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