1 /* 2 * Extensible Firmware Interface 3 * 4 * Based on Extensible Firmware Interface Specification version 0.9 April 30, 1999 5 * 6 * Copyright (C) 1999 VA Linux Systems 7 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com> 8 * Copyright (C) 1999-2003 Hewlett-Packard Co. 9 * David Mosberger-Tang <davidm@hpl.hp.com> 10 * Stephane Eranian <eranian@hpl.hp.com> 11 * (c) Copyright 2006 Hewlett-Packard Development Company, L.P. 12 * Bjorn Helgaas <bjorn.helgaas@hp.com> 13 * 14 * All EFI Runtime Services are not implemented yet as EFI only 15 * supports physical mode addressing on SoftSDV. This is to be fixed 16 * in a future version. --drummond 1999-07-20 17 * 18 * Implemented EFI runtime services and virtual mode calls. --davidm 19 * 20 * Goutham Rao: <goutham.rao@intel.com> 21 * Skip non-WB memory and ignore empty memory ranges. 22 */ 23 #include <linux/module.h> 24 #include <linux/bootmem.h> 25 #include <linux/kernel.h> 26 #include <linux/init.h> 27 #include <linux/types.h> 28 #include <linux/time.h> 29 #include <linux/efi.h> 30 #include <linux/kexec.h> 31 #include <linux/mm.h> 32 33 #include <asm/io.h> 34 #include <asm/kregs.h> 35 #include <asm/meminit.h> 36 #include <asm/pgtable.h> 37 #include <asm/processor.h> 38 #include <asm/mca.h> 39 40 #define EFI_DEBUG 0 41 42 extern efi_status_t efi_call_phys (void *, ...); 43 44 struct efi efi; 45 EXPORT_SYMBOL(efi); 46 static efi_runtime_services_t *runtime; 47 static unsigned long mem_limit = ~0UL, max_addr = ~0UL, min_addr = 0UL; 48 49 #define efi_call_virt(f, args...) (*(f))(args) 50 51 #define STUB_GET_TIME(prefix, adjust_arg) \ 52 static efi_status_t \ 53 prefix##_get_time (efi_time_t *tm, efi_time_cap_t *tc) \ 54 { \ 55 struct ia64_fpreg fr[6]; \ 56 efi_time_cap_t *atc = NULL; \ 57 efi_status_t ret; \ 58 \ 59 if (tc) \ 60 atc = adjust_arg(tc); \ 61 ia64_save_scratch_fpregs(fr); \ 62 ret = efi_call_##prefix((efi_get_time_t *) __va(runtime->get_time), adjust_arg(tm), atc); \ 63 ia64_load_scratch_fpregs(fr); \ 64 return ret; \ 65 } 66 67 #define STUB_SET_TIME(prefix, adjust_arg) \ 68 static efi_status_t \ 69 prefix##_set_time (efi_time_t *tm) \ 70 { \ 71 struct ia64_fpreg fr[6]; \ 72 efi_status_t ret; \ 73 \ 74 ia64_save_scratch_fpregs(fr); \ 75 ret = efi_call_##prefix((efi_set_time_t *) __va(runtime->set_time), adjust_arg(tm)); \ 76 ia64_load_scratch_fpregs(fr); \ 77 return ret; \ 78 } 79 80 #define STUB_GET_WAKEUP_TIME(prefix, adjust_arg) \ 81 static efi_status_t \ 82 prefix##_get_wakeup_time (efi_bool_t *enabled, efi_bool_t *pending, efi_time_t *tm) \ 83 { \ 84 struct ia64_fpreg fr[6]; \ 85 efi_status_t ret; \ 86 \ 87 ia64_save_scratch_fpregs(fr); \ 88 ret = efi_call_##prefix((efi_get_wakeup_time_t *) __va(runtime->get_wakeup_time), \ 89 adjust_arg(enabled), adjust_arg(pending), adjust_arg(tm)); \ 90 ia64_load_scratch_fpregs(fr); \ 91 return ret; \ 92 } 93 94 #define STUB_SET_WAKEUP_TIME(prefix, adjust_arg) \ 95 static efi_status_t \ 96 prefix##_set_wakeup_time (efi_bool_t enabled, efi_time_t *tm) \ 97 { \ 98 struct ia64_fpreg fr[6]; \ 99 efi_time_t *atm = NULL; \ 100 efi_status_t ret; \ 101 \ 102 if (tm) \ 103 atm = adjust_arg(tm); \ 104 ia64_save_scratch_fpregs(fr); \ 105 ret = efi_call_##prefix((efi_set_wakeup_time_t *) __va(runtime->set_wakeup_time), \ 106 enabled, atm); \ 107 ia64_load_scratch_fpregs(fr); \ 108 return ret; \ 109 } 110 111 #define STUB_GET_VARIABLE(prefix, adjust_arg) \ 112 static efi_status_t \ 113 prefix##_get_variable (efi_char16_t *name, efi_guid_t *vendor, u32 *attr, \ 114 unsigned long *data_size, void *data) \ 115 { \ 116 struct ia64_fpreg fr[6]; \ 117 u32 *aattr = NULL; \ 118 efi_status_t ret; \ 119 \ 120 if (attr) \ 121 aattr = adjust_arg(attr); \ 122 ia64_save_scratch_fpregs(fr); \ 123 ret = efi_call_##prefix((efi_get_variable_t *) __va(runtime->get_variable), \ 124 adjust_arg(name), adjust_arg(vendor), aattr, \ 125 adjust_arg(data_size), adjust_arg(data)); \ 126 ia64_load_scratch_fpregs(fr); \ 127 return ret; \ 128 } 129 130 #define STUB_GET_NEXT_VARIABLE(prefix, adjust_arg) \ 131 static efi_status_t \ 132 prefix##_get_next_variable (unsigned long *name_size, efi_char16_t *name, efi_guid_t *vendor) \ 133 { \ 134 struct ia64_fpreg fr[6]; \ 135 efi_status_t ret; \ 136 \ 137 ia64_save_scratch_fpregs(fr); \ 138 ret = efi_call_##prefix((efi_get_next_variable_t *) __va(runtime->get_next_variable), \ 139 adjust_arg(name_size), adjust_arg(name), adjust_arg(vendor)); \ 140 ia64_load_scratch_fpregs(fr); \ 141 return ret; \ 142 } 143 144 #define STUB_SET_VARIABLE(prefix, adjust_arg) \ 145 static efi_status_t \ 146 prefix##_set_variable (efi_char16_t *name, efi_guid_t *vendor, unsigned long attr, \ 147 unsigned long data_size, void *data) \ 148 { \ 149 struct ia64_fpreg fr[6]; \ 150 efi_status_t ret; \ 151 \ 152 ia64_save_scratch_fpregs(fr); \ 153 ret = efi_call_##prefix((efi_set_variable_t *) __va(runtime->set_variable), \ 154 adjust_arg(name), adjust_arg(vendor), attr, data_size, \ 155 adjust_arg(data)); \ 156 ia64_load_scratch_fpregs(fr); \ 157 return ret; \ 158 } 159 160 #define STUB_GET_NEXT_HIGH_MONO_COUNT(prefix, adjust_arg) \ 161 static efi_status_t \ 162 prefix##_get_next_high_mono_count (u32 *count) \ 163 { \ 164 struct ia64_fpreg fr[6]; \ 165 efi_status_t ret; \ 166 \ 167 ia64_save_scratch_fpregs(fr); \ 168 ret = efi_call_##prefix((efi_get_next_high_mono_count_t *) \ 169 __va(runtime->get_next_high_mono_count), adjust_arg(count)); \ 170 ia64_load_scratch_fpregs(fr); \ 171 return ret; \ 172 } 173 174 #define STUB_RESET_SYSTEM(prefix, adjust_arg) \ 175 static void \ 176 prefix##_reset_system (int reset_type, efi_status_t status, \ 177 unsigned long data_size, efi_char16_t *data) \ 178 { \ 179 struct ia64_fpreg fr[6]; \ 180 efi_char16_t *adata = NULL; \ 181 \ 182 if (data) \ 183 adata = adjust_arg(data); \ 184 \ 185 ia64_save_scratch_fpregs(fr); \ 186 efi_call_##prefix((efi_reset_system_t *) __va(runtime->reset_system), \ 187 reset_type, status, data_size, adata); \ 188 /* should not return, but just in case... */ \ 189 ia64_load_scratch_fpregs(fr); \ 190 } 191 192 #define phys_ptr(arg) ((__typeof__(arg)) ia64_tpa(arg)) 193 194 STUB_GET_TIME(phys, phys_ptr) 195 STUB_SET_TIME(phys, phys_ptr) 196 STUB_GET_WAKEUP_TIME(phys, phys_ptr) 197 STUB_SET_WAKEUP_TIME(phys, phys_ptr) 198 STUB_GET_VARIABLE(phys, phys_ptr) 199 STUB_GET_NEXT_VARIABLE(phys, phys_ptr) 200 STUB_SET_VARIABLE(phys, phys_ptr) 201 STUB_GET_NEXT_HIGH_MONO_COUNT(phys, phys_ptr) 202 STUB_RESET_SYSTEM(phys, phys_ptr) 203 204 #define id(arg) arg 205 206 STUB_GET_TIME(virt, id) 207 STUB_SET_TIME(virt, id) 208 STUB_GET_WAKEUP_TIME(virt, id) 209 STUB_SET_WAKEUP_TIME(virt, id) 210 STUB_GET_VARIABLE(virt, id) 211 STUB_GET_NEXT_VARIABLE(virt, id) 212 STUB_SET_VARIABLE(virt, id) 213 STUB_GET_NEXT_HIGH_MONO_COUNT(virt, id) 214 STUB_RESET_SYSTEM(virt, id) 215 216 void 217 efi_gettimeofday (struct timespec *ts) 218 { 219 efi_time_t tm; 220 221 if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS) { 222 memset(ts, 0, sizeof(*ts)); 223 return; 224 } 225 226 ts->tv_sec = mktime(tm.year, tm.month, tm.day, tm.hour, tm.minute, tm.second); 227 ts->tv_nsec = tm.nanosecond; 228 } 229 230 static int 231 is_memory_available (efi_memory_desc_t *md) 232 { 233 if (!(md->attribute & EFI_MEMORY_WB)) 234 return 0; 235 236 switch (md->type) { 237 case EFI_LOADER_CODE: 238 case EFI_LOADER_DATA: 239 case EFI_BOOT_SERVICES_CODE: 240 case EFI_BOOT_SERVICES_DATA: 241 case EFI_CONVENTIONAL_MEMORY: 242 return 1; 243 } 244 return 0; 245 } 246 247 typedef struct kern_memdesc { 248 u64 attribute; 249 u64 start; 250 u64 num_pages; 251 } kern_memdesc_t; 252 253 static kern_memdesc_t *kern_memmap; 254 255 #define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT) 256 257 static inline u64 258 kmd_end(kern_memdesc_t *kmd) 259 { 260 return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT)); 261 } 262 263 static inline u64 264 efi_md_end(efi_memory_desc_t *md) 265 { 266 return (md->phys_addr + efi_md_size(md)); 267 } 268 269 static inline int 270 efi_wb(efi_memory_desc_t *md) 271 { 272 return (md->attribute & EFI_MEMORY_WB); 273 } 274 275 static inline int 276 efi_uc(efi_memory_desc_t *md) 277 { 278 return (md->attribute & EFI_MEMORY_UC); 279 } 280 281 static void 282 walk (efi_freemem_callback_t callback, void *arg, u64 attr) 283 { 284 kern_memdesc_t *k; 285 u64 start, end, voff; 286 287 voff = (attr == EFI_MEMORY_WB) ? PAGE_OFFSET : __IA64_UNCACHED_OFFSET; 288 for (k = kern_memmap; k->start != ~0UL; k++) { 289 if (k->attribute != attr) 290 continue; 291 start = PAGE_ALIGN(k->start); 292 end = (k->start + (k->num_pages << EFI_PAGE_SHIFT)) & PAGE_MASK; 293 if (start < end) 294 if ((*callback)(start + voff, end + voff, arg) < 0) 295 return; 296 } 297 } 298 299 /* 300 * Walks the EFI memory map and calls CALLBACK once for each EFI memory descriptor that 301 * has memory that is available for OS use. 302 */ 303 void 304 efi_memmap_walk (efi_freemem_callback_t callback, void *arg) 305 { 306 walk(callback, arg, EFI_MEMORY_WB); 307 } 308 309 /* 310 * Walks the EFI memory map and calls CALLBACK once for each EFI memory descriptor that 311 * has memory that is available for uncached allocator. 