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).
9 * PPC44x/36-bit changes by Matt Porter (mporter@mvista.com)
11 * Derived from "arch/i386/mm/init.c"
12 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
14 * This program is free software; you can redistribute it and/or
15 * modify it under the terms of the GNU General Public License
16 * as published by the Free Software Foundation; either version
17 * 2 of the License, or (at your option) any later version.
21 #include <linux/config.h>
22 #include <linux/module.h>
23 #include <linux/sched.h>
24 #include <linux/kernel.h>
25 #include <linux/errno.h>
26 #include <linux/string.h>
27 #include <linux/types.h>
29 #include <linux/stddef.h>
30 #include <linux/init.h>
31 #include <linux/bootmem.h>
32 #include <linux/highmem.h>
33 #include <linux/initrd.h>
34 #include <linux/pagemap.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/bootinfo.h>
48 #include "mem_pieces.h"
51 #if defined(CONFIG_KERNEL_START_BOOL) || defined(CONFIG_LOWMEM_SIZE_BOOL)
52 /* The ammount of lowmem must be within 0xF0000000 - KERNELBASE. */
53 #if (CONFIG_LOWMEM_SIZE > (0xF0000000 - KERNELBASE))
54 #error "You must adjust CONFIG_LOWMEM_SIZE or CONFIG_START_KERNEL"
57 #define MAX_LOW_MEM CONFIG_LOWMEM_SIZE
59 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
61 unsigned long total_memory;
62 unsigned long total_lowmem;
64 unsigned long ppc_memstart;
65 unsigned long ppc_memoffset = PAGE_OFFSET;
68 int init_bootmem_done;
72 extern char etext[], _stext[];
73 extern char __init_begin, __init_end;
79 EXPORT_SYMBOL(kmap_prot);
80 EXPORT_SYMBOL(kmap_pte);
84 void set_phys_avail(unsigned long total_ram);
86 /* XXX should be in current.h -- paulus */
87 extern struct task_struct *current_set[NR_CPUS];
90 struct mem_pieces phys_avail;
93 * this tells the system to map all of ram with the segregs
94 * (i.e. page tables) instead of the bats.
97 int __map_without_bats;
98 int __map_without_ltlbs;
100 /* max amount of RAM to use */
101 unsigned long __max_memory;
102 /* max amount of low RAM to map in */
103 unsigned long __max_low_memory = MAX_LOW_MEM;
107 int i,free = 0,total = 0,reserved = 0;
108 int shared = 0, cached = 0;
111 printk("Mem-info:\n");
113 printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
117 if (PageHighMem(mem_map+i))
119 if (PageReserved(mem_map+i))
121 else if (PageSwapCache(mem_map+i))
123 else if (!page_count(mem_map+i))
126 shared += page_count(mem_map+i) - 1;
128 printk("%d pages of RAM\n",total);
129 printk("%d pages of HIGHMEM\n", highmem);
130 printk("%d free pages\n",free);
131 printk("%d reserved pages\n",reserved);
132 printk("%d pages shared\n",shared);
133 printk("%d pages swap cached\n",cached);
136 /* Free up now-unused memory */
137 static void free_sec(unsigned long start, unsigned long end, const char *name)
139 unsigned long cnt = 0;
141 while (start < end) {
142 ClearPageReserved(virt_to_page(start));
143 set_page_count(virt_to_page(start), 1);
149 printk(" %ldk %s", cnt << (PAGE_SHIFT - 10), name);
150 totalram_pages += cnt;
154 void free_initmem(void)
156 #define FREESEC(TYPE) \
157 free_sec((unsigned long)(&__ ## TYPE ## _begin), \
158 (unsigned long)(&__ ## TYPE ## _end), \
161 printk ("Freeing unused kernel memory:");
164 ppc_md.progress = NULL;
168 #ifdef CONFIG_BLK_DEV_INITRD
169 void free_initrd_mem(unsigned long start, unsigned long end)
171 printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
173 for (; start < end; start += PAGE_SIZE) {
174 ClearPageReserved(virt_to_page(start));
175 set_page_count(virt_to_page(start), 1);
183 * Check for command-line options that affect what MMU_init will do.
