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