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
;
70 #ifdef CONFIG_PPC_PMAC
71 unsigned long agp_special_page
;
75 extern char etext
[], _stext
[];
76 extern char __init_begin
, __init_end
;
77 extern char __prep_begin
, __prep_end
;
78 extern char __chrp_begin
, __chrp_end
;
79 extern char __pmac_begin
, __pmac_end
;
80 extern char __openfirmware_begin
, __openfirmware_end
;
86 EXPORT_SYMBOL(kmap_prot
);
87 EXPORT_SYMBOL(kmap_pte
);
91 void set_phys_avail(unsigned long total_ram
);
93 /* XXX should be in current.h -- paulus */
94 extern struct task_struct
*current_set
[NR_CPUS
];
97 struct mem_pieces phys_avail
;
100 * this tells the system to map all of ram with the segregs
101 * (i.e. page tables) instead of the bats.
104 int __map_without_bats
;
105 int __map_without_ltlbs
;
107 /* max amount of RAM to use */
108 unsigned long __max_memory
;
109 /* max amount of low RAM to map in */
110 unsigned long __max_low_memory
= MAX_LOW_MEM
;
114 int i
,free
= 0,total
= 0,reserved
= 0;
115 int shared
= 0, cached
= 0;
118 printk("Mem-info:\n");
120 printk("Free swap: %6ldkB\n", nr_swap_pages
<<(PAGE_SHIFT
-10));
124 if (PageHighMem(mem_map
+i
))
126 if (PageReserved(mem_map
+i
))
128 else if (PageSwapCache(mem_map
+i
))
130 else if (!page_count(mem_map
+i
))
133 shared
+= page_count(mem_map
+i
) - 1;
135 printk("%d pages of RAM\n",total
);
136 printk("%d pages of HIGHMEM\n", highmem
);
137 printk("%d free pages\n",free
);
138 printk("%d reserved pages\n",reserved
);
139 printk("%d pages shared\n",shared
);
140 printk("%d pages swap cached\n",cached
);
143 /* Free up now-unused memory */
144 static void free_sec(unsigned long start
, unsigned long end
, const char *name
)
146 unsigned long cnt
= 0;
148 while (start
< end
) {
149 ClearPageReserved(virt_to_page(start
));
150 set_page_count(virt_to_page(start
), 1);
156 printk(" %ldk %s", cnt
<< (PAGE_SHIFT
- 10), name
);
157 totalram_pages
+= cnt
;
161 void free_initmem(void)
163 #define FREESEC(TYPE) \
164 free_sec((unsigned long)(&__ ## TYPE ## _begin), \
165 (unsigned long)(&__ ## TYPE ## _end), \
168 printk ("Freeing unused kernel memory:");
170 if (_machine
!= _MACH_Pmac
)
172 if (_machine
!= _MACH_chrp
)
174 if (_machine
!= _MACH_prep
)
177 FREESEC(openfirmware
);
179 ppc_md
.progress
= NULL
;
183 #ifdef CONFIG_BLK_DEV_INITRD
184 void free_initrd_mem(unsigned long start
, unsigned long end
)
186 printk ("Freeing initrd memory: %ldk freed\n", (end
- start
) >> 10);
188 for (; start
< end
; start
+= PAGE_SIZE
) {
189 ClearPageReserved(virt_to_page(start
));
190 set_page_count(virt_to_page(start
), 1);
198 * Check for command-line options that affect what MMU_init will do.
202 /* Check for nobats option (used in mapin_ram). */
203 if (strstr(cmd_line
, "nobats")) {
204 __map_without_bats
= 1;
207 if (strstr(cmd_line
, "noltlbs")) {
208 __map_without_ltlbs
= 1;
211 /* Look for mem= option on command line */
212 if (strstr(cmd_line
, "mem=")) {
214 unsigned long maxmem
= 0;
216 for (q
= cmd_line
; (p
= strstr(q
, "mem=")) != 0; ) {
218 if (p
> cmd_line
&& p
[-1] != ' ')
220 maxmem
= simple_strtoul(q
, &q
, 0);
221 if (*q
== 'k' || *q
== 'K') {
224 } else if (*q
== 'm' || *q
== 'M') {
229 __max_memory
= maxmem
;
234 * MMU_init sets up the basic memory mappings for the kernel,
235 * including both RAM and possibly some I/O regions,
236 * and sets up the page tables and the MMU hardware ready to go.
