2 * linux/arch/unicore32/mm/mmu.c
4 * Code specific to PKUnity SoC and UniCore ISA
6 * Copyright (C) 2001-2010 GUAN Xue-tao
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.
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>
22 #include <asm/cputype.h>
23 #include <asm/sections.h>
24 #include <asm/setup.h>
25 #include <asm/sizes.h>
27 #include <asm/memblock.h>
34 * empty_zero_page is a special page that is used for
35 * zero-initialized data and COW.
37 struct page *empty_zero_page;
38 EXPORT_SYMBOL(empty_zero_page);
41 * The pmd table for the upper-most set of pages.
46 EXPORT_SYMBOL(pgprot_user);
48 pgprot_t pgprot_kernel;
49 EXPORT_SYMBOL(pgprot_kernel);
51 static int __init noalign_setup(char *__unused)
53 cr_alignment &= ~CR_A;
54 cr_no_alignment &= ~CR_A;
58 __setup("noalign", noalign_setup);
60 void adjust_cr(unsigned long mask, unsigned long set)
68 local_irq_save(flags);
70 cr_no_alignment = (cr_no_alignment & ~mask) | set;
71 cr_alignment = (cr_alignment & ~mask) | set;
73 set_cr((get_cr() & ~mask) | set);
75 local_irq_restore(flags);
79 unsigned long virtual;
85 #define PROT_PTE_DEVICE (PTE_PRESENT | PTE_YOUNG | \
86 PTE_DIRTY | PTE_READ | PTE_WRITE)
87 #define PROT_SECT_DEVICE (PMD_TYPE_SECT | PMD_PRESENT | \
88 PMD_SECT_READ | PMD_SECT_WRITE)
90 static struct mem_type mem_types[] = {
91 [MT_DEVICE] = { /* Strongly ordered */
92 .prot_pte = PROT_PTE_DEVICE,
93 .prot_l1 = PMD_TYPE_TABLE | PMD_PRESENT,
94 .prot_sect = PROT_SECT_DEVICE,
97 * MT_KUSER: pte for vecpage -- cacheable,
98 * and sect for unigfx mmap -- noncacheable
101 .prot_pte = PTE_PRESENT | PTE_YOUNG | PTE_DIRTY |
102 PTE_CACHEABLE | PTE_READ | PTE_EXEC,
103 .prot_l1 = PMD_TYPE_TABLE | PMD_PRESENT,
104 .prot_sect = PROT_SECT_DEVICE,
106 [MT_HIGH_VECTORS] = {
107 .prot_pte = PTE_PRESENT | PTE_YOUNG | PTE_DIRTY |
108 PTE_CACHEABLE | PTE_READ | PTE_WRITE |
110 .prot_l1 = PMD_TYPE_TABLE | PMD_PRESENT,
113 .prot_pte = PTE_PRESENT | PTE_YOUNG | PTE_DIRTY |
114 PTE_WRITE | PTE_EXEC,
115 .prot_l1 = PMD_TYPE_TABLE | PMD_PRESENT,
116 .prot_sect = PMD_TYPE_SECT | PMD_PRESENT | PMD_SECT_CACHEABLE |
117 PMD_SECT_READ | PMD_SECT_WRITE | PMD_SECT_EXEC,
120 .prot_sect = PMD_TYPE_SECT | PMD_PRESENT | PMD_SECT_CACHEABLE |
125 const struct mem_type *get_mem_type(unsigned int type)
127 return type < ARRAY_SIZE(mem_types) ? &mem_types[type] : NULL;
129 EXPORT_SYMBOL(get_mem_type);
132 * Adjust the PMD section entries according to the CPU in use.
134 static void __init build_mem_type_table(void)
136 pgprot_user = __pgprot(PTE_PRESENT | PTE_YOUNG | PTE_CACHEABLE);
137 pgprot_kernel = __pgprot(PTE_PRESENT | PTE_YOUNG |
138 PTE_DIRTY | PTE_READ | PTE_WRITE |
139 PTE_EXEC | PTE_CACHEABLE);
142 #define vectors_base() (vectors_high() ? 0xffff0000 : 0)
144 static void __init *early_alloc(unsigned long sz)
146 return memblock_alloc(sz, sz);
149 static pte_t * __init early_pte_alloc(pmd_t *pmd, unsigned long addr,
152 if (pmd_none(*pmd)) {
153 pte_t *pte = early_alloc(PTRS_PER_PTE * sizeof(pte_t));
154 __pmd_populate(pmd, __pa(pte) | prot);
156 BUG_ON(pmd_bad(*pmd));
157 return pte_offset_kernel(pmd, addr);
160 static void __init alloc_init_pte(pmd_t *pmd, unsigned long addr,
161 unsigned long end, unsigned long pfn,
162 const struct mem_type *type)
164 pte_t *pte = early_pte_alloc(pmd, addr, type->prot_l1);
166 set_pte(pte, pfn_pte(pfn, __pgprot(type->prot_pte)));
168 } while (pte++, addr += PAGE_SIZE, addr != end);
171 static void __init alloc_init_section(pgd_t *pgd, unsigned long addr,
172 unsigned long end, unsigned long phys,
173 const struct mem_type *type)
175 pmd_t *pmd = pmd_offset((pud_t *)pgd, addr);
178 * Try a section mapping - end, addr and phys must all be aligned
179 * to a section boundary.
