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 Helge Deller (deller@gmx.de)
14 #include <linux/module.h>
16 #include <linux/bootmem.h>
17 #include <linux/delay.h>
18 #include <linux/init.h>
19 #include <linux/pci.h> /* for hppa_dma_ops and pcxl_dma_ops */
20 #include <linux/initrd.h>
21 #include <linux/swap.h>
22 #include <linux/unistd.h>
23 #include <linux/nodemask.h> /* for node_online_map */
24 #include <linux/pagemap.h> /* for release_pages and page_cache_release */
26 #include <asm/pgalloc.h>
28 #include <asm/pdc_chassis.h>
29 #include <asm/mmzone.h>
31 DEFINE_PER_CPU(struct mmu_gather
, mmu_gathers
);
33 extern char _text
; /* start of kernel code, defined by linker */
34 extern int data_start
;
35 extern char _end
; /* end of BSS, defined by linker */
36 extern char __init_begin
, __init_end
;
38 #ifdef CONFIG_DISCONTIGMEM
39 struct node_map_data node_data
[MAX_NUMNODES
] __read_mostly
;
40 bootmem_data_t bmem_data
[MAX_NUMNODES
] __read_mostly
;
41 unsigned char pfnnid_map
[PFNNID_MAP_MAX
] __read_mostly
;
44 static struct resource data_resource
= {
45 .name
= "Kernel data",
46 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
,
49 static struct resource code_resource
= {
50 .name
= "Kernel code",
51 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
,
54 static struct resource pdcdata_resource
= {
55 .name
= "PDC data (Page Zero)",
58 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
,
61 static struct resource sysram_resources
[MAX_PHYSMEM_RANGES
] __read_mostly
;
63 /* The following array is initialized from the firmware specific
64 * information retrieved in kernel/inventory.c.
67 physmem_range_t pmem_ranges
[MAX_PHYSMEM_RANGES
] __read_mostly
;
68 int npmem_ranges __read_mostly
;
71 #define MAX_MEM (~0UL)
73 #define MAX_MEM (3584U*1024U*1024U)
74 #endif /* !__LP64__ */
76 static unsigned long mem_limit __read_mostly
= MAX_MEM
;
78 static void __init
mem_limit_func(void)
82 extern char saved_command_line
[];
84 /* We need this before __setup() functions are called */
87 for (cp
= saved_command_line
; *cp
; ) {
88 if (memcmp(cp
, "mem=", 4) == 0) {
90 limit
= memparse(cp
, &end
);
95 while (*cp
!= ' ' && *cp
)
102 if (limit
< mem_limit
)
106 #define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
108 static void __init
setup_bootmem(void)
110 unsigned long bootmap_size
;
111 unsigned long mem_max
;
112 unsigned long bootmap_pages
;
113 unsigned long bootmap_start_pfn
;
114 unsigned long bootmap_pfn
;
115 #ifndef CONFIG_DISCONTIGMEM
116 physmem_range_t pmem_holes
[MAX_PHYSMEM_RANGES
- 1];
119 int i
, sysram_resource_count
;
121 disable_sr_hashing(); /* Turn off space register hashing */
124 * Sort the ranges. Since the number of ranges is typically
125 * small, and performance is not an issue here, just do
126 * a simple insertion sort.
