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 * Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
10 * Derived from "arch/i386/mm/init.c"
11 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
13 * Dave Engebretsen <engebret@us.ibm.com>
14 * Rework for PPC64 port.
16 * This program is free software; you can redistribute it and/or
17 * modify it under the terms of the GNU General Public License
18 * as published by the Free Software Foundation; either version
19 * 2 of the License, or (at your option) any later version.
23 #include <linux/config.h>
24 #include <linux/signal.h>
25 #include <linux/sched.h>
26 #include <linux/kernel.h>
27 #include <linux/errno.h>
28 #include <linux/string.h>
29 #include <linux/types.h>
30 #include <linux/mman.h>
32 #include <linux/swap.h>
33 #include <linux/stddef.h>
34 #include <linux/vmalloc.h>
35 #include <linux/init.h>
36 #include <linux/delay.h>
37 #include <linux/bootmem.h>
38 #include <linux/highmem.h>
39 #include <linux/idr.h>
40 #include <linux/nodemask.h>
41 #include <linux/module.h>
43 #include <asm/pgalloc.h>
49 #include <asm/mmu_context.h>
50 #include <asm/pgtable.h>
52 #include <asm/uaccess.h>
54 #include <asm/machdep.h>
57 #include <asm/processor.h>
58 #include <asm/mmzone.h>
59 #include <asm/cputable.h>
60 #include <asm/ppcdebug.h>
61 #include <asm/sections.h>
62 #include <asm/system.h>
63 #include <asm/iommu.h>
64 #include <asm/abs_addr.h>
66 #include <asm/imalloc.h>
68 #if PGTABLE_RANGE > USER_VSID_RANGE
69 #warning Limited user VSID range means pagetable space is wasted
72 #if (TASK_SIZE_USER64 < PGTABLE_RANGE) && (TASK_SIZE_USER64 < USER_VSID_RANGE)
73 #warning TASK_SIZE is smaller than it needs to be.
77 unsigned long ioremap_bot = IMALLOC_BASE;
78 static unsigned long phbs_io_bot = PHBS_IO_BASE;
80 extern pgd_t swapper_pg_dir[];
81 extern struct task_struct *current_set[NR_CPUS];
83 unsigned long klimit = (unsigned long)_end;
85 unsigned long _SDR1=0;
88 /* max amount of RAM to use */
89 unsigned long __max_memory;
91 /* info on what we think the IO hole is */
92 unsigned long io_hole_start;
93 unsigned long io_hole_size;
97 unsigned long total = 0, reserved = 0;
98 unsigned long shared = 0, cached = 0;
103 printk("Mem-info:\n");
105 printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
106 for_each_pgdat(pgdat) {
107 for (i = 0; i < pgdat->node_spanned_pages; i++) {
108 page = pgdat_page_nr(pgdat, i);
110 if (PageReserved(page))
112 else if (PageSwapCache(page))
114 else if (page_count(page))
115 shared += page_count(page) - 1;
118 printk("%ld pages of RAM\n", total);
119 printk("%ld reserved pages\n", reserved);
120 printk("%ld pages shared\n", shared);
121 printk("%ld pages swap cached\n", cached);
124 #ifdef CONFIG_PPC_ISERIES
126 void __iomem *ioremap(unsigned long addr, unsigned long size)
128 return (void __iomem *)addr;
131 extern void __iomem *__ioremap(unsigned long addr, unsigned long size,
134 return (void __iomem *)addr;
137 void iounmap(volatile void __iomem *addr)
145 * map_io_page currently only called by __ioremap
146 * map_io_page adds an entry to the ioremap page table
147 * and adds an entry to the HPT, possibly bolting it
149 static int map_io_page(unsigned long ea, unsigned long pa, int flags)
158 spin_lock(&init_mm.page_table_lock);
159 pgdp = pgd_offset_k(ea);
160 pudp = pud_alloc(&init_mm, pgdp, ea);
163 pmdp = pmd_alloc(&init_mm, pudp, ea);
166 ptep = pte_alloc_kernel(&init_mm, pmdp, ea);
169 set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT,
171 spin_unlock(&init_mm.page_table_lock);
173 unsigned long va, vpn, hash, hpteg;
176 * If the mm subsystem is not fully up, we cannot create a
177 * linux page table entry for this mapping. Simply bolt an
178 * entry in the hardware page table.
