projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
mm/MIPS: prepare for killing free_all_bootmem_node()
[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>
5a0e3ad6 17#include <linux/gfp.h>
1da177e4
LT		 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>
ce8420bb 28#include <asm/pgtable.h>
1da177e4
LT		 29#include <asm/tlb.h>
30#include <asm/pdc_chassis.h>
31#include <asm/mmzone.h>
a581c2a4 32#include <asm/sections.h>
1da177e4 33
1da177e4 34extern int data_start;
1da177e4 35
c39f52a9
TG		 36#if PT_NLEVELS == 3
37/* NOTE: This layout exactly conforms to the hybrid L2/L3 page table layout
38 * with the first pmd adjacent to the pgd and below it. gcc doesn't actually
39 * guarantee that global objects will be laid out in memory in the same order
40 * as the order of declaration, so put these in different sections and use
41 * the linker script to order them. */
42pmd_t pmd0[PTRS_PER_PMD] __attribute__ ((__section__ (".data..vm0.pmd"), aligned(PAGE_SIZE)));
43#endif
44
45pgd_t swapper_pg_dir[PTRS_PER_PGD] __attribute__ ((__section__ (".data..vm0.pgd"), aligned(PAGE_SIZE)));
46pte_t pg0[PT_INITIAL * PTRS_PER_PTE] __attribute__ ((__section__ (".data..vm0.pte"), aligned(PAGE_SIZE)));
47
1da177e4 48#ifdef CONFIG_DISCONTIGMEM
8039de10 49struct node_map_data node_data[MAX_NUMNODES] __read_mostly;
91ea8207 50signed char pfnnid_map[PFNNID_MAP_MAX] __read_mostly;
1da177e4
LT		 51#endif
52
53static struct resource data_resource = {
54 .name = "Kernel data",
55 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
56};
57
58static struct resource code_resource = {
59 .name = "Kernel code",
60 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
61};
62
63static struct resource pdcdata_resource = {
64 .name = "PDC data (Page Zero)",
65 .start = 0,
66 .end = 0x9ff,
67 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
68};
69
8039de10 70static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __read_mostly;
1da177e4
LT		 71
72/* The following array is initialized from the firmware specific
73 * information retrieved in kernel/inventory.c.
74 */
75
8039de10
HD		 76physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __read_mostly;
77int npmem_ranges __read_mostly;
1da177e4 78
a8f44e38 79#ifdef CONFIG_64BIT
1da177e4 80#define MAX_MEM (~0UL)
a8f44e38 81#else /* !CONFIG_64BIT */
1da177e4 82#define MAX_MEM (3584U*1024U*1024U)
a8f44e38 83#endif /* !CONFIG_64BIT */
1da177e4 84
8039de10 85static unsigned long mem_limit __read_mostly = MAX_MEM;
1da177e4
LT		 86
87static void __init mem_limit_func(void)
88{
89 char *cp, *end;
90 unsigned long limit;
1da177e4
LT		 91
92 /* We need this before __setup() functions are called */
93
94 limit = MAX_MEM;
668f9931 95 for (cp = boot_command_line; *cp; ) {
1da177e4
LT		 96 if (memcmp(cp, "mem=", 4) == 0) {
97 cp += 4;
98 limit = memparse(cp, &end);
99 if (end != cp)
100 break;
101 cp = end;
102 } else {
103 while (*cp != ' ' && *cp)
104 ++cp;
105 while (*cp == ' ')
106 ++cp;
107 }
108 }
109
110 if (limit < mem_limit)
111 mem_limit = limit;
112}
113
114#define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
115
116static void __init setup_bootmem(void)
117{
118 unsigned long bootmap_size;
119 unsigned long mem_max;
120 unsigned long bootmap_pages;
121 unsigned long bootmap_start_pfn;
122 unsigned long bootmap_pfn;
123#ifndef CONFIG_DISCONTIGMEM
124 physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
125 int npmem_holes;
126#endif
127 int i, sysram_resource_count;
128
129 disable_sr_hashing(); /* Turn off space register hashing */
130
131 /*
132 * Sort the ranges. Since the number of ranges is typically
133 * small, and performance is not an issue here, just do
134 * a simple insertion sort.
