lib, arch: add filter argument to show_mem and fix private implementations
[linux-2.6.git] / arch / parisc / mm / init.c
1 /*
2  *  linux/arch/parisc/mm/init.c
3  *
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)
10  *
11  */
12
13
14 #include <linux/module.h>
15 #include <linux/mm.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 */
26
27 #include <asm/pgalloc.h>
28 #include <asm/pgtable.h>
29 #include <asm/tlb.h>
30 #include <asm/pdc_chassis.h>
31 #include <asm/mmzone.h>
32 #include <asm/sections.h>
33
34 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
35
36 extern int  data_start;
37
38 #ifdef CONFIG_DISCONTIGMEM
39 struct node_map_data node_data[MAX_NUMNODES] __read_mostly;
40 unsigned char pfnnid_map[PFNNID_MAP_MAX] __read_mostly;
41 #endif
42
43 static struct resource data_resource = {
44         .name   = "Kernel data",
45         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM,
46 };
47
48 static struct resource code_resource = {
49         .name   = "Kernel code",
50         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM,
51 };
52
53 static struct resource pdcdata_resource = {
54         .name   = "PDC data (Page Zero)",
55         .start  = 0,
56         .end    = 0x9ff,
57         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM,
58 };
59
60 static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __read_mostly;
61
62 /* The following array is initialized from the firmware specific
63  * information retrieved in kernel/inventory.c.
64  */
65
66 physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __read_mostly;
67 int npmem_ranges __read_mostly;
68
69 #ifdef CONFIG_64BIT
70 #define MAX_MEM         (~0UL)
71 #else /* !CONFIG_64BIT */
72 #define MAX_MEM         (3584U*1024U*1024U)
73 #endif /* !CONFIG_64BIT */
74
75 static unsigned long mem_limit __read_mostly = MAX_MEM;
76
77 static void __init mem_limit_func(void)
78 {
79         char *cp, *end;
80         unsigned long limit;
81
82         /* We need this before __setup() functions are called */
83
84         limit = MAX_MEM;
85         for (cp = boot_command_line; *cp; ) {
86                 if (memcmp(cp, "mem=", 4) == 0) {
87                         cp += 4;
88                         limit = memparse(cp, &end);
89                         if (end != cp)
90                                 break;
91                         cp = end;
92                 } else {
93                         while (*cp != ' ' && *cp)
94                                 ++cp;
95                         while (*cp == ' ')
96                                 ++cp;
97                 }
98         }
99
100         if (limit < mem_limit)
101                 mem_limit = limit;
102 }
103
104 #define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
105
106 static void __init setup_bootmem(void)
107 {
108         unsigned long bootmap_size;
109         unsigned long mem_max;
110         unsigned long bootmap_pages;
111         unsigned long bootmap_start_pfn;
112         unsigned long bootmap_pfn;
113 #ifndef CONFIG_DISCONTIGMEM
114         physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
115         int npmem_holes;
116 #endif
117         int i, sysram_resource_count;
118
119         disable_sr_hashing(); /* Turn off space register hashing */
120
121         /*
122          * Sort the ranges. Since the number of ranges is typically
123          * small, and performance is not an issue here, just do
124          * a simple insertion sort.
125          */
126
127         for (i = 1; i < npmem_ranges; i++) {
128                 int j;
129
130                 for (j = i; j > 0; j--) {
131                         unsigned long tmp;
132
133                         if (pmem_ranges[j-1].start_pfn <
134                             pmem_ranges[j].start_pfn) {
135
136                                 break;
137                         }
138                         tmp = pmem_ranges[j-1].start_pfn;
139                         pmem_ranges[j-1].start_pfn = pmem_ranges[j].start_pfn;
140                         pmem_ranges[j].start_pfn = tmp;
141                         tmp = pmem_ranges[j-1].pages;
142                         pmem_ranges[j-1].pages = pmem_ranges[j].pages;
143                         pmem_ranges[j].pages = tmp;
144                 }
145         }
146
147 #ifndef CONFIG_DISCONTIGMEM
148         /*
149          * Throw out ranges that are too far apart (controlled by
150          * MAX_GAP).
