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>
42 #include <asm/pgalloc.h>
44 #include <asm/abs_addr.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>
67 unsigned long ioremap_bot = IMALLOC_BASE;
68 static unsigned long phbs_io_bot = PHBS_IO_BASE;
70 extern pgd_t swapper_pg_dir[];
71 extern struct task_struct *current_set[NR_CPUS];
73 extern pgd_t ioremap_dir[];
74 pgd_t * ioremap_pgd = (pgd_t *)&ioremap_dir;
76 unsigned long klimit = (unsigned long)_end;
78 unsigned long _SDR1=0;
81 /* max amount of RAM to use */
82 unsigned long __max_memory;
84 /* info on what we think the IO hole is */
85 unsigned long io_hole_start;
86 unsigned long io_hole_size;
90 unsigned long total = 0, reserved = 0;
91 unsigned long shared = 0, cached = 0;
96 printk("Mem-info:\n");
98 printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
99 for_each_pgdat(pgdat) {
100 for (i = 0; i < pgdat->node_spanned_pages; i++) {
101 page = pgdat->node_mem_map + i;
103 if (PageReserved(page))
105 else if (PageSwapCache(page))
107 else if (page_count(page))
108 shared += page_count(page) - 1;
111 printk("%ld pages of RAM\n", total);
112 printk("%ld reserved pages\n", reserved);
113 printk("%ld pages shared\n", shared);
114 printk("%ld pages swap cached\n", cached);
117 #ifdef CONFIG_PPC_ISERIES
119 void __iomem *ioremap(unsigned long addr, unsigned long size)
121 return (void __iomem *)addr;
124 extern void __iomem *__ioremap(unsigned long addr, unsigned long size,
127 return (void __iomem *)addr;
130 void iounmap(volatile void __iomem *addr)
138 * map_io_page currently only called by __ioremap
139 * map_io_page adds an entry to the ioremap page table
140 * and adds an entry to the HPT, possibly bolting it
142 static void map_io_page(unsigned long ea, unsigned long pa, int flags)
150 spin_lock(&ioremap_mm.page_table_lock);
151 pgdp = pgd_offset_i(ea);
152 pmdp = pmd_alloc(&ioremap_mm, pgdp, ea);
153 ptep = pte_alloc_kernel(&ioremap_mm, pmdp, ea);
155 pa = abs_to_phys(pa);
156 set_pte(ptep, pfn_pte(pa >> PAGE_SHIFT, __pgprot(flags)));
157 spin_unlock(&ioremap_mm.page_table_lock);
159 unsigned long va, vpn, hash, hpteg;
162 * If the mm subsystem is not fully up, we cannot create a
163 * linux page table entry for this mapping. Simply bolt an
164 * entry in the hardware page table.
166 vsid = get_kernel_vsid(ea);
167 va = (vsid << 28) | (ea & 0xFFFFFFF);
168 vpn = va >> PAGE_SHIFT;
170 hash = hpt_hash(vpn, 0);
172 hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
174 /* Panic if a pte grpup is full */
175 if (ppc_md.hpte_insert(hpteg, va, pa >> PAGE_SHIFT, 0,
176 _PAGE_NO_CACHE|_PAGE_GUARDED|PP_RWXX,
178 panic("map_io_page: could not insert mapping");
184 static void __iomem * __ioremap_com(unsigned long addr, unsigned long pa,
185 unsigned long ea, unsigned long size,
190 if ((flags & _PAGE_PRESENT) == 0)
191 flags |= pgprot_val(PAGE_KERNEL);
192 if (flags & (_PAGE_NO_CACHE | _PAGE_WRITETHRU))
193 flags |= _PAGE_GUARDED;
195 for (i = 0; i < size; i += PAGE_SIZE) {
196 map_io_page(ea+i, pa+i, flags);
199 return (void __iomem *) (ea + (addr & ~PAGE_MASK));
204 ioremap(unsigned long addr, unsigned long size)
206 return __ioremap(addr, size, _PAGE_NO_CACHE);
210 __ioremap(unsigned long addr, unsigned long size, unsigned long flags)
212 unsigned long pa, ea;
215 * Choose an address to map it to.
216 * Once the imalloc system is running, we use it.
