runtime: unify 386 entry point code

Unify the 386 entry point code as much as possible.

The main function could not be unified because on Windows 386 it is
called _main. Putting main in asm_386.s caused multiple definition
errors when using the external linker.

Add the _lib entry point to various operating systems.  A future CL
will enable c-archive/c-shared mode for those targets.

Fix _rt0_386_windows_lib_go--it was passing arguments as though it
were amd64.

Change-Id: Ic73f1c95cdbcbea87f633f4a29bbc218a5db4f58
Reviewed-on: https://go-review.googlesource.com/70530
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>

src/cmd/vet/all/whitelist/386.txt

@@ -2,6 +2,10 @@
 
 runtime/asm_ARCHSUFF.s: [GOARCH] cannot check cross-package assembly function: Compare is in package bytes
 
+// startup code uses non-standard calling convention and intentionally
+// omits args.
+runtime/asm_386.s: [386] rt0_go: use of 4(SP) points beyond argument frame
+
 // reflect trampolines intentionally omit arg size. Same for morestack.
 runtime/asm_386.s: [386] morestack: use of 4(SP) points beyond argument frame
 runtime/asm_386.s: [386] morestack: use of 8(SP) points beyond argument frame
@@ -21,7 +25,3 @@ runtime/asm_386.s: [386] uint32tofloat64: function uint32tofloat64 missing Go de
 runtime/asm_386.s: [386] float64touint32: function float64touint32 missing Go declaration
 
 runtime/asm_386.s: [386] stackcheck: function stackcheck missing Go declaration
-
-// Clearer using FP than SP, but that requires named offsets.
-runtime/asm_386.s: [386] rt0_go: unknown variable argc
-runtime/asm_386.s: [386] rt0_go: unknown variable argv

src/runtime/asm_386.s

@@ -7,10 +7,87 @@
 #include "funcdata.h"
 #include "textflag.h"
 
+// _rt0_386 is common startup code for most 386 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_386(SB),NOSPLIT,$8
+	MOVL	8(SP), AX	// argc
+	LEAL	12(SP), BX	// argv
+	MOVL	AX, 0(SP)
+	MOVL	BX, 4(SP)
+	JMP	runtime·rt0_go(SB)
+
+// _rt0_386_lib is common startup code for most 386 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 on the stack following the
+// usual C ABI.
+TEXT _rt0_386_lib(SB),NOSPLIT,$0
+	PUSHL	BP
+	MOVL	SP, BP
+	PUSHL	BX
+	PUSHL	SI
+	PUSHL	DI
+
+	MOVL	8(BP), AX
+	MOVL	AX, _rt0_386_lib_argc<>(SB)
+	MOVL	12(BP), AX
+	MOVL	AX, _rt0_386_lib_argv<>(SB)
+
+	// Synchronous initialization.
+	CALL	runtime·libpreinit(SB)
+
+	SUBL	$8, SP
+
+	// Create a new thread to do the runtime initialization.
+	MOVL	_cgo_sys_thread_create(SB), AX
+	TESTL	AX, AX
+	JZ	nocgo
+	MOVL	$_rt0_386_lib_go(SB), BX
+	MOVL	BX, 0(SP)
+	MOVL	$0, 4(SP)
+
+	// TODO: We are calling a C function here so we should be
+	// aligning the stack.
+
+	CALL	AX
+	JMP	restore
+
+nocgo:
+	MOVL	$0x800000, 0(SP)                    // stacksize = 8192KB
+	MOVL	$_rt0_386_lib_go(SB), AX
+	MOVL	AX, 4(SP)                           // fn
+	CALL	runtime·newosproc0(SB)
+
+restore:
+	ADDL	$8, SP
+	POPL	DI
+	POPL	SI
+	POPL	BX
+	POPL	BP
+	RET
+
+// _rt0_386_lib_go initializes the Go runtime.
+// This is started in a separate thread by _rt0_386_lib.
+TEXT _rt0_386_lib_go(SB),NOSPLIT,$8
+	MOVL	_rt0_386_lib_argc<>(SB), AX
+	MOVL	AX, 0(SP)
+	MOVL	_rt0_386_lib_argv<>(SB), AX
+	MOVL	AX, 4(SP)
+	JMP	runtime·rt0_go(SB)
+
+DATA _rt0_386_lib_argc<>(SB)/4, $0
+GLOBL _rt0_386_lib_argc<>(SB),NOPTR, $4
+DATA _rt0_386_lib_argv<>(SB)/4, $0
+GLOBL _rt0_386_lib_argv<>(SB),NOPTR, $4
+
 TEXT runtime·rt0_go(SB),NOSPLIT,$0
-	// copy arguments forward on an even stack
-	MOVL	argc+0(FP), AX
-	MOVL	argv+4(FP), BX
+	// Copy arguments forward on an even stack.
+	// Users of this function jump to it, they don't call it.
+	MOVL	0(SP), AX
+	MOVL	4(SP), BX
 	SUBL	$128, SP		// plenty of scratch
 	ANDL	$~15, SP
 	MOVL	AX, 120(SP)		// save argc, argv away

