Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
[linux-2.6.git] / arch / mips / mm / init.c
1 /*
2  * This file is subject to the terms and conditions of the GNU General Public
3  * License.  See the file "COPYING" in the main directory of this archive
4  * for more details.
5  *
6  * Copyright (C) 1994 - 2000 Ralf Baechle
7  * Copyright (C) 1999, 2000 Silicon Graphics, Inc.
8  * Kevin D. Kissell, kevink@mips.com and Carsten Langgaard, carstenl@mips.com
9  * Copyright (C) 2000 MIPS Technologies, Inc.  All rights reserved.
10  */
11 #include <linux/bug.h>
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/signal.h>
15 #include <linux/sched.h>
16 #include <linux/smp.h>
17 #include <linux/kernel.h>
18 #include <linux/errno.h>
19 #include <linux/string.h>
20 #include <linux/types.h>
21 #include <linux/pagemap.h>
22 #include <linux/ptrace.h>
23 #include <linux/mman.h>
24 #include <linux/mm.h>
25 #include <linux/bootmem.h>
26 #include <linux/highmem.h>
27 #include <linux/swap.h>
28 #include <linux/proc_fs.h>
29 #include <linux/pfn.h>
30 #include <linux/hardirq.h>
31 #include <linux/gfp.h>
32
33 #include <asm/asm-offsets.h>
34 #include <asm/bootinfo.h>
35 #include <asm/cachectl.h>
36 #include <asm/cpu.h>
37 #include <asm/dma.h>
38 #include <asm/kmap_types.h>
39 #include <asm/mmu_context.h>
40 #include <asm/sections.h>
41 #include <asm/pgtable.h>
42 #include <asm/pgalloc.h>
43 #include <asm/tlb.h>
44 #include <asm/fixmap.h>
45
46 /* Atomicity and interruptability */
47 #ifdef CONFIG_MIPS_MT_SMTC
48
49 #include <asm/mipsmtregs.h>
50
51 #define ENTER_CRITICAL(flags) \
52         { \
53         unsigned int mvpflags; \
54         local_irq_save(flags);\
55         mvpflags = dvpe()
56 #define EXIT_CRITICAL(flags) \
57         evpe(mvpflags); \
58         local_irq_restore(flags); \
59         }
60 #else
61
62 #define ENTER_CRITICAL(flags) local_irq_save(flags)
63 #define EXIT_CRITICAL(flags) local_irq_restore(flags)
64
65 #endif /* CONFIG_MIPS_MT_SMTC */
66
67 /*
68  * We have up to 8 empty zeroed pages so we can map one of the right colour
69  * when needed.  This is necessary only on R4000 / R4400 SC and MC versions
70  * where we have to avoid VCED / VECI exceptions for good performance at
71  * any price.  Since page is never written to after the initialization we
72  * don't have to care about aliases on other CPUs.
73  */
74 unsigned long empty_zero_page, zero_page_mask;
75 EXPORT_SYMBOL_GPL(empty_zero_page);
76
77 /*
78  * Not static inline because used by IP27 special magic initialization code
79  */
80 unsigned long setup_zero_pages(void)
81 {
82         unsigned int order;
83         unsigned long size;
84         struct page *page;
85
86         if (cpu_has_vce)
87                 order = 3;
88         else
89                 order = 0;
90
91         empty_zero_page = __get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
92         if (!empty_zero_page)
93                 panic("Oh boy, that early out of memory?");
94
95         page = virt_to_page((void *)empty_zero_page);
96         split_page(page, order);
97         while (page < virt_to_page((void *)(empty_zero_page + (PAGE_SIZE << order)))) {
98                 SetPageReserved(page);
99                 page++;
100         }
101
102         size = PAGE_SIZE << order;
103         zero_page_mask = (size - 1) & PAGE_MASK;
104
105         return 1UL << order;
106 }
107
108 #ifdef CONFIG_MIPS_MT_SMTC
109 static pte_t *kmap_coherent_pte;
110 static void __init kmap_coherent_init(void)
111 {
112         unsigned long vaddr;
113
114         /* cache the first coherent kmap pte */
115         vaddr = __fix_to_virt(FIX_CMAP_BEGIN);
116         kmap_coherent_pte = kmap_get_fixmap_pte(vaddr);
117 }
118 #else
119 static inline void kmap_coherent_init(void) {}
120 #endif
121
122 void *kmap_coherent(struct page *page, unsigned long addr)
123 {
124         enum fixed_addresses idx;
125         unsigned long vaddr, flags, entrylo;
