Merge in v2.6.14 by hand
[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 #ifdef CONFIG_PPC64
50 #include <asm/vdso.h>
51 #endif
52
53 #include "mmu_decl.h"
54
55 #ifndef CPU_FTR_COHERENT_ICACHE
56 #define CPU_FTR_COHERENT_ICACHE 0       /* XXX for now */
57 #define CPU_FTR_NOEXECUTE       0
58 #endif
59
60 int init_bootmem_done;
61 int mem_init_done;
62
63 /*
64  * This is called by /dev/mem to know if a given address has to
65  * be mapped non-cacheable or not
66  */
67 int page_is_ram(unsigned long pfn)
68 {
69         unsigned long paddr = (pfn << PAGE_SHIFT);
70
71 #ifndef CONFIG_PPC64    /* XXX for now */
72         return paddr < __pa(high_memory);
73 #else
74         int i;
75         for (i=0; i < lmb.memory.cnt; i++) {
76                 unsigned long base;
77
78                 base = lmb.memory.region[i].base;
79
80                 if ((paddr >= base) &&
81                         (paddr < (base + lmb.memory.region[i].size))) {
82                         return 1;
83                 }
84         }
85
86         return 0;
87 #endif
88 }
89 EXPORT_SYMBOL(page_is_ram);
90
91 pgprot_t phys_mem_access_prot(struct file *file, unsigned long addr,
92                               unsigned long size, pgprot_t vma_prot)
93 {
94         if (ppc_md.phys_mem_access_prot)
95                 return ppc_md.phys_mem_access_prot(file, addr, size, vma_prot);
96
97         if (!page_is_ram(addr >> PAGE_SHIFT))
98                 vma_prot = __pgprot(pgprot_val(vma_prot)
99                                     | _PAGE_GUARDED | _PAGE_NO_CACHE);
100         return vma_prot;
101 }
102 EXPORT_SYMBOL(phys_mem_access_prot);
103
104 void show_mem(void)
105 {
106         unsigned long total = 0, reserved = 0;
107         unsigned long shared = 0, cached = 0;
108         unsigned long highmem = 0;
109         struct page *page;
110         pg_data_t *pgdat;
111         unsigned long i;
112
113         printk("Mem-info:\n");
114         show_free_areas();
115         printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
116         for_each_pgdat(pgdat) {
117                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
118                         page = pgdat_page_nr(pgdat, i);
119                         total++;
120                         if (PageHighMem(page))
121                                 highmem++;
122                         if (PageReserved(page))
123                                 reserved++;
124                         else if (PageSwapCache(page))
125                                 cached++;
126                         else if (page_count(page))
127                                 shared += page_count(page) - 1;
128                 }
129         }
130         printk("%ld pages of RAM\n", total);
131 #ifdef CONFIG_HIGHMEM
132         printk("%ld pages of HIGHMEM\n", highmem);
133 #endif
134         printk("%ld reserved pages\n", reserved);
135         printk("%ld pages shared\n", shared);
136         printk("%ld pages swap cached\n", cached);
137 }
138
139 /*
140  * Initialize the bootmem system and give it all the memory we
141  * have available.  If we are using highmem, we only put the
142  * lowmem into the bootmem system.
143  */
144 #ifndef CONFIG_NEED_MULTIPLE_NODES
145 void __init do_init_bootmem(void)
146 {
147         unsigned long i;
148         unsigned long start, bootmap_pages;
149         unsigned long total_pages;
150         int boot_mapsize;
151
152         max_pfn = total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT;
153 #ifdef CONFIG_HIGHMEM
154         total_pages = total_lowmem >> PAGE_SHIFT;
155 #endif
156
157         /*
158          * Find an area to use for the bootmem bitmap.  Calculate the size of
159          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
160          * Add 1 additional page in case the address isn't page-aligned.
161          */
162         bootmap_pages = bootmem_bootmap_pages(total_pages);
163
164         start = lmb_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
165         BUG_ON(!start);
166
167         boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
168
169         /* Add all physical memory to the bootmem map, mark each area
170          * present.
171          */
172         for (i = 0; i < lmb.memory.cnt; i++) {
173                 unsigned long base = lmb.memory.region[i].base;
174                 unsigned long size = lmb_size_bytes(&lmb.memory, i);
175 #ifdef CONFIG_HIGHMEM
176                 if (base >= total_lowmem)
177                         continue;
178                 if (base + size > total_lowmem)
179                         size = total_lowmem - base;
180 #endif
181                 free_bootmem(base, size);
182         }
183
184         /* reserve the sections we're already using */
185         for (i = 0; i < lmb.reserved.cnt; i++)
186                 reserve_bootmem(lmb.reserved.region[i].base,
187                                 lmb_size_bytes(&lmb.reserved, i));
188
189         /* XXX need to clip this if using highmem? */
190         for (i = 0; i < lmb.memory.cnt; i++)
191                 memory_present(0, lmb_start_pfn(&lmb.memory, i),
192                                lmb_end_pfn(&lmb.memory, i));
193         init_bootmem_done = 1;
194 }
195
196 /*
197  * paging_init() sets up the page tables - in fact we've already done this.
