2 * linux/arch/arm/mm/mmu.c
4 * Copyright (C) 1995-2005 Russell King
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
10 #include <linux/module.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/init.h>
14 #include <linux/mman.h>
15 #include <linux/nodemask.h>
16 #include <linux/memblock.h>
19 #include <asm/cputype.h>
20 #include <asm/sections.h>
21 #include <asm/cachetype.h>
22 #include <asm/setup.h>
23 #include <asm/sizes.h>
24 #include <asm/smp_plat.h>
26 #include <asm/highmem.h>
27 #include <asm/traps.h>
29 #include <asm/mach/arch.h>
30 #include <asm/mach/map.h>
35 * empty_zero_page is a special page that is used for
36 * zero-initialized data and COW.
38 struct page *empty_zero_page;
39 EXPORT_SYMBOL(empty_zero_page);
42 * The pmd table for the upper-most set of pages.
46 #define CPOLICY_UNCACHED 0
47 #define CPOLICY_BUFFERED 1
48 #define CPOLICY_WRITETHROUGH 2
49 #define CPOLICY_WRITEBACK 3
50 #define CPOLICY_WRITEALLOC 4
52 static unsigned int cachepolicy __initdata = CPOLICY_WRITEBACK;
53 static unsigned int ecc_mask __initdata = 0;
55 pgprot_t pgprot_kernel;
57 EXPORT_SYMBOL(pgprot_user);
58 EXPORT_SYMBOL(pgprot_kernel);
61 const char policy[16];
67 static struct cachepolicy cache_policies[] __initdata = {
71 .pmd = PMD_SECT_UNCACHED,
72 .pte = L_PTE_MT_UNCACHED,
76 .pmd = PMD_SECT_BUFFERED,
77 .pte = L_PTE_MT_BUFFERABLE,
79 .policy = "writethrough",
82 .pte = L_PTE_MT_WRITETHROUGH,
84 .policy = "writeback",
87 .pte = L_PTE_MT_WRITEBACK,
89 .policy = "writealloc",
92 .pte = L_PTE_MT_WRITEALLOC,
97 * These are useful for identifying cache coherency
98 * problems by allowing the cache or the cache and
99 * writebuffer to be turned off. (Note: the write
100 * buffer should not be on and the cache off).
102 static int __init early_cachepolicy(char *p)
106 for (i = 0; i < ARRAY_SIZE(cache_policies); i++) {
107 int len = strlen(cache_policies[i].policy);
109 if (memcmp(p, cache_policies[i].policy, len) == 0) {
111 cr_alignment &= ~cache_policies[i].cr_mask;
112 cr_no_alignment &= ~cache_policies[i].cr_mask;
116 if (i == ARRAY_SIZE(cache_policies))
117 printk(KERN_ERR "ERROR: unknown or unsupported cache policy\n");
119 * This restriction is partly to do with the way we boot; it is
120 * unpredictable to have memory mapped using two different sets of
121 * memory attributes (shared, type, and cache attribs). We can not
122 * change these attributes once the initial assembly has setup the
125 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
126 printk(KERN_WARNING "Only cachepolicy=writeback supported on ARMv6 and later\n");
127 cachepolicy = CPOLICY_WRITEBACK;
130 set_cr(cr_alignment);
133 early_param("cachepolicy", early_cachepolicy);
135 static int __init early_nocache(char *__unused)
137 char *p = "buffered";
138 printk(KERN_WARNING "nocache is deprecated; use cachepolicy=%s\n", p);
139 early_cachepolicy(p);
142 early_param("nocache", early_nocache);
144 static int __init early_nowrite(char *__unused)
146 char *p = "uncached";
147 printk(KERN_WARNING "nowb is deprecated; use cachepolicy=%s\n", p);
148 early_cachepolicy(p);
151 early_param("nowb", early_nowrite);
153 static int __init early_ecc(char *p)
155 if (memcmp(p, "on", 2) == 0)
