[apple/xnu.git] / osfmk / i386 / start.s

/*
 * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
 *
 * @APPLE_LICENSE_HEADER_START@
 * 
 * The contents of this file constitute Original Code as defined in and
 * are subject to the Apple Public Source License Version 1.1 (the
 * "License").  You may not use this file except in compliance with the
 * License.  Please obtain a copy of the License at
 * http://www.apple.com/publicsource and read it before using this file.
 * 
 * This Original Code and all software distributed under the License are
 * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
 * License for the specific language governing rights and limitations
 * under the License.
 * 
 * @APPLE_LICENSE_HEADER_END@
 */
/*
 * @OSF_COPYRIGHT@
 */
/* 
 * Mach Operating System
 * Copyright (c) 1991,1990 Carnegie Mellon University
 * All Rights Reserved.
 * 
 * Permission to use, copy, modify and distribute this software and its
 * documentation is hereby granted, provided that both the copyright
 * notice and this permission notice appear in all copies of the
 * software, derivative works or modified versions, and any portions
 * thereof, and that both notices appear in supporting documentation.
 * 
 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
 * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
 * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
 * 
 * Carnegie Mellon requests users of this software to return to
 * 
 *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
 *  School of Computer Science
 *  Carnegie Mellon University
 *  Pittsburgh PA 15213-3890
 * 
 * any improvements or extensions that they make and grant Carnegie Mellon
 * the rights to redistribute these changes.
 */
/*
 */

#include <platforms.h>
#include <cpus.h>
#include <mach_kdb.h>

#include <i386/asm.h>
#include <i386/proc_reg.h>
#include <assym.s>

#if	NCPUS > 1

#define	CX(addr,reg)	addr(,reg,4)

#else

#define	CPU_NUMBER(reg)
#define	CX(addr,reg)	addr

#endif	/* NCPUS > 1 */

#include <i386/mp.h>

/*
 * GAS won't handle an intersegment jump with a relocatable offset.
 */
#define	LJMP(segment,address)	\
	.byte	0xea		;\
	.long	address		;\
	.word	segment


#define KVTOPHYS	(-KERNELBASE)
#define	KVTOLINEAR	LINEAR_KERNELBASE


#define	PA(addr)	(addr)+KVTOPHYS
#define	VA(addr)	(addr)-KVTOPHYS

	.data
	.align 	2
	.globl	EXT(_kick_buffer_)
EXT(_kick_buffer_):
	.long	1
	.long	3
	.set	.,.+16836
/*
 * Interrupt and bootup stack for initial processor.
 */
	.align	ALIGN
	.globl	EXT(intstack)
EXT(intstack):
	.set	., .+INTSTACK_SIZE
	.globl	EXT(eintstack)
EXT(eintstack:)

#if	NCPUS == 1
	.globl	EXT(int_stack_high)	/* all interrupt stacks */
EXT(int_stack_high):			/* must lie below this */
	.long	EXT(eintstack)		/* address */

	.globl	EXT(int_stack_top)	/* top of interrupt stack */
EXT(int_stack_top):
	.long	EXT(eintstack)
#endif

#if	MACH_KDB
/*
 * Kernel debugger stack for each processor.
 */
	.align	ALIGN
	.globl	EXT(db_stack_store)
EXT(db_stack_store):
	.set	., .+(INTSTACK_SIZE*NCPUS)

/*
 * Stack for last-ditch debugger task for each processor.
 */
	.align	ALIGN
	.globl	EXT(db_task_stack_store)
EXT(db_task_stack_store):
	.set	., .+(INTSTACK_SIZE*NCPUS)
#endif	/* MACH_KDB */

/*
 * per-processor kernel debugger stacks
 */
        .align  ALIGN
        .globl  EXT(kgdb_stack_store)
EXT(kgdb_stack_store):
        .set    ., .+(INTSTACK_SIZE*NCPUS)


/*
 * Pointers to GDT and IDT.  These contain linear addresses.
 */
	.align	ALIGN
	.globl	EXT(gdtptr)
LEXT(gdtptr)
	.word	Times(8,GDTSZ)-1
	.long	EXT(gdt)+KVTOLINEAR

	.align	ALIGN
	.globl	EXT(idtptr)
LEXT(idtptr)
	.word	Times(8,IDTSZ)-1
	.long	EXT(idt)+KVTOLINEAR

#if	NCPUS > 1
	.data
	/*
	 *	start_lock is very special.  We initialize the
	 *	lock at allocation time rather than at run-time.
	 *	Although start_lock should be an instance of a
	 *	hw_lock, we hand-code all manipulation of the lock
	 *	because the hw_lock code may require function calls;
	 *	and we'd rather not introduce another dependency on
	 *	a working stack at this point.
	 */
	.globl	EXT(start_lock)
EXT(start_lock):
	.long	0			/* synchronizes processor startup */

	.globl	EXT(master_is_up)
EXT(master_is_up):
	.long	0			/* 1 when OK for other processors */
					/* to start */
	.globl	EXT(mp_boot_pde)
EXT(mp_boot_pde):
	.long	0
#endif	/* NCPUS > 1 */

/*
 * All CPUs start here.
 *
 * Environment:
 *	protected mode, no paging, flat 32-bit address space.
 *	(Code/data/stack segments have base == 0, limit == 4G)
 */
	.text
	.align	ALIGN
	.globl	EXT(pstart)
	.globl	EXT(_start)
LEXT(_start)
LEXT(pstart)
	mov     %eax, %ebx		/* save pointer to kernbootstruct */
	mov	$0,%ax			/* fs must be zeroed; */
	mov	%ax,%fs			/* some bootstrappers don`t do this */
	mov	%ax,%gs