312 */ 313 void 314 efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg) 315 { 316 walk(callback, arg, EFI_MEMORY_UC); 317 } 318 319 /* 320 * Look for the PAL_CODE region reported by EFI and maps it using an 321 * ITR to enable safe PAL calls in virtual mode. See IA-64 Processor 322 * Abstraction Layer chapter 11 in ADAG 323 */ 324 325 void * 326 efi_get_pal_addr (void) 327 { 328 void *efi_map_start, *efi_map_end, *p; 329 efi_memory_desc_t *md; 330 u64 efi_desc_size; 331 int pal_code_count = 0; 332 u64 vaddr, mask; 333 334 efi_map_start = __va(ia64_boot_param->efi_memmap); 335 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 336 efi_desc_size = ia64_boot_param->efi_memdesc_size; 337 338 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 339 md = p; 340 if (md->type != EFI_PAL_CODE) 341 continue; 342 343 if (++pal_code_count > 1) { 344 printk(KERN_ERR "Too many EFI Pal Code memory ranges, dropped @ %lx\n", 345 md->phys_addr); 346 continue; 347 } 348 /* 349 * The only ITLB entry in region 7 that is used is the one installed by 350 * __start(). That entry covers a 64MB range. 351 */ 352 mask = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1); 353 vaddr = PAGE_OFFSET + md->phys_addr; 354 355 /* 356 * We must check that the PAL mapping won't overlap with the kernel 357 * mapping. 358 * 359 * PAL code is guaranteed to be aligned on a power of 2 between 4k and 360 * 256KB and that only one ITR is needed to map it. This implies that the 361 * PAL code is always aligned on its size, i.e., the closest matching page 362 * size supported by the TLB. Therefore PAL code is guaranteed never to 363 * cross a 64MB unless it is bigger than 64MB (very unlikely!). So for 364 * now the following test is enough to determine whether or not we need a 365 * dedicated ITR for the PAL code. 366 */ 367 if ((vaddr & mask) == (KERNEL_START & mask)) { 368 printk(KERN_INFO "%s: no need to install ITR for PAL code\n", 369 __FUNCTION__); 370 continue; 371 } 372 373 if (efi_md_size(md) > IA64_GRANULE_SIZE) 374 panic("Woah! PAL code size bigger than a granule!"); 375 376 #if EFI_DEBUG 377 mask = ~((1 << IA64_GRANULE_SHIFT) - 1); 378 379 printk(KERN_INFO "CPU %d: mapping PAL code [0x%lx-0x%lx) into [0x%lx-0x%lx)\n", 380 smp_processor_id(), md->phys_addr, 381 md->phys_addr + efi_md_size(md), 382 vaddr & mask, (vaddr & mask) + IA64_GRANULE_SIZE); 383 #endif 384 return __va(md->phys_addr); 385 } 386 printk(KERN_WARNING "%s: no PAL-code memory-descriptor found\n", 387 __FUNCTION__); 388 return NULL; 389 } 390 391 void 392 efi_map_pal_code (void) 393 { 394 void *pal_vaddr = efi_get_pal_addr (); 395 u64 psr; 396 397 if (!pal_vaddr) 398 return; 399 400 /* 401 * Cannot write to CRx with PSR.ic=1 402 */ 403 psr = ia64_clear_ic(); 404 ia64_itr(0x1, IA64_TR_PALCODE, GRANULEROUNDDOWN((unsigned long) pal_vaddr), 405 pte_val(pfn_pte(__pa(pal_vaddr) >> PAGE_SHIFT, PAGE_KERNEL)), 406 IA64_GRANULE_SHIFT); 407 ia64_set_psr(psr); /* restore psr */ 408 ia64_srlz_i(); 409 } 410 411 void __init 412 efi_init (void) 413 { 414 void *efi_map_start, *efi_map_end; 415 efi_config_table_t *config_tables; 416 efi_char16_t *c16; 417 u64 efi_desc_size; 418 char *cp, vendor[100] = "unknown"; 419 int i; 420 421 /* it's too early to be able to use the standard kernel command line support... */ 422 for (cp = boot_command_line; *cp; ) { 423 if (memcmp(cp, "mem=", 4) == 0) { 424 mem_limit = memparse(cp + 4, &cp); 425 } else if (memcmp(cp, "max_addr=", 9) == 0) { 426 max_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp)); 427 } else if (memcmp(cp, "min_addr=", 9) == 0) { 428 min_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp)); 429 } else { 430 while (*cp != ' ' && *cp) 431 ++cp; 432 while (*cp == ' ') 433 ++cp; 434 } 435 } 436 if (min_addr != 0UL) 437 printk(KERN_INFO "Ignoring memory below %luMB\n", min_addr >> 20); 438 if (max_addr != ~0UL) 439 printk(KERN_INFO "Ignoring memory above %luMB\n", max_addr >> 20); 440 441 efi.systab = __va(ia64_boot_param->efi_systab); 442 443 /* 444 * Verify the EFI Table 445 */ 446 if (efi.systab == NULL) 447 panic("Woah! Can't find EFI system table.\n"); 448 if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) 449 panic("Woah! EFI system table signature incorrect\n"); 450 if ((efi.systab->hdr.revision >> 16) == 0) 451 printk(KERN_WARNING "Warning: EFI system table version " 452 "%d.%02d, expected 1.00 or greater\n", 453 efi.systab->hdr.revision >> 16, 454 efi.systab->hdr.revision & 0xffff); 455 456 config_tables = __va(efi.systab->tables); 457 458 /* Show what we know for posterity */ 459 c16 = __va(efi.systab->fw_vendor); 460 if (c16) { 461 for (i = 0;i < (int) sizeof(vendor) - 1 && *c16; ++i) 462 vendor[i] = *c16++; 463 vendor[i] = '\0'; 464 } 465 466 printk(KERN_INFO "EFI v%u.