238 void __init
MMU_init(void)
241 ppc_md
.progress("MMU:enter", 0x111);
243 /* parse args from command line */
247 * Figure out how much memory we have, how much
248 * is lowmem, and how much is highmem. If we were
249 * passed the total memory size from the bootloader,
253 total_memory
= boot_mem_size
;
255 total_memory
= ppc_md
.find_end_of_memory();
257 if (__max_memory
&& total_memory
> __max_memory
)
258 total_memory
= __max_memory
;
259 total_lowmem
= total_memory
;
260 #ifdef CONFIG_FSL_BOOKE
261 /* Freescale Book-E parts expect lowmem to be mapped by fixed TLB
262 * entries, so we need to adjust lowmem to match the amount we can map
263 * in the fixed entries */
264 adjust_total_lowmem();
265 #endif /* CONFIG_FSL_BOOKE */
266 if (total_lowmem
> __max_low_memory
) {
267 total_lowmem
= __max_low_memory
;
268 #ifndef CONFIG_HIGHMEM
269 total_memory
= total_lowmem
;
270 #endif /* CONFIG_HIGHMEM */
272 set_phys_avail(total_lowmem
);
274 /* Initialize the MMU hardware */
276 ppc_md
.progress("MMU:hw init", 0x300);
279 /* Map in all of RAM starting at KERNELBASE */
281 ppc_md
.progress("MMU:mapin", 0x301);
284 #ifdef CONFIG_HIGHMEM
285 ioremap_base
= PKMAP_BASE
;
287 ioremap_base
= 0xfe000000UL
; /* for now, could be 0xfffff000 */
288 #endif /* CONFIG_HIGHMEM */
289 ioremap_bot
= ioremap_base
;
291 /* Map in I/O resources */
293 ppc_md
.progress("MMU:setio", 0x302);
294 if (ppc_md
.setup_io_mappings
)
295 ppc_md
.setup_io_mappings();
297 /* Initialize the context management stuff */
301 ppc_md
.progress("MMU:exit", 0x211);
303 #ifdef CONFIG_BOOTX_TEXT
304 /* By default, we are no longer mapped */
305 boot_text_mapped
= 0;
306 /* Must be done last, or ppc_md.progress will die. */
311 /* This is only called until mem_init is done. */
312 void __init
*early_get_page(void)
316 if (init_bootmem_done
) {
317 p
= alloc_bootmem_pages(PAGE_SIZE
);
319 p
= mem_pieces_find(PAGE_SIZE
, PAGE_SIZE
);
325 * Initialize the bootmem system and give it all the memory we
328 void __init
do_init_bootmem(void)
330 unsigned long start
, size
;
334 * Find an area to use for the bootmem bitmap.
335 * We look for the first area which is at least
336 * 128kB in length (128kB is enough for a bitmap
337 * for 4GB of memory, using 4kB pages), plus 1 page
338 * (in case the address isn't page-aligned).
342 for (i
= 0; i
< phys_avail
.n_regions
; ++i
) {
343 unsigned long a
= phys_avail
.regions
[i
].address
;
344 unsigned long s
= phys_avail
.regions
[i
].size
;
349 if (s
>= 33 * PAGE_SIZE
)
352 start
= PAGE_ALIGN(start
);
354 min_low_pfn
= start
>> PAGE_SHIFT
;
355 max_low_pfn
= (PPC_MEMSTART
+ total_lowmem
) >> PAGE_SHIFT
;
356 max_pfn
= (PPC_MEMSTART
+ total_memory
) >> PAGE_SHIFT
;
357 boot_mapsize
= init_bootmem_node(&contig_page_data
, min_low_pfn
,
358 PPC_MEMSTART
>> PAGE_SHIFT
,
361 /* remove the bootmem bitmap from the available memory */
362 mem_pieces_remove(&phys_avail
, start
, boot_mapsize
, 1);
364 /* add everything in phys_avail into the bootmem map */
365 for (i
= 0; i
< phys_avail
.n_regions
; ++i
)
366 free_bootmem(phys_avail
.regions
[i
].address
,
367 phys_avail
.regions
[i
].size
);
369 init_bootmem_done
= 1;
373 * paging_init() sets up the page tables - in fact we've already done this.
375 void __init
paging_init(void)
377 unsigned long zones_size
[MAX_NR_ZONES
], i
;
379 #ifdef CONFIG_HIGHMEM
380 map_page(PKMAP_BASE
, 0, 0); /* XXX gross */
381 pkmap_page_table
= pte_offset_kernel(pmd_offset(pgd_offset_k
382 (PKMAP_BASE
), PKMAP_BASE
), PKMAP_BASE
);
383 map_page(KMAP_FIX_BEGIN
, 0, 0); /* XXX gross */
384 kmap_pte
= pte_offset_kernel(pmd_offset(pgd_offset_k
385 (KMAP_FIX_BEGIN
), KMAP_FIX_BEGIN
), KMAP_FIX_BEGIN
);
386 kmap_prot
= PAGE_KERNEL
;
387 #endif /* CONFIG_HIGHMEM */
390 * All pages are DMA-able so we put them all in the DMA zone.