135 */
136
137 for (i = 1; i < npmem_ranges; i++) {
138 int j;
139
140 for (j = i; j > 0; j--) {
141 unsigned long tmp;
142
143 if (pmem_ranges[j-1].start_pfn <
144 pmem_ranges[j].start_pfn) {
145
146 break;
147 }
148 tmp = pmem_ranges[j-1].start_pfn;
149 pmem_ranges[j-1].start_pfn = pmem_ranges[j].start_pfn;
150 pmem_ranges[j].start_pfn = tmp;
151 tmp = pmem_ranges[j-1].pages;
152 pmem_ranges[j-1].pages = pmem_ranges[j].pages;
153 pmem_ranges[j].pages = tmp;
154 }
155 }
156
157#ifndef CONFIG_DISCONTIGMEM
158 /*
159 * Throw out ranges that are too far apart (controlled by
160 * MAX_GAP).
161 */
162
163 for (i = 1; i < npmem_ranges; i++) {
164 if (pmem_ranges[i].start_pfn -
165 (pmem_ranges[i-1].start_pfn +
166 pmem_ranges[i-1].pages) > MAX_GAP) {
167 npmem_ranges = i;
168 printk("Large gap in memory detected (%ld pages). "
169 "Consider turning on CONFIG_DISCONTIGMEM\n",
170 pmem_ranges[i].start_pfn -
171 (pmem_ranges[i-1].start_pfn +
172 pmem_ranges[i-1].pages));
173 break;
174 }
175 }
176#endif
177
178 if (npmem_ranges > 1) {
179
180 /* Print the memory ranges */
181
182 printk(KERN_INFO "Memory Ranges:\n");
183
184 for (i = 0; i < npmem_ranges; i++) {
185 unsigned long start;
186 unsigned long size;
187
188 size = (pmem_ranges[i].pages << PAGE_SHIFT);
189 start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
190 printk(KERN_INFO "%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
191 i,start, start + (size - 1), size >> 20);
192 }
193 }
194
195 sysram_resource_count = npmem_ranges;
196 for (i = 0; i < sysram_resource_count; i++) {
197 struct resource *res = &sysram_resources[i];
198 res->name = "System RAM";
199 res->start = pmem_ranges[i].start_pfn << PAGE_SHIFT;
200 res->end = res->start + (pmem_ranges[i].pages << PAGE_SHIFT)-1;
201 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
202 request_resource(&iomem_resource, res);
203 }
204
205 /*
206 * For 32 bit kernels we limit the amount of memory we can
207 * support, in order to preserve enough kernel address space
208 * for other purposes. For 64 bit kernels we don't normally
209 * limit the memory, but this mechanism can be used to
210 * artificially limit the amount of memory (and it is written
211 * to work with multiple memory ranges).
212 */
213
214 mem_limit_func(); /* check for "mem=" argument */
215
216 mem_max = 0;
1da177e4
LT		 217 for (i = 0; i < npmem_ranges; i++) {
218 unsigned long rsize;
219
220 rsize = pmem_ranges[i].pages << PAGE_SHIFT;
221 if ((mem_max + rsize) > mem_limit) {
222 printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
223 if (mem_max == mem_limit)
224 npmem_ranges = i;
225 else {
226 pmem_ranges[i].pages = (mem_limit >> PAGE_SHIFT)
227 - (mem_max >> PAGE_SHIFT);
228 npmem_ranges = i + 1;
229 mem_max = mem_limit;
230 }
1da177e4
LT		 231 break;
232 }
1da177e4
LT		 233 mem_max += rsize;
234 }
235
236 printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
237
238#ifndef CONFIG_DISCONTIGMEM
239 /* Merge the ranges, keeping track of the holes */
240
241 {
242 unsigned long end_pfn;
243 unsigned long hole_pages;
244
245 npmem_holes = 0;
246 end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
247 for (i = 1; i < npmem_ranges; i++) {
248
249 hole_pages = pmem_ranges[i].start_pfn - end_pfn;
250 if (hole_pages) {
251 pmem_holes[npmem_holes].start_pfn = end_pfn;
252 pmem_holes[npmem_holes++].pages = hole_pages;
253 end_pfn += hole_pages;
254 }
255 end_pfn += pmem_ranges[i].pages;
256 }
257
258 pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
259 npmem_ranges = 1;
260 }
261#endif
262
263 bootmap_pages = 0;
264 for (i = 0; i < npmem_ranges; i++)
265 bootmap_pages += bootmem_bootmap_pages(pmem_ranges[i].pages);
266
267 bootmap_start_pfn = PAGE_ALIGN(__pa((unsigned long) &_end)) >> PAGE_SHIFT;
268
269#ifdef CONFIG_DISCONTIGMEM
270 for (i = 0; i < MAX_PHYSMEM_RANGES; i++) {
271 memset(NODE_DATA(i), 0, sizeof(pg_data_t));
b61bfa3c 272 NODE_DATA(i)->bdata = &bootmem_node_data[i];
1da177e4
LT		 273 }
274 memset(pfnnid_map, 0xff, sizeof(pfnnid_map));
275
d9b41e0b
DR		 276 for (i = 0; i < npmem_ranges; i++) {
277 node_set_state(i, N_NORMAL_MEMORY);
1da177e4 278 node_set_online(i);
d9b41e0b 279 }
1da177e4
LT		 280#endif
281
282 /*
283 * Initialize and free the full range of memory in each range.
