blob: 5821e67cf8c2700150af8c2f5ccb85159edaeb63 [file] [log] [blame]
Russell Kingd111e8f2006-09-27 15:27:33 +01001/*
2 * linux/arch/arm/mm/mmu.c
3 *
4 * Copyright (C) 1995-2005 Russell King
5 *
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.
9 */
Russell Kingae8f1542006-09-27 15:38:34 +010010#include <linux/module.h>
Russell Kingd111e8f2006-09-27 15:27:33 +010011#include <linux/kernel.h>
12#include <linux/errno.h>
13#include <linux/init.h>
14#include <linux/bootmem.h>
15#include <linux/mman.h>
16#include <linux/nodemask.h>
17
18#include <asm/mach-types.h>
19#include <asm/setup.h>
20#include <asm/sizes.h>
21#include <asm/tlb.h>
22
23#include <asm/mach/arch.h>
24#include <asm/mach/map.h>
25
26#include "mm.h"
27
28DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
29
Russell King6ae5a6e2006-09-30 10:50:05 +010030extern void _stext, _etext, __data_start, _end;
Russell Kingd111e8f2006-09-27 15:27:33 +010031extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
32
33/*
34 * empty_zero_page is a special page that is used for
35 * zero-initialized data and COW.
36 */
37struct page *empty_zero_page;
38
39/*
40 * The pmd table for the upper-most set of pages.
41 */
42pmd_t *top_pmd;
43
Russell Kingae8f1542006-09-27 15:38:34 +010044#define CPOLICY_UNCACHED 0
45#define CPOLICY_BUFFERED 1
46#define CPOLICY_WRITETHROUGH 2
47#define CPOLICY_WRITEBACK 3
48#define CPOLICY_WRITEALLOC 4
49
50static unsigned int cachepolicy __initdata = CPOLICY_WRITEBACK;
51static unsigned int ecc_mask __initdata = 0;
Imre_Deak44b18692007-02-11 13:45:13 +010052pgprot_t pgprot_user;
Russell Kingae8f1542006-09-27 15:38:34 +010053pgprot_t pgprot_kernel;
54
Imre_Deak44b18692007-02-11 13:45:13 +010055EXPORT_SYMBOL(pgprot_user);
Russell Kingae8f1542006-09-27 15:38:34 +010056EXPORT_SYMBOL(pgprot_kernel);
57
58struct cachepolicy {
59 const char policy[16];
60 unsigned int cr_mask;
61 unsigned int pmd;
62 unsigned int pte;
63};
64
65static struct cachepolicy cache_policies[] __initdata = {
66 {
67 .policy = "uncached",
68 .cr_mask = CR_W|CR_C,
69 .pmd = PMD_SECT_UNCACHED,
70 .pte = 0,
71 }, {
72 .policy = "buffered",
73 .cr_mask = CR_C,
74 .pmd = PMD_SECT_BUFFERED,
75 .pte = PTE_BUFFERABLE,
76 }, {
77 .policy = "writethrough",
78 .cr_mask = 0,
79 .pmd = PMD_SECT_WT,
80 .pte = PTE_CACHEABLE,
81 }, {
82 .policy = "writeback",
83 .cr_mask = 0,
84 .pmd = PMD_SECT_WB,
85 .pte = PTE_BUFFERABLE|PTE_CACHEABLE,
86 }, {
87 .policy = "writealloc",
88 .cr_mask = 0,
89 .pmd = PMD_SECT_WBWA,
90 .pte = PTE_BUFFERABLE|PTE_CACHEABLE,
91 }
92};
93
94/*
95 * These are useful for identifing cache coherency
96 * problems by allowing the cache or the cache and
97 * writebuffer to be turned off. (Note: the write
98 * buffer should not be on and the cache off).
