projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
parisc: add me to Maintainers
[linux-2.6-block.git] / arch / parisc / mm / init.c
CommitLineData
1da177e4
LT		 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)
a8f44e38 9 * Copyright 2006-2007 Helge Deller (deller@gmx.de)
1da177e4
LT		 10 *
11 */
12
1da177e4
LT		 13
14#include <linux/module.h>
15#include <linux/mm.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 */
25
26#include <asm/pgalloc.h>
ce8420bb 27#include <asm/pgtable.h>
1da177e4
LT		 28#include <asm/tlb.h>
29#include <asm/pdc_chassis.h>
30#include <asm/mmzone.h>
a581c2a4 31#include <asm/sections.h>
1da177e4
LT		 32
33DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
34
1da177e4 35extern int data_start;
1da177e4
LT		 36
37#ifdef CONFIG_DISCONTIGMEM
8039de10 38struct node_map_data node_data[MAX_NUMNODES] __read_mostly;
8039de10 39unsigned char pfnnid_map[PFNNID_MAP_MAX] __read_mostly;
1da177e4
LT		 40#endif
41
42static struct resource data_resource = {
43 .name = "Kernel data",
44 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
45};
46
47static struct resource code_resource = {
48 .name = "Kernel code",
49 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
50};
51
52static struct resource pdcdata_resource = {
53 .name = "PDC data (Page Zero)",
54 .start = 0,
55 .end = 0x9ff,
56 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
57};
58
8039de10 59static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __read_mostly;
1da177e4
LT		 60
61/* The following array is initialized from the firmware specific
62 * information retrieved in kernel/inventory.c.
63 */
64
8039de10
HD		 65physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __read_mostly;
66int npmem_ranges __read_mostly;
1da177e4 67
a8f44e38 68#ifdef CONFIG_64BIT
1da177e4 69#define MAX_MEM (~0UL)
a8f44e38 70#else /* !CONFIG_64BIT */
1da177e4 71#define MAX_MEM (3584U*1024U*1024U)
a8f44e38 72#endif /* !CONFIG_64BIT */
1da177e4 73
8039de10 74static unsigned long mem_limit __read_mostly = MAX_MEM;
1da177e4
LT		 75
76static void __init mem_limit_func(void)
77{
78 char *cp, *end;
79 unsigned long limit;
1da177e4
LT		 80
81 /* We need this before __setup() functions are called */
82
83 limit = MAX_MEM;
668f9931 84 for (cp = boot_command_line; *cp; ) {
1da177e4
LT		 85 if (memcmp(cp, "mem=", 4) == 0) {
86 cp += 4;
87 limit = memparse(cp, &end);
88 if (end != cp)
89 break;
90 cp = end;
91 } else {
92 while (*cp != ' ' && *cp)
93 ++cp;
94 while (*cp == ' ')
95 ++cp;
96 }
97 }
98
99 if (limit < mem_limit)
100 mem_limit = limit;
101}
102
103#define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
104
105static void __init setup_bootmem(void)
106{
107 unsigned long bootmap_size;
108 unsigned long mem_max;
109 unsigned long bootmap_pages;
110 unsigned long bootmap_start_pfn;
111 unsigned long bootmap_pfn;
112#ifndef CONFIG_DISCONTIGMEM
113 physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
114 int npmem_holes;
115#endif
116 int i, sysram_resource_count;
117
118 disable_sr_hashing(); /* Turn off space register hashing */
119
120 /*
121 * Sort the ranges. Since the number of ranges is typically
122 * small, and performance is not an issue here, just do
123 * a simple insertion sort.
124 */
125
126 for (i = 1; i < npmem_ranges; i++) {
127 int j;
128
129 for (j = i; j > 0; j--) {
130 unsigned long tmp;
131
132 if (pmem_ranges[j-1].start_pfn <
133 pmem_ranges[j].start_pfn) {
134
135 break;
136 }
137 tmp = pmem_ranges[j-1].start_pfn;
138 pmem_ranges[j-1].start_pfn = pmem_ranges[j].start_pfn;
139 pmem_ranges[j].start_pfn = tmp;
140 tmp = pmem_ranges[j-1].pages;
141 pmem_ranges[j-1].pages = pmem_ranges[j].pages;
142 pmem_ranges[j].pages = tmp;
143 }
144 }
145
146#ifndef CONFIG_DISCONTIGMEM
147 /*
148 * Throw out ranges that are too far apart (controlled by
149 * MAX_GAP).