src/runtime/rt0_android_386.s

@@ -4,21 +4,13 @@
 
 #include "textflag.h"
 
-TEXT _rt0_386_android(SB),NOSPLIT,$8
-	MOVL	8(SP), AX  // argc
-	LEAL	12(SP), BX  // argv
-	MOVL	AX, 0(SP)
-	MOVL	BX, 4(SP)
-	CALL	main(SB)
-	INT	$3
+TEXT _rt0_386_android(SB),NOSPLIT,$0
+	JMP	_rt0_386(SB)
 
 TEXT _rt0_386_android_lib(SB),NOSPLIT,$0
 	PUSHL	$_rt0_386_android_argv(SB)  // argv
 	PUSHL	$1  // argc
-	CALL	_rt0_386_linux_lib(SB)
-	POPL	AX
-	POPL	AX
-	RET
+	JMP	_rt0_386_lib(SB)
 
 DATA _rt0_386_android_argv+0x00(SB)/4,$_rt0_386_android_argv0(SB)
 DATA _rt0_386_android_argv+0x04(SB)/4,$0  // argv terminate

src/runtime/rt0_darwin_386.s

@@ -4,72 +4,14 @@
 
 #include "textflag.h"
 
-TEXT _rt0_386_darwin(SB),NOSPLIT,$8
-	MOVL	8(SP), AX
-	LEAL	12(SP), BX
-	MOVL	AX, 0(SP)
-	MOVL	BX, 4(SP)
-	CALL	main(SB)
-	INT	$3
+TEXT _rt0_386_darwin(SB),NOSPLIT,$0
+	JMP	_rt0_386(SB)
 
-// With -buildmode=c-archive, this symbol is called from a global constructor.
 TEXT _rt0_386_darwin_lib(SB),NOSPLIT,$0
-	PUSHL	BP
-	MOVL	SP, BP
-	PUSHL	BX
-	PUSHL	SI
-	PUSHL	DI
-
-	MOVL	8(BP), AX
-	MOVL	AX, _rt0_386_darwin_lib_argc<>(SB)
-	MOVL	12(BP), AX
-	MOVL	AX, _rt0_386_darwin_lib_argv<>(SB)
-
-	// Synchronous initialization.
-	MOVL	$runtime·libpreinit(SB), AX
-	CALL	AX
-
-	SUBL	$12, SP
-
-	// Create a new thread to do the runtime initialization and return.
-	MOVL	_cgo_sys_thread_create(SB), AX
-	TESTL	AX, AX
-	JZ	nocgo
-	MOVL	$_rt0_386_darwin_lib_go(SB), BX
-	MOVL	BX, 0(SP)
-	MOVL	$0, 4(SP)
-	CALL	AX
-	JMP     restore
-
-nocgo:
-	MOVL	$0x800000, 0(SP)               // stacksize = 8192KB
-	MOVL	$_rt0_386_darwin_lib_go(SB), AX
-	MOVL	AX, 4(SP)                      // fn
-	MOVL	$0, 8(SP)                      // fnarg
-	MOVL	$runtime·newosproc0(SB), AX
-	CALL	AX
-
-restore:
-	ADDL	$12, SP
-	POPL	DI
-	POPL	SI
-	POPL	BX
-	POPL	BP
-	RET
-
-TEXT _rt0_386_darwin_lib_go(SB),NOSPLIT,$12
-	MOVL	_rt0_386_darwin_lib_argc<>(SB), AX
-	MOVL	AX, 0(SP)
-	MOVL	_rt0_386_darwin_lib_argv<>(SB), AX
-	MOVL	AX, 4(SP)
-	MOVL	$runtime·rt0_go(SB), AX
-	CALL	AX
-	RET
-
-DATA _rt0_386_darwin_lib_argc<>(SB)/4, $0
-GLOBL _rt0_386_darwin_lib_argc<>(SB),NOPTR, $4
-DATA _rt0_386_darwin_lib_argv<>(SB)/4, $0
-GLOBL _rt0_386_darwin_lib_argv<>(SB),NOPTR, $4
+	JMP	_rt0_386_lib(SB)
 
