linux/arch/x86/boot/compressed/head_64.S
Ard Biesheuvel a0025f587c x86/boot/64: Clear most of CR4 in startup_64(), except PAE, MCE and LA57
The early 64-bit boot code must be entered with a 1:1 mapping of the
bootable image, but it cannot operate without a 1:1 mapping of all the
assets in memory that it accesses, and therefore, it creates such
mappings for all known assets upfront, and additional ones on demand
when a page fault happens on a memory address.

These mappings are created with the global bit G set, as the flags used
to create page table descriptors are based on __PAGE_KERNEL_LARGE_EXEC
defined by the core kernel, even though the context where these mappings
are used is very different.

This means that the TLB maintenance carried out by the decompressor is
not sufficient if it is entered with CR4.PGE enabled, which has been
observed to happen with the stage0 bootloader of project Oak. While this
is a dubious practice if no global mappings are being used to begin
with, the decompressor is clearly at fault here for creating global
mappings and not performing the appropriate TLB maintenance.

Since commit:

  f97b67a773 ("x86/decompressor: Only call the trampoline when changing paging levels")

CR4 is no longer modified by the decompressor if no change in the number
of paging levels is needed. Before that, CR4 would always be set to a
consistent value with PGE cleared.

So let's reinstate a simplified version of the original logic to put CR4
into a known state, and preserve the PAE, MCE and LA57 bits, none of
which can be modified freely at this point (PAE and LA57 cannot be
changed while running in long mode, and MCE cannot be cleared when
running under some hypervisors).

This effectively clears PGE and works around the project Oak bug.

Fixes: f97b67a773 ("x86/decompressor: Only call the trampoline when ...")
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Link: https://lore.kernel.org/r/20240410151354.506098-2-ardb+git@google.com
2024-04-11 15:37:17 +02:00