156 ecc_mask = PMD_PROTECTION;
157 else if (memcmp(p, "off", 3) == 0)
161 early_param("ecc", early_ecc);
163 static int __init noalign_setup(char *__unused)
165 cr_alignment &= ~CR_A;
166 cr_no_alignment &= ~CR_A;
167 set_cr(cr_alignment);
170 __setup("noalign", noalign_setup);
173 void adjust_cr(unsigned long mask, unsigned long set)
181 local_irq_save(flags);
183 cr_no_alignment = (cr_no_alignment & ~mask) | set;
184 cr_alignment = (cr_alignment & ~mask) | set;
186 set_cr((get_cr() & ~mask) | set);
188 local_irq_restore(flags);
192 #define PROT_PTE_DEVICE L_PTE_PRESENT|L_PTE_YOUNG|L_PTE_DIRTY|L_PTE_XN
193 #define PROT_SECT_DEVICE PMD_TYPE_SECT|PMD_SECT_AP_WRITE
195 static struct mem_type mem_types[] = {
196 [MT_DEVICE] = { /* Strongly ordered / ARMv6 shared device */
197 .prot_pte = PROT_PTE_DEVICE | L_PTE_MT_DEV_SHARED |
199 .prot_l1 = PMD_TYPE_TABLE,
200 .prot_sect = PROT_SECT_DEVICE | PMD_SECT_S,
203 [MT_DEVICE_NONSHARED] = { /* ARMv6 non-shared device */
204 .prot_pte = PROT_PTE_DEVICE | L_PTE_MT_DEV_NONSHARED,
205 .prot_l1 = PMD_TYPE_TABLE,
206 .prot_sect = PROT_SECT_DEVICE,
209 [MT_DEVICE_CACHED] = { /* ioremap_cached */
210 .prot_pte = PROT_PTE_DEVICE | L_PTE_MT_DEV_CACHED,
211 .prot_l1 = PMD_TYPE_TABLE,
212 .prot_sect = PROT_SECT_DEVICE | PMD_SECT_WB,
215 [MT_DEVICE_STRONGLY_ORDERED] = { /* Guaranteed strongly ordered */
216 .prot_pte = PROT_PTE_DEVICE,
217 .prot_l1 = PMD_TYPE_TABLE,
218 .prot_sect = PROT_SECT_DEVICE | PMD_SECT_UNCACHED,
221 [MT_DEVICE_WC] = { /* ioremap_wc */
222 .prot_pte = PROT_PTE_DEVICE | L_PTE_MT_DEV_WC,
223 .prot_l1 = PMD_TYPE_TABLE,
224 .prot_sect = PROT_SECT_DEVICE,
228 .prot_pte = PROT_PTE_DEVICE,
229 .prot_l1 = PMD_TYPE_TABLE,
230 .prot_sect = PMD_TYPE_SECT | PMD_SECT_XN,
234 .prot_sect = PMD_TYPE_SECT | PMD_SECT_XN,
235 .domain = DOMAIN_KERNEL,
238 .prot_sect = PMD_TYPE_SECT | PMD_SECT_XN | PMD_SECT_MINICACHE,
239 .domain = DOMAIN_KERNEL,
242 .prot_pte = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
244 .prot_l1 = PMD_TYPE_TABLE,
245 .domain = DOMAIN_USER,
247 [MT_HIGH_VECTORS] = {
248 .prot_pte = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
249 L_PTE_USER | L_PTE_RDONLY,
250 .prot_l1 = PMD_TYPE_TABLE,
251 .domain = DOMAIN_USER,
254 .prot_pte = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY,
255 .prot_l1 = PMD_TYPE_TABLE,
256 .prot_sect = PMD_TYPE_SECT | PMD_SECT_AP_WRITE,
257 .domain = DOMAIN_KERNEL,
260 .prot_sect = PMD_TYPE_SECT,
261 .domain = DOMAIN_KERNEL,
263 [MT_MEMORY_NONCACHED] = {
264 .prot_pte = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
266 .prot_l1 = PMD_TYPE_TABLE,
267 .prot_sect = PMD_TYPE_SECT | PMD_SECT_AP_WRITE,
268 .domain = DOMAIN_KERNEL,
271 .prot_pte = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
273 .prot_l1 = PMD_TYPE_TABLE,
274 .prot_sect = PMD_TYPE_SECT | PMD_SECT_XN,
275 .domain = DOMAIN_KERNEL,
278 .prot_pte = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY,
279 .prot_l1 = PMD_TYPE_TABLE,
280 .domain = DOMAIN_KERNEL,
284 const struct mem_type *get_mem_type(unsigned int type)
286 return type < ARRAY_SIZE(mem_types) ? &mem_types[type] : NULL;
288 EXPORT_SYMBOL(get_mem_type);
291 * Adjust the PMD section entries according to the CPU in use.
