xref: /openbmc/linux/arch/arm/mm/mmu.c (revision 384740dc)
1 /*
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  */
10 #include <linux/module.h>
11 #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 
28 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
29 
30 extern void _stext, _etext, __data_start, _end;
31 extern 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  */
37 struct page *empty_zero_page;
38 EXPORT_SYMBOL(empty_zero_page);
39 
40 /*
41  * The pmd table for the upper-most set of pages.
42  */
43 pmd_t *top_pmd;
44 
45 #define CPOLICY_UNCACHED	0
46 #define CPOLICY_BUFFERED	1
47 #define CPOLICY_WRITETHROUGH	2
48 #define CPOLICY_WRITEBACK	3
49 #define CPOLICY_WRITEALLOC	4
50 
51 static unsigned int cachepolicy __initdata = CPOLICY_WRITEBACK;
52 static unsigned int ecc_mask __initdata = 0;
53 pgprot_t pgprot_user;
54 pgprot_t pgprot_kernel;
55 
56 EXPORT_SYMBOL(pgprot_user);
57 EXPORT_SYMBOL(pgprot_kernel);
58 
59 struct cachepolicy {
60 	const char	policy[16];
61 	unsigned int	cr_mask;
62 	unsigned int	pmd;
63 	unsigned int	pte;
64 };
65 
66 static struct cachepolicy cache_policies[] __initdata = {
67 	{
68 		.policy		= "uncached",
69 		.cr_mask	= CR_W|CR_C,
70 		.pmd		= PMD_SECT_UNCACHED,
71 		.pte		= 0,
72 	}, {
73 		.policy		= "buffered",
74 		.cr_mask	= CR_C,
75 		.pmd		= PMD_SECT_BUFFERED,
76 		.pte		= PTE_BUFFERABLE,
77 	}, {
78 		.policy		= "writethrough",
79 		.cr_mask	= 0,
80 		.pmd		= PMD_SECT_WT,
81 		.pte		= PTE_CACHEABLE,
82 	}, {
83 		.policy		= "writeback",
84 		.cr_mask	= 0,
85 		.pmd		= PMD_SECT_WB,
86 		.pte		= PTE_BUFFERABLE|PTE_CACHEABLE,
87 	}, {
88 		.policy		= "writealloc",
89 		.cr_mask	= 0,
90 		.pmd		= PMD_SECT_WBWA,
91 		.pte		= PTE_BUFFERABLE|PTE_CACHEABLE,
92 	}
93 };
94 
95 /*
96  * These are useful for identifying cache coherency
97  * problems by allowing the cache or the cache and
98  * writebuffer to be turned off.  (Note: the write
99  * buffer should not be on and the cache off).
100  */
101 static void __init early_cachepolicy(char **p)
102 {
103 	int i;
104 
105 	for (i = 0; i < ARRAY_SIZE(cache_policies); i++) {
106 		int len = strlen(cache_policies[i].policy);
107 
108 		if (memcmp(*p, cache_policies[i].policy, len) == 0) {
109 			cachepolicy = i;
110 			cr_alignment &= ~cache_policies[i].cr_mask;
111 			cr_no_alignment &= ~cache_policies[i].cr_mask;
112 			*p += len;
113 			break;
114 		}
115 	}
116 	if (i == ARRAY_SIZE(cache_policies))
117 		printk(KERN_ERR "ERROR: unknown or unsupported cache policy\n");
118 	if (cpu_architecture() >= CPU_ARCH_ARMv6) {
119 		printk(KERN_WARNING "Only cachepolicy=writeback supported on ARMv6 and later\n");
120 		cachepolicy = CPOLICY_WRITEBACK;
121 	}
122 	flush_cache_all();
123 	set_cr(cr_alignment);
124 }
125 __early_param("cachepolicy=", early_cachepolicy);
126 
127 static void __init early_nocache(char **__unused)
128 {
129 	char *p = "buffered";
130 	printk(KERN_WARNING "nocache is deprecated; use cachepolicy=%s\n", p);
131 	early_cachepolicy(&p);
132 }
133 __early_param("nocache", early_nocache);
134 
135 static void __init early_nowrite(char **__unused)
136 {
137 	char *p = "uncached";
138 	printk(KERN_WARNING "nowb is deprecated; use cachepolicy=%s\n", p);
139 	early_cachepolicy(&p);
140 }
141 __early_param("nowb", early_nowrite);
142 
143 static void __init early_ecc(char **p)
