mm/debug-pagealloc: make debug-pagealloc boottime configurable
[deliverable/linux.git] / arch / powerpc / mm / pgtable_32.c
1 /*
2 * This file contains the routines setting up the linux page tables.
3 * -- paulus
4 *
5 * Derived from arch/ppc/mm/init.c:
6 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
7 *
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
11 *
12 * Derived from "arch/i386/mm/init.c"
13 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
14 *
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.
19 *
20 */
21
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>
31
32 #include <asm/pgtable.h>
33 #include <asm/pgalloc.h>
34 #include <asm/fixmap.h>
35 #include <asm/io.h>
36 #include <asm/setup.h>
37
38 #include "mmu_decl.h"
39
40 unsigned long ioremap_base;
41 unsigned long ioremap_bot;
42 EXPORT_SYMBOL(ioremap_bot); /* aka VMALLOC_END */
43
44 #ifdef CONFIG_6xx
45 #define HAVE_BATS 1
46 #endif
47
48 #if defined(CONFIG_FSL_BOOKE)
49 #define HAVE_TLBCAM 1
50 #endif
51
52 extern char etext[], _stext[];
53
54 #ifdef HAVE_BATS
55 extern phys_addr_t v_mapped_by_bats(unsigned long va);
56 extern unsigned long p_mapped_by_bats(phys_addr_t pa);
57 void setbat(int index, unsigned long virt, phys_addr_t phys,
58 unsigned int size, int flags);
59
60 #else /* !HAVE_BATS */
61 #define v_mapped_by_bats(x) (0UL)
62 #define p_mapped_by_bats(x) (0UL)
63 #endif /* HAVE_BATS */
64
65 #ifdef HAVE_TLBCAM
66 extern unsigned int tlbcam_index;
67 extern phys_addr_t v_mapped_by_tlbcam(unsigned long va);
68 extern unsigned long p_mapped_by_tlbcam(phys_addr_t pa);
69 #else /* !HAVE_TLBCAM */
70 #define v_mapped_by_tlbcam(x) (0UL)
71 #define p_mapped_by_tlbcam(x) (0UL)
72 #endif /* HAVE_TLBCAM */
73
74 #define PGDIR_ORDER (32 + PGD_T_LOG2 - PGDIR_SHIFT)
75
76 pgd_t *pgd_alloc(struct mm_struct *mm)
77 {
78 pgd_t *ret;
79
80 /* pgdir take page or two with 4K pages and a page fraction otherwise */
81 #ifndef CONFIG_PPC_4K_PAGES
82 ret = kzalloc(1 << PGDIR_ORDER, GFP_KERNEL);
83 #else
84 ret = (pgd_t *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
85 PGDIR_ORDER - PAGE_SHIFT);
86 #endif
87 return ret;
88 }
89
90 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
91 {
92 #ifndef CONFIG_PPC_4K_PAGES
93 kfree((void *)pgd);
94 #else
95 free_pages((unsigned long)pgd, PGDIR_ORDER - PAGE_SHIFT);
96 #endif
97 }
98
99 __init_refok pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
100 {
101 pte_t *pte;
102 extern int mem_init_done;
103
104 if (mem_init_done) {
105 pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
106 } else {
107 pte = __va(memblock_alloc(PAGE_SIZE, PAGE_SIZE));
108 if (pte)
109 clear_page(pte);
110 }
111 return pte;
112 }
113
114 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
115 {
116 struct page *ptepage;
117
118 gfp_t flags = GFP_KERNEL | __GFP_REPEAT | __GFP_ZERO;
119
120 ptepage = alloc_pages(flags, 0);
121 if (!ptepage)
122 return NULL;
123 if (!pgtable_page_ctor(ptepage)) {
124 __free_page(ptepage);
125 return NULL;
126 }
127 return ptepage;
128 }
129
130 void __iomem *
131 ioremap(phys_addr_t addr, unsigned long size)
132 {
133 return __ioremap_caller(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED,
134 __builtin_return_address(0));
135 }
136 EXPORT_SYMBOL(ioremap);
137
138 void __iomem *
139 ioremap_wc(phys_addr_t addr, unsigned long size)
140 {
141 return __ioremap_caller(addr, size, _PAGE_NO_CACHE,
142 __builtin_return_address(0));
143 }
144 EXPORT_SYMBOL(ioremap_wc);
145
146 void __iomem *
147 ioremap_prot(phys_addr_t addr, unsigned long size, unsigned long flags)
148 {
149 /* writeable implies dirty for kernel addresses */
150 if (flags & _PAGE_RW)
151 flags |= _PAGE_DIRTY | _PAGE_HWWRITE;
152
153 /* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
154 flags &= ~(_PAGE_USER | _PAGE_EXEC);
155
156 #ifdef _PAGE_BAP_SR
157 /* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format
158 * which means that we just cleared supervisor access... oops ;-) This
159 * restores it
160 */
161 flags |= _PAGE_BAP_SR;
162 #endif
163
164 return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
165 }
166 EXPORT_SYMBOL(ioremap_prot);
167
168 void __iomem *
169 __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
170 {
171 return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
172 }
173
174 void __iomem *
175 __ioremap_caller(phys_addr_t addr, unsigned long size, unsigned long flags,
176 void *caller)
177 {
178 unsigned long v, i;
179 phys_addr_t p;
180 int err;
181
182 /* Make sure we have the base flags */
183 if ((flags & _PAGE_PRESENT) == 0)
184 flags |= PAGE_KERNEL;
185
186 /* Non-cacheable page cannot be coherent */
187 if (flags & _PAGE_NO_CACHE)
188 flags &= ~_PAGE_COHERENT;
189
190 /*
191 * Choose an address to map it to.