284 * Note that the only writing these routines do are to the bootmap,
285 * and we've made sure to locate the bootmap properly so that they
286 * won't be writing over anything important.
287 */
288
289 bootmap_pfn = bootmap_start_pfn;
290 max_pfn = 0;
291 for (i = 0; i < npmem_ranges; i++) {
292 unsigned long start_pfn;
293 unsigned long npages;
294
295 start_pfn = pmem_ranges[i].start_pfn;
296 npages = pmem_ranges[i].pages;
297
298 bootmap_size = init_bootmem_node(NODE_DATA(i),
299 bootmap_pfn,
300 start_pfn,
301 (start_pfn + npages) );
302 free_bootmem_node(NODE_DATA(i),
303 (start_pfn << PAGE_SHIFT),
304 (npages << PAGE_SHIFT) );
305 bootmap_pfn += (bootmap_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
306 if ((start_pfn + npages) > max_pfn)
307 max_pfn = start_pfn + npages;
308 }
309
5cdb8205
GG		 310 /* IOMMU is always used to access "high mem" on those boxes
311 * that can support enough mem that a PCI device couldn't
312 * directly DMA to any physical addresses.
313 * ISA DMA support will need to revisit this.
314 */
315 max_low_pfn = max_pfn;
316
8980a7ba
HD		 317 /* bootmap sizing messed up? */
318 BUG_ON((bootmap_pfn - bootmap_start_pfn) != bootmap_pages);
1da177e4
LT		 319
320 /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
321
322#define PDC_CONSOLE_IO_IODC_SIZE 32768
323
324 reserve_bootmem_node(NODE_DATA(0), 0UL,
72a7fe39
BW		 325 (unsigned long)(PAGE0->mem_free +
326 PDC_CONSOLE_IO_IODC_SIZE), BOOTMEM_DEFAULT);
c51d476a 327 reserve_bootmem_node(NODE_DATA(0), __pa((unsigned long)_text),
72a7fe39 328 (unsigned long)(_end - _text), BOOTMEM_DEFAULT);
1da177e4 329 reserve_bootmem_node(NODE_DATA(0), (bootmap_start_pfn << PAGE_SHIFT),
72a7fe39
BW		 330 ((bootmap_pfn - bootmap_start_pfn) << PAGE_SHIFT),
331 BOOTMEM_DEFAULT);
1da177e4
LT		 332
333#ifndef CONFIG_DISCONTIGMEM
334
335 /* reserve the holes */
336
337 for (i = 0; i < npmem_holes; i++) {
338 reserve_bootmem_node(NODE_DATA(0),
339 (pmem_holes[i].start_pfn << PAGE_SHIFT),
72a7fe39
BW		 340 (pmem_holes[i].pages << PAGE_SHIFT),
341 BOOTMEM_DEFAULT);
1da177e4
LT		 342 }
343#endif
344
345#ifdef CONFIG_BLK_DEV_INITRD
346 if (initrd_start) {
347 printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
348 if (__pa(initrd_start) < mem_max) {
349 unsigned long initrd_reserve;
350
351 if (__pa(initrd_end) > mem_max) {
352 initrd_reserve = mem_max - __pa(initrd_start);
353 } else {
354 initrd_reserve = initrd_end - initrd_start;
355 }
356 initrd_below_start_ok = 1;
357 printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
358
72a7fe39
BW		 359 reserve_bootmem_node(NODE_DATA(0), __pa(initrd_start),
360 initrd_reserve, BOOTMEM_DEFAULT);
1da177e4
LT		 361 }
362 }
363#endif
364
365 data_resource.start = virt_to_phys(&data_start);
c51d476a
KM		 366 data_resource.end = virt_to_phys(_end) - 1;
367 code_resource.start = virt_to_phys(_text);
1da177e4
LT		 368 code_resource.end = virt_to_phys(&data_start)-1;
369
370 /* We don't know which region the kernel will be in, so try
371 * all of them.
