Commit 04487d56 authored by Rémy Oudompheng's avatar Rémy Oudompheng

runtime: convert MemProfile, BlockProfile, ThreadCreateProfile to Go.

LGTM=khr
R=golang-codereviews, bradfitz, khr
CC=golang-codereviews
https://golang.org/cl/123680043
parent cccd66c6
...@@ -75,20 +75,6 @@ func (r *MemProfileRecord) Stack() []uintptr { ...@@ -75,20 +75,6 @@ func (r *MemProfileRecord) Stack() []uintptr {
return r.Stack0[0:] return r.Stack0[0:]
} }
// MemProfile returns n, the number of records in the current memory profile.
// If len(p) >= n, MemProfile copies the profile into p and returns n, true.
// If len(p) < n, MemProfile does not change p and returns n, false.
//
// If inuseZero is true, the profile includes allocation records
// where r.AllocBytes > 0 but r.AllocBytes == r.FreeBytes.
// These are sites where memory was allocated, but it has all
// been released back to the runtime.
//
// Most clients should use the runtime/pprof package or
// the testing package's -test.memprofile flag instead
// of calling MemProfile directly.
func MemProfile(p []MemProfileRecord, inuseZero bool) (n int, ok bool)
// A StackRecord describes a single execution stack. // A StackRecord describes a single execution stack.
type StackRecord struct { type StackRecord struct {
Stack0 [32]uintptr // stack trace for this record; ends at first 0 entry Stack0 [32]uintptr // stack trace for this record; ends at first 0 entry
...@@ -105,14 +91,6 @@ func (r *StackRecord) Stack() []uintptr { ...@@ -105,14 +91,6 @@ func (r *StackRecord) Stack() []uintptr {
return r.Stack0[0:] return r.Stack0[0:]
} }
// ThreadCreateProfile returns n, the number of records in the thread creation profile.
// If len(p) >= n, ThreadCreateProfile copies the profile into p and returns n, true.
// If len(p) < n, ThreadCreateProfile does not change p and returns n, false.
//
// Most clients should use the runtime/pprof package instead
// of calling ThreadCreateProfile directly.
func ThreadCreateProfile(p []StackRecord) (n int, ok bool)
// GoroutineProfile returns n, the number of records in the active goroutine stack profile. // GoroutineProfile returns n, the number of records in the active goroutine stack profile.
// If len(p) >= n, GoroutineProfile copies the profile into p and returns n, true. // If len(p) >= n, GoroutineProfile copies the profile into p and returns n, true.
// If len(p) < n, GoroutineProfile does not change p and returns n, false. // If len(p) < n, GoroutineProfile does not change p and returns n, false.
...@@ -156,15 +134,6 @@ type BlockProfileRecord struct { ...@@ -156,15 +134,6 @@ type BlockProfileRecord struct {
StackRecord StackRecord
} }
// BlockProfile returns n, the number of records in the current blocking profile.
// If len(p) >= n, BlockProfile copies the profile into p and returns n, true.
// If len(p) < n, BlockProfile does not change p and returns n, false.
//
// Most clients should use the runtime/pprof package or
// the testing package's -test.blockprofile flag instead
// of calling BlockProfile directly.
func BlockProfile(p []BlockProfileRecord) (n int, ok bool)
// Stack formats a stack trace of the calling goroutine into buf // Stack formats a stack trace of the calling goroutine into buf
// and returns the number of bytes written to buf. // and returns the number of bytes written to buf.
// If all is true, Stack formats stack traces of all other goroutines // If all is true, Stack formats stack traces of all other goroutines
......
// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package runtime
import (
"unsafe"
)
// Malloc profiling.
// Patterned after tcmalloc's algorithms; shorter code.
// NOTE(rsc): Everything here could use cas if contention became an issue.
var proflock lock
// All memory allocations are local and do not escape outside of the profiler.
// The profiler is forbidden from referring to garbage-collected memory.
var (
mbuckets *bucket // memory profile buckets
bbuckets *bucket // blocking profile buckets
)
// MemProfile returns n, the number of records in the current memory profile.
// If len(p) >= n, MemProfile copies the profile into p and returns n, true.
// If len(p) < n, MemProfile does not change p and returns n, false.
