cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

head_64.S (26441B)


      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	 * Set MSR_AMD64_SEV_ENABLED_BIT in sev_status so that
    193	 * startup32_check_sev_cbit() will do a check. sev_enable() will
    194	 * initialize sev_status with all the bits reported by
    195	 * MSR_AMD_SEV_STATUS later, but only MSR_AMD64_SEV_ENABLED_BIT
    196	 * needs to be set for now.
    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	$CR0_STATE, %eax
    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#ifdef CONFIG_AMD_MEM_ENCRYPT
    451	/*
    452	 * Now that the stage1 interrupt handlers are set up, #VC exceptions from
    453	 * CPUID instructions can be properly handled for SEV-ES guests.
    454	 *
    455	 * For SEV-SNP, the CPUID table also needs to be set up in advance of any
    456	 * CPUID instructions being issued, so go ahead and do that now via
    457	 * sev_enable(), which will also handle the rest of the SEV-related
    458	 * detection/setup to ensure that has been done in advance of any dependent
    459	 * code.
    460	 */
    461	pushq	%rsi
    462	movq	%rsi, %rdi		/* real mode address */
    463	call	sev_enable
    464	popq	%rsi
    465#endif
    466
    467	/*
    468	 * paging_prepare() sets up the trampoline and checks if we need to
    469	 * enable 5-level paging.
    470	 *
    471	 * paging_prepare() returns a two-quadword structure which lands
    472	 * into RDX:RAX:
    473	 *   - Address of the trampoline is returned in RAX.
    474	 *   - Non zero RDX means trampoline needs to enable 5-level
    475	 *     paging.
    476	 *
    477	 * RSI holds real mode data and needs to be preserved across
    478	 * this function call.
    479	 */
    480	pushq	%rsi
    481	movq	%rsi, %rdi		/* real mode address */
    482	call	paging_prepare
    483	popq	%rsi
    484
    485	/* Save the trampoline address in RCX */
    486	movq	%rax, %rcx
    487
    488	/*
    489	 * Load the address of trampoline_return() into RDI.
    490	 * It will be used by the trampoline to return to the main code.
    491	 */
    492	leaq	trampoline_return(%rip), %rdi
    493
    494	/* Switch to compatibility mode (CS.L = 0 CS.D = 1) via far return */
    495	pushq	$__KERNEL32_CS
    496	leaq	TRAMPOLINE_32BIT_CODE_OFFSET(%rax), %rax
    497	pushq	%rax
    498	lretq
    499trampoline_return:
    500	/* Restore the stack, the 32-bit trampoline uses its own stack */
    501	leaq	rva(boot_stack_end)(%rbx), %rsp
    502
    503	/*
    504	 * cleanup_trampoline() would restore trampoline memory.
    505	 *
    506	 * RDI is address of the page table to use instead of page table
    507	 * in trampoline memory (if required).
    508	 *
    509	 * RSI holds real mode data and needs to be preserved across
    510	 * this function call.
    511	 */
    512	pushq	%rsi
    513	leaq	rva(top_pgtable)(%rbx), %rdi
    514	call	cleanup_trampoline
    515	popq	%rsi
    516
    517	/* Zero EFLAGS */
    518	pushq	$0
    519	popfq
    520
    521/*
    522 * Copy the compressed kernel to the end of our buffer
    523 * where decompression in place becomes safe.
    524 */
    525	pushq	%rsi
    526	leaq	(_bss-8)(%rip), %rsi
    527	leaq	rva(_bss-8)(%rbx), %rdi
    528	movl	$(_bss - startup_32), %ecx
    529	shrl	$3, %ecx
    530	std
    531	rep	movsq
    532	cld
    533	popq	%rsi
    534
    535	/*
    536	 * The GDT may get overwritten either during the copy we just did or
    537	 * during extract_kernel below. To avoid any issues, repoint the GDTR
    538	 * to the new copy of the GDT.
    539	 */
    540	leaq	rva(gdt64)(%rbx), %rax
    541	leaq	rva(gdt)(%rbx), %rdx
    542	movq	%rdx, 2(%rax)
    543	lgdt	(%rax)
    544
    545/*
    546 * Jump to the relocated address.
