xref: /openbmc/linux/arch/x86/mm/ioremap.c (revision 603c09f2)
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/slab.h>
15 #include <linux/vmalloc.h>
16 #include <linux/mmiotrace.h>
17 #include <linux/cc_platform.h>
18 #include <linux/efi.h>
19 #include <linux/pgtable.h>
20 
21 #include <asm/set_memory.h>
22 #include <asm/e820/api.h>
23 #include <asm/efi.h>
24 #include <asm/fixmap.h>
25 #include <asm/tlbflush.h>
26 #include <asm/pgalloc.h>
27 #include <asm/memtype.h>
28 #include <asm/setup.h>
29 
30 #include "physaddr.h"
31 
32 /*
33  * Descriptor controlling ioremap() behavior.
34  */
35 struct ioremap_desc {
36 	unsigned int flags;
37 };
38 
39 /*
40  * Fix up the linear direct mapping of the kernel to avoid cache attribute
41  * conflicts.
42  */
43 int ioremap_change_attr(unsigned long vaddr, unsigned long size,
44 			enum page_cache_mode pcm)
45 {
46 	unsigned long nrpages = size >> PAGE_SHIFT;
47 	int err;
48 
49 	switch (pcm) {
50 	case _PAGE_CACHE_MODE_UC:
51 	default:
52 		err = _set_memory_uc(vaddr, nrpages);
53 		break;
54 	case _PAGE_CACHE_MODE_WC:
55 		err = _set_memory_wc(vaddr, nrpages);
56 		break;
57 	case _PAGE_CACHE_MODE_WT:
58 		err = _set_memory_wt(vaddr, nrpages);
59 		break;
60 	case _PAGE_CACHE_MODE_WB:
61 		err = _set_memory_wb(vaddr, nrpages);
62 		break;
63 	}
64 
65 	return err;
66 }
67 
68 /* Does the range (or a subset of) contain normal RAM? */
69 static unsigned int __ioremap_check_ram(struct resource *res)
70 {
71 	unsigned long start_pfn, stop_pfn;
72 	unsigned long i;
73 
74 	if ((res->flags & IORESOURCE_SYSTEM_RAM) != IORESOURCE_SYSTEM_RAM)
75 		return 0;
76 
77 	start_pfn = (res->start + PAGE_SIZE - 1) >> PAGE_SHIFT;
78 	stop_pfn = (res->end + 1) >> PAGE_SHIFT;
79 	if (stop_pfn > start_pfn) {
80 		for (i = 0; i < (stop_pfn - start_pfn); ++i)
81 			if (pfn_valid(start_pfn + i) &&
82 			    !PageReserved(pfn_to_page(start_pfn + i)))
83 				return IORES_MAP_SYSTEM_RAM;
84 	}
85 
86 	return 0;
87 }
88 
89 /*
90  * In a SEV guest, NONE and RESERVED should not be mapped encrypted because
91  * there the whole memory is already encrypted.
92  */
93 static unsigned int __ioremap_check_encrypted(struct resource *res)
94 {
95 	if (!cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
96 		return 0;
97 
98 	switch (res->desc) {
99 	case IORES_DESC_NONE:
100 	case IORES_DESC_RESERVED:
101 		break;
102 	default:
103 		return IORES_MAP_ENCRYPTED;
104 	}
105 
106 	return 0;
107 }
108 
109 /*
110  * The EFI runtime services data area is not covered by walk_mem_res(), but must
111  * be mapped encrypted when SEV is active.