%.02u by %s:", 467 efi.systab->hdr.revision >> 16, efi.systab->hdr.revision & 0xffff, vendor); 468 469 efi.mps = EFI_INVALID_TABLE_ADDR; 470 efi.acpi = EFI_INVALID_TABLE_ADDR; 471 efi.acpi20 = EFI_INVALID_TABLE_ADDR; 472 efi.smbios = EFI_INVALID_TABLE_ADDR; 473 efi.sal_systab = EFI_INVALID_TABLE_ADDR; 474 efi.boot_info = EFI_INVALID_TABLE_ADDR; 475 efi.hcdp = EFI_INVALID_TABLE_ADDR; 476 efi.uga = EFI_INVALID_TABLE_ADDR; 477 478 for (i = 0; i < (int) efi.systab->nr_tables; i++) { 479 if (efi_guidcmp(config_tables[i].guid, MPS_TABLE_GUID) == 0) { 480 efi.mps = config_tables[i].table; 481 printk(" MPS=0x%lx", config_tables[i].table); 482 } else if (efi_guidcmp(config_tables[i].guid, ACPI_20_TABLE_GUID) == 0) { 483 efi.acpi20 = config_tables[i].table; 484 printk(" ACPI 2.0=0x%lx", config_tables[i].table); 485 } else if (efi_guidcmp(config_tables[i].guid, ACPI_TABLE_GUID) == 0) { 486 efi.acpi = config_tables[i].table; 487 printk(" ACPI=0x%lx", config_tables[i].table); 488 } else if (efi_guidcmp(config_tables[i].guid, SMBIOS_TABLE_GUID) == 0) { 489 efi.smbios = config_tables[i].table; 490 printk(" SMBIOS=0x%lx", config_tables[i].table); 491 } else if (efi_guidcmp(config_tables[i].guid, SAL_SYSTEM_TABLE_GUID) == 0) { 492 efi.sal_systab = config_tables[i].table; 493 printk(" SALsystab=0x%lx", config_tables[i].table); 494 } else if (efi_guidcmp(config_tables[i].guid, HCDP_TABLE_GUID) == 0) { 495 efi.hcdp = config_tables[i].table; 496 printk(" HCDP=0x%lx", config_tables[i].table); 497 } 498 } 499 printk("\n"); 500 501 runtime = __va(efi.systab->runtime); 502 efi.get_time = phys_get_time; 503 efi.set_time = phys_set_time; 504 efi.get_wakeup_time = phys_get_wakeup_time; 505 efi.set_wakeup_time = phys_set_wakeup_time; 506 efi.get_variable = phys_get_variable; 507 efi.get_next_variable = phys_get_next_variable; 508 efi.set_variable = phys_set_variable; 509 efi.get_next_high_mono_count = phys_get_next_high_mono_count; 510 efi.reset_system = phys_reset_system; 511 512 efi_map_start = __va(ia64_boot_param->efi_memmap); 513 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 514 efi_desc_size = ia64_boot_param->efi_memdesc_size; 515 516 #if EFI_DEBUG 517 /* print EFI memory map: */ 518 { 519 efi_memory_desc_t *md; 520 void *p; 521 522 for (i = 0, p = efi_map_start; p < efi_map_end; ++i, p += efi_desc_size) { 523 md = p; 524 printk("mem%02u: type=%u, attr=0x%lx, range=[0x%016lx-0x%016lx) (%luMB)\n", 525 i, md->type, md->attribute, md->phys_addr, 526 md->phys_addr + efi_md_size(md), 527 md->num_pages >> (20 - EFI_PAGE_SHIFT)); 528 } 529 } 530 #endif 531 532 efi_map_pal_code(); 533 efi_enter_virtual_mode(); 534 } 535 536 void 537 efi_enter_virtual_mode (void) 538 { 539 void *efi_map_start, *efi_map_end, *p; 540 efi_memory_desc_t *md; 541 efi_status_t status; 542 u64 efi_desc_size; 543 544 efi_map_start = __va(ia64_boot_param->efi_memmap); 545 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 546 efi_desc_size = ia64_boot_param->efi_memdesc_size; 547 548 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 549 md = p; 550 if (md->attribute & EFI_MEMORY_RUNTIME) { 551 /* 552 * Some descriptors have multiple bits set, so the order of 553 * the tests is relevant. 554 */ 555 if (md->attribute & EFI_MEMORY_WB) { 556 md->virt_addr = (u64) __va(md->phys_addr); 557 } else if (md->attribute & EFI_MEMORY_UC) { 558 md->virt_addr = (u64) ioremap(md->phys_addr, 0); 559 } else if (md->attribute & EFI_MEMORY_WC) { 560 #if 0 561 md->virt_addr = ia64_remap(md->phys_addr, (_PAGE_A | _PAGE_P 562 | _PAGE_D 563 | _PAGE_MA_WC 564 | _PAGE_PL_0 565 | _PAGE_AR_RW)); 566 #else 567 printk(KERN_INFO "EFI_MEMORY_WC mapping\n"); 568 md->virt_addr = (u64) ioremap(md->phys_addr, 0); 569 #endif 570 } else if (md->attribute & EFI_MEMORY_WT) { 571 #if 0 572 md->virt_addr = ia64_remap(md->phys_addr, (_PAGE_A | _PAGE_P 573 | _PAGE_D | _PAGE_MA_WT 574 | _PAGE_PL_0 575 | _PAGE_AR_RW)); 576 #else 577 printk(KERN_INFO "EFI_MEMORY_WT mapping\n"); 578 md->virt_addr = (u64) ioremap(md->phys_addr, 0); 579 #endif 580 } 581 } 582 } 583 584 status = efi_call_phys(__va(runtime->set_virtual_address_map), 585 ia64_boot_param->efi_memmap_size, 586 efi_desc_size, ia64_boot_param->efi_memdesc_version, 587 ia64_boot_param->efi_memmap); 588 if (status != EFI_SUCCESS) { 589 printk(KERN_WARNING "warning: unable to switch EFI into virtual mode " 590 "(status=%lu)\n", status); 591 return; 592 } 593 594 /* 595 * Now that EFI is in virtual mode, we call the EFI functions more efficiently: 596 */ 597 efi.get_time = virt_get_time; 