runtime: unify arm entry point code

Change-Id: Id51a2d63f7199b3ff71cedd415345ad20e5bd981
Reviewed-on: https://go-review.googlesource.com/70791
Reviewed-by: Austin Clements <austin@google.com>

src/cmd/vet/all/whitelist/arm.txt

@@ -18,7 +18,4 @@ runtime/tls_arm.s: [arm] save_g: function save_g missing Go declaration
 runtime/tls_arm.s: [arm] load_g: function load_g missing Go declaration
 runtime/tls_arm.s: [arm] _initcgo: function _initcgo missing Go declaration
 
-// Clearer using FP than SP, but that requires named offsets.
-runtime/asm_arm.s: [arm] rt0_go: use of 4(R13) points beyond argument frame
-
 runtime/internal/atomic/asm_arm.s: [arm] cas: function cas missing Go declaration

src/runtime/asm_arm.s

@@ -7,14 +7,112 @@
 #include "funcdata.h"
 #include "textflag.h"
 
+// _rt0_arm is common startup code for most ARM systems when using
+// internal linking. This is the entry point for the program from the
+// kernel for an ordinary -buildmode=exe program. The stack holds the
+// number of arguments and the C-style argv.
+TEXT _rt0_arm(SB),NOSPLIT,$-4
+	MOVW	(R13), R0	// argc
+	MOVW	$4(R13), R1		// argv
+	B	runtime·rt0_go(SB)
+
+// main is common startup code for most ARM systems when using
+// external linking. The C startup code will call the symbol "main"
+// passing argc and argv in the usual C ABI registers R0 and R1.
+TEXT main(SB),NOSPLIT,$-4
+	B	runtime·rt0_go(SB)
+
+// _rt0_arm_lib is common startup code for most ARM systems when
+// using -buildmode=c-archive or -buildmode=c-shared. The linker will
+// arrange to invoke this function as a global constructor (for
+// c-archive) or when the shared library is loaded (for c-shared).
+// We expect argc and argv to be passed in the usual C ABI registers
+// R0 and R1.
+TEXT _rt0_arm_lib(SB),NOSPLIT,$104
+	// Preserve callee-save registers. Raspberry Pi's dlopen(), for example,
+	// actually cares that R11 is preserved.
+	MOVW	R4, 12(R13)
+	MOVW	R5, 16(R13)
+	MOVW	R6, 20(R13)
+	MOVW	R7, 24(R13)
+	MOVW	R8, 28(R13)
+	MOVW	R11, 32(R13)
+
+	// Skip floating point registers on GOARM < 6.
+	MOVB    runtime·goarm(SB), R11
+	CMP	$6, R11
+	BLT	skipfpsave
+	MOVD	F8, (32+8*1)(R13)
+	MOVD	F9, (32+8*2)(R13)
+	MOVD	F10, (32+8*3)(R13)
+	MOVD	F11, (32+8*4)(R13)
+	MOVD	F12, (32+8*5)(R13)
+	MOVD	F13, (32+8*6)(R13)
+	MOVD	F14, (32+8*7)(R13)
+	MOVD	F15, (32+8*8)(R13)
+skipfpsave:
+	// Save argc/argv.
+	MOVW	R0, _rt0_arm_lib_argc<>(SB)
+	MOVW	R1, _rt0_arm_lib_argv<>(SB)
+
+	// Synchronous initialization.
+	CALL	runtime·libpreinit(SB)
+
+	// Create a new thread to do the runtime initialization.
+	MOVW	_cgo_sys_thread_create(SB), R2
+	CMP	$0, R2
+	BEQ	nocgo
+	MOVW	$_rt0_arm_lib_go<>(SB), R0
+	MOVW	$0, R1
+	BL	(R2)
+	B	rr
+nocgo:
+	MOVW	$0x800000, R0                     // stacksize = 8192KB
+	MOVW	$_rt0_arm_lib_go<>(SB), R1  // fn
+	MOVW	R0, 4(R13)
+	MOVW	R1, 8(R13)
+	BL	runtime·newosproc0(SB)
+rr:
+	// Restore callee-save registers and return.
+	MOVB    runtime·goarm(SB), R11
+	CMP	$6, R11
+	BLT	skipfprest
+	MOVD	(32+8*1)(R13), F8
+	MOVD	(32+8*2)(R13), F9
+	MOVD	(32+8*3)(R13), F10
+	MOVD	(32+8*4)(R13), F11
+	MOVD	(32+8*5)(R13), F12
+	MOVD	(32+8*6)(R13), F13
+	MOVD	(32+8*7)(R13), F14
+	MOVD	(32+8*8)(R13), F15
+skipfprest:
+	MOVW	12(R13), R4
+	MOVW	16(R13), R5
+	MOVW	20(R13), R6
+	MOVW	24(R13), R7
+	MOVW	28(R13), R8
+	MOVW	32(R13), R11
+	RET
+
+// _rt0_arm_lib_go initializes the Go runtime.
+// This is started in a separate thread by _rt0_arm_lib.
+TEXT _rt0_arm_lib_go<>(SB),NOSPLIT,$8
+	MOVW	_rt0_arm_lib_argc<>(SB), R0
+	MOVW	_rt0_arm_lib_argv<>(SB), R1
+	B	runtime·rt0_go(SB)
+
+DATA _rt0_arm_lib_argc<>(SB)/4,$0
+GLOBL _rt0_arm_lib_argc<>(SB),NOPTR,$4
+DATA _rt0_arm_lib_argv<>(SB)/4,$0
+GLOBL _rt0_arm_lib_argv<>(SB),NOPTR,$4
+
 // using frame size $-4 means do not save LR on stack.
+// argc is in R0, argv is in R1.
 TEXT runtime·rt0_go(SB),NOSPLIT,$-4
 	MOVW	$0xcafebabe, R12
 
