Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[linux-2.6/libata-dev.git] / arch / powerpc / mm / mem.c
blobf1f7409a418387dcd73d3791eab891ca4ea8e80d
1 /*
2 * PowerPC version
3 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
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)
10 * Derived from "arch/i386/mm/init.c"
11 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
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.
20 #include <linux/export.h>
21 #include <linux/sched.h>
22 #include <linux/kernel.h>
23 #include <linux/errno.h>
24 #include <linux/string.h>
25 #include <linux/gfp.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 #include <linux/memblock.h>
36 #include <linux/hugetlb.h>
37 #include <linux/slab.h>
39 #include <asm/pgalloc.h>
40 #include <asm/prom.h>
41 #include <asm/io.h>
42 #include <asm/mmu_context.h>
43 #include <asm/pgtable.h>
44 #include <asm/mmu.h>
45 #include <asm/smp.h>
46 #include <asm/machdep.h>
47 #include <asm/btext.h>
48 #include <asm/tlb.h>
49 #include <asm/sections.h>
50 #include <asm/sparsemem.h>
51 #include <asm/vdso.h>
52 #include <asm/fixmap.h>
53 #include <asm/swiotlb.h>
54 #include <asm/rtas.h>
56 #include "mmu_decl.h"
58 #ifndef CPU_FTR_COHERENT_ICACHE
59 #define CPU_FTR_COHERENT_ICACHE 0 /* XXX for now */
60 #define CPU_FTR_NOEXECUTE 0
61 #endif
63 int init_bootmem_done;
64 int mem_init_done;
65 unsigned long long memory_limit;
67 #ifdef CONFIG_HIGHMEM
68 pte_t *kmap_pte;
69 pgprot_t kmap_prot;
71 EXPORT_SYMBOL(kmap_prot);
72 EXPORT_SYMBOL(kmap_pte);
74 static inline pte_t *virt_to_kpte(unsigned long vaddr)
76 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
77 vaddr), vaddr), vaddr);
79 #endif
81 int page_is_ram(unsigned long pfn)
83 #ifndef CONFIG_PPC64 /* XXX for now */
84 return pfn < max_pfn;
85 #else
86 unsigned long paddr = (pfn << PAGE_SHIFT);
87 struct memblock_region *reg;
89 for_each_memblock(memory, reg)
90 if (paddr >= reg->base && paddr < (reg->base + reg->size))
91 return 1;
92 return 0;
93 #endif
96 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
97 unsigned long size, pgprot_t vma_prot)
99 if (ppc_md.phys_mem_access_prot)
100 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
102 if (!page_is_ram(pfn))
103 vma_prot = pgprot_noncached(vma_prot);
105 return vma_prot;
107 EXPORT_SYMBOL(phys_mem_access_prot);
109 #ifdef CONFIG_MEMORY_HOTPLUG
111 #ifdef CONFIG_NUMA
112 int memory_add_physaddr_to_nid(u64 start)
114 return hot_add_scn_to_nid(start);
116 #endif
118 int arch_add_memory(int nid, u64 start, u64 size)
120 struct pglist_data *pgdata;
121 struct zone *zone;
122 unsigned long start_pfn = start >> PAGE_SHIFT;
123 unsigned long nr_pages = size >> PAGE_SHIFT;
125 pgdata = NODE_DATA(nid);
127 start = (unsigned long)__va(start);
128 if (create_section_mapping(start, start + size))
129 return -EINVAL;
131 /* this should work for most non-highmem platforms */
132 zone = pgdata->node_zones;
134 return __add_pages(nid, zone, start_pfn, nr_pages);
137 #ifdef CONFIG_MEMORY_HOTREMOVE
138 int arch_remove_memory(u64 start, u64 size)
140 unsigned long start_pfn = start >> PAGE_SHIFT;
141 unsigned long nr_pages = size >> PAGE_SHIFT;
142 struct zone *zone;
144 zone = page_zone(pfn_to_page(start_pfn));
145 return __remove_pages(zone, start_pfn, nr_pages);
147 #endif
148 #endif /* CONFIG_MEMORY_HOTPLUG */
151 * walk_memory_resource() needs to make sure there is no holes in a given
152 * memory range. PPC64 does not maintain the memory layout in /proc/iomem.
153 * Instead it maintains it in memblock.memory structures. Walk through the
154 * memory regions, find holes and callback for contiguous regions.
