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 */
37struct 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 */
45int 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(addr: vaddr, numpages: nrpages);
55 break;
56 case _PAGE_CACHE_MODE_WC:
57 err = _set_memory_wc(addr: vaddr, numpages: nrpages);
58 break;
59 case _PAGE_CACHE_MODE_WT:
60 err = _set_memory_wt(addr: vaddr, numpages: nrpages);
61 break;
62 case _PAGE_CACHE_MODE_WB:
63 err = _set_memory_wb(addr: vaddr, numpages: nrpages);
64 break;
65 }
66
67 return err;
68}
69
70/* Does the range (or a subset of) contain normal RAM? */
71static unsigned int __ioremap_check_ram(struct resource *res)
72{
73 unsigned long start_pfn, stop_pfn;
74 unsigned long pfn;
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_each_valid_pfn(pfn, start_pfn, stop_pfn)
83 if (!PageReserved(pfn_to_page(pfn)))
84 return IORES_MAP_SYSTEM_RAM;
85 }
86
87 return 0;
88}
89
90/*
91 * In a SEV guest, NONE and RESERVED should not be mapped encrypted because
92 * there the whole memory is already encrypted.
93 */
94static unsigned int __ioremap_check_encrypted(struct resource *res)
95{
96 if (!cc_platform_has(attr: CC_ATTR_GUEST_MEM_ENCRYPT))
97 return 0;
98
99 switch (res->desc) {
100 case IORES_DESC_NONE:
101 case IORES_DESC_RESERVED:
102 break;
103 default:
104 return IORES_MAP_ENCRYPTED;
105 }
106
107 return 0;
108}
109
110/*
111 * The EFI runtime services data area is not covered by walk_mem_res(), but must
112 * be mapped encrypted when SEV is active.
113 */
114static void __ioremap_check_other(resource_size_t addr, struct ioremap_desc *desc)
115{
116 if (!cc_platform_has(attr: CC_ATTR_GUEST_MEM_ENCRYPT))
117 return;
118
119 if (x86_platform.hyper.is_private_mmio(addr)) {
120 desc->flags |= IORES_MAP_ENCRYPTED;
121 return;
122 }
123
124 if (!IS_ENABLED(CONFIG_EFI))
125 return;
126
127 if (efi_mem_type(phys_addr: addr) == EFI_RUNTIME_SERVICES_DATA ||
128 (efi_mem_type(phys_addr: addr) == EFI_BOOT_SERVICES_DATA &&
129 efi_mem_attributes(phys_addr: addr) & EFI_MEMORY_RUNTIME))
130 desc->flags |= IORES_MAP_ENCRYPTED;
131}
132
133static int __ioremap_collect_map_flags(struct resource *res, void *arg)
134{
135 struct ioremap_desc *desc = arg;
136
137 if (!(desc->flags & IORES_MAP_SYSTEM_RAM))
138 desc->flags |= __ioremap_check_ram(res);
139
140 if (!(desc->flags & IORES_MAP_ENCRYPTED))
141 desc->flags |= __ioremap_check_encrypted(res);
142
143 return ((desc->flags & (IORES_MAP_SYSTEM_RAM | IORES_MAP_ENCRYPTED)) ==
144 (IORES_MAP_SYSTEM_RAM | IORES_MAP_ENCRYPTED));
145}
146
147/*
148 * To avoid multiple resource walks, this function walks resources marked as
149 * IORESOURCE_MEM and IORESOURCE_BUSY and looking for system RAM and/or a
150 * resource described not as IORES_DESC_NONE (e.g. IORES_DESC_ACPI_TABLES).
151 *
152 * After that, deal with misc other ranges in __ioremap_check_other() which do
153 * not fall into the above category.
