Quelle  mmap-basic.c   Sprache: unbekannt

// SPDX-License-Identifier: GPL-2.0
#include <errno.h>
#include <fcntl.h>
#include <inttypes.h>
#include <stdlib.h>
#include <perf/cpumap.h>

#include "cpumap.h"
#include "debug.h"
#include "event.h"
#include "evlist.h"
#include "evsel.h"
#include "thread_map.h"
#include "tests.h"
#include "util/affinity.h"
#include "util/mmap.h"
#include "util/sample.h"
#include <linux/err.h>
#include <linux/kernel.h>
#include <linux/string.h>
#include <perf/evlist.h>
#include <perf/mmap.h>

/*
 * This test will generate random numbers of calls to some getpid syscalls,
 * then establish an mmap for a group of events that are created to monitor
 * the syscalls.
 *
 * It will receive the events, using mmap, use its PERF_SAMPLE_ID generated
 * sample.id field to map back to its respective perf_evsel instance.
 *
 * Then it checks if the number of syscalls reported as perf events by
 * the kernel corresponds to the number of syscalls made.
 */
static int test__basic_mmap(struct test_suite *test __maybe_unused, int subtest __maybe_unused)
{
 int err = TEST_FAIL;
 union perf_event *event;
 struct perf_thread_map *threads;
 struct perf_cpu_map *cpus;
 struct evlist *evlist;
 cpu_set_t cpu_set;
 const char *syscall_names[] = { "getsid", "getppid", "getpgid", };
 pid_t (*syscalls[])(void) = { (void *)getsid, getppid, (void*)getpgid };
#define nsyscalls ARRAY_SIZE(syscall_names)
 unsigned int nr_events[nsyscalls],
       expected_nr_events[nsyscalls], i, j;
 struct evsel *evsels[nsyscalls], *evsel;
 char sbuf[STRERR_BUFSIZE];
 struct mmap *md;

 threads = thread_map__new_by_tid(getpid());
 if (threads == NULL) {
  pr_debug("thread_map__new\n");
  return -1;
 }

 cpus = perf_cpu_map__new_online_cpus();
 if (cpus == NULL) {
  pr_debug("perf_cpu_map__new\n");
  goto out_free_threads;
 }

 CPU_ZERO(&cpu_set);
 CPU_SET(perf_cpu_map__cpu(cpus, 0).cpu, &cpu_set);
 sched_setaffinity(0, sizeof(cpu_set), &cpu_set);
 if (sched_setaffinity(0, sizeof(cpu_set), &cpu_set) < 0) {
  pr_debug("sched_setaffinity() failed on CPU %d: %s ",
    perf_cpu_map__cpu(cpus, 0).cpu,
    str_error_r(errno, sbuf, sizeof(sbuf)));
  goto out_free_cpus;
 }

 evlist = evlist__new();
 if (evlist == NULL) {
  pr_debug("evlist__new\n");
  goto out_free_cpus;
 }

 perf_evlist__set_maps(&evlist->core, cpus, threads);

 for (i = 0; i < nsyscalls; ++i) {
  char name[64];

  snprintf(name, sizeof(name), "sys_enter_%s", syscall_names[i]);
  evsels[i] = evsel__newtp("syscalls", name);
  if (IS_ERR(evsels[i])) {
   pr_debug("evsel__new(%s)\n", name);
   if (PTR_ERR(evsels[i]) == -EACCES) {
    /* Permissions failure, flag the failure as a skip. */
    err = TEST_SKIP;
   }
   goto out_delete_evlist;
  }

  evsels[i]->core.attr.wakeup_events = 1;
  evsel__set_sample_id(evsels[i], false);

  evlist__add(evlist, evsels[i]);

  if (evsel__open(evsels[i], cpus, threads) < 0) {
   pr_debug("failed to open counter: %s, "
     "tweak /proc/sys/kernel/perf_event_paranoid?\n",
     str_error_r(errno, sbuf, sizeof(sbuf)));
   goto out_delete_evlist;
  }

