Linux 2.6.39-rc2
[pohmelfs.git] / arch / powerpc / mm / pgtable_32.c
blob8dc41c0157fe04c8e85742116381f291329f57ad
1 /*
2 * This file contains the routines setting up the linux page tables.
3 * -- paulus
5 * Derived from arch/ppc/mm/init.c:
6 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
8 * Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
9 * and Cort Dougan (PReP) (cort@cs.nmt.edu)
10 * Copyright (C) 1996 Paul Mackerras
12 * Derived from "arch/i386/mm/init.c"
13 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
22 #include <linux/kernel.h>
23 #include <linux/module.h>
24 #include <linux/types.h>
25 #include <linux/mm.h>
26 #include <linux/vmalloc.h>
27 #include <linux/init.h>
28 #include <linux/highmem.h>
29 #include <linux/memblock.h>
30 #include <linux/slab.h>
32 #include <asm/pgtable.h>
33 #include <asm/pgalloc.h>
34 #include <asm/fixmap.h>
35 #include <asm/io.h>
37 #include "mmu_decl.h"
39 unsigned long ioremap_base;
40 unsigned long ioremap_bot;
41 EXPORT_SYMBOL(ioremap_bot); /* aka VMALLOC_END */
43 #if defined(CONFIG_6xx) || defined(CONFIG_POWER3)
44 #define HAVE_BATS 1
45 #endif
47 #if defined(CONFIG_FSL_BOOKE)
48 #define HAVE_TLBCAM 1
49 #endif
51 extern char etext[], _stext[];
53 #ifdef HAVE_BATS
54 extern phys_addr_t v_mapped_by_bats(unsigned long va);
55 extern unsigned long p_mapped_by_bats(phys_addr_t pa);
56 void setbat(int index, unsigned long virt, phys_addr_t phys,
57 unsigned int size, int flags);
59 #else /* !HAVE_BATS */
60 #define v_mapped_by_bats(x) (0UL)
61 #define p_mapped_by_bats(x) (0UL)
62 #endif /* HAVE_BATS */
64 #ifdef HAVE_TLBCAM
65 extern unsigned int tlbcam_index;
66 extern phys_addr_t v_mapped_by_tlbcam(unsigned long va);
67 extern unsigned long p_mapped_by_tlbcam(phys_addr_t pa);
68 #else /* !HAVE_TLBCAM */
69 #define v_mapped_by_tlbcam(x) (0UL)
70 #define p_mapped_by_tlbcam(x) (0UL)
71 #endif /* HAVE_TLBCAM */
73 #define PGDIR_ORDER (32 + PGD_T_LOG2 - PGDIR_SHIFT)
75 pgd_t *pgd_alloc(struct mm_struct *mm)
77 pgd_t *ret;
79 /* pgdir take page or two with 4K pages and a page fraction otherwise */
80 #ifndef CONFIG_PPC_4K_PAGES
81 ret = kzalloc(1 << PGDIR_ORDER, GFP_KERNEL);
82 #else
83 ret = (pgd_t *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
84 PGDIR_ORDER - PAGE_SHIFT);
85 #endif
86 return ret;
89 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
91 #ifndef CONFIG_PPC_4K_PAGES
92 kfree((void *)pgd);
93 #else
94 free_pages((unsigned long)pgd, PGDIR_ORDER - PAGE_SHIFT);
95 #endif
98 __init_refok pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
100 pte_t *pte;
101 extern int mem_init_done;
102 extern void *early_get_page(void);
104 if (mem_init_done) {
105 pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
106 } else {
107 pte = (pte_t *)early_get_page();
108 if (pte)
109 clear_page(pte);
111 return pte;
114 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
116 struct page *ptepage;
118 gfp_t flags = GFP_KERNEL | __GFP_REPEAT | __GFP_ZERO;
120 ptepage = alloc_pages(flags, 0);
121 if (!ptepage)
122 return NULL;
123 pgtable_page_ctor(ptepage);
124 return ptepage;
127 void __iomem *
128 ioremap(phys_addr_t addr, unsigned long size)
130 return __ioremap_caller(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED,
131 __builtin_return_address(0));
133 EXPORT_SYMBOL(ioremap);
135 void __iomem *
136 ioremap_flags(phys_addr_t addr, unsigned long size, unsigned long flags)
138 /* writeable implies dirty for kernel addresses */
139 if (flags & _PAGE_RW)
140 flags |= _PAGE_DIRTY | _PAGE_HWWRITE;
142 /* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
143 flags &= ~(_PAGE_USER | _PAGE_EXEC);
145 #ifdef _PAGE_BAP_SR
146 /* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format
147 * which means that we just cleared supervisor access... oops ;-) This
148 * restores it
150 flags |= _PAGE_BAP_SR;
151 #endif
153 return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
155 EXPORT_SYMBOL(ioremap_flags);
157 void __iomem *
158 __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
160 return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
163 void __iomem *
164 __ioremap_caller(phys_addr_t addr, unsigned long size, unsigned long flags,
165 void *caller)
167 unsigned long v, i;
168 phys_addr_t p;
169 int err;
171 /* Make sure we have the base flags */
172 if ((flags & _PAGE_PRESENT) == 0)
173 flags |= PAGE_KERNEL;
175 /* Non-cacheable page cannot be coherent */
176 if (flags & _PAGE_NO_CACHE)
177 flags &= ~_PAGE_COHERENT;
180 * Choose an address to map it to.