#if	NCPUS > 1
	jmp	1f
0:	cmpl	$0,PA(EXT(start_lock))
	jne	0b
1:	movb	$1,%eax
	xchgl	%eax,PA(EXT(start_lock)) /* locked */
	testl	%eax,%eax
	jnz	0b

	cmpl	$0,PA(EXT(master_is_up))	/* are we first? */
	jne	EXT(slave_start)		/* no -- system already up. */
	movl	$1,PA(EXT(master_is_up))	/* others become slaves */
#endif	/* NCPUS > 1 */

/*
 * Get startup parameters.
 */

#include <i386/AT386/asm_startup.h>

/*
 * Build initial page table directory and page tables.
 * %ebx holds first available physical address.
 */

	addl	$(NBPG-1),%ebx		/* round first avail physical addr */
	andl	$(-NBPG),%ebx		/* to machine page size */
	leal	-KVTOPHYS(%ebx),%eax	/* convert to virtual address */
	movl	%eax,PA(EXT(kpde))	/* save as kernel page table directory */
	movl	%ebx,%cr3		/* set physical address in CR3 now */

	movl	%ebx,%edi		/* clear page table directory */
	movl	$(PTES_PER_PAGE),%ecx	/* one page of ptes */
	xorl	%eax,%eax
	cld
	rep
	stosl				/* edi now points to next page */

/*
 * Use next few pages for page tables.
 */
	addl	$(KERNELBASEPDE),%ebx	/* point to pde for kernel base */
	movl	%edi,%esi		/* point to end of current pte page */

/*
 * Enter 1-1 mappings for kernel and for kernel page tables.
 */
	movl	$(INTEL_PTE_KERNEL),%eax /* set up pte prototype */
0:
	cmpl	%esi,%edi		/* at end of pte page? */
	jb	1f			/* if so: */
	movl	%edi,%edx		/*    get pte address (physical) */
	andl	$(-NBPG),%edx		/*    mask out offset in page */
	orl	$(INTEL_PTE_KERNEL),%edx /*   add pte bits */
	movl	%edx,(%ebx)		/*    set pde */
	addl	$4,%ebx			/*    point to next pde */
	movl	%edi,%esi		/*    point to */
	addl	$(NBPG),%esi		/*    end of new pte page */
1:
	movl	%eax,(%edi)		/* set pte */
	addl	$4,%edi			/* advance to next pte */
	addl	$(NBPG),%eax		/* advance to next phys page */
	cmpl	%edi,%eax		/* have we mapped this pte page yet? */
	jb	0b			/* loop if not */

/*
 * Zero rest of last pte page.
 */
	xor	%eax,%eax		/* don`t map yet */
2:	cmpl	%esi,%edi		/* at end of pte page? */
	jae	3f
	movl	%eax,(%edi)		/* zero mapping */
	addl	$4,%edi
	jmp	2b
3:

#if	NCPUS > 1
/*
 * Grab (waste?) another page for a bootstrap page directory
 * for the other CPUs.  We don't want the running CPUs to see
 * addresses 0..3fffff mapped 1-1.
 */
	movl	%edi,PA(EXT(mp_boot_pde)) /* save its physical address */
	movl	$(PTES_PER_PAGE),%ecx	/* and clear it */
	rep
	stosl
#endif	/* NCPUS > 1 */
	movl	%edi,PA(EXT(first_avail)) /* save first available phys addr */

/*
 * pmap_bootstrap will enter rest of mappings.
 */

/*
 * Fix initial descriptor tables.
 */
	lea	PA(EXT(idt)),%esi	/* fix IDT */
	movl	$(IDTSZ),%ecx
	movl	$(PA(fix_idt_ret)),%ebx
	jmp	fix_desc_common		/* (cannot use stack) */
fix_idt_ret:

	lea	PA(EXT(gdt)),%esi	/* fix GDT */
	movl	$(GDTSZ),%ecx
	movl	$(PA(fix_gdt_ret)),%ebx
	jmp	fix_desc_common		/* (cannot use stack) */
fix_gdt_ret:

	lea	PA(EXT(ldt)),%esi	/* fix LDT */
	movl	$(LDTSZ),%ecx
	movl	$(PA(fix_ldt_ret)),%ebx
	jmp	fix_desc_common		/* (cannot use stack) */
fix_ldt_ret:

/*
 * Turn on paging.
 */
	movl	%cr3,%eax		/* retrieve kernel PDE phys address */
	movl	KERNELBASEPDE(%eax),%ecx
	movl	%ecx,(%eax)		/* set it also as pte for location */
					/* 0..3fffff, so that the code */
					/* that enters paged mode is mapped */
					/* to identical addresses after */
					/* paged mode is enabled */

	addl	$4,%eax			/* 400000..7fffff */
	movl	KERNELBASEPDE(%eax),%ecx
	movl	%ecx,(%eax)

	movl	$ EXT(pag_start),%ebx	/* first paged code address */

	movl	%cr0,%eax
	orl	$(CR0_PG),%eax		/* set PG bit in CR0 */
	orl	$(CR0_WP),%eax
	movl	%eax,%cr0		/* to enable paging */

	jmp	*%ebx			/* flush prefetch queue */

/*
 * We are now paging, and can run with correct addresses.
 */
LEXT(pag_start)
	lgdt	EXT(gdtptr)		/* load GDT */
	lidt	EXT(idtptr)		/* load IDT */
	LJMP(KERNEL_CS,EXT(vstart))	/* switch to kernel code segment */

