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>
47 #include <asm/sparsemem.h>
49 #include <asm/fixmap.h>
53 #ifndef CPU_FTR_COHERENT_ICACHE
54 #define CPU_FTR_COHERENT_ICACHE 0 /* XXX for now */
55 #define CPU_FTR_NOEXECUTE 0
58 int init_bootmem_done
;
60 unsigned long memory_limit
;
66 EXPORT_SYMBOL(kmap_prot
);
67 EXPORT_SYMBOL(kmap_pte
);
69 static inline pte_t
*virt_to_kpte(unsigned long vaddr
)
71 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr
),
72 vaddr
), vaddr
), vaddr
);
76 int page_is_ram(unsigned long pfn
)
78 unsigned long paddr
= (pfn
<< PAGE_SHIFT
);
80 #ifndef CONFIG_PPC64 /* XXX for now */
81 return paddr
< __pa(high_memory
);
84 for (i
=0; i
< lmb
.memory
.cnt
; i
++) {
87 base
= lmb
.memory
.region
[i
].base
;
89 if ((paddr
>= base
) &&
90 (paddr
< (base
+ lmb
.memory
.region
[i
].size
))) {
99 pgprot_t
phys_mem_access_prot(struct file
*file
, unsigned long pfn
,
100 unsigned long size
, pgprot_t vma_prot
)
102 if (ppc_md
.phys_mem_access_prot
)
103 return ppc_md
.phys_mem_access_prot(file
, pfn
, size
, vma_prot
);
105 if (!page_is_ram(pfn
))
106 vma_prot
= __pgprot(pgprot_val(vma_prot
)
107 | _PAGE_GUARDED
| _PAGE_NO_CACHE
);
110 EXPORT_SYMBOL(phys_mem_access_prot
);
112 #ifdef CONFIG_MEMORY_HOTPLUG
115 int memory_add_physaddr_to_nid(u64 start
)
117 return hot_add_scn_to_nid(start
);
121 int arch_add_memory(int nid
, u64 start
, u64 size
)
123 struct pglist_data
*pgdata
;
125 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
126 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
128 pgdata
= NODE_DATA(nid
);
130 start
= (unsigned long)__va(start
);
131 create_section_mapping(start
, start
+ size
);
133 /* this should work for most non-highmem platforms */
134 zone
= pgdata
->node_zones
;
136 return __add_pages(zone
, start_pfn
, nr_pages
);
139 #ifdef CONFIG_MEMORY_HOTREMOVE
140 int remove_memory(u64 start
, u64 size
)
142 unsigned long start_pfn
, end_pfn
;
145 start_pfn
= start
>> PAGE_SHIFT
;
146 end_pfn
= start_pfn
+ (size
>> PAGE_SHIFT
);
147 ret
= offline_pages(start_pfn
, end_pfn
, 120 * HZ
);
150 /* Arch-specific calls go here - next patch */
154 #endif /* CONFIG_MEMORY_HOTREMOVE */
155 #endif /* CONFIG_MEMORY_HOTPLUG */
158 * walk_memory_resource() needs to make sure there is no holes in a given
159 * memory range. PPC64 does not maintain the memory layout in /proc/iomem.
160 * Instead it maintains it in lmb.memory structures. Walk through the
161 * memory regions, find holes and callback for contiguous regions.
164 walk_memory_resource(unsigned long start_pfn
, unsigned long nr_pages
, void *arg
,
165 int (*func
)(unsigned long, unsigned long, void *))
167 struct lmb_property res
;
168 unsigned long pfn
, len
;
172 res
.base
= (u64
) start_pfn
<< PAGE_SHIFT
;
173 res
.size
= (u64
) nr_pages
<< PAGE_SHIFT
;
175 end
= res
.base
+ res
.size
- 1;
176 while ((res
.base
< end
) && (lmb_find(&res
) >= 0)) {
177 pfn
= (unsigned long)(res
.base
>> PAGE_SHIFT
);
178 len
= (unsigned long)(res
.size
>> PAGE_SHIFT
);
179 ret
= (*func
)(pfn
, len
, arg
);
182 res
.base
+= (res
.size
+ 1);
183 res
.size
= (end
- res
.base
+ 1);
187 EXPORT_SYMBOL_GPL(walk_memory_resource
);
190 * Initialize the bootmem system and give it all the memory we
191 * have available. If we are using highmem, we only put the
192 * lowmem into the bootmem system.
