| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * ARM64 Specific Low-Level ACPI Boot Support |
| 4 | * |
| 5 | * Copyright (C) 2013-2014, Linaro Ltd. |
| 6 | * Author: Al Stone <al.stone@linaro.org> |
| 7 | * Author: Graeme Gregory <graeme.gregory@linaro.org> |
| 8 | * Author: Hanjun Guo <hanjun.guo@linaro.org> |
| 9 | * Author: Tomasz Nowicki <tomasz.nowicki@linaro.org> |
| 10 | * Author: Naresh Bhat <naresh.bhat@linaro.org> |
| 11 | */ |
| 12 | |
| 13 | #define pr_fmt(fmt) "ACPI: " fmt |
| 14 | |
| 15 | #include <linux/acpi.h> |
| 16 | #include <linux/arm-smccc.h> |
| 17 | #include <linux/cpumask.h> |
| 18 | #include <linux/efi.h> |
| 19 | #include <linux/efi-bgrt.h> |
| 20 | #include <linux/init.h> |
| 21 | #include <linux/irq.h> |
| 22 | #include <linux/irqdomain.h> |
| 23 | #include <linux/irq_work.h> |
| 24 | #include <linux/memblock.h> |
| 25 | #include <linux/of_fdt.h> |
| 26 | #include <linux/libfdt.h> |
| 27 | #include <linux/smp.h> |
| 28 | #include <linux/serial_core.h> |
| 29 | #include <linux/suspend.h> |
| 30 | #include <linux/pgtable.h> |
| 31 | |
| 32 | #include <acpi/ghes.h> |
| 33 | #include <acpi/processor.h> |
| 34 | #include <asm/cputype.h> |
| 35 | #include <asm/cpu_ops.h> |
| 36 | #include <asm/daifflags.h> |
| 37 | #include <asm/smp_plat.h> |
| 38 | |
| 39 | int acpi_noirq = 1; /* skip ACPI IRQ initialization */ |
| 40 | int acpi_disabled = 1; |
| 41 | EXPORT_SYMBOL(acpi_disabled); |
| 42 | |
| 43 | int acpi_pci_disabled = 1; /* skip ACPI PCI scan and IRQ initialization */ |
| 44 | EXPORT_SYMBOL(acpi_pci_disabled); |
| 45 | |
| 46 | static bool param_acpi_off __initdata; |
| 47 | static bool param_acpi_on __initdata; |
| 48 | static bool param_acpi_force __initdata; |
| 49 | static bool param_acpi_nospcr __initdata; |
| 50 | |
| 51 | static int __init parse_acpi(char *arg) |
| 52 | { |
| 53 | if (!arg) |
| 54 | return -EINVAL; |
| 55 | |
| 56 | /* "acpi=off" disables both ACPI table parsing and interpreter */ |
| 57 | if (strcmp(arg, "off" ) == 0) |
| 58 | param_acpi_off = true; |
| 59 | else if (strcmp(arg, "on" ) == 0) /* prefer ACPI over DT */ |
| 60 | param_acpi_on = true; |
| 61 | else if (strcmp(arg, "force" ) == 0) /* force ACPI to be enabled */ |
| 62 | param_acpi_force = true; |
| 63 | else if (strcmp(arg, "nospcr" ) == 0) /* disable SPCR as default console */ |
| 64 | param_acpi_nospcr = true; |
| 65 | else |
| 66 | return -EINVAL; /* Core will print when we return error */ |
| 67 | |
| 68 | return 0; |
| 69 | } |
| 70 | early_param("acpi" , parse_acpi); |
| 71 | |
| 72 | static bool __init dt_is_stub(void) |
| 73 | { |
| 74 | int node; |
| 75 | |
| 76 | fdt_for_each_subnode(node, initial_boot_params, 0) { |
| 77 | const char *name = fdt_get_name(fdt: initial_boot_params, nodeoffset: node, NULL); |
| 78 | if (strcmp(name, "chosen" ) == 0) |
| 79 | continue; |
| 80 | if (strcmp(name, "hypervisor" ) == 0 && |
| 81 | of_flat_dt_is_compatible(node, name: "xen,xen" )) |
| 82 | continue; |
| 83 | |
| 84 | return false; |
| 85 | } |
| 86 | |
| 87 | return true; |
| 88 | } |
| 89 | |
| 90 | /* |
| 91 | * __acpi_map_table() will be called before page_init(), so early_ioremap() |
| 92 | * or early_memremap() should be called here to for ACPI table mapping. |
| 93 | */ |
| 94 | void __init __iomem *__acpi_map_table(unsigned long phys, unsigned long size) |
