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/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>
38 #include <asm/pgalloc.h>
41 #include <asm/mmu_context.h>
42 #include <asm/pgtable.h>
45 #include <asm/machdep.h>
46 #include <asm/btext.h>
48 #include <asm/sections.h>
49 #include <asm/sparsemem.h>
51 #include <asm/fixmap.h>
52 #include <asm/swiotlb.h>
56 #ifndef CPU_FTR_COHERENT_ICACHE
57 #define CPU_FTR_COHERENT_ICACHE 0 /* XXX for now */
58 #define CPU_FTR_NOEXECUTE 0
61 int init_bootmem_done
;
63 phys_addr_t memory_limit
;
69 EXPORT_SYMBOL(kmap_prot
);
70 EXPORT_SYMBOL(kmap_pte
);
72 static inline pte_t
*virt_to_kpte(unsigned long vaddr
)
74 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr
),
75 vaddr
), vaddr
), vaddr
);
79 int page_is_ram(unsigned long pfn
)
81 #ifndef CONFIG_PPC64 /* XXX for now */
84 unsigned long paddr
= (pfn
<< PAGE_SHIFT
);
85 struct memblock_region
*reg
;
87 for_each_memblock(memory
, reg
)
88 if (paddr
>= reg
->base
&& paddr
< (reg
->base
+ reg
->size
))
94 pgprot_t
phys_mem_access_prot(struct file
*file
, unsigned long pfn
,
95 unsigned long size
, pgprot_t vma_prot
)
97 if (ppc_md
.phys_mem_access_prot
)
98 return ppc_md
.phys_mem_access_prot(file
, pfn
, size
, vma_prot
);
100 if (!page_is_ram(pfn
))
101 vma_prot
= pgprot_noncached(vma_prot
);
105 EXPORT_SYMBOL(phys_mem_access_prot
);
107 #ifdef CONFIG_MEMORY_HOTPLUG
110 int memory_add_physaddr_to_nid(u64 start
)
112 return hot_add_scn_to_nid(start
);
116 int arch_add_memory(int nid
, u64 start
, u64 size
)
118 struct pglist_data
*pgdata
;
120 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
121 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
123 pgdata
= NODE_DATA(nid
);
125 start
= (unsigned long)__va(start
);
126 create_section_mapping(start
, start
+ size
);
128 /* this should work for most non-highmem platforms */
129 zone
= pgdata
->node_zones
;
131 return __add_pages(nid
, zone
, start_pfn
, nr_pages
);
133 #endif /* CONFIG_MEMORY_HOTPLUG */
136 * walk_memory_resource() needs to make sure there is no holes in a given
137 * memory range. PPC64 does not maintain the memory layout in /proc/iomem.
138 * Instead it maintains it in memblock.memory structures. Walk through the
139 * memory regions, find holes and callback for contiguous regions.
142 walk_system_ram_range(unsigned long start_pfn
, unsigned long nr_pages
,
143 void *arg
, int (*func
)(unsigned long, unsigned long, void *))
145 struct memblock_region
*reg
;
146 unsigned long end_pfn
= start_pfn
+ nr_pages
;
147 unsigned long tstart
, tend
;
150 for_each_memblock(memory
, reg
) {
151 tstart
= max(start_pfn
, memblock_region_memory_base_pfn(reg
));
152 tend
= min(end_pfn
, memblock_region_memory_end_pfn(reg
));
155 ret
= (*func
)(tstart
, tend
- tstart
, arg
);
161 EXPORT_SYMBOL_GPL(walk_system_ram_range
);
164 * Initialize the bootmem system and give it all the memory we
165 * have available. If we are using highmem, we only put the
166 * lowmem into the bootmem system.
168 #ifndef CONFIG_NEED_MULTIPLE_NODES
169 void __init
do_init_bootmem(void)
171 unsigned long start
, bootmap_pages
;
172 unsigned long total_pages
;
173 struct memblock_region
*reg
;
176 max_low_pfn
= max_pfn
= memblock_end_of_DRAM() >> PAGE_SHIFT
;
177 total_pages
= (memblock_end_of_DRAM() - memstart_addr
) >> PAGE_SHIFT
;
178 #ifdef CONFIG_HIGHMEM
179 total_pages
= total_lowmem
>> PAGE_SHIFT
;
180 max_low_pfn
= lowmem_end_addr
>> PAGE_SHIFT
;
184 * Find an area to use for the bootmem bitmap. Calculate the size of
185 * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
186 * Add 1 additional page in case the address isn't page-aligned.