/*
 * Master is now running with correct addresses.
 */
LEXT(vstart)
	mov	$(KERNEL_DS),%ax	/* set kernel data segment */
	mov	%ax,%ds
	mov	%ax,%es
	mov	%ax,%ss
	mov	%ax,EXT(ktss)+TSS_SS0	/* set kernel stack segment */
					/* for traps to kernel */
#if	MACH_KDB
	mov	%ax,EXT(dbtss)+TSS_SS0	/* likewise for debug task switch */
	mov	%cr3,%eax		/* get PDBR into debug TSS */
	mov	%eax,EXT(dbtss)+TSS_PDBR
	mov	$0,%eax
#endif

	movw	$(KERNEL_LDT),%ax	/* get LDT segment */
	lldt	%ax			/* load LDT */
#if	MACH_KDB
	mov	%ax,EXT(ktss)+TSS_LDT	/* store LDT in two TSS, as well... */
	mov	%ax,EXT(dbtss)+TSS_LDT	/*   ...matters if we switch tasks */
#endif
	movw	$(KERNEL_TSS),%ax
	ltr	%ax			/* set up KTSS */

	mov	$ CPU_DATA,%ax
	mov	%ax,%gs

	lea	EXT(eintstack),%esp	/* switch to the bootup stack */
	call	EXT(i386_init)		/* run C code */
	/*NOTREACHED*/
	hlt

#if	NCPUS > 1
/*
 * master_up is used by the master cpu to signify that it is done
 * with the interrupt stack, etc. See the code in pstart and svstart
 * that this interlocks with.
 */
	.align	ALIGN
	.globl	EXT(master_up)
LEXT(master_up)
	pushl	%ebp			/* set up */
	movl	%esp,%ebp		/* stack frame */
	movl	$0,%ecx			/* unlock start_lock */
	xchgl	%ecx,EXT(start_lock)	/* since we are no longer using */
					/* bootstrap stack */
	leave				/* pop stack frame */
	ret

/*
 * We aren't the first.  Call slave_main to initialize the processor
 * and get Mach going on it.
 */
	.align	ALIGN
	.globl	EXT(slave_start)
LEXT(slave_start)
	cli				/* disable interrupts, so we don`t */
					/* need IDT for a while */
	movl	EXT(kpde)+KVTOPHYS,%ebx	/* get PDE virtual address */
	addl	$(KVTOPHYS),%ebx	/* convert to physical address */

	movl	PA(EXT(mp_boot_pde)),%edx /* point to the bootstrap PDE */
	movl	KERNELBASEPDE(%ebx),%eax
					/* point to pte for KERNELBASE */
	movl	%eax,KERNELBASEPDE(%edx)
					/* set in bootstrap PDE */
	movl	%eax,(%edx)		/* set it also as pte for location */
					/* 0..3fffff, so that the code */
					/* that enters paged mode is mapped */
					/* to identical addresses after */
					/* paged mode is enabled */
	movl	%edx,%cr3		/* use bootstrap PDE to enable paging */

	movl	$ EXT(spag_start),%edx	/* first paged code address */

	movl	%cr0,%eax
	orl	$(CR0_PG),%eax		/* set PG bit in CR0 */
	orl	$(CR0_WP),%eax
	movl	%eax,%cr0		/* to enable paging */

	jmp	*%edx			/* flush prefetch queue. */

/*
 * We are now paging, and can run with correct addresses.
 */
LEXT(spag_start)

	lgdt	EXT(gdtptr)		/* load GDT */
	lidt	EXT(idtptr)		/* load IDT */
	LJMP(KERNEL_CS,EXT(svstart))	/* switch to kernel code segment */

/*
 * Slave is now running with correct addresses.
 */
LEXT(svstart)
	mov	$(KERNEL_DS),%ax	/* set kernel data segment */
	mov	%ax,%ds
	mov	%ax,%es
	mov	%ax,%ss

	movl	%ebx,%cr3		/* switch to the real kernel PDE  */

	CPU_NUMBER(%eax)
	movl	CX(EXT(interrupt_stack),%eax),%esp /* get stack */
	addl	$(INTSTACK_SIZE),%esp	/* point to top */
	xorl	%ebp,%ebp		/* for completeness */

	movl	$0,%ecx			/* unlock start_lock */
	xchgl	%ecx,EXT(start_lock)	/* since we are no longer using */
					/* bootstrap stack */

/*
 * switch to the per-cpu descriptor tables
 */

	pushl	%eax			/* pass CPU number */
	call	EXT(mp_desc_init)	/* set up local table */
					/* pointer returned in %eax */
	subl	$4,%esp			/* get space to build pseudo-descriptors */
	
	CPU_NUMBER(%eax)
	movw	$(GDTSZ*8-1),0(%esp)	/* set GDT size in GDT descriptor */
	movl	CX(EXT(mp_gdt),%eax),%edx
	addl	$ KVTOLINEAR,%edx
	movl	%edx,2(%esp)		/* point to local GDT (linear address) */
	lgdt	0(%esp)			/* load new GDT */
	
	movw	$(IDTSZ*8-1),0(%esp)	/* set IDT size in IDT descriptor */
	movl	CX(EXT(mp_idt),%eax),%edx
	addl	$ KVTOLINEAR,%edx
	movl	%edx,2(%esp)		/* point to local IDT (linear address) */
	lidt	0(%esp)			/* load new IDT */
	
	movw	$(KERNEL_LDT),%ax	/* get LDT segment */
	lldt	%ax			/* load LDT */

	movw	$(KERNEL_TSS),%ax
	ltr	%ax			/* load new KTSS */

	mov	$ CPU_DATA,%ax
	mov	%ax,%gs

	call	EXT(slave_main)		/* start MACH */
	/*NOTREACHED*/
	hlt
#endif	/* NCPUS > 1 */

