#include "utils.c"
#include "proflog.h"
#include <string.h>
+#include <assert.h>
#include <stdio.h>
-#if !defined(__APPLE__)
+#if !defined(__APPLE__) && !defined(__FreeBSD__)
#include <malloc.h>
#endif
#include <unistd.h>
#include <stdlib.h>
-#ifdef HOST_WIN32
-#undef HAVE_ZLIB
-#endif
-#if defined (HAVE_ZLIB)
+#if defined (HAVE_SYS_ZLIB)
#include <zlib.h>
#endif
#include <mono/metadata/profiler.h>
#define HASH_SIZE 9371
#define SMALL_HASH_SIZE 31
+#if defined(__native_client__) || defined(__native_client_codegen__)
+volatile int __nacl_thread_suspension_needed = 0;
+void __nacl_suspend_thread_if_needed() {}
+#endif
+
static int debug = 0;
static int collect_traces = 0;
static int show_traces = 0;
static uint64_t startup_time = 0;
static FILE* outfile = NULL;
+static int32_t
+read_int16 (unsigned char *p)
+{
+ int32_t value = *p++;
+ value |= (*p++) << 8;
+ return value;
+}
+
static int32_t
read_int32 (unsigned char *p)
{
ClassDesc *cd = class_hash [slot];
while (cd && cd->klass != klass)
cd = cd->next;
- if (!cd)
- return add_class (klass, "unresolved class");
+ if (!cd) {
+ char buf [128];
+ snprintf (buf, sizeof (buf), "unresolved class %p", (void*)klass);
+ return add_class (klass, buf);
+ }
return cd;
}
intptr_t code;
int len;
int recurse_count;
+ int sample_hits;
uint64_t calls;
uint64_t total_time;
uint64_t callee_time;
MethodDesc *cd = method_hash [slot];
while (cd && cd->method != method)
cd = cd->next;
- if (!cd)
- return add_method (method, "unknown method", 0, 0);
+ if (!cd) {
+ char buf [128];
+ snprintf (buf, sizeof (buf), "unknown method %p", (void*)method);
+ return add_method (method, buf, 0, 0);
+ }
return cd;
}
+static int num_stat_samples = 0;
+static int size_stat_samples = 0;
+uintptr_t *stat_samples = NULL;
+int *stat_sample_desc = NULL;
+
+static void
+add_stat_sample (int type, uintptr_t ip) {
+ if (num_stat_samples == size_stat_samples) {
+ size_stat_samples *= 2;
+ if (!size_stat_samples)
+ size_stat_samples = 32;
+ stat_samples = realloc (stat_samples, size_stat_samples * sizeof (uintptr_t));
+ stat_sample_desc = realloc (stat_sample_desc, size_stat_samples * sizeof (int));
+ }
+ stat_samples [num_stat_samples] = ip;
+ stat_sample_desc [num_stat_samples++] = type;
+}
+
+static MethodDesc*
+lookup_method_by_ip (uintptr_t ip)
+{
+ int i;
+ MethodDesc* m;
+ /* dumb */
+ for (i = 0; i < HASH_SIZE; ++i) {
+ m = method_hash [i];
+ while (m) {
+ //printf ("checking %p against %p-%p\n", (void*)ip, (void*)(m->code), (void*)(m->code + m->len));
+ if (ip >= (uintptr_t)m->code && ip < (uintptr_t)m->code + m->len) {
+ return m;
+ }
+ m = m->next;
+ }
+ }
+ return NULL;
+}
+
+static int
+compare_method_samples (const void *a, const void *b)
+{
+ MethodDesc *const*A = a;
+ MethodDesc *const*B = b;
+ if ((*A)->sample_hits == (*B)->sample_hits)
+ return 0;
+ if ((*B)->sample_hits < (*A)->sample_hits)
+ return -1;
+ return 1;
+}
+
+typedef struct _UnmanagedSymbol UnmanagedSymbol;
+struct _UnmanagedSymbol {
+ UnmanagedSymbol *parent;
+ char *name;
+ int is_binary;
+ uintptr_t addr;
+ uintptr_t size;
+ uintptr_t sample_hits;
+};
+
+static UnmanagedSymbol **usymbols = NULL;
+static int usymbols_size = 0;
+static int usymbols_num = 0;
+
+static int
+compare_usymbol_addr (const void *a, const void *b)
+{
+ UnmanagedSymbol *const*A = a;
+ UnmanagedSymbol *const*B = b;
+ if ((*B)->addr == (*A)->addr)
+ return 0;
+ if ((*B)->addr > (*A)->addr)
+ return -1;
+ return 1;
+}
+
+static int
+compare_usymbol_samples (const void *a, const void *b)
+{
+ UnmanagedSymbol *const*A = a;
+ UnmanagedSymbol *const*B = b;
+ if ((*B)->sample_hits == (*A)->sample_hits)
+ return 0;
+ if ((*B)->sample_hits < (*A)->sample_hits)
+ return -1;
+ return 1;
+}
+
+static void
+add_unmanaged_symbol (uintptr_t addr, char *name, uintptr_t size)
+{
+ UnmanagedSymbol *sym;
+ if (usymbols_num == usymbols_size) {
+ int new_size = usymbols_size * 2;
+ if (!new_size)
+ new_size = 16;
+ usymbols = realloc (usymbols, sizeof (void*) * new_size);
+ usymbols_size = new_size;
+ }
+ sym = calloc (sizeof (UnmanagedSymbol), 1);
+ sym->addr = addr;
+ sym->name = name;
+ sym->size = size;
+ usymbols [usymbols_num++] = sym;
+}
+
+/* only valid after the symbols are sorted */
+static UnmanagedSymbol*
+lookup_unmanaged_symbol (uintptr_t addr)
+{
