2 * linux/arch/parisc/mm/init.c
4 * Copyright (C) 1995 Linus Torvalds
5 * Copyright 1999 SuSE GmbH
6 * changed by Philipp Rumpf
7 * Copyright 1999 Philipp Rumpf (prumpf@tux.org)
8 * Copyright 2004 Randolph Chung (tausq@debian.org)
9 * Copyright 2006-2007 Helge Deller (deller@gmx.de)
14 #include <linux/module.h>
16 #include <linux/bootmem.h>
17 #include <linux/gfp.h>
18 #include <linux/delay.h>
19 #include <linux/init.h>
20 #include <linux/pci.h> /* for hppa_dma_ops and pcxl_dma_ops */
21 #include <linux/initrd.h>
22 #include <linux/swap.h>
23 #include <linux/unistd.h>
24 #include <linux/nodemask.h> /* for node_online_map */
25 #include <linux/pagemap.h> /* for release_pages and page_cache_release */
27 #include <asm/pgalloc.h>
28 #include <asm/pgtable.h>
30 #include <asm/pdc_chassis.h>
31 #include <asm/mmzone.h>
32 #include <asm/sections.h>
34 extern int data_start
;
37 /* NOTE: This layout exactly conforms to the hybrid L2/L3 page table layout
38 * with the first pmd adjacent to the pgd and below it. gcc doesn't actually
39 * guarantee that global objects will be laid out in memory in the same order
40 * as the order of declaration, so put these in different sections and use
41 * the linker script to order them. */
42 pmd_t pmd0
[PTRS_PER_PMD
] __attribute__ ((__section__ (".data..vm0.pmd"), aligned(PAGE_SIZE
)));
45 pgd_t swapper_pg_dir
[PTRS_PER_PGD
] __attribute__ ((__section__ (".data..vm0.pgd"), aligned(PAGE_SIZE
)));
46 pte_t pg0
[PT_INITIAL
* PTRS_PER_PTE
] __attribute__ ((__section__ (".data..vm0.pte"), aligned(PAGE_SIZE
)));
48 #ifdef CONFIG_DISCONTIGMEM
49 struct node_map_data node_data
[MAX_NUMNODES
] __read_mostly
;
50 unsigned char pfnnid_map
[PFNNID_MAP_MAX
] __read_mostly
;
53 static struct resource data_resource
= {
54 .name
= "Kernel data",
55 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
,
58 static struct resource code_resource
= {
59 .name
= "Kernel code",
60 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
,
63 static struct resource pdcdata_resource
= {
64 .name
= "PDC data (Page Zero)",
67 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
,
70 static struct resource sysram_resources
[MAX_PHYSMEM_RANGES
] __read_mostly
;
72 /* The following array is initialized from the firmware specific
73 * information retrieved in kernel/inventory.c.
76 physmem_range_t pmem_ranges
[MAX_PHYSMEM_RANGES
] __read_mostly
;
77 int npmem_ranges __read_mostly
;
80 #define MAX_MEM (~0UL)
81 #else /* !CONFIG_64BIT */
82 #define MAX_MEM (3584U*1024U*1024U)
83 #endif /* !CONFIG_64BIT */
85 static unsigned long mem_limit __read_mostly
= MAX_MEM
;
87 static void __init
mem_limit_func(void)
92 /* We need this before __setup() functions are called */
95 for (cp
= boot_command_line
; *cp
; ) {
96 if (memcmp(cp
, "mem=", 4) == 0) {
98 limit
= memparse(cp
, &end
);
103 while (*cp
!= ' ' && *cp
)
110 if (limit
< mem_limit
)
114 #define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
116 static void __init
setup_bootmem(void)
118 unsigned long bootmap_size
;
119 unsigned long mem_max
;
120 unsigned long bootmap_pages
;
121 unsigned long bootmap_start_pfn
;
122 unsigned long bootmap_pfn
;
123 #ifndef CONFIG_DISCONTIGMEM
124 physmem_range_t pmem_holes
[MAX_PHYSMEM_RANGES
- 1];
127 int i
, sysram_resource_count
;
129 disable_sr_hashing(); /* Turn off space register hashing */
132 * Sort the ranges. Since the number of ranges is typically
133 * small, and performance is not an issue here, just do
134 * a simple insertion sort.