129 for (i
= 1; i
< npmem_ranges
; i
++) {
132 for (j
= i
; j
> 0; j
--) {
135 if (pmem_ranges
[j
-1].start_pfn
<
136 pmem_ranges
[j
].start_pfn
) {
140 tmp
= pmem_ranges
[j
-1].start_pfn
;
141 pmem_ranges
[j
-1].start_pfn
= pmem_ranges
[j
].start_pfn
;
142 pmem_ranges
[j
].start_pfn
= tmp
;
143 tmp
= pmem_ranges
[j
-1].pages
;
144 pmem_ranges
[j
-1].pages
= pmem_ranges
[j
].pages
;
145 pmem_ranges
[j
].pages
= tmp
;
149 #ifndef CONFIG_DISCONTIGMEM
151 * Throw out ranges that are too far apart (controlled by
155 for (i
= 1; i
< npmem_ranges
; i
++) {
156 if (pmem_ranges
[i
].start_pfn
-
157 (pmem_ranges
[i
-1].start_pfn
+
158 pmem_ranges
[i
-1].pages
) > MAX_GAP
) {
160 printk("Large gap in memory detected (%ld pages). "
161 "Consider turning on CONFIG_DISCONTIGMEM\n",
162 pmem_ranges
[i
].start_pfn
-
163 (pmem_ranges
[i
-1].start_pfn
+
164 pmem_ranges
[i
-1].pages
));
170 if (npmem_ranges
> 1) {
172 /* Print the memory ranges */
174 printk(KERN_INFO
"Memory Ranges:\n");
176 for (i
= 0; i
< npmem_ranges
; i
++) {
180 size
= (pmem_ranges
[i
].pages
<< PAGE_SHIFT
);
181 start
= (pmem_ranges
[i
].start_pfn
<< PAGE_SHIFT
);
182 printk(KERN_INFO
"%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
183 i
,start
, start
+ (size
- 1), size
>> 20);
187 sysram_resource_count
= npmem_ranges
;
188 for (i
= 0; i
< sysram_resource_count
; i
++) {
189 struct resource
*res
= &sysram_resources
[i
];
190 res
->name
= "System RAM";
191 res
->start
= pmem_ranges
[i
].start_pfn
<< PAGE_SHIFT
;
192 res
->end
= res
->start
+ (pmem_ranges
[i
].pages
<< PAGE_SHIFT
)-1;
193 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
194 request_resource(&iomem_resource
, res
);
198 * For 32 bit kernels we limit the amount of memory we can
199 * support, in order to preserve enough kernel address space
200 * for other purposes. For 64 bit kernels we don't normally
201 * limit the memory, but this mechanism can be used to
202 * artificially limit the amount of memory (and it is written
203 * to work with multiple memory ranges).
206 mem_limit_func(); /* check for "mem=" argument */
210 for (i
= 0; i
< npmem_ranges
; i
++) {
213 rsize
= pmem_ranges
[i
].pages
<< PAGE_SHIFT
;
214 if ((mem_max
+ rsize
) > mem_limit
) {
215 printk(KERN_WARNING
"Memory truncated to %ld MB\n", mem_limit
>> 20);
216 if (mem_max
== mem_limit
)
219 pmem_ranges
[i
].pages
= (mem_limit
>> PAGE_SHIFT
)
220 - (mem_max
>> PAGE_SHIFT
);
221 npmem_ranges
= i
+ 1;
224 num_physpages
+= pmem_ranges
[i
].pages
;
227 num_physpages
+= pmem_ranges
[i
].pages
;
231 printk(KERN_INFO
"Total Memory: %ld MB\n",mem_max
>> 20);
233 #ifndef CONFIG_DISCONTIGMEM
234 /* Merge the ranges, keeping track of the holes */
237 unsigned long end_pfn
;
238 unsigned long hole_pages
;
241 end_pfn
= pmem_ranges
[0].start_pfn
+ pmem_ranges
[0].pages
;
242 for (i
= 1; i
< npmem_ranges
; i
++) {
244 hole_pages
= pmem_ranges
[i
].start_pfn
- end_pfn
;
246 pmem_holes
[npmem_holes
].start_pfn
= end_pfn
;
247 pmem_holes
[npmem_holes
++].pages
= hole_pages
;
248 end_pfn
+= hole_pages
;
250 end_pfn
+= pmem_ranges
[i
].pages
;
253 pmem_ranges
[0].pages
= end_pfn
- pmem_ranges
[0].start_pfn
;
259 for (i
= 0; i
< npmem_ranges
; i
++)
260 bootmap_pages
+= bootmem_bootmap_pages(pmem_ranges
[i
].pages
);
262 bootmap_start_pfn
= PAGE_ALIGN(__pa((unsigned long) &_end
)) >> PAGE_SHIFT
;
264 #ifdef CONFIG_DISCONTIGMEM
265 for (i
= 0; i
< MAX_PHYSMEM_RANGES
; i
++) {
266 memset(NODE_DATA(i
), 0, sizeof(pg_data_t
));
267 NODE_DATA(i
)->bdata
= &bmem_data
[i
];
269 memset(pfnnid_map
, 0xff, sizeof(pfnnid_map
));
271 for (i
= 0; i
< npmem_ranges
; i
++)
276 * Initialize and free the full range of memory in each range.
