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>
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>
45 #include <asm/bootinfo.h>
47 #include "mem_pieces.h"
50 #if defined(CONFIG_KERNEL_START_BOOL) || defined(CONFIG_LOWMEM_SIZE_BOOL)
51 /* The ammount of lowmem must be within 0xF0000000 - KERNELBASE. */
52 #if (CONFIG_LOWMEM_SIZE > (0xF0000000 - KERNELBASE))
53 #error "You must adjust CONFIG_LOWMEM_SIZE or CONFIG_START_KERNEL"
56 #define MAX_LOW_MEM CONFIG_LOWMEM_SIZE
58 DEFINE_PER_CPU(struct mmu_gather
, mmu_gathers
);
60 unsigned long total_memory
;
61 unsigned long total_lowmem
;
63 unsigned long ppc_memstart
;
64 unsigned long ppc_memoffset
= PAGE_OFFSET
;
67 int init_bootmem_done
;
69 #ifdef CONFIG_PPC_PMAC
70 unsigned long agp_special_page
;
74 extern char etext
[], _stext
[];
75 extern char __init_begin
, __init_end
;
76 extern char __prep_begin
, __prep_end
;
77 extern char __chrp_begin
, __chrp_end
;
78 extern char __pmac_begin
, __pmac_end
;
79 extern char __openfirmware_begin
, __openfirmware_end
;
85 EXPORT_SYMBOL(kmap_prot
);
86 EXPORT_SYMBOL(kmap_pte
);
90 void set_phys_avail(unsigned long total_ram
);
92 /* XXX should be in current.h -- paulus */
93 extern struct task_struct
*current_set
[NR_CPUS
];
96 struct mem_pieces phys_avail
;
99 extern unsigned long sysmap_size
;
102 * this tells the system to map all of ram with the segregs
103 * (i.e. page tables) instead of the bats.
106 int __map_without_bats
;
107 int __map_without_ltlbs
;
109 /* max amount of RAM to use */
110 unsigned long __max_memory
;
111 /* max amount of low RAM to map in */
112 unsigned long __max_low_memory
= MAX_LOW_MEM
;
116 int i
,free
= 0,total
= 0,reserved
= 0;
117 int shared
= 0, cached
= 0;
120 printk("Mem-info:\n");
122 printk("Free swap: %6ldkB\n", nr_swap_pages
<<(PAGE_SHIFT
-10));
126 if (PageHighMem(mem_map
+i
))
128 if (PageReserved(mem_map
+i
))
130 else if (PageSwapCache(mem_map
+i
))
132 else if (!page_count(mem_map
+i
))
135 shared
+= page_count(mem_map
+i
) - 1;
137 printk("%d pages of RAM\n",total
);
138 printk("%d pages of HIGHMEM\n", highmem
);
139 printk("%d free pages\n",free
);
140 printk("%d reserved pages\n",reserved
);
141 printk("%d pages shared\n",shared
);
142 printk("%d pages swap cached\n",cached
);
145 /* Free up now-unused memory */
146 static void free_sec(unsigned long start
, unsigned long end
, const char *name
)
148 unsigned long cnt
= 0;
150 while (start
< end
) {
151 ClearPageReserved(virt_to_page(start
));
152 set_page_count(virt_to_page(start
), 1);
158 printk(" %ldk %s", cnt
<< (PAGE_SHIFT
- 10), name
);
159 totalram_pages
+= cnt
;
163 void free_initmem(void)
165 #define FREESEC(TYPE) \
166 free_sec((unsigned long)(&__ ## TYPE ## _begin), \
167 (unsigned long)(&__ ## TYPE ## _end), \
170 printk ("Freeing unused kernel memory:");
172 if (_machine
!= _MACH_Pmac
)
174 if (_machine
!= _MACH_chrp
)
176 if (_machine
!= _MACH_prep
)
179 FREESEC(openfirmware
);
184 #ifdef CONFIG_BLK_DEV_INITRD
185 void free_initrd_mem(unsigned long start
, unsigned long end
)
187 printk ("Freeing initrd memory: %ldk freed\n", (end
- start
) >> 10);
189 for (; start
< end
; start
+= PAGE_SIZE
) {
190 ClearPageReserved(virt_to_page(start
));
191 set_page_count(virt_to_page(start
), 1);
199 * Check for command-line options that affect what MMU_init will do.