137 for (i
= 1; i
< npmem_ranges
; i
++) {
140 for (j
= i
; j
> 0; j
--) {
143 if (pmem_ranges
[j
-1].start_pfn
<
144 pmem_ranges
[j
].start_pfn
) {
148 tmp
= pmem_ranges
[j
-1].start_pfn
;
149 pmem_ranges
[j
-1].start_pfn
= pmem_ranges
[j
].start_pfn
;
150 pmem_ranges
[j
].start_pfn
= tmp
;
151 tmp
= pmem_ranges
[j
-1].pages
;
152 pmem_ranges
[j
-1].pages
= pmem_ranges
[j
].pages
;
153 pmem_ranges
[j
].pages
= tmp
;
157 #ifndef CONFIG_DISCONTIGMEM
159 * Throw out ranges that are too far apart (controlled by
163 for (i
= 1; i
< npmem_ranges
; i
++) {
164 if (pmem_ranges
[i
].start_pfn
-
165 (pmem_ranges
[i
-1].start_pfn
+
166 pmem_ranges
[i
-1].pages
) > MAX_GAP
) {
168 printk("Large gap in memory detected (%ld pages). "
169 "Consider turning on CONFIG_DISCONTIGMEM\n",
170 pmem_ranges
[i
].start_pfn
-
171 (pmem_ranges
[i
-1].start_pfn
+
172 pmem_ranges
[i
-1].pages
));
178 if (npmem_ranges
> 1) {
180 /* Print the memory ranges */
182 printk(KERN_INFO
"Memory Ranges:\n");
184 for (i
= 0; i
< npmem_ranges
; i
++) {
188 size
= (pmem_ranges
[i
].pages
<< PAGE_SHIFT
);
189 start
= (pmem_ranges
[i
].start_pfn
<< PAGE_SHIFT
);
190 printk(KERN_INFO
"%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
191 i
,start
, start
+ (size
- 1), size
>> 20);
195 sysram_resource_count
= npmem_ranges
;
196 for (i
= 0; i
< sysram_resource_count
; i
++) {
197 struct resource
*res
= &sysram_resources
[i
];
198 res
->name
= "System RAM";
199 res
->start
= pmem_ranges
[i
].start_pfn
<< PAGE_SHIFT
;
200 res
->end
= res
->start
+ (pmem_ranges
[i
].pages
<< PAGE_SHIFT
)-1;
201 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
202 request_resource(&iomem_resource
, res
);
206 * For 32 bit kernels we limit the amount of memory we can
207 * support, in order to preserve enough kernel address space
208 * for other purposes. For 64 bit kernels we don't normally
209 * limit the memory, but this mechanism can be used to
210 * artificially limit the amount of memory (and it is written
211 * to work with multiple memory ranges).
214 mem_limit_func(); /* check for "mem=" argument */
218 for (i
= 0; i
< npmem_ranges
; i
++) {
221 rsize
= pmem_ranges
[i
].pages
<< PAGE_SHIFT
;
222 if ((mem_max
+ rsize
) > mem_limit
) {
223 printk(KERN_WARNING
"Memory truncated to %ld MB\n", mem_limit
>> 20);
224 if (mem_max
== mem_limit
)
227 pmem_ranges
[i
].pages
= (mem_limit
>> PAGE_SHIFT
)
228 - (mem_max
>> PAGE_SHIFT
);
229 npmem_ranges
= i
+ 1;
232 num_physpages
+= pmem_ranges
[i
].pages
;
235 num_physpages
+= pmem_ranges
[i
].pages
;
239 printk(KERN_INFO
"Total Memory: %ld MB\n",mem_max
>> 20);
241 #ifndef CONFIG_DISCONTIGMEM
242 /* Merge the ranges, keeping track of the holes */
245 unsigned long end_pfn
;
246 unsigned long hole_pages
;
249 end_pfn
= pmem_ranges
[0].start_pfn
+ pmem_ranges
[0].pages
;
250 for (i
= 1; i
< npmem_ranges
; i
++) {
252 hole_pages
= pmem_ranges
[i
].start_pfn
- end_pfn
;
254 pmem_holes
[npmem_holes
].start_pfn
= end_pfn
;
255 pmem_holes
[npmem_holes
++].pages
= hole_pages
;
256 end_pfn
+= hole_pages
;
258 end_pfn
+= pmem_ranges
[i
].pages
;
261 pmem_ranges
[0].pages
= end_pfn
- pmem_ranges
[0].start_pfn
;
267 for (i
= 0; i
< npmem_ranges
; i
++)
268 bootmap_pages
+= bootmem_bootmap_pages(pmem_ranges
[i
].pages
);
270 bootmap_start_pfn
= PAGE_ALIGN(__pa((unsigned long) &_end
)) >> PAGE_SHIFT
;
272 #ifdef CONFIG_DISCONTIGMEM
273 for (i
= 0; i
< MAX_PHYSMEM_RANGES
; i
++) {
274 memset(NODE_DATA(i
), 0, sizeof(pg_data_t
));
275 NODE_DATA(i
)->bdata
= &bootmem_node_data
[i
];
277 memset(pfnnid_map
, 0xff, sizeof(pfnnid_map
));
279 for (i
= 0; i
< npmem_ranges
; i
++) {
280 node_set_state(i
, N_NORMAL_MEMORY
);
286 * Initialize and free the full range of memory in each range.