144 {
145 	if (memcmp(*p, "on", 2) == 0) {
146 		ecc_mask = PMD_PROTECTION;
147 		*p += 2;
148 	} else if (memcmp(*p, "off", 3) == 0) {
149 		ecc_mask = 0;
150 		*p += 3;
151 	}
152 }
153 __early_param("ecc=", early_ecc);
154 
155 static int __init noalign_setup(char *__unused)
156 {
157 	cr_alignment &= ~CR_A;
158 	cr_no_alignment &= ~CR_A;
159 	set_cr(cr_alignment);
160 	return 1;
161 }
162 __setup("noalign", noalign_setup);
163 
164 #ifndef CONFIG_SMP
165 void adjust_cr(unsigned long mask, unsigned long set)
166 {
167 	unsigned long flags;
168 
169 	mask &= ~CR_A;
170 
171 	set &= mask;
172 
173 	local_irq_save(flags);
174 
175 	cr_no_alignment = (cr_no_alignment & ~mask) | set;
176 	cr_alignment = (cr_alignment & ~mask) | set;
177 
178 	set_cr((get_cr() & ~mask) | set);
179 
180 	local_irq_restore(flags);
181 }
182 #endif
183 
184 #define PROT_PTE_DEVICE		L_PTE_PRESENT|L_PTE_YOUNG|L_PTE_DIRTY|L_PTE_WRITE
185 #define PROT_SECT_DEVICE	PMD_TYPE_SECT|PMD_SECT_XN|PMD_SECT_AP_WRITE
186 
187 static struct mem_type mem_types[] = {
188 	[MT_DEVICE] = {		  /* Strongly ordered / ARMv6 shared device */
189 		.prot_pte	= PROT_PTE_DEVICE,
190 		.prot_l1	= PMD_TYPE_TABLE,
191 		.prot_sect	= PROT_SECT_DEVICE | PMD_SECT_UNCACHED,
192 		.domain		= DOMAIN_IO,
193 	},
194 	[MT_DEVICE_NONSHARED] = { /* ARMv6 non-shared device */
195 		.prot_pte	= PROT_PTE_DEVICE,
196 		.prot_pte_ext	= PTE_EXT_TEX(2),
197 		.prot_l1	= PMD_TYPE_TABLE,
198 		.prot_sect	= PROT_SECT_DEVICE | PMD_SECT_TEX(2),
199 		.domain		= DOMAIN_IO,
200 	},
201 	[MT_DEVICE_CACHED] = {	  /* ioremap_cached */
202 		.prot_pte	= PROT_PTE_DEVICE | L_PTE_CACHEABLE | L_PTE_BUFFERABLE,
203 		.prot_l1	= PMD_TYPE_TABLE,
204 		.prot_sect	= PROT_SECT_DEVICE | PMD_SECT_WB,
205 		.domain		= DOMAIN_IO,
206 	},
207 	[MT_DEVICE_IXP2000] = {	  /* IXP2400 requires XCB=101 for on-chip I/O */
208 		.prot_pte	= PROT_PTE_DEVICE,
209 		.prot_l1	= PMD_TYPE_TABLE,
210 		.prot_sect	= PROT_SECT_DEVICE | PMD_SECT_BUFFERABLE |
211 				  PMD_SECT_TEX(1),
212 		.domain		= DOMAIN_IO,
213 	},
214 	[MT_DEVICE_WC] = {	/* ioremap_wc */
215 		.prot_pte	= PROT_PTE_DEVICE,
216 		.prot_l1	= PMD_TYPE_TABLE,
217 		.prot_sect	= PROT_SECT_DEVICE,
218 		.domain		= DOMAIN_IO,
219 	},
220 	[MT_CACHECLEAN] = {
221 		.prot_sect = PMD_TYPE_SECT | PMD_SECT_XN,
222 		.domain    = DOMAIN_KERNEL,
223 	},
224 	[MT_MINICLEAN] = {
225 		.prot_sect = PMD_TYPE_SECT | PMD_SECT_XN | PMD_SECT_MINICACHE,
226 		.domain    = DOMAIN_KERNEL,
227 	},
228 	[MT_LOW_VECTORS] = {
229 		.prot_pte  = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
230 				L_PTE_EXEC,
231 		.prot_l1   = PMD_TYPE_TABLE,
232 		.domain    = DOMAIN_USER,
233 	},
234 	[MT_HIGH_VECTORS] = {
235 		.prot_pte  = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
236 				L_PTE_USER | L_PTE_EXEC,
237 		.prot_l1   = PMD_TYPE_TABLE,
238 		.domain    = DOMAIN_USER,
239 	},
240 	[MT_MEMORY] = {
241 		.prot_sect = PMD_TYPE_SECT | PMD_SECT_AP_WRITE,
242 		.domain    = DOMAIN_KERNEL,
243 	},
244 	[MT_ROM] = {
245 		.prot_sect = PMD_TYPE_SECT,
246 		.domain    = DOMAIN_KERNEL,
247 	},
248 };
249 
250 const struct mem_type *get_mem_type(unsigned int type)
251 {
252 	return type < ARRAY_SIZE(mem_types) ? &mem_types[type] : NULL;
253 }
254 
255 /*
256  * Adjust the PMD section entries according to the CPU in use.
