xref: /openbmc/linux/arch/x86/boot/compressed/head_64.S (revision 99a15348)
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 <asm/trapnr.h>
38#include "pgtable.h"
39
40/*
41 * Locally defined symbols should be marked hidden:
42 */
43	.hidden _bss
44	.hidden _ebss
45	.hidden _end
46
47	__HEAD
48
49/*
50 * This macro gives the relative virtual address of X, i.e. the offset of X
51 * from startup_32. This is the same as the link-time virtual address of X,
52 * since startup_32 is at 0, but defining it this way tells the
53 * assembler/linker that we do not want the actual run-time address of X. This
54 * prevents the linker from trying to create unwanted run-time relocation
55 * entries for the reference when the compressed kernel is linked as PIE.
56 *
57 * A reference X(%reg) will result in the link-time VA of X being stored with
58 * the instruction, and a run-time R_X86_64_RELATIVE relocation entry that
59 * adds the 64-bit base address where the kernel is loaded.
60 *
61 * Replacing it with (X-startup_32)(%reg) results in the offset being stored,
62 * and no run-time relocation.
63 *
64 * The macro should be used as a displacement with a base register containing
65 * the run-time address of startup_32 [i.e. rva(X)(%reg)], or as an immediate
66 * [$ rva(X)].
67 *
68 * This macro can only be used from within the .head.text section, since the
69 * expression requires startup_32 to be in the same section as the code being
70 * assembled.
71 */
72#define rva(X) ((X) - startup_32)
73
74	.code32
75SYM_FUNC_START(startup_32)
76	/*
77	 * 32bit entry is 0 and it is ABI so immutable!
78	 * If we come here directly from a bootloader,
79	 * kernel(text+data+bss+brk) ramdisk, zero_page, command line
80	 * all need to be under the 4G limit.
81	 */
82	cld
83	cli
84
85/*
86 * Calculate the delta between where we were compiled to run
87 * at and where we were actually loaded at.  This can only be done
88 * with a short local call on x86.  Nothing  else will tell us what
89 * address we are running at.  The reserved chunk of the real-mode
90 * data at 0x1e4 (defined as a scratch field) are used as the stack
91 * for this calculation. Only 4 bytes are needed.
92 */
93	leal	(BP_scratch+4)(%esi), %esp
94	call	1f
951:	popl	%ebp
96	subl	$ rva(1b), %ebp
97
98	/* Load new GDT with the 64bit segments using 32bit descriptor */
99	leal	rva(gdt)(%ebp), %eax
100	movl	%eax, 2(%eax)
101	lgdt	(%eax)
102
103	/* Load segment registers with our descriptors */
104	movl	$__BOOT_DS, %eax
105	movl	%eax, %ds
106	movl	%eax, %es
107	movl	%eax, %fs
108	movl	%eax, %gs
109	movl	%eax, %ss
110
111	/* Setup a stack and load CS from current GDT */
112	leal	rva(boot_stack_end)(%ebp), %esp
113
114	pushl	$__KERNEL32_CS
115	leal	rva(1f)(%ebp), %eax
116	pushl	%eax
117	lretl
1181:
119
120	/* Setup Exception handling for SEV-ES */
121	call	startup32_load_idt
122
123	/* Make sure cpu supports long mode. */
124	call	verify_cpu
125	testl	%eax, %eax
126	jnz	.Lno_longmode
127
128/*
129 * Compute the delta between where we were compiled to run at
130 * and where the code will actually run at.
131 *
132 * %ebp contains the address we are loaded at by the boot loader and %ebx
133 * contains the address where we should move the kernel image temporarily
134 * for safe in-place decompression.
135 */
136
137#ifdef CONFIG_RELOCATABLE
138	movl	%ebp, %ebx
139
140#ifdef CONFIG_EFI_STUB
141/*
142 * If we were loaded via the EFI LoadImage service, startup_32 will be at an
143 * offset to the start of the space allocated for the image. efi_pe_entry will
144 * set up image_offset to tell us where the image actually starts, so that we
145 * can use the full available buffer.
146 *	image_offset = startup_32 - image_base
147 * Otherwise image_offset will be zero and has no effect on the calculations.
