Merge branch 'i2c-for-linus' of git://jdelvare.pck.nerim.net/jdelvare-2.6
[linux-2.6] / arch / powerpc / mm / mem.c
1 /*
2  *  PowerPC version
3  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
4  *
5  *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
6  *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
7  *    Copyright (C) 1996 Paul Mackerras
8  *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
9  *  PPC44x/36-bit changes by Matt Porter (mporter@mvista.com)
10  *
11  *  Derived from "arch/i386/mm/init.c"
12  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
13  *
14  *  This program is free software; you can redistribute it and/or
15  *  modify it under the terms of the GNU General Public License
16  *  as published by the Free Software Foundation; either version
17  *  2 of the License, or (at your option) any later version.
18  *
19  */
20
21 #include <linux/module.h>
22 #include <linux/sched.h>
23 #include <linux/kernel.h>
24 #include <linux/errno.h>
25 #include <linux/string.h>
26 #include <linux/types.h>
27 #include <linux/mm.h>
28 #include <linux/stddef.h>
29 #include <linux/init.h>
30 #include <linux/bootmem.h>
31 #include <linux/highmem.h>
32 #include <linux/initrd.h>
33 #include <linux/pagemap.h>
34 #include <linux/suspend.h>
35
36 #include <asm/pgalloc.h>
37 #include <asm/prom.h>
38 #include <asm/io.h>
39 #include <asm/mmu_context.h>
40 #include <asm/pgtable.h>
41 #include <asm/mmu.h>
42 #include <asm/smp.h>
43 #include <asm/machdep.h>
44 #include <asm/btext.h>
45 #include <asm/tlb.h>
46 #include <asm/prom.h>
47 #include <asm/lmb.h>
48 #include <asm/sections.h>
49 #include <asm/vdso.h>
50
51 #include "mmu_decl.h"
52
53 #ifndef CPU_FTR_COHERENT_ICACHE
54 #define CPU_FTR_COHERENT_ICACHE 0       /* XXX for now */
55 #define CPU_FTR_NOEXECUTE       0
56 #endif
57
58 int init_bootmem_done;
59 int mem_init_done;
60 unsigned long memory_limit;
61
62 int page_is_ram(unsigned long pfn)
63 {
64         unsigned long paddr = (pfn << PAGE_SHIFT);
65
66 #ifndef CONFIG_PPC64    /* XXX for now */
67         return paddr < __pa(high_memory);
68 #else
69         int i;
70         for (i=0; i < lmb.memory.cnt; i++) {
71                 unsigned long base;
72
73                 base = lmb.memory.region[i].base;
74
75                 if ((paddr >= base) &&
76                         (paddr < (base + lmb.memory.region[i].size))) {
77                         return 1;
78                 }
79         }
80
81         return 0;
82 #endif
83 }
84
85 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
86                               unsigned long size, pgprot_t vma_prot)
87 {
88         if (ppc_md.phys_mem_access_prot)
89                 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
90
91         if (!page_is_ram(pfn))
92                 vma_prot = __pgprot(pgprot_val(vma_prot)
93                                     | _PAGE_GUARDED | _PAGE_NO_CACHE);
94         return vma_prot;
95 }
96 EXPORT_SYMBOL(phys_mem_access_prot);
97
98 #ifdef CONFIG_MEMORY_HOTPLUG
99
100 void online_page(struct page *page)
101 {
102         ClearPageReserved(page);
103         init_page_count(page);
104         __free_page(page);
105         totalram_pages++;
106         num_physpages++;
107 }
108
109 #ifdef CONFIG_NUMA
110 int memory_add_physaddr_to_nid(u64 start)
111 {
112         return hot_add_scn_to_nid(start);
113 }
114 #endif
115
116 int __devinit arch_add_memory(int nid, u64 start, u64 size)
117 {
118         struct pglist_data *pgdata;
119         struct zone *zone;
120         unsigned long start_pfn = start >> PAGE_SHIFT;
121         unsigned long nr_pages = size >> PAGE_SHIFT;
122
123         pgdata = NODE_DATA(nid);
124
125         start = (unsigned long)__va(start);
126         create_section_mapping(start, start + size);
127
128         /* this should work for most non-highmem platforms */
129         zone = pgdata->node_zones;
130
131         return __add_pages(zone, start_pfn, nr_pages);
132
133         return 0;
134 }
135
136 /*
137  * First pass at this code will check to determine if the remove
138  * request is within the RMO.  Do not allow removal within the RMO.
