projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
mm: use two zonelist that are filtered by GFP mask
[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
HD		 38struct node_map_data node_data[MAX_NUMNODES] __read_mostly;
39bootmem_data_t bmem_data[MAX_NUMNODES] __read_mostly;
40unsigned char pfnnid_map[PFNNID_MAP_MAX] __read_mostly;
1da177e4
LT		 41#endif
42
43static struct resource data_resource = {
44 .name = "Kernel data",
45 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
46};
47
48static struct resource code_resource = {
49 .name = "Kernel code",
50 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
51};
52
53static struct resource pdcdata_resource = {
54 .name = "PDC data (Page Zero)",
55 .start = 0,
56 .end = 0x9ff,
57 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
58};
59
8039de10 60static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __read_mostly;
1da177e4
LT		 61
62/* The following array is initialized from the firmware specific
63 * information retrieved in kernel/inventory.c.
64 */
65
8039de10
HD		 66physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __read_mostly;
67int npmem_ranges __read_mostly;
1da177e4 68
a8f44e38 69#ifdef CONFIG_64BIT
1da177e4 70#define MAX_MEM (~0UL)
a8f44e38 71#else /* !CONFIG_64BIT */
1da177e4 72#define MAX_MEM (3584U*1024U*1024U)
a8f44e38 73#endif /* !CONFIG_64BIT */
1da177e4 74
8039de10 75static unsigned long mem_limit __read_mostly = MAX_MEM;
1da177e4
LT		 76
77static void __init mem_limit_func(void)
78{
79 char *cp, *end;
80 unsigned long limit;
1da177e4
LT		 81
82 /* We need this before __setup() functions are called */
83
84 limit = MAX_MEM;
668f9931 85 for (cp = boot_command_line; *cp; ) {
1da177e4
LT		 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
106static 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 = &bmem_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
5cdb8205
GG		 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
1da177e4
LT		 308 if ((bootmap_pfn - bootmap_start_pfn) != bootmap_pages) {
309 printk(KERN_WARNING "WARNING! bootmap sizing is messed up!\n");
310 BUG();
311 }
312
313 /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
314
315#define PDC_CONSOLE_IO_IODC_SIZE 32768
316
317 reserve_bootmem_node(NODE_DATA(0), 0UL,
72a7fe39
BW		 318 (unsigned long)(PAGE0->mem_free +
319 PDC_CONSOLE_IO_IODC_SIZE), BOOTMEM_DEFAULT);
c51d476a 320 reserve_bootmem_node(NODE_DATA(0), __pa((unsigned long)_text),
72a7fe39 321 (unsigned long)(_end - _text), BOOTMEM_DEFAULT);
1da177e4 322 reserve_bootmem_node(NODE_DATA(0), (bootmap_start_pfn << PAGE_SHIFT),
72a7fe39
BW		 323 ((bootmap_pfn - bootmap_start_pfn) << PAGE_SHIFT),
324 BOOTMEM_DEFAULT);
1da177e4
LT		 325
326#ifndef CONFIG_DISCONTIGMEM
327
328 /* reserve the holes */
329
330 for (i = 0; i < npmem_holes; i++) {
331 reserve_bootmem_node(NODE_DATA(0),
332 (pmem_holes[i].start_pfn << PAGE_SHIFT),
72a7fe39
BW		 333 (pmem_holes[i].pages << PAGE_SHIFT),
334 BOOTMEM_DEFAULT);
1da177e4
LT		 335 }
336#endif
337
338#ifdef CONFIG_BLK_DEV_INITRD
339 if (initrd_start) {
340 printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
341 if (__pa(initrd_start) < mem_max) {
342 unsigned long initrd_reserve;
343
344 if (__pa(initrd_end) > mem_max) {
345 initrd_reserve = mem_max - __pa(initrd_start);
346 } else {
347 initrd_reserve = initrd_end - initrd_start;
348 }
349 initrd_below_start_ok = 1;
350 printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
351
72a7fe39
BW		 352 reserve_bootmem_node(NODE_DATA(0), __pa(initrd_start),
353 initrd_reserve, BOOTMEM_DEFAULT);
1da177e4
LT		 354 }
355 }
356#endif
357
358 data_resource.start = virt_to_phys(&data_start);
c51d476a
KM		 359 data_resource.end = virt_to_phys(_end) - 1;
360 code_resource.start = virt_to_phys(_text);
1da177e4
LT		 361 code_resource.end = virt_to_phys(&data_start)-1;
362
363 /* We don't know which region the kernel will be in, so try
364 * all of them.
