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/module.h>
13 #include <linux/slab.h>
14 #include <linux/vmalloc.h>
15 #include <linux/mmiotrace.h>
17 #include <asm/cacheflush.h>
19 #include <asm/fixmap.h>
20 #include <asm/pgtable.h>
21 #include <asm/tlbflush.h>
22 #include <asm/pgalloc.h>
27 static inline int phys_addr_valid(unsigned long addr)
29 return addr < (1UL << boot_cpu_data.x86_phys_bits);
32 unsigned long __phys_addr(unsigned long x)
34 if (x >= __START_KERNEL_map) {
35 x -= __START_KERNEL_map;
36 VIRTUAL_BUG_ON(x >= KERNEL_IMAGE_SIZE);
39 VIRTUAL_BUG_ON(x < PAGE_OFFSET);
41 VIRTUAL_BUG_ON(system_state == SYSTEM_BOOTING ? x > MAXMEM :
46 EXPORT_SYMBOL(__phys_addr);
50 static inline int phys_addr_valid(unsigned long addr)
55 #ifdef CONFIG_DEBUG_VIRTUAL
56 unsigned long __phys_addr(unsigned long x)
58 /* VMALLOC_* aren't constants; not available at the boot time */
59 VIRTUAL_BUG_ON(x < PAGE_OFFSET || (system_state != SYSTEM_BOOTING &&
60 is_vmalloc_addr((void *)x)));
61 return x - PAGE_OFFSET;
63 EXPORT_SYMBOL(__phys_addr);
68 int page_is_ram(unsigned long pagenr)
70 resource_size_t addr, end;
74 * A special case is the first 4Kb of memory;
75 * This is a BIOS owned area, not kernel ram, but generally
76 * not listed as such in the E820 table.
82 * Second special case: Some BIOSen report the PC BIOS
83 * area (640->1Mb) as ram even though it is not.
85 if (pagenr >= (BIOS_BEGIN >> PAGE_SHIFT) &&
86 pagenr < (BIOS_END >> PAGE_SHIFT))
89 for (i = 0; i < e820.nr_map; i++) {
93 if (e820.map[i].type != E820_RAM)
95 addr = (e820.map[i].addr + PAGE_SIZE-1) >> PAGE_SHIFT;
96 end = (e820.map[i].addr + e820.map[i].size) >> PAGE_SHIFT;
99 if ((pagenr >= addr) && (pagenr < end))
105 int pagerange_is_ram(unsigned long start, unsigned long end)
107 int ram_page = 0, not_rampage = 0;
108 unsigned long page_nr;
110 for (page_nr = (start >> PAGE_SHIFT); page_nr < (end >> PAGE_SHIFT);
112 if (page_is_ram(page_nr))
117 if (ram_page == not_rampage)
125 * Fix up the linear direct mapping of the kernel to avoid cache attribute
128 int ioremap_change_attr(unsigned long vaddr, unsigned long size,
129 unsigned long prot_val)
131 unsigned long nrpages = size >> PAGE_SHIFT;
137 err = _set_memory_uc(vaddr, nrpages);
140 err = _set_memory_wc(vaddr, nrpages);
143 err = _set_memory_wb(vaddr, nrpages);
151 * Remap an arbitrary physical address space into the kernel virtual
152 * address space. Needed when the kernel wants to access high addresses
155 * NOTE! We need to allow non-page-aligned mappings too: we will obviously
156 * have to convert them into an offset in a page-aligned mapping, but the
157 * caller shouldn't need to know that small detail.
159 static void __iomem *__ioremap_caller(resource_size_t phys_addr,
160 unsigned long size, unsigned long prot_val, void *caller)
162 unsigned long pfn, offset, vaddr;
163 resource_size_t last_addr;
164 const resource_size_t unaligned_phys_addr = phys_addr;
165 const unsigned long unaligned_size = size;
166 struct vm_struct *area;
167 unsigned long new_prot_val;
170 void __iomem *ret_addr;
172 /* Don't allow wraparound or zero size */
173 last_addr = phys_addr + size - 1;
174 if (!size || last_addr < phys_addr)
177 if (!phys_addr_valid(phys_addr)) {
178 printk(KERN_WARNING "ioremap: invalid physical address %llx\n",
179 (unsigned long long)phys_addr);
185 * Don't remap the low PCI/ISA area, it's always mapped..
187 if (is_ISA_range(phys_addr, last_addr))
188 return (__force void __iomem *)phys_to_virt(phys_addr);
191 * Don't allow anybody to remap normal RAM that we're using..