148 */
149	subl    rva(image_offset)(%ebp), %ebx
150#endif
151
152	movl	BP_kernel_alignment(%esi), %eax
153	decl	%eax
154	addl	%eax, %ebx
155	notl	%eax
156	andl	%eax, %ebx
157	cmpl	$LOAD_PHYSICAL_ADDR, %ebx
158	jae	1f
159#endif
160	movl	$LOAD_PHYSICAL_ADDR, %ebx
1611:
162
163	/* Target address to relocate to for decompression */
164	addl	BP_init_size(%esi), %ebx
165	subl	$ rva(_end), %ebx
166
167/*
168 * Prepare for entering 64 bit mode
169 */
170
171	/* Enable PAE mode */
172	movl	%cr4, %eax
173	orl	$X86_CR4_PAE, %eax
174	movl	%eax, %cr4
175
176 /*
177  * Build early 4G boot pagetable
178  */
179	/*
180	 * If SEV is active then set the encryption mask in the page tables.
181	 * This will insure that when the kernel is copied and decompressed
182	 * it will be done so encrypted.
183	 */
184	call	get_sev_encryption_bit
185	xorl	%edx, %edx
186#ifdef	CONFIG_AMD_MEM_ENCRYPT
187	testl	%eax, %eax
188	jz	1f
189	subl	$32, %eax	/* Encryption bit is always above bit 31 */
190	bts	%eax, %edx	/* Set encryption mask for page tables */
191	/*
192	 * Mark SEV as active in sev_status so that startup32_check_sev_cbit()
193	 * will do a check. The sev_status memory will be fully initialized
194	 * with the contents of MSR_AMD_SEV_STATUS later in
195	 * set_sev_encryption_mask(). For now it is sufficient to know that SEV
196	 * is active.
197	 */
198	movl	$1, rva(sev_status)(%ebp)
1991:
200#endif
201
202	/* Initialize Page tables to 0 */
203	leal	rva(pgtable)(%ebx), %edi
204	xorl	%eax, %eax
205	movl	$(BOOT_INIT_PGT_SIZE/4), %ecx
206	rep	stosl
207
208	/* Build Level 4 */
209	leal	rva(pgtable + 0)(%ebx), %edi
210	leal	0x1007 (%edi), %eax
211	movl	%eax, 0(%edi)
212	addl	%edx, 4(%edi)
213
214	/* Build Level 3 */
215	leal	rva(pgtable + 0x1000)(%ebx), %edi
216	leal	0x1007(%edi), %eax
217	movl	$4, %ecx
2181:	movl	%eax, 0x00(%edi)
219	addl	%edx, 0x04(%edi)
220	addl	$0x00001000, %eax
221	addl	$8, %edi
222	decl	%ecx
223	jnz	1b
224
225	/* Build Level 2 */
226	leal	rva(pgtable + 0x2000)(%ebx), %edi
227	movl	$0x00000183, %eax
228	movl	$2048, %ecx
2291:	movl	%eax, 0(%edi)
230	addl	%edx, 4(%edi)
231	addl	$0x00200000, %eax
232	addl	$8, %edi
233	decl	%ecx
234	jnz	1b
235
236	/* Enable the boot page tables */
237	leal	rva(pgtable)(%ebx), %eax
238	movl	%eax, %cr3
239
240	/* Enable Long mode in EFER (Extended Feature Enable Register) */
241	movl	$MSR_EFER, %ecx
242	rdmsr
243	btsl	$_EFER_LME, %eax
244	wrmsr
245
246	/* After gdt is loaded */
247	xorl	%eax, %eax
248	lldt	%ax
249	movl    $__BOOT_TSS, %eax
250	ltr	%ax
251
252	/*
253	 * Setup for the jump to 64bit mode
254	 *
255	 * When the jump is performed we will be in long mode but
256	 * in 32bit compatibility mode with EFER.LME = 1, CS.L = 0, CS.D = 1
257	 * (and in turn EFER.LMA = 1).	To jump into 64bit mode we use
258	 * the new gdt/idt that has __KERNEL_CS with CS.L = 1.
259	 * We place all of the values on our mini stack so lret can
260	 * used to perform that far jump.
261	 */
262	leal	rva(startup_64)(%ebp), %eax
263#ifdef CONFIG_EFI_MIXED
264	movl	rva(efi32_boot_args)(%ebp), %edi
265	testl	%edi, %edi
266	jz	1f
267	leal	rva(efi64_stub_entry)(%ebp), %eax
268	movl	rva(efi32_boot_args+4)(%ebp), %esi
269	movl	rva(efi32_boot_args+8)(%ebp), %edx	// saved bootparams pointer
270	testl	%edx, %edx
271	jnz	1f
272	/*
273	 * efi_pe_entry uses MS calling convention, which requires 32 bytes of
274	 * shadow space on the stack even if all arguments are passed in
275	 * registers. We also need an additional 8 bytes for the space that
276	 * would be occupied by the return address, and this also results in
277	 * the correct stack alignment for entry.
