cmd/link,runtime: switch openbsd/amd64 to pthreads

This switches openbsd/amd64 to thread creation via pthreads, rather than doing
direct system calls.

Update #36435

Change-Id: I1105d5c392aa3e4c445d99c8cb80b927712e3529
Reviewed-on: https://go-review.googlesource.com/c/go/+/250180
Trust: Joel Sing <[email protected]>
Run-TryBot: Joel Sing <[email protected]>
TryBot-Result: Go Bot <[email protected]>
Reviewed-by: Cherry Zhang <[email protected]>

Author: 4a6f656c

src/cmd/link/internal/ld/data.go

@@ -206,8 +206,8 @@ func (st *relocSymState) relocsym(s loader.Sym, P []byte) {
 		}
 
 		// We need to be able to reference dynimport symbols when linking against
-		// shared libraries, and Solaris, Darwin and AIX need it always
-		if !target.IsSolaris() && !target.IsDarwin() && !target.IsAIX() && rs != 0 && rst == sym.SDYNIMPORT && !target.IsDynlinkingGo() && !ldr.AttrSubSymbol(rs) {
+		// shared libraries, and AIX, Darwin, OpenBSD and Solaris always need it.
+		if !target.IsAIX() && !target.IsDarwin() && !target.IsSolaris() && !target.IsOpenbsd() && rs != 0 && rst == sym.SDYNIMPORT && !target.IsDynlinkingGo() && !ldr.AttrSubSymbol(rs) {
 			if !(target.IsPPC64() && target.IsExternal() && ldr.SymName(rs) == ".TOC.") {
 				st.err.Errorf(s, "unhandled relocation for %s (type %d (%s) rtype %d (%s))", ldr.SymName(rs), rst, rst, rt, sym.RelocName(target.Arch, rt))
 			}

src/cmd/link/internal/ld/lib.go

@@ -1273,6 +1273,7 @@ func (ctxt *Link) hostlink() {
 		}
 	case objabi.Hopenbsd:
 		argv = append(argv, "-Wl,-nopie")
+		argv = append(argv, "-pthread")
 	case objabi.Hwindows:
 		if windowsgui {
 			argv = append(argv, "-mwindows")

src/cmd/link/internal/ld/main.go

@@ -183,6 +183,14 @@ func Main(arch *sys.Arch, theArch Arch) {
 
 	interpreter = *flagInterpreter
 
+	if *flagBuildid == "" && ctxt.Target.IsOpenbsd() {
+		// TODO(jsing): Remove once direct syscalls are no longer in use.
+		// OpenBSD 6.7 onwards will not permit direct syscalls from a
+		// dynamically linked binary unless it identifies the binary
+		// contains a .note.go.buildid ELF note. See issue #36435.
+		*flagBuildid = "go-openbsd"
+	}
+
 	// enable benchmarking
 	var bench *benchmark.Metrics
 	if len(*benchmarkFlag) != 0 {

src/runtime/asm_amd64.s

@@ -181,6 +181,10 @@ needtls:
 	// skip TLS setup on Darwin
 	JMP ok
 #endif
+#ifdef GOOS_openbsd
+	// skip TLS setup on OpenBSD
+	JMP ok
+#endif
 
 	LEAQ	runtime·m0+m_tls(SB), DI
 	CALL	runtime·settls(SB)

src/runtime/defs_openbsd.go

@@ -54,6 +54,8 @@ const (
 	SA_RESTART = C.SA_RESTART
 	SA_ONSTACK = C.SA_ONSTACK
 
+	PTHREAD_CREATE_DETACHED = C.PTHREAD_CREATE_DETACHED
+
 	SIGHUP    = C.SIGHUP
 	SIGINT    = C.SIGINT
 	SIGQUIT   = C.SIGQUIT
@@ -129,3 +131,10 @@ type Timeval C.struct_timeval
 type Itimerval C.struct_itimerval
 
 type KeventT C.struct_kevent
+
+type Pthread C.pthread_t
+type PthreadAttr C.pthread_attr_t
+type PthreadCond C.pthread_cond_t
+type PthreadCondAttr C.pthread_condattr_t
+type PthreadMutex C.pthread_mutex_t
+type PthreadMutexAttr C.pthread_mutexattr_t

src/runtime/defs_openbsd_amd64.go

@@ -30,6 +30,8 @@ const (
 	_SA_RESTART = 0x2
 	_SA_ONSTACK = 0x1
 
+	_PTHREAD_CREATE_DETACHED = 0x1
+
 	_SIGHUP    = 0x1
 	_SIGINT    = 0x2
 	_SIGQUIT   = 0x3
@@ -177,3 +179,10 @@ type keventt struct {
 	data   int64
 	udata  *byte
 }
+
+type pthread uintptr
+type pthreadattr uintptr
+type pthreadcond uintptr
+type pthreadcondattr uintptr
+type pthreadmutex uintptr
+type pthreadmutexattr uintptr

src/runtime/os_openbsd.go

@@ -6,7 +6,6 @@ package runtime
 
 import (
 	"runtime/internal/atomic"
-	"runtime/internal/sys"
 	"unsafe"
 )
 