154 */
155static void __ioremap_check_mem(resource_size_t addr, unsigned long size,
156 struct ioremap_desc *desc)
157{
158 u64 start, end;
159
160 start = (u64)addr;
161 end = start + size - 1;
162 memset(s: desc, c: 0, n: sizeof(struct ioremap_desc));
163
164 walk_mem_res(start, end, arg: desc, func: __ioremap_collect_map_flags);
165
166 __ioremap_check_other(addr, desc);
167}
168
169/*
170 * Remap an arbitrary physical address space into the kernel virtual
171 * address space. It transparently creates kernel huge I/O mapping when
172 * the physical address is aligned by a huge page size (1GB or 2MB) and
173 * the requested size is at least the huge page size.
174 *
175 * NOTE: MTRRs can override PAT memory types with a 4KB granularity.
176 * Therefore, the mapping code falls back to use a smaller page toward 4KB
177 * when a mapping range is covered by non-WB type of MTRRs.
178 *
179 * NOTE! We need to allow non-page-aligned mappings too: we will obviously
180 * have to convert them into an offset in a page-aligned mapping, but the
181 * caller shouldn't need to know that small detail.
182 */
183static void __iomem *
184__ioremap_caller(resource_size_t phys_addr, unsigned long size,
185 enum page_cache_mode pcm, void *caller, bool encrypted)
186{
187 unsigned long offset, vaddr;
188 resource_size_t last_addr;
189 const resource_size_t unaligned_phys_addr = phys_addr;
190 const unsigned long unaligned_size = size;
191 struct ioremap_desc io_desc;
192 struct vm_struct *area;
193 enum page_cache_mode new_pcm;
194 pgprot_t prot;
195 int retval;
196 void __iomem *ret_addr;
197
198 /* Don't allow wraparound or zero size */
199 last_addr = phys_addr + size - 1;
200 if (!size || last_addr < phys_addr)
201 return NULL;
202
203 if (!phys_addr_valid(addr: phys_addr)) {
204 printk(KERN_WARNING "ioremap: invalid physical address %llx\n",
205 (unsigned long long)phys_addr);
206 WARN_ON_ONCE(1);
207 return NULL;
208 }
209
210 __ioremap_check_mem(addr: phys_addr, size, desc: &io_desc);
211
212 /*
213 * Don't allow anybody to remap normal RAM that we're using..
214 */
215 if (io_desc.flags & IORES_MAP_SYSTEM_RAM) {
216 WARN_ONCE(1, "ioremap on RAM at %pa - %pa\n",
217 &phys_addr, &last_addr);
218 return NULL;
219 }
220
221 /*
222 * Mappings have to be page-aligned
223 */
224 offset = phys_addr & ~PAGE_MASK;
225 phys_addr &= PAGE_MASK;
226 size = PAGE_ALIGN(last_addr+1) - phys_addr;
227
228 /*
229 * Mask out any bits not part of the actual physical
230 * address, like memory encryption bits.
231 */
232 phys_addr &= PHYSICAL_PAGE_MASK;
233
234 retval = memtype_reserve(start: phys_addr, end: (u64)phys_addr + size,
235 req_pcm: pcm, ret_pcm: &new_pcm);
236 if (retval) {
237 printk(KERN_ERR "ioremap memtype_reserve failed %d\n", retval);
238 return NULL;
239 }
240
241 if (pcm != new_pcm) {
242 if (!is_new_memtype_allowed(paddr: phys_addr, size, pcm, new_pcm)) {
243 printk(KERN_ERR
244 "ioremap error for 0x%llx-0x%llx, requested 0x%x, got 0x%x\n",
245 (unsigned long long)phys_addr,
246 (unsigned long long)(phys_addr + size),
247 pcm, new_pcm);
248 goto err_free_memtype;
249 }
250 pcm = new_pcm;
251 }
252
253 /*
254 * If the page being mapped is in memory and SEV is active then
255 * make sure the memory encryption attribute is enabled in the
256 * resulting mapping.
257 * In TDX guests, memory is marked private by default. If encryption
258 * is not requested (using encrypted), explicitly set decrypt
259 * attribute in all IOREMAPPED memory.