278	 */
279	subl	$40, %esp
280	leal	rva(efi_pe_entry)(%ebp), %eax
281	movl	%edi, %ecx			// MS calling convention
282	movl	%esi, %edx
2831:
284#endif
285	/* Check if the C-bit position is correct when SEV is active */
286	call	startup32_check_sev_cbit
287
288	pushl	$__KERNEL_CS
289	pushl	%eax
290
291	/* Enter paged protected Mode, activating Long Mode */
292	movl	$(X86_CR0_PG | X86_CR0_PE), %eax /* Enable Paging and Protected mode */
293	movl	%eax, %cr0
294
295	/* Jump from 32bit compatibility mode into 64bit mode. */
296	lret
297SYM_FUNC_END(startup_32)
298
299#ifdef CONFIG_EFI_MIXED
300	.org 0x190
301SYM_FUNC_START(efi32_stub_entry)
302	add	$0x4, %esp		/* Discard return address */
303	popl	%ecx
304	popl	%edx
305	popl	%esi
306
307	call	1f
3081:	pop	%ebp
309	subl	$ rva(1b), %ebp
310
311	movl	%esi, rva(efi32_boot_args+8)(%ebp)
312SYM_INNER_LABEL(efi32_pe_stub_entry, SYM_L_LOCAL)
313	movl	%ecx, rva(efi32_boot_args)(%ebp)
314	movl	%edx, rva(efi32_boot_args+4)(%ebp)
315	movb	$0, rva(efi_is64)(%ebp)
316
317	/* Save firmware GDTR and code/data selectors */
318	sgdtl	rva(efi32_boot_gdt)(%ebp)
319	movw	%cs, rva(efi32_boot_cs)(%ebp)
320	movw	%ds, rva(efi32_boot_ds)(%ebp)
321
322	/* Store firmware IDT descriptor */
323	sidtl	rva(efi32_boot_idt)(%ebp)
324
325	/* Disable paging */
326	movl	%cr0, %eax
327	btrl	$X86_CR0_PG_BIT, %eax
328	movl	%eax, %cr0
329
330	jmp	startup_32
331SYM_FUNC_END(efi32_stub_entry)
332#endif
333
334	.code64
335	.org 0x200
336SYM_CODE_START(startup_64)
337	/*
338	 * 64bit entry is 0x200 and it is ABI so immutable!
339	 * We come here either from startup_32 or directly from a
340	 * 64bit bootloader.
341	 * If we come here from a bootloader, kernel(text+data+bss+brk),
342	 * ramdisk, zero_page, command line could be above 4G.
343	 * We depend on an identity mapped page table being provided
344	 * that maps our entire kernel(text+data+bss+brk), zero page
345	 * and command line.
346	 */
347
348	cld
349	cli
350
351	/* Setup data segments. */
352	xorl	%eax, %eax
353	movl	%eax, %ds
354	movl	%eax, %es
355	movl	%eax, %ss
356	movl	%eax, %fs
357	movl	%eax, %gs
358
359	/*
360	 * Compute the decompressed kernel start address.  It is where
361	 * we were loaded at aligned to a 2M boundary. %rbp contains the
362	 * decompressed kernel start address.
363	 *
364	 * If it is a relocatable kernel then decompress and run the kernel
365	 * from load address aligned to 2MB addr, otherwise decompress and
366	 * run the kernel from LOAD_PHYSICAL_ADDR
367	 *
368	 * We cannot rely on the calculation done in 32-bit mode, since we
369	 * may have been invoked via the 64-bit entry point.
370	 */
371
372	/* Start with the delta to where the kernel will run at. */
373#ifdef CONFIG_RELOCATABLE
374	leaq	startup_32(%rip) /* - $startup_32 */, %rbp
375
376#ifdef CONFIG_EFI_STUB
377/*
378 * If we were loaded via the EFI LoadImage service, startup_32 will be at an
379 * offset to the start of the space allocated for the image. efi_pe_entry will
380 * set up image_offset to tell us where the image actually starts, so that we
381 * can use the full available buffer.
382 *	image_offset = startup_32 - image_base
383 * Otherwise image_offset will be zero and has no effect on the calculations.
