runtime: create library startup for aix/ppc64

As .init_array section aren't available on AIX, the Go runtime
initialization is made with gcc constructor attribute.
However, as cgo tool is building a binary in order to get imported
C symbols, Go symbols imported for this initilization must be ignored.
-Wl,-berok is mandatory otherwize ld will fail to create this binary,
_rt0_aix_ppc64_lib and runtime_rt0_go aren't defined in runtime/cgo.
These two symbols must also be ignored when creating _cgo_import.go.

Change-Id: Icf2e0282f5b50de5fa82007439a428e6147efef1
Reviewed-on: https://go-review.googlesource.com/c/go/+/169118
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

src/cmd/cgo/out.go

@@ -336,6 +336,12 @@ func dynimport(obj string) {
 			fatalf("cannot load imported symbols from XCOFF file %s: %v", obj, err)
 		}
 		for _, s := range sym {
+			if s.Name == "runtime_rt0_go" || s.Name == "_rt0_ppc64_aix_lib" {
+				// These symbols are imported by runtime/cgo but
+				// must not be added to _cgo_import.go as there are
+				// Go symbols.
+				continue
+			}
 			fmt.Fprintf(stdout, "//go:cgo_import_dynamic %s %s %q\n", s.Name, s.Name, s.Library)
 		}
 		lib, err := f.ImportedLibraries()

src/cmd/go/internal/work/security.go

@@ -171,6 +171,7 @@ var validLinkerFlags = []*lazyregexp.Regexp{
 	re(`-Wl,--(no-)?allow-shlib-undefined`),
 	re(`-Wl,--(no-)?as-needed`),
 	re(`-Wl,-Bdynamic`),
+	re(`-Wl,-berok`),
 	re(`-Wl,-Bstatic`),
 	re(`-WL,-O([^@,\-][^,]*)?`),
 	re(`-Wl,-d[ny]`),

src/runtime/cgo/callbacks_aix.go

@@ -8,3 +8,4 @@ package cgo
 // longcall on cgo programs (cf gcc_aix_ppc64.c).
 //go:cgo_export_static __cgo_topofstack
 //go:cgo_export_static runtime.rt0_go
+//go:cgo_export_static _rt0_ppc64_aix_lib

src/runtime/cgo/cgo.go

@@ -20,6 +20,7 @@ package cgo
 #cgo !android,linux LDFLAGS: -lpthread
 #cgo netbsd LDFLAGS: -lpthread
 #cgo openbsd LDFLAGS: -lpthread
+#cgo aix LDFLAGS: -Wl,-berok
 
 #cgo CFLAGS: -Wall -Werror
 

src/runtime/cgo/gcc_aix_ppc64.c

@@ -12,6 +12,7 @@
  */
 extern int __attribute__((longcall)) __cgo_topofstack(void);
 extern int __attribute__((longcall)) runtime_rt0_go(int argc, char **argv);
+extern void __attribute__((longcall)) _rt0_ppc64_aix_lib(void);
 
 int _cgo_topofstack(void) {
 	return __cgo_topofstack();
@@ -20,3 +21,18 @@ int _cgo_topofstack(void) {
 int main(int argc, char **argv) {
 	return runtime_rt0_go(argc, argv);
 }
+
+static void libinit(void) __attribute__ ((constructor));
+
+/*
+ * libinit aims to replace .init_array section which isn't available on aix.
+ * Using __attribute__ ((constructor)) let gcc handles this instead of
+ * adding special code in cmd/link.
+ * However, it will be called for every Go programs which has cgo.
+ * Inside _rt0_ppc64_aix_lib(), runtime.isarchive is checked in order
+ * to know if this program is a c-archive or a simple cgo program.
+ * If it's not set, _rt0_ppc64_ax_lib() returns directly.
+ */
+static void libinit() {
+	_rt0_ppc64_aix_lib();
+}

src/runtime/os2_aix.go

@@ -363,15 +363,34 @@ func syscall6(fn *libFunc, a0, a1, a2, a3, a4, a5 uintptr) (r, err uintptr) {
 	return c.r1, c.err
 }
 