180 vsid = get_kernel_vsid(ea);
181 va = (vsid << 28) | (ea & 0xFFFFFFF);
182 vpn = va >> PAGE_SHIFT;
184 hash = hpt_hash(vpn, 0);
186 hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
188 /* Panic if a pte grpup is full */
189 if (ppc_md.hpte_insert(hpteg, va, pa >> PAGE_SHIFT,
191 _PAGE_NO_CACHE|_PAGE_GUARDED|PP_RWXX)
193 panic("map_io_page: could not insert mapping");
200 static void __iomem * __ioremap_com(unsigned long addr, unsigned long pa,
201 unsigned long ea, unsigned long size,
206 if ((flags & _PAGE_PRESENT) == 0)
207 flags |= pgprot_val(PAGE_KERNEL);
209 for (i = 0; i < size; i += PAGE_SIZE)
210 if (map_io_page(ea+i, pa+i, flags))
213 return (void __iomem *) (ea + (addr & ~PAGE_MASK));
218 ioremap(unsigned long addr, unsigned long size)
220 return __ioremap(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED);
223 void __iomem * __ioremap(unsigned long addr, unsigned long size,
226 unsigned long pa, ea;
230 * Choose an address to map it to.
231 * Once the imalloc system is running, we use it.
232 * Before that, we map using addresses going
233 * up from ioremap_bot. imalloc will use
234 * the addresses from ioremap_bot through
238 pa = addr & PAGE_MASK;
239 size = PAGE_ALIGN(addr + size) - pa;
245 struct vm_struct *area;
246 area = im_get_free_area(size);
249 ea = (unsigned long)(area->addr);
250 ret = __ioremap_com(addr, pa, ea, size, flags);
255 ret = __ioremap_com(addr, pa, ea, size, flags);
262 #define IS_PAGE_ALIGNED(_val) ((_val) == ((_val) & PAGE_MASK))
264 int __ioremap_explicit(unsigned long pa, unsigned long ea,
265 unsigned long size, unsigned long flags)
267 struct vm_struct *area;
270 /* For now, require page-aligned values for pa, ea, and size */
271 if (!IS_PAGE_ALIGNED(pa) || !IS_PAGE_ALIGNED(ea) ||
272 !IS_PAGE_ALIGNED(size)) {
273 printk(KERN_ERR "unaligned value in %s\n", __FUNCTION__);
277 if (!mem_init_done) {
278 /* Two things to consider in this case:
279 * 1) No records will be kept (imalloc, etc) that the region
281 * 2) It won't be easy to iounmap() the region later (because
286 area = im_get_area(ea, size,
287 IM_REGION_UNUSED|IM_REGION_SUBSET|IM_REGION_EXISTS);
289 /* Expected when PHB-dlpar is in play */
292 if (ea != (unsigned long) area->addr) {
293 printk(KERN_ERR "unexpected addr return from "
299 ret = __ioremap_com(pa, pa, ea, size, flags);
301 printk(KERN_ERR "ioremap_explicit() allocation failure !\n");
304 if (ret != (void *) ea) {
305 printk(KERN_ERR "__ioremap_com() returned unexpected addr\n");
313 * Unmap an IO region and remove it from imalloc'd list.
314 * Access to IO memory should be serialized by driver.