187 /* Check for nobats option (used in mapin_ram). */
188 if (strstr(cmd_line, "nobats")) {
189 __map_without_bats = 1;
192 if (strstr(cmd_line, "noltlbs")) {
193 __map_without_ltlbs = 1;
196 /* Look for mem= option on command line */
197 if (strstr(cmd_line, "mem=")) {
199 unsigned long maxmem = 0;
201 for (q = cmd_line; (p = strstr(q, "mem=")) != 0; ) {
203 if (p > cmd_line && p[-1] != ' ')
205 maxmem = simple_strtoul(q, &q, 0);
206 if (*q == 'k' || *q == 'K') {
209 } else if (*q == 'm' || *q == 'M') {
214 __max_memory = maxmem;
219 * MMU_init sets up the basic memory mappings for the kernel,
220 * including both RAM and possibly some I/O regions,
221 * and sets up the page tables and the MMU hardware ready to go.
223 void __init MMU_init(void)
226 ppc_md.progress("MMU:enter", 0x111);
228 /* parse args from command line */
232 * Figure out how much memory we have, how much
233 * is lowmem, and how much is highmem. If we were
234 * passed the total memory size from the bootloader,
238 total_memory = boot_mem_size;
240 total_memory = ppc_md.find_end_of_memory();
242 if (__max_memory && total_memory > __max_memory)
243 total_memory = __max_memory;
244 total_lowmem = total_memory;
245 #ifdef CONFIG_FSL_BOOKE
246 /* Freescale Book-E parts expect lowmem to be mapped by fixed TLB
247 * entries, so we need to adjust lowmem to match the amount we can map
248 * in the fixed entries */
249 adjust_total_lowmem();
250 #endif /* CONFIG_FSL_BOOKE */
251 if (total_lowmem > __max_low_memory) {
252 total_lowmem = __max_low_memory;
253 #ifndef CONFIG_HIGHMEM
254 total_memory = total_lowmem;
255 #endif /* CONFIG_HIGHMEM */
257 set_phys_avail(total_lowmem);
259 /* Initialize the MMU hardware */
261 ppc_md.progress("MMU:hw init", 0x300);
264 /* Map in all of RAM starting at KERNELBASE */
266 ppc_md.progress("MMU:mapin", 0x301);
269 #ifdef CONFIG_HIGHMEM
270 ioremap_base = PKMAP_BASE;
272 ioremap_base = 0xfe000000UL; /* for now, could be 0xfffff000 */
273 #endif /* CONFIG_HIGHMEM */
274 ioremap_bot = ioremap_base;
276 /* Map in I/O resources */
278 ppc_md.progress("MMU:setio", 0x302);
279 if (ppc_md.setup_io_mappings)
280 ppc_md.setup_io_mappings();
282 /* Initialize the context management stuff */
286 ppc_md.progress("MMU:exit", 0x211);
288 #ifdef CONFIG_BOOTX_TEXT
289 /* By default, we are no longer mapped */
290 boot_text_mapped = 0;
291 /* Must be done last, or ppc_md.progress will die. */
296 /* This is only called until mem_init is done. */
297 void __init *early_get_page(void)
301 if (init_bootmem_done) {
302 p = alloc_bootmem_pages(PAGE_SIZE);
304 p = mem_pieces_find(PAGE_SIZE, PAGE_SIZE);
310 * Initialize the bootmem system and give it all the memory we
313 void __init do_init_bootmem(void)
315 unsigned long start, size;
319 * Find an area to use for the bootmem bitmap.
320 * We look for the first area which is at least
321 * 128kB in length (128kB is enough for a bitmap
322 * for 4GB of memory, using 4kB pages), plus 1 page
323 * (in case the address isn't page-aligned).