112  */
113 static void __ioremap_check_other(resource_size_t addr, struct ioremap_desc *desc)
114 {
115 	if (!cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
116 		return;
117 
118 	if (!IS_ENABLED(CONFIG_EFI))
119 		return;
120 
121 	if (efi_mem_type(addr) == EFI_RUNTIME_SERVICES_DATA ||
122 	    (efi_mem_type(addr) == EFI_BOOT_SERVICES_DATA &&
123 	     efi_mem_attributes(addr) & EFI_MEMORY_RUNTIME))
124 		desc->flags |= IORES_MAP_ENCRYPTED;
125 }
126 
127 static int __ioremap_collect_map_flags(struct resource *res, void *arg)
128 {
129 	struct ioremap_desc *desc = arg;
130 
131 	if (!(desc->flags & IORES_MAP_SYSTEM_RAM))
132 		desc->flags |= __ioremap_check_ram(res);
133 
134 	if (!(desc->flags & IORES_MAP_ENCRYPTED))
135 		desc->flags |= __ioremap_check_encrypted(res);
136 
137 	return ((desc->flags & (IORES_MAP_SYSTEM_RAM | IORES_MAP_ENCRYPTED)) ==
138 			       (IORES_MAP_SYSTEM_RAM | IORES_MAP_ENCRYPTED));
139 }
140 
141 /*
142  * To avoid multiple resource walks, this function walks resources marked as
143  * IORESOURCE_MEM and IORESOURCE_BUSY and looking for system RAM and/or a
144  * resource described not as IORES_DESC_NONE (e.g. IORES_DESC_ACPI_TABLES).
145  *
146  * After that, deal with misc other ranges in __ioremap_check_other() which do
147  * not fall into the above category.
148  */
149 static void __ioremap_check_mem(resource_size_t addr, unsigned long size,
150 				struct ioremap_desc *desc)
151 {
152 	u64 start, end;
153 
154 	start = (u64)addr;
155 	end = start + size - 1;
156 	memset(desc, 0, sizeof(struct ioremap_desc));
157 
158 	walk_mem_res(start, end, desc, __ioremap_collect_map_flags);
159 
160 	__ioremap_check_other(addr, desc);
161 }
162 
163 /*
164  * Remap an arbitrary physical address space into the kernel virtual
165  * address space. It transparently creates kernel huge I/O mapping when
166  * the physical address is aligned by a huge page size (1GB or 2MB) and
167  * the requested size is at least the huge page size.
168  *
169  * NOTE: MTRRs can override PAT memory types with a 4KB granularity.
170  * Therefore, the mapping code falls back to use a smaller page toward 4KB
171  * when a mapping range is covered by non-WB type of MTRRs.
172  *
173  * NOTE! We need to allow non-page-aligned mappings too: we will obviously
174  * have to convert them into an offset in a page-aligned mapping, but the
175  * caller shouldn't need to know that small detail.
176  */
177 static void __iomem *
178 __ioremap_caller(resource_size_t phys_addr, unsigned long size,
179 		 enum page_cache_mode pcm, void *caller, bool encrypted)
180 {
181 	unsigned long offset, vaddr;
182 	resource_size_t last_addr;
183 	const resource_size_t unaligned_phys_addr = phys_addr;
184 	const unsigned long unaligned_size = size;
185 	struct ioremap_desc io_desc;
186 	struct vm_struct *area;
187 	enum page_cache_mode new_pcm;
188 	pgprot_t prot;
189 	int retval;
190 	void __iomem *ret_addr;
191 
192 	/* Don't allow wraparound or zero size */
193 	last_addr = phys_addr + size - 1;
194 	if (!size || last_addr < phys_addr)
195 		return NULL;
196 
197 	if (!phys_addr_valid(phys_addr)) {
198 		printk(KERN_WARNING "ioremap: invalid physical address %llx\n",
199 		       (unsigned long long)phys_addr);
200 		WARN_ON_ONCE(1);
201 		return NULL;
202 	}
203 
204 	__ioremap_check_mem(phys_addr, size, &io_desc);
205 
206 	/*
207 	 * Don't allow anybody to remap normal RAM that we're using..
208 	 */
209 	if (io_desc.flags & IORES_MAP_SYSTEM_RAM) {
210 		WARN_ONCE(1, "ioremap on RAM at %pa - %pa\n",
211 			  &phys_addr, &last_addr);
212 		return NULL;
213 	}
214 
215 	/*
216 	 * Mappings have to be page-aligned
217 	 */
218 	offset = phys_addr & ~PAGE_MASK;
219 	phys_addr &= PHYSICAL_PAGE_MASK;
220 	size = PAGE_ALIGN(last_addr+1) - phys_addr;
221 
222 	retval = memtype_reserve(phys_addr, (u64)phys_addr + size,
223 						pcm, &new_pcm);
224 	if (retval) {
225 		printk(KERN_ERR "ioremap memtype_reserve failed %d\n", retval);
226 		return NULL;
227 	}
228 
229 	if (pcm != new_pcm) {
230 		if (!is_new_memtype_allowed(phys_addr, size, pcm, new_pcm)) {
231 			printk(KERN_ERR
232 		"ioremap error for 0x%llx-0x%llx, requested 0x%x, got 0x%x\n",
233 				(unsigned long long)phys_addr,
234 				(unsigned long long)(phys_addr + size),
235 				pcm, new_pcm);
236 			goto err_free_memtype;
237 		}
238 		pcm = new_pcm;
239 	}
240 
241 	/*
242 	 * If the page being mapped is in memory and SEV is active then
243 	 * make sure the memory encryption attribute is enabled in the
244 	 * resulting mapping.