193 for (pfn = phys_addr >> PAGE_SHIFT;
194 (pfn << PAGE_SHIFT) < (last_addr & PAGE_MASK);
197 int is_ram = page_is_ram(pfn);
199 if (is_ram && pfn_valid(pfn) && !PageReserved(pfn_to_page(pfn)))
201 WARN_ON_ONCE(is_ram);
205 * Mappings have to be page-aligned
207 offset = phys_addr & ~PAGE_MASK;
208 phys_addr &= PAGE_MASK;
209 size = PAGE_ALIGN(last_addr+1) - phys_addr;
211 retval = reserve_memtype(phys_addr, (u64)phys_addr + size,
212 prot_val, &new_prot_val);
214 pr_debug("Warning: reserve_memtype returned %d\n", retval);
218 if (prot_val != new_prot_val) {
220 * Do not fallback to certain memory types with certain
222 * - request is uc-, return cannot be write-back
223 * - request is uc-, return cannot be write-combine
224 * - request is write-combine, return cannot be write-back
226 if ((prot_val == _PAGE_CACHE_UC_MINUS &&
227 (new_prot_val == _PAGE_CACHE_WB ||
228 new_prot_val == _PAGE_CACHE_WC)) ||
229 (prot_val == _PAGE_CACHE_WC &&
230 new_prot_val == _PAGE_CACHE_WB)) {
232 "ioremap error for 0x%llx-0x%llx, requested 0x%lx, got 0x%lx\n",
233 (unsigned long long)phys_addr,
234 (unsigned long long)(phys_addr + size),
235 prot_val, new_prot_val);
236 free_memtype(phys_addr, phys_addr + size);
239 prot_val = new_prot_val;
245 prot = PAGE_KERNEL_NOCACHE;
247 case _PAGE_CACHE_UC_MINUS:
248 prot = PAGE_KERNEL_UC_MINUS;
251 prot = PAGE_KERNEL_WC;
261 area = get_vm_area_caller(size, VM_IOREMAP, caller);
264 area->phys_addr = phys_addr;
265 vaddr = (unsigned long) area->addr;
266 if (ioremap_page_range(vaddr, vaddr + size, phys_addr, prot)) {
267 free_memtype(phys_addr, phys_addr + size);
272 if (ioremap_change_attr(vaddr, size, prot_val) < 0) {
273 free_memtype(phys_addr, phys_addr + size);
278 ret_addr = (void __iomem *) (vaddr + offset);
279 mmiotrace_ioremap(unaligned_phys_addr, unaligned_size, ret_addr);
285 * ioremap_nocache - map bus memory into CPU space
286 * @offset: bus address of the memory
287 * @size: size of the resource to map
289 * ioremap_nocache performs a platform specific sequence of operations to
290 * make bus memory CPU accessible via the readb/readw/readl/writeb/
291 * writew/writel functions and the other mmio helpers. The returned
292 * address is not guaranteed to be usable directly as a virtual
295 * This version of ioremap ensures that the memory is marked uncachable
296 * on the CPU as well as honouring existing caching rules from things like
297 * the PCI bus. Note that there are other caches and buffers on many
298 * busses. In particular driver authors should read up on PCI writes
300 * It's useful if some control registers are in such an area and
301 * write combining or read caching is not desirable:
303 * Must be freed with iounmap.
305 void __iomem *ioremap_nocache(resource_size_t phys_addr, unsigned long size)
308 * Ideally, this should be:
309 * pat_enabled ? _PAGE_CACHE_UC : _PAGE_CACHE_UC_MINUS;
311 * Till we fix all X drivers to use ioremap_wc(), we will use
314 unsigned long val = _PAGE_CACHE_UC_MINUS;
316 return __ioremap_caller(phys_addr, size, val,
317 __builtin_return_address(0));
319 EXPORT_SYMBOL(ioremap_nocache);
322 * ioremap_wc - map memory into CPU space write combined
323 * @offset: bus address of the memory
324 * @size: size of the resource to map
326 * This version of ioremap ensures that the memory is marked write combining.
327 * Write combining allows faster writes to some hardware devices.
329 * Must be freed with iounmap.