327 for (i = 0; i < phys_avail.n_regions; ++i) {
328 unsigned long a = phys_avail.regions[i].address;
329 unsigned long s = phys_avail.regions[i].size;
334 if (s >= 33 * PAGE_SIZE)
337 start = PAGE_ALIGN(start);
339 min_low_pfn = start >> PAGE_SHIFT;
340 max_low_pfn = (PPC_MEMSTART + total_lowmem) >> PAGE_SHIFT;
341 max_pfn = (PPC_MEMSTART + total_memory) >> PAGE_SHIFT;
342 boot_mapsize = init_bootmem_node(&contig_page_data, min_low_pfn,
343 PPC_MEMSTART >> PAGE_SHIFT,
346 /* remove the bootmem bitmap from the available memory */
347 mem_pieces_remove(&phys_avail, start, boot_mapsize, 1);
349 /* add everything in phys_avail into the bootmem map */
350 for (i = 0; i < phys_avail.n_regions; ++i)
351 free_bootmem(phys_avail.regions[i].address,
352 phys_avail.regions[i].size);
354 init_bootmem_done = 1;
358 * paging_init() sets up the page tables - in fact we've already done this.
360 void __init paging_init(void)
362 unsigned long zones_size[MAX_NR_ZONES], i;
364 #ifdef CONFIG_HIGHMEM
365 map_page(PKMAP_BASE, 0, 0); /* XXX gross */
366 pkmap_page_table = pte_offset_kernel(pmd_offset(pgd_offset_k
367 (PKMAP_BASE), PKMAP_BASE), PKMAP_BASE);
368 map_page(KMAP_FIX_BEGIN, 0, 0); /* XXX gross */
369 kmap_pte = pte_offset_kernel(pmd_offset(pgd_offset_k
370 (KMAP_FIX_BEGIN), KMAP_FIX_BEGIN), KMAP_FIX_BEGIN);
371 kmap_prot = PAGE_KERNEL;
372 #endif /* CONFIG_HIGHMEM */
375 * All pages are DMA-able so we put them all in the DMA zone.
377 zones_size[ZONE_DMA] = total_lowmem >> PAGE_SHIFT;
378 for (i = 1; i < MAX_NR_ZONES; i++)
381 #ifdef CONFIG_HIGHMEM
382 zones_size[ZONE_HIGHMEM] = (total_memory - total_lowmem) >> PAGE_SHIFT;
383 #endif /* CONFIG_HIGHMEM */
385 free_area_init(zones_size);
388 void __init mem_init(void)
394 #ifdef CONFIG_HIGHMEM
395 unsigned long highmem_mapnr;
397 highmem_mapnr = total_lowmem >> PAGE_SHIFT;
398 #endif /* CONFIG_HIGHMEM */
399 max_mapnr = total_memory >> PAGE_SHIFT;
401 high_memory = (void *) __va(PPC_MEMSTART + total_lowmem);
402 num_physpages = max_mapnr; /* RAM is assumed contiguous */
404 totalram_pages += free_all_bootmem();
406 #ifdef CONFIG_BLK_DEV_INITRD
407 /* if we are booted from BootX with an initial ramdisk,
408 make sure the ramdisk pages aren't reserved. */
410 for (addr = initrd_start; addr < initrd_end; addr += PAGE_SIZE)
411 ClearPageReserved(virt_to_page(addr));
413 #endif /* CONFIG_BLK_DEV_INITRD */
416 /* mark the RTAS pages as reserved */
418 for (addr = (ulong)__va(rtas_data);
419 addr < PAGE_ALIGN((ulong)__va(rtas_data)+rtas_size) ;
421 SetPageReserved(virt_to_page(addr));
423 for (addr = PAGE_OFFSET; addr < (unsigned long)high_memory;
425 if (!PageReserved(virt_to_page(addr)))
427 if (addr < (ulong) etext)
429 else if (addr >= (unsigned long)&__init_begin
430 && addr < (unsigned long)&__init_end)
432 else if (addr < (ulong) klimit)
436 #ifdef CONFIG_HIGHMEM
440 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
441 struct page *page = mem_map + pfn;
443 ClearPageReserved(page);
444 set_page_count(page, 1);
448 totalram_pages += totalhigh_pages;
450 #endif /* CONFIG_HIGHMEM */
452 printk("Memory: %luk available (%dk kernel code, %dk data, %dk init, %ldk highmem)\n",
453 (unsigned long)nr_free_pages()<< (PAGE_SHIFT-10),
454 codepages<< (PAGE_SHIFT-10), datapages<< (PAGE_SHIFT-10),
455 initpages<< (PAGE_SHIFT-10),
456 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10)));
462 * Set phys_avail to the amount of physical memory,
463 * less the kernel text/data/bss.