598 efi.set_time = virt_set_time; 599 efi.get_wakeup_time = virt_get_wakeup_time; 600 efi.set_wakeup_time = virt_set_wakeup_time; 601 efi.get_variable = virt_get_variable; 602 efi.get_next_variable = virt_get_next_variable; 603 efi.set_variable = virt_set_variable; 604 efi.get_next_high_mono_count = virt_get_next_high_mono_count; 605 efi.reset_system = virt_reset_system; 606 } 607 608 /* 609 * Walk the EFI memory map looking for the I/O port range. There can only be one entry of 610 * this type, other I/O port ranges should be described via ACPI. 611 */ 612 u64 613 efi_get_iobase (void) 614 { 615 void *efi_map_start, *efi_map_end, *p; 616 efi_memory_desc_t *md; 617 u64 efi_desc_size; 618 619 efi_map_start = __va(ia64_boot_param->efi_memmap); 620 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 621 efi_desc_size = ia64_boot_param->efi_memdesc_size; 622 623 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 624 md = p; 625 if (md->type == EFI_MEMORY_MAPPED_IO_PORT_SPACE) { 626 if (md->attribute & EFI_MEMORY_UC) 627 return md->phys_addr; 628 } 629 } 630 return 0; 631 } 632 633 static struct kern_memdesc * 634 kern_memory_descriptor (unsigned long phys_addr) 635 { 636 struct kern_memdesc *md; 637 638 for (md = kern_memmap; md->start != ~0UL; md++) { 639 if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT)) 640 return md; 641 } 642 return NULL; 643 } 644 645 static efi_memory_desc_t * 646 efi_memory_descriptor (unsigned long phys_addr) 647 { 648 void *efi_map_start, *efi_map_end, *p; 649 efi_memory_desc_t *md; 650 u64 efi_desc_size; 651 652 efi_map_start = __va(ia64_boot_param->efi_memmap); 653 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 654 efi_desc_size = ia64_boot_param->efi_memdesc_size; 655 656 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 657 md = p; 658 659 if (phys_addr - md->phys_addr < efi_md_size(md)) 660 return md; 661 } 662 return NULL; 663 } 664 665 static int 666 efi_memmap_intersects (unsigned long phys_addr, unsigned long size) 667 { 668 void *efi_map_start, *efi_map_end, *p; 669 efi_memory_desc_t *md; 670 u64 efi_desc_size; 671 unsigned long end; 672 673 efi_map_start = __va(ia64_boot_param->efi_memmap); 674 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 675 efi_desc_size = ia64_boot_param->efi_memdesc_size; 676 677 end = phys_addr + size; 678 679 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 680 md = p; 681 682 if (md->phys_addr < end && efi_md_end(md) > phys_addr) 683 return 1; 684 } 685 return 0; 686 } 687 688 u32 689 efi_mem_type (unsigned long phys_addr) 690 { 691 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr); 692 693 if (md) 694 return md->type; 695 return 0; 696 } 697 698 u64 699 efi_mem_attributes (unsigned long phys_addr) 700 { 701 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr); 702 703 if (md) 704 return md->attribute; 705 return 0; 706 } 707 EXPORT_SYMBOL(efi_mem_attributes); 708 709 u64 710 efi_mem_attribute (unsigned long phys_addr, unsigned long size) 711 { 712 unsigned long end = phys_addr + size; 713 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr); 714 u64 attr; 715 716 if (!md) 717 return 0; 718 719 /* 720 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells 721 * the kernel that firmware needs this region mapped. 722 */ 723 attr = md->attribute & ~EFI_MEMORY_RUNTIME; 724 do { 725 unsigned long md_end = efi_md_end(md); 726 727 if (end <= md_end) 728 return attr; 729 730 md = efi_memory_descriptor(md_end); 731 if (!md || (md->attribute & ~EFI_MEMORY_RUNTIME) != attr) 732 return 0; 733 } while (md); 734 return 0; 735 } 736 737 u64 738 kern_mem_attribute (unsigned long phys_addr, unsigned long size) 739 { 740 unsigned long end = phys_addr + size; 741 struct kern_memdesc *md; 742 u64 attr; 743 744 /* 745 * This is a hack for ioremap calls before we set up kern_memmap. 746 * Maybe we should do efi_memmap_init() earlier instead. 747 */ 748 if (!kern_memmap) { 749 attr = efi_mem_attribute(phys_addr, size); 750 if (attr & EFI_MEMORY_WB) 751 return EFI_MEMORY_WB; 752 return 0; 753 } 754 755 md = kern_memory_descriptor(phys_addr); 756 if (!md) 757 return 0; 758 759 attr = md->attribute; 760 do { 761 unsigned long md_end = kmd_end(md); 762 763 if (end <= md_end) 764 return attr; 765 766 md = kern_memory_descriptor(md_end); 767 if (!md || md->attribute != attr) 768 return 0; 769 } while (md); 770 return 0; 771 } 772 EXPORT_SYMBOL(kern_mem_attribute); 773 774 int 775 valid_phys_addr_range (unsigned long phys_addr, unsigned long size) 776 { 777 u64 attr; 778 779 /* 780 * /dev/mem reads and writes use copy_to_user(), which implicitly 781 * uses