287 * Note that the only writing these routines do are to the bootmap,
288 * and we've made sure to locate the bootmap properly so that they
289 * won't be writing over anything important.
292 bootmap_pfn
= bootmap_start_pfn
;
294 for (i
= 0; i
< npmem_ranges
; i
++) {
295 unsigned long start_pfn
;
296 unsigned long npages
;
298 start_pfn
= pmem_ranges
[i
].start_pfn
;
299 npages
= pmem_ranges
[i
].pages
;
301 bootmap_size
= init_bootmem_node(NODE_DATA(i
),
304 (start_pfn
+ npages
) );
305 free_bootmem_node(NODE_DATA(i
),
306 (start_pfn
<< PAGE_SHIFT
),
307 (npages
<< PAGE_SHIFT
) );
308 bootmap_pfn
+= (bootmap_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
309 if ((start_pfn
+ npages
) > max_pfn
)
310 max_pfn
= start_pfn
+ npages
;
313 /* IOMMU is always used to access "high mem" on those boxes
314 * that can support enough mem that a PCI device couldn't
315 * directly DMA to any physical addresses.
316 * ISA DMA support will need to revisit this.
318 max_low_pfn
= max_pfn
;
320 /* bootmap sizing messed up? */
321 BUG_ON((bootmap_pfn
- bootmap_start_pfn
) != bootmap_pages
);
323 /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
325 #define PDC_CONSOLE_IO_IODC_SIZE 32768
327 reserve_bootmem_node(NODE_DATA(0), 0UL,
328 (unsigned long)(PAGE0
->mem_free
+
329 PDC_CONSOLE_IO_IODC_SIZE
), BOOTMEM_DEFAULT
);
330 reserve_bootmem_node(NODE_DATA(0), __pa((unsigned long)_text
),
331 (unsigned long)(_end
- _text
), BOOTMEM_DEFAULT
);
332 reserve_bootmem_node(NODE_DATA(0), (bootmap_start_pfn
<< PAGE_SHIFT
),
333 ((bootmap_pfn
- bootmap_start_pfn
) << PAGE_SHIFT
),
336 #ifndef CONFIG_DISCONTIGMEM
338 /* reserve the holes */
340 for (i
= 0; i
< npmem_holes
; i
++) {
341 reserve_bootmem_node(NODE_DATA(0),
342 (pmem_holes
[i
].start_pfn
<< PAGE_SHIFT
),
343 (pmem_holes
[i
].pages
<< PAGE_SHIFT
),
348 #ifdef CONFIG_BLK_DEV_INITRD
350 printk(KERN_INFO
"initrd: %08lx-%08lx\n", initrd_start
, initrd_end
);
351 if (__pa(initrd_start
) < mem_max
) {
352 unsigned long initrd_reserve
;
354 if (__pa(initrd_end
) > mem_max
) {
355 initrd_reserve
= mem_max
- __pa(initrd_start
);
357 initrd_reserve
= initrd_end
- initrd_start
;
359 initrd_below_start_ok
= 1;
360 printk(KERN_INFO
"initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start
), __pa(initrd_start
) + initrd_reserve
, mem_max
);
362 reserve_bootmem_node(NODE_DATA(0), __pa(initrd_start
),
363 initrd_reserve
, BOOTMEM_DEFAULT
);
368 data_resource
.start
= virt_to_phys(&data_start
);
369 data_resource
.end
= virt_to_phys(_end
) - 1;
370 code_resource
.start
= virt_to_phys(_text
);
371 code_resource
.end
= virt_to_phys(&data_start
)-1;
373 /* We don't know which region the kernel will be in, so try
376 for (i
= 0; i
< sysram_resource_count
; i
++) {
377 struct resource
*res
= &sysram_resources
[i
];
378 request_resource(res
, &code_resource
);
379 request_resource(res
, &data_resource
);
381 request_resource(&sysram_resources
[0], &pdcdata_resource
);
384 static void __init
map_pages(unsigned long start_vaddr
,
385 unsigned long start_paddr
, unsigned long size
,
386 pgprot_t pgprot
, int force
)
391 unsigned long end_paddr