139  */
140 int __devinit remove_memory(u64 start, u64 size)
141 {
142         struct zone *zone;
143         unsigned long start_pfn, end_pfn, nr_pages;
144
145         start_pfn = start >> PAGE_SHIFT;
146         nr_pages = size >> PAGE_SHIFT;
147         end_pfn = start_pfn + nr_pages;
148
149         printk("%s(): Attempting to remove memoy in range "
150                         "%lx to %lx\n", __func__, start, start+size);
151         /*
152          * check for range within RMO
153          */
154         zone = page_zone(pfn_to_page(start_pfn));
155
156         printk("%s(): memory will be removed from "
157                         "the %s zone\n", __func__, zone->name);
158
159         /*
160          * not handling removing memory ranges that
161          * overlap multiple zones yet
162          */
163         if (end_pfn > (zone->zone_start_pfn + zone->spanned_pages))
164                 goto overlap;
165
166         /* make sure it is NOT in RMO */
167         if ((start < lmb.rmo_size) || ((start+size) < lmb.rmo_size)) {
168                 printk("%s(): range to be removed must NOT be in RMO!\n",
169                         __func__);
170                 goto in_rmo;
171         }
172
173         return __remove_pages(zone, start_pfn, nr_pages);
174
175 overlap:
176         printk("%s(): memory range to be removed overlaps "
177                 "multiple zones!!!\n", __func__);
178 in_rmo:
179         return -1;
180 }
181 #endif /* CONFIG_MEMORY_HOTPLUG */
182
183 void show_mem(void)
184 {
185         unsigned long total = 0, reserved = 0;
186         unsigned long shared = 0, cached = 0;
187         unsigned long highmem = 0;
188         struct page *page;
189         pg_data_t *pgdat;
190         unsigned long i;
191
192         printk("Mem-info:\n");
193         show_free_areas();
194         printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
195         for_each_online_pgdat(pgdat) {
196                 unsigned long flags;
197                 pgdat_resize_lock(pgdat, &flags);
198                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
199                         if (!pfn_valid(pgdat->node_start_pfn + i))
200                                 continue;
201                         page = pgdat_page_nr(pgdat, i);
202                         total++;
203                         if (PageHighMem(page))
204                                 highmem++;
205                         if (PageReserved(page))
206                                 reserved++;
207                         else if (PageSwapCache(page))
208                                 cached++;
209                         else if (page_count(page))
210                                 shared += page_count(page) - 1;
211                 }
212                 pgdat_resize_unlock(pgdat, &flags);
213         }
214         printk("%ld pages of RAM\n", total);
215 #ifdef CONFIG_HIGHMEM
216         printk("%ld pages of HIGHMEM\n", highmem);
217 #endif
218         printk("%ld reserved pages\n", reserved);
219         printk("%ld pages shared\n", shared);
220         printk("%ld pages swap cached\n", cached);
221 }
222
223 /*
224  * Initialize the bootmem system and give it all the memory we
225  * have available.  If we are using highmem, we only put the
226  * lowmem into the bootmem system.
227  */
228 #ifndef CONFIG_NEED_MULTIPLE_NODES
229 void __init do_init_bootmem(void)
230 {
231         unsigned long i;
232         unsigned long start, bootmap_pages;
233         unsigned long total_pages;
234         int boot_mapsize;
235
236         max_pfn = total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT;
237 #ifdef CONFIG_HIGHMEM
238         total_pages = total_lowmem >> PAGE_SHIFT;
239 #endif
240
241         /*
242          * Find an area to use for the bootmem bitmap.  Calculate the size of
243          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
244          * Add 1 additional page in case the address isn't page-aligned.
245          */
246         bootmap_pages = bootmem_bootmap_pages(total_pages);
247
248         start = lmb_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
249
250         boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
251
252         /* Add active regions with valid PFNs */
253         for (i = 0; i < lmb.memory.cnt; i++) {
254                 unsigned long start_pfn, end_pfn;
255                 start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
256                 end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
257                 add_active_range(0, start_pfn, end_pfn);
258         }
259
260         /* Add all physical memory to the bootmem map, mark each area
261          * present.