151          */
152
153         for (i = 1; i < npmem_ranges; i++) {
154                 if (pmem_ranges[i].start_pfn -
155                         (pmem_ranges[i-1].start_pfn +
156                          pmem_ranges[i-1].pages) > MAX_GAP) {
157                         npmem_ranges = i;
158                         printk("Large gap in memory detected (%ld pages). "
159                                "Consider turning on CONFIG_DISCONTIGMEM\n",
160                                pmem_ranges[i].start_pfn -
161                                (pmem_ranges[i-1].start_pfn +
162                                 pmem_ranges[i-1].pages));
163                         break;
164                 }
165         }
166 #endif
167
168         if (npmem_ranges > 1) {
169
170                 /* Print the memory ranges */
171
172                 printk(KERN_INFO "Memory Ranges:\n");
173
174                 for (i = 0; i < npmem_ranges; i++) {
175                         unsigned long start;
176                         unsigned long size;
177
178                         size = (pmem_ranges[i].pages << PAGE_SHIFT);
179                         start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
180                         printk(KERN_INFO "%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
181                                 i,start, start + (size - 1), size >> 20);
182                 }
183         }
184
185         sysram_resource_count = npmem_ranges;
186         for (i = 0; i < sysram_resource_count; i++) {
187                 struct resource *res = &sysram_resources[i];
188                 res->name = "System RAM";
189                 res->start = pmem_ranges[i].start_pfn << PAGE_SHIFT;
190                 res->end = res->start + (pmem_ranges[i].pages << PAGE_SHIFT)-1;
191                 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
192                 request_resource(&iomem_resource, res);
193         }
194
195         /*
196          * For 32 bit kernels we limit the amount of memory we can
197          * support, in order to preserve enough kernel address space
198          * for other purposes. For 64 bit kernels we don't normally
199          * limit the memory, but this mechanism can be used to
200          * artificially limit the amount of memory (and it is written
201          * to work with multiple memory ranges).
202          */
203
204         mem_limit_func();       /* check for "mem=" argument */
205
206         mem_max = 0;
207         num_physpages = 0;
208         for (i = 0; i < npmem_ranges; i++) {
209                 unsigned long rsize;
210
211                 rsize = pmem_ranges[i].pages << PAGE_SHIFT;
212                 if ((mem_max + rsize) > mem_limit) {
213                         printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
214                         if (mem_max == mem_limit)
215                                 npmem_ranges = i;
216                         else {
217                                 pmem_ranges[i].pages =   (mem_limit >> PAGE_SHIFT)
218                                                        - (mem_max >> PAGE_SHIFT);
219                                 npmem_ranges = i + 1;
220                                 mem_max = mem_limit;
221                         }
222                 num_physpages += pmem_ranges[i].pages;
223                         break;
224                 }
225             num_physpages += pmem_ranges[i].pages;
226                 mem_max += rsize;
227         }
228
229         printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
230
231 #ifndef CONFIG_DISCONTIGMEM
232         /* Merge the ranges, keeping track of the holes */
233
234         {
235                 unsigned long end_pfn;
236                 unsigned long hole_pages;
237
238                 npmem_holes = 0;
239                 end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
240                 for (i = 1; i < npmem_ranges; i++) {
241
242                         hole_pages = pmem_ranges[i].start_pfn - end_pfn;
243                         if (hole_pages) {
244                                 pmem_holes[npmem_holes].start_pfn = end_pfn;
245                                 pmem_holes[npmem_holes++].pages = hole_pages;
246                                 end_pfn += hole_pages;
247                         }
248                         end_pfn += pmem_ranges[i].pages;
249                 }
250
251                 pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
252                 npmem_ranges = 1;
253         }
254 #endif
255
256         bootmap_pages = 0;
257         for (i = 0; i < npmem_ranges; i++)
258                 bootmap_pages += bootmem_bootmap_pages(pmem_ranges[i].pages);
259
260         bootmap_start_pfn = PAGE_ALIGN(__pa((unsigned long) &_end)) >> PAGE_SHIFT;
261
262 #ifdef CONFIG_DISCONTIGMEM
263         for (i = 0; i < MAX_PHYSMEM_RANGES; i++) {
264                 memset(NODE_DATA(i), 0, sizeof(pg_data_t));
265                 NODE_DATA(i)->bdata = &bootmem_node_data[i];
266         }
267         memset(pfnnid_map, 0xff, sizeof(pfnnid_map));
268
269         for (i = 0; i < npmem_ranges; i++)
270                 node_set_online(i);
271 #endif
272
273         /*
274          * Initialize and free the full range of memory in each range.
275          * Note that the only writing these routines do are to the bootmap,
276          * and we've made sure to locate the bootmap properly so that they
277          * won't be writing over anything important.