192 * Once the vmalloc system is running, we use it.
193 * Before then, we use space going down from ioremap_base
194 * (ioremap_bot records where we're up to).
195 */
196 p = addr & PAGE_MASK;
197 size = PAGE_ALIGN(addr + size) - p;
198
199 /*
200 * If the address lies within the first 16 MB, assume it's in ISA
201 * memory space
202 */
203 if (p < 16*1024*1024)
204 p += _ISA_MEM_BASE;
205
206 #ifndef CONFIG_CRASH_DUMP
207 /*
208 * Don't allow anybody to remap normal RAM that we're using.
209 * mem_init() sets high_memory so only do the check after that.
210 */
211 if (mem_init_done && (p < virt_to_phys(high_memory)) &&
212 !(__allow_ioremap_reserved && memblock_is_region_reserved(p, size))) {
213 printk("__ioremap(): phys addr 0x%llx is RAM lr %pf\n",
214 (unsigned long long)p, __builtin_return_address(0));
215 return NULL;
216 }
217 #endif
218
219 if (size == 0)
220 return NULL;
221
222 /*
223 * Is it already mapped? Perhaps overlapped by a previous
224 * BAT mapping. If the whole area is mapped then we're done,
225 * otherwise remap it since we want to keep the virt addrs for
226 * each request contiguous.
227 *
228 * We make the assumption here that if the bottom and top
229 * of the range we want are mapped then it's mapped to the
230 * same virt address (and this is contiguous).
231 * -- Cort
232 */
233 if ((v = p_mapped_by_bats(p)) /*&& p_mapped_by_bats(p+size-1)*/ )
234 goto out;
235
236 if ((v = p_mapped_by_tlbcam(p)))
237 goto out;
238
239 if (mem_init_done) {
240 struct vm_struct *area;
241 area = get_vm_area_caller(size, VM_IOREMAP, caller);
242 if (area == 0)
243 return NULL;
244 area->phys_addr = p;
245 v = (unsigned long) area->addr;
246 } else {
247 v = (ioremap_bot -= size);
248 }
249
250 /*
251 * Should check if it is a candidate for a BAT mapping
252 */
253
254 err = 0;
255 for (i = 0; i < size && err == 0; i += PAGE_SIZE)
256 err = map_page(v+i, p+i, flags);
257 if (err) {
258 if (mem_init_done)
259 vunmap((void *)v);
260 return NULL;
261 }
262
263 out:
264 return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
265 }
266 EXPORT_SYMBOL(__ioremap);
267
268 void iounmap(volatile void __iomem *addr)
269 {
270 /*
271 * If mapped by BATs then there is nothing to do.
272 * Calling vfree() generates a benign warning.
273 */
274 if (v_mapped_by_bats((unsigned long)addr)) return;
275
276 if (addr > high_memory && (unsigned long) addr < ioremap_bot)
277 vunmap((void *) (PAGE_MASK & (unsigned long)addr));
278 }
279 EXPORT_SYMBOL(iounmap);
280
281 int map_page(unsigned long va, phys_addr_t pa, int flags)
282 {
283 pmd_t *pd;
284 pte_t *pg;
285 int err = -ENOMEM;
286
287 /* Use upper 10 bits of VA to index the first level map */
288 pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va);
289 /* Use middle 10 bits of VA to index the second-level map */
290 pg = pte_alloc_kernel(pd, va);
291 if (pg != 0) {
292 err = 0;
293 /* The PTE should never be already set nor present in the
294 * hash table
295 */
296 BUG_ON((pte_val(*pg) & (_PAGE_PRESENT | _PAGE_HASHPTE)) &&
297 flags);
298 set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT,
299 __pgprot(flags)));
300 }
301 smp_wmb();
302 return err;
303 }
304
305 /*
306 * Map in a chunk of physical memory starting at start.
