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/types.h>
27 #include <linux/stddef.h>
28 #include <linux/init.h>
29 #include <linux/bootmem.h>
30 #include <linux/highmem.h>
31 #include <linux/initrd.h>
32 #include <linux/pagemap.h>
33 #include <linux/suspend.h>
35 #include <asm/pgalloc.h>
38 #include <asm/mmu_context.h>
39 #include <asm/pgtable.h>
42 #include <asm/machdep.h>
43 #include <asm/btext.h>
46 #include <asm/sections.h>
51 #ifndef CPU_FTR_COHERENT_ICACHE
52 #define CPU_FTR_COHERENT_ICACHE 0 /* XXX for now */
53 #define CPU_FTR_NOEXECUTE 0
56 int init_bootmem_done
;
58 unsigned long memory_limit
;
60 int page_is_ram(unsigned long pfn
)
62 unsigned long paddr
= (pfn
<< PAGE_SHIFT
);
64 #ifndef CONFIG_PPC64 /* XXX for now */
65 return paddr
< __pa(high_memory
);
68 for (i
=0; i
< lmb
.memory
.cnt
; i
++) {
71 base
= lmb
.memory
.region
[i
].base
;
73 if ((paddr
>= base
) &&
74 (paddr
< (base
+ lmb
.memory
.region
[i
].size
))) {
83 pgprot_t
phys_mem_access_prot(struct file
*file
, unsigned long pfn
,
84 unsigned long size
, pgprot_t vma_prot
)
86 if (ppc_md
.phys_mem_access_prot
)
87 return ppc_md
.phys_mem_access_prot(file
, pfn
, size
, vma_prot
);
89 if (!page_is_ram(pfn
))
90 vma_prot
= __pgprot(pgprot_val(vma_prot
)
91 | _PAGE_GUARDED
| _PAGE_NO_CACHE
);
94 EXPORT_SYMBOL(phys_mem_access_prot
);
96 #ifdef CONFIG_MEMORY_HOTPLUG
98 void online_page(struct page
*page
)
100 ClearPageReserved(page
);
101 init_page_count(page
);
108 int memory_add_physaddr_to_nid(u64 start
)
110 return hot_add_scn_to_nid(start
);
114 int __devinit
arch_add_memory(int nid
, u64 start
, u64 size
)
116 struct pglist_data
*pgdata
;
118 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
119 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
121 pgdata
= NODE_DATA(nid
);
123 start
= (unsigned long)__va(start
);
124 create_section_mapping(start
, start
+ size
);
126 /* this should work for most non-highmem platforms */
127 zone
= pgdata
->node_zones
;
129 return __add_pages(zone
, start_pfn
, nr_pages
);
132 #endif /* CONFIG_MEMORY_HOTPLUG */
136 unsigned long total
= 0, reserved
= 0;
137 unsigned long shared
= 0, cached
= 0;
138 unsigned long highmem
= 0;
143 printk("Mem-info:\n");
145 printk("Free swap: %6ldkB\n", nr_swap_pages
<<(PAGE_SHIFT
-10));
146 for_each_online_pgdat(pgdat
) {
148 pgdat_resize_lock(pgdat
, &flags
);
149 for (i
= 0; i
< pgdat
->node_spanned_pages
; i
++) {
150 if (!pfn_valid(pgdat
->node_start_pfn
+ i
))
152 page
= pgdat_page_nr(pgdat
, i
);
154 if (PageHighMem(page
))
156 if (PageReserved(page
))
158 else if (PageSwapCache(page
))
160 else if (page_count(page
))
161 shared
+= page_count(page
) - 1;
163 pgdat_resize_unlock(pgdat
, &flags
);
165 printk("%ld pages of RAM\n", total
);
166 #ifdef CONFIG_HIGHMEM
167 printk("%ld pages of HIGHMEM\n", highmem
);
169 printk("%ld reserved pages\n", reserved
);
170 printk("%ld pages shared\n", shared
);
171 printk("%ld pages swap cached\n", cached
);
175 * Initialize the bootmem system and give it all the memory we
176 * have available. If we are using highmem, we only put the
177 * lowmem into the bootmem system.