/*
 * Convert a descriptor from fake to real format.
 *
 * Calls from assembly code:
 * %ebx = return address (physical) CANNOT USE STACK
 * %esi	= descriptor table address (physical)
 * %ecx = number of descriptors
 *
 * Calls from C:
 * 0(%esp) = return address
 * 4(%esp) = descriptor table address (physical)
 * 8(%esp) = number of descriptors
 *
 * Fake descriptor format:
 *	bytes 0..3		base 31..0
 *	bytes 4..5		limit 15..0
 *	byte  6			access byte 2 | limit 19..16
 *	byte  7			access byte 1
 *
 * Real descriptor format:
 *	bytes 0..1		limit 15..0
 *	bytes 2..3		base 15..0
 *	byte  4			base 23..16
 *	byte  5			access byte 1
 *	byte  6			access byte 2 | limit 19..16
 *	byte  7			base 31..24
 *
 * Fake gate format:
 *	bytes 0..3		offset
 *	bytes 4..5		selector
 *	byte  6			word count << 4 (to match fake descriptor)
 *	byte  7			access byte 1
 *
 * Real gate format:
 *	bytes 0..1		offset 15..0
 *	bytes 2..3		selector
 *	byte  4			word count
 *	byte  5			access byte 1
 *	bytes 6..7		offset 31..16
 */
	.globl	EXT(fix_desc)
LEXT(fix_desc)
	pushl	%ebp			/* set up */
	movl	%esp,%ebp		/* stack frame */
	pushl	%esi			/* save registers */
	pushl	%ebx
	movl	B_ARG0,%esi		/* point to first descriptor */
	movl	B_ARG1,%ecx		/* get number of descriptors */
	lea	0f,%ebx			/* get return address */
	jmp	fix_desc_common		/* call internal routine */
0:	popl	%ebx			/* restore registers */
	popl	%esi
	leave				/* pop stack frame */
	ret				/* return */

fix_desc_common:
0:
	movw	6(%esi),%dx		/* get access byte */
	movb	%dh,%al
	andb	$0x14,%al
	cmpb	$0x04,%al		/* gate or descriptor? */
	je	1f

/* descriptor */
	movl	0(%esi),%eax		/* get base in eax */
	rol	$16,%eax		/* swap 15..0 with 31..16 */
					/* (15..0 in correct place) */
	movb	%al,%dl			/* combine bits 23..16 with ACC1 */
					/* in dh/dl */
	movb	%ah,7(%esi)		/* store bits 31..24 in correct place */
	movw	4(%esi),%ax		/* move limit bits 0..15 to word 0 */
	movl	%eax,0(%esi)		/* store (bytes 0..3 correct) */
	movw	%dx,4(%esi)		/* store bytes 4..5 */
	jmp	2f

/* gate */
1:
	movw	4(%esi),%ax		/* get selector */
	shrb	$4,%dl			/* shift word count to proper place */
	movw	%dx,4(%esi)		/* store word count / ACC1 */
	movw	2(%esi),%dx		/* get offset 16..31 */
	movw	%dx,6(%esi)		/* store in correct place */
	movw	%ax,2(%esi)		/* store selector in correct place */
2:
	addl	$8,%esi			/* bump to next descriptor */
	loop	0b			/* repeat */
	jmp	*%ebx			/* all done */

/*
 * put arg in kbd leds and spin a while
 * eats eax, ecx, edx
 */
#define	K_RDWR		0x60
#define	K_CMD_LEDS	0xed
#define	K_STATUS	0x64
#define	K_IBUF_FULL	0x02		/* input (to kbd) buffer full */
#define	K_OBUF_FULL	0x01		/* output (from kbd) buffer full */

ENTRY(set_kbd_leds)
	mov	S_ARG0,%cl		/* save led value */
	
0:	inb	$(K_STATUS),%al		/* get kbd status */
	testb	$(K_IBUF_FULL),%al	/* input busy? */
	jne	0b			/* loop until not */
	
	mov	$(K_CMD_LEDS),%al	/* K_CMD_LEDS */
	outb	%al,$(K_RDWR)		/* to kbd */

0:	inb	$(K_STATUS),%al		/* get kbd status */
	testb	$(K_OBUF_FULL),%al	/* output present? */
	je	0b			/* loop if not */

	inb	$(K_RDWR),%al		/* read status (and discard) */

0:	inb	$(K_STATUS),%al		/* get kbd status */
	testb	$(K_IBUF_FULL),%al	/* input busy? */
	jne	0b			/* loop until not */
	