365 */
366 for (i = 0; i < sysram_resource_count; i++) {
367 struct resource *res = &sysram_resources[i];
368 request_resource(res, &code_resource);
369 request_resource(res, &data_resource);
370 }
371 request_resource(&sysram_resources[0], &pdcdata_resource);
372}
373
374void free_initmem(void)
375{
2fd83038
HD		 376 unsigned long addr, init_begin, init_end;
377
1da177e4
LT		 378 printk(KERN_INFO "Freeing unused kernel memory: ");
379
81a3de3e 380#ifdef CONFIG_DEBUG_KERNEL
1da177e4
LT		 381 /* Attempt to catch anyone trying to execute code here
382 * by filling the page with BRK insns.
383 *
384 * If we disable interrupts for all CPUs, then IPI stops working.
385 * Kinda breaks the global cache flushing.
386 */
387 local_irq_disable();
388
c51d476a
KM		 389 memset(__init_begin, 0x00,
390 (unsigned long)__init_end - (unsigned long)__init_begin);
1da177e4
LT		 391
392 flush_data_cache();
393 asm volatile("sync" : : );
c51d476a 394 flush_icache_range((unsigned long)__init_begin, (unsigned long)__init_end);
1da177e4
LT		 395 asm volatile("sync" : : );
396
397 local_irq_enable();
398#endif
399
2fd83038
HD		 400 /* align __init_begin and __init_end to page size,
401 ignoring linker script where we might have tried to save RAM */
c51d476a
KM		 402 init_begin = PAGE_ALIGN((unsigned long)(__init_begin));
403 init_end = PAGE_ALIGN((unsigned long)(__init_end));
2fd83038 404 for (addr = init_begin; addr < init_end; addr += PAGE_SIZE) {
1da177e4 405 ClearPageReserved(virt_to_page(addr));
7835e98b 406 init_page_count(virt_to_page(addr));
1da177e4
LT		 407 free_page(addr);
408 num_physpages++;
409 totalram_pages++;
410 }
411
412 /* set up a new led state on systems shipped LED State panel */
413 pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
414
2fd83038 415 printk("%luk freed\n", (init_end - init_begin) >> 10);
1da177e4
LT		 416}
417
1bcdd854
HD		 418
419#ifdef CONFIG_DEBUG_RODATA
420void mark_rodata_ro(void)
421{
1bcdd854
HD		 422 /* rodata memory was already mapped with KERNEL_RO access rights by
423 pagetable_init() and map_pages(). No need to do additional stuff here */
424 printk (KERN_INFO "Write protecting the kernel read-only data: %luk\n",
a581c2a4 425 (unsigned long)(__end_rodata - __start_rodata) >> 10);
1bcdd854
HD		 426}
427#endif
428
429
1da177e4
LT		 430/*
431 * Just an arbitrary offset to serve as a "hole" between mapping areas
432 * (between top of physical memory and a potential pcxl dma mapping
433 * area, and below the vmalloc mapping area).
434 *
435 * The current 32K value just means that there will be a 32K "hole"
436 * between mapping areas. That means that any out-of-bounds memory
437 * accesses will hopefully be caught. The vmalloc() routines leaves
438 * a hole of 4kB between each vmalloced area for the same reason.
