xref: /openbmc/linux/arch/x86/boot/compressed/head_64.S (revision 15e3ae36)
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 *  linux/boot/head.S
4 *
5 *  Copyright (C) 1991, 1992, 1993  Linus Torvalds
6 */
7
8/*
9 *  head.S contains the 32-bit startup code.
10 *
11 * NOTE!!! Startup happens at absolute address 0x00001000, which is also where
12 * the page directory will exist. The startup code will be overwritten by
13 * the page directory. [According to comments etc elsewhere on a compressed
14 * kernel it will end up at 0x1000 + 1Mb I hope so as I assume this. - AC]
15 *
16 * Page 0 is deliberately kept safe, since System Management Mode code in
17 * laptops may need to access the BIOS data stored there.  This is also
18 * useful for future device drivers that either access the BIOS via VM86
19 * mode.
20 */
21
22/*
23 * High loaded stuff by Hans Lermen & Werner Almesberger, Feb. 1996
24 */
25	.code32
26	.text
27
28#include <linux/init.h>
29#include <linux/linkage.h>
30#include <asm/segment.h>
31#include <asm/boot.h>
32#include <asm/msr.h>
33#include <asm/processor-flags.h>
34#include <asm/asm-offsets.h>
35#include <asm/bootparam.h>
36#include "pgtable.h"
37
38/*
39 * Locally defined symbols should be marked hidden:
40 */
41	.hidden _bss
42	.hidden _ebss
43	.hidden _got
44	.hidden _egot
45
46	__HEAD
47	.code32
48SYM_FUNC_START(startup_32)
49	/*
50	 * 32bit entry is 0 and it is ABI so immutable!
51	 * If we come here directly from a bootloader,
52	 * kernel(text+data+bss+brk) ramdisk, zero_page, command line
53	 * all need to be under the 4G limit.
54	 */
55	cld
56	cli
57
58/*
59 * Calculate the delta between where we were compiled to run
60 * at and where we were actually loaded at.  This can only be done
61 * with a short local call on x86.  Nothing  else will tell us what
62 * address we are running at.  The reserved chunk of the real-mode
63 * data at 0x1e4 (defined as a scratch field) are used as the stack
64 * for this calculation. Only 4 bytes are needed.
65 */
66	leal	(BP_scratch+4)(%esi), %esp
67	call	1f
681:	popl	%ebp
69	subl	$1b, %ebp
70
71	/* Load new GDT with the 64bit segments using 32bit descriptor */
72	leal	gdt(%ebp), %eax
73	movl	%eax, 2(%eax)
74	lgdt	(%eax)
75
76	/* Load segment registers with our descriptors */
77	movl	$__BOOT_DS, %eax
78	movl	%eax, %ds
79	movl	%eax, %es
80	movl	%eax, %fs
81	movl	%eax, %gs
82	movl	%eax, %ss
83
84/* setup a stack and make sure cpu supports long mode. */
85	leal	boot_stack_end(%ebp), %esp
86
87	call	verify_cpu
88	testl	%eax, %eax
89	jnz	.Lno_longmode
90
91/*
92 * Compute the delta between where we were compiled to run at
93 * and where the code will actually run at.
94 *
95 * %ebp contains the address we are loaded at by the boot loader and %ebx
96 * contains the address where we should move the kernel image temporarily
97 * for safe in-place decompression.
98 */
99
100#ifdef CONFIG_RELOCATABLE
101	movl	%ebp, %ebx
102
103#ifdef CONFIG_EFI_STUB
104/*
105 * If we were loaded via the EFI LoadImage service, startup_32 will be at an
106 * offset to the start of the space allocated for the image. efi_pe_entry will
107 * set up image_offset to tell us where the image actually starts, so that we
108 * can use the full available buffer.
109 *	image_offset = startup_32 - image_base
110 * Otherwise image_offset will be zero and has no effect on the calculations.