//
// If inuseZero is true, the profile includes allocation records
// where r.AllocBytes > 0 but r.AllocBytes == r.FreeBytes.
// These are sites where memory was allocated, but it has all
// been released back to the runtime.
//
// Most clients should use the runtime/pprof package or
// the testing package's -test.memprofile flag instead
// of calling MemProfile directly.
func MemProfile(p []MemProfileRecord, inuseZero bool) (n int, ok bool) {
golock(&proflock)
clear := true
for b := mbuckets; b != nil; b = b.allnext {
if inuseZero || b.data.mp.alloc_bytes != b.data.mp.free_bytes {
n++
}
if b.data.mp.allocs != 0 || b.data.mp.frees != 0 {
clear = false
}
}
if clear {
// Absolutely no data, suggesting that a garbage collection
// has not yet happened. In order to allow profiling when
// garbage collection is disabled from the beginning of execution,
// accumulate stats as if a GC just happened, and recount buckets.
mprof_GC()
mprof_GC()
n = 0
for b := mbuckets; b != nil; b = b.allnext {
if inuseZero || b.data.mp.alloc_bytes != b.data.mp.free_bytes {
n++
}
}
}
if n <= len(p) {
ok = true
idx := 0
for b := mbuckets; b != nil; b = b.allnext {
if inuseZero || b.data.mp.alloc_bytes != b.data.mp.free_bytes {
record(&p[idx], b)
idx++
}
}
}
gounlock(&proflock)
return
}
func mprof_GC() {
for b := mbuckets; b != nil; b = b.allnext {
b.data.mp.allocs += b.data.mp.prev_allocs
b.data.mp.frees += b.data.mp.prev_frees
b.data.mp.alloc_bytes += b.data.mp.prev_alloc_bytes
b.data.mp.free_bytes += b.data.mp.prev_free_bytes
b.data.mp.prev_allocs = b.data.mp.recent_allocs
b.data.mp.prev_frees = b.data.mp.recent_frees
b.data.mp.prev_alloc_bytes = b.data.mp.recent_alloc_bytes
b.data.mp.prev_free_bytes = b.data.mp.recent_free_bytes
b.data.mp.recent_allocs = 0
b.data.mp.recent_frees = 0
b.data.mp.recent_alloc_bytes = 0
b.data.mp.recent_free_bytes = 0
}
}
// Write b's data to r.
func record(r *MemProfileRecord, b *bucket) {
r.AllocBytes = int64(b.data.mp.alloc_bytes)
r.FreeBytes = int64(b.data.mp.free_bytes)
r.AllocObjects = int64(b.data.mp.allocs)
r.FreeObjects = int64(b.data.mp.frees)
for i := 0; uint(i) < b.nstk && i < len(r.Stack0); i++ {
r.Stack0[i] = *(*uintptr)(add(unsafe.Pointer(&b.stk), uintptr(i)*ptrSize))
}
for i := b.nstk; i < uint(len(r.Stack0)); i++ {
r.Stack0[i] = 0
}
}
// BlockProfile returns n, the number of records in the current blocking profile.
// If len(p) >= n, BlockProfile copies the profile into p and returns n, true.
// If len(p) < n, BlockProfile does not change p and returns n, false.
//
// Most clients should use the runtime/pprof package or
// the testing package's -test.blockprofile flag instead
// of calling BlockProfile directly.
func BlockProfile(p []BlockProfileRecord) (n int, ok bool) {
golock(&proflock)
for b := bbuckets; b != nil; b = b.allnext {
n++
}
if n <= len(p) {
ok = true
idx := 0
for b := bbuckets; b != nil; b = b.allnext {
bp := (*_4_)(unsafe.Pointer(&b.data))
p[idx].Count = int64(bp.count)
p[idx].Cycles = int64(bp.cycles)
i := 0
for uint(i) < b.nstk && i < len(p[idx].Stack0) {
p[idx].Stack0[i] = *(*uintptr)(add(unsafe.Pointer(&b.stk), uintptr(i)*ptrSize))
i++
}
for i < len(p[idx].Stack0) {
p[idx].Stack0[i] = 0
i++
}
idx++
}
}
gounlock(&proflock)
return
}
// ThreadCreateProfile returns n, the number of records in the thread creation profile.
// If len(p) >= n, ThreadCreateProfile copies the profile into p and returns n, true.