257  */
258 static void __init build_mem_type_table(void)
259 {
260 	struct cachepolicy *cp;
261 	unsigned int cr = get_cr();
262 	unsigned int user_pgprot, kern_pgprot;
263 	int cpu_arch = cpu_architecture();
264 	int i;
265 
266 	if (cpu_arch < CPU_ARCH_ARMv6) {
267 #if defined(CONFIG_CPU_DCACHE_DISABLE)
268 		if (cachepolicy > CPOLICY_BUFFERED)
269 			cachepolicy = CPOLICY_BUFFERED;
270 #elif defined(CONFIG_CPU_DCACHE_WRITETHROUGH)
271 		if (cachepolicy > CPOLICY_WRITETHROUGH)
272 			cachepolicy = CPOLICY_WRITETHROUGH;
273 #endif
274 	}
275 	if (cpu_arch < CPU_ARCH_ARMv5) {
276 		if (cachepolicy >= CPOLICY_WRITEALLOC)
277 			cachepolicy = CPOLICY_WRITEBACK;
278 		ecc_mask = 0;
279 	}
280 
281 	/*
282 	 * On non-Xscale3 ARMv5-and-older systems, use CB=01
283 	 * (Uncached/Buffered) for ioremap_wc() mappings.  On XScale3
284 	 * and ARMv6+, use TEXCB=00100 mappings (Inner/Outer Uncacheable
285 	 * in xsc3 parlance, Uncached Normal in ARMv6 parlance).
286 	 */
287 	if (cpu_is_xsc3() || cpu_arch >= CPU_ARCH_ARMv6) {
288 		mem_types[MT_DEVICE_WC].prot_pte_ext |= PTE_EXT_TEX(1);
289 		mem_types[MT_DEVICE_WC].prot_sect |= PMD_SECT_TEX(1);
290 	} else {
291 		mem_types[MT_DEVICE_WC].prot_pte |= L_PTE_BUFFERABLE;
292 		mem_types[MT_DEVICE_WC].prot_sect |= PMD_SECT_BUFFERABLE;
293 	}
294 
295 	/*
296 	 * ARMv5 and lower, bit 4 must be set for page tables.
297 	 * (was: cache "update-able on write" bit on ARM610)
298 	 * However, Xscale cores require this bit to be cleared.
299 	 */
300 	if (cpu_is_xscale()) {
301 		for (i = 0; i < ARRAY_SIZE(mem_types); i++) {
302 			mem_types[i].prot_sect &= ~PMD_BIT4;
303 			mem_types[i].prot_l1 &= ~PMD_BIT4;
304 		}
305 	} else if (cpu_arch < CPU_ARCH_ARMv6) {
306 		for (i = 0; i < ARRAY_SIZE(mem_types); i++) {
307 			if (mem_types[i].prot_l1)
308 				mem_types[i].prot_l1 |= PMD_BIT4;
309 			if (mem_types[i].prot_sect)
310 				mem_types[i].prot_sect |= PMD_BIT4;
311 		}
312 	}
313 
314 	cp = &cache_policies[cachepolicy];
315 	kern_pgprot = user_pgprot = cp->pte;
316 
317 	/*
318 	 * Enable CPU-specific coherency if supported.
319 	 * (Only available on XSC3 at the moment.)
320 	 */
321 	if (arch_is_coherent()) {
322 		if (cpu_is_xsc3()) {
323 			mem_types[MT_MEMORY].prot_sect |= PMD_SECT_S;
324 			mem_types[MT_MEMORY].prot_pte |= L_PTE_SHARED;
325 		}
326 	}
327 
328 	/*
329 	 * ARMv6 and above have extended page tables.