a granule-sized kernel identity mapping. It's really 782 * only safe to do this for regions in kern_memmap. For more 783 * details, see Documentation/ia64/aliasing.txt. 784 */ 785 attr = kern_mem_attribute(phys_addr, size); 786 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC) 787 return 1; 788 return 0; 789 } 790 791 int 792 valid_mmap_phys_addr_range (unsigned long pfn, unsigned long size) 793 { 794 unsigned long phys_addr = pfn << PAGE_SHIFT; 795 u64 attr; 796 797 attr = efi_mem_attribute(phys_addr, size); 798 799 /* 800 * /dev/mem mmap uses normal user pages, so we don't need the entire 801 * granule, but the entire region we're mapping must support the same 802 * attribute. 803 */ 804 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC) 805 return 1; 806 807 /* 808 * Intel firmware doesn't tell us about all the MMIO regions, so 809 * in general we have to allow mmap requests. But if EFI *does* 810 * tell us about anything inside this region, we should deny it. 811 * The user can always map a smaller region to avoid the overlap. 812 */ 813 if (efi_memmap_intersects(phys_addr, size)) 814 return 0; 815 816 return 1; 817 } 818 819 pgprot_t 820 phys_mem_access_prot(struct file *file, unsigned long pfn, unsigned long size, 821 pgprot_t vma_prot) 822 { 823 unsigned long phys_addr = pfn << PAGE_SHIFT; 824 u64 attr; 825 826 /* 827 * For /dev/mem mmap, we use user mappings, but if the region is 828 * in kern_memmap (and hence may be covered by a kernel mapping), 829 * we must use the same attribute as the kernel mapping. 830 */ 831 attr = kern_mem_attribute(phys_addr, size); 832 if (attr & EFI_MEMORY_WB) 833 return pgprot_cacheable(vma_prot); 834 else if (attr & EFI_MEMORY_UC) 835 return pgprot_noncached(vma_prot); 836 837 /* 838 * Some chipsets don't support UC access to memory. If 839 * WB is supported, we prefer that. 840 */ 841 if (efi_mem_attribute(phys_addr, size) & EFI_MEMORY_WB) 842 return pgprot_cacheable(vma_prot); 843 844 return pgprot_noncached(vma_prot); 845 } 846 847 int __init 848 efi_uart_console_only(void) 849 { 850 efi_status_t status; 851 char *s, name[] = "ConOut"; 852 efi_guid_t guid = EFI_GLOBAL_VARIABLE_GUID; 853 efi_char16_t *utf16, name_utf16[32]; 854 unsigned char data[1024]; 855 unsigned long size = sizeof(data); 856 struct efi_generic_dev_path *hdr, *end_addr; 857 int uart = 0; 858 859 /* Convert to UTF-16 */ 860 utf16 = name_utf16; 861 s = name; 862 while (*s) 863 *utf16++ = *s++ & 0x7f; 864 *utf16 = 0; 865 866 status = efi.get_variable(name_utf16, &guid, NULL, &size, data); 867 if (status != EFI_SUCCESS) { 868 printk(KERN_ERR "No EFI %s variable?\n", name); 869 return 0; 870 } 871 872 hdr = (struct efi_generic_dev_path *) data; 873 end_addr = (struct efi_generic_dev_path *) ((u8 *) data + size); 874 while (hdr < end_addr) { 875 if (hdr->type == EFI_DEV_MSG && 876 hdr->sub_type == EFI_DEV_MSG_UART) 877 uart = 1; 878 else if (hdr->type == EFI_DEV_END_PATH || 879 hdr->type == EFI_DEV_END_PATH2) { 880 if (!uart) 881 return 0; 882 if (hdr->sub_type == EFI_DEV_END_ENTIRE) 883 return 1; 884 uart = 0; 885 } 886 hdr = (struct efi_generic_dev_path *) ((u8 *) hdr + hdr->length); 887 } 888 printk(KERN_ERR "Malformed %s value\n", name); 889 return 0; 890 } 891 892 /* 893 * Look for the first granule aligned memory descriptor memory 894 * that is big enough to hold EFI memory map. Make sure this 895 * descriptor is atleast granule sized so it does not get trimmed 896 */ 897 struct kern_memdesc * 898 find_memmap_space (void) 899 { 900 u64 contig_low=0, contig_high=0; 901 u64 as = 0, ae; 902 void *efi_map_start, *efi_map_end, *p, *q; 903 efi_memory_desc_t *md, *pmd = NULL, *check_md; 904 u64 space_needed, efi_desc_size; 905 unsigned long total_mem = 0; 906 907 efi_map_start = __va(ia64_boot_param->efi_memmap); 908 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 909 efi_desc_size = ia64_boot_param->efi_memdesc_size; 910 911 /* 912 * Worst case: we need 3 kernel descriptors for each efi descriptor 913 * (if every entry has a WB part in the middle, and UC head and tail), 914 * plus one for the end marker. 