/*
* set_notify.c - example of how to get notification at the end of a set chain
*
* Copyright (c) 2005-2006 Hewlett-Packard Development Company, L.P.
* Contributed by Stephane Eranian <eranian@hpl.hp.com>
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
* of the Software, and to permit persons to whom the Software is furnished to do so,
* subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A
* PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
* HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
* CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE
* OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*
* This file is part of libpfm, a performance monitoring support library for
* applications on Linux.
*/
#include <sys/types.h>
#include <inttypes.h>
#include <stdio.h>
#include <stdlib.h>
#include <stdarg.h>
#include <errno.h>
#include <unistd.h>
#include <string.h>
#include <signal.h>
#include <fcntl.h>
#include <perfmon/perfmon.h>
#include <perfmon/pfmlib.h>
#include "detect_pmcs.h"
#define NUM_SETS 3
#define THE_TIMEOUT 1
static volatile unsigned long notification_received;
#define NUM_PMCS PFMLIB_MAX_PMCS
#define NUM_PMDS PFMLIB_MAX_PMDS
static int ctx_fd;
static char *event1_name;
static pfarg_setinfo_t setinfo[NUM_SETS];
static pfarg_pmd_t pd[2];
static void fatal_error(char *fmt,...) __attribute__((noreturn));
static void
fatal_error(char *fmt, ...)
{
va_list ap;
va_start(ap, fmt);
vfprintf(stderr, fmt, ap);
va_end(ap);
exit(1);
}
static void
warning(char *fmt, ...)
{
va_list ap;
va_start(ap, fmt);
vfprintf(stderr, fmt, ap);
va_end(ap);
}
static void
sigio_handler(int n, struct siginfo *info, void *data)
{
pfarg_msg_t msg;
unsigned int k;
int r;
retry:
r = read(ctx_fd, &msg, sizeof(msg));
if (r != sizeof(msg)) {
if(r == -1 && errno == EINTR) {
warning("read interrupted, retrying\n");
goto retry;
}
fatal_error("cannot read overflow message: %s\n", strerror(errno));
}
if (msg.type != PFM_MSG_OVFL) {
fatal_error("unexpected msg type: %d\n",msg.type);
}
if (pfm_getinfo_evtsets(ctx_fd, setinfo, NUM_SETS) == -1) {
fatal_error("pfm_getinfo_evtsets: %s", strerror(errno));
}
if (pfm_read_pmds(ctx_fd, pd, 2) == -1) {
fatal_error("pfm_read_pmds: %s", strerror(errno));
}
/*
* XXX: risky to do printf() in signal handler!
*/
printf("Notification %lu: set%u pd[0]=%"PRIx64" pd[1]=%"PRIx64"\n",
notification_received,
pd[0].reg_set,
pd[0].reg_value,
pd[1].reg_value);
for(k=0; k < NUM_SETS; k++)
printf("set%u %"PRIu64" runs\n", setinfo[k].set_id, setinfo[k].set_runs);
/*
* At this point, the counter used for the sampling period has already
* be reset by the kernel because we are in non-blocking mode, self-monitoring.
*/
/*
* increment our notification counter
*/
notification_received++;
/*
* And resume monitoring
*/
if (pfm_restart(ctx_fd) == -1) {
fatal_error("pfm_restart: %s", strerror(errno));
}
}
/*
* infinite loop waiting for notification to get out
*/
void
busyloop(void)
{
/*
* busy loop to burn CPU cycles
*/
for(;notification_received < 3;) ;
}
#ifdef __ia64__
#define FUDGE 1
#else
#define FUDGE 0x100
#endif
/*
* build end marker set
*/
void
setup_end_marker(int fd, unsigned int set_id, uint64_t num_ovfls, int plm_mask)
{
pfarg_setdesc_t setdesc;
pfarg_pmc_t pc[8];
pfmlib_input_param_t inp;
pfmlib_output_param_t outp;
unsigned int i;
int ret;
memset(&setdesc, 0, sizeof(setdesc));
memset(pc, 0, sizeof(pc));
memset(pd, 0, sizeof(pd));
memset(&inp,0, sizeof(inp));
memset(&outp,0, sizeof(outp));
/*
* we use the cycle event twice:
* - first as sampling period to force switch to set 0
* - second as sampling period to force notification
*/
if (pfm_get_cycle_event(&inp.pfp_events[0]) != PFMLIB_SUCCESS)
fatal_error("cannot find cycle event\n");
inp.pfp_events[1] = inp.pfp_events[0];
inp.pfp_dfl_plm = plm_mask;
inp.pfp_event_count = 2;
/*
* build the pfp_unavail_pmcs bitmask by looking
* at what perfmon has available. It is not always
* the case that all PMU registers are actually available
* to applications. For instance, on IA-32 platforms, some
* registers may be reserved for the NMI watchdog timer.
