David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 1 | // SPDX-License-Identifier: GPL-2.0-only |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2 | /* |
| 3 | * EFI stub implementation that is shared by arm and arm64 architectures. |
| 4 | * This should be #included by the EFI stub implementation files. |
| 5 | * |
| 6 | * Copyright (C) 2013,2014 Linaro Limited |
| 7 | * Roy Franz <roy.franz@linaro.org |
| 8 | * Copyright (C) 2013 Red Hat, Inc. |
| 9 | * Mark Salter <msalter@redhat.com> |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 10 | */ |
| 11 | |
| 12 | #include <linux/efi.h> |
| 13 | #include <linux/sort.h> |
| 14 | #include <asm/efi.h> |
| 15 | |
| 16 | #include "efistub.h" |
| 17 | |
| 18 | /* |
| 19 | * This is the base address at which to start allocating virtual memory ranges |
| 20 | * for UEFI Runtime Services. This is in the low TTBR0 range so that we can use |
| 21 | * any allocation we choose, and eliminate the risk of a conflict after kexec. |
| 22 | * The value chosen is the largest non-zero power of 2 suitable for this purpose |
| 23 | * both on 32-bit and 64-bit ARM CPUs, to maximize the likelihood that it can |
| 24 | * be mapped efficiently. |
| 25 | * Since 32-bit ARM could potentially execute with a 1G/3G user/kernel split, |
| 26 | * map everything below 1 GB. (512 MB is a reasonable upper bound for the |
| 27 | * entire footprint of the UEFI runtime services memory regions) |
| 28 | */ |
| 29 | #define EFI_RT_VIRTUAL_BASE SZ_512M |
| 30 | #define EFI_RT_VIRTUAL_SIZE SZ_512M |
| 31 | |
| 32 | #ifdef CONFIG_ARM64 |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 33 | # define EFI_RT_VIRTUAL_LIMIT DEFAULT_MAP_WINDOW_64 |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 34 | #else |
| 35 | # define EFI_RT_VIRTUAL_LIMIT TASK_SIZE |
| 36 | #endif |
| 37 | |
| 38 | static u64 virtmap_base = EFI_RT_VIRTUAL_BASE; |
| 39 | |
| 40 | void efi_char16_printk(efi_system_table_t *sys_table_arg, |
| 41 | efi_char16_t *str) |
| 42 | { |
| 43 | struct efi_simple_text_output_protocol *out; |
| 44 | |
| 45 | out = (struct efi_simple_text_output_protocol *)sys_table_arg->con_out; |
| 46 | out->output_string(out, str); |
| 47 | } |
| 48 | |
| 49 | static struct screen_info *setup_graphics(efi_system_table_t *sys_table_arg) |
| 50 | { |
| 51 | efi_guid_t gop_proto = EFI_GRAPHICS_OUTPUT_PROTOCOL_GUID; |
| 52 | efi_status_t status; |
| 53 | unsigned long size; |
| 54 | void **gop_handle = NULL; |
| 55 | struct screen_info *si = NULL; |
| 56 | |
| 57 | size = 0; |
| 58 | status = efi_call_early(locate_handle, EFI_LOCATE_BY_PROTOCOL, |
| 59 | &gop_proto, NULL, &size, gop_handle); |
| 60 | if (status == EFI_BUFFER_TOO_SMALL) { |
| 61 | si = alloc_screen_info(sys_table_arg); |
| 62 | if (!si) |
| 63 | return NULL; |
| 64 | efi_setup_gop(sys_table_arg, si, &gop_proto, size); |
| 65 | } |
| 66 | return si; |
| 67 | } |
| 68 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 69 | void install_memreserve_table(efi_system_table_t *sys_table_arg) |
| 70 | { |
| 71 | struct linux_efi_memreserve *rsv; |
| 72 | efi_guid_t memreserve_table_guid = LINUX_EFI_MEMRESERVE_TABLE_GUID; |
| 73 | efi_status_t status; |
| 74 | |
