1
0
mirror of https://github.com/golang/go synced 2024-10-01 11:38:34 -06:00

runtime: simplify NetBSD semaphores

NetBSD's semaphore implementation is derived from OpenBSD's, but has
subsequently diverged due to cleanups that were only applied to the
latter (https://golang.org/cl/137960043, https://golang.org/cl/5563).
This CL applies analogous cleanups for NetBSD.

Notably, we can also remove the scary NetBSD deadlock warning.
NetBSD's manual pages document that lwp_unpark on a not-yet-parked LWP
will cause that LWP's next lwp_park system call to return immediately,
so there's no race hazard.

Change-Id: Ib06844c420d2496ac289748eba13eb4700bbbbb2
Reviewed-on: https://go-review.googlesource.com/5564
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Joel Sing <jsing@google.com>
This commit is contained in:
Matthew Dempsky 2015-02-23 17:05:30 +09:00 committed by Joel Sing
parent c123a80063
commit 2fdb728d01

View File

@ -7,8 +7,8 @@ package runtime
import "unsafe"
const (
_ESRCH = 3
_ENOTSUP = 91
_ESRCH = 3
_ETIMEDOUT = 60
// From NetBSD's <sys/time.h>
_CLOCK_REALTIME = 0
@ -46,91 +46,40 @@ func semacreate() uintptr {
func semasleep(ns int64) int32 {
_g_ := getg()
// spin-mutex lock
for {
if xchg(&_g_.m.waitsemalock, 1) == 0 {
break
}
osyield()
// Compute sleep deadline.
var tsp *timespec
if ns >= 0 {
var ts timespec
var nsec int32
ns += nanotime()
ts.set_sec(timediv(ns, 1000000000, &nsec))
ts.set_nsec(nsec)
tsp = &ts
}
for {
// lock held
if _g_.m.waitsemacount == 0 {
// sleep until semaphore != 0 or timeout.
// thrsleep unlocks m.waitsemalock.
if ns < 0 {
// TODO(jsing) - potential deadlock!
//
// There is a potential deadlock here since we
// have to release the waitsemalock mutex
// before we call lwp_park() to suspend the
// thread. This allows another thread to
// release the lock and call lwp_unpark()
// before the thread is actually suspended.
// If this occurs the current thread will end
// up sleeping indefinitely. Unfortunately
// the NetBSD kernel does not appear to provide
// a mechanism for unlocking the userspace
// mutex once the thread is actually parked.
atomicstore(&_g_.m.waitsemalock, 0)
lwp_park(nil, 0, unsafe.Pointer(&_g_.m.waitsemacount), nil)
} else {
var ts timespec
var nsec int32
ns += nanotime()
ts.set_sec(timediv(ns, 1000000000, &nsec))
ts.set_nsec(nsec)
// TODO(jsing) - potential deadlock!
// See above for details.
atomicstore(&_g_.m.waitsemalock, 0)
lwp_park(&ts, 0, unsafe.Pointer(&_g_.m.waitsemacount), nil)
}
// reacquire lock
for {
if xchg(&_g_.m.waitsemalock, 1) == 0 {
break
}
osyield()
v := atomicload(&_g_.m.waitsemacount)
if v > 0 {
if cas(&_g_.m.waitsemacount, v, v-1) {
return 0 // semaphore acquired
}
continue
}
// lock held (again)
if _g_.m.waitsemacount != 0 {
// semaphore is available.
_g_.m.waitsemacount--
// spin-mutex unlock
atomicstore(&_g_.m.waitsemalock, 0)
return 0
}
// semaphore not available.
// if there is a timeout, stop now.
// otherwise keep trying.
if ns >= 0 {
break
// Sleep until unparked by semawakeup or timeout.
ret := lwp_park(tsp, 0, unsafe.Pointer(&_g_.m.waitsemacount), nil)
if ret == _ETIMEDOUT {
return -1
}
}
// lock held but giving up
// spin-mutex unlock
atomicstore(&_g_.m.waitsemalock, 0)
return -1
}
//go:nosplit
func semawakeup(mp *m) {
// spin-mutex lock
for {
if xchg(&mp.waitsemalock, 1) == 0 {
break
}
osyield()
}
mp.waitsemacount++
// TODO(jsing) - potential deadlock, see semasleep() for details.
// Confirm that LWP is parked before unparking...
xadd(&mp.waitsemacount, 1)
// From NetBSD's _lwp_unpark(2) manual:
// "If the target LWP is not currently waiting, it will return
// immediately upon the next call to _lwp_park()."
ret := lwp_unpark(int32(mp.procid), unsafe.Pointer(&mp.waitsemacount))
if ret != 0 && ret != _ESRCH {
// semawakeup can be called on signal stack.
@ -138,9 +87,6 @@ func semawakeup(mp *m) {
print("thrwakeup addr=", &mp.waitsemacount, " sem=", mp.waitsemacount, " ret=", ret, "\n")
})
}
// spin-mutex unlock
atomicstore(&mp.waitsemalock, 0)
}
func newosproc(mp *m, stk unsafe.Pointer) {