x86: mm/numa_32.c calculate_numa_remap_pages should use __init
[linux-2.6/mini2440.git] / arch / x86 / mm / ioremap.c
blobd4c4b2c4dbbe8790e6ddf283bd675097f661bb6c
1 /*
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
7 */
9 #include <linux/bootmem.h>
10 #include <linux/init.h>
11 #include <linux/io.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>
18 #include <asm/e820.h>
19 #include <asm/fixmap.h>
20 #include <asm/pgtable.h>
21 #include <asm/tlbflush.h>
22 #include <asm/pgalloc.h>
23 #include <asm/pat.h>
25 static inline int phys_addr_valid(resource_size_t addr)
27 #ifdef CONFIG_PHYS_ADDR_T_64BIT
28 return !(addr >> boot_cpu_data.x86_phys_bits);
29 #else
30 return 1;
31 #endif
34 #ifdef CONFIG_X86_64
36 unsigned long __phys_addr(unsigned long x)
38 if (x >= __START_KERNEL_map) {
39 x -= __START_KERNEL_map;
40 VIRTUAL_BUG_ON(x >= KERNEL_IMAGE_SIZE);
41 x += phys_base;
42 } else {
43 VIRTUAL_BUG_ON(x < PAGE_OFFSET);
44 x -= PAGE_OFFSET;
45 VIRTUAL_BUG_ON(!phys_addr_valid(x));
47 return x;
49 EXPORT_SYMBOL(__phys_addr);
51 bool __virt_addr_valid(unsigned long x)
53 if (x >= __START_KERNEL_map) {
54 x -= __START_KERNEL_map;
55 if (x >= KERNEL_IMAGE_SIZE)
56 return false;
57 x += phys_base;
58 } else {
59 if (x < PAGE_OFFSET)
60 return false;
61 x -= PAGE_OFFSET;
62 if (!phys_addr_valid(x))
63 return false;
66 return pfn_valid(x >> PAGE_SHIFT);
68 EXPORT_SYMBOL(__virt_addr_valid);
70 #else
72 #ifdef CONFIG_DEBUG_VIRTUAL
73 unsigned long __phys_addr(unsigned long x)
75 /* VMALLOC_* aren't constants */
76 VIRTUAL_BUG_ON(x < PAGE_OFFSET);
77 VIRTUAL_BUG_ON(__vmalloc_start_set && is_vmalloc_addr((void *) x));
78 return x - PAGE_OFFSET;
80 EXPORT_SYMBOL(__phys_addr);
81 #endif
83 bool __virt_addr_valid(unsigned long x)
85 if (x < PAGE_OFFSET)
86 return false;
87 if (__vmalloc_start_set && is_vmalloc_addr((void *) x))
88 return false;
89 if (x >= FIXADDR_START)
90 return false;
91 return pfn_valid((x - PAGE_OFFSET) >> PAGE_SHIFT);
93 EXPORT_SYMBOL(__virt_addr_valid);
95 #endif
97 int page_is_ram(unsigned long pagenr)
99 resource_size_t addr, end;
100 int i;
103 * A special case is the first 4Kb of memory;
104 * This is a BIOS owned area, not kernel ram, but generally
105 * not listed as such in the E820 table.
107 if (pagenr == 0)
108 return 0;
111 * Second special case: Some BIOSen report the PC BIOS
112 * area (640->1Mb) as ram even though it is not.
114 if (pagenr >= (BIOS_BEGIN >> PAGE_SHIFT) &&
115 pagenr < (BIOS_END >> PAGE_SHIFT))
116 return 0;
118 for (i = 0; i < e820.nr_map; i++) {
120 * Not usable memory:
122 if (e820.map[i].type != E820_RAM)
123 continue;
124 addr = (e820.map[i].addr + PAGE_SIZE-1) >> PAGE_SHIFT;
125 end = (e820.map[i].addr + e820.map[i].size) >> PAGE_SHIFT;
128 if ((pagenr >= addr) && (pagenr < end))
129 return 1;
131 return 0;
135 * Fix up the linear direct mapping of the kernel to avoid cache attribute
136 * conflicts.