203 /* Check for nobats option (used in mapin_ram). */
204 if (strstr(cmd_line
, "nobats")) {
205 __map_without_bats
= 1;
208 if (strstr(cmd_line
, "noltlbs")) {
209 __map_without_ltlbs
= 1;
212 /* Look for mem= option on command line */
213 if (strstr(cmd_line
, "mem=")) {
215 unsigned long maxmem
= 0;
217 for (q
= cmd_line
; (p
= strstr(q
, "mem=")) != 0; ) {
219 if (p
> cmd_line
&& p
[-1] != ' ')
221 maxmem
= simple_strtoul(q
, &q
, 0);
222 if (*q
== 'k' || *q
== 'K') {
225 } else if (*q
== 'm' || *q
== 'M') {
230 __max_memory
= maxmem
;
235 * MMU_init sets up the basic memory mappings for the kernel,
236 * including both RAM and possibly some I/O regions,
237 * and sets up the page tables and the MMU hardware ready to go.
239 void __init
MMU_init(void)
242 ppc_md
.progress("MMU:enter", 0x111);
244 /* parse args from command line */
248 * Figure out how much memory we have, how much
249 * is lowmem, and how much is highmem. If we were
250 * passed the total memory size from the bootloader,
254 total_memory
= boot_mem_size
;
256 total_memory
= ppc_md
.find_end_of_memory();
258 if (__max_memory
&& total_memory
> __max_memory
)
259 total_memory
= __max_memory
;
260 total_lowmem
= total_memory
;
261 #ifdef CONFIG_FSL_BOOKE
262 /* Freescale Book-E parts expect lowmem to be mapped by fixed TLB
263 * entries, so we need to adjust lowmem to match the amount we can map
264 * in the fixed entries */
265 adjust_total_lowmem();
266 #endif /* CONFIG_FSL_BOOKE */
267 if (total_lowmem
> __max_low_memory
) {
268 total_lowmem
= __max_low_memory
;
269 #ifndef CONFIG_HIGHMEM
270 total_memory
= total_lowmem
;
271 #endif /* CONFIG_HIGHMEM */
273 set_phys_avail(total_lowmem
);
275 /* Initialize the MMU hardware */
277 ppc_md
.progress("MMU:hw init", 0x300);
280 /* Map in all of RAM starting at KERNELBASE */
282 ppc_md
.progress("MMU:mapin", 0x301);
285 #ifdef CONFIG_HIGHMEM
286 ioremap_base
= PKMAP_BASE
;
288 ioremap_base
= 0xfe000000UL
; /* for now, could be 0xfffff000 */
289 #endif /* CONFIG_HIGHMEM */
290 ioremap_bot
= ioremap_base
;
292 /* Map in I/O resources */
294 ppc_md
.progress("MMU:setio", 0x302);
295 if (ppc_md
.setup_io_mappings
)
296 ppc_md
.setup_io_mappings();
298 /* Initialize the context management stuff */
302 ppc_md
.progress("MMU:exit", 0x211);
304 #ifdef CONFIG_BOOTX_TEXT
305 /* By default, we are no longer mapped */
306 boot_text_mapped
= 0;
307 /* Must be done last, or ppc_md.progress will die. */
312 /* This is only called until mem_init is done. */
313 void __init
*early_get_page(void)
317 if (init_bootmem_done
) {
318 p
= alloc_bootmem_pages(PAGE_SIZE
);
320 p
= mem_pieces_find(PAGE_SIZE
, PAGE_SIZE
);
326 * Initialize the bootmem system and give it all the memory we
329 void __init
do_init_bootmem(void)
331 unsigned long start
, size
;
335 * Find an area to use for the bootmem bitmap.