217 * Before that, we map using addresses going
218 * up from ioremap_bot. imalloc will use
219 * the addresses from ioremap_bot through
220 * IMALLOC_END (0xE000001fffffffff)
223 pa = addr & PAGE_MASK;
224 size = PAGE_ALIGN(addr + size) - pa;
230 struct vm_struct *area;
231 area = im_get_free_area(size);
234 ea = (unsigned long)(area->addr);
240 return __ioremap_com(addr, pa, ea, size, flags);
243 #define IS_PAGE_ALIGNED(_val) ((_val) == ((_val) & PAGE_MASK))
245 int __ioremap_explicit(unsigned long pa, unsigned long ea,
246 unsigned long size, unsigned long flags)
248 struct vm_struct *area;
250 /* For now, require page-aligned values for pa, ea, and size */
251 if (!IS_PAGE_ALIGNED(pa) || !IS_PAGE_ALIGNED(ea) ||
252 !IS_PAGE_ALIGNED(size)) {
253 printk(KERN_ERR "unaligned value in %s\n", __FUNCTION__);
257 if (!mem_init_done) {
258 /* Two things to consider in this case:
259 * 1) No records will be kept (imalloc, etc) that the region
261 * 2) It won't be easy to iounmap() the region later (because
266 area = im_get_area(ea, size,
267 IM_REGION_UNUSED|IM_REGION_SUBSET|IM_REGION_EXISTS);
269 /* Expected when PHB-dlpar is in play */
272 if (ea != (unsigned long) area->addr) {
273 printk(KERN_ERR "unexpected addr return from im_get_area\n");
278 if (__ioremap_com(pa, pa, ea, size, flags) != (void *) ea) {
279 printk(KERN_ERR "__ioremap_com() returned unexpected addr\n");
286 static void unmap_im_area_pte(pmd_t *pmd, unsigned long address,
300 pte = pte_offset_kernel(pmd, address);
301 address &= ~PMD_MASK;
302 end = address + size;
308 page = ptep_get_and_clear(pte);
309 address += PAGE_SIZE;
313 if (pte_present(page))
315 printk(KERN_CRIT "Whee.. Swapped out page in kernel page table\n");
316 } while (address < end);
319 static void unmap_im_area_pmd(pgd_t *dir, unsigned long address,
333 pmd = pmd_offset(dir, address);
334 address &= ~PGDIR_MASK;
335 end = address + size;
336 if (end > PGDIR_SIZE)
340 unmap_im_area_pte(pmd, address, end - address);
341 address = (address + PMD_SIZE) & PMD_MASK;
343 } while (address < end);
347 * Unmap an IO region and remove it from imalloc'd list.
348 * Access to IO memory should be serialized by driver.
349 * This code is modeled after vmalloc code - unmap_vm_area()
351 * XXX what about calls before mem_init_done (ie python_countermeasures())
353 void iounmap(volatile void __iomem *token)
355 unsigned long address, start, end, size;
356 struct mm_struct *mm;
360 if (!mem_init_done) {
364 addr = (void *) ((unsigned long __force) token & PAGE_MASK);
366 if ((size = im_free(addr)) == 0) {
370 address = (unsigned long)addr;
372 end = address + size;
375 spin_lock(&mm->page_table_lock);
377 dir = pgd_offset_i(address);
378 flush_cache_vunmap(address, end);
380 unmap_im_area_pmd(dir, address, end - address);
381 address = (address + PGDIR_SIZE) & PGDIR_MASK;
383 } while (address && (address < end));
384 flush_tlb_kernel_range(start, end);
386 spin_unlock(&mm->page_table_lock);
390 static int iounmap_subset_regions(unsigned long addr, unsigned long size)
392 struct vm_struct *area;
394 /* Check whether subsets of this region exist */
395 area = im_get_area(addr, size, IM_REGION_SUPERSET);
400 iounmap((void __iomem *) area->addr);
401 area = im_get_area(addr, size,
408 int iounmap_explicit(volatile void __iomem *start, unsigned long size)
410 struct vm_struct *area;
414 addr = (unsigned long __force) start & PAGE_MASK;
416 /* Verify that the region either exists or is a subset of an existing
417 * region. In the latter case, split the parent region to create
420 area = im_get_area(addr, size,
421 IM_REGION_EXISTS | IM_REGION_SUBSET);
423 /* Determine whether subset regions exist. If so, unmap */
424 rc = iounmap_subset_regions(addr, size);
427 "%s() cannot unmap nonexistent range 0x%lx\n",