245 	 * In TDX guests, memory is marked private by default. If encryption
246 	 * is not requested (using encrypted), explicitly set decrypt
247 	 * attribute in all IOREMAPPED memory.
248 	 */
249 	prot = PAGE_KERNEL_IO;
250 	if ((io_desc.flags & IORES_MAP_ENCRYPTED) || encrypted)
251 		prot = pgprot_encrypted(prot);
252 	else
253 		prot = pgprot_decrypted(prot);
254 
255 	switch (pcm) {
256 	case _PAGE_CACHE_MODE_UC:
257 	default:
258 		prot = __pgprot(pgprot_val(prot) |
259 				cachemode2protval(_PAGE_CACHE_MODE_UC));
260 		break;
261 	case _PAGE_CACHE_MODE_UC_MINUS:
262 		prot = __pgprot(pgprot_val(prot) |
263 				cachemode2protval(_PAGE_CACHE_MODE_UC_MINUS));
264 		break;
265 	case _PAGE_CACHE_MODE_WC:
266 		prot = __pgprot(pgprot_val(prot) |
267 				cachemode2protval(_PAGE_CACHE_MODE_WC));
268 		break;
269 	case _PAGE_CACHE_MODE_WT:
270 		prot = __pgprot(pgprot_val(prot) |
271 				cachemode2protval(_PAGE_CACHE_MODE_WT));
272 		break;
273 	case _PAGE_CACHE_MODE_WB:
274 		break;
275 	}
276 
277 	/*
278 	 * Ok, go for it..
279 	 */
280 	area = get_vm_area_caller(size, VM_IOREMAP, caller);
281 	if (!area)
282 		goto err_free_memtype;
283 	area->phys_addr = phys_addr;
284 	vaddr = (unsigned long) area->addr;
285 
286 	if (memtype_kernel_map_sync(phys_addr, size, pcm))
287 		goto err_free_area;
288 
289 	if (ioremap_page_range(vaddr, vaddr + size, phys_addr, prot))
290 		goto err_free_area;
291 
292 	ret_addr = (void __iomem *) (vaddr + offset);
293 	mmiotrace_ioremap(unaligned_phys_addr, unaligned_size, ret_addr);
294 
295 	/*
296 	 * Check if the request spans more than any BAR in the iomem resource
297 	 * tree.
298 	 */
299 	if (iomem_map_sanity_check(unaligned_phys_addr, unaligned_size))
300 		pr_warn("caller %pS mapping multiple BARs\n", caller);
301 
302 	return ret_addr;
303 err_free_area:
304 	free_vm_area(area);
305 err_free_memtype:
306 	memtype_free(phys_addr, phys_addr + size);
307 	return NULL;
308 }
309 
310 /**
311  * ioremap     -   map bus memory into CPU space
312  * @phys_addr:    bus address of the memory
313  * @size:      size of the resource to map
314  *
315  * ioremap performs a platform specific sequence of operations to
316  * make bus memory CPU accessible via the readb/readw/readl/writeb/
317  * writew/writel functions and the other mmio helpers. The returned
318  * address is not guaranteed to be usable directly as a virtual
319  * address.
320  *
321  * This version of ioremap ensures that the memory is marked uncachable
322  * on the CPU as well as honouring existing caching rules from things like
323  * the PCI bus. Note that there are other caches and buffers on many
324  * busses. In particular driver authors should read up on PCI writes
325  *
326  * It's useful if some control registers are in such an area and
327  * write combining or read caching is not desirable:
328  *
329  * Must be freed with iounmap.
