Merge master.kernel.org:/pub/scm/linux/kernel/git/paulus/powerpc
[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 unsigned long highstart_pfn, highend_pfn;
65
66 /*
67  * We have up to 8 empty zeroed pages so we can map one of the right colour
68  * when needed.  This is necessary only on R4000 / R4400 SC and MC versions
69  * where we have to avoid VCED / VECI exceptions for good performance at
70  * any price.  Since page is never written to after the initialization we
71  * don't have to care about aliases on other CPUs.
72  */
73 unsigned long empty_zero_page, zero_page_mask;
74
75 /*
76  * Not static inline because used by IP27 special magic initialization code
77  */
78 unsigned long setup_zero_pages(void)
79 {
80         unsigned int order;
81         unsigned long size;
82         struct page *page;
83
84         if (cpu_has_vce)
85                 order = 3;
86         else
87                 order = 0;
88
89         empty_zero_page = __get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
90         if (!empty_zero_page)
91                 panic("Oh boy, that early out of memory?");
92
93         page = virt_to_page((void *)empty_zero_page);
94         split_page(page, order);
95         while (page < virt_to_page((void *)(empty_zero_page + (PAGE_SIZE << order)))) {
96                 SetPageReserved(page);
97                 page++;
98         }
99
100         size = PAGE_SIZE << order;
101         zero_page_mask = (size - 1) & PAGE_MASK;
102
103         return 1UL << order;
104 }
105
106 /*
107  * These are almost like kmap_atomic / kunmap_atmic except they take an
108  * additional address argument as the hint.
109  */
110
111 #define kmap_get_fixmap_pte(vaddr)                                      \
112         pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), (vaddr)), (vaddr)), (vaddr))
113
114 #ifdef CONFIG_MIPS_MT_SMTC
115 static pte_t *kmap_coherent_pte;
116 static void __init kmap_coherent_init(void)
117 {
118         unsigned long vaddr;
119
120         /* cache the first coherent kmap pte */
121         vaddr = __fix_to_virt(FIX_CMAP_BEGIN);
122         kmap_coherent_pte = kmap_get_fixmap_pte(vaddr);
123 }
124 #else
125 static inline void kmap_coherent_init(void) {}
126 #endif
127
128 static inline void *kmap_coherent(struct page *page, unsigned long addr)
129 {
130         enum fixed_addresses idx;
131         unsigned long vaddr, flags, entrylo;
132         unsigned long old_ctx;
133         pte_t pte;
134         int tlbidx;
135
136         inc_preempt_count();
137         idx = (addr >> PAGE_SHIFT) & (FIX_N_COLOURS - 1);
138 #ifdef CONFIG_MIPS_MT_SMTC
139         idx += FIX_N_COLOURS * smp_processor_id();
140 #endif
141         vaddr = __fix_to_virt(FIX_CMAP_END - idx);
142         pte = mk_pte(page, PAGE_KERNEL);
143 #if defined(CONFIG_64BIT_PHYS_ADDR) && defined(CONFIG_CPU_MIPS32_R1)
144         entrylo = pte.pte_high;
145 #else
146         entrylo = pte_val(pte) >> 6;
147 #endif
148
149         ENTER_CRITICAL(flags);
150         old_ctx = read_c0_entryhi();
151         write_c0_entryhi(vaddr & (PAGE_MASK << 1));
152         write_c0_entrylo0(entrylo);
153         write_c0_entrylo1(entrylo);
154 #ifdef CONFIG_MIPS_MT_SMTC
155         set_pte(kmap_coherent_pte - (FIX_CMAP_END - idx), pte);
156         /* preload TLB instead of local_flush_tlb_one() */
157         mtc0_tlbw_hazard();
158         tlb_probe();
159         tlb_probe_hazard();
160         tlbidx = read_c0_index();
161         mtc0_tlbw_hazard();
162         if (tlbidx < 0)
163                 tlb_write_random();
164         else
165                 tlb_write_indexed();
166 #else
167         tlbidx = read_c0_wired();
168         write_c0_wired(tlbidx + 1);
169         write_c0_index(tlbidx);
170         mtc0_tlbw_hazard();
171         tlb_write_indexed();
172 #endif
173         tlbw_use_hazard();
174         write_c0_entryhi(old_ctx);
175         EXIT_CRITICAL(flags);
176
177         return (void*) vaddr;
178 }
179
180 #define UNIQUE_ENTRYHI(idx) (CKSEG0 + ((idx) << (PAGE_SHIFT + 1)))
181
182 static inline void kunmap_coherent(struct page *page)
183 {
184 #ifndef CONFIG_MIPS_MT_SMTC
185         unsigned int wired;
186         unsigned long flags, old_ctx;
187