198  */
199 void __init paging_init(void)
200 {
201         unsigned long zones_size[MAX_NR_ZONES];
202         unsigned long zholes_size[MAX_NR_ZONES];
203         unsigned long total_ram = lmb_phys_mem_size();
204         unsigned long top_of_ram = lmb_end_of_DRAM();
205
206 #ifdef CONFIG_HIGHMEM
207         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
208         pkmap_page_table = pte_offset_kernel(pmd_offset(pgd_offset_k
209                         (PKMAP_BASE), PKMAP_BASE), PKMAP_BASE);
210         map_page(KMAP_FIX_BEGIN, 0, 0); /* XXX gross */
211         kmap_pte = pte_offset_kernel(pmd_offset(pgd_offset_k
212                         (KMAP_FIX_BEGIN), KMAP_FIX_BEGIN), KMAP_FIX_BEGIN);
213         kmap_prot = PAGE_KERNEL;
214 #endif /* CONFIG_HIGHMEM */
215
216         printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
217                top_of_ram, total_ram);
218         printk(KERN_INFO "Memory hole size: %ldMB\n",
219                (top_of_ram - total_ram) >> 20);
220         /*
221          * All pages are DMA-able so we put them all in the DMA zone.
222          */
223         memset(zones_size, 0, sizeof(zones_size));
224         memset(zholes_size, 0, sizeof(zholes_size));
225
226         zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
227         zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
228
229 #ifdef CONFIG_HIGHMEM
230         zones_size[ZONE_DMA] = total_lowmem >> PAGE_SHIFT;
231         zones_size[ZONE_HIGHMEM] = (total_memory - total_lowmem) >> PAGE_SHIFT;
232         zholes_size[ZONE_HIGHMEM] = (top_of_ram - total_ram) >> PAGE_SHIFT;
233 #else
234         zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
235         zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
236 #endif /* CONFIG_HIGHMEM */
237
238         free_area_init_node(0, NODE_DATA(0), zones_size,
239                             __pa(PAGE_OFFSET) >> PAGE_SHIFT, zholes_size);
240 }
241 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
242
243 void __init mem_init(void)
244 {
245 #ifdef CONFIG_NEED_MULTIPLE_NODES
246         int nid;
247 #endif
248         pg_data_t *pgdat;
249         unsigned long i;
250         struct page *page;
251         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
252
253         num_physpages = max_pfn;        /* RAM is assumed contiguous */
254         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
255
256 #ifdef CONFIG_NEED_MULTIPLE_NODES
257         for_each_online_node(nid) {
258                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
259                         printk("freeing bootmem node %x\n", nid);
260                         totalram_pages +=
261                                 free_all_bootmem_node(NODE_DATA(nid));
262                 }
263         }
264 #else
265         max_mapnr = num_physpages;
266         totalram_pages += free_all_bootmem();
267 #endif
268         for_each_pgdat(pgdat) {
269                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
270                         page = pgdat_page_nr(pgdat, i);
271                         if (PageReserved(page))
272                                 reservedpages++;
273                 }
274         }
275
276         codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
277         datasize = (unsigned long)&__init_begin - (unsigned long)&_sdata;
278         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
279         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
280
281 #ifdef CONFIG_HIGHMEM
282         {
283                 unsigned long pfn, highmem_mapnr;
284
285                 highmem_mapnr = total_lowmem >> PAGE_SHIFT;
286                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
287                         struct page *page = pfn_to_page(pfn);
288
289                         ClearPageReserved(page);
290                         set_page_count(page, 1);
291                         __free_page(page);
292                         totalhigh_pages++;
293                 }
294                 totalram_pages += totalhigh_pages;
295                 printk(KERN_INFO "High memory: %luk\n",
296                        totalhigh_pages << (PAGE_SHIFT-10));
297         }
298 #endif /* CONFIG_HIGHMEM */
299
300         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
301                "%luk reserved, %luk data, %luk bss, %luk init)\n",
302                 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
303                 num_physpages << (PAGE_SHIFT-10),
304                 codesize >> 10,
305                 reservedpages << (PAGE_SHIFT-10),
306                 datasize >> 10,
307                 bsssize >> 10,
308                 initsize >> 10);
309
310         mem_init_done = 1;
311
312 #ifdef CONFIG_PPC64
313         /* Initialize the vDSO */
314         vdso_init();
315 #endif
316 }
317
318 /*
319  * This is called when a page has been modified by the kernel.
320  * It just marks the page as not i-cache clean.  We do the i-cache
321  * flush later when the page is given to a user process, if necessary.