194 #ifndef CONFIG_NEED_MULTIPLE_NODES
195 void __init
do_init_bootmem(void)
198 unsigned long start
, bootmap_pages
;
199 unsigned long total_pages
;
202 max_low_pfn
= max_pfn
= lmb_end_of_DRAM() >> PAGE_SHIFT
;
203 total_pages
= (lmb_end_of_DRAM() - memstart_addr
) >> PAGE_SHIFT
;
204 #ifdef CONFIG_HIGHMEM
205 total_pages
= total_lowmem
>> PAGE_SHIFT
;
206 max_low_pfn
= lowmem_end_addr
>> PAGE_SHIFT
;
210 * Find an area to use for the bootmem bitmap. Calculate the size of
211 * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
212 * Add 1 additional page in case the address isn't page-aligned.
214 bootmap_pages
= bootmem_bootmap_pages(total_pages
);
216 start
= lmb_alloc(bootmap_pages
<< PAGE_SHIFT
, PAGE_SIZE
);
218 min_low_pfn
= MEMORY_START
>> PAGE_SHIFT
;
219 boot_mapsize
= init_bootmem_node(NODE_DATA(0), start
>> PAGE_SHIFT
, min_low_pfn
, max_low_pfn
);
221 /* Add active regions with valid PFNs */
222 for (i
= 0; i
< lmb
.memory
.cnt
; i
++) {
223 unsigned long start_pfn
, end_pfn
;
224 start_pfn
= lmb
.memory
.region
[i
].base
>> PAGE_SHIFT
;
225 end_pfn
= start_pfn
+ lmb_size_pages(&lmb
.memory
, i
);
226 add_active_range(0, start_pfn
, end_pfn
);
229 /* Add all physical memory to the bootmem map, mark each area
232 #ifdef CONFIG_HIGHMEM
233 free_bootmem_with_active_regions(0, lowmem_end_addr
>> PAGE_SHIFT
);
235 /* reserve the sections we're already using */
236 for (i
= 0; i
< lmb
.reserved
.cnt
; i
++) {
237 unsigned long addr
= lmb
.reserved
.region
[i
].base
+
238 lmb_size_bytes(&lmb
.reserved
, i
) - 1;
239 if (addr
< lowmem_end_addr
)
240 reserve_bootmem(lmb
.reserved
.region
[i
].base
,
241 lmb_size_bytes(&lmb
.reserved
, i
),
243 else if (lmb
.reserved
.region
[i
].base
< lowmem_end_addr
) {
244 unsigned long adjusted_size
= lowmem_end_addr
-
245 lmb
.reserved
.region
[i
].base
;
246 reserve_bootmem(lmb
.reserved
.region
[i
].base
,
247 adjusted_size
, BOOTMEM_DEFAULT
);
251 free_bootmem_with_active_regions(0, max_pfn
);
253 /* reserve the sections we're already using */
254 for (i
= 0; i
< lmb
.reserved
.cnt
; i
++)
255 reserve_bootmem(lmb
.reserved
.region
[i
].base
,
256 lmb_size_bytes(&lmb
.reserved
, i
),
260 /* XXX need to clip this if using highmem? */
261 sparse_memory_present_with_active_regions(0);
263 init_bootmem_done
= 1;
266 /* mark pages that don't exist as nosave */
267 static int __init
mark_nonram_nosave(void)
269 unsigned long lmb_next_region_start_pfn
,
273 for (i
= 0; i
< lmb
.memory
.cnt
- 1; i
++) {
275 (lmb
.memory
.region
[i
].base
>> PAGE_SHIFT
) +
276 (lmb
.memory
.region
[i
].size
>> PAGE_SHIFT
);
277 lmb_next_region_start_pfn
=
278 lmb
.memory
.region
[i
+1].base
>> PAGE_SHIFT
;
280 if (lmb_region_max_pfn
< lmb_next_region_start_pfn
)
281 register_nosave_region(lmb_region_max_pfn
,
282 lmb_next_region_start_pfn
);
289 * paging_init() sets up the page tables - in fact we've already done this.
