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