 TEXT main(SB),NOSPLIT,$0
+	// Remove the return address from the stack.
+	// rt0_go doesn't expect it to be there.
+	ADDL	$4, SP
 	JMP	runtime·rt0_go(SB)

src/runtime/rt0_freebsd_386.s

@@ -4,13 +4,14 @@
 
 #include "textflag.h"
 
-TEXT _rt0_386_freebsd(SB),NOSPLIT,$8
-	MOVL	8(SP), AX
-	LEAL	12(SP), BX
-	MOVL	AX, 0(SP)
-	MOVL	BX, 4(SP)
-	CALL	main(SB)
-	INT	$3
+TEXT _rt0_386_freebsd(SB),NOSPLIT,$0
+	JMP	_rt0_386(SB)
+
+TEXT _rt0_386_freebsd_lib(SB),NOSPLIT,$0
+	JMP	_rt0_386_lib(SB)
 
 TEXT main(SB),NOSPLIT,$0
+	// Remove the return address from the stack.
+	// rt0_go doesn't expect it to be there.
+	ADDL	$4, SP
 	JMP	runtime·rt0_go(SB)

src/runtime/rt0_linux_386.s

@@ -4,72 +4,14 @@
 
 #include "textflag.h"
 
-TEXT _rt0_386_linux(SB),NOSPLIT,$8
-	MOVL	8(SP), AX
-	LEAL	12(SP), BX
-	MOVL	AX, 0(SP)
-	MOVL	BX, 4(SP)
-	CALL	main(SB)
-	INT	$3
+TEXT _rt0_386_linux(SB),NOSPLIT,$0
+	JMP	_rt0_386(SB)
 
-// When building with -buildmode=c-shared, this symbol is called when the shared
-// library is loaded.
 TEXT _rt0_386_linux_lib(SB),NOSPLIT,$0
-	PUSHL	BP
-	MOVL	SP, BP
-	PUSHL	BX
-	PUSHL	SI
-	PUSHL	DI
-
-	MOVL	8(BP), AX
-	MOVL	AX, _rt0_386_linux_lib_argc<>(SB)
-	MOVL	12(BP), AX
-	MOVL	AX, _rt0_386_linux_lib_argv<>(SB)
-
-	// Synchronous initialization.
-	MOVL	$runtime·libpreinit(SB), AX
-	CALL	AX
-
-	SUBL	$8, SP
-
-	// Create a new thread to do the runtime initialization.
-	MOVL	_cgo_sys_thread_create(SB), AX
-	TESTL	AX, AX
-	JZ	nocgo
-	MOVL	$_rt0_386_linux_lib_go(SB), BX
-	MOVL	BX, 0(SP)
-	MOVL	$0, 4(SP)
-	CALL	AX
-	JMP	restore
-
-nocgo:
-	MOVL	$0x800000, 0(SP)                    // stacksize = 8192KB
-	MOVL	$_rt0_386_linux_lib_go(SB), AX
-	MOVL	AX, 4(SP)                           // fn
-	MOVL	$runtime·newosproc0(SB), AX
-	CALL	AX
-
-restore:
-	ADDL	$8, SP
-	POPL	DI
-	POPL	SI
-	POPL	BX
-	POPL	BP
-	RET
-
-TEXT _rt0_386_linux_lib_go(SB),NOSPLIT,$12
-	MOVL	_rt0_386_linux_lib_argc<>(SB), AX
-	MOVL	AX, 0(SP)
-	MOVL	_rt0_386_linux_lib_argv<>(SB), AX
-	MOVL	AX, 4(SP)
-	MOVL	$runtime·rt0_go(SB), AX
-	CALL	AX
-	RET
-
-DATA _rt0_386_linux_lib_argc<>(SB)/4, $0
-GLOBL _rt0_386_linux_lib_argc<>(SB),NOPTR, $4
-DATA _rt0_386_linux_lib_argv<>(SB)/4, $0
-GLOBL _rt0_386_linux_lib_argv<>(SB),NOPTR, $4
+	JMP	_rt0_386_lib(SB)
 