150 */
151
152 for (i = 1; i < npmem_ranges; i++) {
153 if (pmem_ranges[i].start_pfn -
154 (pmem_ranges[i-1].start_pfn +
155 pmem_ranges[i-1].pages) > MAX_GAP) {
156 npmem_ranges = i;
157 printk("Large gap in memory detected (%ld pages). "
158 "Consider turning on CONFIG_DISCONTIGMEM\n",
159 pmem_ranges[i].start_pfn -
160 (pmem_ranges[i-1].start_pfn +
161 pmem_ranges[i-1].pages));
162 break;
163 }
164 }
165#endif
166
167 if (npmem_ranges > 1) {
168
169 /* Print the memory ranges */
170
171 printk(KERN_INFO "Memory Ranges:\n");
172
173 for (i = 0; i < npmem_ranges; i++) {
174 unsigned long start;
175 unsigned long size;
176
177 size = (pmem_ranges[i].pages << PAGE_SHIFT);
178 start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
179 printk(KERN_INFO "%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
180 i,start, start + (size - 1), size >> 20);
181 }
182 }
183
184 sysram_resource_count = npmem_ranges;
185 for (i = 0; i < sysram_resource_count; i++) {
186 struct resource *res = &sysram_resources[i];
187 res->name = "System RAM";
188 res->start = pmem_ranges[i].start_pfn << PAGE_SHIFT;
189 res->end = res->start + (pmem_ranges[i].pages << PAGE_SHIFT)-1;
190 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
191 request_resource(&iomem_resource, res);
192 }
193
194 /*
195 * For 32 bit kernels we limit the amount of memory we can
196 * support, in order to preserve enough kernel address space
197 * for other purposes. For 64 bit kernels we don't normally
198 * limit the memory, but this mechanism can be used to
199 * artificially limit the amount of memory (and it is written
200 * to work with multiple memory ranges).
201 */
202
203 mem_limit_func(); /* check for "mem=" argument */
204
205 mem_max = 0;
206 num_physpages = 0;
207 for (i = 0; i < npmem_ranges; i++) {
208 unsigned long rsize;
209
210 rsize = pmem_ranges[i].pages << PAGE_SHIFT;
211 if ((mem_max + rsize) > mem_limit) {
212 printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
213 if (mem_max == mem_limit)
214 npmem_ranges = i;
215 else {
216 pmem_ranges[i].pages = (mem_limit >> PAGE_SHIFT)
217 - (mem_max >> PAGE_SHIFT);
218 npmem_ranges = i + 1;
219 mem_max = mem_limit;
220 }
221 num_physpages += pmem_ranges[i].pages;
222 break;
223 }
224 num_physpages += pmem_ranges[i].pages;
225 mem_max += rsize;
226 }
227
228 printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
229
230#ifndef CONFIG_DISCONTIGMEM
231 /* Merge the ranges, keeping track of the holes */
232
233 {
234 unsigned long end_pfn;
235 unsigned long hole_pages;
236
237 npmem_holes = 0;
238 end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
239 for (i = 1; i < npmem_ranges; i++) {
240
241 hole_pages = pmem_ranges[i].start_pfn - end_pfn;
242 if (hole_pages) {
243 pmem_holes[npmem_holes].start_pfn = end_pfn;
244 pmem_holes[npmem_holes++].pages = hole_pages;
245 end_pfn += hole_pages;
246 }
247 end_pfn += pmem_ranges[i].pages;
248 }
249
250 pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
251 npmem_ranges = 1;
252 }
253#endif
254
255 bootmap_pages = 0;
256 for (i = 0; i < npmem_ranges; i++)
257 bootmap_pages += bootmem_bootmap_pages(pmem_ranges[i].pages);
258
259 bootmap_start_pfn = PAGE_ALIGN(__pa((unsigned long) &_end)) >> PAGE_SHIFT;
260
261#ifdef CONFIG_DISCONTIGMEM
262 for (i = 0; i < MAX_PHYSMEM_RANGES; i++) {
263 memset(NODE_DATA(i), 0, sizeof(pg_data_t));
b61bfa3c 264 NODE_DATA(i)->bdata = &bootmem_node_data[i];
1da177e4
LT		 265 }
266 memset(pfnnid_map, 0xff, sizeof(pfnnid_map));
267
268 for (i = 0; i < npmem_ranges; i++)
269 node_set_online(i);
270#endif
271
272 /*
273 * Initialize and free the full range of memory in each range.
274 * Note that the only writing these routines do are to the bootmap,
275 * and we've made sure to locate the bootmap properly so that they
276 * won't be writing over anything important.
277 */
278
279 bootmap_pfn = bootmap_start_pfn;
280 max_pfn = 0;
281 for (i = 0; i < npmem_ranges; i++) {
282 unsigned long start_pfn;
283 unsigned long npages;
284
285 start_pfn = pmem_ranges[i].start_pfn;
286 npages = pmem_ranges[i].pages;
287
288 bootmap_size = init_bootmem_node(NODE_DATA(i),
289 bootmap_pfn,
290 start_pfn,
291 (start_pfn + npages) );
292 free_bootmem_node(NODE_DATA(i),
293 (start_pfn << PAGE_SHIFT),
294 (npages << PAGE_SHIFT) );
295 bootmap_pfn += (bootmap_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
296 if ((start_pfn + npages) > max_pfn)
297 max_pfn = start_pfn + npages;
298 }
299
5cdb8205
GG		 300 /* IOMMU is always used to access "high mem" on those boxes
301 * that can support enough mem that a PCI device couldn't
302 * directly DMA to any physical addresses.