	mov	%cl,%al			/* move led value */
	outb	%al,$(K_RDWR)		/* to kbd */

	movl	$10000000,%ecx		/* spin */
0:	nop
	nop
	loop	0b			/* a while */

	ret
Commit	Line	Data
1c79356b A	1	/*
	2	* Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
	3	*
	4	* @APPLE_LICENSE_HEADER_START@
	5	*
e5568f75 A	6	* The contents of this file constitute Original Code as defined in and
	7	* are subject to the Apple Public Source License Version 1.1 (the
	8	* "License"). You may not use this file except in compliance with the
	9	* License. Please obtain a copy of the License at
	10	* http://www.apple.com/publicsource and read it before using this file.
1c79356b	11	*
e5568f75 A	12	* This Original Code and all software distributed under the License are
e5568f75 A	13	* distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
1c79356b A	14	* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
1c79356b A	15	* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
e5568f75 A	16	* FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
	17	* License for the specific language governing rights and limitations
	18	* under the License.
1c79356b A	19	*
	20	* @APPLE_LICENSE_HEADER_END@
	21	*/
	22	/*
	23	* @OSF_COPYRIGHT@
	24	*/
	25	/*
	26	* Mach Operating System
	27	* Copyright (c) 1991,1990 Carnegie Mellon University
	28	* All Rights Reserved.
	29	*
	30	* Permission to use, copy, modify and distribute this software and its
	31	* documentation is hereby granted, provided that both the copyright
	32	* notice and this permission notice appear in all copies of the
	33	* software, derivative works or modified versions, and any portions
	34	* thereof, and that both notices appear in supporting documentation.
	35	*
	36	* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
	37	* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
	38	* ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
	39	*
	40	* Carnegie Mellon requests users of this software to return to
	41	*
	42	* Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
	43	* School of Computer Science
	44	* Carnegie Mellon University
	45	* Pittsburgh PA 15213-3890
	46	*
	47	* any improvements or extensions that they make and grant Carnegie Mellon
	48	* the rights to redistribute these changes.
	49	*/
	50	/*
	51	*/
	52
	53	#include <platforms.h>
	54	#include <cpus.h>
	55	#include <mach_kdb.h>
	56
	57	#include <i386/asm.h>
	58	#include <i386/proc_reg.h>
	59	#include <assym.s>
	60
	61	#if NCPUS > 1
	62
	63	#define CX(addr,reg) addr(,reg,4)
	64
	65	#else
	66
	67	#define CPU_NUMBER(reg)
	68	#define CX(addr,reg) addr
	69
	70	#endif /* NCPUS > 1 */
	71
55e303ae	72	#include <i386/mp.h>
1c79356b A	73
	74	/*
	75	* GAS won't handle an intersegment jump with a relocatable offset.
	76	*/
	77	#define LJMP(segment,address) \
	78	.byte 0xea ;\
	79	.long address ;\
	80	.word segment
	81
	82
	83
	84	#define KVTOPHYS (-KERNELBASE)
	85	#define KVTOLINEAR LINEAR_KERNELBASE
	86
	87
	88	#define PA(addr) (addr)+KVTOPHYS
	89	#define VA(addr) (addr)-KVTOPHYS
	90
	91	.data
	92	.align 2
	93	.globl EXT(_kick_buffer_)
	94	EXT(_kick_buffer_):
	95	.long 1
	96	.long 3
	97	.set .,.+16836
	98	/*
	99	* Interrupt and bootup stack for initial processor.
	100	*/
	101	.align ALIGN
	102	.globl EXT(intstack)
	103	EXT(intstack):
	104	.set ., .+INTSTACK_SIZE
	105	.globl EXT(eintstack)
	106	EXT(eintstack:)
	107
	108	#if NCPUS == 1
	109	.globl EXT(int_stack_high) /* all interrupt stacks */
	110	EXT(int_stack_high): /* must lie below this */
	111	.long EXT(eintstack) /* address */
	112
	113	.globl EXT(int_stack_top) /* top of interrupt stack */
	114	EXT(int_stack_top):
	115	.long EXT(eintstack)
	116	#endif
	117
	118	#if MACH_KDB
	119	/*
	120	* Kernel debugger stack for each processor.
	121	*/
	122	.align ALIGN
	123	.globl EXT(db_stack_store)
	124	EXT(db_stack_store):
	125	.set ., .+(INTSTACK_SIZE*NCPUS)
	126
	127	/*
	128	* Stack for last-ditch debugger task for each processor.
	129	*/
	130	.align ALIGN
	131	.globl EXT(db_task_stack_store)
	132	EXT(db_task_stack_store):
	133	.set ., .+(INTSTACK_SIZE*NCPUS)
	134	#endif /* MACH_KDB */
	135
	136	/*
137	* per-processor kernel debugger stacks
138	*/
139	.align ALIGN
140	.globl EXT(kgdb_stack_store)
141	EXT(kgdb_stack_store):
142	.set ., .+(INTSTACK_SIZE*NCPUS)
143
144
145	/*
146	* Pointers to GDT and IDT. These contain linear addresses.
147	*/
148	.align ALIGN
149	.globl EXT(gdtptr)
150	LEXT(gdtptr)
151	.word Times(8,GDTSZ)-1
152	.long EXT(gdt)+KVTOLINEAR
153
154	.align ALIGN
155	.globl EXT(idtptr)
156	LEXT(idtptr)
157	.word Times(8,IDTSZ)-1
158	.long EXT(idt)+KVTOLINEAR
159
160	#if NCPUS > 1
161	.data
162	/*
163	* start_lock is very special. We initialize the