181 * Once the vmalloc system is running, we use it.
182 * Before then, we use space going down from ioremap_base
183 * (ioremap_bot records where we're up to).
185 p = addr & PAGE_MASK;
186 size = PAGE_ALIGN(addr + size) - p;
189 * If the address lies within the first 16 MB, assume it's in ISA
190 * memory space
192 if (p < 16*1024*1024)
193 p += _ISA_MEM_BASE;
195 #ifndef CONFIG_CRASH_DUMP
197 * Don't allow anybody to remap normal RAM that we're using.
198 * mem_init() sets high_memory so only do the check after that.
200 if (mem_init_done && (p < virt_to_phys(high_memory)) &&
201 !(__allow_ioremap_reserved && memblock_is_region_reserved(p, size))) {
202 printk("__ioremap(): phys addr 0x%llx is RAM lr %p\n",
203 (unsigned long long)p, __builtin_return_address(0));
204 return NULL;
206 #endif
208 if (size == 0)
209 return NULL;
212 * Is it already mapped? Perhaps overlapped by a previous
213 * BAT mapping. If the whole area is mapped then we're done,
214 * otherwise remap it since we want to keep the virt addrs for
215 * each request contiguous.
217 * We make the assumption here that if the bottom and top
218 * of the range we want are mapped then it's mapped to the
219 * same virt address (and this is contiguous).
220 * -- Cort
222 if ((v = p_mapped_by_bats(p)) /*&& p_mapped_by_bats(p+size-1)*/ )
223 goto out;
225 if ((v = p_mapped_by_tlbcam(p)))
226 goto out;
228 if (mem_init_done) {
229 struct vm_struct *area;
230 area = get_vm_area_caller(size, VM_IOREMAP, caller);
231 if (area == 0)
232 return NULL;
233 area->phys_addr = p;
234 v = (unsigned long) area->addr;
235 } else {
236 v = (ioremap_bot -= size);
240 * Should check if it is a candidate for a BAT mapping
243 err = 0;
244 for (i = 0; i < size && err == 0; i += PAGE_SIZE)
245 err = map_page(v+i, p+i, flags);
246 if (err) {
247 if (mem_init_done)
248 vunmap((void *)v);
249 return NULL;
252 out:
253 return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
255 EXPORT_SYMBOL(__ioremap);
257 void iounmap(volatile void __iomem *addr)
260 * If mapped by BATs then there is nothing to do.
261 * Calling vfree() generates a benign warning.
263 if (v_mapped_by_bats((unsigned long)addr)) return;
265 if (addr > high_memory && (unsigned long) addr < ioremap_bot)
266 vunmap((void *) (PAGE_MASK & (unsigned long)addr));
268 EXPORT_SYMBOL(iounmap);
270 int map_page(unsigned long va, phys_addr_t pa, int flags)
272 pmd_t *pd;
273 pte_t *pg;
274 int err = -ENOMEM;
276 /* Use upper 10 bits of VA to index the first level map */
277 pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va);
278 /* Use middle 10 bits of VA to index the second-level map */
279 pg = pte_alloc_kernel(pd, va);
280 if (pg != 0) {
281 err = 0;
282 /* The PTE should never be already set nor present in the
283 * hash table
285 BUG_ON((pte_val(*pg) & (_PAGE_PRESENT | _PAGE_HASHPTE)) &&
286 flags);
287 set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT,
288 __pgprot(flags)));
290 return err;
294 * Map in a chunk of physical memory starting at start.