278          */
279
280         bootmap_pfn = bootmap_start_pfn;
281         max_pfn = 0;
282         for (i = 0; i < npmem_ranges; i++) {
283                 unsigned long start_pfn;
284                 unsigned long npages;
285
286                 start_pfn = pmem_ranges[i].start_pfn;
287                 npages = pmem_ranges[i].pages;
288
289                 bootmap_size = init_bootmem_node(NODE_DATA(i),
290                                                 bootmap_pfn,
291                                                 start_pfn,
292                                                 (start_pfn + npages) );
293                 free_bootmem_node(NODE_DATA(i),
294                                   (start_pfn << PAGE_SHIFT),
295                                   (npages << PAGE_SHIFT) );
296                 bootmap_pfn += (bootmap_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
297                 if ((start_pfn + npages) > max_pfn)
298                         max_pfn = start_pfn + npages;
299         }
300
301         /* IOMMU is always used to access "high mem" on those boxes
302          * that can support enough mem that a PCI device couldn't
303          * directly DMA to any physical addresses.
304          * ISA DMA support will need to revisit this.
305          */
306         max_low_pfn = max_pfn;
307
308         /* bootmap sizing messed up? */
309         BUG_ON((bootmap_pfn - bootmap_start_pfn) != bootmap_pages);
310
311         /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
312
313 #define PDC_CONSOLE_IO_IODC_SIZE 32768
314
315         reserve_bootmem_node(NODE_DATA(0), 0UL,
316                         (unsigned long)(PAGE0->mem_free +
317                                 PDC_CONSOLE_IO_IODC_SIZE), BOOTMEM_DEFAULT);
318         reserve_bootmem_node(NODE_DATA(0), __pa((unsigned long)_text),
319                         (unsigned long)(_end - _text), BOOTMEM_DEFAULT);
320         reserve_bootmem_node(NODE_DATA(0), (bootmap_start_pfn << PAGE_SHIFT),
321                         ((bootmap_pfn - bootmap_start_pfn) << PAGE_SHIFT),
322                         BOOTMEM_DEFAULT);
323
324 #ifndef CONFIG_DISCONTIGMEM
325
326         /* reserve the holes */
327
328         for (i = 0; i < npmem_holes; i++) {
329                 reserve_bootmem_node(NODE_DATA(0),
330                                 (pmem_holes[i].start_pfn << PAGE_SHIFT),
331                                 (pmem_holes[i].pages << PAGE_SHIFT),
332                                 BOOTMEM_DEFAULT);
333         }
334 #endif
335
336 #ifdef CONFIG_BLK_DEV_INITRD
337         if (initrd_start) {
338                 printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
339                 if (__pa(initrd_start) < mem_max) {
340                         unsigned long initrd_reserve;
341
342                         if (__pa(initrd_end) > mem_max) {
343                                 initrd_reserve = mem_max - __pa(initrd_start);
344                         } else {
345                                 initrd_reserve = initrd_end - initrd_start;
346                         }
347                         initrd_below_start_ok = 1;
348                         printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
349
350                         reserve_bootmem_node(NODE_DATA(0), __pa(initrd_start),
351                                         initrd_reserve, BOOTMEM_DEFAULT);
352                 }
353         }
354 #endif
355
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;
360
361         /* We don't know which region the kernel will be in, so try
362          * all of them.
363          */
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);
368         }
369         request_resource(&sysram_resources[0], &pdcdata_resource);
370 }
371
372 void free_initmem(void)
373 {
374         unsigned long addr;
375         unsigned long init_begin = (unsigned long)__init_begin;
376         unsigned long init_end = (unsigned long)__init_end;
377
378 #ifdef CONFIG_DEBUG_KERNEL
379         /* Attempt to catch anyone trying to execute code here
380          * by filling the page with BRK insns.
381          */
382         memset((void *)init_begin, 0x00, init_end - init_begin);
383         flush_icache_range(init_begin, init_end);
384 #endif
385         
386         /* align __init_begin and __init_end to page size,
387            ignoring linker script where we might have tried to save RAM */
388         init_begin = PAGE_ALIGN(init_begin);
389         init_end = PAGE_ALIGN(init_end);
390         for (addr = init_begin; addr < init_end; addr += PAGE_SIZE) {
391                 ClearPageReserved(virt_to_page(addr));
392                 init_page_count(virt_to_page(addr));
393                 free_page(addr);
394                 num_physpages++;
395                 totalram_pages++;
396         }
397
398         /* set up a new led state on systems shipped LED State panel */
399         pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
400         
401         printk(KERN_INFO "Freeing unused kernel memory: %luk freed\n",
402                 (init_end - init_begin) >> 10);
403 }
404
405
406 #ifdef CONFIG_DEBUG_RODATA
407 void mark_rodata_ro(void)
408 {
409         /* rodata memory was already mapped with KERNEL_RO access rights by
410            pagetable_init() and map_pages(). No need to do additional stuff here */
411         printk (KERN_INFO "Write protecting the kernel read-only data: %luk\n",
412                 (unsigned long)(__end_rodata - __start_rodata) >> 10);
413 }
414 #endif
415
416
417 /*
418  * Just an arbitrary offset to serve as a "hole" between mapping areas
419  * (between top of physical memory and a potential pcxl dma mapping
420  * area, and below the vmalloc mapping area).