303 * ISA DMA support will need to revisit this.
304 */
305 max_low_pfn = max_pfn;
306
8980a7ba
HD		 307 /* bootmap sizing messed up? */
308 BUG_ON((bootmap_pfn - bootmap_start_pfn) != bootmap_pages);
1da177e4
LT		 309
310 /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
311
312#define PDC_CONSOLE_IO_IODC_SIZE 32768
313
314 reserve_bootmem_node(NODE_DATA(0), 0UL,
72a7fe39
BW		 315 (unsigned long)(PAGE0->mem_free +
316 PDC_CONSOLE_IO_IODC_SIZE), BOOTMEM_DEFAULT);
c51d476a 317 reserve_bootmem_node(NODE_DATA(0), __pa((unsigned long)_text),
72a7fe39 318 (unsigned long)(_end - _text), BOOTMEM_DEFAULT);
1da177e4 319 reserve_bootmem_node(NODE_DATA(0), (bootmap_start_pfn << PAGE_SHIFT),
72a7fe39
BW		 320 ((bootmap_pfn - bootmap_start_pfn) << PAGE_SHIFT),
321 BOOTMEM_DEFAULT);
1da177e4
LT		 322
323#ifndef CONFIG_DISCONTIGMEM
324
325 /* reserve the holes */
326
327 for (i = 0; i < npmem_holes; i++) {
328 reserve_bootmem_node(NODE_DATA(0),
329 (pmem_holes[i].start_pfn << PAGE_SHIFT),
72a7fe39
BW		 330 (pmem_holes[i].pages << PAGE_SHIFT),
331 BOOTMEM_DEFAULT);
1da177e4
LT		 332 }
333#endif
334
335#ifdef CONFIG_BLK_DEV_INITRD
336 if (initrd_start) {
337 printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
338 if (__pa(initrd_start) < mem_max) {
339 unsigned long initrd_reserve;
340
341 if (__pa(initrd_end) > mem_max) {
342 initrd_reserve = mem_max - __pa(initrd_start);
343 } else {
344 initrd_reserve = initrd_end - initrd_start;
345 }
346 initrd_below_start_ok = 1;
347 printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
348
72a7fe39
BW		 349 reserve_bootmem_node(NODE_DATA(0), __pa(initrd_start),
350 initrd_reserve, BOOTMEM_DEFAULT);
1da177e4
LT		 351 }
352 }
353#endif
354
355 data_resource.start = virt_to_phys(&data_start);
c51d476a
KM		 356 data_resource.end = virt_to_phys(_end) - 1;
357 code_resource.start = virt_to_phys(_text);
1da177e4
LT		 358 code_resource.end = virt_to_phys(&data_start)-1;
359
360 /* We don't know which region the kernel will be in, so try
361 * all of them.
362 */
363 for (i = 0; i < sysram_resource_count; i++) {
364 struct resource *res = &sysram_resources[i];
365 request_resource(res, &code_resource);
366 request_resource(res, &data_resource);
367 }
368 request_resource(&sysram_resources[0], &pdcdata_resource);
369}
370
371void free_initmem(void)
372{
4fb11781
KM		 373 unsigned long addr;
374 unsigned long init_begin = (unsigned long)__init_begin;
375 unsigned long init_end = (unsigned long)__init_end;
1da177e4 376
81a3de3e 377#ifdef CONFIG_DEBUG_KERNEL
1da177e4
LT		 378 /* Attempt to catch anyone trying to execute code here
379 * by filling the page with BRK insns.