384 */
385	movl    image_offset(%rip), %eax
386	subq	%rax, %rbp
387#endif
388
389	movl	BP_kernel_alignment(%rsi), %eax
390	decl	%eax
391	addq	%rax, %rbp
392	notq	%rax
393	andq	%rax, %rbp
394	cmpq	$LOAD_PHYSICAL_ADDR, %rbp
395	jae	1f
396#endif
397	movq	$LOAD_PHYSICAL_ADDR, %rbp
3981:
399
400	/* Target address to relocate to for decompression */
401	movl	BP_init_size(%rsi), %ebx
402	subl	$ rva(_end), %ebx
403	addq	%rbp, %rbx
404
405	/* Set up the stack */
406	leaq	rva(boot_stack_end)(%rbx), %rsp
407
408	/*
409	 * At this point we are in long mode with 4-level paging enabled,
410	 * but we might want to enable 5-level paging or vice versa.
411	 *
412	 * The problem is that we cannot do it directly. Setting or clearing
413	 * CR4.LA57 in long mode would trigger #GP. So we need to switch off
414	 * long mode and paging first.
415	 *
416	 * We also need a trampoline in lower memory to switch over from
417	 * 4- to 5-level paging for cases when the bootloader puts the kernel
418	 * above 4G, but didn't enable 5-level paging for us.
419	 *
420	 * The same trampoline can be used to switch from 5- to 4-level paging
421	 * mode, like when starting 4-level paging kernel via kexec() when
422	 * original kernel worked in 5-level paging mode.
423	 *
424	 * For the trampoline, we need the top page table to reside in lower
425	 * memory as we don't have a way to load 64-bit values into CR3 in
426	 * 32-bit mode.
427	 *
428	 * We go though the trampoline even if we don't have to: if we're
429	 * already in a desired paging mode. This way the trampoline code gets
430	 * tested on every boot.
431	 */
432
433	/* Make sure we have GDT with 32-bit code segment */
434	leaq	gdt64(%rip), %rax
435	addq	%rax, 2(%rax)
436	lgdt	(%rax)
437
438	/* Reload CS so IRET returns to a CS actually in the GDT */
439	pushq	$__KERNEL_CS
440	leaq	.Lon_kernel_cs(%rip), %rax
441	pushq	%rax
442	lretq
443
444.Lon_kernel_cs:
445
446	pushq	%rsi
447	call	load_stage1_idt
448	popq	%rsi
449
450	/*
451	 * paging_prepare() sets up the trampoline and checks if we need to
452	 * enable 5-level paging.
453	 *
454	 * paging_prepare() returns a two-quadword structure which lands
455	 * into RDX:RAX:
456	 *   - Address of the trampoline is returned in RAX.
457	 *   - Non zero RDX means trampoline needs to enable 5-level
458	 *     paging.
459	 *
460	 * RSI holds real mode data and needs to be preserved across
461	 * this function call.
462	 */
463	pushq	%rsi
464	movq	%rsi, %rdi		/* real mode address */
465	call	paging_prepare
466	popq	%rsi
467
468	/* Save the trampoline address in RCX */
469	movq	%rax, %rcx
470
471	/*
472	 * Load the address of trampoline_return() into RDI.
473	 * It will be used by the trampoline to return to the main code.
474	 */
475	leaq	trampoline_return(%rip), %rdi
476
477	/* Switch to compatibility mode (CS.L = 0 CS.D = 1) via far return */
478	pushq	$__KERNEL32_CS
479	leaq	TRAMPOLINE_32BIT_CODE_OFFSET(%rax), %rax
480	pushq	%rax
481	lretq
482trampoline_return:
483	/* Restore the stack, the 32-bit trampoline uses its own stack */
484	leaq	rva(boot_stack_end)(%rbx), %rsp
485
486	/*
487	 * cleanup_trampoline() would restore trampoline memory.
488	 *
489	 * RDI is address of the page table to use instead of page table
490	 * in trampoline memory (if required).
491	 *
492	 * RSI holds real mode data and needs to be preserved across
493	 * this function call.
494	 */
495	pushq	%rsi
496	leaq	rva(top_pgtable)(%rbx), %rdi
497	call	cleanup_trampoline
498	popq	%rsi
499
500	/* Zero EFLAGS */
501	pushq	$0
502	popfq
503
504/*
505 * Copy the compressed kernel to the end of our buffer
506 * where decompression in place becomes safe.
507 */
508	pushq	%rsi
509	leaq	(_bss-8)(%rip), %rsi
510	leaq	rva(_bss-8)(%rbx), %rdi
511	movl	$(_bss - startup_32), %ecx
512	shrl	$3, %ecx
513	std
514	rep	movsq
515	cld
516	popq	%rsi
517
518	/*
519	 * The GDT may get overwritten either during the copy we just did or
520	 * during extract_kernel below. To avoid any issues, repoint the GDTR
521	 * to the new copy of the GDT.
