1
0
mirror of https://github.com/golang/go synced 2024-11-19 17:14:44 -07:00
go/src/runtime/mgcwork.go

407 lines
11 KiB
Go
Raw Normal View History

// 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"
const (
_Debugwbufs = false // if true check wbufs consistency
_WorkbufSize = 4096 // in bytes; larger values result in less contention
)
// Garbage collector work pool abstraction.
//
// This implements a producer/consumer model for pointers to grey
// objects. A grey object is one that is marked and on a work
// queue. A black object is marked and not on a work queue.
//
// Write barriers, root discovery, stack scanning, and object scanning
// produce pointers to grey objects. Scanning consumes pointers to
// grey objects, thus blackening them, and then scans them,
// potentially producing new pointers to grey objects.
// A wbufptr holds a workbuf*, but protects it from write barriers.
// workbufs never live on the heap, so write barriers are unnecessary.
// Write barriers on workbuf pointers may also be dangerous in the GC.
type wbufptr uintptr
func wbufptrOf(w *workbuf) wbufptr {
return wbufptr(unsafe.Pointer(w))
}
func (wp wbufptr) ptr() *workbuf {
return (*workbuf)(unsafe.Pointer(wp))
}
// A gcWork provides the interface to produce and consume work for the
// garbage collector.
//
runtime: replace per-M workbuf cache with per-P gcWork cache Currently, each M has a cache of the most recently used *workbuf. This is used primarily by the write barrier so it doesn't have to access the global workbuf lists on every write barrier. It's also used by stack scanning because it's convenient. This cache is important for write barrier performance, but this particular approach has several downsides. It's faster than no cache, but far from optimal (as the benchmarks below show). It's complex: access to the cache is sprinkled through most of the workbuf list operations and it requires special care to transform into and back out of the gcWork cache that's actually used for scanning and marking. It requires atomic exchanges to take ownership of the cached workbuf and to return it to the M's cache even though it's almost always used by only the current M. Since it's per-M, flushing these caches is O(# of Ms), which may be high. And it has some significant subtleties: for example, in general the cache shouldn't be used after the harvestwbufs() in mark termination because it could hide work from mark termination, but stack scanning can happen after this and *will* use the cache (but it turns out this is okay because it will always be followed by a getfull(), which drains the cache). This change replaces this cache with a per-P gcWork object. This gcWork cache can be used directly by scanning and marking (as long as preemption is disabled, which is a general requirement of gcWork). Since it's per-P, it doesn't require synchronization, which simplifies things and means the only atomic operations in the write barrier are occasionally fetching new work buffers and setting a mark bit if the object isn't already marked. This cache can be flushed in O(# of Ps), which is generally small. It follows a simple flushing rule: the cache can be used during any phase, but during mark termination it must be flushed before allowing preemption. This also makes the dispose during mutator assist no longer necessary, which eliminates the vast majority of gcWork dispose calls and reduces contention on the global workbuf lists. And it's a lot faster on some benchmarks: benchmark old ns/op new ns/op delta BenchmarkBinaryTree17 11963668673 11206112763 -6.33% BenchmarkFannkuch11 2643217136 2649182499 +0.23% BenchmarkFmtFprintfEmpty 70.4 70.2 -0.28% BenchmarkFmtFprintfString 364 307 -15.66% BenchmarkFmtFprintfInt 317 282 -11.04% BenchmarkFmtFprintfIntInt 512 483 -5.66% BenchmarkFmtFprintfPrefixedInt 404 380 -5.94% BenchmarkFmtFprintfFloat 521 479 -8.06% BenchmarkFmtManyArgs 2164 1894 -12.48% BenchmarkGobDecode 30366146 22429593 -26.14% BenchmarkGobEncode 29867472 26663152 -10.73% BenchmarkGzip 391236616 396779490 +1.42% BenchmarkGunzip 96639491 96297024 -0.35% BenchmarkHTTPClientServer 100110 70763 -29.31% BenchmarkJSONEncode 51866051 52511382 +1.24% BenchmarkJSONDecode 103813138 86094963 -17.07% BenchmarkMandelbrot200 4121834 4120886 -0.02% BenchmarkGoParse 16472789 5879949 -64.31% BenchmarkRegexpMatchEasy0_32 140 140 +0.00% BenchmarkRegexpMatchEasy0_1K 394 394 +0.00% BenchmarkRegexpMatchEasy1_32 120 120 +0.00% BenchmarkRegexpMatchEasy1_1K 621 614 -1.13% BenchmarkRegexpMatchMedium_32 209 202 -3.35% BenchmarkRegexpMatchMedium_1K 54889 55175 +0.52% BenchmarkRegexpMatchHard_32 2682 2675 -0.26% BenchmarkRegexpMatchHard_1K 79383 79524 +0.18% BenchmarkRevcomp 584116718 584595320 +0.08% BenchmarkTemplate 125400565 109620196 -12.58% BenchmarkTimeParse 386 387 +0.26% BenchmarkTimeFormat 580 447 -22.93% (Best out of 10 runs. The delta of averages is similar.) This also puts us in a good position to flush these caches when nearing the end of concurrent marking, which will let us increase the size of the work buffers while still controlling mark termination pause time. Change-Id: I2dd94c8517a19297a98ec280203cccaa58792522 Reviewed-on: https://go-review.googlesource.com/9178 Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-04-19 13:22:20 -06:00
// A gcWork can be used on the stack as follows:
//
// var gcw gcWork
// disable preemption
// .. call gcw.put() to produce and gcw.get() to consume ..