330  */
331 void __iomem *ioremap(resource_size_t phys_addr, unsigned long size)
332 {
333 	/*
334 	 * Ideally, this should be:
335 	 *	pat_enabled() ? _PAGE_CACHE_MODE_UC : _PAGE_CACHE_MODE_UC_MINUS;
336 	 *
337 	 * Till we fix all X drivers to use ioremap_wc(), we will use
338 	 * UC MINUS. Drivers that are certain they need or can already
339 	 * be converted over to strong UC can use ioremap_uc().
340 	 */
341 	enum page_cache_mode pcm = _PAGE_CACHE_MODE_UC_MINUS;
342 
343 	return __ioremap_caller(phys_addr, size, pcm,
344 				__builtin_return_address(0), false);
345 }
346 EXPORT_SYMBOL(ioremap);
347 
348 /**
349  * ioremap_uc     -   map bus memory into CPU space as strongly uncachable
350  * @phys_addr:    bus address of the memory
351  * @size:      size of the resource to map
352  *
353  * ioremap_uc performs a platform specific sequence of operations to
354  * make bus memory CPU accessible via the readb/readw/readl/writeb/
355  * writew/writel functions and the other mmio helpers. The returned
356  * address is not guaranteed to be usable directly as a virtual
357  * address.
358  *
359  * This version of ioremap ensures that the memory is marked with a strong
360  * preference as completely uncachable on the CPU when possible. For non-PAT
361  * systems this ends up setting page-attribute flags PCD=1, PWT=1. For PAT
362  * systems this will set the PAT entry for the pages as strong UC.  This call
363  * will honor existing caching rules from things like the PCI bus. Note that
364  * there are other caches and buffers on many busses. In particular driver
365  * authors should read up on PCI writes.
366  *
367  * It's useful if some control registers are in such an area and
368  * write combining or read caching is not desirable:
369  *
370  * Must be freed with iounmap.
371  */
372 void __iomem *ioremap_uc(resource_size_t phys_addr, unsigned long size)
373 {
374 	enum page_cache_mode pcm = _PAGE_CACHE_MODE_UC;
375 
376 	return __ioremap_caller(phys_addr, size, pcm,
377 				__builtin_return_address(0), false);
378 }
379 EXPORT_SYMBOL_GPL(ioremap_uc);
380 
381 /**
382  * ioremap_wc	-	map memory into CPU space write combined
383  * @phys_addr:	bus address of the memory
384  * @size:	size of the resource to map
385  *
386  * This version of ioremap ensures that the memory is marked write combining.
387  * Write combining allows faster writes to some hardware devices.
388  *
389  * Must be freed with iounmap.
390  */
391 void __iomem *ioremap_wc(resource_size_t phys_addr, unsigned long size)
392 {
393 	return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WC,
394 					__builtin_return_address(0), false);
395 }
396 EXPORT_SYMBOL(ioremap_wc);
397 
398 /**
399  * ioremap_wt	-	map memory into CPU space write through
400  * @phys_addr:	bus address of the memory
401  * @size:	size of the resource to map
402  *
403  * This version of ioremap ensures that the memory is marked write through.
404  * Write through stores data into memory while keeping the cache up-to-date.
405  *
406  * Must be freed with iounmap.
407  */
408 void __iomem *ioremap_wt(resource_size_t phys_addr, unsigned long size)
409 {
410 	return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WT,
411 					__builtin_return_address(0), false);
412 }
413 EXPORT_SYMBOL(ioremap_wt);
414 
415 void __iomem *ioremap_encrypted(resource_size_t phys_addr, unsigned long size)
416 {
417 	return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WB,
418 				__builtin_return_address(0), true);
419 }
420 EXPORT_SYMBOL(ioremap_encrypted);
421 
422 void __iomem *ioremap_cache(resource_size_t phys_addr, unsigned long size)
423 {
424 	return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WB,
425 				__builtin_return_address(0), false);
426 }
427 EXPORT_SYMBOL(ioremap_cache);
428 
429 void __iomem *ioremap_prot(resource_size_t phys_addr, unsigned long size,
430 				unsigned long prot_val)
431 {
432 	return __ioremap_caller(phys_addr, size,
433 				pgprot2cachemode(__pgprot(prot_val)),
434 				__builtin_return_address(0), false);
435 }
436 EXPORT_SYMBOL(ioremap_prot);
437 
438 /**
439  * iounmap - Free a IO remapping
440  * @addr: virtual address from ioremap_*
441  *
442  * Caller must ensure there is only one unmapping for the same pointer.
