xref: /openbmc/linux/arch/x86/boot/compressed/head_64.S (revision 2791f47d)
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 <asm/desc_defs.h>
37#include "pgtable.h"
38
39/*
40 * Locally defined symbols should be marked hidden:
41 */
42	.hidden _bss
43	.hidden _ebss
44	.hidden _end
45
46	__HEAD
47
48/*
49 * This macro gives the relative virtual address of X, i.e. the offset of X
50 * from startup_32. This is the same as the link-time virtual address of X,
51 * since startup_32 is at 0, but defining it this way tells the
52 * assembler/linker that we do not want the actual run-time address of X. This
53 * prevents the linker from trying to create unwanted run-time relocation
54 * entries for the reference when the compressed kernel is linked as PIE.
55 *
56 * A reference X(%reg) will result in the link-time VA of X being stored with
57 * the instruction, and a run-time R_X86_64_RELATIVE relocation entry that
58 * adds the 64-bit base address where the kernel is loaded.
59 *
60 * Replacing it with (X-startup_32)(%reg) results in the offset being stored,
61 * and no run-time relocation.
62 *
63 * The macro should be used as a displacement with a base register containing
64 * the run-time address of startup_32 [i.e. rva(X)(%reg)], or as an immediate
65 * [$ rva(X)].
66 *
67 * This macro can only be used from within the .head.text section, since the
68 * expression requires startup_32 to be in the same section as the code being
69 * assembled.
70 */
71#define rva(X) ((X) - startup_32)
72
73	.code32
74SYM_FUNC_START(startup_32)
75	/*
76	 * 32bit entry is 0 and it is ABI so immutable!
77	 * If we come here directly from a bootloader,
78	 * kernel(text+data+bss+brk) ramdisk, zero_page, command line
79	 * all need to be under the 4G limit.
80	 */
81	cld
82	cli
83
84/*
85 * Calculate the delta between where we were compiled to run
86 * at and where we were actually loaded at.  This can only be done
87 * with a short local call on x86.  Nothing  else will tell us what
88 * address we are running at.  The reserved chunk of the real-mode
89 * data at 0x1e4 (defined as a scratch field) are used as the stack
90 * for this calculation. Only 4 bytes are needed.
91 */
92	leal	(BP_scratch+4)(%esi), %esp
93	call	1f
941:	popl	%ebp
95	subl	$ rva(1b), %ebp
96
97	/* Load new GDT with the 64bit segments using 32bit descriptor */
98	leal	rva(gdt)(%ebp), %eax
99	movl	%eax, 2(%eax)
100	lgdt	(%eax)
101
102	/* Load segment registers with our descriptors */
103	movl	$__BOOT_DS, %eax
104	movl	%eax, %ds
105	movl	%eax, %es
106	movl	%eax, %fs
107	movl	%eax, %gs
108	movl	%eax, %ss
109
110/* setup a stack and make sure cpu supports long mode. */
111	leal	rva(boot_stack_end)(%ebp), %esp
112
113	call	verify_cpu
114	testl	%eax, %eax
115	jnz	.Lno_longmode
116
117/*
118 * Compute the delta between where we were compiled to run at
119 * and where the code will actually run at.
120 *
121 * %ebp contains the address we are loaded at by the boot loader and %ebx
122 * contains the address where we should move the kernel image temporarily
123 * for safe in-place decompression.
124 */
125
126#ifdef CONFIG_RELOCATABLE
127	movl	%ebp, %ebx
128
129#ifdef CONFIG_EFI_STUB
130/*
131 * If we were loaded via the EFI LoadImage service, startup_32 will be at an
132 * offset to the start of the space allocated for the image. efi_pe_entry will
133 * set up image_offset to tell us where the image actually starts, so that we
134 * can use the full available buffer.
135 *	image_offset = startup_32 - image_base
136 * Otherwise image_offset will be zero and has no effect on the calculations.