293 static void __init build_mem_type_table(void)
295 struct cachepolicy *cp;
296 unsigned int cr = get_cr();
297 unsigned int user_pgprot, kern_pgprot, vecs_pgprot;
298 int cpu_arch = cpu_architecture();
301 if (cpu_arch < CPU_ARCH_ARMv6) {
302 #if defined(CONFIG_CPU_DCACHE_DISABLE)
303 if (cachepolicy > CPOLICY_BUFFERED)
304 cachepolicy = CPOLICY_BUFFERED;
305 #elif defined(CONFIG_CPU_DCACHE_WRITETHROUGH)
306 if (cachepolicy > CPOLICY_WRITETHROUGH)
307 cachepolicy = CPOLICY_WRITETHROUGH;
310 if (cpu_arch < CPU_ARCH_ARMv5) {
311 if (cachepolicy >= CPOLICY_WRITEALLOC)
312 cachepolicy = CPOLICY_WRITEBACK;
315 #ifndef CONFIG_PLAT_RK
317 cachepolicy = CPOLICY_WRITEALLOC;
321 * Strip out features not present on earlier architectures.
322 * Pre-ARMv5 CPUs don't have TEX bits. Pre-ARMv6 CPUs or those
323 * without extended page tables don't have the 'Shared' bit.
325 if (cpu_arch < CPU_ARCH_ARMv5)
326 for (i = 0; i < ARRAY_SIZE(mem_types); i++)
327 mem_types[i].prot_sect &= ~PMD_SECT_TEX(7);
328 if ((cpu_arch < CPU_ARCH_ARMv6 || !(cr & CR_XP)) && !cpu_is_xsc3())
329 for (i = 0; i < ARRAY_SIZE(mem_types); i++)
330 mem_types[i].prot_sect &= ~PMD_SECT_S;
333 * ARMv5 and lower, bit 4 must be set for page tables (was: cache
334 * "update-able on write" bit on ARM610). However, Xscale and
335 * Xscale3 require this bit to be cleared.
337 if (cpu_is_xscale() || cpu_is_xsc3()) {
338 for (i = 0; i < ARRAY_SIZE(mem_types); i++) {
339 mem_types[i].prot_sect &= ~PMD_BIT4;
340 mem_types[i].prot_l1 &= ~PMD_BIT4;
342 } else if (cpu_arch < CPU_ARCH_ARMv6) {
343 for (i = 0; i < ARRAY_SIZE(mem_types); i++) {
344 if (mem_types[i].prot_l1)
345 mem_types[i].prot_l1 |= PMD_BIT4;
346 if (mem_types[i].prot_sect)
347 mem_types[i].prot_sect |= PMD_BIT4;
352 * Mark the device areas according to the CPU/architecture.
354 if (cpu_is_xsc3() || (cpu_arch >= CPU_ARCH_ARMv6 && (cr & CR_XP))) {
355 if (!cpu_is_xsc3()) {
357 * Mark device regions on ARMv6+ as execute-never
358 * to prevent speculative instruction fetches.
360 mem_types[MT_DEVICE].prot_sect |= PMD_SECT_XN;
361 mem_types[MT_DEVICE_NONSHARED].prot_sect |= PMD_SECT_XN;
362 mem_types[MT_DEVICE_CACHED].prot_sect |= PMD_SECT_XN;
363 mem_types[MT_DEVICE_WC].prot_sect |= PMD_SECT_XN;
364 mem_types[MT_DEVICE_STRONGLY_ORDERED].prot_sect |= PMD_SECT_XN;
366 if (cpu_arch >= CPU_ARCH_ARMv7 && (cr & CR_TRE)) {
368 * For ARMv7 with TEX remapping,
369 * - shared device is SXCB=1100
370 * - nonshared device is SXCB=0100
371 * - write combine device mem is SXCB=0001
372 * (Uncached Normal memory)
374 mem_types[MT_DEVICE].prot_sect |= PMD_SECT_TEX(1);
375 mem_types[MT_DEVICE_NONSHARED].prot_sect |= PMD_SECT_TEX(1);
376 mem_types[MT_DEVICE_WC].prot_sect |= PMD_SECT_BUFFERABLE;
377 } else if (cpu_is_xsc3()) {
380 * - shared device is TEXCB=00101
381 * - nonshared device is TEXCB=01000
382 * - write combine device mem is TEXCB=00100
383 * (Inner/Outer Uncacheable in xsc3 parlance)
385 mem_types[MT_DEVICE].prot_sect |= PMD_SECT_TEX(1) | PMD_SECT_BUFFERED;
386 mem_types[MT_DEVICE_NONSHARED].prot_sect |= PMD_SECT_TEX(2);
387 mem_types[MT_DEVICE_WC].prot_sect |= PMD_SECT_TEX(1);
390 * For ARMv6 and ARMv7 without TEX remapping,
391 * - shared device is TEXCB=00001
392 * - nonshared device is TEXCB=01000
393 * - write combine device mem is TEXCB=00100
394 * (Uncached Normal in ARMv6 parlance).