126         unsigned long old_ctx;
127         pte_t pte;
128         int tlbidx;
129
130         BUG_ON(Page_dcache_dirty(page));
131
132         inc_preempt_count();
133         idx = (addr >> PAGE_SHIFT) & (FIX_N_COLOURS - 1);
134 #ifdef CONFIG_MIPS_MT_SMTC
135         idx += FIX_N_COLOURS * smp_processor_id() +
136                 (in_interrupt() ? (FIX_N_COLOURS * NR_CPUS) : 0);
137 #else
138         idx += in_interrupt() ? FIX_N_COLOURS : 0;
139 #endif
140         vaddr = __fix_to_virt(FIX_CMAP_END - idx);
141         pte = mk_pte(page, PAGE_KERNEL);
142 #if defined(CONFIG_64BIT_PHYS_ADDR) && defined(CONFIG_CPU_MIPS32)
143         entrylo = pte.pte_high;
144 #else
145         entrylo = pte_to_entrylo(pte_val(pte));
146 #endif
147
148         ENTER_CRITICAL(flags);
149         old_ctx = read_c0_entryhi();
150         write_c0_entryhi(vaddr & (PAGE_MASK << 1));
151         write_c0_entrylo0(entrylo);
152         write_c0_entrylo1(entrylo);
153 #ifdef CONFIG_MIPS_MT_SMTC
154         set_pte(kmap_coherent_pte - (FIX_CMAP_END - idx), pte);
155         /* preload TLB instead of local_flush_tlb_one() */
156         mtc0_tlbw_hazard();
157         tlb_probe();
158         tlb_probe_hazard();
159         tlbidx = read_c0_index();
160         mtc0_tlbw_hazard();
161         if (tlbidx < 0)
162                 tlb_write_random();
163         else
164                 tlb_write_indexed();
165 #else
166         tlbidx = read_c0_wired();
167         write_c0_wired(tlbidx + 1);
168         write_c0_index(tlbidx);
169         mtc0_tlbw_hazard();
170         tlb_write_indexed();
171 #endif
172         tlbw_use_hazard();
173         write_c0_entryhi(old_ctx);
174         EXIT_CRITICAL(flags);
175
176         return (void*) vaddr;
177 }
178
179 #define UNIQUE_ENTRYHI(idx) (CKSEG0 + ((idx) << (PAGE_SHIFT + 1)))
180
181 void kunmap_coherent(void)
182 {
183 #ifndef CONFIG_MIPS_MT_SMTC
184         unsigned int wired;
185         unsigned long flags, old_ctx;
186
187         ENTER_CRITICAL(flags);
188         old_ctx = read_c0_entryhi();
189         wired = read_c0_wired() - 1;
190         write_c0_wired(wired);
191         write_c0_index(wired);
192         write_c0_entryhi(UNIQUE_ENTRYHI(wired));
193         write_c0_entrylo0(0);
194         write_c0_entrylo1(0);
195         mtc0_tlbw_hazard();
196         tlb_write_indexed();
197         tlbw_use_hazard();
198         write_c0_entryhi(old_ctx);
199         EXIT_CRITICAL(flags);
200 #endif
201         dec_preempt_count();
202         preempt_check_resched();
203 }
204
205 void copy_user_highpage(struct page *to, struct page *from,
206         unsigned long vaddr, struct vm_area_struct *vma)
207 {
208         void *vfrom, *vto;
209
210         vto = kmap_atomic(to, KM_USER1);
211         if (cpu_has_dc_aliases &&
212             page_mapped(from) && !Page_dcache_dirty(from)) {
213                 vfrom = kmap_coherent(from, vaddr);
214                 copy_page(vto, vfrom);
215                 kunmap_coherent();
216         } else {
217                 vfrom = kmap_atomic(from, KM_USER0);
218                 copy_page(vto, vfrom);
219                 kunmap_atomic(vfrom, KM_USER0);
220         }
221         if ((!cpu_has_ic_fills_f_dc) ||
222             pages_do_alias((unsigned long)vto, vaddr & PAGE_MASK))
223                 flush_data_cache_page((unsigned long)vto);
224         kunmap_atomic(vto, KM_USER1);
225         /* Make sure this page is cleared on other CPU's too before using it */
226         smp_wmb();
227 }
228
229 void copy_to_user_page(struct vm_area_struct *vma,
230         struct page *page, unsigned long vaddr, void *dst, const void *src,
231         unsigned long len)
232 {
233         if (cpu_has_dc_aliases &&
234             page_mapped(page) && !Page_dcache_dirty(page)) {
235                 void *vto = kmap_coherent(page, vaddr) + (vaddr & ~PAGE_MASK);
236                 memcpy(vto, src, len);
237                 kunmap_coherent();