// gcw.dispose()
// enable preemption
//
runtime: replace per-M workbuf cache with per-P gcWork cache Currently, each M has a cache of the most recently used *workbuf. This is used primarily by the write barrier so it doesn't have to access the global workbuf lists on every write barrier. It's also used by stack scanning because it's convenient. This cache is important for write barrier performance, but this particular approach has several downsides. It's faster than no cache, but far from optimal (as the benchmarks below show). It's complex: access to the cache is sprinkled through most of the workbuf list operations and it requires special care to transform into and back out of the gcWork cache that's actually used for scanning and marking. It requires atomic exchanges to take ownership of the cached workbuf and to return it to the M's cache even though it's almost always used by only the current M. Since it's per-M, flushing these caches is O(# of Ms), which may be high. And it has some significant subtleties: for example, in general the cache shouldn't be used after the harvestwbufs() in mark termination because it could hide work from mark termination, but stack scanning can happen after this and *will* use the cache (but it turns out this is okay because it will always be followed by a getfull(), which drains the cache). This change replaces this cache with a per-P gcWork object. This gcWork cache can be used directly by scanning and marking (as long as preemption is disabled, which is a general requirement of gcWork). Since it's per-P, it doesn't require synchronization, which simplifies things and means the only atomic operations in the write barrier are occasionally fetching new work buffers and setting a mark bit if the object isn't already marked. This cache can be flushed in O(# of Ps), which is generally small. It follows a simple flushing rule: the cache can be used during any phase, but during mark termination it must be flushed before allowing preemption. This also makes the dispose during mutator assist no longer necessary, which eliminates the vast majority of gcWork dispose calls and reduces contention on the global workbuf lists. And it's a lot faster on some benchmarks: benchmark old ns/op new ns/op delta BenchmarkBinaryTree17 11963668673 11206112763 -6.33% BenchmarkFannkuch11 2643217136 2649182499 +0.23% BenchmarkFmtFprintfEmpty 70.4 70.2 -0.28% BenchmarkFmtFprintfString 364 307 -15.66% BenchmarkFmtFprintfInt 317 282 -11.04% BenchmarkFmtFprintfIntInt 512 483 -5.66% BenchmarkFmtFprintfPrefixedInt 404 380 -5.94% BenchmarkFmtFprintfFloat 521 479 -8.06% BenchmarkFmtManyArgs 2164 1894 -12.48% BenchmarkGobDecode 30366146 22429593 -26.14% BenchmarkGobEncode 29867472 26663152 -10.73% BenchmarkGzip 391236616 396779490 +1.42% BenchmarkGunzip 96639491 96297024 -0.35% BenchmarkHTTPClientServer 100110 70763 -29.31% BenchmarkJSONEncode 51866051 52511382 +1.24% BenchmarkJSONDecode 103813138 86094963 -17.07% BenchmarkMandelbrot200 4121834 4120886 -0.02% BenchmarkGoParse 16472789 5879949 -64.31% BenchmarkRegexpMatchEasy0_32 140 140 +0.00% BenchmarkRegexpMatchEasy0_1K 394 394 +0.00% BenchmarkRegexpMatchEasy1_32 120 120 +0.00% BenchmarkRegexpMatchEasy1_1K 621 614 -1.13% BenchmarkRegexpMatchMedium_32 209 202 -3.35% BenchmarkRegexpMatchMedium_1K 54889 55175 +0.52% BenchmarkRegexpMatchHard_32 2682 2675 -0.26% BenchmarkRegexpMatchHard_1K 79383 79524 +0.18% BenchmarkRevcomp 584116718 584595320 +0.08% BenchmarkTemplate 125400565 109620196 -12.58% BenchmarkTimeParse 386 387 +0.26% BenchmarkTimeFormat 580 447 -22.93% (Best out of 10 runs. The delta of averages is similar.) This also puts us in a good position to flush these caches when nearing the end of concurrent marking, which will let us increase the size of the work buffers while still controlling mark termination pause time. Change-Id: I2dd94c8517a19297a98ec280203cccaa58792522 Reviewed-on: https://go-review.googlesource.com/9178 Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-04-19 13:22:20 -06:00
// Or from the per-P gcWork cache:
//
// (preemption must be disabled)
// gcw := &getg().m.p.ptr().gcw
// .. call gcw.put() to produce and gcw.get() to consume ..
// if gcBlackenPromptly {
runtime: replace per-M workbuf cache with per-P gcWork cache Currently, each M has a cache of the most recently used *workbuf. This is used primarily by the write barrier so it doesn't have to access the global workbuf lists on every write barrier. It's also used by stack scanning because it's convenient. This cache is important for write barrier performance, but this particular approach has several downsides. It's faster than no cache, but far from optimal (as the benchmarks below show). It's complex: access to the cache is sprinkled through most of the workbuf list operations and it requires special care to transform into and back out of the gcWork cache that's actually used for scanning and marking. It requires atomic exchanges to take ownership of the cached workbuf and to return it to the M's cache even though it's almost always used by only the current M. Since it's per-M, flushing these caches is O(# of Ms), which may be high. And it has some significant subtleties: for example, in general the cache shouldn't be used after the harvestwbufs() in mark termination because it could hide work from mark termination, but stack scanning can happen after this and *will* use the cache (but it turns out this is okay because it will always be followed by a getfull(), which drains the cache). This change replaces this cache with a per-P gcWork object. This gcWork cache can be used directly by scanning and marking (as long as preemption is disabled, which is a general requirement of gcWork). Since it's per-P, it doesn't require synchronization, which simplifies things and means the only atomic operations in the write barrier are occasionally fetching new work buffers and setting a mark bit if the object isn't already marked. This cache can be flushed in O(# of Ps), which