99 */
100static void __init early_cachepolicy(char **p)
101{
102 int i;
103
104 for (i = 0; i < ARRAY_SIZE(cache_policies); i++) {
105 int len = strlen(cache_policies[i].policy);
106
107 if (memcmp(*p, cache_policies[i].policy, len) == 0) {
108 cachepolicy = i;
109 cr_alignment &= ~cache_policies[i].cr_mask;
110 cr_no_alignment &= ~cache_policies[i].cr_mask;
111 *p += len;
112 break;
113 }
114 }
115 if (i == ARRAY_SIZE(cache_policies))
116 printk(KERN_ERR "ERROR: unknown or unsupported cache policy\n");
117 flush_cache_all();
118 set_cr(cr_alignment);
119}
120__early_param("cachepolicy=", early_cachepolicy);
121
122static void __init early_nocache(char **__unused)
123{
124 char *p = "buffered";
125 printk(KERN_WARNING "nocache is deprecated; use cachepolicy=%s\n", p);
126 early_cachepolicy(&p);
127}
128__early_param("nocache", early_nocache);
129
130static void __init early_nowrite(char **__unused)
131{
132 char *p = "uncached";
133 printk(KERN_WARNING "nowb is deprecated; use cachepolicy=%s\n", p);
134 early_cachepolicy(&p);
135}
136__early_param("nowb", early_nowrite);
137
138static void __init early_ecc(char **p)
139{
140 if (memcmp(*p, "on", 2) == 0) {
141 ecc_mask = PMD_PROTECTION;
142 *p += 2;
143 } else if (memcmp(*p, "off", 3) == 0) {
144 ecc_mask = 0;
145 *p += 3;
146 }
147}
148__early_param("ecc=", early_ecc);
149
150static int __init noalign_setup(char *__unused)
151{
152 cr_alignment &= ~CR_A;
153 cr_no_alignment &= ~CR_A;
154 set_cr(cr_alignment);
155 return 1;
156}
157__setup("noalign", noalign_setup);
158
Russell King255d1f82006-12-18 00:12:47 +0000159#ifndef CONFIG_SMP
160void adjust_cr(unsigned long mask, unsigned long set)
161{
162 unsigned long flags;
163
164 mask &= ~CR_A;
165
166 set &= mask;
167
168 local_irq_save(flags);
169
170 cr_no_alignment = (cr_no_alignment & ~mask) | set;
171 cr_alignment = (cr_alignment & ~mask) | set;
172
173 set_cr((get_cr() & ~mask) | set);
174
175 local_irq_restore(flags);
176}
177#endif
178
Russell King2497f0a2007-04-21 09:59:44 +0100179struct mem_type {
Russell Kingae8f1542006-09-27 15:38:34 +0100180 unsigned int prot_pte;
181 unsigned int prot_l1;
182 unsigned int prot_sect;
183 unsigned int domain;
184};
185
Russell King2497f0a2007-04-21 09:59:44 +0100186static struct mem_type mem_types[] __initdata = {
Russell Kingae8f1542006-09-27 15:38:34 +0100187 [MT_DEVICE] = {
188 .prot_pte = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
189 L_PTE_WRITE,
190 .prot_l1 = PMD_TYPE_TABLE,
191 .prot_sect = PMD_TYPE_SECT | PMD_BIT4 | PMD_SECT_UNCACHED |
192 PMD_SECT_AP_WRITE,
193 .domain = DOMAIN_IO,
194 },
195 [MT_CACHECLEAN] = {
196 .prot_sect = PMD_TYPE_SECT | PMD_BIT4,
197 .domain = DOMAIN_KERNEL,
198 },
199 [MT_MINICLEAN] = {
200 .prot_sect = PMD_TYPE_SECT | PMD_BIT4 | PMD_SECT_MINICACHE,
201 .domain = DOMAIN_KERNEL,
202 },
203 [MT_LOW_VECTORS] = {
204 .prot_pte = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
205 L_PTE_EXEC,
206 .prot_l1 = PMD_TYPE_TABLE,
207 .domain = DOMAIN_USER,
208 },
209 [MT_HIGH_VECTORS] = {
210 .prot_pte = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
211 L_PTE_USER | L_PTE_EXEC,
212 .prot_l1 = PMD_TYPE_TABLE,
213 .domain = DOMAIN_USER,
214 },
215 [MT_MEMORY] = {
216 .prot_sect = PMD_TYPE_SECT | PMD_BIT4 | PMD_SECT_AP_WRITE,
217 .domain = DOMAIN_KERNEL,
218 },
219 [MT_ROM] = {
220 .prot_sect = PMD_TYPE_SECT | PMD_BIT4,
221 .domain = DOMAIN_KERNEL,
222 },
223 [MT_IXP2000_DEVICE] = { /* IXP2400 requires XCB=101 for on-chip I/O */
224 .prot_pte = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
225 L_PTE_WRITE,
226 .prot_l1 = PMD_TYPE_TABLE,
227 .prot_sect = PMD_TYPE_SECT | PMD_BIT4 | PMD_SECT_UNCACHED |
228 PMD_SECT_AP_WRITE | PMD_SECT_BUFFERABLE |
229 PMD_SECT_TEX(1),
230 .domain = DOMAIN_IO,
231 },
232 [MT_NONSHARED_DEVICE] = {
233 .prot_l1 = PMD_TYPE_TABLE,
234 .prot_sect = PMD_TYPE_SECT | PMD_BIT4 | PMD_SECT_NONSHARED_DEV |
235 PMD_SECT_AP_WRITE,
236 .domain = DOMAIN_IO,
237 }
238};
239
240/*
241 * Adjust the PMD section entries according to the CPU in use.