111 */
112	subl    image_offset(%ebp), %ebx
113#endif
114
115	movl	BP_kernel_alignment(%esi), %eax
116	decl	%eax
117	addl	%eax, %ebx
118	notl	%eax
119	andl	%eax, %ebx
120	cmpl	$LOAD_PHYSICAL_ADDR, %ebx
121	jae	1f
122#endif
123	movl	$LOAD_PHYSICAL_ADDR, %ebx
1241:
125
126	/* Target address to relocate to for decompression */
127	addl	BP_init_size(%esi), %ebx
128	subl	$_end, %ebx
129
130/*
131 * Prepare for entering 64 bit mode
132 */
133
134	/* Enable PAE mode */
135	movl	%cr4, %eax
136	orl	$X86_CR4_PAE, %eax
137	movl	%eax, %cr4
138
139 /*
140  * Build early 4G boot pagetable
141  */
142	/*
143	 * If SEV is active then set the encryption mask in the page tables.
144	 * This will insure that when the kernel is copied and decompressed
145	 * it will be done so encrypted.
146	 */
147	call	get_sev_encryption_bit
148	xorl	%edx, %edx
149	testl	%eax, %eax
150	jz	1f
151	subl	$32, %eax	/* Encryption bit is always above bit 31 */
152	bts	%eax, %edx	/* Set encryption mask for page tables */
1531:
154
155	/* Initialize Page tables to 0 */
156	leal	pgtable(%ebx), %edi
157	xorl	%eax, %eax
158	movl	$(BOOT_INIT_PGT_SIZE/4), %ecx
159	rep	stosl
160
161	/* Build Level 4 */
162	leal	pgtable + 0(%ebx), %edi
163	leal	0x1007 (%edi), %eax
164	movl	%eax, 0(%edi)
165	addl	%edx, 4(%edi)
166
167	/* Build Level 3 */
168	leal	pgtable + 0x1000(%ebx), %edi
169	leal	0x1007(%edi), %eax
170	movl	$4, %ecx
1711:	movl	%eax, 0x00(%edi)
172	addl	%edx, 0x04(%edi)
173	addl	$0x00001000, %eax
174	addl	$8, %edi
175	decl	%ecx
176	jnz	1b
177
178	/* Build Level 2 */
179	leal	pgtable + 0x2000(%ebx), %edi
180	movl	$0x00000183, %eax
181	movl	$2048, %ecx
1821:	movl	%eax, 0(%edi)
183	addl	%edx, 4(%edi)
184	addl	$0x00200000, %eax
185	addl	$8, %edi
186	decl	%ecx
187	jnz	1b
188
189	/* Enable the boot page tables */
190	leal	pgtable(%ebx), %eax
191	movl	%eax, %cr3
192
193	/* Enable Long mode in EFER (Extended Feature Enable Register) */
194	movl	$MSR_EFER, %ecx
195	rdmsr
196	btsl	$_EFER_LME, %eax
197	wrmsr
198
199	/* After gdt is loaded */
200	xorl	%eax, %eax
201	lldt	%ax
202	movl    $__BOOT_TSS, %eax
203	ltr	%ax
204
205	/*
206	 * Setup for the jump to 64bit mode
207	 *
208	 * When the jump is performend we will be in long mode but
209	 * in 32bit compatibility mode with EFER.LME = 1, CS.L = 0, CS.D = 1
210	 * (and in turn EFER.LMA = 1).	To jump into 64bit mode we use
211	 * the new gdt/idt that has __KERNEL_CS with CS.L = 1.
212	 * We place all of the values on our mini stack so lret can
213	 * used to perform that far jump.
