arch/x86/platform/efi/efi_64.c

   1 /*
   2  * x86_64 specific EFI support functions
   3  * Based on Extensible Firmware Interface Specification version 1.0
   4  *
   5  * Copyright (C) 2005-2008 Intel Co.
   6  *      Fenghua Yu <fenghua.yu@intel.com>
   7  *      Bibo Mao <bibo.mao@intel.com>
   8  *      Chandramouli Narayanan <mouli@linux.intel.com>
   9  *      Huang Ying <ying.huang@intel.com>
  10  *
  11  * Code to convert EFI to E820 map has been implemented in elilo bootloader
  12  * based on a EFI patch by Edgar Hucek. Based on the E820 map, the page table
  13  * is setup appropriately for EFI runtime code.
  14  * - mouli 06/14/2007.
  15  *
  16  */
  17
  18 #include <linux/kernel.h>
  19 #include <linux/init.h>
  20 #include <linux/mm.h>
  21 #include <linux/types.h>
  22 #include <linux/spinlock.h>
  23 #include <linux/bootmem.h>
  24 #include <linux/ioport.h>
  25 #include <linux/module.h>
  26 #include <linux/efi.h>
  27 #include <linux/uaccess.h>
  28 #include <linux/io.h>
  29 #include <linux/reboot.h>
  30 #include <linux/slab.h>
  31
  32 #include <asm/setup.h>
  33 #include <asm/page.h>
  34 #include <asm/e820.h>
  35 #include <asm/pgtable.h>
  36 #include <asm/tlbflush.h>
  37 #include <asm/proto.h>
  38 #include <asm/efi.h>
  39 #include <asm/cacheflush.h>
  40 #include <asm/fixmap.h>
  41 #include <asm/realmode.h>
  42 #include <asm/time.h>
  43 #include <asm/nospec-branch.h>
  44
  45 /*
  46  * We allocate runtime services regions bottom-up, starting from -4G, i.e.
  47  * 0xffff_ffff_0000_0000 and limit EFI VA mapping space to 64G.
  48  */
  49 static u64 efi_va = EFI_VA_START;
  50
  51 /*
  52  * Scratch space used for switching the pagetable in the EFI stub
  53  */
  54 struct efi_scratch {
  55         u64 r15;
  56         u64 prev_cr3;
  57         pgd_t *efi_pgt;
  58         bool use_pgd;
  59         u64 phys_stack;
  60 } __packed;
  61
  62 static void __init early_code_mapping_set_exec(int executable)
  63 {
  64         efi_memory_desc_t *md;
  65         void *p;
  66
  67         if (!(__supported_pte_mask & _PAGE_NX))
  68                 return;
  69
  70         /* Make EFI service code area executable */
  71         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
  72                 md = p;
  73                 if (md->type == EFI_RUNTIME_SERVICES_CODE ||
  74                     md->type == EFI_BOOT_SERVICES_CODE)
  75                         efi_set_executable(md, executable);
  76         }
  77 }
  78
  79 pgd_t * __init efi_call_phys_prolog(void)
  80 {
  81         unsigned long vaddress;
  82         pgd_t *save_pgd;
  83
  84         int pgd;
  85         int n_pgds;
  86
  87         if (!efi_enabled(EFI_OLD_MEMMAP))
  88                 return NULL;
  89
  90         early_code_mapping_set_exec(1);
  91
  92         n_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT), PGDIR_SIZE);
  93         save_pgd = kmalloc(n_pgds * sizeof(pgd_t), GFP_KERNEL);
  94
  95         for (pgd = 0; pgd < n_pgds; pgd++) {
  96                 save_pgd[pgd] = *pgd_offset_k(pgd * PGDIR_SIZE);
  97                 vaddress = (unsigned long)__va(pgd * PGDIR_SIZE);
  98                 set_pgd(pgd_offset_k(pgd * PGDIR_SIZE), *pgd_offset_k(vaddress));
  99         }
 100         __flush_tlb_all();
 101
 102         return save_pgd;
 103 }
 104
 105 void __init efi_call_phys_epilog(pgd_t *save_pgd)
 106 {
 107         /*
 108          * After the lock is released, the original page table is restored.