238         } else {
239                 memcpy(dst, src, len);
240                 if (cpu_has_dc_aliases)
241                         SetPageDcacheDirty(page);
242         }
243         if ((vma->vm_flags & VM_EXEC) && !cpu_has_ic_fills_f_dc)
244                 flush_cache_page(vma, vaddr, page_to_pfn(page));
245 }
246
247 void copy_from_user_page(struct vm_area_struct *vma,
248         struct page *page, unsigned long vaddr, void *dst, const void *src,
249         unsigned long len)
250 {
251         if (cpu_has_dc_aliases &&
252             page_mapped(page) && !Page_dcache_dirty(page)) {
253                 void *vfrom = kmap_coherent(page, vaddr) + (vaddr & ~PAGE_MASK);
254                 memcpy(dst, vfrom, len);
255                 kunmap_coherent();
256         } else {
257                 memcpy(dst, src, len);
258                 if (cpu_has_dc_aliases)
259                         SetPageDcacheDirty(page);
260         }
261 }
262
263 void __init fixrange_init(unsigned long start, unsigned long end,
264         pgd_t *pgd_base)
265 {
266 #if defined(CONFIG_HIGHMEM) || defined(CONFIG_MIPS_MT_SMTC)
267         pgd_t *pgd;
268         pud_t *pud;
269         pmd_t *pmd;
270         pte_t *pte;
271         int i, j, k;
272         unsigned long vaddr;
273
274         vaddr = start;
275         i = __pgd_offset(vaddr);
276         j = __pud_offset(vaddr);
277         k = __pmd_offset(vaddr);
278         pgd = pgd_base + i;
279
280         for ( ; (i < PTRS_PER_PGD) && (vaddr != end); pgd++, i++) {
281                 pud = (pud_t *)pgd;
282                 for ( ; (j < PTRS_PER_PUD) && (vaddr != end); pud++, j++) {
283                         pmd = (pmd_t *)pud;
284                         for (; (k < PTRS_PER_PMD) && (vaddr != end); pmd++, k++) {
285                                 if (pmd_none(*pmd)) {
286                                         pte = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
287                                         set_pmd(pmd, __pmd((unsigned long)pte));
288                                         BUG_ON(pte != pte_offset_kernel(pmd, 0));
289                                 }
290                                 vaddr += PMD_SIZE;
291                         }
292                         k = 0;
293                 }
294                 j = 0;
295         }
296 #endif
297 }
298
299 #ifndef CONFIG_NEED_MULTIPLE_NODES
300 int page_is_ram(unsigned long pagenr)
301 {
302         int i;
303
304         for (i = 0; i < boot_mem_map.nr_map; i++) {
305                 unsigned long addr, end;
306
307                 if (boot_mem_map.map[i].type != BOOT_MEM_RAM)
308                         /* not usable memory */
309                         continue;
310
311                 addr = PFN_UP(boot_mem_map.map[i].addr);
312                 end = PFN_DOWN(boot_mem_map.map[i].addr +
313                                boot_mem_map.map[i].size);
314
315                 if (pagenr >= addr && pagenr < end)
316                         return 1;
317         }
318
319         return 0;
320 }
321
322 void __init paging_init(void)
323 {
324         unsigned long max_zone_pfns[MAX_NR_ZONES];
325         unsigned long lastpfn __maybe_unused;
326
327         pagetable_init();
328
329 #ifdef CONFIG_HIGHMEM
330         kmap_init();
331 #endif
332         kmap_coherent_init();
333
334 #ifdef CONFIG_ZONE_DMA
335         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
336 #endif
337 #ifdef CONFIG_ZONE_DMA32
338         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
339 #endif
340         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
341         lastpfn = max_low_pfn;
342 #ifdef CONFIG_HIGHMEM
343         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
344         lastpfn = highend_pfn;
345
346         if (cpu_has_dc_aliases && max_low_pfn != highend_pfn) {
347                 printk(KERN_WARNING "This processor doesn't support highmem."