    547 */
    548	leaq	rva(.Lrelocated)(%rbx), %rax
    549	jmp	*%rax
    550SYM_CODE_END(startup_64)
    551
    552#ifdef CONFIG_EFI_STUB
    553	.org 0x390
    554SYM_FUNC_START(efi64_stub_entry)
    555	and	$~0xf, %rsp			/* realign the stack */
    556	movq	%rdx, %rbx			/* save boot_params pointer */
    557	call	efi_main
    558	movq	%rbx,%rsi
    559	leaq	rva(startup_64)(%rax), %rax
    560	jmp	*%rax
    561SYM_FUNC_END(efi64_stub_entry)
    562SYM_FUNC_ALIAS(efi_stub_entry, efi64_stub_entry)
    563#endif
    564
    565	.text
    566SYM_FUNC_START_LOCAL_NOALIGN(.Lrelocated)
    567
    568/*
    569 * Clear BSS (stack is currently empty)
    570 */
    571	xorl	%eax, %eax
    572	leaq    _bss(%rip), %rdi
    573	leaq    _ebss(%rip), %rcx
    574	subq	%rdi, %rcx
    575	shrq	$3, %rcx
    576	rep	stosq
    577
    578	pushq	%rsi
    579	call	load_stage2_idt
    580
    581	/* Pass boot_params to initialize_identity_maps() */
    582	movq	(%rsp), %rdi
    583	call	initialize_identity_maps
    584	popq	%rsi
    585
    586/*
    587 * Do the extraction, and jump to the new kernel..
    588 */
    589	pushq	%rsi			/* Save the real mode argument */
    590	movq	%rsi, %rdi		/* real mode address */
    591	leaq	boot_heap(%rip), %rsi	/* malloc area for uncompression */
    592	leaq	input_data(%rip), %rdx  /* input_data */
    593	movl	input_len(%rip), %ecx	/* input_len */
    594	movq	%rbp, %r8		/* output target address */
    595	movl	output_len(%rip), %r9d	/* decompressed length, end of relocs */
    596	call	extract_kernel		/* returns kernel location in %rax */
    597	popq	%rsi
    598
    599/*
    600 * Jump to the decompressed kernel.
    601 */
    602	jmp	*%rax
    603SYM_FUNC_END(.Lrelocated)
    604
    605	.code32
    606/*
    607 * This is the 32-bit trampoline that will be copied over to low memory.
    608 *
    609 * RDI contains the return address (might be above 4G).
    610 * ECX contains the base address of the trampoline memory.
    611 * Non zero RDX means trampoline needs to enable 5-level paging.
    612 */
    613SYM_CODE_START(trampoline_32bit_src)
    614	/* Set up data and stack segments */
    615	movl	$__KERNEL_DS, %eax
    616	movl	%eax, %ds
    617	movl	%eax, %ss
    618
    619	/* Set up new stack */
    620	leal	TRAMPOLINE_32BIT_STACK_END(%ecx), %esp
    621
    622	/* Disable paging */
    623	movl	%cr0, %eax
    624	btrl	$X86_CR0_PG_BIT, %eax
    625	movl	%eax, %cr0
    626
    627	/* Check what paging mode we want to be in after the trampoline */
    628	testl	%edx, %edx
    629	jz	1f
    630
    631	/* We want 5-level paging: don't touch CR3 if it already points to 5-level page tables */
    632	movl	%cr4, %eax
    633	testl	$X86_CR4_LA57, %eax
    634	jnz	3f
    635	jmp	2f
    6361:
    637	/* We want 4-level paging: don't touch CR3 if it already points to 4-level page tables */
    638	movl	%cr4, %eax
    639	testl	$X86_CR4_LA57, %eax
    640	jz	3f
    6412:
    642	/* Point CR3 to the trampoline's new top level page table */
    643	leal	TRAMPOLINE_32BIT_PGTABLE_OFFSET(%ecx), %eax
    644	movl	%eax, %cr3
    6453:
    646	/* Set EFER.LME=1 as a precaution in case hypervsior pulls the rug */
    647	pushl	%ecx
    648	pushl	%edx
    649	movl	$MSR_EFER, %ecx
    650	rdmsr
    651	btsl	$_EFER_LME, %eax
    652	/* Avoid writing EFER if no change was made (for TDX guest) */
    653	jc	1f
    654	wrmsr
    6551:	popl	%edx
    656	popl	%ecx
    657
    658#ifdef CONFIG_X86_MCE
    659	/*
    660	 * Preserve CR4.MCE if the kernel will enable #MC support.
    661	 * Clearing MCE may fault in some environments (that also force #MC
    662	 * support). Any machine check that occurs before #MC support is fully
    663	 * configured will crash the system regardless of the CR4.MCE value set
    664	 * here.