396 mem_types[MT_DEVICE].prot_sect |= PMD_SECT_BUFFERED;
397 mem_types[MT_DEVICE_NONSHARED].prot_sect |= PMD_SECT_TEX(2);
398 mem_types[MT_DEVICE_WC].prot_sect |= PMD_SECT_TEX(1);
402 * On others, write combining is "Uncached/Buffered"
404 mem_types[MT_DEVICE_WC].prot_sect |= PMD_SECT_BUFFERABLE;
408 * Now deal with the memory-type mappings
410 cp = &cache_policies[cachepolicy];
411 vecs_pgprot = kern_pgprot = user_pgprot = cp->pte;
414 * Only use write-through for non-SMP systems
416 if (!is_smp() && cpu_arch >= CPU_ARCH_ARMv5 && cachepolicy > CPOLICY_WRITETHROUGH)
417 vecs_pgprot = cache_policies[CPOLICY_WRITETHROUGH].pte;
420 * Enable CPU-specific coherency if supported.
421 * (Only available on XSC3 at the moment.)
423 if (arch_is_coherent() && cpu_is_xsc3()) {
424 mem_types[MT_MEMORY].prot_sect |= PMD_SECT_S;
425 mem_types[MT_MEMORY].prot_pte |= L_PTE_SHARED;
426 mem_types[MT_MEMORY_NONCACHED].prot_sect |= PMD_SECT_S;
427 mem_types[MT_MEMORY_NONCACHED].prot_pte |= L_PTE_SHARED;
430 * ARMv6 and above have extended page tables.
432 if (cpu_arch >= CPU_ARCH_ARMv6 && (cr & CR_XP)) {
434 * Mark cache clean areas and XIP ROM read only
435 * from SVC mode and no access from userspace.
437 mem_types[MT_ROM].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
438 mem_types[MT_MINICLEAN].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
439 mem_types[MT_CACHECLEAN].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
443 * Mark memory with the "shared" attribute
446 user_pgprot |= L_PTE_SHARED;
447 kern_pgprot |= L_PTE_SHARED;
448 vecs_pgprot |= L_PTE_SHARED;
449 mem_types[MT_DEVICE_WC].prot_sect |= PMD_SECT_S;
450 mem_types[MT_DEVICE_WC].prot_pte |= L_PTE_SHARED;
451 mem_types[MT_DEVICE_CACHED].prot_sect |= PMD_SECT_S;
452 mem_types[MT_DEVICE_CACHED].prot_pte |= L_PTE_SHARED;
453 mem_types[MT_MEMORY].prot_sect |= PMD_SECT_S;
454 mem_types[MT_MEMORY].prot_pte |= L_PTE_SHARED;
455 mem_types[MT_MEMORY_NONCACHED].prot_sect |= PMD_SECT_S;
456 mem_types[MT_MEMORY_NONCACHED].prot_pte |= L_PTE_SHARED;
461 * Non-cacheable Normal - intended for memory areas that must
462 * not cause dirty cache line writebacks when used
464 if (cpu_arch >= CPU_ARCH_ARMv6) {
465 if (cpu_arch >= CPU_ARCH_ARMv7 && (cr & CR_TRE)) {
466 /* Non-cacheable Normal is XCB = 001 */
467 mem_types[MT_MEMORY_NONCACHED].prot_sect |=
470 /* For both ARMv6 and non-TEX-remapping ARMv7 */
471 mem_types[MT_MEMORY_NONCACHED].prot_sect |=
475 mem_types[MT_MEMORY_NONCACHED].prot_sect |= PMD_SECT_BUFFERABLE;
478 for (i = 0; i < 16; i++) {
479 pteval_t v = pgprot_val(protection_map[i]);
480 protection_map[i] = __pgprot(v | user_pgprot);
483 mem_types[MT_LOW_VECTORS].prot_pte |= vecs_pgprot;
484 mem_types[MT_HIGH_VECTORS].prot_pte |= vecs_pgprot;
486 pgprot_user = __pgprot(L_PTE_PRESENT | L_PTE_YOUNG | user_pgprot);
487 pgprot_kernel = __pgprot(L_PTE_PRESENT | L_PTE_YOUNG |
488 L_PTE_DIRTY | kern_pgprot);
490 mem_types[MT_LOW_VECTORS].prot_l1 |= ecc_mask;
491 mem_types[MT_HIGH_VECTORS].prot_l1 |= ecc_mask;
492 mem_types[MT_MEMORY].prot_sect |= ecc_mask | cp->pmd;
493 mem_types[MT_MEMORY].prot_pte |= kern_pgprot;
