// Copyright 2009 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.

#include "asm_386.h"

TEXT _rt0_386(SB),7,$0
	// Linux, Windows start the FPU in extended double precision.
	// Other operating systems use double precision.
	// Change to double precision to match them,
	// and to match other hardware that only has double.
	PUSHL $0x27F
	FLDCW	0(SP)
	POPL AX

	// copy arguments forward on an even stack
	MOVL	0(SP), AX		// argc
	LEAL	4(SP), BX		// argv
	SUBL	$128, SP		// plenty of scratch
	ANDL	$~15, SP
	MOVL	AX, 120(SP)		// save argc, argv away
	MOVL	BX, 124(SP)

	// set default stack bounds.
	// initcgo may update stackguard.
	MOVL	$runtime·g0(SB), BP
	LEAL	(-64*1024+104)(SP), BX
	MOVL	BX, g_stackguard(BP)
	MOVL	SP, g_stackbase(BP)
	
	// if there is an initcgo, call it to let it
	// initialize and to set up GS.  if not,
	// we set up GS ourselves.
	MOVL	initcgo(SB), AX
	TESTL	AX, AX
	JZ	needtls
	PUSHL	BP
	CALL	AX
	POPL	BP
	// skip runtime·ldt0setup(SB) and tls test after initcgo for non-windows
	CMPL runtime·iswindows(SB), $0
	JEQ ok
needtls:
	// skip runtime·ldt0setup(SB) and tls test on Plan 9 in all cases
	CMPL	runtime·isplan9(SB), $1
	JEQ	ok

	// set up %gs
	CALL	runtime·ldt0setup(SB)

	// store through it, to make sure it works
	get_tls(BX)
	MOVL	$0x123, g(BX)
	MOVL	runtime·tls0(SB), AX
	CMPL	AX, $0x123
	JEQ	ok
	MOVL	AX, 0	// abort
ok:
	// set up m and g "registers"
	get_tls(BX)
	LEAL	runtime·g0(SB), CX
	MOVL	CX, g(BX)
	LEAL	runtime·m0(SB), AX
	MOVL	AX, m(BX)

	// save m->g0 = g0
	MOVL	CX, m_g0(AX)

	CALL	runtime·emptyfunc(SB)	// fault if stack check is wrong

	// convention is D is always cleared
	CLD

	CALL	runtime·check(SB)

	// saved argc, argv
	MOVL	120(SP), AX
	MOVL	AX, 0(SP)
	MOVL	124(SP), AX
	MOVL	AX, 4(SP)
	CALL	runtime·args(SB)
	CALL	runtime·osinit(SB)
	CALL	runtime·schedinit(SB)

	// create a new goroutine to start program
	PUSHL	$runtime·main(SB)	// entry
	PUSHL	$0	// arg size
	CALL	runtime·newproc(SB)
	POPL	AX
	POPL	AX

	// start this M
	CALL	runtime·mstart(SB)

	INT $3
	RET

TEXT runtime·breakpoint(SB),7,$0
	INT $3
	RET

/*
 *  go-routine
 */

// void gosave(Gobuf*)
// save state in Gobuf; setjmp
TEXT runtime·gosave(SB), 7, $0
	MOVL	4(SP), AX		// gobuf
	LEAL	4(SP), BX		// caller's SP
	MOVL	BX, gobuf_sp(AX)
	MOVL	0(SP), BX		// caller's PC
	MOVL	BX, gobuf_pc(AX)
	get_tls(CX)
	MOVL	g(CX), BX
	MOVL	BX, gobuf_g(AX)
	RET

// void gogo(Gobuf*, uintptr)
// restore state from Gobuf; longjmp
TEXT runtime·gogo(SB), 7, $0
	MOVL	8(SP), AX		// return 2nd arg
	MOVL	4(SP), BX		// gobuf
	MOVL	gobuf_g(BX), DX
	MOVL	0(DX), CX		// make sure g != nil
	get_tls(CX)
	MOVL	DX, g(CX)
	MOVL	gobuf_sp(BX), SP	// restore SP
	MOVL	gobuf_pc(BX), BX
	JMP	BX

// void gogocall(Gobuf*, void (*fn)(void))
// restore state from Gobuf but then call fn.
// (call fn, returning to state in Gobuf)
TEXT runtime·gogocall(SB), 7, $0
	MOVL	8(SP), AX		// fn
	MOVL	4(SP), BX		// gobuf
	MOVL	gobuf_g(BX), DX
	get_tls(CX)
	MOVL	DX, g(CX)
	MOVL	0(DX), CX		// make sure g != nil
	MOVL	gobuf_sp(BX), SP	// restore SP
	MOVL	gobuf_pc(BX), BX
	PUSHL	BX
	JMP	AX
	POPL	BX	// not reached