is generally small. It follows a simple flushing rule: the cache can be used during any phase, but during mark termination it must be flushed before allowing preemption. This also makes the dispose during mutator assist no longer necessary, which eliminates the vast majority of gcWork dispose calls and reduces contention on the global workbuf lists. And it's a lot faster on some benchmarks: benchmark old ns/op new ns/op delta BenchmarkBinaryTree17 11963668673 11206112763 -6.33% BenchmarkFannkuch11 2643217136 2649182499 +0.23% BenchmarkFmtFprintfEmpty 70.4 70.2 -0.28% BenchmarkFmtFprintfString 364 307 -15.66% BenchmarkFmtFprintfInt 317 282 -11.04% BenchmarkFmtFprintfIntInt 512 483 -5.66% BenchmarkFmtFprintfPrefixedInt 404 380 -5.94% BenchmarkFmtFprintfFloat 521 479 -8.06% BenchmarkFmtManyArgs 2164 1894 -12.48% BenchmarkGobDecode 30366146 22429593 -26.14% BenchmarkGobEncode 29867472 26663152 -10.73% BenchmarkGzip 391236616 396779490 +1.42% BenchmarkGunzip 96639491 96297024 -0.35% BenchmarkHTTPClientServer 100110 70763 -29.31% BenchmarkJSONEncode 51866051 52511382 +1.24% BenchmarkJSONDecode 103813138 86094963 -17.07% BenchmarkMandelbrot200 4121834 4120886 -0.02% BenchmarkGoParse 16472789 5879949 -64.31% BenchmarkRegexpMatchEasy0_32 140 140 +0.00% BenchmarkRegexpMatchEasy0_1K 394 394 +0.00% BenchmarkRegexpMatchEasy1_32 120 120 +0.00% BenchmarkRegexpMatchEasy1_1K 621 614 -1.13% BenchmarkRegexpMatchMedium_32 209 202 -3.35% BenchmarkRegexpMatchMedium_1K 54889 55175 +0.52% BenchmarkRegexpMatchHard_32 2682 2675 -0.26% BenchmarkRegexpMatchHard_1K 79383 79524 +0.18% BenchmarkRevcomp 584116718 584595320 +0.08% BenchmarkTemplate 125400565 109620196 -12.58% BenchmarkTimeParse 386 387 +0.26% BenchmarkTimeFormat 580 447 -22.93% (Best out of 10 runs. The delta of averages is similar.) This also puts us in a good position to flush these caches when nearing the end of concurrent marking, which will let us increase the size of the work buffers while still controlling mark termination pause time. Change-Id: I2dd94c8517a19297a98ec280203cccaa58792522 Reviewed-on: https://go-review.googlesource.com/9178 Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-04-19 13:22:20 -06:00
// gcw.dispose()
// }
//
// It's important that any use of gcWork during the mark phase prevent
// the garbage collector from transitioning to mark termination since
// gcWork may locally hold GC work buffers. This can be done by
// disabling preemption (systemstack or acquirem).
type gcWork struct {
// Invariant: wbuf is never full or empty
wbuf wbufptr
// Bytes marked (blackened) on this gcWork. This is aggregated
// into work.bytesMarked by dispose.
bytesMarked uint64
// Scan work performed on this gcWork. This is aggregated into
// gcController by dispose and may also be flushed by callers.
scanWork int64
}
// put enqueues a pointer for the garbage collector to trace.
runtime: eliminate one heapBitsForObject from scanobject scanobject with ptrmask!=nil is only ever called with the base pointer of a heap object. Currently, scanobject calls heapBitsForObject, which goes to a great deal of trouble to check that the pointer points into the heap and to find the base of the object it points to, both of which are completely unnecessary in this case. Replace this call to heapBitsForObject with much simpler logic to fetch the span and compute the heap bits. Benchmark results with five runs: name old mean new mean delta BenchmarkBinaryTree17 9.21s × (0.95,1.02) 8.55s × (0.91,1.03) -7.16% (p=0.022) BenchmarkFannkuch11 2.65s × (1.00,1.00) 2.62s × (1.00,1.00) -1.10% (p=0.000) BenchmarkFmtFprintfEmpty 73.2ns × (0.99,1.01) 71.7ns × (1.00,1.01) -1.99% (p=0.004) BenchmarkFmtFprintfString 302ns × (0.99,1.00) 292ns × (0.98,1.02) -3.31% (p=0.020) BenchmarkFmtFprintfInt 281ns × (0.98,1.01) 279ns × (0.96,1.02) ~ (p=0.596) BenchmarkFmtFprintfIntInt 482ns × (0.98,1.01) 488ns × (0.95,1.02) ~ (p=0.419) BenchmarkFmtFprintfPrefixedInt 382ns × (0.99,1.01) 365ns × (0.96,1.02) -4.35% (p=0.015) BenchmarkFmtFprintfFloat 475ns × (0.99,1.01) 472ns × (1.00,1.00) ~ (p=0.108) BenchmarkFmtManyArgs 1.89µs × (1.00,1.01) 1.90µs × (0.94,1.02) ~ (p=0.883) BenchmarkGobDecode 22.4ms × (0.99,1.01) 21.9ms × (0.92,1.04) ~ (p=0.332) BenchmarkGobEncode 24.7ms × (0.98,1.02) 23.9ms × (0.87,1.07) ~ (p=0.407) BenchmarkGzip 397ms × (0.99,1.01) 398ms × (0.99,1.01) ~ (p=0.718) BenchmarkGunzip 96.7ms × (1.00,1.00) 96.9ms × (1.00,1.00) ~ (p=0.230) BenchmarkHTTPClientServer 71.5µs × (0.98,1.01) 68.5µs × (0.92,1.06) ~ (p=0.243) BenchmarkJSONEncode 46.1ms × (0.98,1.01) 44.9ms × (0.98,1.03) -2.51% (p=0.040) BenchmarkJSONDecode 86.1ms × (0.99,1.01) 86.5ms × (0.99,1.01) ~ (p=0.343) BenchmarkMandelbrot200 4.12ms × (1.00,1.00) 4.13ms × (1.00,1.00) +0.23% (p=0.000) BenchmarkGoParse 5.89ms × (0.96,1.03) 5.82ms × (0.96,1.04) ~ (p=0.522) BenchmarkRegexpMatchEasy0_32 141ns × (0.99,1.01) 142ns × (1.00,1.00) ~ (p=0.178) BenchmarkRegexpMatchEasy0_1K 408ns × (1.00,1.00) 392ns × (0.99,1.00) -3.83% (p=0.000) BenchmarkRegexpMatchEasy1_32 122ns × (1.00,1.00) 122ns × (1.00,1.00) ~ (p=0.178) BenchmarkRegexpMatchEasy1_1K 626ns × (1.00,1.01) 624ns × (0.99,1.00) ~ (p=0.122) BenchmarkRegexpMatchMedium_32 202ns × (0.99,1.00) 205ns × (0.99,1.01) +1.58% (p=0.001) BenchmarkRegexpMatchMedium_1K 54.4µs × (1.00,1.00) 55.5µs × (1.00,1.00) +1.86% (p=0.000) BenchmarkRegexpMatchHard_32 2.68µs × (1.00,1.00) 2.71µs × (1.00,1.00) +0.97% (p=0.002) BenchmarkRegexpMatchHard_1K 79.8µs × (1.00,1.01) 80.5µs × (1.00,1.01) +0.94% (p=0.003) BenchmarkRevcomp 590ms × (0.99,1.01) 585ms × (1.00,1.00) ~ (p=0.066) BenchmarkTemplate 111ms × (0.97,1.02) 112ms × (0.99,1.01) ~ (p=0.201) BenchmarkTimeParse 392ns × (1.00,1.00) 385ns × (1.00,1.00) -1.69% (p=0.000) BenchmarkTimeFormat 449ns × (0.98,1.01) 448ns × (0.99,1.01) ~ (p=0.550) Change-Id: Ie7c3830c481d96c9043e7bf26853c6c1d05dc9f4 Reviewed-on: https://go-review.googlesource.com/9364 Reviewed-by: Rick Hudson <rlh@golang.org>
2015-04-26 16:27:17 -06:00
// obj must point to the beginning of a heap object.