    665	 */
    666	movl	%cr4, %eax
    667	andl	$X86_CR4_MCE, %eax
    668#else
    669	movl	$0, %eax
    670#endif
    671
    672	/* Enable PAE and LA57 (if required) paging modes */
    673	orl	$X86_CR4_PAE, %eax
    674	testl	%edx, %edx
    675	jz	1f
    676	orl	$X86_CR4_LA57, %eax
    6771:
    678	movl	%eax, %cr4
    679
    680	/* Calculate address of paging_enabled() once we are executing in the trampoline */
    681	leal	.Lpaging_enabled - trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_OFFSET(%ecx), %eax
    682
    683	/* Prepare the stack for far return to Long Mode */
    684	pushl	$__KERNEL_CS
    685	pushl	%eax
    686
    687	/* Enable paging again. */
    688	movl	%cr0, %eax
    689	btsl	$X86_CR0_PG_BIT, %eax
    690	movl	%eax, %cr0
    691
    692	lret
    693SYM_CODE_END(trampoline_32bit_src)
    694
    695	.code64
    696SYM_FUNC_START_LOCAL_NOALIGN(.Lpaging_enabled)
    697	/* Return from the trampoline */
    698	jmp	*%rdi
    699SYM_FUNC_END(.Lpaging_enabled)
    700
    701	/*
    702         * The trampoline code has a size limit.
    703         * Make sure we fail to compile if the trampoline code grows
    704         * beyond TRAMPOLINE_32BIT_CODE_SIZE bytes.
    705	 */
    706	.org	trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_SIZE
    707
    708	.code32
    709SYM_FUNC_START_LOCAL_NOALIGN(.Lno_longmode)
    710	/* This isn't an x86-64 CPU, so hang intentionally, we cannot continue */
    7111:
    712	hlt
    713	jmp     1b
    714SYM_FUNC_END(.Lno_longmode)
    715
    716#include "../../kernel/verify_cpu.S"
    717
    718	.data
    719SYM_DATA_START_LOCAL(gdt64)
    720	.word	gdt_end - gdt - 1
    721	.quad   gdt - gdt64
    722SYM_DATA_END(gdt64)
    723	.balign	8
    724SYM_DATA_START_LOCAL(gdt)
    725	.word	gdt_end - gdt - 1
    726	.long	0
    727	.word	0
    728	.quad	0x00cf9a000000ffff	/* __KERNEL32_CS */
    729	.quad	0x00af9a000000ffff	/* __KERNEL_CS */
    730	.quad	0x00cf92000000ffff	/* __KERNEL_DS */
    731	.quad	0x0080890000000000	/* TS descriptor */
    732	.quad   0x0000000000000000	/* TS continued */
    733SYM_DATA_END_LABEL(gdt, SYM_L_LOCAL, gdt_end)
    734
    735SYM_DATA_START(boot_idt_desc)
    736	.word	boot_idt_end - boot_idt - 1
    737	.quad	0
    738SYM_DATA_END(boot_idt_desc)
    739	.balign 8
    740SYM_DATA_START(boot_idt)
    741	.rept	BOOT_IDT_ENTRIES
    742	.quad	0
    743	.quad	0
    744	.endr
    745SYM_DATA_END_LABEL(boot_idt, SYM_L_GLOBAL, boot_idt_end)
    746
    747#ifdef CONFIG_AMD_MEM_ENCRYPT
    748SYM_DATA_START(boot32_idt_desc)
    749	.word   boot32_idt_end - boot32_idt - 1
    750	.long   0
    751SYM_DATA_END(boot32_idt_desc)
    752	.balign 8
    753SYM_DATA_START(boot32_idt)
    754	.rept 32
    755	.quad 0
    756	.endr
    757SYM_DATA_END_LABEL(boot32_idt, SYM_L_GLOBAL, boot32_idt_end)
    758#endif
    759
    760#ifdef CONFIG_EFI_STUB
    761SYM_DATA(image_offset, .long 0)
    762#endif
    763#ifdef CONFIG_EFI_MIXED
    764SYM_DATA_LOCAL(efi32_boot_args, .long 0, 0, 0)
    765SYM_DATA(efi_is64, .byte 1)
    766
    767#define ST32_boottime		60 // offsetof(efi_system_table_32_t, boottime)
    768#define BS32_handle_protocol	88 // offsetof(efi_boot_services_32_t, handle_protocol)
    769#define LI32_image_base		32 // offsetof(efi_loaded_image_32_t, image_base)
    770
    771	__HEAD
    772	.code32
    773SYM_FUNC_START(efi32_pe_entry)
    774/*
    775 * efi_status_t efi32_pe_entry(efi_handle_t image_handle,
    776 *			       efi_system_table_32_t *sys_table)
    777 */
    778
    779	pushl	%ebp
    780	movl	%esp, %ebp
    781	pushl	%eax				// dummy push to allocate loaded_image
    782
    783	pushl	%ebx				// save callee-save registers
    784	pushl	%edi
    785
    786	call	verify_cpu			// check for long mode support
    787	testl	%eax, %eax
    788	movl	$0x80000003, %eax		// EFI_UNSUPPORTED
    789	jnz	2f
    790
    791	call	1f
    7921:	pop	%ebx
    793	subl	$ rva(1b), %ebx
    794
    795	/* Get the loaded image protocol pointer from the image handle */
    796	leal	-4(%ebp), %eax
    797	pushl	%eax				// &loaded_image
    798	leal	rva(loaded_image_proto)(%ebx), %eax
    799	pushl	%eax				// pass the GUID address
    800	pushl	8(%ebp)				// pass the image handle
    801
    802	/*
    803	 * Note the alignment of the stack frame.
