = {};
struct event_ring_info *rings;
size_t tmp_buf_sz = 0;
void *tmp_buf = NULL;
struct pollfd *pfds;
__u32 map_info_len;
bool do_all = true;
map_info_len = sizeof(map_info);
map_fd = map_parse_fd_and_info(&argc, &argv, &map_info, &map_info_len);
if (map_fd < 0)
return -1;
if (map_info.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
p_err("map is not a perf event array");
goto err_close_map;
}
while (argc) {
if (argc < 2)
BAD_ARG();
if (is_prefix(*argv, "cpu")) {
char *endptr;
NEXT_ARG();
cpu = strtoul(*argv, &endptr, 0);
if (*endptr) {
p_err("can't parse %s as CPU ID", **argv);
goto err_close_map;
}
NEXT_ARG();
} else if (is_prefix(*argv, "index")) {
char *endptr;
NEXT_ARG();
index = strtoul(*argv, &endptr, 0);
if (*endptr) {
p_err("can't parse %s as index", **argv);
goto err_close_map;
}
NEXT_ARG();
} else {
BAD_ARG();
}
do_all = false;
}
if (!do_all) {
if (index == -1 || cpu == -1) {
p_err("cpu and index must be specified together");
goto err_close_map;
}
nfds = 1;
} else {
nfds = min(get_possible_cpus(), map_info.max_entries);
cpu = 0;
index = 0;
}
rings = calloc(nfds, sizeof(rings[0]));
if (!rings)
goto err_close_map;
pfds = calloc(nfds, sizeof(pfds[0]));
if (!pfds)
goto err_free_rings;
for (i = 0; i < nfds; i++) {
rings[i].cpu = cpu + i;
rings[i].key = index + i;
rings[i].fd = bpf_perf_event_open(map_fd, rings[i].key,
rings[i].cpu);
if (rings[i].fd < 0)
goto err_close_fds_prev;
rings[i].mem = perf_event_mmap(rings[i].fd);
if (!rings[i].mem)
goto err_close_fds_current;
pfds[i].fd = rings[i].fd;
pfds[i].events = POLLIN;
}
signal(SIGINT, int_exit);
signal(SIGHUP, int_exit);
signal(SIGTERM, int_exit);
if (json_output)
jsonw_start_array(json_wtr);
while (!stop) {
poll(pfds, nfds, 200);
for (i = 0; i < nfds; i++)
perf_event_read(&rings[i], &tmp_buf, &tmp_buf_sz);
}
free(tmp_buf);
if (json_output)
jsonw_end_array(json_wtr);
for (i = 0; i < nfds; i++) {
perf_event_unmap(rings[i].mem);
close(rings[i].fd);
}
free(pfds);
free(rings);
close(map_fd);
return 0;
err_close_fds_prev:
while (i--) {
perf_event_unmap(rings[i].mem);
err_close_fds_current:
close(rings[i].fd);
}
free(pfds);
err_free_rings:
free(rings);
err_close_map:
close(map_fd);
return -1;
}
static void create_counter(int counter, int cpu, pid_t pid)
{
char *filter = filters[counter];
struct perf_event_attr *attr = attrs + counter;
struct perf_header_attr *h_attr;
int track = !counter; /* only the first counter needs these */
int ret;
struct {
u64 count;
u64 time_enabled;
u64 time_running;
u64 id;
} read_data;
attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
PERF_FORMAT_TOTAL_TIME_RUNNING |
PERF_FORMAT_ID;
attr->sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID;
if (freq) {
attr->sample_type |= PERF_SAMPLE_PERIOD;
attr->freq = 1;
attr->sample_freq = freq;
}
if (no_samples)
attr->sample_freq = 0;
if (inherit_stat)
attr->inherit_stat = 1;
if (sample_address)
attr->sample_type |= PERF_SAMPLE_ADDR;
if (call_graph)
attr->sample_type |= PERF_SAMPLE_CALLCHAIN;
if (raw_samples) {
attr->sample_type |= PERF_SAMPLE_TIME;