439 */
440
441 /* Leave room for gateway page expansion */
442#if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
443#error KERNEL_MAP_START is in gateway reserved region
444#endif
445#define MAP_START (KERNEL_MAP_START)
446
447#define VM_MAP_OFFSET (32*1024)
448#define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
449 & ~(VM_MAP_OFFSET-1)))
450
8039de10 451void *vmalloc_start __read_mostly;
1da177e4
LT		 452EXPORT_SYMBOL(vmalloc_start);
453
454#ifdef CONFIG_PA11
8039de10 455unsigned long pcxl_dma_start __read_mostly;
1da177e4
LT		 456#endif
457
458void __init mem_init(void)
459{
ce8420bb 460 int codesize, reservedpages, datasize, initsize;
ce8420bb 461
1da177e4
LT		 462 high_memory = __va((max_pfn << PAGE_SHIFT));
463
464#ifndef CONFIG_DISCONTIGMEM
465 max_mapnr = page_to_pfn(virt_to_page(high_memory - 1)) + 1;
466 totalram_pages += free_all_bootmem();
467#else
468 {
469 int i;
470
471 for (i = 0; i < npmem_ranges; i++)
472 totalram_pages += free_all_bootmem_node(NODE_DATA(i));
473 }
474#endif
475
53faf291
KM		 476 codesize = (unsigned long)_etext - (unsigned long)_text;
477 datasize = (unsigned long)_edata - (unsigned long)_etext;
478 initsize = (unsigned long)__init_end - (unsigned long)__init_begin;
ce8420bb
HD		 479
480 reservedpages = 0;
53faf291
KM		 481{
482 unsigned long pfn;
483#ifdef CONFIG_DISCONTIGMEM
484 int i;
485
486 for (i = 0; i < npmem_ranges; i++) {
487 for (pfn = node_start_pfn(i); pfn < node_end_pfn(i); pfn++) {
488 if (PageReserved(pfn_to_page(pfn)))
489 reservedpages++;
490 }
491 }
492#else /* !CONFIG_DISCONTIGMEM */
493 for (pfn = 0; pfn < max_pfn; pfn++) {
ce8420bb
HD		 494 /*
495 * Only count reserved RAM pages
496 */
53faf291 497 if (PageReserved(pfn_to_page(pfn)))
ce8420bb 498 reservedpages++;
53faf291
KM		 499 }
500#endif
501}
1da177e4
LT		 502
503#ifdef CONFIG_PA11
504 if (hppa_dma_ops == &pcxl_dma_ops) {
505 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
506 vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start + PCXL_DMA_MAP_SIZE);
507 } else {
508 pcxl_dma_start = 0;
509 vmalloc_start = SET_MAP_OFFSET(MAP_START);
510 }
511#else
512 vmalloc_start = SET_MAP_OFFSET(MAP_START);
513#endif
514
53faf291
KM		 515 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init)\n",
516 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
ce8420bb
HD		 517 num_physpages << (PAGE_SHIFT-10),
518 codesize >> 10,
519 reservedpages << (PAGE_SHIFT-10),
520 datasize >> 10,
53faf291
KM		 521 initsize >> 10
522 );
ce8420bb
HD		 523
524#ifdef CONFIG_DEBUG_KERNEL /* double-sanity-check paranoia */
525 printk("virtual kernel memory layout:\n"
526 " vmalloc : 0x%p - 0x%p (%4ld MB)\n"
53faf291 527 " memory : 0x%p - 0x%p (%4ld MB)\n"
ce8420bb
HD		 528 " .init : 0x%p - 0x%p (%4ld kB)\n"
529 " .data : 0x%p - 0x%p (%4ld kB)\n"
530 " .text : 0x%p - 0x%p (%4ld kB)\n",
531
532 (void*)VMALLOC_START, (void*)VMALLOC_END,
533 (VMALLOC_END - VMALLOC_START) >> 20,
534
535 __va(0), high_memory,
536 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
537
53faf291
KM		 538 __init_begin, __init_end,
539 ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
ce8420bb 540
53faf291
KM		 541 _etext, _edata,
542 ((unsigned long)_edata - (unsigned long)_etext) >> 10,
ce8420bb 543
53faf291
KM		 544 _text, _etext,
545 ((unsigned long)_etext - (unsigned long)_text) >> 10);
ce8420bb 546#endif
1da177e4
LT		 547}
548
8039de10 549unsigned long *empty_zero_page __read_mostly;
1da177e4
LT		 550
551void show_mem(void)
552{
553 int i,free = 0,total = 0,reserved = 0;
554 int shared = 0, cached = 0;
555
556 printk(KERN_INFO "Mem-info:\n");
557 show_free_areas();
558 printk(KERN_INFO "Free swap: %6ldkB\n",
559 nr_swap_pages<<(PAGE_SHIFT-10));
560#ifndef CONFIG_DISCONTIGMEM
561 i = max_mapnr;
562 while (i-- > 0) {
563 total++;
564 if (PageReserved(mem_map+i))
565 reserved++;