1da177e4 380 */
20dbc9f7 381 memset((void *)init_begin, 0x00, init_end - init_begin);
4fb11781 382 flush_icache_range(init_begin, init_end);
1da177e4
LT		 383#endif
384
2fd83038
HD		 385 /* align __init_begin and __init_end to page size,
386 ignoring linker script where we might have tried to save RAM */
4fb11781
KM		 387 init_begin = PAGE_ALIGN(init_begin);
388 init_end = PAGE_ALIGN(init_end);
2fd83038 389 for (addr = init_begin; addr < init_end; addr += PAGE_SIZE) {
1da177e4 390 ClearPageReserved(virt_to_page(addr));
7835e98b 391 init_page_count(virt_to_page(addr));
1da177e4
LT		 392 free_page(addr);
393 num_physpages++;
394 totalram_pages++;
395 }
396
397 /* set up a new led state on systems shipped LED State panel */
398 pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
399
4fb11781
KM		 400 printk(KERN_INFO "Freeing unused kernel memory: %luk freed\n",
401 (init_end - init_begin) >> 10);
1da177e4
LT		 402}
403
1bcdd854
HD		 404
405#ifdef CONFIG_DEBUG_RODATA
406void mark_rodata_ro(void)
407{
1bcdd854
HD		 408 /* rodata memory was already mapped with KERNEL_RO access rights by
409 pagetable_init() and map_pages(). No need to do additional stuff here */
410 printk (KERN_INFO "Write protecting the kernel read-only data: %luk\n",
a581c2a4 411 (unsigned long)(__end_rodata - __start_rodata) >> 10);
1bcdd854
HD		 412}
413#endif
414
415
1da177e4
LT		 416/*
417 * Just an arbitrary offset to serve as a "hole" between mapping areas
418 * (between top of physical memory and a potential pcxl dma mapping
419 * area, and below the vmalloc mapping area).
420 *
421 * The current 32K value just means that there will be a 32K "hole"
422 * between mapping areas. That means that any out-of-bounds memory
423 * accesses will hopefully be caught. The vmalloc() routines leaves
424 * a hole of 4kB between each vmalloced area for the same reason.
425 */
426
427 /* Leave room for gateway page expansion */
428#if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
429#error KERNEL_MAP_START is in gateway reserved region
430#endif
431#define MAP_START (KERNEL_MAP_START)
432
433#define VM_MAP_OFFSET (32*1024)
434#define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
435 & ~(VM_MAP_OFFSET-1)))
436
8039de10 437void *vmalloc_start __read_mostly;
1da177e4
LT		 438EXPORT_SYMBOL(vmalloc_start);
439
440#ifdef CONFIG_PA11
8039de10 441unsigned long pcxl_dma_start __read_mostly;
1da177e4
LT		 442#endif
443
444void __init mem_init(void)
445{
ce8420bb 446 int codesize, reservedpages, datasize, initsize;
ce8420bb 447
48d27cb2
HD		 448 /* Do sanity checks on page table constants */
449 BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
450 BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
451 BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
452 BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
453 > BITS_PER_LONG);
454
1da177e4
LT		 455 high_memory = __va((max_pfn << PAGE_SHIFT));
456
457#ifndef CONFIG_DISCONTIGMEM
458 max_mapnr = page_to_pfn(virt_to_page(high_memory - 1)) + 1;
459 totalram_pages += free_all_bootmem();
460#else
461 {
462 int i;
463
464 for (i = 0; i < npmem_ranges; i++)
465 totalram_pages += free_all_bootmem_node(NODE_DATA(i));
466 }
467#endif
468
53faf291
KM		 469 codesize = (unsigned long)_etext - (unsigned long)_text;
470 datasize = (unsigned long)_edata - (unsigned long)_etext;
471 initsize = (unsigned long)__init_end - (unsigned long)__init_begin;
ce8420bb
HD		 472
473 reservedpages = 0;
53faf291
KM		 474{
475 unsigned long pfn;
476#ifdef CONFIG_DISCONTIGMEM
477 int i;
478
479 for (i = 0; i < npmem_ranges; i++) {
480 for (pfn = node_start_pfn(i); pfn < node_end_pfn(i); pfn++) {
481 if (PageReserved(pfn_to_page(pfn)))
482 reservedpages++;
483 }
484 }
485#else /* !CONFIG_DISCONTIGMEM */
486 for (pfn = 0; pfn < max_pfn; pfn++) {
ce8420bb
HD		 487 /*
488 * Only count reserved RAM pages
489 */
53faf291 490 if (PageReserved(pfn_to_page(pfn)))
ce8420bb 491 reservedpages++;
53faf291
KM		 492 }
493#endif
494}
1da177e4
LT		 495
496#ifdef CONFIG_PA11