137 */
138	subl    rva(image_offset)(%ebp), %ebx
139#endif
140
141	movl	BP_kernel_alignment(%esi), %eax
142	decl	%eax
143	addl	%eax, %ebx
144	notl	%eax
145	andl	%eax, %ebx
146	cmpl	$LOAD_PHYSICAL_ADDR, %ebx
147	jae	1f
148#endif
149	movl	$LOAD_PHYSICAL_ADDR, %ebx
1501:
151
152	/* Target address to relocate to for decompression */
153	addl	BP_init_size(%esi), %ebx
154	subl	$ rva(_end), %ebx
155
156/*
157 * Prepare for entering 64 bit mode
158 */
159
160	/* Enable PAE mode */
161	movl	%cr4, %eax
162	orl	$X86_CR4_PAE, %eax
163	movl	%eax, %cr4
164
165 /*
166  * Build early 4G boot pagetable
167  */
168	/*
169	 * If SEV is active then set the encryption mask in the page tables.
170	 * This will insure that when the kernel is copied and decompressed
171	 * it will be done so encrypted.
172	 */
173	call	get_sev_encryption_bit
174	xorl	%edx, %edx
175	testl	%eax, %eax
176	jz	1f
177	subl	$32, %eax	/* Encryption bit is always above bit 31 */
178	bts	%eax, %edx	/* Set encryption mask for page tables */
1791:
180
181	/* Initialize Page tables to 0 */
182	leal	rva(pgtable)(%ebx), %edi
183	xorl	%eax, %eax
184	movl	$(BOOT_INIT_PGT_SIZE/4), %ecx
185	rep	stosl
186
187	/* Build Level 4 */
188	leal	rva(pgtable + 0)(%ebx), %edi
189	leal	0x1007 (%edi), %eax
190	movl	%eax, 0(%edi)
191	addl	%edx, 4(%edi)
192
193	/* Build Level 3 */
194	leal	rva(pgtable + 0x1000)(%ebx), %edi
195	leal	0x1007(%edi), %eax
196	movl	$4, %ecx
1971:	movl	%eax, 0x00(%edi)
198	addl	%edx, 0x04(%edi)
199	addl	$0x00001000, %eax
200	addl	$8, %edi
201	decl	%ecx
202	jnz	1b
203
204	/* Build Level 2 */
205	leal	rva(pgtable + 0x2000)(%ebx), %edi
206	movl	$0x00000183, %eax
207	movl	$2048, %ecx
2081:	movl	%eax, 0(%edi)
209	addl	%edx, 4(%edi)
210	addl	$0x00200000, %eax
211	addl	$8, %edi
212	decl	%ecx
213	jnz	1b
214
215	/* Enable the boot page tables */
216	leal	rva(pgtable)(%ebx), %eax
217	movl	%eax, %cr3
218
219	/* Enable Long mode in EFER (Extended Feature Enable Register) */
220	movl	$MSR_EFER, %ecx
221	rdmsr
222	btsl	$_EFER_LME, %eax
223	wrmsr
224
225	/* After gdt is loaded */
226	xorl	%eax, %eax
227	lldt	%ax
228	movl    $__BOOT_TSS, %eax
229	ltr	%ax
230
231	/*
232	 * Setup for the jump to 64bit mode
233	 *
234	 * When the jump is performend we will be in long mode but
235	 * in 32bit compatibility mode with EFER.LME = 1, CS.L = 0, CS.D = 1
236	 * (and in turn EFER.LMA = 1).	To jump into 64bit mode we use
237	 * the new gdt/idt that has __KERNEL_CS with CS.L = 1.
238	 * We place all of the values on our mini stack so lret can
239	 * used to perform that far jump.
240	 */
241	leal	rva(startup_64)(%ebp), %eax
242#ifdef CONFIG_EFI_MIXED
243	movl	rva(efi32_boot_args)(%ebp), %edi
244	cmp	$0, %edi
245	jz	1f
246	leal	rva(efi64_stub_entry)(%ebp), %eax
247	movl	rva(efi32_boot_args+4)(%ebp), %esi
248	movl	rva(efi32_boot_args+8)(%ebp), %edx	// saved bootparams pointer
249	cmpl	$0, %edx
250	jnz	1f
251	/*
252	 * efi_pe_entry uses MS calling convention, which requires 32 bytes of
253	 * shadow space on the stack even if all arguments are passed in
254	 * registers. We also need an additional 8 bytes for the space that
255	 * would be occupied by the return address, and this also results in
256	 * the correct stack alignment for entry.