331 void __iomem *ioremap_wc(unsigned long phys_addr, unsigned long size)
334 return __ioremap_caller(phys_addr, size, _PAGE_CACHE_WC,
335 __builtin_return_address(0));
337 return ioremap_nocache(phys_addr, size);
339 EXPORT_SYMBOL(ioremap_wc);
341 void __iomem *ioremap_cache(resource_size_t phys_addr, unsigned long size)
343 return __ioremap_caller(phys_addr, size, _PAGE_CACHE_WB,
344 __builtin_return_address(0));
346 EXPORT_SYMBOL(ioremap_cache);
348 static void __iomem *ioremap_default(resource_size_t phys_addr,
356 * - WB for WB-able memory and no other conflicting mappings
357 * - UC_MINUS for non-WB-able memory with no other conflicting mappings
358 * - Inherit from confliting mappings otherwise
360 err = reserve_memtype(phys_addr, phys_addr + size, -1, &flags);
364 ret = (void *) __ioremap_caller(phys_addr, size, flags,
365 __builtin_return_address(0));
367 free_memtype(phys_addr, phys_addr + size);
368 return (void __iomem *)ret;
371 void __iomem *ioremap_prot(resource_size_t phys_addr, unsigned long size,
372 unsigned long prot_val)
374 return __ioremap_caller(phys_addr, size, (prot_val & _PAGE_CACHE_MASK),
375 __builtin_return_address(0));
377 EXPORT_SYMBOL(ioremap_prot);
380 * iounmap - Free a IO remapping
381 * @addr: virtual address from ioremap_*
383 * Caller must ensure there is only one unmapping for the same pointer.
385 void iounmap(volatile void __iomem *addr)
387 struct vm_struct *p, *o;
389 if ((void __force *)addr <= high_memory)
393 * __ioremap special-cases the PCI/ISA range by not instantiating a
394 * vm_area and by simply returning an address into the kernel mapping
395 * of ISA space. So handle that here.
397 if ((void __force *)addr >= phys_to_virt(ISA_START_ADDRESS) &&
398 (void __force *)addr < phys_to_virt(ISA_END_ADDRESS))
401 addr = (volatile void __iomem *)
402 (PAGE_MASK & (unsigned long __force)addr);
404 mmiotrace_iounmap(addr);
406 /* Use the vm area unlocked, assuming the caller
407 ensures there isn't another iounmap for the same address
408 in parallel. Reuse of the virtual address is prevented by
409 leaving it in the global lists until we're done with it.
410 cpa takes care of the direct mappings. */
411 read_lock(&vmlist_lock);
412 for (p = vmlist; p; p = p->next) {
413 if (p->addr == (void __force *)addr)
416 read_unlock(&vmlist_lock);
419 printk(KERN_ERR "iounmap: bad address %p\n", addr);
424 free_memtype(p->phys_addr, p->phys_addr + get_vm_area_size(p));
426 /* Finally remove it */
427 o = remove_vm_area((void __force *)addr);
428 BUG_ON(p != o || o == NULL);
431 EXPORT_SYMBOL(iounmap);
434 * Convert a physical pointer to a virtual kernel pointer for /dev/mem
437 void *xlate_dev_mem_ptr(unsigned long phys)
440 unsigned long start = phys & PAGE_MASK;
442 /* If page is RAM, we can use __va. Otherwise ioremap and unmap. */
443 if (page_is_ram(start >> PAGE_SHIFT))
446 addr = (void __force *)ioremap_default(start, PAGE_SIZE);
448 addr = (void *)((unsigned long)addr | (phys & ~PAGE_MASK));
453 void unxlate_dev_mem_ptr(unsigned long phys, void *addr)
455 if (page_is_ram(phys >> PAGE_SHIFT))
458 iounmap((void __iomem *)((unsigned long)addr & PAGE_MASK));
462 static int __initdata early_ioremap_debug;
464 static int __init early_ioremap_debug_setup(char *str)
466 early_ioremap_debug = 1;
470 early_param("early_ioremap_debug", early_ioremap_debug_setup);
472 static __initdata int after_paging_init;
473 static pte_t bm_pte[PAGE_SIZE/sizeof(pte_t)] __page_aligned_bss;
475 static inline pmd_t * __init early_ioremap_pmd(unsigned long addr)
477 /* Don't assume we're using swapper_pg_dir at this point */
478 pgd_t *base = __va(read_cr3());
479 pgd_t *pgd = &base[pgd_index(addr)];
480 pud_t *pud = pud_offset(pgd, addr);
481 pmd_t *pmd = pmd_offset(pud, addr);
486 static inline pte_t * __init early_ioremap_pte(unsigned long addr)
488 return &bm_pte[pte_index(addr)];
491 void __init early_ioremap_init(void)
495 if (early_ioremap_debug)
496 printk(KERN_INFO "early_ioremap_init()\n");
498 pmd = early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN));
499 memset(bm_pte, 0, sizeof(bm_pte));
