2 * Re-map IO memory to kernel address space so that we can access it.
3 * This is needed for high PCI addresses that aren't mapped in the
4 * 640k-1MB IO memory area on PC's
6 * (C) Copyright 1995 1996 Linus Torvalds
9 #include <linux/bootmem.h>
10 #include <linux/init.h>
12 #include <linux/ioport.h>
13 #include <linux/slab.h>
14 #include <linux/vmalloc.h>
15 #include <linux/mmiotrace.h>
16 #include <linux/mem_encrypt.h>
17 #include <linux/efi.h>
19 #include <asm/set_memory.h>
20 #include <asm/e820/api.h>
21 #include <asm/fixmap.h>
22 #include <asm/pgtable.h>
23 #include <asm/tlbflush.h>
24 #include <asm/pgalloc.h>
26 #include <asm/setup.h>
31 * Fix up the linear direct mapping of the kernel to avoid cache attribute
34 int ioremap_change_attr(unsigned long vaddr, unsigned long size,
35 enum page_cache_mode pcm)
37 unsigned long nrpages = size >> PAGE_SHIFT;
41 case _PAGE_CACHE_MODE_UC:
43 err = _set_memory_uc(vaddr, nrpages);
45 case _PAGE_CACHE_MODE_WC:
46 err = _set_memory_wc(vaddr, nrpages);
48 case _PAGE_CACHE_MODE_WT:
49 err = _set_memory_wt(vaddr, nrpages);
51 case _PAGE_CACHE_MODE_WB:
52 err = _set_memory_wb(vaddr, nrpages);
59 static int __ioremap_check_ram(unsigned long start_pfn, unsigned long nr_pages,
64 for (i = 0; i < nr_pages; ++i)
65 if (pfn_valid(start_pfn + i) &&
66 !PageReserved(pfn_to_page(start_pfn + i)))
73 * Remap an arbitrary physical address space into the kernel virtual
74 * address space. It transparently creates kernel huge I/O mapping when
75 * the physical address is aligned by a huge page size (1GB or 2MB) and
76 * the requested size is at least the huge page size.
78 * NOTE: MTRRs can override PAT memory types with a 4KB granularity.
79 * Therefore, the mapping code falls back to use a smaller page toward 4KB
80 * when a mapping range is covered by non-WB type of MTRRs.
82 * NOTE! We need to allow non-page-aligned mappings too: we will obviously
83 * have to convert them into an offset in a page-aligned mapping, but the
84 * caller shouldn't need to know that small detail.
86 static void __iomem *__ioremap_caller(resource_size_t phys_addr,
87 unsigned long size, enum page_cache_mode pcm, void *caller)
89 unsigned long offset, vaddr;
90 resource_size_t pfn, last_pfn, last_addr;
91 const resource_size_t unaligned_phys_addr = phys_addr;
92 const unsigned long unaligned_size = size;
93 struct vm_struct *area;
94 enum page_cache_mode new_pcm;
97 void __iomem *ret_addr;
99 /* Don't allow wraparound or zero size */
100 last_addr = phys_addr + size - 1;
101 if (!size || last_addr < phys_addr)
104 if (!phys_addr_valid(phys_addr)) {
105 printk(KERN_WARNING "ioremap: invalid physical address %llx\n",
106 (unsigned long long)phys_addr);
112 * Don't allow anybody to remap normal RAM that we're using..
114 pfn = phys_addr >> PAGE_SHIFT;
115 last_pfn = last_addr >> PAGE_SHIFT;
116 if (walk_system_ram_range(pfn, last_pfn - pfn + 1, NULL,
117 __ioremap_check_ram) == 1) {
118 WARN_ONCE(1, "ioremap on RAM at %pa - %pa\n",
119 &phys_addr, &last_addr);
124 * Mappings have to be page-aligned
126 offset = phys_addr & ~PAGE_MASK;
127 phys_addr &= PAGE_MASK;
128 size = PAGE_ALIGN(last_addr+1) - phys_addr;
131 * Mask out any bits not part of the actual physical
132 * address, like memory encryption bits.