372 */
373 for (i = 0; i < sysram_resource_count; i++) {
374 struct resource *res = &sysram_resources[i];
375 request_resource(res, &code_resource);
376 request_resource(res, &data_resource);
377 }
378 request_resource(&sysram_resources[0], &pdcdata_resource);
379}
380
d7dd2ff1
JB		 381static void __init map_pages(unsigned long start_vaddr,
382 unsigned long start_paddr, unsigned long size,
383 pgprot_t pgprot, int force)
384{
385 pgd_t *pg_dir;
386 pmd_t *pmd;
387 pte_t *pg_table;
388 unsigned long end_paddr;
389 unsigned long start_pmd;
390 unsigned long start_pte;
391 unsigned long tmp1;
392 unsigned long tmp2;
393 unsigned long address;
394 unsigned long vaddr;
395 unsigned long ro_start;
396 unsigned long ro_end;
397 unsigned long fv_addr;
398 unsigned long gw_addr;
399 extern const unsigned long fault_vector_20;
400 extern void * const linux_gateway_page;
401
402 ro_start = __pa((unsigned long)_text);
403 ro_end = __pa((unsigned long)&data_start);
404 fv_addr = __pa((unsigned long)&fault_vector_20) & PAGE_MASK;
405 gw_addr = __pa((unsigned long)&linux_gateway_page) & PAGE_MASK;
406
407 end_paddr = start_paddr + size;
408
409 pg_dir = pgd_offset_k(start_vaddr);
410
411#if PTRS_PER_PMD == 1
412 start_pmd = 0;
413#else
414 start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
415#endif
416 start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
417
418 address = start_paddr;
419 vaddr = start_vaddr;
420 while (address < end_paddr) {
421#if PTRS_PER_PMD == 1
422 pmd = (pmd_t *)__pa(pg_dir);
423#else
424 pmd = (pmd_t *)pgd_address(*pg_dir);
425
426 /*
427 * pmd is physical at this point
428 */
429
430 if (!pmd) {
431 pmd = (pmd_t *) alloc_bootmem_low_pages_node(NODE_DATA(0), PAGE_SIZE << PMD_ORDER);
432 pmd = (pmd_t *) __pa(pmd);
433 }
434
435 pgd_populate(NULL, pg_dir, __va(pmd));
436#endif
437 pg_dir++;
438
439 /* now change pmd to kernel virtual addresses */
440
441 pmd = (pmd_t *)__va(pmd) + start_pmd;
442 for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) {
443
444 /*
445 * pg_table is physical at this point
446 */
447
448 pg_table = (pte_t *)pmd_address(*pmd);
449 if (!pg_table) {
450 pg_table = (pte_t *)
451 alloc_bootmem_low_pages_node(NODE_DATA(0), PAGE_SIZE);
452 pg_table = (pte_t *) __pa(pg_table);
453 }
454
455 pmd_populate_kernel(NULL, pmd, __va(pg_table));
456
457 /* now change pg_table to kernel virtual addresses */
458
459 pg_table = (pte_t *) __va(pg_table) + start_pte;
460 for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) {
461 pte_t pte;
462
463 /*
464 * Map the fault vector writable so we can
465 * write the HPMC checksum.