// void mcall(void (*fn)(G*))
// Switch to m->g0's stack, call fn(g).
// Fn must never return.  It should gogo(&g->sched)
// to keep running g.
TEXT runtime·mcall(SB), 7, $0
	MOVL	fn+0(FP), DI
	
	get_tls(CX)
	MOVL	g(CX), AX	// save state in g->gobuf
	MOVL	0(SP), BX	// caller's PC
	MOVL	BX, (g_sched+gobuf_pc)(AX)
	LEAL	4(SP), BX	// caller's SP
	MOVL	BX, (g_sched+gobuf_sp)(AX)
	MOVL	AX, (g_sched+gobuf_g)(AX)

	// switch to m->g0 & its stack, call fn
	MOVL	m(CX), BX
	MOVL	m_g0(BX), SI
	CMPL	SI, AX	// if g == m->g0 call badmcall
	JNE	2(PC)
	CALL	runtime·badmcall(SB)
	MOVL	SI, g(CX)	// g = m->g0
	MOVL	(g_sched+gobuf_sp)(SI), SP	// sp = m->g0->gobuf.sp
	PUSHL	AX
	CALL	DI
	POPL	AX
	CALL	runtime·badmcall2(SB)
	RET

/*
 * support for morestack
 */

// Called during function prolog when more stack is needed.
TEXT runtime·morestack(SB),7,$0
	// Cannot grow scheduler stack (m->g0).
	get_tls(CX)
	MOVL	m(CX), BX
	MOVL	m_g0(BX), SI
	CMPL	g(CX), SI
	JNE	2(PC)
	INT	$3

	// frame size in DX
	// arg size in AX
	// Save in m.
	MOVL	DX, m_moreframesize(BX)
	MOVL	AX, m_moreargsize(BX)

	// Called from f.
	// Set m->morebuf to f's caller.
	MOVL	4(SP), DI	// f's caller's PC
	MOVL	DI, (m_morebuf+gobuf_pc)(BX)
	LEAL	8(SP), CX	// f's caller's SP
	MOVL	CX, (m_morebuf+gobuf_sp)(BX)
	MOVL	CX, m_moreargp(BX)
	get_tls(CX)
	MOVL	g(CX), SI
	MOVL	SI, (m_morebuf+gobuf_g)(BX)

	// Set m->morepc to f's PC.
	MOVL	0(SP), AX
	MOVL	AX, m_morepc(BX)

	// Call newstack on m->g0's stack.
	MOVL	m_g0(BX), BP
	MOVL	BP, g(CX)
	MOVL	(g_sched+gobuf_sp)(BP), AX
	MOVL	-4(AX), BX	// fault if CALL would, before smashing SP
	MOVL	AX, SP
	CALL	runtime·newstack(SB)
	MOVL	$0, 0x1003	// crash if newstack returns
	RET

// Called from reflection library.  Mimics morestack,
// reuses stack growth code to create a frame
// with the desired args running the desired function.
//
// func call(fn *byte, arg *byte, argsize uint32).
TEXT reflect·call(SB), 7, $0
	get_tls(CX)
	MOVL	m(CX), BX

	// Save our caller's state as the PC and SP to
	// restore when returning from f.
	MOVL	0(SP), AX	// our caller's PC
	MOVL	AX, (m_morebuf+gobuf_pc)(BX)
	LEAL	4(SP), AX	// our caller's SP
	MOVL	AX, (m_morebuf+gobuf_sp)(BX)
	MOVL	g(CX), AX
	MOVL	AX, (m_morebuf+gobuf_g)(BX)

	// Set up morestack arguments to call f on a new stack.
	// We set f's frame size to 1, as a hint to newstack
	// that this is a call from reflect·call.
	// If it turns out that f needs a larger frame than
	// the default stack, f's usual stack growth prolog will
	// allocate a new segment (and recopy the arguments).
	MOVL	4(SP), AX	// fn
	MOVL	8(SP), DX	// arg frame
	MOVL	12(SP), CX	// arg size

	MOVL	AX, m_morepc(BX)	// f's PC
	MOVL	DX, m_moreargp(BX)	// f's argument pointer
	MOVL	CX, m_moreargsize(BX)	// f's argument size
	MOVL	$1, m_moreframesize(BX)	// f's frame size

	// Call newstack on m->g0's stack.
	MOVL	m_g0(BX), BP
	get_tls(CX)
	MOVL	BP, g(CX)
	MOVL	(g_sched+gobuf_sp)(BP), SP
	CALL	runtime·newstack(SB)
	MOVL	$0, 0x1103	// crash if newstack returns
	RET


// Return point when leaving stack.
TEXT runtime·lessstack(SB), 7, $0
	// Save return value in m->cret
	get_tls(CX)
	MOVL	m(CX), BX
	MOVL	AX, m_cret(BX)