 TEXT main(SB),NOSPLIT,$0
+	// Remove the return address from the stack.
+	// rt0_go doesn't expect it to be there.
+	ADDL	$4, SP
 	JMP	runtime·rt0_go(SB)

src/runtime/rt0_nacl_386.s

@@ -15,8 +15,10 @@ TEXT _rt0_386_nacl(SB),NOSPLIT,$8
 	LEAL	argv+16(FP), BX
 	MOVL	AX, 0(SP)
 	MOVL	BX, 4(SP)
-	CALL	main(SB)
-	INT	$3
+	JMP	runtime·rt0_go(SB)
 
 TEXT main(SB),NOSPLIT,$0
+	// Remove the return address from the stack.
+	// rt0_go doesn't expect it to be there.
+	ADDL	$4, SP
 	JMP	runtime·rt0_go(SB)

src/runtime/rt0_netbsd_386.s

@@ -4,13 +4,14 @@
 
 #include "textflag.h"
 
-TEXT _rt0_386_netbsd(SB),NOSPLIT,$8
-	MOVL	8(SP), AX
-	LEAL	12(SP), BX
-	MOVL	AX, 0(SP)
-	MOVL	BX, 4(SP)
-	CALL	main(SB)
-	INT	$3
+TEXT _rt0_386_netbsd(SB),NOSPLIT,$0
+	JMP	_rt0_386(SB)
+
+TEXT _rt0_386_netbsd_lib(SB),NOSPLIT,$0
+	JMP	_rt0_386_lib(SB)
 
 TEXT main(SB),NOSPLIT,$0
+	// Remove the return address from the stack.
+	// rt0_go doesn't expect it to be there.
+	ADDL	$4, SP
 	JMP	runtime·rt0_go(SB)

src/runtime/rt0_openbsd_386.s

@@ -4,13 +4,14 @@
 
 #include "textflag.h"
 
-TEXT _rt0_386_openbsd(SB),NOSPLIT,$8
-	MOVL	8(SP), AX
-	LEAL	12(SP), BX
-	MOVL	AX, 0(SP)
-	MOVL	BX, 4(SP)
-	CALL	main(SB)
-	INT	$3
+TEXT _rt0_386_openbsd(SB),NOSPLIT,$0
+	JMP	_rt0_386(SB)
+
+TEXT _rt0_386_openbsd_lib(SB),NOSPLIT,$0
+	JMP	_rt0_386_lib(SB)
 
 TEXT main(SB),NOSPLIT,$0
+	// Remove the return address from the stack.
+	// rt0_go doesn't expect it to be there.
+	ADDL	$4, SP
 	JMP	runtime·rt0_go(SB)

src/runtime/rt0_plan9_386.s

@@ -14,7 +14,7 @@ TEXT _rt0_386_plan9(SB),NOSPLIT,$12
 	MOVL	AX, 0(SP)
 	LEAL	inargv+0(FP), AX
 	MOVL	AX, 4(SP)
-	CALL	runtime·rt0_go(SB)
+	JMP	runtime·rt0_go(SB)
 
 GLOBL _tos(SB), NOPTR, $4
 GLOBL _privates(SB), NOPTR, $4

src/runtime/rt0_windows_386.s

@@ -4,13 +4,8 @@
 
 #include "textflag.h"
 
-TEXT _rt0_386_windows(SB),NOSPLIT,$12
-	MOVL	12(SP), AX
-	LEAL	16(SP), BX
-	MOVL	AX, 4(SP)
-	MOVL	BX, 8(SP)
-	MOVL	$-1, 0(SP) // return PC for main
-	JMP	_main(SB)
+TEXT _rt0_386_windows(SB),NOSPLIT,$0
+	JMP	_rt0_386(SB)
 
 // When building with -buildmode=(c-shared or c-archive), this
 // symbol is called. For dynamic libraries it is called when the
@@ -41,10 +36,12 @@ TEXT _rt0_386_windows_lib(SB),NOSPLIT,$0x1C
 	RET
 
 TEXT _rt0_386_windows_lib_go(SB),NOSPLIT,$0
-	MOVL  $0, DI
-	MOVL	$0, SI
-	MOVL	$runtime·rt0_go(SB), AX
-	JMP	AX
+	PUSHL	$0
+	PUSHL	$0
+	JMP	runtime·rt0_go(SB)
 
 TEXT _main(SB),NOSPLIT,$0
+	// Remove the return address from the stack.
+	// rt0_go doesn't expect it to be there.
+	ADDL	$4, SP
 	JMP	runtime·rt0_go(SB)