466 */
467 if (force)
468 pte = __mk_pte(address, pgprot);
469 else if (core_kernel_text(vaddr) &&
470 address != fv_addr)
471 pte = __mk_pte(address, PAGE_KERNEL_EXEC);
472 else
473#if defined(CONFIG_PARISC_PAGE_SIZE_4KB)
474 if (address >= ro_start && address < ro_end
475 && address != fv_addr
476 && address != gw_addr)
477 pte = __mk_pte(address, PAGE_KERNEL_RO);
478 else
479#endif
480 pte = __mk_pte(address, pgprot);
481
482 if (address >= end_paddr) {
483 if (force)
484 break;
485 else
486 pte_val(pte) = 0;
487 }
488
489 set_pte(pg_table, pte);
490
491 address += PAGE_SIZE;
492 vaddr += PAGE_SIZE;
493 }
494 start_pte = 0;
495
496 if (address >= end_paddr)
497 break;
498 }
499 start_pmd = 0;
500 }
501}
502
1da177e4
LT		 503void free_initmem(void)
504{
4fb11781
KM		 505 unsigned long init_begin = (unsigned long)__init_begin;
506 unsigned long init_end = (unsigned long)__init_end;
1da177e4 507
d7dd2ff1
JB		 508 /* The init text pages are marked R-X. We have to
509 * flush the icache and mark them RW-
510 *
511 * This is tricky, because map_pages is in the init section.
512 * Do a dummy remap of the data section first (the data
513 * section is already PAGE_KERNEL) to pull in the TLB entries
514 * for map_kernel */
515 map_pages(init_begin, __pa(init_begin), init_end - init_begin,
516 PAGE_KERNEL_RWX, 1);
517 /* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
518 * map_pages */
519 map_pages(init_begin, __pa(init_begin), init_end - init_begin,
520 PAGE_KERNEL, 1);
521
522 /* force the kernel to see the new TLB entries */
523 __flush_tlb_range(0, init_begin, init_end);
1da177e4
LT		 524 /* Attempt to catch anyone trying to execute code here
525 * by filling the page with BRK insns.
1da177e4 526 */
20dbc9f7 527 memset((void *)init_begin, 0x00, init_end - init_begin);
d7dd2ff1
JB		 528 /* finally dump all the instructions which were cached, since the
529 * pages are no-longer executable */
4fb11781 530 flush_icache_range(init_begin, init_end);
1da177e4 531
7d2c7747 532 free_initmem_default(-1);
1da177e4
LT		 533
534 /* set up a new led state on systems shipped LED State panel */
535 pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
1da177e4
LT		 536}
537
1bcdd854
HD		 538
539#ifdef CONFIG_DEBUG_RODATA
540void mark_rodata_ro(void)
541{
1bcdd854
HD		 542 /* rodata memory was already mapped with KERNEL_RO access rights by
543 pagetable_init() and map_pages(). No need to do additional stuff here */
544 printk (KERN_INFO "Write protecting the kernel read-only data: %luk\n",
a581c2a4 545 (unsigned long)(__end_rodata - __start_rodata) >> 10);
1bcdd854
HD		 546}
547#endif
548
549
1da177e4
LT		 550/*
551 * Just an arbitrary offset to serve as a "hole" between mapping areas
552 * (between top of physical memory and a potential pcxl dma mapping
553 * area, and below the vmalloc mapping area).
554 *
555 * The current 32K value just means that there will be a 32K "hole"
556 * between mapping areas. That means that any out-of-bounds memory
557 * accesses will hopefully be caught. The vmalloc() routines leaves
558 * a hole of 4kB between each vmalloced area for the same reason.