  nr_events[i] = 0;
  expected_nr_events[i] = 1 + rand() % 127;
 }

 if (evlist__mmap(evlist, 128) < 0) {
  pr_debug("failed to mmap events: %d (%s)\n", errno,
    str_error_r(errno, sbuf, sizeof(sbuf)));
  goto out_delete_evlist;
 }

 for (i = 0; i < nsyscalls; ++i)
  for (j = 0; j < expected_nr_events[i]; ++j) {
   syscalls[i]();
  }

 md = &evlist->mmap[0];
 if (perf_mmap__read_init(&md->core) < 0)
  goto out_init;

 while ((event = perf_mmap__read_event(&md->core)) != NULL) {
  struct perf_sample sample;

  if (event->header.type != PERF_RECORD_SAMPLE) {
   pr_debug("unexpected %s event\n",
     perf_event__name(event->header.type));
   goto out_delete_evlist;
  }

  perf_sample__init(&sample, /*all=*/false);
  err = evlist__parse_sample(evlist, event, &sample);
  if (err) {
   pr_err("Can't parse sample, err = %d\n", err);
   perf_sample__exit(&sample);
   goto out_delete_evlist;
  }

  err = -1;
  evsel = evlist__id2evsel(evlist, sample.id);
  perf_sample__exit(&sample);
  if (evsel == NULL) {
   pr_debug("event with id %" PRIu64
     " doesn't map to an evsel\n", sample.id);
   goto out_delete_evlist;
  }
  nr_events[evsel->core.idx]++;
  perf_mmap__consume(&md->core);
 }
 perf_mmap__read_done(&md->core);

out_init:
 err = 0;
 evlist__for_each_entry(evlist, evsel) {
  if (nr_events[evsel->core.idx] != expected_nr_events[evsel->core.idx]) {
   pr_debug("expected %d %s events, got %d\n",
     expected_nr_events[evsel->core.idx],
     evsel__name(evsel), nr_events[evsel->core.idx]);
   err = -1;
   goto out_delete_evlist;
  }
 }

out_delete_evlist:
 evlist__delete(evlist);
out_free_cpus:
 perf_cpu_map__put(cpus);
out_free_threads:
 perf_thread_map__put(threads);
 return err;
}

enum user_read_state {
 USER_READ_ENABLED,
 USER_READ_DISABLED,
 USER_READ_UNKNOWN,
};

static enum user_read_state set_user_read(struct perf_pmu *pmu, enum user_read_state enabled)
{
 char buf[2] = {0, '\n'};
 ssize_t len;
 int events_fd, rdpmc_fd;
 enum user_read_state old_user_read = USER_READ_UNKNOWN;

 if (enabled == USER_READ_UNKNOWN)
  return USER_READ_UNKNOWN;

 events_fd = perf_pmu__event_source_devices_fd();
 if (events_fd < 0)
  return USER_READ_UNKNOWN;

 rdpmc_fd = perf_pmu__pathname_fd(events_fd, pmu->name, "rdpmc", O_RDWR);
 if (rdpmc_fd < 0) {
  close(events_fd);
  return USER_READ_UNKNOWN;
 }

 len = read(rdpmc_fd, buf, sizeof(buf));
 if (len != sizeof(buf))
  pr_debug("%s read failed\n", __func__);

 // Note, on Intel hybrid disabling on 1 PMU will implicitly disable on
 // all the core PMUs.
 old_user_read = (buf[0] == '1') ? USER_READ_ENABLED : USER_READ_DISABLED;

 if (enabled != old_user_read) {
  buf[0] = (enabled == USER_READ_ENABLED) ? '1' : '0';
  len = write(rdpmc_fd, buf, sizeof(buf));
  if (len != sizeof(buf))
   pr_debug("%s write failed\n", __func__);
 }
 close(rdpmc_fd);
 close(events_fd);
 return old_user_read;
}