+func exit1(code int32)
+
 //go:nosplit
 func exit(code int32) {
-	syscall1(&libc_exit, uintptr(code))
+	_g_ := getg()
+
+	// Check the validity of g because without a g during
+	// newosproc0.
+	if _g_ != nil {
+		syscall1(&libc_exit, uintptr(code))
+		return
+	}
+	exit1(code)
 }
 
+func write1(fd, p uintptr, n int32) int32
+
 //go:nosplit
 func write(fd uintptr, p unsafe.Pointer, n int32) int32 {
-	r, _ := syscall3(&libc_write, uintptr(fd), uintptr(p), uintptr(n))
-	return int32(r)
+	_g_ := getg()
+
+	// Check the validity of g because without a g during
+	// newosproc0.
+	if _g_ != nil {
+		r, _ := syscall3(&libc_write, uintptr(fd), uintptr(p), uintptr(n))
+		return int32(r)
+	}
+	return write1(fd, uintptr(p), n)
 
 }
 
@@ -428,13 +447,24 @@ func madvise(addr unsafe.Pointer, n uintptr, flags int32) {
 	}
 }
 
+func sigaction1(sig, new, old uintptr)
+
 //go:nosplit
 func sigaction(sig uintptr, new, old *sigactiont) {
-	r, err := syscall3(&libc_sigaction, sig, uintptr(unsafe.Pointer(new)), uintptr(unsafe.Pointer(old)))
-	if int32(r) == -1 {
-		println("Sigaction failed for sig: ", sig, " with error:", hex(err))
-		throw("syscall sigaction")
+	_g_ := getg()
+
+	// Check the validity of g because without a g during
+	// runtime.libpreinit.
+	if _g_ != nil {
+		r, err := syscall3(&libc_sigaction, sig, uintptr(unsafe.Pointer(new)), uintptr(unsafe.Pointer(old)))
+		if int32(r) == -1 {
+			println("Sigaction failed for sig: ", sig, " with error:", hex(err))
+			throw("syscall sigaction")
+		}
+		return
 	}
+
+	sigaction1(sig, uintptr(unsafe.Pointer(new)), uintptr(unsafe.Pointer(old)))
 }
 
 //go:nosplit
@@ -574,16 +604,36 @@ func pthread_attr_destroy(attr *pthread_attr) int32 {
 	return int32(r)
 }
 
+func pthread_attr_init1(attr uintptr) int32
+
 //go:nosplit
 func pthread_attr_init(attr *pthread_attr) int32 {
-	r, _ := syscall1(&libpthread_attr_init, uintptr(unsafe.Pointer(attr)))
-	return int32(r)
+	_g_ := getg()
+
+	// Check the validity of g because without a g during
+	// newosproc0.
+	if _g_ != nil {
+		r, _ := syscall1(&libpthread_attr_init, uintptr(unsafe.Pointer(attr)))
+		return int32(r)
+	}
+
+	return pthread_attr_init1(uintptr(unsafe.Pointer(attr)))
 }
 
+func pthread_attr_setdetachstate1(attr uintptr, state int32) int32
+
 //go:nosplit
 func pthread_attr_setdetachstate(attr *pthread_attr, state int32) int32 {
-	r, _ := syscall2(&libpthread_attr_setdetachstate, uintptr(unsafe.Pointer(attr)), uintptr(state))
-	return int32(r)
+	_g_ := getg()
+
+	// Check the validity of g because without a g during
+	// newosproc0.
+	if _g_ != nil {
+		r, _ := syscall2(&libpthread_attr_setdetachstate, uintptr(unsafe.Pointer(attr)), uintptr(state))
+		return int32(r)
+	}
+
+	return pthread_attr_setdetachstate1(uintptr(unsafe.Pointer(attr)), state)
 }
 
 //go:nosplit
@@ -598,16 +648,36 @@ func pthread_attr_getstacksize(attr *pthread_attr, size *uint64) int32 {
 	return int32(r)
 }
 
+func pthread_attr_setstacksize1(attr uintptr, size uint64) int32
+
 //go:nosplit
 func pthread_attr_setstacksize(attr *pthread_attr, size uint64) int32 {
-	r, _ := syscall2(&libpthread_attr_setstacksize, uintptr(unsafe.Pointer(attr)), uintptr(size))
-	return int32(r)
+	_g_ := getg()
+
+	// Check the validity of g because without a g during
+	// newosproc0.
+	if _g_ != nil {
+		r, _ := syscall2(&libpthread_attr_setstacksize, uintptr(unsafe.Pointer(attr)), uintptr(size))
+		return int32(r)
+	}
+
+	return pthread_attr_setstacksize1(uintptr(unsafe.Pointer(attr)), size)
 }
 