214	 */
215	pushl	$__KERNEL_CS
216	leal	startup_64(%ebp), %eax
217#ifdef CONFIG_EFI_MIXED
218	movl	efi32_boot_args(%ebp), %edi
219	cmp	$0, %edi
220	jz	1f
221	leal	efi64_stub_entry(%ebp), %eax
222	movl	efi32_boot_args+4(%ebp), %esi
223	movl	efi32_boot_args+8(%ebp), %edx	// saved bootparams pointer
224	cmpl	$0, %edx
225	jnz	1f
226	leal	efi_pe_entry(%ebp), %eax
227	movl	%edi, %ecx			// MS calling convention
228	movl	%esi, %edx
2291:
230#endif
231	pushl	%eax
232
233	/* Enter paged protected Mode, activating Long Mode */
234	movl	$(X86_CR0_PG | X86_CR0_PE), %eax /* Enable Paging and Protected mode */
235	movl	%eax, %cr0
236
237	/* Jump from 32bit compatibility mode into 64bit mode. */
238	lret
239SYM_FUNC_END(startup_32)
240
241#ifdef CONFIG_EFI_MIXED
242	.org 0x190
243SYM_FUNC_START(efi32_stub_entry)
244	add	$0x4, %esp		/* Discard return address */
245	popl	%ecx
246	popl	%edx
247	popl	%esi
248
249	call	1f
2501:	pop	%ebp
251	subl	$1b, %ebp
252
253	movl	%esi, efi32_boot_args+8(%ebp)
254SYM_INNER_LABEL(efi32_pe_stub_entry, SYM_L_LOCAL)
255	movl	%ecx, efi32_boot_args(%ebp)
256	movl	%edx, efi32_boot_args+4(%ebp)
257	movb	$0, efi_is64(%ebp)
258
259	/* Save firmware GDTR and code/data selectors */
260	sgdtl	efi32_boot_gdt(%ebp)
261	movw	%cs, efi32_boot_cs(%ebp)
262	movw	%ds, efi32_boot_ds(%ebp)
263
264	/* Disable paging */
265	movl	%cr0, %eax
266	btrl	$X86_CR0_PG_BIT, %eax
267	movl	%eax, %cr0
268
269	jmp	startup_32
270SYM_FUNC_END(efi32_stub_entry)
271#endif
272
273	.code64
274	.org 0x200
275SYM_CODE_START(startup_64)
276	/*
277	 * 64bit entry is 0x200 and it is ABI so immutable!
278	 * We come here either from startup_32 or directly from a
279	 * 64bit bootloader.
280	 * If we come here from a bootloader, kernel(text+data+bss+brk),
281	 * ramdisk, zero_page, command line could be above 4G.
282	 * We depend on an identity mapped page table being provided
283	 * that maps our entire kernel(text+data+bss+brk), zero page
284	 * and command line.
285	 */
286
287	cld
288	cli
289
290	/* Setup data segments. */
291	xorl	%eax, %eax
292	movl	%eax, %ds
293	movl	%eax, %es
294	movl	%eax, %ss
295	movl	%eax, %fs
296	movl	%eax, %gs
297
298	/*
299	 * Compute the decompressed kernel start address.  It is where
300	 * we were loaded at aligned to a 2M boundary. %rbp contains the
301	 * decompressed kernel start address.
302	 *
303	 * If it is a relocatable kernel then decompress and run the kernel
304	 * from load address aligned to 2MB addr, otherwise decompress and
305	 * run the kernel from LOAD_PHYSICAL_ADDR
306	 *
307	 * We cannot rely on the calculation done in 32-bit mode, since we
308	 * may have been invoked via the 64-bit entry point.
309	 */
310
311	/* Start with the delta to where the kernel will run at. */
312#ifdef CONFIG_RELOCATABLE
313	leaq	startup_32(%rip) /* - $startup_32 */, %rbp
314
315#ifdef CONFIG_EFI_STUB
316/*
317 * If we were loaded via the EFI LoadImage service, startup_32 will be at an
318 * offset to the start of the space allocated for the image. efi_pe_entry will
319 * set up image_offset to tell us where the image actually starts, so that we
320 * can use the full available buffer.
321 *	image_offset = startup_32 - image_base
322 * Otherwise image_offset will be zero and has no effect on the calculations.
323 */
324	movl    image_offset(%rip), %eax
325	subq	%rax, %rbp
326#endif
327
328	movl	BP_kernel_alignment(%rsi), %eax
329	decl	%eax
330	addq	%rax, %rbp
331	notq	%rax
332	andq	%rax, %rbp
333	cmpq	$LOAD_PHYSICAL_ADDR, %rbp
334	jae	1f
335#endif
336	movq	$LOAD_PHYSICAL_ADDR, %rbp
3371:
338
339	/* Target address to relocate to for decompression */
340	movl	BP_init_size(%rsi), %ebx
341	subl	$_end, %ebx
342	addq	%rbp, %rbx
343
344	/* Set up the stack */
345	leaq	boot_stack_end(%rbx), %rsp
346
347	/*
348	 * paging_prepare() and cleanup_trampoline() below can have GOT
349	 * references. Adjust the table with address we are running at.
