Merge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
[sfrench/cifs-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/export.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 #include <linux/kcore.h>
33 #include <linux/export.h>
34 #include <linux/initrd.h>
35
36 #include <asm/asm-offsets.h>
37 #include <asm/bootinfo.h>
38 #include <asm/cachectl.h>
39 #include <asm/cpu.h>
40 #include <asm/dma.h>
41 #include <asm/kmap_types.h>
42 #include <asm/maar.h>
43 #include <asm/mmu_context.h>
44 #include <asm/sections.h>
45 #include <asm/pgtable.h>
46 #include <asm/pgalloc.h>
47 #include <asm/tlb.h>
48 #include <asm/fixmap.h>
49 #include <asm/maar.h>
50
51 /*
52  * We have up to 8 empty zeroed pages so we can map one of the right colour
53  * when needed.  This is necessary only on R4000 / R4400 SC and MC versions
54  * where we have to avoid VCED / VECI exceptions for good performance at
55  * any price.  Since page is never written to after the initialization we
56  * don't have to care about aliases on other CPUs.
57  */
58 unsigned long empty_zero_page, zero_page_mask;
59 EXPORT_SYMBOL_GPL(empty_zero_page);
60 EXPORT_SYMBOL(zero_page_mask);
61
62 /*
63  * Not static inline because used by IP27 special magic initialization code
64  */
65 void setup_zero_pages(void)
66 {
67         unsigned int order, i;
68         struct page *page;
69
70         if (cpu_has_vce)
71                 order = 3;
72         else
73                 order = 0;
74
75         empty_zero_page = __get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
76         if (!empty_zero_page)
77                 panic("Oh boy, that early out of memory?");
78
79         page = virt_to_page((void *)empty_zero_page);
80         split_page(page, order);
81         for (i = 0; i < (1 << order); i++, page++)
82                 mark_page_reserved(page);
83
84         zero_page_mask = ((PAGE_SIZE << order) - 1) & PAGE_MASK;
85 }
86
87 static void *__kmap_pgprot(struct page *page, unsigned long addr, pgprot_t prot)
88 {
89         enum fixed_addresses idx;
90         unsigned long vaddr, flags, entrylo;
91         unsigned long old_ctx;
92         pte_t pte;
93         int tlbidx;
94
95         BUG_ON(Page_dcache_dirty(page));
96
97         preempt_disable();
98         pagefault_disable();
99         idx = (addr >> PAGE_SHIFT) & (FIX_N_COLOURS - 1);
100         idx += in_interrupt() ? FIX_N_COLOURS : 0;
101         vaddr = __fix_to_virt(FIX_CMAP_END - idx);
102         pte = mk_pte(page, prot);
103 #if defined(CONFIG_XPA)
104         entrylo = pte_to_entrylo(pte.pte_high);
105 #elif defined(CONFIG_PHYS_ADDR_T_64BIT) && defined(CONFIG_CPU_MIPS32)
106         entrylo = pte.pte_high;
107 #else
108         entrylo = pte_to_entrylo(pte_val(pte));
109 #endif
110
111         local_irq_save(flags);
112         old_ctx = read_c0_entryhi();
113         write_c0_entryhi(vaddr & (PAGE_MASK << 1));
114         write_c0_entrylo0(entrylo);
115         write_c0_entrylo1(entrylo);
116 #ifdef CONFIG_XPA
117         if (cpu_has_xpa) {
118                 entrylo = (pte.pte_low & _PFNX_MASK);
119                 writex_c0_entrylo0(entrylo);
120                 writex_c0_entrylo1(entrylo);
121         }
122 #endif
123         tlbidx = num_wired_entries();
124         write_c0_wired(tlbidx + 1);
125         write_c0_index(tlbidx);
126         mtc0_tlbw_hazard();
127         tlb_write_indexed();
128         tlbw_use_hazard();
129         write_c0_entryhi(old_ctx);
130         local_irq_restore(flags);
131
132         return (void*) vaddr;
133 }
134
135 void *kmap_coherent(struct page *page, unsigned long addr)
136 {
137         return __kmap_pgprot(page, addr, PAGE_KERNEL);
138 }
139
140 void *kmap_noncoherent(struct page *page, unsigned long addr)
141 {
142         return __kmap_pgprot(page, addr, PAGE_KERNEL_NC);
143 }
144
145 void kunmap_coherent(void)
146 {
147         unsigned int wired;
148         unsigned long flags, old_ctx;
149
150         local_irq_save(flags);
151         old_ctx = read_c0_entryhi();
152         wired = num_wired_entries() - 1;
153         write_c0_wired(wired);
154         write_c0_index(wired);
155         write_c0_entryhi(UNIQUE_ENTRYHI(wired));
156         write_c0_entrylo0(0);
157         write_c0_entrylo1(0);
158         mtc0_tlbw_hazard();
159         tlb_write_indexed();
160         tlbw_use_hazard();
161         write_c0_entryhi(old_ctx);
162         local_irq_restore(flags);
163         pagefault_enable();
164         preempt_enable();
165 }
166
167 void copy_user_highpage(struct page *to, struct page *from,
168         unsigned long vaddr, struct vm_area_struct *vma)
169 {
170         void *vfrom, *vto;