// If len(p) < n, ThreadCreateProfile does not change p and returns n, false.
//
// Most clients should use the runtime/pprof package instead
// of calling ThreadCreateProfile directly.
func ThreadCreateProfile(p []StackRecord) (n int, ok bool) {
first := (*m)(goatomicloadp(unsafe.Pointer(&allm)))
for mp := first; mp != nil; mp = mp.alllink {
n++
}
if n <= len(p) {
ok = true
i := 0
for mp := first; mp != nil; mp = mp.alllink {
for s := range mp.createstack {
p[i].Stack0[s] = uintptr(mp.createstack[s])
}
i++
}
}
return
}
...@@ -14,7 +14,7 @@ package runtime ...@@ -14,7 +14,7 @@ package runtime
#include "type.h" #include "type.h"
// NOTE(rsc): Everything here could use cas if contention became an issue. // NOTE(rsc): Everything here could use cas if contention became an issue.
static Lock proflock; extern Lock runtime·proflock;
// All memory allocations are local and do not escape outside of the profiler. // All memory allocations are local and do not escape outside of the profiler.
// The profiler is forbidden from referring to garbage-collected memory. // The profiler is forbidden from referring to garbage-collected memory.
...@@ -25,8 +25,8 @@ enum { ...@@ -25,8 +25,8 @@ enum {
BuckHashSize = 179999, BuckHashSize = 179999,
}; };
static Bucket **buckhash; static Bucket **buckhash;
static Bucket *mbuckets; // memory profile buckets extern Bucket *runtime·mbuckets; // memory profile buckets
static Bucket *bbuckets; // blocking profile buckets extern Bucket *runtime·bbuckets; // blocking profile buckets
static uintptr bucketmem; static uintptr bucketmem;
// Return the bucket for stk[0:nstk], allocating new bucket if needed. // Return the bucket for stk[0:nstk], allocating new bucket if needed.
...@@ -77,11 +77,11 @@ stkbucket(int32 typ, uintptr size, uintptr *stk, int32 nstk, bool alloc) ...@@ -77,11 +77,11 @@ stkbucket(int32 typ, uintptr size, uintptr *stk, int32 nstk, bool alloc)
b->next = buckhash[i]; b->next = buckhash[i];
buckhash[i] = b; buckhash[i] = b;
if(typ == MProf) { if(typ == MProf) {
b->allnext = mbuckets; b->allnext = runtime·mbuckets;
mbuckets = b; runtime·mbuckets = b;
} else { } else {
b->allnext = bbuckets; b->allnext = runtime·bbuckets;
bbuckets = b; runtime·bbuckets = b;
} }
return b; return b;
} }
...@@ -91,7 +91,7 @@ MProf_GC(void) ...@@ -91,7 +91,7 @@ MProf_GC(void)
{ {
Bucket *b; Bucket *b;
for(b=mbuckets; b; b=b->allnext) { for(b=runtime·mbuckets; b; b=b->allnext) {
b->data.mp.allocs += b->data.mp.prev_allocs; b->data.mp.allocs += b->data.mp.prev_allocs;
b->data.mp.frees += b->data.mp.prev_frees; b->data.mp.frees += b->data.mp.prev_frees;
b->data.mp.alloc_bytes += b->data.mp.prev_alloc_bytes; b->data.mp.alloc_bytes += b->data.mp.prev_alloc_bytes;
...@@ -113,9 +113,9 @@ MProf_GC(void) ...@@ -113,9 +113,9 @@ MProf_GC(void)
void void
runtime·MProf_GC(void) runtime·MProf_GC(void)
{ {
runtime·lock(&proflock); runtime·lock(&runtime·proflock);
MProf_GC(); MProf_GC();
runtime·unlock(&proflock); runtime·unlock(&runtime·proflock);
} }
// Called by malloc to record a profiled block. // Called by malloc to record a profiled block.