315 * This code is modeled after vmalloc code - unmap_vm_area()
317 * XXX what about calls before mem_init_done (ie python_countermeasures())
319 void iounmap(volatile void __iomem *token)
326 addr = (void *) ((unsigned long __force) token & PAGE_MASK);
331 static int iounmap_subset_regions(unsigned long addr, unsigned long size)
333 struct vm_struct *area;
335 /* Check whether subsets of this region exist */
336 area = im_get_area(addr, size, IM_REGION_SUPERSET);
341 iounmap((void __iomem *) area->addr);
342 area = im_get_area(addr, size,
349 int iounmap_explicit(volatile void __iomem *start, unsigned long size)
351 struct vm_struct *area;
355 addr = (unsigned long __force) start & PAGE_MASK;
357 /* Verify that the region either exists or is a subset of an existing
358 * region. In the latter case, split the parent region to create
361 area = im_get_area(addr, size,
362 IM_REGION_EXISTS | IM_REGION_SUBSET);
364 /* Determine whether subset regions exist. If so, unmap */
365 rc = iounmap_subset_regions(addr, size);
368 "%s() cannot unmap nonexistent range 0x%lx\n",
373 iounmap((void __iomem *) area->addr);
376 * FIXME! This can't be right:
378 * Maybe it should be "iounmap(area);"
385 EXPORT_SYMBOL(ioremap);
386 EXPORT_SYMBOL(__ioremap);
387 EXPORT_SYMBOL(iounmap);
389 void free_initmem(void)
393 addr = (unsigned long)__init_begin;
394 for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) {
395 memset((void *)addr, 0xcc, PAGE_SIZE);
396 ClearPageReserved(virt_to_page(addr));
397 set_page_count(virt_to_page(addr), 1);
401 printk ("Freeing unused kernel memory: %luk freed\n",
402 ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10);
405 #ifdef CONFIG_BLK_DEV_INITRD
406 void free_initrd_mem(unsigned long start, unsigned long end)
409 printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
410 for (; start < end; start += PAGE_SIZE) {
411 ClearPageReserved(virt_to_page(start));
412 set_page_count(virt_to_page(start), 1);
419 static DEFINE_SPINLOCK(mmu_context_lock);
420 static DEFINE_IDR(mmu_context_idr);
422 int init_new_context(struct task_struct *tsk, struct mm_struct *mm)
428 if (!idr_pre_get(&mmu_context_idr, GFP_KERNEL))
431 spin_lock(&mmu_context_lock);
432 err = idr_get_new_above(&mmu_context_idr, NULL, 1, &index);
433 spin_unlock(&mmu_context_lock);
440 if (index > MAX_CONTEXT) {
441 idr_remove(&mmu_context_idr, index);
445 mm->context.id = index;
450 void destroy_context(struct mm_struct *mm)
452 spin_lock(&mmu_context_lock);
453 idr_remove(&mmu_context_idr, mm->context.id);
454 spin_unlock(&mmu_context_lock);
456 mm->context.id = NO_CONTEXT;
460 * Do very early mm setup.
462 void __init mm_init_ppc64(void)
464 #ifndef CONFIG_PPC_ISERIES
468 ppc64_boot_msg(0x100, "MM Init");
470 /* This is the story of the IO hole... please, keep seated,
471 * unfortunately, we are out of oxygen masks at the moment.
472 * So we need some rough way to tell where your big IO hole
473 * is. On pmac, it's between 2G and 4G, on POWER3, it's around
474 * that area as well, on POWER4 we don't have one, etc...
475 * We need that as a "hint" when sizing the TCE table on POWER3
476 * So far, the simplest way that seem work well enough for us it
477 * to just assume that the first discontinuity in our physical
478 * RAM layout is the IO hole. That may not be correct in the future
479 * (and isn't on iSeries but then we don't care ;)
482 #ifndef CONFIG_PPC_ISERIES
483 for (i = 1; i < lmb.memory.cnt; i++) {
484 unsigned long base, prevbase, prevsize;
486 prevbase = lmb.memory.region[i-1].base;
487 prevsize = lmb.memory.region[i-1].size;
488 base = lmb.memory.region[i].base;
489 if (base > (prevbase + prevsize)) {
490 io_hole_start = prevbase + prevsize;
491 io_hole_size = base - (prevbase + prevsize);
495 #endif /* CONFIG_PPC_ISERIES */
497 printk("IO Hole assumed to be %lx -> %lx\n",
498 io_hole_start, io_hole_start + io_hole_size - 1);
500 ppc64_boot_msg(0x100, "MM Init Done");
504 * This is called by /dev/mem to know if a given address has to
505 * be mapped non-cacheable or not
507 int page_is_ram(unsigned long pfn)
510 unsigned long paddr = (pfn << PAGE_SHIFT);
512 for (i=0; i < lmb.memory.cnt; i++) {
515 base = lmb.memory.region[i].base;
517 if ((paddr >= base) &&
518 (paddr < (base + lmb.memory.region[i].size))) {
525 EXPORT_SYMBOL(page_is_ram);
528 * Initialize the bootmem system and give it all the memory we
531 #ifndef CONFIG_NEED_MULTIPLE_NODES
532 void __init do_init_bootmem(void)
535 unsigned long start, bootmap_pages;
536 unsigned long total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT;
540 * Find an area to use for the bootmem bitmap. Calculate the size of
541 * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
542 * Add 1 additional page in case the address isn't page-aligned.
