Pull throttle into release branch
[linux-2.6] / 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/init.h>
12 #include <linux/module.h>
13 #include <linux/signal.h>
14 #include <linux/sched.h>
15 #include <linux/kernel.h>
16 #include <linux/errno.h>
17 #include <linux/string.h>
18 #include <linux/types.h>
19 #include <linux/pagemap.h>
20 #include <linux/ptrace.h>
21 #include <linux/mman.h>
22 #include <linux/mm.h>
23 #include <linux/bootmem.h>
24 #include <linux/highmem.h>
25 #include <linux/swap.h>
26 #include <linux/proc_fs.h>
27 #include <linux/pfn.h>
28
29 #include <asm/bootinfo.h>
30 #include <asm/cachectl.h>
31 #include <asm/cpu.h>
32 #include <asm/dma.h>
33 #include <asm/kmap_types.h>
34 #include <asm/mmu_context.h>
35 #include <asm/sections.h>
36 #include <asm/pgtable.h>
37 #include <asm/pgalloc.h>
38 #include <asm/tlb.h>
39 #include <asm/fixmap.h>
40
41 /* Atomicity and interruptability */
42 #ifdef CONFIG_MIPS_MT_SMTC
43
44 #include <asm/mipsmtregs.h>
45
46 #define ENTER_CRITICAL(flags) \
47         { \
48         unsigned int mvpflags; \
49         local_irq_save(flags);\
50         mvpflags = dvpe()
51 #define EXIT_CRITICAL(flags) \
52         evpe(mvpflags); \
53         local_irq_restore(flags); \
54         }
55 #else
56
57 #define ENTER_CRITICAL(flags) local_irq_save(flags)
58 #define EXIT_CRITICAL(flags) local_irq_restore(flags)
59
60 #endif /* CONFIG_MIPS_MT_SMTC */
61
62 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
63
64 /*
65  * We have up to 8 empty zeroed pages so we can map one of the right colour
66  * when needed.  This is necessary only on R4000 / R4400 SC and MC versions
67  * where we have to avoid VCED / VECI exceptions for good performance at
68  * any price.  Since page is never written to after the initialization we
69  * don't have to care about aliases on other CPUs.
70  */
71 unsigned long empty_zero_page, zero_page_mask;
72
73 /*
74  * Not static inline because used by IP27 special magic initialization code
75  */
76 unsigned long setup_zero_pages(void)
77 {
78         unsigned int order;
79         unsigned long size;
80         struct page *page;
81
82         if (cpu_has_vce)
83                 order = 3;
84         else
85                 order = 0;
86
87         empty_zero_page = __get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
88         if (!empty_zero_page)
89                 panic("Oh boy, that early out of memory?");
90
91         page = virt_to_page((void *)empty_zero_page);
92         split_page(page, order);
93         while (page < virt_to_page((void *)(empty_zero_page + (PAGE_SIZE << order)))) {
94                 SetPageReserved(page);
95                 page++;
96         }
97
98         size = PAGE_SIZE << order;
99         zero_page_mask = (size - 1) & PAGE_MASK;
100
101         return 1UL << order;
102 }
103
104 /*
105  * These are almost like kmap_atomic / kunmap_atmic except they take an
106  * additional address argument as the hint.
