| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | #include "misc.h" |
| 3 | #include <asm/bootparam.h> |
| 4 | #include <asm/bootparam_utils.h> |
| 5 | #include <asm/e820/types.h> |
| 6 | #include <asm/pgtable.h> |
| 7 | #include <asm/processor.h> |
| 8 | #include "../string.h" |
| 9 | #include "efi.h" |
| 10 | |
| 11 | #define BIOS_START_MIN 0x20000U /* 128K, less than this is insane */ |
| 12 | #define BIOS_START_MAX 0x9f000U /* 640K, absolute maximum */ |
| 13 | |
| 14 | /* __pgtable_l5_enabled needs to be in .data to avoid being cleared along with .bss */ |
| 15 | unsigned int __section(".data" ) __pgtable_l5_enabled; |
| 16 | unsigned int __section(".data" ) pgdir_shift = 39; |
| 17 | unsigned int __section(".data" ) ptrs_per_p4d = 1; |
| 18 | |
| 19 | /* Buffer to preserve trampoline memory */ |
| 20 | static char trampoline_save[TRAMPOLINE_32BIT_SIZE]; |
| 21 | |
| 22 | /* |
| 23 | * Trampoline address will be printed by extract_kernel() for debugging |
| 24 | * purposes. |
| 25 | * |
| 26 | * Avoid putting the pointer into .bss as it will be cleared between |
| 27 | * configure_5level_paging() and extract_kernel(). |
| 28 | */ |
| 29 | unsigned long *trampoline_32bit __section(".data" ); |
| 30 | |
| 31 | int cmdline_find_option_bool(const char *option); |
| 32 | |
| 33 | static unsigned long find_trampoline_placement(void) |
| 34 | { |
| 35 | unsigned long bios_start = 0, ebda_start = 0; |
| 36 | struct boot_e820_entry *entry; |
| 37 | char *signature; |
| 38 | int i; |
| 39 | |
| 40 | /* |
| 41 | * Find a suitable spot for the trampoline. |
| 42 | * This code is based on reserve_bios_regions(). |
| 43 | */ |
| 44 | |
| 45 | /* |
| 46 | * EFI systems may not provide legacy ROM. The memory may not be mapped |
| 47 | * at all. |
| 48 | * |
| 49 | * Only look for values in the legacy ROM for non-EFI system. |
| 50 | */ |
| 51 | signature = (char *)&boot_params_ptr->efi_info.efi_loader_signature; |
| 52 | if (strncmp(cs: signature, EFI32_LOADER_SIGNATURE, count: 4) && |
| 53 | strncmp(cs: signature, EFI64_LOADER_SIGNATURE, count: 4)) { |
| 54 | ebda_start = *(unsigned short *)0x40e << 4; |
| 55 | bios_start = *(unsigned short *)0x413 << 10; |
| 56 | } |
| 57 | |
| 58 | if (bios_start < BIOS_START_MIN || bios_start > BIOS_START_MAX) |
| 59 | bios_start = BIOS_START_MAX; |
| 60 | |
| 61 | if (ebda_start > BIOS_START_MIN && ebda_start < bios_start) |
| 62 | bios_start = ebda_start; |
| 63 | |
| 64 | bios_start = round_down(bios_start, PAGE_SIZE); |
| 65 | |
| 66 | /* Find the first usable memory region under bios_start. */ |
| 67 | for (i = boot_params_ptr->e820_entries - 1; i >= 0; i--) { |
| 68 | unsigned long new = bios_start; |
| 69 | |
| 70 | entry = &boot_params_ptr->e820_table[i]; |
| 71 | |
| 72 | /* Skip all entries above bios_start. */ |
| 73 | if (bios_start <= entry->addr) |
| 74 | continue; |
| 75 | |
| 76 | /* Skip non-RAM entries. */ |
| 77 | if (entry->type != E820_TYPE_RAM) |
| 78 | continue; |
| 79 | |
| 80 | /* Adjust bios_start to the end of the entry if needed. */ |
| 81 | if (bios_start > entry->addr + entry->size) |
| 82 | new = entry->addr + entry->size; |
| 83 | |
| 84 | /* Keep bios_start page-aligned. */ |
| 85 | new = round_down(new, PAGE_SIZE); |
| 86 | |
| 87 | /* Skip the entry if it's too small. */ |
| 88 | if (new - TRAMPOLINE_32BIT_SIZE < entry->addr) |
| 89 | continue; |
| 90 | |
| 91 | /* Protect against underflow. */ |
| 92 | if (new - TRAMPOLINE_32BIT_SIZE > bios_start) |
| 93 | break; |
| 94 | |
| 95 | bios_start = new; |
| 96 | break; |
| 97 | } |
| 98 | |
| 99 | /* Place the trampoline just below the end of low memory */ |
| 100 | return bios_start - TRAMPOLINE_32BIT_SIZE; |
| 101 | } |
| 102 | |
| 103 | asmlinkage void configure_5level_paging(struct boot_params *bp, void *pgtable) |