350	 *
351	 * Zero RAX for adjust_got: the GOT was not adjusted before;
352	 * there's no adjustment to undo.
353	 */
354	xorq	%rax, %rax
355
356	/*
357	 * Calculate the address the binary is loaded at and use it as
358	 * a GOT adjustment.
359	 */
360	call	1f
3611:	popq	%rdi
362	subq	$1b, %rdi
363
364	call	.Ladjust_got
365
366	/*
367	 * At this point we are in long mode with 4-level paging enabled,
368	 * but we might want to enable 5-level paging or vice versa.
369	 *
370	 * The problem is that we cannot do it directly. Setting or clearing
371	 * CR4.LA57 in long mode would trigger #GP. So we need to switch off
372	 * long mode and paging first.
373	 *
374	 * We also need a trampoline in lower memory to switch over from
375	 * 4- to 5-level paging for cases when the bootloader puts the kernel
376	 * above 4G, but didn't enable 5-level paging for us.
377	 *
378	 * The same trampoline can be used to switch from 5- to 4-level paging
379	 * mode, like when starting 4-level paging kernel via kexec() when
380	 * original kernel worked in 5-level paging mode.
381	 *
382	 * For the trampoline, we need the top page table to reside in lower
383	 * memory as we don't have a way to load 64-bit values into CR3 in
384	 * 32-bit mode.
385	 *
386	 * We go though the trampoline even if we don't have to: if we're
387	 * already in a desired paging mode. This way the trampoline code gets
388	 * tested on every boot.
389	 */
390
391	/* Make sure we have GDT with 32-bit code segment */
392	leaq	gdt64(%rip), %rax
393	addq	%rax, 2(%rax)
394	lgdt	(%rax)
395
396	/*
397	 * paging_prepare() sets up the trampoline and checks if we need to
398	 * enable 5-level paging.
399	 *
400	 * paging_prepare() returns a two-quadword structure which lands
401	 * into RDX:RAX:
402	 *   - Address of the trampoline is returned in RAX.
403	 *   - Non zero RDX means trampoline needs to enable 5-level
404	 *     paging.
405	 *
406	 * RSI holds real mode data and needs to be preserved across
407	 * this function call.
408	 */
409	pushq	%rsi
410	movq	%rsi, %rdi		/* real mode address */
411	call	paging_prepare
412	popq	%rsi
413
414	/* Save the trampoline address in RCX */
415	movq	%rax, %rcx
416
417	/*
418	 * Load the address of trampoline_return() into RDI.
419	 * It will be used by the trampoline to return to the main code.
420	 */
421	leaq	trampoline_return(%rip), %rdi
422
423	/* Switch to compatibility mode (CS.L = 0 CS.D = 1) via far return */
424	pushq	$__KERNEL32_CS
425	leaq	TRAMPOLINE_32BIT_CODE_OFFSET(%rax), %rax
426	pushq	%rax
427	lretq
428trampoline_return:
429	/* Restore the stack, the 32-bit trampoline uses its own stack */
430	leaq	boot_stack_end(%rbx), %rsp
431
432	/*
433	 * cleanup_trampoline() would restore trampoline memory.
434	 *
435	 * RDI is address of the page table to use instead of page table
436	 * in trampoline memory (if required).
437	 *
438	 * RSI holds real mode data and needs to be preserved across
439	 * this function call.
440	 */
441	pushq	%rsi
442	leaq	top_pgtable(%rbx), %rdi
443	call	cleanup_trampoline
444	popq	%rsi
445
446	/* Zero EFLAGS */
447	pushq	$0
448	popfq
449
450	/*
451	 * Previously we've adjusted the GOT with address the binary was
452	 * loaded at. Now we need to re-adjust for relocation address.
453	 *
454	 * Calculate the address the binary is loaded at, so that we can
455	 * undo the previous GOT adjustment.
456	 */
457	call	1f
4581:	popq	%rax
459	subq	$1b, %rax
460
461	/* The new adjustment is the relocation address */
462	movq	%rbx, %rdi
463	call	.Ladjust_got
464
465/*
466 * Copy the compressed kernel to the end of our buffer
467 * where decompression in place becomes safe.