559 */
560
561 /* Leave room for gateway page expansion */
562#if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
563#error KERNEL_MAP_START is in gateway reserved region
564#endif
565#define MAP_START (KERNEL_MAP_START)
566
567#define VM_MAP_OFFSET (32*1024)
568#define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
569 & ~(VM_MAP_OFFSET-1)))
570
4255f0d2
HD		 571void *parisc_vmalloc_start __read_mostly;
572EXPORT_SYMBOL(parisc_vmalloc_start);
1da177e4
LT		 573
574#ifdef CONFIG_PA11
8039de10 575unsigned long pcxl_dma_start __read_mostly;
1da177e4
LT		 576#endif
577
578void __init mem_init(void)
579{
48d27cb2
HD		 580 /* Do sanity checks on page table constants */
581 BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
582 BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
583 BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
584 BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
585 > BITS_PER_LONG);
586
1da177e4
LT		 587 high_memory = __va((max_pfn << PAGE_SHIFT));
588
589#ifndef CONFIG_DISCONTIGMEM
590 max_mapnr = page_to_pfn(virt_to_page(high_memory - 1)) + 1;
0c988534 591 free_all_bootmem();
1da177e4
LT		 592#else
593 {
594 int i;
595
596 for (i = 0; i < npmem_ranges; i++)
0c988534 597 free_all_bootmem_node(NODE_DATA(i));
1da177e4
LT		 598 }
599#endif
600
1da177e4
LT		 601#ifdef CONFIG_PA11
602 if (hppa_dma_ops == &pcxl_dma_ops) {
603 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
4255f0d2
HD		 604 parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
605 + PCXL_DMA_MAP_SIZE);
1da177e4
LT		 606 } else {
607 pcxl_dma_start = 0;
4255f0d2 608 parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
1da177e4
LT		 609 }
610#else
4255f0d2 611 parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
1da177e4
LT		 612#endif
613
7d2c7747 614 mem_init_print_info(NULL);
ce8420bb
HD		 615#ifdef CONFIG_DEBUG_KERNEL /* double-sanity-check paranoia */
616 printk("virtual kernel memory layout:\n"
617 " vmalloc : 0x%p - 0x%p (%4ld MB)\n"
53faf291 618 " memory : 0x%p - 0x%p (%4ld MB)\n"
ce8420bb
HD		 619 " .init : 0x%p - 0x%p (%4ld kB)\n"
620 " .data : 0x%p - 0x%p (%4ld kB)\n"
621 " .text : 0x%p - 0x%p (%4ld kB)\n",
622
623 (void*)VMALLOC_START, (void*)VMALLOC_END,
624 (VMALLOC_END - VMALLOC_START) >> 20,
625
626 __va(0), high_memory,
627 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
628
53faf291
KM		 629 __init_begin, __init_end,
630 ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
ce8420bb 631
53faf291
KM		 632 _etext, _edata,
633 ((unsigned long)_edata - (unsigned long)_etext) >> 10,
ce8420bb 634
53faf291
KM		 635 _text, _etext,
636 ((unsigned long)_etext - (unsigned long)_text) >> 10);
ce8420bb 637#endif
1da177e4
LT		 638}
639
8039de10 640unsigned long *empty_zero_page __read_mostly;
22febf1f 641EXPORT_SYMBOL(empty_zero_page);
1da177e4 642
b2b755b5 643void show_mem(unsigned int filter)
1da177e4
LT		 644{
645 int i,free = 0,total = 0,reserved = 0;
646 int shared = 0, cached = 0;
647
648 printk(KERN_INFO "Mem-info:\n");
7bf02ea2 649 show_free_areas(filter);
4b59e6c4
DR		 650 if (filter & SHOW_MEM_FILTER_PAGE_COUNT)
651 return;
1da177e4
LT		 652#ifndef CONFIG_DISCONTIGMEM
653 i = max_mapnr;
654 while (i-- > 0) {
655 total++;
656 if (PageReserved(mem_map+i))
657 reserved++;
658 else if (PageSwapCache(mem_map+i))
659 cached++;
660 else if (!page_count(&mem_map[i]))
661 free++;
662 else
663 shared += page_count(&mem_map[i]) - 1;
664 }
665#else
666 for (i = 0; i < npmem_ranges; i++) {
667 int j;
668
669 for (j = node_start_pfn(i); j < node_end_pfn(i); j++) {
670 struct page *p;
208d54e5 671 unsigned long flags;
1da177e4 672
208d54e5 673 pgdat_resize_lock(NODE_DATA(i), &flags);
408fde81 674 p = nid_page_nr(i, j) - node_start_pfn(i);
1da177e4
LT		 675
676 total++;
677 if (PageReserved(p))
678 reserved++;
679 else if (PageSwapCache(p))
680 cached++;
681 else if (!page_count(p))
682 free++;
683 else
684 shared += page_count(p) - 1;
208d54e5 685 pgdat_resize_unlock(NODE_DATA(i), &flags);
1da177e4
LT		 686 }
687 }
688#endif
689 printk(KERN_INFO "%d pages of RAM\n", total);
690 printk(KERN_INFO "%d reserved pages\n", reserved);
691 printk(KERN_INFO "%d pages shared\n", shared);
692 printk(KERN_INFO "%d pages swap cached\n", cached);
693
694
695#ifdef CONFIG_DISCONTIGMEM
696 {
697 struct zonelist *zl;
54a6eb5c 698 int i, j;
1da177e4
LT		 699
700 for (i = 0; i < npmem_ranges; i++) {
4413a0f6 701 zl = node_zonelist(i, 0);
1da177e4 702 for (j = 0; j < MAX_NR_ZONES; j++) {
dd1a239f 703 struct zoneref *z;
54a6eb5c 704 struct zone *zone;
1da177e4
LT		 705
706 printk("Zone list for zone %d on node %d: ", j, i);
54a6eb5c
MG		 707 for_each_zone_zonelist(zone, z, zl, j)
708 printk("[%d/%s] ", zone_to_nid(zone),
709 zone->name);
1da177e4
LT		 710 printk("\n");
711 }
712 }
713 }
714#endif
715}
716
1da177e4
LT		 717/*
718 * pagetable_init() sets up the page tables
719 *
720 * Note that gateway_init() places the Linux gateway page at page 0.