 	// copy arguments forward on an even stack
 	// use R13 instead of SP to avoid linker rewriting the offsets
-	MOVW	0(R13), R0		// argc
-	MOVW	4(R13), R1		// argv
 	SUB	$64, R13		// plenty of scratch
 	AND	$~7, R13
 	MOVW	R0, 60(R13)		// save argc, argv away

src/runtime/rt0_android_arm.s

@@ -10,13 +10,10 @@ TEXT _rt0_arm_android(SB),NOSPLIT,$-4
 	MOVW		$_rt0_arm_linux1(SB), R4
 	B		(R4)
 
-// When building with -buildmode=c-shared, this symbol is called when the shared
-// library is loaded.
 TEXT _rt0_arm_android_lib(SB),NOSPLIT,$0
 	MOVW	$1, R0                          // argc
 	MOVW	$_rt0_arm_android_argv(SB), R1  // **argv
-	BL _rt0_arm_linux_lib(SB)
+	B	_rt0_arm_lib(SB)
-	RET
 
 DATA _rt0_arm_android_argv+0x00(SB)/4,$_rt0_arm_android_argv0(SB)
 DATA _rt0_arm_android_argv+0x04(SB)/4,$0 // end argv

src/runtime/rt0_darwin_arm.s

@@ -4,94 +4,8 @@
 
 #include "textflag.h"
 
-TEXT _rt0_arm_darwin(SB),7,$-4
+TEXT _rt0_arm_darwin(SB),7,$0
-	// prepare arguments for main (_rt0_go)
+	B	_rt0_asm(SB)
-	MOVW	(R13), R0	// argc
-	MOVW	$4(R13), R1		// argv
-	MOVW	$main(SB), R4
-	B		(R4)
 