134 phys_addr &= PHYSICAL_PAGE_MASK;
136 retval = reserve_memtype(phys_addr, (u64)phys_addr + size,
139 printk(KERN_ERR "ioremap reserve_memtype failed %d\n", retval);
143 if (pcm != new_pcm) {
144 if (!is_new_memtype_allowed(phys_addr, size, pcm, new_pcm)) {
146 "ioremap error for 0x%llx-0x%llx, requested 0x%x, got 0x%x\n",
147 (unsigned long long)phys_addr,
148 (unsigned long long)(phys_addr + size),
150 goto err_free_memtype;
155 prot = PAGE_KERNEL_IO;
157 case _PAGE_CACHE_MODE_UC:
159 prot = __pgprot(pgprot_val(prot) |
160 cachemode2protval(_PAGE_CACHE_MODE_UC));
162 case _PAGE_CACHE_MODE_UC_MINUS:
163 prot = __pgprot(pgprot_val(prot) |
164 cachemode2protval(_PAGE_CACHE_MODE_UC_MINUS));
166 case _PAGE_CACHE_MODE_WC:
167 prot = __pgprot(pgprot_val(prot) |
168 cachemode2protval(_PAGE_CACHE_MODE_WC));
170 case _PAGE_CACHE_MODE_WT:
171 prot = __pgprot(pgprot_val(prot) |
172 cachemode2protval(_PAGE_CACHE_MODE_WT));
174 case _PAGE_CACHE_MODE_WB:
181 area = get_vm_area_caller(size, VM_IOREMAP, caller);
183 goto err_free_memtype;
184 area->phys_addr = phys_addr;
185 vaddr = (unsigned long) area->addr;
187 if (kernel_map_sync_memtype(phys_addr, size, pcm))
190 if (ioremap_page_range(vaddr, vaddr + size, phys_addr, prot))
193 ret_addr = (void __iomem *) (vaddr + offset);
194 mmiotrace_ioremap(unaligned_phys_addr, unaligned_size, ret_addr);
197 * Check if the request spans more than any BAR in the iomem resource
200 if (iomem_map_sanity_check(unaligned_phys_addr, unaligned_size))
201 pr_warn("caller %pS mapping multiple BARs\n", caller);
207 free_memtype(phys_addr, phys_addr + size);
212 * ioremap_nocache - map bus memory into CPU space
213 * @phys_addr: bus address of the memory
214 * @size: size of the resource to map
216 * ioremap_nocache performs a platform specific sequence of operations to
217 * make bus memory CPU accessible via the readb/readw/readl/writeb/
218 * writew/writel functions and the other mmio helpers. The returned
219 * address is not guaranteed to be usable directly as a virtual
222 * This version of ioremap ensures that the memory is marked uncachable
223 * on the CPU as well as honouring existing caching rules from things like
224 * the PCI bus. Note that there are other caches and buffers on many
225 * busses. In particular driver authors should read up on PCI writes
227 * It's useful if some control registers are in such an area and
228 * write combining or read caching is not desirable:
230 * Must be freed with iounmap.
232 void __iomem *ioremap_nocache(resource_size_t phys_addr, unsigned long size)
235 * Ideally, this should be:
236 * pat_enabled() ? _PAGE_CACHE_MODE_UC : _PAGE_CACHE_MODE_UC_MINUS;
238 * Till we fix all X drivers to use ioremap_wc(), we will use
239 * UC MINUS. Drivers that are certain they need or can already
240 * be converted over to strong UC can use ioremap_uc().
242 enum page_cache_mode pcm = _PAGE_CACHE_MODE_UC_MINUS;
244 return __ioremap_caller(phys_addr, size, pcm,
245 __builtin_return_address(0));
247 EXPORT_SYMBOL(ioremap_nocache);
250 * ioremap_uc - map bus memory into CPU space as strongly uncachable
251 * @phys_addr: bus address of the memory
252 * @size: size of the resource to map
254 * ioremap_uc performs a platform specific sequence of operations to
255 * make bus memory CPU accessible via the readb/readw/readl/writeb/
256 * writew/writel functions and the other mmio helpers. The returned
257 * address is not guaranteed to be usable directly as a virtual
260 * This version of ioremap ensures that the memory is marked with a strong
261 * preference as completely uncachable on the CPU when possible. For non-PAT
262 * systems this ends up setting page-attribute flags PCD=1, PWT=1. For PAT
263 * systems this will set the PAT entry for the pages as strong UC. This call
264 * will honor existing caching rules from things like the PCI bus. Note that
265 * there are other caches and buffers on many busses. In particular driver
266 * authors should read up on PCI writes.