544 bootmap_pages = bootmem_bootmap_pages(total_pages);
546 start = lmb_alloc(bootmap_pages<<PAGE_SHIFT, PAGE_SIZE);
549 boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
551 max_pfn = max_low_pfn;
553 /* Add all physical memory to the bootmem map, mark each area
556 for (i=0; i < lmb.memory.cnt; i++)
557 free_bootmem(lmb.memory.region[i].base,
558 lmb_size_bytes(&lmb.memory, i));
560 /* reserve the sections we're already using */
561 for (i=0; i < lmb.reserved.cnt; i++)
562 reserve_bootmem(lmb.reserved.region[i].base,
563 lmb_size_bytes(&lmb.reserved, i));
565 for (i=0; i < lmb.memory.cnt; i++)
566 memory_present(0, lmb_start_pfn(&lmb.memory, i),
567 lmb_end_pfn(&lmb.memory, i));
571 * paging_init() sets up the page tables - in fact we've already done this.
573 void __init paging_init(void)
575 unsigned long zones_size[MAX_NR_ZONES];
576 unsigned long zholes_size[MAX_NR_ZONES];
577 unsigned long total_ram = lmb_phys_mem_size();
578 unsigned long top_of_ram = lmb_end_of_DRAM();
580 printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
581 top_of_ram, total_ram);
582 printk(KERN_INFO "Memory hole size: %ldMB\n",
583 (top_of_ram - total_ram) >> 20);
585 * All pages are DMA-able so we put them all in the DMA zone.
587 memset(zones_size, 0, sizeof(zones_size));
588 memset(zholes_size, 0, sizeof(zholes_size));
590 zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
591 zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
593 free_area_init_node(0, NODE_DATA(0), zones_size,
594 __pa(PAGE_OFFSET) >> PAGE_SHIFT, zholes_size);
596 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
598 static struct kcore_list kcore_vmem;
600 static int __init setup_kcore(void)
604 for (i=0; i < lmb.memory.cnt; i++) {
605 unsigned long base, size;
606 struct kcore_list *kcore_mem;
608 base = lmb.memory.region[i].base;
609 size = lmb.memory.region[i].size;
611 /* GFP_ATOMIC to avoid might_sleep warnings during boot */
612 kcore_mem = kmalloc(sizeof(struct kcore_list), GFP_ATOMIC);
614 panic("mem_init: kmalloc failed\n");
616 kclist_add(kcore_mem, __va(base), size);
619 kclist_add(&kcore_vmem, (void *)VMALLOC_START, VMALLOC_END-VMALLOC_START);
623 module_init(setup_kcore);
625 void __init mem_init(void)
627 #ifdef CONFIG_NEED_MULTIPLE_NODES
633 unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
635 num_physpages = max_low_pfn; /* RAM is assumed contiguous */
636 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
638 #ifdef CONFIG_NEED_MULTIPLE_NODES
639 for_each_online_node(nid) {
640 if (NODE_DATA(nid)->node_spanned_pages != 0) {
641 printk("freeing bootmem node %x\n", nid);
643 free_all_bootmem_node(NODE_DATA(nid));
647 max_mapnr = num_physpages;
648 totalram_pages += free_all_bootmem();
651 for_each_pgdat(pgdat) {
652 for (i = 0; i < pgdat->node_spanned_pages; i++) {
653 page = pgdat_page_nr(pgdat, i);
654 if (PageReserved(page))
659 codesize = (unsigned long)&_etext - (unsigned long)&_stext;
660 initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
661 datasize = (unsigned long)&_edata - (unsigned long)&__init_end;
662 bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
664 printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
665 "%luk reserved, %luk data, %luk bss, %luk init)\n",
666 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
667 num_physpages << (PAGE_SHIFT-10),
669 reservedpages << (PAGE_SHIFT-10),
676 /* Initialize the vDSO */
681 * This is called when a page has been modified by the kernel.
682 * It just marks the page as not i-cache clean. We do the i-cache
683 * flush later when the page is given to a user process, if necessary.
685 void flush_dcache_page(struct page *page)
687 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
689 /* avoid an atomic op if possible */
690 if (test_bit(PG_arch_1, &page->flags))
691 clear_bit(PG_arch_1, &page->flags);
693 EXPORT_SYMBOL(flush_dcache_page);
695 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
699 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
702 * We shouldnt have to do this, but some versions of glibc
703 * require it (ld.so assumes zero filled pages are icache clean)
707 /* avoid an atomic op if possible */
708 if (test_bit(PG_arch_1, &pg->flags))
709 clear_bit(PG_arch_1, &pg->flags);
711 EXPORT_SYMBOL(clear_user_page);
713 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
716 copy_page(vto, vfrom);
719 * We should be able to use the following optimisation, however
720 * there are two problems.