;
392 unsigned long start_pmd
;
393 unsigned long start_pte
;
396 unsigned long address
;
398 unsigned long ro_start
;
399 unsigned long ro_end
;
400 unsigned long fv_addr
;
401 unsigned long gw_addr
;
402 extern const unsigned long fault_vector_20
;
403 extern void * const linux_gateway_page
;
405 ro_start
= __pa((unsigned long)_text
);
406 ro_end
= __pa((unsigned long)&data_start
);
407 fv_addr
= __pa((unsigned long)&fault_vector_20
) & PAGE_MASK
;
408 gw_addr
= __pa((unsigned long)&linux_gateway_page
) & PAGE_MASK
;
410 end_paddr
= start_paddr
+ size
;
412 pg_dir
= pgd_offset_k(start_vaddr
);
414 #if PTRS_PER_PMD == 1
417 start_pmd
= ((start_vaddr
>> PMD_SHIFT
) & (PTRS_PER_PMD
- 1));
419 start_pte
= ((start_vaddr
>> PAGE_SHIFT
) & (PTRS_PER_PTE
- 1));
421 address
= start_paddr
;
423 while (address
< end_paddr
) {
424 #if PTRS_PER_PMD == 1
425 pmd
= (pmd_t
*)__pa(pg_dir
);
427 pmd
= (pmd_t
*)pgd_address(*pg_dir
);
430 * pmd is physical at this point
434 pmd
= (pmd_t
*) alloc_bootmem_low_pages_node(NODE_DATA(0), PAGE_SIZE
<< PMD_ORDER
);
435 pmd
= (pmd_t
*) __pa(pmd
);
438 pgd_populate(NULL
, pg_dir
, __va(pmd
));
442 /* now change pmd to kernel virtual addresses */
444 pmd
= (pmd_t
*)__va(pmd
) + start_pmd
;
445 for (tmp1
= start_pmd
; tmp1
< PTRS_PER_PMD
; tmp1
++, pmd
++) {
448 * pg_table is physical at this point
451 pg_table
= (pte_t
*)pmd_address(*pmd
);
454 alloc_bootmem_low_pages_node(NODE_DATA(0), PAGE_SIZE
);
455 pg_table
= (pte_t
*) __pa(pg_table
);
458 pmd_populate_kernel(NULL
, pmd
, __va(pg_table
));
460 /* now change pg_table to kernel virtual addresses */
462 pg_table
= (pte_t
*) __va(pg_table
) + start_pte
;
463 for (tmp2
= start_pte
; tmp2
< PTRS_PER_PTE
; tmp2
++, pg_table
++) {
467 * Map the fault vector writable so we can
468 * write the HPMC checksum.
471 pte
= __mk_pte(address
, pgprot
);
472 else if (core_kernel_text(vaddr
) &&
474 pte
= __mk_pte(address
, PAGE_KERNEL_EXEC
);
476 #if defined(CONFIG_PARISC_PAGE_SIZE_4KB)
477 if (address
>= ro_start
&& address
< ro_end
478 && address
!= fv_addr
479 && address
!= gw_addr
)
480 pte
= __mk_pte(address
, PAGE_KERNEL_RO
);
483 pte
= __mk_pte(address
, pgprot
);
485 if (address
>= end_paddr
) {
492 set_pte(pg_table
, pte
);
494 address
+= PAGE_SIZE
;
499 if (address
>= end_paddr
)
506 void free_initmem(void)
509 unsigned long init_begin
= (unsigned long)__init_begin
;
510 unsigned long init_end
= (unsigned long)__init_end
;
512 /* The init text pages are marked R-X. We have to
513 * flush the icache and mark them RW-
515 * This is tricky, because map_pages is in the init section.
516 * Do a dummy remap of the data section first (the data
517 * section is already PAGE_KERNEL) to pull in the TLB entries
519 map_pages(init_begin
, __pa(init_begin
), init_end
- init_begin
,
521 /* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
523 map_pages(init_begin
, __pa(init_begin
), init_end
- init_begin
,
526 /* force the kernel to see the new TLB entries */
527 __flush_tlb_range(0, init_begin
, init_end
);
528 /* Attempt to catch anyone trying to execute code here
529 * by filling the page with BRK insns.