268 * It's useful if some control registers are in such an area and
269 * write combining or read caching is not desirable:
271 * Must be freed with iounmap.
273 void __iomem *ioremap_uc(resource_size_t phys_addr, unsigned long size)
275 enum page_cache_mode pcm = _PAGE_CACHE_MODE_UC;
277 return __ioremap_caller(phys_addr, size, pcm,
278 __builtin_return_address(0));
280 EXPORT_SYMBOL_GPL(ioremap_uc);
283 * ioremap_wc - map memory into CPU space write combined
284 * @phys_addr: bus address of the memory
285 * @size: size of the resource to map
287 * This version of ioremap ensures that the memory is marked write combining.
288 * Write combining allows faster writes to some hardware devices.
290 * Must be freed with iounmap.
292 void __iomem *ioremap_wc(resource_size_t phys_addr, unsigned long size)
294 return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WC,
295 __builtin_return_address(0));
297 EXPORT_SYMBOL(ioremap_wc);
300 * ioremap_wt - map memory into CPU space write through
301 * @phys_addr: bus address of the memory
302 * @size: size of the resource to map
304 * This version of ioremap ensures that the memory is marked write through.
305 * Write through stores data into memory while keeping the cache up-to-date.
307 * Must be freed with iounmap.
309 void __iomem *ioremap_wt(resource_size_t phys_addr, unsigned long size)
311 return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WT,
312 __builtin_return_address(0));
314 EXPORT_SYMBOL(ioremap_wt);
316 void __iomem *ioremap_cache(resource_size_t phys_addr, unsigned long size)
318 return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WB,
319 __builtin_return_address(0));
321 EXPORT_SYMBOL(ioremap_cache);
323 void __iomem *ioremap_prot(resource_size_t phys_addr, unsigned long size,
324 unsigned long prot_val)
326 return __ioremap_caller(phys_addr, size,
327 pgprot2cachemode(__pgprot(prot_val)),
328 __builtin_return_address(0));
330 EXPORT_SYMBOL(ioremap_prot);
333 * iounmap - Free a IO remapping
334 * @addr: virtual address from ioremap_*
336 * Caller must ensure there is only one unmapping for the same pointer.
338 void iounmap(volatile void __iomem *addr)
340 struct vm_struct *p, *o;
342 if ((void __force *)addr <= high_memory)
346 * The PCI/ISA range special-casing was removed from __ioremap()
347 * so this check, in theory, can be removed. However, there are
348 * cases where iounmap() is called for addresses not obtained via
349 * ioremap() (vga16fb for example). Add a warning so that these
350 * cases can be caught and fixed.
352 if ((void __force *)addr >= phys_to_virt(ISA_START_ADDRESS) &&
353 (void __force *)addr < phys_to_virt(ISA_END_ADDRESS)) {
354 WARN(1, "iounmap() called for ISA range not obtained using ioremap()\n");
358 mmiotrace_iounmap(addr);
360 addr = (volatile void __iomem *)
361 (PAGE_MASK & (unsigned long __force)addr);
363 /* Use the vm area unlocked, assuming the caller
364 ensures there isn't another iounmap for the same address
365 in parallel. Reuse of the virtual address is prevented by
366 leaving it in the global lists until we're done with it.
367 cpa takes care of the direct mappings. */
368 p = find_vm_area((void __force *)addr);
371 printk(KERN_ERR "iounmap: bad address %p\n", addr);
376 free_memtype(p->phys_addr, p->phys_addr + get_vm_area_size(p));
378 /* Finally remove it */
379 o = remove_vm_area((void __force *)addr);
380 BUG_ON(p != o || o == NULL);
383 EXPORT_SYMBOL(iounmap);
385 int __init arch_ioremap_pud_supported(void)
388 return boot_cpu_has(X86_FEATURE_GBPAGES);
394 int __init arch_ioremap_pmd_supported(void)
396 return boot_cpu_has(X86_FEATURE_PSE);
400 * Convert a physical pointer to a virtual kernel pointer for /dev/mem
403 void *xlate_dev_mem_ptr(phys_addr_t phys)
405 unsigned long start = phys & PAGE_MASK;
406 unsigned long offset = phys & ~PAGE_MASK;
409 /* memremap() maps if RAM, otherwise falls back to ioremap() */
410 vaddr = memremap(start, PAGE_SIZE, MEMREMAP_WB);
412 /* Only add the offset on success and return NULL if memremap() failed */
419 void unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
421 memunmap((void *)((unsigned long)addr & PAGE_MASK));
425 * Examine the physical address to determine if it is an area of memory
426 * that should be mapped decrypted. If the memory is not part of the
427 * kernel usable area it was accessed and created decrypted, so these
428 * areas should be mapped decrypted. And since the encryption key can
429 * change across reboots, persistent memory should also be mapped
432 static bool memremap_should_map_decrypted(resource_size_t phys_addr,
438 * Check if the address is part of a persistent memory region.