443  */
444 void iounmap(volatile void __iomem *addr)
445 {
446 	struct vm_struct *p, *o;
447 
448 	if ((void __force *)addr <= high_memory)
449 		return;
450 
451 	/*
452 	 * The PCI/ISA range special-casing was removed from __ioremap()
453 	 * so this check, in theory, can be removed. However, there are
454 	 * cases where iounmap() is called for addresses not obtained via
455 	 * ioremap() (vga16fb for example). Add a warning so that these
456 	 * cases can be caught and fixed.
457 	 */
458 	if ((void __force *)addr >= phys_to_virt(ISA_START_ADDRESS) &&
459 	    (void __force *)addr < phys_to_virt(ISA_END_ADDRESS)) {
460 		WARN(1, "iounmap() called for ISA range not obtained using ioremap()\n");
461 		return;
462 	}
463 
464 	mmiotrace_iounmap(addr);
465 
466 	addr = (volatile void __iomem *)
467 		(PAGE_MASK & (unsigned long __force)addr);
468 
469 	/* Use the vm area unlocked, assuming the caller
470 	   ensures there isn't another iounmap for the same address
471 	   in parallel. Reuse of the virtual address is prevented by
472 	   leaving it in the global lists until we're done with it.
473 	   cpa takes care of the direct mappings. */
474 	p = find_vm_area((void __force *)addr);
475 
476 	if (!p) {
477 		printk(KERN_ERR "iounmap: bad address %p\n", addr);
478 		dump_stack();
479 		return;
480 	}
481 
482 	memtype_free(p->phys_addr, p->phys_addr + get_vm_area_size(p));
483 
484 	/* Finally remove it */
485 	o = remove_vm_area((void __force *)addr);
486 	BUG_ON(p != o || o == NULL);
487 	kfree(p);
488 }
489 EXPORT_SYMBOL(iounmap);
490 
491 /*
492  * Convert a physical pointer to a virtual kernel pointer for /dev/mem
493  * access
494  */
495 void *xlate_dev_mem_ptr(phys_addr_t phys)
496 {
497 	unsigned long start  = phys &  PAGE_MASK;
498 	unsigned long offset = phys & ~PAGE_MASK;
499 	void *vaddr;
500 
501 	/* memremap() maps if RAM, otherwise falls back to ioremap() */
502 	vaddr = memremap(start, PAGE_SIZE, MEMREMAP_WB);
503 
504 	/* Only add the offset on success and return NULL if memremap() failed */
505 	if (vaddr)
506 		vaddr += offset;
507 
508 	return vaddr;
509 }
510 
511 void unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
512 {
513 	memunmap((void *)((unsigned long)addr & PAGE_MASK));
514 }
515 
516 #ifdef CONFIG_AMD_MEM_ENCRYPT
517 /*
518  * Examine the physical address to determine if it is an area of memory
519  * that should be mapped decrypted.  If the memory is not part of the
520  * kernel usable area it was accessed and created decrypted, so these
521  * areas should be mapped decrypted. And since the encryption key can
522  * change across reboots, persistent memory should also be mapped
523  * decrypted.
524  *
525  * If SEV is active, that implies that BIOS/UEFI also ran encrypted so
526  * only persistent memory should be mapped decrypted.
527  */
528 static bool memremap_should_map_decrypted(resource_size_t phys_addr,
529 					  unsigned long size)
530 {
531 	int is_pmem;
532 
533 	/*
534 	 * Check if the address is part of a persistent memory region.
535 	 * This check covers areas added by E820, EFI and ACPI.
536 	 */
537 	is_pmem = region_intersects(phys_addr, size, IORESOURCE_MEM,
538 				    IORES_DESC_PERSISTENT_MEMORY);
539 	if (is_pmem != REGION_DISJOINT)
540 		return true;
541 
542 	/*
543 	 * Check if the non-volatile attribute is set for an EFI
544 	 * reserved area.