522	 */
523	leaq	rva(gdt64)(%rbx), %rax
524	leaq	rva(gdt)(%rbx), %rdx
525	movq	%rdx, 2(%rax)
526	lgdt	(%rax)
527
528/*
529 * Jump to the relocated address.
530 */
531	leaq	rva(.Lrelocated)(%rbx), %rax
532	jmp	*%rax
533SYM_CODE_END(startup_64)
534
535#ifdef CONFIG_EFI_STUB
536	.org 0x390
537SYM_FUNC_START(efi64_stub_entry)
538	and	$~0xf, %rsp			/* realign the stack */
539	movq	%rdx, %rbx			/* save boot_params pointer */
540	call	efi_main
541	movq	%rbx,%rsi
542	leaq	rva(startup_64)(%rax), %rax
543	jmp	*%rax
544SYM_FUNC_END(efi64_stub_entry)
545SYM_FUNC_ALIAS(efi_stub_entry, efi64_stub_entry)
546#endif
547
548	.text
549SYM_FUNC_START_LOCAL_NOALIGN(.Lrelocated)
550
551/*
552 * Clear BSS (stack is currently empty)
553 */
554	xorl	%eax, %eax
555	leaq    _bss(%rip), %rdi
556	leaq    _ebss(%rip), %rcx
557	subq	%rdi, %rcx
558	shrq	$3, %rcx
559	rep	stosq
560
561/*
562 * If running as an SEV guest, the encryption mask is required in the
563 * page-table setup code below. When the guest also has SEV-ES enabled
564 * set_sev_encryption_mask() will cause #VC exceptions, but the stage2
565 * handler can't map its GHCB because the page-table is not set up yet.
566 * So set up the encryption mask here while still on the stage1 #VC
567 * handler. Then load stage2 IDT and switch to the kernel's own
568 * page-table.
569 */
570	pushq	%rsi
571	call	set_sev_encryption_mask
572	call	load_stage2_idt
573
574	/* Pass boot_params to initialize_identity_maps() */
575	movq	(%rsp), %rdi
576	call	initialize_identity_maps
577	popq	%rsi
578
579/*
580 * Do the extraction, and jump to the new kernel..
581 */
582	pushq	%rsi			/* Save the real mode argument */
583	movq	%rsi, %rdi		/* real mode address */
584	leaq	boot_heap(%rip), %rsi	/* malloc area for uncompression */
585	leaq	input_data(%rip), %rdx  /* input_data */
586	movl	input_len(%rip), %ecx	/* input_len */
587	movq	%rbp, %r8		/* output target address */
588	movl	output_len(%rip), %r9d	/* decompressed length, end of relocs */
589	call	extract_kernel		/* returns kernel location in %rax */
590	popq	%rsi
591
592/*
593 * Jump to the decompressed kernel.
594 */
595	jmp	*%rax
596SYM_FUNC_END(.Lrelocated)
597
598	.code32
599/*
600 * This is the 32-bit trampoline that will be copied over to low memory.
601 *
602 * RDI contains the return address (might be above 4G).
603 * ECX contains the base address of the trampoline memory.
604 * Non zero RDX means trampoline needs to enable 5-level paging.