242 */
243static void __init build_mem_type_table(void)
244{
245 struct cachepolicy *cp;
246 unsigned int cr = get_cr();
247 unsigned int user_pgprot, kern_pgprot;
248 int cpu_arch = cpu_architecture();
249 int i;
250
251#if defined(CONFIG_CPU_DCACHE_DISABLE)
252 if (cachepolicy > CPOLICY_BUFFERED)
253 cachepolicy = CPOLICY_BUFFERED;
254#elif defined(CONFIG_CPU_DCACHE_WRITETHROUGH)
255 if (cachepolicy > CPOLICY_WRITETHROUGH)
256 cachepolicy = CPOLICY_WRITETHROUGH;
257#endif
258 if (cpu_arch < CPU_ARCH_ARMv5) {
259 if (cachepolicy >= CPOLICY_WRITEALLOC)
260 cachepolicy = CPOLICY_WRITEBACK;
261 ecc_mask = 0;
262 }
263
264 /*
265 * Xscale must not have PMD bit 4 set for section mappings.
266 */
267 if (cpu_is_xscale())
268 for (i = 0; i < ARRAY_SIZE(mem_types); i++)
269 mem_types[i].prot_sect &= ~PMD_BIT4;
270
271 /*
272 * ARMv5 and lower, excluding Xscale, bit 4 must be set for
273 * page tables.
274 */
275 if (cpu_arch < CPU_ARCH_ARMv6 && !cpu_is_xscale())
276 for (i = 0; i < ARRAY_SIZE(mem_types); i++)
277 if (mem_types[i].prot_l1)
278 mem_types[i].prot_l1 |= PMD_BIT4;
279
280 cp = &cache_policies[cachepolicy];
281 kern_pgprot = user_pgprot = cp->pte;
282
283 /*
284 * Enable CPU-specific coherency if supported.
285 * (Only available on XSC3 at the moment.)
286 */
287 if (arch_is_coherent()) {
288 if (cpu_is_xsc3()) {
289 mem_types[MT_MEMORY].prot_sect |= PMD_SECT_S;
Lennert Buytenhek0e5fdca72006-12-02 00:03:47 +0100290 mem_types[MT_MEMORY].prot_pte |= L_PTE_SHARED;
Russell Kingae8f1542006-09-27 15:38:34 +0100291 }
292 }
293
294 /*
295 * ARMv6 and above have extended page tables.
296 */
297 if (cpu_arch >= CPU_ARCH_ARMv6 && (cr & CR_XP)) {
298 /*
299 * bit 4 becomes XN which we must clear for the
300 * kernel memory mapping.
301 */
302 mem_types[MT_MEMORY].prot_sect &= ~PMD_SECT_XN;
303 mem_types[MT_ROM].prot_sect &= ~PMD_SECT_XN;
304
305 /*
306 * Mark cache clean areas and XIP ROM read only
307 * from SVC mode and no access from userspace.