500 pmd_populate_kernel(&init_mm, pmd, bm_pte);
503 * The boot-ioremap range spans multiple pmds, for which
504 * we are not prepared:
506 if (pmd != early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END))) {
508 printk(KERN_WARNING "pmd %p != %p\n",
509 pmd, early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END)));
510 printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
511 fix_to_virt(FIX_BTMAP_BEGIN));
512 printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_END): %08lx\n",
513 fix_to_virt(FIX_BTMAP_END));
515 printk(KERN_WARNING "FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
516 printk(KERN_WARNING "FIX_BTMAP_BEGIN: %d\n",
521 void __init early_ioremap_clear(void)
525 if (early_ioremap_debug)
526 printk(KERN_INFO "early_ioremap_clear()\n");
528 pmd = early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN));
530 paravirt_release_pte(__pa(bm_pte) >> PAGE_SHIFT);
534 void __init early_ioremap_reset(void)
536 enum fixed_addresses idx;
537 unsigned long addr, phys;
540 after_paging_init = 1;
541 for (idx = FIX_BTMAP_BEGIN; idx >= FIX_BTMAP_END; idx--) {
542 addr = fix_to_virt(idx);
543 pte = early_ioremap_pte(addr);
544 if (pte_present(*pte)) {
545 phys = pte_val(*pte) & PAGE_MASK;
546 set_fixmap(idx, phys);
551 static void __init __early_set_fixmap(enum fixed_addresses idx,
552 unsigned long phys, pgprot_t flags)
554 unsigned long addr = __fix_to_virt(idx);
557 if (idx >= __end_of_fixed_addresses) {
561 pte = early_ioremap_pte(addr);
563 if (pgprot_val(flags))
564 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
566 pte_clear(&init_mm, addr, pte);
567 __flush_tlb_one(addr);
570 static inline void __init early_set_fixmap(enum fixed_addresses idx,
573 if (after_paging_init)
574 set_fixmap(idx, phys);
576 __early_set_fixmap(idx, phys, PAGE_KERNEL);
579 static inline void __init early_clear_fixmap(enum fixed_addresses idx)
581 if (after_paging_init)
584 __early_set_fixmap(idx, 0, __pgprot(0));
588 static int __initdata early_ioremap_nested;
590 static int __init check_early_ioremap_leak(void)
592 if (!early_ioremap_nested)
595 "Debug warning: early ioremap leak of %d areas detected.\n",
596 early_ioremap_nested);
598 "please boot with early_ioremap_debug and report the dmesg.\n");
602 late_initcall(check_early_ioremap_leak);
604 void __init *early_ioremap(unsigned long phys_addr, unsigned long size)
606 unsigned long offset, last_addr;
607 unsigned int nrpages, nesting;
608 enum fixed_addresses idx0, idx;
610 WARN_ON(system_state != SYSTEM_BOOTING);
612 nesting = early_ioremap_nested;
613 if (early_ioremap_debug) {
614 printk(KERN_INFO "early_ioremap(%08lx, %08lx) [%d] => ",
615 phys_addr, size, nesting);
619 /* Don't allow wraparound or zero size */
620 last_addr = phys_addr + size - 1;
621 if (!size || last_addr < phys_addr) {
626 if (nesting >= FIX_BTMAPS_NESTING) {
630 early_ioremap_nested++;
632 * Mappings have to be page-aligned
634 offset = phys_addr & ~PAGE_MASK;
635 phys_addr &= PAGE_MASK;
636 size = PAGE_ALIGN(last_addr + 1) - phys_addr;
639 * Mappings have to fit in the FIX_BTMAP area.
641 nrpages = size >> PAGE_SHIFT;
642 if (nrpages > NR_FIX_BTMAPS) {
650 idx0 = FIX_BTMAP_BEGIN - NR_FIX_BTMAPS*nesting;
652 while (nrpages > 0) {
653 early_set_fixmap(idx, phys_addr);
654 phys_addr += PAGE_SIZE;
658 if (early_ioremap_debug)
659 printk(KERN_CONT "%08lx + %08lx\n", offset, fix_to_virt(idx0));
661 return (void *) (offset + fix_to_virt(idx0));
664 void __init early_iounmap(void *addr, unsigned long size)
666 unsigned long virt_addr;
667 unsigned long offset;
668 unsigned int nrpages;
669 enum fixed_addresses idx;
672 nesting = --early_ioremap_nested;
673 if (WARN_ON(nesting < 0))
676 if (early_ioremap_debug) {
677 printk(KERN_INFO "early_iounmap(%p, %08lx) [%d]\n", addr,
682 virt_addr = (unsigned long)addr;
683 if (virt_addr < fix_to_virt(FIX_BTMAP_BEGIN)) {
687 offset = virt_addr & ~PAGE_MASK;
688 nrpages = PAGE_ALIGN(offset + size - 1) >> PAGE_SHIFT;
690 idx = FIX_BTMAP_BEGIN - NR_FIX_BTMAPS*nesting;
691 while (nrpages > 0) {
692 early_clear_fixmap(idx);
698 void __this_fixmap_does_not_exist(void)