260 */
261 prot = PAGE_KERNEL_IO;
262 if ((io_desc.flags & IORES_MAP_ENCRYPTED) || encrypted)
263 prot = pgprot_encrypted(prot);
264 else
265 prot = pgprot_decrypted(prot);
266
267 switch (pcm) {
268 case _PAGE_CACHE_MODE_UC:
269 default:
270 prot = __pgprot(pgprot_val(prot) |
271 cachemode2protval(_PAGE_CACHE_MODE_UC));
272 break;
273 case _PAGE_CACHE_MODE_UC_MINUS:
274 prot = __pgprot(pgprot_val(prot) |
275 cachemode2protval(_PAGE_CACHE_MODE_UC_MINUS));
276 break;
277 case _PAGE_CACHE_MODE_WC:
278 prot = __pgprot(pgprot_val(prot) |
279 cachemode2protval(_PAGE_CACHE_MODE_WC));
280 break;
281 case _PAGE_CACHE_MODE_WT:
282 prot = __pgprot(pgprot_val(prot) |
283 cachemode2protval(_PAGE_CACHE_MODE_WT));
284 break;
285 case _PAGE_CACHE_MODE_WB:
286 break;
287 }
288
289 /*
290 * Ok, go for it..
291 */
292 area = get_vm_area_caller(size, VM_IOREMAP, caller);
293 if (!area)
294 goto err_free_memtype;
295 area->phys_addr = phys_addr;
296 vaddr = (unsigned long) area->addr;
297
298 if (memtype_kernel_map_sync(base: phys_addr, size, pcm))
299 goto err_free_area;
300
301 if (ioremap_page_range(addr: vaddr, end: vaddr + size, phys_addr, prot))
302 goto err_free_area;
303
304 ret_addr = (void __iomem *) (vaddr + offset);
305 mmiotrace_ioremap(offset: unaligned_phys_addr, size: unaligned_size, addr: ret_addr);
306
307 /*
308 * Check if the request spans more than any BAR in the iomem resource
309 * tree.
310 */
311 if (iomem_map_sanity_check(addr: unaligned_phys_addr, size: unaligned_size))
312 pr_warn("caller %pS mapping multiple BARs\n", caller);
313
314 return ret_addr;
315err_free_area:
316 free_vm_area(area);
317err_free_memtype:
318 memtype_free(start: phys_addr, end: phys_addr + size);
319 return NULL;
320}
321
322/**
323 * ioremap - map bus memory into CPU space
324 * @phys_addr: bus address of the memory
325 * @size: size of the resource to map
326 *
327 * ioremap performs a platform specific sequence of operations to
328 * make bus memory CPU accessible via the readb/readw/readl/writeb/
329 * writew/writel functions and the other mmio helpers. The returned
330 * address is not guaranteed to be usable directly as a virtual
331 * address.
332 *
333 * This version of ioremap ensures that the memory is marked uncachable
334 * on the CPU as well as honouring existing caching rules from things like
335 * the PCI bus. Note that there are other caches and buffers on many
336 * busses. In particular driver authors should read up on PCI writes
337 *
338 * It's useful if some control registers are in such an area and
339 * write combining or read caching is not desirable:
340 *
341 * Must be freed with iounmap.
342 */
343void __iomem *ioremap(resource_size_t phys_addr, unsigned long size)
344{
345 /*
346 * Ideally, this should be:
347 * pat_enabled() ? _PAGE_CACHE_MODE_UC : _PAGE_CACHE_MODE_UC_MINUS;
348 *
349 * Till we fix all X drivers to use ioremap_wc(), we will use
350 * UC MINUS. Drivers that are certain they need or can already
351 * be converted over to strong UC can use ioremap_uc().