257	 */
258	subl	$40, %esp
259	leal	rva(efi_pe_entry)(%ebp), %eax
260	movl	%edi, %ecx			// MS calling convention
261	movl	%esi, %edx
2621:
263#endif
264	pushl	$__KERNEL_CS
265	pushl	%eax
266
267	/* Enter paged protected Mode, activating Long Mode */
268	movl	$(X86_CR0_PG | X86_CR0_PE), %eax /* Enable Paging and Protected mode */
269	movl	%eax, %cr0
270
271	/* Jump from 32bit compatibility mode into 64bit mode. */
272	lret
273SYM_FUNC_END(startup_32)
274
275#ifdef CONFIG_EFI_MIXED
276	.org 0x190
277SYM_FUNC_START(efi32_stub_entry)
278	add	$0x4, %esp		/* Discard return address */
279	popl	%ecx
280	popl	%edx
281	popl	%esi
282
283	call	1f
2841:	pop	%ebp
285	subl	$ rva(1b), %ebp
286
287	movl	%esi, rva(efi32_boot_args+8)(%ebp)
288SYM_INNER_LABEL(efi32_pe_stub_entry, SYM_L_LOCAL)
289	movl	%ecx, rva(efi32_boot_args)(%ebp)
290	movl	%edx, rva(efi32_boot_args+4)(%ebp)
291	movb	$0, rva(efi_is64)(%ebp)
292
293	/* Save firmware GDTR and code/data selectors */
294	sgdtl	rva(efi32_boot_gdt)(%ebp)
295	movw	%cs, rva(efi32_boot_cs)(%ebp)
296	movw	%ds, rva(efi32_boot_ds)(%ebp)
297
298	/* Disable paging */
299	movl	%cr0, %eax
300	btrl	$X86_CR0_PG_BIT, %eax
301	movl	%eax, %cr0
302
303	jmp	startup_32
304SYM_FUNC_END(efi32_stub_entry)
305#endif
306
307	.code64
308	.org 0x200
309SYM_CODE_START(startup_64)
310	/*
311	 * 64bit entry is 0x200 and it is ABI so immutable!
312	 * We come here either from startup_32 or directly from a
313	 * 64bit bootloader.
314	 * If we come here from a bootloader, kernel(text+data+bss+brk),
315	 * ramdisk, zero_page, command line could be above 4G.
316	 * We depend on an identity mapped page table being provided
317	 * that maps our entire kernel(text+data+bss+brk), zero page
318	 * and command line.
319	 */
320
321	cld
322	cli
323
324	/* Setup data segments. */
325	xorl	%eax, %eax
326	movl	%eax, %ds
327	movl	%eax, %es
328	movl	%eax, %ss
329	movl	%eax, %fs
330	movl	%eax, %gs
331
332	/*
333	 * Compute the decompressed kernel start address.  It is where
334	 * we were loaded at aligned to a 2M boundary. %rbp contains the
335	 * decompressed kernel start address.
336	 *
337	 * If it is a relocatable kernel then decompress and run the kernel
338	 * from load address aligned to 2MB addr, otherwise decompress and
339	 * run the kernel from LOAD_PHYSICAL_ADDR
340	 *
341	 * We cannot rely on the calculation done in 32-bit mode, since we
342	 * may have been invoked via the 64-bit entry point.
343	 */
344
345	/* Start with the delta to where the kernel will run at. */
346#ifdef CONFIG_RELOCATABLE
347	leaq	startup_32(%rip) /* - $startup_32 */, %rbp
348
349#ifdef CONFIG_EFI_STUB
350/*
351 * If we were loaded via the EFI LoadImage service, startup_32 will be at an
352 * offset to the start of the space allocated for the image. efi_pe_entry will
353 * set up image_offset to tell us where the image actually starts, so that we
354 * can use the full available buffer.
355 *	image_offset = startup_32 - image_base
356 * Otherwise image_offset will be zero and has no effect on the calculations.