605 */
606SYM_CODE_START(trampoline_32bit_src)
607	/* Set up data and stack segments */
608	movl	$__KERNEL_DS, %eax
609	movl	%eax, %ds
610	movl	%eax, %ss
611
612	/* Set up new stack */
613	leal	TRAMPOLINE_32BIT_STACK_END(%ecx), %esp
614
615	/* Disable paging */
616	movl	%cr0, %eax
617	btrl	$X86_CR0_PG_BIT, %eax
618	movl	%eax, %cr0
619
620	/* Check what paging mode we want to be in after the trampoline */
621	testl	%edx, %edx
622	jz	1f
623
624	/* We want 5-level paging: don't touch CR3 if it already points to 5-level page tables */
625	movl	%cr4, %eax
626	testl	$X86_CR4_LA57, %eax
627	jnz	3f
628	jmp	2f
6291:
630	/* We want 4-level paging: don't touch CR3 if it already points to 4-level page tables */
631	movl	%cr4, %eax
632	testl	$X86_CR4_LA57, %eax
633	jz	3f
6342:
635	/* Point CR3 to the trampoline's new top level page table */
636	leal	TRAMPOLINE_32BIT_PGTABLE_OFFSET(%ecx), %eax
637	movl	%eax, %cr3
6383:
639	/* Set EFER.LME=1 as a precaution in case hypervsior pulls the rug */
640	pushl	%ecx
641	pushl	%edx
642	movl	$MSR_EFER, %ecx
643	rdmsr
644	btsl	$_EFER_LME, %eax
645	wrmsr
646	popl	%edx
647	popl	%ecx
648
649	/* Enable PAE and LA57 (if required) paging modes */
650	movl	$X86_CR4_PAE, %eax
651	testl	%edx, %edx
652	jz	1f
653	orl	$X86_CR4_LA57, %eax
6541:
655	movl	%eax, %cr4
656
657	/* Calculate address of paging_enabled() once we are executing in the trampoline */
658	leal	.Lpaging_enabled - trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_OFFSET(%ecx), %eax
659
660	/* Prepare the stack for far return to Long Mode */
661	pushl	$__KERNEL_CS
662	pushl	%eax
663
664	/* Enable paging again */
665	movl	$(X86_CR0_PG | X86_CR0_PE), %eax
666	movl	%eax, %cr0
667
668	lret
669SYM_CODE_END(trampoline_32bit_src)
670
671	.code64
672SYM_FUNC_START_LOCAL_NOALIGN(.Lpaging_enabled)
673	/* Return from the trampoline */
674	jmp	*%rdi
675SYM_FUNC_END(.Lpaging_enabled)
676
677	/*
678         * The trampoline code has a size limit.
679         * Make sure we fail to compile if the trampoline code grows
680         * beyond TRAMPOLINE_32BIT_CODE_SIZE bytes.
681	 */
682	.org	trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_SIZE
683
684	.code32
685SYM_FUNC_START_LOCAL_NOALIGN(.Lno_longmode)
686	/* This isn't an x86-64 CPU, so hang intentionally, we cannot continue */
6871:
688	hlt
689	jmp     1b
690SYM_FUNC_END(.Lno_longmode)
691
692#include "../../kernel/verify_cpu.S"
693
694	.data
695SYM_DATA_START_LOCAL(gdt64)
696	.word	gdt_end - gdt - 1
697	.quad   gdt - gdt64
698SYM_DATA_END(gdt64)
699	.balign	8
700SYM_DATA_START_LOCAL(gdt)
701	.word	gdt_end - gdt - 1
702	.long	0
703	.word	0
704	.quad	0x00cf9a000000ffff	/* __KERNEL32_CS */
705	.quad	0x00af9a000000ffff	/* __KERNEL_CS */
706	.quad	0x00cf92000000ffff	/* __KERNEL_DS */
707	.quad	0x0080890000000000	/* TS descriptor */
708	.quad   0x0000000000000000	/* TS continued */
709SYM_DATA_END_LABEL(gdt, SYM_L_LOCAL, gdt_end)
710
711SYM_DATA_START(boot_idt_desc)
712	.word	boot_idt_end - boot_idt - 1
713	.quad	0
714SYM_DATA_END(boot_idt_desc)
715	.balign 8
716SYM_DATA_START(boot_idt)
717	.rept	BOOT_IDT_ENTRIES
718	.quad	0
719	.quad	0
720	.endr
721SYM_DATA_END_LABEL(boot_idt, SYM_L_GLOBAL, boot_idt_end)
722
723#ifdef CONFIG_AMD_MEM_ENCRYPT
724SYM_DATA_START(boot32_idt_desc)
725	.word   boot32_idt_end - boot32_idt - 1
726	.long   0
727SYM_DATA_END(boot32_idt_desc)
728	.balign 8
729SYM_DATA_START(boot32_idt)
730	.rept 32
731	.quad 0
732	.endr
733SYM_DATA_END_LABEL(boot32_idt, SYM_L_GLOBAL, boot32_idt_end)
734#endif
735
736#ifdef CONFIG_EFI_STUB
737SYM_DATA(image_offset, .long 0)
738#endif
739#ifdef CONFIG_EFI_MIXED
740SYM_DATA_LOCAL(efi32_boot_args, .long 0, 0, 0)
741SYM_DATA(efi_is64, .byte 1)
742
743#define ST32_boottime		60 // offsetof(efi_system_table_32_t, boottime)
744#define BS32_handle_protocol	88 // offsetof(efi_boot_services_32_t, handle_protocol)
745#define LI32_image_base		32 // offsetof(efi_loaded_image_32_t, image_base)
746
747	__HEAD
748	.code32
749SYM_FUNC_START(efi32_pe_entry)
750/*
751 * efi_status_t efi32_pe_entry(efi_handle_t image_handle,
752 *			       efi_system_table_32_t *sys_table)
753 */
754
755	pushl	%ebp
756	movl	%esp, %ebp
757	pushl	%eax				// dummy push to allocate loaded_image
758
759	pushl	%ebx				// save callee-save registers
760	pushl	%edi
761
762	call	verify_cpu			// check for long mode support
763	testl	%eax, %eax
764	movl	$0x80000003, %eax		// EFI_UNSUPPORTED
765	jnz	2f
766
767	call	1f
7681:	pop	%ebx
769	subl	$ rva(1b), %ebx
770
771	/* Get the loaded image protocol pointer from the image handle */
772	leal	-4(%ebp), %eax
773	pushl	%eax				// &loaded_image
774	leal	rva(loaded_image_proto)(%ebx), %eax
775	pushl	%eax				// pass the GUID address
776	pushl	8(%ebp)				// pass the image handle
777
778	/*
779	 * Note the alignment of the stack frame.