497 if (hppa_dma_ops == &pcxl_dma_ops) {
498 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
499 vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start + PCXL_DMA_MAP_SIZE);
500 } else {
501 pcxl_dma_start = 0;
502 vmalloc_start = SET_MAP_OFFSET(MAP_START);
503 }
504#else
505 vmalloc_start = SET_MAP_OFFSET(MAP_START);
506#endif
507
53faf291 508 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init)\n",
cc013a88 509 nr_free_pages() << (PAGE_SHIFT-10),
ce8420bb
HD		 510 num_physpages << (PAGE_SHIFT-10),
511 codesize >> 10,
512 reservedpages << (PAGE_SHIFT-10),
513 datasize >> 10,
53faf291
KM		 514 initsize >> 10
515 );
ce8420bb
HD		 516
517#ifdef CONFIG_DEBUG_KERNEL /* double-sanity-check paranoia */
518 printk("virtual kernel memory layout:\n"
519 " vmalloc : 0x%p - 0x%p (%4ld MB)\n"
53faf291 520 " memory : 0x%p - 0x%p (%4ld MB)\n"
ce8420bb
HD		 521 " .init : 0x%p - 0x%p (%4ld kB)\n"
522 " .data : 0x%p - 0x%p (%4ld kB)\n"
523 " .text : 0x%p - 0x%p (%4ld kB)\n",
524
525 (void*)VMALLOC_START, (void*)VMALLOC_END,
526 (VMALLOC_END - VMALLOC_START) >> 20,
527
528 __va(0), high_memory,
529 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
530
53faf291
KM		 531 __init_begin, __init_end,
532 ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
ce8420bb 533
53faf291
KM		 534 _etext, _edata,
535 ((unsigned long)_edata - (unsigned long)_etext) >> 10,
ce8420bb 536
53faf291
KM		 537 _text, _etext,
538 ((unsigned long)_etext - (unsigned long)_text) >> 10);
ce8420bb 539#endif
1da177e4
LT		 540}
541
8039de10 542unsigned long *empty_zero_page __read_mostly;
22febf1f 543EXPORT_SYMBOL(empty_zero_page);
1da177e4
LT		 544
545void show_mem(void)
546{
547 int i,free = 0,total = 0,reserved = 0;
548 int shared = 0, cached = 0;
549
550 printk(KERN_INFO "Mem-info:\n");
551 show_free_areas();
1da177e4
LT		 552#ifndef CONFIG_DISCONTIGMEM
553 i = max_mapnr;
554 while (i-- > 0) {
555 total++;
556 if (PageReserved(mem_map+i))
557 reserved++;
558 else if (PageSwapCache(mem_map+i))
559 cached++;
560 else if (!page_count(&mem_map[i]))
561 free++;
562 else
563 shared += page_count(&mem_map[i]) - 1;
564 }
565#else
566 for (i = 0; i < npmem_ranges; i++) {
567 int j;
568
569 for (j = node_start_pfn(i); j < node_end_pfn(i); j++) {
570 struct page *p;
208d54e5 571 unsigned long flags;
1da177e4 572
208d54e5 573 pgdat_resize_lock(NODE_DATA(i), &flags);
408fde81 574 p = nid_page_nr(i, j) - node_start_pfn(i);
1da177e4
LT		 575
576 total++;
577 if (PageReserved(p))
578 reserved++;
579 else if (PageSwapCache(p))
580 cached++;
581 else if (!page_count(p))
582 free++;
583 else
584 shared += page_count(p) - 1;
208d54e5 585 pgdat_resize_unlock(NODE_DATA(i), &flags);
1da177e4
LT		 586 }
587 }
588#endif
589 printk(KERN_INFO "%d pages of RAM\n", total);
590 printk(KERN_INFO "%d reserved pages\n", reserved);
591 printk(KERN_INFO "%d pages shared\n", shared);
592 printk(KERN_INFO "%d pages swap cached\n", cached);
593
594
595#ifdef CONFIG_DISCONTIGMEM
596 {
597 struct zonelist *zl;
54a6eb5c 598 int i, j;
1da177e4
LT		 599
600 for (i = 0; i < npmem_ranges; i++) {
4413a0f6 601 zl = node_zonelist(i, 0);
1da177e4 602 for (j = 0; j < MAX_NR_ZONES; j++) {
dd1a239f 603 struct zoneref *z;
54a6eb5c 604 struct zone *zone;
1da177e4
LT		 605
606 printk("Zone list for zone %d on node %d: ", j, i);
54a6eb5c
MG		 607 for_each_zone_zonelist(zone, z, zl, j)
608 printk("[%d/%s] ", zone_to_nid(zone),
609 zone->name);
1da177e4
LT		 610 printk("\n");
611 }
612 }
613 }
614#endif
615}
616
617
618static void __init map_pages(unsigned long start_vaddr, unsigned long start_paddr, unsigned long size, pgprot_t pgprot)
619{
620 pgd_t *pg_dir;
621 pmd_t *pmd;
622 pte_t *pg_table;
623 unsigned long end_paddr;
624 unsigned long start_pmd;
625 unsigned long start_pte;
626 unsigned long tmp1;
627 unsigned long tmp2;
628 unsigned long address;
629 unsigned long ro_start;
630 unsigned long ro_end;
631 unsigned long fv_addr;
632 unsigned long gw_addr;
633 extern const unsigned long fault_vector_20;
634 extern void * const linux_gateway_page;