439 * This check covers areas added by E820, EFI and ACPI.
441 is_pmem = region_intersects(phys_addr, size, IORESOURCE_MEM,
442 IORES_DESC_PERSISTENT_MEMORY);
443 if (is_pmem != REGION_DISJOINT)
447 * Check if the non-volatile attribute is set for an EFI
450 if (efi_enabled(EFI_BOOT)) {
451 switch (efi_mem_type(phys_addr)) {
452 case EFI_RESERVED_TYPE:
453 if (efi_mem_attributes(phys_addr) & EFI_MEMORY_NV)
461 /* Check if the address is outside kernel usable area */
462 switch (e820__get_entry_type(phys_addr, phys_addr + size - 1)) {
463 case E820_TYPE_RESERVED:
466 case E820_TYPE_UNUSABLE:
477 * Examine the physical address to determine if it is EFI data. Check
478 * it against the boot params structure and EFI tables and memory types.
480 static bool memremap_is_efi_data(resource_size_t phys_addr,
485 /* Check if the address is part of EFI boot/runtime data */
486 if (!efi_enabled(EFI_BOOT))
489 paddr = boot_params.efi_info.efi_memmap_hi;
491 paddr |= boot_params.efi_info.efi_memmap;
492 if (phys_addr == paddr)
495 paddr = boot_params.efi_info.efi_systab_hi;
497 paddr |= boot_params.efi_info.efi_systab;
498 if (phys_addr == paddr)
501 if (efi_is_table_address(phys_addr))
504 switch (efi_mem_type(phys_addr)) {
505 case EFI_BOOT_SERVICES_DATA:
506 case EFI_RUNTIME_SERVICES_DATA:
516 * Examine the physical address to determine if it is boot data by checking
517 * it against the boot params setup_data chain.
519 static bool memremap_is_setup_data(resource_size_t phys_addr,
522 struct setup_data *data;
523 u64 paddr, paddr_next;
525 paddr = boot_params.hdr.setup_data;
529 if (phys_addr == paddr)
532 data = memremap(paddr, sizeof(*data),
533 MEMREMAP_WB | MEMREMAP_DEC);
535 paddr_next = data->next;
540 if ((phys_addr > paddr) && (phys_addr < (paddr + len)))
550 * Examine the physical address to determine if it is boot data by checking
551 * it against the boot params setup_data chain (early boot version).
553 static bool __init early_memremap_is_setup_data(resource_size_t phys_addr,
556 struct setup_data *data;
557 u64 paddr, paddr_next;
559 paddr = boot_params.hdr.setup_data;
563 if (phys_addr == paddr)
566 data = early_memremap_decrypted(paddr, sizeof(*data));
568 paddr_next = data->next;
571 early_memunmap(data, sizeof(*data));
573 if ((phys_addr > paddr) && (phys_addr < (paddr + len)))
583 * Architecture function to determine if RAM remap is allowed. By default, a
584 * RAM remap will map the data as encrypted. Determine if a RAM remap should
585 * not be done so that the data will be mapped decrypted.
587 bool arch_memremap_can_ram_remap(resource_size_t phys_addr, unsigned long size,
593 if (flags & MEMREMAP_ENC)
596 if (flags & MEMREMAP_DEC)
599 if (memremap_is_setup_data(phys_addr, size) ||
600 memremap_is_efi_data(phys_addr, size) ||
601 memremap_should_map_decrypted(phys_addr, size))
608 * Architecture override of __weak function to adjust the protection attributes
609 * used when remapping memory. By default, early_memremap() will map the data
610 * as encrypted. Determine if an encrypted mapping should not be done and set
611 * the appropriate protection attributes.