171
172         vto = kmap_atomic(to);
173         if (cpu_has_dc_aliases &&
174             page_mapcount(from) && !Page_dcache_dirty(from)) {
175                 vfrom = kmap_coherent(from, vaddr);
176                 copy_page(vto, vfrom);
177                 kunmap_coherent();
178         } else {
179                 vfrom = kmap_atomic(from);
180                 copy_page(vto, vfrom);
181                 kunmap_atomic(vfrom);
182         }
183         if ((!cpu_has_ic_fills_f_dc) ||
184             pages_do_alias((unsigned long)vto, vaddr & PAGE_MASK))
185                 flush_data_cache_page((unsigned long)vto);
186         kunmap_atomic(vto);
187         /* Make sure this page is cleared on other CPU's too before using it */
188         smp_wmb();
189 }
190
191 void copy_to_user_page(struct vm_area_struct *vma,
192         struct page *page, unsigned long vaddr, void *dst, const void *src,
193         unsigned long len)
194 {
195         if (cpu_has_dc_aliases &&
196             page_mapcount(page) && !Page_dcache_dirty(page)) {
197                 void *vto = kmap_coherent(page, vaddr) + (vaddr & ~PAGE_MASK);
198                 memcpy(vto, src, len);
199                 kunmap_coherent();
200         } else {
201                 memcpy(dst, src, len);
202                 if (cpu_has_dc_aliases)
203                         SetPageDcacheDirty(page);
204         }
205         if (vma->vm_flags & VM_EXEC)
206                 flush_cache_page(vma, vaddr, page_to_pfn(page));
207 }
208
209 void copy_from_user_page(struct vm_area_struct *vma,
210         struct page *page, unsigned long vaddr, void *dst, const void *src,
211         unsigned long len)
212 {
213         if (cpu_has_dc_aliases &&
214             page_mapcount(page) && !Page_dcache_dirty(page)) {
215                 void *vfrom = kmap_coherent(page, vaddr) + (vaddr & ~PAGE_MASK);
216                 memcpy(dst, vfrom, len);
217                 kunmap_coherent();
218         } else {
219                 memcpy(dst, src, len);
220                 if (cpu_has_dc_aliases)
221                         SetPageDcacheDirty(page);
222         }
223 }
224 EXPORT_SYMBOL_GPL(copy_from_user_page);
225
226 void __init fixrange_init(unsigned long start, unsigned long end,
227         pgd_t *pgd_base)
228 {
229 #ifdef CONFIG_HIGHMEM
230         pgd_t *pgd;
231         pud_t *pud;
232         pmd_t *pmd;
233         pte_t *pte;
234         int i, j, k;
235         unsigned long vaddr;
236
237         vaddr = start;
238         i = __pgd_offset(vaddr);
239         j = __pud_offset(vaddr);
240         k = __pmd_offset(vaddr);
241         pgd = pgd_base + i;
242
243         for ( ; (i < PTRS_PER_PGD) && (vaddr < end); pgd++, i++) {
244                 pud = (pud_t *)pgd;
245                 for ( ; (j < PTRS_PER_PUD) && (vaddr < end); pud++, j++) {
246                         pmd = (pmd_t *)pud;
247                         for (; (k < PTRS_PER_PMD) && (vaddr < end); pmd++, k++) {
248                                 if (pmd_none(*pmd)) {
249                                         pte = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
250                                         set_pmd(pmd, __pmd((unsigned long)pte));
251                                         BUG_ON(pte != pte_offset_kernel(pmd, 0));
252                                 }
253                                 vaddr += PMD_SIZE;
254                         }
255                         k = 0;
256                 }
257                 j = 0;
258         }
259 #endif
260 }
261
262 unsigned __weak platform_maar_init(unsigned num_pairs)
263 {
264         struct maar_config cfg[BOOT_MEM_MAP_MAX];
265         unsigned i, num_configured, num_cfg = 0;
266
267         for (i = 0; i < boot_mem_map.nr_map; i++) {
268                 switch (boot_mem_map.map[i].type) {
269                 case BOOT_MEM_RAM:
270                 case BOOT_MEM_INIT_RAM:
271                         break;
272                 default:
273                         continue;
274                 }
275
276                 /* Round lower up */
277                 cfg[num_cfg].lower = boot_mem_map.map[i].addr;
278                 cfg[num_cfg].lower = (cfg[num_cfg].lower + 0xffff) & ~0xffff;
279
280                 /* Round upper down */
281                 cfg[num_cfg].upper = boot_mem_map.map[i].addr +
282                                         boot_mem_map.map[i].size;
283                 cfg[num_cfg].upper = (cfg[num_cfg].upper & ~0xffff) - 1;
284
285                 cfg[num_cfg].attrs = MIPS_MAAR_S;
286                 num_cfg++;
287         }
288
289         num_configured = maar_config(cfg, num_cfg, num_pairs);
