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>
34 #include <linux/lmb.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/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 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 #ifdef CONFIG_MEMORY_HOTREMOVE
133 int remove_memory(u64 start
, u64 size
)
135 unsigned long start_pfn
, end_pfn
;
138 start_pfn
= start
>> PAGE_SHIFT
;
139 end_pfn
= start_pfn
+ (size
>> PAGE_SHIFT
);
140 ret
= offline_pages(start_pfn
, end_pfn
, 120 * HZ
);
143 /* Arch-specific calls go here - next patch */
147 #endif /* CONFIG_MEMORY_HOTREMOVE */
150 * walk_memory_resource() needs to make sure there is no holes in a given
151 * memory range. On PPC64, since this range comes from /sysfs, the range
152 * is guaranteed to be valid, non-overlapping and can not contain any
153 * holes. By the time we get here (memory add or remove), /proc/device-tree
154 * is updated and correct. Only reason we need to check against device-tree
155 * would be if we allow user-land to specify a memory range through a
156 * system call/ioctl etc. instead of doing offline/online through /sysfs.
159 walk_memory_resource(unsigned long start_pfn
, unsigned long nr_pages
, void *arg
,
160 int (*func
)(unsigned long, unsigned long, void *))
162 return (*func
)(start_pfn
, nr_pages
, arg
);
165 #endif /* CONFIG_MEMORY_HOTPLUG */
169 unsigned long total
= 0, reserved
= 0;
170 unsigned long shared
= 0, cached
= 0;
171 unsigned long highmem
= 0;
176 printk("Mem-info:\n");
178 for_each_online_pgdat(pgdat
) {
180 pgdat_resize_lock(pgdat
, &flags
);
181 for (i
= 0; i
< pgdat
->node_spanned_pages
; i
++) {
182 if (!pfn_valid(pgdat
->node_start_pfn
+ i
))
184 page
= pgdat_page_nr(pgdat
, i
);
186 if (PageHighMem(page
))
188 if (PageReserved(page
))
190 else if (PageSwapCache(page
))
192 else if (page_count(page
))
193 shared
+= page_count(page
) - 1;
195 pgdat_resize_unlock(pgdat
, &flags
);
197 printk("%ld pages of RAM\n", total
);
198 #ifdef CONFIG_HIGHMEM
199 printk("%ld pages of HIGHMEM\n", highmem
);
201 printk("%ld reserved pages\n", reserved
);
202 printk("%ld pages shared\n", shared
);
203 printk("%ld pages swap cached\n", cached
);
207 * Initialize the bootmem system and give it all the memory we
208 * have available. If we are using highmem, we only put the
209 * lowmem into the bootmem system.
211 #ifndef CONFIG_NEED_MULTIPLE_NODES
212 void __init
do_init_bootmem(void)
215 unsigned long start
, bootmap_pages
;
216 unsigned long total_pages
;
219 max_pfn
= total_pages
= lmb_end_of_DRAM() >> PAGE_SHIFT
;
220 #ifdef CONFIG_HIGHMEM
221 total_pages
= total_lowmem
>> PAGE_SHIFT
;
225 * Find an area to use for the bootmem bitmap. Calculate the size of
226 * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
227 * Add 1 additional page in case the address isn't page-aligned.
229 bootmap_pages
= bootmem_bootmap_pages(total_pages
);
231 start
= lmb_alloc(bootmap_pages
<< PAGE_SHIFT
, PAGE_SIZE
);
233 boot_mapsize
= init_bootmem(start
>> PAGE_SHIFT
, total_pages
);
235 /* Add active regions with valid PFNs */
236 for (i
= 0; i
< lmb
.memory
.cnt
; i
++) {
237 unsigned long start_pfn
, end_pfn
;
238 start_pfn
= lmb
.memory
.region
[i
].base
>> PAGE_SHIFT
;
239 end_pfn
= start_pfn
+ lmb_size_pages(&lmb
.memory
, i
);
240 add_active_range(0, start_pfn
, end_pfn
);
243 /* Add all physical memory to the bootmem map, mark each area
246 #ifdef CONFIG_HIGHMEM
247 free_bootmem_with_active_regions(0, total_lowmem
>> PAGE_SHIFT
);
249 /* reserve the sections we're already using */
250 for (i
= 0; i
< lmb
.reserved
.cnt
; i
++) {
251 unsigned long addr
= lmb
.reserved
.region
[i
].base
+
252 lmb_size_bytes(&lmb
.reserved
, i
) - 1;
253 if (addr
< total_lowmem
)
254 reserve_bootmem(lmb
.reserved
.region
[i
].base
,
255 lmb_size_bytes(&lmb
.reserved
, i
),
257 else if (lmb
.reserved
.region
[i
].base
< total_lowmem
) {
258 unsigned long adjusted_size
= total_lowmem
-
259 lmb
.reserved
.region
[i
].base
;
260 reserve_bootmem(lmb
.reserved
.region
[i
].base
,
261 adjusted_size
, BOOTMEM_DEFAULT
);
265 free_bootmem_with_active_regions(0, max_pfn
);
267 /* reserve the sections we're already using */
268 for (i
= 0; i
< lmb
.reserved
.cnt
; i
++)
269 reserve_bootmem(lmb
.reserved
.region
[i
].base
,
270 lmb_size_bytes(&lmb
.reserved
, i
),
274 /* XXX need to clip this if using highmem? */
275 sparse_memory_present_with_active_regions(0);
277 init_bootmem_done
= 1;
280 /* mark pages that don't exist as nosave */
281 static int __init
mark_nonram_nosave(void)
283 unsigned long lmb_next_region_start_pfn
,
287 for (i
= 0; i
< lmb
.memory
.cnt
- 1; i
++) {
289 (lmb
.memory
.region
[i
].base
>> PAGE_SHIFT
) +
290 (lmb
.memory
.region
[i
].size
>> PAGE_SHIFT
);
291 lmb_next_region_start_pfn
=
292 lmb
.memory
.region
[i
+1].base
>> PAGE_SHIFT
;
294 if (lmb_region_max_pfn
< lmb_next_region_start_pfn
)
295 register_nosave_region(lmb_region_max_pfn
,
296 lmb_next_region_start_pfn
);
303 * paging_init() sets up the page tables - in fact we've already done this.