330 	 */
331 	if (cpu_arch >= CPU_ARCH_ARMv6 && (cr & CR_XP)) {
332 		/*
333 		 * Mark cache clean areas and XIP ROM read only
334 		 * from SVC mode and no access from userspace.
335 		 */
336 		mem_types[MT_ROM].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
337 		mem_types[MT_MINICLEAN].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
338 		mem_types[MT_CACHECLEAN].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
339 
340 		/*
341 		 * Mark the device area as "shared device"
342 		 */
343 		mem_types[MT_DEVICE].prot_pte |= L_PTE_BUFFERABLE;
344 		mem_types[MT_DEVICE].prot_sect |= PMD_SECT_BUFFERED;
345 
346 #ifdef CONFIG_SMP
347 		/*
348 		 * Mark memory with the "shared" attribute for SMP systems
349 		 */
350 		user_pgprot |= L_PTE_SHARED;
351 		kern_pgprot |= L_PTE_SHARED;
352 		mem_types[MT_MEMORY].prot_sect |= PMD_SECT_S;
353 #endif
354 	}
355 
356 	for (i = 0; i < 16; i++) {
357 		unsigned long v = pgprot_val(protection_map[i]);
358 		v = (v & ~(L_PTE_BUFFERABLE|L_PTE_CACHEABLE)) | user_pgprot;
359 		protection_map[i] = __pgprot(v);
360 	}
361 
362 	mem_types[MT_LOW_VECTORS].prot_pte |= kern_pgprot;
363 	mem_types[MT_HIGH_VECTORS].prot_pte |= kern_pgprot;
364 
365 	if (cpu_arch >= CPU_ARCH_ARMv5) {
366 #ifndef CONFIG_SMP
367 		/*
368 		 * Only use write-through for non-SMP systems
369 		 */
370 		mem_types[MT_LOW_VECTORS].prot_pte &= ~L_PTE_BUFFERABLE;
371 		mem_types[MT_HIGH_VECTORS].prot_pte &= ~L_PTE_BUFFERABLE;
372 #endif
373 	} else {
374 		mem_types[MT_MINICLEAN].prot_sect &= ~PMD_SECT_TEX(1);
375 	}
376 
377 	pgprot_user   = __pgprot(L_PTE_PRESENT | L_PTE_YOUNG | user_pgprot);
378 	pgprot_kernel = __pgprot(L_PTE_PRESENT | L_PTE_YOUNG |
379 				 L_PTE_DIRTY | L_PTE_WRITE |
380 				 L_PTE_EXEC | kern_pgprot);
381 
382 	mem_types[MT_LOW_VECTORS].prot_l1 |= ecc_mask;
383 	mem_types[MT_HIGH_VECTORS].prot_l1 |= ecc_mask;
384 	mem_types[MT_MEMORY].prot_sect |= ecc_mask | cp->pmd;
385 	mem_types[MT_ROM].prot_sect |= cp->pmd;
386 
387 	switch (cp->pmd) {
388 	case PMD_SECT_WT:
389 		mem_types[MT_CACHECLEAN].prot_sect |= PMD_SECT_WT;
390 		break;
391 	case PMD_SECT_WB:
392 	case PMD_SECT_WBWA:
393 		mem_types[MT_CACHECLEAN].prot_sect |= PMD_SECT_WB;
394 		break;
395 	}
396 	printk("Memory policy: ECC %sabled, Data cache %s\n",
397 		ecc_mask ? "en" : "dis", cp->policy);
398 
399 	for (i = 0; i < ARRAY_SIZE(mem_types); i++) {
400 		struct mem_type *t = &mem_types[i];
401 		if (t->prot_l1)
402 			t->prot_l1 |= PMD_DOMAIN(t->domain);
403 		if (t->prot_sect)
404 			t->prot_sect |= PMD_DOMAIN(t->domain);
405 	}
406 }
407 
408 #define vectors_base()	(vectors_high() ? 0xffff0000 : 0)
409 
410 static void __init alloc_init_pte(pmd_t *pmd, unsigned long addr,
411 				  unsigned long end, unsigned long pfn,
412 				  const struct mem_type *type)
413 {
414 	pte_t *pte;
415 
416 	if (pmd_none(*pmd)) {
417 		pte = alloc_bootmem_low_pages(2 * PTRS_PER_PTE * sizeof(pte_t));
418 		__pmd_populate(pmd, __pa(pte) | type->prot_l1);
419 	}
420 
421 	pte = pte_offset_kernel(pmd, addr);
422 	do {
423 		set_pte_ext(pte, pfn_pte(pfn, __pgprot(type->prot_pte)),
424 			    type->prot_pte_ext);
425 		pfn++;
426 	} while (pte++, addr += PAGE_SIZE, addr != end);
427 }
428 
429 static void __init alloc_init_section(pgd_t *pgd, unsigned long addr,
430 				      unsigned long end, unsigned long phys,
431 				      const struct mem_type *type)
432 {
433 	pmd_t *pmd = pmd_offset(pgd, addr);
434 
435 	/*
436 	 * Try a section mapping - end, addr and phys must all be aligned
437 	 * to a section boundary.  Note that PMDs refer to the individual
438 	 * L1 entries, whereas PGDs refer to a group of L1 entries making
439 	 * up one logical pointer to an L2 table.