+ int r = usymbols_num - 1;
+ int l = 0;
+ UnmanagedSymbol *sym;
+ int last_best = -1;
+ while (r >= l) {
+ int m = (l + r) / 2;
+ sym = usymbols [m];
+ if (addr == sym->addr)
+ return sym;
+ if (addr < sym->addr) {
+ r = m - 1;
+ } else if (addr > sym->addr) {
+ l = m + 1;
+ last_best = m;
+ }
+ }
+ if (last_best >= 0 && (addr - usymbols [last_best]->addr) < 4096)
+ return usymbols [last_best];
+ return NULL;
+}
+
+/* we use the same structure for binaries */
+static UnmanagedSymbol **ubinaries = NULL;
+static int ubinaries_size = 0;
+static int ubinaries_num = 0;
+
+static void
+add_unmanaged_binary (uintptr_t addr, char *name, uintptr_t size)
+{
+ UnmanagedSymbol *sym;
+ if (ubinaries_num == ubinaries_size) {
+ int new_size = ubinaries_size * 2;
+ if (!new_size)
+ new_size = 16;
+ ubinaries = realloc (ubinaries, sizeof (void*) * new_size);
+ ubinaries_size = new_size;
+ }
+ sym = calloc (sizeof (UnmanagedSymbol), 1);
+ sym->addr = addr;
+ sym->name = name;
+ sym->size = size;
+ sym->is_binary = 1;
+ ubinaries [ubinaries_num++] = sym;
+}
+
+static UnmanagedSymbol*
+lookup_unmanaged_binary (uintptr_t addr)
+{
+ int i;
+ for (i = 0; i < ubinaries_num; ++i) {
+ UnmanagedSymbol *ubin = ubinaries [i];
+ if (addr >= ubin->addr && addr < ubin->addr + ubin->size) {
+ return ubin;
+ }
+ }
+ return NULL;
+}
+
+static const char*
+sample_type_name (int type)
+{
+ switch (type) {
+ case SAMPLE_CYCLES: return "cycles";
+ case SAMPLE_INSTRUCTIONS: return "instructions retired";
+ case SAMPLE_CACHE_MISSES: return "cache misses";
+ case SAMPLE_CACHE_REFS: return "cache references";
+ case SAMPLE_BRANCHES: return "executed branches";
+ case SAMPLE_BRANCH_MISSES: return "unpredicted branches";
+ }
+ return "unknown";
+}
+
+static void
+set_usym_parent (UnmanagedSymbol** cachedus, int count)
+{
+ int i;
+ for (i = 0; i < count; ++i) {
+ UnmanagedSymbol *ubin = lookup_unmanaged_binary (cachedus [i]->addr);
+ if (ubin == cachedus [i])
+ continue;
+ cachedus [i]->parent = ubin;
+ }
+}
+
+static void
+print_usym (UnmanagedSymbol* um)
+{
+ if (um->parent)
+ fprintf (outfile, "\t%6d %6.2f %-36s in %s\n", um->sample_hits, um->sample_hits*100.0/num_stat_samples, um->name, um->parent->name);
+ else
+ fprintf (outfile, "\t%6d %6.2f %s\n", um->sample_hits, um->sample_hits*100.0/num_stat_samples, um->name);
+}
+
+static int
+sym_percent (uintptr_t sample_hits)
+{
+ double pc;
+ if (verbose)
+ return 1;
+ pc = sample_hits*100.0/num_stat_samples;
+ return pc >= 0.1;
+}
+
+static void
+dump_samples (void)
+{
+ int i, u;
+ int count = 0, msize = 0;
+ int unmanaged_hits = 0;
+ int unresolved_hits = 0;
+ MethodDesc** cachedm = NULL;
+ int ucount = 0, usize = 0;
+ UnmanagedSymbol** cachedus = NULL;
+ if (!num_stat_samples)
+ return;
+ qsort (usymbols, usymbols_num, sizeof (UnmanagedSymbol*), compare_usymbol_addr);
+ for (i = 0; i < num_stat_samples; ++i) {
+ MethodDesc *m = lookup_method_by_ip (stat_samples [i]);
+ if (m) {
+ if (!m->sample_hits) {
+ if (count == msize) {
+ msize *= 2;
+ if (!msize)
+ msize = 4;
+ cachedm = realloc (cachedm, sizeof (void*) * msize);
+ }
+ cachedm [count++] = m;
+ }
+ m->sample_hits++;
+ } else {
+ UnmanagedSymbol *usym = lookup_unmanaged_symbol (stat_samples [i]);
+ if (!usym) {
+ unresolved_hits++;
+ //printf ("unmanaged hit at %p\n", (void*)stat_samples [i]);
+ usym = lookup_unmanaged_binary (stat_samples [i]);
+ }
+ if (usym) {
+ if (!usym->sample_hits) {
+ if (ucount == usize) {
+ usize *= 2;
+ if (!usize)
+ usize = 4;
+ cachedus = realloc (cachedus, sizeof (void*) * usize);
+ }
+ cachedus [ucount++] = usym;
+ }
+ usym->sample_hits++;
+ }
+ unmanaged_hits++;
+ }
+ }
+ qsort (cachedm, count, sizeof (MethodDesc*), compare_method_samples);
+ qsort (cachedus, ucount, sizeof (UnmanagedSymbol*), compare_usymbol_samples);
+ set_usym_parent (cachedus, ucount);
+ fprintf (outfile, "\nStatistical samples summary\n");
+ fprintf (outfile, "\tSample type: %s\n", sample_type_name (stat_sample_desc [0]));
+ fprintf (outfile, "\tUnmanaged hits: %6d (%4.1f%%)\n", unmanaged_hits, (100.0*unmanaged_hits)/num_stat_samples);
+ fprintf (outfile, "\tManaged hits: %6d (%4.1f%%)\n", num_stat_samples - unmanaged_hits, (100.0*(num_stat_samples-unmanaged_hits))/num_stat_samples);
+ fprintf (outfile, "\tUnresolved hits: %6d (%4.1f%%)\n", unresolved_hits, (100.0*unresolved_hits)/num_stat_samples);