468 */
469	pushq	%rsi
470	leaq	(_bss-8)(%rip), %rsi
471	leaq	(_bss-8)(%rbx), %rdi
472	movq	$_bss /* - $startup_32 */, %rcx
473	shrq	$3, %rcx
474	std
475	rep	movsq
476	cld
477	popq	%rsi
478
479	/*
480	 * The GDT may get overwritten either during the copy we just did or
481	 * during extract_kernel below. To avoid any issues, repoint the GDTR
482	 * to the new copy of the GDT.
483	 */
484	leaq	gdt64(%rbx), %rax
485	leaq	gdt(%rbx), %rdx
486	movq	%rdx, 2(%rax)
487	lgdt	(%rax)
488
489/*
490 * Jump to the relocated address.
491 */
492	leaq	.Lrelocated(%rbx), %rax
493	jmp	*%rax
494SYM_CODE_END(startup_64)
495
496#ifdef CONFIG_EFI_STUB
497	.org 0x390
498SYM_FUNC_START(efi64_stub_entry)
499SYM_FUNC_START_ALIAS(efi_stub_entry)
500	and	$~0xf, %rsp			/* realign the stack */
501	movq	%rdx, %rbx			/* save boot_params pointer */
502	call	efi_main
503	movq	%rbx,%rsi
504	leaq	startup_64(%rax), %rax
505	jmp	*%rax
506SYM_FUNC_END(efi64_stub_entry)
507SYM_FUNC_END_ALIAS(efi_stub_entry)
508#endif
509
510	.text
511SYM_FUNC_START_LOCAL_NOALIGN(.Lrelocated)
512
513/*
514 * Clear BSS (stack is currently empty)
515 */
516	xorl	%eax, %eax
517	leaq    _bss(%rip), %rdi
518	leaq    _ebss(%rip), %rcx
519	subq	%rdi, %rcx
520	shrq	$3, %rcx
521	rep	stosq
522
523/*
524 * Do the extraction, and jump to the new kernel..
525 */
526	pushq	%rsi			/* Save the real mode argument */
527	movq	%rsi, %rdi		/* real mode address */
528	leaq	boot_heap(%rip), %rsi	/* malloc area for uncompression */
529	leaq	input_data(%rip), %rdx  /* input_data */
530	movl	$z_input_len, %ecx	/* input_len */
531	movq	%rbp, %r8		/* output target address */
532	movl	$z_output_len, %r9d	/* decompressed length, end of relocs */
533	call	extract_kernel		/* returns kernel location in %rax */
534	popq	%rsi
535
536/*
537 * Jump to the decompressed kernel.
538 */
539	jmp	*%rax
540SYM_FUNC_END(.Lrelocated)
541
542/*
543 * Adjust the global offset table
544 *
545 * RAX is the previous adjustment of the table to undo (use 0 if it's the
546 * first time we touch GOT).
547 * RDI is the new adjustment to apply.
548 */
549.Ladjust_got:
550	/* Walk through the GOT adding the address to the entries */
551	leaq	_got(%rip), %rdx
552	leaq	_egot(%rip), %rcx
5531:
554	cmpq	%rcx, %rdx
555	jae	2f
556	subq	%rax, (%rdx)	/* Undo previous adjustment */
557	addq	%rdi, (%rdx)	/* Apply the new adjustment */
558	addq	$8, %rdx
559	jmp	1b
5602:
561	ret
562
563	.code32
564/*
565 * This is the 32-bit trampoline that will be copied over to low memory.
566 *
567 * RDI contains the return address (might be above 4G).
568 * ECX contains the base address of the trampoline memory.
569 * Non zero RDX means trampoline needs to enable 5-level paging.