421  *
422  * The current 32K value just means that there will be a 32K "hole"
423  * between mapping areas. That means that  any out-of-bounds memory
424  * accesses will hopefully be caught. The vmalloc() routines leaves
425  * a hole of 4kB between each vmalloced area for the same reason.
426  */
427
428  /* Leave room for gateway page expansion */
429 #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
430 #error KERNEL_MAP_START is in gateway reserved region
431 #endif
432 #define MAP_START (KERNEL_MAP_START)
433
434 #define VM_MAP_OFFSET  (32*1024)
435 #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
436                                      & ~(VM_MAP_OFFSET-1)))
437
438 void *parisc_vmalloc_start __read_mostly;
439 EXPORT_SYMBOL(parisc_vmalloc_start);
440
441 #ifdef CONFIG_PA11
442 unsigned long pcxl_dma_start __read_mostly;
443 #endif
444
445 void __init mem_init(void)
446 {
447         int codesize, reservedpages, datasize, initsize;
448
449         /* Do sanity checks on page table constants */
450         BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
451         BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
452         BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
453         BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
454                         > BITS_PER_LONG);
455
456         high_memory = __va((max_pfn << PAGE_SHIFT));
457
458 #ifndef CONFIG_DISCONTIGMEM
459         max_mapnr = page_to_pfn(virt_to_page(high_memory - 1)) + 1;
460         totalram_pages += free_all_bootmem();
461 #else
462         {
463                 int i;
464
465                 for (i = 0; i < npmem_ranges; i++)
466                         totalram_pages += free_all_bootmem_node(NODE_DATA(i));
467         }
468 #endif
469
470         codesize = (unsigned long)_etext - (unsigned long)_text;
471         datasize = (unsigned long)_edata - (unsigned long)_etext;
472         initsize = (unsigned long)__init_end - (unsigned long)__init_begin;
473
474         reservedpages = 0;
475 {
476         unsigned long pfn;
477 #ifdef CONFIG_DISCONTIGMEM
478         int i;
479
480         for (i = 0; i < npmem_ranges; i++) {
481                 for (pfn = node_start_pfn(i); pfn < node_end_pfn(i); pfn++) {
482                         if (PageReserved(pfn_to_page(pfn)))
483                                 reservedpages++;
484                 }
485         }
486 #else /* !CONFIG_DISCONTIGMEM */
487         for (pfn = 0; pfn < max_pfn; pfn++) {
488                 /*
489                  * Only count reserved RAM pages
490                  */
491                 if (PageReserved(pfn_to_page(pfn)))
492                         reservedpages++;
493         }
494 #endif
495 }
496
497 #ifdef CONFIG_PA11
498         if (hppa_dma_ops == &pcxl_dma_ops) {
499                 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
500                 parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
501                                                 + PCXL_DMA_MAP_SIZE);
502         } else {
503                 pcxl_dma_start = 0;
504                 parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
505         }
506 #else
507         parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
508 #endif
509
510         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init)\n",
511                 nr_free_pages() << (PAGE_SHIFT-10),
512                 num_physpages << (PAGE_SHIFT-10),
513                 codesize >> 10,
514                 reservedpages << (PAGE_SHIFT-10),
515                 datasize >> 10,
516                 initsize >> 10
517         );
518
519 #ifdef CONFIG_DEBUG_KERNEL /* double-sanity-check paranoia */
520         printk("virtual kernel memory layout:\n"
521                "    vmalloc : 0x%p - 0x%p   (%4ld MB)\n"
522                "    memory  : 0x%p - 0x%p   (%4ld MB)\n"
523                "      .init : 0x%p - 0x%p   (%4ld kB)\n"
524                "      .data : 0x%p - 0x%p   (%4ld kB)\n"
525                "      .text : 0x%p - 0x%p   (%4ld kB)\n",
526
527                (void*)VMALLOC_START, (void*)VMALLOC_END,
528                (VMALLOC_END - VMALLOC_START) >> 20,
529
530                __va(0), high_memory,
531                ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
532
533                __init_begin, __init_end,
534                ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
535
536                _etext, _edata,
537                ((unsigned long)_edata - (unsigned long)_etext) >> 10,
538
539                _text, _etext,
540                ((unsigned long)_etext - (unsigned long)_text) >> 10);
541 #endif
542 }
543
544 unsigned long *empty_zero_page __read_mostly;
545 EXPORT_SYMBOL(empty_zero_page);
546
547 void show_mem(unsigned int filter)
548 {
549         int i,free = 0,total = 0,reserved = 0;
550         int shared = 0, cached = 0;