//go:nowritebarrier
func (ww *gcWork) put(obj uintptr) {
w := (*gcWork)(noescape(unsafe.Pointer(ww))) // TODO: remove when escape analysis is fixed
wbuf := w.wbuf.ptr()
if wbuf == nil {
wbuf = getempty(42)
w.wbuf = wbufptrOf(wbuf)
}
wbuf.obj[wbuf.nobj] = obj
wbuf.nobj++
if wbuf.nobj == len(wbuf.obj) {
putfull(wbuf, 50)
w.wbuf = 0
}
}
// tryGet dequeues a pointer for the garbage collector to trace.
//
// If there are no pointers remaining in this gcWork or in the global
// queue, tryGet returns 0. Note that there may still be pointers in
// other gcWork instances or other caches.
//go:nowritebarrier
func (ww *gcWork) tryGet() uintptr {
w := (*gcWork)(noescape(unsafe.Pointer(ww))) // TODO: remove when escape analysis is fixed
wbuf := w.wbuf.ptr()
if wbuf == nil {
wbuf = trygetfull(74)
if wbuf == nil {
return 0
}
w.wbuf = wbufptrOf(wbuf)
}
wbuf.nobj--
obj := wbuf.obj[wbuf.nobj]
if wbuf.nobj == 0 {
putempty(wbuf, 86)
w.wbuf = 0
}
return obj
}
// get dequeues a pointer for the garbage collector to trace, blocking
// if necessary to ensure all pointers from all queues and caches have
// been retrieved. get returns 0 if there are no pointers remaining.
//go:nowritebarrier
func (ww *gcWork) get() uintptr {
w := (*gcWork)(noescape(unsafe.Pointer(ww))) // TODO: remove when escape analysis is fixed
wbuf := w.wbuf.ptr()
if wbuf == nil {
wbuf = getfull(103)
if wbuf == nil {
return 0
}
wbuf.checknonempty()
w.wbuf = wbufptrOf(wbuf)
}
// TODO: This might be a good place to add prefetch code
wbuf.nobj--
obj := wbuf.obj[wbuf.nobj]
if wbuf.nobj == 0 {
putempty(wbuf, 115)
w.wbuf = 0
}
return obj
}
// dispose returns any cached pointers to the global queue.
// The buffers are being put on the full queue so that the
// write barriers will not simply reacquire them before the
// GC can inspect them. This helps reduce the mutator's
// ability to hide pointers during the concurrent mark phase.
//
//go:nowritebarrier
func (w *gcWork) dispose() {
if wbuf := w.wbuf; wbuf != 0 {
if wbuf.ptr().nobj == 0 {
throw("dispose: workbuf is empty")
}
putfull(wbuf.ptr(), 166)
w.wbuf = 0
}
if w.bytesMarked != 0 {
// dispose happens relatively infrequently. If this
// atomic becomes a problem, we should first try to
// dispose less and if necessary aggregate in a per-P
// counter.
xadd64(&work.bytesMarked, int64(w.bytesMarked))
w.bytesMarked = 0
}
if w.scanWork != 0 {
xaddint64(&gcController.scanWork, w.scanWork)
w.scanWork = 0
}
}
// balance moves some work that's cached in this gcWork back on the
// global queue.
//go:nowritebarrier
func (w *gcWork) balance() {
if wbuf := w.wbuf; wbuf != 0 && wbuf.ptr().nobj > 4 {
w.wbuf = wbufptrOf(handoff(wbuf.ptr()))
}
}
// empty returns true if w has no mark work available.
//go:nowritebarrier
func (w *gcWork) empty() bool {
wbuf := w.wbuf
return wbuf == 0 || wbuf.ptr().nobj == 0
}
// Internally, the GC work pool is kept in arrays in work buffers.