138 int ioremap_change_attr(unsigned long vaddr, unsigned long size,
139 unsigned long prot_val)
141 unsigned long nrpages = size >> PAGE_SHIFT;
142 int err;
144 switch (prot_val) {
145 case _PAGE_CACHE_UC:
146 default:
147 err = _set_memory_uc(vaddr, nrpages);
148 break;
149 case _PAGE_CACHE_WC:
150 err = _set_memory_wc(vaddr, nrpages);
151 break;
152 case _PAGE_CACHE_WB:
153 err = _set_memory_wb(vaddr, nrpages);
154 break;
157 return err;
161 * Remap an arbitrary physical address space into the kernel virtual
162 * address space. Needed when the kernel wants to access high addresses
163 * directly.
165 * NOTE! We need to allow non-page-aligned mappings too: we will obviously
166 * have to convert them into an offset in a page-aligned mapping, but the
167 * caller shouldn't need to know that small detail.
169 static void __iomem *__ioremap_caller(resource_size_t phys_addr,
170 unsigned long size, unsigned long prot_val, void *caller)
172 unsigned long pfn, offset, vaddr;
173 resource_size_t last_addr;
174 const resource_size_t unaligned_phys_addr = phys_addr;
175 const unsigned long unaligned_size = size;
176 struct vm_struct *area;
177 unsigned long new_prot_val;
178 pgprot_t prot;
179 int retval;
180 void __iomem *ret_addr;
182 /* Don't allow wraparound or zero size */
183 last_addr = phys_addr + size - 1;
184 if (!size || last_addr < phys_addr)
185 return NULL;
187 if (!phys_addr_valid(phys_addr)) {
188 printk(KERN_WARNING "ioremap: invalid physical address %llx\n",
189 (unsigned long long)phys_addr);
190 WARN_ON_ONCE(1);
191 return NULL;
195 * Don't remap the low PCI/ISA area, it's always mapped..
197 if (is_ISA_range(phys_addr, last_addr))
198 return (__force void __iomem *)phys_to_virt(phys_addr);
201 * Check if the request spans more than any BAR in the iomem resource
202 * tree.
204 WARN_ONCE(iomem_map_sanity_check(phys_addr, size),
205 KERN_INFO "Info: mapping multiple BARs. Your kernel is fine.");
208 * Don't allow anybody to remap normal RAM that we're using..
210 for (pfn = phys_addr >> PAGE_SHIFT;
211 (pfn << PAGE_SHIFT) < (last_addr & PAGE_MASK);
212 pfn++) {
214 int is_ram = page_is_ram(pfn);
216 if (is_ram && pfn_valid(pfn) && !PageReserved(pfn_to_page(pfn)))
217 return NULL;
218 WARN_ON_ONCE(is_ram);
222 * Mappings have to be page-aligned
224 offset = phys_addr & ~PAGE_MASK;
225 phys_addr &= PAGE_MASK;
226 size = PAGE_ALIGN(last_addr+1) - phys_addr;
228 retval = reserve_memtype(phys_addr, (u64)phys_addr + size,
229 prot_val, &new_prot_val);
230 if (retval) {
231 pr_debug("Warning: reserve_memtype returned %d\n", retval);
232 return NULL;
235 if (prot_val != new_prot_val) {
237 * Do not fallback to certain memory types with certain
238 * requested type:
239 * - request is uc-, return cannot be write-back
240 * - request is uc-, return cannot be write-combine
241 * - request is write-combine, return cannot be write-back
243 if ((prot_val == _PAGE_CACHE_UC_MINUS &&
244 (new_prot_val == _PAGE_CACHE_WB ||
245 new_prot_val == _PAGE_CACHE_WC)) ||
246 (prot_val == _PAGE_CACHE_WC &&
247 new_prot_val == _PAGE_CACHE_WB)) {
248 pr_debug(
249 "ioremap error for 0x%llx-0x%llx, requested 0x%lx, got 0x%lx\n",
250 (unsigned long long)phys_addr,
251 (unsigned long long)(phys_addr + size),
252 prot_val, new_prot_val);
253 free_memtype(phys_addr, phys_addr + size);
254 return NULL;
256 prot_val = new_prot_val;
259 switch (prot_val) {
260 case _PAGE_CACHE_UC:
261 default:
262 prot = PAGE_KERNEL_IO_NOCACHE;
263 break;
264 case _PAGE_CACHE_UC_MINUS:
265 prot = PAGE_KERNEL_IO_UC_MINUS;
266 break;
267 case _PAGE_CACHE_WC:
268 prot = PAGE_KERNEL_IO_WC;
269 break;
270 case _PAGE_CACHE_WB:
271 prot = PAGE_KERNEL_IO;
272 break;
276 * Ok, go for it..