181 if (((addr | end | phys) & ~SECTION_MASK) == 0) {
185 set_pmd(pmd, __pmd(phys | type->prot_sect));
186 phys += SECTION_SIZE;
187 } while (pmd++, addr += SECTION_SIZE, addr != end);
192 * No need to loop; pte's aren't interested in the
193 * individual L1 entries.
195 alloc_init_pte(pmd, addr, end, __phys_to_pfn(phys), type);
200 * Create the page directory entries and any necessary
201 * page tables for the mapping specified by `md'. We
202 * are able to cope here with varying sizes and address
203 * offsets, and we take full advantage of sections.
205 static void __init create_mapping(struct map_desc *md)
207 unsigned long phys, addr, length, end;
208 const struct mem_type *type;
211 if (md->virtual != vectors_base() && md->virtual < TASK_SIZE) {
212 printk(KERN_WARNING "BUG: not creating mapping for "
213 "0x%08llx at 0x%08lx in user region\n",
214 __pfn_to_phys((u64)md->pfn), md->virtual);
218 if ((md->type == MT_DEVICE || md->type == MT_ROM) &&
219 md->virtual >= PAGE_OFFSET && md->virtual < VMALLOC_END) {
220 printk(KERN_WARNING "BUG: mapping for 0x%08llx at 0x%08lx "
221 "overlaps vmalloc space\n",
222 __pfn_to_phys((u64)md->pfn), md->virtual);
225 type = &mem_types[md->type];
227 addr = md->virtual & PAGE_MASK;
228 phys = (unsigned long)__pfn_to_phys(md->pfn);
229 length = PAGE_ALIGN(md->length + (md->virtual & ~PAGE_MASK));
231 if (type->prot_l1 == 0 && ((addr | phys | length) & ~SECTION_MASK)) {
232 printk(KERN_WARNING "BUG: map for 0x%08lx at 0x%08lx can not "
233 "be mapped using pages, ignoring.\n",
234 __pfn_to_phys(md->pfn), addr);
238 pgd = pgd_offset_k(addr);
241 unsigned long next = pgd_addr_end(addr, end);
243 alloc_init_section(pgd, addr, next, phys, type);
247 } while (pgd++, addr != end);
250 static void * __initdata vmalloc_min = (void *)(VMALLOC_END - SZ_128M);
253 * vmalloc=size forces the vmalloc area to be exactly 'size'
254 * bytes. This can be used to increase (or decrease) the vmalloc
255 * area - the default is 128m.
257 static int __init early_vmalloc(char *arg)
259 unsigned long vmalloc_reserve = memparse(arg, NULL);
261 if (vmalloc_reserve < SZ_16M) {
262 vmalloc_reserve = SZ_16M;
264 "vmalloc area too small, limiting to %luMB\n",
265 vmalloc_reserve >> 20);
268 if (vmalloc_reserve > VMALLOC_END - (PAGE_OFFSET + SZ_32M)) {
269 vmalloc_reserve = VMALLOC_END - (PAGE_OFFSET + SZ_32M);
271 "vmalloc area is too big, limiting to %luMB\n",
272 vmalloc_reserve >> 20);
275 vmalloc_min = (void *)(VMALLOC_END - vmalloc_reserve);
278 early_param("vmalloc", early_vmalloc);
280 static phys_addr_t lowmem_limit __initdata = SZ_1G;
282 static void __init sanity_check_meminfo(void)
286 lowmem_limit = __pa(vmalloc_min - 1) + 1;
287 memblock_set_current_limit(lowmem_limit);
289 for (i = 0, j = 0; i < meminfo.nr_banks; i++) {
290 struct membank *bank = &meminfo.bank[j];
291 *bank = meminfo.bank[i];
294 meminfo.nr_banks = j;
297 static inline void prepare_page_table(void)
303 * Clear out all the mappings below the kernel image.
305 for (addr = 0; addr < MODULES_VADDR; addr += PGDIR_SIZE)
306 pmd_clear(pmd_off_k(addr));
308 for ( ; addr < PAGE_OFFSET; addr += PGDIR_SIZE)
309 pmd_clear(pmd_off_k(addr));
312 * Find the end of the first block of lowmem.
314 end = memblock.memory.regions[0].base + memblock.memory.regions[0].size;
315 if (end >= lowmem_limit)
319 * Clear out all the kernel space mappings, except for the first
320 * memory bank, up to the end of the vmalloc region.
