#include <zlib.h>
#endif
-#if defined(__linux__) && defined (ENABLE_PERF_EVENTS)
-
-#include <linux/perf_event.h>
-
-#define USE_PERF_EVENTS 1
-
-static int read_perf_mmap (MonoProfiler* prof, int cpu);
-
-#endif
-
#define BUFFER_SIZE (4096 * 16)
/* Worst-case size in bytes of a 64-bit value encoded with LEB128. */
static volatile int runtime_inited = 0;
static int command_port = 0;
static int heapshot_requested = 0;
-static int sample_type = 0;
static int sample_freq = 0;
static int do_mono_sample = 0;
static int in_shutdown = 0;
* type: TYPE_SAMPLE
* exinfo: one of TYPE_SAMPLE_HIT, TYPE_SAMPLE_USYM, TYPE_SAMPLE_UBIN, TYPE_SAMPLE_COUNTERS_DESC, TYPE_SAMPLE_COUNTERS
* if exinfo == TYPE_SAMPLE_HIT
- * [sample_type: byte] type of sample (SAMPLE_*)
* [thread: sleb128] thread id as difference from ptr_base
* [count: uleb128] number of following instruction addresses
* [ip: sleb128]* instruction pointer as difference from ptr_base
return 1;
}
-#if USE_PERF_EVENTS
-
-typedef struct {
- int perf_fd;
- unsigned int prev_pos;
- void *mmap_base;
- struct perf_event_mmap_page *page_desc;
-} PerfData ;
-
-static PerfData *perf_data = NULL;
-static int num_perf;
-#define PERF_PAGES_SHIFT 4
-static int num_pages = 1 << PERF_PAGES_SHIFT;
-static unsigned int mmap_mask;
-
-typedef struct {
- struct perf_event_header h;
- uint64_t ip;
- uint32_t pid;
- uint32_t tid;
- uint64_t timestamp;
- uint64_t period;
- uint64_t nframes;
-} PSample;
-
-static int
-perf_event_syscall (struct perf_event_attr *attr, pid_t pid, int cpu, int group_fd, unsigned long flags)
-{
- attr->size = PERF_ATTR_SIZE_VER0;
- //printf ("perf attr size: %d\n", attr->size);
-#if defined(__x86_64__)
- return syscall(/*__NR_perf_event_open*/ 298, attr, pid, cpu, group_fd, flags);
-#elif defined(__i386__)
- return syscall(/*__NR_perf_event_open*/ 336, attr, pid, cpu, group_fd, flags);
-#elif defined(__arm__) || defined (__aarch64__)
- return syscall(/*__NR_perf_event_open*/ 364, attr, pid, cpu, group_fd, flags);
-#else
- return -1;
-#endif
-}
-
-static int
-setup_perf_map (PerfData *perf)
-{
- perf->mmap_base = mmap (NULL, (num_pages + 1) * getpagesize (), PROT_READ|PROT_WRITE, MAP_SHARED, perf->perf_fd, 0);
- if (perf->mmap_base == MAP_FAILED) {
- if (do_debug)
- printf ("failed mmap\n");
- return 0;
- }
- perf->page_desc = perf->mmap_base;
- if (do_debug)
- printf ("mmap version: %d\n", perf->page_desc->version);
- return 1;
-}
-
-static void
-dump_perf_hits (MonoProfiler *prof, void *buf, int size)
-{
- int count = 1;
- int mbt_count = 0;
- void *end = (char*)buf + size;
- int samples = 0;
- int pid = getpid ();
-
- while (buf < end) {
- PSample *s = buf;
- if (s->h.size == 0)
- break;
- if (pid != s->pid) {
- if (do_debug)
- printf ("event for different pid: %d\n", s->pid);
- buf = (char*)buf + s->h.size;
- continue;
- }
- /*ip = (void*)s->ip;
- printf ("sample: %d, size: %d, ip: %p (%s), timestamp: %llu, nframes: %llu\n",
- s->h.type, s->h.size, ip, symbol_for (ip), s->timestamp, s->nframes);*/
-
- InterlockedIncrement (&sample_hits_ctr);
-
- ENTER_LOG (&sample_hits_ctr, logbuffer,
- EVENT_SIZE /* event */ +
- BYTE_SIZE /* type */ +
- LEB128_SIZE /* tid */ +
- LEB128_SIZE /* count */ +
- count * (
- LEB128_SIZE /* ip */
- ) +
- LEB128_SIZE /* managed count */ +
- mbt_count * (
- LEB128_SIZE /* method */
- )
- );
-
- emit_event (logbuffer, TYPE_SAMPLE | TYPE_SAMPLE_HIT);
- emit_byte (logbuffer, sample_type);
- /*
- * No useful thread ID to write here, since throughout the
- * profiler we use pthread_self () but the ID we get from
- * perf is the kernel's thread ID.
