xref: /openbmc/linux/arch/x86/boot/compressed/head_64.S (revision 39b6f3aa)
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/pgtable_types.h>
31#include <asm/page_types.h>
32#include <asm/boot.h>
33#include <asm/msr.h>
34#include <asm/processor-flags.h>
35#include <asm/asm-offsets.h>
36
37	__HEAD
38	.code32
39ENTRY(startup_32)
40	/*
41	 * 32bit entry is 0 and it is ABI so immutable!
42	 * If we come here directly from a bootloader,
43	 * kernel(text+data+bss+brk) ramdisk, zero_page, command line
44	 * all need to be under the 4G limit.
45	 */
46	cld
47	/*
48	 * Test KEEP_SEGMENTS flag to see if the bootloader is asking
49	 * us to not reload segments
50	 */
51	testb $(1<<6), BP_loadflags(%esi)
52	jnz 1f
53
54	cli
55	movl	$(__BOOT_DS), %eax
56	movl	%eax, %ds
57	movl	%eax, %es
58	movl	%eax, %ss
591:
60
61/*
62 * Calculate the delta between where we were compiled to run
63 * at and where we were actually loaded at.  This can only be done
64 * with a short local call on x86.  Nothing  else will tell us what
65 * address we are running at.  The reserved chunk of the real-mode
66 * data at 0x1e4 (defined as a scratch field) are used as the stack
67 * for this calculation. Only 4 bytes are needed.
68 */
69	leal	(BP_scratch+4)(%esi), %esp
70	call	1f
711:	popl	%ebp
72	subl	$1b, %ebp
73
74/* setup a stack and make sure cpu supports long mode. */
75	movl	$boot_stack_end, %eax
76	addl	%ebp, %eax
77	movl	%eax, %esp
78
79	call	verify_cpu
80	testl	%eax, %eax
81	jnz	no_longmode
82
83/*
84 * Compute the delta between where we were compiled to run at
85 * and where the code will actually run at.
86 *
87 * %ebp contains the address we are loaded at by the boot loader and %ebx
88 * contains the address where we should move the kernel image temporarily
89 * for safe in-place decompression.
90 */
91
92#ifdef CONFIG_RELOCATABLE
93	movl	%ebp, %ebx
94	movl	BP_kernel_alignment(%esi), %eax
95	decl	%eax
96	addl	%eax, %ebx
97	notl	%eax
98	andl	%eax, %ebx
99#else
100	movl	$LOAD_PHYSICAL_ADDR, %ebx
101#endif
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#else
275	movq	$LOAD_PHYSICAL_ADDR, %rbp
276#endif
277
278	/* Target address to relocate to for decompression */
279	leaq	z_extract_offset(%rbp), %rbx
280
281	/* Set up the stack */
282	leaq	boot_stack_end(%rbx), %rsp
283
284	/* Zero EFLAGS */
285	pushq	$0
286	popfq
287
288/*
289 * Copy the compressed kernel to the end of our buffer
290 * where decompression in place becomes safe.
291 */
292	pushq	%rsi
293	leaq	(_bss-8)(%rip), %rsi
294	leaq	(_bss-8)(%rbx), %rdi
295	movq	$_bss /* - $startup_32 */, %rcx
296	shrq	$3, %rcx
297	std
298	rep	movsq
299	cld
300	popq	%rsi
301
302/*
303 * Jump to the relocated address.
304 */
305	leaq	relocated(%rbx), %rax
306	jmp	*%rax
307
308	.text
309relocated:
310
311/*
312 * Clear BSS (stack is currently empty)
313 */
314	xorl	%eax, %eax
315	leaq    _bss(%rip), %rdi
316	leaq    _ebss(%rip), %rcx
317	subq	%rdi, %rcx
318	shrq	$3, %rcx
319	rep	stosq
320
321/*
322 * Adjust our own GOT
323 */
324	leaq	_got(%rip), %rdx
325	leaq	_egot(%rip), %rcx
3261:
327	cmpq	%rcx, %rdx
328	jae	2f
329	addq	%rbx, (%rdx)
330	addq	$8, %rdx
331	jmp	1b
3322:
333
334/*
335 * Do the decompression, and jump to the new kernel..
336 */
337	pushq	%rsi			/* Save the real mode argument */
338	movq	%rsi, %rdi		/* real mode address */
339	leaq	boot_heap(%rip), %rsi	/* malloc area for uncompression */
340	leaq	input_data(%rip), %rdx  /* input_data */
341	movl	$z_input_len, %ecx	/* input_len */
342	movq	%rbp, %r8		/* output target address */
343	call	decompress_kernel
344	popq	%rsi
345
346/*
347 * Jump to the decompressed kernel.
348 */
349	jmp	*%rbp
350
351	.code32
352no_longmode:
353	/* This isn't an x86-64 CPU so hang */
3541:
355	hlt
356	jmp     1b
357
358#include "../../kernel/verify_cpu.S"
359
360	.data
361gdt:
362	.word	gdt_end - gdt
363	.long	gdt
364	.word	0
365	.quad	0x0000000000000000	/* NULL descriptor */
366	.quad	0x00af9a000000ffff	/* __KERNEL_CS */
367	.quad	0x00cf92000000ffff	/* __KERNEL_DS */
368	.quad	0x0080890000000000	/* TS descriptor */
369	.quad   0x0000000000000000	/* TS continued */
370gdt_end:
371
372/*
373 * Stack and heap for uncompression
374 */
375	.bss
376	.balign 4
377boot_heap:
378	.fill BOOT_HEAP_SIZE, 1, 0
379boot_stack:
380	.fill BOOT_STACK_SIZE, 1, 0
381boot_stack_end:
382
383/*
384 * Space for page tables (not in .bss so not zeroed)
385 */
386	.section ".pgtable","a",@nobits
387	.balign 4096
388pgtable:
389	.fill 6*4096, 1, 0
390