mpboot.S revision 126246
1/*-
2 * Copyright (c) 2003 Peter Wemm
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * $FreeBSD: head/sys/amd64/amd64/mpboot.S 126246 2004-02-25 23:12:39Z peter $
27 */
28
29#include <machine/asmacros.h>		/* miscellaneous asm macros */
30#include <machine/specialreg.h>
31
32#include "assym.s"
33
34	.data				/* So we can modify it */
35
36	.p2align 4,0
37	.globl	mptramp_start
38mptramp_start:
39	.code16
40	/*
41	 * The AP enters here in response to the startup IPI.
42	 * We are in real mode. %cs is the only segment register set.
43	 */
44	cli				/* make sure no interrupts */
45	mov	%cs, %ax		/* copy %cs to %ds.  Remember these */
46	mov	%ax, %ds		/* are offsets rather than selectors */
47	mov	%ax, %ss
48
49	/*
50	 * Find relocation base and patch the gdt descript and ljmp targets
51	 */
52	xorl	%ebx,%ebx
53	mov	%cs, %bx
54	sall	$4, %ebx		/* %ebx is now our relocation base */
55	orl	%ebx, lgdt_desc-mptramp_start+2
56	orl	%ebx, jmp_32-mptramp_start+2
57	orl	%ebx, jmp_64-mptramp_start+1
58
59	/*
60	 * Load the descriptor table pointer.  We'll need it when running
61	 * in 16 bit protected mode.
62	 */
63	lgdt	lgdt_desc-mptramp_start
64
65	/* Enable protected mode */
66	movl	$CR0_PE, %eax
67	mov	%eax, %cr0
68
69	/*
70	 * Now execute a far jump to turn on protected mode.  This
71	 * causes the segment registers to turn into selectors and causes
72	 * %cs to be loaded from the gdt.
73	 *
74	 * The following instruction is:
75	 * ljmpl $bootcode-gdt, $protmode-mptramp_start
76	 * but gas cannot assemble that.  And besides, we patch the targets
77	 * in early startup and its a little clearer what we are patching.
78	 */
79jmp_32:
80	.byte	0x66			/* size override to 32 bits */
81	.byte	0xea			/* opcode for far jump */
82	.long	protmode-mptramp_start	/* offset in segment */
83	.word	bootcode-gdt		/* index in gdt for 32 bit code */
84
85	/*
86	 * At this point, we are running in 32 bit legacy protected mode.
87	 */
88	.code32
89protmode:
90	mov	$bootdata-gdt, %eax
91	mov	%ax, %ds
92
93	/* Turn on the PAE, PSE and PGE bits for when paging is enabled */
94	mov	%cr4, %eax
95	orl	$(CR4_PAE | CR4_PSE), %eax
96	mov	%eax, %cr4
97
98	/*
99	 * Enable EFER.LME so that we get long mode when all the prereqs are
100	 * in place.  In this case, it turns on when CR0_PG is finally enabled.
101	 * Pick up a few other EFER bits that we'll use need we're here.
102	 */
103	movl	$MSR_EFER, %ecx
104	rdmsr
105#if 0	/* not till we test the NX cpuid bits */
106	orl	$EFER_LME | EFER_SCE | EFER_NXE, %eax
107#else
108	orl	$EFER_LME | EFER_SCE, %eax
109#endif
110	wrmsr
111
112	/*
113	 * Point to the embedded page tables for startup.  Note that this
114	 * only gets accessed after we're actually in 64 bit mode, however
115	 * we can only set the bottom 32 bits of %cr3 in this state.  This
116	 * means we are required to use a temporary page table that is below
117	 * the 4GB limit.  %ebx is still our relocation base.  We could just
118	 * subtract 3 * PAGE_SIZE, but that would be too easy.
119	 */
120	leal	mptramp_pagetables-mptramp_start(%ebx),%eax
121	movl	(%eax), %eax
122	mov	%eax, %cr3
123
124	/*
125	 * Finally, switch to long bit mode by enabling paging.  We have
126	 * to be very careful here because all the segmentation disappears
127	 * out from underneath us.  The spec says we can depend on the
128	 * subsequent pipelined branch to execute, but *only if* everthing
129	 * is still identity mapped.  If any mappings change, the pipeline
130	 * will flush.