290         if (num_configured < num_cfg)
291                 pr_warn("Not enough MAAR pairs (%u) for all bootmem regions (%u)\n",
292                         num_pairs, num_cfg);
293
294         return num_configured;
295 }
296
297 void maar_init(void)
298 {
299         unsigned num_maars, used, i;
300         phys_addr_t lower, upper, attr;
301         static struct {
302                 struct maar_config cfgs[3];
303                 unsigned used;
304         } recorded = { { { 0 } }, 0 };
305
306         if (!cpu_has_maar)
307                 return;
308
309         /* Detect the number of MAARs */
310         write_c0_maari(~0);
311         back_to_back_c0_hazard();
312         num_maars = read_c0_maari() + 1;
313
314         /* MAARs should be in pairs */
315         WARN_ON(num_maars % 2);
316
317         /* Set MAARs using values we recorded already */
318         if (recorded.used) {
319                 used = maar_config(recorded.cfgs, recorded.used, num_maars / 2);
320                 BUG_ON(used != recorded.used);
321         } else {
322                 /* Configure the required MAARs */
323                 used = platform_maar_init(num_maars / 2);
324         }
325
326         /* Disable any further MAARs */
327         for (i = (used * 2); i < num_maars; i++) {
328                 write_c0_maari(i);
329                 back_to_back_c0_hazard();
330                 write_c0_maar(0);
331                 back_to_back_c0_hazard();
332         }
333
334         if (recorded.used)
335                 return;
336
337         pr_info("MAAR configuration:\n");
338         for (i = 0; i < num_maars; i += 2) {
339                 write_c0_maari(i);
340                 back_to_back_c0_hazard();
341                 upper = read_c0_maar();
342
343                 write_c0_maari(i + 1);
344                 back_to_back_c0_hazard();
345                 lower = read_c0_maar();
346
347                 attr = lower & upper;
348                 lower = (lower & MIPS_MAAR_ADDR) << 4;
349                 upper = ((upper & MIPS_MAAR_ADDR) << 4) | 0xffff;
350
351                 pr_info("  [%d]: ", i / 2);
352                 if (!(attr & MIPS_MAAR_VL)) {
353                         pr_cont("disabled\n");
354                         continue;
355                 }
356
357                 pr_cont("%pa-%pa", &lower, &upper);
358
359                 if (attr & MIPS_MAAR_S)
360                         pr_cont(" speculate");
361
362                 pr_cont("\n");
363
364                 /* Record the setup for use on secondary CPUs */
365                 if (used <= ARRAY_SIZE(recorded.cfgs)) {
366                         recorded.cfgs[recorded.used].lower = lower;
367                         recorded.cfgs[recorded.used].upper = upper;
368                         recorded.cfgs[recorded.used].attrs = attr;
369                         recorded.used++;
370                 }
371         }
372 }
373
374 #ifndef CONFIG_NEED_MULTIPLE_NODES
375 int page_is_ram(unsigned long pagenr)
376 {
377         int i;
378
379         for (i = 0; i < boot_mem_map.nr_map; i++) {
380                 unsigned long addr, end;
381
382                 switch (boot_mem_map.map[i].type) {
383                 case BOOT_MEM_RAM:
384                 case BOOT_MEM_INIT_RAM:
385                         break;
386                 default:
387                         /* not usable memory */
388                         continue;
389                 }
390
391                 addr = PFN_UP(boot_mem_map.map[i].addr);
392                 end = PFN_DOWN(boot_mem_map.map[i].addr +
393                                boot_mem_map.map[i].size);
394
395                 if (pagenr >= addr && pagenr < end)
396                         return 1;
397         }
398
399         return 0;
400 }
401
402 void __init paging_init(void)
403 {
404         unsigned long max_zone_pfns[MAX_NR_ZONES];
405         unsigned long lastpfn __maybe_unused;
406
407         pagetable_init();
408
409 #ifdef CONFIG_HIGHMEM
410         kmap_init();
411 #endif
412 #ifdef CONFIG_ZONE_DMA
413         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
414 #endif
415 #ifdef CONFIG_ZONE_DMA32
416         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
417 #endif
418         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
419         lastpfn = max_low_pfn;
420 #ifdef CONFIG_HIGHMEM
421         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
422         lastpfn = highend_pfn;
423
424         if (cpu_has_dc_aliases && max_low_pfn != highend_pfn) {
425                 printk(KERN_WARNING "This processor doesn't support highmem."