305 void __init
paging_init(void)
307 unsigned long total_ram
= lmb_phys_mem_size();
308 unsigned long top_of_ram
= lmb_end_of_DRAM();
309 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
311 #ifdef CONFIG_HIGHMEM
312 map_page(PKMAP_BASE
, 0, 0); /* XXX gross */
313 pkmap_page_table
= pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k
314 (PKMAP_BASE
), PKMAP_BASE
), PKMAP_BASE
), PKMAP_BASE
);
315 map_page(KMAP_FIX_BEGIN
, 0, 0); /* XXX gross */
316 kmap_pte
= pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k
317 (KMAP_FIX_BEGIN
), KMAP_FIX_BEGIN
), KMAP_FIX_BEGIN
),
319 kmap_prot
= PAGE_KERNEL
;
320 #endif /* CONFIG_HIGHMEM */
322 printk(KERN_DEBUG
"Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
323 top_of_ram
, total_ram
);
324 printk(KERN_DEBUG
"Memory hole size: %ldMB\n",
325 (top_of_ram
- total_ram
) >> 20);
326 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
327 #ifdef CONFIG_HIGHMEM
328 max_zone_pfns
[ZONE_DMA
] = total_lowmem
>> PAGE_SHIFT
;
329 max_zone_pfns
[ZONE_HIGHMEM
] = top_of_ram
>> PAGE_SHIFT
;
331 max_zone_pfns
[ZONE_DMA
] = top_of_ram
>> PAGE_SHIFT
;
333 free_area_init_nodes(max_zone_pfns
);
335 mark_nonram_nosave();
337 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
339 void __init
mem_init(void)
341 #ifdef CONFIG_NEED_MULTIPLE_NODES
347 unsigned long reservedpages
= 0, codesize
, initsize
, datasize
, bsssize
;
349 num_physpages
= lmb
.memory
.size
>> PAGE_SHIFT
;
350 high_memory
= (void *) __va(max_low_pfn
* PAGE_SIZE
);
352 #ifdef CONFIG_NEED_MULTIPLE_NODES
353 for_each_online_node(nid
) {
354 if (NODE_DATA(nid
)->node_spanned_pages
!= 0) {
355 printk("freeing bootmem node %d\n", nid
);
357 free_all_bootmem_node(NODE_DATA(nid
));
362 totalram_pages
+= free_all_bootmem();
364 for_each_online_pgdat(pgdat
) {
365 for (i
= 0; i
< pgdat
->node_spanned_pages
; i
++) {
366 if (!pfn_valid(pgdat
->node_start_pfn
+ i
))
368 page
= pgdat_page_nr(pgdat
, i
);
369 if (PageReserved(page
))
374 codesize
= (unsigned long)&_sdata
- (unsigned long)&_stext
;
375 datasize
= (unsigned long)&_edata
- (unsigned long)&_sdata
;
376 initsize
= (unsigned long)&__init_end
- (unsigned long)&__init_begin
;
377 bsssize
= (unsigned long)&__bss_stop
- (unsigned long)&__bss_start
;
379 #ifdef CONFIG_HIGHMEM
381 unsigned long pfn
, highmem_mapnr
;
383 highmem_mapnr
= total_lowmem
>> PAGE_SHIFT
;
384 for (pfn
= highmem_mapnr
; pfn
< max_mapnr
; ++pfn
) {
385 struct page
*page
= pfn_to_page(pfn
);
386 if (lmb_is_reserved(pfn
<< PAGE_SHIFT
))
388 ClearPageReserved(page
);
389 init_page_count(page
);
394 totalram_pages
+= totalhigh_pages
;
395 printk(KERN_DEBUG
"High memory: %luk\n",
396 totalhigh_pages
<< (PAGE_SHIFT
-10));
398 #endif /* CONFIG_HIGHMEM */
400 printk(KERN_INFO
"Memory: %luk/%luk available (%luk kernel code, "
401 "%luk reserved, %luk data, %luk bss, %luk init)\n",
402 (unsigned long)nr_free_pages() << (PAGE_SHIFT
-10),
403 num_physpages
<< (PAGE_SHIFT
-10),
405 reservedpages
<< (PAGE_SHIFT
-10),
414 * This is called when a page has been modified by the kernel.