357 */
358	movl    image_offset(%rip), %eax
359	subq	%rax, %rbp
360#endif
361
362	movl	BP_kernel_alignment(%rsi), %eax
363	decl	%eax
364	addq	%rax, %rbp
365	notq	%rax
366	andq	%rax, %rbp
367	cmpq	$LOAD_PHYSICAL_ADDR, %rbp
368	jae	1f
369#endif
370	movq	$LOAD_PHYSICAL_ADDR, %rbp
3711:
372
373	/* Target address to relocate to for decompression */
374	movl	BP_init_size(%rsi), %ebx
375	subl	$ rva(_end), %ebx
376	addq	%rbp, %rbx
377
378	/* Set up the stack */
379	leaq	rva(boot_stack_end)(%rbx), %rsp
380
381	/*
382	 * At this point we are in long mode with 4-level paging enabled,
383	 * but we might want to enable 5-level paging or vice versa.
384	 *
385	 * The problem is that we cannot do it directly. Setting or clearing
386	 * CR4.LA57 in long mode would trigger #GP. So we need to switch off
387	 * long mode and paging first.
388	 *
389	 * We also need a trampoline in lower memory to switch over from
390	 * 4- to 5-level paging for cases when the bootloader puts the kernel
391	 * above 4G, but didn't enable 5-level paging for us.
392	 *
393	 * The same trampoline can be used to switch from 5- to 4-level paging
394	 * mode, like when starting 4-level paging kernel via kexec() when
395	 * original kernel worked in 5-level paging mode.
396	 *
397	 * For the trampoline, we need the top page table to reside in lower
398	 * memory as we don't have a way to load 64-bit values into CR3 in
399	 * 32-bit mode.
400	 *
401	 * We go though the trampoline even if we don't have to: if we're
402	 * already in a desired paging mode. This way the trampoline code gets
403	 * tested on every boot.
404	 */
405
406	/* Make sure we have GDT with 32-bit code segment */
407	leaq	gdt64(%rip), %rax
408	addq	%rax, 2(%rax)
409	lgdt	(%rax)
410
411	/* Reload CS so IRET returns to a CS actually in the GDT */
412	pushq	$__KERNEL_CS
413	leaq	.Lon_kernel_cs(%rip), %rax
414	pushq	%rax
415	lretq
416
417.Lon_kernel_cs:
418
419	pushq	%rsi
420	call	load_stage1_idt
421	popq	%rsi
422
423	/*
424	 * paging_prepare() sets up the trampoline and checks if we need to
425	 * enable 5-level paging.
426	 *
427	 * paging_prepare() returns a two-quadword structure which lands
428	 * into RDX:RAX:
429	 *   - Address of the trampoline is returned in RAX.
430	 *   - Non zero RDX means trampoline needs to enable 5-level
431	 *     paging.
432	 *
433	 * RSI holds real mode data and needs to be preserved across
434	 * this function call.
435	 */
436	pushq	%rsi
437	movq	%rsi, %rdi		/* real mode address */
438	call	paging_prepare
439	popq	%rsi
440
441	/* Save the trampoline address in RCX */
442	movq	%rax, %rcx
443
444	/*
445	 * Load the address of trampoline_return() into RDI.
446	 * It will be used by the trampoline to return to the main code.
447	 */
448	leaq	trampoline_return(%rip), %rdi
449
450	/* Switch to compatibility mode (CS.L = 0 CS.D = 1) via far return */
451	pushq	$__KERNEL32_CS
452	leaq	TRAMPOLINE_32BIT_CODE_OFFSET(%rax), %rax
453	pushq	%rax
454	lretq
455trampoline_return:
456	/* Restore the stack, the 32-bit trampoline uses its own stack */
457	leaq	rva(boot_stack_end)(%rbx), %rsp
458
459	/*
460	 * cleanup_trampoline() would restore trampoline memory.
461	 *
462	 * RDI is address of the page table to use instead of page table
463	 * in trampoline memory (if required).
464	 *
465	 * RSI holds real mode data and needs to be preserved across
466	 * this function call.
467	 */
468	pushq	%rsi
469	leaq	rva(top_pgtable)(%rbx), %rdi
470	call	cleanup_trampoline
471	popq	%rsi
472
473	/* Zero EFLAGS */
474	pushq	$0
475	popfq
476
477/*
478 * Copy the compressed kernel to the end of our buffer
479 * where decompression in place becomes safe.