- */
- emit_ptr (logbuffer, 0);
- emit_value (logbuffer, count);
- emit_ptr (logbuffer, (void*)(uintptr_t)s->ip);
- /* no support here yet for the managed backtrace */
- emit_uvalue (logbuffer, mbt_count);
-
- EXIT_LOG_EXPLICIT (TRUE, FALSE);
-
- add_code_pointer (s->ip);
- buf = (char*)buf + s->h.size;
- samples++;
- }
- if (do_debug)
- printf ("dumped %d samples\n", samples);
- dump_unmanaged_coderefs (prof);
-}
-
-/* read events from the ring buffer */
-static int
-read_perf_mmap (MonoProfiler* prof, int cpu)
-{
- PerfData *perf = perf_data + cpu;
- unsigned char *buf;
- unsigned char *data = (unsigned char*)perf->mmap_base + getpagesize ();
- unsigned int head = perf->page_desc->data_head;
- int diff, size;
- unsigned int old;
-
- mono_memory_read_barrier ();
-
- old = perf->prev_pos;
- diff = head - old;
- if (diff < 0) {
- if (do_debug)
- printf ("lost mmap events: old: %d, head: %d\n", old, head);
- old = head;
- }
- size = head - old;
- if ((old & mmap_mask) + size != (head & mmap_mask)) {
- buf = data + (old & mmap_mask);
- size = mmap_mask + 1 - (old & mmap_mask);
- old += size;
- /* size bytes at buf */
- if (do_debug)
- printf ("found1 bytes of events: %d\n", size);
- dump_perf_hits (prof, buf, size);
- }
- buf = data + (old & mmap_mask);
- size = head - old;
- /* size bytes at buf */
- if (do_debug)
- printf ("found bytes of events: %d\n", size);
- dump_perf_hits (prof, buf, size);
- old += size;
- perf->prev_pos = old;
- perf->page_desc->data_tail = old;
- return 0;
-}
-
-static int
-setup_perf_event_for_cpu (PerfData *perf, int cpu)
-{
- struct perf_event_attr attr;
- memset (&attr, 0, sizeof (attr));
- attr.type = PERF_TYPE_HARDWARE;
- switch (sample_type) {
- case SAMPLE_CYCLES: attr.config = PERF_COUNT_HW_CPU_CYCLES; break;
- case SAMPLE_INSTRUCTIONS: attr.config = PERF_COUNT_HW_INSTRUCTIONS; break;
- case SAMPLE_CACHE_MISSES: attr.config = PERF_COUNT_HW_CACHE_MISSES; break;
- case SAMPLE_CACHE_REFS: attr.config = PERF_COUNT_HW_CACHE_REFERENCES; break;
- case SAMPLE_BRANCHES: attr.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS; break;
- case SAMPLE_BRANCH_MISSES: attr.config = PERF_COUNT_HW_BRANCH_MISSES; break;
- default: attr.config = PERF_COUNT_HW_CPU_CYCLES; break;
- }
- attr.sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID | PERF_SAMPLE_PERIOD | PERF_SAMPLE_TIME;
-// attr.sample_type |= PERF_SAMPLE_CALLCHAIN;
- attr.read_format = PERF_FORMAT_TOTAL_TIME_ENABLED | PERF_FORMAT_TOTAL_TIME_RUNNING | PERF_FORMAT_ID;
- attr.inherit = 1;
- attr.freq = 1;
- attr.sample_freq = sample_freq;
-
- perf->perf_fd = perf_event_syscall (&attr, getpid (), cpu, -1, 0);