566 else if (PageSwapCache(mem_map+i))
567 cached++;
568 else if (!page_count(&mem_map[i]))
569 free++;
570 else
571 shared += page_count(&mem_map[i]) - 1;
572 }
573#else
574 for (i = 0; i < npmem_ranges; i++) {
575 int j;
576
577 for (j = node_start_pfn(i); j < node_end_pfn(i); j++) {
578 struct page *p;
208d54e5 579 unsigned long flags;
1da177e4 580
208d54e5 581 pgdat_resize_lock(NODE_DATA(i), &flags);
408fde81 582 p = nid_page_nr(i, j) - node_start_pfn(i);
1da177e4
LT		 583
584 total++;
585 if (PageReserved(p))
586 reserved++;
587 else if (PageSwapCache(p))
588 cached++;
589 else if (!page_count(p))
590 free++;
591 else
592 shared += page_count(p) - 1;
208d54e5 593 pgdat_resize_unlock(NODE_DATA(i), &flags);
1da177e4
LT		 594 }
595 }
596#endif
597 printk(KERN_INFO "%d pages of RAM\n", total);
598 printk(KERN_INFO "%d reserved pages\n", reserved);
599 printk(KERN_INFO "%d pages shared\n", shared);
600 printk(KERN_INFO "%d pages swap cached\n", cached);
601
602
603#ifdef CONFIG_DISCONTIGMEM
604 {
605 struct zonelist *zl;
54a6eb5c 606 int i, j;
1da177e4
LT		 607
608 for (i = 0; i < npmem_ranges; i++) {
54a6eb5c 609 zl = node_zonelist(i);
1da177e4 610 for (j = 0; j < MAX_NR_ZONES; j++) {
54a6eb5c
MG		 611 struct zone **z;
612 struct zone *zone;
1da177e4
LT		 613
614 printk("Zone list for zone %d on node %d: ", j, i);
54a6eb5c
MG		 615 for_each_zone_zonelist(zone, z, zl, j)
616 printk("[%d/%s] ", zone_to_nid(zone),
617 zone->name);
1da177e4
LT		 618 printk("\n");
619 }
620 }
621 }
622#endif
623}
624
625
626static void __init map_pages(unsigned long start_vaddr, unsigned long start_paddr, unsigned long size, pgprot_t pgprot)
627{
628 pgd_t *pg_dir;
629 pmd_t *pmd;
630 pte_t *pg_table;
631 unsigned long end_paddr;
632 unsigned long start_pmd;
633 unsigned long start_pte;
634 unsigned long tmp1;
635 unsigned long tmp2;
636 unsigned long address;
637 unsigned long ro_start;
638 unsigned long ro_end;
639 unsigned long fv_addr;
640 unsigned long gw_addr;
641 extern const unsigned long fault_vector_20;
642 extern void * const linux_gateway_page;
643
c51d476a 644 ro_start = __pa((unsigned long)_text);
1da177e4
LT		 645 ro_end = __pa((unsigned long)&data_start);
646 fv_addr = __pa((unsigned long)&fault_vector_20) & PAGE_MASK;
647 gw_addr = __pa((unsigned long)&linux_gateway_page) & PAGE_MASK;
648
649 end_paddr = start_paddr + size;
650
651 pg_dir = pgd_offset_k(start_vaddr);
652
653#if PTRS_PER_PMD == 1
654 start_pmd = 0;
655#else
656 start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
657#endif
658 start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
659
660 address = start_paddr;
661 while (address < end_paddr) {
662#if PTRS_PER_PMD == 1
663 pmd = (pmd_t *)__pa(pg_dir);
664#else
665 pmd = (pmd_t *)pgd_address(*pg_dir);
666
667 /*
668 * pmd is physical at this point
669 */
670
671 if (!pmd) {
672 pmd = (pmd_t *) alloc_bootmem_low_pages_node(NODE_DATA(0),PAGE_SIZE << PMD_ORDER);
673 pmd = (pmd_t *) __pa(pmd);
674 }
675
676 pgd_populate(NULL, pg_dir, __va(pmd));
677#endif
678 pg_dir++;
679
680 /* now change pmd to kernel virtual addresses */
681
682 pmd = (pmd_t *)__va(pmd) + start_pmd;
683 for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++,pmd++) {
684
685 /*
686 * pg_table is physical at this point
687 */
688
689 pg_table = (pte_t *)pmd_address(*pmd);
690 if (!pg_table) {
691 pg_table = (pte_t *)
692 alloc_bootmem_low_pages_node(NODE_DATA(0),PAGE_SIZE);
693 pg_table = (pte_t *) __pa(pg_table);
694 }
695
696 pmd_populate_kernel(NULL, pmd, __va(pg_table));
697
698 /* now change pg_table to kernel virtual addresses */
699
700 pg_table = (pte_t *) __va(pg_table) + start_pte;
701 for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++,pg_table++) {
702 pte_t pte;
703
704 /*
705 * Map the fault vector writable so we can
706 * write the HPMC checksum.
