1 /* 2 * Extensible Firmware Interface 3 * 4 * Based on Extensible Firmware Interface Specification version 0.9 5 * April 30, 1999 6 * 7 * Copyright (C) 1999 VA Linux Systems 8 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com> 9 * Copyright (C) 1999-2003 Hewlett-Packard Co. 10 * David Mosberger-Tang <davidm@hpl.hp.com> 11 * Stephane Eranian <eranian@hpl.hp.com> 12 * (c) Copyright 2006 Hewlett-Packard Development Company, L.P. 13 * Bjorn Helgaas <bjorn.helgaas@hp.com> 14 * 15 * All EFI Runtime Services are not implemented yet as EFI only 16 * supports physical mode addressing on SoftSDV. This is to be fixed 17 * in a future version. --drummond 1999-07-20 18 * 19 * Implemented EFI runtime services and virtual mode calls. --davidm 20 * 21 * Goutham Rao: <goutham.rao@intel.com> 22 * Skip non-WB memory and ignore empty memory ranges. 23 */ 24 #include <linux/module.h> 25 #include <linux/bootmem.h> 26 #include <linux/crash_dump.h> 27 #include <linux/kernel.h> 28 #include <linux/init.h> 29 #include <linux/types.h> 30 #include <linux/slab.h> 31 #include <linux/time.h> 32 #include <linux/efi.h> 33 #include <linux/kexec.h> 34 #include <linux/mm.h> 35 36 #include <asm/io.h> 37 #include <asm/kregs.h> 38 #include <asm/meminit.h> 39 #include <asm/pgtable.h> 40 #include <asm/processor.h> 41 #include <asm/mca.h> 42 #include <asm/setup.h> 43 #include <asm/tlbflush.h> 44 45 #define EFI_DEBUG 0 46 47 static __initdata unsigned long palo_phys; 48 49 static __initdata efi_config_table_type_t arch_tables[] = { 50 {PROCESSOR_ABSTRACTION_LAYER_OVERWRITE_GUID, "PALO", &palo_phys}, 51 {NULL_GUID, NULL, 0}, 52 }; 53 54 extern efi_status_t efi_call_phys (void *, ...); 55 56 static efi_runtime_services_t *runtime; 57 static u64 mem_limit = ~0UL, max_addr = ~0UL, min_addr = 0UL; 58 59 #define efi_call_virt(f, args...) (*(f))(args) 60 61 #define STUB_GET_TIME(prefix, adjust_arg) \ 62 static efi_status_t \ 63 prefix##_get_time (efi_time_t *tm, efi_time_cap_t *tc) \ 64 { \ 65 struct ia64_fpreg fr[6]; \ 66 efi_time_cap_t *atc = NULL; \ 67 efi_status_t ret; \ 68 \ 69 if (tc) \ 70 atc = adjust_arg(tc); \ 71 ia64_save_scratch_fpregs(fr); \ 72 ret = efi_call_##prefix((efi_get_time_t *) __va(runtime->get_time), \ 73 adjust_arg(tm), atc); \ 74 ia64_load_scratch_fpregs(fr); \ 75 return ret; \ 76 } 77 78 #define STUB_SET_TIME(prefix, adjust_arg) \ 79 static efi_status_t \ 80 prefix##_set_time (efi_time_t *tm) \ 81 { \ 82 struct ia64_fpreg fr[6]; \ 83 efi_status_t ret; \ 84 \ 85 ia64_save_scratch_fpregs(fr); \ 86 ret = efi_call_##prefix((efi_set_time_t *) __va(runtime->set_time), \ 87 adjust_arg(tm)); \ 88 ia64_load_scratch_fpregs(fr); \ 89 return ret; \ 90 } 91 92 #define STUB_GET_WAKEUP_TIME(prefix, adjust_arg) \ 93 static efi_status_t \ 94 prefix##_get_wakeup_time (efi_bool_t *enabled, efi_bool_t *pending, \ 95 efi_time_t *tm) \ 96 { \ 97 struct ia64_fpreg fr[6]; \ 98 efi_status_t ret; \ 99 \ 100 ia64_save_scratch_fpregs(fr); \ 101 ret = efi_call_##prefix( \ 102 (efi_get_wakeup_time_t *) __va(runtime->get_wakeup_time), \ 103 adjust_arg(enabled), adjust_arg(pending), adjust_arg(tm)); \ 104 ia64_load_scratch_fpregs(fr); \ 105 return ret; \ 106 } 107 108 #define STUB_SET_WAKEUP_TIME(prefix, adjust_arg) \ 109 static efi_status_t \ 110 prefix##_set_wakeup_time (efi_bool_t enabled, efi_time_t *tm) \ 111 { \ 112 struct ia64_fpreg fr[6]; \ 113 efi_time_t *atm = NULL; \ 114 efi_status_t ret; \ 115 \ 116 if (tm) \ 117 atm = adjust_arg(tm); \ 118 ia64_save_scratch_fpregs(fr); \ 119 ret = efi_call_##prefix( \ 120 (efi_set_wakeup_time_t *) __va(runtime->set_wakeup_time), \ 121 enabled, atm); \ 122 ia64_load_scratch_fpregs(fr); \ 123 return ret; \ 124 } 125 126 #define STUB_GET_VARIABLE(prefix, adjust_arg) \ 127 static efi_status_t \ 128 prefix##_get_variable (efi_char16_t *name, efi_guid_t *vendor, u32 *attr, \ 129 unsigned long *data_size, void *data) \ 130 { \ 131 struct ia64_fpreg fr[6]; \ 132 u32 *aattr = NULL; \ 133 efi_status_t ret; \ 134 \ 135 if (attr) \ 136 aattr = adjust_arg(attr); \ 137 ia64_save_scratch_fpregs(fr); \ 138 ret = efi_call_##prefix( \ 139 (efi_get_variable_t *) __va(runtime->get_variable), \ 140 adjust_arg(name), adjust_arg(vendor), aattr, \ 141 adjust_arg(data_size), adjust_arg(data)); \ 142 ia64_load_scratch_fpregs(fr); \ 143 return ret; \ 144 } 145 146 #define STUB_GET_NEXT_VARIABLE(prefix, adjust_arg) \ 147 static efi_status_t \ 148 prefix##_get_next_variable (unsigned long *name_size, efi_char16_t *name, \ 149 efi_guid_t *vendor) \ 150 { \ 151 struct ia64_fpreg fr[6]; \ 152 efi_status_t ret; \ 153 \ 154 ia64_save_scratch_fpregs(fr); \ 155 ret = efi_call_##prefix( \ 156 (efi_get_next_variable_t *) __va(runtime->get_next_variable), \ 157 adjust_arg(name_size), adjust_arg(name), adjust_arg(vendor)); \ 158 ia64_load_scratch_fpregs(fr); \ 159 return ret; \ 160 } 161 162 #define STUB_SET_VARIABLE(prefix, adjust_arg) \ 163 static efi_status_t \ 164 prefix##_set_variable (efi_char16_t *name, efi_guid_t *vendor, \ 165 u32 attr, unsigned long data_size, \ 166 void *data) \ 167 { \ 168 struct ia64_fpreg fr[6]; \ 169 efi_status_t ret; \ 170 \ 171 ia64_save_scratch_fpregs(fr); \ 172 ret = efi_call_##prefix( \ 173 (efi_set_variable_t *) __va(runtime->set_variable), \ 174 adjust_arg(name), adjust_arg(vendor), attr, data_size, \ 175 adjust_arg(data)); \ 176 ia64_load_scratch_fpregs(fr); \ 177 return ret; \ 178 } 179 180 #define STUB_GET_NEXT_HIGH_MONO_COUNT(prefix, adjust_arg) \ 181 static efi_status_t \ 182 prefix##_get_next_high_mono_count (u32 *count) \ 183 { \ 184 struct ia64_fpreg fr[6]; \ 185 efi_status_t ret; \ 186 \ 187 ia64_save_scratch_fpregs(fr); \ 188 ret = efi_call_##prefix((efi_get_next_high_mono_count_t *) \ 189 __va(runtime->get_next_high_mono_count), \ 190 adjust_arg(count)); \ 191 ia64_load_scratch_fpregs(fr); \ 192 return ret; \ 193 } 194 195 #define STUB_RESET_SYSTEM(prefix, adjust_arg) \ 196 static void \ 197 prefix##_reset_system (int reset_type, efi_status_t status, \ 198 unsigned long data_size, efi_char16_t *data) \ 199 { \ 200 struct ia64_fpreg fr[6]; \ 201 efi_char16_t *adata = NULL; \ 202 \ 203 if (data) \ 204 adata = adjust_arg(data); \ 205 \ 206 ia64_save_scratch_fpregs(fr); \ 207 efi_call_##prefix( \ 208 (efi_reset_system_t *) __va(runtime->reset_system), \ 209 reset_type, status, data_size, adata); \ 210 /* should not return, but just in case... */ \ 211 ia64_load_scratch_fpregs(fr); \ 212 } 213 214 #define phys_ptr(arg) ((__typeof__(arg)) ia64_tpa(arg)) 215 216 STUB_GET_TIME(phys, phys_ptr) 217 STUB_SET_TIME(phys, phys_ptr) 218 STUB_GET_WAKEUP_TIME(phys, phys_ptr) 219 STUB_SET_WAKEUP_TIME(phys, phys_ptr) 220 STUB_GET_VARIABLE(phys, phys_ptr) 221 STUB_GET_NEXT_VARIABLE(phys, phys_ptr) 222 STUB_SET_VARIABLE(phys, phys_ptr) 223 STUB_GET_NEXT_HIGH_MONO_COUNT(phys, phys_ptr) 224 STUB_RESET_SYSTEM(phys, phys_ptr) 225 226 #define id(arg) arg 227 228 STUB_GET_TIME(virt, id) 229 STUB_SET_TIME(virt, id) 230 STUB_GET_WAKEUP_TIME(virt, id) 231 STUB_SET_WAKEUP_TIME(virt, id) 232 STUB_GET_VARIABLE(virt, id) 233 STUB_GET_NEXT_VARIABLE(virt, id) 234 STUB_SET_VARIABLE(virt, id) 235 STUB_GET_NEXT_HIGH_MONO_COUNT(virt, id) 236 STUB_RESET_SYSTEM(virt, id) 237 238 void 239 efi_gettimeofday (struct timespec *ts) 240 { 241 efi_time_t tm; 242 243 if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS) { 244 memset(ts, 0, sizeof(*ts)); 245 return; 246 } 247 248 ts->tv_sec = mktime(tm.year, tm.month, tm.day, 249 tm.hour, tm.minute, tm.second); 250 ts->tv_nsec = tm.nanosecond; 251 } 252 253 static int 254 is_memory_available (efi_memory_desc_t *md) 255 { 256 if (!(md->attribute & EFI_MEMORY_WB)) 257 return 0; 258 259 switch (md->type) { 260 case EFI_LOADER_CODE: 261 case EFI_LOADER_DATA: 262 case EFI_BOOT_SERVICES_CODE: 263 case EFI_BOOT_SERVICES_DATA: 264 case EFI_CONVENTIONAL_MEMORY: 265 return 1; 266 } 267 return 0; 268 } 269 270 typedef struct kern_memdesc { 271 u64 attribute; 272 u64 start; 273 u64 num_pages; 274 } kern_memdesc_t; 275 276 static kern_memdesc_t *kern_memmap; 277 278 #define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT) 279 280 static inline u64 281 kmd_end(kern_memdesc_t *kmd) 282 { 283 return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT)); 284 } 285 286 static inline u64 287 efi_md_end(efi_memory_desc_t *md) 288 { 289 return (md->phys_addr + efi_md_size(md)); 290 } 291 292 static inline int 293 efi_wb(efi_memory_desc_t *md) 294 { 295 return (md->attribute & EFI_MEMORY_WB); 296 } 297 298 static inline int 299 efi_uc(efi_memory_desc_t *md) 300 { 301 return (md->attribute & EFI_MEMORY_UC); 302 } 303 304 static void 305 walk (efi_freemem_callback_t callback, void *arg, u64 attr) 306 { 307 kern_memdesc_t *k; 308 u64 start, end, voff; 309 310 voff = (attr == EFI_MEMORY_WB) ? PAGE_OFFSET : __IA64_UNCACHED_OFFSET; 311 for (k = kern_memmap; k->start != ~0UL; k++) { 312 if (k->attribute != attr) 313 continue; 314 start = PAGE_ALIGN(k->start); 315 end = (k->start + (k->num_pages << EFI_PAGE_SHIFT)) & PAGE_MASK; 316 if (start < end) 317 if ((*callback)(start + voff, end + voff, arg) < 0) 318 return; 319 } 320 } 321 322 /* 323 * Walk the EFI memory map and call CALLBACK once for each EFI memory 324 * descriptor that has memory that is available for OS use. 325 */ 326 void 327 efi_memmap_walk (efi_freemem_callback_t callback, void *arg) 328 { 329 walk(callback, arg, EFI_MEMORY_WB); 330 } 331 332 /* 333 * Walk the EFI memory map and call CALLBACK once for each EFI memory 334 * descriptor that has memory that is available for uncached allocator. 