	// Call oldstack on m->g0's stack.
	MOVL	m_g0(BX), BP
	MOVL	BP, g(CX)
	MOVL	(g_sched+gobuf_sp)(BP), SP
	CALL	runtime·oldstack(SB)
	MOVL	$0, 0x1004	// crash if oldstack returns
	RET


// bool cas(int32 *val, int32 old, int32 new)
// Atomically:
//	if(*val == old){
//		*val = new;
//		return 1;
//	}else
//		return 0;
TEXT runtime·cas(SB), 7, $0
	MOVL	4(SP), BX
	MOVL	8(SP), AX
	MOVL	12(SP), CX
	LOCK
	CMPXCHGL	CX, 0(BX)
	JZ 3(PC)
	MOVL	$0, AX
	RET
	MOVL	$1, AX
	RET

// bool casp(void **p, void *old, void *new)
// Atomically:
//	if(*p == old){
//		*p = new;
//		return 1;
//	}else
//		return 0;
TEXT runtime·casp(SB), 7, $0
	MOVL	4(SP), BX
	MOVL	8(SP), AX
	MOVL	12(SP), CX
	LOCK
	CMPXCHGL	CX, 0(BX)
	JZ 3(PC)
	MOVL	$0, AX
	RET
	MOVL	$1, AX
	RET

// uint32 xadd(uint32 volatile *val, int32 delta)
// Atomically:
//	*val += delta;
//	return *val;
TEXT runtime·xadd(SB), 7, $0
	MOVL	4(SP), BX
	MOVL	8(SP), AX
	MOVL	AX, CX
	LOCK
	XADDL	AX, 0(BX)
	ADDL	CX, AX
	RET

TEXT runtime·xchg(SB), 7, $0
	MOVL	4(SP), BX
	MOVL	8(SP), AX
	XCHGL	AX, 0(BX)
	RET

TEXT runtime·procyield(SB),7,$0
	MOVL	4(SP), AX
again:
	PAUSE
	SUBL	$1, AX
	JNZ	again
	RET

TEXT runtime·atomicstorep(SB), 7, $0
	MOVL	4(SP), BX
	MOVL	8(SP), AX
	XCHGL	AX, 0(BX)
	RET

TEXT runtime·atomicstore(SB), 7, $0
	MOVL	4(SP), BX
	MOVL	8(SP), AX
	XCHGL	AX, 0(BX)
	RET

// void jmpdefer(fn, sp);
// called from deferreturn.
// 1. pop the caller
// 2. sub 5 bytes from the callers return
// 3. jmp to the argument
TEXT runtime·jmpdefer(SB), 7, $0
	MOVL	4(SP), AX	// fn
	MOVL	8(SP), BX	// caller sp
	LEAL	-4(BX), SP	// caller sp after CALL
	SUBL	$5, (SP)	// return to CALL again
	JMP	AX	// but first run the deferred function

// Dummy function to use in saved gobuf.PC,
// to match SP pointing at a return address.
// The gobuf.PC is unused by the contortions here
// but setting it to return will make the traceback code work.
TEXT return<>(SB),7,$0
	RET

// asmcgocall(void(*fn)(void*), void *arg)
// Call fn(arg) on the scheduler stack,
// aligned appropriately for the gcc ABI.
// See cgocall.c for more details.
TEXT runtime·asmcgocall(SB),7,$0
	MOVL	fn+0(FP), AX
	MOVL	arg+4(FP), BX
	MOVL	SP, DX

	// Figure out if we need to switch to m->g0 stack.
	// We get called to create new OS threads too, and those
	// come in on the m->g0 stack already.
	get_tls(CX)
	MOVL	m(CX), BP
	MOVL	m_g0(BP), SI
	MOVL	g(CX), DI
	CMPL	SI, DI
	JEQ	6(PC)
	MOVL	SP, (g_sched+gobuf_sp)(DI)
	MOVL	$return<>(SB), (g_sched+gobuf_pc)(DI)
	MOVL	DI, (g_sched+gobuf_g)(DI)
	MOVL	SI, g(CX)
	MOVL	(g_sched+gobuf_sp)(SI), SP

	// Now on a scheduling stack (a pthread-created stack).
	SUBL	$32, SP
	ANDL	$~15, SP	// alignment, perhaps unnecessary
	MOVL	DI, 8(SP)	// save g
	MOVL	DX, 4(SP)	// save SP
	MOVL	BX, 0(SP)	// first argument in x86-32 ABI
	CALL	AX

	// Restore registers, g, stack pointer.
	get_tls(CX)
	MOVL	8(SP), DI
	MOVL	DI, g(CX)
	MOVL	4(SP), SP
	RET