613 pgprot_t __init early_memremap_pgprot_adjust(resource_size_t phys_addr,
620 if (early_memremap_is_setup_data(phys_addr, size) ||
621 memremap_is_efi_data(phys_addr, size) ||
622 memremap_should_map_decrypted(phys_addr, size))
623 prot = pgprot_decrypted(prot);
625 prot = pgprot_encrypted(prot);
630 bool phys_mem_access_encrypted(unsigned long phys_addr, unsigned long size)
632 return arch_memremap_can_ram_remap(phys_addr, size, 0);
635 #ifdef CONFIG_AMD_MEM_ENCRYPT
636 /* Remap memory with encryption */
637 void __init *early_memremap_encrypted(resource_size_t phys_addr,
640 return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_ENC);
644 * Remap memory with encryption and write-protected - cannot be called
645 * before pat_init() is called
647 void __init *early_memremap_encrypted_wp(resource_size_t phys_addr,
650 /* Be sure the write-protect PAT entry is set for write-protect */
651 if (__pte2cachemode_tbl[_PAGE_CACHE_MODE_WP] != _PAGE_CACHE_MODE_WP)
654 return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_ENC_WP);
657 /* Remap memory without encryption */
658 void __init *early_memremap_decrypted(resource_size_t phys_addr,
661 return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_NOENC);
665 * Remap memory without encryption and write-protected - cannot be called
666 * before pat_init() is called
668 void __init *early_memremap_decrypted_wp(resource_size_t phys_addr,
671 /* Be sure the write-protect PAT entry is set for write-protect */
672 if (__pte2cachemode_tbl[_PAGE_CACHE_MODE_WP] != _PAGE_CACHE_MODE_WP)
675 return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_NOENC_WP);
677 #endif /* CONFIG_AMD_MEM_ENCRYPT */
679 static pte_t bm_pte[PAGE_SIZE/sizeof(pte_t)] __page_aligned_bss;
681 static inline pmd_t * __init early_ioremap_pmd(unsigned long addr)
683 /* Don't assume we're using swapper_pg_dir at this point */
684 pgd_t *base = __va(read_cr3_pa());
685 pgd_t *pgd = &base[pgd_index(addr)];
686 p4d_t *p4d = p4d_offset(pgd, addr);
687 pud_t *pud = pud_offset(p4d, addr);
688 pmd_t *pmd = pmd_offset(pud, addr);
693 static inline pte_t * __init early_ioremap_pte(unsigned long addr)
695 return &bm_pte[pte_index(addr)];
698 bool __init is_early_ioremap_ptep(pte_t *ptep)
700 return ptep >= &bm_pte[0] && ptep < &bm_pte[PAGE_SIZE/sizeof(pte_t)];
703 void __init early_ioremap_init(void)
708 BUILD_BUG_ON((fix_to_virt(0) + PAGE_SIZE) & ((1 << PMD_SHIFT) - 1));
710 WARN_ON((fix_to_virt(0) + PAGE_SIZE) & ((1 << PMD_SHIFT) - 1));
713 early_ioremap_setup();
715 pmd = early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN));
716 memset(bm_pte, 0, sizeof(bm_pte));
717 pmd_populate_kernel(&init_mm, pmd, bm_pte);
720 * The boot-ioremap range spans multiple pmds, for which
721 * we are not prepared:
723 #define __FIXADDR_TOP (-PAGE_SIZE)
724 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
725 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
727 if (pmd != early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END))) {
729 printk(KERN_WARNING "pmd %p != %p\n",
730 pmd, early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END)));
731 printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
732 fix_to_virt(FIX_BTMAP_BEGIN));
733 printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_END): %08lx\n",
734 fix_to_virt(FIX_BTMAP_END));
736 printk(KERN_WARNING "FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
737 printk(KERN_WARNING "FIX_BTMAP_BEGIN: %d\n",
742 void __init __early_set_fixmap(enum fixed_addresses idx,
743 phys_addr_t phys, pgprot_t flags)
745 unsigned long addr = __fix_to_virt(idx);
748 if (idx >= __end_of_fixed_addresses) {
752 pte = early_ioremap_pte(addr);
754 if (pgprot_val(flags))
755 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
757 pte_clear(&init_mm, addr, pte);
758 __flush_tlb_one_kernel(addr);