291 void __init
paging_init(void)
293 unsigned long total_ram
= lmb_phys_mem_size();
294 phys_addr_t top_of_ram
= lmb_end_of_DRAM();
295 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
298 unsigned long v
= __fix_to_virt(__end_of_fixed_addresses
- 1);
299 unsigned long end
= __fix_to_virt(FIX_HOLE
);
301 for (; v
< end
; v
+= PAGE_SIZE
)
302 map_page(v
, 0, 0); /* XXX gross */
305 #ifdef CONFIG_HIGHMEM
306 map_page(PKMAP_BASE
, 0, 0); /* XXX gross */
307 pkmap_page_table
= virt_to_kpte(PKMAP_BASE
);
309 kmap_pte
= virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN
));
310 kmap_prot
= PAGE_KERNEL
;
311 #endif /* CONFIG_HIGHMEM */
313 printk(KERN_DEBUG
"Top of RAM: 0x%llx, Total RAM: 0x%lx\n",
314 (unsigned long long)top_of_ram
, total_ram
);
315 printk(KERN_DEBUG
"Memory hole size: %ldMB\n",
316 (long int)((top_of_ram
- total_ram
) >> 20));
317 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
318 #ifdef CONFIG_HIGHMEM
319 max_zone_pfns
[ZONE_DMA
] = lowmem_end_addr
>> PAGE_SHIFT
;
320 max_zone_pfns
[ZONE_HIGHMEM
] = top_of_ram
>> PAGE_SHIFT
;
322 max_zone_pfns
[ZONE_DMA
] = top_of_ram
>> PAGE_SHIFT
;
324 free_area_init_nodes(max_zone_pfns
);
326 mark_nonram_nosave();
328 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
330 void __init
mem_init(void)
332 #ifdef CONFIG_NEED_MULTIPLE_NODES
338 unsigned long reservedpages
= 0, codesize
, initsize
, datasize
, bsssize
;
340 num_physpages
= lmb
.memory
.size
>> PAGE_SHIFT
;
341 high_memory
= (void *) __va(max_low_pfn
* PAGE_SIZE
);
343 #ifdef CONFIG_NEED_MULTIPLE_NODES
344 for_each_online_node(nid
) {
345 if (NODE_DATA(nid
)->node_spanned_pages
!= 0) {
346 printk("freeing bootmem node %d\n", nid
);
348 free_all_bootmem_node(NODE_DATA(nid
));
353 totalram_pages
+= free_all_bootmem();
355 for_each_online_pgdat(pgdat
) {
356 for (i
= 0; i
< pgdat
->node_spanned_pages
; i
++) {
357 if (!pfn_valid(pgdat
->node_start_pfn
+ i
))
359 page
= pgdat_page_nr(pgdat
, i
);
360 if (PageReserved(page
))
365 codesize
= (unsigned long)&_sdata
- (unsigned long)&_stext
;
366 datasize
= (unsigned long)&_edata
- (unsigned long)&_sdata
;
367 initsize
= (unsigned long)&__init_end
- (unsigned long)&__init_begin
;
368 bsssize
= (unsigned long)&__bss_stop
- (unsigned long)&__bss_start
;
370 #ifdef CONFIG_HIGHMEM
372 unsigned long pfn
, highmem_mapnr
;
374 highmem_mapnr
= lowmem_end_addr
>> PAGE_SHIFT
;
375 for (pfn
= highmem_mapnr
; pfn
< max_mapnr
; ++pfn
) {
376 struct page
*page
= pfn_to_page(pfn
);
377 if (lmb_is_reserved(pfn
<< PAGE_SHIFT
))
379 ClearPageReserved(page
);
380 init_page_count(page
);
385 totalram_pages
+= totalhigh_pages
;
386 printk(KERN_DEBUG
"High memory: %luk\n",
387 totalhigh_pages
<< (PAGE_SHIFT
-10));
389 #endif /* CONFIG_HIGHMEM */
391 printk(KERN_INFO
"Memory: %luk/%luk available (%luk kernel code, "
392 "%luk reserved, %luk data, %luk bss, %luk init)\n",
393 (unsigned long)nr_free_pages() << (PAGE_SHIFT
-10),
394 num_physpages
<< (PAGE_SHIFT
-10),
396 reservedpages
<< (PAGE_SHIFT
-10),
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
)
422 void *start
= kmap_atomic(page
, KM_PPC_SYNC_ICACHE
);
423 __flush_dcache_icache(start
);
424 kunmap_atomic(start
, KM_PPC_SYNC_ICACHE
);
425 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
426 /* On 8xx there is no need to kmap since highmem is not supported */
427 __flush_dcache_icache(page_address(page
));
429 __flush_dcache_icache_phys(page_to_pfn(page
) << PAGE_SHIFT
);
433 void clear_user_page(void *page
, unsigned long vaddr
, struct page
*pg
)
438 * We shouldnt have to do this, but some versions of glibc
439 * require it (ld.so assumes zero filled pages are icache clean)
442 flush_dcache_page(pg
);
444 EXPORT_SYMBOL(clear_user_page
);
446 void copy_user_page(void *vto
, void *vfrom
, unsigned long vaddr
,
449 copy_page(vto
, vfrom
);
452 * We should be able to use the following optimisation, however
453 * there are two problems.