322  */
323 void flush_dcache_page(struct page *page)
324 {
325         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
326                 return;
327         /* avoid an atomic op if possible */
328         if (test_bit(PG_arch_1, &page->flags))
329                 clear_bit(PG_arch_1, &page->flags);
330 }
331 EXPORT_SYMBOL(flush_dcache_page);
332
333 void flush_dcache_icache_page(struct page *page)
334 {
335 #ifdef CONFIG_BOOKE
336         void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
337         __flush_dcache_icache(start);
338         kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
339 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
340         /* On 8xx there is no need to kmap since highmem is not supported */
341         __flush_dcache_icache(page_address(page)); 
342 #else
343         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
344 #endif
345
346 }
347 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
348 {
349         clear_page(page);
350
351         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
352                 return;
353         /*
354          * We shouldnt have to do this, but some versions of glibc
355          * require it (ld.so assumes zero filled pages are icache clean)
356          * - Anton
357          */
358
359         /* avoid an atomic op if possible */
360         if (test_bit(PG_arch_1, &pg->flags))
361                 clear_bit(PG_arch_1, &pg->flags);
362 }
363 EXPORT_SYMBOL(clear_user_page);
364
365 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
366                     struct page *pg)
367 {
368         copy_page(vto, vfrom);
369
370         /*
371          * We should be able to use the following optimisation, however
372          * there are two problems.
373          * Firstly a bug in some versions of binutils meant PLT sections
374          * were not marked executable.
375          * Secondly the first word in the GOT section is blrl, used
376          * to establish the GOT address. Until recently the GOT was
377          * not marked executable.
378          * - Anton
379          */
380 #if 0
381         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
382                 return;
383 #endif
384
385         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
386                 return;
387
388         /* avoid an atomic op if possible */
389         if (test_bit(PG_arch_1, &pg->flags))
390                 clear_bit(PG_arch_1, &pg->flags);
391 }
392
393 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
394                              unsigned long addr, int len)
395 {
396         unsigned long maddr;
397
398         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
399         flush_icache_range(maddr, maddr + len);
400         kunmap(page);
401 }
402 EXPORT_SYMBOL(flush_icache_user_range);
403
404 /*
405  * This is called at the end of handling a user page fault, when the
406  * fault has been handled by updating a PTE in the linux page tables.
407  * We use it to preload an HPTE into the hash table corresponding to
408  * the updated linux PTE.
409  * 
410  * This must always be called with the mm->page_table_lock held
411  */
412 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
413                       pte_t pte)
414 {
415         /* handle i-cache coherency */
416         unsigned long pfn = pte_pfn(pte);
417 #ifdef CONFIG_PPC32
418         pmd_t *pmd;
419 #else
420         unsigned long vsid;
421         void *pgdir;
422         pte_t *ptep;
423         int local = 0;
424         cpumask_t tmp;
425         unsigned long flags;
426 #endif
427
428         /* handle i-cache coherency */
429         if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
430             !cpu_has_feature(CPU_FTR_NOEXECUTE) &&
431             pfn_valid(pfn)) {
432                 struct page *page = pfn_to_page(pfn);
433                 if (!PageReserved(page)
434                     && !test_bit(PG_arch_1, &page->flags)) {
435                         if (vma->vm_mm == current->active_mm) {
436 #ifdef CONFIG_8xx
437                         /* On 8xx, cache control instructions (particularly 
438                          * "dcbst" from flush_dcache_icache) fault as write 
439                          * operation if there is an unpopulated TLB entry 
440                          * for the address in question. To workaround that, 
441                          * we invalidate the TLB here, thus avoiding dcbst 
442                          * misbehaviour.
443                          */
444                                 _tlbie(address);
445 #endif
446                                 __flush_dcache_icache((void *) address);
447                         } else
448                                 flush_dcache_icache_page(page);
449                         set_bit(PG_arch_1, &page->flags);
450                 }
451         }
452
453 #ifdef CONFIG_PPC_STD_MMU
454         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
455         if (!pte_young(pte) || address >= TASK_SIZE)
456                 return;
457 #ifdef CONFIG_PPC32
458         if (Hash == 0)
459                 return;
460         pmd = pmd_offset(pgd_offset(vma->vm_mm, address), address);
461         if (!pmd_none(*pmd))
462                 add_hash_page(vma->vm_mm->context, address, pmd_val(*pmd));
463 #else
464         pgdir = vma->vm_mm->pgd;
465         if (pgdir == NULL)
466                 return;
467
468         ptep = find_linux_pte(pgdir, address);
469         if (!ptep)
470                 return;
471
472         vsid = get_vsid(vma->vm_mm->context.id, address);
473
474         local_irq_save(flags);
475         tmp = cpumask_of_cpu(smp_processor_id());
476         if (cpus_equal(vma->vm_mm->cpu_vm_mask, tmp))
477                 local = 1;
478
479         __hash_page(address, 0, vsid, ptep, 0x300, local);
480         local_irq_restore(flags);
481 #endif
482 #endif
483 }