+ fprintf (outfile, "\t%6s %6s %s\n", "Hits", "%", "Method name");
+ i = 0;
+ u = 0;
+ while (i < count || u < ucount) {
+ if (i < count) {
+ MethodDesc *m = cachedm [i];
+ if (u < ucount) {
+ UnmanagedSymbol *um = cachedus [u];
+ if (um->sample_hits > m->sample_hits) {
+ if (!sym_percent (um->sample_hits))
+ break;
+ print_usym (um);
+ u++;
+ continue;
+ }
+ }
+ if (!sym_percent (m->sample_hits))
+ break;
+ fprintf (outfile, "\t%6d %6.2f %s\n", m->sample_hits, m->sample_hits*100.0/num_stat_samples, m->name);
+ i++;
+ continue;
+ }
+ if (u < ucount) {
+ UnmanagedSymbol *um = cachedus [u];
+ if (!sym_percent (um->sample_hits))
+ break;
+ print_usym (um);
+ u++;
+ continue;
+ }
+ }
+}
+
+typedef struct _HeapClassDesc HeapClassDesc;
typedef struct {
+ HeapClassDesc *klass;
+ uint64_t count;
+} HeapClassRevRef;
+
+struct _HeapClassDesc {
ClassDesc *klass;
int64_t count;
int64_t total_size;
-} HeapClassDesc;
+ HeapClassRevRef *rev_hash;
+ int rev_hash_size;
+ int rev_count;
+ uintptr_t pinned_references;
+ uintptr_t root_references;
+};
+
+static int
+add_rev_class_hashed (HeapClassRevRef *rev_hash, uintptr_t size, HeapClassDesc *hklass, uint64_t value)
+{
+ uintptr_t i;
+ uintptr_t start_pos;
+ start_pos = (hklass->klass->klass >> 2) % size;
+ assert (start_pos < size);
+ i = start_pos;
+ do {
+ if (rev_hash [i].klass == hklass) {
+ rev_hash [i].count += value;
+ return 0;
+ } else if (!rev_hash [i].klass) {
+ rev_hash [i].klass = hklass;
+ rev_hash [i].count += value;
+ start_pos = 0;
+ for (i = 0; i < size; ++i)
+ if (rev_hash [i].klass && rev_hash [i].klass->klass == hklass->klass)
+ start_pos ++;
+ assert (start_pos == 1);
+ return 1;
+ }
+ /* wrap around */
+ if (++i == size)
+ i = 0;
+ } while (i != start_pos);
+ /* should not happen */
+ printf ("failed revref store\n");
+ return 0;
+}
+
+static void
+add_heap_class_rev (HeapClassDesc *from, HeapClassDesc *to)
+{
+ uintptr_t i;
+ if (to->rev_count * 2 >= to->rev_hash_size) {
+ HeapClassRevRef *n;
+ uintptr_t old_size = to->rev_hash_size;
+ to->rev_hash_size *= 2;
+ if (to->rev_hash_size == 0)
+ to->rev_hash_size = 4;
+ n = calloc (sizeof (HeapClassRevRef) * to->rev_hash_size, 1);
+ for (i = 0; i < old_size; ++i) {
+ if (to->rev_hash [i].klass)
+ add_rev_class_hashed (n, to->rev_hash_size, to->rev_hash [i].klass, to->rev_hash [i].count);
+ }
+ if (to->rev_hash)
+ free (to->rev_hash);
+ to->rev_hash = n;
+ }
+ to->rev_count += add_rev_class_hashed (to->rev_hash, to->rev_hash_size, from, 1);
+}
+
+typedef struct {
+ uintptr_t objaddr;
+ HeapClassDesc *hklass;
+ uintptr_t num_refs;
+ uintptr_t refs [0];
+} HeapObjectDesc;
typedef struct _HeapShot HeapShot;
struct _HeapShot {
int hash_size;
HeapClassDesc **class_hash;
HeapClassDesc **sorted;
+ HeapObjectDesc **objects_hash;
+ uintptr_t objects_count;
+ uintptr_t objects_hash_size;
+ uintptr_t num_roots;
+ uintptr_t *roots;
+ uintptr_t *roots_extra;
+ int *roots_types;
};
static HeapShot *heap_shots = NULL;
-static HeapShot *current_heap_shot = NULL;
+static int num_heap_shots = 0;
static HeapShot*
new_heap_shot (uint64_t timestamp)
{
- HeapShot *p;
HeapShot *hs = calloc (sizeof (HeapShot), 1);
hs->hash_size = 4;
hs->class_hash = calloc (sizeof (void*), hs->hash_size);
hs->timestamp = timestamp;
- /* append it to the list */
- p = heap_shots;
- while (p && p->next)
- p = p->next;
- if (p)
- p->next = hs;
- else
- heap_shots = hs;
+ num_heap_shots++;
+ hs->next = heap_shots;
+ heap_shots = hs;
return hs;
}
}
static int
-add_heap_hashed (HeapClassDesc **hash, int hsize, ClassDesc *klass, uint64_t size, uint64_t count)
+add_heap_hashed (HeapClassDesc **hash, HeapClassDesc **retv, uintptr_t hsize, ClassDesc *klass, uint64_t size, uint64_t count)
{
- int i;
- unsigned int start_pos;
+ uintptr_t i;
+ uintptr_t start_pos;
start_pos = ((uintptr_t)klass->klass >> 2) % hsize;
i = start_pos;
do {
if (hash [i] && hash [i]->klass == klass) {
hash [i]->total_size += size;
hash [i]->count += count;
+ *retv = hash [i];
return 0;
} else if (!hash [i]) {
+ if (*retv) {
+ hash [i] = *retv;
+ return 1;
+ }
hash [i] = calloc (sizeof (HeapClassDesc), 1);
hash [i]->klass = klass;
hash [i]->total_size += size;
hash [i]->count += count;
+ *retv = hash [i];
return 1;
}
/* wrap around */
return 0;