432 iounmap((void __iomem *) area->addr);
435 * FIXME! This can't be right:
437 * Maybe it should be "iounmap(area);"
444 void free_initmem(void)
448 addr = (unsigned long)__init_begin;
449 for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) {
450 ClearPageReserved(virt_to_page(addr));
451 set_page_count(virt_to_page(addr), 1);
455 printk ("Freeing unused kernel memory: %luk freed\n",
456 ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10);
459 #ifdef CONFIG_BLK_DEV_INITRD
460 void free_initrd_mem(unsigned long start, unsigned long end)
463 printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
464 for (; start < end; start += PAGE_SIZE) {
465 ClearPageReserved(virt_to_page(start));
466 set_page_count(virt_to_page(start), 1);
473 static DEFINE_SPINLOCK(mmu_context_lock);
474 static DEFINE_IDR(mmu_context_idr);
476 int init_new_context(struct task_struct *tsk, struct mm_struct *mm)
481 #ifdef CONFIG_HUGETLB_PAGE
482 /* We leave htlb_segs as it was, but for a fork, we need to
483 * clear the huge_pgdir. */
484 mm->context.huge_pgdir = NULL;
488 if (!idr_pre_get(&mmu_context_idr, GFP_KERNEL))
491 spin_lock(&mmu_context_lock);
492 err = idr_get_new_above(&mmu_context_idr, NULL, 1, &index);
493 spin_unlock(&mmu_context_lock);
500 if (index > MAX_CONTEXT) {
501 idr_remove(&mmu_context_idr, index);
505 mm->context.id = index;
510 void destroy_context(struct mm_struct *mm)
512 spin_lock(&mmu_context_lock);
513 idr_remove(&mmu_context_idr, mm->context.id);
514 spin_unlock(&mmu_context_lock);
516 mm->context.id = NO_CONTEXT;
518 hugetlb_mm_free_pgd(mm);
522 * Do very early mm setup.
524 void __init mm_init_ppc64(void)
526 #ifndef CONFIG_PPC_ISERIES
530 ppc64_boot_msg(0x100, "MM Init");
532 /* This is the story of the IO hole... please, keep seated,
533 * unfortunately, we are out of oxygen masks at the moment.
534 * So we need some rough way to tell where your big IO hole
535 * is. On pmac, it's between 2G and 4G, on POWER3, it's around
536 * that area as well, on POWER4 we don't have one, etc...
537 * We need that as a "hint" when sizing the TCE table on POWER3
538 * So far, the simplest way that seem work well enough for us it
539 * to just assume that the first discontinuity in our physical
540 * RAM layout is the IO hole. That may not be correct in the future
541 * (and isn't on iSeries but then we don't care ;)
544 #ifndef CONFIG_PPC_ISERIES
545 for (i = 1; i < lmb.memory.cnt; i++) {
546 unsigned long base, prevbase, prevsize;
548 prevbase = lmb.memory.region[i-1].physbase;
549 prevsize = lmb.memory.region[i-1].size;
550 base = lmb.memory.region[i].physbase;
551 if (base > (prevbase + prevsize)) {
552 io_hole_start = prevbase + prevsize;
553 io_hole_size = base - (prevbase + prevsize);
557 #endif /* CONFIG_PPC_ISERIES */
559 printk("IO Hole assumed to be %lx -> %lx\n",
560 io_hole_start, io_hole_start + io_hole_size - 1);
562 ppc64_boot_msg(0x100, "MM Init Done");
566 * This is called by /dev/mem to know if a given address has to
567 * be mapped non-cacheable or not
569 int page_is_ram(unsigned long pfn)
572 unsigned long paddr = (pfn << PAGE_SHIFT);
574 for (i=0; i < lmb.memory.cnt; i++) {
577 #ifdef CONFIG_MSCHUNKS
578 base = lmb.memory.region[i].physbase;
580 base = lmb.memory.region[i].base;
582 if ((paddr >= base) &&
583 (paddr < (base + lmb.memory.region[i].size))) {
590 EXPORT_SYMBOL(page_is_ram);
593 * Initialize the bootmem system and give it all the memory we
596 #ifndef CONFIG_DISCONTIGMEM
597 void __init do_init_bootmem(void)
600 unsigned long start, bootmap_pages;
601 unsigned long total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT;
605 * Find an area to use for the bootmem bitmap. Calculate the size of
606 * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
607 * Add 1 additional page in case the address isn't page-aligned.