531 memset((void *)init_begin
, 0x00, init_end
- init_begin
);
532 /* finally dump all the instructions which were cached, since the
533 * pages are no-longer executable */
534 flush_icache_range(init_begin
, init_end
);
536 for (addr
= init_begin
; addr
< init_end
; addr
+= PAGE_SIZE
) {
537 ClearPageReserved(virt_to_page(addr
));
538 init_page_count(virt_to_page(addr
));
544 /* set up a new led state on systems shipped LED State panel */
545 pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE
);
547 printk(KERN_INFO
"Freeing unused kernel memory: %luk freed\n",
548 (init_end
- init_begin
) >> 10);
552 #ifdef CONFIG_DEBUG_RODATA
553 void mark_rodata_ro(void)
555 /* rodata memory was already mapped with KERNEL_RO access rights by
556 pagetable_init() and map_pages(). No need to do additional stuff here */
557 printk (KERN_INFO
"Write protecting the kernel read-only data: %luk\n",
558 (unsigned long)(__end_rodata
- __start_rodata
) >> 10);
564 * Just an arbitrary offset to serve as a "hole" between mapping areas
565 * (between top of physical memory and a potential pcxl dma mapping
566 * area, and below the vmalloc mapping area).
568 * The current 32K value just means that there will be a 32K "hole"
569 * between mapping areas. That means that any out-of-bounds memory
570 * accesses will hopefully be caught. The vmalloc() routines leaves
571 * a hole of 4kB between each vmalloced area for the same reason.
574 /* Leave room for gateway page expansion */
575 #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
576 #error KERNEL_MAP_START is in gateway reserved region
578 #define MAP_START (KERNEL_MAP_START)
580 #define VM_MAP_OFFSET (32*1024)
581 #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
582 & ~(VM_MAP_OFFSET-1)))
584 void *parisc_vmalloc_start __read_mostly
;
585 EXPORT_SYMBOL(parisc_vmalloc_start
);
588 unsigned long pcxl_dma_start __read_mostly
;
591 void __init
mem_init(void)
593 int codesize
, reservedpages
, datasize
, initsize
;
595 /* Do sanity checks on page table constants */
596 BUILD_BUG_ON(PTE_ENTRY_SIZE
!= sizeof(pte_t
));
597 BUILD_BUG_ON(PMD_ENTRY_SIZE
!= sizeof(pmd_t
));
598 BUILD_BUG_ON(PGD_ENTRY_SIZE
!= sizeof(pgd_t
));
599 BUILD_BUG_ON(PAGE_SHIFT
+ BITS_PER_PTE
+ BITS_PER_PMD
+ BITS_PER_PGD
602 high_memory
= __va((max_pfn
<< PAGE_SHIFT
));
604 #ifndef CONFIG_DISCONTIGMEM
605 max_mapnr
= page_to_pfn(virt_to_page(high_memory
- 1)) + 1;
606 totalram_pages
+= free_all_bootmem();
611 for (i
= 0; i
< npmem_ranges
; i
++)
612 totalram_pages
+= free_all_bootmem_node(NODE_DATA(i
));
616 codesize
= (unsigned long)_etext
- (unsigned long)_text
;
617 datasize
= (unsigned long)_edata
- (unsigned long)_etext
;
618 initsize
= (unsigned long)__init_end
- (unsigned long)__init_begin
;
623 #ifdef CONFIG_DISCONTIGMEM
626 for (i
= 0; i
< npmem_ranges
; i
++) {
627 for (pfn
= node_start_pfn(i
); pfn
< node_end_pfn(i
); pfn
++) {
628 if (PageReserved(pfn_to_page(pfn
)))
632 #else /* !CONFIG_DISCONTIGMEM */
633 for (pfn
= 0; pfn
< max_pfn
; pfn
++) {
635 * Only count reserved RAM pages
637 if (PageReserved(pfn_to_page(pfn
)))
644 if (hppa_dma_ops
== &pcxl_dma_ops
) {
645 pcxl_dma_start
= (unsigned long)SET_MAP_OFFSET(MAP_START
);
646 parisc_vmalloc_start
= SET_MAP_OFFSET(pcxl_dma_start
647 + PCXL_DMA_MAP_SIZE
);
650 parisc_vmalloc_start
= SET_MAP_OFFSET(MAP_START
);
653 parisc_vmalloc_start
= SET_MAP_OFFSET(MAP_START
);
656 printk(KERN_INFO
"Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init)\n",
657 nr_free_pages() << (PAGE_SHIFT
-10),
658 num_physpages
<< (PAGE_SHIFT
-10),