attr->sample_type |= PERF_SAMPLE_RAW;
attr->sample_type |= PERF_SAMPLE_CPU;
}
attr->mmap = track;
attr->comm = track;
attr->inherit = (cpu < 0) && inherit;
attr->disabled = 1;
try_again:
fd[nr_cpu][counter] = sys_perf_event_open(attr, pid, cpu, group_fd, 0);
if (fd[nr_cpu][counter] < 0) {
int err = errno;
if (err == EPERM || err == EACCES)
die("Permission error - are you root?\n");
else if (err == ENODEV && profile_cpu != -1)
die("No such device - did you specify an out-of-range profile CPU?\n");
/*
* If it's cycles then fall back to hrtimer
* based cpu-clock-tick sw counter, which
* is always available even if no PMU support:
*/
if (attr->type == PERF_TYPE_HARDWARE
&& attr->config == PERF_COUNT_HW_CPU_CYCLES) {
if (verbose)
warning(" ... trying to fall back to cpu-clock-ticks\n");
attr->type = PERF_TYPE_SOFTWARE;
attr->config = PERF_COUNT_SW_CPU_CLOCK;
goto try_again;
}
printf("\n");
error("perfcounter syscall returned with %d (%s)\n",
fd[nr_cpu][counter], strerror(err));
#if defined(__i386__) || defined(__x86_64__)
if (attr->type == PERF_TYPE_HARDWARE && err == EOPNOTSUPP)
die("No hardware sampling interrupt available. No APIC? If so then you can boot the kernel with the \"lapic\" boot parameter to force-enable it.\n");
#endif
die("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
exit(-1);
}
h_attr = get_header_attr(attr, counter);
if (h_attr == NULL)
die("nomem\n");
if (!file_new) {
if (memcmp(&h_attr->attr, attr, sizeof(*attr))) {
fprintf(stderr, "incompatible append\n");
exit(-1);
}
}
if (read(fd[nr_cpu][counter], &read_data, sizeof(read_data)) == -1) {
perror("Unable to read perf file descriptor\n");
exit(-1);
}
if (perf_header_attr__add_id(h_attr, read_data.id) < 0) {
pr_warning("Not enough memory to add id\n");
exit(-1);
}
assert(fd[nr_cpu][counter] >= 0);
fcntl(fd[nr_cpu][counter], F_SETFL, O_NONBLOCK);
/*
* First counter acts as the group leader:
*/
if (group && group_fd == -1)
group_fd = fd[nr_cpu][counter];
if (multiplex && multiplex_fd == -1)
multiplex_fd = fd[nr_cpu][counter];
if (multiplex && fd[nr_cpu][counter] != multiplex_fd) {
ret = ioctl(fd[nr_cpu][counter], PERF_EVENT_IOC_SET_OUTPUT, multiplex_fd);
assert(ret != -1);
} else {
event_array[nr_poll].fd = fd[nr_cpu][counter];
event_array[nr_poll].events = POLLIN;
nr_poll++;
mmap_array[nr_cpu][counter].counter = counter;
mmap_array[nr_cpu][counter].prev = 0;
mmap_array[nr_cpu][counter].mask = mmap_pages*page_size - 1;
mmap_array[nr_cpu][counter].base = mmap(NULL, (mmap_pages+1)*page_size,
PROT_READ|PROT_WRITE, MAP_SHARED, fd[nr_cpu][counter], 0);
if (mmap_array[nr_cpu][counter].base == MAP_FAILED) {
error("failed to mmap with %d (%s)\n", errno, strerror(errno));
exit(-1);
}
}
if (filter != NULL) {
ret = ioctl(fd[nr_cpu][counter],
PERF_EVENT_IOC_SET_FILTER, filter);
if (ret) {
error("failed to set filter with %d (%s)\n", errno,
strerror(errno));
exit(-1);
}
}
ioctl(fd[nr_cpu][counter], PERF_EVENT_IOC_ENABLE);
}
int main(int argc, char *argv[])
{
QCoreApplication app(argc, argv);
QElapsedTimer timer;