// The gcWork interface caches a work buffer until full (or empty) to
// avoid contending on the global work buffer lists.
type workbufhdr struct {
node lfnode // must be first
nobj int
inuse bool // This workbuf is in use by some gorotuine and is not on the work.empty/full queues.
log [4]int // line numbers forming a history of ownership changes to workbuf
}
type workbuf struct {
workbufhdr
// account for the above fields
obj [(_WorkbufSize - unsafe.Sizeof(workbufhdr{})) / ptrSize]uintptr
}
// workbuf factory routines. These funcs are used to manage the
runtime: replace per-M workbuf cache with per-P gcWork cache Currently, each M has a cache of the most recently used *workbuf. This is used primarily by the write barrier so it doesn't have to access the global workbuf lists on every write barrier. It's also used by stack scanning because it's convenient. This cache is important for write barrier performance, but this particular approach has several downsides. It's faster than no cache, but far from optimal (as the benchmarks below show). It's complex: access to the cache is sprinkled through most of the workbuf list operations and it requires special care to transform into and back out of the gcWork cache that's actually used for scanning and marking. It requires atomic exchanges to take ownership of the cached workbuf and to return it to the M's cache even though it's almost always used by only the current M. Since it's per-M, flushing these caches is O(# of Ms), which may be high. And it has some significant subtleties: for example, in general the cache shouldn't be used after the harvestwbufs() in mark termination because it could hide work from mark termination, but stack scanning can happen after this and *will* use the cache (but it turns out this is okay because it will always be followed by a getfull(), which drains the cache). This change replaces this cache with a per-P gcWork object. This gcWork cache can be used directly by scanning and marking (as long as preemption is disabled, which is a general requirement of gcWork). Since it's per-P, it doesn't require synchronization, which simplifies things and means the only atomic operations in the write barrier are occasionally fetching new work buffers and setting a mark bit if the object isn't already marked. This cache can be flushed in O(# of Ps), which is generally small. It follows a simple flushing rule: the cache can be used during any phase, but during mark termination it must be flushed before allowing preemption. This also makes the dispose during mutator assist no longer necessary, which eliminates the vast majority of gcWork dispose calls and reduces contention on the global workbuf lists. And it's a lot faster on some benchmarks: benchmark old ns/op new ns/op delta BenchmarkBinaryTree17 11963668673 11206112763 -6.33% BenchmarkFannkuch11 2643217136 2649182499 +0.23% BenchmarkFmtFprintfEmpty 70.4 70.2 -0.28% BenchmarkFmtFprintfString 364 307 -15.66% BenchmarkFmtFprintfInt 317 282 -11.04% BenchmarkFmtFprintfIntInt 512 483 -5.66% BenchmarkFmtFprintfPrefixedInt 404 380 -5.94% BenchmarkFmtFprintfFloat 521 479 -8.06% BenchmarkFmtManyArgs 2164 1894 -12.48% BenchmarkGobDecode 30366146 22429593 -26.14% BenchmarkGobEncode 29867472 26663152 -10.73% BenchmarkGzip 391236616 396779490 +1.42% BenchmarkGunzip 96639491 96297024 -0.35% BenchmarkHTTPClientServer 100110 70763 -29.31% BenchmarkJSONEncode 51866051 52511382 +1.24% BenchmarkJSONDecode 103813138 86094963 -17.07% BenchmarkMandelbrot200 4121834 4120886 -0.02% BenchmarkGoParse 16472789 5879949 -64.31% BenchmarkRegexpMatchEasy0_32 140 140 +0.00% BenchmarkRegexpMatchEasy0_1K 394 394 +0.00% BenchmarkRegexpMatchEasy1_32 120 120 +0.00% BenchmarkRegexpMatchEasy1_1K 621 614 -1.13% BenchmarkRegexpMatchMedium_32 209 202 -3.35% BenchmarkRegexpMatchMedium_1K 54889 55175 +0.52% BenchmarkRegexpMatchHard_32 2682 2675 -0.26% BenchmarkRegexpMatchHard_1K 79383 79524 +0.18% BenchmarkRevcomp 584116718 584595320 +0.08% BenchmarkTemplate 125400565 109620196 -12.58% BenchmarkTimeParse 386 387 +0.26% BenchmarkTimeFormat 580 447 -22.93% (Best out of 10 runs. The delta of averages is similar.) This also puts us in a good position to flush these caches when nearing the end of concurrent marking, which will let us increase the size of the work buffers while still controlling mark termination pause time. Change-Id: I2dd94c8517a19297a98ec280203cccaa58792522 Reviewed-on: https://go-review.googlesource.com/9178 Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-04-19 13:22:20 -06:00
// workbufs.
// If the GC asks for some work these are the only routines that
// make wbufs available to the GC.
// Each of the gets and puts also take an distinct integer that is used
// to record a brief history of changes to ownership of the workbuf.
// The convention is to use a unique line number but any encoding
// is permissible. For example if you want to pass in 2 bits of information
// you could simple add lineno1*100000+lineno2.
// logget records the past few values of entry to aid in debugging.
// logget checks the buffer b is not currently in use.
func (b *workbuf) logget(entry int) {
if !_Debugwbufs {
return
}
if b.inuse {
println("runtime: logget fails log entry=", entry,
"b.log[0]=", b.log[0], "b.log[1]=", b.log[1],
"b.log[2]=", b.log[2], "b.log[3]=", b.log[3])
throw("logget: get not legal")
}
b.inuse = true
copy(b.log[1:], b.log[:])
b.log[0] = entry
}
// logput records the past few values of entry to aid in debugging.
// logput checks the buffer b is currently in use.