277 * Note that the only writing these routines do are to the bootmap,
278 * and we've made sure to locate the bootmap properly so that they
279 * won't be writing over anything important.
282 bootmap_pfn
= bootmap_start_pfn
;
284 for (i
= 0; i
< npmem_ranges
; i
++) {
285 unsigned long start_pfn
;
286 unsigned long npages
;
288 start_pfn
= pmem_ranges
[i
].start_pfn
;
289 npages
= pmem_ranges
[i
].pages
;
291 bootmap_size
= init_bootmem_node(NODE_DATA(i
),
294 (start_pfn
+ npages
) );
295 free_bootmem_node(NODE_DATA(i
),
296 (start_pfn
<< PAGE_SHIFT
),
297 (npages
<< PAGE_SHIFT
) );
298 bootmap_pfn
+= (bootmap_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
299 if ((start_pfn
+ npages
) > max_pfn
)
300 max_pfn
= start_pfn
+ npages
;
303 /* IOMMU is always used to access "high mem" on those boxes
304 * that can support enough mem that a PCI device couldn't
305 * directly DMA to any physical addresses.
306 * ISA DMA support will need to revisit this.
308 max_low_pfn
= max_pfn
;
310 if ((bootmap_pfn
- bootmap_start_pfn
) != bootmap_pages
) {
311 printk(KERN_WARNING
"WARNING! bootmap sizing is messed up!\n");
315 /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
317 #define PDC_CONSOLE_IO_IODC_SIZE 32768
319 reserve_bootmem_node(NODE_DATA(0), 0UL,
320 (unsigned long)(PAGE0
->mem_free
+ PDC_CONSOLE_IO_IODC_SIZE
));
321 reserve_bootmem_node(NODE_DATA(0),__pa((unsigned long)&_text
),
322 (unsigned long)(&_end
- &_text
));
323 reserve_bootmem_node(NODE_DATA(0), (bootmap_start_pfn
<< PAGE_SHIFT
),
324 ((bootmap_pfn
- bootmap_start_pfn
) << PAGE_SHIFT
));
326 #ifndef CONFIG_DISCONTIGMEM
328 /* reserve the holes */
330 for (i
= 0; i
< npmem_holes
; i
++) {
331 reserve_bootmem_node(NODE_DATA(0),
332 (pmem_holes
[i
].start_pfn
<< PAGE_SHIFT
),
333 (pmem_holes
[i
].pages
<< PAGE_SHIFT
));
337 #ifdef CONFIG_BLK_DEV_INITRD
339 printk(KERN_INFO
"initrd: %08lx-%08lx\n", initrd_start
, initrd_end
);
340 if (__pa(initrd_start
) < mem_max
) {
341 unsigned long initrd_reserve
;
343 if (__pa(initrd_end
) > mem_max
) {
344 initrd_reserve
= mem_max
- __pa(initrd_start
);
346 initrd_reserve
= initrd_end
- initrd_start
;
348 initrd_below_start_ok
= 1;
349 printk(KERN_INFO
"initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start
), __pa(initrd_start
) + initrd_reserve
, mem_max
);
351 reserve_bootmem_node(NODE_DATA(0),__pa(initrd_start
), initrd_reserve
);
356 data_resource
.start
= virt_to_phys(&data_start
);
357 data_resource
.end
= virt_to_phys(&_end
)-1;
358 code_resource
.start
= virt_to_phys(&_text
);
359 code_resource
.end
= virt_to_phys(&data_start
)-1;
361 /* We don't know which region the kernel will be in, so try
364 for (i
= 0; i
< sysram_resource_count
; i
++) {
365 struct resource
*res
= &sysram_resources
[i
];
366 request_resource(res
, &code_resource
);
367 request_resource(res
, &data_resource
);
369 request_resource(&sysram_resources
[0], &pdcdata_resource
);
372 void free_initmem(void)
374 unsigned long addr
, init_begin
, init_end
;
376 printk(KERN_INFO
"Freeing unused kernel memory: ");
378 #ifdef CONFIG_DEBUG_KERNEL
379 /* Attempt to catch anyone trying to execute code here
380 * by filling the page with BRK insns.