545 	 */
546 	if (efi_enabled(EFI_BOOT)) {
547 		switch (efi_mem_type(phys_addr)) {
548 		case EFI_RESERVED_TYPE:
549 			if (efi_mem_attributes(phys_addr) & EFI_MEMORY_NV)
550 				return true;
551 			break;
552 		default:
553 			break;
554 		}
555 	}
556 
557 	/* Check if the address is outside kernel usable area */
558 	switch (e820__get_entry_type(phys_addr, phys_addr + size - 1)) {
559 	case E820_TYPE_RESERVED:
560 	case E820_TYPE_ACPI:
561 	case E820_TYPE_NVS:
562 	case E820_TYPE_UNUSABLE:
563 		/* For SEV, these areas are encrypted */
564 		if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
565 			break;
566 		fallthrough;
567 
568 	case E820_TYPE_PRAM:
569 		return true;
570 	default:
571 		break;
572 	}
573 
574 	return false;
575 }
576 
577 /*
578  * Examine the physical address to determine if it is EFI data. Check
579  * it against the boot params structure and EFI tables and memory types.
580  */
581 static bool memremap_is_efi_data(resource_size_t phys_addr,
582 				 unsigned long size)
583 {
584 	u64 paddr;
585 
586 	/* Check if the address is part of EFI boot/runtime data */
587 	if (!efi_enabled(EFI_BOOT))
588 		return false;
589 
590 	paddr = boot_params.efi_info.efi_memmap_hi;
591 	paddr <<= 32;
592 	paddr |= boot_params.efi_info.efi_memmap;
593 	if (phys_addr == paddr)
594 		return true;
595 
596 	paddr = boot_params.efi_info.efi_systab_hi;
597 	paddr <<= 32;
598 	paddr |= boot_params.efi_info.efi_systab;
599 	if (phys_addr == paddr)
600 		return true;
601 
602 	if (efi_is_table_address(phys_addr))
603 		return true;
604 
605 	switch (efi_mem_type(phys_addr)) {
606 	case EFI_BOOT_SERVICES_DATA:
607 	case EFI_RUNTIME_SERVICES_DATA:
608 		return true;
609 	default:
610 		break;
611 	}
612 
613 	return false;
614 }
615 
616 /*
617  * Examine the physical address to determine if it is boot data by checking
618  * it against the boot params setup_data chain.
619  */
620 static bool memremap_is_setup_data(resource_size_t phys_addr,
621 				   unsigned long size)
622 {
623 	struct setup_indirect *indirect;
624 	struct setup_data *data;
625 	u64 paddr, paddr_next;
626 
627 	paddr = boot_params.hdr.setup_data;
628 	while (paddr) {
629 		unsigned int len;
630 
631 		if (phys_addr == paddr)
632 			return true;
633 
634 		data = memremap(paddr, sizeof(*data),
635 				MEMREMAP_WB | MEMREMAP_DEC);
636 		if (!data) {
637 			pr_warn("failed to memremap setup_data entry\n");
638 			return false;
639 		}
640 
641 		paddr_next = data->next;
642 		len = data->len;
643 
644 		if ((phys_addr > paddr) && (phys_addr < (paddr + len))) {
645 			memunmap(data);
646 			return true;
647 		}
648 
649 		if (data->type == SETUP_INDIRECT) {
650 			memunmap(data);
651 			data = memremap(paddr, sizeof(*data) + len,
652 					MEMREMAP_WB | MEMREMAP_DEC);
653 			if (!data) {
654 				pr_warn("failed to memremap indirect setup_data\n");
655 				return false;
656 			}
657 
658 			indirect = (struct setup_indirect *)data->data;
659 
660 			if (indirect->type != SETUP_INDIRECT) {
661 				paddr = indirect->addr;
662 				len = indirect->len;
663 			}
664 		}
665 
666 		memunmap(data);
667 
668 		if ((phys_addr > paddr) && (phys_addr < (paddr + len)))
669 			return true;
670 
671 		paddr = paddr_next;
672 	}
673 
674 	return false;
675 }
676 
677 /*
678  * Examine the physical address to determine if it is boot data by checking
679  * it against the boot params setup_data chain (early boot version).