551
552         printk(KERN_INFO "Mem-info:\n");
553         show_free_areas();
554 #ifndef CONFIG_DISCONTIGMEM
555         i = max_mapnr;
556         while (i-- > 0) {
557                 total++;
558                 if (PageReserved(mem_map+i))
559                         reserved++;
560                 else if (PageSwapCache(mem_map+i))
561                         cached++;
562                 else if (!page_count(&mem_map[i]))
563                         free++;
564                 else
565                         shared += page_count(&mem_map[i]) - 1;
566         }
567 #else
568         for (i = 0; i < npmem_ranges; i++) {
569                 int j;
570
571                 for (j = node_start_pfn(i); j < node_end_pfn(i); j++) {
572                         struct page *p;
573                         unsigned long flags;
574
575                         pgdat_resize_lock(NODE_DATA(i), &flags);
576                         p = nid_page_nr(i, j) - node_start_pfn(i);
577
578                         total++;
579                         if (PageReserved(p))
580                                 reserved++;
581                         else if (PageSwapCache(p))
582                                 cached++;
583                         else if (!page_count(p))
584                                 free++;
585                         else
586                                 shared += page_count(p) - 1;
587                         pgdat_resize_unlock(NODE_DATA(i), &flags);
588                 }
589         }
590 #endif
591         printk(KERN_INFO "%d pages of RAM\n", total);
592         printk(KERN_INFO "%d reserved pages\n", reserved);
593         printk(KERN_INFO "%d pages shared\n", shared);
594         printk(KERN_INFO "%d pages swap cached\n", cached);
595
596
597 #ifdef CONFIG_DISCONTIGMEM
598         {
599                 struct zonelist *zl;
600                 int i, j;
601
602                 for (i = 0; i < npmem_ranges; i++) {
603                         zl = node_zonelist(i, 0);
604                         for (j = 0; j < MAX_NR_ZONES; j++) {
605                                 struct zoneref *z;
606                                 struct zone *zone;
607
608                                 printk("Zone list for zone %d on node %d: ", j, i);
609                                 for_each_zone_zonelist(zone, z, zl, j)
610                                         printk("[%d/%s] ", zone_to_nid(zone),
611                                                                 zone->name);
612                                 printk("\n");
613                         }
614                 }
615         }
616 #endif
617 }
618
619
620 static void __init map_pages(unsigned long start_vaddr, unsigned long start_paddr, unsigned long size, pgprot_t pgprot)
621 {
622         pgd_t *pg_dir;
623         pmd_t *pmd;
624         pte_t *pg_table;
625         unsigned long end_paddr;
626         unsigned long start_pmd;
627         unsigned long start_pte;
628         unsigned long tmp1;
629         unsigned long tmp2;
630         unsigned long address;
631         unsigned long ro_start;
632         unsigned long ro_end;
633         unsigned long fv_addr;
634         unsigned long gw_addr;
635         extern const unsigned long fault_vector_20;
636         extern void * const linux_gateway_page;
637
638         ro_start = __pa((unsigned long)_text);
639         ro_end   = __pa((unsigned long)&data_start);
640         fv_addr  = __pa((unsigned long)&fault_vector_20) & PAGE_MASK;
641         gw_addr  = __pa((unsigned long)&linux_gateway_page) & PAGE_MASK;
642
643         end_paddr = start_paddr + size;
644
645         pg_dir = pgd_offset_k(start_vaddr);
646
647 #if PTRS_PER_PMD == 1
648         start_pmd = 0;
649 #else
650         start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
651 #endif
652         start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
653
654         address = start_paddr;
655         while (address < end_paddr) {
656 #if PTRS_PER_PMD == 1
657                 pmd = (pmd_t *)__pa(pg_dir);
658 #else
659                 pmd = (pmd_t *)pgd_address(*pg_dir);
660
661                 /*
662                  * pmd is physical at this point
663                  */
664
665                 if (!pmd) {
666                         pmd = (pmd_t *) alloc_bootmem_low_pages_node(NODE_DATA(0),PAGE_SIZE << PMD_ORDER);
667                         pmd = (pmd_t *) __pa(pmd);
668                 }
669
670                 pgd_populate(NULL, pg_dir, __va(pmd));
671 #endif
672                 pg_dir++;
673
674                 /* now change pmd to kernel virtual addresses */
675
676                 pmd = (pmd_t *)__va(pmd) + start_pmd;
677                 for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++,pmd++) {
678
679                         /*
680                          * pg_table is physical at this point
