xref: /openbmc/linux/arch/x86/boot/compressed/head_64.S (revision 8a10bc9d)
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	cmpl	$LOAD_PHYSICAL_ADDR, %ebx
98	jge	1f
99#endif
100	movl	$LOAD_PHYSICAL_ADDR, %ebx
1011:
102
103	/* Target address to relocate to for decompression */
104	addl	$z_extract_offset, %ebx
105
106/*
107 * Prepare for entering 64 bit mode
108 */
109
110	/* Load new GDT with the 64bit segments using 32bit descriptor */
111	leal	gdt(%ebp), %eax
112	movl	%eax, gdt+2(%ebp)
113	lgdt	gdt(%ebp)
114
115	/* Enable PAE mode */
116	movl	$(X86_CR4_PAE), %eax
117	movl	%eax, %cr4
118
119 /*
120  * Build early 4G boot pagetable
121  */
122	/* Initialize Page tables to 0 */
123	leal	pgtable(%ebx), %edi
124	xorl	%eax, %eax
125	movl	$((4096*6)/4), %ecx
126	rep	stosl
127
128	/* Build Level 4 */
129	leal	pgtable + 0(%ebx), %edi
130	leal	0x1007 (%edi), %eax
131	movl	%eax, 0(%edi)
132
133	/* Build Level 3 */
134	leal	pgtable + 0x1000(%ebx), %edi
135	leal	0x1007(%edi), %eax
136	movl	$4, %ecx
1371:	movl	%eax, 0x00(%edi)
138	addl	$0x00001000, %eax
139	addl	$8, %edi
140	decl	%ecx
141	jnz	1b
142
143	/* Build Level 2 */
144	leal	pgtable + 0x2000(%ebx), %edi
145	movl	$0x00000183, %eax
146	movl	$2048, %ecx
1471:	movl	%eax, 0(%edi)
148	addl	$0x00200000, %eax
149	addl	$8, %edi
150	decl	%ecx
151	jnz	1b
152
153	/* Enable the boot page tables */
154	leal	pgtable(%ebx), %eax
155	movl	%eax, %cr3
156
157	/* Enable Long mode in EFER (Extended Feature Enable Register) */
158	movl	$MSR_EFER, %ecx
159	rdmsr
160	btsl	$_EFER_LME, %eax
161	wrmsr
162
163	/* After gdt is loaded */
164	xorl	%eax, %eax
165	lldt	%ax
166	movl    $0x20, %eax
167	ltr	%ax
168
169	/*
170	 * Setup for the jump to 64bit mode
171	 *
172	 * When the jump is performend we will be in long mode but
173	 * in 32bit compatibility mode with EFER.LME = 1, CS.L = 0, CS.D = 1
174	 * (and in turn EFER.LMA = 1).	To jump into 64bit mode we use
175	 * the new gdt/idt that has __KERNEL_CS with CS.L = 1.
176	 * We place all of the values on our mini stack so lret can
177	 * used to perform that far jump.
178	 */
179	pushl	$__KERNEL_CS
180	leal	startup_64(%ebp), %eax
181	pushl	%eax
182
183	/* Enter paged protected Mode, activating Long Mode */
184	movl	$(X86_CR0_PG | X86_CR0_PE), %eax /* Enable Paging and Protected mode */
185	movl	%eax, %cr0
186
187	/* Jump from 32bit compatibility mode into 64bit mode. */
188	lret
189ENDPROC(startup_32)
190
191	.code64
192	.org 0x200
193ENTRY(startup_64)
194	/*
195	 * 64bit entry is 0x200 and it is ABI so immutable!
196	 * We come here either from startup_32 or directly from a
197	 * 64bit bootloader.
198	 * If we come here from a bootloader, kernel(text+data+bss+brk),
199	 * ramdisk, zero_page, command line could be above 4G.
200	 * We depend on an identity mapped page table being provided
201	 * that maps our entire kernel(text+data+bss+brk), zero page
202	 * and command line.
203	 */
204#ifdef CONFIG_EFI_STUB
205	/*
206	 * The entry point for the PE/COFF executable is efi_pe_entry, so
207	 * only legacy boot loaders will execute this jmp.
208	 */
209	jmp	preferred_addr
210
211ENTRY(efi_pe_entry)
212	mov	%rcx, %rdi
213	mov	%rdx, %rsi
214	pushq	%rdi
215	pushq	%rsi
216	call	make_boot_params
217	cmpq	$0,%rax
218	je	1f
219	mov	%rax, %rdx
220	popq	%rsi
221	popq	%rdi
222
223ENTRY(efi_stub_entry)
224	call	efi_main
225	movq	%rax,%rsi
226	cmpq	$0,%rax
227	jne	2f
2281:
229	/* EFI init failed, so hang. */
230	hlt
231	jmp	1b
2322:
233	call	3f
2343:
235	popq	%rax
236	subq	$3b, %rax
237	subq	BP_pref_address(%rsi), %rax
238	add	BP_code32_start(%esi), %eax
239	leaq	preferred_addr(%rax), %rax
240	jmp	*%rax
241
242preferred_addr:
243#endif
244
245	/* Setup data segments. */
246	xorl	%eax, %eax
247	movl	%eax, %ds
248	movl	%eax, %es
249	movl	%eax, %ss
250	movl	%eax, %fs
251	movl	%eax, %gs
252
253	/*
254	 * Compute the decompressed kernel start address.  It is where
255	 * we were loaded at aligned to a 2M boundary. %rbp contains the
256	 * decompressed kernel start address.