}
-static void
-add_heap_shot_obj (HeapShot *hs, ClassDesc *klass, uint64_t size)
+static HeapClassDesc*
+add_heap_shot_class (HeapShot *hs, ClassDesc *klass, uint64_t size)
{
+ HeapClassDesc *res;
int i;
if (hs->class_count * 2 >= hs->hash_size) {
HeapClassDesc **n;
hs->hash_size = 4;
n = calloc (sizeof (void*) * hs->hash_size, 1);
for (i = 0; i < old_size; ++i) {
+ res = hs->class_hash [i];
if (hs->class_hash [i])
- add_heap_hashed (n, hs->hash_size, hs->class_hash [i]->klass, hs->class_hash [i]->total_size, hs->class_hash [i]->count);
+ add_heap_hashed (n, &res, hs->hash_size, hs->class_hash [i]->klass, hs->class_hash [i]->total_size, hs->class_hash [i]->count);
}
if (hs->class_hash)
free (hs->class_hash);
hs->class_hash = n;
}
- hs->class_count += add_heap_hashed (hs->class_hash, hs->hash_size, klass, size, 1);
+ res = NULL;
+ hs->class_count += add_heap_hashed (hs->class_hash, &res, hs->hash_size, klass, size, 1);
+ //if (res->count == 1)
+ // printf ("added heap class: %s\n", res->klass->name);
+ return res;
}
+static HeapObjectDesc*
+alloc_heap_obj (uintptr_t objaddr, HeapClassDesc *hklass, uintptr_t num_refs)
+{
+ HeapObjectDesc* ho = calloc (sizeof (HeapObjectDesc) + num_refs * sizeof (uintptr_t), 1);
+ ho->objaddr = objaddr;
+ ho->hklass = hklass;
+ ho->num_refs = num_refs;
+ return ho;
+}
+
+static uintptr_t
+heap_shot_find_obj_slot (HeapShot *hs, uintptr_t objaddr)
+{
+ uintptr_t i;
+ uintptr_t start_pos;
+ HeapObjectDesc **hash = hs->objects_hash;
+ start_pos = ((uintptr_t)objaddr >> 3) % hs->objects_hash_size;
+ i = start_pos;
+ do {
+ if (hash [i] && hash [i]->objaddr == objaddr) {
+ return i;
+ } else if (!hash [i]) {
+ break; /* fail */
+ }
+ /* wrap around */
+ if (++i == hs->objects_hash_size)
+ i = 0;
+ } while (i != start_pos);
+ /* should not happen */
+ //printf ("failed heap obj slot\n");
+ return -1;
+}
+
+static HeapObjectDesc*
+heap_shot_obj_add_refs (HeapShot *hs, uintptr_t objaddr, uintptr_t num, uintptr_t *ref_offset)
+{
+ HeapObjectDesc **hash = hs->objects_hash;
+ uintptr_t i = heap_shot_find_obj_slot (hs, objaddr);
+ if (i >= 0) {
+ HeapObjectDesc* ho = alloc_heap_obj (objaddr, hash [i]->hklass, hash [i]->num_refs + num);
+ *ref_offset = hash [i]->num_refs;
+ memcpy (ho->refs, hash [i]->refs, hash [i]->num_refs * sizeof (uintptr_t));
+ free (hash [i]);
+ hash [i] = ho;
+ return ho;
+ }
+ /* should not happen */
+ printf ("failed heap obj update\n");
+ return NULL;
+
+}
+
+static uintptr_t
+add_heap_hashed_obj (HeapObjectDesc **hash, uintptr_t hsize, HeapObjectDesc *obj)
+{
+ uintptr_t i;
+ uintptr_t start_pos;
+ start_pos = ((uintptr_t)obj->objaddr >> 3) % hsize;
+ i = start_pos;
+ do {
+ if (hash [i] && hash [i]->objaddr == obj->objaddr) {
+ printf ("duplicate object!\n");
+ return 0;
+ } else if (!hash [i]) {
+ hash [i] = obj;
+ return 1;
+ }
+ /* wrap around */
+ if (++i == hsize)
+ i = 0;
+ } while (i != start_pos);
+ /* should not happen */
+ printf ("failed heap obj store\n");
+ return 0;
+}
+
+static void
+add_heap_shot_obj (HeapShot *hs, HeapObjectDesc *obj)
+{
+ uintptr_t i;
+ if (hs->objects_count * 2 >= hs->objects_hash_size) {
+ HeapObjectDesc **n;
+ uintptr_t old_size = hs->objects_hash_size;
+ hs->objects_hash_size *= 2;
+ if (hs->objects_hash_size == 0)
+ hs->objects_hash_size = 4;
+ n = calloc (sizeof (void*) * hs->objects_hash_size, 1);
+ for (i = 0; i < old_size; ++i) {
+ if (hs->objects_hash [i])
+ add_heap_hashed_obj (n, hs->objects_hash_size, hs->objects_hash [i]);
+ }
+ if (hs->objects_hash)
+ free (hs->objects_hash);
+ hs->objects_hash = n;
+ }
+ hs->objects_count += add_heap_hashed_obj (hs->objects_hash, hs->objects_hash_size, obj);
+}
+
+static void
+heap_shot_resolve_reverse_refs (HeapShot *hs)
+{
+ uintptr_t i;
+ for (i = 0; i < hs->objects_hash_size; ++i) {
+ uintptr_t r;
+ HeapObjectDesc *ho = hs->objects_hash [i];
+ if (!ho)
+ continue;
+ for (r = 0; r < ho->num_refs; ++r) {
+ uintptr_t oi = heap_shot_find_obj_slot (hs, ho->refs [r]);
+ add_heap_class_rev (ho->hklass, hs->objects_hash [oi]->hklass);
+ }
+ }
+}
+
+#define MARK_GRAY 1
+#define MARK_BLACK 2
+
+static void
+heap_shot_mark_objects (HeapShot *hs)
+{
+ uintptr_t i, oi, r;
+ unsigned char *marks;
+ HeapObjectDesc *obj, *ref;
+ int marked_some;
+ uintptr_t num_marked = 0, num_unmarked;