352 */
353 enum page_cache_mode pcm = _PAGE_CACHE_MODE_UC_MINUS;
354
355 return __ioremap_caller(phys_addr, size, pcm,
356 caller: __builtin_return_address(0), encrypted: false);
357}
358EXPORT_SYMBOL(ioremap);
359
360/**
361 * ioremap_uc - map bus memory into CPU space as strongly uncachable
362 * @phys_addr: bus address of the memory
363 * @size: size of the resource to map
364 *
365 * ioremap_uc performs a platform specific sequence of operations to
366 * make bus memory CPU accessible via the readb/readw/readl/writeb/
367 * writew/writel functions and the other mmio helpers. The returned
368 * address is not guaranteed to be usable directly as a virtual
369 * address.
370 *
371 * This version of ioremap ensures that the memory is marked with a strong
372 * preference as completely uncachable on the CPU when possible. For non-PAT
373 * systems this ends up setting page-attribute flags PCD=1, PWT=1. For PAT
374 * systems this will set the PAT entry for the pages as strong UC. This call
375 * will honor existing caching rules from things like the PCI bus. Note that
376 * there are other caches and buffers on many busses. In particular driver
377 * authors should read up on PCI writes.
378 *
379 * It's useful if some control registers are in such an area and
380 * write combining or read caching is not desirable:
381 *
382 * Must be freed with iounmap.
383 */
384void __iomem *ioremap_uc(resource_size_t phys_addr, unsigned long size)
385{
386 enum page_cache_mode pcm = _PAGE_CACHE_MODE_UC;
387
388 return __ioremap_caller(phys_addr, size, pcm,
389 caller: __builtin_return_address(0), encrypted: false);
390}
391EXPORT_SYMBOL_GPL(ioremap_uc);
392
393/**
394 * ioremap_wc - map memory into CPU space write combined
395 * @phys_addr: bus address of the memory
396 * @size: size of the resource to map
397 *
398 * This version of ioremap ensures that the memory is marked write combining.
399 * Write combining allows faster writes to some hardware devices.
400 *
401 * Must be freed with iounmap.
402 */
403void __iomem *ioremap_wc(resource_size_t phys_addr, unsigned long size)
404{
405 return __ioremap_caller(phys_addr, size, pcm: _PAGE_CACHE_MODE_WC,
406 caller: __builtin_return_address(0), encrypted: false);
407}
408EXPORT_SYMBOL(ioremap_wc);
409
410/**
411 * ioremap_wt - map memory into CPU space write through
412 * @phys_addr: bus address of the memory
413 * @size: size of the resource to map
414 *
415 * This version of ioremap ensures that the memory is marked write through.
416 * Write through stores data into memory while keeping the cache up-to-date.
417 *
418 * Must be freed with iounmap.
419 */
420void __iomem *ioremap_wt(resource_size_t phys_addr, unsigned long size)
421{
422 return __ioremap_caller(phys_addr, size, pcm: _PAGE_CACHE_MODE_WT,
423 caller: __builtin_return_address(0), encrypted: false);
424}
425EXPORT_SYMBOL(ioremap_wt);
426
427void __iomem *ioremap_encrypted(resource_size_t phys_addr, unsigned long size)
428{
429 return __ioremap_caller(phys_addr, size, pcm: _PAGE_CACHE_MODE_WB,
430 caller: __builtin_return_address(0), encrypted: true);
431}
432EXPORT_SYMBOL(ioremap_encrypted);
433
434void __iomem *ioremap_cache(resource_size_t phys_addr, unsigned long size)
435{
436 return __ioremap_caller(phys_addr, size, pcm: _PAGE_CACHE_MODE_WB,
437 caller: __builtin_return_address(0), encrypted: false);
438}
439EXPORT_SYMBOL(ioremap_cache);
440
441void __iomem *ioremap_prot(resource_size_t phys_addr, unsigned long size,
442 pgprot_t prot)
443{
444 return __ioremap_caller(phys_addr, size,
445 pcm: pgprot2cachemode(pgprot: prot),
446 caller: __builtin_return_address(0), encrypted: false);
447}
448EXPORT_SYMBOL(ioremap_prot);
449
450/**
451 * iounmap - Free a IO remapping
452 * @addr: virtual address from ioremap_*
453 *
454 * Caller must ensure there is only one unmapping for the same pointer.
