1/*
2 * linux/arch/unicore32/mm/mmu.c
3 *
4 * Code specific to PKUnity SoC and UniCore ISA
5 *
6 * Copyright (C) 2001-2010 GUAN Xue-tao
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 as
10 * published by the Free Software Foundation.
11 */
12#include <linux/module.h>
13#include <linux/kernel.h>
14#include <linux/errno.h>
15#include <linux/init.h>
16#include <linux/mman.h>
17#include <linux/nodemask.h>
18#include <linux/memblock.h>
19#include <linux/fs.h>
20#include <linux/bootmem.h>
21#include <linux/io.h>
22
23#include <asm/cputype.h>
24#include <asm/sections.h>
25#include <asm/setup.h>
26#include <asm/sizes.h>
27#include <asm/tlb.h>
28#include <asm/memblock.h>
29
30#include <mach/map.h>
31
32#include "mm.h"
33
34/*
35 * empty_zero_page is a special page that is used for
36 * zero-initialized data and COW.
37 */
38struct page *empty_zero_page;
39EXPORT_SYMBOL(empty_zero_page);
40
41/*
42 * The pmd table for the upper-most set of pages.
43 */
44pmd_t *top_pmd;
45
46pgprot_t pgprot_user;
47EXPORT_SYMBOL(pgprot_user);
48
49pgprot_t pgprot_kernel;
50EXPORT_SYMBOL(pgprot_kernel);
51
52static int __init noalign_setup(char *__unused)
53{
54	cr_alignment &= ~CR_A;
55	cr_no_alignment &= ~CR_A;
56	set_cr(cr_alignment);
57	return 1;
58}
59__setup("noalign", noalign_setup);
60
61void adjust_cr(unsigned long mask, unsigned long set)
62{
63	unsigned long flags;
64
65	mask &= ~CR_A;
66
67	set &= mask;
68
69	local_irq_save(flags);
70
71	cr_no_alignment = (cr_no_alignment & ~mask) | set;
72	cr_alignment = (cr_alignment & ~mask) | set;
73
74	set_cr((get_cr() & ~mask) | set);
75
76	local_irq_restore(flags);
77}
78
79struct map_desc {
80	unsigned long virtual;
81	unsigned long pfn;
82	unsigned long length;
83	unsigned int type;
84};
85
86#define PROT_PTE_DEVICE		(PTE_PRESENT | PTE_YOUNG |	\
87				PTE_DIRTY | PTE_READ | PTE_WRITE)
88#define PROT_SECT_DEVICE	(PMD_TYPE_SECT | PMD_PRESENT |	\
89				PMD_SECT_READ | PMD_SECT_WRITE)
90
91static struct mem_type mem_types[] = {
92	[MT_DEVICE] = {		  /* Strongly ordered */
93		.prot_pte	= PROT_PTE_DEVICE,
94		.prot_l1	= PMD_TYPE_TABLE | PMD_PRESENT,
95		.prot_sect	= PROT_SECT_DEVICE,
96	},
97	/*
98	 * MT_KUSER: pte for vecpage -- cacheable,
99	 *       and sect for unigfx mmap -- noncacheable
100	 */
101	[MT_KUSER] = {
102		.prot_pte  = PTE_PRESENT | PTE_YOUNG | PTE_DIRTY |
103				PTE_CACHEABLE | PTE_READ | PTE_EXEC,
104		.prot_l1   = PMD_TYPE_TABLE | PMD_PRESENT,
105		.prot_sect = PROT_SECT_DEVICE,
106	},
107	[MT_HIGH_VECTORS] = {
108		.prot_pte  = PTE_PRESENT | PTE_YOUNG | PTE_DIRTY |
109				PTE_CACHEABLE | PTE_READ | PTE_WRITE |
110				PTE_EXEC,
111		.prot_l1   = PMD_TYPE_TABLE | PMD_PRESENT,
112	},
113	[MT_MEMORY] = {
114		.prot_pte  = PTE_PRESENT | PTE_YOUNG | PTE_DIRTY |
115				PTE_WRITE | PTE_EXEC,
116		.prot_l1   = PMD_TYPE_TABLE | PMD_PRESENT,
117		.prot_sect = PMD_TYPE_SECT | PMD_PRESENT | PMD_SECT_CACHEABLE |
118				PMD_SECT_READ | PMD_SECT_WRITE | PMD_SECT_EXEC,
119	},
120	[MT_ROM] = {
121		.prot_sect = PMD_TYPE_SECT | PMD_PRESENT | PMD_SECT_CACHEABLE |
122				PMD_SECT_READ,
123	},
124};
125
126const struct mem_type *get_mem_type(unsigned int type)
127{
128	return type < ARRAY_SIZE(mem_types) ? &mem_types[type] : NULL;
129}
130EXPORT_SYMBOL(get_mem_type);
131
132/*
133 * Adjust the PMD section entries according to the CPU in use.