+ for (i = 0; i < hs->num_roots; ++i) {
+ HeapClassDesc *cd;
+ oi = heap_shot_find_obj_slot (hs, hs->roots [i]);
+ if (oi == -1) {
+ continue;
+ }
+ obj = hs->objects_hash [oi];
+ cd = obj->hklass;
+ if (hs->roots_types [i] & MONO_PROFILE_GC_ROOT_PINNING)
+ cd->pinned_references++;
+ cd->root_references++;
+ }
+ if (!debug)
+ return;
+ /* consistency checks: it seems not all the objects are walked in the heap in some cases */
+ marks = calloc (hs->objects_hash_size, 1);
+ if (!marks)
+ return;
+ for (i = 0; i < hs->num_roots; ++i) {
+ oi = heap_shot_find_obj_slot (hs, hs->roots [i]);
+ if (oi == -1) {
+ fprintf (outfile, "root type 0x%x for obj %p (%s) not found in heap\n", hs->roots_types [i], (void*)hs->roots [i], lookup_class (hs->roots_extra [i])->name);
+ continue;
+ }
+ obj = hs->objects_hash [oi];
+ if (!marks [oi]) {
+ marks [oi] = obj->num_refs? MARK_GRAY: MARK_BLACK;
+ num_marked++;
+ }
+ }
+ marked_some = 1;
+ while (marked_some) {
+ marked_some = 0;
+ for (i = 0; i < hs->objects_hash_size; ++i) {
+ if (marks [i] != MARK_GRAY)
+ continue;
+ marks [i] = MARK_BLACK;
+ obj = hs->objects_hash [i];
+ for (r = 0; r < obj->num_refs; ++r) {
+ oi = heap_shot_find_obj_slot (hs, obj->refs [r]);
+ if (oi == -1) {
+ fprintf (outfile, "referenced obj %p not found in heap\n", (void*)obj->refs [r]);
+ continue;
+ }
+ ref = hs->objects_hash [oi];
+ if (!marks [oi]) {
+ marks [oi] = ref->num_refs? MARK_GRAY: MARK_BLACK;
+ }
+ }
+ marked_some++;
+ }
+ }
+
+ num_unmarked = 0;
+ for (i = 0; i < hs->objects_hash_size; ++i) {
+ if (hs->objects_hash [i] && !marks [i]) {
+ num_unmarked++;
+ fprintf (outfile, "object %p (%s) unmarked\n", (void*)hs->objects_hash [i], hs->objects_hash [i]->hklass->klass->name);
+ }
+ }
+ fprintf (outfile, "Total unmarked: %d/%d\n", num_unmarked, hs->objects_count);
+ free (marks);
+}
+
+static void
+heap_shot_free_objects (HeapShot *hs)
+{
+ uintptr_t i;
+ for (i = 0; i < hs->objects_hash_size; ++i) {
+ HeapObjectDesc *ho = hs->objects_hash [i];
+ if (ho)
+ free (ho);
+ }
+ if (hs->objects_hash)
+ free (hs->objects_hash);
+ hs->objects_hash = NULL;
+ hs->objects_hash_size = 0;
+ hs->objects_count = 0;
+}
+
+
struct _BackTrace {
BackTrace *next;
unsigned int hash;
typedef struct {
FILE *file;
-#if defined (HAVE_ZLIB)
+#if defined (HAVE_SYS_ZLIB)
gzFile *gzfile;
#endif
unsigned char *buf;
int version_major;
int version_minor;
int timer_overhead;
+ int pid;
+ int port;
uint64_t startup_time;
ThreadContext *threads;
ThreadContext *current;
MonitorDesc *monitor;
int stack_size;
int stack_id;
+ HeapShot *current_heap_shot;
+ uintptr_t num_roots;
+ uintptr_t size_roots;
+ uintptr_t *roots;
+ uintptr_t *roots_extra;
+ int *roots_types;
+ uint64_t gc_start_times [3];
};
static void
load_data (ProfContext *ctx, int size)
{
ensure_buffer (ctx, size);
-#if defined (HAVE_ZLIB)
- if (ctx->gzfile)
- return gzread (ctx->gzfile, ctx->buf, size) == size;
- else
+#if defined (HAVE_SYS_ZLIB)
+ if (ctx->gzfile) {
+ int r = gzread (ctx->gzfile, ctx->buf, size);
+ if (r == 0)
+ return size == 0? 1: 0;
+ return r == size;
+ } else
#endif
- return fread (ctx->buf, size, 1, ctx->file);
+ {
+ int r = fread (ctx->buf, size, 1, ctx->file);
+ if (r == 0)
+ return size == 0? 1: 0;
+ return r;
+ }
}
static ThreadContext*
return bt;
}
+static void
+thread_add_root (ThreadContext *ctx, uintptr_t obj, int root_type, uintptr_t extra_info)
+{
+ if (ctx->num_roots == ctx->size_roots) {
+ int new_size = ctx->size_roots * 2;
+ if (!new_size)
+ new_size = 4;
+ ctx->roots = realloc (ctx->roots, new_size * sizeof (uintptr_t));
+ ctx->roots_extra = realloc (ctx->roots_extra, new_size * sizeof (uintptr_t));
+ ctx->roots_types = realloc (ctx->roots_types, new_size * sizeof (int));
+ ctx->size_roots = new_size;
+ }
+ ctx->roots_types [ctx->num_roots] = root_type;
+ ctx->roots_extra [ctx->num_roots] = extra_info;
+ ctx->roots [ctx->num_roots++] = obj;
+}
+
static int
compare_callc (const void *a, const void *b)
{
}
typedef struct {
- uint64_t start_time; /* move this per-thread? */
uint64_t total_time;
uint64_t max_time;
int count;
typedef struct {
uint64_t created;
uint64_t destroyed;
+ uint64_t live;
uint64_t max_live;
TraceDesc traces;
} HandleInfo;
}
}
+static const char*
+get_root_name (int rtype)
+{
+ switch (rtype & MONO_PROFILE_GC_ROOT_TYPEMASK) {