455 */
456void iounmap(volatile void __iomem *addr)
457{
458 struct vm_struct *p, *o;
459
460 if (WARN_ON_ONCE(!is_ioremap_addr((void __force *)addr)))
461 return;
462
463 /*
464 * The PCI/ISA range special-casing was removed from __ioremap()
465 * so this check, in theory, can be removed. However, there are
466 * cases where iounmap() is called for addresses not obtained via
467 * ioremap() (vga16fb for example). Add a warning so that these
468 * cases can be caught and fixed.
469 */
470 if ((void __force *)addr >= phys_to_virt(ISA_START_ADDRESS) &&
471 (void __force *)addr < phys_to_virt(ISA_END_ADDRESS)) {
472 WARN(1, "iounmap() called for ISA range not obtained using ioremap()\n");
473 return;
474 }
475
476 mmiotrace_iounmap(addr);
477
478 addr = (volatile void __iomem *)
479 (PAGE_MASK & (unsigned long __force)addr);
480
481 /* Use the vm area unlocked, assuming the caller
482 ensures there isn't another iounmap for the same address
483 in parallel. Reuse of the virtual address is prevented by
484 leaving it in the global lists until we're done with it.
485 cpa takes care of the direct mappings. */
486 p = find_vm_area(addr: (void __force *)addr);
487
488 if (!p) {
489 printk(KERN_ERR "iounmap: bad address %p\n", addr);
490 dump_stack();
491 return;
492 }
493
494 kmsan_iounmap_page_range(start: (unsigned long)addr,
495 end: (unsigned long)addr + get_vm_area_size(area: p));
496 memtype_free(start: p->phys_addr, end: p->phys_addr + get_vm_area_size(area: p));
497
498 /* Finally remove it */
499 o = remove_vm_area(addr: (void __force *)addr);
500 BUG_ON(p != o || o == NULL);
501 kfree(objp: p);
502}
503EXPORT_SYMBOL(iounmap);
504
505void *arch_memremap_wb(phys_addr_t phys_addr, size_t size, unsigned long flags)
506{
507 if ((flags & MEMREMAP_DEC) || cc_platform_has(attr: CC_ATTR_HOST_MEM_ENCRYPT))
508 return (void __force *)ioremap_cache(phys_addr, size);
509
510 return (void __force *)ioremap_encrypted(phys_addr, size);
511}
512
513/*
514 * Convert a physical pointer to a virtual kernel pointer for /dev/mem
515 * access
516 */
517void *xlate_dev_mem_ptr(phys_addr_t phys)
518{
519 unsigned long start = phys & PAGE_MASK;
520 unsigned long offset = phys & ~PAGE_MASK;
521 void *vaddr;
522
523 /* memremap() maps if RAM, otherwise falls back to ioremap() */
524 vaddr = memremap(offset: start, PAGE_SIZE, flags: MEMREMAP_WB);
525
526 /* Only add the offset on success and return NULL if memremap() failed */
527 if (vaddr)
528 vaddr += offset;
529
530 return vaddr;
531}
532
533void unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
534{
535 memunmap(addr: (void *)((unsigned long)addr & PAGE_MASK));
536}
537
538#ifdef CONFIG_AMD_MEM_ENCRYPT
539/*
540 * Examine the physical address to determine if it is an area of memory
541 * that should be mapped decrypted. If the memory is not part of the
542 * kernel usable area it was accessed and created decrypted, so these
543 * areas should be mapped decrypted. And since the encryption key can
544 * change across reboots, persistent memory should also be mapped
545 * decrypted.
546 *
547 * If SEV is active, that implies that BIOS/UEFI also ran encrypted so
548 * only persistent memory should be mapped decrypted.
549 */
550static bool memremap_should_map_decrypted(resource_size_t phys_addr,
551 unsigned long size)
552{
553 int is_pmem;
554
555 /*
556 * Check if the address is part of a persistent memory region.
557 * This check covers areas added by E820, EFI and ACPI.