440 	 */
441 	if (((addr | end | phys) & ~SECTION_MASK) == 0) {
442 		pmd_t *p = pmd;
443 
444 		if (addr & SECTION_SIZE)
445 			pmd++;
446 
447 		do {
448 			*pmd = __pmd(phys | type->prot_sect);
449 			phys += SECTION_SIZE;
450 		} while (pmd++, addr += SECTION_SIZE, addr != end);
451 
452 		flush_pmd_entry(p);
453 	} else {
454 		/*
455 		 * No need to loop; pte's aren't interested in the
456 		 * individual L1 entries.
457 		 */
458 		alloc_init_pte(pmd, addr, end, __phys_to_pfn(phys), type);
459 	}
460 }
461 
462 static void __init create_36bit_mapping(struct map_desc *md,
463 					const struct mem_type *type)
464 {
465 	unsigned long phys, addr, length, end;
466 	pgd_t *pgd;
467 
468 	addr = md->virtual;
469 	phys = (unsigned long)__pfn_to_phys(md->pfn);
470 	length = PAGE_ALIGN(md->length);
471 
472 	if (!(cpu_architecture() >= CPU_ARCH_ARMv6 || cpu_is_xsc3())) {
473 		printk(KERN_ERR "MM: CPU does not support supersection "
474 		       "mapping for 0x%08llx at 0x%08lx\n",
475 		       __pfn_to_phys((u64)md->pfn), addr);
476 		return;
477 	}
478 
479 	/* N.B.	ARMv6 supersections are only defined to work with domain 0.
480 	 *	Since domain assignments can in fact be arbitrary, the
481 	 *	'domain == 0' check below is required to insure that ARMv6
482 	 *	supersections are only allocated for domain 0 regardless
483 	 *	of the actual domain assignments in use.
484 	 */
485 	if (type->domain) {
486 		printk(KERN_ERR "MM: invalid domain in supersection "
487 		       "mapping for 0x%08llx at 0x%08lx\n",
488 		       __pfn_to_phys((u64)md->pfn), addr);
489 		return;
490 	}
491 
492 	if ((addr | length | __pfn_to_phys(md->pfn)) & ~SUPERSECTION_MASK) {
493 		printk(KERN_ERR "MM: cannot create mapping for "
494 		       "0x%08llx at 0x%08lx invalid alignment\n",
495 		       __pfn_to_phys((u64)md->pfn), addr);
496 		return;
497 	}
498 
499 	/*
500 	 * Shift bits [35:32] of address into bits [23:20] of PMD
501 	 * (See ARMv6 spec).
502 	 */
503 	phys |= (((md->pfn >> (32 - PAGE_SHIFT)) & 0xF) << 20);
504 
505 	pgd = pgd_offset_k(addr);
506 	end = addr + length;
507 	do {
508 		pmd_t *pmd = pmd_offset(pgd, addr);
509 		int i;
510 
511 		for (i = 0; i < 16; i++)
512 			*pmd++ = __pmd(phys | type->prot_sect | PMD_SECT_SUPER);
513 
514 		addr += SUPERSECTION_SIZE;
515 		phys += SUPERSECTION_SIZE;
516 		pgd += SUPERSECTION_SIZE >> PGDIR_SHIFT;
517 	} while (addr != end);
518 }
519 
520 /*
521  * Create the page directory entries and any necessary
522  * page tables for the mapping specified by `md'.  We
523  * are able to cope here with varying sizes and address
524  * offsets, and we take full advantage of sections and
525  * supersections.