188 bootmap_pages
= bootmem_bootmap_pages(total_pages
);
190 start
= memblock_alloc(bootmap_pages
<< PAGE_SHIFT
, PAGE_SIZE
);
192 min_low_pfn
= MEMORY_START
>> PAGE_SHIFT
;
193 boot_mapsize
= init_bootmem_node(NODE_DATA(0), start
>> PAGE_SHIFT
, min_low_pfn
, max_low_pfn
);
195 /* Add active regions with valid PFNs */
196 for_each_memblock(memory
, reg
) {
197 unsigned long start_pfn
, end_pfn
;
198 start_pfn
= memblock_region_memory_base_pfn(reg
);
199 end_pfn
= memblock_region_memory_end_pfn(reg
);
200 add_active_range(0, start_pfn
, end_pfn
);
203 /* Add all physical memory to the bootmem map, mark each area
206 #ifdef CONFIG_HIGHMEM
207 free_bootmem_with_active_regions(0, lowmem_end_addr
>> PAGE_SHIFT
);
209 /* reserve the sections we're already using */
210 for_each_memblock(reserved
, reg
) {
211 unsigned long top
= reg
->base
+ reg
->size
- 1;
212 if (top
< lowmem_end_addr
)
213 reserve_bootmem(reg
->base
, reg
->size
, BOOTMEM_DEFAULT
);
214 else if (reg
->base
< lowmem_end_addr
) {
215 unsigned long trunc_size
= lowmem_end_addr
- reg
->base
;
216 reserve_bootmem(reg
->base
, trunc_size
, BOOTMEM_DEFAULT
);
220 free_bootmem_with_active_regions(0, max_pfn
);
222 /* reserve the sections we're already using */
223 for_each_memblock(reserved
, reg
)
224 reserve_bootmem(reg
->base
, reg
->size
, BOOTMEM_DEFAULT
);
226 /* XXX need to clip this if using highmem? */
227 sparse_memory_present_with_active_regions(0);
229 init_bootmem_done
= 1;
232 /* mark pages that don't exist as nosave */
233 static int __init
mark_nonram_nosave(void)
235 struct memblock_region
*reg
, *prev
= NULL
;
237 for_each_memblock(memory
, reg
) {
239 memblock_region_memory_end_pfn(prev
) < memblock_region_memory_base_pfn(reg
))
240 register_nosave_region(memblock_region_memory_end_pfn(prev
),
241 memblock_region_memory_base_pfn(reg
));
248 * paging_init() sets up the page tables - in fact we've already done this.
250 void __init
paging_init(void)
252 unsigned long total_ram
= memblock_phys_mem_size();
253 phys_addr_t top_of_ram
= memblock_end_of_DRAM();
254 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
257 unsigned long v
= __fix_to_virt(__end_of_fixed_addresses
- 1);
258 unsigned long end
= __fix_to_virt(FIX_HOLE
);
260 for (; v
< end
; v
+= PAGE_SIZE
)
261 map_page(v
, 0, 0); /* XXX gross */
264 #ifdef CONFIG_HIGHMEM
265 map_page(PKMAP_BASE
, 0, 0); /* XXX gross */
266 pkmap_page_table
= virt_to_kpte(PKMAP_BASE
);
268 kmap_pte
= virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN
));
269 kmap_prot
= PAGE_KERNEL
;
270 #endif /* CONFIG_HIGHMEM */
272 printk(KERN_DEBUG
"Top of RAM: 0x%llx, Total RAM: 0x%lx\n",
273 (unsigned long long)top_of_ram
, total_ram
);
274 printk(KERN_DEBUG
"Memory hole size: %ldMB\n",
275 (long int)((top_of_ram
- total_ram
) >> 20));
276 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
277 #ifdef CONFIG_HIGHMEM
278 max_zone_pfns
[ZONE_DMA
] = lowmem_end_addr
>> PAGE_SHIFT
;
279 max_zone_pfns
[ZONE_HIGHMEM
] = top_of_ram
>> PAGE_SHIFT
;
281 max_zone_pfns
[ZONE_DMA
] = top_of_ram
>> PAGE_SHIFT
;
