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>
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>
42 #include <asm/mmu_context.h>
43 #include <asm/pgtable.h>
46 #include <asm/machdep.h>
47 #include <asm/btext.h>
49 #include <asm/sections.h>
50 #include <asm/sparsemem.h>
52 #include <asm/fixmap.h>
53 #include <asm/swiotlb.h>
58 #ifndef CPU_FTR_COHERENT_ICACHE
59 #define CPU_FTR_COHERENT_ICACHE 0 /* XXX for now */
60 #define CPU_FTR_NOEXECUTE 0
63 int init_bootmem_done;
65 unsigned long long memory_limit;
69 EXPORT_SYMBOL(kmap_pte);
71 EXPORT_SYMBOL(kmap_prot);
73 static inline pte_t *virt_to_kpte(unsigned long vaddr)
75 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
76 vaddr), vaddr), vaddr);
80 int page_is_ram(unsigned long pfn)
82 #ifndef CONFIG_PPC64 /* XXX for now */
85 unsigned long paddr = (pfn << PAGE_SHIFT);
86 struct memblock_region *reg;
88 for_each_memblock(memory, reg)
89 if (paddr >= reg->base && paddr < (reg->base + reg->size))
95 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
96 unsigned long size, pgprot_t vma_prot)
98 if (ppc_md.phys_mem_access_prot)
99 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
101 if (!page_is_ram(pfn))
102 vma_prot = pgprot_noncached(vma_prot);
106 EXPORT_SYMBOL(phys_mem_access_prot);
108 #ifdef CONFIG_MEMORY_HOTPLUG
111 int memory_add_physaddr_to_nid(u64 start)
113 return hot_add_scn_to_nid(start);
117 int arch_add_memory(int nid, u64 start, u64 size)
119 struct pglist_data *pgdata;
121 unsigned long start_pfn = start >> PAGE_SHIFT;
122 unsigned long nr_pages = size >> PAGE_SHIFT;
124 pgdata = NODE_DATA(nid);
126 start = (unsigned long)__va(start);
127 if (create_section_mapping(start, start + size))
130 /* this should work for most non-highmem platforms */
131 zone = pgdata->node_zones;
133 return __add_pages(nid, zone, start_pfn, nr_pages);
136 #ifdef CONFIG_MEMORY_HOTREMOVE
137 int arch_remove_memory(u64 start, u64 size)
139 unsigned long start_pfn = start >> PAGE_SHIFT;
140 unsigned long nr_pages = size >> PAGE_SHIFT;
144 zone = page_zone(pfn_to_page(start_pfn));
145 ret = __remove_pages(zone, start_pfn, nr_pages);
146 if (!ret && (ppc_md.remove_memory))
147 ret = ppc_md.remove_memory(start, size);
152 #endif /* CONFIG_MEMORY_HOTPLUG */
155 * walk_memory_resource() needs to make sure there is no holes in a given
156 * memory range. PPC64 does not maintain the memory layout in /proc/iomem.
157 * Instead it maintains it in memblock.memory structures. Walk through the
158 * memory regions, find holes and callback for contiguous regions.
161 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
162 void *arg, int (*func)(unsigned long, unsigned long, void *))
164 struct memblock_region *reg;
165 unsigned long end_pfn = start_pfn + nr_pages;
166 unsigned long tstart, tend;
169 for_each_memblock(memory, reg) {
170 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
171 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
174 ret = (*func)(tstart, tend - tstart, arg);
180 EXPORT_SYMBOL_GPL(walk_system_ram_range);
183 * Initialize the bootmem system and give it all the memory we
184 * have available. If we are using highmem, we only put the
185 * lowmem into the bootmem system.
187 #ifndef CONFIG_NEED_MULTIPLE_NODES
188 void __init do_init_bootmem(void)
190 unsigned long start, bootmap_pages;
191 unsigned long total_pages;
192 struct memblock_region *reg;
195 max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
196 total_pages = (memblock_end_of_DRAM() - memstart_addr) >> PAGE_SHIFT;
197 #ifdef CONFIG_HIGHMEM
198 total_pages = total_lowmem >> PAGE_SHIFT;
199 max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
203 * Find an area to use for the bootmem bitmap. Calculate the size of
204 * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
205 * Add 1 additional page in case the address isn't page-aligned.