382 * If we disable interrupts for all CPUs, then IPI stops working.
383 * Kinda breaks the global cache flushing.
387 memset(&__init_begin
, 0x00,
388 (unsigned long)&__init_end
- (unsigned long)&__init_begin
);
391 asm volatile("sync" : : );
392 flush_icache_range((unsigned long)&__init_begin
, (unsigned long)&__init_end
);
393 asm volatile("sync" : : );
398 /* align __init_begin and __init_end to page size,
399 ignoring linker script where we might have tried to save RAM */
400 init_begin
= PAGE_ALIGN((unsigned long)(&__init_begin
));
401 init_end
= PAGE_ALIGN((unsigned long)(&__init_end
));
402 for (addr
= init_begin
; addr
< init_end
; addr
+= PAGE_SIZE
) {
403 ClearPageReserved(virt_to_page(addr
));
404 init_page_count(virt_to_page(addr
));
410 /* set up a new led state on systems shipped LED State panel */
411 pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE
);
413 printk("%luk freed\n", (init_end
- init_begin
) >> 10);
417 #ifdef CONFIG_DEBUG_RODATA
418 void mark_rodata_ro(void)
420 extern char __start_rodata
, __end_rodata
;
421 /* rodata memory was already mapped with KERNEL_RO access rights by
422 pagetable_init() and map_pages(). No need to do additional stuff here */
423 printk (KERN_INFO
"Write protecting the kernel read-only data: %luk\n",
424 (unsigned long)(&__end_rodata
- &__start_rodata
) >> 10);
430 * Just an arbitrary offset to serve as a "hole" between mapping areas
431 * (between top of physical memory and a potential pcxl dma mapping
432 * area, and below the vmalloc mapping area).
434 * The current 32K value just means that there will be a 32K "hole"
435 * between mapping areas. That means that any out-of-bounds memory
436 * accesses will hopefully be caught. The vmalloc() routines leaves
437 * a hole of 4kB between each vmalloced area for the same reason.
440 /* Leave room for gateway page expansion */
441 #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
442 #error KERNEL_MAP_START is in gateway reserved region
444 #define MAP_START (KERNEL_MAP_START)
446 #define VM_MAP_OFFSET (32*1024)
447 #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
448 & ~(VM_MAP_OFFSET-1)))
450 void *vmalloc_start __read_mostly
;
451 EXPORT_SYMBOL(vmalloc_start
);
454 unsigned long pcxl_dma_start __read_mostly
;
457 void __init
mem_init(void)
459 high_memory
= __va((max_pfn
<< PAGE_SHIFT
));
461 #ifndef CONFIG_DISCONTIGMEM
462 max_mapnr
= page_to_pfn(virt_to_page(high_memory
- 1)) + 1;
463 totalram_pages
+= free_all_bootmem();
468 for (i
= 0; i
< npmem_ranges
; i
++)
469 totalram_pages
+= free_all_bootmem_node(NODE_DATA(i
));
473 printk(KERN_INFO
"Memory: %luk available\n", num_physpages
<< (PAGE_SHIFT
-10));
476 if (hppa_dma_ops
== &pcxl_dma_ops
) {
477 pcxl_dma_start
= (unsigned long)SET_MAP_OFFSET(MAP_START
);
478 vmalloc_start
= SET_MAP_OFFSET(pcxl_dma_start
+ PCXL_DMA_MAP_SIZE
);
481 vmalloc_start
= SET_MAP_OFFSET(MAP_START
);
484 vmalloc_start
= SET_MAP_OFFSET(MAP_START
);
489 unsigned long *empty_zero_page __read_mostly
;
493 int i
,free
= 0,total
= 0,reserved
= 0;
494 int shared
= 0, cached
= 0;
496 printk(KERN_INFO
"Mem-info:\n");
498 printk(KERN_INFO