 109          */
 110         int pgd_idx;
 111         int nr_pgds;
 112
 113         if (!save_pgd)
 114                 return;
 115
 116         nr_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT) , PGDIR_SIZE);
 117
 118         for (pgd_idx = 0; pgd_idx < nr_pgds; pgd_idx++)
 119                 set_pgd(pgd_offset_k(pgd_idx * PGDIR_SIZE), save_pgd[pgd_idx]);
 120
 121         kfree(save_pgd);
 122
 123         __flush_tlb_all();
 124         early_code_mapping_set_exec(0);
 125 }
 126
 127 /*
 128  * Add low kernel mappings for passing arguments to EFI functions.
 129  */
 130 void efi_sync_low_kernel_mappings(void)
 131 {
 132         unsigned num_pgds;
 133         pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
 134
 135         if (efi_enabled(EFI_OLD_MEMMAP))
 136                 return;
 137
 138         num_pgds = pgd_index(MODULES_END - 1) - pgd_index(PAGE_OFFSET);
 139
 140         memcpy(pgd + pgd_index(PAGE_OFFSET),
 141                 init_mm.pgd + pgd_index(PAGE_OFFSET),
 142                 sizeof(pgd_t) * num_pgds);
 143 }
 144
 145 int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
 146 {
 147         unsigned long text;
 148         struct page *page;
 149         unsigned npages;
 150         pgd_t *pgd;
 151
 152         if (efi_enabled(EFI_OLD_MEMMAP))
 153                 return 0;
 154
 155         efi_scratch.efi_pgt = (pgd_t *)(unsigned long)real_mode_header->trampoline_pgd;
 156         pgd = __va(efi_scratch.efi_pgt);
 157
 158         /*
 159          * It can happen that the physical address of new_memmap lands in memory
 160          * which is not mapped in the EFI page table. Therefore we need to go
 161          * and ident-map those pages containing the map before calling
 162          * phys_efi_set_virtual_address_map().
 163          */
 164         if (kernel_map_pages_in_pgd(pgd, pa_memmap, pa_memmap, num_pages, _PAGE_NX)) {
 165                 pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
 166                 return 1;
 167         }
 168
 169         efi_scratch.use_pgd = true;
 170
 171         /*
 172          * When making calls to the firmware everything needs to be 1:1
 173          * mapped and addressable with 32-bit pointers. Map the kernel
 174          * text and allocate a new stack because we can't rely on the
 175          * stack pointer being < 4GB.
 176          */
 177         if (!IS_ENABLED(CONFIG_EFI_MIXED))
 178                 return 0;
 179
 180         page = alloc_page(GFP_KERNEL|__GFP_DMA32);
 181         if (!page)
 182                 panic("Unable to allocate EFI runtime stack < 4GB\n");
 183
 184         efi_scratch.phys_stack = virt_to_phys(page_address(page));
 185         efi_scratch.phys_stack += PAGE_SIZE; /* stack grows down */
 186
 187         npages = (_end - _text) >> PAGE_SHIFT;
 188         text = __pa(_text);
 189
 190         if (kernel_map_pages_in_pgd(pgd, text >> PAGE_SHIFT, text, npages, 0)) {
 191                 pr_err("Failed to map kernel text 1:1\n");
 192                 return 1;
 193         }
 194
 195         return 0;
 196 }
 197
 198 void __init efi_cleanup_page_tables(unsigned long pa_memmap, unsigned num_pages)
 199 {
 200         pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
 201
 202         kernel_unmap_pages_in_pgd(pgd, pa_memmap, num_pages);
 203 }
 204
 205 static void __init __map_region(efi_memory_desc_t *md, u64 va)
 206 {
 207         pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
 208         unsigned long pf = 0;
 209
 210         if (!(md->attribute & EFI_MEMORY_WB))
 211                 pf |= _PAGE_PCD;
 212
 213         if (kernel_map_pages_in_pgd(pgd, md->phys_addr, va, md->num_pages, pf))
 214                 pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
 215                            md->phys_addr, va);
 216 }
 217
 218 void __init efi_map_region(efi_memory_desc_t *md)
 219 {
 220         unsigned long size = md->num_pages << PAGE_SHIFT;
 221         u64 pa = md->phys_addr;
 222
 223         if (efi_enabled(EFI_OLD_MEMMAP))
 224                 return old_map_region(md);
 225
 226         /*
 227          * Make sure the 1:1 mappings are present as a catch-all for b0rked
 228          * firmware which doesn't update all internal pointers after switching
 229          * to virtual mode and would otherwise crap on us.