480 */
481	pushq	%rsi
482	leaq	(_bss-8)(%rip), %rsi
483	leaq	rva(_bss-8)(%rbx), %rdi
484	movl	$(_bss - startup_32), %ecx
485	shrl	$3, %ecx
486	std
487	rep	movsq
488	cld
489	popq	%rsi
490
491	/*
492	 * The GDT may get overwritten either during the copy we just did or
493	 * during extract_kernel below. To avoid any issues, repoint the GDTR
494	 * to the new copy of the GDT.
495	 */
496	leaq	rva(gdt64)(%rbx), %rax
497	leaq	rva(gdt)(%rbx), %rdx
498	movq	%rdx, 2(%rax)
499	lgdt	(%rax)
500
501/*
502 * Jump to the relocated address.
503 */
504	leaq	rva(.Lrelocated)(%rbx), %rax
505	jmp	*%rax
506SYM_CODE_END(startup_64)
507
508#ifdef CONFIG_EFI_STUB
509	.org 0x390
510SYM_FUNC_START(efi64_stub_entry)
511SYM_FUNC_START_ALIAS(efi_stub_entry)
512	and	$~0xf, %rsp			/* realign the stack */
513	movq	%rdx, %rbx			/* save boot_params pointer */
514	call	efi_main
515	movq	%rbx,%rsi
516	leaq	rva(startup_64)(%rax), %rax
517	jmp	*%rax
518SYM_FUNC_END(efi64_stub_entry)
519SYM_FUNC_END_ALIAS(efi_stub_entry)
520#endif
521
522	.text
523SYM_FUNC_START_LOCAL_NOALIGN(.Lrelocated)
524
525/*
526 * Clear BSS (stack is currently empty)
527 */
528	xorl	%eax, %eax
529	leaq    _bss(%rip), %rdi
530	leaq    _ebss(%rip), %rcx
531	subq	%rdi, %rcx
532	shrq	$3, %rcx
533	rep	stosq
534
535/*
536 * If running as an SEV guest, the encryption mask is required in the
537 * page-table setup code below. When the guest also has SEV-ES enabled
538 * set_sev_encryption_mask() will cause #VC exceptions, but the stage2
539 * handler can't map its GHCB because the page-table is not set up yet.
540 * So set up the encryption mask here while still on the stage1 #VC
541 * handler. Then load stage2 IDT and switch to the kernel's own
542 * page-table.
543 */
544	pushq	%rsi
545	call	set_sev_encryption_mask
546	call	load_stage2_idt
547	call	initialize_identity_maps
548	popq	%rsi
549
550/*
551 * Do the extraction, and jump to the new kernel..
552 */
553	pushq	%rsi			/* Save the real mode argument */
554	movq	%rsi, %rdi		/* real mode address */
555	leaq	boot_heap(%rip), %rsi	/* malloc area for uncompression */
556	leaq	input_data(%rip), %rdx  /* input_data */
557	movl	input_len(%rip), %ecx	/* input_len */
558	movq	%rbp, %r8		/* output target address */
559	movl	output_len(%rip), %r9d	/* decompressed length, end of relocs */
560	call	extract_kernel		/* returns kernel location in %rax */
561	popq	%rsi
562
563/*
564 * Jump to the decompressed kernel.
565 */
566	jmp	*%rax
567SYM_FUNC_END(.Lrelocated)
568
569	.code32
570/*
571 * This is the 32-bit trampoline that will be copied over to low memory.
572 *
573 * RDI contains the return address (might be above 4G).
574 * ECX contains the base address of the trampoline memory.
575 * Non zero RDX means trampoline needs to enable 5-level paging.