@@ -47,9 +46,6 @@ func raiseproc(sig uint32)
 func getthrid() int32
 func thrkill(tid int32, sig int)
 
-//go:noescape
-func tfork(param *tforkt, psize uintptr, mm *m, gg *g, fn uintptr) int32
-
 //go:noescape
 func thrsleep(ident uintptr, clock_id int32, tsp *timespec, lock uintptr, abort *uint32) int32
 
@@ -183,36 +179,6 @@ func semawakeup(mp *m) {
 	}
 }
 
-// May run with m.p==nil, so write barriers are not allowed.
-//go:nowritebarrier
-func newosproc(mp *m) {
-	stk := unsafe.Pointer(mp.g0.stack.hi)
-	if false {
-		print("newosproc stk=", stk, " m=", mp, " g=", mp.g0, " id=", mp.id, " ostk=", &mp, "\n")
-	}
-
-	// Stack pointer must point inside stack area (as marked with MAP_STACK),
-	// rather than at the top of it.
-	param := tforkt{
-		tf_tcb:   unsafe.Pointer(&mp.tls[0]),
-		tf_tid:   nil, // minit will record tid
-		tf_stack: uintptr(stk) - sys.PtrSize,
-	}
-
-	var oset sigset
-	sigprocmask(_SIG_SETMASK, &sigset_all, &oset)
-	ret := tfork(&param, unsafe.Sizeof(param), mp, mp.g0, funcPC(mstart))
-	sigprocmask(_SIG_SETMASK, &oset, nil)
-
-	if ret < 0 {
-		print("runtime: failed to create new OS thread (have ", mcount()-1, " already; errno=", -ret, ")\n")
-		if ret == -_EAGAIN {
-			println("runtime: may need to increase max user processes (ulimit -p)")
-		}
-		throw("runtime.newosproc")
-	}
-}
-
 func osinit() {
 	ncpu = getncpu()
 	physPageSize = getPageSize()

src/runtime/os_openbsd_libc.go

@@ -0,0 +1,58 @@
+// Copyright 2020 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.
+
+// +build openbsd,amd64
+
+package runtime
+
+import (
+	"unsafe"
+)
+
+var failThreadCreate = []byte("runtime: failed to create new OS thread\n")
+
+// mstart_stub provides glue code to call mstart from pthread_create.
+func mstart_stub()
+
+// May run with m.p==nil, so write barriers are not allowed.
+//go:nowritebarrierrec
+func newosproc(mp *m) {
+	if false {
+		print("newosproc m=", mp, " g=", mp.g0, " id=", mp.id, " ostk=", &mp, "\n")
+	}
+
+	// Initialize an attribute object.
+	var attr pthreadattr
+	if err := pthread_attr_init(&attr); err != 0 {
+		write(2, unsafe.Pointer(&failThreadCreate[0]), int32(len(failThreadCreate)))
+		exit(1)
+	}
+
+	// Find out OS stack size for our own stack guard.
+	var stacksize uintptr
+	if pthread_attr_getstacksize(&attr, &stacksize) != 0 {
+		write(2, unsafe.Pointer(&failThreadCreate[0]), int32(len(failThreadCreate)))
+		exit(1)
+	}
+	mp.g0.stack.hi = stacksize // for mstart
+
+	// Tell the pthread library we won't join with this thread.
+	if pthread_attr_setdetachstate(&attr, _PTHREAD_CREATE_DETACHED) != 0 {
+		write(2, unsafe.Pointer(&failThreadCreate[0]), int32(len(failThreadCreate)))
+		exit(1)
+	}
+
+	// Finally, create the thread. It starts at mstart_stub, which does some low-level
+	// setup and then calls mstart.
+	var oset sigset
+	sigprocmask(_SIG_SETMASK, &sigset_all, &oset)
+	err := pthread_create(&attr, funcPC(mstart_stub), unsafe.Pointer(mp))
+	sigprocmask(_SIG_SETMASK, &oset, nil)
+	if err != 0 {
+		write(2, unsafe.Pointer(&failThreadCreate[0]), int32(len(failThreadCreate)))
+		exit(1)
+	}
+
+	pthread_attr_destroy(&attr)
+}

src/runtime/os_openbsd_syscall.go

@@ -0,0 +1,45 @@
+// Copyright 2011 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.
+
+// +build openbsd,!amd64
+
+package runtime
+
+import (
+	"runtime/internal/sys"
+	"unsafe"
+)
+
+//go:noescape
+func tfork(param *tforkt, psize uintptr, mm *m, gg *g, fn uintptr) int32
+
+// May run with m.p==nil, so write barriers are not allowed.
+//go:nowritebarrier
+func newosproc(mp *m) {
+	stk := unsafe.Pointer(mp.g0.stack.hi)
+	if false {
+		print("newosproc stk=", stk, " m=", mp, " g=", mp.g0, " id=", mp.id, " ostk=", &mp, "\n")
+	}
+
+	// Stack pointer must point inside stack area (as marked with MAP_STACK),
+	// rather than at the top of it.
+	param := tforkt{
+		tf_tcb:   unsafe.Pointer(&mp.tls[0]),
+		tf_tid:   nil, // minit will record tid
+		tf_stack: uintptr(stk) - sys.PtrSize,
+	}
+
+	var oset sigset
+	sigprocmask(_SIG_SETMASK, &sigset_all, &oset)
+	ret := tfork(&param, unsafe.Sizeof(param), mp, mp.g0, funcPC(mstart))
+	sigprocmask(_SIG_SETMASK, &oset, nil)
+
+	if ret < 0 {
+		print("runtime: failed to create new OS thread (have ", mcount()-1, " already; errno=", -ret, ")\n")
+		if ret == -_EAGAIN {
+			println("runtime: may need to increase max user processes (ulimit -p)")
+		}
+		throw("runtime.newosproc")
+	}
+}