609 bootmap_pages = bootmem_bootmap_pages(total_pages);
611 start = abs_to_phys(lmb_alloc(bootmap_pages<<PAGE_SHIFT, PAGE_SIZE));
614 boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
616 max_pfn = max_low_pfn;
618 /* add all physical memory to the bootmem map. Also find the first */
619 for (i=0; i < lmb.memory.cnt; i++) {
620 unsigned long physbase, size;
622 physbase = lmb.memory.region[i].physbase;
623 size = lmb.memory.region[i].size;
624 free_bootmem(physbase, size);
627 /* reserve the sections we're already using */
628 for (i=0; i < lmb.reserved.cnt; i++) {
629 unsigned long physbase = lmb.reserved.region[i].physbase;
630 unsigned long size = lmb.reserved.region[i].size;
632 reserve_bootmem(physbase, size);
637 * paging_init() sets up the page tables - in fact we've already done this.
639 void __init paging_init(void)
641 unsigned long zones_size[MAX_NR_ZONES];
642 unsigned long zholes_size[MAX_NR_ZONES];
643 unsigned long total_ram = lmb_phys_mem_size();
644 unsigned long top_of_ram = lmb_end_of_DRAM();
646 printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
647 top_of_ram, total_ram);
648 printk(KERN_INFO "Memory hole size: %ldMB\n",
649 (top_of_ram - total_ram) >> 20);
651 * All pages are DMA-able so we put them all in the DMA zone.
653 memset(zones_size, 0, sizeof(zones_size));
654 memset(zholes_size, 0, sizeof(zholes_size));
656 zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
657 zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
659 free_area_init_node(0, &contig_page_data, zones_size,
660 __pa(PAGE_OFFSET) >> PAGE_SHIFT, zholes_size);
661 mem_map = contig_page_data.node_mem_map;
663 #endif /* CONFIG_DISCONTIGMEM */
665 static struct kcore_list kcore_vmem;
667 static int __init setup_kcore(void)
671 for (i=0; i < lmb.memory.cnt; i++) {
672 unsigned long physbase, size;
673 struct kcore_list *kcore_mem;
675 physbase = lmb.memory.region[i].physbase;
676 size = lmb.memory.region[i].size;
678 /* GFP_ATOMIC to avoid might_sleep warnings during boot */
679 kcore_mem = kmalloc(sizeof(struct kcore_list), GFP_ATOMIC);
681 panic("mem_init: kmalloc failed\n");
683 kclist_add(kcore_mem, __va(physbase), size);
686 kclist_add(&kcore_vmem, (void *)VMALLOC_START, VMALLOC_END-VMALLOC_START);
690 module_init(setup_kcore);
692 void __init mem_init(void)
694 #ifdef CONFIG_DISCONTIGMEM
700 unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
702 num_physpages = max_low_pfn; /* RAM is assumed contiguous */
703 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
705 #ifdef CONFIG_DISCONTIGMEM
706 for_each_online_node(nid) {
707 if (NODE_DATA(nid)->node_spanned_pages != 0) {
708 printk("freeing bootmem node %x\n", nid);
710 free_all_bootmem_node(NODE_DATA(nid));
714 max_mapnr = num_physpages;
715 totalram_pages += free_all_bootmem();
718 for_each_pgdat(pgdat) {
719 for (i = 0; i < pgdat->node_spanned_pages; i++) {
720 page = pgdat->node_mem_map + i;
721 if (PageReserved(page))
726 codesize = (unsigned long)&_etext - (unsigned long)&_stext;
727 initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
728 datasize = (unsigned long)&_edata - (unsigned long)&__init_end;
729 bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
731 printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
732 "%luk reserved, %luk data, %luk bss, %luk init)\n",
733 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
734 num_physpages << (PAGE_SHIFT-10),
736 reservedpages << (PAGE_SHIFT-10),
743 #ifdef CONFIG_PPC_ISERIES
749 * This is called when a page has been modified by the kernel.