-// When linking with -buildmode=c-archive or -buildmode=c-shared,
+TEXT _rt0_arm_darwin_lib(SB),NOSPLIT,$0
-// this symbol is called from a global initialization function.
+	B	_rt0_arm_lib(SB)
-//
-// Note that all currently shipping darwin/arm platforms require
-// cgo and do not support c-shared.
-TEXT _rt0_arm_darwin_lib(SB),NOSPLIT,$104
-	// Preserve callee-save registers.
-	MOVW    R4, 12(R13)
-	MOVW    R5, 16(R13)
-	MOVW    R6, 20(R13)
-	MOVW    R7, 24(R13)
-	MOVW    R8, 28(R13)
-	MOVW    R11, 32(R13)
-
-	MOVD	F8, (32+8*1)(R13)
-	MOVD	F9, (32+8*2)(R13)
-	MOVD	F10, (32+8*3)(R13)
-	MOVD	F11, (32+8*4)(R13)
-	MOVD	F12, (32+8*5)(R13)
-	MOVD	F13, (32+8*6)(R13)
-	MOVD	F14, (32+8*7)(R13)
-	MOVD	F15, (32+8*8)(R13)
-
-	MOVW  R0, _rt0_arm_darwin_lib_argc<>(SB)
-	MOVW  R1, _rt0_arm_darwin_lib_argv<>(SB)
-
-	// Synchronous initialization.
-	MOVW	$runtime·libpreinit(SB), R3
-	CALL	(R3)
-
-	// Create a new thread to do the runtime initialization and return.
-	MOVW  _cgo_sys_thread_create(SB), R3
-	CMP   $0, R3
-	B.EQ  nocgo
-	MOVW  $_rt0_arm_darwin_lib_go(SB), R0
-	MOVW  $0, R1
-	BL    (R3)
-	B rr
-nocgo:
-	MOVW  $0x400000, R0
-	MOVW  R0, (R13) // stacksize
-	MOVW  $_rt0_arm_darwin_lib_go(SB), R0
-	MOVW  R0, 4(R13) // fn
-	MOVW  $0, R0
-	MOVW  R0, 8(R13) // fnarg
-	MOVW  $runtime·newosproc0(SB), R3
-	BL    (R3)
-rr:
-	// Restore callee-save registers and return.
-	MOVW    12(R13), R4
-	MOVW    16(R13), R5
-	MOVW    20(R13), R6
-	MOVW    24(R13), R7
-	MOVW    28(R13), R8
-	MOVW    32(R13), R11
-	MOVD	(32+8*1)(R13), F8
-	MOVD	(32+8*2)(R13), F9
-	MOVD	(32+8*3)(R13), F10
-	MOVD	(32+8*4)(R13), F11
-	MOVD	(32+8*5)(R13), F12
-	MOVD	(32+8*6)(R13), F13
-	MOVD	(32+8*7)(R13), F14
-	MOVD	(32+8*8)(R13), F15
-	RET
-
-
-TEXT _rt0_arm_darwin_lib_go(SB),NOSPLIT,$0
-	MOVW  _rt0_arm_darwin_lib_argc<>(SB), R0
-	MOVW  _rt0_arm_darwin_lib_argv<>(SB), R1
-	MOVW  R0,  (R13)
-	MOVW  R1, 4(R13)
-	MOVW  $runtime·rt0_go(SB), R4
-	B     (R4)
-
-DATA  _rt0_arm_darwin_lib_argc<>(SB)/4, $0
-GLOBL _rt0_arm_darwin_lib_argc<>(SB),NOPTR, $4
-DATA  _rt0_arm_darwin_lib_argv<>(SB)/4, $0
-GLOBL _rt0_arm_darwin_lib_argv<>(SB),NOPTR, $4
-
-TEXT main(SB),NOSPLIT,$-8
-	// save argc and argv onto stack
-	MOVM.DB.W [R0-R1], (R13)
-	MOVW	$runtime·rt0_go(SB), R4
-	B		(R4)

src/runtime/rt0_freebsd_arm.s

@@ -4,13 +4,8 @@
 
 #include "textflag.h"
 