915 */ 916 space_needed = sizeof(kern_memdesc_t) * 917 (3 * (ia64_boot_param->efi_memmap_size/efi_desc_size) + 1); 918 919 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) { 920 md = p; 921 if (!efi_wb(md)) { 922 continue; 923 } 924 if (pmd == NULL || !efi_wb(pmd) || efi_md_end(pmd) != md->phys_addr) { 925 contig_low = GRANULEROUNDUP(md->phys_addr); 926 contig_high = efi_md_end(md); 927 for (q = p + efi_desc_size; q < efi_map_end; q += efi_desc_size) { 928 check_md = q; 929 if (!efi_wb(check_md)) 930 break; 931 if (contig_high != check_md->phys_addr) 932 break; 933 contig_high = efi_md_end(check_md); 934 } 935 contig_high = GRANULEROUNDDOWN(contig_high); 936 } 937 if (!is_memory_available(md) || md->type == EFI_LOADER_DATA) 938 continue; 939 940 /* Round ends inward to granule boundaries */ 941 as = max(contig_low, md->phys_addr); 942 ae = min(contig_high, efi_md_end(md)); 943 944 /* keep within max_addr= and min_addr= command line arg */ 945 as = max(as, min_addr); 946 ae = min(ae, max_addr); 947 if (ae <= as) 948 continue; 949 950 /* avoid going over mem= command line arg */ 951 if (total_mem + (ae - as) > mem_limit) 952 ae -= total_mem + (ae - as) - mem_limit; 953 954 if (ae <= as) 955 continue; 956 957 if (ae - as > space_needed) 958 break; 959 } 960 if (p >= efi_map_end) 961 panic("Can't allocate space for kernel memory descriptors"); 962 963 return __va(as); 964 } 965 966 /* 967 * Walk the EFI memory map and gather all memory available for kernel 968 * to use. We can allocate partial granules only if the unavailable 969 * parts exist, and are WB. 970 */ 971 unsigned long 972 efi_memmap_init(unsigned long *s, unsigned long *e) 973 { 974 struct kern_memdesc *k, *prev = NULL; 975 u64 contig_low=0, contig_high=0; 976 u64 as, ae, lim; 977 void *efi_map_start, *efi_map_end, *p, *q; 978 efi_memory_desc_t *md, *pmd = NULL, *check_md; 979 u64 efi_desc_size; 980 unsigned long total_mem = 0; 981 982 k = kern_memmap = find_memmap_space(); 983 984 efi_map_start = __va(ia64_boot_param->efi_memmap); 985 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 986 efi_desc_size = ia64_boot_param->efi_memdesc_size; 987 988 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) { 989 md = p; 990 if (!efi_wb(md)) { 991 if (efi_uc(md) && (md->type == EFI_CONVENTIONAL_MEMORY || 992 md->type == EFI_BOOT_SERVICES_DATA)) { 993 k->attribute = EFI_MEMORY_UC; 994 k->start = md->phys_addr; 995 k->num_pages = md->num_pages; 996 k++; 997 } 998 continue; 999 } 1000 if (pmd == NULL || !efi_wb(pmd) || efi_md_end(pmd) != md->phys_addr) { 1001 contig_low = GRANULEROUNDUP(md->phys_addr); 1002 contig_high = efi_md_end(md); 1003 for (q = p + efi_desc_size; q < efi_map_end; q += efi_desc_size) { 1004 check_md = q; 1005 if (!efi_wb(check_md)) 1006 break; 1007 if (contig_high != check_md->phys_addr) 1008 break; 1009 contig_high = efi_md_end(check_md); 1010 } 1011 contig_high = GRANULEROUNDDOWN(contig_high); 1012 } 1013 if (!is_memory_available(md)) 1014 continue; 1015 1016 #ifdef CONFIG_CRASH_DUMP 1017 /* saved_max_pfn should ignore max_addr= command line arg */ 1018 if (saved_max_pfn < (efi_md_end(md) >> PAGE_SHIFT)) 1019 saved_max_pfn = (efi_md_end(md) >> PAGE_SHIFT); 1020 #endif 1021 /* 1022 * Round ends inward to granule boundaries 1023 * Give trimmings to uncached allocator 1024 */ 1025 if (md->phys_addr < contig_low) { 1026 lim = min(efi_md_end(md), contig_low); 1027 if (efi_uc(md)) { 1028 if (k > kern_memmap && (k-1)->attribute == EFI_MEMORY_UC && 1029 kmd_end(k-1) == md->phys_addr) { 1030 (k-1)->num_pages += (lim - md->phys_addr) >> EFI_PAGE_SHIFT; 1031 } else { 1032 k->attribute = EFI_MEMORY_UC; 1033 k->start = md->phys_addr; 1034 k->num_pages = (lim - md->phys_addr) >> EFI_PAGE_SHIFT; 1035 k++; 1036 } 1037 } 1038 as = contig_low; 1039 } else 1040 as = md->phys_addr; 1041 1042 if (efi_md_end(md) > contig_high) { 1043 lim = max(md->phys_addr, contig_high); 1044 if (efi_uc(md)) { 1045 if (lim == md->phys_addr && k > kern_memmap && 1046 (k-1)->attribute == EFI_MEMORY_UC && 1047 kmd_end(k-1) == md->phys_addr) { 1048 (k-1)->num_pages += md->num_pages; 1049 } else { 1050 k->attribute = EFI_MEMORY_UC; 1051 k->start = lim; 1052 k->num_pages = (efi_md_end(md) - lim) >> EFI_PAGE_SHIFT; 1053 k++; 1054 } 1055 } 1056 ae = contig_high; 1057 } else 1058 ae = efi_md_end(md); 1059 1060 /* keep within max_addr= and min_addr= command line arg */ 1061 as = max(as, min_addr); 1062 ae = min(ae, max_addr); 1063 if (ae <= as) 1064 continue; 1065 1066 /* avoid going over mem= command line arg */ 1067 if (total_mem + (ae - as) > mem_limit) 1068 ae -= total_mem + (ae - as) - mem_limit; 1069 1070 if (ae <= as) 1071 continue; 1072 if (prev && kmd_end(prev) == md->phys_addr) { 1073 prev->num_pages += (ae - as) >> EFI_PAGE_SHIFT; 1074 total_mem += ae - as; 1075 continue; 1076 } 1077 k->attribute = EFI_MEMORY_WB; 1078 k->start = as; 1079 k->num_pages = (ae - as) >> EFI_PAGE_SHIFT; 1080 total_mem += ae - as; 1081 prev = k++; 1082 } 1083 k->start = ~0L; /* end-marker */ 1084 1085 /* reserve the memory we are using for kern_memmap */ 1086 *s = (u64)kern_memmap; 1087 *e = (u64)++k; 1088 1089 