xref: /openbmc/linux/arch/x86/boot/compressed/head_64.S (revision cd5d5810)
1/*
2 *  linux/boot/head.S
3 *
4 *  Copyright (C) 1991, 1992, 1993  Linus Torvalds
5 */
6
7/*
8 *  head.S contains the 32-bit startup code.
9 *
10 * NOTE!!! Startup happens at absolute address 0x00001000, which is also where
11 * the page directory will exist. The startup code will be overwritten by
12 * the page directory. [According to comments etc elsewhere on a compressed
13 * kernel it will end up at 0x1000 + 1Mb I hope so as I assume this. - AC]
14 *
15 * Page 0 is deliberately kept safe, since System Management Mode code in
16 * laptops may need to access the BIOS data stored there.  This is also
17 * useful for future device drivers that either access the BIOS via VM86
18 * mode.
19 */
20
21/*
22 * High loaded stuff by Hans Lermen & Werner Almesberger, Feb. 1996
23 */
24	.code32
25	.text
26
27#include <linux/init.h>
28#include <linux/linkage.h>
29#include <asm/segment.h>
30#include <asm/boot.h>
31#include <asm/msr.h>
32#include <asm/processor-flags.h>
33#include <asm/asm-offsets.h>
34
35	__HEAD
36	.code32
37ENTRY(startup_32)
38	/*
39	 * 32bit entry is 0 and it is ABI so immutable!
40	 * If we come here directly from a bootloader,
41	 * kernel(text+data+bss+brk) ramdisk, zero_page, command line
42	 * all need to be under the 4G limit.
43	 */
44	cld
45	/*
46	 * Test KEEP_SEGMENTS flag to see if the bootloader is asking
47	 * us to not reload segments
48	 */
49	testb $(1<<6), BP_loadflags(%esi)
50	jnz 1f
51
52	cli
53	movl	$(__BOOT_DS), %eax
54	movl	%eax, %ds
55	movl	%eax, %es
56	movl	%eax, %ss
571:
58
59/*
60 * Calculate the delta between where we were compiled to run
61 * at and where we were actually loaded at.  This can only be done
62 * with a short local call on x86.  Nothing  else will tell us what
63 * address we are running at.  The reserved chunk of the real-mode
64 * data at 0x1e4 (defined as a scratch field) are used as the stack
65 * for this calculation. Only 4 bytes are needed.
66 */
67	leal	(BP_scratch+4)(%esi), %esp
68	call	1f
691:	popl	%ebp
70	subl	$1b, %ebp
71
72/* setup a stack and make sure cpu supports long mode. */
73	movl	$boot_stack_end, %eax
74	addl	%ebp, %eax
75	movl	%eax, %esp
76
77	call	verify_cpu
78	testl	%eax, %eax
79	jnz	no_longmode
80
81/*
82 * Compute the delta between where we were compiled to run at
83 * and where the code will actually run at.
84 *
85 * %ebp contains the address we are loaded at by the boot loader and %ebx
86 * contains the address where we should move the kernel image temporarily
87 * for safe in-place decompression.