348                        " %ldk highmem ignored\n",
349                        (highend_pfn - max_low_pfn) << (PAGE_SHIFT - 10));
350                 max_zone_pfns[ZONE_HIGHMEM] = max_low_pfn;
351                 lastpfn = max_low_pfn;
352         }
353 #endif
354
355         free_area_init_nodes(max_zone_pfns);
356 }
357
358 #ifdef CONFIG_64BIT
359 static struct kcore_list kcore_kseg0;
360 #endif
361
362 void __init mem_init(void)
363 {
364         unsigned long codesize, reservedpages, datasize, initsize;
365         unsigned long tmp, ram;
366
367 #ifdef CONFIG_HIGHMEM
368 #ifdef CONFIG_DISCONTIGMEM
369 #error "CONFIG_HIGHMEM and CONFIG_DISCONTIGMEM dont work together yet"
370 #endif
371         max_mapnr = highend_pfn;
372 #else
373         max_mapnr = max_low_pfn;
374 #endif
375         high_memory = (void *) __va(max_low_pfn << PAGE_SHIFT);
376
377         totalram_pages += free_all_bootmem();
378         totalram_pages -= setup_zero_pages();   /* Setup zeroed pages.  */
379
380         reservedpages = ram = 0;
381         for (tmp = 0; tmp < max_low_pfn; tmp++)
382                 if (page_is_ram(tmp)) {
383                         ram++;
384                         if (PageReserved(pfn_to_page(tmp)))
385                                 reservedpages++;
386                 }
387         num_physpages = ram;
388
389 #ifdef CONFIG_HIGHMEM
390         for (tmp = highstart_pfn; tmp < highend_pfn; tmp++) {
391                 struct page *page = pfn_to_page(tmp);
392
393                 if (!page_is_ram(tmp)) {
394                         SetPageReserved(page);
395                         continue;
396                 }
397                 ClearPageReserved(page);
398                 init_page_count(page);
399                 __free_page(page);
400                 totalhigh_pages++;
401         }
402         totalram_pages += totalhigh_pages;
403         num_physpages += totalhigh_pages;
404 #endif
405
406         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
407         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
408         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
409
410 #ifdef CONFIG_64BIT
411         if ((unsigned long) &_text > (unsigned long) CKSEG0)
412                 /* The -4 is a hack so that user tools don't have to handle
413                    the overflow.  */
414                 kclist_add(&kcore_kseg0, (void *) CKSEG0,
415                                 0x80000000 - 4, KCORE_TEXT);
416 #endif
417
418         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
419                "%ldk reserved, %ldk data, %ldk init, %ldk highmem)\n",
420                nr_free_pages() << (PAGE_SHIFT-10),
421                ram << (PAGE_SHIFT-10),
422                codesize >> 10,
423                reservedpages << (PAGE_SHIFT-10),
424                datasize >> 10,
425                initsize >> 10,
426                totalhigh_pages << (PAGE_SHIFT-10));
427 }
428 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
429
430 void free_init_pages(const char *what, unsigned long begin, unsigned long end)
431 {
432         unsigned long pfn;
433
434         for (pfn = PFN_UP(begin); pfn < PFN_DOWN(end); pfn++) {
435                 struct page *page = pfn_to_page(pfn);
436                 void *addr = phys_to_virt(PFN_PHYS(pfn));
437
438                 ClearPageReserved(page);
439                 init_page_count(page);
440                 memset(addr, POISON_FREE_INITMEM, PAGE_SIZE);
441                 __free_page(page);
442                 totalram_pages++;
443         }
444         printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10);
445 }
446
447 #ifdef CONFIG_BLK_DEV_INITRD
448 void free_initrd_mem(unsigned long start, unsigned long end)
449 {
450         free_init_pages("initrd memory",
451                         virt_to_phys((void *)start),
452                         virt_to_phys((void *)end));
453 }
454 #endif
455
456 void __init_refok free_initmem(void)
457 {
458         prom_free_prom_memory();
459         free_init_pages("unused kernel memory",
460                         __pa_symbol(&__init_begin),
461                         __pa_symbol(&__init_end));
462 }
463
464 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
465 unsigned long pgd_current[NR_CPUS];
466 #endif
467 /*
468  * On 64-bit we've got three-level pagetables with a slightly
469  * different layout ...
470  */
471 #define __page_aligned(order) __attribute__((__aligned__(PAGE_SIZE<<order)))
472
473 /*
474  * gcc 3.3 and older have trouble determining that PTRS_PER_PGD and PGD_ORDER
475  * are constants.  So we use the variants from asm-offset.h until that gcc
476  * will officially be retired.
477  */
478 pgd_t swapper_pg_dir[_PTRS_PER_PGD] __page_aligned(_PGD_ORDER);
479 #ifndef __PAGETABLE_PMD_FOLDED
480 pmd_t invalid_pmd_table[PTRS_PER_PMD] __page_aligned(PMD_ORDER);
481 #endif
482 pte_t invalid_pte_table[PTRS_PER_PTE] __page_aligned(PTE_ORDER);