| 104 | { |
| 105 | void (*toggle_la57)(void *cr3); |
| 106 | bool l5_required = false; |
| 107 | |
| 108 | /* Initialize boot_params. Required for cmdline_find_option_bool(). */ |
| 109 | sanitize_boot_params(boot_params: bp); |
| 110 | boot_params_ptr = bp; |
| 111 | |
| 112 | /* |
| 113 | * Check if LA57 is desired and supported. |
| 114 | * |
| 115 | * There are several parts to the check: |
| 116 | * - if user asked to disable 5-level paging: no5lvl in cmdline |
| 117 | * - if the machine supports 5-level paging: |
| 118 | * + CPUID leaf 7 is supported |
| 119 | * + the leaf has the feature bit set |
| 120 | */ |
| 121 | if (!cmdline_find_option_bool(option: "no5lvl" ) && |
| 122 | native_cpuid_eax(op: 0) >= 7 && (native_cpuid_ecx(op: 7) & BIT(16))) { |
| 123 | l5_required = true; |
| 124 | |
| 125 | /* Initialize variables for 5-level paging */ |
| 126 | __pgtable_l5_enabled = 1; |
| 127 | pgdir_shift = 48; |
| 128 | ptrs_per_p4d = 512; |
| 129 | } |
| 130 | |
| 131 | /* |
| 132 | * The trampoline will not be used if the paging mode is already set to |
| 133 | * the desired one. |
| 134 | */ |
| 135 | if (l5_required == !!(native_read_cr4() & X86_CR4_LA57)) |
| 136 | return; |
| 137 | |
| 138 | trampoline_32bit = (unsigned long *)find_trampoline_placement(); |
| 139 | |
| 140 | /* Preserve trampoline memory */ |
| 141 | memcpy(trampoline_save, trampoline_32bit, TRAMPOLINE_32BIT_SIZE); |
| 142 | |
| 143 | /* Clear trampoline memory first */ |
| 144 | memset(trampoline_32bit, 0, TRAMPOLINE_32BIT_SIZE); |
| 145 | |
| 146 | /* Copy trampoline code in place */ |
| 147 | toggle_la57 = memcpy(trampoline_32bit + |
| 148 | TRAMPOLINE_32BIT_CODE_OFFSET / sizeof(unsigned long), |
| 149 | &trampoline_32bit_src, TRAMPOLINE_32BIT_CODE_SIZE); |
| 150 | |
| 151 | /* |
| 152 | * Avoid the need for a stack in the 32-bit trampoline code, by using |
| 153 | * LJMP rather than LRET to return back to long mode. LJMP takes an |
| 154 | * immediate absolute address, which needs to be adjusted based on the |
| 155 | * placement of the trampoline. |
| 156 | */ |
| 157 | *(u32 *)((u8 *)toggle_la57 + trampoline_ljmp_imm_offset) += |
| 158 | (unsigned long)toggle_la57; |
| 159 | |
| 160 | /* |
| 161 | * The code below prepares page table in trampoline memory. |
| 162 | * |
| 163 | * The new page table will be used by trampoline code for switching |
| 164 | * from 4- to 5-level paging or vice versa. |
| 165 | */ |
| 166 | |
| 167 | if (l5_required) { |
| 168 | /* |
| 169 | * For 4- to 5-level paging transition, set up current CR3 as |
| 170 | * the first and the only entry in a new top-level page table. |
| 171 | */ |
| 172 | *trampoline_32bit = native_read_cr3_pa() | _PAGE_TABLE_NOENC; |
| 173 | } else { |
| 174 | u64 *new_cr3; |
| 175 | pgd_t *pgdp; |
| 176 | |
| 177 | /* |
| 178 | * For 5- to 4-level paging transition, copy page table pointed |
| 179 | * by first entry in the current top-level page table as our |
| 180 | * new top-level page table. |
| 181 | * |
| 182 | * We cannot just point to the page table from trampoline as it |
| 183 | * may be above 4G. |
| 184 | */ |
| 185 | pgdp = (pgd_t *)native_read_cr3_pa(); |
| 186 | new_cr3 = (u64 *)(native_pgd_val(pgd: pgdp[0]) & PTE_PFN_MASK); |
| 187 | memcpy(trampoline_32bit, new_cr3, PAGE_SIZE); |
| 188 | } |
| 189 | |
| 190 | toggle_la57(trampoline_32bit); |
| 191 | |
| 192 | /* |
| 193 | * Move the top level page table out of trampoline memory. |
| 194 | */ |
| 195 | memcpy(pgtable, trampoline_32bit, PAGE_SIZE); |
| 196 | native_write_cr3(val: (unsigned long)pgtable); |
| 197 | |
| 198 | /* Restore trampoline memory */ |
| 199 | memcpy(trampoline_32bit, trampoline_save, TRAMPOLINE_32BIT_SIZE); |
| 200 | } |
| 201 | |