| 95 | { |
| 96 | if (!size) |
| 97 | return NULL; |
| 98 | |
| 99 | return early_memremap(phys_addr: phys, size); |
| 100 | } |
| 101 | |
| 102 | void __init __acpi_unmap_table(void __iomem *map, unsigned long size) |
| 103 | { |
| 104 | if (!map || !size) |
| 105 | return; |
| 106 | |
| 107 | early_memunmap(addr: map, size); |
| 108 | } |
| 109 | |
| 110 | bool __init acpi_psci_present(void) |
| 111 | { |
| 112 | return acpi_gbl_FADT.arm_boot_flags & ACPI_FADT_PSCI_COMPLIANT; |
| 113 | } |
| 114 | |
| 115 | /* Whether HVC must be used instead of SMC as the PSCI conduit */ |
| 116 | bool acpi_psci_use_hvc(void) |
| 117 | { |
| 118 | return acpi_gbl_FADT.arm_boot_flags & ACPI_FADT_PSCI_USE_HVC; |
| 119 | } |
| 120 | |
| 121 | /* |
| 122 | * acpi_fadt_sanity_check() - Check FADT presence and carry out sanity |
| 123 | * checks on it |
| 124 | * |
| 125 | * Return 0 on success, <0 on failure |
| 126 | */ |
| 127 | static int __init acpi_fadt_sanity_check(void) |
| 128 | { |
| 129 | struct acpi_table_header *table; |
| 130 | struct acpi_table_fadt *fadt; |
| 131 | acpi_status status; |
| 132 | int ret = 0; |
| 133 | |
| 134 | /* |
| 135 | * FADT is required on arm64; retrieve it to check its presence |
| 136 | * and carry out revision and ACPI HW reduced compliance tests |
| 137 | */ |
| 138 | status = acpi_get_table(ACPI_SIG_FADT, instance: 0, out_table: &table); |
| 139 | if (ACPI_FAILURE(status)) { |
| 140 | const char *msg = acpi_format_exception(exception: status); |
| 141 | |
| 142 | pr_err("Failed to get FADT table, %s\n" , msg); |
| 143 | return -ENODEV; |
| 144 | } |
| 145 | |
| 146 | fadt = (struct acpi_table_fadt *)table; |
| 147 | |
| 148 | /* |
| 149 | * Revision in table header is the FADT Major revision, and there |
| 150 | * is a minor revision of FADT which was introduced by ACPI 5.1, |
| 151 | * we only deal with ACPI 5.1 or newer revision to get GIC and SMP |
| 152 | * boot protocol configuration data. |
| 153 | */ |
| 154 | if (table->revision < 5 || |
| 155 | (table->revision == 5 && fadt->minor_revision < 1)) { |
| 156 | pr_err(FW_BUG "Unsupported FADT revision %d.%d, should be 5.1+\n" , |
| 157 | table->revision, fadt->minor_revision); |
| 158 | |
| 159 | if (!fadt->arm_boot_flags) { |
| 160 | ret = -EINVAL; |
| 161 | goto out; |
| 162 | } |
| 163 | pr_err("FADT has ARM boot flags set, assuming 5.1\n" ); |
| 164 | } |
| 165 | |
| 166 | if (!(fadt->flags & ACPI_FADT_HW_REDUCED)) { |
| 167 | pr_err("FADT not ACPI hardware reduced compliant\n" ); |
| 168 | ret = -EINVAL; |
| 169 | } |
| 170 | |
| 171 | out: |
| 172 | /* |
| 173 | * acpi_get_table() creates FADT table mapping that |
| 174 | * should be released after parsing and before resuming boot |
| 175 | */ |
| 176 | acpi_put_table(table); |
| 177 | return ret; |
| 178 | } |
| 179 | |
| 180 | /* |
| 181 | * acpi_boot_table_init() called from setup_arch(), always. |
| 182 | * 1. find RSDP and get its address, and then find XSDT |
| 183 | * 2. extract all tables and checksums them all |
| 184 | * 3. check ACPI FADT revision |
| 185 | * 4. check ACPI FADT HW reduced flag |
| 186 | * |
| 187 | * We can parse ACPI boot-time tables such as MADT after |
| 188 | * this function is called. |
| 189 | * |
| 190 | * On return ACPI is enabled if either: |
| 191 | * |