"Free swap: %6ldkB\n",
499 nr_swap_pages
<<(PAGE_SHIFT
-10));
500 #ifndef CONFIG_DISCONTIGMEM
504 if (PageReserved(mem_map
+i
))
506 else if (PageSwapCache(mem_map
+i
))
508 else if (!page_count(&mem_map
[i
]))
511 shared
+= page_count(&mem_map
[i
]) - 1;
514 for (i
= 0; i
< npmem_ranges
; i
++) {
517 for (j
= node_start_pfn(i
); j
< node_end_pfn(i
); j
++) {
521 pgdat_resize_lock(NODE_DATA(i
), &flags
);
522 p
= nid_page_nr(i
, j
) - node_start_pfn(i
);
527 else if (PageSwapCache(p
))
529 else if (!page_count(p
))
532 shared
+= page_count(p
) - 1;
533 pgdat_resize_unlock(NODE_DATA(i
), &flags
);
537 printk(KERN_INFO
"%d pages of RAM\n", total
);
538 printk(KERN_INFO
"%d reserved pages\n", reserved
);
539 printk(KERN_INFO
"%d pages shared\n", shared
);
540 printk(KERN_INFO
"%d pages swap cached\n", cached
);
543 #ifdef CONFIG_DISCONTIGMEM
548 for (i
= 0; i
< npmem_ranges
; i
++) {
549 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
550 zl
= NODE_DATA(i
)->node_zonelists
+ j
;
552 printk("Zone list for zone %d on node %d: ", j
, i
);
553 for (k
= 0; zl
->zones
[k
] != NULL
; k
++)
554 printk("[%d/%s] ", zl
->zones
[k
]->zone_pgdat
->node_id
, zl
->zones
[k
]->name
);
563 static void __init
map_pages(unsigned long start_vaddr
, unsigned long start_paddr
, unsigned long size
, pgprot_t pgprot
)
568 unsigned long end_paddr
;
569 unsigned long start_pmd
;
570 unsigned long start_pte
;
573 unsigned long address
;
574 unsigned long ro_start
;
575 unsigned long ro_end
;
576 unsigned long fv_addr
;
577 unsigned long gw_addr
;
578 extern const unsigned long fault_vector_20
;
579 extern void * const linux_gateway_page
;
581 ro_start
= __pa((unsigned long)&_text
);
582 ro_end
= __pa((unsigned long)&data_start
);
583 fv_addr
= __pa((unsigned long)&fault_vector_20
) & PAGE_MASK
;
584 gw_addr
= __pa((unsigned long)&linux_gateway_page
) & PAGE_MASK
;
586 end_paddr
= start_paddr
+ size
;
588 pg_dir
= pgd_offset_k(start_vaddr
);
590 #if PTRS_PER_PMD == 1
593 start_pmd
= ((start_vaddr
>> PMD_SHIFT
) & (PTRS_PER_PMD
- 1));
595 start_pte
= ((start_vaddr
>> PAGE_SHIFT
) & (PTRS_PER_PTE
- 1));
597 address
= start_paddr
;
598 while (address
< end_paddr
) {
599 #if PTRS_PER_PMD == 1
600 pmd
= (pmd_t
*)__pa(pg_dir
);
602 pmd
= (pmd_t
*)pgd_address(*pg_dir
);
605 * pmd is physical at this point
609 pmd
= (pmd_t
*) alloc_bootmem_low_pages_node(NODE_DATA(0),PAGE_SIZE
<< PMD_ORDER
);
610 pmd
= (pmd_t
*) __pa(pmd
);
613 pgd_populate(NULL
, pg_dir
, __va(pmd
));
617 /* now change pmd to kernel virtual addresses */
619 pmd
= (pmd_t
*)__va(pmd
) + start_pmd
;
620 for (tmp1
= start_pmd
; tmp1
< PTRS_PER_PMD
; tmp1
++,pmd
++) {
623 * pg_table is physical at this point
626 pg_table
= (pte_t
*)pmd_address(*pmd
);
629 alloc_bootmem_low_pages_node(NODE_DATA(0),PAGE_SIZE
);
630 pg_table
= (pte_t
*) __pa(pg_table
);
633 pmd_populate_kernel(NULL
, pmd
, __va(pg_table
));
635 /* now change pg_table to kernel virtual addresses */
637 pg_table
= (pte_t
*) __va(pg_table
) + start_pte
;
638 for (tmp2
= start_pte
; tmp2
< PTRS_PER_PTE
; tmp2
++,pg_table
++) {
642 * Map the fault vector writable so we can
643 * write the HPMC checksum.