570 */
571SYM_CODE_START(trampoline_32bit_src)
572	/* Set up data and stack segments */
573	movl	$__KERNEL_DS, %eax
574	movl	%eax, %ds
575	movl	%eax, %ss
576
577	/* Set up new stack */
578	leal	TRAMPOLINE_32BIT_STACK_END(%ecx), %esp
579
580	/* Disable paging */
581	movl	%cr0, %eax
582	btrl	$X86_CR0_PG_BIT, %eax
583	movl	%eax, %cr0
584
585	/* Check what paging mode we want to be in after the trampoline */
586	cmpl	$0, %edx
587	jz	1f
588
589	/* We want 5-level paging: don't touch CR3 if it already points to 5-level page tables */
590	movl	%cr4, %eax
591	testl	$X86_CR4_LA57, %eax
592	jnz	3f
593	jmp	2f
5941:
595	/* We want 4-level paging: don't touch CR3 if it already points to 4-level page tables */
596	movl	%cr4, %eax
597	testl	$X86_CR4_LA57, %eax
598	jz	3f
5992:
600	/* Point CR3 to the trampoline's new top level page table */
601	leal	TRAMPOLINE_32BIT_PGTABLE_OFFSET(%ecx), %eax
602	movl	%eax, %cr3
6033:
604	/* Set EFER.LME=1 as a precaution in case hypervsior pulls the rug */
605	pushl	%ecx
606	pushl	%edx
607	movl	$MSR_EFER, %ecx
608	rdmsr
609	btsl	$_EFER_LME, %eax
610	wrmsr
611	popl	%edx
612	popl	%ecx
613
614	/* Enable PAE and LA57 (if required) paging modes */
615	movl	$X86_CR4_PAE, %eax
616	cmpl	$0, %edx
617	jz	1f
618	orl	$X86_CR4_LA57, %eax
6191:
620	movl	%eax, %cr4
621
622	/* Calculate address of paging_enabled() once we are executing in the trampoline */
623	leal	.Lpaging_enabled - trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_OFFSET(%ecx), %eax
624
625	/* Prepare the stack for far return to Long Mode */
626	pushl	$__KERNEL_CS
627	pushl	%eax
628
629	/* Enable paging again */
630	movl	$(X86_CR0_PG | X86_CR0_PE), %eax
631	movl	%eax, %cr0
632
633	lret
634SYM_CODE_END(trampoline_32bit_src)
635
636	.code64
637SYM_FUNC_START_LOCAL_NOALIGN(.Lpaging_enabled)
638	/* Return from the trampoline */
639	jmp	*%rdi
640SYM_FUNC_END(.Lpaging_enabled)
641
642	/*
643         * The trampoline code has a size limit.
644         * Make sure we fail to compile if the trampoline code grows
645         * beyond TRAMPOLINE_32BIT_CODE_SIZE bytes.
646	 */
647	.org	trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_SIZE
648
649	.code32
650SYM_FUNC_START_LOCAL_NOALIGN(.Lno_longmode)
651	/* This isn't an x86-64 CPU, so hang intentionally, we cannot continue */
6521:
653	hlt
654	jmp     1b
655SYM_FUNC_END(.Lno_longmode)
656
657#include "../../kernel/verify_cpu.S"
658
659	.data
660SYM_DATA_START_LOCAL(gdt64)
661	.word	gdt_end - gdt - 1
662	.quad   gdt - gdt64
663SYM_DATA_END(gdt64)
664	.balign	8
665SYM_DATA_START_LOCAL(gdt)
666	.word	gdt_end - gdt - 1
667	.long	0
668	.word	0
669	.quad	0x00cf9a000000ffff	/* __KERNEL32_CS */
670	.quad	0x00af9a000000ffff	/* __KERNEL_CS */
671	.quad	0x00cf92000000ffff	/* __KERNEL_DS */
672	.quad	0x0080890000000000	/* TS descriptor */
673	.quad   0x0000000000000000	/* TS continued */
674SYM_DATA_END_LABEL(gdt, SYM_L_LOCAL, gdt_end)
675
676#ifdef CONFIG_EFI_STUB
677SYM_DATA(image_offset, .long 0)
678#endif
679
680#ifdef CONFIG_EFI_MIXED
681SYM_DATA_LOCAL(efi32_boot_args, .long 0, 0, 0)
682SYM_DATA(efi_is64, .byte 1)
683
684#define ST32_boottime		60 // offsetof(efi_system_table_32_t, boottime)
685#define BS32_handle_protocol	88 // offsetof(efi_boot_services_32_t, handle_protocol)
686#define LI32_image_base		32 // offsetof(efi_loaded_image_32_t, image_base)
687
688	.text
689	.code32
690SYM_FUNC_START(efi32_pe_entry)
691/*
692 * efi_status_t efi32_pe_entry(efi_handle_t image_handle,
693 *			       efi_system_table_32_t *sys_table)
694 */
695
696	pushl	%ebp
697	movl	%esp, %ebp
698	pushl	%eax				// dummy push to allocate loaded_image
699
700	pushl	%ebx				// save callee-save registers
701	pushl	%edi
702
703	call	verify_cpu			// check for long mode support
704	testl	%eax, %eax
705	movl	$0x80000003, %eax		// EFI_UNSUPPORTED
706	jnz	2f
707
708	call	1f
7091:	pop	%ebx
710	subl	$1b, %ebx
711
712	/* Get the loaded image protocol pointer from the image handle */
713	leal	-4(%ebp), %eax
714	pushl	%eax				// &loaded_image
715	leal	loaded_image_proto(%ebx), %eax
716	pushl	%eax				// pass the GUID address
717	pushl	8(%ebp)				// pass the image handle
718
719	/*
720	 * Note the alignment of the stack frame.