681                          */
682
683                         pg_table = (pte_t *)pmd_address(*pmd);
684                         if (!pg_table) {
685                                 pg_table = (pte_t *)
686                                         alloc_bootmem_low_pages_node(NODE_DATA(0),PAGE_SIZE);
687                                 pg_table = (pte_t *) __pa(pg_table);
688                         }
689
690                         pmd_populate_kernel(NULL, pmd, __va(pg_table));
691
692                         /* now change pg_table to kernel virtual addresses */
693
694                         pg_table = (pte_t *) __va(pg_table) + start_pte;
695                         for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++,pg_table++) {
696                                 pte_t pte;
697
698                                 /*
699                                  * Map the fault vector writable so we can
700                                  * write the HPMC checksum.
701                                  */
702 #if defined(CONFIG_PARISC_PAGE_SIZE_4KB)
703                                 if (address >= ro_start && address < ro_end
704                                                         && address != fv_addr
705                                                         && address != gw_addr)
706                                     pte = __mk_pte(address, PAGE_KERNEL_RO);
707                                 else
708 #endif
709                                     pte = __mk_pte(address, pgprot);
710
711                                 if (address >= end_paddr)
712                                         pte_val(pte) = 0;
713
714                                 set_pte(pg_table, pte);
715
716                                 address += PAGE_SIZE;
717                         }
718                         start_pte = 0;
719
720                         if (address >= end_paddr)
721                             break;
722                 }
723                 start_pmd = 0;
724         }
725 }
726
727 /*
728  * pagetable_init() sets up the page tables
729  *
730  * Note that gateway_init() places the Linux gateway page at page 0.
731  * Since gateway pages cannot be dereferenced this has the desirable
732  * side effect of trapping those pesky NULL-reference errors in the
733  * kernel.
734  */
735 static void __init pagetable_init(void)
736 {
737         int range;
738
739         /* Map each physical memory range to its kernel vaddr */
740
741         for (range = 0; range < npmem_ranges; range++) {
742                 unsigned long start_paddr;
743                 unsigned long end_paddr;
744                 unsigned long size;
745
746                 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
747                 end_paddr = start_paddr + (pmem_ranges[range].pages << PAGE_SHIFT);
748                 size = pmem_ranges[range].pages << PAGE_SHIFT;
749
750                 map_pages((unsigned long)__va(start_paddr), start_paddr,
751                         size, PAGE_KERNEL);
752         }
753
754 #ifdef CONFIG_BLK_DEV_INITRD
755         if (initrd_end && initrd_end > mem_limit) {
756                 printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
757                 map_pages(initrd_start, __pa(initrd_start),
758                         initrd_end - initrd_start, PAGE_KERNEL);
759         }
760 #endif
761
762         empty_zero_page = alloc_bootmem_pages(PAGE_SIZE);
763         memset(empty_zero_page, 0, PAGE_SIZE);
764 }
765
766 static void __init gateway_init(void)
767 {
768         unsigned long linux_gateway_page_addr;
769         /* FIXME: This is 'const' in order to trick the compiler
770            into not treating it as DP-relative data. */
771         extern void * const linux_gateway_page;
772
773         linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
774
775         /*
776          * Setup Linux Gateway page.
777          *
778          * The Linux gateway page will reside in kernel space (on virtual
779          * page 0), so it doesn't need to be aliased into user space.
780          */
781
782         map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
783                 PAGE_SIZE, PAGE_GATEWAY);
784 }
785
786 #ifdef CONFIG_HPUX
787 void
788 map_hpux_gateway_page(struct task_struct *tsk, struct mm_struct *mm)
789 {
790         pgd_t *pg_dir;
791         pmd_t *pmd;
792         pte_t *pg_table;
793         unsigned long start_pmd;
794         unsigned long start_pte;
795         unsigned long address;
796         unsigned long hpux_gw_page_addr;
797         /* FIXME: This is 'const' in order to trick the compiler
798            into not treating it as DP-relative data. */
799         extern void * const hpux_gateway_page;
800
801         hpux_gw_page_addr = HPUX_GATEWAY_ADDR & PAGE_MASK;
802
803         /*
804          * Setup HP-UX Gateway page.