+func pthread_create1(tid, attr, fn, arg uintptr) int32
+
 //go:nosplit
 func pthread_create(tid *pthread, attr *pthread_attr, fn *funcDescriptor, arg unsafe.Pointer) int32 {
-	r, _ := syscall4(&libpthread_create, uintptr(unsafe.Pointer(tid)), uintptr(unsafe.Pointer(attr)), uintptr(unsafe.Pointer(fn)), uintptr(arg))
-	return int32(r)
+	_g_ := getg()
+
+	// Check the validity of g because without a g during
+	// newosproc0.
+	if _g_ != nil {
+		r, _ := syscall4(&libpthread_create, uintptr(unsafe.Pointer(tid)), uintptr(unsafe.Pointer(attr)), uintptr(unsafe.Pointer(fn)), uintptr(arg))
+		return int32(r)
+	}
+
+	return pthread_create1(uintptr(unsafe.Pointer(tid)), uintptr(unsafe.Pointer(attr)), uintptr(unsafe.Pointer(fn)), uintptr(arg))
 }
 
 // On multi-thread program, sigprocmask must not be called.

src/runtime/os_aix.go

@@ -97,6 +97,66 @@ func osinit() {
 	setupSystemConf()
 }
 
+// newosproc0 is a version of newosproc that can be called before the runtime
+// is initialized.
+//
+// This function is not safe to use after initialization as it does not pass an M as fnarg.
+//
+//go:nosplit
+func newosproc0(stacksize uintptr, fn *funcDescriptor) {
+	var (
+		attr pthread_attr
+		oset sigset
+		tid  pthread
+	)
+
+	if pthread_attr_init(&attr) != 0 {
+		write(2, unsafe.Pointer(&failthreadcreate[0]), int32(len(failthreadcreate)))
+		exit(1)
+	}
+
+	if pthread_attr_setstacksize(&attr, threadStackSize) != 0 {
+		write(2, unsafe.Pointer(&failthreadcreate[0]), int32(len(failthreadcreate)))
+		exit(1)
+	}
+
+	if pthread_attr_setdetachstate(&attr, _PTHREAD_CREATE_DETACHED) != 0 {
+		write(2, unsafe.Pointer(&failthreadcreate[0]), int32(len(failthreadcreate)))
+		exit(1)
+	}
+
+	// Disable signals during create, so that the new thread starts
+	// with signals disabled. It will enable them in minit.
+	sigprocmask(_SIG_SETMASK, &sigset_all, &oset)
+	var ret int32
+	for tries := 0; tries < 20; tries++ {
+		// pthread_create can fail with EAGAIN for no reasons
+		// but it will be ok if it retries.
+		ret = pthread_create(&tid, &attr, fn, nil)
+		if ret != _EAGAIN {
+			break
+		}
+		usleep(uint32(tries+1) * 1000) // Milliseconds.
+	}
+	sigprocmask(_SIG_SETMASK, &oset, nil)
+	if ret != 0 {
+		write(2, unsafe.Pointer(&failthreadcreate[0]), int32(len(failthreadcreate)))
+		exit(1)
+	}
+
+}
+
+var failthreadcreate = []byte("runtime: failed to create new OS thread\n")
+
+// Called to do synchronous initialization of Go code built with
+// -buildmode=c-archive or -buildmode=c-shared.
+// None of the Go runtime is initialized.
+//go:nosplit
+//go:nowritebarrierrec
+func libpreinit() {
+	initsig(true)
+}
+
 // Ms related functions
 func mpreinit(mp *m) {
 	mp.gsignal = malg(32 * 1024) // AIX wants >= 8K
@@ -213,7 +273,13 @@ func setsig(i uint32, fn uintptr) {
 //go:nosplit
 //go:nowritebarrierrec
 func setsigstack(i uint32) {
-	throw("Not yet implemented\n")
+	var sa sigactiont
+	sigaction(uintptr(i), nil, &sa)
+	if sa.sa_flags&_SA_ONSTACK != 0 {
+		return
+	}
+	sa.sa_flags |= _SA_ONSTACK
+	sigaction(uintptr(i), &sa, nil)
 }
 