 230          */
 231         __map_region(md, md->phys_addr);
 232
 233         /*
 234          * Enforce the 1:1 mapping as the default virtual address when
 235          * booting in EFI mixed mode, because even though we may be
 236          * running a 64-bit kernel, the firmware may only be 32-bit.
 237          */
 238         if (!efi_is_native () && IS_ENABLED(CONFIG_EFI_MIXED)) {
 239                 md->virt_addr = md->phys_addr;
 240                 return;
 241         }
 242
 243         efi_va -= size;
 244
 245         /* Is PA 2M-aligned? */
 246         if (!(pa & (PMD_SIZE - 1))) {
 247                 efi_va &= PMD_MASK;
 248         } else {
 249                 u64 pa_offset = pa & (PMD_SIZE - 1);
 250                 u64 prev_va = efi_va;
 251
 252                 /* get us the same offset within this 2M page */
 253                 efi_va = (efi_va & PMD_MASK) + pa_offset;
 254
 255                 if (efi_va > prev_va)
 256                         efi_va -= PMD_SIZE;
 257         }
 258
 259         if (efi_va < EFI_VA_END) {
 260                 pr_warn(FW_WARN "VA address range overflow!\n");
 261                 return;
 262         }
 263
 264         /* Do the VA map */
 265         __map_region(md, efi_va);
 266         md->virt_addr = efi_va;
 267 }
 268
 269 /*
 270  * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
 271  * md->virt_addr is the original virtual address which had been mapped in kexec
 272  * 1st kernel.
 273  */
 274 void __init efi_map_region_fixed(efi_memory_desc_t *md)
 275 {
 276         __map_region(md, md->virt_addr);
 277 }
 278
 279 void __iomem *__init efi_ioremap(unsigned long phys_addr, unsigned long size,
 280                                  u32 type, u64 attribute)
 281 {
 282         unsigned long last_map_pfn;
 283
 284         if (type == EFI_MEMORY_MAPPED_IO)
 285                 return ioremap(phys_addr, size);
 286
 287         last_map_pfn = init_memory_mapping(phys_addr, phys_addr + size);
 288         if ((last_map_pfn << PAGE_SHIFT) < phys_addr + size) {
 289                 unsigned long top = last_map_pfn << PAGE_SHIFT;
 290                 efi_ioremap(top, size - (top - phys_addr), type, attribute);
 291         }
 292
 293         if (!(attribute & EFI_MEMORY_WB))
 294                 efi_memory_uc((u64)(unsigned long)__va(phys_addr), size);
 295
 296         return (void __iomem *)__va(phys_addr);
 297 }
 298
 299 void __init parse_efi_setup(u64 phys_addr, u32 data_len)
 300 {
 301         efi_setup = phys_addr + sizeof(struct setup_data);
 302 }
 303
 304 void __init efi_runtime_mkexec(void)
 305 {
 306         if (!efi_enabled(EFI_OLD_MEMMAP))
 307                 return;
 308
 309         if (__supported_pte_mask & _PAGE_NX)
 310                 runtime_code_page_mkexec();
 311 }
 312
 313 void __init efi_dump_pagetable(void)
 314 {
 315 #ifdef CONFIG_EFI_PGT_DUMP
 316         pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
 317
 318         ptdump_walk_pgd_level(NULL, pgd);
 319 #endif
 320 }
 321
 322 #ifdef CONFIG_EFI_MIXED
 323 extern efi_status_t efi64_thunk(u32, ...);
 324
 325 #define runtime_service32(func)                                          \
 326 ({                                                                       \
 327         u32 table = (u32)(unsigned long)efi.systab;                      \
 328         u32 *rt, *___f;                                                  \
 329                                                                          \
 330         rt = (u32 *)(table + offsetof(efi_system_table_32_t, runtime));  \
 331         ___f = (u32 *)(*rt + offsetof(efi_runtime_services_32_t, func)); \
 332         *___f;                                                           \
 333 })
 334
 335 /*
 336  * Switch to the EFI page tables early so that we can access the 1:1
 337  * runtime services mappings which are not mapped in any other page
 338  * tables. This function must be called before runtime_service32().
 339  *
 340  * Also, disable interrupts because the IDT points to 64-bit handlers,
 341  * which aren't going to function correctly when we switch to 32-bit.