308 */
309 mem_types[MT_ROM].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
310 mem_types[MT_MINICLEAN].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
311 mem_types[MT_CACHECLEAN].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
312
313 /*
314 * Mark the device area as "shared device"
315 */
316 mem_types[MT_DEVICE].prot_pte |= L_PTE_BUFFERABLE;
317 mem_types[MT_DEVICE].prot_sect |= PMD_SECT_BUFFERED;
318
Russell Kingae8f1542006-09-27 15:38:34 +0100319#ifdef CONFIG_SMP
320 /*
321 * Mark memory with the "shared" attribute for SMP systems
322 */
323 user_pgprot |= L_PTE_SHARED;
324 kern_pgprot |= L_PTE_SHARED;
325 mem_types[MT_MEMORY].prot_sect |= PMD_SECT_S;
326#endif
327 }
328
329 for (i = 0; i < 16; i++) {
330 unsigned long v = pgprot_val(protection_map[i]);
331 v = (v & ~(L_PTE_BUFFERABLE|L_PTE_CACHEABLE)) | user_pgprot;
332 protection_map[i] = __pgprot(v);
333 }
334
335 mem_types[MT_LOW_VECTORS].prot_pte |= kern_pgprot;
336 mem_types[MT_HIGH_VECTORS].prot_pte |= kern_pgprot;
337
338 if (cpu_arch >= CPU_ARCH_ARMv5) {
339#ifndef CONFIG_SMP
340 /*
341 * Only use write-through for non-SMP systems
342 */
343 mem_types[MT_LOW_VECTORS].prot_pte &= ~L_PTE_BUFFERABLE;
344 mem_types[MT_HIGH_VECTORS].prot_pte &= ~L_PTE_BUFFERABLE;
345#endif
346 } else {
347 mem_types[MT_MINICLEAN].prot_sect &= ~PMD_SECT_TEX(1);
348 }
349
Imre_Deak44b18692007-02-11 13:45:13 +0100350 pgprot_user = __pgprot(L_PTE_PRESENT | L_PTE_YOUNG | user_pgprot);
Russell Kingae8f1542006-09-27 15:38:34 +0100351 pgprot_kernel = __pgprot(L_PTE_PRESENT | L_PTE_YOUNG |
352 L_PTE_DIRTY | L_PTE_WRITE |
353 L_PTE_EXEC | kern_pgprot);
354
355 mem_types[MT_LOW_VECTORS].prot_l1 |= ecc_mask;
356 mem_types[MT_HIGH_VECTORS].prot_l1 |= ecc_mask;
357 mem_types[MT_MEMORY].prot_sect |= ecc_mask | cp->pmd;
358 mem_types[MT_ROM].prot_sect |= cp->pmd;
359
360 switch (cp->pmd) {
361 case PMD_SECT_WT:
362 mem_types[MT_CACHECLEAN].prot_sect |= PMD_SECT_WT;
363 break;
364 case PMD_SECT_WB:
365 case PMD_SECT_WBWA:
366 mem_types[MT_CACHECLEAN].prot_sect |= PMD_SECT_WB;
367 break;
368 }
369 printk("Memory policy: ECC %sabled, Data cache %s\n",
370 ecc_mask ? "en" : "dis", cp->policy);
Russell King2497f0a2007-04-21 09:59:44 +0100371
372 for (i = 0; i < ARRAY_SIZE(mem_types); i++) {
373 struct mem_type *t = &mem_types[i];
374 if (t->prot_l1)
375 t->prot_l1 |= PMD_DOMAIN(t->domain);
376 if (t->prot_sect)
377 t->prot_sect |= PMD_DOMAIN(t->domain);
378 }
Russell Kingae8f1542006-09-27 15:38:34 +0100379}
380
381#define vectors_base() (vectors_high() ? 0xffff0000 : 0)
382
Russell King24e6c692007-04-21 10:21:28 +0100383static void __init alloc_init_pte(pmd_t *pmd, unsigned long addr,
384 unsigned long end, unsigned long pfn,
385 const struct mem_type *type)
Russell Kingae8f1542006-09-27 15:38:34 +0100386{
Russell King24e6c692007-04-21 10:21:28 +0100387 pte_t *pte;
Russell Kingae8f1542006-09-27 15:38:34 +0100388
Russell King24e6c692007-04-21 10:21:28 +0100389 if (pmd_none(*pmd)) {
390 pte = alloc_bootmem_low_pages(2 * PTRS_PER_PTE * sizeof(pte_t));