576 */
577SYM_CODE_START(trampoline_32bit_src)
578	/* Set up data and stack segments */
579	movl	$__KERNEL_DS, %eax
580	movl	%eax, %ds
581	movl	%eax, %ss
582
583	/* Set up new stack */
584	leal	TRAMPOLINE_32BIT_STACK_END(%ecx), %esp
585
586	/* Disable paging */
587	movl	%cr0, %eax
588	btrl	$X86_CR0_PG_BIT, %eax
589	movl	%eax, %cr0
590
591	/* Check what paging mode we want to be in after the trampoline */
592	cmpl	$0, %edx
593	jz	1f
594
595	/* We want 5-level paging: don't touch CR3 if it already points to 5-level page tables */
596	movl	%cr4, %eax
597	testl	$X86_CR4_LA57, %eax
598	jnz	3f
599	jmp	2f
6001:
601	/* We want 4-level paging: don't touch CR3 if it already points to 4-level page tables */
602	movl	%cr4, %eax
603	testl	$X86_CR4_LA57, %eax
604	jz	3f
6052:
606	/* Point CR3 to the trampoline's new top level page table */
607	leal	TRAMPOLINE_32BIT_PGTABLE_OFFSET(%ecx), %eax
608	movl	%eax, %cr3
6093:
610	/* Set EFER.LME=1 as a precaution in case hypervsior pulls the rug */
611	pushl	%ecx
612	pushl	%edx
613	movl	$MSR_EFER, %ecx
614	rdmsr
615	btsl	$_EFER_LME, %eax
616	wrmsr
617	popl	%edx
618	popl	%ecx
619
620	/* Enable PAE and LA57 (if required) paging modes */
621	movl	$X86_CR4_PAE, %eax
622	cmpl	$0, %edx
623	jz	1f
624	orl	$X86_CR4_LA57, %eax
6251:
626	movl	%eax, %cr4
627
628	/* Calculate address of paging_enabled() once we are executing in the trampoline */
629	leal	.Lpaging_enabled - trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_OFFSET(%ecx), %eax
630
631	/* Prepare the stack for far return to Long Mode */
632	pushl	$__KERNEL_CS
633	pushl	%eax
634
635	/* Enable paging again */
636	movl	$(X86_CR0_PG | X86_CR0_PE), %eax
637	movl	%eax, %cr0
638
639	lret
640SYM_CODE_END(trampoline_32bit_src)
641
642	.code64
643SYM_FUNC_START_LOCAL_NOALIGN(.Lpaging_enabled)
644	/* Return from the trampoline */
645	jmp	*%rdi
646SYM_FUNC_END(.Lpaging_enabled)
647
648	/*
649         * The trampoline code has a size limit.
650         * Make sure we fail to compile if the trampoline code grows
651         * beyond TRAMPOLINE_32BIT_CODE_SIZE bytes.
652	 */
653	.org	trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_SIZE
654
655	.code32
656SYM_FUNC_START_LOCAL_NOALIGN(.Lno_longmode)
657	/* This isn't an x86-64 CPU, so hang intentionally, we cannot continue */
6581:
659	hlt
660	jmp     1b
661SYM_FUNC_END(.Lno_longmode)
662
663#include "../../kernel/verify_cpu.S"
664
665	.data
666SYM_DATA_START_LOCAL(gdt64)
667	.word	gdt_end - gdt - 1
668	.quad   gdt - gdt64
669SYM_DATA_END(gdt64)
670	.balign	8
671SYM_DATA_START_LOCAL(gdt)
672	.word	gdt_end - gdt - 1
673	.long	0
674	.word	0
675	.quad	0x00cf9a000000ffff	/* __KERNEL32_CS */
676	.quad	0x00af9a000000ffff	/* __KERNEL_CS */
677	.quad	0x00cf92000000ffff	/* __KERNEL_DS */
678	.quad	0x0080890000000000	/* TS descriptor */
679	.quad   0x0000000000000000	/* TS continued */
680SYM_DATA_END_LABEL(gdt, SYM_L_LOCAL, gdt_end)
681
682SYM_DATA_START(boot_idt_desc)
683	.word	boot_idt_end - boot_idt - 1
684	.quad	0
685SYM_DATA_END(boot_idt_desc)
686	.balign 8
687SYM_DATA_START(boot_idt)
688	.rept	BOOT_IDT_ENTRIES
689	.quad	0
690	.quad	0
691	.endr
692SYM_DATA_END_LABEL(boot_idt, SYM_L_GLOBAL, boot_idt_end)
693
694#ifdef CONFIG_EFI_STUB
695SYM_DATA(image_offset, .long 0)
696#endif
697#ifdef CONFIG_EFI_MIXED
698SYM_DATA_LOCAL(efi32_boot_args, .long 0, 0, 0)
699SYM_DATA(efi_is64, .byte 1)
700
701#define ST32_boottime		60 // offsetof(efi_system_table_32_t, boottime)
702#define BS32_handle_protocol	88 // offsetof(efi_boot_services_32_t, handle_protocol)
703#define LI32_image_base		32 // offsetof(efi_loaded_image_32_t, image_base)
704
705	__HEAD
706	.code32
707SYM_FUNC_START(efi32_pe_entry)
708/*
709 * efi_status_t efi32_pe_entry(efi_handle_t image_handle,
710 *			       efi_system_table_32_t *sys_table)
711 */
712
713	pushl	%ebp
714	movl	%esp, %ebp
715	pushl	%eax				// dummy push to allocate loaded_image
716
717	pushl	%ebx				// save callee-save registers
718	pushl	%edi
719
720	call	verify_cpu			// check for long mode support
721	testl	%eax, %eax
722	movl	$0x80000003, %eax		// EFI_UNSUPPORTED
723	jnz	2f
724
725	call	1f
7261:	pop	%ebx
727	subl	$ rva(1b), %ebx
728
729	/* Get the loaded image protocol pointer from the image handle */
730	leal	-4(%ebp), %eax
731	pushl	%eax				// &loaded_image
732	leal	rva(loaded_image_proto)(%ebx), %eax
733	pushl	%eax				// pass the GUID address
734	pushl	8(%ebp)				// pass the image handle
735
736	/*
737	 * Note the alignment of the stack frame.