415 * It just marks the page as not i-cache clean. We do the i-cache
416 * flush later when the page is given to a user process, if necessary.
418 void flush_dcache_page(struct page
*page
)
420 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE
))
422 /* avoid an atomic op if possible */
423 if (test_bit(PG_arch_1
, &page
->flags
))
424 clear_bit(PG_arch_1
, &page
->flags
);
426 EXPORT_SYMBOL(flush_dcache_page
);
428 void flush_dcache_icache_page(struct page
*page
)
431 void *start
= kmap_atomic(page
, KM_PPC_SYNC_ICACHE
);
432 __flush_dcache_icache(start
);
433 kunmap_atomic(start
, KM_PPC_SYNC_ICACHE
);
434 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
435 /* On 8xx there is no need to kmap since highmem is not supported */
436 __flush_dcache_icache(page_address(page
));
438 __flush_dcache_icache_phys(page_to_pfn(page
) << PAGE_SHIFT
);
442 void clear_user_page(void *page
, unsigned long vaddr
, struct page
*pg
)
447 * We shouldnt have to do this, but some versions of glibc
448 * require it (ld.so assumes zero filled pages are icache clean)
451 flush_dcache_page(pg
);
453 EXPORT_SYMBOL(clear_user_page
);
455 void copy_user_page(void *vto
, void *vfrom
, unsigned long vaddr
,
458 copy_page(vto
, vfrom
);
461 * We should be able to use the following optimisation, however
462 * there are two problems.
463 * Firstly a bug in some versions of binutils meant PLT sections
464 * were not marked executable.
465 * Secondly the first word in the GOT section is blrl, used
466 * to establish the GOT address. Until recently the GOT was
467 * not marked executable.
471 if (!vma
->vm_file
&& ((vma
->vm_flags
& VM_EXEC
) == 0))
475 flush_dcache_page(pg
);
478 void flush_icache_user_range(struct vm_area_struct
*vma
, struct page
*page
,
479 unsigned long addr
, int len
)
483 maddr
= (unsigned long) kmap(page
) + (addr
& ~PAGE_MASK
);
484 flush_icache_range(maddr
, maddr
+ len
);
487 EXPORT_SYMBOL(flush_icache_user_range
);
490 * This is called at the end of handling a user page fault, when the
491 * fault has been handled by updating a PTE in the linux page tables.
492 * We use it to preload an HPTE into the hash table corresponding to
493 * the updated linux PTE.
495 * This must always be called with the pte lock held.
497 void update_mmu_cache(struct vm_area_struct
*vma
, unsigned long address
,
500 #ifdef CONFIG_PPC_STD_MMU
501 unsigned long access
= 0, trap
;
503 unsigned long pfn
= pte_pfn(pte
);
505 /* handle i-cache coherency */
506 if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE
) &&
507 !cpu_has_feature(CPU_FTR_NOEXECUTE
) &&
509 struct page
*page
= pfn_to_page(pfn
);
511 /* On 8xx, cache control instructions (particularly
512 * "dcbst" from flush_dcache_icache) fault as write
513 * operation if there is an unpopulated TLB entry
514 * for the address in question. To workaround that,
515 * we invalidate the TLB here, thus avoiding dcbst
518 _tlbie(address
, 0 /* 8xx doesn't care about PID */);
520 /* The _PAGE_USER test should really be _PAGE_EXEC, but
521 * older glibc versions execute some code from no-exec
522 * pages, which for now we are supporting. If exec-only
523 * pages are ever implemented, this will have to change.
525 if (!PageReserved(page
) && (pte_val(pte
) & _PAGE_USER
)
526 && !test_bit(PG_arch_1
, &page
->flags
)) {
527 if (vma
->vm_mm
== current
->active_mm
) {
528 __flush_dcache_icache((void *) address
);
530 flush_dcache_icache_page(page
);
531 set_bit(PG_arch_1
, &page
->flags
);
535 #ifdef CONFIG_PPC_STD_MMU
536 /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
537 if (!pte_young(pte
) || address
>= TASK_SIZE
)
540 /* We try to figure out if we are coming from an instruction
541 * access fault and pass that down to __hash_page so we avoid
542 * double-faulting on execution of fresh text. We have to test
543 * for regs NULL since init will get here first thing at boot
545 * We also avoid filling the hash if not coming from a fault
547 if (current
->thread
.regs
== NULL
)
549 trap
= TRAP(current
->thread
.regs
);
551 access
|= _PAGE_EXEC
;
552 else if (trap
!= 0x300)
554 hash_preload(vma
->vm_mm
, address
, access
, trap
);
555 #endif /* CONFIG_PPC_STD_MMU */