// cgocallback(void (*fn)(void*), void *frame, uintptr framesize)
// See cgocall.c for more details.
TEXT runtime·cgocallback(SB),7,$12
	MOVL	fn+0(FP), AX
	MOVL	frame+4(FP), BX
	MOVL	framesize+8(FP), DX

	// Save current m->g0->sched.sp on stack and then set it to SP.
	get_tls(CX)
	MOVL	m(CX), BP
	MOVL	m_g0(BP), SI
	PUSHL	(g_sched+gobuf_sp)(SI)
	MOVL	SP, (g_sched+gobuf_sp)(SI)

	// Switch to m->curg stack and call runtime.cgocallbackg
	// with the three arguments.  Because we are taking over
	// the execution of m->curg but *not* resuming what had
	// been running, we need to save that information (m->curg->gobuf)
	// so that we can restore it when we're done. 
	// We can restore m->curg->gobuf.sp easily, because calling
	// runtime.cgocallbackg leaves SP unchanged upon return.
	// To save m->curg->gobuf.pc, we push it onto the stack.
	// This has the added benefit that it looks to the traceback
	// routine like cgocallbackg is going to return to that
	// PC (because we defined cgocallbackg to have
	// a frame size of 12, the same amount that we use below),
	// so that the traceback will seamlessly trace back into
	// the earlier calls.
	MOVL	m_curg(BP), SI
	MOVL	SI, g(CX)
	MOVL	(g_sched+gobuf_sp)(SI), DI  // prepare stack as DI

	// Push gobuf.pc
	MOVL	(g_sched+gobuf_pc)(SI), BP
	SUBL	$4, DI
	MOVL	BP, 0(DI)

	// Push arguments to cgocallbackg.
	// Frame size here must match the frame size above
	// to trick traceback routines into doing the right thing.
	SUBL	$12, DI
	MOVL	AX, 0(DI)
	MOVL	BX, 4(DI)
	MOVL	DX, 8(DI)
	
	// Switch stack and make the call.
	MOVL	DI, SP
	CALL	runtime·cgocallbackg(SB)

	// Restore g->gobuf (== m->curg->gobuf) from saved values.
	get_tls(CX)
	MOVL	g(CX), SI
	MOVL	12(SP), BP
	MOVL	BP, (g_sched+gobuf_pc)(SI)
	LEAL	(12+4)(SP), DI
	MOVL	DI, (g_sched+gobuf_sp)(SI)

	// Switch back to m->g0's stack and restore m->g0->sched.sp.
	// (Unlike m->curg, the g0 goroutine never uses sched.pc,
	// so we do not have to restore it.)
	MOVL	m(CX), BP
	MOVL	m_g0(BP), SI
	MOVL	SI, g(CX)
	MOVL	(g_sched+gobuf_sp)(SI), SP
	POPL	(g_sched+gobuf_sp)(SI)

	// Done!
	RET

// check that SP is in range [g->stackbase, g->stackguard)
TEXT runtime·stackcheck(SB), 7, $0
	get_tls(CX)
	MOVL	g(CX), AX
	CMPL	g_stackbase(AX), SP
	JHI	2(PC)
	INT	$3
	CMPL	SP, g_stackguard(AX)
	JHI	2(PC)
	INT	$3
	RET

TEXT runtime·memclr(SB),7,$0
	MOVL	4(SP), DI		// arg 1 addr
	MOVL	8(SP), CX		// arg 2 count
	MOVL	CX, BX
	ANDL	$3, BX
	SHRL	$2, CX
	MOVL	$0, AX
	CLD
	REP
	STOSL
	MOVL	BX, CX
	REP
	STOSB
	RET

TEXT runtime·getcallerpc(SB),7,$0
	MOVL	x+0(FP),AX		// addr of first arg
	MOVL	-4(AX),AX		// get calling pc
	RET

TEXT runtime·setcallerpc(SB),7,$0
	MOVL	x+0(FP),AX		// addr of first arg
	MOVL	x+4(FP), BX
	MOVL	BX, -4(AX)		// set calling pc
	RET

TEXT runtime·getcallersp(SB), 7, $0
	MOVL	sp+0(FP), AX
	RET

TEXT runtime·ldt0setup(SB),7,$16
	// set up ldt 7 to point at tls0
	// ldt 1 would be fine on Linux, but on OS X, 7 is as low as we can go.
	// the entry number is just a hint.  setldt will set up GS with what it used.
	MOVL	$7, 0(SP)
	LEAL	runtime·tls0(SB), AX
	MOVL	AX, 4(SP)
	MOVL	$32, 8(SP)	// sizeof(tls array)
	CALL	runtime·setldt(SB)
	RET

TEXT runtime·emptyfunc(SB),0,$0
	RET

TEXT runtime·abort(SB),7,$0
	INT $0x3

GLOBL runtime·tls0(SB), $32