134 */
135static void __init build_mem_type_table(void)
136{
137	pgprot_user   = __pgprot(PTE_PRESENT | PTE_YOUNG | PTE_CACHEABLE);
138	pgprot_kernel = __pgprot(PTE_PRESENT | PTE_YOUNG |
139				 PTE_DIRTY | PTE_READ | PTE_WRITE |
140				 PTE_EXEC | PTE_CACHEABLE);
141}
142
143#define vectors_base()	(vectors_high() ? 0xffff0000 : 0)
144
145static void __init *early_alloc(unsigned long sz)
146{
147	void *ptr = __va(memblock_alloc(sz, sz));
148	memset(ptr, 0, sz);
149	return ptr;
150}
151
152static pte_t * __init early_pte_alloc(pmd_t *pmd, unsigned long addr,
153		unsigned long prot)
154{
155	if (pmd_none(*pmd)) {
156		pte_t *pte = early_alloc(PTRS_PER_PTE * sizeof(pte_t));
157		__pmd_populate(pmd, __pa(pte) | prot);
158	}
159	BUG_ON(pmd_bad(*pmd));
160	return pte_offset_kernel(pmd, addr);
161}
162
163static void __init alloc_init_pte(pmd_t *pmd, unsigned long addr,
164				  unsigned long end, unsigned long pfn,
165				  const struct mem_type *type)
166{
167	pte_t *pte = early_pte_alloc(pmd, addr, type->prot_l1);
168	do {
169		set_pte(pte, pfn_pte(pfn, __pgprot(type->prot_pte)));
170		pfn++;
171	} while (pte++, addr += PAGE_SIZE, addr != end);
172}
173
174static void __init alloc_init_section(pgd_t *pgd, unsigned long addr,
175				      unsigned long end, unsigned long phys,
176				      const struct mem_type *type)
177{
178	pmd_t *pmd = pmd_offset((pud_t *)pgd, addr);
179
180	/*
181	 * Try a section mapping - end, addr and phys must all be aligned
182	 * to a section boundary.
183	 */
184	if (((addr | end | phys) & ~SECTION_MASK) == 0) {
185		pmd_t *p = pmd;
186
187		do {
188			set_pmd(pmd, __pmd(phys | type->prot_sect));
189			phys += SECTION_SIZE;
190		} while (pmd++, addr += SECTION_SIZE, addr != end);
191
192		flush_pmd_entry(p);
193	} else {
194		/*
195		 * No need to loop; pte's aren't interested in the
196		 * individual L1 entries.
197		 */
198		alloc_init_pte(pmd, addr, end, __phys_to_pfn(phys), type);
199	}
200}
201
202/*
203 * Create the page directory entries and any necessary
204 * page tables for the mapping specified by `md'.  We
205 * are able to cope here with varying sizes and address
206 * offsets, and we take full advantage of sections.
207 */
208static void __init create_mapping(struct map_desc *md)
209{
210	unsigned long phys, addr, length, end;
211	const struct mem_type *type;
212	pgd_t *pgd;
213
214	if (md->virtual != vectors_base() && md->virtual < TASK_SIZE) {
215		printk(KERN_WARNING "BUG: not creating mapping for "
216		       "0x%08llx at 0x%08lx in user region\n",
217		       __pfn_to_phys((u64)md->pfn), md->virtual);
218		return;
219	}
220
221	if ((md->type == MT_DEVICE || md->type == MT_ROM) &&
222	    md->virtual >= PAGE_OFFSET && md->virtual < VMALLOC_END) {
223		printk(KERN_WARNING "BUG: mapping for 0x%08llx at 0x%08lx "
224		       "overlaps vmalloc space\n",
225		       __pfn_to_phys((u64)md->pfn), md->virtual);
226	}
227
228	type = &mem_types[md->type];
229
230	addr = md->virtual & PAGE_MASK;
231	phys = (unsigned long)__pfn_to_phys(md->pfn);
232	length = PAGE_ALIGN(md->length + (md->virtual & ~PAGE_MASK));
233
234	if (type->prot_l1 == 0 && ((addr | phys | length) & ~SECTION_MASK)) {
235		printk(KERN_WARNING "BUG: map for 0x%08lx at 0x%08lx can not "
236		       "be mapped using pages, ignoring.\n",
237		       __pfn_to_phys(md->pfn), addr);
238		return;
239	}
240
241	pgd = pgd_offset_k(addr);
242	end = addr + length;
243	do {
244		unsigned long next = pgd_addr_end(addr, end);
245
246		alloc_init_section(pgd, addr, next, phys, type);
247
248		phys += next - addr;
249		addr = next;
250	} while (pgd++, addr != end);
251}
252
253static void * __initdata vmalloc_min = (void *)(VMALLOC_END - SZ_128M);
254
255/*
256 * vmalloc=size forces the vmalloc area to be exactly 'size'
257 * bytes. This can be used to increase (or decrease) the vmalloc
258 * area - the default is 128m.
