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/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>
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 #include <linux/lmb.h>
36 #include <asm/pgalloc.h>
39 #include <asm/mmu_context.h>
40 #include <asm/pgtable.h>
43 #include <asm/machdep.h>
44 #include <asm/btext.h>
46 #include <asm/sections.h>
47 #include <asm/sparsemem.h>
49 #include <asm/fixmap.h>
53 #ifndef CPU_FTR_COHERENT_ICACHE
54 #define CPU_FTR_COHERENT_ICACHE 0 /* XXX for now */
55 #define CPU_FTR_NOEXECUTE 0
58 int init_bootmem_done;
60 unsigned long memory_limit;
66 EXPORT_SYMBOL(kmap_prot);
67 EXPORT_SYMBOL(kmap_pte);
69 static inline pte_t *virt_to_kpte(unsigned long vaddr)
71 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
72 vaddr), vaddr), vaddr);
76 int page_is_ram(unsigned long pfn)
78 #ifndef CONFIG_PPC64 /* XXX for now */
81 unsigned long paddr = (pfn << PAGE_SHIFT);
83 for (i=0; i < lmb.memory.cnt; i++) {
86 base = lmb.memory.region[i].base;
88 if ((paddr >= base) &&
89 (paddr < (base + lmb.memory.region[i].size))) {
98 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
99 unsigned long size, pgprot_t vma_prot)
101 if (ppc_md.phys_mem_access_prot)
102 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
104 if (!page_is_ram(pfn))
105 vma_prot = pgprot_noncached(vma_prot);
109 EXPORT_SYMBOL(phys_mem_access_prot);
111 #ifdef CONFIG_MEMORY_HOTPLUG
114 int memory_add_physaddr_to_nid(u64 start)
116 return hot_add_scn_to_nid(start);
120 int arch_add_memory(int nid, u64 start, u64 size)
122 struct pglist_data *pgdata;
124 unsigned long start_pfn = start >> PAGE_SHIFT;
125 unsigned long nr_pages = size >> PAGE_SHIFT;
127 pgdata = NODE_DATA(nid);
129 start = (unsigned long)__va(start);
130 create_section_mapping(start, start + size);
132 /* this should work for most non-highmem platforms */
133 zone = pgdata->node_zones;
135 return __add_pages(nid, zone, start_pfn, nr_pages);
137 #endif /* CONFIG_MEMORY_HOTPLUG */
140 * walk_memory_resource() needs to make sure there is no holes in a given
141 * memory range. PPC64 does not maintain the memory layout in /proc/iomem.
142 * Instead it maintains it in lmb.memory structures. Walk through the
143 * memory regions, find holes and callback for contiguous regions.
146 walk_memory_resource(unsigned long start_pfn, unsigned long nr_pages, void *arg,
147 int (*func)(unsigned long, unsigned long, void *))
149 struct lmb_property res;
150 unsigned long pfn, len;
154 res.base = (u64) start_pfn << PAGE_SHIFT;
155 res.size = (u64) nr_pages << PAGE_SHIFT;
157 end = res.base + res.size - 1;
158 while ((res.base < end) && (lmb_find(&res) >= 0)) {
159 pfn = (unsigned long)(res.base >> PAGE_SHIFT);
160 len = (unsigned long)(res.size >> PAGE_SHIFT);
161 ret = (*func)(pfn, len, arg);
164 res.base += (res.size + 1);
165 res.size = (end - res.base + 1);
169 EXPORT_SYMBOL_GPL(walk_memory_resource);
172 * Initialize the bootmem system and give it all the memory we
173 * have available. If we are using highmem, we only put the
174 * lowmem into the bootmem system.
176 #ifndef CONFIG_NEED_MULTIPLE_NODES
177 void __init do_init_bootmem(void)
180 unsigned long start, bootmap_pages;
181 unsigned long total_pages;
184 max_low_pfn = max_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
185 total_pages = (lmb_end_of_DRAM() - memstart_addr) >> PAGE_SHIFT;
186 #ifdef CONFIG_HIGHMEM
187 total_pages = total_lowmem >> PAGE_SHIFT;
188 max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
192 * Find an area to use for the bootmem bitmap. Calculate the size of
193 * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
194 * Add 1 additional page in case the address isn't page-aligned.