283 free_area_init_nodes(max_zone_pfns
);
285 mark_nonram_nosave();
287 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
289 void __init
mem_init(void)
291 #ifdef CONFIG_NEED_MULTIPLE_NODES
297 unsigned long reservedpages
= 0, codesize
, initsize
, datasize
, bsssize
;
299 #ifdef CONFIG_SWIOTLB
300 if (ppc_swiotlb_enable
)
304 num_physpages
= memblock_phys_mem_size() >> PAGE_SHIFT
;
305 high_memory
= (void *) __va(max_low_pfn
* PAGE_SIZE
);
307 #ifdef CONFIG_NEED_MULTIPLE_NODES
308 for_each_online_node(nid
) {
309 if (NODE_DATA(nid
)->node_spanned_pages
!= 0) {
310 printk("freeing bootmem node %d\n", nid
);
312 free_all_bootmem_node(NODE_DATA(nid
));
317 totalram_pages
+= free_all_bootmem();
319 for_each_online_pgdat(pgdat
) {
320 for (i
= 0; i
< pgdat
->node_spanned_pages
; i
++) {
321 if (!pfn_valid(pgdat
->node_start_pfn
+ i
))
323 page
= pgdat_page_nr(pgdat
, i
);
324 if (PageReserved(page
))
329 codesize
= (unsigned long)&_sdata
- (unsigned long)&_stext
;
330 datasize
= (unsigned long)&_edata
- (unsigned long)&_sdata
;
331 initsize
= (unsigned long)&__init_end
- (unsigned long)&__init_begin
;
332 bsssize
= (unsigned long)&__bss_stop
- (unsigned long)&__bss_start
;
334 #ifdef CONFIG_HIGHMEM
336 unsigned long pfn
, highmem_mapnr
;
338 highmem_mapnr
= lowmem_end_addr
>> PAGE_SHIFT
;
339 for (pfn
= highmem_mapnr
; pfn
< max_mapnr
; ++pfn
) {
340 struct page
*page
= pfn_to_page(pfn
);
341 if (memblock_is_reserved(pfn
<< PAGE_SHIFT
))
343 ClearPageReserved(page
);
344 init_page_count(page
);
349 totalram_pages
+= totalhigh_pages
;
350 printk(KERN_DEBUG
"High memory: %luk\n",
351 totalhigh_pages
<< (PAGE_SHIFT
-10));
353 #endif /* CONFIG_HIGHMEM */
355 printk(KERN_INFO
"Memory: %luk/%luk available (%luk kernel code, "
356 "%luk reserved, %luk data, %luk bss, %luk init)\n",
357 nr_free_pages() << (PAGE_SHIFT
-10),
358 num_physpages
<< (PAGE_SHIFT
-10),
360 reservedpages
<< (PAGE_SHIFT
-10),
366 pr_info("Kernel virtual memory layout:\n");
367 pr_info(" * 0x%08lx..0x%08lx : fixmap\n", FIXADDR_START
, FIXADDR_TOP
);
368 #ifdef CONFIG_HIGHMEM
369 pr_info(" * 0x%08lx..0x%08lx : highmem PTEs\n",
370 PKMAP_BASE
, PKMAP_ADDR(LAST_PKMAP
));
371 #endif /* CONFIG_HIGHMEM */
372 #ifdef CONFIG_NOT_COHERENT_CACHE
373 pr_info(" * 0x%08lx..0x%08lx : consistent mem\n",
374 IOREMAP_TOP
, IOREMAP_TOP
+ CONFIG_CONSISTENT_SIZE
);
375 #endif /* CONFIG_NOT_COHERENT_CACHE */
376 pr_info(" * 0x%08lx..0x%08lx : early ioremap\n",
377 ioremap_bot
, IOREMAP_TOP
);
378 pr_info(" * 0x%08lx..0x%08lx : vmalloc & ioremap\n",
379 VMALLOC_START
, VMALLOC_END
);
380 #endif /* CONFIG_PPC32 */
385 #ifdef CONFIG_BLK_DEV_INITRD
386 void __init
free_initrd_mem(unsigned long start
, unsigned long end
)
391 start
= _ALIGN_DOWN(start
, PAGE_SIZE
);
392 end
= _ALIGN_UP(end
, PAGE_SIZE
);
393 pr_info("Freeing initrd memory: %ldk freed\n", (end
- start
) >> 10);
395 for (; start
< end
; start
+= PAGE_SIZE
) {
396 ClearPageReserved(virt_to_page(start
));
397 init_page_count(virt_to_page(start
));
405 * This is called when a page has been modified by the kernel.