157 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
158 void *arg, int (*func)(unsigned long, unsigned long, void *))
160 struct memblock_region *reg;
161 unsigned long end_pfn = start_pfn + nr_pages;
162 unsigned long tstart, tend;
163 int ret = -1;
165 for_each_memblock(memory, reg) {
166 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
167 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
168 if (tstart >= tend)
169 continue;
170 ret = (*func)(tstart, tend - tstart, arg);
171 if (ret)
172 break;
174 return ret;
176 EXPORT_SYMBOL_GPL(walk_system_ram_range);
179 * Initialize the bootmem system and give it all the memory we
180 * have available. If we are using highmem, we only put the
181 * lowmem into the bootmem system.
183 #ifndef CONFIG_NEED_MULTIPLE_NODES
184 void __init do_init_bootmem(void)
186 unsigned long start, bootmap_pages;
187 unsigned long total_pages;
188 struct memblock_region *reg;
189 int boot_mapsize;
191 max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
192 total_pages = (memblock_end_of_DRAM() - memstart_addr) >> PAGE_SHIFT;
193 #ifdef CONFIG_HIGHMEM
194 total_pages = total_lowmem >> PAGE_SHIFT;
195 max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
196 #endif
199 * Find an area to use for the bootmem bitmap. Calculate the size of
200 * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
201 * Add 1 additional page in case the address isn't page-aligned.
203 bootmap_pages = bootmem_bootmap_pages(total_pages);
205 start = memblock_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
207 min_low_pfn = MEMORY_START >> PAGE_SHIFT;
208 boot_mapsize = init_bootmem_node(NODE_DATA(0), start >> PAGE_SHIFT, min_low_pfn, max_low_pfn);
210 /* Place all memblock_regions in the same node and merge contiguous
211 * memblock_regions
213 memblock_set_node(0, (phys_addr_t)ULLONG_MAX, 0);
215 /* Add all physical memory to the bootmem map, mark each area
216 * present.
218 #ifdef CONFIG_HIGHMEM
219 free_bootmem_with_active_regions(0, lowmem_end_addr >> PAGE_SHIFT);
221 /* reserve the sections we're already using */
222 for_each_memblock(reserved, reg) {
223 unsigned long top = reg->base + reg->size - 1;
224 if (top < lowmem_end_addr)
225 reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
226 else if (reg->base < lowmem_end_addr) {
227 unsigned long trunc_size = lowmem_end_addr - reg->base;
228 reserve_bootmem(reg->base, trunc_size, BOOTMEM_DEFAULT);
231 #else
232 free_bootmem_with_active_regions(0, max_pfn);
234 /* reserve the sections we're already using */
235 for_each_memblock(reserved, reg)
236 reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
237 #endif
238 /* XXX need to clip this if using highmem? */
239 sparse_memory_present_with_active_regions(0);
241 init_bootmem_done = 1;
244 /* mark pages that don't exist as nosave */
245 static int __init mark_nonram_nosave(void)
247 struct memblock_region *reg, *prev = NULL;
249 for_each_memblock(memory, reg) {
250 if (prev &&
251 memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
252 register_nosave_region(memblock_region_memory_end_pfn(prev),
253 memblock_region_memory_base_pfn(reg));
254 prev = reg;
256 return 0;
260 * paging_init() sets up the page tables - in fact we've already done this.
262 void __init paging_init(void)
264 unsigned long long total_ram = memblock_phys_mem_size();
265 phys_addr_t top_of_ram = memblock_end_of_DRAM();
266 unsigned long max_zone_pfns[MAX_NR_ZONES];
268 #ifdef CONFIG_PPC32
269 unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
270 unsigned long end = __fix_to_virt(FIX_HOLE);
272 for (; v < end; v += PAGE_SIZE)
273 map_page(v, 0, 0); /* XXX gross */
274 #endif
276 #ifdef CONFIG_HIGHMEM
277 map_page(PKMAP_BASE, 0, 0); /* XXX gross */
278 pkmap_page_table = virt_to_kpte(PKMAP_BASE);
280 kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
281 kmap_prot = PAGE_KERNEL;
282 #endif /* CONFIG_HIGHMEM */
284 printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
285 (unsigned long long)top_of_ram, total_ram);
286 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
287 (long int)((top_of_ram - total_ram) >> 20));
288 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
289 #ifdef CONFIG_HIGHMEM
290 max_zone_pfns[ZONE_DMA] = lowmem_end_addr >> PAGE_SHIFT;
291 max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
292 #else
293 max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
294 #endif
295 free_area_init_nodes(max_zone_pfns);