645 #if defined(CONFIG_PARISC_PAGE_SIZE_4KB)
646 if (address
>= ro_start
&& address
< ro_end
647 && address
!= fv_addr
648 && address
!= gw_addr
)
649 pte
= __mk_pte(address
, PAGE_KERNEL_RO
);
652 pte
= __mk_pte(address
, pgprot
);
654 if (address
>= end_paddr
)
657 set_pte(pg_table
, pte
);
659 address
+= PAGE_SIZE
;
663 if (address
>= end_paddr
)
671 * pagetable_init() sets up the page tables
673 * Note that gateway_init() places the Linux gateway page at page 0.
674 * Since gateway pages cannot be dereferenced this has the desirable
675 * side effect of trapping those pesky NULL-reference errors in the
678 static void __init
pagetable_init(void)
682 /* Map each physical memory range to its kernel vaddr */
684 for (range
= 0; range
< npmem_ranges
; range
++) {
685 unsigned long start_paddr
;
686 unsigned long end_paddr
;
689 start_paddr
= pmem_ranges
[range
].start_pfn
<< PAGE_SHIFT
;
690 end_paddr
= start_paddr
+ (pmem_ranges
[range
].pages
<< PAGE_SHIFT
);
691 size
= pmem_ranges
[range
].pages
<< PAGE_SHIFT
;
693 map_pages((unsigned long)__va(start_paddr
), start_paddr
,
697 #ifdef CONFIG_BLK_DEV_INITRD
698 if (initrd_end
&& initrd_end
> mem_limit
) {
699 printk(KERN_INFO
"initrd: mapping %08lx-%08lx\n", initrd_start
, initrd_end
);
700 map_pages(initrd_start
, __pa(initrd_start
),
701 initrd_end
- initrd_start
, PAGE_KERNEL
);
705 empty_zero_page
= alloc_bootmem_pages(PAGE_SIZE
);
706 memset(empty_zero_page
, 0, PAGE_SIZE
);
709 static void __init
gateway_init(void)
711 unsigned long linux_gateway_page_addr
;
712 /* FIXME: This is 'const' in order to trick the compiler
713 into not treating it as DP-relative data. */
714 extern void * const linux_gateway_page
;
716 linux_gateway_page_addr
= LINUX_GATEWAY_ADDR
& PAGE_MASK
;
719 * Setup Linux Gateway page.
721 * The Linux gateway page will reside in kernel space (on virtual
722 * page 0), so it doesn't need to be aliased into user space.
725 map_pages(linux_gateway_page_addr
, __pa(&linux_gateway_page
),
726 PAGE_SIZE
, PAGE_GATEWAY
);
731 map_hpux_gateway_page(struct task_struct
*tsk
, struct mm_struct
*mm
)
736 unsigned long start_pmd
;
737 unsigned long start_pte
;
738 unsigned long address
;
739 unsigned long hpux_gw_page_addr
;
740 /* FIXME: This is 'const' in order to trick the compiler
741 into not treating it as DP-relative data. */
742 extern void * const hpux_gateway_page
;
744 hpux_gw_page_addr
= HPUX_GATEWAY_ADDR
& PAGE_MASK
;
747 * Setup HP-UX Gateway page.