188         ENTER_CRITICAL(flags);
189         old_ctx = read_c0_entryhi();
190         wired = read_c0_wired() - 1;
191         write_c0_wired(wired);
192         write_c0_index(wired);
193         write_c0_entryhi(UNIQUE_ENTRYHI(wired));
194         write_c0_entrylo0(0);
195         write_c0_entrylo1(0);
196         mtc0_tlbw_hazard();
197         tlb_write_indexed();
198         tlbw_use_hazard();
199         write_c0_entryhi(old_ctx);
200         EXIT_CRITICAL(flags);
201 #endif
202         dec_preempt_count();
203         preempt_check_resched();
204 }
205
206 void copy_user_highpage(struct page *to, struct page *from,
207         unsigned long vaddr, struct vm_area_struct *vma)
208 {
209         void *vfrom, *vto;
210
211         vto = kmap_atomic(to, KM_USER1);
212         if (cpu_has_dc_aliases) {
213                 vfrom = kmap_coherent(from, vaddr);
214                 copy_page(vto, vfrom);
215                 kunmap_coherent(from);
216         } else {
217                 vfrom = kmap_atomic(from, KM_USER0);
218                 copy_page(vto, vfrom);
219                 kunmap_atomic(vfrom, KM_USER0);
220         }
221         if (((vma->vm_flags & VM_EXEC) && !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 EXPORT_SYMBOL(copy_user_highpage);
230
231 void copy_to_user_page(struct vm_area_struct *vma,
232         struct page *page, unsigned long vaddr, void *dst, const void *src,
233         unsigned long len)
234 {
235         if (cpu_has_dc_aliases) {
236                 void *vto = kmap_coherent(page, vaddr) + (vaddr & ~PAGE_MASK);
237                 memcpy(vto, src, len);
238                 kunmap_coherent(page);
239         } else
240                 memcpy(dst, src, len);
241         if ((vma->vm_flags & VM_EXEC) && !cpu_has_ic_fills_f_dc)
242                 flush_cache_page(vma, vaddr, page_to_pfn(page));
243 }
244
245 EXPORT_SYMBOL(copy_to_user_page);
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                 void *vfrom =
253                         kmap_coherent(page, vaddr) + (vaddr & ~PAGE_MASK);
254                 memcpy(dst, vfrom, len);
255                 kunmap_coherent(page);
256         } else
257                 memcpy(dst, src, len);
258 }
259
260 EXPORT_SYMBOL(copy_from_user_page);
261
262
263 #ifdef CONFIG_HIGHMEM
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 extern void pagetable_init(void);
318
319 static int __init page_is_ram(unsigned long pagenr)
320 {
321         int i;
322
323         for (i = 0; i < boot_mem_map.nr_map; i++) {
324                 unsigned long addr, end;
325
326                 if (boot_mem_map.map[i].type != BOOT_MEM_RAM)
327                         /* not usable memory */
328                         continue;
329
330                 addr = PFN_UP(boot_mem_map.map[i].addr);
331                 end = PFN_DOWN(boot_mem_map.map[i].addr +
332                                boot_mem_map.map[i].size);
333
334                 if (pagenr >= addr && pagenr < end)
335                         return 1;
336         }
337
338         return 0;
339 }
340
341 void __init paging_init(void)
342 {
343         unsigned long zones_size[MAX_NR_ZONES] = { 0, };
344 #ifndef CONFIG_FLATMEM
345         unsigned long zholes_size[MAX_NR_ZONES] = { 0, };
346         unsigned long i, j, pfn;
347 #endif
348
349         pagetable_init();
350
351 #ifdef CONFIG_HIGHMEM
352         kmap_init();
353 #endif
354         kmap_coherent_init();
355
356 #ifdef CONFIG_ISA
357         if (max_low_pfn >= MAX_DMA_PFN)
358                 if (min_low_pfn >= MAX_DMA_PFN) {
359                         zones_size[ZONE_DMA] = 0;
360                         zones_size[ZONE_NORMAL] = max_low_pfn - min_low_pfn;
361                 } else {
362                         zones_size[ZONE_DMA] = MAX_DMA_PFN - min_low_pfn;
363                         zones_size[ZONE_NORMAL] = max_low_pfn - MAX_DMA_PFN;
364                 }
365         else
366 #endif
367         zones_size[ZONE_DMA] = max_low_pfn - min_low_pfn;
368
369 #ifdef CONFIG_HIGHMEM
370         zones_size[ZONE_HIGHMEM] = highend_pfn - highstart_pfn;
371
372         if (cpu_has_dc_aliases && zones_size[ZONE_HIGHMEM]) {
373                 printk(KERN_WARNING "This processor doesn't support highmem."