750 * It just marks the page as not i-cache clean. We do the i-cache
751 * flush later when the page is given to a user process, if necessary.
753 void flush_dcache_page(struct page *page)
755 if (cur_cpu_spec->cpu_features & CPU_FTR_COHERENT_ICACHE)
757 /* avoid an atomic op if possible */
758 if (test_bit(PG_arch_1, &page->flags))
759 clear_bit(PG_arch_1, &page->flags);
762 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
766 if (cur_cpu_spec->cpu_features & CPU_FTR_COHERENT_ICACHE)
769 * We shouldnt have to do this, but some versions of glibc
770 * require it (ld.so assumes zero filled pages are icache clean)
774 /* avoid an atomic op if possible */
775 if (test_bit(PG_arch_1, &pg->flags))
776 clear_bit(PG_arch_1, &pg->flags);
779 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
782 copy_page(vto, vfrom);
785 * We should be able to use the following optimisation, however
786 * there are two problems.
787 * Firstly a bug in some versions of binutils meant PLT sections
788 * were not marked executable.
789 * Secondly the first word in the GOT section is blrl, used
790 * to establish the GOT address. Until recently the GOT was
791 * not marked executable.
795 if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
799 if (cur_cpu_spec->cpu_features & CPU_FTR_COHERENT_ICACHE)
802 /* avoid an atomic op if possible */
803 if (test_bit(PG_arch_1, &pg->flags))
804 clear_bit(PG_arch_1, &pg->flags);
807 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
808 unsigned long addr, int len)
812 maddr = (unsigned long)page_address(page) + (addr & ~PAGE_MASK);
813 flush_icache_range(maddr, maddr + len);
817 * This is called at the end of handling a user page fault, when the
818 * fault has been handled by updating a PTE in the linux page tables.
819 * We use it to preload an HPTE into the hash table corresponding to
820 * the updated linux PTE.
822 * This must always be called with the mm->page_table_lock held
824 void update_mmu_cache(struct vm_area_struct *vma, unsigned long ea,
834 /* handle i-cache coherency */
835 if (!(cur_cpu_spec->cpu_features & CPU_FTR_COHERENT_ICACHE) &&
836 !(cur_cpu_spec->cpu_features & CPU_FTR_NOEXECUTE)) {
837 unsigned long pfn = pte_pfn(pte);
838 if (pfn_valid(pfn)) {
839 struct page *page = pfn_to_page(pfn);
840 if (!PageReserved(page)
841 && !test_bit(PG_arch_1, &page->flags)) {
842 __flush_dcache_icache(page_address(page));
843 set_bit(PG_arch_1, &page->flags);
848 /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
852 pgdir = vma->vm_mm->pgd;
856 ptep = find_linux_pte(pgdir, ea);
860 vsid = get_vsid(vma->vm_mm->context.id, ea);
862 local_irq_save(flags);
863 tmp = cpumask_of_cpu(smp_processor_id());
864 if (cpus_equal(vma->vm_mm->cpu_vm_mask, tmp))
867 __hash_page(ea, pte_val(pte) & (_PAGE_USER|_PAGE_RW), vsid, ptep,
869 local_irq_restore(flags);
872 void __iomem * reserve_phb_iospace(unsigned long size)
874 void __iomem *virt_addr;
876 if (phbs_io_bot >= IMALLOC_BASE)
877 panic("reserve_phb_iospace(): phb io space overflow\n");
879 virt_addr = (void __iomem *) phbs_io_bot;
885 kmem_cache_t *zero_cache;
887 static void zero_ctor(void *pte, kmem_cache_t *cache, unsigned long flags)
889 memset(pte, 0, PAGE_SIZE);
892 void pgtable_cache_init(void)
894 zero_cache = kmem_cache_create("zero",
897 SLAB_HWCACHE_ALIGN | SLAB_MUST_HWCACHE_ALIGN,
901 panic("pgtable_cache_init(): could not create zero_cache!\n");