392 zones_size
[ZONE_DMA
] = total_lowmem
>> PAGE_SHIFT
;
393 for (i
= 1; i
< MAX_NR_ZONES
; i
++)
396 #ifdef CONFIG_HIGHMEM
397 zones_size
[ZONE_HIGHMEM
] = (total_memory
- total_lowmem
) >> PAGE_SHIFT
;
398 #endif /* CONFIG_HIGHMEM */
400 free_area_init(zones_size
);
403 void __init
mem_init(void)
409 #ifdef CONFIG_HIGHMEM
410 unsigned long highmem_mapnr
;
412 highmem_mapnr
= total_lowmem
>> PAGE_SHIFT
;
413 #endif /* CONFIG_HIGHMEM */
414 max_mapnr
= total_memory
>> PAGE_SHIFT
;
416 high_memory
= (void *) __va(PPC_MEMSTART
+ total_lowmem
);
417 num_physpages
= max_mapnr
; /* RAM is assumed contiguous */
419 totalram_pages
+= free_all_bootmem();
421 #ifdef CONFIG_BLK_DEV_INITRD
422 /* if we are booted from BootX with an initial ramdisk,
423 make sure the ramdisk pages aren't reserved. */
425 for (addr
= initrd_start
; addr
< initrd_end
; addr
+= PAGE_SIZE
)
426 ClearPageReserved(virt_to_page(addr
));
428 #endif /* CONFIG_BLK_DEV_INITRD */
431 /* mark the RTAS pages as reserved */
433 for (addr
= (ulong
)__va(rtas_data
);
434 addr
< PAGE_ALIGN((ulong
)__va(rtas_data
)+rtas_size
) ;
436 SetPageReserved(virt_to_page(addr
));
438 #ifdef CONFIG_PPC_PMAC
439 if (agp_special_page
)
440 SetPageReserved(virt_to_page(agp_special_page
));
442 for (addr
= PAGE_OFFSET
; addr
< (unsigned long)high_memory
;
444 if (!PageReserved(virt_to_page(addr
)))
446 if (addr
< (ulong
) etext
)
448 else if (addr
>= (unsigned long)&__init_begin
449 && addr
< (unsigned long)&__init_end
)
451 else if (addr
< (ulong
) klimit
)
455 #ifdef CONFIG_HIGHMEM
459 for (pfn
= highmem_mapnr
; pfn
< max_mapnr
; ++pfn
) {
460 struct page
*page
= mem_map
+ pfn
;
462 ClearPageReserved(page
);
463 set_page_count(page
, 1);
467 totalram_pages
+= totalhigh_pages
;
469 #endif /* CONFIG_HIGHMEM */
471 printk("Memory: %luk available (%dk kernel code, %dk data, %dk init, %ldk highmem)\n",
472 (unsigned long)nr_free_pages()<< (PAGE_SHIFT
-10),
473 codepages
<< (PAGE_SHIFT
-10), datapages
<< (PAGE_SHIFT
-10),
474 initpages
<< (PAGE_SHIFT
-10),
475 (unsigned long) (totalhigh_pages
<< (PAGE_SHIFT
-10)));
477 #ifdef CONFIG_PPC_PMAC
478 if (agp_special_page
)
479 printk(KERN_INFO
"AGP special page: 0x%08lx\n", agp_special_page
);
486 * Set phys_avail to the amount of physical memory,
487 * less the kernel text/data/bss.
490 set_phys_avail(unsigned long total_memory
)
492 unsigned long kstart
, ksize
;
495 * Initially, available physical memory is equivalent to all
499 phys_avail
.regions
[0].address
= PPC_MEMSTART
;
500 phys_avail
.regions
[0].size
= total_memory
;
501 phys_avail
.n_regions
= 1;
504 * Map out the kernel text/data/bss from the available physical
508 kstart
= __pa(_stext
); /* should be 0 */
509 ksize
= PAGE_ALIGN(klimit
- _stext
);
511 mem_pieces_remove(&phys_avail
, kstart
, ksize
, 0);
512 mem_pieces_remove(&phys_avail
, 0, 0x4000, 0);
514 #if defined(CONFIG_BLK_DEV_INITRD)
515 /* Remove the init RAM disk from the available memory. */
517 mem_pieces_remove(&phys_avail
, __pa(initrd_start
),
518 initrd_end
- initrd_start
, 1);
520 #endif /* CONFIG_BLK_DEV_INITRD */
522 /* remove the RTAS pages from the available memory */
524 mem_pieces_remove(&phys_avail
, rtas_data
, rtas_size
, 1);
526 #ifdef CONFIG_PPC_PMAC
527 /* Because of some uninorth weirdness, we need a page of
528 * memory as high as possible (it must be outside of the
529 * bus address seen as the AGP aperture). It will be used
530 * by the r128 DRM driver
532 * FIXME: We need to make sure that page doesn't overlap any of the\
533 * above. This could be done by improving mem_pieces_find to be able
534 * to do a backward search from the end of the list.