297 mark_nonram_nosave();
299 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
301 void __init mem_init(void)
303 #ifdef CONFIG_NEED_MULTIPLE_NODES
304 int nid;
305 #endif
306 pg_data_t *pgdat;
307 unsigned long i;
308 struct page *page;
309 unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
311 #ifdef CONFIG_SWIOTLB
312 swiotlb_init(0);
313 #endif
315 num_physpages = memblock_phys_mem_size() >> PAGE_SHIFT;
316 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
318 #ifdef CONFIG_NEED_MULTIPLE_NODES
319 for_each_online_node(nid) {
320 if (NODE_DATA(nid)->node_spanned_pages != 0) {
321 printk("freeing bootmem node %d\n", nid);
322 totalram_pages +=
323 free_all_bootmem_node(NODE_DATA(nid));
326 #else
327 max_mapnr = max_pfn;
328 totalram_pages += free_all_bootmem();
329 #endif
330 for_each_online_pgdat(pgdat) {
331 for (i = 0; i < pgdat->node_spanned_pages; i++) {
332 if (!pfn_valid(pgdat->node_start_pfn + i))
333 continue;
334 page = pgdat_page_nr(pgdat, i);
335 if (PageReserved(page))
336 reservedpages++;
340 codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
341 datasize = (unsigned long)&_edata - (unsigned long)&_sdata;
342 initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
343 bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
345 #ifdef CONFIG_HIGHMEM
347 unsigned long pfn, highmem_mapnr;
349 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
350 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
351 phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
352 struct page *page = pfn_to_page(pfn);
353 if (memblock_is_reserved(paddr))
354 continue;
355 ClearPageReserved(page);
356 init_page_count(page);
357 __free_page(page);
358 totalhigh_pages++;
359 reservedpages--;
361 totalram_pages += totalhigh_pages;
362 printk(KERN_DEBUG "High memory: %luk\n",
363 totalhigh_pages << (PAGE_SHIFT-10));
365 #endif /* CONFIG_HIGHMEM */
367 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
369 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
370 * functions.... do it here for the non-smp case.
372 per_cpu(next_tlbcam_idx, smp_processor_id()) =
373 (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
374 #endif
376 printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
377 "%luk reserved, %luk data, %luk bss, %luk init)\n",
378 nr_free_pages() << (PAGE_SHIFT-10),
379 num_physpages << (PAGE_SHIFT-10),
380 codesize >> 10,
381 reservedpages << (PAGE_SHIFT-10),
382 datasize >> 10,
383 bsssize >> 10,
384 initsize >> 10);
386 #ifdef CONFIG_PPC32
387 pr_info("Kernel virtual memory layout:\n");
388 pr_info(" * 0x%08lx..0x%08lx : fixmap\n", FIXADDR_START, FIXADDR_TOP);
389 #ifdef CONFIG_HIGHMEM
390 pr_info(" * 0x%08lx..0x%08lx : highmem PTEs\n",
391 PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
392 #endif /* CONFIG_HIGHMEM */
393 #ifdef CONFIG_NOT_COHERENT_CACHE
394 pr_info(" * 0x%08lx..0x%08lx : consistent mem\n",
395 IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
396 #endif /* CONFIG_NOT_COHERENT_CACHE */
397 pr_info(" * 0x%08lx..0x%08lx : early ioremap\n",
398 ioremap_bot, IOREMAP_TOP);
399 pr_info(" * 0x%08lx..0x%08lx : vmalloc & ioremap\n",
400 VMALLOC_START, VMALLOC_END);
401 #endif /* CONFIG_PPC32 */
403 mem_init_done = 1;
406 void free_initmem(void)
408 unsigned long addr;
410 ppc_md.progress = ppc_printk_progress;
412 addr = (unsigned long)__init_begin;
413 for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) {
414 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
415 ClearPageReserved(virt_to_page(addr));
416 init_page_count(virt_to_page(addr));
417 free_page(addr);
418 totalram_pages++;
420 pr_info("Freeing unused kernel memory: %luk freed\n",
421 ((unsigned long)__init_end -
422 (unsigned long)__init_begin) >> 10);
425 #ifdef CONFIG_BLK_DEV_INITRD
426 void __init free_initrd_mem(unsigned long start, unsigned long end)
428 if (start >= end)
429 return;
431 start = _ALIGN_DOWN(start, PAGE_SIZE);
432 end = _ALIGN_UP(end, PAGE_SIZE);
433 pr_info("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
435 for (; start < end; start += PAGE_SIZE) {
436 ClearPageReserved(virt_to_page(start));
437 init_page_count(virt_to_page(start));
438 free_page(start);
439 totalram_pages++;
442 #endif
445 * This is called when a page has been modified by the kernel.
446 * It just marks the page as not i-cache clean. We do the i-cache
447 * flush later when the page is given to a user process, if necessary.