    804	 *   sys_table
    805	 *   handle             <-- 16-byte aligned on entry by ABI
    806	 *   return address
    807	 *   frame pointer
    808	 *   loaded_image       <-- local variable
    809	 *   saved %ebx		<-- 16-byte aligned here
    810	 *   saved %edi
    811	 *   &loaded_image
    812	 *   &loaded_image_proto
    813	 *   handle             <-- 16-byte aligned for call to handle_protocol
    814	 */
    815
    816	movl	12(%ebp), %eax			// sys_table
    817	movl	ST32_boottime(%eax), %eax	// sys_table->boottime
    818	call	*BS32_handle_protocol(%eax)	// sys_table->boottime->handle_protocol
    819	addl	$12, %esp			// restore argument space
    820	testl	%eax, %eax
    821	jnz	2f
    822
    823	movl	8(%ebp), %ecx			// image_handle
    824	movl	12(%ebp), %edx			// sys_table
    825	movl	-4(%ebp), %esi			// loaded_image
    826	movl	LI32_image_base(%esi), %esi	// loaded_image->image_base
    827	movl	%ebx, %ebp			// startup_32 for efi32_pe_stub_entry
    828	/*
    829	 * We need to set the image_offset variable here since startup_32() will
    830	 * use it before we get to the 64-bit efi_pe_entry() in C code.
    831	 */
    832	subl	%esi, %ebx
    833	movl	%ebx, rva(image_offset)(%ebp)	// save image_offset
    834	jmp	efi32_pe_stub_entry
    835
    8362:	popl	%edi				// restore callee-save registers
    837	popl	%ebx
    838	leave
    839	RET
    840SYM_FUNC_END(efi32_pe_entry)
    841
    842	.section ".rodata"
    843	/* EFI loaded image protocol GUID */
    844	.balign 4
    845SYM_DATA_START_LOCAL(loaded_image_proto)
    846	.long	0x5b1b31a1
    847	.word	0x9562, 0x11d2
    848	.byte	0x8e, 0x3f, 0x00, 0xa0, 0xc9, 0x69, 0x72, 0x3b
    849SYM_DATA_END(loaded_image_proto)
    850#endif
    851
    852#ifdef CONFIG_AMD_MEM_ENCRYPT
    853	__HEAD
    854	.code32
    855/*
    856 * Write an IDT entry into boot32_idt
    857 *
    858 * Parameters:
    859 *
    860 * %eax:	Handler address
    861 * %edx:	Vector number
    862 *
    863 * Physical offset is expected in %ebp
    864 */
    865SYM_FUNC_START(startup32_set_idt_entry)
    866	push    %ebx
    867	push    %ecx
    868
    869	/* IDT entry address to %ebx */
    870	leal    rva(boot32_idt)(%ebp), %ebx
    871	shl	$3, %edx
    872	addl    %edx, %ebx
    873
    874	/* Build IDT entry, lower 4 bytes */
    875	movl    %eax, %edx
    876	andl    $0x0000ffff, %edx	# Target code segment offset [15:0]
    877	movl    $__KERNEL32_CS, %ecx	# Target code segment selector
    878	shl     $16, %ecx
    879	orl     %ecx, %edx
    880
    881	/* Store lower 4 bytes to IDT */
    882	movl    %edx, (%ebx)
    883
    884	/* Build IDT entry, upper 4 bytes */
    885	movl    %eax, %edx
    886	andl    $0xffff0000, %edx	# Target code segment offset [31:16]
    887	orl     $0x00008e00, %edx	# Present, Type 32-bit Interrupt Gate
    888
    889	/* Store upper 4 bytes to IDT */
    890	movl    %edx, 4(%ebx)
    891
    892	pop     %ecx
    893	pop     %ebx
    894	RET
    895SYM_FUNC_END(startup32_set_idt_entry)
    896#endif
    897
    898SYM_FUNC_START(startup32_load_idt)
    899#ifdef CONFIG_AMD_MEM_ENCRYPT
    900	/* #VC handler */
    901	leal    rva(startup32_vc_handler)(%ebp), %eax
    902	movl    $X86_TRAP_VC, %edx
    903	call    startup32_set_idt_entry
    904
    905	/* Load IDT */
    906	leal	rva(boot32_idt)(%ebp), %eax
    907	movl	%eax, rva(boot32_idt_desc+2)(%ebp)
    908	lidt    rva(boot32_idt_desc)(%ebp)
    909#endif
    910	RET
    911SYM_FUNC_END(startup32_load_idt)
    912
    913/*
    914 * Check for the correct C-bit position when the startup_32 boot-path is used.