-TEXT _rt0_arm_freebsd(SB),NOSPLIT,$-4
+TEXT _rt0_arm_freebsd(SB),NOSPLIT,$0
-	MOVW	(R13), R0		// argc
+	B	_rt0_arm(SB)
-	MOVW	$4(R13), R1		// argv
-	MOVM.DB.W [R0-R1], (R13)
-	B	runtime·rt0_go(SB)
 
-TEXT main(SB),NOSPLIT,$-4
+TEXT _rt0_arm_freebsd_lib(SB),NOSPLIT,$0
-	MOVM.DB.W [R0-R1], (R13)
+	B	_rt0_arm_lib(SB)
-	MOVW	$runtime·rt0_go(SB), R4
-	B	(R4)

src/runtime/rt0_linux_arm.s

@@ -12,84 +12,8 @@ TEXT _rt0_arm_linux(SB),NOSPLIT,$-4
 
 // When building with -buildmode=c-shared, this symbol is called when the shared
 // library is loaded.
-TEXT _rt0_arm_linux_lib(SB),NOSPLIT,$104
+TEXT _rt0_arm_linux_lib(SB),NOSPLIT,$0
-	// Preserve callee-save registers. Raspberry Pi's dlopen(), for example,
+	B	_rt0_arm_lib(SB)
-	// actually cares that R11 is preserved.
-	MOVW	R4, 12(R13)
-	MOVW	R5, 16(R13)
-	MOVW	R6, 20(R13)
-	MOVW	R7, 24(R13)
-	MOVW	R8, 28(R13)
-	MOVW	R11, 32(R13)
-
-	// Skip floating point registers on GOARM < 6.
-	MOVB    runtime·goarm(SB), R11
-	CMP $6, R11
-	BLT skipfpsave
-	MOVD	F8, (32+8*1)(R13)
-	MOVD	F9, (32+8*2)(R13)
-	MOVD	F10, (32+8*3)(R13)
-	MOVD	F11, (32+8*4)(R13)
-	MOVD	F12, (32+8*5)(R13)
-	MOVD	F13, (32+8*6)(R13)
-	MOVD	F14, (32+8*7)(R13)
-	MOVD	F15, (32+8*8)(R13)
-skipfpsave:
-	// Save argc/argv.
-	MOVW	R0, _rt0_arm_linux_lib_argc<>(SB)
-	MOVW	R1, _rt0_arm_linux_lib_argv<>(SB)
-
-	// Synchronous initialization.
-	MOVW	$runtime·libpreinit(SB), R2
-	CALL	(R2)
-
-	// Create a new thread to do the runtime initialization.
-	MOVW	_cgo_sys_thread_create(SB), R2
-	CMP	$0, R2
-	BEQ	nocgo
-	MOVW	$_rt0_arm_linux_lib_go<>(SB), R0
-	MOVW	$0, R1
-	BL	(R2)
-	B	rr
-nocgo:
-	MOVW	$0x800000, R0                     // stacksize = 8192KB
-	MOVW	$_rt0_arm_linux_lib_go<>(SB), R1  // fn
-	MOVW	R0, 4(R13)
-	MOVW	R1, 8(R13)
-	BL	runtime·newosproc0(SB)
-rr:
-	// Restore callee-save registers and return.
-	MOVB    runtime·goarm(SB), R11
-	CMP $6, R11
-	BLT skipfprest
-	MOVD	(32+8*1)(R13), F8
-	MOVD	(32+8*2)(R13), F9
-	MOVD	(32+8*3)(R13), F10
-	MOVD	(32+8*4)(R13), F11
-	MOVD	(32+8*5)(R13), F12
-	MOVD	(32+8*6)(R13), F13
-	MOVD	(32+8*7)(R13), F14
-	MOVD	(32+8*8)(R13), F15
-skipfprest:
-	MOVW	12(R13), R4
-	MOVW	16(R13), R5
-	MOVW	20(R13), R6
-	MOVW	24(R13), R7
-	MOVW	28(R13), R8
-	MOVW	32(R13), R11
-	RET
-
-TEXT _rt0_arm_linux_lib_go<>(SB),NOSPLIT,$8
-	MOVW	_rt0_arm_linux_lib_argc<>(SB), R0
-	MOVW	_rt0_arm_linux_lib_argv<>(SB), R1
-	MOVW	R0, 0(R13)
-	MOVW	R1, 4(R13)
-	B	runtime·rt0_go(SB)
-
-DATA _rt0_arm_linux_lib_argc<>(SB)/4,$0
-GLOBL _rt0_arm_linux_lib_argc<>(SB),NOPTR,$4
-DATA _rt0_arm_linux_lib_argv<>(SB)/4,$0
-GLOBL _rt0_arm_linux_lib_argv<>(SB),NOPTR,$4
 