635
c51d476a 636 ro_start = __pa((unsigned long)_text);
1da177e4
LT		 637 ro_end = __pa((unsigned long)&data_start);
638 fv_addr = __pa((unsigned long)&fault_vector_20) & PAGE_MASK;
639 gw_addr = __pa((unsigned long)&linux_gateway_page) & PAGE_MASK;
640
641 end_paddr = start_paddr + size;
642
643 pg_dir = pgd_offset_k(start_vaddr);
644
645#if PTRS_PER_PMD == 1
646 start_pmd = 0;
647#else
648 start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
649#endif
650 start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
651
652 address = start_paddr;
653 while (address < end_paddr) {
654#if PTRS_PER_PMD == 1
655 pmd = (pmd_t *)__pa(pg_dir);
656#else
657 pmd = (pmd_t *)pgd_address(*pg_dir);
658
659 /*
660 * pmd is physical at this point
661 */
662
663 if (!pmd) {
664 pmd = (pmd_t *) alloc_bootmem_low_pages_node(NODE_DATA(0),PAGE_SIZE << PMD_ORDER);
665 pmd = (pmd_t *) __pa(pmd);
666 }
667
668 pgd_populate(NULL, pg_dir, __va(pmd));
669#endif
670 pg_dir++;
671
672 /* now change pmd to kernel virtual addresses */
673
674 pmd = (pmd_t *)__va(pmd) + start_pmd;
675 for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++,pmd++) {
676
677 /*
678 * pg_table is physical at this point
679 */
680
681 pg_table = (pte_t *)pmd_address(*pmd);
682 if (!pg_table) {
683 pg_table = (pte_t *)
684 alloc_bootmem_low_pages_node(NODE_DATA(0),PAGE_SIZE);
685 pg_table = (pte_t *) __pa(pg_table);
686 }
687
688 pmd_populate_kernel(NULL, pmd, __va(pg_table));
689
690 /* now change pg_table to kernel virtual addresses */
691
692 pg_table = (pte_t *) __va(pg_table) + start_pte;
693 for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++,pg_table++) {
694 pte_t pte;
695
696 /*
697 * Map the fault vector writable so we can
698 * write the HPMC checksum.
699 */
2fd83038 700#if defined(CONFIG_PARISC_PAGE_SIZE_4KB)
1da177e4
LT		 701 if (address >= ro_start && address < ro_end
702 && address != fv_addr
703 && address != gw_addr)
704 pte = __mk_pte(address, PAGE_KERNEL_RO);
705 else
2fd83038 706#endif
1da177e4
LT		 707 pte = __mk_pte(address, pgprot);
708
709 if (address >= end_paddr)
710 pte_val(pte) = 0;
711
712 set_pte(pg_table, pte);
713
714 address += PAGE_SIZE;
715 }
716 start_pte = 0;
717
718 if (address >= end_paddr)
719 break;
720 }
721 start_pmd = 0;
722 }
723}
724
725/*
726 * pagetable_init() sets up the page tables
727 *
728 * Note that gateway_init() places the Linux gateway page at page 0.
729 * Since gateway pages cannot be dereferenced this has the desirable
730 * side effect of trapping those pesky NULL-reference errors in the
731 * kernel.
732 */
733static void __init pagetable_init(void)
734{
735 int range;
736
737 /* Map each physical memory range to its kernel vaddr */
738
739 for (range = 0; range < npmem_ranges; range++) {
740 unsigned long start_paddr;
741 unsigned long end_paddr;
742 unsigned long size;
743
744 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
745 end_paddr = start_paddr + (pmem_ranges[range].pages << PAGE_SHIFT);
746 size = pmem_ranges[range].pages << PAGE_SHIFT;
747
748 map_pages((unsigned long)__va(start_paddr), start_paddr,
749 size, PAGE_KERNEL);
750 }
751
752#ifdef CONFIG_BLK_DEV_INITRD
753 if (initrd_end && initrd_end > mem_limit) {
1bcdd854 754 printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
1da177e4
LT		 755 map_pages(initrd_start, __pa(initrd_start),
756 initrd_end - initrd_start, PAGE_KERNEL);
757 }
758#endif
759
760 empty_zero_page = alloc_bootmem_pages(PAGE_SIZE);
761 memset(empty_zero_page, 0, PAGE_SIZE);
762}
763
764static void __init gateway_init(void)
765{
766 unsigned long linux_gateway_page_addr;
767 /* FIXME: This is 'const' in order to trick the compiler
768 into not treating it as DP-relative data. */
769 extern void * const linux_gateway_page;
770
771 linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
772
773 /*
774 * Setup Linux Gateway page.
775 *
776 * The Linux gateway page will reside in kernel space (on virtual
777 * page 0), so it doesn't need to be aliased into user space.