660 reservedpages
<< (PAGE_SHIFT
-10),
665 #ifdef CONFIG_DEBUG_KERNEL /* double-sanity-check paranoia */
666 printk("virtual kernel memory layout:\n"
667 " vmalloc : 0x%p - 0x%p (%4ld MB)\n"
668 " memory : 0x%p - 0x%p (%4ld MB)\n"
669 " .init : 0x%p - 0x%p (%4ld kB)\n"
670 " .data : 0x%p - 0x%p (%4ld kB)\n"
671 " .text : 0x%p - 0x%p (%4ld kB)\n",
673 (void*)VMALLOC_START
, (void*)VMALLOC_END
,
674 (VMALLOC_END
- VMALLOC_START
) >> 20,
676 __va(0), high_memory
,
677 ((unsigned long)high_memory
- (unsigned long)__va(0)) >> 20,
679 __init_begin
, __init_end
,
680 ((unsigned long)__init_end
- (unsigned long)__init_begin
) >> 10,
683 ((unsigned long)_edata
- (unsigned long)_etext
) >> 10,
686 ((unsigned long)_etext
- (unsigned long)_text
) >> 10);
690 unsigned long *empty_zero_page __read_mostly
;
691 EXPORT_SYMBOL(empty_zero_page
);
693 void show_mem(unsigned int filter
)
695 int i
,free
= 0,total
= 0,reserved
= 0;
696 int shared
= 0, cached
= 0;
698 printk(KERN_INFO
"Mem-info:\n");
699 show_free_areas(filter
);
700 #ifndef CONFIG_DISCONTIGMEM
704 if (PageReserved(mem_map
+i
))
706 else if (PageSwapCache(mem_map
+i
))
708 else if (!page_count(&mem_map
[i
]))
711 shared
+= page_count(&mem_map
[i
]) - 1;
714 for (i
= 0; i
< npmem_ranges
; i
++) {
717 for (j
= node_start_pfn(i
); j
< node_end_pfn(i
); j
++) {
721 pgdat_resize_lock(NODE_DATA(i
), &flags
);
722 p
= nid_page_nr(i
, j
) - node_start_pfn(i
);
727 else if (PageSwapCache(p
))
729 else if (!page_count(p
))
732 shared
+= page_count(p
) - 1;
733 pgdat_resize_unlock(NODE_DATA(i
), &flags
);
737 printk(KERN_INFO
"%d pages of RAM\n", total
);
738 printk(KERN_INFO
"%d reserved pages\n", reserved
);
739 printk(KERN_INFO
"%d pages shared\n", shared
);
740 printk(KERN_INFO
"%d pages swap cached\n", cached
);
743 #ifdef CONFIG_DISCONTIGMEM
748 for (i
= 0; i
< npmem_ranges
; i
++) {
749 zl
= node_zonelist(i
, 0);
750 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
754 printk("Zone list for zone %d on node %d: ", j
, i
);
755 for_each_zone_zonelist(zone
, z
, zl
, j
)
756 printk("[%d/%s] ", zone_to_nid(zone
),
766 * pagetable_init() sets up the page tables
768 * Note that gateway_init() places the Linux gateway page at page 0.
769 * Since gateway pages cannot be dereferenced this has the desirable
770 * side effect of trapping those pesky NULL-reference errors in the
773 static void __init
pagetable_init(void)
777 /* Map each physical memory range to its kernel vaddr */
779 for (range
= 0; range
< npmem_ranges
; range
++) {
780 unsigned long start_paddr
;
781 unsigned long end_paddr
;
784 start_paddr
= pmem_ranges
[range
].start_pfn
<< PAGE_SHIFT
;
785 end_paddr
= start_paddr
+ (pmem_ranges
[range
].pages
<< PAGE_SHIFT
);
786 size
= pmem_ranges
[range
].pages
<< PAGE_SHIFT
;
788 map_pages((unsigned long)__va(start_paddr
), start_paddr
,
789 size
, PAGE_KERNEL
, 0);
792 #ifdef CONFIG_BLK_DEV_INITRD
793 if (initrd_end
&& initrd_end
> mem_limit
) {
794 printk(KERN_INFO
"initrd: mapping %08lx-%08lx\n", initrd_start
, initrd_end
);
795 map_pages(initrd_start
, __pa(initrd_start
),
796 initrd_end
- initrd_start
, PAGE_KERNEL
, 0);
800 empty_zero_page
= alloc_bootmem_pages(PAGE_SIZE
);
801 memset(empty_zero_page
, 0, PAGE_SIZE
);
804 static void __init
gateway_init(void)
806 unsigned long linux_gateway_page_addr
;
807 /* FIXME: This is 'const' in order to trick the compiler
808 into not treating it as DP-relative data. */
809 extern void * const linux_gateway_page
;
811 linux_gateway_page_addr
= LINUX_GATEWAY_ADDR
& PAGE_MASK
;
814 * Setup Linux Gateway page.