207 bootmap_pages = bootmem_bootmap_pages(total_pages);
209 start = memblock_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
211 min_low_pfn = MEMORY_START >> PAGE_SHIFT;
212 boot_mapsize = init_bootmem_node(NODE_DATA(0), start >> PAGE_SHIFT, min_low_pfn, max_low_pfn);
214 /* Place all memblock_regions in the same node and merge contiguous
217 memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
219 /* Add all physical memory to the bootmem map, mark each area
222 #ifdef CONFIG_HIGHMEM
223 free_bootmem_with_active_regions(0, lowmem_end_addr >> PAGE_SHIFT);
225 /* reserve the sections we're already using */
226 for_each_memblock(reserved, reg) {
227 unsigned long top = reg->base + reg->size - 1;
228 if (top < lowmem_end_addr)
229 reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
230 else if (reg->base < lowmem_end_addr) {
231 unsigned long trunc_size = lowmem_end_addr - reg->base;
232 reserve_bootmem(reg->base, trunc_size, BOOTMEM_DEFAULT);
236 free_bootmem_with_active_regions(0, max_pfn);
238 /* reserve the sections we're already using */
239 for_each_memblock(reserved, reg)
240 reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
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 struct memblock_region *reg, *prev = NULL;
253 for_each_memblock(memory, reg) {
255 memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
256 register_nosave_region(memblock_region_memory_end_pfn(prev),
257 memblock_region_memory_base_pfn(reg));
262 #else /* CONFIG_NEED_MULTIPLE_NODES */
263 static int __init mark_nonram_nosave(void)
269 static bool zone_limits_final;
271 static unsigned long max_zone_pfns[MAX_NR_ZONES] = {
272 [0 ... MAX_NR_ZONES - 1] = ~0UL
276 * Restrict the specified zone and all more restrictive zones
277 * to be below the specified pfn. May not be called after
280 void __init limit_zone_pfn(enum zone_type zone, unsigned long pfn_limit)
284 if (WARN_ON(zone_limits_final))
287 for (i = zone; i >= 0; i--) {
288 if (max_zone_pfns[i] > pfn_limit)
289 max_zone_pfns[i] = pfn_limit;
294 * Find the least restrictive zone that is entirely below the
295 * specified pfn limit. Returns < 0 if no suitable zone is found.
297 * pfn_limit must be u64 because it can exceed 32 bits even on 32-bit
298 * systems -- the DMA limit can be higher than any possible real pfn.
300 int dma_pfn_limit_to_zone(u64 pfn_limit)
302 enum zone_type top_zone = ZONE_NORMAL;
305 #ifdef CONFIG_HIGHMEM
306 top_zone = ZONE_HIGHMEM;
309 for (i = top_zone; i >= 0; i--) {
310 if (max_zone_pfns[i] <= pfn_limit)
318 * paging_init() sets up the page tables - in fact we've already done this.
320 void __init paging_init(void)
322 unsigned long long total_ram = memblock_phys_mem_size();
323 phys_addr_t top_of_ram = memblock_end_of_DRAM();
324 enum zone_type top_zone;
327 unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
328 unsigned long end = __fix_to_virt(FIX_HOLE);
330 for (; v < end; v += PAGE_SIZE)
331 map_page(v, 0, 0); /* XXX gross */
334 #ifdef CONFIG_HIGHMEM
335 map_page(PKMAP_BASE, 0, 0); /* XXX gross */
336 pkmap_page_table = virt_to_kpte(PKMAP_BASE);
338 kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
339 kmap_prot = PAGE_KERNEL;
340 #endif /* CONFIG_HIGHMEM */
342 printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
343 (unsigned long long)top_of_ram, total_ram);
344 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
345 (long int)((top_of_ram - total_ram) >> 20));
347 #ifdef CONFIG_HIGHMEM
348 top_zone = ZONE_HIGHMEM;
349 limit_zone_pfn(ZONE_NORMAL, lowmem_end_addr >> PAGE_SHIFT);
351 top_zone = ZONE_NORMAL;
354 limit_zone_pfn(top_zone, top_of_ram >> PAGE_SHIFT);
355 zone_limits_final = true;
356 free_area_init_nodes(max_zone_pfns);
358 mark_nonram_nosave();
361 static void __init register_page_bootmem_info(void)
365 for_each_online_node(i)
366 register_page_bootmem_info_node(NODE_DATA(i));
369 void __init mem_init(void)
372 * book3s is limited to 16 page sizes due to encoding this in
373 * a 4-bit field for slices.
375 BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
377 #ifdef CONFIG_SWIOTLB
381 register_page_bootmem_info();
382 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
383 set_max_mapnr(max_pfn);
386 #ifdef CONFIG_HIGHMEM
388 unsigned long pfn, highmem_mapnr;
390 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
391 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
392 phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
393 struct page *page = pfn_to_page(pfn);
394 if (!memblock_is_reserved(paddr))
395 free_highmem_page(page);
398 #endif /* CONFIG_HIGHMEM */
400 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
402 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
403 * functions.... do it here for the non-smp case.