307 */
308 void __init __mapin_ram_chunk(unsigned long offset, unsigned long top)
309 {
310 unsigned long v, s, f;
311 phys_addr_t p;
312 int ktext;
313
314 s = offset;
315 v = PAGE_OFFSET + s;
316 p = memstart_addr + s;
317 for (; s < top; s += PAGE_SIZE) {
318 ktext = ((char *) v >= _stext && (char *) v < etext);
319 f = ktext ? PAGE_KERNEL_TEXT : PAGE_KERNEL;
320 map_page(v, p, f);
321 #ifdef CONFIG_PPC_STD_MMU_32
322 if (ktext)
323 hash_preload(&init_mm, v, 0, 0x300);
324 #endif
325 v += PAGE_SIZE;
326 p += PAGE_SIZE;
327 }
328 }
329
330 void __init mapin_ram(void)
331 {
332 unsigned long s, top;
333
334 #ifndef CONFIG_WII
335 top = total_lowmem;
336 s = mmu_mapin_ram(top);
337 __mapin_ram_chunk(s, top);
338 #else
339 if (!wii_hole_size) {
340 s = mmu_mapin_ram(total_lowmem);
341 __mapin_ram_chunk(s, total_lowmem);
342 } else {
343 top = wii_hole_start;
344 s = mmu_mapin_ram(top);
345 __mapin_ram_chunk(s, top);
346
347 top = memblock_end_of_DRAM();
348 s = wii_mmu_mapin_mem2(top);
349 __mapin_ram_chunk(s, top);
350 }
351 #endif
352 }
353
354 /* Scan the real Linux page tables and return a PTE pointer for
355 * a virtual address in a context.
356 * Returns true (1) if PTE was found, zero otherwise. The pointer to
357 * the PTE pointer is unmodified if PTE is not found.
358 */
359 int
360 get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
361 {
362 pgd_t *pgd;
363 pud_t *pud;
364 pmd_t *pmd;
365 pte_t *pte;
366 int retval = 0;
367
368 pgd = pgd_offset(mm, addr & PAGE_MASK);
369 if (pgd) {
370 pud = pud_offset(pgd, addr & PAGE_MASK);
371 if (pud && pud_present(*pud)) {
372 pmd = pmd_offset(pud, addr & PAGE_MASK);
373 if (pmd_present(*pmd)) {
374 pte = pte_offset_map(pmd, addr & PAGE_MASK);
375 if (pte) {
376 retval = 1;
377 *ptep = pte;
378 if (pmdp)
379 *pmdp = pmd;
380 /* XXX caller needs to do pte_unmap, yuck */
381 }
382 }
383 }
384 }
385 return(retval);
386 }
387
388 #ifdef CONFIG_DEBUG_PAGEALLOC
389
390 static int __change_page_attr(struct page *page, pgprot_t prot)
391 {
392 pte_t *kpte;
393 pmd_t *kpmd;
394 unsigned long address;
395
396 BUG_ON(PageHighMem(page));
397 address = (unsigned long)page_address(page);
398
399 if (v_mapped_by_bats(address) || v_mapped_by_tlbcam(address))
400 return 0;
401 if (!get_pteptr(&init_mm, address, &kpte, &kpmd))
402 return -EINVAL;
403 __set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0);
404 wmb();
405 flush_tlb_page(NULL, address);
406 pte_unmap(kpte);
407
408 return 0;
409 }
410
411 /*
412 * Change the page attributes of an page in the linear mapping.
413 *
414 * THIS CONFLICTS WITH BAT MAPPINGS, DEBUG USE ONLY
415 */
416 static int change_page_attr(struct page *page, int numpages, pgprot_t prot)
417 {
418 int i, err = 0;
419 unsigned long flags;
420
421 local_irq_save(flags);
422 for (i = 0; i < numpages; i++, page++) {
423 err = __change_page_attr(page, prot);
424 if (err)
425 break;
426 }
427 local_irq_restore(flags);
428 return err;
429 }
430
431
432 void __kernel_map_pages(struct page *page, int numpages, int enable)
433 {
434 if (PageHighMem(page))
435 return;
436
437 change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
438 }
439 #endif /* CONFIG_DEBUG_PAGEALLOC */
440
441 static int fixmaps;
442
443 void __set_fixmap (enum fixed_addresses idx, phys_addr_t phys, pgprot_t flags)
444 {
445 unsigned long address = __fix_to_virt(idx);
446
447 if (idx >= __end_of_fixed_addresses) {
448 BUG();
449 return;
450 }
451
452 map_page(address, phys, pgprot_val(flags));
453 fixmaps++;
454 }
455
456 void __this_fixmap_does_not_exist(void)
457 {
458 WARN_ON(1);
459 }
This page took 0.038996 seconds and 5 git commands to generate.