816 * The Linux gateway page will reside in kernel space (on virtual
817 * page 0), so it doesn't need to be aliased into user space.
820 map_pages(linux_gateway_page_addr
, __pa(&linux_gateway_page
),
821 PAGE_SIZE
, PAGE_GATEWAY
, 1);
826 map_hpux_gateway_page(struct task_struct
*tsk
, struct mm_struct
*mm
)
831 unsigned long start_pmd
;
832 unsigned long start_pte
;
833 unsigned long address
;
834 unsigned long hpux_gw_page_addr
;
835 /* FIXME: This is 'const' in order to trick the compiler
836 into not treating it as DP-relative data. */
837 extern void * const hpux_gateway_page
;
839 hpux_gw_page_addr
= HPUX_GATEWAY_ADDR
& PAGE_MASK
;
842 * Setup HP-UX Gateway page.
844 * The HP-UX gateway page resides in the user address space,
845 * so it needs to be aliased into each process.
848 pg_dir
= pgd_offset(mm
,hpux_gw_page_addr
);
850 #if PTRS_PER_PMD == 1
853 start_pmd
= ((hpux_gw_page_addr
>> PMD_SHIFT
) & (PTRS_PER_PMD
- 1));
855 start_pte
= ((hpux_gw_page_addr
>> PAGE_SHIFT
) & (PTRS_PER_PTE
- 1));
857 address
= __pa(&hpux_gateway_page
);
858 #if PTRS_PER_PMD == 1
859 pmd
= (pmd_t
*)__pa(pg_dir
);
861 pmd
= (pmd_t
*) pgd_address(*pg_dir
);
864 * pmd is physical at this point
868 pmd
= (pmd_t
*) get_zeroed_page(GFP_KERNEL
);
869 pmd
= (pmd_t
*) __pa(pmd
);
872 __pgd_val_set(*pg_dir
, PxD_FLAG_PRESENT
| PxD_FLAG_VALID
| (unsigned long) pmd
);
874 /* now change pmd to kernel virtual addresses */
876 pmd
= (pmd_t
*)__va(pmd
) + start_pmd
;
879 * pg_table is physical at this point
882 pg_table
= (pte_t
*) pmd_address(*pmd
);
884 pg_table
= (pte_t
*) __pa(get_zeroed_page(GFP_KERNEL
));
886 __pmd_val_set(*pmd
, PxD_FLAG_PRESENT
| PxD_FLAG_VALID
| (unsigned long) pg_table
);
888 /* now change pg_table to kernel virtual addresses */
890 pg_table
= (pte_t
*) __va(pg_table
) + start_pte
;
891 set_pte(pg_table
, __mk_pte(address
, PAGE_GATEWAY
));
893 EXPORT_SYMBOL(map_hpux_gateway_page
);
896 void __init
paging_init(void)
903 flush_cache_all_local(); /* start with known state */
904 flush_tlb_all_local(NULL
);
906 for (i
= 0; i
< npmem_ranges
; i
++) {
907 unsigned long zones_size
[MAX_NR_ZONES
] = { 0, };
909 zones_size
[ZONE_NORMAL
] = pmem_ranges
[i
].pages
;
911 #ifdef CONFIG_DISCONTIGMEM
912 /* Need to initialize the pfnnid_map before we can initialize
916 for (j
= (pmem_ranges
[i
].start_pfn
>> PFNNID_SHIFT
);
917 j
<= ((pmem_ranges
[i
].start_pfn
+ pmem_ranges
[i
].pages
) >> PFNNID_SHIFT
);
924 free_area_init_node(i
, zones_size
,
925 pmem_ranges
[i
].start_pfn
, NULL
);
932 * Currently, all PA20 chips have 18 bit protection IDs, which is the
933 * limiting factor (space ids are 32 bits).
936 #define NR_SPACE_IDS 262144
941 * Currently we have a one-to-one relationship between space IDs and
942 * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
943 * support 15 bit protection IDs, so that is the limiting factor.
944 * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
945 * probably not worth the effort for a special case here.