| 75 | status = efi_call_early(allocate_pool, EFI_LOADER_DATA, sizeof(*rsv), |
| 76 | (void **)&rsv); |
| 77 | if (status != EFI_SUCCESS) { |
| 78 | pr_efi_err(sys_table_arg, "Failed to allocate memreserve entry!\n"); |
| 79 | return; |
| 80 | } |
| 81 | |
| 82 | rsv->next = 0; |
| 83 | rsv->size = 0; |
| 84 | atomic_set(&rsv->count, 0); |
| 85 | |
| 86 | status = efi_call_early(install_configuration_table, |
| 87 | &memreserve_table_guid, |
| 88 | rsv); |
| 89 | if (status != EFI_SUCCESS) |
| 90 | pr_efi_err(sys_table_arg, "Failed to install memreserve config table!\n"); |
| 91 | } |
| 92 | |
| 93 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 94 | /* |
| 95 | * This function handles the architcture specific differences between arm and |
| 96 | * arm64 regarding where the kernel image must be loaded and any memory that |
| 97 | * must be reserved. On failure it is required to free all |
| 98 | * all allocations it has made. |
| 99 | */ |
| 100 | efi_status_t handle_kernel_image(efi_system_table_t *sys_table, |
| 101 | unsigned long *image_addr, |
| 102 | unsigned long *image_size, |
| 103 | unsigned long *reserve_addr, |
| 104 | unsigned long *reserve_size, |
| 105 | unsigned long dram_base, |
| 106 | efi_loaded_image_t *image); |
| 107 | /* |
| 108 | * EFI entry point for the arm/arm64 EFI stubs. This is the entrypoint |
| 109 | * that is described in the PE/COFF header. Most of the code is the same |
| 110 | * for both archictectures, with the arch-specific code provided in the |
| 111 | * handle_kernel_image() function. |
| 112 | */ |
| 113 | unsigned long efi_entry(void *handle, efi_system_table_t *sys_table, |
| 114 | unsigned long *image_addr) |
| 115 | { |
| 116 | efi_loaded_image_t *image; |
| 117 | efi_status_t status; |
| 118 | unsigned long image_size = 0; |
| 119 | unsigned long dram_base; |
| 120 | /* addr/point and size pairs for memory management*/ |
| 121 | unsigned long initrd_addr; |
| 122 | u64 initrd_size = 0; |
| 123 | unsigned long fdt_addr = 0; /* Original DTB */ |
| 124 | unsigned long fdt_size = 0; |
| 125 | char *cmdline_ptr = NULL; |
| 126 | int cmdline_size = 0; |
| 127 | unsigned long new_fdt_addr; |
| 128 | efi_guid_t loaded_image_proto = LOADED_IMAGE_PROTOCOL_GUID; |
| 129 | unsigned long reserve_addr = 0; |
| 130 | unsigned long reserve_size = 0; |
| 131 | enum efi_secureboot_mode secure_boot; |
| 132 | struct screen_info *si; |
| 133 | |
| 134 | /* Check if we were booted by the EFI firmware */ |
| 135 | if (sys_table->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) |
| 136 | goto fail; |
| 137 | |
| 138 | status = check_platform_features(sys_table); |
| 139 | if (status != EFI_SUCCESS) |
| 140 | goto fail; |
| 141 | |
| 142 | /* |
| 143 | * Get a handle to the loaded image protocol. This is used to get |
| 144 | * information about the running image, such as size and the command |
| 145 | * line. |
| 146 | */ |
| 147 | status = sys_table->boottime->handle_protocol(handle, |
| 148 | &loaded_image_proto, (void *)&image); |
| 149 | if (status != EFI_SUCCESS) { |
| 150 | pr_efi_err(sys_table, "Failed to get loaded image protocol\n"); |
| 151 | goto fail; |
| 152 | } |
| 153 | |
| 154 | dram_base = get_dram_base(sys_table); |
| 155 | if (dram_base == EFI_ERROR) { |