805          *
806          * The HP-UX gateway page resides in the user address space,
807          * so it needs to be aliased into each process.
808          */
809
810         pg_dir = pgd_offset(mm,hpux_gw_page_addr);
811
812 #if PTRS_PER_PMD == 1
813         start_pmd = 0;
814 #else
815         start_pmd = ((hpux_gw_page_addr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
816 #endif
817         start_pte = ((hpux_gw_page_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
818
819         address = __pa(&hpux_gateway_page);
820 #if PTRS_PER_PMD == 1
821         pmd = (pmd_t *)__pa(pg_dir);
822 #else
823         pmd = (pmd_t *) pgd_address(*pg_dir);
824
825         /*
826          * pmd is physical at this point
827          */
828
829         if (!pmd) {
830                 pmd = (pmd_t *) get_zeroed_page(GFP_KERNEL);
831                 pmd = (pmd_t *) __pa(pmd);
832         }
833
834         __pgd_val_set(*pg_dir, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pmd);
835 #endif
836         /* now change pmd to kernel virtual addresses */
837
838         pmd = (pmd_t *)__va(pmd) + start_pmd;
839
840         /*
841          * pg_table is physical at this point
842          */
843
844         pg_table = (pte_t *) pmd_address(*pmd);
845         if (!pg_table)
846                 pg_table = (pte_t *) __pa(get_zeroed_page(GFP_KERNEL));
847
848         __pmd_val_set(*pmd, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pg_table);
849
850         /* now change pg_table to kernel virtual addresses */
851
852         pg_table = (pte_t *) __va(pg_table) + start_pte;
853         set_pte(pg_table, __mk_pte(address, PAGE_GATEWAY));
854 }
855 EXPORT_SYMBOL(map_hpux_gateway_page);
856 #endif
857
858 void __init paging_init(void)
859 {
860         int i;
861
862         setup_bootmem();
863         pagetable_init();
864         gateway_init();
865         flush_cache_all_local(); /* start with known state */
866         flush_tlb_all_local(NULL);
867
868         for (i = 0; i < npmem_ranges; i++) {
869                 unsigned long zones_size[MAX_NR_ZONES] = { 0, };
870
871                 zones_size[ZONE_NORMAL] = pmem_ranges[i].pages;
872
873 #ifdef CONFIG_DISCONTIGMEM
874                 /* Need to initialize the pfnnid_map before we can initialize
875                    the zone */
876                 {
877                     int j;
878                     for (j = (pmem_ranges[i].start_pfn >> PFNNID_SHIFT);
879                          j <= ((pmem_ranges[i].start_pfn + pmem_ranges[i].pages) >> PFNNID_SHIFT);
880                          j++) {
881                         pfnnid_map[j] = i;
882                     }
883                 }
884 #endif
885
886                 free_area_init_node(i, zones_size,
887                                 pmem_ranges[i].start_pfn, NULL);
888         }
889 }
890
891 #ifdef CONFIG_PA20
892
893 /*
894  * Currently, all PA20 chips have 18 bit protection IDs, which is the
895  * limiting factor (space ids are 32 bits).
896  */
897
898 #define NR_SPACE_IDS 262144
899
900 #else
901
902 /*
903  * Currently we have a one-to-one relationship between space IDs and
904  * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
905  * support 15 bit protection IDs, so that is the limiting factor.
906  * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
907  * probably not worth the effort for a special case here.