179 #ifndef CONFIG_NEED_MULTIPLE_NODES
180 void __init
do_init_bootmem(void)
183 unsigned long start
, bootmap_pages
;
184 unsigned long total_pages
;
187 max_pfn
= total_pages
= lmb_end_of_DRAM() >> PAGE_SHIFT
;
188 #ifdef CONFIG_HIGHMEM
189 total_pages
= total_lowmem
>> PAGE_SHIFT
;
193 * Find an area to use for the bootmem bitmap. Calculate the size of
194 * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
195 * Add 1 additional page in case the address isn't page-aligned.
197 bootmap_pages
= bootmem_bootmap_pages(total_pages
);
199 start
= lmb_alloc(bootmap_pages
<< PAGE_SHIFT
, PAGE_SIZE
);
201 boot_mapsize
= init_bootmem(start
>> PAGE_SHIFT
, total_pages
);
203 /* Add active regions with valid PFNs */
204 for (i
= 0; i
< lmb
.memory
.cnt
; i
++) {
205 unsigned long start_pfn
, end_pfn
;
206 start_pfn
= lmb
.memory
.region
[i
].base
>> PAGE_SHIFT
;
207 end_pfn
= start_pfn
+ lmb_size_pages(&lmb
.memory
, i
);
208 add_active_range(0, start_pfn
, end_pfn
);
211 /* Add all physical memory to the bootmem map, mark each area
214 #ifdef CONFIG_HIGHMEM
215 free_bootmem_with_active_regions(0, total_lowmem
>> PAGE_SHIFT
);
217 /* reserve the sections we're already using */
218 for (i
= 0; i
< lmb
.reserved
.cnt
; i
++) {
219 unsigned long addr
= lmb
.reserved
.region
[i
].base
+
220 lmb_size_bytes(&lmb
.reserved
, i
) - 1;
221 if (addr
< total_lowmem
)
222 reserve_bootmem(lmb
.reserved
.region
[i
].base
,
223 lmb_size_bytes(&lmb
.reserved
, i
),
225 else if (lmb
.reserved
.region
[i
].base
< total_lowmem
) {
226 unsigned long adjusted_size
= total_lowmem
-
227 lmb
.reserved
.region
[i
].base
;
228 reserve_bootmem(lmb
.reserved
.region
[i
].base
,
229 adjusted_size
, BOOTMEM_DEFAULT
);
233 free_bootmem_with_active_regions(0, max_pfn
);
235 /* reserve the sections we're already using */
236 for (i
= 0; i
< lmb
.reserved
.cnt
; i
++)
237 reserve_bootmem(lmb
.reserved
.region
[i
].base
,
238 lmb_size_bytes(&lmb
.reserved
, i
),
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 unsigned long lmb_next_region_start_pfn
,
255 for (i
= 0; i
< lmb
.memory
.cnt
- 1; i
++) {
257 (lmb
.memory
.region
[i
].base
>> PAGE_SHIFT
) +
258 (lmb
.memory
.region
[i
].size
>> PAGE_SHIFT
);
259 lmb_next_region_start_pfn
=
260 lmb
.memory
.region
[i
+1].base
>> PAGE_SHIFT
;
262 if (lmb_region_max_pfn
< lmb_next_region_start_pfn
)
263 register_nosave_region(lmb_region_max_pfn
,
264 lmb_next_region_start_pfn
);
271 * paging_init() sets up the page tables - in fact we've already done this.