| 156 | pr_efi_err(sys_table, "Failed to find DRAM base\n"); |
| 157 | goto fail; |
| 158 | } |
| 159 | |
| 160 | /* |
| 161 | * Get the command line from EFI, using the LOADED_IMAGE |
| 162 | * protocol. We are going to copy the command line into the |
| 163 | * device tree, so this can be allocated anywhere. |
| 164 | */ |
| 165 | cmdline_ptr = efi_convert_cmdline(sys_table, image, &cmdline_size); |
| 166 | if (!cmdline_ptr) { |
| 167 | pr_efi_err(sys_table, "getting command line via LOADED_IMAGE_PROTOCOL\n"); |
| 168 | goto fail; |
| 169 | } |
| 170 | |
| 171 | if (IS_ENABLED(CONFIG_CMDLINE_EXTEND) || |
| 172 | IS_ENABLED(CONFIG_CMDLINE_FORCE) || |
| 173 | cmdline_size == 0) |
| 174 | efi_parse_options(CONFIG_CMDLINE); |
| 175 | |
| 176 | if (!IS_ENABLED(CONFIG_CMDLINE_FORCE) && cmdline_size > 0) |
| 177 | efi_parse_options(cmdline_ptr); |
| 178 | |
| 179 | pr_efi(sys_table, "Booting Linux Kernel...\n"); |
| 180 | |
| 181 | si = setup_graphics(sys_table); |
| 182 | |
| 183 | status = handle_kernel_image(sys_table, image_addr, &image_size, |
| 184 | &reserve_addr, |
| 185 | &reserve_size, |
| 186 | dram_base, image); |
| 187 | if (status != EFI_SUCCESS) { |
| 188 | pr_efi_err(sys_table, "Failed to relocate kernel\n"); |
| 189 | goto fail_free_cmdline; |
| 190 | } |
| 191 | |
| 192 | /* Ask the firmware to clear memory on unclean shutdown */ |
| 193 | efi_enable_reset_attack_mitigation(sys_table); |
| 194 | |
| 195 | secure_boot = efi_get_secureboot(sys_table); |
| 196 | |
| 197 | /* |
| 198 | * Unauthenticated device tree data is a security hazard, so ignore |
| 199 | * 'dtb=' unless UEFI Secure Boot is disabled. We assume that secure |
| 200 | * boot is enabled if we can't determine its state. |
| 201 | */ |
| 202 | if (!IS_ENABLED(CONFIG_EFI_ARMSTUB_DTB_LOADER) || |
| 203 | secure_boot != efi_secureboot_mode_disabled) { |
| 204 | if (strstr(cmdline_ptr, "dtb=")) |
| 205 | pr_efi(sys_table, "Ignoring DTB from command line.\n"); |
| 206 | } else { |
| 207 | status = handle_cmdline_files(sys_table, image, cmdline_ptr, |
| 208 | "dtb=", |
| 209 | ~0UL, &fdt_addr, &fdt_size); |
| 210 | |
| 211 | if (status != EFI_SUCCESS) { |
| 212 | pr_efi_err(sys_table, "Failed to load device tree!\n"); |
| 213 | goto fail_free_image; |
| 214 | } |
| 215 | } |
| 216 | |
| 217 | if (fdt_addr) { |
| 218 | pr_efi(sys_table, "Using DTB from command line\n"); |
| 219 | } else { |
| 220 | /* Look for a device tree configuration table entry. */ |
| 221 | fdt_addr = (uintptr_t)get_fdt(sys_table, &fdt_size); |
| 222 | if (fdt_addr) |
| 223 | pr_efi(sys_table, "Using DTB from configuration table\n"); |
| 224 | } |
| 225 | |
| 226 | if (!fdt_addr) |
| 227 | pr_efi(sys_table, "Generating empty DTB\n"); |
| 228 | |
| 229 | status = handle_cmdline_files(sys_table, image, cmdline_ptr, "initrd=", |
| 230 | efi_get_max_initrd_addr(dram_base, |
| 231 | *image_addr), |
| 232 | (unsigned long *)&initrd_addr, |
| 233 | (unsigned long *)&initrd_size); |
| 234 | if (status != EFI_SUCCESS) |
| 235 | pr_efi_err(sys_table, "Failed initrd from command line!\n"); |
| 236 | |
| 237 | efi_random_get_seed(sys_table); |
| 238 | |
| 239 | /* hibernation expects the runtime regions to stay in the same place */ |