707 */
2fd83038 708#if defined(CONFIG_PARISC_PAGE_SIZE_4KB)
1da177e4
LT		 709 if (address >= ro_start && address < ro_end
710 && address != fv_addr
711 && address != gw_addr)
712 pte = __mk_pte(address, PAGE_KERNEL_RO);
713 else
2fd83038 714#endif
1da177e4
LT		 715 pte = __mk_pte(address, pgprot);
716
717 if (address >= end_paddr)
718 pte_val(pte) = 0;
719
720 set_pte(pg_table, pte);
721
722 address += PAGE_SIZE;
723 }
724 start_pte = 0;
725
726 if (address >= end_paddr)
727 break;
728 }
729 start_pmd = 0;
730 }
731}
732
733/*
734 * pagetable_init() sets up the page tables
735 *
736 * Note that gateway_init() places the Linux gateway page at page 0.
737 * Since gateway pages cannot be dereferenced this has the desirable
738 * side effect of trapping those pesky NULL-reference errors in the
739 * kernel.
740 */
741static void __init pagetable_init(void)
742{
743 int range;
744
745 /* Map each physical memory range to its kernel vaddr */
746
747 for (range = 0; range < npmem_ranges; range++) {
748 unsigned long start_paddr;
749 unsigned long end_paddr;
750 unsigned long size;
751
752 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
753 end_paddr = start_paddr + (pmem_ranges[range].pages << PAGE_SHIFT);
754 size = pmem_ranges[range].pages << PAGE_SHIFT;
755
756 map_pages((unsigned long)__va(start_paddr), start_paddr,
757 size, PAGE_KERNEL);
758 }
759
760#ifdef CONFIG_BLK_DEV_INITRD
761 if (initrd_end && initrd_end > mem_limit) {
1bcdd854 762 printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
1da177e4
LT		 763 map_pages(initrd_start, __pa(initrd_start),
764 initrd_end - initrd_start, PAGE_KERNEL);
765 }
766#endif
767
768 empty_zero_page = alloc_bootmem_pages(PAGE_SIZE);
769 memset(empty_zero_page, 0, PAGE_SIZE);
770}
771
772static void __init gateway_init(void)
773{
774 unsigned long linux_gateway_page_addr;
775 /* FIXME: This is 'const' in order to trick the compiler
776 into not treating it as DP-relative data. */
777 extern void * const linux_gateway_page;
778
779 linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
780
781 /*
782 * Setup Linux Gateway page.
783 *
784 * The Linux gateway page will reside in kernel space (on virtual
785 * page 0), so it doesn't need to be aliased into user space.
786 */
787
788 map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
789 PAGE_SIZE, PAGE_GATEWAY);
790}
791
792#ifdef CONFIG_HPUX
793void
794map_hpux_gateway_page(struct task_struct *tsk, struct mm_struct *mm)
795{
796 pgd_t *pg_dir;
797 pmd_t *pmd;
798 pte_t *pg_table;
799 unsigned long start_pmd;
800 unsigned long start_pte;
801 unsigned long address;
802 unsigned long hpux_gw_page_addr;
803 /* FIXME: This is 'const' in order to trick the compiler
804 into not treating it as DP-relative data. */
805 extern void * const hpux_gateway_page;
806
807 hpux_gw_page_addr = HPUX_GATEWAY_ADDR & PAGE_MASK;
808
809 /*
810 * Setup HP-UX Gateway page.
811 *
812 * The HP-UX gateway page resides in the user address space,
813 * so it needs to be aliased into each process.