- if (do_debug)
- printf ("perf fd: %d, freq: %d, event: %llu\n", perf->perf_fd, sample_freq, attr.config);
- if (perf->perf_fd < 0) {
- if (perf->perf_fd == -EPERM) {
- fprintf (stderr, "Perf syscall denied, do \"echo 1 > /proc/sys/kernel/perf_event_paranoid\" as root to enable.\n");
- } else {
- if (do_debug)
- perror ("open perf event");
- }
- return 0;
- }
- if (!setup_perf_map (perf)) {
- close (perf->perf_fd);
- perf->perf_fd = -1;
- return 0;
- }
- return 1;
-}
-
-static int
-setup_perf_event (void)
-{
- int i, count = 0;
- mmap_mask = num_pages * getpagesize () - 1;
- num_perf = mono_cpu_count ();
- perf_data = g_calloc (num_perf, sizeof (PerfData));
- for (i = 0; i < num_perf; ++i) {
- count += setup_perf_event_for_cpu (perf_data + i, i);
- }
- if (count)
- return 1;
- g_free (perf_data);
- perf_data = NULL;
- return 0;
-}
-
-#endif /* USE_PERF_EVENTS */
-
typedef struct MonoCounterAgent {
MonoCounter *counter;
// MonoCounterAgent specific data :
g_free (cur);
}
-#if USE_PERF_EVENTS
- if (perf_data) {
- int i;
- for (i = 0; i < num_perf; ++i)
- read_perf_mmap (prof, i);
- }
-#endif
-
/*
* Ensure that we empty the LLS completely, even if some nodes are
* not immediately removed upon calling mono_lls_remove (), by
add_to_fd_set (fd_set *set, int fd, int *max_fd)
{
/*
- * This should only trigger for the basic FDs (server socket, pipe, perf
- * events) at startup if for some mysterious reason they're too large. In
- * this case, the profiler really can't function, and we're better off
- * printing an error and exiting.
+ * This should only trigger for the basic FDs (server socket, pipes) at
+ * startup if for some mysterious reason they're too large. In this case,
+ * the profiler really can't function, and we're better off printing an
+ * error and exiting.
*/
if (fd >= FD_SETSIZE) {
fprintf (stderr, "File descriptor is out of bounds for fd_set: %d\n", fd);
for (gint i = 0; i < command_sockets->len; i++)
add_to_fd_set (&rfds, g_array_index (command_sockets, int, i), &max_fd);
-#if USE_PERF_EVENTS
- if (perf_data) {
- for (int i = 0; i < num_perf; ++i) {
- if (perf_data [i].perf_fd < 0)
- continue;
-
- add_to_fd_set (&rfds, perf_data [i].perf_fd, &max_fd);
- }
- }
-#endif
-
struct timeval tv = { .tv_sec = 1, .tv_usec = 0 };
// Sleep for 1sec or until a file descriptor has data.
buffer_unlock_excl ();
-#if USE_PERF_EVENTS
- if (perf_data) {
- for (int i = 0; i < num_perf; ++i) {
- if (perf_data [i].perf_fd < 0 || !FD_ISSET (perf_data [i].perf_fd, &rfds))
- continue;
-
- read_perf_mmap (prof, i);
- }
- }
-#endif
-
// Are we shutting down?