107  */
108
109 #define kmap_get_fixmap_pte(vaddr)                                      \
110         pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), (vaddr)), (vaddr)), (vaddr))
111
112 #ifdef CONFIG_MIPS_MT_SMTC
113 static pte_t *kmap_coherent_pte;
114 static void __init kmap_coherent_init(void)
115 {
116         unsigned long vaddr;
117
118         /* cache the first coherent kmap pte */
119         vaddr = __fix_to_virt(FIX_CMAP_BEGIN);
120         kmap_coherent_pte = kmap_get_fixmap_pte(vaddr);
121 }
122 #else
123 static inline void kmap_coherent_init(void) {}
124 #endif
125
126 void *kmap_coherent(struct page *page, unsigned long addr)
127 {
128         enum fixed_addresses idx;
129         unsigned long vaddr, flags, entrylo;
130         unsigned long old_ctx;
131         pte_t pte;
132         int tlbidx;
133
134         inc_preempt_count();
135         idx = (addr >> PAGE_SHIFT) & (FIX_N_COLOURS - 1);
136 #ifdef CONFIG_MIPS_MT_SMTC
137         idx += FIX_N_COLOURS * smp_processor_id();
138 #endif
139         vaddr = __fix_to_virt(FIX_CMAP_END - idx);
140         pte = mk_pte(page, PAGE_KERNEL);
141 #if defined(CONFIG_64BIT_PHYS_ADDR) && defined(CONFIG_CPU_MIPS32_R1)
142         entrylo = pte.pte_high;
143 #else
144         entrylo = pte_val(pte) >> 6;
145 #endif
146
147         ENTER_CRITICAL(flags);
148         old_ctx = read_c0_entryhi();
149         write_c0_entryhi(vaddr & (PAGE_MASK << 1));
150         write_c0_entrylo0(entrylo);
151         write_c0_entrylo1(entrylo);
152 #ifdef CONFIG_MIPS_MT_SMTC
153         set_pte(kmap_coherent_pte - (FIX_CMAP_END - idx), pte);
154         /* preload TLB instead of local_flush_tlb_one() */
155         mtc0_tlbw_hazard();
156         tlb_probe();
157         tlb_probe_hazard();
158         tlbidx = read_c0_index();
159         mtc0_tlbw_hazard();
160         if (tlbidx < 0)
161                 tlb_write_random();
162         else
163                 tlb_write_indexed();
164 #else
165         tlbidx = read_c0_wired();
166         write_c0_wired(tlbidx + 1);
167         write_c0_index(tlbidx);
168         mtc0_tlbw_hazard();
169         tlb_write_indexed();
170 #endif
171         tlbw_use_hazard();
172         write_c0_entryhi(old_ctx);
173         EXIT_CRITICAL(flags);
174
175         return (void*) vaddr;
176 }
177
178 #define UNIQUE_ENTRYHI(idx) (CKSEG0 + ((idx) << (PAGE_SHIFT + 1)))
179
180 void kunmap_coherent(void)
181 {
182 #ifndef CONFIG_MIPS_MT_SMTC
183         unsigned int wired;
184         unsigned long flags, old_ctx;
185
186         ENTER_CRITICAL(flags);
187         old_ctx = read_c0_entryhi();
188         wired = read_c0_wired() - 1;
189         write_c0_wired(wired);
190         write_c0_index(wired);
191         write_c0_entryhi(UNIQUE_ENTRYHI(wired));
192         write_c0_entrylo0(0);
193         write_c0_entrylo1(0);
194         mtc0_tlbw_hazard();
195         tlb_write_indexed();
196         tlbw_use_hazard();
197         write_c0_entryhi(old_ctx);
198         EXIT_CRITICAL(flags);
199 #endif
200         dec_preempt_count();
201         preempt_check_resched();
202 }
203
204 void copy_user_highpage(struct page *to, struct page *from,
205         unsigned long vaddr, struct vm_area_struct *vma)
206 {
207         void *vfrom, *vto;
208
209         vto = kmap_atomic(to, KM_USER1);
210         if (cpu_has_dc_aliases) {
211                 vfrom = kmap_coherent(from, vaddr);
212                 copy_page(vto, vfrom);
213                 kunmap_coherent();
214         } else {
215                 vfrom = kmap_atomic(from, KM_USER0);
216                 copy_page(vto, vfrom);
217                 kunmap_atomic(vfrom, KM_USER0);
218         }
219         if (((vma->vm_flags & VM_EXEC) && !cpu_has_ic_fills_f_dc) ||
220             pages_do_alias((unsigned long)vto, vaddr & PAGE_MASK))
221                 flush_data_cache_page((unsigned long)vto);
222         kunmap_atomic(vto, KM_USER1);
223         /* Make sure this page is cleared on other CPU's too before using it */
224         smp_wmb();
225 }
226
227 EXPORT_SYMBOL(copy_user_highpage);
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                 void *vto = kmap_coherent(page, vaddr) + (vaddr & ~PAGE_MASK);
235                 memcpy(vto, src, len);
236                 kunmap_coherent();
237         } else
238                 memcpy(dst, src, len);
239         if ((vma->vm_flags & VM_EXEC) && !cpu_has_ic_fills_f_dc)