88 */
89
90#ifdef CONFIG_RELOCATABLE
91	movl	%ebp, %ebx
92	movl	BP_kernel_alignment(%esi), %eax
93	decl	%eax
94	addl	%eax, %ebx
95	notl	%eax
96	andl	%eax, %ebx
97#else
98	movl	$LOAD_PHYSICAL_ADDR, %ebx
99#endif
100
101	/* Target address to relocate to for decompression */
102	addl	$z_extract_offset, %ebx
103
104/*
105 * Prepare for entering 64 bit mode
106 */
107
108	/* Load new GDT with the 64bit segments using 32bit descriptor */
109	leal	gdt(%ebp), %eax
110	movl	%eax, gdt+2(%ebp)
111	lgdt	gdt(%ebp)
112
113	/* Enable PAE mode */
114	movl	$(X86_CR4_PAE), %eax
115	movl	%eax, %cr4
116
117 /*
118  * Build early 4G boot pagetable
119  */
120	/* Initialize Page tables to 0 */
121	leal	pgtable(%ebx), %edi
122	xorl	%eax, %eax
123	movl	$((4096*6)/4), %ecx
124	rep	stosl
125
126	/* Build Level 4 */
127	leal	pgtable + 0(%ebx), %edi
128	leal	0x1007 (%edi), %eax
129	movl	%eax, 0(%edi)
130
131	/* Build Level 3 */
132	leal	pgtable + 0x1000(%ebx), %edi
133	leal	0x1007(%edi), %eax
134	movl	$4, %ecx
1351:	movl	%eax, 0x00(%edi)
136	addl	$0x00001000, %eax
137	addl	$8, %edi
138	decl	%ecx
139	jnz	1b
140
141	/* Build Level 2 */
142	leal	pgtable + 0x2000(%ebx), %edi
143	movl	$0x00000183, %eax
144	movl	$2048, %ecx
1451:	movl	%eax, 0(%edi)
146	addl	$0x00200000, %eax
147	addl	$8, %edi
148	decl	%ecx
149	jnz	1b
150
151	/* Enable the boot page tables */
152	leal	pgtable(%ebx), %eax
153	movl	%eax, %cr3
154
155	/* Enable Long mode in EFER (Extended Feature Enable Register) */
156	movl	$MSR_EFER, %ecx
157	rdmsr
158	btsl	$_EFER_LME, %eax
159	wrmsr
160
161	/* After gdt is loaded */
162	xorl	%eax, %eax
163	lldt	%ax
164	movl    $0x20, %eax
165	ltr	%ax
166
167	/*
168	 * Setup for the jump to 64bit mode
169	 *
170	 * When the jump is performend we will be in long mode but
171	 * in 32bit compatibility mode with EFER.LME = 1, CS.L = 0, CS.D = 1
172	 * (and in turn EFER.LMA = 1).	To jump into 64bit mode we use
173	 * the new gdt/idt that has __KERNEL_CS with CS.L = 1.
174	 * We place all of the values on our mini stack so lret can
175	 * used to perform that far jump.
176	 */
177	pushl	$__KERNEL_CS
178	leal	startup_64(%ebp), %eax
179	pushl	%eax
180
181	/* Enter paged protected Mode, activating Long Mode */
182	movl	$(X86_CR0_PG | X86_CR0_PE), %eax /* Enable Paging and Protected mode */
183	movl	%eax, %cr0
184
185	/* Jump from 32bit compatibility mode into 64bit mode. */
186	lret
187ENDPROC(startup_32)
188
189	.code64
190	.org 0x200
191ENTRY(startup_64)
192	/*
193	 * 64bit entry is 0x200 and it is ABI so immutable!
194	 * We come here either from startup_32 or directly from a
195	 * 64bit bootloader.
196	 * If we come here from a bootloader, kernel(text+data+bss+brk),
197	 * ramdisk, zero_page, command line could be above 4G.
198	 * We depend on an identity mapped page table being provided
199	 * that maps our entire kernel(text+data+bss+brk), zero page
200	 * and command line.
201	 */
202#ifdef CONFIG_EFI_STUB
203	/*
204	 * The entry point for the PE/COFF executable is efi_pe_entry, so
205	 * only legacy boot loaders will execute this jmp.
206	 */
207	jmp	preferred_addr
208
209ENTRY(efi_pe_entry)
210	mov	%rcx, %rdi
211	mov	%rdx, %rsi
212	pushq	%rdi
213	pushq	%rsi
214	call	make_boot_params
215	cmpq	$0,%rax
216	je	1f
217	mov	%rax, %rdx
218	popq	%rsi
219	popq	%rdi
220
221ENTRY(efi_stub_entry)
222	call	efi_main
223	movq	%rax,%rsi
224	cmpq	$0,%rax
225	jne	2f
2261:
227	/* EFI init failed, so hang. */
228	hlt
229	jmp	1b
2302:
231	call	3f
2323:
233	popq	%rax
234	subq	$3b, %rax
235	subq	BP_pref_address(%rsi), %rax
236	add	BP_code32_start(%esi), %eax
237	leaq	preferred_addr(%rax), %rax
238	jmp	*%rax
239
240preferred_addr:
241#endif
242
243	/* Setup data segments. */
244	xorl	%eax, %eax
245	movl	%eax, %ds
246	movl	%eax, %es
247	movl	%eax, %ss
248	movl	%eax, %fs
249	movl	%eax, %gs
250
251	/*
252	 * Compute the decompressed kernel start address.  It is where
253	 * we were loaded at aligned to a 2M boundary. %rbp contains the
254	 * decompressed kernel start address.