335 */ 336 void 337 efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg) 338 { 339 walk(callback, arg, EFI_MEMORY_UC); 340 } 341 342 /* 343 * Look for the PAL_CODE region reported by EFI and map it using an 344 * ITR to enable safe PAL calls in virtual mode. See IA-64 Processor 345 * Abstraction Layer chapter 11 in ADAG 346 */ 347 void * 348 efi_get_pal_addr (void) 349 { 350 void *efi_map_start, *efi_map_end, *p; 351 efi_memory_desc_t *md; 352 u64 efi_desc_size; 353 int pal_code_count = 0; 354 u64 vaddr, mask; 355 356 efi_map_start = __va(ia64_boot_param->efi_memmap); 357 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 358 efi_desc_size = ia64_boot_param->efi_memdesc_size; 359 360 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 361 md = p; 362 if (md->type != EFI_PAL_CODE) 363 continue; 364 365 if (++pal_code_count > 1) { 366 printk(KERN_ERR "Too many EFI Pal Code memory ranges, " 367 "dropped @ %llx\n", md->phys_addr); 368 continue; 369 } 370 /* 371 * The only ITLB entry in region 7 that is used is the one 372 * installed by __start(). That entry covers a 64MB range. 373 */ 374 mask = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1); 375 vaddr = PAGE_OFFSET + md->phys_addr; 376 377 /* 378 * We must check that the PAL mapping won't overlap with the 379 * kernel mapping. 380 * 381 * PAL code is guaranteed to be aligned on a power of 2 between 382 * 4k and 256KB and that only one ITR is needed to map it. This 383 * implies that the PAL code is always aligned on its size, 384 * i.e., the closest matching page size supported by the TLB. 385 * Therefore PAL code is guaranteed never to cross a 64MB unless 386 * it is bigger than 64MB (very unlikely!). So for now the 387 * following test is enough to determine whether or not we need 388 * a dedicated ITR for the PAL code. 389 */ 390 if ((vaddr & mask) == (KERNEL_START & mask)) { 391 printk(KERN_INFO "%s: no need to install ITR for PAL code\n", 392 __func__); 393 continue; 394 } 395 396 if (efi_md_size(md) > IA64_GRANULE_SIZE) 397 panic("Whoa! PAL code size bigger than a granule!"); 398 399 #if EFI_DEBUG 400 mask = ~((1 << IA64_GRANULE_SHIFT) - 1); 401 402 printk(KERN_INFO "CPU %d: mapping PAL code " 403 "[0x%lx-0x%lx) into [0x%lx-0x%lx)\n", 404 smp_processor_id(), md->phys_addr, 405 md->phys_addr + efi_md_size(md), 406 vaddr & mask, (vaddr & mask) + IA64_GRANULE_SIZE); 407 #endif 408 return __va(md->phys_addr); 409 } 410 printk(KERN_WARNING "%s: no PAL-code memory-descriptor found\n", 411 __func__); 412 return NULL; 413 } 414 415 416 static u8 __init palo_checksum(u8 *buffer, u32 length) 417 { 418 u8 sum = 0; 419 u8 *end = buffer + length; 420 421 while (buffer < end) 422 sum = (u8) (sum + *(buffer++)); 423 424 return sum; 425 } 426 427 /* 428 * Parse and handle PALO table which is published at: 429 * http://www.dig64.org/home/DIG64_PALO_R1_0.pdf 430 */ 431 static void __init handle_palo(unsigned long phys_addr) 432 { 433 struct palo_table *palo = __va(phys_addr); 434 u8 checksum; 435 436 if (strncmp(palo->signature, PALO_SIG, sizeof(PALO_SIG) - 1)) { 437 printk(KERN_INFO "PALO signature incorrect.\n"); 438 return; 439 } 440 441 checksum = palo_checksum((u8 *)palo, palo->length); 442 if (checksum) { 443 printk(KERN_INFO "PALO checksum incorrect.\n"); 444 return; 445 } 446 447 setup_ptcg_sem(palo->max_tlb_purges, NPTCG_FROM_PALO); 448 } 449 450 void 451 efi_map_pal_code (void) 452 { 453 void *pal_vaddr = efi_get_pal_addr (); 454 u64 psr; 455 456 if (!pal_vaddr) 457 return; 458 459 /* 460 * Cannot write to CRx with PSR.ic=1 461 */ 462 psr = ia64_clear_ic(); 463 ia64_itr(0x1, IA64_TR_PALCODE, 464 GRANULEROUNDDOWN((unsigned long) pal_vaddr), 465 pte_val(pfn_pte(__pa(pal_vaddr) >> PAGE_SHIFT, PAGE_KERNEL)), 466 IA64_GRANULE_SHIFT); 467 paravirt_dv_serialize_data(); 468 ia64_set_psr(psr); /* restore psr */ 469 } 470 471 void __init 472 efi_init (void) 473 { 474 void *efi_map_start, *efi_map_end; 475 efi_char16_t *c16; 476 u64 efi_desc_size; 477 char *cp, vendor[100] = "unknown"; 478 int i; 479 480 /* 481 * It's too early to be able to use the standard kernel command line 482 * support... 483 */ 484 for (cp = boot_command_line; *cp; ) { 485 if (memcmp(cp, "mem=", 4) == 0) { 486 mem_limit = memparse(cp + 4, &cp); 487 } else if (memcmp(cp, "max_addr=", 9) == 0) { 488 max_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp)); 489 } else if (memcmp(cp, "min_addr=", 9) == 0) { 490 min_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp)); 491 } else { 492 while (*cp != ' ' && *cp) 493 ++cp; 494 while (*cp == ' ') 495 ++cp; 496 } 497 } 498 if (min_addr != 0UL) 499 printk(KERN_INFO "Ignoring memory below %lluMB\n", 500 min_addr >> 20); 501 if (max_addr != ~0UL) 502 printk(KERN_INFO "Ignoring memory above %lluMB\n", 503 max_addr >> 20); 504 505 efi.systab = __va(ia64_boot_param->efi_systab); 506 507 /* 508 * Verify the EFI Table 509 */ 510 if (efi.systab == NULL) 511 panic("Whoa! Can't find EFI system table.\n"); 512 if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) 513 panic("Whoa! EFI system table signature incorrect\n"); 514 if ((efi.systab->hdr.revision >> 16) == 0) 515 printk(KERN_WARNING "Warning: EFI system table version " 516 "%d.%02d, expected 1.00 or greater\n", 517 efi.systab->hdr.revision >> 16, 518 efi.systab->hdr.revision & 0xffff); 519 520 /* Show what we know for posterity */ 521 c16 = __va(efi.systab->fw_vendor); 522 if (c16) { 523 for (i = 0;i < (int) sizeof(vendor) - 1 && *c16; ++i) 524 vendor[i] = *c16++; 525 vendor[i] = '\0'; 526 } 527 528 printk(KERN_INFO "EFI v%u.%.02u by %s:", 529 efi.systab->hdr.revision >> 16, 530 efi.systab->hdr.revision & 0xffff, vendor); 531 532 palo_phys = EFI_INVALID_TABLE_ADDR; 533 534 if (efi_config_init(arch_tables) != 0) 535 return; 536 537 if (palo_phys != EFI_INVALID_TABLE_ADDR) 538 handle_palo(palo_phys); 539 540 runtime = __va(efi.systab->runtime); 541 efi.get_time = phys_get_time; 542 efi.set_time = phys_set_time; 543 efi.get_wakeup_time = phys_get_wakeup_time; 544 efi.set_wakeup_time = phys_set_wakeup_time; 545 efi.get_variable = phys_get_variable; 546 efi.get_next_variable = phys_get_next_variable; 547 efi.set_variable = phys_set_variable; 548 efi.get_next_high_mono_count = phys_get_next_high_mono_count; 549 efi.reset_system = phys_reset_system; 550 551 efi_map_start = __va(ia64_boot_param->efi_memmap); 552 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 553 efi_desc_size = ia64_boot_param->efi_memdesc_size; 554 555 #if EFI_DEBUG 556 /* print EFI memory map: */ 557 { 558 efi_memory_desc_t *md; 559 void *p; 560 561 for (i = 0, p = efi_map_start; p < efi_map_end; 562 ++i, p += efi_desc_size) 563 { 564 const char *unit; 565 unsigned long size; 566 567 md = p; 568 size = md->num_pages << EFI_PAGE_SHIFT; 569 570 if ((size >> 40) > 0) { 571 size >>= 40; 572 unit = "TB"; 573 } else if ((size >> 30) > 0) { 574 size >>= 30; 575 unit = "GB"; 576 } else if ((size >> 20) > 0) { 577 size >>= 20; 578 unit = "MB"; 579 } else { 580 size >>= 10; 581 unit = "KB"; 582 } 583 584 printk("mem%02d: type=%2u, attr=0x%016lx, " 585 "range=[0x%016lx-0x%016lx) (%4lu%s)\n", 586 i, md->type, md->attribute, md->phys_addr, 587 md->phys_addr + efi_md_size(md), size, unit); 588 } 589 } 590 #endif 591 592 efi_map_pal_code(); 593 efi_enter_virtual_mode(); 594 } 595 596 void 597 efi_enter_virtual_mode (void) 598 { 599 void *efi_map_start, *efi_map_end, *p; 600 efi_memory_desc_t *md; 601 efi_status_t status; 602 u64 efi_desc_size; 603 604 efi_map_start = __va(ia64_boot_param->efi_memmap); 605 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 606 efi_desc_size = ia64_boot_param->efi_memdesc_size; 607 608 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 609 md = p; 610 if (md->attribute & EFI_MEMORY_RUNTIME) { 611 /* 612 * Some descriptors have multiple bits set, so the 613 * order of the tests is relevant. 614 */ 615 if (md->attribute & EFI_MEMORY_WB) { 616 md->virt_addr = (u64) __va(md->phys_addr); 617 } else if (md->attribute & EFI_MEMORY_UC) { 618 md->virt_addr = (u64) ioremap(md->phys_addr, 0); 619 } else if (md->attribute & EFI_MEMORY_WC) { 620 #if 0 621 md->virt_addr = ia64_remap(md->phys_addr, 622 (_PAGE_A | 623 _PAGE_P | 624 _PAGE_D | 625 _PAGE_MA_WC | 626 _PAGE_PL_0 | 627 _PAGE_AR_RW)); 628 #else 629 printk(KERN_INFO "EFI_MEMORY_WC mapping\n"); 630 md->virt_addr = (u64) ioremap(md->phys_addr, 0); 631 #endif 632 } else if (md->attribute & EFI_MEMORY_WT) { 633 #if 0 634 md->virt_addr = ia64_remap(md->phys_addr, 635 (_PAGE_A | 636 _PAGE_P | 637 _PAGE_D | 638 _PAGE_MA_WT | 639 _PAGE_PL_0 | 640 _PAGE_AR_RW)); 641 #else 642 printk(KERN_INFO "EFI_MEMORY_WT mapping\n"); 643 md->virt_addr = (u64) ioremap(md->phys_addr, 0); 644 #endif 645 } 646 } 647 } 648 649 status = efi_call_phys(__va(runtime->set_virtual_address_map), 650 ia64_boot_param->efi_memmap_size, 651 efi_desc_size, 652 ia64_boot_param->efi_memdesc_version, 653 ia64_boot_param->efi_memmap); 654 if (status != EFI_SUCCESS) { 655 printk(KERN_WARNING "warning: unable to switch EFI into " 656 "virtual mode (status=%lu)\n", status); 657 return; 658 } 659 660 /* 661 * Now that EFI is in virtual mode, we call the EFI functions more 662 * efficiently: 663 */ 664 efi.get_time = virt_get_time; 665 efi.set_time = virt_set_time; 666 efi.get_wakeup_time = virt_get_wakeup_time; 667 efi.set_wakeup_time = virt_set_wakeup_time; 668 efi.get_variable = virt_get_variable; 669 efi.get_next_variable = virt_get_next_variable; 670 efi.set_variable = virt_set_variable; 671 efi.get_next_high_mono_count = virt_get_next_high_mono_count; 672 efi.reset_system = virt_reset_system; 673 } 674 675 /* 676 * Walk the EFI memory map looking for the I/O port range. There can only be 677 * one entry of this type, other I/O port ranges should be described via ACPI. 678 */ 679 u64 680 efi_get_iobase (void) 681 { 682 void *efi_map_start, *efi_map_end, *p; 683 efi_memory_desc_t *md; 684 u64 efi_desc_size; 685 686 efi_map_start = __va(ia64_boot_param->efi_memmap); 687 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 688 efi_desc_size = ia64_boot_param->efi_memdesc_size; 689 690 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 691 md = p; 692 if (md->type == EFI_MEMORY_MAPPED_IO_PORT_SPACE) { 693 if (md->attribute & EFI_MEMORY_UC) 694 return md->phys_addr; 695 } 696 } 697 return 0; 698 } 699 700 static struct kern_memdesc * 701 kern_memory_descriptor (unsigned long phys_addr) 702 { 703 struct kern_memdesc *md; 704 705 for (md = kern_memmap; md->start != ~0UL; md++) { 706 if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT)) 707 return md; 708 } 709 return NULL; 710 } 711 712 static efi_memory_desc_t * 713 efi_memory_descriptor (unsigned long phys_addr) 714 { 715 void *efi_map_start, *efi_map_end, *p; 716 efi_memory_desc_t *md; 717 u64 efi_desc_size; 718 719 efi_map_start = __va(ia64_boot_param->efi_memmap); 720 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 721 efi_desc_size = ia64_boot_param->efi_memdesc_size; 722 723 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 724 md = p; 725 726 if (phys_addr - md->phys_addr < efi_md_size(md)) 727 return md; 728 } 729 return NULL; 730 } 731 732 static int 733 efi_memmap_intersects (unsigned long phys_addr, unsigned long size) 734 { 735 void *efi_map_start, *efi_map_end, *p; 736 efi_memory_desc_t *md; 737 u64 efi_desc_size; 738 unsigned long end; 739 740 efi_map_start = __va(ia64_boot_param->efi_memmap); 741 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 742 efi_desc_size = ia64_boot_param->efi_memdesc_size; 743 744 end = phys_addr + size; 745 746 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 747 md = p; 748 if (md->phys_addr < end && efi_md_end(md) > phys_addr) 749 return 1; 750 } 751 return 0; 752 } 753 754 u32 755 efi_mem_type (unsigned long phys_addr) 756 { 757 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr); 758 759 if (md) 760 return md->type; 761 return 0; 762 } 763 764 u64 765 efi_mem_attributes (unsigned long phys_addr) 766 { 767 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr); 768 769 if (md) 770 return md->attribute; 771 return 0; 772 } 773 EXPORT_SYMBOL(efi_mem_attributes); 774 775 u64 776 efi_mem_attribute (unsigned long phys_addr, unsigned long size) 777 { 778 unsigned long end = phys_addr + size; 779 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr); 780 u64 attr; 781 782 if (!md) 783 return 0; 784 785 /* 786 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells 787 * the kernel that firmware needs this region mapped. 788 */ 789 attr = md->attribute & ~EFI_MEMORY_RUNTIME; 790 do { 791 unsigned long md_end = efi_md_end(md); 792 793 if (end <= md_end) 794 return attr; 795 796 md = efi_memory_descriptor(md_end); 797 if (!md || (md->attribute & ~EFI_MEMORY_RUNTIME) != attr) 798 return 0; 799 } while (md); 800 return 0; /* never reached */ 801 } 802 803 u64 804 kern_mem_attribute (unsigned long phys_addr, unsigned long size) 805 { 806 unsigned long end = phys_addr + size; 807 struct kern_memdesc *md; 808 u64 attr; 809 810 /* 811 * This is a hack for ioremap calls before we set up kern_memmap. 812 * Maybe we should do efi_memmap_init() earlier instead. 813 */ 814 if (!kern_memmap) { 815 attr = efi_mem_attribute(phys_addr, size); 816 if (attr & EFI_MEMORY_WB) 817 return EFI_MEMORY_WB; 818 return 0; 819 } 820 821 md = kern_memory_descriptor(phys_addr); 822 if (!md) 823 return 0; 824 825 attr = md->attribute; 826 do { 827 unsigned long md_end = kmd_end(md); 828 829 if (end <= md_end) 830 return attr; 831 832 md = kern_memory_descriptor(md_end); 833 if (!md || md->attribute != attr) 834 return 0; 835 } while (md); 836 return 0; /* never reached */ 837 } 838 EXPORT_SYMBOL(kern_mem_attribute); 839 840 int 841 valid_phys_addr_range (phys_addr_t phys_addr, unsigned long size) 842 { 843 u64 attr; 844 845 /* 846 * /dev/mem reads and writes use copy_to_user(), which implicitly 847 * uses a granule-sized kernel identity mapping. It's really 848 * only safe to do this for regions in kern_memmap. For more 849 * details, see Documentation/ia64/aliasing.txt. 850 */ 851 attr = kern_mem_attribute(phys_addr, size); 852 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC) 853 return 1; 854 return 0; 855 } 856 857 int 858 valid_mmap_phys_addr_range (unsigned long pfn, unsigned long size) 859 { 860 unsigned long phys_addr = pfn << PAGE_SHIFT; 861 u64 attr; 862 863 attr = efi_mem_attribute(phys_addr, size); 864 865 /* 866 * /dev/mem mmap uses normal user pages, so we don't need the entire 867 * granule, but the entire region we're mapping must support the same 868 * attribute. 869 */ 870 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC) 871 return 1; 872 873 /* 874 * Intel firmware doesn't tell us about all the MMIO regions, so 875 * in general we have to allow mmap requests. But if EFI *does* 876 * tell us about anything inside this region, we should deny it. 877 * The user can always map a smaller region to avoid the overlap. 878 */ 879 if (efi_memmap_intersects(phys_addr, size)) 880 return 0; 881 882 return 1; 883 } 884 885 pgprot_t 886 phys_mem_access_prot(struct file *file, unsigned long pfn, unsigned long size, 887 pgprot_t vma_prot) 888 { 889 unsigned long phys_addr = pfn << PAGE_SHIFT; 890 u64 attr; 891 892 /* 893 * For /dev/mem mmap, we use user mappings, but if the region is 894 * in kern_memmap (and hence may be covered by a kernel mapping), 895 * we must use the same attribute as the kernel mapping. 896 */ 897 attr = kern_mem_attribute(phys_addr, size); 898 if (attr & EFI_MEMORY_WB) 899 return pgprot_cacheable(vma_prot); 900 else if (attr & EFI_MEMORY_UC) 901 return pgprot_noncached(vma_prot); 902 903 /* 904 * Some chipsets don't support UC access to memory. If 905 * WB is supported, we prefer that. 906 */ 907 if (efi_mem_attribute(phys_addr, size) & EFI_MEMORY_WB) 908 return pgprot_cacheable(vma_prot); 909 910 return pgprot_noncached(vma_prot); 911 } 912 913 int __init 914 efi_uart_console_only(void) 915 { 916 efi_status_t status; 917 char *s, name[] = "ConOut"; 918 efi_guid_t guid = EFI_GLOBAL_VARIABLE_GUID; 919 efi_char16_t *utf16, name_utf16[32]; 920 unsigned char data[1024]; 921 unsigned long size = sizeof(data); 922 struct efi_generic_dev_path *hdr, *end_addr; 923 int uart = 0; 924 925 /* Convert to UTF-16 */ 926 utf16 = name_utf16; 927 s = name; 928 while (*s) 929 *utf16++ = *s++ & 0x7f; 930 *utf16 = 0; 931 932 status = efi.get_variable(name_utf16, &guid, NULL, &size, data); 933 if (status != EFI_SUCCESS) { 934 printk(KERN_ERR "No EFI %s variable?\n", name); 935 return 0; 936 } 937 938 hdr = (struct efi_generic_dev_path *) data; 939 end_addr = (struct efi_generic_dev_path *) ((u8 *) data + size); 940 while (hdr < end_addr) { 941 if (hdr->type == EFI_DEV_MSG && 942 hdr->sub_type == EFI_DEV_MSG_UART) 943 uart = 1; 944 else if (hdr->type == EFI_DEV_END_PATH || 945 hdr->type == EFI_DEV_END_PATH2) { 946 if (!uart) 947 return 0; 948 if (hdr->sub_type == EFI_DEV_END_ENTIRE) 949 return 1; 950 uart = 0; 951 } 952 hdr = (struct efi_generic_dev_path *)((u8 *) hdr + hdr->length); 953 } 954 printk(KERN_ERR "Malformed %s value\n", name); 955 return 0; 956 } 957 958 /* 959 * Look for the first granule aligned memory descriptor memory 960 * that is big enough to hold EFI memory map. Make sure this 961 * descriptor is atleast granule sized so it does not get trimmed 962 */ 963 struct kern_memdesc * 964 find_memmap_space (void) 965 { 966 u64 contig_low=0, contig_high=0; 967 u64 as = 0, ae; 968 void *efi_map_start, *efi_map_end, *p, *q; 969 efi_memory_desc_t *md, *pmd = NULL, *check_md; 970 u64 space_needed, efi_desc_size; 971 unsigned long total_mem = 0; 972 973 efi_map_start = __va(ia64_boot_param->efi_memmap); 974 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 975 efi_desc_size = ia64_boot_param->efi_memdesc_size; 976 977 /* 978 * Worst case: we need 3 kernel descriptors for each efi descriptor 979 * (if every entry has a WB part in the middle, and UC head and tail), 980 * plus one for the end marker. 981 */ 982 space_needed = sizeof(kern_memdesc_t) * 983 (3 * (ia64_boot_param->efi_memmap_size/efi_desc_size) + 1); 984 985 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) { 986 md = p; 987 if (!efi_wb(md)) { 988 continue; 989 } 990 if (pmd == NULL || !efi_wb(pmd) || 991 efi_md_end(pmd) != md->phys_addr) { 992 contig_low = GRANULEROUNDUP(md->phys_addr); 993 contig_high = efi_md_end(md); 994 for (q = p + efi_desc_size; q < efi_map_end; 995 q += efi_desc_size) { 996 check_md = q; 997 if (!efi_wb(check_md)) 998 break; 999 if (contig_high != check_md->phys_addr) 1000 break; 1001 contig_high = efi_md_end(check_md); 1002 } 1003 contig_high = GRANULEROUNDDOWN(contig_high); 1004 } 1005 if (!is_memory_available(md) || md->type == EFI_LOADER_DATA) 1006 continue; 1007 1008 /* Round ends inward to granule boundaries */ 1009 as = max(contig_low, md->phys_addr); 1010 ae = min(contig_high, efi_md_end(md)); 1011 1012 /* keep within max_addr= and min_addr= command line arg */ 1013 as = max(as, min_addr); 1014 ae = min(ae, max_addr); 1015 if (ae <= as) 1016 continue; 1017 1018 /* avoid going over mem= command line arg */ 1019 if (total_mem + (ae - as) > mem_limit) 1020 ae -= total_mem + (ae - as) - mem_limit; 1021 1022 if (ae <= as) 1023 continue; 1024 1025 if (ae - as > space_needed) 1026 break; 1027 } 1028 if (p >= efi_map_end) 1029 panic("Can't allocate space for kernel memory descriptors"); 1030 1031 return __va(as); 1032 } 1033 1034 /* 1035 * Walk the EFI memory map and gather all memory available for kernel 1036 * to use. We can allocate partial granules only if the unavailable 1037 * parts exist, and are WB. 1038 */ 1039 unsigned long 1040 efi_memmap_init(u64 *s, u64 *e) 1041 { 1042 struct kern_memdesc *k, *prev = NULL; 1043 u64 contig_low=0, contig_high=0; 1044 u64 as, ae, lim; 1045 void *efi_map_start, *efi_map_end, *p, *q; 1046 efi_memory_desc_t *md, *pmd = NULL, *check_md; 1047 u64 efi_desc_size; 1048 unsigned long total_mem = 0; 1049 1050 k = kern_memmap = find_memmap_space(); 1051 1052 efi_map_start = __va(ia64_boot_param->efi_memmap); 1053 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 1054 efi_desc_size = ia64_boot_param->efi_memdesc_size; 1055 1056 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) { 1057 md = p; 1058 if (!efi_wb(md)) { 1059 if (efi_uc(md) && 1060 (md->type == EFI_CONVENTIONAL_MEMORY || 1061 md->type == EFI_BOOT_SERVICES_DATA)) { 1062 k->attribute = EFI_MEMORY_UC; 1063 k->start = md->phys_addr; 1064 k->num_pages = md->num_pages; 1065 k++; 1066 } 1067 continue; 1068 } 1069 if (pmd == NULL || !efi_wb(pmd) || 1070 efi_md_end(pmd) != md->phys_addr) { 1071 contig_low = GRANULEROUNDUP(md->phys_addr); 1072 contig_high = efi_md_end(md); 1073 for (q = p + efi_desc_size; q < efi_map_end; 1074 q += efi_desc_size) { 1075 check_md = q; 1076 if (!efi_wb(check_md)) 1077 break; 1078 if (contig_high != check_md->phys_addr) 1079 break; 1080 contig_high = efi_md_end(check_md); 1081 } 1082 contig_high = GRANULEROUNDDOWN(contig_high); 1083 } 1084 if (!is_memory_available(md)) 1085 continue; 1086 1087 /* 1088 * Round ends inward to granule boundaries 1089 * Give trimmings to uncached allocator 1090 */ 1091 if (md->phys_addr < contig_low) { 1092 lim = min(efi_md_end(md), contig_low); 1093 if (efi_uc(md)) { 1094 if (k > kern_memmap && 1095 (k-1)->attribute == EFI_MEMORY_UC && 1096 kmd_end(k-1) == md->phys_addr) { 1097 (k-1)->num_pages += 1098 (lim - md->phys_addr) 1099 >> EFI_PAGE_SHIFT; 1100 } else { 1101 k->attribute = EFI_MEMORY_UC; 1102 k->start = md->phys_addr; 1103 k->num_pages = (lim - md->phys_addr) 1104 >> EFI_PAGE_SHIFT; 1105 k++; 1106 } 1107 } 1108 as = contig_low; 1109 } else 1110 as = md->phys_addr; 1111 1112 if (efi_md_end(md) > contig_high) { 1113 lim = max(md->phys_addr, contig_high); 1114 if (efi_uc(md)) { 1115 if (lim == md->phys_addr && k > kern_memmap && 1116 (k-1)->attribute == EFI_MEMORY_UC && 1117 kmd_end(k-1) == md->phys_addr) { 1118 (k-1)->num_pages += md->num_pages; 1119 } else { 1120 k->attribute = EFI_MEMORY_UC; 1121 k->start = lim; 1122 k->num_pages = (efi_md_end(md) - lim) 1123 >> EFI_PAGE_SHIFT; 1124 k++; 1125 } 1126 } 1127 ae = contig_high; 1128 } else 1129 ae = efi_md_end(md); 1130 1131 /* keep within max_addr= and min_addr= command line arg */ 1132 as = max(as, min_addr); 1133 ae = min(ae, max_addr); 1134 if (ae <= as) 1135 continue; 1136 1137 /* avoid going over mem= command line arg */ 1138 if (total_mem + (ae - as) > mem_limit) 1139 ae -= total_mem + (ae - as) - mem_limit; 1140 1141 if (ae <= as) 1142 continue; 1143 if (prev && kmd_end(prev) == md->phys_addr) { 1144 prev->num_pages += (ae - as) >> EFI_PAGE_SHIFT; 1145 total_mem += ae - as; 1146 continue; 1147 } 1148 k->attribute = EFI_MEMORY_WB; 1149 k->start = as; 1150 k->num_pages = (ae - as) >> EFI_PAGE_SHIFT; 1151 total_mem += ae - as; 1152 prev = k++; 1153 } 1154 k->start = ~0L; /* end-marker */ 1155 1156 /* reserve the memory we are using for kern_memmap */ 1157 *s = (u64)kern_memmap; 1158 *e = (u64)++k; 1159 1160 return total_mem; 1161 } 1162 1163 void 1164 efi_initialize_iomem_resources(struct resource *code_resource, 1165 struct resource *data_resource, 1166 struct resource *bss_resource) 1167 { 1168 struct resource *res; 1169 void *efi_map_start, *efi_map_end, *p; 1170 efi_memory_desc_t *md; 1171 u64 efi_desc_size; 1172 char *name; 1173 unsigned long flags; 1174 1175 efi_map_start = __va(ia64_boot_param->efi_memmap); 1176 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 1177 efi_desc_size = ia64_boot_param->efi_memdesc_size; 1178 1179 res = NULL; 1180 1181 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 1182 md = p; 1183 1184 if (md->num_pages == 0) /* should not happen */ 1185 continue; 1186 1187 flags = IORESOURCE_MEM | IORESOURCE_BUSY; 1188 switch (md->type) { 1189 1190 case EFI_MEMORY_MAPPED_IO: 1191 case EFI_MEMORY_MAPPED_IO_PORT_SPACE: 1192 continue; 1193 1194 case EFI_LOADER_CODE: 1195 case EFI_LOADER_DATA: 1196 case EFI_BOOT_SERVICES_DATA: 1197 case EFI_BOOT_SERVICES_CODE: 1198 case EFI_CONVENTIONAL_MEMORY: 1199 if (md->attribute & EFI_MEMORY_WP) { 1200 name = "System ROM"; 1201 flags |= IORESOURCE_READONLY; 1202 } else if (md->attribute == EFI_MEMORY_UC) 1203 name = "Uncached RAM"; 1204 else 1205 name = "System RAM"; 1206 break; 1207 1208 case EFI_ACPI_MEMORY_NVS: 1209 name = "ACPI Non-volatile Storage"; 1210 break; 1211 1212 case EFI_UNUSABLE_MEMORY: 1213 name = "reserved"; 1214 flags |= IORESOURCE_DISABLED; 1215 break; 1216 1217 case EFI_RESERVED_TYPE: 1218 case EFI_RUNTIME_SERVICES_CODE: 1219 case EFI_RUNTIME_SERVICES_DATA: 1220 case EFI_ACPI_RECLAIM_MEMORY: 1221 default: 1222 name = "reserved"; 1223 break; 1224 } 1225 1226 if ((res = kzalloc(sizeof(struct resource), 1227 GFP_KERNEL)) == NULL) { 1228 printk(KERN_ERR 1229 "failed to allocate resource for iomem\n"); 1230 return; 1231 } 1232 1233 res->name = name; 1234 res->start = md->phys_addr; 1235 res->end = md->phys_addr + efi_md_size(md) - 1; 1236 res->flags = flags; 1237 1238 if (insert_resource(&iomem_resource, res) < 0) 1239 kfree(res); 1240 else { 1241 /* 1242 * We don't know which region contains 1243 * kernel data so we try it repeatedly and 1244 * let the resource manager test it. 1245 */ 1246 insert_resource(res, code_resource); 1247 insert_resource(res, data_resource); 1248 insert_resource(res, bss_resource); 1249 #ifdef CONFIG_KEXEC 1250 insert_resource(res, &efi_memmap_res); 1251 insert_resource(res, &boot_param_res); 1252 if (crashk_res.end > crashk_res.start) 1253 insert_resource(res, &crashk_res); 1254 #endif 1255 } 1256 } 1257 } 1258 1259 #ifdef CONFIG_KEXEC 1260 /* find a block of memory aligned to 64M exclude reserved regions 1261 rsvd_regions are sorted 1262 */ 1263 unsigned long __init 1264 kdump_find_rsvd_region (unsigned long size, struct rsvd_region *r, int n) 1265 { 1266 int i; 1267 u64 start, end; 1268 u64 alignment = 1UL << _PAGE_SIZE_64M; 1269 void *efi_map_start, *efi_map_end, *p; 1270 efi_memory_desc_t *md; 1271 u64 efi_desc_size; 1272 1273 efi_map_start = __va(ia64_boot_param->efi_memmap); 1274 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 1275 efi_desc_size = ia64_boot_param->efi_memdesc_size; 1276 1277 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 1278 md = p; 1279 if (!efi_wb(md)) 1280 continue; 1281 start = ALIGN(md->phys_addr, alignment); 1282 end = efi_md_end(md); 1283 for (i = 0; i < n; i++) { 1284 if (__pa(r[i].start) >= start && __pa(r[i].end) < end) { 1285 if (__pa(r[i].start) > start + size) 1286 return start; 1287 start = ALIGN(__pa(r[i].end), alignment); 1288 if (i < n-1 && 1289 __pa(r[i+1].start) < start + size) 1290 continue; 1291 else 1292 break; 1293 } 1294 } 1295 if (end > start + size) 1296 return start; 1297 } 1298 1299 printk(KERN_WARNING 1300 "Cannot reserve 0x%lx byte of memory for crashdump\n", size); 1301 return ~0UL; 1302 } 1303 #endif 1304 1305 #ifdef CONFIG_CRASH_DUMP 1306 /* locate the size find a the descriptor at a certain address */ 1307 unsigned long __init 1308 vmcore_find_descriptor_size (unsigned long address) 1309 { 1310 void *efi_map_start, *efi_map_end, *p; 1311 efi_memory_desc_t *md; 1312 u64 efi_desc_size; 1313 unsigned long ret = 0; 1314 1315 efi_map_start = __va(ia64_boot_param->efi_memmap); 1316 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size; 1317 efi_desc_size = ia64_boot_param->efi_memdesc_size; 1318 1319 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) { 1320 md = p; 1321 if (efi_wb(md) && md->type == EFI_LOADER_DATA 1322 && md->phys_addr == address) { 1323 ret = efi_md_size(md); 1324 break; 1325 } 1326 } 1327 1328 if (ret == 0) 1329 printk(KERN_WARNING "Cannot locate EFI vmcore descriptor\n"); 1330 1331 return ret; 1332 } 1333 #endif 1334