449 void flush_dcache_page(struct page *page)
451 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
452 return;
453 /* avoid an atomic op if possible */
454 if (test_bit(PG_arch_1, &page->flags))
455 clear_bit(PG_arch_1, &page->flags);
457 EXPORT_SYMBOL(flush_dcache_page);
459 void flush_dcache_icache_page(struct page *page)
461 #ifdef CONFIG_HUGETLB_PAGE
462 if (PageCompound(page)) {
463 flush_dcache_icache_hugepage(page);
464 return;
466 #endif
467 #ifdef CONFIG_BOOKE
469 void *start = kmap_atomic(page);
470 __flush_dcache_icache(start);
471 kunmap_atomic(start);
473 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
474 /* On 8xx there is no need to kmap since highmem is not supported */
475 __flush_dcache_icache(page_address(page));
476 #else
477 __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
478 #endif
480 EXPORT_SYMBOL(flush_dcache_icache_page);
482 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
484 clear_page(page);
487 * We shouldn't have to do this, but some versions of glibc
488 * require it (ld.so assumes zero filled pages are icache clean)
489 * - Anton
491 flush_dcache_page(pg);
493 EXPORT_SYMBOL(clear_user_page);
495 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
496 struct page *pg)
498 copy_page(vto, vfrom);
501 * We should be able to use the following optimisation, however
502 * there are two problems.
503 * Firstly a bug in some versions of binutils meant PLT sections
504 * were not marked executable.
505 * Secondly the first word in the GOT section is blrl, used
506 * to establish the GOT address. Until recently the GOT was
507 * not marked executable.
508 * - Anton
510 #if 0
511 if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
512 return;
513 #endif
515 flush_dcache_page(pg);
518 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
519 unsigned long addr, int len)
521 unsigned long maddr;
523 maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
524 flush_icache_range(maddr, maddr + len);
525 kunmap(page);
527 EXPORT_SYMBOL(flush_icache_user_range);
530 * This is called at the end of handling a user page fault, when the
531 * fault has been handled by updating a PTE in the linux page tables.
532 * We use it to preload an HPTE into the hash table corresponding to
533 * the updated linux PTE.
535 * This must always be called with the pte lock held.
537 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
538 pte_t *ptep)
540 #ifdef CONFIG_PPC_STD_MMU
541 unsigned long access = 0, trap;
543 /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
544 if (!pte_young(*ptep) || address >= TASK_SIZE)
545 return;
547 /* We try to figure out if we are coming from an instruction
548 * access fault and pass that down to __hash_page so we avoid
549 * double-faulting on execution of fresh text. We have to test
550 * for regs NULL since init will get here first thing at boot
552 * We also avoid filling the hash if not coming from a fault
554 if (current->thread.regs == NULL)
555 return;
556 trap = TRAP(current->thread.regs);
557 if (trap == 0x400)
558 access |= _PAGE_EXEC;
559 else if (trap != 0x300)
560 return;
561 hash_preload(vma->vm_mm, address, access, trap);
562 #endif /* CONFIG_PPC_STD_MMU */
563 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
564 && defined(CONFIG_HUGETLB_PAGE)
565 if (is_vm_hugetlb_page(vma))
566 book3e_hugetlb_preload(vma, address, *ptep);
567 #endif
571 * System memory should not be in /proc/iomem but various tools expect it
572 * (eg kdump).
574 static int add_system_ram_resources(void)
576 struct memblock_region *reg;
578 for_each_memblock(memory, reg) {
579 struct resource *res;
580 unsigned long base = reg->base;
581 unsigned long size = reg->size;
583 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
584 WARN_ON(!res);
586 if (res) {
587 res->name = "System RAM";
588 res->start = base;
589 res->end = base + size - 1;
590 res->flags = IORESOURCE_MEM;
591 WARN_ON(request_resource(&iomem_resource, res) < 0);
595 return 0;
597 subsys_initcall(add_system_ram_resources);
599 #ifdef CONFIG_STRICT_DEVMEM
601 * devmem_is_allowed(): check to see if /dev/mem access to a certain address
602 * is valid. The argument is a physical page number.
604 * Access has to be given to non-kernel-ram areas as well, these contain the
605 * PCI mmio resources as well as potential bios/acpi data regions.
607 int devmem_is_allowed(unsigned long pfn)
609 if (iomem_is_exclusive(pfn << PAGE_SHIFT))
610 return 0;
611 if (!page_is_ram(pfn))
612 return 1;
613 if (page_is_rtas_user_buf(pfn))
614 return 1;
615 return 0;
617 #endif /* CONFIG_STRICT_DEVMEM */