164	* lock at allocation time rather than at run-time.
165	* Although start_lock should be an instance of a
166	* hw_lock, we hand-code all manipulation of the lock
167	* because the hw_lock code may require function calls;
168	* and we'd rather not introduce another dependency on
169	* a working stack at this point.
170	*/
171	.globl EXT(start_lock)
172	EXT(start_lock):
173	.long 0 /* synchronizes processor startup */
174
175	.globl EXT(master_is_up)
176	EXT(master_is_up):
177	.long 0 /* 1 when OK for other processors */
178	/* to start */
179	.globl EXT(mp_boot_pde)
180	EXT(mp_boot_pde):
181	.long 0
182	#endif /* NCPUS > 1 */
183
184	/*
185	* All CPUs start here.
186	*
187	* Environment:
188	* protected mode, no paging, flat 32-bit address space.
189	* (Code/data/stack segments have base == 0, limit == 4G)
190	*/
191	.text
192	.align ALIGN
193	.globl EXT(pstart)
194	.globl EXT(_start)
195	LEXT(_start)
196	LEXT(pstart)
55e303ae	197	mov %eax, %ebx /* save pointer to kernbootstruct */
1c79356b A	198	mov $0,%ax /* fs must be zeroed; */
	199	mov %ax,%fs /* some bootstrappers don`t do this */
	200	mov %ax,%gs
	201
	202	#if NCPUS > 1
	203	jmp 1f
	204	0: cmpl $0,PA(EXT(start_lock))
	205	jne 0b
	206	1: movb $1,%eax
	207	xchgl %eax,PA(EXT(start_lock)) /* locked */
	208	testl %eax,%eax
	209	jnz 0b
	210
	211	cmpl $0,PA(EXT(master_is_up)) /* are we first? */
	212	jne EXT(slave_start) /* no -- system already up. */
	213	movl $1,PA(EXT(master_is_up)) /* others become slaves */
	214	#endif /* NCPUS > 1 */
	215
	216	/*
	217	* Get startup parameters.
	218	*/
	219
	220	#include <i386/AT386/asm_startup.h>
	221
	222	/*
	223	* Build initial page table directory and page tables.
	224	* %ebx holds first available physical address.
	225	*/
	226
	227	addl $(NBPG-1),%ebx /* round first avail physical addr */
	228	andl $(-NBPG),%ebx /* to machine page size */
	229	leal -KVTOPHYS(%ebx),%eax /* convert to virtual address */
	230	movl %eax,PA(EXT(kpde)) /* save as kernel page table directory */
	231	movl %ebx,%cr3 /* set physical address in CR3 now */
	232
	233	movl %ebx,%edi /* clear page table directory */
	234	movl $(PTES_PER_PAGE),%ecx /* one page of ptes */
	235	xorl %eax,%eax
	236	cld
	237	rep
	238	stosl /* edi now points to next page */
	239
	240	/*
	241	* Use next few pages for page tables.
	242	*/
	243	addl $(KERNELBASEPDE),%ebx /* point to pde for kernel base */
	244	movl %edi,%esi /* point to end of current pte page */
	245
	246	/*
	247	* Enter 1-1 mappings for kernel and for kernel page tables.
	248	*/
	249	movl $(INTEL_PTE_KERNEL),%eax /* set up pte prototype */
	250	0:
	251	cmpl %esi,%edi /* at end of pte page? */
	252	jb 1f /* if so: */
	253	movl %edi,%edx /* get pte address (physical) */
	254	andl $(-NBPG),%edx /* mask out offset in page */
	255	orl $(INTEL_PTE_KERNEL),%edx /* add pte bits */
	256	movl %edx,(%ebx) /* set pde */
	257	addl $4,%ebx /* point to next pde */
	258	movl %edi,%esi /* point to */
	259	addl $(NBPG),%esi /* end of new pte page */
	260	1:
	261	movl %eax,(%edi) /* set pte */
262	addl $4,%edi /* advance to next pte */
263	addl $(NBPG),%eax /* advance to next phys page */
264	cmpl %edi,%eax /* have we mapped this pte page yet? */
265	jb 0b /* loop if not */
266
267	/*
268	* Zero rest of last pte page.
269	*/
270	xor %eax,%eax /* don`t map yet */
271	2: cmpl %esi,%edi /* at end of pte page? */
272	jae 3f
273	movl %eax,(%edi) /* zero mapping */
274	addl $4,%edi
275	jmp 2b
276	3:
277
278	#if NCPUS > 1
279	/*
280	* Grab (waste?) another page for a bootstrap page directory
281	* for the other CPUs. We don't want the running CPUs to see
282	* addresses 0..3fffff mapped 1-1.
283	*/
284	movl %edi,PA(EXT(mp_boot_pde)) /* save its physical address */
285	movl $(PTES_PER_PAGE),%ecx /* and clear it */
286	rep
287	stosl
288	#endif /* NCPUS > 1 */
289	movl %edi,PA(EXT(first_avail)) /* save first available phys addr */
290
291	/*
292	* pmap_bootstrap will enter rest of mappings.
293	*/
294
295	/*
296	* Fix initial descriptor tables.
297	*/
298	lea PA(EXT(idt)),%esi /* fix IDT */
299	movl $(IDTSZ),%ecx
300	movl $(PA(fix_idt_ret)),%ebx
301	jmp fix_desc_common /* (cannot use stack) */
302	fix_idt_ret:
303
304	lea PA(EXT(gdt)),%esi /* fix GDT */
305	movl $(GDTSZ),%ecx
306	movl $(PA(fix_gdt_ret)),%ebx
307	jmp fix_desc_common /* (cannot use stack) */
308	fix_gdt_ret:
309
310	lea PA(EXT(ldt)),%esi /* fix LDT */
311	movl $(LDTSZ),%ecx
312	movl $(PA(fix_ldt_ret)),%ebx
313	jmp fix_desc_common /* (cannot use stack) */
314	fix_ldt_ret:
315
316	/*
317	* Turn on paging.
318	*/
319	movl %cr3,%eax /* retrieve kernel PDE phys address */