196 bootmap_pages = bootmem_bootmap_pages(total_pages);
198 start = lmb_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
200 min_low_pfn = MEMORY_START >> PAGE_SHIFT;
201 boot_mapsize = init_bootmem_node(NODE_DATA(0), start >> PAGE_SHIFT, min_low_pfn, max_low_pfn);
203 /* Add active regions with valid PFNs */
204 for (i = 0; i < lmb.memory.cnt; i++) {
205 unsigned long start_pfn, end_pfn;
206 start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
207 end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
208 add_active_range(0, start_pfn, end_pfn);
211 /* Add all physical memory to the bootmem map, mark each area
214 #ifdef CONFIG_HIGHMEM
215 free_bootmem_with_active_regions(0, lowmem_end_addr >> PAGE_SHIFT);
217 /* reserve the sections we're already using */
218 for (i = 0; i < lmb.reserved.cnt; i++) {
219 unsigned long addr = lmb.reserved.region[i].base +
220 lmb_size_bytes(&lmb.reserved, i) - 1;
221 if (addr < lowmem_end_addr)
222 reserve_bootmem(lmb.reserved.region[i].base,
223 lmb_size_bytes(&lmb.reserved, i),
225 else if (lmb.reserved.region[i].base < lowmem_end_addr) {
226 unsigned long adjusted_size = lowmem_end_addr -
227 lmb.reserved.region[i].base;
228 reserve_bootmem(lmb.reserved.region[i].base,
229 adjusted_size, BOOTMEM_DEFAULT);
233 free_bootmem_with_active_regions(0, max_pfn);
235 /* reserve the sections we're already using */
236 for (i = 0; i < lmb.reserved.cnt; i++)
237 reserve_bootmem(lmb.reserved.region[i].base,
238 lmb_size_bytes(&lmb.reserved, i),
242 /* XXX need to clip this if using highmem? */
243 sparse_memory_present_with_active_regions(0);
245 init_bootmem_done = 1;
248 /* mark pages that don't exist as nosave */
249 static int __init mark_nonram_nosave(void)
251 unsigned long lmb_next_region_start_pfn,
255 for (i = 0; i < lmb.memory.cnt - 1; i++) {
257 (lmb.memory.region[i].base >> PAGE_SHIFT) +
258 (lmb.memory.region[i].size >> PAGE_SHIFT);
259 lmb_next_region_start_pfn =
260 lmb.memory.region[i+1].base >> PAGE_SHIFT;
262 if (lmb_region_max_pfn < lmb_next_region_start_pfn)
263 register_nosave_region(lmb_region_max_pfn,
264 lmb_next_region_start_pfn);
271 * paging_init() sets up the page tables - in fact we've already done this.
273 void __init paging_init(void)
275 unsigned long total_ram = lmb_phys_mem_size();
276 phys_addr_t top_of_ram = lmb_end_of_DRAM();
277 unsigned long max_zone_pfns[MAX_NR_ZONES];
280 unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
281 unsigned long end = __fix_to_virt(FIX_HOLE);
283 for (; v < end; v += PAGE_SIZE)
284 map_page(v, 0, 0); /* XXX gross */
287 #ifdef CONFIG_HIGHMEM
288 map_page(PKMAP_BASE, 0, 0); /* XXX gross */
289 pkmap_page_table = virt_to_kpte(PKMAP_BASE);
291 kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
292 kmap_prot = PAGE_KERNEL;
293 #endif /* CONFIG_HIGHMEM */
295 printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%lx\n",
296 (unsigned long long)top_of_ram, total_ram);
297 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
298 (long int)((top_of_ram - total_ram) >> 20));
299 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
300 #ifdef CONFIG_HIGHMEM
301 max_zone_pfns[ZONE_DMA] = lowmem_end_addr >> PAGE_SHIFT;
302 max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
304 max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
306 free_area_init_nodes(max_zone_pfns);
308 mark_nonram_nosave();
310 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
312 void __init mem_init(void)
314 #ifdef CONFIG_NEED_MULTIPLE_NODES
320 unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
322 num_physpages = lmb.memory.size >> PAGE_SHIFT;
323 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
325 #ifdef CONFIG_NEED_MULTIPLE_NODES
326 for_each_online_node(nid) {
327 if (NODE_DATA(nid)->node_spanned_pages != 0) {
328 printk("freeing bootmem node %d\n", nid);
330 free_all_bootmem_node(NODE_DATA(nid));
335 totalram_pages += free_all_bootmem();
337 for_each_online_pgdat(pgdat) {
338 for (i = 0; i < pgdat->node_spanned_pages; i++) {
339 if (!pfn_valid(pgdat->node_start_pfn + i))
341 page = pgdat_page_nr(pgdat, i);
342 if (PageReserved(page))
347 codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
348 datasize = (unsigned long)&_edata - (unsigned long)&_sdata;
349 initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
350 bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
352 #ifdef CONFIG_HIGHMEM
354 unsigned long pfn, highmem_mapnr;
356 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
357 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
358 struct page *page = pfn_to_page(pfn);
359 if (lmb_is_reserved(pfn << PAGE_SHIFT))
361 ClearPageReserved(page);
362 init_page_count(page);
367 totalram_pages += totalhigh_pages;
368 printk(KERN_DEBUG "High memory: %luk\n",
369 totalhigh_pages << (PAGE_SHIFT-10));
371 #endif /* CONFIG_HIGHMEM */
373 printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
374 "%luk reserved, %luk data, %luk bss, %luk init)\n",
375 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
376 num_physpages << (PAGE_SHIFT-10),
378 reservedpages << (PAGE_SHIFT-10),
387 * This is called when a page has been modified by the kernel.