494 mem_types[MT_MEMORY_NONCACHED].prot_sect |= ecc_mask;
495 mem_types[MT_ROM].prot_sect |= cp->pmd;
499 mem_types[MT_CACHECLEAN].prot_sect |= PMD_SECT_WT;
503 mem_types[MT_CACHECLEAN].prot_sect |= PMD_SECT_WB;
506 printk("Memory policy: ECC %sabled, Data cache %s\n",
507 ecc_mask ? "en" : "dis", cp->policy);
509 for (i = 0; i < ARRAY_SIZE(mem_types); i++) {
510 struct mem_type *t = &mem_types[i];
512 t->prot_l1 |= PMD_DOMAIN(t->domain);
514 t->prot_sect |= PMD_DOMAIN(t->domain);
518 #ifdef CONFIG_ARM_DMA_MEM_BUFFERABLE
519 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
520 unsigned long size, pgprot_t vma_prot)
523 return pgprot_noncached(vma_prot);
524 else if (file->f_flags & O_SYNC)
525 return pgprot_writecombine(vma_prot);
528 EXPORT_SYMBOL(phys_mem_access_prot);
531 #define vectors_base() (vectors_high() ? 0xffff0000 : 0)
533 static void __init *early_alloc(unsigned long sz)
535 void *ptr = __va(memblock_alloc(sz, sz));
540 static pte_t * __init early_pte_alloc(pmd_t *pmd, unsigned long addr, unsigned long prot)
542 if (pmd_none(*pmd)) {
543 pte_t *pte = early_alloc(PTE_HWTABLE_OFF + PTE_HWTABLE_SIZE);
544 __pmd_populate(pmd, __pa(pte), prot);
546 BUG_ON(pmd_bad(*pmd));
547 return pte_offset_kernel(pmd, addr);
550 static void __init alloc_init_pte(pmd_t *pmd, unsigned long addr,
551 unsigned long end, unsigned long pfn,
552 const struct mem_type *type)
554 pte_t *pte = early_pte_alloc(pmd, addr, type->prot_l1);
556 set_pte_ext(pte, pfn_pte(pfn, __pgprot(type->prot_pte)), 0);
558 } while (pte++, addr += PAGE_SIZE, addr != end);
561 static void __init alloc_init_section(pud_t *pud, unsigned long addr,
562 unsigned long end, phys_addr_t phys,
563 const struct mem_type *type)
565 pmd_t *pmd = pmd_offset(pud, addr);
568 * Try a section mapping - end, addr and phys must all be aligned
569 * to a section boundary. Note that PMDs refer to the individual
570 * L1 entries, whereas PGDs refer to a group of L1 entries making
571 * up one logical pointer to an L2 table.
573 if (((addr | end | phys) & ~SECTION_MASK) == 0) {
576 if (addr & SECTION_SIZE)
580 *pmd = __pmd(phys | type->prot_sect);
581 phys += SECTION_SIZE;
582 } while (pmd++, addr += SECTION_SIZE, addr != end);
587 * No need to loop; pte's aren't interested in the
588 * individual L1 entries.
590 alloc_init_pte(pmd, addr, end, __phys_to_pfn(phys), type);
594 static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
595 unsigned long phys, const struct mem_type *type)
597 pud_t *pud = pud_offset(pgd, addr);
601 next = pud_addr_end(addr, end);
602 alloc_init_section(pud, addr, next, phys, type);
604 } while (pud++, addr = next, addr != end);
607 static void __init create_36bit_mapping(struct map_desc *md,
608 const struct mem_type *type)
610 unsigned long addr, length, end;
615 phys = __pfn_to_phys(md->pfn);
616 length = PAGE_ALIGN(md->length);
618 if (!(cpu_architecture() >= CPU_ARCH_ARMv6 || cpu_is_xsc3())) {
619 printk(KERN_ERR "MM: CPU does not support supersection "
620 "mapping for 0x%08llx at 0x%08lx\n",
621 (long long)__pfn_to_phys((u64)md->pfn), addr);
625 /* N.B. ARMv6 supersections are only defined to work with domain 0.