func (b *workbuf) logput(entry int) {
if !_Debugwbufs {
return
}
if !b.inuse {
runtime: reduce latency by aggressively ending mark phase Some latency regressions have crept into our system over the past few weeks. This CL fixes those by having the mark phase more aggressively blacken objects so that the mark termination phase, a STW phase, has less work to do. Three approaches were taken when the mark phase believes it has no more work to do, ie all the work buffers are empty. If things have gone well the mark phase is correct and there is in fact little or no work. In that case the following items will take very little time. If the mark phase is wrong this CL will ferret that work out and give the mark phase a chance to deal with it concurrently before mark termination begins. When the mark phase first appears to be out of work, it does three things: 1) It switches from allocating white to allocating black to reduce the number of unmarked objects reachable only from stacks. 2) It flushes and disables per-P GC work caches so all work must be in globally visible work buffers. 3) It rescans the global roots---the BSS and data segments---so there are fewer objects to blacken during mark termination. We do not rescan stacks at this point, though that could be done in a later CL. After these steps, it again drains the global work buffers. On a lightly loaded machine the garbage benchmark has reduced the number of GC cycles with latency > 10 ms from 83 out of 4083 cycles down to 2 out of 3995 cycles. Maximum latency was reduced from 60+ msecs down to 20 ms. Change-Id: I152285b48a7e56c5083a02e8e4485dd39c990492 Reviewed-on: https://go-review.googlesource.com/10590 Reviewed-by: Austin Clements <austin@google.com>
2015-06-01 16:16:03 -06:00
println("runtime: logput fails log entry=", entry,
"b.log[0]=", b.log[0], "b.log[1]=", b.log[1],
"b.log[2]=", b.log[2], "b.log[3]=", b.log[3])
throw("logput: put not legal")
}
b.inuse = false
copy(b.log[1:], b.log[:])
b.log[0] = entry
}
func (b *workbuf) checknonempty() {
if b.nobj == 0 {
println("runtime: nonempty check fails",
"b.log[0]=", b.log[0], "b.log[1]=", b.log[1],
"b.log[2]=", b.log[2], "b.log[3]=", b.log[3])
throw("workbuf is empty")
}
}
func (b *workbuf) checkempty() {
if b.nobj != 0 {
println("runtime: empty check fails",
"b.log[0]=", b.log[0], "b.log[1]=", b.log[1],
"b.log[2]=", b.log[2], "b.log[3]=", b.log[3])
throw("workbuf is not empty")
}
}
// getempty pops an empty work buffer off the work.empty list,
// allocating new buffers if none are available.
// entry is used to record a brief history of ownership.
//go:nowritebarrier
func getempty(entry int) *workbuf {
var b *workbuf
if work.empty != 0 {
b = (*workbuf)(lfstackpop(&work.empty))
if b != nil {
b.checkempty()
}
}
if b == nil {
b = (*workbuf)(persistentalloc(unsafe.Sizeof(*b), _CacheLineSize, &memstats.gc_sys))
}
b.logget(entry)
return b
}
// putempty puts a workbuf onto the work.empty list.
// Upon entry this go routine owns b. The lfstackpush relinquishes ownership.
//go:nowritebarrier
func putempty(b *workbuf, entry int) {
b.checkempty()
b.logput(entry)
lfstackpush(&work.empty, &b.node)
}
// putfull puts the workbuf on the work.full list for the GC.
// putfull accepts partially full buffers so the GC can avoid competing
// with the mutators for ownership of partially full buffers.
//go:nowritebarrier
func putfull(b *workbuf, entry int) {
b.checknonempty()
b.logput(entry)
lfstackpush(&work.full, &b.node)
}
// trygetfull tries to get a full or partially empty workbuffer.
// If one is not immediately available return nil
//go:nowritebarrier
func trygetfull(entry int) *workbuf {
b := (*workbuf)(lfstackpop(&work.full))
if b != nil {
b.logget(entry)
b.checknonempty()
return b
}
return b
}
// Get a full work buffer off the work.full list.
// If nothing is available wait until all the other gc helpers have
// finished and then return nil.
// getfull acts as a barrier for work.nproc helpers. As long as one
// gchelper is actively marking objects it
// may create a workbuffer that the other helpers can work on.
// The for loop either exits when a work buffer is found
// or when _all_ of the work.nproc GC helpers are in the loop
// looking for work and thus not capable of creating new work.
// This is in fact the termination condition for the STW mark
// phase.
//go:nowritebarrier
func getfull(entry int) *workbuf {
b := (*workbuf)(lfstackpop(&work.full))
if b != nil {
b.logget(entry)
b.checknonempty()
return b
}
runtime: reduce latency by aggressively ending mark phase Some latency regressions have crept into our system over the past few weeks. This CL fixes those by having the mark phase more aggressively blacken objects so that the mark termination phase, a STW phase, has less work to do. Three approaches were taken when the mark phase believes it has no more work to do, ie all the work buffers are empty. If things have gone well the mark phase is correct and there is in fact little or no work. In that case the following items will take very little time. If the mark phase is wrong this CL will ferret that work out and give the mark phase a chance to deal with it concurrently before mark termination begins. When the mark phase first appears to be out of work, it does three things: 1) It switches from allocating white to allocating black to reduce the number of unmarked objects reachable only from stacks. 2) It flushes and disables per-P GC work caches so all work must be in globally visible work buffers. 3) It rescans the global roots---the BSS and data segments---so there are fewer objects to blacken during mark termination. We do not rescan stacks at this point, though that could be done in a later CL. After these steps, it again drains the global work buffers. On a lightly loaded machine the garbage benchmark has reduced the number of GC cycles with latency > 10 ms from 83 out of 4083 cycles down to 2 out of 3995 cycles. Maximum latency was reduced from 60+ msecs down to 20 ms. Change-Id: I152285b48a7e56c5083a02e8e4485dd39c990492 Reviewed-on: https://go-review.googlesource.com/10590 Reviewed-by: Austin Clements <austin@google.com>