466 set_phys_avail(unsigned long total_memory)
468 unsigned long kstart, ksize;
471 * Initially, available physical memory is equivalent to all
475 phys_avail.regions[0].address = PPC_MEMSTART;
476 phys_avail.regions[0].size = total_memory;
477 phys_avail.n_regions = 1;
480 * Map out the kernel text/data/bss from the available physical
484 kstart = __pa(_stext); /* should be 0 */
485 ksize = PAGE_ALIGN(klimit - _stext);
487 mem_pieces_remove(&phys_avail, kstart, ksize, 0);
488 mem_pieces_remove(&phys_avail, 0, 0x4000, 0);
490 #if defined(CONFIG_BLK_DEV_INITRD)
491 /* Remove the init RAM disk from the available memory. */
493 mem_pieces_remove(&phys_avail, __pa(initrd_start),
494 initrd_end - initrd_start, 1);
496 #endif /* CONFIG_BLK_DEV_INITRD */
498 /* remove the RTAS pages from the available memory */
500 mem_pieces_remove(&phys_avail, rtas_data, rtas_size, 1);
504 /* Mark some memory as reserved by removing it from phys_avail. */
505 void __init reserve_phys_mem(unsigned long start, unsigned long size)
507 mem_pieces_remove(&phys_avail, start, size, 1);
511 * This is called when a page has been modified by the kernel.
512 * It just marks the page as not i-cache clean. We do the i-cache
513 * flush later when the page is given to a user process, if necessary.
515 void flush_dcache_page(struct page *page)
517 clear_bit(PG_arch_1, &page->flags);
520 void flush_dcache_icache_page(struct page *page)
523 void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
524 __flush_dcache_icache(start);
525 kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
526 #elif defined(CONFIG_8xx)
527 /* On 8xx there is no need to kmap since highmem is not supported */
528 __flush_dcache_icache(page_address(page));
530 __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
534 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
537 clear_bit(PG_arch_1, &pg->flags);
540 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
543 copy_page(vto, vfrom);
544 clear_bit(PG_arch_1, &pg->flags);
547 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
548 unsigned long addr, int len)
552 maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
553 flush_icache_range(maddr, maddr + len);
558 * This is called at the end of handling a user page fault, when the
559 * fault has been handled by updating a PTE in the linux page tables.
560 * We use it to preload an HPTE into the hash table corresponding to
561 * the updated linux PTE.
563 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
566 /* handle i-cache coherency */
567 unsigned long pfn = pte_pfn(pte);
569 if (pfn_valid(pfn)) {
570 struct page *page = pfn_to_page(pfn);
572 /* On 8xx, the TLB handlers work in 2 stages:
573 * First, a zeroed entry is loaded by TLBMiss handler,
574 * which causes the TLBError handler to be triggered.
575 * That means the zeroed TLB has to be invalidated
576 * whenever a page miss occurs.
580 if (!PageReserved(page)
581 && !test_bit(PG_arch_1, &page->flags)) {
582 if (vma->vm_mm == current->active_mm)
583 __flush_dcache_icache((void *) address);
585 flush_dcache_icache_page(page);
586 set_bit(PG_arch_1, &page->flags);
590 #ifdef CONFIG_PPC_STD_MMU
591 /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
592 if (Hash != 0 && pte_young(pte)) {
593 struct mm_struct *mm;
596 mm = (address < TASK_SIZE)? vma->vm_mm: &init_mm;
597 pmd = pmd_offset(pgd_offset(mm, address), address);
599 add_hash_page(mm->context, address, pmd_val(*pmd));
605 * This is called by /dev/mem to know if a given address has to
606 * be mapped non-cacheable or not
608 int page_is_ram(unsigned long pfn)
610 return pfn < max_pfn;
613 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
614 unsigned long size, pgprot_t vma_prot)
616 if (ppc_md.phys_mem_access_prot)
617 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
619 if (!page_is_ram(pfn))
620 vma_prot = __pgprot(pgprot_val(vma_prot)
621 | _PAGE_GUARDED | _PAGE_NO_CACHE);
624 EXPORT_SYMBOL(phys_mem_access_prot);