336 * We look for the first area which is at least
337 * 128kB in length (128kB is enough for a bitmap
338 * for 4GB of memory, using 4kB pages), plus 1 page
339 * (in case the address isn't page-aligned).
343 for (i
= 0; i
< phys_avail
.n_regions
; ++i
) {
344 unsigned long a
= phys_avail
.regions
[i
].address
;
345 unsigned long s
= phys_avail
.regions
[i
].size
;
350 if (s
>= 33 * PAGE_SIZE
)
353 start
= PAGE_ALIGN(start
);
355 min_low_pfn
= start
>> PAGE_SHIFT
;
356 max_low_pfn
= (PPC_MEMSTART
+ total_lowmem
) >> PAGE_SHIFT
;
357 max_pfn
= (PPC_MEMSTART
+ total_memory
) >> PAGE_SHIFT
;
358 boot_mapsize
= init_bootmem_node(&contig_page_data
, min_low_pfn
,
359 PPC_MEMSTART
>> PAGE_SHIFT
,
362 /* remove the bootmem bitmap from the available memory */
363 mem_pieces_remove(&phys_avail
, start
, boot_mapsize
, 1);
365 /* add everything in phys_avail into the bootmem map */
366 for (i
= 0; i
< phys_avail
.n_regions
; ++i
)
367 free_bootmem(phys_avail
.regions
[i
].address
,
368 phys_avail
.regions
[i
].size
);
370 init_bootmem_done
= 1;
374 * paging_init() sets up the page tables - in fact we've already done this.
376 void __init
paging_init(void)
378 unsigned long zones_size
[MAX_NR_ZONES
], i
;
380 #ifdef CONFIG_HIGHMEM
381 map_page(PKMAP_BASE
, 0, 0); /* XXX gross */
382 pkmap_page_table
= pte_offset_kernel(pmd_offset(pgd_offset_k
383 (PKMAP_BASE
), PKMAP_BASE
), PKMAP_BASE
);
384 map_page(KMAP_FIX_BEGIN
, 0, 0); /* XXX gross */
385 kmap_pte
= pte_offset_kernel(pmd_offset(pgd_offset_k
386 (KMAP_FIX_BEGIN
), KMAP_FIX_BEGIN
), KMAP_FIX_BEGIN
);
387 kmap_prot
= PAGE_KERNEL
;
388 #endif /* CONFIG_HIGHMEM */
391 * All pages are DMA-able so we put them all in the DMA zone.
393 zones_size
[ZONE_DMA
] = total_lowmem
>> PAGE_SHIFT
;
394 for (i
= 1; i
< MAX_NR_ZONES
; i
++)
397 #ifdef CONFIG_HIGHMEM
398 zones_size
[ZONE_HIGHMEM
] = (total_memory
- total_lowmem
) >> PAGE_SHIFT
;
399 #endif /* CONFIG_HIGHMEM */
401 free_area_init(zones_size
);
404 void __init
mem_init(void)
410 #ifdef CONFIG_HIGHMEM
411 unsigned long highmem_mapnr
;
413 highmem_mapnr
= total_lowmem
>> PAGE_SHIFT
;
414 highmem_start_page
= mem_map
+ highmem_mapnr
;
415 #endif /* CONFIG_HIGHMEM */
416 max_mapnr
= total_memory
>> PAGE_SHIFT
;
418 high_memory
= (void *) __va(PPC_MEMSTART
+ total_lowmem
);
419 num_physpages
= max_mapnr
; /* RAM is assumed contiguous */
421 totalram_pages
+= free_all_bootmem();
423 #ifdef CONFIG_BLK_DEV_INITRD
424 /* if we are booted from BootX with an initial ramdisk,
425 make sure the ramdisk pages aren't reserved. */
427 for (addr
= initrd_start
; addr
< initrd_end
; addr
+= PAGE_SIZE
)
428 ClearPageReserved(virt_to_page(addr
));
430 #endif /* CONFIG_BLK_DEV_INITRD */
433 /* mark the RTAS pages as reserved */
435 for (addr
= (ulong
)__va(rtas_data
);
436 addr
< PAGE_ALIGN((ulong
)__va(rtas_data
)+rtas_size
) ;
438 SetPageReserved(virt_to_page(addr
));
440 #ifdef CONFIG_PPC_PMAC
441 if (agp_special_page
)
442 SetPageReserved(virt_to_page(agp_special_page
));
445 for (addr
= (unsigned long)sysmap
;
446 addr
