[WATCHDOG] Merge code clean-up's from Alan Cox.
[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/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/kernel.h>
17 #include <linux/errno.h>
18 #include <linux/string.h>
19 #include <linux/types.h>
20 #include <linux/pagemap.h>
21 #include <linux/ptrace.h>
22 #include <linux/mman.h>
23 #include <linux/mm.h>
24 #include <linux/bootmem.h>
25 #include <linux/highmem.h>
26 #include <linux/swap.h>
27 #include <linux/proc_fs.h>
28 #include <linux/pfn.h>
29
30 #include <asm/asm-offsets.h>
31 #include <asm/bootinfo.h>
32 #include <asm/cachectl.h>
33 #include <asm/cpu.h>
34 #include <asm/dma.h>
35 #include <asm/kmap_types.h>
36 #include <asm/mmu_context.h>
37 #include <asm/sections.h>
38 #include <asm/pgtable.h>
39 #include <asm/pgalloc.h>
40 #include <asm/tlb.h>
41 #include <asm/fixmap.h>
42
43 /* Atomicity and interruptability */
44 #ifdef CONFIG_MIPS_MT_SMTC
45
46 #include <asm/mipsmtregs.h>
47
48 #define ENTER_CRITICAL(flags) \
49         { \
50         unsigned int mvpflags; \
51         local_irq_save(flags);\
52         mvpflags = dvpe()
53 #define EXIT_CRITICAL(flags) \
54         evpe(mvpflags); \
55         local_irq_restore(flags); \
56         }
57 #else
58
59 #define ENTER_CRITICAL(flags) local_irq_save(flags)
60 #define EXIT_CRITICAL(flags) local_irq_restore(flags)
61
62 #endif /* CONFIG_MIPS_MT_SMTC */
63
64 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
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 EXPORT_SYMBOL_GPL(empty_zero_page);
75
76 /*
77  * Not static inline because used by IP27 special magic initialization code
78  */
79 unsigned long setup_zero_pages(void)
80 {
81         unsigned int order;
82         unsigned long size;
83         struct page *page;
84
85         if (cpu_has_vce)
86                 order = 3;
87         else
88                 order = 0;
89
90         empty_zero_page = __get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
91         if (!empty_zero_page)
92                 panic("Oh boy, that early out of memory?");
93
94         page = virt_to_page((void *)empty_zero_page);
95         split_page(page, order);
96         while (page < virt_to_page((void *)(empty_zero_page + (PAGE_SIZE << order)))) {
97                 SetPageReserved(page);
98                 page++;
99         }
100
101         size = PAGE_SIZE << order;
102         zero_page_mask = (size - 1) & PAGE_MASK;
103
104         return 1UL << order;
105 }
106
107 /*
108  * These are almost like kmap_atomic / kunmap_atmic except they take an
109  * additional address argument as the hint.
110  */
111
112 #define kmap_get_fixmap_pte(vaddr)                                      \
113         pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), (vaddr)), (vaddr)), (vaddr))
114
115 #ifdef CONFIG_MIPS_MT_SMTC
116 static pte_t *kmap_coherent_pte;
117 static void __init kmap_coherent_init(void)
118 {
119         unsigned long vaddr;
120
121         /* cache the first coherent kmap pte */
122         vaddr = __fix_to_virt(FIX_CMAP_BEGIN);
123         kmap_coherent_pte = kmap_get_fixmap_pte(vaddr);
124 }
125 #else
126 static inline void kmap_coherent_init(void) {}
127 #endif
128
129 void *kmap_coherent(struct page *page, unsigned long addr)
130 {
131         enum fixed_addresses idx;
132         unsigned long vaddr, flags, entrylo;
133         unsigned long old_ctx;
134         pte_t pte;
135         int tlbidx;
136
137         BUG_ON(Page_dcache_dirty(page));
138
139         inc_preempt_count();
140         idx = (addr >> PAGE_SHIFT) & (FIX_N_COLOURS - 1);
141 #ifdef CONFIG_MIPS_MT_SMTC
142         idx += FIX_N_COLOURS * smp_processor_id();
143 #endif
144         vaddr = __fix_to_virt(FIX_CMAP_END - idx);
145         pte = mk_pte(page, PAGE_KERNEL);
146 #if defined(CONFIG_64BIT_PHYS_ADDR) && defined(CONFIG_CPU_MIPS32)
147         entrylo = pte.pte_high;