+ case MONO_PROFILE_GC_ROOT_STACK: return "stack";
+ case MONO_PROFILE_GC_ROOT_FINALIZER: return "finalizer";
+ case MONO_PROFILE_GC_ROOT_HANDLE: return "handle";
+ case MONO_PROFILE_GC_ROOT_OTHER: return "other";
+ case MONO_PROFILE_GC_ROOT_MISC: return "misc";
+ default: return "unknown";
+ }
+}
+
static MethodDesc**
decode_bt (MethodDesc** sframes, int *size, unsigned char *p, unsigned char **endp, intptr_t ptr_base)
{
int len, i;
ThreadContext *thread;
- file_offset = ftell (ctx->file);
+#ifdef HAVE_SYS_ZLIB
+ if (ctx->gzfile)
+ file_offset = gztell (ctx->gzfile);
+ else
+#endif
+ file_offset = ftell (ctx->file);
if (!load_data (ctx, 48))
return 0;
p = ctx->buf;
- if (read_int32 (p) != BUF_ID)
+ if (read_int32 (p) != BUF_ID) {
+ fprintf (outfile, "Incorrect buffer id: 0x%x\n", read_int32 (p));
+ for (i = 0; i < 48; ++i) {
+ fprintf (outfile, "0x%x%s", p [i], i % 8?" ":"\n");
+ }
return 0;
+ }
len = read_int32 (p + 4);
time_base = read_int64 (p + 8);
ptr_base = read_int64 (p + 16);
uint64_t ev = decode_uleb128 (p, &p);
int gen = decode_uleb128 (p, &p);
if (debug)
- fprintf (outfile, "gc event for gen%d: %s at %llu\n", gen - 1, gc_event_name (ev), time_base);
+ fprintf (outfile, "gc event for gen%d: %s at %llu (thread: 0x%x)\n", gen, gc_event_name (ev), time_base, thread->thread_id);
if (gen > 2) {
fprintf (outfile, "incorrect gc gen: %d\n", gen);
break;
}
if (ev == MONO_GC_EVENT_START) {
- gc_info [gen].start_time = time_base;
+ thread->gc_start_times [gen] = time_base;
gc_info [gen].count++;
} else if (ev == MONO_GC_EVENT_END) {
- tdiff = time_base - gc_info [gen].start_time;
+ tdiff = time_base - thread->gc_start_times [gen];
gc_info [gen].total_time += tdiff;
if (tdiff > gc_info [gen].max_time)
gc_info [gen].max_time = tdiff;
if (htype > 3)
return 0;
handle_info [htype].created++;
+ handle_info [htype].live++;
add_trace_thread (thread, &handle_info [htype].traces, 1);
/* FIXME: we don't take into account timing here */
- if (handle_info [htype].created > handle_info [htype].max_live)
- handle_info [htype].max_live = handle_info [htype].created;
+ if (handle_info [htype].live > handle_info [htype].max_live)
+ handle_info [htype].max_live = handle_info [htype].live;
if (num_tracked_objects)
track_handle (OBJ_ADDR (objdiff), htype, handle);
if (debug)
uint32_t handle = decode_uleb128 (p, &p);
if (htype > 3)
return 0;
- handle_info [htype].created--;
+ handle_info [htype].destroyed ++;
+ handle_info [htype].live--;
if (debug)
fprintf (outfile, "handle (%s) %u destroyed\n", get_handle_name (htype), handle);
}
intptr_t codediff = decode_sleb128 (p, &p);
int codelen = decode_uleb128 (p, &p);
if (debug)
- fprintf (outfile, "jitted method %p (%s), size: %d\n", (void*)(method_base), p, codelen);
+ fprintf (outfile, "jitted method %p (%s), size: %d, code: %p\n", (void*)(method_base), p, codelen, (void*)(ptr_base + codediff));
add_method (method_base, (char*)p, ptr_base + codediff, codelen);
while (*p) p++;
p++;
case TYPE_HEAP: {
int subtype = *p & 0xf0;
if (subtype == TYPE_HEAP_OBJECT) {
+ HeapObjectDesc *ho;
int i;
intptr_t objdiff = decode_sleb128 (p + 1, &p);
intptr_t ptrdiff = decode_sleb128 (p, &p);
uint64_t size = decode_uleb128 (p, &p);
- int num = decode_uleb128 (p, &p);
+ uintptr_t num = decode_uleb128 (p, &p);
+ uintptr_t ref_offset;
+ uintptr_t last_obj_offset = 0;
ClassDesc *cd = lookup_class (ptr_base + ptrdiff);
+ if (size) {
+ HeapClassDesc *hcd = add_heap_shot_class (thread->current_heap_shot, cd, size);
+ if (collect_traces) {
+ ho = alloc_heap_obj (OBJ_ADDR (objdiff), hcd, num);
+ add_heap_shot_obj (thread->current_heap_shot, ho);
+ ref_offset = 0;
+ }
+ } else {
+ if (collect_traces)
+ ho = heap_shot_obj_add_refs (thread->current_heap_shot, OBJ_ADDR (objdiff), num, &ref_offset);
+ }
for (i = 0; i < num; ++i) {
/* FIXME: use object distance to measure how good
* the GC is at keeping related objects close
*/
+ uintptr_t offset = ctx->data_version > 1? last_obj_offset + decode_uleb128 (p, &p): -1;
intptr_t obj1diff = decode_sleb128 (p, &p);
+ last_obj_offset = offset;
+ if (collect_traces)
+ ho->refs [ref_offset + i] = OBJ_ADDR (obj1diff);
if (num_tracked_objects)
track_obj_reference (OBJ_ADDR (obj1diff), OBJ_ADDR (objdiff), cd);
}
if (debug && size)