296 void __init __mapin_ram_chunk(unsigned long offset, unsigned long top)
298 unsigned long v, s, f;
299 phys_addr_t p;
300 int ktext;
302 s = offset;
303 v = PAGE_OFFSET + s;
304 p = memstart_addr + s;
305 for (; s < top; s += PAGE_SIZE) {
306 ktext = ((char *) v >= _stext && (char *) v < etext);
307 f = ktext ? PAGE_KERNEL_TEXT : PAGE_KERNEL;
308 map_page(v, p, f);
309 #ifdef CONFIG_PPC_STD_MMU_32
310 if (ktext)
311 hash_preload(&init_mm, v, 0, 0x300);
312 #endif
313 v += PAGE_SIZE;
314 p += PAGE_SIZE;
318 void __init mapin_ram(void)
320 unsigned long s, top;
322 #ifndef CONFIG_WII
323 top = total_lowmem;
324 s = mmu_mapin_ram(top);
325 __mapin_ram_chunk(s, top);
326 #else
327 if (!wii_hole_size) {
328 s = mmu_mapin_ram(total_lowmem);
329 __mapin_ram_chunk(s, total_lowmem);
330 } else {
331 top = wii_hole_start;
332 s = mmu_mapin_ram(top);
333 __mapin_ram_chunk(s, top);
335 top = memblock_end_of_DRAM();
336 s = wii_mmu_mapin_mem2(top);
337 __mapin_ram_chunk(s, top);
339 #endif
342 /* Scan the real Linux page tables and return a PTE pointer for
343 * a virtual address in a context.
344 * Returns true (1) if PTE was found, zero otherwise. The pointer to
345 * the PTE pointer is unmodified if PTE is not found.
348 get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
350 pgd_t *pgd;
351 pud_t *pud;
352 pmd_t *pmd;
353 pte_t *pte;
354 int retval = 0;
356 pgd = pgd_offset(mm, addr & PAGE_MASK);
357 if (pgd) {
358 pud = pud_offset(pgd, addr & PAGE_MASK);
359 if (pud && pud_present(*pud)) {
360 pmd = pmd_offset(pud, addr & PAGE_MASK);
361 if (pmd_present(*pmd)) {
362 pte = pte_offset_map(pmd, addr & PAGE_MASK);
363 if (pte) {
364 retval = 1;
365 *ptep = pte;
366 if (pmdp)
367 *pmdp = pmd;
368 /* XXX caller needs to do pte_unmap, yuck */
373 return(retval);
376 #ifdef CONFIG_DEBUG_PAGEALLOC
378 static int __change_page_attr(struct page *page, pgprot_t prot)
380 pte_t *kpte;
381 pmd_t *kpmd;
382 unsigned long address;
384 BUG_ON(PageHighMem(page));
385 address = (unsigned long)page_address(page);
387 if (v_mapped_by_bats(address) || v_mapped_by_tlbcam(address))
388 return 0;
389 if (!get_pteptr(&init_mm, address, &kpte, &kpmd))
390 return -EINVAL;
391 __set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0);
392 wmb();
393 flush_tlb_page(NULL, address);
394 pte_unmap(kpte);
396 return 0;
400 * Change the page attributes of an page in the linear mapping.
402 * THIS CONFLICTS WITH BAT MAPPINGS, DEBUG USE ONLY
404 static int change_page_attr(struct page *page, int numpages, pgprot_t prot)
406 int i, err = 0;
407 unsigned long flags;
409 local_irq_save(flags);
410 for (i = 0; i < numpages; i++, page++) {
411 err = __change_page_attr(page, prot);
412 if (err)
413 break;
415 local_irq_restore(flags);
416 return err;
420 void kernel_map_pages(struct page *page, int numpages, int enable)
422 if (PageHighMem(page))
423 return;
425 change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
427 #endif /* CONFIG_DEBUG_PAGEALLOC */
429 static int fixmaps;
431 void __set_fixmap (enum fixed_addresses idx, phys_addr_t phys, pgprot_t flags)
433 unsigned long address = __fix_to_virt(idx);
435 if (idx >= __end_of_fixed_addresses) {
436 BUG();
437 return;
440 map_page(address, phys, pgprot_val(flags));
441 fixmaps++;
444 void __this_fixmap_does_not_exist(void)
446 WARN_ON(1);