131	 */
132	mov	%cr0, %eax
133	orl	$CR0_PG, %eax
134	mov	%eax, %cr0
135
136	/*
137	 * At this point paging is enabled, and we are in "compatability" mode.
138	 * We do another far jump to reload %cs with the 64 bit selector.
139	 * %cr3 points to a 4-level page table page.
140	 * We cannot yet jump all the way to the kernel because we can only
141	 * specify a 32 bit linear address.  So, yet another trampoline.
142	 *
143	 * The following instruction is:
144	 * ljmp $kernelcode-gdt, $tramp_64-mptramp_start
145	 * but gas cannot assemble that.  And besides, we patch the targets
146	 * in early startup and its a little clearer what we are patching.
147	 */
148jmp_64:
149	.byte	0xea			/* opcode for far jump */
150	.long	tramp_64-mptramp_start	/* offset in segment */
151	.word	kernelcode-gdt		/* index in gdt for 64 bit code */
152
153	/*
154	 * Yeehar!  We're running in 64 bit mode!  We can mostly ignore our
155	 * segment registers, and get on with it.
156	 * Note that we are running at the correct virtual address, but with
157	 * a 1:1 1GB mirrored mapping over entire address space.  We had better
158	 * switch to a real %cr3 promptly so that we can get to the direct map
159	 * space. Remember that jmp is relative and that we've been relocated,
160	 * so use an indirect jump.
161	 */
162	.code64
163tramp_64:
164	movabsq	$entry_64,%rax		/* 64 bit immediate load */
165	jmp	*%rax
166
167	.p2align 4,0
168gdt:
169	/*
170	 * All segment descriptor tables start with a null descriptor
171	 */
172	.long	0x00000000
173	.long	0x00000000
174
175	/*
176	 * This is the 64 bit long mode code descriptor.  There is no
177	 * 64 bit data descriptor.
178	 */
179kernelcode:
180	.long	0x00000000
181	.long	0x00209800
182
183	/*
184	 * This is the descriptor for the 32 bit boot code.
185	 * %cs:  +A, +R, -C, DPL=0, +P, +D, +G
186	 * Accessed, Readable, Present, 32 bit, 4G granularity
187	 */
188bootcode:
189	.long	0x0000ffff
190	.long	0x00cf9b00
191
192	/*
193	 * This is the descriptor for the 32 bit boot data.
194	 * We load it into %ds and %ss.  The bits for each selector
195	 * are interpreted slightly differently.
196	 * %ds:  +A, +W, -E, DPL=0, +P, +D, +G
197	 * %ss:  +A, +W, -E, DPL=0, +P, +B, +G
198	 * Accessed, Writeable, Expand up, Present, 32 bit, 4GB
199	 * For %ds, +D means 'default operand size is 32 bit'.
200	 * For %ss, +B means the stack register is %esp rather than %sp.
201	 */
202bootdata:
203	.long	0x0000ffff
204	.long	0x00cf9300
205
206gdtend:
207
208	/*
209	 * The address of our page table pages that the boot code
210	 * uses to trampoline up to kernel address space.
211	 */
212	.globl	mptramp_pagetables
213mptramp_pagetables:
214	.long	0
215
216	/*
217	 * The pseudo descriptor for lgdt to use.
218	 */
219lgdt_desc:
220	.word	gdtend-gdt		/* Length */
221	.long	gdt-mptramp_start	/* Offset plus %ds << 4 */
222
223	.globl	mptramp_end
224mptramp_end:
225
226	/*
227	 * From here on down is executed in the kernel .text section.
228	 *
229	 * Load a real %cr3 that has all the direct map stuff and switches
230	 * off the 1GB replicated mirror.  Load a stack pointer and jump
231	 * into AP startup code in C.
232	 */
233	.text
234	.code64
235	.p2align 4,0
236entry_64:
237	movq	KPML4phys, %rax
238	movq	%rax, %cr3
239	movq	bootSTK, %rsp
240	jmp	init_secondary
241