148 #else
149         entrylo = pte_val(pte) >> 6;
150 #endif
151
152         ENTER_CRITICAL(flags);
153         old_ctx = read_c0_entryhi();
154         write_c0_entryhi(vaddr & (PAGE_MASK << 1));
155         write_c0_entrylo0(entrylo);
156         write_c0_entrylo1(entrylo);
157 #ifdef CONFIG_MIPS_MT_SMTC
158         set_pte(kmap_coherent_pte - (FIX_CMAP_END - idx), pte);
159         /* preload TLB instead of local_flush_tlb_one() */
160         mtc0_tlbw_hazard();
161         tlb_probe();
162         tlb_probe_hazard();
163         tlbidx = read_c0_index();
164         mtc0_tlbw_hazard();
165         if (tlbidx < 0)
166                 tlb_write_random();
167         else
168                 tlb_write_indexed();
169 #else
170         tlbidx = read_c0_wired();
171         write_c0_wired(tlbidx + 1);
172         write_c0_index(tlbidx);
173         mtc0_tlbw_hazard();
174         tlb_write_indexed();
175 #endif
176         tlbw_use_hazard();
177         write_c0_entryhi(old_ctx);
178         EXIT_CRITICAL(flags);
179
180         return (void*) vaddr;
181 }
182
183 #define UNIQUE_ENTRYHI(idx) (CKSEG0 + ((idx) << (PAGE_SHIFT + 1)))
184
185 void kunmap_coherent(void)
186 {
187 #ifndef CONFIG_MIPS_MT_SMTC
188         unsigned int wired;
189         unsigned long flags, old_ctx;
190
191         ENTER_CRITICAL(flags);
192         old_ctx = read_c0_entryhi();
193         wired = read_c0_wired() - 1;
194         write_c0_wired(wired);
195         write_c0_index(wired);
196         write_c0_entryhi(UNIQUE_ENTRYHI(wired));
197         write_c0_entrylo0(0);
198         write_c0_entrylo1(0);
199         mtc0_tlbw_hazard();
200         tlb_write_indexed();
201         tlbw_use_hazard();
202         write_c0_entryhi(old_ctx);
203         EXIT_CRITICAL(flags);
204 #endif
205         dec_preempt_count();
206         preempt_check_resched();
207 }
208
209 void copy_user_highpage(struct page *to, struct page *from,
210         unsigned long vaddr, struct vm_area_struct *vma)
211 {
212         void *vfrom, *vto;
213
214         vto = kmap_atomic(to, KM_USER1);
215         if (cpu_has_dc_aliases &&
216             page_mapped(from) && !Page_dcache_dirty(from)) {
217                 vfrom = kmap_coherent(from, vaddr);
218                 copy_page(vto, vfrom);
219                 kunmap_coherent();
220         } else {
221                 vfrom = kmap_atomic(from, KM_USER0);
222                 copy_page(vto, vfrom);
223                 kunmap_atomic(vfrom, KM_USER0);
224         }
225         if ((!cpu_has_ic_fills_f_dc) ||
226             pages_do_alias((unsigned long)vto, vaddr & PAGE_MASK))
227                 flush_data_cache_page((unsigned long)vto);
228         kunmap_atomic(vto, KM_USER1);
229         /* Make sure this page is cleared on other CPU's too before using it */
230         smp_wmb();
231 }
232
233 void copy_to_user_page(struct vm_area_struct *vma,
234         struct page *page, unsigned long vaddr, void *dst, const void *src,
235         unsigned long len)
236 {
237         if (cpu_has_dc_aliases &&
238             page_mapped(page) && !Page_dcache_dirty(page)) {
239                 void *vto = kmap_coherent(page, vaddr) + (vaddr & ~PAGE_MASK);
240                 memcpy(vto, src, len);
241                 kunmap_coherent();
242         } else {
243                 memcpy(dst, src, len);
244                 if (cpu_has_dc_aliases)
245                         SetPageDcacheDirty(page);
246         }
247         if ((vma->vm_flags & VM_EXEC) && !cpu_has_ic_fills_f_dc)
248                 flush_cache_page(vma, vaddr, page_to_pfn(page));
249 }
250
251 void copy_from_user_page(struct vm_area_struct *vma,
252         struct page *page, unsigned long vaddr, void *dst, const void *src,
253         unsigned long len)
254 {
255         if (cpu_has_dc_aliases &&
256             page_mapped(page) && !Page_dcache_dirty(page)) {