454 * Firstly a bug in some versions of binutils meant PLT sections
455 * were not marked executable.
456 * Secondly the first word in the GOT section is blrl, used
457 * to establish the GOT address. Until recently the GOT was
458 * not marked executable.
462 if (!vma
->vm_file
&& ((vma
->vm_flags
& VM_EXEC
) == 0))
466 flush_dcache_page(pg
);
469 void flush_icache_user_range(struct vm_area_struct
*vma
, struct page
*page
,
470 unsigned long addr
, int len
)
474 maddr
= (unsigned long) kmap(page
) + (addr
& ~PAGE_MASK
);
475 flush_icache_range(maddr
, maddr
+ len
);
478 EXPORT_SYMBOL(flush_icache_user_range
);
481 * This is called at the end of handling a user page fault, when the
482 * fault has been handled by updating a PTE in the linux page tables.
483 * We use it to preload an HPTE into the hash table corresponding to
484 * the updated linux PTE.
486 * This must always be called with the pte lock held.
488 void update_mmu_cache(struct vm_area_struct
*vma
, unsigned long address
,
491 #ifdef CONFIG_PPC_STD_MMU
492 unsigned long access
= 0, trap
;
494 unsigned long pfn
= pte_pfn(pte
);
496 /* handle i-cache coherency */
497 if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE
) &&
498 !cpu_has_feature(CPU_FTR_NOEXECUTE
) &&
500 struct page
*page
= pfn_to_page(pfn
);
502 /* On 8xx, cache control instructions (particularly
503 * "dcbst" from flush_dcache_icache) fault as write
504 * operation if there is an unpopulated TLB entry
505 * for the address in question. To workaround that,
506 * we invalidate the TLB here, thus avoiding dcbst
509 _tlbie(address
, 0 /* 8xx doesn't care about PID */);
511 /* The _PAGE_USER test should really be _PAGE_EXEC, but
512 * older glibc versions execute some code from no-exec
513 * pages, which for now we are supporting. If exec-only
514 * pages are ever implemented, this will have to change.
516 if (!PageReserved(page
) && (pte_val(pte
) & _PAGE_USER
)
517 && !test_bit(PG_arch_1
, &page
->flags
)) {
518 if (vma
->vm_mm
== current
->active_mm
) {
519 __flush_dcache_icache((void *) address
);
521 flush_dcache_icache_page(page
);
522 set_bit(PG_arch_1
, &page
->flags
);
526 #ifdef CONFIG_PPC_STD_MMU
527 /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
528 if (!pte_young(pte
) || address
>= TASK_SIZE
)
531 /* We try to figure out if we are coming from an instruction
532 * access fault and pass that down to __hash_page so we avoid
533 * double-faulting on execution of fresh text. We have to test
534 * for regs NULL since init will get here first thing at boot
536 * We also avoid filling the hash if not coming from a fault
538 if (current
->thread
.regs
== NULL
)
540 trap
= TRAP(current
->thread
.regs
);
542 access
|= _PAGE_EXEC
;
543 else if (trap
!= 0x300)
545 hash_preload(vma
->vm_mm
, address
, access
, trap
);
546 #endif /* CONFIG_PPC_STD_MMU */