perf_event_attr attr_inst, attr_cpu, attr_miss;
memset(&attr_inst, 0, sizeof(attr_inst));
attr_inst.size = sizeof(attr_inst);
attr_inst.type = PERF_TYPE_HARDWARE;
attr_inst.config = PERF_COUNT_HW_INSTRUCTIONS;
memset(&attr_cpu, 0, sizeof(attr_cpu));
attr_cpu.size = sizeof(attr_cpu);
attr_cpu.type = PERF_TYPE_SOFTWARE;
attr_cpu.config = PERF_COUNT_SW_TASK_CLOCK;
memset(&attr_miss, 0, sizeof(attr_miss));
attr_miss.size = sizeof(attr_miss);
attr_miss.type = PERF_TYPE_HW_CACHE;
attr_miss.config = (PERF_COUNT_HW_CACHE_L1D) | (PERF_COUNT_HW_CACHE_OP_READ << 8) | (PERF_COUNT_HW_CACHE_RESULT_MISS << 16);
pid_t tid = gettid();
int fd_inst = sys_perf_event_open(&attr_inst, tid, -1, -1, 0);
int fd_cpu = sys_perf_event_open(&attr_cpu, tid, -1, -1, 0);
int fd_miss = sys_perf_event_open(&attr_cpu, tid, -1, -1, 0);
int n = 50;
QVector<Sample> samples(n);
int sz = 100000;
QVector<int> idx_rnd(sz); // 100 000 * sizeof(int) = 400 kio, L1 32kio, LLC 4mb
QVector<int> idx_lin(sz);
QVector<int> buf_large(sz);
QVector<int> buf_small(10);
for (int i = 0; i < sz; i++) {
idx_rnd[i] = i;
idx_lin[i] = i;
}
std::random_shuffle(idx_rnd.begin(), idx_rnd.end());
for (int i = 0; i < n; i++) {
//qDebug() << "before";
timer.restart();
qint64 ret = 0;
qint64 val_inst0, val_inst1;
qint64 val_cpu0, val_cpu1;
qint64 val_miss0, val_miss1;
ret |= read(fd_inst, &val_inst0, sizeof(val_inst0));
ret |= read(fd_cpu, &val_cpu0, sizeof(val_cpu0));
ret = read(fd_miss, &val_miss0, sizeof(val_miss0));
assert(ret > 0);
bool slow = (i % 6) == 0;
if (slow) {
// Tests::read_data(buf_large, idx_rnd);
Tests::factorial(100);
} else {
// Tests::read_data(buf_small, idx_lin);
Tests::factorial(10);
}
// do_compute(work[i % work.size()] * scale);
ret |= read(fd_inst, &val_inst1, sizeof(val_inst1));
ret |= read(fd_cpu, &val_cpu1, sizeof(val_cpu1));
ret |= read(fd_miss, &val_miss1, sizeof(val_miss1));
qint64 delta = timer.nsecsElapsed() / 1000;
//qDebug() << "after" << delta;
samples[i].slow = slow;
samples[i].delta = delta;
samples[i].inst = (val_inst1 - val_inst0) / 1000;
samples[i].cpu = (val_cpu1 - val_cpu0) / 1000;
samples[i].miss = (val_miss1 - val_miss0);
}
//run through all of them:
for (int i = 0; i < n; i++) {
Sample s = samples[i];
qDebug() << s;
}
//write the csv:
QFile file("sample.csv");
if (file.open(QIODevice::WriteOnly | QIODevice::Text)) {
QTextStream out(&file);
for (const Sample &s: samples) {
out << s.slow << "," << s.delta << "," << s.inst << "," << s.cpu << "," << s.miss << "\n";
}
file.close();
}
return 0;
}
/*
* Create an event group that contains both a sampled hardware
* (cpu-cycles) and software (intel_cqm/llc_occupancy/) event. We then
* wait for the hardware perf counter to overflow and generate a PMI,
* which triggers an event read for both of the events in the group.
*
* Since reading Intel CQM event counters requires sending SMP IPIs, the
* CQM pmu needs to handle the above situation gracefully, and return
* the last read counter value to avoid triggering a WARN_ON_ONCE() in
* smp_call_function_many() caused by sending IPIs from NMI context.