    915 *
    916 * The check makes use of the fact that all memory is encrypted when paging is
    917 * disabled. The function creates 64 bits of random data using the RDRAND
    918 * instruction. RDRAND is mandatory for SEV guests, so always available. If the
    919 * hypervisor violates that the kernel will crash right here.
    920 *
    921 * The 64 bits of random data are stored to a memory location and at the same
    922 * time kept in the %eax and %ebx registers. Since encryption is always active
    923 * when paging is off the random data will be stored encrypted in main memory.
    924 *
    925 * Then paging is enabled. When the C-bit position is correct all memory is
    926 * still mapped encrypted and comparing the register values with memory will
    927 * succeed. An incorrect C-bit position will map all memory unencrypted, so that
    928 * the compare will use the encrypted random data and fail.
    929 */
    930SYM_FUNC_START(startup32_check_sev_cbit)
    931#ifdef CONFIG_AMD_MEM_ENCRYPT
    932	pushl	%eax
    933	pushl	%ebx
    934	pushl	%ecx
    935	pushl	%edx
    936
    937	/* Check for non-zero sev_status */
    938	movl	rva(sev_status)(%ebp), %eax
    939	testl	%eax, %eax
    940	jz	4f
    941
    942	/*
    943	 * Get two 32-bit random values - Don't bail out if RDRAND fails
    944	 * because it is better to prevent forward progress if no random value
    945	 * can be gathered.
    946	 */
    9471:	rdrand	%eax
    948	jnc	1b
    9492:	rdrand	%ebx
    950	jnc	2b
    951
    952	/* Store to memory and keep it in the registers */
    953	movl	%eax, rva(sev_check_data)(%ebp)
    954	movl	%ebx, rva(sev_check_data+4)(%ebp)
    955
    956	/* Enable paging to see if encryption is active */
    957	movl	%cr0, %edx			 /* Backup %cr0 in %edx */
    958	movl	$(X86_CR0_PG | X86_CR0_PE), %ecx /* Enable Paging and Protected mode */
    959	movl	%ecx, %cr0
    960
    961	cmpl	%eax, rva(sev_check_data)(%ebp)
    962	jne	3f
    963	cmpl	%ebx, rva(sev_check_data+4)(%ebp)
    964	jne	3f
    965
    966	movl	%edx, %cr0	/* Restore previous %cr0 */
    967
    968	jmp	4f
    969
    9703:	/* Check failed - hlt the machine */
    971	hlt
    972	jmp	3b
    973
    9744:
    975	popl	%edx
    976	popl	%ecx
    977	popl	%ebx
    978	popl	%eax
    979#endif
    980	RET
    981SYM_FUNC_END(startup32_check_sev_cbit)
    982
    983/*
    984 * Stack and heap for uncompression
    985 */
    986	.bss
    987	.balign 4
    988SYM_DATA_LOCAL(boot_heap,	.fill BOOT_HEAP_SIZE, 1, 0)
    989
    990SYM_DATA_START_LOCAL(boot_stack)
    991	.fill BOOT_STACK_SIZE, 1, 0
    992	.balign 16
    993SYM_DATA_END_LABEL(boot_stack, SYM_L_LOCAL, boot_stack_end)
    994
    995/*
    996 * Space for page tables (not in .bss so not zeroed)
    997 */
    998	.section ".pgtable","aw",@nobits
    999	.balign 4096
   1000SYM_DATA_LOCAL(pgtable,		.fill BOOT_PGT_SIZE, 1, 0)
   1001
   1002/*
   1003 * The page table is going to be used instead of page table in the trampoline
   1004 * memory.
   1005 */
   1006SYM_DATA_LOCAL(top_pgtable,	.fill PAGE_SIZE, 1, 0)