| 240 | if (!IS_ENABLED(CONFIG_HIBERNATION) && !nokaslr()) { |
| 241 | /* |
| 242 | * Randomize the base of the UEFI runtime services region. |
| 243 | * Preserve the 2 MB alignment of the region by taking a |
| 244 | * shift of 21 bit positions into account when scaling |
| 245 | * the headroom value using a 32-bit random value. |
| 246 | */ |
| 247 | static const u64 headroom = EFI_RT_VIRTUAL_LIMIT - |
| 248 | EFI_RT_VIRTUAL_BASE - |
| 249 | EFI_RT_VIRTUAL_SIZE; |
| 250 | u32 rnd; |
| 251 | |
| 252 | status = efi_get_random_bytes(sys_table, sizeof(rnd), |
| 253 | (u8 *)&rnd); |
| 254 | if (status == EFI_SUCCESS) { |
| 255 | virtmap_base = EFI_RT_VIRTUAL_BASE + |
| 256 | (((headroom >> 21) * rnd) >> (32 - 21)); |
| 257 | } |
| 258 | } |
| 259 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 260 | install_memreserve_table(sys_table); |
| 261 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 262 | new_fdt_addr = fdt_addr; |
| 263 | status = allocate_new_fdt_and_exit_boot(sys_table, handle, |
| 264 | &new_fdt_addr, efi_get_max_fdt_addr(dram_base), |
| 265 | initrd_addr, initrd_size, cmdline_ptr, |
| 266 | fdt_addr, fdt_size); |
| 267 | |
| 268 | /* |
| 269 | * If all went well, we need to return the FDT address to the |
| 270 | * calling function so it can be passed to kernel as part of |
| 271 | * the kernel boot protocol. |
| 272 | */ |
| 273 | if (status == EFI_SUCCESS) |
| 274 | return new_fdt_addr; |
| 275 | |
| 276 | pr_efi_err(sys_table, "Failed to update FDT and exit boot services\n"); |
| 277 | |
| 278 | efi_free(sys_table, initrd_size, initrd_addr); |
| 279 | efi_free(sys_table, fdt_size, fdt_addr); |
| 280 | |
| 281 | fail_free_image: |
| 282 | efi_free(sys_table, image_size, *image_addr); |
| 283 | efi_free(sys_table, reserve_size, reserve_addr); |
| 284 | fail_free_cmdline: |
| 285 | free_screen_info(sys_table, si); |
| 286 | efi_free(sys_table, cmdline_size, (unsigned long)cmdline_ptr); |
| 287 | fail: |
| 288 | return EFI_ERROR; |
| 289 | } |
| 290 | |
| 291 | static int cmp_mem_desc(const void *l, const void *r) |
| 292 | { |
| 293 | const efi_memory_desc_t *left = l, *right = r; |
| 294 | |
| 295 | return (left->phys_addr > right->phys_addr) ? 1 : -1; |
| 296 | } |
| 297 | |
| 298 | /* |
| 299 | * Returns whether region @left ends exactly where region @right starts, |
| 300 | * or false if either argument is NULL. |
| 301 | */ |
| 302 | static bool regions_are_adjacent(efi_memory_desc_t *left, |
| 303 | efi_memory_desc_t *right) |
| 304 | { |
| 305 | u64 left_end; |
| 306 | |
| 307 | if (left == NULL || right == NULL) |
| 308 | return false; |
| 309 | |
| 310 | left_end = left->phys_addr + left->num_pages * EFI_PAGE_SIZE; |
| 311 | |
| 312 | return left_end == right->phys_addr; |
| 313 | } |
| 314 | |
| 315 | /* |
| 316 | * Returns whether region @left and region @right have compatible memory type |
| 317 | * mapping attributes, and are both EFI_MEMORY_RUNTIME regions. |
| 318 | */ |
| 319 | static bool regions_have_compatible_memory_type_attrs(efi_memory_desc_t *left, |
| 320 | efi_memory_desc_t *right) |
| 321 | { |
| 322 | static const u64 mem_type_mask = EFI_MEMORY_WB | EFI_MEMORY_WT | |
| 323 | EFI_MEMORY_WC | EFI_MEMORY_UC | |