273 void __init
paging_init(void)
275 unsigned long total_ram
= lmb_phys_mem_size();
276 unsigned long top_of_ram
= lmb_end_of_DRAM();
277 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
279 #ifdef CONFIG_HIGHMEM
280 map_page(PKMAP_BASE
, 0, 0); /* XXX gross */
281 pkmap_page_table
= pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k
282 (PKMAP_BASE
), PKMAP_BASE
), PKMAP_BASE
), PKMAP_BASE
);
283 map_page(KMAP_FIX_BEGIN
, 0, 0); /* XXX gross */
284 kmap_pte
= pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k
285 (KMAP_FIX_BEGIN
), KMAP_FIX_BEGIN
), KMAP_FIX_BEGIN
),
287 kmap_prot
= PAGE_KERNEL
;
288 #endif /* CONFIG_HIGHMEM */
290 printk(KERN_DEBUG
"Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
291 top_of_ram
, total_ram
);
292 printk(KERN_DEBUG
"Memory hole size: %ldMB\n",
293 (top_of_ram
- total_ram
) >> 20);
294 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
295 #ifdef CONFIG_HIGHMEM
296 max_zone_pfns
[ZONE_DMA
] = total_lowmem
>> PAGE_SHIFT
;
297 max_zone_pfns
[ZONE_HIGHMEM
] = top_of_ram
>> PAGE_SHIFT
;
299 max_zone_pfns
[ZONE_DMA
] = top_of_ram
>> PAGE_SHIFT
;
301 free_area_init_nodes(max_zone_pfns
);
303 mark_nonram_nosave();
305 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
307 void __init
mem_init(void)
309 #ifdef CONFIG_NEED_MULTIPLE_NODES
315 unsigned long reservedpages
= 0, codesize
, initsize
, datasize
, bsssize
;
317 num_physpages
= lmb
.memory
.size
>> PAGE_SHIFT
;
318 high_memory
= (void *) __va(max_low_pfn
* PAGE_SIZE
);
320 #ifdef CONFIG_NEED_MULTIPLE_NODES
321 for_each_online_node(nid
) {
322 if (NODE_DATA(nid
)->node_spanned_pages
!= 0) {
323 printk("freeing bootmem node %d\n", nid
);
325 free_all_bootmem_node(NODE_DATA(nid
));
330 totalram_pages
+= free_all_bootmem();
332 for_each_online_pgdat(pgdat
) {
333 for (i
= 0; i
< pgdat
->node_spanned_pages
; i
++) {
334 if (!pfn_valid(pgdat
->node_start_pfn
+ i
))
336 page
= pgdat_page_nr(pgdat
, i
);
337 if (PageReserved(page
))
342 codesize
= (unsigned long)&_sdata
- (unsigned long)&_stext
;
343 datasize
= (unsigned long)&_edata
- (unsigned long)&_sdata
;
344 initsize
= (unsigned long)&__init_end
- (unsigned long)&__init_begin
;
345 bsssize
= (unsigned long)&__bss_stop
- (unsigned long)&__bss_start
;
347 #ifdef CONFIG_HIGHMEM
349 unsigned long pfn
, highmem_mapnr
;
351 highmem_mapnr
= total_lowmem
>> PAGE_SHIFT
;
352 for (pfn
= highmem_mapnr
; pfn
< max_mapnr
; ++pfn
) {
353 struct page
*page
= pfn_to_page(pfn
);
354 if (lmb_is_reserved(pfn
<< PAGE_SHIFT
))
356 ClearPageReserved(page
);
357 init_page_count(page
);
362 totalram_pages
+= totalhigh_pages
;
363 printk(KERN_DEBUG
"High memory: %luk\n",
364 totalhigh_pages
<< (PAGE_SHIFT
-10));
366 #endif /* CONFIG_HIGHMEM */
368 printk(KERN_INFO
"Memory: %luk/%luk available (%luk kernel code, "
369 "%luk reserved, %luk data, %luk bss, %luk init)\n",
370 (unsigned long)nr_free_pages() << (PAGE_SHIFT
-10),
371 num_physpages
<< (PAGE_SHIFT
-10),
373 reservedpages
<< (PAGE_SHIFT
-10),
382 * This is called when a page has been modified by the kernel.
383 * It just marks the page as not i-cache clean. We do the i-cache
384 * flush later when the page is given to a user process, if necessary.
