The unbounded list-based defer pool can grow infinitely.
This can happen if a goroutine routinely allocates a defer;
then blocks on one P; and then unblocked, scheduled and
frees the defer on another P.
The scenario was reported on golang-nuts list.
We've been here several times. Any unbounded local caches
are bad and grow to infinite size. This change introduces
central defer pool; local pools become fixed-size
with the only purpose of amortizing accesses to the
central pool.
Change-Id: Iadcfb113ccecf912e1b64afc07926f0de9de2248
Reviewed-on: https://go-review.googlesource.com/3741
Reviewed-by: Keith Randall <khr@golang.org>
poolcleanup()
}
+ // Clear central defer pools.
+ // Leave per-P pools alone, they have strictly bounded size.
+ lock(&sched.deferlock)
+ for i := range sched.deferpool {
+ // disconnect cached list before dropping it on the floor,
+ // so that a dangling ref to one entry does not pin all of them.
+ var d, dlink *_defer
+ for d = sched.deferpool[i]; d != nil; d = dlink {
+ dlink = d.link
+ d.link = nil
+ }
+ sched.deferpool[i] = nil
+ }
+ unlock(&sched.deferlock)
+
for _, p := range &allp {
if p == nil {
break
}
c.sudogcache = nil
}
-
- // clear defer pools
- for i := range p.deferpool {
- // disconnect cached list before dropping it on the floor,
- // so that a dangling ref to one entry does not pin all of them.
- var d, dlink *_defer
- for d = p.deferpool[i]; d != nil; d = dlink {
- dlink = d.link
- d.link = nil
- }
- p.deferpool[i] = nil
- }
}
}
mp := acquirem()
if sc < uintptr(len(p{}.deferpool)) {
pp := mp.p
- d = pp.deferpool[sc]
- if d != nil {
- pp.deferpool[sc] = d.link
+ if len(pp.deferpool[sc]) == 0 {
+ lock(&sched.deferlock)
+ for len(pp.deferpool[sc]) < cap(pp.deferpool[sc])/2 && sched.deferpool[sc] != nil {
+ d := sched.deferpool[sc]
+ sched.deferpool[sc] = d.link
+ d.link = nil
+ pp.deferpool[sc] = append(pp.deferpool[sc], d)
+ }
+ unlock(&sched.deferlock)
+ }
+ if ln := len(pp.deferpool[sc]); ln > 0 {
+ d = pp.deferpool[sc][ln-1]
+ pp.deferpool[sc][ln-1] = nil
+ pp.deferpool[sc] = pp.deferpool[sc][:ln-1]
}
}
if d == nil {
if sc < uintptr(len(p{}.deferpool)) {
mp := acquirem()
pp := mp.p
+ if len(pp.deferpool[sc]) == cap(pp.deferpool[sc]) {
+ // Transfer half of local cache to the central cache.
+ var first, last *_defer
+ for len(pp.deferpool[sc]) > cap(pp.deferpool[sc])/2 {
+ ln := len(pp.deferpool[sc])
+ d := pp.deferpool[sc][ln-1]
+ pp.deferpool[sc][ln-1] = nil
+ pp.deferpool[sc] = pp.deferpool[sc][:ln-1]
+ if first == nil {
+ first = d
+ } else {
+ last.link = d
+ }
+ last = d
+ }
+ lock(&sched.deferlock)
+ last.link = sched.deferpool[sc]
+ sched.deferpool[sc] = first
+ unlock(&sched.deferlock)
+ }
*d = _defer{}
- d.link = pp.deferpool[sc]
- pp.deferpool[sc] = d
+ pp.deferpool[sc] = append(pp.deferpool[sc], d)
releasem(mp)
}
}
p = newP()
p.id = i
p.status = _Pgcstop
+ for i := range p.deferpool {
+ p.deferpool[i] = p.deferpoolbuf[i][:0]
+ }
atomicstorep(unsafe.Pointer(&allp[i]), unsafe.Pointer(p))
}
if p.mcache == nil {
}
sched.runqsize++
}
+ for i := range p.deferpool {
+ for j := range p.deferpoolbuf[i] {
+ p.deferpoolbuf[i][j] = nil
+ }
+ p.deferpool[i] = p.deferpoolbuf[i][:0]
+ }
+
freemcache(p.mcache)
p.mcache = nil
gfpurge(p)
syscalltick uint32 // incremented on every system call
m *m // back-link to associated m (nil if idle)
mcache *mcache
- deferpool [5]*_defer // pool of available defer structs of different sizes (see panic.c)
+
+ deferpool [5][]*_defer // pool of available defer structs of different sizes (see panic.go)
+ deferpoolbuf [5][32]*_defer
// Cache of goroutine ids, amortizes accesses to runtime·sched.goidgen.
goidcache uint64
gfree *g
ngfree int32
+ deferlock mutex
+ deferpool [5]*_defer // central pool of available defer structs of different sizes
+
gcwaiting uint32 // gc is waiting to run
stopwait int32
stopnote note