| 192 | * - ACPI tables are initialized and sanity checks passed |
| 193 | * - acpi=force was passed in the command line and ACPI was not disabled |
| 194 | * explicitly through acpi=off command line parameter |
| 195 | * |
| 196 | * ACPI is disabled on function return otherwise |
| 197 | */ |
| 198 | void __init acpi_boot_table_init(void) |
| 199 | { |
| 200 | /* |
| 201 | * Enable ACPI instead of device tree unless |
| 202 | * - ACPI has been disabled explicitly (acpi=off), or |
| 203 | * - the device tree is not empty (it has more than just a /chosen node, |
| 204 | * and a /hypervisor node when running on Xen) |
| 205 | * and ACPI has not been [force] enabled (acpi=on|force) |
| 206 | */ |
| 207 | if (param_acpi_off || |
| 208 | (!param_acpi_on && !param_acpi_force && !dt_is_stub())) |
| 209 | goto done; |
| 210 | |
| 211 | /* |
| 212 | * ACPI is disabled at this point. Enable it in order to parse |
| 213 | * the ACPI tables and carry out sanity checks |
| 214 | */ |
| 215 | enable_acpi(); |
| 216 | |
| 217 | /* |
| 218 | * If ACPI tables are initialized and FADT sanity checks passed, |
| 219 | * leave ACPI enabled and carry on booting; otherwise disable ACPI |
| 220 | * on initialization error. |
| 221 | * If acpi=force was passed on the command line it forces ACPI |
| 222 | * to be enabled even if its initialization failed. |
| 223 | */ |
| 224 | if (acpi_table_init() || acpi_fadt_sanity_check()) { |
| 225 | pr_err("Failed to init ACPI tables\n" ); |
| 226 | if (!param_acpi_force) |
| 227 | disable_acpi(); |
| 228 | } |
| 229 | |
| 230 | done: |
| 231 | if (acpi_disabled) { |
| 232 | if (earlycon_acpi_spcr_enable) |
| 233 | early_init_dt_scan_chosen_stdout(); |
| 234 | } else { |
| 235 | #ifdef CONFIG_HIBERNATION |
| 236 | struct acpi_table_header *facs = NULL; |
| 237 | acpi_get_table(ACPI_SIG_FACS, instance: 1, out_table: &facs); |
| 238 | if (facs) { |
| 239 | swsusp_hardware_signature = |
| 240 | ((struct acpi_table_facs *)facs)->hardware_signature; |
| 241 | acpi_put_table(table: facs); |
| 242 | } |
| 243 | #endif |
| 244 | |
| 245 | /* |
| 246 | * For varying privacy and security reasons, sometimes need |
| 247 | * to completely silence the serial console output, and only |
| 248 | * enable it when needed. |
| 249 | * But there are many existing systems that depend on this |
| 250 | * behaviour, use acpi=nospcr to disable console in ACPI SPCR |
| 251 | * table as default serial console. |
| 252 | */ |
| 253 | acpi_parse_spcr(enable_earlycon: earlycon_acpi_spcr_enable, |
| 254 | enable_console: !param_acpi_nospcr); |
| 255 | |
| 256 | if (IS_ENABLED(CONFIG_ACPI_BGRT)) |
| 257 | acpi_table_parse(ACPI_SIG_BGRT, handler: acpi_parse_bgrt); |
| 258 | } |
| 259 | } |
| 260 | |
| 261 | static pgprot_t __acpi_get_writethrough_mem_attribute(void) |
| 262 | { |
| 263 | /* |
| 264 | * Although UEFI specifies the use of Normal Write-through for |
| 265 | * EFI_MEMORY_WT, it is seldom used in practice and not implemented |
| 266 | * by most (all?) CPUs. Rather than allocate a MAIR just for this |
| 267 | * purpose, emit a warning and use Normal Non-cacheable instead. |
| 268 | */ |
| 269 | pr_warn_once("No MAIR allocation for EFI_MEMORY_WT; treating as Normal Non-cacheable\n" ); |
| 270 | return __pgprot(PROT_NORMAL_NC); |
| 271 | } |
| 272 | |