426                        " %ldk highmem ignored\n",
427                        (highend_pfn - max_low_pfn) << (PAGE_SHIFT - 10));
428                 max_zone_pfns[ZONE_HIGHMEM] = max_low_pfn;
429                 lastpfn = max_low_pfn;
430         }
431 #endif
432
433         free_area_init_nodes(max_zone_pfns);
434 }
435
436 #ifdef CONFIG_64BIT
437 static struct kcore_list kcore_kseg0;
438 #endif
439
440 static inline void mem_init_free_highmem(void)
441 {
442 #ifdef CONFIG_HIGHMEM
443         unsigned long tmp;
444
445         if (cpu_has_dc_aliases)
446                 return;
447
448         for (tmp = highstart_pfn; tmp < highend_pfn; tmp++) {
449                 struct page *page = pfn_to_page(tmp);
450
451                 if (!page_is_ram(tmp))
452                         SetPageReserved(page);
453                 else
454                         free_highmem_page(page);
455         }
456 #endif
457 }
458
459 void __init mem_init(void)
460 {
461 #ifdef CONFIG_HIGHMEM
462 #ifdef CONFIG_DISCONTIGMEM
463 #error "CONFIG_HIGHMEM and CONFIG_DISCONTIGMEM dont work together yet"
464 #endif
465         max_mapnr = highend_pfn ? highend_pfn : max_low_pfn;
466 #else
467         max_mapnr = max_low_pfn;
468 #endif
469         high_memory = (void *) __va(max_low_pfn << PAGE_SHIFT);
470
471         maar_init();
472         free_all_bootmem();
473         setup_zero_pages();     /* Setup zeroed pages.  */
474         mem_init_free_highmem();
475         mem_init_print_info(NULL);
476
477 #ifdef CONFIG_64BIT
478         if ((unsigned long) &_text > (unsigned long) CKSEG0)
479                 /* The -4 is a hack so that user tools don't have to handle
480                    the overflow.  */
481                 kclist_add(&kcore_kseg0, (void *) CKSEG0,
482                                 0x80000000 - 4, KCORE_TEXT);
483 #endif
484 }
485 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
486
487 void free_init_pages(const char *what, unsigned long begin, unsigned long end)
488 {
489         unsigned long pfn;
490
491         for (pfn = PFN_UP(begin); pfn < PFN_DOWN(end); pfn++) {
492                 struct page *page = pfn_to_page(pfn);
493                 void *addr = phys_to_virt(PFN_PHYS(pfn));
494
495                 memset(addr, POISON_FREE_INITMEM, PAGE_SIZE);
496                 free_reserved_page(page);
497         }
498         printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10);
499 }
500
501 #ifdef CONFIG_BLK_DEV_INITRD
502 void free_initrd_mem(unsigned long start, unsigned long end)
503 {
504         free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM,
505                            "initrd");
506 }
507 #endif
508
509 void (*free_init_pages_eva)(void *begin, void *end) = NULL;
510
511 void __ref free_initmem(void)
512 {
513         prom_free_prom_memory();
514         /*
515          * Let the platform define a specific function to free the
516          * init section since EVA may have used any possible mapping
517          * between virtual and physical addresses.
518          */
519         if (free_init_pages_eva)
520                 free_init_pages_eva((void *)&__init_begin, (void *)&__init_end);
521         else
522                 free_initmem_default(POISON_FREE_INITMEM);
523 }
524
525 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
526 unsigned long pgd_current[NR_CPUS];
527 #endif
528
529 /*
530  * gcc 3.3 and older have trouble determining that PTRS_PER_PGD and PGD_ORDER
531  * are constants.  So we use the variants from asm-offset.h until that gcc
532  * will officially be retired.
533  *
534  * Align swapper_pg_dir in to 64K, allows its address to be loaded
535  * with a single LUI instruction in the TLB handlers.  If we used
536  * __aligned(64K), its size would get rounded up to the alignment
537  * size, and waste space.  So we place it in its own section and align
538  * it in the linker script.
539  */
540 pgd_t swapper_pg_dir[_PTRS_PER_PGD] __section(.bss..swapper_pg_dir);
541 #ifndef __PAGETABLE_PUD_FOLDED
542 pud_t invalid_pud_table[PTRS_PER_PUD] __page_aligned_bss;
543 #endif
544 #ifndef __PAGETABLE_PMD_FOLDED
545 pmd_t invalid_pmd_table[PTRS_PER_PMD] __page_aligned_bss;
546 EXPORT_SYMBOL_GPL(invalid_pmd_table);
547 #endif
548 pte_t invalid_pte_table[PTRS_PER_PTE] __page_aligned_bss;
549 EXPORT_SYMBOL(invalid_pte_table);