405 per_cpu(next_tlbcam_idx, smp_processor_id()) =
406 (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
409 mem_init_print_info(NULL);
411 pr_info("Kernel virtual memory layout:\n");
412 pr_info(" * 0x%08lx..0x%08lx : fixmap\n", FIXADDR_START, FIXADDR_TOP);
413 #ifdef CONFIG_HIGHMEM
414 pr_info(" * 0x%08lx..0x%08lx : highmem PTEs\n",
415 PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
416 #endif /* CONFIG_HIGHMEM */
417 #ifdef CONFIG_NOT_COHERENT_CACHE
418 pr_info(" * 0x%08lx..0x%08lx : consistent mem\n",
419 IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
420 #endif /* CONFIG_NOT_COHERENT_CACHE */
421 pr_info(" * 0x%08lx..0x%08lx : early ioremap\n",
422 ioremap_bot, IOREMAP_TOP);
423 pr_info(" * 0x%08lx..0x%08lx : vmalloc & ioremap\n",
424 VMALLOC_START, VMALLOC_END);
425 #endif /* CONFIG_PPC32 */
430 void free_initmem(void)
432 ppc_md.progress = ppc_printk_progress;
433 free_initmem_default(POISON_FREE_INITMEM);
436 #ifdef CONFIG_BLK_DEV_INITRD
437 void __init free_initrd_mem(unsigned long start, unsigned long end)
439 free_reserved_area((void *)start, (void *)end, -1, "initrd");
444 * This is called when a page has been modified by the kernel.
445 * It just marks the page as not i-cache clean. We do the i-cache
446 * flush later when the page is given to a user process, if necessary.
448 void flush_dcache_page(struct page *page)
450 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
452 /* avoid an atomic op if possible */
453 if (test_bit(PG_arch_1, &page->flags))
454 clear_bit(PG_arch_1, &page->flags);
456 EXPORT_SYMBOL(flush_dcache_page);
458 void flush_dcache_icache_page(struct page *page)
460 #ifdef CONFIG_HUGETLB_PAGE
461 if (PageCompound(page)) {
462 flush_dcache_icache_hugepage(page);
468 void *start = kmap_atomic(page);
469 __flush_dcache_icache(start);
470 kunmap_atomic(start);
472 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
473 /* On 8xx there is no need to kmap since highmem is not supported */
474 __flush_dcache_icache(page_address(page));
476 __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
479 EXPORT_SYMBOL(flush_dcache_icache_page);
481 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
486 * We shouldn't have to do this, but some versions of glibc
487 * require it (ld.so assumes zero filled pages are icache clean)
490 flush_dcache_page(pg);
492 EXPORT_SYMBOL(clear_user_page);
494 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
497 copy_page(vto, vfrom);
500 * We should be able to use the following optimisation, however
501 * there are two problems.
502 * Firstly a bug in some versions of binutils meant PLT sections
503 * were not marked executable.
504 * Secondly the first word in the GOT section is blrl, used
505 * to establish the GOT address. Until recently the GOT was
506 * not marked executable.
510 if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
514 flush_dcache_page(pg);
517 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
518 unsigned long addr, int len)
522 maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
523 flush_icache_range(maddr, maddr + len);
526 EXPORT_SYMBOL(flush_icache_user_range);
529 * This is called at the end of handling a user page fault, when the
530 * fault has been handled by updating a PTE in the linux page tables.
531 * We use it to preload an HPTE into the hash table corresponding to
532 * the updated linux PTE.
534 * This must always be called with the pte lock held.
536 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
539 #ifdef CONFIG_PPC_STD_MMU
541 * We don't need to worry about _PAGE_PRESENT here because we are
542 * called with either mm->page_table_lock held or ptl lock held
544 unsigned long access = 0, trap;
546 /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
547 if (!pte_young(*ptep) || address >= TASK_SIZE)
550 /* We try to figure out if we are coming from an instruction
551 * access fault and pass that down to __hash_page so we avoid
552 * double-faulting on execution of fresh text. We have to test
553 * for regs NULL since init will get here first thing at boot
555 * We also avoid filling the hash if not coming from a fault
557 if (current->thread.regs == NULL)
559 trap = TRAP(current->thread.regs);
561 access |= _PAGE_EXEC;
562 else if (trap != 0x300)
564 hash_preload(vma->vm_mm, address, access, trap);
565 #endif /* CONFIG_PPC_STD_MMU */
566 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
567 && defined(CONFIG_HUGETLB_PAGE)
568 if (is_vm_hugetlb_page(vma))
569 book3e_hugetlb_preload(vma, address, *ptep);
574 * System memory should not be in /proc/iomem but various tools expect it
577 static int __init add_system_ram_resources(void)
579 struct memblock_region *reg;
581 for_each_memblock(memory, reg) {
582 struct resource *res;
583 unsigned long base = reg->base;
584 unsigned long size = reg->size;
586 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
590 res->name = "System RAM";
592 res->end = base + size - 1;
593 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
594 WARN_ON(request_resource(&iomem_resource, res) < 0);
600 subsys_initcall(add_system_ram_resources);
602 #ifdef CONFIG_STRICT_DEVMEM
604 * devmem_is_allowed(): check to see if /dev/mem access to a certain address
605 * is valid. The argument is a physical page number.
607 * Access has to be given to non-kernel-ram areas as well, these contain the
608 * PCI mmio resources as well as potential bios/acpi data regions.
610 int devmem_is_allowed(unsigned long pfn)
612 if (iomem_is_exclusive(pfn << PAGE_SHIFT))
614 if (!page_is_ram(pfn))
616 if (page_is_rtas_user_buf(pfn))
620 #endif /* CONFIG_STRICT_DEVMEM */