626 * Since domain assignments can in fact be arbitrary, the
627 * 'domain == 0' check below is required to insure that ARMv6
628 * supersections are only allocated for domain 0 regardless
629 * of the actual domain assignments in use.
632 printk(KERN_ERR "MM: invalid domain in supersection "
633 "mapping for 0x%08llx at 0x%08lx\n",
634 (long long)__pfn_to_phys((u64)md->pfn), addr);
638 if ((addr | length | __pfn_to_phys(md->pfn)) & ~SUPERSECTION_MASK) {
639 printk(KERN_ERR "MM: cannot create mapping for 0x%08llx"
640 " at 0x%08lx invalid alignment\n",
641 (long long)__pfn_to_phys((u64)md->pfn), addr);
646 * Shift bits [35:32] of address into bits [23:20] of PMD
649 phys |= (((md->pfn >> (32 - PAGE_SHIFT)) & 0xF) << 20);
651 pgd = pgd_offset_k(addr);
654 pud_t *pud = pud_offset(pgd, addr);
655 pmd_t *pmd = pmd_offset(pud, addr);
658 for (i = 0; i < 16; i++)
659 *pmd++ = __pmd(phys | type->prot_sect | PMD_SECT_SUPER);
661 addr += SUPERSECTION_SIZE;
662 phys += SUPERSECTION_SIZE;
663 pgd += SUPERSECTION_SIZE >> PGDIR_SHIFT;
664 } while (addr != end);
668 * Create the page directory entries and any necessary
669 * page tables for the mapping specified by `md'. We
670 * are able to cope here with varying sizes and address
671 * offsets, and we take full advantage of sections and
674 static void __init create_mapping(struct map_desc *md)
676 unsigned long addr, length, end;
678 const struct mem_type *type;
681 if (md->virtual != vectors_base() && md->virtual < TASK_SIZE) {
682 printk(KERN_WARNING "BUG: not creating mapping for 0x%08llx"
683 " at 0x%08lx in user region\n",
684 (long long)__pfn_to_phys((u64)md->pfn), md->virtual);
688 if ((md->type == MT_DEVICE || md->type == MT_ROM) &&
689 md->virtual >= PAGE_OFFSET && md->virtual < VMALLOC_END) {
690 printk(KERN_WARNING "BUG: mapping for 0x%08llx"
691 " at 0x%08lx overlaps vmalloc space\n",
692 (long long)__pfn_to_phys((u64)md->pfn), md->virtual);
695 type = &mem_types[md->type];
698 * Catch 36-bit addresses
700 if (md->pfn >= 0x100000) {
701 create_36bit_mapping(md, type);
705 addr = md->virtual & PAGE_MASK;
706 phys = __pfn_to_phys(md->pfn);
707 length = PAGE_ALIGN(md->length + (md->virtual & ~PAGE_MASK));
709 if (type->prot_l1 == 0 && ((addr | phys | length) & ~SECTION_MASK)) {
710 printk(KERN_WARNING "BUG: map for 0x%08llx at 0x%08lx can not "
711 "be mapped using pages, ignoring.\n",
712 (long long)__pfn_to_phys(md->pfn), addr);
716 pgd = pgd_offset_k(addr);
719 unsigned long next = pgd_addr_end(addr, end);
721 alloc_init_pud(pgd, addr, next, phys, type);
725 } while (pgd++, addr != end);
729 * Create the architecture specific mappings
731 void __init iotable_init(struct map_desc *io_desc, int nr)
735 for (i = 0; i < nr; i++)
736 create_mapping(io_desc + i);
739 #if defined(CONFIG_PLAT_RK)
740 static void * __initdata vmalloc_min = (void *)(VMALLOC_END - SZ_512M);
742 static void * __initdata vmalloc_min = (void *)(VMALLOC_END - SZ_128M);
746 * vmalloc=size forces the vmalloc area to be exactly 'size'
747 * bytes. This can be used to increase (or decrease) the vmalloc
748 * area - the default is 128m.