391 __pmd_populate(pmd, __pa(pte) | type->prot_l1);
392 }
Russell Kingae8f1542006-09-27 15:38:34 +0100393
Russell King24e6c692007-04-21 10:21:28 +0100394 pte = pte_offset_kernel(pmd, addr);
395 do {
396 set_pte_ext(pte, pfn_pte(pfn, __pgprot(type->prot_pte)), 0);
397 pfn++;
398 } while (pte++, addr += PAGE_SIZE, addr != end);
Russell Kingae8f1542006-09-27 15:38:34 +0100399}
400
Russell King24e6c692007-04-21 10:21:28 +0100401static void __init alloc_init_section(pgd_t *pgd, unsigned long addr,
402 unsigned long end, unsigned long phys,
403 const struct mem_type *type)
Russell Kingae8f1542006-09-27 15:38:34 +0100404{
Russell King24e6c692007-04-21 10:21:28 +0100405 pmd_t *pmd = pmd_offset(pgd, addr);
Russell Kingae8f1542006-09-27 15:38:34 +0100406
Russell King24e6c692007-04-21 10:21:28 +0100407 /*
408 * Try a section mapping - end, addr and phys must all be aligned
409 * to a section boundary. Note that PMDs refer to the individual
410 * L1 entries, whereas PGDs refer to a group of L1 entries making
411 * up one logical pointer to an L2 table.
412 */
413 if (((addr | end | phys) & ~SECTION_MASK) == 0) {
414 pmd_t *p = pmd;
Russell Kingae8f1542006-09-27 15:38:34 +0100415
Russell King24e6c692007-04-21 10:21:28 +0100416 if (addr & SECTION_SIZE)
417 pmd++;
418
419 do {
420 *pmd = __pmd(phys | type->prot_sect);
421 phys += SECTION_SIZE;
422 } while (pmd++, addr += SECTION_SIZE, addr != end);
423
424 flush_pmd_entry(p);
425 } else {
426 /*
427 * No need to loop; pte's aren't interested in the
428 * individual L1 entries.
429 */
430 alloc_init_pte(pmd, addr, end, __phys_to_pfn(phys), type);
Russell Kingae8f1542006-09-27 15:38:34 +0100431 }
Russell Kingae8f1542006-09-27 15:38:34 +0100432}
433
Russell King4a56c1e2007-04-21 10:16:48 +0100434static void __init create_36bit_mapping(struct map_desc *md,
435 const struct mem_type *type)
436{
437 unsigned long phys, addr, length, end;
438 pgd_t *pgd;
439
440 addr = md->virtual;
441 phys = (unsigned long)__pfn_to_phys(md->pfn);
442 length = PAGE_ALIGN(md->length);
443
444 if (!(cpu_architecture() >= CPU_ARCH_ARMv6 || cpu_is_xsc3())) {
445 printk(KERN_ERR "MM: CPU does not support supersection "
446 "mapping for 0x%08llx at 0x%08lx\n",
447 __pfn_to_phys((u64)md->pfn), addr);
448 return;
449 }
450
451 /* N.B. ARMv6 supersections are only defined to work with domain 0.
452 * Since domain assignments can in fact be arbitrary, the
453 * 'domain == 0' check below is required to insure that ARMv6
454 * supersections are only allocated for domain 0 regardless
455 * of the actual domain assignments in use.
456 */
457 if (type->domain) {
458 printk(KERN_ERR "MM: invalid domain in supersection "
459 "mapping for 0x%08llx at 0x%08lx\n",
460 __pfn_to_phys((u64)md->pfn), addr);
461 return;
462 }
463
464 if ((addr | length | __pfn_to_phys(md->pfn)) & ~SUPERSECTION_MASK) {
465 printk(KERN_ERR "MM: cannot create mapping for "
466 "0x%08llx at 0x%08lx invalid alignment\n",
467 __pfn_to_phys((u64)md->pfn), addr);
468 return;
469 }
470
471 /*
472 * Shift bits [35:32] of address into bits [23:20] of PMD
473 * (See ARMv6 spec).
