runtime: add library startup support for ppc64le

I have been running this patch inside Google against Go 1.6 for the last month.

The new tests will probably break the builders but let's see
exactly how they break.

Change-Id: Ia65cf7d3faecffeeb4b06e9b80875c0e57d86d9e
Reviewed-on: https://go-review.googlesource.com/23452Reviewed-by: Ian Lance Taylor <iant@golang.org>

src/cmd/dist/test.go

@@ -680,7 +680,8 @@ func (t *tester) supportedBuildmode(mode string) bool {
 		}
 		switch pair {
 		case "darwin-386", "darwin-amd64", "darwin-arm", "darwin-arm64",
-			"linux-amd64", "linux-386", "windows-amd64", "windows-386":
+			"linux-amd64", "linux-386", "windows-amd64", "windows-386",
+			"linux-ppc64le":
 			return true
 		}
 		return false
@@ -688,7 +689,8 @@ func (t *tester) supportedBuildmode(mode string) bool {
 		switch pair {
 		case "linux-386", "linux-amd64", "linux-arm", "linux-arm64",
 			"darwin-amd64", "darwin-386",
-			"android-arm", "android-arm64", "android-386":
+			"android-arm", "android-arm64", "android-386",
+			"linux-ppc64le":
 			return true
 		}
 		return false

src/runtime/rt0_linux_ppc64le.s

@@ -4,6 +4,135 @@
 TEXT _rt0_ppc64le_linux(SB),NOSPLIT,$0
 	BR _main<>(SB)
 
+TEXT _rt0_ppc64le_linux_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.
+	MOVD	R2, 24(R1) // Save TOC in caller's frame.
+	MOVDU	R1, -320(R1) // Allocate frame.
+	
+	// Preserve callee-save registers.
+	MOVD	R14, 24(R1)
+	MOVD	R15, 32(R1)
+	MOVD	R16, 40(R1)
+	MOVD	R17, 48(R1)
+	MOVD	R18, 56(R1)
+	MOVD	R19, 64(R1)
+	MOVD	R20, 72(R1)
+	MOVD	R21, 80(R1)
+	MOVD	R22, 88(R1)
+	MOVD	R23, 96(R1)
+	MOVD	R24, 104(R1)
+	MOVD	R25, 112(R1)
+	MOVD	R26, 120(R1)
+	MOVD	R27, 128(R1)
+	MOVD	R28, 136(R1)
+	MOVD	R29, 144(R1)
+	MOVD	g, 152(R1) // R30
+	MOVD	R31, 160(R1)
+	FMOVD	F14, 168(R1)
+	FMOVD	F15, 176(R1)
+	FMOVD	F16, 184(R1)
+	FMOVD	F17, 192(R1)
+	FMOVD	F18, 200(R1)
+	FMOVD	F19, 208(R1)
+	FMOVD	F20, 216(R1)
+	FMOVD	F21, 224(R1)
+	FMOVD	F22, 232(R1)
+	FMOVD	F23, 240(R1)
+	FMOVD	F24, 248(R1)
+	FMOVD	F25, 256(R1)
+	FMOVD	F26, 264(R1)
+	FMOVD	F27, 272(R1)
+	FMOVD	F28, 280(R1)
+	FMOVD	F29, 288(R1)
+	FMOVD	F30, 296(R1)
+	FMOVD	F31, 304(R1)
+
+	MOVD	R3, _rt0_ppc64le_linux_lib_argc<>(SB)
+	MOVD	R4, _rt0_ppc64le_linux_lib_argv<>(SB)
+
+	// Synchronous initialization.
+	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_ppc64le_linux_lib_go(SB), R3
+	MOVD	$0, R4
+	MOVD	R12, CTR
+	BL	(CTR)
+	BR	done
+
+nocgo:
+	MOVD	$0x800000, R12                     // stacksize = 8192KB
+	MOVD	R12, 8(R1)
+	MOVD	$_rt0_ppc64le_linux_lib_go(SB), R12
+	MOVD	R12, 16(R1)
+	MOVD	$runtime·newosproc0(SB),R12
+	MOVD	R12, CTR
+	BL	(CTR)
+
+done:
+	// Restore saved registers.
+	MOVD	24(R1), R14
+	MOVD	32(R1), R15
+	MOVD	40(R1), R16
+	MOVD	48(R1), R17
+	MOVD	56(R1), R18
+	MOVD	64(R1), R19
+	MOVD	72(R1), R20
+	MOVD	80(R1), R21
+	MOVD	88(R1), R22
+	MOVD	96(R1), R23
+	MOVD	104(R1), R24
+	MOVD	112(R1), R25
+	MOVD	120(R1), R26
+	MOVD	128(R1), R27
+	MOVD	136(R1), R28
+	MOVD	144(R1), R29
+	MOVD	152(R1), g // R30
+	MOVD	160(R1), R31
+	FMOVD	168(R1), F14
+	FMOVD	176(R1), F15
+	FMOVD	184(R1), F16
+	FMOVD	192(R1), F17
+	FMOVD	200(R1), F18
+	FMOVD	208(R1), F19
+	FMOVD	216(R1), F20
+	FMOVD	224(R1), F21
+	FMOVD	232(R1), F22
+	FMOVD	240(R1), F23
+	FMOVD	248(R1), F24
+	FMOVD	256(R1), F25
+	FMOVD	264(R1), F26
+	FMOVD	272(R1), F27
+	FMOVD	280(R1), F28
+	FMOVD	288(R1), F29
+	FMOVD	296(R1), F30
+	FMOVD	304(R1), F31
+
+	ADD	$320, R1
+	MOVD	24(R1), R2
+	MOVD	16(R1), R0
+	MOVD	R0, LR
+	RET
+
+TEXT _rt0_ppc64le_linux_lib_go(SB),NOSPLIT,$0
+	MOVD	_rt0_ppc64le_linux_lib_argc<>(SB), R3
+	MOVD	_rt0_ppc64le_linux_lib_argv<>(SB), R4
+	MOVD	$runtime·rt0_go(SB), R12
+	MOVD	R12, CTR
+	BR	(CTR)
+
+DATA _rt0_ppc64le_linux_lib_argc<>(SB)/8, $0
+GLOBL _rt0_ppc64le_linux_lib_argc<>(SB),NOPTR, $8
+DATA _rt0_ppc64le_linux_lib_argv<>(SB)/8, $0
+GLOBL _rt0_ppc64le_linux_lib_argv<>(SB),NOPTR, $8
+
 TEXT _main<>(SB),NOSPLIT,$-8
 	// In a statically linked binary, the stack contains argc,
 	// argv as argc string pointers followed by a NULL, envv as a