src/runtime/proc.go

@@ -1222,6 +1222,11 @@ func mStackIsSystemAllocated() bool {
 	switch GOOS {
 	case "aix", "darwin", "plan9", "illumos", "ios", "solaris", "windows":
 		return true
+	case "openbsd":
+		switch GOARCH {
+		case "amd64":
+			return true
+		}
 	}
 	return false
 }

src/runtime/sys_darwin.go

@@ -6,50 +6,6 @@ package runtime
 
 import "unsafe"
 
-// Call fn with arg as its argument. Return what fn returns.
-// fn is the raw pc value of the entry point of the desired function.
-// Switches to the system stack, if not already there.
-// Preserves the calling point as the location where a profiler traceback will begin.
-//go:nosplit
-func libcCall(fn, arg unsafe.Pointer) int32 {
-	// Leave caller's PC/SP/G around for traceback.
-	gp := getg()
-	var mp *m
-	if gp != nil {
-		mp = gp.m
-	}
-	if mp != nil && mp.libcallsp == 0 {
-		mp.libcallg.set(gp)
-		mp.libcallpc = getcallerpc()
-		// sp must be the last, because once async cpu profiler finds
-		// all three values to be non-zero, it will use them
-		mp.libcallsp = getcallersp()
-	} else {
-		// Make sure we don't reset libcallsp. This makes
-		// libcCall reentrant; We remember the g/pc/sp for the
-		// first call on an M, until that libcCall instance
-		// returns.  Reentrance only matters for signals, as
-		// libc never calls back into Go.  The tricky case is
-		// where we call libcX from an M and record g/pc/sp.
-		// Before that call returns, a signal arrives on the
-		// same M and the signal handling code calls another
-		// libc function.  We don't want that second libcCall
-		// from within the handler to be recorded, and we
-		// don't want that call's completion to zero
-		// libcallsp.
-		// We don't need to set libcall* while we're in a sighandler
-		// (even if we're not currently in libc) because we block all
-		// signals while we're handling a signal. That includes the
-		// profile signal, which is the one that uses the libcall* info.
-		mp = nil
-	}
-	res := asmcgocall(fn, arg)
-	if mp != nil {
-		mp.libcallsp = 0
-	}
-	return res
-}
-
 // The X versions of syscall expect the libc call to return a 64-bit result.
 // Otherwise (the non-X version) expects a 32-bit result.
 // This distinction is required because an error is indicated by returning -1,

src/runtime/sys_libc.go

@@ -0,0 +1,53 @@
+// Copyright 2018 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.
+
+// +build darwin openbsd,amd64
+
+package runtime
+
+import "unsafe"
+
+// Call fn with arg as its argument. Return what fn returns.
+// fn is the raw pc value of the entry point of the desired function.
+// Switches to the system stack, if not already there.
+// Preserves the calling point as the location where a profiler traceback will begin.
+//go:nosplit
+func libcCall(fn, arg unsafe.Pointer) int32 {
+	// Leave caller's PC/SP/G around for traceback.
+	gp := getg()
+	var mp *m
+	if gp != nil {
+		mp = gp.m
+	}
+	if mp != nil && mp.libcallsp == 0 {
+		mp.libcallg.set(gp)
+		mp.libcallpc = getcallerpc()
+		// sp must be the last, because once async cpu profiler finds
+		// all three values to be non-zero, it will use them
+		mp.libcallsp = getcallersp()
+	} else {
+		// Make sure we don't reset libcallsp. This makes
+		// libcCall reentrant; We remember the g/pc/sp for the
+		// first call on an M, until that libcCall instance
+		// returns.  Reentrance only matters for signals, as
+		// libc never calls back into Go.  The tricky case is
+		// where we call libcX from an M and record g/pc/sp.
+		// Before that call returns, a signal arrives on the
+		// same M and the signal handling code calls another
+		// libc function.  We don't want that second libcCall
+		// from within the handler to be recorded, and we
+		// don't want that call's completion to zero
+		// libcallsp.
+		// We don't need to set libcall* while we're in a sighandler
+		// (even if we're not currently in libc) because we block all
+		// signals while we're handling a signal. That includes the
+		// profile signal, which is the one that uses the libcall* info.
+		mp = nil
+	}
+	res := asmcgocall(fn, arg)
+	if mp != nil {
+		mp.libcallsp = 0
+	}
+	return res
+}