948 #define NR_SPACE_IDS 32768
950 #endif /* !CONFIG_PA20 */
952 #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
953 #define SID_ARRAY_SIZE (NR_SPACE_IDS / (8 * sizeof(long)))
955 static unsigned long space_id
[SID_ARRAY_SIZE
] = { 1 }; /* disallow space 0 */
956 static unsigned long dirty_space_id
[SID_ARRAY_SIZE
];
957 static unsigned long space_id_index
;
958 static unsigned long free_space_ids
= NR_SPACE_IDS
- 1;
959 static unsigned long dirty_space_ids
= 0;
961 static DEFINE_SPINLOCK(sid_lock
);
963 unsigned long alloc_sid(void)
967 spin_lock(&sid_lock
);
969 if (free_space_ids
== 0) {
970 if (dirty_space_ids
!= 0) {
971 spin_unlock(&sid_lock
);
972 flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
973 spin_lock(&sid_lock
);
975 BUG_ON(free_space_ids
== 0);
980 index
= find_next_zero_bit(space_id
, NR_SPACE_IDS
, space_id_index
);
981 space_id
[index
>> SHIFT_PER_LONG
] |= (1L << (index
& (BITS_PER_LONG
- 1)));
982 space_id_index
= index
;
984 spin_unlock(&sid_lock
);
986 return index
<< SPACEID_SHIFT
;
989 void free_sid(unsigned long spaceid
)
991 unsigned long index
= spaceid
>> SPACEID_SHIFT
;
992 unsigned long *dirty_space_offset
;
994 dirty_space_offset
= dirty_space_id
+ (index
>> SHIFT_PER_LONG
);
995 index
&= (BITS_PER_LONG
- 1);
997 spin_lock(&sid_lock
);
999 BUG_ON(*dirty_space_offset
& (1L << index
)); /* attempt to free space id twice */
1001 *dirty_space_offset
|= (1L << index
);
1004 spin_unlock(&sid_lock
);
1009 static void get_dirty_sids(unsigned long *ndirtyptr
,unsigned long *dirty_array
)
1013 /* NOTE: sid_lock must be held upon entry */
1015 *ndirtyptr
= dirty_space_ids
;
1016 if (dirty_space_ids
!= 0) {
1017 for (i
= 0; i
< SID_ARRAY_SIZE
; i
++) {
1018 dirty_array
[i
] = dirty_space_id
[i
];
1019 dirty_space_id
[i
] = 0;
1021 dirty_space_ids
= 0;
1027 static void recycle_sids(unsigned long ndirty
,unsigned long *dirty_array
)
1031 /* NOTE: sid_lock must be held upon entry */
1034 for (i
= 0; i
< SID_ARRAY_SIZE
; i
++) {
1035 space_id
[i
] ^= dirty_array
[i
];
1038 free_space_ids
+= ndirty
;
1043 #else /* CONFIG_SMP */
1045 static void recycle_sids(void)
1049 /* NOTE: sid_lock must be held upon entry */
1051 if (dirty_space_ids
!= 0) {
1052 for (i
= 0; i
< SID_ARRAY_SIZE
; i
++) {
1053 space_id
[i
] ^= dirty_space_id
[i
];
1054 dirty_space_id
[i
] = 0;
1057 free_space_ids
+= dirty_space_ids
;
1058 dirty_space_ids
= 0;
1065 * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
1066 * purged, we can safely reuse the space ids that were released but
1067 * not flushed from the tlb.
1072 static unsigned long recycle_ndirty
;
1073 static unsigned long recycle_dirty_array
[SID_ARRAY_SIZE
];
1074 static unsigned int recycle_inuse
;
1076 void flush_tlb_all(void)
1081 spin_lock(&sid_lock
);
1082 if (dirty_space_ids
> RECYCLE_THRESHOLD
) {
1083 BUG_ON(recycle_inuse
); /* FIXME: Use a semaphore/wait queue here */
1084 get_dirty_sids(&recycle_ndirty
,recycle_dirty_array
);
1088 spin_unlock(&sid_lock
);
1089 on_each_cpu(flush_tlb_all_local
, NULL
, 1);
1091 spin_lock(&sid_lock
);
1092 recycle_sids(recycle_ndirty
,recycle_dirty_array
);
1094 spin_unlock(&sid_lock
);
1098 void flush_tlb_all(void)
1100 spin_lock(&sid_lock
);
1101 flush_tlb_all_local(NULL
);
1103 spin_unlock(&sid_lock
);
1107 #ifdef CONFIG_BLK_DEV_INITRD
1108 void free_initrd_mem(unsigned long start
, unsigned long end
)
1112 printk(KERN_INFO
"Freeing initrd memory: %ldk freed\n", (end
- start
) >> 10);
1113 for (; start
< end
; start
+= PAGE_SIZE
) {
1114 ClearPageReserved(virt_to_page(start
));
1115 init_page_count(virt_to_page(start
));