374                        " %ldk highmem ignored\n", zones_size[ZONE_HIGHMEM]);
375                 zones_size[ZONE_HIGHMEM] = 0;
376         }
377 #endif
378
379 #ifdef CONFIG_FLATMEM
380         free_area_init(zones_size);
381 #else
382         pfn = 0;
383         for (i = 0; i < MAX_NR_ZONES; i++)
384                 for (j = 0; j < zones_size[i]; j++, pfn++)
385                         if (!page_is_ram(pfn))
386                                 zholes_size[i]++;
387         free_area_init_node(0, NODE_DATA(0), zones_size, 0, zholes_size);
388 #endif
389 }
390
391 static struct kcore_list kcore_mem, kcore_vmalloc;
392 #ifdef CONFIG_64BIT
393 static struct kcore_list kcore_kseg0;
394 #endif
395
396 void __init mem_init(void)
397 {
398         unsigned long codesize, reservedpages, datasize, initsize;
399         unsigned long tmp, ram;
400
401 #ifdef CONFIG_HIGHMEM
402 #ifdef CONFIG_DISCONTIGMEM
403 #error "CONFIG_HIGHMEM and CONFIG_DISCONTIGMEM dont work together yet"
404 #endif
405         max_mapnr = highend_pfn;
406 #else
407         max_mapnr = max_low_pfn;
408 #endif
409         high_memory = (void *) __va(max_low_pfn << PAGE_SHIFT);
410
411         totalram_pages += free_all_bootmem();
412         totalram_pages -= setup_zero_pages();   /* Setup zeroed pages.  */
413
414         reservedpages = ram = 0;
415         for (tmp = 0; tmp < max_low_pfn; tmp++)
416                 if (page_is_ram(tmp)) {
417                         ram++;
418                         if (PageReserved(pfn_to_page(tmp)))
419                                 reservedpages++;
420                 }
421         num_physpages = ram;
422
423 #ifdef CONFIG_HIGHMEM
424         for (tmp = highstart_pfn; tmp < highend_pfn; tmp++) {
425                 struct page *page = mem_map + tmp;
426
427                 if (!page_is_ram(tmp)) {
428                         SetPageReserved(page);
429                         continue;
430                 }
431                 ClearPageReserved(page);
432 #ifdef CONFIG_LIMITED_DMA
433                 set_page_address(page, lowmem_page_address(page));
434 #endif
435                 init_page_count(page);
436                 __free_page(page);
437                 totalhigh_pages++;
438         }
439         totalram_pages += totalhigh_pages;
440         num_physpages += totalhigh_pages;
441 #endif
442
443         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
444         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
445         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
446
447 #ifdef CONFIG_64BIT
448         if ((unsigned long) &_text > (unsigned long) CKSEG0)
449                 /* The -4 is a hack so that user tools don't have to handle
450                    the overflow.  */
451                 kclist_add(&kcore_kseg0, (void *) CKSEG0, 0x80000000 - 4);
452 #endif
453         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
454         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
455                    VMALLOC_END-VMALLOC_START);
456
457         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
458                "%ldk reserved, %ldk data, %ldk init, %ldk highmem)\n",
459                (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
460                ram << (PAGE_SHIFT-10),
461                codesize >> 10,
462                reservedpages << (PAGE_SHIFT-10),
463                datasize >> 10,
464                initsize >> 10,
465                (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10)));
466 }
467 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
468
469 void free_init_pages(const char *what, unsigned long begin, unsigned long end)
470 {
471         unsigned long pfn;
472
473         for (pfn = PFN_UP(begin); pfn < PFN_DOWN(end); pfn++) {
474                 struct page *page = pfn_to_page(pfn);
475                 void *addr = phys_to_virt(PFN_PHYS(pfn));
476
477                 ClearPageReserved(page);
478                 init_page_count(page);
479                 memset(addr, POISON_FREE_INITMEM, PAGE_SIZE);
480                 __free_page(page);
481                 totalram_pages++;
482         }
483         printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10);
484 }
485
486 #ifdef CONFIG_BLK_DEV_INITRD
487 void free_initrd_mem(unsigned long start, unsigned long end)
488 {
489         free_init_pages("initrd memory",
490                         virt_to_phys((void *)start),
491                         virt_to_phys((void *)end));
492 }
493 #endif
494
495 void free_initmem(void)
496 {
497         prom_free_prom_memory();
498         free_init_pages("unused kernel memory",
499                         __pa_symbol(&__init_begin),
500                         __pa_symbol(&__init_end));
501 }
502
503 unsigned long pgd_current[NR_CPUS];
504 /*
505  * On 64-bit we've got three-level pagetables with a slightly
506  * different layout ...
507  */
508 #define __page_aligned(order) __attribute__((__aligned__(PAGE_SIZE<<order)))
509 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned(PGD_ORDER);
510 #ifdef CONFIG_64BIT
511 #ifdef MODULE_START
512 pgd_t module_pg_dir[PTRS_PER_PGD] __page_aligned(PGD_ORDER);
513 #endif
514 pmd_t invalid_pmd_table[PTRS_PER_PMD] __page_aligned(PMD_ORDER);
515 #endif
516 pte_t invalid_pte_table[PTRS_PER_PTE] __page_aligned(PTE_ORDER);