| 273 | pgprot_t __acpi_get_mem_attribute(phys_addr_t addr) |
| 274 | { |
| 275 | /* |
| 276 | * According to "Table 8 Map: EFI memory types to AArch64 memory |
| 277 | * types" of UEFI 2.5 section 2.3.6.1, each EFI memory type is |
| 278 | * mapped to a corresponding MAIR attribute encoding. |
| 279 | * The EFI memory attribute advises all possible capabilities |
| 280 | * of a memory region. |
| 281 | */ |
| 282 | |
| 283 | u64 attr; |
| 284 | |
| 285 | attr = efi_mem_attributes(phys_addr: addr); |
| 286 | if (attr & EFI_MEMORY_WB) |
| 287 | return PAGE_KERNEL; |
| 288 | if (attr & EFI_MEMORY_WC) |
| 289 | return __pgprot(PROT_NORMAL_NC); |
| 290 | if (attr & EFI_MEMORY_WT) |
| 291 | return __acpi_get_writethrough_mem_attribute(); |
| 292 | return __pgprot(PROT_DEVICE_nGnRnE); |
| 293 | } |
| 294 | |
| 295 | void __iomem *acpi_os_ioremap(acpi_physical_address phys, acpi_size size) |
| 296 | { |
| 297 | efi_memory_desc_t *md, *region = NULL; |
| 298 | pgprot_t prot; |
| 299 | |
| 300 | if (WARN_ON_ONCE(!efi_enabled(EFI_MEMMAP))) |
| 301 | return NULL; |
| 302 | |
| 303 | for_each_efi_memory_desc(md) { |
| 304 | u64 end = md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT); |
| 305 | |
| 306 | if (phys < md->phys_addr || phys >= end) |
| 307 | continue; |
| 308 | |
| 309 | if (phys + size > end) { |
| 310 | pr_warn(FW_BUG "requested region covers multiple EFI memory regions\n" ); |
| 311 | return NULL; |
| 312 | } |
| 313 | region = md; |
| 314 | break; |
| 315 | } |
| 316 | |
| 317 | /* |
| 318 | * It is fine for AML to remap regions that are not represented in the |
| 319 | * EFI memory map at all, as it only describes normal memory, and MMIO |
| 320 | * regions that require a virtual mapping to make them accessible to |
| 321 | * the EFI runtime services. |
| 322 | */ |
| 323 | prot = __pgprot(PROT_DEVICE_nGnRnE); |
| 324 | if (region) { |
| 325 | switch (region->type) { |
| 326 | case EFI_LOADER_CODE: |
| 327 | case EFI_LOADER_DATA: |
| 328 | case EFI_BOOT_SERVICES_CODE: |
| 329 | case EFI_BOOT_SERVICES_DATA: |
| 330 | case EFI_CONVENTIONAL_MEMORY: |
| 331 | case EFI_PERSISTENT_MEMORY: |
| 332 | if (memblock_is_map_memory(addr: phys) || |
| 333 | !memblock_is_region_memory(base: phys, size)) { |
| 334 | pr_warn(FW_BUG "requested region covers kernel memory @ %pa\n" , &phys); |
| 335 | return NULL; |
| 336 | } |
| 337 | /* |
| 338 | * Mapping kernel memory is permitted if the region in |
| 339 | * question is covered by a single memblock with the |
| 340 | * NOMAP attribute set: this enables the use of ACPI |
| 341 | * table overrides passed via initramfs, which are |
| 342 | * reserved in memory using arch_reserve_mem_area() |
| 343 | * below. As this particular use case only requires |
| 344 | * read access, fall through to the R/O mapping case. |
| 345 | */ |
| 346 | fallthrough; |
| 347 | |
| 348 | case EFI_RUNTIME_SERVICES_CODE: |
| 349 | /* |
| 350 | * This would be unusual, but not problematic per se, |
| 351 | * as long as we take care not to create a writable |
| 352 | * mapping for executable code. |
| 353 | */ |
| 354 | prot = PAGE_KERNEL_RO; |
| 355 | break; |
| 356 | |
| 357 | case EFI_ACPI_RECLAIM_MEMORY: |
| 358 | /* |
| 359 | * ACPI reclaim memory is used to pass firmware tables |
| 360 | * and other data that is intended for consumption by |