526  */
527 void __init create_mapping(struct map_desc *md)
528 {
529 	unsigned long phys, addr, length, end;
530 	const struct mem_type *type;
531 	pgd_t *pgd;
532 
533 	if (md->virtual != vectors_base() && md->virtual < TASK_SIZE) {
534 		printk(KERN_WARNING "BUG: not creating mapping for "
535 		       "0x%08llx at 0x%08lx in user region\n",
536 		       __pfn_to_phys((u64)md->pfn), md->virtual);
537 		return;
538 	}
539 
540 	if ((md->type == MT_DEVICE || md->type == MT_ROM) &&
541 	    md->virtual >= PAGE_OFFSET && md->virtual < VMALLOC_END) {
542 		printk(KERN_WARNING "BUG: mapping for 0x%08llx at 0x%08lx "
543 		       "overlaps vmalloc space\n",
544 		       __pfn_to_phys((u64)md->pfn), md->virtual);
545 	}
546 
547 	type = &mem_types[md->type];
548 
549 	/*
550 	 * Catch 36-bit addresses
551 	 */
552 	if (md->pfn >= 0x100000) {
553 		create_36bit_mapping(md, type);
554 		return;
555 	}
556 
557 	addr = md->virtual & PAGE_MASK;
558 	phys = (unsigned long)__pfn_to_phys(md->pfn);
559 	length = PAGE_ALIGN(md->length + (md->virtual & ~PAGE_MASK));
560 
561 	if (type->prot_l1 == 0 && ((addr | phys | length) & ~SECTION_MASK)) {
562 		printk(KERN_WARNING "BUG: map for 0x%08lx at 0x%08lx can not "
563 		       "be mapped using pages, ignoring.\n",
564 		       __pfn_to_phys(md->pfn), addr);
565 		return;
566 	}
567 
568 	pgd = pgd_offset_k(addr);
569 	end = addr + length;
570 	do {
571 		unsigned long next = pgd_addr_end(addr, end);
572 
573 		alloc_init_section(pgd, addr, next, phys, type);
574 
575 		phys += next - addr;
576 		addr = next;
577 	} while (pgd++, addr != end);
578 }
579 
580 /*
581  * Create the architecture specific mappings
582  */
583 void __init iotable_init(struct map_desc *io_desc, int nr)
584 {
585 	int i;
586 
587 	for (i = 0; i < nr; i++)
588 		create_mapping(io_desc + i);
589 }
590 
591 static int __init check_membank_valid(struct membank *mb)
592 {
593 	/*
594 	 * Check whether this memory region has non-zero size.
595 	 */
596 	if (mb->size == 0)
597 		return 0;
598 
599 	/*
600 	 * Check whether this memory region would entirely overlap
601 	 * the vmalloc area.
602 	 */
603 	if (phys_to_virt(mb->start) >= VMALLOC_MIN) {
604 		printk(KERN_NOTICE "Ignoring RAM at %.8lx-%.8lx "
605 			"(vmalloc region overlap).\n",
606 			mb->start, mb->start + mb->size - 1);
607 		return 0;
608 	}
609 
610 	/*
611 	 * Check whether this memory region would partially overlap
612 	 * the vmalloc area.
613 	 */
614 	if (phys_to_virt(mb->start + mb->size) < phys_to_virt(mb->start) ||
615 	    phys_to_virt(mb->start + mb->size) > VMALLOC_MIN) {
616 		unsigned long newsize = VMALLOC_MIN - phys_to_virt(mb->start);
617 
618 		printk(KERN_NOTICE "Truncating RAM at %.8lx-%.8lx "
619 			"to -%.8lx (vmalloc region overlap).\n",
620 			mb->start, mb->start + mb->size - 1,
621 			mb->start + newsize - 1);
622 		mb->size = newsize;
623 	}
624 
625 	return 1;
626 }
627 
628 static void __init sanity_check_meminfo(struct meminfo *mi)
629 {
630 	int i;
631 	int j;
632 
633 	for (i = 0, j = 0; i < mi->nr_banks; i++) {
634 		if (check_membank_valid(&mi->bank[i]))
635 			mi->bank[j++] = mi->bank[i];
636 	}
637 	mi->nr_banks = j;
638 }
639 
640 static inline void prepare_page_table(struct meminfo *mi)
641 {
642 	unsigned long addr;
643 
644 	/*
645 	 * Clear out all the mappings below the kernel image.