 TEXT _rt0_arm_linux1(SB),NOSPLIT,$-4
 	// We first need to detect the kernel ABI, and warn the user
@@ -98,16 +22,12 @@ TEXT _rt0_arm_linux1(SB),NOSPLIT,$-4
 	// SIGILL is received.
 	// If you get a SIGILL here, you have the wrong kernel.
 
-	// Save argc and argv
+	// Save argc and argv (syscall will clobber at least R0).
 	MOVM.DB.W [R0-R1], (R13)
 
 	// do an EABI syscall
 	MOVW	$20, R7 // sys_getpid
 	SWI	$0 // this will trigger SIGILL on OABI systems
-	
+
+	MOVM.IA.W (R13), [R0-R1]
 	B	runtime·rt0_go(SB)
-
-TEXT main(SB),NOSPLIT,$-4
-	MOVW	$_rt0_arm_linux1(SB), R4
-	B		(R4)
-

src/runtime/rt0_nacl_arm.s

@@ -13,8 +13,4 @@
 TEXT _rt0_arm_nacl(SB),NOSPLIT,$-4
 	MOVW	8(R13), R0
 	MOVW	$12(R13), R1
-	MOVM.DB.W [R0-R1], (R13)
-	B	main(SB)
-
-TEXT main(SB),NOSPLIT,$0
 	B	runtime·rt0_go(SB)

src/runtime/rt0_netbsd_arm.s

@@ -4,8 +4,8 @@
 
 #include "textflag.h"
 
-TEXT _rt0_arm_netbsd(SB),NOSPLIT,$-4
+TEXT _rt0_arm_netbsd(SB),NOSPLIT,$0
-	MOVW	(R13), R0		// argc
+	B	_rt0_arm(SB)
-	MOVW	$4(R13), R1		// argv
+
-	MOVM.DB.W [R0-R1], (R13)
+TEXT _rt0_arm_netbsd_lib(SB),NOSPLIT,$0
-	B runtime·rt0_go(SB)
+	B	_rt0_arm_lib(SB)

src/runtime/rt0_openbsd_arm.s

@@ -4,8 +4,8 @@
 
 #include "textflag.h"
 
-TEXT _rt0_arm_openbsd(SB),NOSPLIT,$-4
+TEXT _rt0_arm_openbsd(SB),NOSPLIT,$0
-	MOVW	(R13), R0		// argc
+	B	_rt0_arm(SB)
-	MOVW	$4(R13), R1		// argv
+
-	MOVM.DB.W [R0-R1], (R13)
+TEXT _rt0_arm_openbsd_lib(SB),NOSPLIT,$0
-	B	runtime·rt0_go(SB)
+	B	_rt0_arm_lib(SB)

src/runtime/rt0_plan9_arm.s

@@ -10,8 +10,6 @@ TEXT _rt0_arm_plan9(SB),NOSPLIT,$-4
 	MOVW	R0, _tos(SB)
 	MOVW	0(R13), R0
 	MOVW	$4(R13), R1
-	MOVW.W	R1, -4(R13)
-	MOVW.W	R0, -4(R13)
 	B	runtime·rt0_go(SB)
 
 GLOBL _tos(SB), NOPTR, $4