814 */
815
816 pg_dir = pgd_offset(mm,hpux_gw_page_addr);
817
818#if PTRS_PER_PMD == 1
819 start_pmd = 0;
820#else
821 start_pmd = ((hpux_gw_page_addr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
822#endif
823 start_pte = ((hpux_gw_page_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
824
825 address = __pa(&hpux_gateway_page);
826#if PTRS_PER_PMD == 1
827 pmd = (pmd_t *)__pa(pg_dir);
828#else
829 pmd = (pmd_t *) pgd_address(*pg_dir);
830
831 /*
832 * pmd is physical at this point
833 */
834
835 if (!pmd) {
836 pmd = (pmd_t *) get_zeroed_page(GFP_KERNEL);
837 pmd = (pmd_t *) __pa(pmd);
838 }
839
840 __pgd_val_set(*pg_dir, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pmd);
841#endif
842 /* now change pmd to kernel virtual addresses */
843
844 pmd = (pmd_t *)__va(pmd) + start_pmd;
845
846 /*
847 * pg_table is physical at this point
848 */
849
850 pg_table = (pte_t *) pmd_address(*pmd);
851 if (!pg_table)
852 pg_table = (pte_t *) __pa(get_zeroed_page(GFP_KERNEL));
853
854 __pmd_val_set(*pmd, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pg_table);
855
856 /* now change pg_table to kernel virtual addresses */
857
858 pg_table = (pte_t *) __va(pg_table) + start_pte;
859 set_pte(pg_table, __mk_pte(address, PAGE_GATEWAY));
860}
861EXPORT_SYMBOL(map_hpux_gateway_page);
862#endif
863
1da177e4
LT		 864void __init paging_init(void)
865{
866 int i;
867
868 setup_bootmem();
869 pagetable_init();
870 gateway_init();
871 flush_cache_all_local(); /* start with known state */
ce33941f 872 flush_tlb_all_local(NULL);
1da177e4
LT		 873
874 for (i = 0; i < npmem_ranges; i++) {
f06a9684 875 unsigned long zones_size[MAX_NR_ZONES] = { 0, };
1da177e4 876
00592837 877 zones_size[ZONE_NORMAL] = pmem_ranges[i].pages;
1da177e4
LT		 878
879#ifdef CONFIG_DISCONTIGMEM
880 /* Need to initialize the pfnnid_map before we can initialize
881 the zone */
882 {
883 int j;
884 for (j = (pmem_ranges[i].start_pfn >> PFNNID_SHIFT);
885 j <= ((pmem_ranges[i].start_pfn + pmem_ranges[i].pages) >> PFNNID_SHIFT);
886 j++) {
887 pfnnid_map[j] = i;
888 }
889 }
890#endif
891
892 free_area_init_node(i, NODE_DATA(i), zones_size,
893 pmem_ranges[i].start_pfn, NULL);
894 }
895}
896
897#ifdef CONFIG_PA20
898
899/*
7022672e 900 * Currently, all PA20 chips have 18 bit protection IDs, which is the
1da177e4
LT		 901 * limiting factor (space ids are 32 bits).
902 */
903
904#define NR_SPACE_IDS 262144
905
906#else
907
908/*
7022672e
SA		 909 * Currently we have a one-to-one relationship between space IDs and
910 * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
911 * support 15 bit protection IDs, so that is the limiting factor.
912 * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
1da177e4
LT		 913 * probably not worth the effort for a special case here.