 342  */
 343 #define efi_thunk(f, ...)                                               \
 344 ({                                                                      \
 345         efi_status_t __s;                                               \
 346         unsigned long flags;                                            \
 347         u32 func;                                                       \
 348                                                                         \
 349         efi_sync_low_kernel_mappings();                                 \
 350         local_irq_save(flags);                                          \
 351         firmware_restrict_branch_speculation_start();                   \
 352                                                                         \
 353         efi_scratch.prev_cr3 = read_cr3();                              \
 354         write_cr3((unsigned long)efi_scratch.efi_pgt);                  \
 355         __flush_tlb_all();                                              \
 356                                                                         \
 357         func = runtime_service32(f);                                    \
 358         __s = efi64_thunk(func, __VA_ARGS__);                   \
 359                                                                         \
 360         write_cr3(efi_scratch.prev_cr3);                                \
 361         __flush_tlb_all();                                              \
 362         firmware_restrict_branch_speculation_end();                     \
 363         local_irq_restore(flags);                                       \
 364                                                                         \
 365         __s;                                                            \
 366 })
 367
 368 efi_status_t efi_thunk_set_virtual_address_map(
 369         void *phys_set_virtual_address_map,
 370         unsigned long memory_map_size,
 371         unsigned long descriptor_size,
 372         u32 descriptor_version,
 373         efi_memory_desc_t *virtual_map)
 374 {
 375         efi_status_t status;
 376         unsigned long flags;
 377         u32 func;
 378
 379         efi_sync_low_kernel_mappings();
 380         local_irq_save(flags);
 381
 382         efi_scratch.prev_cr3 = read_cr3();
 383         write_cr3((unsigned long)efi_scratch.efi_pgt);
 384         __flush_tlb_all();
 385
 386         func = (u32)(unsigned long)phys_set_virtual_address_map;
 387         status = efi64_thunk(func, memory_map_size, descriptor_size,
 388                              descriptor_version, virtual_map);
 389
 390         write_cr3(efi_scratch.prev_cr3);
 391         __flush_tlb_all();
 392         local_irq_restore(flags);
 393
 394         return status;
 395 }
 396
 397 static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
 398 {
 399         efi_status_t status;
 400         u32 phys_tm, phys_tc;
 401
 402         spin_lock(&rtc_lock);
 403
 404         phys_tm = virt_to_phys(tm);
 405         phys_tc = virt_to_phys(tc);
 406
 407         status = efi_thunk(get_time, phys_tm, phys_tc);
 408
 409         spin_unlock(&rtc_lock);
 410
 411         return status;
 412 }
 413
 414 static efi_status_t efi_thunk_set_time(efi_time_t *tm)
 415 {
 416         efi_status_t status;
 417         u32 phys_tm;
 418
 419         spin_lock(&rtc_lock);
 420
 421         phys_tm = virt_to_phys(tm);
 422
 423         status = efi_thunk(set_time, phys_tm);
 424
 425         spin_unlock(&rtc_lock);
 426
 427         return status;
 428 }
 429
 430 static efi_status_t
 431 efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
 432                           efi_time_t *tm)
 433 {
 434         efi_status_t status;
 435         u32 phys_enabled, phys_pending, phys_tm;
 436
 437         spin_lock(&rtc_lock);
 438
 439         phys_enabled = virt_to_phys(enabled);
 440         phys_pending = virt_to_phys(pending);
 441         phys_tm = virt_to_phys(tm);
 442
 443         status = efi_thunk(get_wakeup_time, phys_enabled,
 444                              phys_pending, phys_tm);
 445
 446         spin_unlock(&rtc_lock);
 447
 448         return status;
 449 }
 450
 451 static efi_status_t
 452 efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
 453 {
 454         efi_status_t status;
 455         u32 phys_tm;
 456
 457         spin_lock(&rtc_lock);
 458
 459         phys_tm = virt_to_phys(tm);
 460
 461         status = efi_thunk(set_wakeup_time, enabled, phys_tm);
 462
 463         spin_unlock(&rtc_lock);
 464
 465         return status;
 466 }
 467
 468
 469 static efi_status_t
 470 efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
 471                        u32 *attr, unsigned long *data_size, void *data)
 472 {
 473         efi_status_t status;
 474         u32 phys_name, phys_vendor, phys_attr;