...@@ -127,11 +127,11 @@ runtime·MProf_Malloc(void *p, uintptr size) ...@@ -127,11 +127,11 @@ runtime·MProf_Malloc(void *p, uintptr size)
int32 nstk; int32 nstk;
nstk = runtime·callers(1, stk, nelem(stk)); nstk = runtime·callers(1, stk, nelem(stk));
runtime·lock(&proflock); runtime·lock(&runtime·proflock);
b = stkbucket(MProf, size, stk, nstk, true); b = stkbucket(MProf, size, stk, nstk, true);
b->data.mp.recent_allocs++; b->data.mp.recent_allocs++;
b->data.mp.recent_alloc_bytes += size; b->data.mp.recent_alloc_bytes += size;
runtime·unlock(&proflock); runtime·unlock(&runtime·proflock);
// Setprofilebucket locks a bunch of other mutexes, so we call it outside of proflock. // Setprofilebucket locks a bunch of other mutexes, so we call it outside of proflock.
// This reduces potential contention and chances of deadlocks. // This reduces potential contention and chances of deadlocks.
...@@ -158,11 +158,11 @@ runtime·mprofMalloc_m(void) ...@@ -158,11 +158,11 @@ runtime·mprofMalloc_m(void)
nstk = runtime·callers(1, stk, nelem(stk)); nstk = runtime·callers(1, stk, nelem(stk));
else else
nstk = runtime·gcallers(g->m->curg, 1, stk, nelem(stk)); nstk = runtime·gcallers(g->m->curg, 1, stk, nelem(stk));
runtime·lock(&proflock); runtime·lock(&runtime·proflock);
b = stkbucket(MProf, size, stk, nstk, true); b = stkbucket(MProf, size, stk, nstk, true);
b->data.mp.recent_allocs++; b->data.mp.recent_allocs++;
b->data.mp.recent_alloc_bytes += size; b->data.mp.recent_alloc_bytes += size;
runtime·unlock(&proflock); runtime·unlock(&runtime·proflock);
// Setprofilebucket locks a bunch of other mutexes, so we call it outside of proflock. // Setprofilebucket locks a bunch of other mutexes, so we call it outside of proflock.
// This reduces potential contention and chances of deadlocks. // This reduces potential contention and chances of deadlocks.
...@@ -175,7 +175,7 @@ runtime·mprofMalloc_m(void) ...@@ -175,7 +175,7 @@ runtime·mprofMalloc_m(void)
void void
runtime·MProf_Free(Bucket *b, uintptr size, bool freed) runtime·MProf_Free(Bucket *b, uintptr size, bool freed)
{ {
runtime·lock(&proflock); runtime·lock(&runtime·proflock);
if(freed) { if(freed) {
b->data.mp.recent_frees++; b->data.mp.recent_frees++;
b->data.mp.recent_free_bytes += size; b->data.mp.recent_free_bytes += size;
...@@ -183,7 +183,7 @@ runtime·MProf_Free(Bucket *b, uintptr size, bool freed) ...@@ -183,7 +183,7 @@ runtime·MProf_Free(Bucket *b, uintptr size, bool freed)
b->data.mp.prev_frees++; b->data.mp.prev_frees++;
b->data.mp.prev_free_bytes += size; b->data.mp.prev_free_bytes += size;
} }
runtime·unlock(&proflock); runtime·unlock(&runtime·proflock);
} }
int64 runtime·blockprofilerate; // in CPU ticks int64 runtime·blockprofilerate; // in CPU ticks
...@@ -219,74 +219,11 @@ runtime·blockevent(int64 cycles, int32 skip) ...@@ -219,74 +219,11 @@ runtime·blockevent(int64 cycles, int32 skip)
return; return;
nstk = runtime·callers(skip, stk, nelem(stk)); nstk = runtime·callers(skip, stk, nelem(stk));
runtime·lock(&proflock); runtime·lock(&runtime·proflock);
b = stkbucket(BProf, 0, stk, nstk, true); b = stkbucket(BProf, 0, stk, nstk, true);
b->data.bp.count++; b->data.bp.count++;
b->data.bp.cycles += cycles; b->data.bp.cycles += cycles;
runtime·unlock(&proflock); runtime·unlock(&runtime·proflock);
}
// Go interface to profile data. (Declared in debug.go)
// Must match MemProfileRecord in debug.go.
typedef struct Record Record;
struct Record {
int64 alloc_bytes, free_bytes;
int64 alloc_objects, free_objects;
uintptr stk[32];
};
// Write b's data to r.