257                 void *vfrom = kmap_coherent(page, vaddr) + (vaddr & ~PAGE_MASK);
258                 memcpy(dst, vfrom, len);
259                 kunmap_coherent();
260         } else {
261                 memcpy(dst, src, len);
262                 if (cpu_has_dc_aliases)
263                         SetPageDcacheDirty(page);
264         }
265 }
266
267 #ifdef CONFIG_HIGHMEM
268 unsigned long highstart_pfn, highend_pfn;
269
270 pte_t *kmap_pte;
271 pgprot_t kmap_prot;
272
273 static void __init kmap_init(void)
274 {
275         unsigned long kmap_vstart;
276
277         /* cache the first kmap pte */
278         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
279         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
280
281         kmap_prot = PAGE_KERNEL;
282 }
283 #endif /* CONFIG_HIGHMEM */
284
285 void __init fixrange_init(unsigned long start, unsigned long end,
286         pgd_t *pgd_base)
287 {
288 #if defined(CONFIG_HIGHMEM) || defined(CONFIG_MIPS_MT_SMTC)
289         pgd_t *pgd;
290         pud_t *pud;
291         pmd_t *pmd;
292         pte_t *pte;
293         int i, j, k;
294         unsigned long vaddr;
295
296         vaddr = start;
297         i = __pgd_offset(vaddr);
298         j = __pud_offset(vaddr);
299         k = __pmd_offset(vaddr);
300         pgd = pgd_base + i;
301
302         for ( ; (i < PTRS_PER_PGD) && (vaddr != end); pgd++, i++) {
303                 pud = (pud_t *)pgd;
304                 for ( ; (j < PTRS_PER_PUD) && (vaddr != end); pud++, j++) {
305                         pmd = (pmd_t *)pud;
306                         for (; (k < PTRS_PER_PMD) && (vaddr != end); pmd++, k++) {
307                                 if (pmd_none(*pmd)) {
308                                         pte = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
309                                         set_pmd(pmd, __pmd((unsigned long)pte));
310                                         if (pte != pte_offset_kernel(pmd, 0))
311                                                 BUG();
312                                 }
313                                 vaddr += PMD_SIZE;
314                         }
315                         k = 0;
316                 }
317                 j = 0;
318         }
319 #endif
320 }
321
322 #ifndef CONFIG_NEED_MULTIPLE_NODES
323 static int __init page_is_ram(unsigned long pagenr)
324 {
325         int i;
326
327         for (i = 0; i < boot_mem_map.nr_map; i++) {
328                 unsigned long addr, end;
329
330                 if (boot_mem_map.map[i].type != BOOT_MEM_RAM)
331                         /* not usable memory */
332                         continue;
333
334                 addr = PFN_UP(boot_mem_map.map[i].addr);
335                 end = PFN_DOWN(boot_mem_map.map[i].addr +
336                                boot_mem_map.map[i].size);
337
338                 if (pagenr >= addr && pagenr < end)
339                         return 1;
340         }
341
342         return 0;
343 }
344
345 void __init paging_init(void)
346 {
347         unsigned long max_zone_pfns[MAX_NR_ZONES];
348         unsigned long lastpfn;
349
350         pagetable_init();
351
352 #ifdef CONFIG_HIGHMEM
353         kmap_init();
354 #endif
355         kmap_coherent_init();
356
357 #ifdef CONFIG_ZONE_DMA
358         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
359 #endif
360 #ifdef CONFIG_ZONE_DMA32
361         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
362 #endif
363         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
364         lastpfn = max_low_pfn;
365 #ifdef CONFIG_HIGHMEM
366         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
367         lastpfn = highend_pfn;
368
369         if (cpu_has_dc_aliases && max_low_pfn != highend_pfn) {
370                 printk(KERN_WARNING "This processor doesn't support highmem."