474 */
475 phys |= (((md->pfn >> (32 - PAGE_SHIFT)) & 0xF) << 20);
476
477 pgd = pgd_offset_k(addr);
478 end = addr + length;
479 do {
480 pmd_t *pmd = pmd_offset(pgd, addr);
481 int i;
482
483 for (i = 0; i < 16; i++)
484 *pmd++ = __pmd(phys | type->prot_sect | PMD_SECT_SUPER);
485
486 addr += SUPERSECTION_SIZE;
487 phys += SUPERSECTION_SIZE;
488 pgd += SUPERSECTION_SIZE >> PGDIR_SHIFT;
489 } while (addr != end);
490}
491
Russell Kingae8f1542006-09-27 15:38:34 +0100492/*
493 * Create the page directory entries and any necessary
494 * page tables for the mapping specified by `md'. We
495 * are able to cope here with varying sizes and address
496 * offsets, and we take full advantage of sections and
497 * supersections.
498 */
499void __init create_mapping(struct map_desc *md)
500{
Russell King24e6c692007-04-21 10:21:28 +0100501 unsigned long phys, addr, length, end;
Russell Kingd5c98172007-04-21 10:05:32 +0100502 const struct mem_type *type;
Russell King24e6c692007-04-21 10:21:28 +0100503 pgd_t *pgd;
Russell Kingae8f1542006-09-27 15:38:34 +0100504
505 if (md->virtual != vectors_base() && md->virtual < TASK_SIZE) {
506 printk(KERN_WARNING "BUG: not creating mapping for "
507 "0x%08llx at 0x%08lx in user region\n",
508 __pfn_to_phys((u64)md->pfn), md->virtual);
509 return;
510 }
511
512 if ((md->type == MT_DEVICE || md->type == MT_ROM) &&
513 md->virtual >= PAGE_OFFSET && md->virtual < VMALLOC_END) {
514 printk(KERN_WARNING "BUG: mapping for 0x%08llx at 0x%08lx "
515 "overlaps vmalloc space\n",
516 __pfn_to_phys((u64)md->pfn), md->virtual);
517 }
518
Russell Kingd5c98172007-04-21 10:05:32 +0100519 type = &mem_types[md->type];
Russell Kingae8f1542006-09-27 15:38:34 +0100520
521 /*
522 * Catch 36-bit addresses
523 */
Russell King4a56c1e2007-04-21 10:16:48 +0100524 if (md->pfn >= 0x100000) {
525 create_36bit_mapping(md, type);
526 return;
Russell Kingae8f1542006-09-27 15:38:34 +0100527 }
528
Russell King24e6c692007-04-21 10:21:28 +0100529 addr = md->virtual;
530 phys = (unsigned long)__pfn_to_phys(md->pfn);
531 length = PAGE_ALIGN(md->length);
Russell Kingae8f1542006-09-27 15:38:34 +0100532
Russell King24e6c692007-04-21 10:21:28 +0100533 if (type->prot_l1 == 0 && ((addr | phys | length) & ~SECTION_MASK)) {
Russell Kingae8f1542006-09-27 15:38:34 +0100534 printk(KERN_WARNING "BUG: map for 0x%08lx at 0x%08lx can not "
535 "be mapped using pages, ignoring.\n",
Russell King24e6c692007-04-21 10:21:28 +0100536 __pfn_to_phys(md->pfn), addr);
Russell Kingae8f1542006-09-27 15:38:34 +0100537 return;
538 }
539
Russell King24e6c692007-04-21 10:21:28 +0100540 pgd = pgd_offset_k(addr);
541 end = addr + length;
542 do {
543 unsigned long next = pgd_addr_end(addr, end);
Russell Kingae8f1542006-09-27 15:38:34 +0100544
Russell King24e6c692007-04-21 10:21:28 +0100545 alloc_init_section(pgd, addr, next, phys, type);
Russell Kingae8f1542006-09-27 15:38:34 +0100546
Russell King24e6c692007-04-21 10:21:28 +0100547 phys += next - addr;
548 addr = next;
549 } while (pgd++, addr != end);
Russell Kingae8f1542006-09-27 15:38:34 +0100550}
551
552/*
553 * Create the architecture specific mappings
554 */
555void __init iotable_init(struct map_desc *io_desc, int nr)
556{
557 int i;
558
559 for (i = 0; i < nr; i++)
560 create_mapping(io_desc + i);
561}
562
Russell Kingd111e8f2006-09-27 15:27:33 +0100563static inline void prepare_page_table(struct meminfo *mi)
564{
565 unsigned long addr;
566
567 /*
568 * Clear out all the mappings below the kernel image.