914 */
915
916#define NR_SPACE_IDS 32768
917
918#endif /* !CONFIG_PA20 */
919
920#define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
921#define SID_ARRAY_SIZE (NR_SPACE_IDS / (8 * sizeof(long)))
922
923static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
924static unsigned long dirty_space_id[SID_ARRAY_SIZE];
925static unsigned long space_id_index;
926static unsigned long free_space_ids = NR_SPACE_IDS - 1;
927static unsigned long dirty_space_ids = 0;
928
929static DEFINE_SPINLOCK(sid_lock);
930
931unsigned long alloc_sid(void)
932{
933 unsigned long index;
934
935 spin_lock(&sid_lock);
936
937 if (free_space_ids == 0) {
938 if (dirty_space_ids != 0) {
939 spin_unlock(&sid_lock);
940 flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
941 spin_lock(&sid_lock);
942 }
2fd83038 943 BUG_ON(free_space_ids == 0);
1da177e4
LT		 944 }
945
946 free_space_ids--;
947
948 index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
949 space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1)));
950 space_id_index = index;
951
952 spin_unlock(&sid_lock);
953
954 return index << SPACEID_SHIFT;
955}
956
957void free_sid(unsigned long spaceid)
958{
959 unsigned long index = spaceid >> SPACEID_SHIFT;
960 unsigned long *dirty_space_offset;
961
962 dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG);
963 index &= (BITS_PER_LONG - 1);
964
965 spin_lock(&sid_lock);
966
2fd83038 967 BUG_ON(*dirty_space_offset & (1L << index)); /* attempt to free space id twice */
1da177e4
LT		 968
969 *dirty_space_offset |= (1L << index);
970 dirty_space_ids++;
971
972 spin_unlock(&sid_lock);
973}
974
975
976#ifdef CONFIG_SMP
977static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
978{
979 int i;
980
981 /* NOTE: sid_lock must be held upon entry */
982
983 *ndirtyptr = dirty_space_ids;
984 if (dirty_space_ids != 0) {
985 for (i = 0; i < SID_ARRAY_SIZE; i++) {
986 dirty_array[i] = dirty_space_id[i];
987 dirty_space_id[i] = 0;
988 }
989 dirty_space_ids = 0;
990 }
991
992 return;
993}
994
995static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
996{
997 int i;
998
999 /* NOTE: sid_lock must be held upon entry */
1000
1001 if (ndirty != 0) {
1002 for (i = 0; i < SID_ARRAY_SIZE; i++) {
1003 space_id[i] ^= dirty_array[i];
1004 }
1005
1006 free_space_ids += ndirty;
1007 space_id_index = 0;
1008 }
1009}
1010
1011#else /* CONFIG_SMP */
1012
1013static void recycle_sids(void)
1014{
1015 int i;
1016
1017 /* NOTE: sid_lock must be held upon entry */
1018
1019 if (dirty_space_ids != 0) {
1020 for (i = 0; i < SID_ARRAY_SIZE; i++) {
1021 space_id[i] ^= dirty_space_id[i];
1022 dirty_space_id[i] = 0;
1023 }
1024
1025 free_space_ids += dirty_space_ids;
1026 dirty_space_ids = 0;
1027 space_id_index = 0;
1028 }
1029}
1030#endif
1031
1032/*
1033 * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
1034 * purged, we can safely reuse the space ids that were released but
1035 * not flushed from the tlb.
1036 */
1037
1038#ifdef CONFIG_SMP
1039
1040static unsigned long recycle_ndirty;
1041static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
2fd83038 1042static unsigned int recycle_inuse;
1da177e4
LT		 1043
1044void flush_tlb_all(void)
1045{
1046 int do_recycle;
1047
1048 do_recycle = 0;
1049 spin_lock(&sid_lock);
1050 if (dirty_space_ids > RECYCLE_THRESHOLD) {
2fd83038 1051 BUG_ON(recycle_inuse); /* FIXME: Use a semaphore/wait queue here */
1da177e4
LT		 1052 get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
1053 recycle_inuse++;
1054 do_recycle++;
1055 }
1056 spin_unlock(&sid_lock);
ce33941f 1057 on_each_cpu(flush_tlb_all_local, NULL, 1, 1);
1da177e4
LT		 1058 if (do_recycle) {
1059 spin_lock(&sid_lock);
1060 recycle_sids(recycle_ndirty,recycle_dirty_array);
1061 recycle_inuse = 0;
1062 spin_unlock(&sid_lock);
1063 }
1064}
1065#else
1066void flush_tlb_all(void)
1067{
1068 spin_lock(&sid_lock);
1b2425e3 1069 flush_tlb_all_local(NULL);
1da177e4
LT		 1070 recycle_sids();
1071 spin_unlock(&sid_lock);
1072}
1073#endif
1074
1075#ifdef CONFIG_BLK_DEV_INITRD
1076void free_initrd_mem(unsigned long start, unsigned long end)
1077{
94c3e87a
HD		 1078 if (start >= end)
1079 return;
1080 printk(KERN_INFO "Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
1da177e4
LT		 1081 for (; start < end; start += PAGE_SIZE) {
1082 ClearPageReserved(virt_to_page(start));
7835e98b 1083 init_page_count(virt_to_page(start));
1da177e4
LT		 1084 free_page(start);
1085 num_physpages++;
1086 totalram_pages++;
1087 }
1da177e4
LT		 1088}
1089#endif
Linux 6.x block layer and io_uring tree(s)