262          */
263 #ifdef CONFIG_HIGHMEM
264         free_bootmem_with_active_regions(0, total_lowmem >> PAGE_SHIFT);
265 #else
266         free_bootmem_with_active_regions(0, max_pfn);
267 #endif
268
269         /* reserve the sections we're already using */
270         for (i = 0; i < lmb.reserved.cnt; i++)
271                 reserve_bootmem(lmb.reserved.region[i].base,
272                                 lmb_size_bytes(&lmb.reserved, i));
273
274         /* XXX need to clip this if using highmem? */
275         sparse_memory_present_with_active_regions(0);
276
277         init_bootmem_done = 1;
278 }
279
280 /* mark pages that don't exist as nosave */
281 static int __init mark_nonram_nosave(void)
282 {
283         unsigned long lmb_next_region_start_pfn,
284                       lmb_region_max_pfn;
285         int i;
286
287         for (i = 0; i < lmb.memory.cnt - 1; i++) {
288                 lmb_region_max_pfn =
289                         (lmb.memory.region[i].base >> PAGE_SHIFT) +
290                         (lmb.memory.region[i].size >> PAGE_SHIFT);
291                 lmb_next_region_start_pfn =
292                         lmb.memory.region[i+1].base >> PAGE_SHIFT;
293
294                 if (lmb_region_max_pfn < lmb_next_region_start_pfn)
295                         register_nosave_region(lmb_region_max_pfn,
296                                                lmb_next_region_start_pfn);
297         }
298
299         return 0;
300 }
301
302 /*
303  * paging_init() sets up the page tables - in fact we've already done this.
304  */
305 void __init paging_init(void)
306 {
307         unsigned long total_ram = lmb_phys_mem_size();
308         unsigned long top_of_ram = lmb_end_of_DRAM();
309         unsigned long max_zone_pfns[MAX_NR_ZONES];
310
311 #ifdef CONFIG_HIGHMEM
312         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
313         pkmap_page_table = pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k
314                         (PKMAP_BASE), PKMAP_BASE), PKMAP_BASE), PKMAP_BASE);
315         map_page(KMAP_FIX_BEGIN, 0, 0); /* XXX gross */
316         kmap_pte = pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k
317                         (KMAP_FIX_BEGIN), KMAP_FIX_BEGIN), KMAP_FIX_BEGIN),
318                          KMAP_FIX_BEGIN);
319         kmap_prot = PAGE_KERNEL;
320 #endif /* CONFIG_HIGHMEM */
321
322         printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
323                top_of_ram, total_ram);
324         printk(KERN_DEBUG "Memory hole size: %ldMB\n",
325                (top_of_ram - total_ram) >> 20);
326         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
327 #ifdef CONFIG_HIGHMEM
328         max_zone_pfns[ZONE_DMA] = total_lowmem >> PAGE_SHIFT;
329         max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
330 #else
331         max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
332 #endif
333         free_area_init_nodes(max_zone_pfns);
334
335         mark_nonram_nosave();
336 }
337 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
338
339 void __init mem_init(void)
340 {
341 #ifdef CONFIG_NEED_MULTIPLE_NODES
342         int nid;
343 #endif
344         pg_data_t *pgdat;
345         unsigned long i;
346         struct page *page;
347         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
348
349         num_physpages = lmb.memory.size >> PAGE_SHIFT;
350         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
351
352 #ifdef CONFIG_NEED_MULTIPLE_NODES
353         for_each_online_node(nid) {
354                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
355                         printk("freeing bootmem node %d\n", nid);
356                         totalram_pages +=
357                                 free_all_bootmem_node(NODE_DATA(nid));
358                 }
359         }
360 #else
361         max_mapnr = max_pfn;
362         totalram_pages += free_all_bootmem();
363 #endif
364         for_each_online_pgdat(pgdat) {
365                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
366                         if (!pfn_valid(pgdat->node_start_pfn + i))
367                                 continue;
368                         page = pgdat_page_nr(pgdat, i);
369                         if (PageReserved(page))
370                                 reservedpages++;
371                 }
372         }
373
374         codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
375         datasize = (unsigned long)&_edata - (unsigned long)&_sdata;
376         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
377         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
378
379 #ifdef CONFIG_HIGHMEM
380         {
381                 unsigned long pfn, highmem_mapnr;
382
383                 highmem_mapnr = total_lowmem >> PAGE_SHIFT;
384                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
385                         struct page *page = pfn_to_page(pfn);
386
387                         ClearPageReserved(page);
388                         init_page_count(page);
389                         __free_page(page);
390                         totalhigh_pages++;
391                 }
392                 totalram_pages += totalhigh_pages;
393                 printk(KERN_DEBUG "High memory: %luk\n",
394                        totalhigh_pages << (PAGE_SHIFT-10));
395         }
396 #endif /* CONFIG_HIGHMEM */
397
398         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
399                "%luk reserved, %luk data, %luk bss, %luk init)\n",
400                 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
401                 num_physpages << (PAGE_SHIFT-10),
402                 codesize >> 10,
403                 reservedpages << (PAGE_SHIFT-10),
404                 datasize >> 10,
405                 bsssize >> 10,
406                 initsize >> 10);
407
408         mem_init_done = 1;
409 }
410
411 /*
412  * This is called when a page has been modified by the kernel.