fprintf (outfile, "traced object %p, size %llu (%s), refs: %d\n", (void*)OBJ_ADDR (objdiff), size, cd->name, num);
- if (size)
- add_heap_shot_obj (current_heap_shot, cd, size);
+ } else if (subtype == TYPE_HEAP_ROOT) {
+ uintptr_t num = decode_uleb128 (p + 1, &p);
+ uintptr_t gc_num = decode_uleb128 (p, &p);
+ int i;
+ for (i = 0; i < num; ++i) {
+ intptr_t objdiff = decode_sleb128 (p, &p);
+ int root_type = decode_uleb128 (p, &p);
+ /* we just discard the extra info for now */
+ uintptr_t extra_info = decode_uleb128 (p, &p);
+ if (debug)
+ fprintf (outfile, "object %p is a %s root\n", (void*)OBJ_ADDR (objdiff), get_root_name (root_type));
+ if (collect_traces)
+ thread_add_root (thread, OBJ_ADDR (objdiff), root_type, extra_info);
+ }
} else if (subtype == TYPE_HEAP_END) {
uint64_t tdiff = decode_uleb128 (p + 1, &p);
LOG_TIME (time_base, tdiff);
time_base += tdiff;
if (debug)
fprintf (outfile, "heap shot end\n");
- current_heap_shot = NULL;
+ if (collect_traces) {
+ HeapShot *hs = thread->current_heap_shot;
+ if (hs && thread->num_roots) {
+ /* transfer the root ownershipt to the heapshot */
+ hs->num_roots = thread->num_roots;
+ hs->roots = thread->roots;
+ hs->roots_extra = thread->roots_extra;
+ hs->roots_types = thread->roots_types;
+ } else {
+ free (thread->roots);
+ free (thread->roots_extra);
+ free (thread->roots_types);
+ }
+ thread->num_roots = 0;
+ thread->size_roots = 0;
+ thread->roots = NULL;
+ thread->roots_extra = NULL;
+ thread->roots_types = NULL;
+ heap_shot_resolve_reverse_refs (hs);
+ heap_shot_mark_objects (hs);
+ heap_shot_free_objects (hs);
+ }
+ thread->current_heap_shot = NULL;
} else if (subtype == TYPE_HEAP_START) {
uint64_t tdiff = decode_uleb128 (p + 1, &p);
LOG_TIME (time_base, tdiff);
time_base += tdiff;
if (debug)
fprintf (outfile, "heap shot start\n");
- current_heap_shot = new_heap_shot (time_base);
+ thread->current_heap_shot = new_heap_shot (time_base);
}
break;
}
}
break;
}
+ case TYPE_SAMPLE: {
+ int subtype = *p & 0xf0;
+ if (subtype == TYPE_SAMPLE_HIT) {
+ int i;
+ int sample_type = decode_uleb128 (p + 1, &p);
+ uint64_t tstamp = decode_uleb128 (p, &p);
+ int count = decode_uleb128 (p, &p);
+ for (i = 0; i < count; ++i) {
+ uintptr_t ip = ptr_base + decode_sleb128 (p, &p);
+ add_stat_sample (sample_type, ip);
+ if (debug)
+ fprintf (outfile, "sample hit, type: %d at %p\n", sample_type, (void*)ip);
+ }
+ } else if (subtype == TYPE_SAMPLE_USYM) {
+ /* un unmanaged symbol description */
+ uintptr_t addr = ptr_base + decode_sleb128 (p + 1, &p);
+ uintptr_t size = decode_uleb128 (p, &p);
+ char *name;
+ name = pstrdup ((char*)p);
+ add_unmanaged_symbol (addr, name, size);
+ if (debug)
+ fprintf (outfile, "unmanaged symbol %s at %p\n", name, (void*)addr);
+ while (*p) p++;
+ p++;
+ } else if (subtype == TYPE_SAMPLE_UBIN) {
+ /* un unmanaged binary loaded in memory */
+ uint64_t tdiff = decode_uleb128 (p + 1, &p);
+ uintptr_t addr = decode_sleb128 (p, &p);
+ uint64_t offset = decode_uleb128 (p, &p);
+ uintptr_t size = decode_uleb128 (p, &p);
+ char *name;
+ LOG_TIME (time_base, tdiff);
+ time_base += tdiff;
+ name = pstrdup ((char*)p);
+ add_unmanaged_binary (addr, name, size);
+ if (debug)
+ fprintf (outfile, "unmanaged binary %s at %p\n", name, (void*)addr);
+ while (*p) p++;
+ p++;
+ } else {
+ return 0;
+ }
+ break;
+ }
default:
- fprintf (outfile, "unhandled profiler event: 0x%x\n", *p);
+ fprintf (outfile, "unhandled profiler event: 0x%x at file offset: %llu + %d (len: %d\n)\n", *p, file_offset, p - ctx->buf, len);
exit (1);
}
}
printf ("Cannot open file: %s\n", name);
exit (1);
}
-#if defined (HAVE_ZLIB)
- ctx->gzfile = gzdopen (fileno (ctx->file), "rb");
+#if defined (HAVE_SYS_ZLIB)
+ if (ctx->file != stdin)
+ ctx->gzfile = gzdopen (fileno (ctx->file), "rb");
#endif
if (!load_data (ctx, 32))
return NULL;
p = ctx->buf;
- if (read_int32 (p) != LOG_HEADER_ID || p [6] != LOG_DATA_VERSION)
+ if (read_int32 (p) != LOG_HEADER_ID || p [6] > LOG_DATA_VERSION)
return NULL;
ctx->version_major = p [4];
ctx->version_minor = p [5];
return NULL;
ctx->startup_time = read_int64 (p + 8);
ctx->timer_overhead = read_int32 (p + 16);
+ ctx->pid = read_int32 (p + 24);
+ ctx->port = read_int16 (p + 28);
return ctx;
}
fprintf (outfile, "\tProfiler version: %d.%d\n", ctx->version_major, ctx->version_minor);