721 * Since gateway pages cannot be dereferenced this has the desirable
722 * side effect of trapping those pesky NULL-reference errors in the
723 * kernel.
724 */
725static void __init pagetable_init(void)
726{
727 int range;
728
729 /* Map each physical memory range to its kernel vaddr */
730
731 for (range = 0; range < npmem_ranges; range++) {
732 unsigned long start_paddr;
733 unsigned long end_paddr;
734 unsigned long size;
735
736 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
737 end_paddr = start_paddr + (pmem_ranges[range].pages << PAGE_SHIFT);
738 size = pmem_ranges[range].pages << PAGE_SHIFT;
739
740 map_pages((unsigned long)__va(start_paddr), start_paddr,
d7dd2ff1 741 size, PAGE_KERNEL, 0);
1da177e4
LT		 742 }
743
744#ifdef CONFIG_BLK_DEV_INITRD
745 if (initrd_end && initrd_end > mem_limit) {
1bcdd854 746 printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
1da177e4 747 map_pages(initrd_start, __pa(initrd_start),
d7dd2ff1 748 initrd_end - initrd_start, PAGE_KERNEL, 0);
1da177e4
LT		 749 }
750#endif
751
752 empty_zero_page = alloc_bootmem_pages(PAGE_SIZE);
753 memset(empty_zero_page, 0, PAGE_SIZE);
754}
755
756static void __init gateway_init(void)
757{
758 unsigned long linux_gateway_page_addr;
759 /* FIXME: This is 'const' in order to trick the compiler
760 into not treating it as DP-relative data. */
761 extern void * const linux_gateway_page;
762
763 linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
764
765 /*
766 * Setup Linux Gateway page.
767 *
768 * The Linux gateway page will reside in kernel space (on virtual
769 * page 0), so it doesn't need to be aliased into user space.
770 */
771
772 map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
d7dd2ff1 773 PAGE_SIZE, PAGE_GATEWAY, 1);
1da177e4
LT		 774}
775
776#ifdef CONFIG_HPUX
777void
778map_hpux_gateway_page(struct task_struct *tsk, struct mm_struct *mm)
779{
780 pgd_t *pg_dir;
781 pmd_t *pmd;
782 pte_t *pg_table;
783 unsigned long start_pmd;
784 unsigned long start_pte;
785 unsigned long address;
786 unsigned long hpux_gw_page_addr;
787 /* FIXME: This is 'const' in order to trick the compiler
788 into not treating it as DP-relative data. */
789 extern void * const hpux_gateway_page;
790
791 hpux_gw_page_addr = HPUX_GATEWAY_ADDR & PAGE_MASK;
792
793 /*
794 * Setup HP-UX Gateway page.
795 *
796 * The HP-UX gateway page resides in the user address space,
797 * so it needs to be aliased into each process.