259 */
260static int __init early_vmalloc(char *arg)
261{
262	unsigned long vmalloc_reserve = memparse(arg, NULL);
263
264	if (vmalloc_reserve < SZ_16M) {
265		vmalloc_reserve = SZ_16M;
266		printk(KERN_WARNING
267			"vmalloc area too small, limiting to %luMB\n",
268			vmalloc_reserve >> 20);
269	}
270
271	if (vmalloc_reserve > VMALLOC_END - (PAGE_OFFSET + SZ_32M)) {
272		vmalloc_reserve = VMALLOC_END - (PAGE_OFFSET + SZ_32M);
273		printk(KERN_WARNING
274			"vmalloc area is too big, limiting to %luMB\n",
275			vmalloc_reserve >> 20);
276	}
277
278	vmalloc_min = (void *)(VMALLOC_END - vmalloc_reserve);
279	return 0;
280}
281early_param("vmalloc", early_vmalloc);
282
283static phys_addr_t lowmem_limit __initdata = SZ_1G;
284
285static void __init sanity_check_meminfo(void)
286{
287	int i, j;
288
289	lowmem_limit = __pa(vmalloc_min - 1) + 1;
290	memblock_set_current_limit(lowmem_limit);
291
292	for (i = 0, j = 0; i < meminfo.nr_banks; i++) {
293		struct membank *bank = &meminfo.bank[j];
294		*bank = meminfo.bank[i];
295		j++;
296	}
297	meminfo.nr_banks = j;
298}
299
300static inline void prepare_page_table(void)
301{
302	unsigned long addr;
303	phys_addr_t end;
304
305	/*
306	 * Clear out all the mappings below the kernel image.
307	 */
308	for (addr = 0; addr < MODULES_VADDR; addr += PGDIR_SIZE)
309		pmd_clear(pmd_off_k(addr));
310
311	for ( ; addr < PAGE_OFFSET; addr += PGDIR_SIZE)
312		pmd_clear(pmd_off_k(addr));
313
314	/*
315	 * Find the end of the first block of lowmem.
316	 */
317	end = memblock.memory.regions[0].base + memblock.memory.regions[0].size;
318	if (end >= lowmem_limit)
319		end = lowmem_limit;
320
321	/*
322	 * Clear out all the kernel space mappings, except for the first
323	 * memory bank, up to the end of the vmalloc region.
324	 */
325	for (addr = __phys_to_virt(end);
326	     addr < VMALLOC_END; addr += PGDIR_SIZE)
327		pmd_clear(pmd_off_k(addr));
328}
329
330/*
331 * Reserve the special regions of memory
332 */
333void __init uc32_mm_memblock_reserve(void)
334{
335	/*
336	 * Reserve the page tables.  These are already in use,
337	 * and can only be in node 0.
338	 */
339	memblock_reserve(__pa(swapper_pg_dir), PTRS_PER_PGD * sizeof(pgd_t));
340}
341
342/*
343 * Set up device the mappings.  Since we clear out the page tables for all
344 * mappings above VMALLOC_END, we will remove any debug device mappings.
345 * This means you have to be careful how you debug this function, or any
346 * called function.  This means you can't use any function or debugging
347 * method which may touch any device, otherwise the kernel _will_ crash.
348 */
349static void __init devicemaps_init(void)
350{
351	struct map_desc map;
352	unsigned long addr;
353	void *vectors;
354
355	/*
356	 * Allocate the vector page early.
357	 */
358	vectors = early_alloc(PAGE_SIZE);
359
360	for (addr = VMALLOC_END; addr; addr += PGDIR_SIZE)
361		pmd_clear(pmd_off_k(addr));
362
363	/*
364	 * Create a mapping for the machine vectors at the high-vectors
365	 * location (0xffff0000).  If we aren't using high-vectors, also
366	 * create a mapping at the low-vectors virtual address.
367	 */
368	map.pfn = __phys_to_pfn(virt_to_phys(vectors));
369	map.virtual = VECTORS_BASE;
370	map.length = PAGE_SIZE;
371	map.type = MT_HIGH_VECTORS;
372	create_mapping(&map);
373
374	/*
375	 * Create a mapping for the kuser page at the special
376	 * location (0xbfff0000) to the same vectors location.