569 */
570 for (addr = 0; addr < MODULE_START; addr += PGDIR_SIZE)
571 pmd_clear(pmd_off_k(addr));
572
573#ifdef CONFIG_XIP_KERNEL
574 /* The XIP kernel is mapped in the module area -- skip over it */
575 addr = ((unsigned long)&_etext + PGDIR_SIZE - 1) & PGDIR_MASK;
576#endif
577 for ( ; addr < PAGE_OFFSET; addr += PGDIR_SIZE)
578 pmd_clear(pmd_off_k(addr));
579
580 /*
581 * Clear out all the kernel space mappings, except for the first
582 * memory bank, up to the end of the vmalloc region.
583 */
584 for (addr = __phys_to_virt(mi->bank[0].start + mi->bank[0].size);
585 addr < VMALLOC_END; addr += PGDIR_SIZE)
586 pmd_clear(pmd_off_k(addr));
587}
588
589/*
590 * Reserve the various regions of node 0
591 */
592void __init reserve_node_zero(pg_data_t *pgdat)
593{
594 unsigned long res_size = 0;
595
596 /*
597 * Register the kernel text and data with bootmem.
598 * Note that this can only be in node 0.
599 */
600#ifdef CONFIG_XIP_KERNEL
601 reserve_bootmem_node(pgdat, __pa(&__data_start), &_end - &__data_start);
602#else
603 reserve_bootmem_node(pgdat, __pa(&_stext), &_end - &_stext);
604#endif
605
606 /*
607 * Reserve the page tables. These are already in use,
608 * and can only be in node 0.
609 */
610 reserve_bootmem_node(pgdat, __pa(swapper_pg_dir),
611 PTRS_PER_PGD * sizeof(pgd_t));
612
613 /*
614 * Hmm... This should go elsewhere, but we really really need to
615 * stop things allocating the low memory; ideally we need a better
616 * implementation of GFP_DMA which does not assume that DMA-able
617 * memory starts at zero.
618 */
619 if (machine_is_integrator() || machine_is_cintegrator())
620 res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
621
622 /*
623 * These should likewise go elsewhere. They pre-reserve the
624 * screen memory region at the start of main system memory.
625 */
626 if (machine_is_edb7211())
627 res_size = 0x00020000;
628 if (machine_is_p720t())
629 res_size = 0x00014000;
630
Ben Dooksbbf6f282006-12-07 20:47:58 +0100631 /* H1940 and RX3715 need to reserve this for suspend */
632
633 if (machine_is_h1940() || machine_is_rx3715()) {
Ben Dooks90733412006-12-06 01:50:24 +0100634 reserve_bootmem_node(pgdat, 0x30003000, 0x1000);
635 reserve_bootmem_node(pgdat, 0x30081000, 0x1000);
636 }
637
Russell Kingd111e8f2006-09-27 15:27:33 +0100638#ifdef CONFIG_SA1111
639 /*
640 * Because of the SA1111 DMA bug, we want to preserve our
641 * precious DMA-able memory...
642 */
643 res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
644#endif
645 if (res_size)
646 reserve_bootmem_node(pgdat, PHYS_OFFSET, res_size);
647}
648
649/*
650 * Set up device the mappings. Since we clear out the page tables for all
651 * mappings above VMALLOC_END, we will remove any debug device mappings.
652 * This means you have to be careful how you debug this function, or any
653 * called function. This means you can't use any function or debugging
654 * method which may touch any device, otherwise the kernel _will_ crash.
655 */
656static void __init devicemaps_init(struct machine_desc *mdesc)
657{
658 struct map_desc map;
659 unsigned long addr;
660 void *vectors;
661
662 /*
663 * Allocate the vector page early.
664 */
665 vectors = alloc_bootmem_low_pages(PAGE_SIZE);
666 BUG_ON(!vectors);
667
668 for (addr = VMALLOC_END; addr; addr += PGDIR_SIZE)
669 pmd_clear(pmd_off_k(addr));
670
671 /*
672 * Map the kernel if it is XIP.
673 * It is always first in the modulearea.