780	 *   sys_table
781	 *   handle             <-- 16-byte aligned on entry by ABI
782	 *   return address
783	 *   frame pointer
784	 *   loaded_image       <-- local variable
785	 *   saved %ebx		<-- 16-byte aligned here
786	 *   saved %edi
787	 *   &loaded_image
788	 *   &loaded_image_proto
789	 *   handle             <-- 16-byte aligned for call to handle_protocol
790	 */
791
792	movl	12(%ebp), %eax			// sys_table
793	movl	ST32_boottime(%eax), %eax	// sys_table->boottime
794	call	*BS32_handle_protocol(%eax)	// sys_table->boottime->handle_protocol
795	addl	$12, %esp			// restore argument space
796	testl	%eax, %eax
797	jnz	2f
798
799	movl	8(%ebp), %ecx			// image_handle
800	movl	12(%ebp), %edx			// sys_table
801	movl	-4(%ebp), %esi			// loaded_image
802	movl	LI32_image_base(%esi), %esi	// loaded_image->image_base
803	movl	%ebx, %ebp			// startup_32 for efi32_pe_stub_entry
804	/*
805	 * We need to set the image_offset variable here since startup_32() will
806	 * use it before we get to the 64-bit efi_pe_entry() in C code.
807	 */
808	subl	%esi, %ebx
809	movl	%ebx, rva(image_offset)(%ebp)	// save image_offset
810	jmp	efi32_pe_stub_entry
811
8122:	popl	%edi				// restore callee-save registers
813	popl	%ebx
814	leave
815	RET
816SYM_FUNC_END(efi32_pe_entry)
817
818	.section ".rodata"
819	/* EFI loaded image protocol GUID */
820	.balign 4
821SYM_DATA_START_LOCAL(loaded_image_proto)
822	.long	0x5b1b31a1
823	.word	0x9562, 0x11d2
824	.byte	0x8e, 0x3f, 0x00, 0xa0, 0xc9, 0x69, 0x72, 0x3b
825SYM_DATA_END(loaded_image_proto)
826#endif
827
828#ifdef CONFIG_AMD_MEM_ENCRYPT
829	__HEAD
830	.code32
831/*
832 * Write an IDT entry into boot32_idt
833 *
834 * Parameters:
835 *
836 * %eax:	Handler address
837 * %edx:	Vector number
838 *
839 * Physical offset is expected in %ebp
840 */
841SYM_FUNC_START(startup32_set_idt_entry)
842	push    %ebx
843	push    %ecx
844
845	/* IDT entry address to %ebx */
846	leal    rva(boot32_idt)(%ebp), %ebx
847	shl	$3, %edx
848	addl    %edx, %ebx
849
850	/* Build IDT entry, lower 4 bytes */
851	movl    %eax, %edx
852	andl    $0x0000ffff, %edx	# Target code segment offset [15:0]
853	movl    $__KERNEL32_CS, %ecx	# Target code segment selector
854	shl     $16, %ecx
855	orl     %ecx, %edx
856
857	/* Store lower 4 bytes to IDT */
858	movl    %edx, (%ebx)
859
860	/* Build IDT entry, upper 4 bytes */
861	movl    %eax, %edx
862	andl    $0xffff0000, %edx	# Target code segment offset [31:16]
863	orl     $0x00008e00, %edx	# Present, Type 32-bit Interrupt Gate
864
865	/* Store upper 4 bytes to IDT */
866	movl    %edx, 4(%ebx)
867
868	pop     %ecx
869	pop     %ebx
870	RET
871SYM_FUNC_END(startup32_set_idt_entry)
872#endif
873
874SYM_FUNC_START(startup32_load_idt)
875#ifdef CONFIG_AMD_MEM_ENCRYPT
876	/* #VC handler */
877	leal    rva(startup32_vc_handler)(%ebp), %eax
878	movl    $X86_TRAP_VC, %edx
879	call    startup32_set_idt_entry
880
881	/* Load IDT */
882	leal	rva(boot32_idt)(%ebp), %eax
883	movl	%eax, rva(boot32_idt_desc+2)(%ebp)
884	lidt    rva(boot32_idt_desc)(%ebp)
885#endif
886	RET
887SYM_FUNC_END(startup32_load_idt)
888
889/*
890 * Check for the correct C-bit position when the startup_32 boot-path is used.