< PAGE_ALIGN((unsigned long)sysmap
+sysmap_size
) ;
448 SetPageReserved(virt_to_page(addr
));
450 for (addr
= PAGE_OFFSET
; addr
< (unsigned long)high_memory
;
452 if (!PageReserved(virt_to_page(addr
)))
454 if (addr
< (ulong
) etext
)
456 else if (addr
>= (unsigned long)&__init_begin
457 && addr
< (unsigned long)&__init_end
)
459 else if (addr
< (ulong
) klimit
)
463 #ifdef CONFIG_HIGHMEM
467 for (pfn
= highmem_mapnr
; pfn
< max_mapnr
; ++pfn
) {
468 struct page
*page
= mem_map
+ pfn
;
470 ClearPageReserved(page
);
471 set_bit(PG_highmem
, &page
->flags
);
472 set_page_count(page
, 1);
476 totalram_pages
+= totalhigh_pages
;
478 #endif /* CONFIG_HIGHMEM */
480 printk("Memory: %luk available (%dk kernel code, %dk data, %dk init, %ldk highmem)\n",
481 (unsigned long)nr_free_pages()<< (PAGE_SHIFT
-10),
482 codepages
<< (PAGE_SHIFT
-10), datapages
<< (PAGE_SHIFT
-10),
483 initpages
<< (PAGE_SHIFT
-10),
484 (unsigned long) (totalhigh_pages
<< (PAGE_SHIFT
-10)));
486 printk("System.map loaded at 0x%08x for debugger, size: %ld bytes\n",
487 (unsigned int)sysmap
, sysmap_size
);
488 #ifdef CONFIG_PPC_PMAC
489 if (agp_special_page
)
490 printk(KERN_INFO
"AGP special page: 0x%08lx\n", agp_special_page
);
493 /* Make sure all our pagetable pages have page->mapping
494 and page->index set correctly. */
495 for (addr
= KERNELBASE
; addr
!= 0; addr
+= PGDIR_SIZE
) {
497 pmd_t
*pmd
= pmd_offset(pgd_offset_k(addr
), addr
);
498 if (pmd_present(*pmd
)) {
500 pg
->mapping
= (void *) &init_mm
;
509 * Set phys_avail to the amount of physical memory,
510 * less the kernel text/data/bss.
513 set_phys_avail(unsigned long total_memory
)
515 unsigned long kstart
, ksize
;
518 * Initially, available physical memory is equivalent to all
522 phys_avail
.regions
[0].address
= PPC_MEMSTART
;
523 phys_avail
.regions
[0].size
= total_memory
;
524 phys_avail
.n_regions
= 1;
527 * Map out the kernel text/data/bss from the available physical
531 kstart
= __pa(_stext
); /* should be 0 */
532 ksize
= PAGE_ALIGN(klimit
- _stext
);
534 mem_pieces_remove(&phys_avail
, kstart
, ksize
, 0);
535 mem_pieces_remove(&phys_avail
, 0, 0x4000, 0);
537 #if defined(CONFIG_BLK_DEV_INITRD)
538 /* Remove the init RAM disk from the available memory. */
540 mem_pieces_remove(&phys_avail
, __pa(initrd_start
),
541 initrd_end
- initrd_start
, 1);
543 #endif /* CONFIG_BLK_DEV_INITRD */
545 /* remove the RTAS pages from the available memory */
547 mem_pieces_remove(&phys_avail
, rtas_data
, rtas_size
, 1);
549 /* remove the sysmap pages from the available memory */
551 mem_pieces_remove(&phys_avail
, __pa(sysmap
), sysmap_size
, 1);
552 #ifdef CONFIG_PPC_PMAC
553 /* Because of some uninorth weirdness, we need a page of
554 * memory as high as possible (it must be outside of the
555 * bus address seen as the AGP aperture). It will be used
556 * by the r128 DRM driver
558 * FIXME: We need to make sure that page doesn't overlap any of the\
559 * above. This could be done by improving mem_pieces_find to be able
560 * to do a backward search from the end of the list.