static int test_stat_user_read(u64 event, enum user_read_state enabled)
{
 struct perf_pmu *pmu = NULL;
 struct perf_thread_map *threads = perf_thread_map__new_dummy();
 int ret = TEST_OK;

 pr_err("User space counter reading %" PRIu64 "\n", event);
 if (!threads) {
  pr_err("User space counter reading [Failed to create threads]\n");
  return TEST_FAIL;
 }
 perf_thread_map__set_pid(threads, 0, 0);

 while ((pmu = perf_pmus__scan_core(pmu)) != NULL) {
  enum user_read_state saved_user_read_state = set_user_read(pmu, enabled);
  struct perf_event_attr attr = {
   .type = PERF_TYPE_HARDWARE,
   .config = perf_pmus__supports_extended_type()
   ? event | ((u64)pmu->type << PERF_PMU_TYPE_SHIFT)
    : event,
#ifdef __aarch64__
   .config1 = 0x2,  /* Request user access */
#endif
  };
  struct perf_evsel *evsel = NULL;
  int err;
  struct perf_event_mmap_page *pc;
  bool mapped = false, opened = false, rdpmc_supported;
  struct perf_counts_values counts = { .val = 0 };


  pr_debug("User space counter reading for PMU %s\n", pmu->name);
  /*
 * Restrict scheduling to only use the rdpmc on the CPUs the
 * event can be on. If the test doesn't run on the CPU of the
 * event then the event will be disabled and the pc->index test
 * will fail.
 */
  if (pmu->cpus != NULL)
   cpu_map__set_affinity(pmu->cpus);

  /* Make the evsel. */
  evsel = perf_evsel__new(&attr);
  if (!evsel) {
   pr_err("User space counter reading for PMU %s [Failed to allocate evsel]\n",
    pmu->name);
   ret = TEST_FAIL;
   goto cleanup;
  }

  err = perf_evsel__open(evsel, NULL, threads);
  if (err) {
   pr_err("User space counter reading for PMU %s [Failed to open evsel]\n",
    pmu->name);
   ret = TEST_SKIP;
   goto cleanup;
  }
  opened = true;
  err = perf_evsel__mmap(evsel, 0);
  if (err) {
   pr_err("User space counter reading for PMU %s [Failed to mmap evsel]\n",
    pmu->name);
   ret = TEST_FAIL;
   goto cleanup;
  }
  mapped = true;

  pc = perf_evsel__mmap_base(evsel, 0, 0);
  if (!pc) {
   pr_err("User space counter reading for PMU %s [Failed to get mmaped address]\n",
    pmu->name);
   ret = TEST_FAIL;
   goto cleanup;
  }

  if (saved_user_read_state == USER_READ_UNKNOWN)
   rdpmc_supported = pc->cap_user_rdpmc && pc->index;
  else
   rdpmc_supported = (enabled == USER_READ_ENABLED);

  if (rdpmc_supported && (!pc->cap_user_rdpmc || !pc->index)) {
   pr_err("User space counter reading for PMU %s [Failed unexpected supported counter access %d %d]\n",
    pmu->name, pc->cap_user_rdpmc, pc->index);
   ret = TEST_FAIL;
   goto cleanup;
  }

  if (!rdpmc_supported && pc->cap_user_rdpmc) {
   pr_err("User space counter reading for PMU %s [Failed unexpected unsupported counter access %d]\n",
    pmu->name, pc->cap_user_rdpmc);
   ret = TEST_FAIL;
   goto cleanup;
  }

  if (rdpmc_supported && pc->pmc_width < 32) {
   pr_err("User space counter reading for PMU %s [Failed width not set %d]\n",
    pmu->name, pc->pmc_width);
   ret = TEST_FAIL;
   goto cleanup;
  }

  perf_evsel__read(evsel, 0, 0, &counts);
  if (counts.val == 0) {
   pr_err("User space counter reading for PMU %s [Failed read]\n", pmu->name);
   ret = TEST_FAIL;
   goto cleanup;
  }