278 area = get_vm_area_caller(size, VM_IOREMAP, caller);
279 if (!area)
280 return NULL;
281 area->phys_addr = phys_addr;
282 vaddr = (unsigned long) area->addr;
284 if (kernel_map_sync_memtype(phys_addr, size, prot_val)) {
285 free_memtype(phys_addr, phys_addr + size);
286 free_vm_area(area);
287 return NULL;
290 if (ioremap_page_range(vaddr, vaddr + size, phys_addr, prot)) {
291 free_memtype(phys_addr, phys_addr + size);
292 free_vm_area(area);
293 return NULL;
296 ret_addr = (void __iomem *) (vaddr + offset);
297 mmiotrace_ioremap(unaligned_phys_addr, unaligned_size, ret_addr);
299 return ret_addr;
303 * ioremap_nocache - map bus memory into CPU space
304 * @offset: bus address of the memory
305 * @size: size of the resource to map
307 * ioremap_nocache performs a platform specific sequence of operations to
308 * make bus memory CPU accessible via the readb/readw/readl/writeb/
309 * writew/writel functions and the other mmio helpers. The returned
310 * address is not guaranteed to be usable directly as a virtual
311 * address.
313 * This version of ioremap ensures that the memory is marked uncachable
314 * on the CPU as well as honouring existing caching rules from things like
315 * the PCI bus. Note that there are other caches and buffers on many
316 * busses. In particular driver authors should read up on PCI writes
318 * It's useful if some control registers are in such an area and
319 * write combining or read caching is not desirable:
321 * Must be freed with iounmap.
323 void __iomem *ioremap_nocache(resource_size_t phys_addr, unsigned long size)
326 * Ideally, this should be:
327 * pat_enabled ? _PAGE_CACHE_UC : _PAGE_CACHE_UC_MINUS;
329 * Till we fix all X drivers to use ioremap_wc(), we will use
330 * UC MINUS.
332 unsigned long val = _PAGE_CACHE_UC_MINUS;
334 return __ioremap_caller(phys_addr, size, val,
335 __builtin_return_address(0));
337 EXPORT_SYMBOL(ioremap_nocache);
340 * ioremap_wc - map memory into CPU space write combined
341 * @offset: bus address of the memory
342 * @size: size of the resource to map
344 * This version of ioremap ensures that the memory is marked write combining.
345 * Write combining allows faster writes to some hardware devices.
347 * Must be freed with iounmap.