240                 flush_cache_page(vma, vaddr, page_to_pfn(page));
241 }
242
243 EXPORT_SYMBOL(copy_to_user_page);
244
245 void copy_from_user_page(struct vm_area_struct *vma,
246         struct page *page, unsigned long vaddr, void *dst, const void *src,
247         unsigned long len)
248 {
249         if (cpu_has_dc_aliases) {
250                 void *vfrom =
251                         kmap_coherent(page, vaddr) + (vaddr & ~PAGE_MASK);
252                 memcpy(dst, vfrom, len);
253                 kunmap_coherent();
254         } else
255                 memcpy(dst, src, len);
256 }
257
258 EXPORT_SYMBOL(copy_from_user_page);
259
260
261 #ifdef CONFIG_HIGHMEM
262 unsigned long highstart_pfn, highend_pfn;
263
264 pte_t *kmap_pte;
265 pgprot_t kmap_prot;
266
267 static void __init kmap_init(void)
268 {
269         unsigned long kmap_vstart;
270
271         /* cache the first kmap pte */
272         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
273         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
274
275         kmap_prot = PAGE_KERNEL;
276 }
277 #endif /* CONFIG_HIGHMEM */
278
279 void __init fixrange_init(unsigned long start, unsigned long end,
280         pgd_t *pgd_base)
281 {
282 #if defined(CONFIG_HIGHMEM) || defined(CONFIG_MIPS_MT_SMTC)
283         pgd_t *pgd;
284         pud_t *pud;
285         pmd_t *pmd;
286         pte_t *pte;
287         int i, j, k;
288         unsigned long vaddr;
289
290         vaddr = start;
291         i = __pgd_offset(vaddr);
292         j = __pud_offset(vaddr);
293         k = __pmd_offset(vaddr);
294         pgd = pgd_base + i;
295
296         for ( ; (i < PTRS_PER_PGD) && (vaddr != end); pgd++, i++) {
297                 pud = (pud_t *)pgd;
298                 for ( ; (j < PTRS_PER_PUD) && (vaddr != end); pud++, j++) {
299                         pmd = (pmd_t *)pud;
300                         for (; (k < PTRS_PER_PMD) && (vaddr != end); pmd++, k++) {
301                                 if (pmd_none(*pmd)) {
302                                         pte = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
303                                         set_pmd(pmd, __pmd((unsigned long)pte));
304                                         if (pte != pte_offset_kernel(pmd, 0))
305                                                 BUG();
306                                 }
307                                 vaddr += PMD_SIZE;
308                         }
309                         k = 0;
310                 }
311                 j = 0;
312         }
313 #endif
314 }
315
316 #ifndef CONFIG_NEED_MULTIPLE_NODES
317 static int __init page_is_ram(unsigned long pagenr)
318 {
319         int i;
320
321         for (i = 0; i < boot_mem_map.nr_map; i++) {
322                 unsigned long addr, end;
323
324                 if (boot_mem_map.map[i].type != BOOT_MEM_RAM)
325                         /* not usable memory */
326                         continue;
327
328                 addr = PFN_UP(boot_mem_map.map[i].addr);
329                 end = PFN_DOWN(boot_mem_map.map[i].addr +
330                                boot_mem_map.map[i].size);
331
332                 if (pagenr >= addr && pagenr < end)
333                         return 1;
334         }
335
336         return 0;
337 }
338
339 void __init paging_init(void)
340 {
341         unsigned long zones_size[MAX_NR_ZONES] = { 0, };
342 #ifndef CONFIG_FLATMEM
343         unsigned long zholes_size[MAX_NR_ZONES] = { 0, };
344         unsigned long i, j, pfn;
345 #endif
346
347         pagetable_init();
348
349 #ifdef CONFIG_HIGHMEM
350         kmap_init();
351 #endif
352         kmap_coherent_init();
353
354 #ifdef CONFIG_ZONE_DMA
355         if (min_low_pfn < MAX_DMA_PFN && MAX_DMA_PFN <= max_low_pfn) {
356                 zones_size[ZONE_DMA] = MAX_DMA_PFN - min_low_pfn;
357                 zones_size[ZONE_NORMAL] = max_low_pfn - MAX_DMA_PFN;
358         } else if (max_low_pfn < MAX_DMA_PFN)
359                 zones_size[ZONE_DMA] = max_low_pfn - min_low_pfn;
360         else
361 #endif
362         zones_size[ZONE_NORMAL] = max_low_pfn - min_low_pfn;
363
364 #ifdef CONFIG_HIGHMEM
365         zones_size[ZONE_HIGHMEM] = highend_pfn - highstart_pfn;
366
367         if (cpu_has_dc_aliases && zones_size[ZONE_HIGHMEM]) {
368                 printk(KERN_WARNING "This processor doesn't support highmem."