257	 *
258	 * If it is a relocatable kernel then decompress and run the kernel
259	 * from load address aligned to 2MB addr, otherwise decompress and
260	 * run the kernel from LOAD_PHYSICAL_ADDR
261	 *
262	 * We cannot rely on the calculation done in 32-bit mode, since we
263	 * may have been invoked via the 64-bit entry point.
264	 */
265
266	/* Start with the delta to where the kernel will run at. */
267#ifdef CONFIG_RELOCATABLE
268	leaq	startup_32(%rip) /* - $startup_32 */, %rbp
269	movl	BP_kernel_alignment(%rsi), %eax
270	decl	%eax
271	addq	%rax, %rbp
272	notq	%rax
273	andq	%rax, %rbp
274	cmpq	$LOAD_PHYSICAL_ADDR, %rbp
275	jge	1f
276#endif
277	movq	$LOAD_PHYSICAL_ADDR, %rbp
2781:
279
280	/* Target address to relocate to for decompression */
281	leaq	z_extract_offset(%rbp), %rbx
282
283	/* Set up the stack */
284	leaq	boot_stack_end(%rbx), %rsp
285
286	/* Zero EFLAGS */
287	pushq	$0
288	popfq
289
290/*
291 * Copy the compressed kernel to the end of our buffer
292 * where decompression in place becomes safe.
293 */
294	pushq	%rsi
295	leaq	(_bss-8)(%rip), %rsi
296	leaq	(_bss-8)(%rbx), %rdi
297	movq	$_bss /* - $startup_32 */, %rcx
298	shrq	$3, %rcx
299	std
300	rep	movsq
301	cld
302	popq	%rsi
303
304/*
305 * Jump to the relocated address.
306 */
307	leaq	relocated(%rbx), %rax
308	jmp	*%rax
309
310	.text
311relocated:
312
313/*
314 * Clear BSS (stack is currently empty)
315 */
316	xorl	%eax, %eax
317	leaq    _bss(%rip), %rdi
318	leaq    _ebss(%rip), %rcx
319	subq	%rdi, %rcx
320	shrq	$3, %rcx
321	rep	stosq
322
323/*
324 * Adjust our own GOT
325 */
326	leaq	_got(%rip), %rdx
327	leaq	_egot(%rip), %rcx
3281:
329	cmpq	%rcx, %rdx
330	jae	2f
331	addq	%rbx, (%rdx)
332	addq	$8, %rdx
333	jmp	1b
3342:
335
336/*
337 * Do the decompression, and jump to the new kernel..
338 */
339	pushq	%rsi			/* Save the real mode argument */
340	movq	%rsi, %rdi		/* real mode address */
341	leaq	boot_heap(%rip), %rsi	/* malloc area for uncompression */
342	leaq	input_data(%rip), %rdx  /* input_data */
343	movl	$z_input_len, %ecx	/* input_len */
344	movq	%rbp, %r8		/* output target address */
345	movq	$z_output_len, %r9	/* decompressed length */
346	call	decompress_kernel	/* returns kernel location in %rax */
347	popq	%rsi
348
349/*
350 * Jump to the decompressed kernel.
351 */
352	jmp	*%rax
353
354	.code32
355no_longmode:
356	/* This isn't an x86-64 CPU so hang */
3571:
358	hlt
359	jmp     1b
360
361#include "../../kernel/verify_cpu.S"
362
363	.data
364gdt:
365	.word	gdt_end - gdt
366	.long	gdt
367	.word	0
368	.quad	0x0000000000000000	/* NULL descriptor */
369	.quad	0x00af9a000000ffff	/* __KERNEL_CS */
370	.quad	0x00cf92000000ffff	/* __KERNEL_DS */
371	.quad	0x0080890000000000	/* TS descriptor */
372	.quad   0x0000000000000000	/* TS continued */
373gdt_end:
374
375/*
376 * Stack and heap for uncompression
377 */
378	.bss
379	.balign 4
380boot_heap:
381	.fill BOOT_HEAP_SIZE, 1, 0
382boot_stack:
383	.fill BOOT_STACK_SIZE, 1, 0
384boot_stack_end:
385
386/*
387 * Space for page tables (not in .bss so not zeroed)
388 */
389	.section ".pgtable","a",@nobits
390	.balign 4096
391pgtable:
392	.fill 6*4096, 1, 0
393