This repairs one of the several causes of pauses uncovered
by a GC microbenchmark. A pause can occur when a goroutine's
quantum expires "at the same time" a GC is needed. The
current M switches to running a GC worker, which means that
the amount of available work has expanded by one. The GC
worker, however, does not call ready, and does not itself
conditionally wake a P (a "normal" thread would do this).
This is also true if M switches to a traceReader.
This is problem 4 in this list:
https://github.com/golang/go/issues/27732#issuecomment-
423301252
Updates #27732.
Change-Id: I6905365cac8504cde6faab2420f4421536551f0b
Reviewed-on: https://go-review.googlesource.com/c/go/+/146817
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
var gp *g
var inheritTime bool
+
+ // Normal goroutines will check for need to wakeP in ready,
+ // but GCworkers and tracereaders will not, so the check must
+ // be done here instead.
+ tryWakeP := false
if trace.enabled || trace.shutdown {
gp = traceReader()
if gp != nil {
casgstatus(gp, _Gwaiting, _Grunnable)
traceGoUnpark(gp, 0)
+ tryWakeP = true
}
}
if gp == nil && gcBlackenEnabled != 0 {
gp = gcController.findRunnableGCWorker(_g_.m.p.ptr())
+ tryWakeP = tryWakeP || gp != nil
}
if gp == nil {
// Check the global runnable queue once in a while to ensure fairness.
}
}
+ // If about to schedule a not-normal goroutine (a GCworker or tracereader),
+ // wake a P if there is one.
+ if tryWakeP {
+ if atomic.Load(&sched.npidle) != 0 && atomic.Load(&sched.nmspinning) == 0 {
+ wakep()
+ }
+ }
if gp.lockedm != 0 {
// Hands off own p to the locked m,
// then blocks waiting for a new p.