if (FD_ISSET (prof->pipes [0], &rfds)) {
char c;
);
emit_event_time (logbuffer, TYPE_SAMPLE | TYPE_SAMPLE_HIT, sample->time);
- emit_byte (logbuffer, sample_type);
+ emit_byte (logbuffer, SAMPLE_CYCLES);
emit_ptr (logbuffer, (void *) sample->tid);
emit_value (logbuffer, 1);
prof->gzfile = gzdopen (fileno (prof->file), "wb");
#endif
-#if USE_PERF_EVENTS
- setup_perf_event ();
-
- if (!perf_data) {
- /* FIXME: warn if different freq or sample type */
- do_mono_sample = 1;
- }
-#endif
-
/*
* If you hit this assert while increasing MAX_FRAMES, you need to increase
* SAMPLE_BLOCK_SIZE as well.
return p;
}
-typedef struct {
- const char *name;
- int sample_mode;
-} SampleMode;
-
-static const SampleMode sample_modes [] = {
- {"cycles", SAMPLE_CYCLES},
- {"instr", SAMPLE_INSTRUCTIONS},
- {"cachemiss", SAMPLE_CACHE_MISSES},
- {"cacherefs", SAMPLE_CACHE_REFS},
- {"branches", SAMPLE_BRANCHES},
- {"branchmiss", SAMPLE_BRANCH_MISSES},
- {NULL, 0}
-};
-
static void
-set_sample_mode (char* val, int allow_empty)
+set_sample_freq (char *val)
{
- char *end;
- char *maybe_freq = NULL;
- unsigned int count;
- const SampleMode *smode = sample_modes;
-#ifndef USE_PERF_EVENTS
do_mono_sample = 1;
-#endif
- if (allow_empty && !val) {
- sample_type = SAMPLE_CYCLES;
- sample_freq = 100;
- return;
- }
- if (strcmp (val, "mono") == 0) {
- do_mono_sample = 1;
- sample_type = SAMPLE_CYCLES;
- g_free (val);
+ sample_freq = 100;
+
+ if (!val)
return;
- }
- for (smode = sample_modes; smode->name; smode++) {
- int l = strlen (smode->name);
- if (strncmp (val, smode->name, l) == 0) {
- sample_type = smode->sample_mode;
- maybe_freq = val + l;
- break;
- }
- }
- if (!smode->name)
- usage (1);
- if (*maybe_freq == '/') {
- count = strtoul (maybe_freq + 1, &end, 10);
- if (maybe_freq + 1 == end)
+
+ char *p = val;
+
+ // Is it only the frequency (new option style)?
+ if (isdigit (*p))
+ goto parse;
+
+ // Skip the sample type for backwards compatibility.
+ while (isalpha (*p))
+ p++;
+
+ // Skip the forward slash only if we got a sample type.
+ if (p != val && *p == '/') {
+ p++;
+
+ char *end;
+
+ parse:
+ sample_freq = strtoul (p, &end, 10);
+
+ if (p == end)
usage (1);
- sample_freq = count;
- } else if (*maybe_freq != 0) {
- usage (1);
- } else {
- sample_freq = 100;
+
+ p = end;
}
+
+ if (*p)
+ usage (1);
+
g_free (val);
}
events &= ~MONO_PROFILE_GC_MOVES;
events &= ~MONO_PROFILE_ENTER_LEAVE;
nocalls = 1;
- set_sample_mode (val, 1);
+ set_sample_freq (val);
continue;
}
if ((opt = match_option (p, "zip", NULL)) != p) {
if (do_coverage)
mono_profiler_install_coverage_filter (coverage_filter);
- if (do_mono_sample && sample_type == SAMPLE_CYCLES && sample_freq) {
+ if (do_mono_sample && sample_freq) {
events |= MONO_PROFILE_STATISTICAL;
mono_profiler_set_statistical_mode (sampling_mode, sample_freq);
mono_profiler_install_statistical (mono_sample_hit);
projects / mono.git / commitdiff