738	 *   sys_table
739	 *   handle             <-- 16-byte aligned on entry by ABI
740	 *   return address
741	 *   frame pointer
742	 *   loaded_image       <-- local variable
743	 *   saved %ebx		<-- 16-byte aligned here
744	 *   saved %edi
745	 *   &loaded_image
746	 *   &loaded_image_proto
747	 *   handle             <-- 16-byte aligned for call to handle_protocol
748	 */
749
750	movl	12(%ebp), %eax			// sys_table
751	movl	ST32_boottime(%eax), %eax	// sys_table->boottime
752	call	*BS32_handle_protocol(%eax)	// sys_table->boottime->handle_protocol
753	addl	$12, %esp			// restore argument space
754	testl	%eax, %eax
755	jnz	2f
756
757	movl	8(%ebp), %ecx			// image_handle
758	movl	12(%ebp), %edx			// sys_table
759	movl	-4(%ebp), %esi			// loaded_image
760	movl	LI32_image_base(%esi), %esi	// loaded_image->image_base
761	movl	%ebx, %ebp			// startup_32 for efi32_pe_stub_entry
762	/*
763	 * We need to set the image_offset variable here since startup_32() will
764	 * use it before we get to the 64-bit efi_pe_entry() in C code.
765	 */
766	subl	%esi, %ebx
767	movl	%ebx, rva(image_offset)(%ebp)	// save image_offset
768	jmp	efi32_pe_stub_entry
769
7702:	popl	%edi				// restore callee-save registers
771	popl	%ebx
772	leave
773	ret
774SYM_FUNC_END(efi32_pe_entry)
775
776	.section ".rodata"
777	/* EFI loaded image protocol GUID */
778	.balign 4
779SYM_DATA_START_LOCAL(loaded_image_proto)
780	.long	0x5b1b31a1
781	.word	0x9562, 0x11d2
782	.byte	0x8e, 0x3f, 0x00, 0xa0, 0xc9, 0x69, 0x72, 0x3b
783SYM_DATA_END(loaded_image_proto)
784#endif
785
786/*
787 * Stack and heap for uncompression
788 */
789	.bss
790	.balign 4
791SYM_DATA_LOCAL(boot_heap,	.fill BOOT_HEAP_SIZE, 1, 0)
792
793SYM_DATA_START_LOCAL(boot_stack)
794	.fill BOOT_STACK_SIZE, 1, 0
795	.balign 16
796SYM_DATA_END_LABEL(boot_stack, SYM_L_LOCAL, boot_stack_end)
797
798/*
799 * Space for page tables (not in .bss so not zeroed)
800 */
801	.section ".pgtable","aw",@nobits
802	.balign 4096
803SYM_DATA_LOCAL(pgtable,		.fill BOOT_PGT_SIZE, 1, 0)
804
805/*
806 * The page table is going to be used instead of page table in the trampoline
807 * memory.
808 */
809SYM_DATA_LOCAL(top_pgtable,	.fill PAGE_SIZE, 1, 0)
810