fprintf (outfile, "\tData version: %d\n", ctx->data_version);
fprintf (outfile, "\tMean timer overhead: %d nanoseconds\n", ctx->timer_overhead);
- fprintf (outfile, "\tProgram startup: %s\n", t);
+ fprintf (outfile, "\tProgram startup: %s", t);
+ if (ctx->pid)
+ fprintf (outfile, "\tProgram ID: %d\n", ctx->pid);
+ if (ctx->port)
+ fprintf (outfile, "\tServer listening on: %d\n", ctx->port);
}
static void
}
}
+static void
+dump_jit (void)
+{
+ int i;
+ int code_size = 0;
+ int compiled_methods = 0;
+ MethodDesc* m;
+ fprintf (outfile, "\nJIT summary\n");
+ for (i = 0; i < HASH_SIZE; ++i) {
+ m = method_hash [i];
+ for (m = method_hash [i]; m; m = m->next) {
+ if (!m->code)
+ continue;
+ compiled_methods++;
+ code_size += m->len;
+ }
+ }
+ fprintf (outfile, "\tCompiled methods: %d\n", compiled_methods);
+ fprintf (outfile, "\tGenerated code size: %d\n", code_size);
+}
+
static void
dump_allocations (void)
{
return 1;
}
+static int
+compare_rev_class (const void *a, const void *b)
+{
+ const HeapClassRevRef *A = a;
+ const HeapClassRevRef *B = b;
+ if (B->count == A->count)
+ return 0;
+ if (B->count < A->count)
+ return -1;
+ return 1;
+}
+
+static void
+dump_rev_claases (HeapClassRevRef *revs, int count)
+{
+ int j;
+ if (!show_traces)
+ return;
+ if (!count)
+ return;
+ for (j = 0; j < count; ++j) {
+ HeapClassDesc *cd = revs [j].klass;
+ fprintf (outfile, "\t\t%llu references from: %s\n", revs [j].count, cd->klass->name);
+ }
+}
+
static void
heap_shot_summary (HeapShot *hs, int hs_num, HeapShot *last_hs)
{
}
hs->sorted = sorted;
qsort (sorted, ccount, sizeof (void*), compare_heap_class);
- fprintf (outfile, "\n\tHeap shot %d at %.3f secs: size: %llu, object count: %llu, class count: %d\n",
- hs_num, (hs->timestamp - startup_time)/1000000000.0, size, count, ccount);
+ fprintf (outfile, "\n\tHeap shot %d at %.3f secs: size: %llu, object count: %llu, class count: %d, roots: %d\n",
+ hs_num, (hs->timestamp - startup_time)/1000000000.0, size, count, ccount, hs->num_roots);
if (!verbose && ccount > 30)
ccount = 30;
fprintf (outfile, "\t%10s %10s %8s Class name\n", "Bytes", "Count", "Average");
for (i = 0; i < ccount; ++i) {
+ HeapClassRevRef *rev_sorted;
+ int j, k;
HeapClassDesc *ocd = NULL;
cd = sorted [i];
if (last_hs)
} else {
fprintf (outfile, "\n");
}
+ if (!collect_traces)
+ continue;
+ rev_sorted = malloc (cd->rev_count * sizeof (HeapClassRevRef));
+ k = 0;
+ for (j = 0; j < cd->rev_hash_size; ++j) {
+ if (cd->rev_hash [j].klass)
+ rev_sorted [k++] = cd->rev_hash [j];
+ }
+ assert (cd->rev_count == k);
+ qsort (rev_sorted, cd->rev_count, sizeof (HeapClassRevRef), compare_rev_class);
+ if (cd->root_references)
+ fprintf (outfile, "\t\t%d root references (%d pinning)\n", cd->root_references, cd->pinned_references);
+ dump_rev_claases (rev_sorted, cd->rev_count);
+ free (rev_sorted);
}
+ free (sorted);
+}
+
+static int
+compare_heap_shots (const void *a, const void *b)
+{
+ HeapShot *const*A = a;
+ HeapShot *const*B = b;
+ if ((*B)->timestamp == (*A)->timestamp)
+ return 0;
+ if ((*B)->timestamp > (*A)->timestamp)
+ return -1;
+ return 1;
}
static void
dump_heap_shots (void)
{
+ HeapShot **hs_sorted;
HeapShot *hs;
HeapShot *last_hs = NULL;
int i;
if (!heap_shots)
return;
+ hs_sorted = malloc (num_heap_shots * sizeof (void*));
fprintf (outfile, "\nHeap shot summary\n");
i = 0;
- for (hs = heap_shots; hs; hs = hs->next) {
- heap_shot_summary (hs, i++, last_hs);
+ for (hs = heap_shots; hs; hs = hs->next)
+ hs_sorted [i++] = hs;
+ qsort (hs_sorted, num_heap_shots, sizeof (void*), compare_heap_shots);
+ for (i = 0; i < num_heap_shots; ++i) {
+ hs = hs_sorted [i];
+ heap_shot_summary (hs, i, last_hs);
last_hs = hs;
}
}
}
}
-static const char *reports = "header,gc,alloc,call,metadata,exception,monitor,thread,heapshot";
+static const char *reports = "header,jit,gc,sample,alloc,call,metadata,exception,monitor,thread,heapshot";
static const char*
match_option (const char *p, const char *opt)
dump_gcs ();
continue;
}
+ if ((opt = match_option (p, "jit")) != p) {
+ if (!parse_only)
+ dump_jit ();
+ continue;
+ }
if ((opt = match_option (p, "alloc")) != p) {
if (!parse_only)
dump_allocations ();
dump_heap_shots ();
continue;
}
+ if ((opt = match_option (p, "sample")) != p) {
+ if (!parse_only)
+ dump_samples ();
+ continue;
+ }
return 0;
}
return 1;