650 lines
17 KiB
ArmAsm

/* SPDX-License-Identifier: GPL-2.0 */
/*
* linux/boot/head.S
*
* Copyright (C) 1991, 1992, 1993 Linus Torvalds
*/
/*
* head.S contains the 32-bit startup code.
*
* NOTE!!! Startup happens at absolute address 0x00001000, which is also where
* the page directory will exist. The startup code will be overwritten by
* the page directory. [According to comments etc elsewhere on a compressed
* kernel it will end up at 0x1000 + 1Mb I hope so as I assume this. - AC]
*
* Page 0 is deliberately kept safe, since System Management Mode code in
* laptops may need to access the BIOS data stored there. This is also
* useful for future device drivers that either access the BIOS via VM86
* mode.
*/
/*
* High loaded stuff by Hans Lermen & Werner Almesberger, Feb. 1996
*/
.code32
.text
#include <linux/init.h>
#include <linux/linkage.h>
#include <asm/segment.h>
#include <asm/boot.h>
#include <asm/msr.h>
#include <asm/processor-flags.h>
#include <asm/asm-offsets.h>
#include <asm/bootparam.h>
#include <asm/desc_defs.h>
#include <asm/trapnr.h>
#include "pgtable.h"
/*
* Fix alignment at 16 bytes. Following CONFIG_FUNCTION_ALIGNMENT will result
* in assembly errors due to trying to move .org backward due to the excessive
* alignment.
*/
#undef __ALIGN
#define __ALIGN .balign 16, 0x90
/*
* Locally defined symbols should be marked hidden:
*/
.hidden _bss
.hidden _ebss
.hidden _end
__HEAD
/*
* This macro gives the relative virtual address of X, i.e. the offset of X
* from startup_32. This is the same as the link-time virtual address of X,
* since startup_32 is at 0, but defining it this way tells the
* assembler/linker that we do not want the actual run-time address of X. This
* prevents the linker from trying to create unwanted run-time relocation
* entries for the reference when the compressed kernel is linked as PIE.
*
* A reference X(%reg) will result in the link-time VA of X being stored with
* the instruction, and a run-time R_X86_64_RELATIVE relocation entry that
* adds the 64-bit base address where the kernel is loaded.
*
* Replacing it with (X-startup_32)(%reg) results in the offset being stored,
* and no run-time relocation.
*
* The macro should be used as a displacement with a base register containing
* the run-time address of startup_32 [i.e. rva(X)(%reg)], or as an immediate
* [$ rva(X)].
*
* This macro can only be used from within the .head.text section, since the
* expression requires startup_32 to be in the same section as the code being
* assembled.
*/
#define rva(X) ((X) - startup_32)
.code32
SYM_FUNC_START(startup_32)
/*
* 32bit entry is 0 and it is ABI so immutable!
* If we come here directly from a bootloader,
* kernel(text+data+bss+brk) ramdisk, zero_page, command line
* all need to be under the 4G limit.
*/
cld
cli
/*
* Calculate the delta between where we were compiled to run
* at and where we were actually loaded at. This can only be done
* with a short local call on x86. Nothing else will tell us what
* address we are running at. The reserved chunk of the real-mode
* data at 0x1e4 (defined as a scratch field) are used as the stack
* for this calculation. Only 4 bytes are needed.
*/
leal (BP_scratch+4)(%esi), %esp
call 1f
1: popl %ebp
subl $ rva(1b), %ebp
/* Load new GDT with the 64bit segments using 32bit descriptor */
leal rva(gdt)(%ebp), %eax
movl %eax, 2(%eax)
lgdt (%eax)
/* Load segment registers with our descriptors */
movl $__BOOT_DS, %eax
movl %eax, %ds
movl %eax, %es
movl %eax, %fs
movl %eax, %gs
movl %eax, %ss
/* Setup a stack and load CS from current GDT */
leal rva(boot_stack_end)(%ebp), %esp
pushl $__KERNEL32_CS
leal rva(1f)(%ebp), %eax
pushl %eax
lretl
1:
/* Setup Exception handling for SEV-ES */
#ifdef CONFIG_AMD_MEM_ENCRYPT
call startup32_load_idt
#endif
/* Make sure cpu supports long mode. */
call verify_cpu
testl %eax, %eax
jnz .Lno_longmode
/*
* Compute the delta between where we were compiled to run at
* and where the code will actually run at.
*
* %ebp contains the address we are loaded at by the boot loader and %ebx
* contains the address where we should move the kernel image temporarily