320	movl KERNELBASEPDE(%eax),%ecx
321	movl %ecx,(%eax) /* set it also as pte for location */
322	/* 0..3fffff, so that the code */
323	/* that enters paged mode is mapped */
324	/* to identical addresses after */
325	/* paged mode is enabled */
326
327	addl $4,%eax /* 400000..7fffff */
328	movl KERNELBASEPDE(%eax),%ecx
329	movl %ecx,(%eax)
330
9bccf70c	331	movl $ EXT(pag_start),%ebx /* first paged code address */
1c79356b A	332
	333	movl %cr0,%eax
	334	orl $(CR0_PG),%eax /* set PG bit in CR0 */
	335	orl $(CR0_WP),%eax
	336	movl %eax,%cr0 /* to enable paging */
	337
	338	jmp %ebx / flush prefetch queue */
	339
	340	/*
	341	* We are now paging, and can run with correct addresses.
	342	*/
	343	LEXT(pag_start)
	344	lgdt EXT(gdtptr) /* load GDT */
	345	lidt EXT(idtptr) /* load IDT */
	346	LJMP(KERNEL_CS,EXT(vstart)) /* switch to kernel code segment */
	347
	348	/*
	349	* Master is now running with correct addresses.
	350	*/
	351	LEXT(vstart)
	352	mov $(KERNEL_DS),%ax /* set kernel data segment */
	353	mov %ax,%ds
	354	mov %ax,%es
	355	mov %ax,%ss
	356	mov %ax,EXT(ktss)+TSS_SS0 /* set kernel stack segment */
	357	/* for traps to kernel */
	358	#if MACH_KDB
	359	mov %ax,EXT(dbtss)+TSS_SS0 /* likewise for debug task switch */
	360	mov %cr3,%eax /* get PDBR into debug TSS */
	361	mov %eax,EXT(dbtss)+TSS_PDBR
	362	mov $0,%eax
	363	#endif
	364
	365	movw $(KERNEL_LDT),%ax /* get LDT segment */
	366	lldt %ax /* load LDT */
	367	#if MACH_KDB
	368	mov %ax,EXT(ktss)+TSS_LDT /* store LDT in two TSS, as well... */
	369	mov %ax,EXT(dbtss)+TSS_LDT /* ...matters if we switch tasks */
	370	#endif
	371	movw $(KERNEL_TSS),%ax
	372	ltr %ax /* set up KTSS */
	373
9bccf70c	374	mov $ CPU_DATA,%ax
1c79356b A	375	mov %ax,%gs
	376
	377	lea EXT(eintstack),%esp /* switch to the bootup stack */
55e303ae	378	call EXT(i386_init) /* run C code */
1c79356b A	379	/NOTREACHED/
	380	hlt
	381
	382	#if NCPUS > 1
	383	/*
	384	* master_up is used by the master cpu to signify that it is done
	385	* with the interrupt stack, etc. See the code in pstart and svstart
	386	* that this interlocks with.
	387	*/
	388	.align ALIGN
	389	.globl EXT(master_up)
	390	LEXT(master_up)
	391	pushl %ebp /* set up */
	392	movl %esp,%ebp /* stack frame */
	393	movl $0,%ecx /* unlock start_lock */
	394	xchgl %ecx,EXT(start_lock) /* since we are no longer using */
	395	/* bootstrap stack */
	396	leave /* pop stack frame */
	397	ret
	398
	399	/*
	400	* We aren't the first. Call slave_main to initialize the processor
	401	* and get Mach going on it.
	402	*/
	403	.align ALIGN
	404	.globl EXT(slave_start)
	405	LEXT(slave_start)
	406	cli /* disable interrupts, so we don`t */
	407	/* need IDT for a while */
	408	movl EXT(kpde)+KVTOPHYS,%ebx /* get PDE virtual address */
	409	addl $(KVTOPHYS),%ebx /* convert to physical address */
	410
	411	movl PA(EXT(mp_boot_pde)),%edx /* point to the bootstrap PDE */
	412	movl KERNELBASEPDE(%ebx),%eax
	413	/* point to pte for KERNELBASE */
	414	movl %eax,KERNELBASEPDE(%edx)
	415	/* set in bootstrap PDE */
	416	movl %eax,(%edx) /* set it also as pte for location */
	417	/* 0..3fffff, so that the code */
	418	/* that enters paged mode is mapped */
	419	/* to identical addresses after */
	420	/* paged mode is enabled */
	421	movl %edx,%cr3 /* use bootstrap PDE to enable paging */
	422
9bccf70c	423	movl $ EXT(spag_start),%edx /* first paged code address */
1c79356b A	424
	425	movl %cr0,%eax
	426	orl $(CR0_PG),%eax /* set PG bit in CR0 */
	427	orl $(CR0_WP),%eax
	428	movl %eax,%cr0 /* to enable paging */
	429
	430	jmp %edx / flush prefetch queue. */
	431
	432	/*
	433	* We are now paging, and can run with correct addresses.
	434	*/
	435	LEXT(spag_start)
	436
	437	lgdt EXT(gdtptr) /* load GDT */
	438	lidt EXT(idtptr) /* load IDT */
	439	LJMP(KERNEL_CS,EXT(svstart)) /* switch to kernel code segment */
	440
	441	/*
	442	* Slave is now running with correct addresses.
	443	*/
	444	LEXT(svstart)
	445	mov $(KERNEL_DS),%ax /* set kernel data segment */
	446	mov %ax,%ds
	447	mov %ax,%es
	448	mov %ax,%ss
	449
	450	movl %ebx,%cr3 /* switch to the real kernel PDE */
	451
	452	CPU_NUMBER(%eax)
	453	movl CX(EXT(interrupt_stack),%eax),%esp /* get stack */
	454	addl $(INTSTACK_SIZE),%esp /* point to top */
	455	xorl %ebp,%ebp /* for completeness */
	456
	457	movl $0,%ecx /* unlock start_lock */
	458	xchgl %ecx,EXT(start_lock) /* since we are no longer using */
	459	/* bootstrap stack */
	460
	461	/*
	462	* switch to the per-cpu descriptor tables
	463	*/
	464
	465	pushl %eax /* pass CPU number */
	466	call EXT(mp_desc_init) /* set up local table */
	467	/* pointer returned in %eax */
	468	subl $4,%esp /* get space to build pseudo-descriptors */