406 * It just marks the page as not i-cache clean. We do the i-cache
407 * flush later when the page is given to a user process, if necessary.
409 void flush_dcache_page(struct page
*page
)
411 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE
))
413 /* avoid an atomic op if possible */
414 if (test_bit(PG_arch_1
, &page
->flags
))
415 clear_bit(PG_arch_1
, &page
->flags
);
417 EXPORT_SYMBOL(flush_dcache_page
);
419 void flush_dcache_icache_page(struct page
*page
)
421 #ifdef CONFIG_HUGETLB_PAGE
422 if (PageCompound(page
)) {
423 flush_dcache_icache_hugepage(page
);
429 void *start
= kmap_atomic(page
, KM_PPC_SYNC_ICACHE
);
430 __flush_dcache_icache(start
);
431 kunmap_atomic(start
, KM_PPC_SYNC_ICACHE
);
433 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
434 /* On 8xx there is no need to kmap since highmem is not supported */
435 __flush_dcache_icache(page_address(page
));
437 __flush_dcache_icache_phys(page_to_pfn(page
) << PAGE_SHIFT
);
441 void clear_user_page(void *page
, unsigned long vaddr
, struct page
*pg
)
446 * We shouldn't have to do this, but some versions of glibc
447 * require it (ld.so assumes zero filled pages are icache clean)
450 flush_dcache_page(pg
);
452 EXPORT_SYMBOL(clear_user_page
);
454 void copy_user_page(void *vto
, void *vfrom
, unsigned long vaddr
,
457 copy_page(vto
, vfrom
);
460 * We should be able to use the following optimisation, however
461 * there are two problems.
462 * Firstly a bug in some versions of binutils meant PLT sections
463 * were not marked executable.
464 * Secondly the first word in the GOT section is blrl, used
465 * to establish the GOT address. Until recently the GOT was
466 * not marked executable.
470 if (!vma
->vm_file
&& ((vma
->vm_flags
& VM_EXEC
) == 0))
474 flush_dcache_page(pg
);
477 void flush_icache_user_range(struct vm_area_struct
*vma
, struct page
*page
,
478 unsigned long addr
, int len
)
482 maddr
= (unsigned long) kmap(page
) + (addr
& ~PAGE_MASK
);
483 flush_icache_range(maddr
, maddr
+ len
);
486 EXPORT_SYMBOL(flush_icache_user_range
);
489 * This is called at the end of handling a user page fault, when the
490 * fault has been handled by updating a PTE in the linux page tables.
491 * We use it to preload an HPTE into the hash table corresponding to
492 * the updated linux PTE.
494 * This must always be called with the pte lock held.
496 void update_mmu_cache(struct vm_area_struct
*vma
, unsigned long address
,
499 #ifdef CONFIG_PPC_STD_MMU
500 unsigned long access
= 0, trap
;
502 /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
503 if (!pte_young(*ptep
) || address
>= TASK_SIZE
)
506 /* We try to figure out if we are coming from an instruction
507 * access fault and pass that down to __hash_page so we avoid
508 * double-faulting on execution of fresh text. We have to test
509 * for regs NULL since init will get here first thing at boot
511 * We also avoid filling the hash if not coming from a fault
513 if (current
->thread
.regs
== NULL
)
515 trap
= TRAP(current
->thread
.regs
);
517 access
|= _PAGE_EXEC
;
518 else if (trap
!= 0x300)
520 hash_preload(vma
->vm_mm
, address
, access
, trap
);
521 #endif /* CONFIG_PPC_STD_MMU */