 //go:nosplit

src/runtime/rt0_aix_ppc64.s

@@ -47,3 +47,153 @@ TEXT _main(SB),NOSPLIT,$-8
 	MOVD R12, CTR
 	BR (CTR)
 
+
+TEXT _rt0_ppc64_aix_lib(SB),NOSPLIT,$-8
+	// Start with standard C stack frame layout and linkage.
+	MOVD	LR, R0
+	MOVD	R0, 16(R1) // Save LR in caller's frame.
+	MOVW	CR, R0	   // Save CR in caller's frame
+	MOVD	R0, 8(R1)
+
+	MOVDU	R1, -344(R1) // Allocate frame.
+
+	// Preserve callee-save registers.
+	MOVD	R14, 48(R1)
+	MOVD	R15, 56(R1)
+	MOVD	R16, 64(R1)
+	MOVD	R17, 72(R1)
+	MOVD	R18, 80(R1)
+	MOVD	R19, 88(R1)
+	MOVD	R20, 96(R1)
+	MOVD	R21,104(R1)
+	MOVD	R22, 112(R1)
+	MOVD	R23, 120(R1)
+	MOVD	R24, 128(R1)
+	MOVD	R25, 136(R1)
+	MOVD	R26, 144(R1)
+	MOVD	R27, 152(R1)
+	MOVD	R28, 160(R1)
+	MOVD	R29, 168(R1)
+	MOVD	g, 176(R1) // R30
+	MOVD	R31, 184(R1)
+	FMOVD	F14, 192(R1)
+	FMOVD	F15, 200(R1)
+	FMOVD	F16, 208(R1)
+	FMOVD	F17, 216(R1)
+	FMOVD	F18, 224(R1)
+	FMOVD	F19, 232(R1)
+	FMOVD	F20, 240(R1)
+	FMOVD	F21, 248(R1)
+	FMOVD	F22, 256(R1)
+	FMOVD	F23, 264(R1)
+	FMOVD	F24, 272(R1)
+	FMOVD	F25, 280(R1)
+	FMOVD	F26, 288(R1)
+	FMOVD	F27, 296(R1)
+	FMOVD	F28, 304(R1)
+	FMOVD	F29, 312(R1)
+	FMOVD	F30, 320(R1)
+	FMOVD	F31, 328(R1)
+
+	// Synchronous initialization.
+	MOVD	$runtime·reginit(SB), R12
+	MOVD	R12, CTR
+	BL	(CTR)
+
+	MOVBZ	runtime·isarchive(SB), R3	// Check buildmode = c-archive
+	CMP		$0, R3
+	BEQ		done
+
+	MOVD	R14, _rt0_ppc64_aix_lib_argc<>(SB)
+	MOVD	R15, _rt0_ppc64_aix_lib_argv<>(SB)
+
+	MOVD	$runtime·libpreinit(SB), R12
+	MOVD	R12, CTR
+	BL	(CTR)
+
+	// Create a new thread to do the runtime initialization and return.
+	MOVD	_cgo_sys_thread_create(SB), R12
+	CMP	$0, R12
+	BEQ	nocgo
+	MOVD	$_rt0_ppc64_aix_lib_go(SB), R3
+	MOVD	$0, R4
+	MOVD	R2, 40(R1)
+	MOVD	8(R12), R2
+	MOVD	(R12), R12
+	MOVD	R12, CTR
+	BL	(CTR)
+	MOVD	40(R1), R2
+	BR	done
+
+nocgo:
+	MOVD	$0x800000, R12					   // stacksize = 8192KB
+	MOVD	R12, 8(R1)
+	MOVD	$_rt0_ppc64_aix_lib_go(SB), R12
+	MOVD	R12, 16(R1)
+	MOVD	$runtime·newosproc0(SB),R12
+	MOVD	R12, CTR
+	BL	(CTR)
+
+done:
+	// Restore saved registers.
+	MOVD	48(R1), R14
+	MOVD	56(R1), R15
+	MOVD	64(R1), R16
+	MOVD	72(R1), R17
+	MOVD	80(R1), R18
+	MOVD	88(R1), R19
+	MOVD	96(R1), R20
+	MOVD	104(R1), R21
+	MOVD	112(R1), R22
+	MOVD	120(R1), R23
+	MOVD	128(R1), R24
+	MOVD	136(R1), R25
+	MOVD	144(R1), R26
+	MOVD	152(R1), R27
+	MOVD	160(R1), R28
+	MOVD	168(R1), R29
+	MOVD	176(R1), g // R30
+	MOVD	184(R1), R31
+	FMOVD	196(R1), F14
+	FMOVD	200(R1), F15
+	FMOVD	208(R1), F16
+	FMOVD	216(R1), F17
+	FMOVD	224(R1), F18
+	FMOVD	232(R1), F19
+	FMOVD	240(R1), F20
+	FMOVD	248(R1), F21
+	FMOVD	256(R1), F22
+	FMOVD	264(R1), F23
+	FMOVD	272(R1), F24
+	FMOVD	280(R1), F25
+	FMOVD	288(R1), F26
+	FMOVD	296(R1), F27
+	FMOVD	304(R1), F28
+	FMOVD	312(R1), F29
+	FMOVD	320(R1), F30
+	FMOVD	328(R1), F31
+
+	ADD	$344, R1
+
+	MOVD	8(R1), R0
+	MOVFL	R0, $0xff
+	MOVD	16(R1), R0
+	MOVD	R0, LR
+	RET
+
+DATA	_rt0_ppc64_aix_lib_go+0(SB)/8, $__rt0_ppc64_aix_lib_go(SB)
+DATA	_rt0_ppc64_aix_lib_go+8(SB)/8, $TOC(SB)
+DATA	_rt0_ppc64_aix_lib_go+16(SB)/8, $0
+GLOBL	_rt0_ppc64_aix_lib_go(SB), NOPTR, $24
+
+TEXT __rt0_ppc64_aix_lib_go(SB),NOSPLIT,$0
+	MOVD	_rt0_ppc64_aix_lib_argc<>(SB), R3
+	MOVD	_rt0_ppc64_aix_lib_argv<>(SB), R4
+	MOVD	$runtime·rt0_go(SB), R12
+	MOVD	R12, CTR
+	BR	(CTR)
+
+DATA _rt0_ppc64_aix_lib_argc<>(SB)/8, $0
+GLOBL _rt0_ppc64_aix_lib_argc<>(SB),NOPTR, $8
+DATA _rt0_ppc64_aix_lib_argv<>(SB)/8, $0
+GLOBL _rt0_ppc64_aix_lib_argv<>(SB),NOPTR, $8