* for safe in-place decompression.
*/
#ifdef CONFIG_RELOCATABLE
movl %ebp, %ebx
movl BP_kernel_alignment(%esi), %eax
decl %eax
addl %eax, %ebx
notl %eax
andl %eax, %ebx
cmpl $LOAD_PHYSICAL_ADDR, %ebx
jae 1f
#endif
movl $LOAD_PHYSICAL_ADDR, %ebx
1:
/* Target address to relocate to for decompression */
addl BP_init_size(%esi), %ebx
subl $ rva(_end), %ebx
/*
* Prepare for entering 64 bit mode
*/
/* Enable PAE mode */
movl %cr4, %eax
orl $X86_CR4_PAE, %eax
movl %eax, %cr4
/*
* Build early 4G boot pagetable
*/
/*
* If SEV is active then set the encryption mask in the page tables.
* This will ensure that when the kernel is copied and decompressed
* it will be done so encrypted.
*/
xorl %edx, %edx
#ifdef CONFIG_AMD_MEM_ENCRYPT
call get_sev_encryption_bit
xorl %edx, %edx
testl %eax, %eax
jz 1f
subl $32, %eax /* Encryption bit is always above bit 31 */
bts %eax, %edx /* Set encryption mask for page tables */
/*
* Set MSR_AMD64_SEV_ENABLED_BIT in sev_status so that
* startup32_check_sev_cbit() will do a check. sev_enable() will
* initialize sev_status with all the bits reported by
* MSR_AMD_SEV_STATUS later, but only MSR_AMD64_SEV_ENABLED_BIT
* needs to be set for now.
*/
movl $1, rva(sev_status)(%ebp)
1:
#endif
/* Initialize Page tables to 0 */
leal rva(pgtable)(%ebx), %edi
xorl %eax, %eax
movl $(BOOT_INIT_PGT_SIZE/4), %ecx
rep stosl
/* Build Level 4 */
leal rva(pgtable + 0)(%ebx), %edi
leal 0x1007 (%edi), %eax
movl %eax, 0(%edi)
addl %edx, 4(%edi)
/* Build Level 3 */
leal rva(pgtable + 0x1000)(%ebx), %edi
leal 0x1007(%edi), %eax
movl $4, %ecx
1: movl %eax, 0x00(%edi)
addl %edx, 0x04(%edi)
addl $0x00001000, %eax
addl $8, %edi
decl %ecx
jnz 1b
/* Build Level 2 */
leal rva(pgtable + 0x2000)(%ebx), %edi
movl $0x00000183, %eax
movl $2048, %ecx
1: movl %eax, 0(%edi)
addl %edx, 4(%edi)
addl $0x00200000, %eax
addl $8, %edi
decl %ecx
jnz 1b
/* Enable the boot page tables */
leal rva(pgtable)(%ebx), %eax
movl %eax, %cr3
/* Enable Long mode in EFER (Extended Feature Enable Register) */
movl $MSR_EFER, %ecx
rdmsr
btsl $_EFER_LME, %eax
wrmsr
/* After gdt is loaded */
xorl %eax, %eax
lldt %ax
movl $__BOOT_TSS, %eax
ltr %ax
#ifdef CONFIG_AMD_MEM_ENCRYPT
/* Check if the C-bit position is correct when SEV is active */
call startup32_check_sev_cbit
#endif
/*
* Setup for the jump to 64bit mode
*
* When the jump is performed we will be in long mode but
* in 32bit compatibility mode with EFER.LME = 1, CS.L = 0, CS.D = 1
* (and in turn EFER.LMA = 1). To jump into 64bit mode we use
* the new gdt/idt that has __KERNEL_CS with CS.L = 1.
* We place all of the values on our mini stack so lret can
* used to perform that far jump.
*/
leal rva(startup_64)(%ebp), %eax
#ifdef CONFIG_EFI_MIXED
cmpb $1, rva(efi_is64)(%ebp)
je 1f
leal rva(startup_64_mixed_mode)(%ebp), %eax
1:
#endif
pushl $__KERNEL_CS
pushl %eax
/* Enter paged protected Mode, activating Long Mode */
movl $CR0_STATE, %eax
movl %eax, %cr0
/* Jump from 32bit compatibility mode into 64bit mode. */
lret
SYM_FUNC_END(startup_32)
.code64
.org 0x200
SYM_CODE_START(startup_64)
/*
* 64bit entry is 0x200 and it is ABI so immutable!
* We come here either from startup_32 or directly from a
* 64bit bootloader.
* If we come here from a bootloader, kernel(text+data+bss+brk),
* ramdisk, zero_page, command line could be above 4G.
* We depend on an identity mapped page table being provided
* that maps our entire kernel(text+data+bss+brk), zero page
* and command line.
*/
cld
cli
/* Setup data segments. */
xorl %eax, %eax
movl %eax, %ds
movl %eax, %es
movl %eax, %ss
movl %eax, %fs
movl %eax, %gs
/*
* Compute the decompressed kernel start address. It is where
* we were loaded at aligned to a 2M boundary. %rbp contains the
* decompressed kernel start address.
*
* If it is a relocatable kernel then decompress and run the kernel
* from load address aligned to 2MB addr, otherwise decompress and
* run the kernel from LOAD_PHYSICAL_ADDR
*
* We cannot rely on the calculation done in 32-bit mode, since we
* may have been invoked via the 64-bit entry point.
*/
/* Start with the delta to where the kernel will run at. */