674 */
675#ifdef CONFIG_XIP_KERNEL
676 map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
677 map.virtual = MODULE_START;
678 map.length = ((unsigned long)&_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
679 map.type = MT_ROM;
680 create_mapping(&map);
681#endif
682
683 /*
684 * Map the cache flushing regions.
685 */
686#ifdef FLUSH_BASE
687 map.pfn = __phys_to_pfn(FLUSH_BASE_PHYS);
688 map.virtual = FLUSH_BASE;
689 map.length = SZ_1M;
690 map.type = MT_CACHECLEAN;
691 create_mapping(&map);
692#endif
693#ifdef FLUSH_BASE_MINICACHE
694 map.pfn = __phys_to_pfn(FLUSH_BASE_PHYS + SZ_1M);
695 map.virtual = FLUSH_BASE_MINICACHE;
696 map.length = SZ_1M;
697 map.type = MT_MINICLEAN;
698 create_mapping(&map);
699#endif
700
701 /*
702 * Create a mapping for the machine vectors at the high-vectors
703 * location (0xffff0000). If we aren't using high-vectors, also
704 * create a mapping at the low-vectors virtual address.
705 */
706 map.pfn = __phys_to_pfn(virt_to_phys(vectors));
707 map.virtual = 0xffff0000;
708 map.length = PAGE_SIZE;
709 map.type = MT_HIGH_VECTORS;
710 create_mapping(&map);
711
712 if (!vectors_high()) {
713 map.virtual = 0;
714 map.type = MT_LOW_VECTORS;
715 create_mapping(&map);
716 }
717
718 /*
719 * Ask the machine support to map in the statically mapped devices.
720 */
721 if (mdesc->map_io)
722 mdesc->map_io();
723
724 /*
725 * Finally flush the caches and tlb to ensure that we're in a
726 * consistent state wrt the writebuffer. This also ensures that
727 * any write-allocated cache lines in the vector page are written
728 * back. After this point, we can start to touch devices again.
729 */
730 local_flush_tlb_all();
731 flush_cache_all();
732}
733
734/*
735 * paging_init() sets up the page tables, initialises the zone memory
736 * maps, and sets up the zero page, bad page and bad page tables.
737 */
738void __init paging_init(struct meminfo *mi, struct machine_desc *mdesc)
739{
740 void *zero_page;
741
742 build_mem_type_table();
743 prepare_page_table(mi);
744 bootmem_init(mi);
745 devicemaps_init(mdesc);
746
747 top_pmd = pmd_off_k(0xffff0000);
748
749 /*
750 * allocate the zero page. Note that we count on this going ok.
751 */
752 zero_page = alloc_bootmem_low_pages(PAGE_SIZE);
753 memzero(zero_page, PAGE_SIZE);
754 empty_zero_page = virt_to_page(zero_page);
755 flush_dcache_page(empty_zero_page);
756}
Russell Kingae8f1542006-09-27 15:38:34 +0100757
758/*
759 * In order to soft-boot, we need to insert a 1:1 mapping in place of
760 * the user-mode pages. This will then ensure that we have predictable
761 * results when turning the mmu off
762 */
763void setup_mm_for_reboot(char mode)
764{
765 unsigned long base_pmdval;
766 pgd_t *pgd;
767 int i;
768
769 if (current->mm && current->mm->pgd)
770 pgd = current->mm->pgd;
771 else
772 pgd = init_mm.pgd;
773
774 base_pmdval = PMD_SECT_AP_WRITE | PMD_SECT_AP_READ | PMD_TYPE_SECT;
775 if (cpu_architecture() <= CPU_ARCH_ARMv5TEJ && !cpu_is_xscale())
776 base_pmdval |= PMD_BIT4;
777
778 for (i = 0; i < FIRST_USER_PGD_NR + USER_PTRS_PER_PGD; i++, pgd++) {
779 unsigned long pmdval = (i << PGDIR_SHIFT) | base_pmdval;
780 pmd_t *pmd;
781
782 pmd = pmd_off(pgd, i << PGDIR_SHIFT);
783 pmd[0] = __pmd(pmdval);
784 pmd[1] = __pmd(pmdval + (1 << (PGDIR_SHIFT - 1)));
785 flush_pmd_entry(pmd);
786 }
787}