The timerpMask optimization updates a mask of Ps (potentially)
containing timers in pidleget / pidleput. For correctness, it depends on
the assumption that new timers can only be added to a P's own heap.
addtimer violates this assumption if it is preempted after computing pp.
That G may then run on a different P, but adding a timer to the original
P's heap.
Avoid this by disabling preemption while pp is in use.
Other uses of doaddtimer should be OK:
* moveTimers: always moves to the current P's heap
* modtimer, cleantimers, addAdjustedTimers, runtimer: does not add net
new timers to the heap while locked
For #44868
Fixes #45731
Change-Id: I4a5d080865e854931d0a3a09a51ca36879101d72
Reviewed-on: https://go-review.googlesource.com/c/go/+/300610
Trust: Michael Pratt <mpratt@google.com>
Run-TryBot: Michael Pratt <mpratt@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-on: https://go-review.googlesource.com/c/go/+/313129
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Michael Pratt <mpratt@google.com>
when := t.when
+ // Disable preemption while using pp to avoid changing another P's heap.
+ mp := acquirem()
+
pp := getg().m.p.ptr()
lock(&pp.timersLock)
cleantimers(pp)
unlock(&pp.timersLock)
wakeNetPoller(when)
+
+ releasem(mp)
}
// doaddtimer adds t to the current P's heap.
var tr Timer
tr.Stop()
}
+
+// Test that zero duration timers aren't missed by the scheduler. Regression test for issue 44868.
+func TestZeroTimer(t *testing.T) {
+ if testing.Short() {
+ t.Skip("-short")
+ }
+
+ for i := 0; i < 1000000; i++ {
+ s := Now()
+ ti := NewTimer(0)
+ <-ti.C
+ if diff := Since(s); diff > 2*Second {
+ t.Errorf("Expected time to get value from Timer channel in less than 2 sec, took %v", diff)
+ }
+ }
+}