#ifdef CONFIG_RELOCATABLE
leaq startup_32(%rip) /* - $startup_32 */, %rbp
movl BP_kernel_alignment(%rsi), %eax
decl %eax
addq %rax, %rbp
notq %rax
andq %rax, %rbp
cmpq $LOAD_PHYSICAL_ADDR, %rbp
jae 1f
#endif
movq $LOAD_PHYSICAL_ADDR, %rbp
1:
/* Target address to relocate to for decompression */
movl BP_init_size(%rsi), %ebx
subl $ rva(_end), %ebx
addq %rbp, %rbx
/* Set up the stack */
leaq rva(boot_stack_end)(%rbx), %rsp
/*
* At this point we are in long mode with 4-level paging enabled,
* but we might want to enable 5-level paging or vice versa.
*
* The problem is that we cannot do it directly. Setting or clearing
* CR4.LA57 in long mode would trigger #GP. So we need to switch off
* long mode and paging first.
*
* We also need a trampoline in lower memory to switch over from
* 4- to 5-level paging for cases when the bootloader puts the kernel
* above 4G, but didn't enable 5-level paging for us.
*
* The same trampoline can be used to switch from 5- to 4-level paging
* mode, like when starting 4-level paging kernel via kexec() when
* original kernel worked in 5-level paging mode.
*
* For the trampoline, we need the top page table to reside in lower
* memory as we don't have a way to load 64-bit values into CR3 in
* 32-bit mode.
*/
/* Make sure we have GDT with 32-bit code segment */
leaq gdt64(%rip), %rax
addq %rax, 2(%rax)
lgdt (%rax)
/* Reload CS so IRET returns to a CS actually in the GDT */
pushq $__KERNEL_CS
leaq .Lon_kernel_cs(%rip), %rax
pushq %rax
lretq
.Lon_kernel_cs:
/*
* RSI holds a pointer to a boot_params structure provided by the
* loader, and this needs to be preserved across C function calls. So
* move it into a callee saved register.
*/
movq %rsi, %r15
call load_stage1_idt
#ifdef CONFIG_AMD_MEM_ENCRYPT
/*
* Now that the stage1 interrupt handlers are set up, #VC exceptions from
* CPUID instructions can be properly handled for SEV-ES guests.
*
* For SEV-SNP, the CPUID table also needs to be set up in advance of any
* CPUID instructions being issued, so go ahead and do that now via
* sev_enable(), which will also handle the rest of the SEV-related
* detection/setup to ensure that has been done in advance of any dependent
* code. Pass the boot_params pointer as the first argument.
*/
movq %r15, %rdi
call sev_enable
#endif
/* Preserve only the CR4 bits that must be preserved, and clear the rest */
movq %cr4, %rax
andl $(X86_CR4_PAE | X86_CR4_MCE | X86_CR4_LA57), %eax
movq %rax, %cr4
/*
* configure_5level_paging() updates the number of paging levels using
* a trampoline in 32-bit addressable memory if the current number does
* not match the desired number.
*
* Pass the boot_params pointer as the first argument. The second
* argument is the relocated address of the page table to use instead
* of the page table in trampoline memory (if required).
*/
movq %r15, %rdi
leaq rva(top_pgtable)(%rbx), %rsi
call configure_5level_paging
/* Zero EFLAGS */
pushq $0
popfq
/*
* Copy the compressed kernel to the end of our buffer
* where decompression in place becomes safe.
*/
leaq (_bss-8)(%rip), %rsi
leaq rva(_bss-8)(%rbx), %rdi
movl $(_bss - startup_32), %ecx
shrl $3, %ecx
std
rep movsq
cld
/*
* The GDT may get overwritten either during the copy we just did or
* during extract_kernel below. To avoid any issues, repoint the GDTR
* to the new copy of the GDT.
*/
leaq rva(gdt64)(%rbx), %rax
leaq rva(gdt)(%rbx), %rdx
movq %rdx, 2(%rax)
lgdt (%rax)
/*
* Jump to the relocated address.
*/
leaq rva(.Lrelocated)(%rbx), %rax
jmp *%rax
SYM_CODE_END(startup_64)
.text
SYM_FUNC_START_LOCAL_NOALIGN(.Lrelocated)
/*
* Clear BSS (stack is currently empty)
*/
xorl %eax, %eax
leaq _bss(%rip), %rdi
leaq _ebss(%rip), %rcx
subq %rdi, %rcx
shrq $3, %rcx
rep stosq
call load_stage2_idt
/* Pass boot_params to initialize_identity_maps() */
movq %r15, %rdi
call initialize_identity_maps
/*
* Do the extraction, and jump to the new kernel..
*/
/* pass struct boot_params pointer and output target address */
movq %r15, %rdi
movq %rbp, %rsi
call extract_kernel /* returns kernel entry point in %rax */
/*
* Jump to the decompressed kernel.
*/
movq %r15, %rsi
jmp *%rax