2015-06-01 16:16:03 -06:00
incnwait := xadd(&work.nwait, +1)
if incnwait > work.nproc {
println("runtime: work.nwait=", incnwait, "work.nproc=", work.nproc)
throw("work.nwait > work.nproc")
}
for i := 0; ; i++ {
if work.full != 0 {
runtime: reduce latency by aggressively ending mark phase Some latency regressions have crept into our system over the past few weeks. This CL fixes those by having the mark phase more aggressively blacken objects so that the mark termination phase, a STW phase, has less work to do. Three approaches were taken when the mark phase believes it has no more work to do, ie all the work buffers are empty. If things have gone well the mark phase is correct and there is in fact little or no work. In that case the following items will take very little time. If the mark phase is wrong this CL will ferret that work out and give the mark phase a chance to deal with it concurrently before mark termination begins. When the mark phase first appears to be out of work, it does three things: 1) It switches from allocating white to allocating black to reduce the number of unmarked objects reachable only from stacks. 2) It flushes and disables per-P GC work caches so all work must be in globally visible work buffers. 3) It rescans the global roots---the BSS and data segments---so there are fewer objects to blacken during mark termination. We do not rescan stacks at this point, though that could be done in a later CL. After these steps, it again drains the global work buffers. On a lightly loaded machine the garbage benchmark has reduced the number of GC cycles with latency > 10 ms from 83 out of 4083 cycles down to 2 out of 3995 cycles. Maximum latency was reduced from 60+ msecs down to 20 ms. Change-Id: I152285b48a7e56c5083a02e8e4485dd39c990492 Reviewed-on: https://go-review.googlesource.com/10590 Reviewed-by: Austin Clements <austin@google.com>
2015-06-01 16:16:03 -06:00
decnwait := xadd(&work.nwait, -1)
if decnwait == work.nproc {
println("runtime: work.nwait=", decnwait, "work.nproc=", work.nproc)
throw("work.nwait > work.nproc")
}
b = (*workbuf)(lfstackpop(&work.full))
if b != nil {
b.logget(entry)
b.checknonempty()
return b
}
runtime: reduce latency by aggressively ending mark phase Some latency regressions have crept into our system over the past few weeks. This CL fixes those by having the mark phase more aggressively blacken objects so that the mark termination phase, a STW phase, has less work to do. Three approaches were taken when the mark phase believes it has no more work to do, ie all the work buffers are empty. If things have gone well the mark phase is correct and there is in fact little or no work. In that case the following items will take very little time. If the mark phase is wrong this CL will ferret that work out and give the mark phase a chance to deal with it concurrently before mark termination begins. When the mark phase first appears to be out of work, it does three things: 1) It switches from allocating white to allocating black to reduce the number of unmarked objects reachable only from stacks. 2) It flushes and disables per-P GC work caches so all work must be in globally visible work buffers. 3) It rescans the global roots---the BSS and data segments---so there are fewer objects to blacken during mark termination. We do not rescan stacks at this point, though that could be done in a later CL. After these steps, it again drains the global work buffers. On a lightly loaded machine the garbage benchmark has reduced the number of GC cycles with latency > 10 ms from 83 out of 4083 cycles down to 2 out of 3995 cycles. Maximum latency was reduced from 60+ msecs down to 20 ms. Change-Id: I152285b48a7e56c5083a02e8e4485dd39c990492 Reviewed-on: https://go-review.googlesource.com/10590 Reviewed-by: Austin Clements <austin@google.com>
2015-06-01 16:16:03 -06:00
incnwait := xadd(&work.nwait, +1)
if incnwait > work.nproc {
println("runtime: work.nwait=", incnwait, "work.nproc=", work.nproc)
throw("work.nwait > work.nproc")
}
}
runtime: perform concurrent scan in GC workers Currently the concurrent root scan is performed in its entirety by the GC coordinator before entering concurrent mark (which enables GC workers). This scan is done sequentially, which can prolong the scan phase, delay the mark phase, and means that the scan phase does not obey the 25% CPU goal. Furthermore, there's no need to complete the root scan before starting marking (in fact, we already allow GC assists to happen during the scan phase), so this acts as an unnecessary barrier between root scanning and marking. This change shifts the root scan work out of the GC coordinator and in to the GC workers. The coordinator simply sets up the scan state and enqueues the right number of root scan jobs. The GC workers then drain the root scan jobs prior to draining heap scan jobs. This parallelizes the root scan process, makes it obey the 25% CPU goal, and effectively eliminates root scanning as an isolated phase, allowing the