| 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
| 2 | |
| 3 | #include <linux/linkage.h> |
| 4 | #include <asm/segment.h> |
| 5 | #include <asm/boot.h> |
| 6 | #include <asm/msr.h> |
| 7 | #include <asm/processor-flags.h> |
| 8 | |
| 9 | /* |
| 10 | * This is the 32-bit trampoline that will be copied over to low memory. It |
| 11 | * will be called using the ordinary 64-bit calling convention from code |
| 12 | * running in 64-bit mode. |
| 13 | * |
| 14 | * Return address is at the top of the stack (might be above 4G). |
| 15 | * The first argument (EDI) contains the address of the temporary PGD level |
| 16 | * page table in 32-bit addressable memory which will be programmed into |
| 17 | * register CR3. |
| 18 | */ |
| 19 | |
| 20 | .section ".rodata" , "a" , @progbits |
| 21 | SYM_CODE_START(trampoline_32bit_src) |
| 22 | /* |
| 23 | * Preserve callee save 64-bit registers on the stack: this is |
| 24 | * necessary because the architecture does not guarantee that GPRs will |
| 25 | * retain their full 64-bit values across a 32-bit mode switch. |
| 26 | */ |
| 27 | pushq %r15 |
| 28 | pushq %r14 |
| 29 | pushq %r13 |
| 30 | pushq %r12 |
| 31 | pushq %rbp |
| 32 | pushq %rbx |
| 33 | |
| 34 | /* Preserve top half of RSP in a legacy mode GPR to avoid truncation */ |
| 35 | movq %rsp, %rbx |
| 36 | shrq $32, %rbx |
| 37 | |
| 38 | /* Switch to compatibility mode (CS.L = 0 CS.D = 1) via far return */ |
| 39 | pushq $__KERNEL32_CS |
| 40 | leaq 0f(%rip), %rax |
| 41 | pushq %rax |
| 42 | lretq |
| 43 | |
| 44 | /* |
| 45 | * The 32-bit code below will do a far jump back to long mode and end |
| 46 | * up here after reconfiguring the number of paging levels. First, the |
| 47 | * stack pointer needs to be restored to its full 64-bit value before |
| 48 | * the callee save register contents can be popped from the stack. |
| 49 | */ |
| 50 | .Lret: |
| 51 | shlq $32, %rbx |
| 52 | orq %rbx, %rsp |
| 53 | |
| 54 | /* Restore the preserved 64-bit registers */ |
| 55 | popq %rbx |
| 56 | popq %rbp |
| 57 | popq %r12 |
| 58 | popq %r13 |
| 59 | popq %r14 |
| 60 | popq %r15 |
| 61 | retq |
| 62 | |
| 63 | .code32 |
| 64 | 0: |
| 65 | /* Disable paging */ |
| 66 | movl %cr0, %eax |
| 67 | btrl $X86_CR0_PG_BIT, %eax |
| 68 | movl %eax, %cr0 |
| 69 | |
| 70 | /* Point CR3 to the trampoline's new top level page table */ |
| 71 | movl %edi, %cr3 |
| 72 | |
| 73 | /* Set EFER.LME=1 as a precaution in case hypervsior pulls the rug */ |
| 74 | movl $MSR_EFER, %ecx |
| 75 | rdmsr |
| 76 | btsl $_EFER_LME, %eax |
| 77 | /* Avoid writing EFER if no change was made (for TDX guest) */ |
| 78 | jc 1f |
| 79 | wrmsr |
| 80 | 1: |
| 81 | /* Toggle CR4.LA57 */ |
| 82 | movl %cr4, %eax |
| 83 | btcl $X86_CR4_LA57_BIT, %eax |
| 84 | movl %eax, %cr4 |
| 85 | |
| 86 | /* Enable paging again. */ |
| 87 | movl %cr0, %eax |
| 88 | btsl $X86_CR0_PG_BIT, %eax |
| 89 | movl %eax, %cr0 |
| 90 | |
| 91 | /* |
| 92 | * Return to the 64-bit calling code using LJMP rather than LRET, to |
| 93 | * avoid the need for a 32-bit addressable stack. The destination |
| 94 | * address will be adjusted after the template code is copied into a |
| 95 | * 32-bit addressable buffer. |
| 96 | */ |
| 97 | .Ljmp: ljmpl $__KERNEL_CS, $(.Lret - trampoline_32bit_src) |
| 98 | SYM_CODE_END(trampoline_32bit_src) |
| 99 | |
| 100 | /* |
| 101 | * This symbol is placed right after trampoline_32bit_src() so its address can |
| 102 | * be used to infer the size of the trampoline code. |
| 103 | */ |
| 104 | SYM_DATA(trampoline_ljmp_imm_offset, .word .Ljmp + 1 - trampoline_32bit_src) |
| 105 | |
| 106 | /* |
| 107 | * The trampoline code has a size limit. |
| 108 | * Make sure we fail to compile if the trampoline code grows |
| 109 | * beyond TRAMPOLINE_32BIT_CODE_SIZE bytes. |
| 110 | */ |
| 111 | .org trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_SIZE |
| 112 | |