377	 */
378	map.pfn = __phys_to_pfn(virt_to_phys(vectors));
379	map.virtual = KUSER_VECPAGE_BASE;
380	map.length = PAGE_SIZE;
381	map.type = MT_KUSER;
382	create_mapping(&map);
383
384	/*
385	 * Finally flush the caches and tlb to ensure that we're in a
386	 * consistent state wrt the writebuffer.  This also ensures that
387	 * any write-allocated cache lines in the vector page are written
388	 * back.  After this point, we can start to touch devices again.
389	 */
390	local_flush_tlb_all();
391	flush_cache_all();
392}
393
394static void __init map_lowmem(void)
395{
396	struct memblock_region *reg;
397
398	/* Map all the lowmem memory banks. */
399	for_each_memblock(memory, reg) {
400		phys_addr_t start = reg->base;
401		phys_addr_t end = start + reg->size;
402		struct map_desc map;
403
404		if (end > lowmem_limit)
405			end = lowmem_limit;
406		if (start >= end)
407			break;
408
409		map.pfn = __phys_to_pfn(start);
410		map.virtual = __phys_to_virt(start);
411		map.length = end - start;
412		map.type = MT_MEMORY;
413
414		create_mapping(&map);
415	}
416}
417
418/*
419 * paging_init() sets up the page tables, initialises the zone memory
420 * maps, and sets up the zero page, bad page and bad page tables.
421 */
422void __init paging_init(void)
423{
424	void *zero_page;
425
426	build_mem_type_table();
427	sanity_check_meminfo();
428	prepare_page_table();
429	map_lowmem();
430	devicemaps_init();
431
432	top_pmd = pmd_off_k(0xffff0000);
433
434	/* allocate the zero page. */
435	zero_page = early_alloc(PAGE_SIZE);
436
437	bootmem_init();
438
439	empty_zero_page = virt_to_page(zero_page);
440	__flush_dcache_page(NULL, empty_zero_page);
441}
442
443/*
444 * In order to soft-boot, we need to insert a 1:1 mapping in place of
445 * the user-mode pages.  This will then ensure that we have predictable
446 * results when turning the mmu off
447 */
448void setup_mm_for_reboot(void)
449{
450	unsigned long base_pmdval;
451	pgd_t *pgd;
452	int i;
453
454	/*
455	 * We need to access to user-mode page tables here. For kernel threads
456	 * we don't have any user-mode mappings so we use the context that we
457	 * "borrowed".
458	 */
459	pgd = current->active_mm->pgd;
460
461	base_pmdval = PMD_SECT_WRITE | PMD_SECT_READ | PMD_TYPE_SECT;
462
463	for (i = 0; i < FIRST_USER_PGD_NR + USER_PTRS_PER_PGD; i++, pgd++) {
464		unsigned long pmdval = (i << PGDIR_SHIFT) | base_pmdval;
465		pmd_t *pmd;
466
467		pmd = pmd_off(pgd, i << PGDIR_SHIFT);
468		set_pmd(pmd, __pmd(pmdval));
469		flush_pmd_entry(pmd);
470	}
471
472	local_flush_tlb_all();
473}
474
475/*
476 * Take care of architecture specific things when placing a new PTE into
477 * a page table, or changing an existing PTE.  Basically, there are two
478 * things that we need to take care of:
479 *
480 *  1. If PG_dcache_clean is not set for the page, we need to ensure
481 *     that any cache entries for the kernels virtual memory
482 *     range are written back to the page.
483 *  2. If we have multiple shared mappings of the same space in
484 *     an object, we need to deal with the cache aliasing issues.
485 *
486 * Note that the pte lock will be held.
487 */
488void update_mmu_cache(struct vm_area_struct *vma, unsigned long addr,
489	pte_t *ptep)
490{
491	unsigned long pfn = pte_pfn(*ptep);
492	struct address_space *mapping;
493	struct page *page;
494
495	if (!pfn_valid(pfn))
496		return;
497
498	/*
499	 * The zero page is never written to, so never has any dirty
500	 * cache lines, and therefore never needs to be flushed.
501	 */
502	page = pfn_to_page(pfn);
503	if (page == ZERO_PAGE(0))
504		return;
505
506	mapping = page_mapping(page);
507	if (!test_and_set_bit(PG_dcache_clean, &page->flags))
508		__flush_dcache_page(mapping, page);
509	if (mapping)
510		if (vma->vm_flags & VM_EXEC)
511			__flush_icache_all();
512}
513