255	 *
256	 * If it is a relocatable kernel then decompress and run the kernel
257	 * from load address aligned to 2MB addr, otherwise decompress and
258	 * run the kernel from LOAD_PHYSICAL_ADDR
259	 *
260	 * We cannot rely on the calculation done in 32-bit mode, since we
261	 * may have been invoked via the 64-bit entry point.
262	 */
263
264	/* Start with the delta to where the kernel will run at. */
265#ifdef CONFIG_RELOCATABLE
266	leaq	startup_32(%rip) /* - $startup_32 */, %rbp
267	movl	BP_kernel_alignment(%rsi), %eax
268	decl	%eax
269	addq	%rax, %rbp
270	notq	%rax
271	andq	%rax, %rbp
272#else
273	movq	$LOAD_PHYSICAL_ADDR, %rbp
274#endif
275
276	/* Target address to relocate to for decompression */
277	leaq	z_extract_offset(%rbp), %rbx
278
279	/* Set up the stack */
280	leaq	boot_stack_end(%rbx), %rsp
281
282	/* Zero EFLAGS */
283	pushq	$0
284	popfq
285
286/*
287 * Copy the compressed kernel to the end of our buffer
288 * where decompression in place becomes safe.
289 */
290	pushq	%rsi
291	leaq	(_bss-8)(%rip), %rsi
292	leaq	(_bss-8)(%rbx), %rdi
293	movq	$_bss /* - $startup_32 */, %rcx
294	shrq	$3, %rcx
295	std
296	rep	movsq
297	cld
298	popq	%rsi
299
300/*
301 * Jump to the relocated address.
302 */
303	leaq	relocated(%rbx), %rax
304	jmp	*%rax
305
306	.text
307relocated:
308
309/*
310 * Clear BSS (stack is currently empty)
311 */
312	xorl	%eax, %eax
313	leaq    _bss(%rip), %rdi
314	leaq    _ebss(%rip), %rcx
315	subq	%rdi, %rcx
316	shrq	$3, %rcx
317	rep	stosq
318
319/*
320 * Adjust our own GOT
321 */
322	leaq	_got(%rip), %rdx
323	leaq	_egot(%rip), %rcx
3241:
325	cmpq	%rcx, %rdx
326	jae	2f
327	addq	%rbx, (%rdx)
328	addq	$8, %rdx
329	jmp	1b
3302:
331
332/*
333 * Do the decompression, and jump to the new kernel..
334 */
335	pushq	%rsi			/* Save the real mode argument */
336	movq	%rsi, %rdi		/* real mode address */
337	leaq	boot_heap(%rip), %rsi	/* malloc area for uncompression */
338	leaq	input_data(%rip), %rdx  /* input_data */
339	movl	$z_input_len, %ecx	/* input_len */
340	movq	%rbp, %r8		/* output target address */
341	movq	$z_output_len, %r9	/* decompressed length */
342	call	decompress_kernel
343	popq	%rsi
344
345/*
346 * Jump to the decompressed kernel.
347 */
348	jmp	*%rbp
349
350	.code32
351no_longmode:
352	/* This isn't an x86-64 CPU so hang */
3531:
354	hlt
355	jmp     1b
356
357#include "../../kernel/verify_cpu.S"
358
359	.data
360gdt:
361	.word	gdt_end - gdt
362	.long	gdt
363	.word	0
364	.quad	0x0000000000000000	/* NULL descriptor */
365	.quad	0x00af9a000000ffff	/* __KERNEL_CS */
366	.quad	0x00cf92000000ffff	/* __KERNEL_DS */
367	.quad	0x0080890000000000	/* TS descriptor */
368	.quad   0x0000000000000000	/* TS continued */
369gdt_end:
370
371/*
372 * Stack and heap for uncompression
373 */
374	.bss
375	.balign 4
376boot_heap:
377	.fill BOOT_HEAP_SIZE, 1, 0
378boot_stack:
379	.fill BOOT_STACK_SIZE, 1, 0
380boot_stack_end:
381
382/*
383 * Space for page tables (not in .bss so not zeroed)
384 */
385	.section ".pgtable","a",@nobits
386	.balign 4096
387pgtable:
388	.fill 6*4096, 1, 0
389