749 * The HP-UX gateway page resides in the user address space,
750 * so it needs to be aliased into each process.
753 pg_dir
= pgd_offset(mm
,hpux_gw_page_addr
);
755 #if PTRS_PER_PMD == 1
758 start_pmd
= ((hpux_gw_page_addr
>> PMD_SHIFT
) & (PTRS_PER_PMD
- 1));
760 start_pte
= ((hpux_gw_page_addr
>> PAGE_SHIFT
) & (PTRS_PER_PTE
- 1));
762 address
= __pa(&hpux_gateway_page
);
763 #if PTRS_PER_PMD == 1
764 pmd
= (pmd_t
*)__pa(pg_dir
);
766 pmd
= (pmd_t
*) pgd_address(*pg_dir
);
769 * pmd is physical at this point
773 pmd
= (pmd_t
*) get_zeroed_page(GFP_KERNEL
);
774 pmd
= (pmd_t
*) __pa(pmd
);
777 __pgd_val_set(*pg_dir
, PxD_FLAG_PRESENT
| PxD_FLAG_VALID
| (unsigned long) pmd
);
779 /* now change pmd to kernel virtual addresses */
781 pmd
= (pmd_t
*)__va(pmd
) + start_pmd
;
784 * pg_table is physical at this point
787 pg_table
= (pte_t
*) pmd_address(*pmd
);
789 pg_table
= (pte_t
*) __pa(get_zeroed_page(GFP_KERNEL
));
791 __pmd_val_set(*pmd
, PxD_FLAG_PRESENT
| PxD_FLAG_VALID
| (unsigned long) pg_table
);
793 /* now change pg_table to kernel virtual addresses */
795 pg_table
= (pte_t
*) __va(pg_table
) + start_pte
;
796 set_pte(pg_table
, __mk_pte(address
, PAGE_GATEWAY
));
798 EXPORT_SYMBOL(map_hpux_gateway_page
);
801 void __init
paging_init(void)
808 flush_cache_all_local(); /* start with known state */
809 flush_tlb_all_local(NULL
);
811 for (i
= 0; i
< npmem_ranges
; i
++) {
812 unsigned long zones_size
[MAX_NR_ZONES
] = { 0, 0, 0 };
814 /* We have an IOMMU, so all memory can go into a single
816 zones_size
[ZONE_DMA
] = pmem_ranges
[i
].pages
;
818 #ifdef CONFIG_DISCONTIGMEM
819 /* Need to initialize the pfnnid_map before we can initialize
823 for (j
= (pmem_ranges
[i
].start_pfn
>> PFNNID_SHIFT
);
824 j
<= ((pmem_ranges
[i
].start_pfn
+ pmem_ranges
[i
].pages
) >> PFNNID_SHIFT
);
831 free_area_init_node(i
, NODE_DATA(i
), zones_size
,
832 pmem_ranges
[i
].start_pfn
, NULL
);
839 * Currently, all PA20 chips have 18 bit protection id's, which is the
840 * limiting factor (space ids are 32 bits).
843 #define NR_SPACE_IDS 262144
848 * Currently we have a one-to-one relationship between space id's and
849 * protection id's. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
850 * support 15 bit protection id's, so that is the limiting factor.
851 * PCXT' has 18 bit protection id's, but only 16 bit spaceids, so it's
852 * probably not worth the effort for a special case here.