562 if (_machine
== _MACH_Pmac
&& find_devices("uni-north-agp")) {
563 agp_special_page
= (total_memory
- PAGE_SIZE
);
564 mem_pieces_remove(&phys_avail
, agp_special_page
, PAGE_SIZE
, 0);
565 agp_special_page
= (unsigned long)__va(agp_special_page
);
567 #endif /* CONFIG_PPC_PMAC */
570 /* Mark some memory as reserved by removing it from phys_avail. */
571 void __init
reserve_phys_mem(unsigned long start
, unsigned long size
)
573 mem_pieces_remove(&phys_avail
, start
, size
, 1);
577 * This is called when a page has been modified by the kernel.
578 * It just marks the page as not i-cache clean. We do the i-cache
579 * flush later when the page is given to a user process, if necessary.
581 void flush_dcache_page(struct page
*page
)
583 clear_bit(PG_arch_1
, &page
->flags
);
586 void flush_dcache_icache_page(struct page
*page
)
589 __flush_dcache_icache(kmap(page
));
592 __flush_dcache_icache_phys(page_to_pfn(page
) << PAGE_SHIFT
);
596 void clear_user_page(void *page
, unsigned long vaddr
, struct page
*pg
)
599 clear_bit(PG_arch_1
, &pg
->flags
);
602 void copy_user_page(void *vto
, void *vfrom
, unsigned long vaddr
,
605 copy_page(vto
, vfrom
);
606 clear_bit(PG_arch_1
, &pg
->flags
);
609 void flush_icache_user_range(struct vm_area_struct
*vma
, struct page
*page
,
610 unsigned long addr
, int len
)
614 maddr
= (unsigned long) kmap(page
) + (addr
& ~PAGE_MASK
);
615 flush_icache_range(maddr
, maddr
+ len
);
620 * This is called at the end of handling a user page fault, when the
621 * fault has been handled by updating a PTE in the linux page tables.
622 * We use it to preload an HPTE into the hash table corresponding to
623 * the updated linux PTE.
625 void update_mmu_cache(struct vm_area_struct
*vma
, unsigned long address
,
628 /* handle i-cache coherency */
629 unsigned long pfn
= pte_pfn(pte
);
631 if (pfn_valid(pfn
)) {
632 struct page
*page
= pfn_to_page(pfn
);
633 if (!PageReserved(page
)
634 && !test_bit(PG_arch_1
, &page
->flags
)) {
635 if (vma
->vm_mm
== current
->active_mm
)
636 __flush_dcache_icache((void *) address
);
638 flush_dcache_icache_page(page
);
639 set_bit(PG_arch_1
, &page
->flags
);
643 #ifdef CONFIG_PPC_STD_MMU
644 /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
645 if (Hash
!= 0 && pte_young(pte
)) {
646 struct mm_struct
*mm
;
649 mm
= (address
< TASK_SIZE
)? vma
->vm_mm
: &init_mm
;
650 pmd
= pmd_offset(pgd_offset(mm
, address
), address
);
652 add_hash_page(mm
->context
, address
, pmd_val(*pmd
));