646 	 */
647 	for (addr = 0; addr < MODULE_START; addr += PGDIR_SIZE)
648 		pmd_clear(pmd_off_k(addr));
649 
650 #ifdef CONFIG_XIP_KERNEL
651 	/* The XIP kernel is mapped in the module area -- skip over it */
652 	addr = ((unsigned long)&_etext + PGDIR_SIZE - 1) & PGDIR_MASK;
653 #endif
654 	for ( ; addr < PAGE_OFFSET; addr += PGDIR_SIZE)
655 		pmd_clear(pmd_off_k(addr));
656 
657 	/*
658 	 * Clear out all the kernel space mappings, except for the first
659 	 * memory bank, up to the end of the vmalloc region.
660 	 */
661 	for (addr = __phys_to_virt(mi->bank[0].start + mi->bank[0].size);
662 	     addr < VMALLOC_END; addr += PGDIR_SIZE)
663 		pmd_clear(pmd_off_k(addr));
664 }
665 
666 /*
667  * Reserve the various regions of node 0
668  */
669 void __init reserve_node_zero(pg_data_t *pgdat)
670 {
671 	unsigned long res_size = 0;
672 
673 	/*
674 	 * Register the kernel text and data with bootmem.
675 	 * Note that this can only be in node 0.
676 	 */
677 #ifdef CONFIG_XIP_KERNEL
678 	reserve_bootmem_node(pgdat, __pa(&__data_start), &_end - &__data_start,
679 			BOOTMEM_DEFAULT);
680 #else
681 	reserve_bootmem_node(pgdat, __pa(&_stext), &_end - &_stext,
682 			BOOTMEM_DEFAULT);
683 #endif
684 
685 	/*
686 	 * Reserve the page tables.  These are already in use,
687 	 * and can only be in node 0.
688 	 */
689 	reserve_bootmem_node(pgdat, __pa(swapper_pg_dir),
690 			     PTRS_PER_PGD * sizeof(pgd_t), BOOTMEM_DEFAULT);
691 
692 	/*
693 	 * Hmm... This should go elsewhere, but we really really need to
694 	 * stop things allocating the low memory; ideally we need a better
695 	 * implementation of GFP_DMA which does not assume that DMA-able
696 	 * memory starts at zero.
697 	 */
698 	if (machine_is_integrator() || machine_is_cintegrator())
699 		res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
700 
701 	/*
702 	 * These should likewise go elsewhere.  They pre-reserve the
703 	 * screen memory region at the start of main system memory.
704 	 */
705 	if (machine_is_edb7211())
706 		res_size = 0x00020000;
707 	if (machine_is_p720t())
708 		res_size = 0x00014000;
709 
710 	/* H1940 and RX3715 need to reserve this for suspend */
711 
712 	if (machine_is_h1940() || machine_is_rx3715()) {
713 		reserve_bootmem_node(pgdat, 0x30003000, 0x1000,
714 				BOOTMEM_DEFAULT);
715 		reserve_bootmem_node(pgdat, 0x30081000, 0x1000,
716 				BOOTMEM_DEFAULT);
717 	}
718 
719 #ifdef CONFIG_SA1111
720 	/*
721 	 * Because of the SA1111 DMA bug, we want to preserve our
722 	 * precious DMA-able memory...
723 	 */
724 	res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
725 #endif
726 	if (res_size)
727 		reserve_bootmem_node(pgdat, PHYS_OFFSET, res_size,
728 				BOOTMEM_DEFAULT);
729 }
730 
731 /*
732  * Set up device the mappings.  Since we clear out the page tables for all
733  * mappings above VMALLOC_END, we will remove any debug device mappings.
734  * This means you have to be careful how you debug this function, or any
735  * called function.  This means you can't use any function or debugging
736  * method which may touch any device, otherwise the kernel _will_ crash.
737  */
738 static void __init devicemaps_init(struct machine_desc *mdesc)
739 {
740 	struct map_desc map;
741 	unsigned long addr;
742 	void *vectors;
743 
744 	/*
745 	 * Allocate the vector page early.
746 	 */
747 	vectors = alloc_bootmem_low_pages(PAGE_SIZE);
748 	BUG_ON(!vectors);
749 
750 	for (addr = VMALLOC_END; addr; addr += PGDIR_SIZE)
751 		pmd_clear(pmd_off_k(addr));
752 
753 	/*
754 	 * Map the kernel if it is XIP.
755 	 * It is always first in the modulearea.