*/
int test__intel_cqm_count_nmi_context(int subtest __maybe_unused)
{
struct perf_evlist *evlist = NULL;
struct perf_evsel *evsel = NULL;
struct perf_event_attr pe;
int i, fd[2], flag, ret;
size_t mmap_len;
void *event;
pid_t pid;
int err = TEST_FAIL;
flag = perf_event_open_cloexec_flag();
evlist = perf_evlist__new();
if (!evlist) {
pr_debug("perf_evlist__new failed\n");
return TEST_FAIL;
}
ret = parse_events(evlist, "intel_cqm/llc_occupancy/", NULL);
if (ret) {
pr_debug("parse_events failed, is \"intel_cqm/llc_occupancy/\" available?\n");
err = TEST_SKIP;
goto out;
}
evsel = perf_evlist__first(evlist);
if (!evsel) {
pr_debug("perf_evlist__first failed\n");
goto out;
}
memset(&pe, 0, sizeof(pe));
pe.size = sizeof(pe);
pe.type = PERF_TYPE_HARDWARE;
pe.config = PERF_COUNT_HW_CPU_CYCLES;
pe.read_format = PERF_FORMAT_GROUP;
pe.sample_period = 128;
pe.sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_READ;
pid = spawn();
fd[0] = sys_perf_event_open(&pe, pid, -1, -1, flag);
if (fd[0] < 0) {
pr_debug("failed to open event\n");
goto out;
}
memset(&pe, 0, sizeof(pe));
pe.size = sizeof(pe);
pe.type = evsel->attr.type;
pe.config = evsel->attr.config;
fd[1] = sys_perf_event_open(&pe, pid, -1, fd[0], flag);
if (fd[1] < 0) {
pr_debug("failed to open event\n");
goto out;
}
/*
* Pick a power-of-two number of pages + 1 for the meta-data
* page (struct perf_event_mmap_page). See tools/perf/design.txt.
*/
mmap_len = page_size * 65;
event = mmap(NULL, mmap_len, PROT_READ, MAP_SHARED, fd[0], 0);
if (event == (void *)(-1)) {
pr_debug("failed to mmap %d\n", errno);
goto out;
}
sleep(1);
err = TEST_OK;
munmap(event, mmap_len);
for (i = 0; i < 2; i++)
close(fd[i]);
kill(pid, SIGKILL);
wait(NULL);
out:
perf_evlist__delete(evlist);
return err;
}
void perf_event::create_perf_event(char *eventname, int _cpu)
{
struct perf_event_attr attr;
int ret;
int err;
struct {
__u64 count;
__u64 time_enabled;
__u64 time_running;
__u64 id;
} read_data;
if (perf_fd != -1)
clear();
memset(&attr, 0, sizeof(attr));
attr.read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
PERF_FORMAT_TOTAL_TIME_RUNNING |
PERF_FORMAT_ID;
attr.sample_freq = 0;
attr.sample_period = 1;
attr.sample_type |= PERF_SAMPLE_RAW | PERF_SAMPLE_CPU | PERF_SAMPLE_TIME;
attr.mmap = 1;
attr.comm = 1;
attr.inherit = 0;
attr.disabled = 1;
attr.type = PERF_TYPE_TRACEPOINT;
attr.config = trace_type;
if (attr.config <= 0)
return;
perf_fd = sys_perf_event_open(&attr, -1, _cpu, -1, 0);
if (perf_fd < 0) {
err = errno;
reset_display();
if (err == EMFILE)
fprintf(stderr, _("Too many open files, please increase the limit of open file descriptors.\n"));
else {
fprintf(stderr, _("PowerTOP %s needs the kernel to support the 'perf' subsystem\n"), POWERTOP_VERSION);
fprintf(stderr, _("as well as support for trace points in the kernel:\n"));
fprintf(stderr, "CONFIG_PERF_EVENTS=y\nCONFIG_PERF_COUNTERS=y\nCONFIG_TRACEPOINTS=y\nCONFIG_TRACING=y\n");
}
exit(EXIT_FAILURE);
}
if (read(perf_fd, &read_data, sizeof(read_data)) == -1) {
reset_display();
perror("Unable to read perf file descriptor\n");
exit(-1);
}
fcntl(perf_fd, F_SETFL, O_NONBLOCK);
perf_mmap = mmap(NULL, (bufsize+1)*getpagesize(),
PROT_READ | PROT_WRITE, MAP_SHARED, perf_fd, 0);
if (perf_mmap == MAP_FAILED) {
fprintf(stderr, "failed to mmap with %d (%s)\n", errno, strerror(errno));
return;
}
ret = ioctl(perf_fd, PERF_EVENT_IOC_ENABLE, 0);
if (ret < 0) {
fprintf(stderr, "failed to enable perf \n");
}
pc = (perf_event_mmap_page *)perf_mmap;
data_mmap = (unsigned char *)perf_mmap + getpagesize();
}