908  */
909
910 #define NR_SPACE_IDS 32768
911
912 #endif  /* !CONFIG_PA20 */
913
914 #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
915 #define SID_ARRAY_SIZE  (NR_SPACE_IDS / (8 * sizeof(long)))
916
917 static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
918 static unsigned long dirty_space_id[SID_ARRAY_SIZE];
919 static unsigned long space_id_index;
920 static unsigned long free_space_ids = NR_SPACE_IDS - 1;
921 static unsigned long dirty_space_ids = 0;
922
923 static DEFINE_SPINLOCK(sid_lock);
924
925 unsigned long alloc_sid(void)
926 {
927         unsigned long index;
928
929         spin_lock(&sid_lock);
930
931         if (free_space_ids == 0) {
932                 if (dirty_space_ids != 0) {
933                         spin_unlock(&sid_lock);
934                         flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
935                         spin_lock(&sid_lock);
936                 }
937                 BUG_ON(free_space_ids == 0);
938         }
939
940         free_space_ids--;
941
942         index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
943         space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1)));
944         space_id_index = index;
945
946         spin_unlock(&sid_lock);
947
948         return index << SPACEID_SHIFT;
949 }
950
951 void free_sid(unsigned long spaceid)
952 {
953         unsigned long index = spaceid >> SPACEID_SHIFT;
954         unsigned long *dirty_space_offset;
955
956         dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG);
957         index &= (BITS_PER_LONG - 1);
958
959         spin_lock(&sid_lock);
960
961         BUG_ON(*dirty_space_offset & (1L << index)); /* attempt to free space id twice */
962
963         *dirty_space_offset |= (1L << index);
964         dirty_space_ids++;
965
966         spin_unlock(&sid_lock);
967 }
968
969
970 #ifdef CONFIG_SMP
971 static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
972 {
973         int i;
974
975         /* NOTE: sid_lock must be held upon entry */
976
977         *ndirtyptr = dirty_space_ids;
978         if (dirty_space_ids != 0) {
979             for (i = 0; i < SID_ARRAY_SIZE; i++) {
980                 dirty_array[i] = dirty_space_id[i];
981                 dirty_space_id[i] = 0;
982             }
983             dirty_space_ids = 0;
984         }
985
986         return;
987 }
988
989 static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
990 {
991         int i;
992
993         /* NOTE: sid_lock must be held upon entry */
994
995         if (ndirty != 0) {
996                 for (i = 0; i < SID_ARRAY_SIZE; i++) {
997                         space_id[i] ^= dirty_array[i];
998                 }
999
1000                 free_space_ids += ndirty;
1001                 space_id_index = 0;
1002         }
1003 }
1004
1005 #else /* CONFIG_SMP */
1006
1007 static void recycle_sids(void)
1008 {
1009         int i;
1010
1011         /* NOTE: sid_lock must be held upon entry */
1012
1013         if (dirty_space_ids != 0) {
1014                 for (i = 0; i < SID_ARRAY_SIZE; i++) {
1015                         space_id[i] ^= dirty_space_id[i];
1016                         dirty_space_id[i] = 0;
1017                 }
1018
1019                 free_space_ids += dirty_space_ids;
1020                 dirty_space_ids = 0;
1021                 space_id_index = 0;
1022         }
1023 }
1024 #endif
1025
1026 /*
1027  * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
1028  * purged, we can safely reuse the space ids that were released but
1029  * not flushed from the tlb.
1030  */
1031
1032 #ifdef CONFIG_SMP
1033
1034 static unsigned long recycle_ndirty;
1035 static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
1036 static unsigned int recycle_inuse;
1037
1038 void flush_tlb_all(void)
1039 {
1040         int do_recycle;
1041
1042         do_recycle = 0;
1043         spin_lock(&sid_lock);
1044         if (dirty_space_ids > RECYCLE_THRESHOLD) {
1045             BUG_ON(recycle_inuse);  /* FIXME: Use a semaphore/wait queue here */
1046             get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
1047             recycle_inuse++;
1048             do_recycle++;
1049         }
1050         spin_unlock(&sid_lock);
1051         on_each_cpu(flush_tlb_all_local, NULL, 1);
1052         if (do_recycle) {
1053             spin_lock(&sid_lock);
1054             recycle_sids(recycle_ndirty,recycle_dirty_array);
1055             recycle_inuse = 0;
1056             spin_unlock(&sid_lock);
1057         }
1058 }
1059 #else
1060 void flush_tlb_all(void)
1061 {
1062         spin_lock(&sid_lock);
1063         flush_tlb_all_local(NULL);
1064         recycle_sids();
1065         spin_unlock(&sid_lock);
1066 }
1067 #endif
1068
1069 #ifdef CONFIG_BLK_DEV_INITRD
1070 void free_initrd_mem(unsigned long start, unsigned long end)
1071 {
1072         if (start >= end)
1073                 return;
1074         printk(KERN_INFO "Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
1075         for (; start < end; start += PAGE_SIZE) {
1076                 ClearPageReserved(virt_to_page(start));
1077                 init_page_count(virt_to_page(start));
1078                 free_page(start);
1079                 num_physpages++;
1080                 totalram_pages++;
1081         }
1082 }
1083 #endif