558 */
559 is_pmem = region_intersects(offset: phys_addr, size, IORESOURCE_MEM,
560 desc: IORES_DESC_PERSISTENT_MEMORY);
561 if (is_pmem != REGION_DISJOINT)
562 return true;
563
564 /*
565 * Check if the non-volatile attribute is set for an EFI
566 * reserved area.
567 */
568 if (efi_enabled(EFI_BOOT)) {
569 switch (efi_mem_type(phys_addr)) {
570 case EFI_RESERVED_TYPE:
571 if (efi_mem_attributes(phys_addr) & EFI_MEMORY_NV)
572 return true;
573 break;
574 default:
575 break;
576 }
577 }
578
579 /* Check if the address is outside kernel usable area */
580 switch (e820__get_entry_type(start: phys_addr, end: phys_addr + size - 1)) {
581 case E820_TYPE_RESERVED:
582 case E820_TYPE_ACPI:
583 case E820_TYPE_NVS:
584 case E820_TYPE_UNUSABLE:
585 /* For SEV, these areas are encrypted */
586 if (cc_platform_has(attr: CC_ATTR_GUEST_MEM_ENCRYPT))
587 break;
588 fallthrough;
589
590 case E820_TYPE_PRAM:
591 return true;
592 default:
593 break;
594 }
595
596 return false;
597}
598
599/*
600 * Examine the physical address to determine if it is EFI data. Check
601 * it against the boot params structure and EFI tables and memory types.
602 */
603static bool memremap_is_efi_data(resource_size_t phys_addr)
604{
605 u64 paddr;
606
607 /* Check if the address is part of EFI boot/runtime data */
608 if (!efi_enabled(EFI_BOOT))
609 return false;
610
611 paddr = boot_params.efi_info.efi_memmap_hi;
612 paddr <<= 32;
613 paddr |= boot_params.efi_info.efi_memmap;
614 if (phys_addr == paddr)
615 return true;
616
617 paddr = boot_params.efi_info.efi_systab_hi;
618 paddr <<= 32;
619 paddr |= boot_params.efi_info.efi_systab;
620 if (phys_addr == paddr)
621 return true;
622
623 if (efi_is_table_address(phys_addr))
624 return true;
625
626 switch (efi_mem_type(phys_addr)) {
627 case EFI_BOOT_SERVICES_DATA:
628 case EFI_RUNTIME_SERVICES_DATA:
629 return true;
630 default:
631 break;
632 }
633
634 return false;
635}
636
637/*
638 * Examine the physical address to determine if it is boot data by checking
639 * it against the boot params setup_data chain.
640 */
641static bool __ref __memremap_is_setup_data(resource_size_t phys_addr, bool early)
642{
643 unsigned int setup_data_sz = sizeof(struct setup_data);
644 struct setup_indirect *indirect;
645 struct setup_data *data;
646 u64 paddr, paddr_next;
647
648 paddr = boot_params.hdr.setup_data;
649 while (paddr) {
650 unsigned int len, size;
651
652 if (phys_addr == paddr)
653 return true;
654
655 if (early)
656 data = early_memremap_decrypted(phys_addr: paddr, size: setup_data_sz);
657 else
658 data = memremap(offset: paddr, size: setup_data_sz, flags: MEMREMAP_WB | MEMREMAP_DEC);
659 if (!data) {
660 pr_warn("failed to remap setup_data entry\n");
661 return false;
662 }
663
664 size = setup_data_sz;
665
666 paddr_next = data->next;
667 len = data->len;
668
669 if ((phys_addr > paddr) &&
670 (phys_addr < (paddr + setup_data_sz + len))) {
671 if (early)
672 early_memunmap(addr: data, size: setup_data_sz);
673 else
674 memunmap(addr: data);
675 return true;
676 }
677
678 if (data->type == SETUP_INDIRECT) {
679 size += len;
680 if (early) {
681 early_memunmap(addr: data, size: setup_data_sz);
682 data = early_memremap_decrypted(phys_addr: paddr, size);
683 } else {
684 memunmap(addr: data);