413  * It just marks the page as not i-cache clean.  We do the i-cache
414  * flush later when the page is given to a user process, if necessary.
415  */
416 void flush_dcache_page(struct page *page)
417 {
418         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
419                 return;
420         /* avoid an atomic op if possible */
421         if (test_bit(PG_arch_1, &page->flags))
422                 clear_bit(PG_arch_1, &page->flags);
423 }
424 EXPORT_SYMBOL(flush_dcache_page);
425
426 void flush_dcache_icache_page(struct page *page)
427 {
428 #ifdef CONFIG_BOOKE
429         void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
430         __flush_dcache_icache(start);
431         kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
432 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
433         /* On 8xx there is no need to kmap since highmem is not supported */
434         __flush_dcache_icache(page_address(page)); 
435 #else
436         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
437 #endif
438
439 }
440 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
441 {
442         clear_page(page);
443
444         /*
445          * We shouldnt have to do this, but some versions of glibc
446          * require it (ld.so assumes zero filled pages are icache clean)
447          * - Anton
448          */
449         flush_dcache_page(pg);
450 }
451 EXPORT_SYMBOL(clear_user_page);
452
453 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
454                     struct page *pg)
455 {
456         copy_page(vto, vfrom);
457
458         /*
459          * We should be able to use the following optimisation, however
460          * there are two problems.
461          * Firstly a bug in some versions of binutils meant PLT sections
462          * were not marked executable.
463          * Secondly the first word in the GOT section is blrl, used
464          * to establish the GOT address. Until recently the GOT was
465          * not marked executable.
466          * - Anton
467          */
468 #if 0
469         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
470                 return;
471 #endif
472
473         flush_dcache_page(pg);
474 }
475
476 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
477                              unsigned long addr, int len)
478 {
479         unsigned long maddr;
480
481         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
482         flush_icache_range(maddr, maddr + len);
483         kunmap(page);
484 }
485 EXPORT_SYMBOL(flush_icache_user_range);
486
487 /*
488  * This is called at the end of handling a user page fault, when the
489  * fault has been handled by updating a PTE in the linux page tables.
490  * We use it to preload an HPTE into the hash table corresponding to
491  * the updated linux PTE.
492  * 
493  * This must always be called with the pte lock held.
494  */
495 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
496                       pte_t pte)
497 {
498 #ifdef CONFIG_PPC_STD_MMU
499         unsigned long access = 0, trap;
500 #endif
501         unsigned long pfn = pte_pfn(pte);
502
503         /* handle i-cache coherency */
504         if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
505             !cpu_has_feature(CPU_FTR_NOEXECUTE) &&
506             pfn_valid(pfn)) {
507                 struct page *page = pfn_to_page(pfn);
508 #ifdef CONFIG_8xx
509                 /* On 8xx, cache control instructions (particularly
510                  * "dcbst" from flush_dcache_icache) fault as write
511                  * operation if there is an unpopulated TLB entry
512                  * for the address in question. To workaround that,
513                  * we invalidate the TLB here, thus avoiding dcbst
514                  * misbehaviour.
515                  */
516                 _tlbie(address);
517 #endif
518                 if (!PageReserved(page)
519                     && !test_bit(PG_arch_1, &page->flags)) {
520                         if (vma->vm_mm == current->active_mm) {
521                                 __flush_dcache_icache((void *) address);
522                         } else
523                                 flush_dcache_icache_page(page);
524                         set_bit(PG_arch_1, &page->flags);
525                 }
526         }
527
528 #ifdef CONFIG_PPC_STD_MMU
529         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
530         if (!pte_young(pte) || address >= TASK_SIZE)
531                 return;
532
533         /* We try to figure out if we are coming from an instruction
534          * access fault and pass that down to __hash_page so we avoid
535          * double-faulting on execution of fresh text. We have to test
536          * for regs NULL since init will get here first thing at boot
537          *
538          * We also avoid filling the hash if not coming from a fault
539          */
540         if (current->thread.regs == NULL)
541                 return;
542         trap = TRAP(current->thread.regs);
543         if (trap == 0x400)
544                 access |= _PAGE_EXEC;
545         else if (trap != 0x300)
546                 return;
547         hash_preload(vma->vm_mm, address, access, trap);
548 #endif /* CONFIG_PPC_STD_MMU */
549 }