386 void flush_dcache_page(struct page
*page
)
388 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE
))
390 /* avoid an atomic op if possible */
391 if (test_bit(PG_arch_1
, &page
->flags
))
392 clear_bit(PG_arch_1
, &page
->flags
);
394 EXPORT_SYMBOL(flush_dcache_page
);
396 void flush_dcache_icache_page(struct page
*page
)
399 void *start
= kmap_atomic(page
, KM_PPC_SYNC_ICACHE
);
400 __flush_dcache_icache(start
);
401 kunmap_atomic(start
, KM_PPC_SYNC_ICACHE
);
402 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
403 /* On 8xx there is no need to kmap since highmem is not supported */
404 __flush_dcache_icache(page_address(page
));
406 __flush_dcache_icache_phys(page_to_pfn(page
) << PAGE_SHIFT
);
410 void clear_user_page(void *page
, unsigned long vaddr
, struct page
*pg
)
415 * We shouldnt have to do this, but some versions of glibc
416 * require it (ld.so assumes zero filled pages are icache clean)
419 flush_dcache_page(pg
);
421 EXPORT_SYMBOL(clear_user_page
);
423 void copy_user_page(void *vto
, void *vfrom
, unsigned long vaddr
,
426 copy_page(vto
, vfrom
);
429 * We should be able to use the following optimisation, however
430 * there are two problems.
431 * Firstly a bug in some versions of binutils meant PLT sections
432 * were not marked executable.
433 * Secondly the first word in the GOT section is blrl, used
434 * to establish the GOT address. Until recently the GOT was
435 * not marked executable.
439 if (!vma
->vm_file
&& ((vma
->vm_flags
& VM_EXEC
) == 0))
443 flush_dcache_page(pg
);
446 void flush_icache_user_range(struct vm_area_struct
*vma
, struct page
*page
,
447 unsigned long addr
, int len
)
451 maddr
= (unsigned long) kmap(page
) + (addr
& ~PAGE_MASK
);
452 flush_icache_range(maddr
, maddr
+ len
);
455 EXPORT_SYMBOL(flush_icache_user_range
);
458 * This is called at the end of handling a user page fault, when the
459 * fault has been handled by updating a PTE in the linux page tables.
460 * We use it to preload an HPTE into the hash table corresponding to
461 * the updated linux PTE.
463 * This must always be called with the pte lock held.
465 void update_mmu_cache(struct vm_area_struct
*vma
, unsigned long address
,
468 #ifdef CONFIG_PPC_STD_MMU
469 unsigned long access
= 0, trap
;
471 unsigned long pfn
= pte_pfn(pte
);
473 /* handle i-cache coherency */
474 if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE
) &&
475 !cpu_has_feature(CPU_FTR_NOEXECUTE
) &&
477 struct page
*page
= pfn_to_page(pfn
);
479 /* On 8xx, cache control instructions (particularly
480 * "dcbst" from flush_dcache_icache) fault as write
481 * operation if there is an unpopulated TLB entry
482 * for the address in question. To workaround that,
483 * we invalidate the TLB here, thus avoiding dcbst
486 _tlbie(address
, 0 /* 8xx doesn't care about PID */);
488 if (!PageReserved(page
)
489 && !test_bit(PG_arch_1
, &page
->flags
)) {
490 if (vma
->vm_mm
== current
->active_mm
) {
491 __flush_dcache_icache((void *) address
);
493 flush_dcache_icache_page(page
);
494 set_bit(PG_arch_1
, &page
->flags
);
498 #ifdef CONFIG_PPC_STD_MMU
499 /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
500 if (!pte_young(pte
) || address
>= TASK_SIZE
)
503 /* We try to figure out if we are coming from an instruction
504 * access fault and pass that down to __hash_page so we avoid
505 * double-faulting on execution of fresh text. We have to test
506 * for regs NULL since init will get here first thing at boot
508 * We also avoid filling the hash if not coming from a fault
510 if (current
->thread
.regs
== NULL
)
512 trap
= TRAP(current
->thread
.regs
);
514 access
|= _PAGE_EXEC
;
515 else if (trap
!= 0x300)
517 hash_preload(vma
->vm_mm
, address
, access
, trap
);
518 #endif /* CONFIG_PPC_STD_MMU */