src/runtime/symtab.go

@@ -445,6 +445,9 @@ func moduledataverify1(datap *moduledata) {
 			for j := 0; j <= i; j++ {
 				print("\t", hex(datap.ftab[j].entry), " ", funcname(funcInfo{(*_func)(unsafe.Pointer(&datap.pclntable[datap.ftab[j].funcoff])), datap}), "\n")
 			}
+			if GOOS == "aix" && isarchive {
+				println("-Wl,-bnoobjreorder is mandatory on aix/ppc64 with c-archive")
+			}
 			throw("invalid runtime symbol table")
 		}
 	}

src/runtime/sys_aix_ppc64.s

@@ -216,17 +216,22 @@ TEXT runtime·_tstart(SB),NOSPLIT,$0
 	MOVD R0, R3
 	RET
 
+
+#define CSYSCALL()			\
+	MOVD	0(R12), R12		\
+	MOVD	R2, 40(R1)		\
+	MOVD	0(R12), R0		\
+	MOVD	8(R12), R2		\
+	MOVD	R0, CTR			\
+	BL	(CTR)			\
+	MOVD	40(R1), R2		\
+	BL runtime·reginit(SB)
+
+
 // Runs on OS stack, called from runtime·osyield.
 TEXT runtime·osyield1(SB),NOSPLIT,$0
 	MOVD	$libc_sched_yield(SB), R12
-	MOVD	0(R12), R12
-	MOVD	R2, 40(R1)
-	MOVD	0(R12), R0
-	MOVD	8(R12), R2
-	MOVD	R0, CTR
-	BL	(CTR)
-	MOVD	40(R1), R2
-	BL runtime·reginit(SB)
+	CSYSCALL()
 	RET
 