778 */
779
780 map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
781 PAGE_SIZE, PAGE_GATEWAY);
782}
783
784#ifdef CONFIG_HPUX
785void
786map_hpux_gateway_page(struct task_struct *tsk, struct mm_struct *mm)
787{
788 pgd_t *pg_dir;
789 pmd_t *pmd;
790 pte_t *pg_table;
791 unsigned long start_pmd;
792 unsigned long start_pte;
793 unsigned long address;
794 unsigned long hpux_gw_page_addr;
795 /* FIXME: This is 'const' in order to trick the compiler
796 into not treating it as DP-relative data. */
797 extern void * const hpux_gateway_page;
798
799 hpux_gw_page_addr = HPUX_GATEWAY_ADDR & PAGE_MASK;
800
801 /*
802 * Setup HP-UX Gateway page.
803 *
804 * The HP-UX gateway page resides in the user address space,
805 * so it needs to be aliased into each process.
806 */
807
808 pg_dir = pgd_offset(mm,hpux_gw_page_addr);
809
810#if PTRS_PER_PMD == 1
811 start_pmd = 0;
812#else
813 start_pmd = ((hpux_gw_page_addr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
814#endif
815 start_pte = ((hpux_gw_page_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
816
817 address = __pa(&hpux_gateway_page);
818#if PTRS_PER_PMD == 1
819 pmd = (pmd_t *)__pa(pg_dir);
820#else
821 pmd = (pmd_t *) pgd_address(*pg_dir);
822
823 /*
824 * pmd is physical at this point
825 */
826
827 if (!pmd) {
828 pmd = (pmd_t *) get_zeroed_page(GFP_KERNEL);
829 pmd = (pmd_t *) __pa(pmd);
830 }
831
832 __pgd_val_set(*pg_dir, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pmd);
833#endif
834 /* now change pmd to kernel virtual addresses */
835
836 pmd = (pmd_t *)__va(pmd) + start_pmd;
837
838 /*
839 * pg_table is physical at this point
840 */
841
842 pg_table = (pte_t *) pmd_address(*pmd);
843 if (!pg_table)
844 pg_table = (pte_t *) __pa(get_zeroed_page(GFP_KERNEL));
845
846 __pmd_val_set(*pmd, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pg_table);
847
848 /* now change pg_table to kernel virtual addresses */
849
850 pg_table = (pte_t *) __va(pg_table) + start_pte;
851 set_pte(pg_table, __mk_pte(address, PAGE_GATEWAY));
852}
853EXPORT_SYMBOL(map_hpux_gateway_page);
854#endif
855
1da177e4
LT		 856void __init paging_init(void)
857{
858 int i;
859
860 setup_bootmem();
861 pagetable_init();
862 gateway_init();
863 flush_cache_all_local(); /* start with known state */
ce33941f 864 flush_tlb_all_local(NULL);
1da177e4
LT		 865
866 for (i = 0; i < npmem_ranges; i++) {
f06a9684 867 unsigned long zones_size[MAX_NR_ZONES] = { 0, };
1da177e4 868
00592837 869 zones_size[ZONE_NORMAL] = pmem_ranges[i].pages;
1da177e4
LT		 870
871#ifdef CONFIG_DISCONTIGMEM
872 /* Need to initialize the pfnnid_map before we can initialize
873 the zone */
874 {
875 int j;
876 for (j = (pmem_ranges[i].start_pfn >> PFNNID_SHIFT);
877 j <= ((pmem_ranges[i].start_pfn + pmem_ranges[i].pages) >> PFNNID_SHIFT);
878 j++) {
879 pfnnid_map[j] = i;
880 }
881 }
882#endif
883
9109fb7b 884 free_area_init_node(i, zones_size,
1da177e4
LT		 885 pmem_ranges[i].start_pfn, NULL);
886 }
887}
888
889#ifdef CONFIG_PA20
890
891/*
7022672e 892 * Currently, all PA20 chips have 18 bit protection IDs, which is the
1da177e4
LT		 893 * limiting factor (space ids are 32 bits).
894 */
895
896#define NR_SPACE_IDS 262144
897
898#else
899
900/*
7022672e
SA		 901 * Currently we have a one-to-one relationship between space IDs and
902 * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
903 * support 15 bit protection IDs, so that is the limiting factor.
904 * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
1da177e4
LT		 905 * probably not worth the effort for a special case here.