680  */
681 static bool __init early_memremap_is_setup_data(resource_size_t phys_addr,
682 						unsigned long size)
683 {
684 	struct setup_indirect *indirect;
685 	struct setup_data *data;
686 	u64 paddr, paddr_next;
687 
688 	paddr = boot_params.hdr.setup_data;
689 	while (paddr) {
690 		unsigned int len, size;
691 
692 		if (phys_addr == paddr)
693 			return true;
694 
695 		data = early_memremap_decrypted(paddr, sizeof(*data));
696 		if (!data) {
697 			pr_warn("failed to early memremap setup_data entry\n");
698 			return false;
699 		}
700 
701 		size = sizeof(*data);
702 
703 		paddr_next = data->next;
704 		len = data->len;
705 
706 		if ((phys_addr > paddr) && (phys_addr < (paddr + len))) {
707 			early_memunmap(data, sizeof(*data));
708 			return true;
709 		}
710 
711 		if (data->type == SETUP_INDIRECT) {
712 			size += len;
713 			early_memunmap(data, sizeof(*data));
714 			data = early_memremap_decrypted(paddr, size);
715 			if (!data) {
716 				pr_warn("failed to early memremap indirect setup_data\n");
717 				return false;
718 			}
719 
720 			indirect = (struct setup_indirect *)data->data;
721 
722 			if (indirect->type != SETUP_INDIRECT) {
723 				paddr = indirect->addr;
724 				len = indirect->len;
725 			}
726 		}
727 
728 		early_memunmap(data, size);
729 
730 		if ((phys_addr > paddr) && (phys_addr < (paddr + len)))
731 			return true;
732 
733 		paddr = paddr_next;
734 	}
735 
736 	return false;
737 }
738 
739 /*
740  * Architecture function to determine if RAM remap is allowed. By default, a
741  * RAM remap will map the data as encrypted. Determine if a RAM remap should
742  * not be done so that the data will be mapped decrypted.
743  */
744 bool arch_memremap_can_ram_remap(resource_size_t phys_addr, unsigned long size,
745 				 unsigned long flags)
746 {
747 	if (!cc_platform_has(CC_ATTR_MEM_ENCRYPT))
748 		return true;
749 
750 	if (flags & MEMREMAP_ENC)
751 		return true;
752 
753 	if (flags & MEMREMAP_DEC)
754 		return false;
755 
756 	if (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT)) {
757 		if (memremap_is_setup_data(phys_addr, size) ||
758 		    memremap_is_efi_data(phys_addr, size))
759 			return false;
760 	}
761 
762 	return !memremap_should_map_decrypted(phys_addr, size);
763 }
764 
765 /*
766  * Architecture override of __weak function to adjust the protection attributes
767  * used when remapping memory. By default, early_memremap() will map the data
768  * as encrypted. Determine if an encrypted mapping should not be done and set
769  * the appropriate protection attributes.
770  */
771 pgprot_t __init early_memremap_pgprot_adjust(resource_size_t phys_addr,
772 					     unsigned long size,
773 					     pgprot_t prot)
774 {
775 	bool encrypted_prot;
776 
777 	if (!cc_platform_has(CC_ATTR_MEM_ENCRYPT))
778 		return prot;
779 
780 	encrypted_prot = true;
781 
782 	if (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT)) {
783 		if (early_memremap_is_setup_data(phys_addr, size) ||
784 		    memremap_is_efi_data(phys_addr, size))
785 			encrypted_prot = false;
786 	}
787 
788 	if (encrypted_prot && memremap_should_map_decrypted(phys_addr, size))
789 		encrypted_prot = false;
790 
791 	return encrypted_prot ? pgprot_encrypted(prot)
792 			      : pgprot_decrypted(prot);
793 }
794 
795 bool phys_mem_access_encrypted(unsigned long phys_addr, unsigned long size)
796 {
797 	return arch_memremap_can_ram_remap(phys_addr, size, 0);
798 }
799 
800 /* Remap memory with encryption */
801 void __init *early_memremap_encrypted(resource_size_t phys_addr,
802 				      unsigned long size)
803 {
804 	return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_ENC);
805 }
806 
807 /*