| 324 | EFI_MEMORY_RUNTIME; |
| 325 | |
| 326 | return ((left->attribute ^ right->attribute) & mem_type_mask) == 0; |
| 327 | } |
| 328 | |
| 329 | /* |
| 330 | * efi_get_virtmap() - create a virtual mapping for the EFI memory map |
| 331 | * |
| 332 | * This function populates the virt_addr fields of all memory region descriptors |
| 333 | * in @memory_map whose EFI_MEMORY_RUNTIME attribute is set. Those descriptors |
| 334 | * are also copied to @runtime_map, and their total count is returned in @count. |
| 335 | */ |
| 336 | void efi_get_virtmap(efi_memory_desc_t *memory_map, unsigned long map_size, |
| 337 | unsigned long desc_size, efi_memory_desc_t *runtime_map, |
| 338 | int *count) |
| 339 | { |
| 340 | u64 efi_virt_base = virtmap_base; |
| 341 | efi_memory_desc_t *in, *prev = NULL, *out = runtime_map; |
| 342 | int l; |
| 343 | |
| 344 | /* |
| 345 | * To work around potential issues with the Properties Table feature |
| 346 | * introduced in UEFI 2.5, which may split PE/COFF executable images |
| 347 | * in memory into several RuntimeServicesCode and RuntimeServicesData |
| 348 | * regions, we need to preserve the relative offsets between adjacent |
| 349 | * EFI_MEMORY_RUNTIME regions with the same memory type attributes. |
| 350 | * The easiest way to find adjacent regions is to sort the memory map |
| 351 | * before traversing it. |
| 352 | */ |
| 353 | if (IS_ENABLED(CONFIG_ARM64)) |
| 354 | sort(memory_map, map_size / desc_size, desc_size, cmp_mem_desc, |
| 355 | NULL); |
| 356 | |
| 357 | for (l = 0; l < map_size; l += desc_size, prev = in) { |
| 358 | u64 paddr, size; |
| 359 | |
| 360 | in = (void *)memory_map + l; |
| 361 | if (!(in->attribute & EFI_MEMORY_RUNTIME)) |
| 362 | continue; |
| 363 | |
| 364 | paddr = in->phys_addr; |
| 365 | size = in->num_pages * EFI_PAGE_SIZE; |
| 366 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 367 | if (novamap()) { |
| 368 | in->virt_addr = in->phys_addr; |
| 369 | continue; |
| 370 | } |
| 371 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 372 | /* |
| 373 | * Make the mapping compatible with 64k pages: this allows |
| 374 | * a 4k page size kernel to kexec a 64k page size kernel and |
| 375 | * vice versa. |
| 376 | */ |
| 377 | if ((IS_ENABLED(CONFIG_ARM64) && |
| 378 | !regions_are_adjacent(prev, in)) || |
| 379 | !regions_have_compatible_memory_type_attrs(prev, in)) { |
| 380 | |
| 381 | paddr = round_down(in->phys_addr, SZ_64K); |
| 382 | size += in->phys_addr - paddr; |
| 383 | |
| 384 | /* |
| 385 | * Avoid wasting memory on PTEs by choosing a virtual |
| 386 | * base that is compatible with section mappings if this |
| 387 | * region has the appropriate size and physical |
| 388 | * alignment. (Sections are 2 MB on 4k granule kernels) |
| 389 | */ |
| 390 | if (IS_ALIGNED(in->phys_addr, SZ_2M) && size >= SZ_2M) |
| 391 | efi_virt_base = round_up(efi_virt_base, SZ_2M); |
| 392 | else |
| 393 | efi_virt_base = round_up(efi_virt_base, SZ_64K); |
| 394 | } |
| 395 | |
| 396 | in->virt_addr = efi_virt_base + in->phys_addr - paddr; |
| 397 | efi_virt_base += size; |
| 398 | |
| 399 | memcpy(out, in, desc_size); |
| 400 | out = (void *)out + desc_size; |
| 401 | ++*count; |
| 402 | } |
| 403 | } |