SYM_FUNC_END(.Lrelocated)
/*
* This is the 32-bit trampoline that will be copied over to low memory. It
* will be called using the ordinary 64-bit calling convention from code
* running in 64-bit mode.
*
* Return address is at the top of the stack (might be above 4G).
* The first argument (EDI) contains the address of the temporary PGD level
* page table in 32-bit addressable memory which will be programmed into
* register CR3.
*/
.section ".rodata", "a", @progbits
SYM_CODE_START(trampoline_32bit_src)
/*
* Preserve callee save 64-bit registers on the stack: this is
* necessary because the architecture does not guarantee that GPRs will
* retain their full 64-bit values across a 32-bit mode switch.
*/
pushq %r15
pushq %r14
pushq %r13
pushq %r12
pushq %rbp
pushq %rbx
/* Preserve top half of RSP in a legacy mode GPR to avoid truncation */
movq %rsp, %rbx
shrq $32, %rbx
/* Switch to compatibility mode (CS.L = 0 CS.D = 1) via far return */
pushq $__KERNEL32_CS
leaq 0f(%rip), %rax
pushq %rax
lretq
/*
* The 32-bit code below will do a far jump back to long mode and end
* up here after reconfiguring the number of paging levels. First, the
* stack pointer needs to be restored to its full 64-bit value before
* the callee save register contents can be popped from the stack.
*/
.Lret:
shlq $32, %rbx
orq %rbx, %rsp
/* Restore the preserved 64-bit registers */
popq %rbx
popq %rbp
popq %r12
popq %r13
popq %r14
popq %r15
retq
.code32
0:
/* Disable paging */
movl %cr0, %eax
btrl $X86_CR0_PG_BIT, %eax
movl %eax, %cr0
/* Point CR3 to the trampoline's new top level page table */
movl %edi, %cr3
/* Set EFER.LME=1 as a precaution in case hypervsior pulls the rug */
movl $MSR_EFER, %ecx
rdmsr
btsl $_EFER_LME, %eax
/* Avoid writing EFER if no change was made (for TDX guest) */
jc 1f
wrmsr
1:
/* Toggle CR4.LA57 */
movl %cr4, %eax
btcl $X86_CR4_LA57_BIT, %eax
movl %eax, %cr4
/* Enable paging again. */
movl %cr0, %eax
btsl $X86_CR0_PG_BIT, %eax
movl %eax, %cr0
/*
* Return to the 64-bit calling code using LJMP rather than LRET, to
* avoid the need for a 32-bit addressable stack. The destination
* address will be adjusted after the template code is copied into a
* 32-bit addressable buffer.
*/
.Ljmp: ljmpl $__KERNEL_CS, $(.Lret - trampoline_32bit_src)
SYM_CODE_END(trampoline_32bit_src)
/*
* This symbol is placed right after trampoline_32bit_src() so its address can
* be used to infer the size of the trampoline code.
*/
SYM_DATA(trampoline_ljmp_imm_offset, .word .Ljmp + 1 - trampoline_32bit_src)
/*
* The trampoline code has a size limit.
* Make sure we fail to compile if the trampoline code grows
* beyond TRAMPOLINE_32BIT_CODE_SIZE bytes.
*/
.org trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_SIZE
.text
SYM_FUNC_START_LOCAL_NOALIGN(.Lno_longmode)
/* This isn't an x86-64 CPU, so hang intentionally, we cannot continue */
1:
hlt
jmp 1b
SYM_FUNC_END(.Lno_longmode)
.globl verify_cpu
#include "../../kernel/verify_cpu.S"
.data
SYM_DATA_START_LOCAL(gdt64)
.word gdt_end - gdt - 1
.quad gdt - gdt64
SYM_DATA_END(gdt64)
.balign 8
SYM_DATA_START_LOCAL(gdt)
.word gdt_end - gdt - 1
.long 0
.word 0
.quad 0x00cf9a000000ffff /* __KERNEL32_CS */
.quad 0x00af9a000000ffff /* __KERNEL_CS */
.quad 0x00cf92000000ffff /* __KERNEL_DS */
.quad 0x0080890000000000 /* TS descriptor */
.quad 0x0000000000000000 /* TS continued */
SYM_DATA_END_LABEL(gdt, SYM_L_LOCAL, gdt_end)
SYM_DATA_START(boot_idt_desc)
.word boot_idt_end - boot_idt - 1
.quad 0
SYM_DATA_END(boot_idt_desc)
.balign 8
SYM_DATA_START(boot_idt)
.rept BOOT_IDT_ENTRIES
.quad 0
.quad 0
.endr
SYM_DATA_END_LABEL(boot_idt, SYM_L_GLOBAL, boot_idt_end)
/*
* Stack and heap for uncompression
*/
.bss
.balign 4
SYM_DATA_START_LOCAL(boot_stack)
.fill BOOT_STACK_SIZE, 1, 0
.balign 16
SYM_DATA_END_LABEL(boot_stack, SYM_L_LOCAL, boot_stack_end)
/*
* Space for page tables (not in .bss so not zeroed)
*/
.section ".pgtable","aw",@nobits
.balign 4096
SYM_DATA_LOCAL(pgtable, .fill BOOT_PGT_SIZE, 1, 0)
/*
* The page table is going to be used instead of page table in the trampoline
* memory.
*/
SYM_DATA_LOCAL(top_pgtable, .fill PAGE_SIZE, 1, 0)