	469
	470	CPU_NUMBER(%eax)
	471	movw $(GDTSZ8-1),0(%esp) / set GDT size in GDT descriptor */
	472	movl CX(EXT(mp_gdt),%eax),%edx
9bccf70c	473	addl $ KVTOLINEAR,%edx
1c79356b A	474	movl %edx,2(%esp) /* point to local GDT (linear address) */
	475	lgdt 0(%esp) /* load new GDT */
	476
	477	movw $(IDTSZ8-1),0(%esp) / set IDT size in IDT descriptor */
	478	movl CX(EXT(mp_idt),%eax),%edx
9bccf70c	479	addl $ KVTOLINEAR,%edx
1c79356b A	480	movl %edx,2(%esp) /* point to local IDT (linear address) */
	481	lidt 0(%esp) /* load new IDT */
	482
55e303ae A	483	movw $(KERNEL_LDT),%ax /* get LDT segment */
	484	lldt %ax /* load LDT */
	485
1c79356b A	486	movw $(KERNEL_TSS),%ax
	487	ltr %ax /* load new KTSS */
	488
9bccf70c	489	mov $ CPU_DATA,%ax
1c79356b A	490	mov %ax,%gs
	491
	492	call EXT(slave_main) /* start MACH */
	493	/NOTREACHED/
	494	hlt
	495	#endif /* NCPUS > 1 */
	496
	497	/*
	498	* Convert a descriptor from fake to real format.
	499	*
	500	* Calls from assembly code:
	501	* %ebx = return address (physical) CANNOT USE STACK
	502	* %esi = descriptor table address (physical)
	503	* %ecx = number of descriptors
	504	*
	505	* Calls from C:
	506	* 0(%esp) = return address
	507	* 4(%esp) = descriptor table address (physical)
	508	* 8(%esp) = number of descriptors
	509	*
	510	* Fake descriptor format:
	511	* bytes 0..3 base 31..0
	512	* bytes 4..5 limit 15..0
	513	* byte 6 access byte 2 \| limit 19..16
	514	* byte 7 access byte 1
	515	*
	516	* Real descriptor format:
	517	* bytes 0..1 limit 15..0
	518	* bytes 2..3 base 15..0
	519	* byte 4 base 23..16
	520	* byte 5 access byte 1
	521	* byte 6 access byte 2 \| limit 19..16
	522	* byte 7 base 31..24
	523	*
	524	* Fake gate format:
	525	* bytes 0..3 offset
	526	* bytes 4..5 selector
	527	* byte 6 word count << 4 (to match fake descriptor)
	528	* byte 7 access byte 1
	529	*
	530	* Real gate format:
	531	* bytes 0..1 offset 15..0
	532	* bytes 2..3 selector
	533	* byte 4 word count
	534	* byte 5 access byte 1
	535	* bytes 6..7 offset 31..16
	536	*/
	537	.globl EXT(fix_desc)
	538	LEXT(fix_desc)
	539	pushl %ebp /* set up */
	540	movl %esp,%ebp /* stack frame */
	541	pushl %esi /* save registers */
	542	pushl %ebx
	543	movl B_ARG0,%esi /* point to first descriptor */
	544	movl B_ARG1,%ecx /* get number of descriptors */
	545	lea 0f,%ebx /* get return address */
	546	jmp fix_desc_common /* call internal routine */
	547	0: popl %ebx /* restore registers */
	548	popl %esi
	549	leave /* pop stack frame */
	550	ret /* return */
	551
	552	fix_desc_common:
	553	0:
554	movw 6(%esi),%dx /* get access byte */
555	movb %dh,%al
556	andb $0x14,%al
557	cmpb $0x04,%al /* gate or descriptor? */
558	je 1f
559
560	/* descriptor */
561	movl 0(%esi),%eax /* get base in eax */
562	rol $16,%eax /* swap 15..0 with 31..16 */
563	/* (15..0 in correct place) */
564	movb %al,%dl /* combine bits 23..16 with ACC1 */
565	/* in dh/dl */
566	movb %ah,7(%esi) /* store bits 31..24 in correct place */
567	movw 4(%esi),%ax /* move limit bits 0..15 to word 0 */
568	movl %eax,0(%esi) /* store (bytes 0..3 correct) */
569	movw %dx,4(%esi) /* store bytes 4..5 */
570	jmp 2f
571
572	/* gate */
573	1:
574	movw 4(%esi),%ax /* get selector */
575	shrb $4,%dl /* shift word count to proper place */
576	movw %dx,4(%esi) /* store word count / ACC1 */
577	movw 2(%esi),%dx /* get offset 16..31 */
578	movw %dx,6(%esi) /* store in correct place */
579	movw %ax,2(%esi) /* store selector in correct place */
580	2:
581	addl $8,%esi /* bump to next descriptor */
582	loop 0b /* repeat */
583	jmp %ebx / all done */
584
585	/*
586	* put arg in kbd leds and spin a while
587	* eats eax, ecx, edx
588	*/
589	#define K_RDWR 0x60
590	#define K_CMD_LEDS 0xed
591	#define K_STATUS 0x64
592	#define K_IBUF_FULL 0x02 /* input (to kbd) buffer full */
593	#define K_OBUF_FULL 0x01 /* output (from kbd) buffer full */
594
595	ENTRY(set_kbd_leds)
596	mov S_ARG0,%cl /* save led value */
597
598	0: inb $(K_STATUS),%al /* get kbd status */
599	testb $(K_IBUF_FULL),%al /* input busy? */
600	jne 0b /* loop until not */
601
602	mov $(K_CMD_LEDS),%al /* K_CMD_LEDS */
603	outb %al,$(K_RDWR) /* to kbd */
604
605	0: inb $(K_STATUS),%al /* get kbd status */
606	testb $(K_OBUF_FULL),%al /* output present? */
607	je 0b /* loop if not */
608
609	inb $(K_RDWR),%al /* read status (and discard) */
610
611	0: inb $(K_STATUS),%al /* get kbd status */
612	testb $(K_IBUF_FULL),%al /* input busy? */
613	jne 0b /* loop until not */
614
615	mov %cl,%al /* move led value */
616	outb %al,$(K_RDWR) /* to kbd */
617
618	movl $10000000,%ecx /* spin */
619	0: nop
620	nop
621	loop 0b /* a while */
622
623	ret