536 if (_machine
== _MACH_Pmac
&& find_devices("uni-north-agp")) {
537 agp_special_page
= (total_memory
- PAGE_SIZE
);
538 mem_pieces_remove(&phys_avail
, agp_special_page
, PAGE_SIZE
, 0);
539 agp_special_page
= (unsigned long)__va(agp_special_page
);
541 #endif /* CONFIG_PPC_PMAC */
544 /* Mark some memory as reserved by removing it from phys_avail. */
545 void __init
reserve_phys_mem(unsigned long start
, unsigned long size
)
547 mem_pieces_remove(&phys_avail
, start
, size
, 1);
551 * This is called when a page has been modified by the kernel.
552 * It just marks the page as not i-cache clean. We do the i-cache
553 * flush later when the page is given to a user process, if necessary.
555 void flush_dcache_page(struct page
*page
)
557 clear_bit(PG_arch_1
, &page
->flags
);
560 void flush_dcache_icache_page(struct page
*page
)
563 __flush_dcache_icache(kmap(page
));
566 /* On 8xx there is no need to kmap since highmem is not supported */
567 __flush_dcache_icache(page_address(page
));
569 __flush_dcache_icache_phys(page_to_pfn(page
) << PAGE_SHIFT
);
573 void clear_user_page(void *page
, unsigned long vaddr
, struct page
*pg
)
576 clear_bit(PG_arch_1
, &pg
->flags
);
579 void copy_user_page(void *vto
, void *vfrom
, unsigned long vaddr
,
582 copy_page(vto
, vfrom
);
583 clear_bit(PG_arch_1
, &pg
->flags
);
586 void flush_icache_user_range(struct vm_area_struct
*vma
, struct page
*page
,
587 unsigned long addr
, int len
)
591 maddr
= (unsigned long) kmap(page
) + (addr
& ~PAGE_MASK
);
592 flush_icache_range(maddr
, maddr
+ len
);
597 * This is called at the end of handling a user page fault, when the
598 * fault has been handled by updating a PTE in the linux page tables.
599 * We use it to preload an HPTE into the hash table corresponding to
600 * the updated linux PTE.
602 void update_mmu_cache(struct vm_area_struct
*vma
, unsigned long address
,
605 /* handle i-cache coherency */
606 unsigned long pfn
= pte_pfn(pte
);
608 if (pfn_valid(pfn
)) {
609 struct page
*page
= pfn_to_page(pfn
);
610 if (!PageReserved(page
)
611 && !test_bit(PG_arch_1
, &page
->flags
)) {
612 if (vma
->vm_mm
== current
->active_mm
) {
614 /* On 8xx, cache control instructions (particularly
615 * "dcbst" from flush_dcache_icache) fault as write
616 * operation if there is an unpopulated TLB entry
617 * for the address in question. To workaround that,
618 * we invalidate the TLB here, thus avoiding dcbst
623 __flush_dcache_icache((void *) address
);
625 flush_dcache_icache_page(page
);
626 set_bit(PG_arch_1
, &page
->flags
);
630 #ifdef CONFIG_PPC_STD_MMU
631 /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
632 if (Hash
!= 0 && pte_young(pte
)) {
633 struct mm_struct
*mm
;
636 mm
= (address
< TASK_SIZE
)? vma
->vm_mm
: &init_mm
;
637 pmd
= pmd_offset(pgd_offset(mm
, address
), address
);
639 add_hash_page(mm
->context
, address
, pmd_val(*pmd
));
645 * This is called by /dev/mem to know if a given address has to
646 * be mapped non-cacheable or not
648 int page_is_ram(unsigned long pfn
)
650 unsigned long paddr
= (pfn
<< PAGE_SHIFT
);
652 return paddr
< __pa(high_memory
);
655 pgprot_t
phys_mem_access_prot(struct file
*file
, unsigned long addr
,
656 unsigned long size
, pgprot_t vma_prot
)
658 if (ppc_md
.phys_mem_access_prot
)
659 return ppc_md
.phys_mem_access_prot(file
, addr
, size
, vma_prot
);
661 if (!page_is_ram(addr
>> PAGE_SHIFT
))
662 vma_prot
= __pgprot(pgprot_val(vma_prot
)
663 | _PAGE_GUARDED
| _PAGE_NO_CACHE
);
666 EXPORT_SYMBOL(phys_mem_access_prot
);