369                        " %ldk highmem ignored\n", zones_size[ZONE_HIGHMEM]);
370                 zones_size[ZONE_HIGHMEM] = 0;
371         }
372 #endif
373
374 #ifdef CONFIG_FLATMEM
375         free_area_init(zones_size);
376 #else
377         pfn = min_low_pfn;
378         for (i = 0; i < MAX_NR_ZONES; i++)
379                 for (j = 0; j < zones_size[i]; j++, pfn++)
380                         if (!page_is_ram(pfn))
381                                 zholes_size[i]++;
382         free_area_init_node(0, NODE_DATA(0), zones_size, 0, zholes_size);
383 #endif
384 }
385
386 static struct kcore_list kcore_mem, kcore_vmalloc;
387 #ifdef CONFIG_64BIT
388 static struct kcore_list kcore_kseg0;
389 #endif
390
391 void __init mem_init(void)
392 {
393         unsigned long codesize, reservedpages, datasize, initsize;
394         unsigned long tmp, ram;
395
396 #ifdef CONFIG_HIGHMEM
397 #ifdef CONFIG_DISCONTIGMEM
398 #error "CONFIG_HIGHMEM and CONFIG_DISCONTIGMEM dont work together yet"
399 #endif
400         max_mapnr = highend_pfn;
401 #else
402         max_mapnr = max_low_pfn;
403 #endif
404         high_memory = (void *) __va(max_low_pfn << PAGE_SHIFT);
405
406         totalram_pages += free_all_bootmem();
407         totalram_pages -= setup_zero_pages();   /* Setup zeroed pages.  */
408
409         reservedpages = ram = 0;
410         for (tmp = 0; tmp < max_low_pfn; tmp++)
411                 if (page_is_ram(tmp)) {
412                         ram++;
413                         if (PageReserved(pfn_to_page(tmp)))
414                                 reservedpages++;
415                 }
416         num_physpages = ram;
417
418 #ifdef CONFIG_HIGHMEM
419         for (tmp = highstart_pfn; tmp < highend_pfn; tmp++) {
420                 struct page *page = mem_map + tmp;
421
422                 if (!page_is_ram(tmp)) {
423                         SetPageReserved(page);
424                         continue;
425                 }
426                 ClearPageReserved(page);
427                 init_page_count(page);
428                 __free_page(page);
429                 totalhigh_pages++;
430         }
431         totalram_pages += totalhigh_pages;
432         num_physpages += totalhigh_pages;
433 #endif
434
435         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
436         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
437         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
438
439 #ifdef CONFIG_64BIT
440         if ((unsigned long) &_text > (unsigned long) CKSEG0)
441                 /* The -4 is a hack so that user tools don't have to handle
442                    the overflow.  */
443                 kclist_add(&kcore_kseg0, (void *) CKSEG0, 0x80000000 - 4);
444 #endif
445         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
446         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
447                    VMALLOC_END-VMALLOC_START);
448
449         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
450                "%ldk reserved, %ldk data, %ldk init, %ldk highmem)\n",
451                (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
452                ram << (PAGE_SHIFT-10),
453                codesize >> 10,
454                reservedpages << (PAGE_SHIFT-10),
455                datasize >> 10,
456                initsize >> 10,
457                (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10)));
458 }
459 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
460
461 void free_init_pages(const char *what, unsigned long begin, unsigned long end)
462 {
463         unsigned long pfn;
464
465         for (pfn = PFN_UP(begin); pfn < PFN_DOWN(end); pfn++) {
466                 struct page *page = pfn_to_page(pfn);
467                 void *addr = phys_to_virt(PFN_PHYS(pfn));
468
469                 ClearPageReserved(page);
470                 init_page_count(page);
471                 memset(addr, POISON_FREE_INITMEM, PAGE_SIZE);
472                 __free_page(page);
473                 totalram_pages++;
474         }
475         printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10);
476 }
477
478 #ifdef CONFIG_BLK_DEV_INITRD
479 void free_initrd_mem(unsigned long start, unsigned long end)
480 {
481         free_init_pages("initrd memory",
482                         virt_to_phys((void *)start),
483                         virt_to_phys((void *)end));
484 }
485 #endif
486
487 void free_initmem(void)
488 {
489         prom_free_prom_memory();
490         free_init_pages("unused kernel memory",
491                         __pa_symbol(&__init_begin),
492                         __pa_symbol(&__init_end));
493 }
494
495 unsigned long pgd_current[NR_CPUS];
496 /*
497  * On 64-bit we've got three-level pagetables with a slightly
498  * different layout ...
499  */
500 #define __page_aligned(order) __attribute__((__aligned__(PAGE_SIZE<<order)))
501 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned(PGD_ORDER);
502 #ifdef CONFIG_64BIT
503 #ifdef MODULE_START
504 pgd_t module_pg_dir[PTRS_PER_PGD] __page_aligned(PGD_ORDER);
505 #endif
506 pmd_t invalid_pmd_table[PTRS_PER_PMD] __page_aligned(PMD_ORDER);
507 #endif
508 pte_t invalid_pte_table[PTRS_PER_PTE] __page_aligned(PTE_ORDER);