808  * Remap memory with encryption and write-protected - cannot be called
809  * before pat_init() is called
810  */
811 void __init *early_memremap_encrypted_wp(resource_size_t phys_addr,
812 					 unsigned long size)
813 {
814 	if (!x86_has_pat_wp())
815 		return NULL;
816 	return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_ENC_WP);
817 }
818 
819 /* Remap memory without encryption */
820 void __init *early_memremap_decrypted(resource_size_t phys_addr,
821 				      unsigned long size)
822 {
823 	return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_NOENC);
824 }
825 
826 /*
827  * Remap memory without encryption and write-protected - cannot be called
828  * before pat_init() is called
829  */
830 void __init *early_memremap_decrypted_wp(resource_size_t phys_addr,
831 					 unsigned long size)
832 {
833 	if (!x86_has_pat_wp())
834 		return NULL;
835 	return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_NOENC_WP);
836 }
837 #endif	/* CONFIG_AMD_MEM_ENCRYPT */
838 
839 static pte_t bm_pte[PAGE_SIZE/sizeof(pte_t)] __page_aligned_bss;
840 
841 static inline pmd_t * __init early_ioremap_pmd(unsigned long addr)
842 {
843 	/* Don't assume we're using swapper_pg_dir at this point */
844 	pgd_t *base = __va(read_cr3_pa());
845 	pgd_t *pgd = &base[pgd_index(addr)];
846 	p4d_t *p4d = p4d_offset(pgd, addr);
847 	pud_t *pud = pud_offset(p4d, addr);
848 	pmd_t *pmd = pmd_offset(pud, addr);
849 
850 	return pmd;
851 }
852 
853 static inline pte_t * __init early_ioremap_pte(unsigned long addr)
854 {
855 	return &bm_pte[pte_index(addr)];
856 }
857 
858 bool __init is_early_ioremap_ptep(pte_t *ptep)
859 {
860 	return ptep >= &bm_pte[0] && ptep < &bm_pte[PAGE_SIZE/sizeof(pte_t)];
861 }
862 
863 void __init early_ioremap_init(void)
864 {
865 	pmd_t *pmd;
866 
867 #ifdef CONFIG_X86_64
868 	BUILD_BUG_ON((fix_to_virt(0) + PAGE_SIZE) & ((1 << PMD_SHIFT) - 1));
869 #else
870 	WARN_ON((fix_to_virt(0) + PAGE_SIZE) & ((1 << PMD_SHIFT) - 1));
871 #endif
872 
873 	early_ioremap_setup();
874 
875 	pmd = early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN));
876 	memset(bm_pte, 0, sizeof(bm_pte));
877 	pmd_populate_kernel(&init_mm, pmd, bm_pte);
878 
879 	/*
880 	 * The boot-ioremap range spans multiple pmds, for which
881 	 * we are not prepared:
882 	 */
883 #define __FIXADDR_TOP (-PAGE_SIZE)
884 	BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
885 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
886 #undef __FIXADDR_TOP
887 	if (pmd != early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END))) {
888 		WARN_ON(1);
889 		printk(KERN_WARNING "pmd %p != %p\n",
890 		       pmd, early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END)));
891 		printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
892 			fix_to_virt(FIX_BTMAP_BEGIN));
893 		printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_END):   %08lx\n",
894 			fix_to_virt(FIX_BTMAP_END));
895 
896 		printk(KERN_WARNING "FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
897 		printk(KERN_WARNING "FIX_BTMAP_BEGIN:     %d\n",
898 		       FIX_BTMAP_BEGIN);
899 	}
900 }
901 
902 void __init __early_set_fixmap(enum fixed_addresses idx,
903 			       phys_addr_t phys, pgprot_t flags)
904 {
905 	unsigned long addr = __fix_to_virt(idx);
906 	pte_t *pte;
907 
908 	if (idx >= __end_of_fixed_addresses) {
909 		BUG();
910 		return;
911 	}
912 	pte = early_ioremap_pte(addr);
913 
914 	/* Sanitize 'prot' against any unsupported bits: */
915 	pgprot_val(flags) &= __supported_pte_mask;
916 
917 	if (pgprot_val(flags))
918 		set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
919 	else
920 		pte_clear(&init_mm, addr, pte);
921 	flush_tlb_one_kernel(addr);
922 }
923