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