*
* With this bitmap, the library knows which registers NOT to
* use. Of source, it is possible that no valid assignement may
* be possible if certina PMU registers are not available.
*/
detect_unavail_pmcs(fd, &inp.pfp_unavail_pmcs);
if ((ret=pfm_dispatch_events(&inp, NULL, &outp, NULL)) != PFMLIB_SUCCESS)
fatal_error("Cannot configure events: %s\n", pfm_strerror(ret));
for (i=0; i < outp.pfp_pmc_count; i++) {
pc[i].reg_num = outp.pfp_pmcs[i].reg_num;
pc[i].reg_value = outp.pfp_pmcs[i].reg_value;
pc[i].reg_set = set_id;
}
for (i=0; i < outp.pfp_pmd_count; i++) {
pd[i].reg_num = outp.pfp_pmds[i].reg_num;
pd[i].reg_set = set_id;
}
/*
* first cycle overflow: no notification, simply trigger a switch
*/
pd[0].reg_flags = 0;
pd[0].reg_value = -1;
pd[0].reg_long_reset = -1;
pd[0].reg_short_reset = -1;
pd[0].reg_ovfl_switch_cnt = 1;
/*
* second cycle overflow: generate notification, switch on restart
*/
pd[1].reg_flags = PFM_REGFL_OVFL_NOTIFY;
pd[1].reg_value = -num_ovfls*FUDGE;
pd[1].reg_long_reset = -num_ovfls*FUDGE;
pd[1].reg_short_reset = -num_ovfls*FUDGE;
pd[1].reg_ovfl_switch_cnt = 1;
/*
* set uses overflow switch
*/
setdesc.set_id = set_id;
setdesc.set_flags = PFM_SETFL_OVFL_SWITCH;
setdesc.set_timeout = 0;
if (pfm_create_evtsets(fd, &setdesc, 1) == -1)
fatal_error("pfm_create_evtsets error errno %d\n",errno);
if (pfm_write_pmcs(fd, pc, outp.pfp_pmc_count) == -1)
fatal_error("pfm_write_pmcs error errno %d\n",errno);
/*
* To be read, each PMD must be either written or declared
* as being part of a sample (reg_smpl_pmds)
*/
if (pfm_write_pmds(fd, pd, outp.pfp_pmd_count) == -1)
fatal_error("pfm_write_pmds error errno %d\n",errno);
}
int
main(int argc, char **argv)
{
pfarg_ctx_t ctx[1];
pfmlib_input_param_t inp;
pfmlib_output_param_t outp;
pfarg_pmc_t pc[NUM_PMCS];
pfarg_pmd_t pd[NUM_PMDS];
pfarg_load_t load_args;
pfarg_setdesc_t setdesc;
pfmlib_options_t pfmlib_options;
struct sigaction act;
uint64_t num_ovfls;
unsigned int i, k;
size_t len;
int ret;
/*
* pass options to library (optional)
*/
memset(&pfmlib_options, 0, sizeof(pfmlib_options));
pfmlib_options.pfm_debug = 0; /* set to 1 for debug */
pfmlib_options.pfm_verbose = 1; /* set to 1 for verbose */
pfm_set_options(&pfmlib_options);
/*
* Initialize pfm library (required before we can use it)
*/
ret = pfm_initialize();
if (ret != PFMLIB_SUCCESS)
fatal_error("Cannot initialize library: %s\n", pfm_strerror(ret));
num_ovfls = argc > 1 ? strtoull(argv[1], NULL, 10) : 3;
printf("chain contains %d sets, time switching every %u seconds\n"
"notification every %"PRIu64" times the end of the chain is reached\n",
NUM_SETS,
THE_TIMEOUT,
num_ovfls);
/*
* Install the signal handler (SIGIO)
*
* SA_SIGINFO required on some platforms
* to get siginfo passed to handler.