906 */
907
908#define NR_SPACE_IDS 32768
909
910#endif /* !CONFIG_PA20 */
911
912#define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
913#define SID_ARRAY_SIZE (NR_SPACE_IDS / (8 * sizeof(long)))
914
915static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
916static unsigned long dirty_space_id[SID_ARRAY_SIZE];
917static unsigned long space_id_index;
918static unsigned long free_space_ids = NR_SPACE_IDS - 1;
919static unsigned long dirty_space_ids = 0;
920
921static DEFINE_SPINLOCK(sid_lock);
922
923unsigned long alloc_sid(void)
924{
925 unsigned long index;
926
927 spin_lock(&sid_lock);
928
929 if (free_space_ids == 0) {
930 if (dirty_space_ids != 0) {
931 spin_unlock(&sid_lock);
932 flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
933 spin_lock(&sid_lock);
934 }
2fd83038 935 BUG_ON(free_space_ids == 0);
1da177e4
LT		 936 }
937
938 free_space_ids--;
939
940 index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
941 space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1)));
942 space_id_index = index;
943
944 spin_unlock(&sid_lock);
945
946 return index << SPACEID_SHIFT;
947}
948
949void free_sid(unsigned long spaceid)
950{
951 unsigned long index = spaceid >> SPACEID_SHIFT;
952 unsigned long *dirty_space_offset;
953
954 dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG);
955 index &= (BITS_PER_LONG - 1);
956
957 spin_lock(&sid_lock);
958
2fd83038 959 BUG_ON(*dirty_space_offset & (1L << index)); /* attempt to free space id twice */
1da177e4
LT		 960
961 *dirty_space_offset |= (1L << index);
962 dirty_space_ids++;
963
964 spin_unlock(&sid_lock);
965}
966
967
968#ifdef CONFIG_SMP
969static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
970{
971 int i;
972
973 /* NOTE: sid_lock must be held upon entry */
974
975 *ndirtyptr = dirty_space_ids;
976 if (dirty_space_ids != 0) {
977 for (i = 0; i < SID_ARRAY_SIZE; i++) {
978 dirty_array[i] = dirty_space_id[i];
979 dirty_space_id[i] = 0;
980 }
981 dirty_space_ids = 0;
982 }
983
984 return;
985}
986
987static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
988{
989 int i;
990
991 /* NOTE: sid_lock must be held upon entry */
992
993 if (ndirty != 0) {
994 for (i = 0; i < SID_ARRAY_SIZE; i++) {
995 space_id[i] ^= dirty_array[i];
996 }
997
998 free_space_ids += ndirty;
999 space_id_index = 0;
1000 }
1001}
1002
1003#else /* CONFIG_SMP */
1004
1005static void recycle_sids(void)
1006{
1007 int i;
1008
1009 /* NOTE: sid_lock must be held upon entry */
1010
1011 if (dirty_space_ids != 0) {
1012 for (i = 0; i < SID_ARRAY_SIZE; i++) {
1013 space_id[i] ^= dirty_space_id[i];
1014 dirty_space_id[i] = 0;
1015 }
1016
1017 free_space_ids += dirty_space_ids;
1018 dirty_space_ids = 0;
1019 space_id_index = 0;
1020 }
1021}
1022#endif
1023
1024/*
1025 * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
1026 * purged, we can safely reuse the space ids that were released but
1027 * not flushed from the tlb.
1028 */
1029
1030#ifdef CONFIG_SMP
1031
1032static unsigned long recycle_ndirty;
1033static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
2fd83038 1034static unsigned int recycle_inuse;
1da177e4
LT		 1035
1036void flush_tlb_all(void)
1037{
1038 int do_recycle;
1039
1040 do_recycle = 0;
1041 spin_lock(&sid_lock);
1042 if (dirty_space_ids > RECYCLE_THRESHOLD) {
2fd83038 1043 BUG_ON(recycle_inuse); /* FIXME: Use a semaphore/wait queue here */
1da177e4
LT		 1044 get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
1045 recycle_inuse++;
1046 do_recycle++;
1047 }
1048 spin_unlock(&sid_lock);
15c8b6c1 1049 on_each_cpu(flush_tlb_all_local, NULL, 1);
1da177e4
LT		 1050 if (do_recycle) {
1051 spin_lock(&sid_lock);
1052 recycle_sids(recycle_ndirty,recycle_dirty_array);
1053 recycle_inuse = 0;
1054 spin_unlock(&sid_lock);
1055 }
1056}
1057#else
1058void flush_tlb_all(void)
1059{
1060 spin_lock(&sid_lock);
1b2425e3 1061 flush_tlb_all_local(NULL);
1da177e4
LT		 1062 recycle_sids();
1063 spin_unlock(&sid_lock);
1064}
1065#endif
1066
1067#ifdef CONFIG_BLK_DEV_INITRD
1068void free_initrd_mem(unsigned long start, unsigned long end)
1069{
94c3e87a
HD		 1070 if (start >= end)
1071 return;
1072 printk(KERN_INFO "Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
1da177e4
LT		 1073 for (; start < end; start += PAGE_SIZE) {
1074 ClearPageReserved(virt_to_page(start));
7835e98b 1075 init_page_count(virt_to_page(start));
1da177e4
LT		 1076 free_page(start);
1077 num_physpages++;
1078 totalram_pages++;
1079 }
1da177e4
LT		 1080}
1081#endif
Linux 6.x block layer and io_uring tree(s)