721	 *   sys_table
722	 *   handle             <-- 16-byte aligned on entry by ABI
723	 *   return address
724	 *   frame pointer
725	 *   loaded_image       <-- local variable
726	 *   saved %ebx		<-- 16-byte aligned here
727	 *   saved %edi
728	 *   &loaded_image
729	 *   &loaded_image_proto
730	 *   handle             <-- 16-byte aligned for call to handle_protocol
731	 */
732
733	movl	12(%ebp), %eax			// sys_table
734	movl	ST32_boottime(%eax), %eax	// sys_table->boottime
735	call	*BS32_handle_protocol(%eax)	// sys_table->boottime->handle_protocol
736	addl	$12, %esp			// restore argument space
737	testl	%eax, %eax
738	jnz	2f
739
740	movl	8(%ebp), %ecx			// image_handle
741	movl	12(%ebp), %edx			// sys_table
742	movl	-4(%ebp), %esi			// loaded_image
743	movl	LI32_image_base(%esi), %esi	// loaded_image->image_base
744	movl	%ebx, %ebp			// startup_32 for efi32_pe_stub_entry
745	/*
746	 * We need to set the image_offset variable here since startup_32() will
747	 * use it before we get to the 64-bit efi_pe_entry() in C code.
748	 */
749	subl	%esi, %ebx
750	movl	%ebx, image_offset(%ebp)	// save image_offset
751	jmp	efi32_pe_stub_entry
752
7532:	popl	%edi				// restore callee-save registers
754	popl	%ebx
755	leave
756	ret
757SYM_FUNC_END(efi32_pe_entry)
758
759	.section ".rodata"
760	/* EFI loaded image protocol GUID */
761	.balign 4
762SYM_DATA_START_LOCAL(loaded_image_proto)
763	.long	0x5b1b31a1
764	.word	0x9562, 0x11d2
765	.byte	0x8e, 0x3f, 0x00, 0xa0, 0xc9, 0x69, 0x72, 0x3b
766SYM_DATA_END(loaded_image_proto)
767#endif
768
769/*
770 * Stack and heap for uncompression
771 */
772	.bss
773	.balign 4
774SYM_DATA_LOCAL(boot_heap,	.fill BOOT_HEAP_SIZE, 1, 0)
775
776SYM_DATA_START_LOCAL(boot_stack)
777	.fill BOOT_STACK_SIZE, 1, 0
778SYM_DATA_END_LABEL(boot_stack, SYM_L_LOCAL, boot_stack_end)
779
780/*
781 * Space for page tables (not in .bss so not zeroed)
782 */
783	.section ".pgtable","aw",@nobits
784	.balign 4096
785SYM_DATA_LOCAL(pgtable,		.fill BOOT_PGT_SIZE, 1, 0)
786
787/*
788 * The page table is going to be used instead of page table in the trampoline
789 * memory.
790 */
791SYM_DATA_LOCAL(top_pgtable,	.fill PAGE_SIZE, 1, 0)
792