685 data = memremap(offset: paddr, size, flags: MEMREMAP_WB | MEMREMAP_DEC);
686 }
687 if (!data) {
688 pr_warn("failed to remap indirect setup_data\n");
689 return false;
690 }
691
692 indirect = (struct setup_indirect *)data->data;
693
694 if (indirect->type != SETUP_INDIRECT) {
695 paddr = indirect->addr;
696 len = indirect->len;
697 }
698 }
699
700 if (early)
701 early_memunmap(addr: data, size);
702 else
703 memunmap(addr: data);
704
705 if ((phys_addr > paddr) && (phys_addr < (paddr + len)))
706 return true;
707
708 paddr = paddr_next;
709 }
710
711 return false;
712}
713
714static bool memremap_is_setup_data(resource_size_t phys_addr)
715{
716 return __memremap_is_setup_data(phys_addr, early: false);
717}
718
719static bool __init early_memremap_is_setup_data(resource_size_t phys_addr)
720{
721 return __memremap_is_setup_data(phys_addr, early: true);
722}
723
724/*
725 * Architecture function to determine if RAM remap is allowed. By default, a
726 * RAM remap will map the data as encrypted. Determine if a RAM remap should
727 * not be done so that the data will be mapped decrypted.
728 */
729bool arch_memremap_can_ram_remap(resource_size_t phys_addr, unsigned long size,
730 unsigned long flags)
731{
732 if (!cc_platform_has(attr: CC_ATTR_MEM_ENCRYPT))
733 return true;
734
735 if (flags & MEMREMAP_ENC)
736 return true;
737
738 if (flags & MEMREMAP_DEC)
739 return false;
740
741 if (cc_platform_has(attr: CC_ATTR_HOST_MEM_ENCRYPT)) {
742 if (memremap_is_setup_data(phys_addr) ||
743 memremap_is_efi_data(phys_addr))
744 return false;
745 }
746
747 return !memremap_should_map_decrypted(phys_addr, size);
748}
749
750/*
751 * Architecture override of __weak function to adjust the protection attributes
752 * used when remapping memory. By default, early_memremap() will map the data
753 * as encrypted. Determine if an encrypted mapping should not be done and set
754 * the appropriate protection attributes.
755 */
756pgprot_t __init early_memremap_pgprot_adjust(resource_size_t phys_addr,
757 unsigned long size,
758 pgprot_t prot)
759{
760 bool encrypted_prot;
761
762 if (!cc_platform_has(attr: CC_ATTR_MEM_ENCRYPT))
763 return prot;
764
765 encrypted_prot = true;
766
767 if (cc_platform_has(attr: CC_ATTR_HOST_MEM_ENCRYPT)) {
768 if (early_memremap_is_setup_data(phys_addr) ||
769 memremap_is_efi_data(phys_addr))
770 encrypted_prot = false;
771 }
772
773 if (encrypted_prot && memremap_should_map_decrypted(phys_addr, size))
774 encrypted_prot = false;
775
776 return encrypted_prot ? pgprot_encrypted(prot)
777 : pgprot_decrypted(prot);
778}
779
780bool phys_mem_access_encrypted(unsigned long phys_addr, unsigned long size)
781{
782 return arch_memremap_can_ram_remap(phys_addr, size, flags: 0);
783}
784
785/* Remap memory with encryption */
786void __init *early_memremap_encrypted(resource_size_t phys_addr,
787 unsigned long size)
788{
789 return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_ENC);
790}
791
792/*
793 * Remap memory with encryption and write-protected - cannot be called
794 * before pat_init() is called
795 */
796void __init *early_memremap_encrypted_wp(resource_size_t phys_addr,
797 unsigned long size)
798{
799 if (!x86_has_pat_wp())
800 return NULL;
801 return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_ENC_WP);
802}
803