  for (int i = 0; i < 5; i++) {
   volatile int count = 0x10000 << i;
   __u64 start, end, last = 0;

   pr_debug("\tloop = %u, ", count);

   perf_evsel__read(evsel, 0, 0, &counts);
   start = counts.val;

   while (count--) ;

   perf_evsel__read(evsel, 0, 0, &counts);
   end = counts.val;

   if ((end - start) < last) {
    pr_err("User space counter reading for PMU %s [Failed invalid counter data: end=%llu start=%llu last= %llu]\n",
     pmu->name, end, start, last);
    ret = TEST_FAIL;
    goto cleanup;
   }
   last = end - start;
   pr_debug("count = %llu\n", last);
  }
  pr_debug("User space counter reading for PMU %s [Success]\n", pmu->name);
cleanup:
  if (mapped)
   perf_evsel__munmap(evsel);
  if (opened)
   perf_evsel__close(evsel);
  perf_evsel__delete(evsel);

  /* If the affinity was changed, then put it back to all CPUs. */
  if (pmu->cpus != NULL) {
   struct perf_cpu_map *cpus = cpu_map__online();

   cpu_map__set_affinity(cpus);
   perf_cpu_map__put(cpus);
  }
  set_user_read(pmu, saved_user_read_state);
 }
 perf_thread_map__put(threads);
 return ret;
}

static int test__mmap_user_read_instr(struct test_suite *test __maybe_unused,
          int subtest __maybe_unused)
{
 return test_stat_user_read(PERF_COUNT_HW_INSTRUCTIONS, USER_READ_ENABLED);
}

static int test__mmap_user_read_cycles(struct test_suite *test __maybe_unused,
           int subtest __maybe_unused)
{
 return test_stat_user_read(PERF_COUNT_HW_CPU_CYCLES, USER_READ_ENABLED);
}

static int test__mmap_user_read_instr_disabled(struct test_suite *test __maybe_unused,
            int subtest __maybe_unused)
{
 return test_stat_user_read(PERF_COUNT_HW_INSTRUCTIONS, USER_READ_DISABLED);
}

static int test__mmap_user_read_cycles_disabled(struct test_suite *test __maybe_unused,
      int subtest __maybe_unused)
{
 return test_stat_user_read(PERF_COUNT_HW_CPU_CYCLES, USER_READ_DISABLED);
}

static struct test_case tests__basic_mmap[] = {
 TEST_CASE_REASON("Read samples using the mmap interface",
    basic_mmap,
    "permissions"),
 TEST_CASE_REASON_EXCLUSIVE("User space counter reading of instructions",
    mmap_user_read_instr,
#if defined(__i386__) || defined(__x86_64__) || defined(__aarch64__) || \
    (defined(__riscv) && __riscv_xlen == 64)
    "permissions"
#else
    "unsupported"
#endif
  ),
 TEST_CASE_REASON_EXCLUSIVE("User space counter reading of cycles",
    mmap_user_read_cycles,
#if defined(__i386__) || defined(__x86_64__) || defined(__aarch64__) || \
    (defined(__riscv) && __riscv_xlen == 64)
    "permissions"
#else
    "unsupported"
#endif
  ),
 TEST_CASE_REASON_EXCLUSIVE("User space counter disabling instructions",
    mmap_user_read_instr_disabled,
#if defined(__i386__) || defined(__x86_64__) || defined(__aarch64__) || \
    (defined(__riscv) && __riscv_xlen == 64)
    "permissions"
#else
    "unsupported"
#endif
  ),
 TEST_CASE_REASON_EXCLUSIVE("User space counter disabling cycles",
    mmap_user_read_cycles_disabled,
#if defined(__i386__) || defined(__x86_64__) || defined(__aarch64__) || \
    (defined(__riscv) && __riscv_xlen == 64)
    "permissions"
#else
    "unsupported"
#endif
  ),
 { .name = NULL, }
};

struct test_suite suite__basic_mmap = {
 .desc = "mmap interface tests",
 .test_cases = tests__basic_mmap,
};