static void
record(Record *r, Bucket *b)
{
int32 i;
r->alloc_bytes = b->data.mp.alloc_bytes;
r->free_bytes = b->data.mp.free_bytes;
r->alloc_objects = b->data.mp.allocs;
r->free_objects = b->data.mp.frees;
for(i=0; i<b->nstk && i<nelem(r->stk); i++)
r->stk[i] = b->stk[i];
for(; i<nelem(r->stk); i++)
r->stk[i] = 0;
}
func MemProfile(p Slice, include_inuse_zero bool) (n int, ok bool) {
Bucket *b;
Record *r;
bool clear;
runtime·lock(&proflock);
n = 0;
clear = true;
for(b=mbuckets; b; b=b->allnext) {
if(include_inuse_zero || b->data.mp.alloc_bytes != b->data.mp.free_bytes)
n++;
if(b->data.mp.allocs != 0 || b->data.mp.frees != 0)
clear = false;
}
if(clear) {
// Absolutely no data, suggesting that a garbage collection
// has not yet happened. In order to allow profiling when
// garbage collection is disabled from the beginning of execution,
// accumulate stats as if a GC just happened, and recount buckets.
MProf_GC();
MProf_GC();
n = 0;
for(b=mbuckets; b; b=b->allnext)
if(include_inuse_zero || b->data.mp.alloc_bytes != b->data.mp.free_bytes)
n++;
}
ok = false;
if(n <= p.len) {
ok = true;
r = (Record*)p.array;
for(b=mbuckets; b; b=b->allnext)
if(include_inuse_zero || b->data.mp.alloc_bytes != b->data.mp.free_bytes)
record(r++, b);
}
runtime·unlock(&proflock);
} }
void void
...@@ -294,45 +231,15 @@ runtime·iterate_memprof(void (*callback)(Bucket*, uintptr, uintptr*, uintptr, u ...@@ -294,45 +231,15 @@ runtime·iterate_memprof(void (*callback)(Bucket*, uintptr, uintptr*, uintptr, u
{ {
Bucket *b; Bucket *b;
runtime·lock(&proflock); runtime·lock(&runtime·proflock);
for(b=mbuckets; b; b=b->allnext) { for(b=runtime·mbuckets; b; b=b->allnext) {
callback(b, b->nstk, b->stk, b->size, b->data.mp.allocs, b->data.mp.frees); callback(b, b->nstk, b->stk, b->size, b->data.mp.allocs, b->data.mp.frees);
} }
runtime·unlock(&proflock); runtime·unlock(&runtime·proflock);
} }
// Must match BlockProfileRecord in debug.go. // Go interface to profile data. (Declared in debug.go)
typedef struct BRecord BRecord;
struct BRecord {
int64 count;
int64 cycles;
uintptr stk[32];
};
func BlockProfile(p Slice) (n int, ok bool) {
Bucket *b;
BRecord *r;
int32 i;
runtime·lock(&proflock);
n = 0;
for(b=bbuckets; b; b=b->allnext)
n++;
ok = false;
if(n <= p.len) {
ok = true;
r = (BRecord*)p.array;
for(b=bbuckets; b; b=b->allnext, r++) {
r->count = b->data.bp.count;
r->cycles = b->data.bp.cycles;
for(i=0; i<b->nstk && i<nelem(r->stk); i++)
r->stk[i] = b->stk[i];
for(; i<nelem(r->stk); i++)
r->stk[i] = 0;
}
}
runtime·unlock(&proflock);
}
// Must match StackRecord in debug.go. // Must match StackRecord in debug.go.
typedef struct TRecord TRecord; typedef struct TRecord TRecord;
...@@ -340,25 +247,6 @@ struct TRecord { ...@@ -340,25 +247,6 @@ struct TRecord {
uintptr stk[32]; uintptr stk[32];
}; };
func ThreadCreateProfile(p Slice) (n int, ok bool) {
TRecord *r;
M *first, *mp;
first = runtime·atomicloadp(&runtime·allm);
n = 0;
for(mp=first; mp; mp=mp->alllink)
n++;
ok = false;
if(n <= p.len) {
ok = true;
r = (TRecord*)p.array;
for(mp=first; mp; mp=mp->alllink) {
runtime·memmove(r->stk, mp->createstack, sizeof r->stk);
r++;
}
}
}
func Stack(b Slice, all bool) (n int) { func Stack(b Slice, all bool) (n int) {
uintptr pc, sp; uintptr pc, sp;
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment