1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Re-map IO memory to kernel address space so that we can access it. 4 * This is needed for high PCI addresses that aren't mapped in the 5 * 640k-1MB IO memory area on PC's 6 * 7 * (C) Copyright 1995 1996 Linus Torvalds 8 */ 9 10 #include <linux/memblock.h> 11 #include <linux/init.h> 12 #include <linux/io.h> 13 #include <linux/ioport.h> 14 #include <linux/ioremap.h> 15 #include <linux/slab.h> 16 #include <linux/vmalloc.h> 17 #include <linux/mmiotrace.h> 18 #include <linux/cc_platform.h> 19 #include <linux/efi.h> 20 #include <linux/pgtable.h> 21 #include <linux/kmsan.h> 22 23 #include <asm/set_memory.h> 24 #include <asm/e820/api.h> 25 #include <asm/efi.h> 26 #include <asm/fixmap.h> 27 #include <asm/tlbflush.h> 28 #include <asm/pgalloc.h> 29 #include <asm/memtype.h> 30 #include <asm/setup.h> 31 32 #include "physaddr.h" 33 34 /* 35 * Descriptor controlling ioremap() behavior. 36 */ 37 struct ioremap_desc { 38 unsigned int flags; 39 }; 40 41 /* 42 * Fix up the linear direct mapping of the kernel to avoid cache attribute 43 * conflicts. 44 */ 45 int ioremap_change_attr(unsigned long vaddr, unsigned long size, 46 enum page_cache_mode pcm) 47 { 48 unsigned long nrpages = size >> PAGE_SHIFT; 49 int err; 50 51 switch (pcm) { 52 case _PAGE_CACHE_MODE_UC: 53 default: 54 err = _set_memory_uc(vaddr, nrpages); 55 break; 56 case _PAGE_CACHE_MODE_WC: 57 err = _set_memory_wc(vaddr, nrpages); 58 break; 59 case _PAGE_CACHE_MODE_WT: 60 err = _set_memory_wt(vaddr, nrpages); 61 break; 62 case _PAGE_CACHE_MODE_WB: 63 err = _set_memory_wb(vaddr, nrpages); 64 break; 65 } 66 67 return err; 68 } 69 70 /* Does the range (or a subset of) contain normal RAM? */ 71 static unsigned int __ioremap_check_ram(struct resource *res) 72 { 73 unsigned long start_pfn, stop_pfn; 74 unsigned long i; 75 76 if ((res->flags & IORESOURCE_SYSTEM_RAM) != IORESOURCE_SYSTEM_RAM) 77 return 0; 78 79 start_pfn = (res->start + PAGE_SIZE - 1) >> PAGE_SHIFT; 80 stop_pfn = (res->end + 1) >> PAGE_SHIFT; 81 if (stop_pfn > start_pfn) { 82 for (i = 0; i < (stop_pfn - start_pfn); ++i) 83 if (pfn_valid(start_pfn + i) && 84 !PageReserved(pfn_to_page(start_pfn + i))) 85 return IORES_MAP_SYSTEM_RAM; 86 } 87 88 return 0; 89 } 90 91 /* 92 * In a SEV guest, NONE and RESERVED should not be mapped encrypted because 93 * there the whole memory is already encrypted. 94 */ 95 static unsigned int __ioremap_check_encrypted(struct resource *res) 96 { 97 if (!cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) 98 return 0; 99 100 switch (res->desc) { 101 case IORES_DESC_NONE: 102 case IORES_DESC_RESERVED: 103 break; 104 default: 105 return IORES_MAP_ENCRYPTED; 106 } 107 108 return 0; 109 } 110 111 /* 112 * The EFI runtime services data area is not covered by walk_mem_res(), but must 113 * be mapped encrypted when SEV is active. 114 */ 115 static void __ioremap_check_other(resource_size_t addr, struct ioremap_desc *desc) 116 { 117 if (!cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) 118 return; 119 120 if (x86_platform.hyper.is_private_mmio(addr)) { 121 desc->flags |= IORES_MAP_ENCRYPTED; 122 return; 123 } 124 125 if (!IS_ENABLED(CONFIG_EFI)) 126 return; 127 128 if (efi_mem_type(addr) == EFI_RUNTIME_SERVICES_DATA || 129 (efi_mem_type(addr) == EFI_BOOT_SERVICES_DATA && 130 efi_mem_attributes(addr) & EFI_MEMORY_RUNTIME)) 131 desc->flags |= IORES_MAP_ENCRYPTED; 132 } 133 134 static int __ioremap_collect_map_flags(struct resource *res, void *arg) 135 { 136 struct ioremap_desc *desc = arg; 137 138 if (!(desc->flags & IORES_MAP_SYSTEM_RAM)) 139 desc->flags |= __ioremap_check_ram(res); 140 141 if (!(desc->flags & IORES_MAP_ENCRYPTED)) 142 desc->flags |= __ioremap_check_encrypted(res); 143 144 return ((desc->flags & (IORES_MAP_SYSTEM_RAM | IORES_MAP_ENCRYPTED)) == 145 (IORES_MAP_SYSTEM_RAM | IORES_MAP_ENCRYPTED)); 146 } 147 148 /* 149 * To avoid multiple resource walks, this function walks resources marked as 150 * IORESOURCE_MEM and IORESOURCE_BUSY and looking for system RAM and/or a 151 * resource described not as IORES_DESC_NONE (e.g. IORES_DESC_ACPI_TABLES). 152 * 153 * After that, deal with misc other ranges in __ioremap_check_other() which do 154 * not fall into the above category. 155 */ 156 static void __ioremap_check_mem(resource_size_t addr, unsigned long size, 157 struct ioremap_desc *desc) 158 { 159 u64 start, end; 160 161 start = (u64)addr; 162 end = start + size - 1; 163 memset(desc, 0, sizeof(struct ioremap_desc)); 164 165 walk_mem_res(start, end, desc, __ioremap_collect_map_flags); 166 167 __ioremap_check_other(addr, desc); 168 } 169 170 /* 171 * Remap an arbitrary physical address space into the kernel virtual 172 * address space. It transparently creates kernel huge I/O mapping when 173 * the physical address is aligned by a huge page size (1GB or 2MB) and 174 * the requested size is at least the huge page size. 175 * 176 * NOTE: MTRRs can override PAT memory types with a 4KB granularity. 177 * Therefore, the mapping code falls back to use a smaller page toward 4KB 178 * when a mapping range is covered by non-WB type of MTRRs. 179 * 180 * NOTE! We need to allow non-page-aligned mappings too: we will obviously 181 * have to convert them into an offset in a page-aligned mapping, but the 182 * caller shouldn't need to know that small detail. 183 */ 184 static void __iomem * 185 __ioremap_caller(resource_size_t phys_addr, unsigned long size, 186 enum page_cache_mode pcm, void *caller, bool encrypted) 187 { 188 unsigned long offset, vaddr; 189 resource_size_t last_addr; 190 const resource_size_t unaligned_phys_addr = phys_addr; 191 const unsigned long unaligned_size = size; 192 struct ioremap_desc io_desc; 193 struct vm_struct *area; 194 enum page_cache_mode new_pcm; 195 pgprot_t prot; 196 int retval; 197 void __iomem *ret_addr; 198 199 /* Don't allow wraparound or zero size */ 200 last_addr = phys_addr + size - 1; 201 if (!size || last_addr < phys_addr) 202 return NULL; 203 204 if (!phys_addr_valid(phys_addr)) { 205 printk(KERN_WARNING "ioremap: invalid physical address %llx\n", 206 (unsigned long long)phys_addr); 207 WARN_ON_ONCE(1); 208 return NULL; 209 } 210 211 __ioremap_check_mem(phys_addr, size, &io_desc); 212 213 /* 214 * Don't allow anybody to remap normal RAM that we're using.. 215 */ 216 if (io_desc.flags & IORES_MAP_SYSTEM_RAM) { 217 WARN_ONCE(1, "ioremap on RAM at %pa - %pa\n", 218 &phys_addr, &last_addr); 219 return NULL; 220 } 221 222 /* 223 * Mappings have to be page-aligned 224 */ 225 offset = phys_addr & ~PAGE_MASK; 226 phys_addr &= PAGE_MASK; 227 size = PAGE_ALIGN(last_addr+1) - phys_addr; 228 229 /* 230 * Mask out any bits not part of the actual physical 231 * address, like memory encryption bits. 232 */ 233 phys_addr &= PHYSICAL_PAGE_MASK; 234 235 retval = memtype_reserve(phys_addr, (u64)phys_addr + size, 236 pcm, &new_pcm); 237 if (retval) { 238 printk(KERN_ERR "ioremap memtype_reserve failed %d\n", retval); 239 return NULL; 240 } 241 242 if (pcm != new_pcm) { 243 if (!is_new_memtype_allowed(phys_addr, size, pcm, new_pcm)) { 244 printk(KERN_ERR 245 "ioremap error for 0x%llx-0x%llx, requested 0x%x, got 0x%x\n", 246 (unsigned long long)phys_addr, 247 (unsigned long long)(phys_addr + size), 248 pcm, new_pcm); 249 goto err_free_memtype; 250 } 251 pcm = new_pcm; 252 } 253 254 /* 255 * If the page being mapped is in memory and SEV is active then 256 * make sure the memory encryption attribute is enabled in the 257 * resulting mapping. 258 * In TDX guests, memory is marked private by default. If encryption 259 * is not requested (using encrypted), explicitly set decrypt 260 * attribute in all IOREMAPPED memory. 261 */ 262 prot = PAGE_KERNEL_IO; 263 if ((io_desc.flags & IORES_MAP_ENCRYPTED) || encrypted) 264 prot = pgprot_encrypted(prot); 265 else 266 prot = pgprot_decrypted(prot); 267 268 switch (pcm) { 269 case _PAGE_CACHE_MODE_UC: 270 default: 271 prot = __pgprot(pgprot_val(prot) | 272 cachemode2protval(_PAGE_CACHE_MODE_UC)); 273 break; 274 case _PAGE_CACHE_MODE_UC_MINUS: 275 prot = __pgprot(pgprot_val(prot) | 276 cachemode2protval(_PAGE_CACHE_MODE_UC_MINUS)); 277 break; 278 case _PAGE_CACHE_MODE_WC: 279 prot = __pgprot(pgprot_val(prot) | 280 cachemode2protval(_PAGE_CACHE_MODE_WC)); 281 break; 282 case _PAGE_CACHE_MODE_WT: 283 prot = __pgprot(pgprot_val(prot) | 284 cachemode2protval(_PAGE_CACHE_MODE_WT)); 285 break; 286 case _PAGE_CACHE_MODE_WB: 287 break; 288 } 289 290 /* 291 * Ok, go for it.. 292 */ 293 area = get_vm_area_caller(size, VM_IOREMAP, caller); 294 if (!area) 295 goto err_free_memtype; 296 area->phys_addr = phys_addr; 297 vaddr = (unsigned long) area->addr; 298 299 if (memtype_kernel_map_sync(phys_addr, size, pcm)) 300 goto err_free_area; 301 302 if (ioremap_page_range(vaddr, vaddr + size, phys_addr, prot)) 303 goto err_free_area; 304 305 ret_addr = (void __iomem *) (vaddr + offset); 306 mmiotrace_ioremap(unaligned_phys_addr, unaligned_size, ret_addr); 307 308 /* 309 * Check if the request spans more than any BAR in the iomem resource 310 * tree. 311 */ 312 if (iomem_map_sanity_check(unaligned_phys_addr, unaligned_size)) 313 pr_warn("caller %pS mapping multiple BARs\n", caller); 314 315 return ret_addr; 316 err_free_area: 317 free_vm_area(area); 318 err_free_memtype: 319 memtype_free(phys_addr, phys_addr + size); 320 return NULL; 321 } 322 323 /** 324 * ioremap - map bus memory into CPU space 325 * @phys_addr: bus address of the memory 326 * @size: size of the resource to map 327 * 328 * ioremap performs a platform specific sequence of operations to 329 * make bus memory CPU accessible via the readb/readw/readl/writeb/ 330 * writew/writel functions and the other mmio helpers. The returned 331 * address is not guaranteed to be usable directly as a virtual 332 * address. 333 * 334 * This version of ioremap ensures that the memory is marked uncachable 335 * on the CPU as well as honouring existing caching rules from things like 336 * the PCI bus. Note that there are other caches and buffers on many 337 * busses. In particular driver authors should read up on PCI writes 338 * 339 * It's useful if some control registers are in such an area and 340 * write combining or read caching is not desirable: 341 * 342 * Must be freed with iounmap. 343 */ 344 void __iomem *ioremap(resource_size_t phys_addr, unsigned long size) 345 { 346 /* 347 * Ideally, this should be: 348 * pat_enabled() ? _PAGE_CACHE_MODE_UC : _PAGE_CACHE_MODE_UC_MINUS; 349 * 350 * Till we fix all X drivers to use ioremap_wc(), we will use 351 * UC MINUS. Drivers that are certain they need or can already 352 * be converted over to strong UC can use ioremap_uc(). 353 */ 354 enum page_cache_mode pcm = _PAGE_CACHE_MODE_UC_MINUS; 355 356 return __ioremap_caller(phys_addr, size, pcm, 357 __builtin_return_address(0), false); 358 } 359 EXPORT_SYMBOL(ioremap); 360 361 /** 362 * ioremap_uc - map bus memory into CPU space as strongly uncachable 363 * @phys_addr: bus address of the memory 364 * @size: size of the resource to map 365 * 366 * ioremap_uc performs a platform specific sequence of operations to 367 * make bus memory CPU accessible via the readb/readw/readl/writeb/ 368 * writew/writel functions and the other mmio helpers. The returned 369 * address is not guaranteed to be usable directly as a virtual 370 * address. 371 * 372 * This version of ioremap ensures that the memory is marked with a strong 373 * preference as completely uncachable on the CPU when possible. For non-PAT 374 * systems this ends up setting page-attribute flags PCD=1, PWT=1. For PAT 375 * systems this will set the PAT entry for the pages as strong UC. This call 376 * will honor existing caching rules from things like the PCI bus. Note that 377 * there are other caches and buffers on many busses. In particular driver 378 * authors should read up on PCI writes. 379 * 380 * It's useful if some control registers are in such an area and 381 * write combining or read caching is not desirable: 382 * 383 * Must be freed with iounmap. 384 */ 385 void __iomem *ioremap_uc(resource_size_t phys_addr, unsigned long size) 386 { 387 enum page_cache_mode pcm = _PAGE_CACHE_MODE_UC; 388 389 return __ioremap_caller(phys_addr, size, pcm, 390 __builtin_return_address(0), false); 391 } 392 EXPORT_SYMBOL_GPL(ioremap_uc); 393 394 /** 395 * ioremap_wc - map memory into CPU space write combined 396 * @phys_addr: bus address of the memory 397 * @size: size of the resource to map 398 * 399 * This version of ioremap ensures that the memory is marked write combining. 400 * Write combining allows faster writes to some hardware devices. 401 * 402 * Must be freed with iounmap. 403 */ 404 void __iomem *ioremap_wc(resource_size_t phys_addr, unsigned long size) 405 { 406 return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WC, 407 __builtin_return_address(0), false); 408 } 409 EXPORT_SYMBOL(ioremap_wc); 410 411 /** 412 * ioremap_wt - map memory into CPU space write through 413 * @phys_addr: bus address of the memory 414 * @size: size of the resource to map 415 * 416 * This version of ioremap ensures that the memory is marked write through. 417 * Write through stores data into memory while keeping the cache up-to-date. 418 * 419 * Must be freed with iounmap. 420 */ 421 void __iomem *ioremap_wt(resource_size_t phys_addr, unsigned long size) 422 { 423 return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WT, 424 __builtin_return_address(0), false); 425 } 426 EXPORT_SYMBOL(ioremap_wt); 427 428 void __iomem *ioremap_encrypted(resource_size_t phys_addr, unsigned long size) 429 { 430 return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WB, 431 __builtin_return_address(0), true); 432 } 433 EXPORT_SYMBOL(ioremap_encrypted); 434 435 void __iomem *ioremap_cache(resource_size_t phys_addr, unsigned long size) 436 { 437 return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WB, 438 __builtin_return_address(0), false); 439 } 440 EXPORT_SYMBOL(ioremap_cache); 441 442 void __iomem *ioremap_prot(resource_size_t phys_addr, unsigned long size, 443 unsigned long prot_val) 444 { 445 return __ioremap_caller(phys_addr, size, 446 pgprot2cachemode(__pgprot(prot_val)), 447 __builtin_return_address(0), false); 448 } 449 EXPORT_SYMBOL(ioremap_prot); 450 451 /** 452 * iounmap - Free a IO remapping 453 * @addr: virtual address from ioremap_* 454 * 455 * Caller must ensure there is only one unmapping for the same pointer. 456 */ 457 void iounmap(volatile void __iomem *addr) 458 { 459 struct vm_struct *p, *o; 460 461 if (WARN_ON_ONCE(!is_ioremap_addr((void __force *)addr))) 462 return; 463 464 /* 465 * The PCI/ISA range special-casing was removed from __ioremap() 466 * so this check, in theory, can be removed. However, there are 467 * cases where iounmap() is called for addresses not obtained via 468 * ioremap() (vga16fb for example). Add a warning so that these 469 * cases can be caught and fixed. 470 */ 471 if ((void __force *)addr >= phys_to_virt(ISA_START_ADDRESS) && 472 (void __force *)addr < phys_to_virt(ISA_END_ADDRESS)) { 473 WARN(1, "iounmap() called for ISA range not obtained using ioremap()\n"); 474 return; 475 } 476 477 mmiotrace_iounmap(addr); 478 479 addr = (volatile void __iomem *) 480 (PAGE_MASK & (unsigned long __force)addr); 481 482 /* Use the vm area unlocked, assuming the caller 483 ensures there isn't another iounmap for the same address 484 in parallel. Reuse of the virtual address is prevented by 485 leaving it in the global lists until we're done with it. 486 cpa takes care of the direct mappings. */ 487 p = find_vm_area((void __force *)addr); 488 489 if (!p) { 490 printk(KERN_ERR "iounmap: bad address %p\n", addr); 491 dump_stack(); 492 return; 493 } 494 495 kmsan_iounmap_page_range((unsigned long)addr, 496 (unsigned long)addr + get_vm_area_size(p)); 497 memtype_free(p->phys_addr, p->phys_addr + get_vm_area_size(p)); 498 499 /* Finally remove it */ 500 o = remove_vm_area((void __force *)addr); 501 BUG_ON(p != o || o == NULL); 502 kfree(p); 503 } 504 EXPORT_SYMBOL(iounmap); 505 506 /* 507 * Convert a physical pointer to a virtual kernel pointer for /dev/mem 508 * access 509 */ 510 void *xlate_dev_mem_ptr(phys_addr_t phys) 511 { 512 unsigned long start = phys & PAGE_MASK; 513 unsigned long offset = phys & ~PAGE_MASK; 514 void *vaddr; 515 516 /* memremap() maps if RAM, otherwise falls back to ioremap() */ 517 vaddr = memremap(start, PAGE_SIZE, MEMREMAP_WB); 518 519 /* Only add the offset on success and return NULL if memremap() failed */ 520 if (vaddr) 521 vaddr += offset; 522 523 return vaddr; 524 } 525 526 void unxlate_dev_mem_ptr(phys_addr_t phys, void *addr) 527 { 528 memunmap((void *)((unsigned long)addr & PAGE_MASK)); 529 } 530 531 #ifdef CONFIG_AMD_MEM_ENCRYPT 532 /* 533 * Examine the physical address to determine if it is an area of memory 534 * that should be mapped decrypted. If the memory is not part of the 535 * kernel usable area it was accessed and created decrypted, so these 536 * areas should be mapped decrypted. And since the encryption key can 537 * change across reboots, persistent memory should also be mapped 538 * decrypted. 539 * 540 * If SEV is active, that implies that BIOS/UEFI also ran encrypted so 541 * only persistent memory should be mapped decrypted. 542 */ 543 static bool memremap_should_map_decrypted(resource_size_t phys_addr, 544 unsigned long size) 545 { 546 int is_pmem; 547 548 /* 549 * Check if the address is part of a persistent memory region. 550 * This check covers areas added by E820, EFI and ACPI. 551 */ 552 is_pmem = region_intersects(phys_addr, size, IORESOURCE_MEM, 553 IORES_DESC_PERSISTENT_MEMORY); 554 if (is_pmem != REGION_DISJOINT) 555 return true; 556 557 /* 558 * Check if the non-volatile attribute is set for an EFI 559 * reserved area. 560 */ 561 if (efi_enabled(EFI_BOOT)) { 562 switch (efi_mem_type(phys_addr)) { 563 case EFI_RESERVED_TYPE: 564 if (efi_mem_attributes(phys_addr) & EFI_MEMORY_NV) 565 return true; 566 break; 567 default: 568 break; 569 } 570 } 571 572 /* Check if the address is outside kernel usable area */ 573 switch (e820__get_entry_type(phys_addr, phys_addr + size - 1)) { 574 case E820_TYPE_RESERVED: 575 case E820_TYPE_ACPI: 576 case E820_TYPE_NVS: 577 case E820_TYPE_UNUSABLE: 578 /* For SEV, these areas are encrypted */ 579 if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) 580 break; 581 fallthrough; 582 583 case E820_TYPE_PRAM: 584 return true; 585 default: 586 break; 587 } 588 589 return false; 590 } 591 592 /* 593 * Examine the physical address to determine if it is EFI data. Check 594 * it against the boot params structure and EFI tables and memory types. 595 */ 596 static bool memremap_is_efi_data(resource_size_t phys_addr) 597 { 598 u64 paddr; 599 600 /* Check if the address is part of EFI boot/runtime data */ 601 if (!efi_enabled(EFI_BOOT)) 602 return false; 603 604 paddr = boot_params.efi_info.efi_memmap_hi; 605 paddr <<= 32; 606 paddr |= boot_params.efi_info.efi_memmap; 607 if (phys_addr == paddr) 608 return true; 609 610 paddr = boot_params.efi_info.efi_systab_hi; 611 paddr <<= 32; 612 paddr |= boot_params.efi_info.efi_systab; 613 if (phys_addr == paddr) 614 return true; 615 616 if (efi_is_table_address(phys_addr)) 617 return true; 618 619 switch (efi_mem_type(phys_addr)) { 620 case EFI_BOOT_SERVICES_DATA: 621 case EFI_RUNTIME_SERVICES_DATA: 622 return true; 623 default: 624 break; 625 } 626 627 return false; 628 } 629 630 /* 631 * Examine the physical address to determine if it is boot data by checking 632 * it against the boot params setup_data chain. 633 */ 634 static bool __ref __memremap_is_setup_data(resource_size_t phys_addr, bool early) 635 { 636 unsigned int setup_data_sz = sizeof(struct setup_data); 637 struct setup_indirect *indirect; 638 struct setup_data *data; 639 u64 paddr, paddr_next; 640 641 paddr = boot_params.hdr.setup_data; 642 while (paddr) { 643 unsigned int len, size; 644 645 if (phys_addr == paddr) 646 return true; 647 648 if (early) 649 data = early_memremap_decrypted(paddr, setup_data_sz); 650 else 651 data = memremap(paddr, setup_data_sz, MEMREMAP_WB | MEMREMAP_DEC); 652 if (!data) { 653 pr_warn("failed to remap setup_data entry\n"); 654 return false; 655 } 656 657 size = setup_data_sz; 658 659 paddr_next = data->next; 660 len = data->len; 661 662 if ((phys_addr > paddr) && 663 (phys_addr < (paddr + setup_data_sz + len))) { 664 if (early) 665 early_memunmap(data, setup_data_sz); 666 else 667 memunmap(data); 668 return true; 669 } 670 671 if (data->type == SETUP_INDIRECT) { 672 size += len; 673 if (early) { 674 early_memunmap(data, setup_data_sz); 675 data = early_memremap_decrypted(paddr, size); 676 } else { 677 memunmap(data); 678 data = memremap(paddr, size, MEMREMAP_WB | MEMREMAP_DEC); 679 } 680 if (!data) { 681 pr_warn("failed to remap indirect setup_data\n"); 682 return false; 683 } 684 685 indirect = (struct setup_indirect *)data->data; 686 687 if (indirect->type != SETUP_INDIRECT) { 688 paddr = indirect->addr; 689 len = indirect->len; 690 } 691 } 692 693 if (early) 694 early_memunmap(data, size); 695 else 696 memunmap(data); 697 698 if ((phys_addr > paddr) && (phys_addr < (paddr + len))) 699 return true; 700 701 paddr = paddr_next; 702 } 703 704 return false; 705 } 706 707 static bool memremap_is_setup_data(resource_size_t phys_addr) 708 { 709 return __memremap_is_setup_data(phys_addr, false); 710 } 711 712 static bool __init early_memremap_is_setup_data(resource_size_t phys_addr) 713 { 714 return __memremap_is_setup_data(phys_addr, true); 715 } 716 717 /* 718 * Architecture function to determine if RAM remap is allowed. By default, a 719 * RAM remap will map the data as encrypted. Determine if a RAM remap should 720 * not be done so that the data will be mapped decrypted. 721 */ 722 bool arch_memremap_can_ram_remap(resource_size_t phys_addr, unsigned long size, 723 unsigned long flags) 724 { 725 if (!cc_platform_has(CC_ATTR_MEM_ENCRYPT)) 726 return true; 727 728 if (flags & MEMREMAP_ENC) 729 return true; 730 731 if (flags & MEMREMAP_DEC) 732 return false; 733 734 if (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT)) { 735 if (memremap_is_setup_data(phys_addr) || 736 memremap_is_efi_data(phys_addr)) 737 return false; 738 } 739 740 return !memremap_should_map_decrypted(phys_addr, size); 741 } 742 743 /* 744 * Architecture override of __weak function to adjust the protection attributes 745 * used when remapping memory. By default, early_memremap() will map the data 746 * as encrypted. Determine if an encrypted mapping should not be done and set 747 * the appropriate protection attributes. 748 */ 749 pgprot_t __init early_memremap_pgprot_adjust(resource_size_t phys_addr, 750 unsigned long size, 751 pgprot_t prot) 752 { 753 bool encrypted_prot; 754 755 if (!cc_platform_has(CC_ATTR_MEM_ENCRYPT)) 756 return prot; 757 758 encrypted_prot = true; 759 760 if (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT)) { 761 if (early_memremap_is_setup_data(phys_addr) || 762 memremap_is_efi_data(phys_addr)) 763 encrypted_prot = false; 764 } 765 766 if (encrypted_prot && memremap_should_map_decrypted(phys_addr, size)) 767 encrypted_prot = false; 768 769 return encrypted_prot ? pgprot_encrypted(prot) 770 : pgprot_decrypted(prot); 771 } 772 773 bool phys_mem_access_encrypted(unsigned long phys_addr, unsigned long size) 774 { 775 return arch_memremap_can_ram_remap(phys_addr, size, 0); 776 } 777 778 /* Remap memory with encryption */ 779 void __init *early_memremap_encrypted(resource_size_t phys_addr, 780 unsigned long size) 781 { 782 return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_ENC); 783 } 784 785 /* 786 * Remap memory with encryption and write-protected - cannot be called 787 * before pat_init() is called 788 */ 789 void __init *early_memremap_encrypted_wp(resource_size_t phys_addr, 790 unsigned long size) 791 { 792 if (!x86_has_pat_wp()) 793 return NULL; 794 return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_ENC_WP); 795 } 796 797 /* Remap memory without encryption */ 798 void __init *early_memremap_decrypted(resource_size_t phys_addr, 799 unsigned long size) 800 { 801 return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_NOENC); 802 } 803 804 /* 805 * Remap memory without encryption and write-protected - cannot be called 806 * before pat_init() is called 807 */ 808 void __init *early_memremap_decrypted_wp(resource_size_t phys_addr, 809 unsigned long size) 810 { 811 if (!x86_has_pat_wp()) 812 return NULL; 813 return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_NOENC_WP); 814 } 815 #endif /* CONFIG_AMD_MEM_ENCRYPT */ 816 817 static pte_t bm_pte[PAGE_SIZE/sizeof(pte_t)] __page_aligned_bss; 818 819 static inline pmd_t * __init early_ioremap_pmd(unsigned long addr) 820 { 821 /* Don't assume we're using swapper_pg_dir at this point */ 822 pgd_t *base = __va(read_cr3_pa()); 823 pgd_t *pgd = &base[pgd_index(addr)]; 824 p4d_t *p4d = p4d_offset(pgd, addr); 825 pud_t *pud = pud_offset(p4d, addr); 826 pmd_t *pmd = pmd_offset(pud, addr); 827 828 return pmd; 829 } 830 831 static inline pte_t * __init early_ioremap_pte(unsigned long addr) 832 { 833 return &bm_pte[pte_index(addr)]; 834 } 835 836 bool __init is_early_ioremap_ptep(pte_t *ptep) 837 { 838 return ptep >= &bm_pte[0] && ptep < &bm_pte[PAGE_SIZE/sizeof(pte_t)]; 839 } 840 841 void __init early_ioremap_init(void) 842 { 843 pmd_t *pmd; 844 845 #ifdef CONFIG_X86_64 846 BUILD_BUG_ON((fix_to_virt(0) + PAGE_SIZE) & ((1 << PMD_SHIFT) - 1)); 847 #else 848 WARN_ON((fix_to_virt(0) + PAGE_SIZE) & ((1 << PMD_SHIFT) - 1)); 849 #endif 850 851 early_ioremap_setup(); 852 853 pmd = early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)); 854 memset(bm_pte, 0, sizeof(bm_pte)); 855 pmd_populate_kernel(&init_mm, pmd, bm_pte); 856 857 /* 858 * The boot-ioremap range spans multiple pmds, for which 859 * we are not prepared: 860 */ 861 #define __FIXADDR_TOP (-PAGE_SIZE) 862 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 863 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 864 #undef __FIXADDR_TOP 865 if (pmd != early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END))) { 866 WARN_ON(1); 867 printk(KERN_WARNING "pmd %p != %p\n", 868 pmd, early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END))); 869 printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 870 fix_to_virt(FIX_BTMAP_BEGIN)); 871 printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_END): %08lx\n", 872 fix_to_virt(FIX_BTMAP_END)); 873 874 printk(KERN_WARNING "FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 875 printk(KERN_WARNING "FIX_BTMAP_BEGIN: %d\n", 876 FIX_BTMAP_BEGIN); 877 } 878 } 879 880 void __init __early_set_fixmap(enum fixed_addresses idx, 881 phys_addr_t phys, pgprot_t flags) 882 { 883 unsigned long addr = __fix_to_virt(idx); 884 pte_t *pte; 885 886 if (idx >= __end_of_fixed_addresses) { 887 BUG(); 888 return; 889 } 890 pte = early_ioremap_pte(addr); 891 892 /* Sanitize 'prot' against any unsupported bits: */ 893 pgprot_val(flags) &= __supported_pte_mask; 894 895 if (pgprot_val(flags)) 896 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags)); 897 else 898 pte_clear(&init_mm, addr, pte); 899 flush_tlb_one_kernel(addr); 900 } 901