349 void __iomem *ioremap_wc(resource_size_t phys_addr, unsigned long size)
351 if (pat_enabled)
352 return __ioremap_caller(phys_addr, size, _PAGE_CACHE_WC,
353 __builtin_return_address(0));
354 else
355 return ioremap_nocache(phys_addr, size);
357 EXPORT_SYMBOL(ioremap_wc);
359 void __iomem *ioremap_cache(resource_size_t phys_addr, unsigned long size)
361 return __ioremap_caller(phys_addr, size, _PAGE_CACHE_WB,
362 __builtin_return_address(0));
364 EXPORT_SYMBOL(ioremap_cache);
366 static void __iomem *ioremap_default(resource_size_t phys_addr,
367 unsigned long size)
369 unsigned long flags;
370 void __iomem *ret;
371 int err;
374 * - WB for WB-able memory and no other conflicting mappings
375 * - UC_MINUS for non-WB-able memory with no other conflicting mappings
376 * - Inherit from confliting mappings otherwise
378 err = reserve_memtype(phys_addr, phys_addr + size,
379 _PAGE_CACHE_WB, &flags);
380 if (err < 0)
381 return NULL;
383 ret = __ioremap_caller(phys_addr, size, flags,
384 __builtin_return_address(0));
386 free_memtype(phys_addr, phys_addr + size);
387 return ret;
390 void __iomem *ioremap_prot(resource_size_t phys_addr, unsigned long size,
391 unsigned long prot_val)
393 return __ioremap_caller(phys_addr, size, (prot_val & _PAGE_CACHE_MASK),
394 __builtin_return_address(0));
396 EXPORT_SYMBOL(ioremap_prot);
399 * iounmap - Free a IO remapping
400 * @addr: virtual address from ioremap_*
402 * Caller must ensure there is only one unmapping for the same pointer.
404 void iounmap(volatile void __iomem *addr)
406 struct vm_struct *p, *o;
408 if ((void __force *)addr <= high_memory)
409 return;
412 * __ioremap special-cases the PCI/ISA range by not instantiating a
413 * vm_area and by simply returning an address into the kernel mapping
414 * of ISA space. So handle that here.
416 if ((void __force *)addr >= phys_to_virt(ISA_START_ADDRESS) &&
417 (void __force *)addr < phys_to_virt(ISA_END_ADDRESS))
418 return;
420 addr = (volatile void __iomem *)
421 (PAGE_MASK & (unsigned long __force)addr);
423 mmiotrace_iounmap(addr);
425 /* Use the vm area unlocked, assuming the caller
426 ensures there isn't another iounmap for the same address
427 in parallel. Reuse of the virtual address is prevented by
428 leaving it in the global lists until we're done with it.
429 cpa takes care of the direct mappings. */
430 read_lock(&vmlist_lock);
431 for (p = vmlist; p; p = p->next) {
432 if (p->addr == (void __force *)addr)
433 break;
435 read_unlock(&vmlist_lock);
437 if (!p) {
438 printk(KERN_ERR "iounmap: bad address %p\n", addr);
439 dump_stack();
440 return;
443 free_memtype(p->phys_addr, p->phys_addr + get_vm_area_size(p));
445 /* Finally remove it */
446 o = remove_vm_area((void __force *)addr);
447 BUG_ON(p != o || o == NULL);
448 kfree(p);
450 EXPORT_SYMBOL(iounmap);
453 * Convert a physical pointer to a virtual kernel pointer for /dev/mem
454 * access
456 void *xlate_dev_mem_ptr(unsigned long phys)
458 void *addr;
459 unsigned long start = phys & PAGE_MASK;
461 /* If page is RAM, we can use __va. Otherwise ioremap and unmap. */
462 if (page_is_ram(start >> PAGE_SHIFT))
463 return __va(phys);
465 addr = (void __force *)ioremap_default(start, PAGE_SIZE);
466 if (addr)
467 addr = (void *)((unsigned long)addr | (phys & ~PAGE_MASK));
469 return addr;
472 void unxlate_dev_mem_ptr(unsigned long phys, void *addr)
474 if (page_is_ram(phys >> PAGE_SHIFT))
475 return;
477 iounmap((void __iomem *)((unsigned long)addr & PAGE_MASK));
478 return;
481 static int __initdata early_ioremap_debug;
483 static int __init early_ioremap_debug_setup(char *str)
485 early_ioremap_debug = 1;
487 return 0;
489 early_param("early_ioremap_debug", early_ioremap_debug_setup);
491 static __initdata int after_paging_init;
492 static pte_t bm_pte[PAGE_SIZE/sizeof(pte_t)] __page_aligned_bss;
494 static inline pmd_t * __init early_ioremap_pmd(unsigned long addr)