*/
memset(&act, 0, sizeof(act));
act.sa_handler = (sig_t)sigio_handler;
act.sa_flags = SA_SIGINFO;
sigaction (SIGIO, &act, 0);
memset(pc, 0, sizeof(pc));
memset(ctx, 0, sizeof(ctx));
memset(&load_args, 0, sizeof(load_args));
memset(&inp,0, sizeof(inp));
memset(&outp,0, sizeof(outp));
memset(&setdesc,0, sizeof(setdesc));
if (pfm_get_cycle_event(&inp.pfp_events[0]) != PFMLIB_SUCCESS)
fatal_error("cannot find cycle event\n");
pfm_get_max_event_name_len(&len);
event1_name = malloc(len+1);
if (event1_name == NULL) {
fatal_error("cannot allocate event name\n");
}
pfm_get_full_event_name(&inp.pfp_events[1], event1_name, len+1);
/*
* set the default privilege mode for all counters:
* PFM_PLM3 : user level only
*/
inp.pfp_dfl_plm = PFM_PLM3;
/*
* how many counters we use
*/
inp.pfp_event_count = 1;
/*
* now create the context for self monitoring/per-task
*/
ctx_fd = pfm_create_context(ctx, NULL, NULL, 0);
if (ctx_fd == -1) {
if (errno == ENOSYS)
fatal_error("Your kernel does not have performance monitoring support!\n");
fatal_error("Can't create PFM context %s\n", strerror(errno));
}
/*
* build the pfp_unavail_pmcs bitmask by looking
* at what perfmon has available. It is not always
* the case that all PMU registers are actually available
* to applications. For instance, on IA-32 platforms, some
* registers may be reserved for the NMI watchdog timer.
*
* With this bitmap, the library knows which registers NOT to
* use. Of source, it is possible that no valid assignement may
* be possible if certina PMU registers are not available.
*/
detect_unavail_pmcs(ctx_fd, &inp.pfp_unavail_pmcs);
/*
* let the library figure out the values for the PMCS
*/
if ((ret=pfm_dispatch_events(&inp, NULL, &outp, NULL)) != PFMLIB_SUCCESS)
fatal_error("Cannot configure events: %s\n", pfm_strerror(ret));
/*
* Now prepare the argument to initialize the PMDs and PMCS.
* We use pfp_pmc_count to determine the number of registers to
* setup. Note that this field can be >= pfp_event_count.
*/
for (i=0; i < outp.pfp_pmc_count; i++) {
pc[i].reg_num = outp.pfp_pmcs[i].reg_num;
pc[i].reg_value = outp.pfp_pmcs[i].reg_value;
}
for (i=0; i < outp.pfp_pmd_count; i++)
pd[i].reg_num = outp.pfp_pmds[i].reg_num;
pd[0].reg_value = 0;
pd[0].reg_long_reset = 0;
pd[0].reg_short_reset = 0;
pd[0].reg_ovfl_switch_cnt = 0;
for(k=0; k < NUM_SETS; k++) {
setdesc.set_id = setinfo[k].set_id = k;
setdesc.set_flags = PFM_SETFL_TIME_SWITCH;
setdesc.set_timeout = THE_TIMEOUT * 1000000000; /* in nsecs */
for (i=0; i < outp.pfp_pmc_count; i++)
pc[i].reg_set = k;
for (i=0; i < outp.pfp_pmd_count; i++)
pd[i].reg_set = k;
if (pfm_create_evtsets(ctx_fd, &setdesc, 1) == -1)
fatal_error("pfm_create_evtsets error errno %d\n",errno);
if (pfm_write_pmcs(ctx_fd, pc, outp.pfp_pmc_count) == -1)
fatal_error("pfm_write_pmcs error errno %d\n",errno);
/*
* To be read, each PMD must be either written or declared
* as being part of a sample (reg_smpl_pmds)
*/
if (pfm_write_pmds(ctx_fd, pd, outp.pfp_pmd_count) == -1)
fatal_error("pfm_write_pmds error errno %d\n",errno);
}
setup_end_marker(ctx_fd, k, num_ovfls, inp.pfp_dfl_plm);
/*
* we want to monitor ourself
*/
load_args.load_pid = getpid();
load_args.load_set = 0;
if (pfm_load_context(ctx_fd, &load_args) == -1) {
fatal_error("pfm_load_context error errno %d\n",errno);
}
/*
* setup asynchronous notification on the file descriptor
*/
ret = fcntl(ctx_fd, F_SETFL, fcntl(ctx_fd, F_GETFL, 0) | O_ASYNC);
if (ret == -1) {
fatal_error("cannot set ASYNC: %s\n", strerror(errno));
}
/*
* get ownership of the descriptor
*/
ret = fcntl(ctx_fd, F_SETOWN, getpid());
if (ret == -1) {
fatal_error("cannot setown: %s\n", strerror(errno));
}
/*
* Let's roll now
*/
pfm_start(ctx_fd, NULL);
busyloop();
pfm_stop(ctx_fd);
close(ctx_fd);
free(event1_name);
return 0;
}