804/* Remap memory without encryption */
805void __init *early_memremap_decrypted(resource_size_t phys_addr,
806 unsigned long size)
807{
808 return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_NOENC);
809}
810
811/*
812 * Remap memory without encryption and write-protected - cannot be called
813 * before pat_init() is called
814 */
815void __init *early_memremap_decrypted_wp(resource_size_t phys_addr,
816 unsigned long size)
817{
818 if (!x86_has_pat_wp())
819 return NULL;
820 return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_NOENC_WP);
821}
822#endif /* CONFIG_AMD_MEM_ENCRYPT */
823
824static pte_t bm_pte[PAGE_SIZE/sizeof(pte_t)] __page_aligned_bss;
825
826static inline pmd_t * __init early_ioremap_pmd(unsigned long addr)
827{
828 /* Don't assume we're using swapper_pg_dir at this point */
829 pgd_t *base = __va(read_cr3_pa());
830 pgd_t *pgd = &base[pgd_index(addr)];
831 p4d_t *p4d = p4d_offset(pgd, address: addr);
832 pud_t *pud = pud_offset(p4d, address: addr);
833 pmd_t *pmd = pmd_offset(pud, address: addr);
834
835 return pmd;
836}
837
838static inline pte_t * __init early_ioremap_pte(unsigned long addr)
839{
840 return &bm_pte[pte_index(address: addr)];
841}
842
843bool __init is_early_ioremap_ptep(pte_t *ptep)
844{
845 return ptep >= &bm_pte[0] && ptep < &bm_pte[PAGE_SIZE/sizeof(pte_t)];
846}
847
848void __init early_ioremap_init(void)
849{
850 pmd_t *pmd;
851
852#ifdef CONFIG_X86_64
853 BUILD_BUG_ON((fix_to_virt(0) + PAGE_SIZE) & ((1 << PMD_SHIFT) - 1));
854#else
855 WARN_ON((fix_to_virt(0) + PAGE_SIZE) & ((1 << PMD_SHIFT) - 1));
856#endif
857
858 early_ioremap_setup();
859
860 pmd = early_ioremap_pmd(addr: fix_to_virt(idx: FIX_BTMAP_BEGIN));
861 memset(s: bm_pte, c: 0, n: sizeof(bm_pte));
862 pmd_populate_kernel(mm: &init_mm, pmd, pte: bm_pte);
863
864 /*
865 * The boot-ioremap range spans multiple pmds, for which
866 * we are not prepared:
867 */
868#define __FIXADDR_TOP (-PAGE_SIZE)
869 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
870 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
871#undef __FIXADDR_TOP
872 if (pmd != early_ioremap_pmd(addr: fix_to_virt(idx: FIX_BTMAP_END))) {
873 WARN_ON(1);
874 printk(KERN_WARNING "pmd %p != %p\n",
875 pmd, early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END)));
876 printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
877 fix_to_virt(FIX_BTMAP_BEGIN));
878 printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_END): %08lx\n",
879 fix_to_virt(FIX_BTMAP_END));
880
881 printk(KERN_WARNING "FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
882 printk(KERN_WARNING "FIX_BTMAP_BEGIN: %d\n",
883 FIX_BTMAP_BEGIN);
884 }
885}
886
887void __init __early_set_fixmap(enum fixed_addresses idx,
888 phys_addr_t phys, pgprot_t flags)
889{
890 unsigned long addr = __fix_to_virt(idx);
891 pte_t *pte;
892
893 if (idx >= __end_of_fixed_addresses) {
894 BUG();
895 return;
896 }
897 pte = early_ioremap_pte(addr);
898
899 /* Sanitize 'prot' against any unsupported bits: */
900 pgprot_val(flags) &= __supported_pte_mask;
901
902 if (pgprot_val(flags))
903 set_pte(ptep: pte, pte: pfn_pte(page_nr: phys >> PAGE_SHIFT, pgprot: flags));
904 else
905 pte_clear(mm: &init_mm, addr, ptep: pte);
906 flush_tlb_one_kernel(addr);
907}
908

source code of linux/arch/x86/mm/ioremap.c