322 for (addr = __phys_to_virt(end);
323 addr < VMALLOC_END; addr += PGDIR_SIZE)
324 pmd_clear(pmd_off_k(addr));
328 * Reserve the special regions of memory
330 void __init uc32_mm_memblock_reserve(void)
333 * Reserve the page tables. These are already in use,
334 * and can only be in node 0.
336 memblock_reserve(__pa(swapper_pg_dir), PTRS_PER_PGD * sizeof(pgd_t));
340 * Set up device the mappings. Since we clear out the page tables for all
341 * mappings above VMALLOC_END, we will remove any debug device mappings.
342 * This means you have to be careful how you debug this function, or any
343 * called function. This means you can't use any function or debugging
344 * method which may touch any device, otherwise the kernel _will_ crash.
346 static void __init devicemaps_init(void)
353 * Allocate the vector page early.
355 vectors = early_alloc(PAGE_SIZE);
357 for (addr = VMALLOC_END; addr; addr += PGDIR_SIZE)
358 pmd_clear(pmd_off_k(addr));
361 * Create a mapping for the machine vectors at the high-vectors
362 * location (0xffff0000). If we aren't using high-vectors, also
363 * create a mapping at the low-vectors virtual address.
365 map.pfn = __phys_to_pfn(virt_to_phys(vectors));
366 map.virtual = VECTORS_BASE;
367 map.length = PAGE_SIZE;
368 map.type = MT_HIGH_VECTORS;
369 create_mapping(&map);
372 * Create a mapping for the kuser page at the special
373 * location (0xbfff0000) to the same vectors location.
375 map.pfn = __phys_to_pfn(virt_to_phys(vectors));
376 map.virtual = KUSER_VECPAGE_BASE;
377 map.length = PAGE_SIZE;
379 create_mapping(&map);
382 * Finally flush the caches and tlb to ensure that we're in a
383 * consistent state wrt the writebuffer. This also ensures that
384 * any write-allocated cache lines in the vector page are written
385 * back. After this point, we can start to touch devices again.
387 local_flush_tlb_all();
391 static void __init map_lowmem(void)
393 struct memblock_region *reg;
395 /* Map all the lowmem memory banks. */
396 for_each_memblock(memory, reg) {
397 phys_addr_t start = reg->base;
398 phys_addr_t end = start + reg->size;
401 if (end > lowmem_limit)
406 map.pfn = __phys_to_pfn(start);
407 map.virtual = __phys_to_virt(start);
408 map.length = end - start;
409 map.type = MT_MEMORY;
411 create_mapping(&map);
416 * paging_init() sets up the page tables, initialises the zone memory
417 * maps, and sets up the zero page, bad page and bad page tables.
419 void __init paging_init(void)
423 build_mem_type_table();
424 sanity_check_meminfo();
425 prepare_page_table();
429 top_pmd = pmd_off_k(0xffff0000);
431 /* allocate the zero page. */
432 zero_page = early_alloc(PAGE_SIZE);
436 empty_zero_page = virt_to_page(zero_page);
437 __flush_dcache_page(NULL, empty_zero_page);
441 * In order to soft-boot, we need to insert a 1:1 mapping in place of
442 * the user-mode pages. This will then ensure that we have predictable
443 * results when turning the mmu off
445 void setup_mm_for_reboot(void)
447 unsigned long base_pmdval;
452 * We need to access to user-mode page tables here. For kernel threads
453 * we don't have any user-mode mappings so we use the context that we
456 pgd = current->active_mm->pgd;
458 base_pmdval = PMD_SECT_WRITE | PMD_SECT_READ | PMD_TYPE_SECT;
460 for (i = 0; i < FIRST_USER_PGD_NR + USER_PTRS_PER_PGD; i++, pgd++) {
461 unsigned long pmdval = (i << PGDIR_SHIFT) | base_pmdval;
464 pmd = pmd_off(pgd, i << PGDIR_SHIFT);
465 set_pmd(pmd, __pmd(pmdval));
466 flush_pmd_entry(pmd);
469 local_flush_tlb_all();
473 * Take care of architecture specific things when placing a new PTE into
474 * a page table, or changing an existing PTE. Basically, there are two
475 * things that we need to take care of:
477 * 1. If PG_dcache_clean is not set for the page, we need to ensure
478 * that any cache entries for the kernels virtual memory
479 * range are written back to the page.
480 * 2. If we have multiple shared mappings of the same space in
481 * an object, we need to deal with the cache aliasing issues.
483 * Note that the pte lock will be held.
485 void update_mmu_cache(struct vm_area_struct *vma, unsigned long addr,
488 unsigned long pfn = pte_pfn(*ptep);
489 struct address_space *mapping;
496 * The zero page is never written to, so never has any dirty
497 * cache lines, and therefore never needs to be flushed.
499 page = pfn_to_page(pfn);
500 if (page == ZERO_PAGE(0))
503 mapping = page_mapping_file(page);
504 if (!test_and_set_bit(PG_dcache_clean, &page->flags))
505 __flush_dcache_page(mapping, page);
507 if (vma->vm_flags & VM_EXEC)
508 __flush_icache_all();