855 #define NR_SPACE_IDS 32768
857 #endif /* !CONFIG_PA20 */
859 #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
860 #define SID_ARRAY_SIZE (NR_SPACE_IDS / (8 * sizeof(long)))
862 static unsigned long space_id
[SID_ARRAY_SIZE
] = { 1 }; /* disallow space 0 */
863 static unsigned long dirty_space_id
[SID_ARRAY_SIZE
];
864 static unsigned long space_id_index
;
865 static unsigned long free_space_ids
= NR_SPACE_IDS
- 1;
866 static unsigned long dirty_space_ids
= 0;
868 static DEFINE_SPINLOCK(sid_lock
);
870 unsigned long alloc_sid(void)
874 spin_lock(&sid_lock
);
876 if (free_space_ids
== 0) {
877 if (dirty_space_ids
!= 0) {
878 spin_unlock(&sid_lock
);
879 flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
880 spin_lock(&sid_lock
);
882 BUG_ON(free_space_ids
== 0);
887 index
= find_next_zero_bit(space_id
, NR_SPACE_IDS
, space_id_index
);
888 space_id
[index
>> SHIFT_PER_LONG
] |= (1L << (index
& (BITS_PER_LONG
- 1)));
889 space_id_index
= index
;
891 spin_unlock(&sid_lock
);
893 return index
<< SPACEID_SHIFT
;
896 void free_sid(unsigned long spaceid
)
898 unsigned long index
= spaceid
>> SPACEID_SHIFT
;
899 unsigned long *dirty_space_offset
;
901 dirty_space_offset
= dirty_space_id
+ (index
>> SHIFT_PER_LONG
);
902 index
&= (BITS_PER_LONG
- 1);
904 spin_lock(&sid_lock
);
906 BUG_ON(*dirty_space_offset
& (1L << index
)); /* attempt to free space id twice */
908 *dirty_space_offset
|= (1L << index
);
911 spin_unlock(&sid_lock
);
916 static void get_dirty_sids(unsigned long *ndirtyptr
,unsigned long *dirty_array
)
920 /* NOTE: sid_lock must be held upon entry */
922 *ndirtyptr
= dirty_space_ids
;
923 if (dirty_space_ids
!= 0) {
924 for (i
= 0; i
< SID_ARRAY_SIZE
; i
++) {
925 dirty_array
[i
] = dirty_space_id
[i
];
926 dirty_space_id
[i
] = 0;
934 static void recycle_sids(unsigned long ndirty
,unsigned long *dirty_array
)
938 /* NOTE: sid_lock must be held upon entry */
941 for (i
= 0; i
< SID_ARRAY_SIZE
; i
++) {
942 space_id
[i
] ^= dirty_array
[i
];
945 free_space_ids
+= ndirty
;
950 #else /* CONFIG_SMP */
952 static void recycle_sids(void)
956 /* NOTE: sid_lock must be held upon entry */
958 if (dirty_space_ids
!= 0) {
959 for (i
= 0; i
< SID_ARRAY_SIZE
; i
++) {
960 space_id
[i
] ^= dirty_space_id
[i
];
961 dirty_space_id
[i
] = 0;
964 free_space_ids
+= dirty_space_ids
;
972 * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
973 * purged, we can safely reuse the space ids that were released but
974 * not flushed from the tlb.
979 static unsigned long recycle_ndirty
;
980 static unsigned long recycle_dirty_array
[SID_ARRAY_SIZE
];
981 static unsigned int recycle_inuse
;
983 void flush_tlb_all(void)
988 spin_lock(&sid_lock
);
989 if (dirty_space_ids
> RECYCLE_THRESHOLD
) {
990 BUG_ON(recycle_inuse
); /* FIXME: Use a semaphore/wait queue here */
991 get_dirty_sids(&recycle_ndirty
,recycle_dirty_array
);
995 spin_unlock(&sid_lock
);
996 on_each_cpu(flush_tlb_all_local
, NULL
, 1, 1);
998 spin_lock(&sid_lock
);
999 recycle_sids(recycle_ndirty
,recycle_dirty_array
);
1001 spin_unlock(&sid_lock
);
1005 void flush_tlb_all(void)
1007 spin_lock(&sid_lock
);
1008 flush_tlb_all_local(NULL
);
1010 spin_unlock(&sid_lock
);
1014 #ifdef CONFIG_BLK_DEV_INITRD
1015 void free_initrd_mem(unsigned long start
, unsigned long end
)
1019 printk(KERN_INFO
"Freeing initrd memory: %ldk freed\n", (end
- start
) >> 10);
1020 for (; start
< end
; start
+= PAGE_SIZE
) {
1021 ClearPageReserved(virt_to_page(start
));
1022 init_page_count(virt_to_page(start
));