496 /* Don't assume we're using swapper_pg_dir at this point */
497 pgd_t *base = __va(read_cr3());
498 pgd_t *pgd = &base[pgd_index(addr)];
499 pud_t *pud = pud_offset(pgd, addr);
500 pmd_t *pmd = pmd_offset(pud, addr);
502 return pmd;
505 static inline pte_t * __init early_ioremap_pte(unsigned long addr)
507 return &bm_pte[pte_index(addr)];
510 static unsigned long slot_virt[FIX_BTMAPS_SLOTS] __initdata;
512 void __init early_ioremap_init(void)
514 pmd_t *pmd;
515 int i;
517 if (early_ioremap_debug)
518 printk(KERN_INFO "early_ioremap_init()\n");
520 for (i = 0; i < FIX_BTMAPS_SLOTS; i++)
521 slot_virt[i] = __fix_to_virt(FIX_BTMAP_BEGIN - NR_FIX_BTMAPS*i);
523 pmd = early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN));
524 memset(bm_pte, 0, sizeof(bm_pte));
525 pmd_populate_kernel(&init_mm, pmd, bm_pte);
528 * The boot-ioremap range spans multiple pmds, for which
529 * we are not prepared:
531 if (pmd != early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END))) {
532 WARN_ON(1);
533 printk(KERN_WARNING "pmd %p != %p\n",
534 pmd, early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END)));
535 printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
536 fix_to_virt(FIX_BTMAP_BEGIN));
537 printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_END): %08lx\n",
538 fix_to_virt(FIX_BTMAP_END));
540 printk(KERN_WARNING "FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
541 printk(KERN_WARNING "FIX_BTMAP_BEGIN: %d\n",
542 FIX_BTMAP_BEGIN);
546 void __init early_ioremap_reset(void)
548 after_paging_init = 1;
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);
555 pte_t *pte;
557 if (idx >= __end_of_fixed_addresses) {
558 BUG();
559 return;
561 pte = early_ioremap_pte(addr);
563 if (pgprot_val(flags))
564 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
565 else
566 pte_clear(&init_mm, addr, pte);
567 __flush_tlb_one(addr);
570 static inline void __init early_set_fixmap(enum fixed_addresses idx,
571 unsigned long phys, pgprot_t prot)
573 if (after_paging_init)
574 __set_fixmap(idx, phys, prot);
575 else
576 __early_set_fixmap(idx, phys, prot);
579 static inline void __init early_clear_fixmap(enum fixed_addresses idx)
581 if (after_paging_init)
582 clear_fixmap(idx);
583 else
584 __early_set_fixmap(idx, 0, __pgprot(0));
587 static void __iomem *prev_map[FIX_BTMAPS_SLOTS] __initdata;
588 static unsigned long prev_size[FIX_BTMAPS_SLOTS] __initdata;
590 static int __init check_early_ioremap_leak(void)
592 int count = 0;
593 int i;
595 for (i = 0; i < FIX_BTMAPS_SLOTS; i++)
596 if (prev_map[i])
597 count++;
599 if (!count)
600 return 0;
601 WARN(1, KERN_WARNING
602 "Debug warning: early ioremap leak of %d areas detected.\n",
603 count);
604 printk(KERN_WARNING
605 "please boot with early_ioremap_debug and report the dmesg.\n");
607 return 1;
609 late_initcall(check_early_ioremap_leak);
611 static void __init __iomem *
612 __early_ioremap(unsigned long phys_addr, unsigned long size, pgprot_t prot)
614 unsigned long offset, last_addr;
615 unsigned int nrpages;
616 enum fixed_addresses idx0, idx;
617 int i, slot;
619 WARN_ON(system_state != SYSTEM_BOOTING);
621 slot = -1;
622 for (i = 0; i < FIX_BTMAPS_SLOTS; i++) {
623 if (!prev_map[i]) {
624 slot = i;
625 break;
629 if (slot < 0) {
630 printk(KERN_INFO "early_iomap(%08lx, %08lx) not found slot\n",
631 phys_addr, size);
632 WARN_ON(1);
633 return NULL;
636 if (early_ioremap_debug) {
637 printk(KERN_INFO "early_ioremap(%08lx, %08lx) [%d] => ",
638 phys_addr, size, slot);
639 dump_stack();
642 /* Don't allow wraparound or zero size */
643 last_addr = phys_addr + size - 1;
644 if (!size || last_addr < phys_addr) {
645 WARN_ON(1);
646 return NULL;
649 prev_size[slot] = size;
651 * Mappings have to be page-aligned
653 offset = phys_addr & ~PAGE_MASK;
654 phys_addr &= PAGE_MASK;
655 size = PAGE_ALIGN(last_addr + 1) - phys_addr;
658 * Mappings have to fit in the FIX_BTMAP area.