 475         u32 phys_data_size, phys_data;
 476
 477         phys_data_size = virt_to_phys(data_size);
 478         phys_vendor = virt_to_phys(vendor);
 479         phys_name = virt_to_phys(name);
 480         phys_attr = virt_to_phys(attr);
 481         phys_data = virt_to_phys(data);
 482
 483         status = efi_thunk(get_variable, phys_name, phys_vendor,
 484                            phys_attr, phys_data_size, phys_data);
 485
 486         return status;
 487 }
 488
 489 static efi_status_t
 490 efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
 491                        u32 attr, unsigned long data_size, void *data)
 492 {
 493         u32 phys_name, phys_vendor, phys_data;
 494         efi_status_t status;
 495
 496         phys_name = virt_to_phys(name);
 497         phys_vendor = virt_to_phys(vendor);
 498         phys_data = virt_to_phys(data);
 499
 500         /* If data_size is > sizeof(u32) we've got problems */
 501         status = efi_thunk(set_variable, phys_name, phys_vendor,
 502                            attr, data_size, phys_data);
 503
 504         return status;
 505 }
 506
 507 static efi_status_t
 508 efi_thunk_get_next_variable(unsigned long *name_size,
 509                             efi_char16_t *name,
 510                             efi_guid_t *vendor)
 511 {
 512         efi_status_t status;
 513         u32 phys_name_size, phys_name, phys_vendor;
 514
 515         phys_name_size = virt_to_phys(name_size);
 516         phys_vendor = virt_to_phys(vendor);
 517         phys_name = virt_to_phys(name);
 518
 519         status = efi_thunk(get_next_variable, phys_name_size,
 520                            phys_name, phys_vendor);
 521
 522         return status;
 523 }
 524
 525 static efi_status_t
 526 efi_thunk_get_next_high_mono_count(u32 *count)
 527 {
 528         efi_status_t status;
 529         u32 phys_count;
 530
 531         phys_count = virt_to_phys(count);
 532         status = efi_thunk(get_next_high_mono_count, phys_count);
 533
 534         return status;
 535 }
 536
 537 static void
 538 efi_thunk_reset_system(int reset_type, efi_status_t status,
 539                        unsigned long data_size, efi_char16_t *data)
 540 {
 541         u32 phys_data;
 542
 543         phys_data = virt_to_phys(data);
 544
 545         efi_thunk(reset_system, reset_type, status, data_size, phys_data);
 546 }
 547
 548 static efi_status_t
 549 efi_thunk_update_capsule(efi_capsule_header_t **capsules,
 550                          unsigned long count, unsigned long sg_list)
 551 {
 552         /*
 553          * To properly support this function we would need to repackage
 554          * 'capsules' because the firmware doesn't understand 64-bit
 555          * pointers.
 556          */
 557         return EFI_UNSUPPORTED;
 558 }
 559
 560 static efi_status_t
 561 efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
 562                               u64 *remaining_space,
 563                               u64 *max_variable_size)
 564 {
 565         efi_status_t status;
 566         u32 phys_storage, phys_remaining, phys_max;
 567
 568         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 569                 return EFI_UNSUPPORTED;
 570
 571         phys_storage = virt_to_phys(storage_space);
 572         phys_remaining = virt_to_phys(remaining_space);
 573         phys_max = virt_to_phys(max_variable_size);
 574
 575         status = efi_thunk(query_variable_info, attr, phys_storage,
 576                            phys_remaining, phys_max);
 577
 578         return status;
 579 }
 580
 581 static efi_status_t
 582 efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
 583                              unsigned long count, u64 *max_size,
 584                              int *reset_type)
 585 {
 586         /*
 587          * To properly support this function we would need to repackage
 588          * 'capsules' because the firmware doesn't understand 64-bit
 589          * pointers.
 590          */
 591         return EFI_UNSUPPORTED;
 592 }
 593
 594 void efi_thunk_runtime_setup(void)
 595 {
 596         efi.get_time = efi_thunk_get_time;
 597         efi.set_time = efi_thunk_set_time;
 598         efi.get_wakeup_time = efi_thunk_get_wakeup_time;
 599         efi.set_wakeup_time = efi_thunk_set_wakeup_time;
 600         efi.get_variable = efi_thunk_get_variable;
 601         efi.get_next_variable = efi_thunk_get_next_variable;
 602         efi.set_variable = efi_thunk_set_variable;
 603         efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
 604         efi.reset_system = efi_thunk_reset_system;
 605         efi.query_variable_info = efi_thunk_query_variable_info;
 606         efi.update_capsule = efi_thunk_update_capsule;
 607         efi.query_capsule_caps = efi_thunk_query_capsule_caps;
 608 }
 609 #endif /* CONFIG_EFI_MIXED */