756 	 */
757 #ifdef CONFIG_XIP_KERNEL
758 	map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
759 	map.virtual = MODULE_START;
760 	map.length = ((unsigned long)&_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
761 	map.type = MT_ROM;
762 	create_mapping(&map);
763 #endif
764 
765 	/*
766 	 * Map the cache flushing regions.
767 	 */
768 #ifdef FLUSH_BASE
769 	map.pfn = __phys_to_pfn(FLUSH_BASE_PHYS);
770 	map.virtual = FLUSH_BASE;
771 	map.length = SZ_1M;
772 	map.type = MT_CACHECLEAN;
773 	create_mapping(&map);
774 #endif
775 #ifdef FLUSH_BASE_MINICACHE
776 	map.pfn = __phys_to_pfn(FLUSH_BASE_PHYS + SZ_1M);
777 	map.virtual = FLUSH_BASE_MINICACHE;
778 	map.length = SZ_1M;
779 	map.type = MT_MINICLEAN;
780 	create_mapping(&map);
781 #endif
782 
783 	/*
784 	 * Create a mapping for the machine vectors at the high-vectors
785 	 * location (0xffff0000).  If we aren't using high-vectors, also
786 	 * create a mapping at the low-vectors virtual address.
787 	 */
788 	map.pfn = __phys_to_pfn(virt_to_phys(vectors));
789 	map.virtual = 0xffff0000;
790 	map.length = PAGE_SIZE;
791 	map.type = MT_HIGH_VECTORS;
792 	create_mapping(&map);
793 
794 	if (!vectors_high()) {
795 		map.virtual = 0;
796 		map.type = MT_LOW_VECTORS;
797 		create_mapping(&map);
798 	}
799 
800 	/*
801 	 * Ask the machine support to map in the statically mapped devices.
802 	 */
803 	if (mdesc->map_io)
804 		mdesc->map_io();
805 
806 	/*
807 	 * Finally flush the caches and tlb to ensure that we're in a
808 	 * consistent state wrt the writebuffer.  This also ensures that
809 	 * any write-allocated cache lines in the vector page are written
810 	 * back.  After this point, we can start to touch devices again.
811 	 */
812 	local_flush_tlb_all();
813 	flush_cache_all();
814 }
815 
816 /*
817  * paging_init() sets up the page tables, initialises the zone memory
818  * maps, and sets up the zero page, bad page and bad page tables.
819  */
820 void __init paging_init(struct meminfo *mi, struct machine_desc *mdesc)
821 {
822 	void *zero_page;
823 
824 	build_mem_type_table();
825 	sanity_check_meminfo(mi);
826 	prepare_page_table(mi);
827 	bootmem_init(mi);
828 	devicemaps_init(mdesc);
829 
830 	top_pmd = pmd_off_k(0xffff0000);
831 
832 	/*
833 	 * allocate the zero page.  Note that we count on this going ok.
834 	 */
835 	zero_page = alloc_bootmem_low_pages(PAGE_SIZE);
836 	memzero(zero_page, PAGE_SIZE);
837 	empty_zero_page = virt_to_page(zero_page);
838 	flush_dcache_page(empty_zero_page);
839 }
840 
841 /*
842  * In order to soft-boot, we need to insert a 1:1 mapping in place of
843  * the user-mode pages.  This will then ensure that we have predictable
844  * results when turning the mmu off
845  */
846 void setup_mm_for_reboot(char mode)
847 {
848 	unsigned long base_pmdval;
849 	pgd_t *pgd;
850 	int i;
851 
852 	if (current->mm && current->mm->pgd)
853 		pgd = current->mm->pgd;
854 	else
855 		pgd = init_mm.pgd;
856 
857 	base_pmdval = PMD_SECT_AP_WRITE | PMD_SECT_AP_READ | PMD_TYPE_SECT;
858 	if (cpu_architecture() <= CPU_ARCH_ARMv5TEJ && !cpu_is_xscale())
859 		base_pmdval |= PMD_BIT4;
860 
861 	for (i = 0; i < FIRST_USER_PGD_NR + USER_PTRS_PER_PGD; i++, pgd++) {
862 		unsigned long pmdval = (i << PGDIR_SHIFT) | base_pmdval;
863 		pmd_t *pmd;
864 
865 		pmd = pmd_off(pgd, i << PGDIR_SHIFT);
866 		pmd[0] = __pmd(pmdval);
867 		pmd[1] = __pmd(pmdval + (1 << (PGDIR_SHIFT - 1)));
868 		flush_pmd_entry(pmd);
869 	}
870 }
871