return total_mem; 1090 } 1091 1092 void 1093 efi_initialize_iomem_resources(struct resource *code_resource, 1094 struct resource *data_resource, 1095 struct resource *bss_resource) 1096 { 1097 struct resource *res; 1098 void *efi_map_start, *efi_map_end, *p; 1099 efi_memory_desc_t *md; 1100 u64 efi_desc_size; 1101 char *name; 1102 unsigned long flags; 1103 1104 efi_map_start = __va(ia64_boot_param->efi_memmap); 1105 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 1106 efi_desc_size = ia64_boot_param->efi_memdesc_size; 1107 1108 res = NULL; 1109 1110 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 1111 md = p; 1112 1113 if (md->num_pages == 0) /* should not happen */ 1114 continue; 1115 1116 flags = IORESOURCE_MEM | IORESOURCE_BUSY; 1117 switch (md->type) { 1118 1119 case EFI_MEMORY_MAPPED_IO: 1120 case EFI_MEMORY_MAPPED_IO_PORT_SPACE: 1121 continue; 1122 1123 case EFI_LOADER_CODE: 1124 case EFI_LOADER_DATA: 1125 case EFI_BOOT_SERVICES_DATA: 1126 case EFI_BOOT_SERVICES_CODE: 1127 case EFI_CONVENTIONAL_MEMORY: 1128 if (md->attribute & EFI_MEMORY_WP) { 1129 name = "System ROM"; 1130 flags |= IORESOURCE_READONLY; 1131 } else { 1132 name = "System RAM"; 1133 } 1134 break; 1135 1136 case EFI_ACPI_MEMORY_NVS: 1137 name = "ACPI Non-volatile Storage"; 1138 break; 1139 1140 case EFI_UNUSABLE_MEMORY: 1141 name = "reserved"; 1142 flags |= IORESOURCE_DISABLED; 1143 break; 1144 1145 case EFI_RESERVED_TYPE: 1146 case EFI_RUNTIME_SERVICES_CODE: 1147 case EFI_RUNTIME_SERVICES_DATA: 1148 case EFI_ACPI_RECLAIM_MEMORY: 1149 default: 1150 name = "reserved"; 1151 break; 1152 } 1153 1154 if ((res = kzalloc(sizeof(struct resource), GFP_KERNEL)) == NULL) { 1155 printk(KERN_ERR "failed to alocate resource for iomem\n"); 1156 return; 1157 } 1158 1159 res->name = name; 1160 res->start = md->phys_addr; 1161 res->end = md->phys_addr + efi_md_size(md) - 1; 1162 res->flags = flags; 1163 1164 if (insert_resource(&iomem_resource, res) < 0) 1165 kfree(res); 1166 else { 1167 /* 1168 * We don't know which region contains 1169 * kernel data so we try it repeatedly and 1170 * let the resource manager test it. 1171 */ 1172 insert_resource(res, code_resource); 1173 insert_resource(res, data_resource); 1174 insert_resource(res, bss_resource); 1175 #ifdef CONFIG_KEXEC 1176 insert_resource(res, &efi_memmap_res); 1177 insert_resource(res, &boot_param_res); 1178 if (crashk_res.end > crashk_res.start) 1179 insert_resource(res, &crashk_res); 1180 #endif 1181 } 1182 } 1183 } 1184 1185 #ifdef CONFIG_KEXEC 1186 /* find a block of memory aligned to 64M exclude reserved regions 1187 rsvd_regions are sorted 1188 */ 1189 unsigned long __init 1190 kdump_find_rsvd_region (unsigned long size, 1191 struct rsvd_region *r, int n) 1192 { 1193 int i; 1194 u64 start, end; 1195 u64 alignment = 1UL << _PAGE_SIZE_64M; 1196 void *efi_map_start, *efi_map_end, *p; 1197 efi_memory_desc_t *md; 1198 u64 efi_desc_size; 1199 1200 efi_map_start = __va(ia64_boot_param->efi_memmap); 1201 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 1202 efi_desc_size = ia64_boot_param->efi_memdesc_size; 1203 1204 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 1205 md = p; 1206 if (!efi_wb(md)) 1207 continue; 1208 start = ALIGN(md->phys_addr, alignment); 1209 end = efi_md_end(md); 1210 for (i = 0; i < n; i++) { 1211 if (__pa(r[i].start) >= start && __pa(r[i].end) < end) { 1212 if (__pa(r[i].start) > start + size) 1213 return start; 1214 start = ALIGN(__pa(r[i].end), alignment); 1215 if (i < n-1 && __pa(r[i+1].start) < start + size) 1216 continue; 1217 else 1218 break; 1219 } 1220 } 1221 if (end > start + size) 1222 return start; 1223 } 1224 1225 printk(KERN_WARNING "Cannot reserve 0x%lx byte of memory for crashdump\n", 1226 size); 1227 return ~0UL; 1228 } 1229 #endif 1230 1231 #ifdef CONFIG_PROC_VMCORE 1232 /* locate the size find a the descriptor at a certain address */ 1233 unsigned long __init 1234 vmcore_find_descriptor_size (unsigned long address) 1235 { 1236 void *efi_map_start, *efi_map_end, *p; 1237 efi_memory_desc_t *md; 1238 u64 efi_desc_size; 1239 unsigned long ret = 0; 1240 1241 efi_map_start = __va(ia64_boot_param->efi_memmap); 1242 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 1243 efi_desc_size = ia64_boot_param->efi_memdesc_size; 1244 1245 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 1246 md = p; 1247 if (efi_wb(md) && md->type == EFI_LOADER_DATA 1248 && md->phys_addr == address) { 1249 ret = efi_md_size(md); 1250 break; 1251 } 1252 } 1253 1254 if (ret == 0) 1255 printk(KERN_WARNING "Cannot locate EFI vmcore descriptor\n"); 1256 1257 return ret; 1258 } 1259 #endif 1260