750 static int __init early_vmalloc(char *arg)
752 unsigned long vmalloc_reserve = memparse(arg, NULL);
754 if (vmalloc_reserve < SZ_16M) {
755 vmalloc_reserve = SZ_16M;
757 "vmalloc area too small, limiting to %luMB\n",
758 vmalloc_reserve >> 20);
761 if (vmalloc_reserve > VMALLOC_END - (PAGE_OFFSET + SZ_32M)) {
762 vmalloc_reserve = VMALLOC_END - (PAGE_OFFSET + SZ_32M);
764 "vmalloc area is too big, limiting to %luMB\n",
765 vmalloc_reserve >> 20);
768 vmalloc_min = (void *)(VMALLOC_END - vmalloc_reserve);
771 early_param("vmalloc", early_vmalloc);
773 static phys_addr_t lowmem_limit __initdata = 0;
775 void __init sanity_check_meminfo(void)
777 int i, j, highmem = 0;
779 for (i = 0, j = 0; i < meminfo.nr_banks; i++) {
780 struct membank *bank = &meminfo.bank[j];
781 *bank = meminfo.bank[i];
783 #ifdef CONFIG_HIGHMEM
784 if (__va(bank->start) >= vmalloc_min ||
785 __va(bank->start) < (void *)PAGE_OFFSET)
788 bank->highmem = highmem;
791 * Split those memory banks which are partially overlapping
792 * the vmalloc area greatly simplifying things later.
794 if (__va(bank->start) < vmalloc_min &&
795 bank->size > vmalloc_min - __va(bank->start)) {
796 if (meminfo.nr_banks >= NR_BANKS) {
797 printk(KERN_CRIT "NR_BANKS too low, "
798 "ignoring high memory\n");
800 memmove(bank + 1, bank,
801 (meminfo.nr_banks - i) * sizeof(*bank));
804 bank[1].size -= vmalloc_min - __va(bank->start);
805 bank[1].start = __pa(vmalloc_min - 1) + 1;
806 bank[1].highmem = highmem = 1;
809 bank->size = vmalloc_min - __va(bank->start);
812 bank->highmem = highmem;
815 * Check whether this memory bank would entirely overlap
818 if (__va(bank->start) >= vmalloc_min ||
819 __va(bank->start) < (void *)PAGE_OFFSET) {
820 printk(KERN_NOTICE "Ignoring RAM at %.8llx-%.8llx "
821 "(vmalloc region overlap).\n",
822 (unsigned long long)bank->start,
823 (unsigned long long)bank->start + bank->size - 1);
828 * Check whether this memory bank would partially overlap
831 if (__va(bank->start + bank->size) > vmalloc_min ||
832 __va(bank->start + bank->size) < __va(bank->start)) {
833 unsigned long newsize = vmalloc_min - __va(bank->start);
834 printk(KERN_NOTICE "Truncating RAM at %.8llx-%.8llx "
835 "to -%.8llx (vmalloc region overlap).\n",
836 (unsigned long long)bank->start,
837 (unsigned long long)bank->start + bank->size - 1,
838 (unsigned long long)bank->start + newsize - 1);
839 bank->size = newsize;
842 if (!bank->highmem && bank->start + bank->size > lowmem_limit)
843 lowmem_limit = bank->start + bank->size;
847 #ifdef CONFIG_HIGHMEM
849 const char *reason = NULL;
851 if (cache_is_vipt_aliasing()) {
853 * Interactions between kmap and other mappings
854 * make highmem support with aliasing VIPT caches
857 reason = "with VIPT aliasing cache";
860 printk(KERN_CRIT "HIGHMEM is not supported %s, ignoring high memory\n",
862 while (j > 0 && meminfo.bank[j - 1].highmem)
867 meminfo.nr_banks = j;
868 memblock_set_current_limit(lowmem_limit);
871 static inline void prepare_page_table(void)
877 * Clear out all the mappings below the kernel image.
879 for (addr = 0; addr < MODULES_VADDR; addr += PGDIR_SIZE)
880 pmd_clear(pmd_off_k(addr));
882 #ifdef CONFIG_XIP_KERNEL
883 /* The XIP kernel is mapped in the module area -- skip over it */
884 addr = ((unsigned long)_etext + PGDIR_SIZE - 1) & PGDIR_MASK;
886 for ( ; addr < PAGE_OFFSET; addr += PGDIR_SIZE)
887 pmd_clear(pmd_off_k(addr));
890 * Find the end of the first block of lowmem.
892 end = memblock.memory.regions[0].base + memblock.memory.regions[0].size;
893 if (end >= lowmem_limit)
897 * Clear out all the kernel space mappings, except for the first
898 * memory bank, up to the end of the vmalloc region.