system to smoothly transition from root scanning to heap marking. This also eliminates a major non-STW responsibility of the GC coordinator, which will make it easier to switch to a decentralized state machine. Finally, it puts us in a good position to perform root scanning in assists as well, which will help satisfy assists at the beginning of the GC cycle. This is mostly straightforward. One tricky aspect is that we have to deal with preemption deadlock: where two non-preemptible gorountines are trying to preempt each other to perform a stack scan. Given the context where this happens, the only instance of this is two background workers trying to scan each other. We avoid this by simply not scanning the stacks of background workers during the concurrent phase; this is safe because we'll scan them during mark termination (and their stacks are *very* small and should not contain any new pointers). This change also switches the root marking during mark termination to use the same gcDrain-based code path as concurrent mark. This shouldn't affect performance because STW root marking was already parallel and tasks switched to heap marking immediately when no more root marking tasks were available. However, it simplifies the code and unifies these code paths. This has negligible effect on the go1 benchmarks. It slightly slows down the garbage benchmark, possibly by making GC run slightly more frequently. name old time/op new time/op delta XBenchGarbage-12 5.10ms ± 1% 5.24ms ± 1% +2.87% (p=0.000 n=18+18) name old time/op new time/op delta BinaryTree17-12 3.25s ± 3% 3.20s ± 5% -1.57% (p=0.013 n=20+20) Fannkuch11-12 2.45s ± 1% 2.46s ± 1% +0.38% (p=0.019 n=20+18) FmtFprintfEmpty-12 49.7ns ± 3% 49.9ns ± 4% ~ (p=0.851 n=19+20) FmtFprintfString-12 170ns ± 2% 170ns ± 1% ~ (p=0.775 n=20+19) FmtFprintfInt-12 161ns ± 1% 160ns ± 1% -0.78% (p=0.000 n=19+18) FmtFprintfIntInt-12 267ns ± 1% 270ns ± 1% +1.04% (p=0.000 n=19+19) FmtFprintfPrefixedInt-12 238ns ± 2% 238ns ± 1% ~ (p=0.133 n=18+19) FmtFprintfFloat-12 311ns ± 1% 310ns ± 2% -0.35% (p=0.023 n=20+19) FmtManyArgs-12 1.08µs ± 1% 1.06µs ± 1% -2.31% (p=0.000 n=20+20) GobDecode-12 8.65ms ± 1% 8.63ms ± 1% ~ (p=0.377 n=18+20) GobEncode-12 6.49ms ± 1% 6.52ms ± 1% +0.37% (p=0.015 n=20+20) Gzip-12 319ms ± 3% 318ms ± 1% ~ (p=0.975 n=19+17) Gunzip-12 41.9ms ± 1% 42.1ms ± 2% +0.65% (p=0.004 n=19+20) HTTPClientServer-12 61.7µs ± 1% 62.6µs ± 1% +1.40% (p=0.000 n=18+20) JSONEncode-12 16.8ms ± 1% 16.9ms ± 1% ~ (p=0.239 n=20+18) JSONDecode-12 58.4ms ± 1% 60.7ms ± 1% +3.85% (p=0.000 n=19+20) Mandelbrot200-12 3.86ms ± 0% 3.86ms ± 1% ~ (p=0.092 n=18+19) GoParse-12 3.75ms ± 2% 3.75ms ± 2% ~ (p=0.708 n=19+20) RegexpMatchEasy0_32-12 100ns ± 1% 100ns ± 2% +0.60% (p=0.010 n=17+20) RegexpMatchEasy0_1K-12 341ns ± 1% 342ns ± 2% ~ (p=0.203 n=20+19) RegexpMatchEasy1_32-12 82.5ns ± 2% 83.2ns ± 2% +0.83% (p=0.007 n=19+19) RegexpMatchEasy1_1K-12 495ns ± 1% 495ns ± 2% ~ (p=0.970 n=19+18) RegexpMatchMedium_32-12 130ns ± 2% 130ns ± 2% +0.59% (p=0.039 n=19+20) RegexpMatchMedium_1K-12 39.2µs ± 1% 39.3µs ± 1% ~ (p=0.214 n=18+18) RegexpMatchHard_32-12 2.03µs ± 2% 2.02µs ± 1% ~ (p=0.166 n=18+19) RegexpMatchHard_1K-12 61.0µs ± 1% 60.9µs ± 1% ~ (p=0.169 n=20+18) Revcomp-12 533ms ± 1% 535ms ± 1% ~ (p=0.071 n=19+17) Template-12 68.1ms ± 2% 73.0ms ± 1% +7.26% (p=0.000 n=19+20) TimeParse-12 355ns ± 2% 356ns ± 2% ~ (p=0.530 n=19+20) TimeFormat-12 357ns ± 2% 347ns ± 1% -2.59% (p=0.000 n=20+19) [Geo mean] 62.1µs 62.3µs +0.31% name old speed new speed delta GobDecode-12 88.7MB/s ± 1% 88.9MB/s ± 1% ~ (p=0.377 n=18+20) GobEncode-12 118MB/s ± 1% 118MB/s ± 1% -0.37% (p=0.015 n=20+20) Gzip-12 60.9MB/s ± 3% 60.9MB/s ± 1% ~ (p=0.944 n=19+17) Gunzip-12 464MB/s ± 1% 461MB/s ± 2% -0.64% (p=0.004 n=19+20) JSONEncode-12 115MB/s ± 1% 115MB/s ± 1% ~ (p=0.236 n=20+18) JSONDecode-12 33.2MB/s ± 1% 32.0MB/s ± 1% -3.71% (p=0.000 n=19+20) GoParse-12 15.5MB/s ± 2% 15.5MB/s ± 2% ~ (p=0.702 n=19+20) RegexpMatchEasy0_32-12 320MB/s ± 1% 318MB/s ± 2% ~ (p=0.094 n=18+20) RegexpMatchEasy0_1K-12 3.00GB/s ± 1% 2.99GB/s ± 1% ~ (p=0.194 n=20+19) RegexpMatchEasy1_32-12 388MB/s ± 2% 385MB/s ± 2% -0.83% (p=0.008 n=19+19) RegexpMatchEasy1_1K-12 2.07GB/s ± 1% 2.07GB/s ± 1% ~ (p=0.964 n=19+18) RegexpMatchMedium_32-12 7.68MB/s ± 1% 7.64MB/s ± 2% -0.57% (p=0.020 n=19+20) RegexpMatchMedium_1K-12 26.1MB/s ± 1% 26.1MB/s ± 1% ~ (p=0.211 n=18+18) RegexpMatchHard_32-12 15.8MB/s ± 1% 15.8MB/s ± 1% ~ (p=0.180 n=18+19) RegexpMatchHard_1K-12 16.8MB/s ± 1% 16.8MB/s ± 2% ~ (p=0.236 n=20+19) Revcomp-12 477MB/s ± 1% 475MB/s ± 1% ~ (p=0.071 n=19+17) Template-12 28.5MB/s ± 2% 26.6MB/s ± 1% -6.77% (p=0.000 n=19+20) [Geo mean] 100MB/s 99.0MB/s -0.82% Change-Id: I875bf6ceb306d1ee2f470cabf88aa6ede27c47a0 Reviewed-on: https://go-review.googlesource.com/16059 Reviewed-by: Rick Hudson <rlh@golang.org> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-10-19 11:46:32 -06:00
if work.nwait == work.nproc && work.markrootNext >= work.markrootJobs {
return nil
}
_g_ := getg()
if i < 10 {
_g_.m.gcstats.nprocyield++
procyield(20)
} else if i < 20 {
_g_.m.gcstats.nosyield++
osyield()
} else {
_g_.m.gcstats.nsleep++
usleep(100)
}
}
}
//go:nowritebarrier
func handoff(b *workbuf) *workbuf {
// Make new buffer with half of b's pointers.
b1 := getempty(915)
n := b.nobj / 2
b.nobj -= n
b1.nobj = n
memmove(unsafe.Pointer(&b1.obj[0]), unsafe.Pointer(&b.obj[b.nobj]), uintptr(n)*unsafe.Sizeof(b1.obj[0]))
_g_ := getg()
_g_.m.gcstats.nhandoff++
_g_.m.gcstats.nhandoffcnt += uint64(n)
// Put b on full list - let first half of b get stolen.
putfull(b, 942)
return b1
}