| 361 | * the OS only, which may decide it wants to reclaim |
| 362 | * that memory and use it for something else. We never |
| 363 | * do that, but we usually add it to the linear map |
| 364 | * anyway, in which case we should use the existing |
| 365 | * mapping. |
| 366 | */ |
| 367 | if (memblock_is_map_memory(addr: phys)) |
| 368 | return (void __iomem *)__phys_to_virt(phys); |
| 369 | fallthrough; |
| 370 | |
| 371 | default: |
| 372 | if (region->attribute & EFI_MEMORY_WB) |
| 373 | prot = PAGE_KERNEL; |
| 374 | else if (region->attribute & EFI_MEMORY_WC) |
| 375 | prot = __pgprot(PROT_NORMAL_NC); |
| 376 | else if (region->attribute & EFI_MEMORY_WT) |
| 377 | prot = __acpi_get_writethrough_mem_attribute(); |
| 378 | } |
| 379 | } |
| 380 | return ioremap_prot(offset: phys, size, prot); |
| 381 | } |
| 382 | |
| 383 | /* |
| 384 | * Claim Synchronous External Aborts as a firmware first notification. |
| 385 | * |
| 386 | * Used by KVM and the arch do_sea handler. |
| 387 | * @regs may be NULL when called from process context. |
| 388 | */ |
| 389 | int apei_claim_sea(struct pt_regs *regs) |
| 390 | { |
| 391 | int err = -ENOENT; |
| 392 | bool return_to_irqs_enabled; |
| 393 | unsigned long current_flags; |
| 394 | |
| 395 | if (!IS_ENABLED(CONFIG_ACPI_APEI_GHES)) |
| 396 | return err; |
| 397 | |
| 398 | current_flags = local_daif_save_flags(); |
| 399 | |
| 400 | /* current_flags isn't useful here as daif doesn't tell us about pNMI */ |
| 401 | return_to_irqs_enabled = !irqs_disabled_flags(arch_local_save_flags()); |
| 402 | |
| 403 | if (regs) |
| 404 | return_to_irqs_enabled = !regs_irqs_disabled(regs); |
| 405 | |
| 406 | /* |
| 407 | * SEA can interrupt SError, mask it and describe this as an NMI so |
| 408 | * that APEI defers the handling. |
| 409 | */ |
| 410 | local_daif_restore(DAIF_ERRCTX); |
| 411 | nmi_enter(); |
| 412 | err = ghes_notify_sea(); |
| 413 | nmi_exit(); |
| 414 | |
| 415 | /* |
| 416 | * APEI NMI-like notifications are deferred to irq_work. Unless |
| 417 | * we interrupted irqs-masked code, we can do that now. |
| 418 | */ |
| 419 | if (!err) { |
| 420 | if (return_to_irqs_enabled) { |
| 421 | local_daif_restore(DAIF_PROCCTX_NOIRQ); |
| 422 | __irq_enter(); |
| 423 | irq_work_run(); |
| 424 | __irq_exit(); |
| 425 | } else { |
| 426 | pr_warn_ratelimited("APEI work queued but not completed\n" ); |
| 427 | err = -EINPROGRESS; |
| 428 | } |
| 429 | } |
| 430 | |
| 431 | local_daif_restore(current_flags); |
| 432 | |
| 433 | return err; |
| 434 | } |
| 435 | |
| 436 | void arch_reserve_mem_area(acpi_physical_address addr, size_t size) |
| 437 | { |
| 438 | memblock_mark_nomap(base: addr, size); |
| 439 | } |
| 440 | |
| 441 | #ifdef CONFIG_ACPI_HOTPLUG_CPU |
| 442 | int acpi_map_cpu(acpi_handle handle, phys_cpuid_t physid, u32 apci_id, |
| 443 | int *pcpu) |
| 444 | { |
| 445 | /* If an error code is passed in this stub can't fix it */ |
| 446 | if (*pcpu < 0) { |
| 447 | pr_warn_once("Unable to map CPU to valid ID\n" ); |
| 448 | return *pcpu; |
| 449 | } |
| 450 | |
| 451 | return 0; |
| 452 | } |
| 453 | EXPORT_SYMBOL(acpi_map_cpu); |
| 454 | |
| 455 | int acpi_unmap_cpu(int cpu) |
| 456 | { |
| 457 | return 0; |
| 458 | } |
| 459 | EXPORT_SYMBOL(acpi_unmap_cpu); |
| 460 | #endif /* CONFIG_ACPI_HOTPLUG_CPU */ |
| 461 | |