371                        " %ldk highmem ignored\n",
372                        (highend_pfn - max_low_pfn) << (PAGE_SHIFT - 10));
373                 max_zone_pfns[ZONE_HIGHMEM] = max_low_pfn;
374                 lastpfn = max_low_pfn;
375         }
376 #endif
377
378         free_area_init_nodes(max_zone_pfns);
379 }
380
381 static struct kcore_list kcore_mem, kcore_vmalloc;
382 #ifdef CONFIG_64BIT
383 static struct kcore_list kcore_kseg0;
384 #endif
385
386 void __init mem_init(void)
387 {
388         unsigned long codesize, reservedpages, datasize, initsize;
389         unsigned long tmp, ram;
390
391 #ifdef CONFIG_HIGHMEM
392 #ifdef CONFIG_DISCONTIGMEM
393 #error "CONFIG_HIGHMEM and CONFIG_DISCONTIGMEM dont work together yet"
394 #endif
395         max_mapnr = highend_pfn;
396 #else
397         max_mapnr = max_low_pfn;
398 #endif
399         high_memory = (void *) __va(max_low_pfn << PAGE_SHIFT);
400
401         totalram_pages += free_all_bootmem();
402         totalram_pages -= setup_zero_pages();   /* Setup zeroed pages.  */
403
404         reservedpages = ram = 0;
405         for (tmp = 0; tmp < max_low_pfn; tmp++)
406                 if (page_is_ram(tmp)) {
407                         ram++;
408                         if (PageReserved(pfn_to_page(tmp)))
409                                 reservedpages++;
410                 }
411         num_physpages = ram;
412
413 #ifdef CONFIG_HIGHMEM
414         for (tmp = highstart_pfn; tmp < highend_pfn; tmp++) {
415                 struct page *page = pfn_to_page(tmp);
416
417                 if (!page_is_ram(tmp)) {
418                         SetPageReserved(page);
419                         continue;
420                 }
421                 ClearPageReserved(page);
422                 init_page_count(page);
423                 __free_page(page);
424                 totalhigh_pages++;
425         }
426         totalram_pages += totalhigh_pages;
427         num_physpages += totalhigh_pages;
428 #endif
429
430         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
431         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
432         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
433
434 #ifdef CONFIG_64BIT
435         if ((unsigned long) &_text > (unsigned long) CKSEG0)
436                 /* The -4 is a hack so that user tools don't have to handle
437                    the overflow.  */
438                 kclist_add(&kcore_kseg0, (void *) CKSEG0, 0x80000000 - 4);
439 #endif
440         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
441         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
442                    VMALLOC_END-VMALLOC_START);
443
444         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
445                "%ldk reserved, %ldk data, %ldk init, %ldk highmem)\n",
446                (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
447                ram << (PAGE_SHIFT-10),
448                codesize >> 10,
449                reservedpages << (PAGE_SHIFT-10),
450                datasize >> 10,
451                initsize >> 10,
452                (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10)));
453 }
454 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
455
456 void free_init_pages(const char *what, unsigned long begin, unsigned long end)
457 {
458         unsigned long pfn;
459
460         for (pfn = PFN_UP(begin); pfn < PFN_DOWN(end); pfn++) {
461                 struct page *page = pfn_to_page(pfn);
462                 void *addr = phys_to_virt(PFN_PHYS(pfn));
463
464                 ClearPageReserved(page);
465                 init_page_count(page);
466                 memset(addr, POISON_FREE_INITMEM, PAGE_SIZE);
467                 __free_page(page);
468                 totalram_pages++;
469         }
470         printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10);
471 }
472
473 #ifdef CONFIG_BLK_DEV_INITRD
474 void free_initrd_mem(unsigned long start, unsigned long end)
475 {
476         free_init_pages("initrd memory",
477                         virt_to_phys((void *)start),
478                         virt_to_phys((void *)end));
479 }
480 #endif
481
482 void __init_refok free_initmem(void)
483 {
484         prom_free_prom_memory();
485         free_init_pages("unused kernel memory",
486                         __pa_symbol(&__init_begin),
487                         __pa_symbol(&__init_end));
488 }
489
490 unsigned long pgd_current[NR_CPUS];
491 /*
492  * On 64-bit we've got three-level pagetables with a slightly
493  * different layout ...
494  */
495 #define __page_aligned(order) __attribute__((__aligned__(PAGE_SIZE<<order)))
496
497 /*
498  * gcc 3.3 and older have trouble determining that PTRS_PER_PGD and PGD_ORDER
499  * are constants.  So we use the variants from asm-offset.h until that gcc
500  * will officially be retired.
501  */
502 pgd_t swapper_pg_dir[_PTRS_PER_PGD] __page_aligned(_PGD_ORDER);
503 #ifdef CONFIG_64BIT
504 #ifdef MODULE_START
505 pgd_t module_pg_dir[PTRS_PER_PGD] __page_aligned(PGD_ORDER);
506 #endif
507 pmd_t invalid_pmd_table[PTRS_PER_PMD] __page_aligned(PMD_ORDER);
508 #endif
509 pte_t invalid_pte_table[PTRS_PER_PTE] __page_aligned(PTE_ORDER);