891 *
892 * The check makes use of the fact that all memory is encrypted when paging is
893 * disabled. The function creates 64 bits of random data using the RDRAND
894 * instruction. RDRAND is mandatory for SEV guests, so always available. If the
895 * hypervisor violates that the kernel will crash right here.
896 *
897 * The 64 bits of random data are stored to a memory location and at the same
898 * time kept in the %eax and %ebx registers. Since encryption is always active
899 * when paging is off the random data will be stored encrypted in main memory.
900 *
901 * Then paging is enabled. When the C-bit position is correct all memory is
902 * still mapped encrypted and comparing the register values with memory will
903 * succeed. An incorrect C-bit position will map all memory unencrypted, so that
904 * the compare will use the encrypted random data and fail.
905 */
906SYM_FUNC_START(startup32_check_sev_cbit)
907#ifdef CONFIG_AMD_MEM_ENCRYPT
908	pushl	%eax
909	pushl	%ebx
910	pushl	%ecx
911	pushl	%edx
912
913	/* Check for non-zero sev_status */
914	movl	rva(sev_status)(%ebp), %eax
915	testl	%eax, %eax
916	jz	4f
917
918	/*
919	 * Get two 32-bit random values - Don't bail out if RDRAND fails
920	 * because it is better to prevent forward progress if no random value
921	 * can be gathered.
922	 */
9231:	rdrand	%eax
924	jnc	1b
9252:	rdrand	%ebx
926	jnc	2b
927
928	/* Store to memory and keep it in the registers */
929	movl	%eax, rva(sev_check_data)(%ebp)
930	movl	%ebx, rva(sev_check_data+4)(%ebp)
931
932	/* Enable paging to see if encryption is active */
933	movl	%cr0, %edx			 /* Backup %cr0 in %edx */
934	movl	$(X86_CR0_PG | X86_CR0_PE), %ecx /* Enable Paging and Protected mode */
935	movl	%ecx, %cr0
936
937	cmpl	%eax, rva(sev_check_data)(%ebp)
938	jne	3f
939	cmpl	%ebx, rva(sev_check_data+4)(%ebp)
940	jne	3f
941
942	movl	%edx, %cr0	/* Restore previous %cr0 */
943
944	jmp	4f
945
9463:	/* Check failed - hlt the machine */
947	hlt
948	jmp	3b
949
9504:
951	popl	%edx
952	popl	%ecx
953	popl	%ebx
954	popl	%eax
955#endif
956	RET
957SYM_FUNC_END(startup32_check_sev_cbit)
958
959/*
960 * Stack and heap for uncompression
961 */
962	.bss
963	.balign 4
964SYM_DATA_LOCAL(boot_heap,	.fill BOOT_HEAP_SIZE, 1, 0)
965
966SYM_DATA_START_LOCAL(boot_stack)
967	.fill BOOT_STACK_SIZE, 1, 0
968	.balign 16
969SYM_DATA_END_LABEL(boot_stack, SYM_L_LOCAL, boot_stack_end)
970
971/*
972 * Space for page tables (not in .bss so not zeroed)
973 */
974	.section ".pgtable","aw",@nobits
975	.balign 4096
976SYM_DATA_LOCAL(pgtable,		.fill BOOT_PGT_SIZE, 1, 0)
977
978/*
979 * The page table is going to be used instead of page table in the trampoline
980 * memory.
981 */
982SYM_DATA_LOCAL(top_pgtable,	.fill PAGE_SIZE, 1, 0)
983