388 * It just marks the page as not i-cache clean. We do the i-cache
389 * flush later when the page is given to a user process, if necessary.
391 void flush_dcache_page(struct page *page)
393 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
395 /* avoid an atomic op if possible */
396 if (test_bit(PG_arch_1, &page->flags))
397 clear_bit(PG_arch_1, &page->flags);
399 EXPORT_SYMBOL(flush_dcache_page);
401 void flush_dcache_icache_page(struct page *page)
404 void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
405 __flush_dcache_icache(start);
406 kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
407 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
408 /* On 8xx there is no need to kmap since highmem is not supported */
409 __flush_dcache_icache(page_address(page));
411 __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
415 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
420 * We shouldnt have to do this, but some versions of glibc
421 * require it (ld.so assumes zero filled pages are icache clean)
424 flush_dcache_page(pg);
426 EXPORT_SYMBOL(clear_user_page);
428 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
431 copy_page(vto, vfrom);
434 * We should be able to use the following optimisation, however
435 * there are two problems.
436 * Firstly a bug in some versions of binutils meant PLT sections
437 * were not marked executable.
438 * Secondly the first word in the GOT section is blrl, used
439 * to establish the GOT address. Until recently the GOT was
440 * not marked executable.
444 if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
448 flush_dcache_page(pg);
451 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
452 unsigned long addr, int len)
456 maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
457 flush_icache_range(maddr, maddr + len);
460 EXPORT_SYMBOL(flush_icache_user_range);
463 * This is called at the end of handling a user page fault, when the
464 * fault has been handled by updating a PTE in the linux page tables.
465 * We use it to preload an HPTE into the hash table corresponding to
466 * the updated linux PTE.
468 * This must always be called with the pte lock held.
470 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
473 #ifdef CONFIG_PPC_STD_MMU
474 unsigned long access = 0, trap;
476 unsigned long pfn = pte_pfn(pte);
478 /* handle i-cache coherency */
479 if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
480 !cpu_has_feature(CPU_FTR_NOEXECUTE) &&
482 struct page *page = pfn_to_page(pfn);
484 /* On 8xx, cache control instructions (particularly
485 * "dcbst" from flush_dcache_icache) fault as write
486 * operation if there is an unpopulated TLB entry
487 * for the address in question. To workaround that,
488 * we invalidate the TLB here, thus avoiding dcbst
491 _tlbil_va(address, 0 /* 8xx doesn't care about PID */);
493 /* The _PAGE_USER test should really be _PAGE_EXEC, but
494 * older glibc versions execute some code from no-exec
495 * pages, which for now we are supporting. If exec-only
496 * pages are ever implemented, this will have to change.
498 if (!PageReserved(page) && (pte_val(pte) & _PAGE_USER)
499 && !test_bit(PG_arch_1, &page->flags)) {
500 if (vma->vm_mm == current->active_mm) {
501 __flush_dcache_icache((void *) address);
503 flush_dcache_icache_page(page);
504 set_bit(PG_arch_1, &page->flags);
508 #ifdef CONFIG_PPC_STD_MMU
509 /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
510 if (!pte_young(pte) || address >= TASK_SIZE)
513 /* We try to figure out if we are coming from an instruction
514 * access fault and pass that down to __hash_page so we avoid
515 * double-faulting on execution of fresh text. We have to test
516 * for regs NULL since init will get here first thing at boot
518 * We also avoid filling the hash if not coming from a fault
520 if (current->thread.regs == NULL)
522 trap = TRAP(current->thread.regs);
524 access |= _PAGE_EXEC;
525 else if (trap != 0x300)
527 hash_preload(vma->vm_mm, address, access, trap);
528 #endif /* CONFIG_PPC_STD_MMU */