900 for (addr = __phys_to_virt(end);
901 addr < VMALLOC_END; addr += PGDIR_SIZE)
902 pmd_clear(pmd_off_k(addr));
906 * Reserve the special regions of memory
908 void __init arm_mm_memblock_reserve(void)
911 * Reserve the page tables. These are already in use,
912 * and can only be in node 0.
914 memblock_reserve(__pa(swapper_pg_dir), PTRS_PER_PGD * sizeof(pgd_t));
918 * Because of the SA1111 DMA bug, we want to preserve our
919 * precious DMA-able memory...
921 memblock_reserve(PHYS_OFFSET, __pa(swapper_pg_dir) - PHYS_OFFSET);
926 * Set up device the mappings. Since we clear out the page tables for all
927 * mappings above VMALLOC_END, we will remove any debug device mappings.
928 * This means you have to be careful how you debug this function, or any
929 * called function. This means you can't use any function or debugging
930 * method which may touch any device, otherwise the kernel _will_ crash.
932 static void __init devicemaps_init(struct machine_desc *mdesc)
938 * Allocate the vector page early.
940 vectors_page = early_alloc(PAGE_SIZE);
942 for (addr = VMALLOC_END; addr; addr += PGDIR_SIZE)
943 pmd_clear(pmd_off_k(addr));
946 * Map the kernel if it is XIP.
947 * It is always first in the modulearea.
949 #ifdef CONFIG_XIP_KERNEL
950 map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
951 map.virtual = MODULES_VADDR;
952 map.length = ((unsigned long)_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
954 create_mapping(&map);
958 * Map the cache flushing regions.
961 map.pfn = __phys_to_pfn(FLUSH_BASE_PHYS);
962 map.virtual = FLUSH_BASE;
964 map.type = MT_CACHECLEAN;
965 create_mapping(&map);
967 #ifdef FLUSH_BASE_MINICACHE
968 map.pfn = __phys_to_pfn(FLUSH_BASE_PHYS + SZ_1M);
969 map.virtual = FLUSH_BASE_MINICACHE;
971 map.type = MT_MINICLEAN;
972 create_mapping(&map);
976 * Create a mapping for the machine vectors at the high-vectors
977 * location (0xffff0000). If we aren't using high-vectors, also
978 * create a mapping at the low-vectors virtual address.
980 map.pfn = __phys_to_pfn(virt_to_phys(vectors_page));
981 map.virtual = 0xffff0000;
982 map.length = PAGE_SIZE;
983 map.type = MT_HIGH_VECTORS;
984 create_mapping(&map);
986 if (!vectors_high()) {
988 map.type = MT_LOW_VECTORS;
989 create_mapping(&map);
993 * Ask the machine support to map in the statically mapped devices.
999 * Finally flush the caches and tlb to ensure that we're in a
1000 * consistent state wrt the writebuffer. This also ensures that
1001 * any write-allocated cache lines in the vector page are written
1002 * back. After this point, we can start to touch devices again.
1004 local_flush_tlb_all();
1008 static void __init kmap_init(void)
1010 #ifdef CONFIG_HIGHMEM
1011 pkmap_page_table = early_pte_alloc(pmd_off_k(PKMAP_BASE),
1012 PKMAP_BASE, _PAGE_KERNEL_TABLE);
1016 static void __init map_lowmem(void)
1018 struct memblock_region *reg;
1020 /* Map all the lowmem memory banks. */
1021 for_each_memblock(memory, reg) {
1022 phys_addr_t start = reg->base;
1023 phys_addr_t end = start + reg->size;
1024 struct map_desc map;
1026 if (end > lowmem_limit)
1031 map.pfn = __phys_to_pfn(start);
1032 map.virtual = __phys_to_virt(start);
1033 map.length = end - start;
1034 map.type = MT_MEMORY;
1036 create_mapping(&map);
1041 * paging_init() sets up the page tables, initialises the zone memory
1042 * maps, and sets up the zero page, bad page and bad page tables.
1044 void __init paging_init(struct machine_desc *mdesc)
1048 memblock_set_current_limit(lowmem_limit);
1050 build_mem_type_table();
1051 prepare_page_table();
1053 devicemaps_init(mdesc);
1056 top_pmd = pmd_off_k(0xffff0000);
1058 /* allocate the zero page. */
1059 zero_page = early_alloc(PAGE_SIZE);
1063 empty_zero_page = virt_to_page(zero_page);
1064 __flush_dcache_page(NULL, empty_zero_page);