660 nrpages = size >> PAGE_SHIFT;
661 if (nrpages > NR_FIX_BTMAPS) {
662 WARN_ON(1);
663 return NULL;
667 * Ok, go for it..
669 idx0 = FIX_BTMAP_BEGIN - NR_FIX_BTMAPS*slot;
670 idx = idx0;
671 while (nrpages > 0) {
672 early_set_fixmap(idx, phys_addr, prot);
673 phys_addr += PAGE_SIZE;
674 --idx;
675 --nrpages;
677 if (early_ioremap_debug)
678 printk(KERN_CONT "%08lx + %08lx\n", offset, slot_virt[slot]);
680 prev_map[slot] = (void __iomem *)(offset + slot_virt[slot]);
681 return prev_map[slot];
684 /* Remap an IO device */
685 void __init __iomem *early_ioremap(unsigned long phys_addr, unsigned long size)
687 return __early_ioremap(phys_addr, size, PAGE_KERNEL_IO);
690 /* Remap memory */
691 void __init __iomem *early_memremap(unsigned long phys_addr, unsigned long size)
693 return __early_ioremap(phys_addr, size, PAGE_KERNEL);
696 void __init early_iounmap(void __iomem *addr, unsigned long size)
698 unsigned long virt_addr;
699 unsigned long offset;
700 unsigned int nrpages;
701 enum fixed_addresses idx;
702 int i, slot;
704 slot = -1;
705 for (i = 0; i < FIX_BTMAPS_SLOTS; i++) {
706 if (prev_map[i] == addr) {
707 slot = i;
708 break;
712 if (slot < 0) {
713 printk(KERN_INFO "early_iounmap(%p, %08lx) not found slot\n",
714 addr, size);
715 WARN_ON(1);
716 return;
719 if (prev_size[slot] != size) {
720 printk(KERN_INFO "early_iounmap(%p, %08lx) [%d] size not consistent %08lx\n",
721 addr, size, slot, prev_size[slot]);
722 WARN_ON(1);
723 return;
726 if (early_ioremap_debug) {
727 printk(KERN_INFO "early_iounmap(%p, %08lx) [%d]\n", addr,
728 size, slot);
729 dump_stack();
732 virt_addr = (unsigned long)addr;
733 if (virt_addr < fix_to_virt(FIX_BTMAP_BEGIN)) {
734 WARN_ON(1);
735 return;
737 offset = virt_addr & ~PAGE_MASK;
738 nrpages = PAGE_ALIGN(offset + size - 1) >> PAGE_SHIFT;
740 idx = FIX_BTMAP_BEGIN - NR_FIX_BTMAPS*slot;
741 while (nrpages > 0) {
742 early_clear_fixmap(idx);
743 --idx;
744 --nrpages;
746 prev_map[slot] = NULL;