 
@@ -236,26 +241,75 @@ TEXT runtime·sigprocmask1(SB),NOSPLIT,$0-24
 	MOVD	new+8(FP), R4
 	MOVD	old+16(FP), R5
 	MOVD	$libpthread_sigthreadmask(SB), R12
-	MOVD	0(R12), R12
-	MOVD	R2, 40(R1)
-	MOVD	0(R12), R0
-	MOVD	8(R12), R2
-	MOVD	R0, CTR
-	BL	(CTR)
-	MOVD	40(R1), R2
-	BL runtime·reginit(SB)
+	CSYSCALL()
 	RET
 
 // Runs on OS stack, called from runtime·usleep.
-TEXT runtime·usleep1(SB),NOSPLIT,$0-8
+TEXT runtime·usleep1(SB),NOSPLIT,$0-4
 	MOVW	us+0(FP), R3
 	MOVD	$libc_usleep(SB), R12
-	MOVD	0(R12), R12
-	MOVD	R2, 40(R1)
-	MOVD	0(R12), R0
-	MOVD	8(R12), R2
-	MOVD	R0, CTR
-	BL	(CTR)
-	MOVD	40(R1), R2
-	BL runtime·reginit(SB)
+	CSYSCALL()
+	RET
+
+// Runs on OS stack, called from runtime·exit.
+TEXT runtime·exit1(SB),NOSPLIT,$0-4
+	MOVW	code+0(FP), R3
+	MOVD	$libc_exit(SB), R12
+	CSYSCALL()
+	RET
+
+// Runs on OS stack, called from runtime·write.
+TEXT runtime·write1(SB),NOSPLIT,$0-28
+	MOVD	fd+0(FP), R3
+	MOVD	p+8(FP), R4
+	MOVW	n+16(FP), R5
+	MOVD	$libc_write(SB), R12
+	CSYSCALL()
+	MOVW	R3, ret+24(FP)
+	RET
+
+// Runs on OS stack, called from runtime·pthread_attr_init.
+TEXT runtime·pthread_attr_init1(SB),NOSPLIT,$0-12
+	MOVD	attr+0(FP), R3
+	MOVD	$libpthread_attr_init(SB), R12
+	CSYSCALL()
+	MOVW	R3, ret+8(FP)
+	RET
+
+// Runs on OS stack, called from runtime·pthread_attr_setstacksize.
+TEXT runtime·pthread_attr_setstacksize1(SB),NOSPLIT,$0-20
+	MOVD	attr+0(FP), R3
+	MOVD	size+8(FP), R4
+	MOVD	$libpthread_attr_setstacksize(SB), R12
+	CSYSCALL()
+	MOVW	R3, ret+16(FP)
+	RET
+
+// Runs on OS stack, called from runtime·pthread_setdetachstate.
+TEXT runtime·pthread_attr_setdetachstate1(SB),NOSPLIT,$0-20
+	MOVD	attr+0(FP), R3
+	MOVW	state+8(FP), R4
+	MOVD	$libpthread_attr_setdetachstate(SB), R12
+	CSYSCALL()
+	MOVW	R3, ret+16(FP)
+	RET
+
+// Runs on OS stack, called from runtime·pthread_create.
+TEXT runtime·pthread_create1(SB),NOSPLIT,$0-36
+	MOVD	tid+0(FP), R3
+	MOVD	attr+8(FP), R4
+	MOVD	fn+16(FP), R5
+	MOVD	arg+24(FP), R6
+	MOVD	$libpthread_create(SB), R12
+	CSYSCALL()
+	MOVW	R3, ret+32(FP)
+	RET
+
+// Runs on OS stack, called from runtime·sigaction.
+TEXT runtime·sigaction1(SB),NOSPLIT,$0-24
+	MOVD	sig+0(FP), R3
+	MOVD	new+8(FP), R4
+	MOVD	old+16(FP), R5
+	MOVD	$libc_sigaction(SB), R12
+	CSYSCALL()
 	RET