721 * Firstly a bug in some versions of binutils meant PLT sections
722 * were not marked executable.
723 * Secondly the first word in the GOT section is blrl, used
724 * to establish the GOT address. Until recently the GOT was
725 * not marked executable.
729 if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
733 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
736 /* avoid an atomic op if possible */
737 if (test_bit(PG_arch_1, &pg->flags))
738 clear_bit(PG_arch_1, &pg->flags);
741 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
742 unsigned long addr, int len)
746 maddr = (unsigned long)page_address(page) + (addr & ~PAGE_MASK);
747 flush_icache_range(maddr, maddr + len);
749 EXPORT_SYMBOL(flush_icache_user_range);
752 * This is called at the end of handling a user page fault, when the
753 * fault has been handled by updating a PTE in the linux page tables.
754 * We use it to preload an HPTE into the hash table corresponding to
755 * the updated linux PTE.
757 * This must always be called with the mm->page_table_lock held
759 void update_mmu_cache(struct vm_area_struct *vma, unsigned long ea,
769 /* handle i-cache coherency */
770 if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
771 !cpu_has_feature(CPU_FTR_NOEXECUTE)) {
772 unsigned long pfn = pte_pfn(pte);
773 if (pfn_valid(pfn)) {
774 struct page *page = pfn_to_page(pfn);
775 if (!PageReserved(page)
776 && !test_bit(PG_arch_1, &page->flags)) {
777 __flush_dcache_icache(page_address(page));
778 set_bit(PG_arch_1, &page->flags);
783 /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
787 pgdir = vma->vm_mm->pgd;
791 ptep = find_linux_pte(pgdir, ea);
795 vsid = get_vsid(vma->vm_mm->context.id, ea);
797 local_irq_save(flags);
798 tmp = cpumask_of_cpu(smp_processor_id());
799 if (cpus_equal(vma->vm_mm->cpu_vm_mask, tmp))
802 __hash_page(ea, pte_val(pte) & (_PAGE_USER|_PAGE_RW), vsid, ptep,
804 local_irq_restore(flags);
807 void __iomem * reserve_phb_iospace(unsigned long size)
809 void __iomem *virt_addr;
811 if (phbs_io_bot >= IMALLOC_BASE)
812 panic("reserve_phb_iospace(): phb io space overflow\n");
814 virt_addr = (void __iomem *) phbs_io_bot;
820 static void zero_ctor(void *addr, kmem_cache_t *cache, unsigned long flags)
822 memset(addr, 0, kmem_cache_size(cache));
825 static const int pgtable_cache_size[2] = {
826 PTE_TABLE_SIZE, PMD_TABLE_SIZE
828 static const char *pgtable_cache_name[ARRAY_SIZE(pgtable_cache_size)] = {
829 "pgd_pte_cache", "pud_pmd_cache",
832 kmem_cache_t *pgtable_cache[ARRAY_SIZE(pgtable_cache_size)];
834 void pgtable_cache_init(void)
838 BUILD_BUG_ON(PTE_TABLE_SIZE != pgtable_cache_size[PTE_CACHE_NUM]);
839 BUILD_BUG_ON(PMD_TABLE_SIZE != pgtable_cache_size[PMD_CACHE_NUM]);
840 BUILD_BUG_ON(PUD_TABLE_SIZE != pgtable_cache_size[PUD_CACHE_NUM]);
841 BUILD_BUG_ON(PGD_TABLE_SIZE != pgtable_cache_size[PGD_CACHE_NUM]);
843 for (i = 0; i < ARRAY_SIZE(pgtable_cache_size); i++) {
844 int size = pgtable_cache_size[i];
845 const char *name = pgtable_cache_name[i];
847 pgtable_cache[i] = kmem_cache_create(name,
850 | SLAB_MUST_HWCACHE_ALIGN,
853 if (! pgtable_cache[i])
854 panic("pgtable_cache_init(): could not create %s!\n",
859 pgprot_t phys_mem_access_prot(struct file *file, unsigned long addr,
860 unsigned long size, pgprot_t vma_prot)
862 if (ppc_md.phys_mem_access_prot)
863 return ppc_md.phys_mem_access_prot(file, addr, size, vma_prot);
865 if (!page_is_ram(addr >> PAGE_SHIFT))
866 vma_prot = __pgprot(pgprot_val(vma_prot)
867 | _PAGE_GUARDED | _PAGE_NO_CACHE);
870 EXPORT_SYMBOL(phys_mem_access_prot);