798 */
799
800 pg_dir = pgd_offset(mm,hpux_gw_page_addr);
801
802#if PTRS_PER_PMD == 1
803 start_pmd = 0;
804#else
805 start_pmd = ((hpux_gw_page_addr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
806#endif
807 start_pte = ((hpux_gw_page_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
808
809 address = __pa(&hpux_gateway_page);
810#if PTRS_PER_PMD == 1
811 pmd = (pmd_t *)__pa(pg_dir);
812#else
813 pmd = (pmd_t *) pgd_address(*pg_dir);
814
815 /*
816 * pmd is physical at this point
817 */
818
819 if (!pmd) {
820 pmd = (pmd_t *) get_zeroed_page(GFP_KERNEL);
821 pmd = (pmd_t *) __pa(pmd);
822 }
823
824 __pgd_val_set(*pg_dir, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pmd);
825#endif
826 /* now change pmd to kernel virtual addresses */
827
828 pmd = (pmd_t *)__va(pmd) + start_pmd;
829
830 /*
831 * pg_table is physical at this point
832 */
833
834 pg_table = (pte_t *) pmd_address(*pmd);
835 if (!pg_table)
836 pg_table = (pte_t *) __pa(get_zeroed_page(GFP_KERNEL));
837
838 __pmd_val_set(*pmd, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pg_table);
839
840 /* now change pg_table to kernel virtual addresses */
841
842 pg_table = (pte_t *) __va(pg_table) + start_pte;
843 set_pte(pg_table, __mk_pte(address, PAGE_GATEWAY));
844}
845EXPORT_SYMBOL(map_hpux_gateway_page);
846#endif
847
1da177e4
LT		 848void __init paging_init(void)
849{
850 int i;
851
852 setup_bootmem();
853 pagetable_init();
854 gateway_init();
855 flush_cache_all_local(); /* start with known state */
ce33941f 856 flush_tlb_all_local(NULL);
1da177e4
LT		 857
858 for (i = 0; i < npmem_ranges; i++) {
f06a9684 859 unsigned long zones_size[MAX_NR_ZONES] = { 0, };
1da177e4 860
00592837 861 zones_size[ZONE_NORMAL] = pmem_ranges[i].pages;
1da177e4
LT		 862
863#ifdef CONFIG_DISCONTIGMEM
864 /* Need to initialize the pfnnid_map before we can initialize
865 the zone */
866 {
867 int j;
868 for (j = (pmem_ranges[i].start_pfn >> PFNNID_SHIFT);
869 j <= ((pmem_ranges[i].start_pfn + pmem_ranges[i].pages) >> PFNNID_SHIFT);
870 j++) {
871 pfnnid_map[j] = i;
872 }
873 }
874#endif
875
9109fb7b 876 free_area_init_node(i, zones_size,
1da177e4
LT		 877 pmem_ranges[i].start_pfn, NULL);
878 }
879}
880
881#ifdef CONFIG_PA20
882
883/*
7022672e 884 * Currently, all PA20 chips have 18 bit protection IDs, which is the
1da177e4
LT		 885 * limiting factor (space ids are 32 bits).
886 */
887
888#define NR_SPACE_IDS 262144
889
890#else
891
892/*
7022672e
SA		 893 * Currently we have a one-to-one relationship between space IDs and
894 * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
895 * support 15 bit protection IDs, so that is the limiting factor.
896 * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
1da177e4
LT		 897 * probably not worth the effort for a special case here.
898 */
899
900#define NR_SPACE_IDS 32768
901
902#endif /* !CONFIG_PA20 */
903
904#define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
905#define SID_ARRAY_SIZE (NR_SPACE_IDS / (8 * sizeof(long)))
906
907static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
908static unsigned long dirty_space_id[SID_ARRAY_SIZE];
909static unsigned long space_id_index;
910static unsigned long free_space_ids = NR_SPACE_IDS - 1;
911static unsigned long dirty_space_ids = 0;
912
913static DEFINE_SPINLOCK(sid_lock);
914
915unsigned long alloc_sid(void)
916{
917 unsigned long index;
918
919 spin_lock(&sid_lock);
920
921 if (free_space_ids == 0) {
922 if (dirty_space_ids != 0) {
923 spin_unlock(&sid_lock);
924 flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
925 spin_lock(&sid_lock);
926 }
2fd83038 927 BUG_ON(free_space_ids == 0);
1da177e4
LT		 928 }
929
930 free_space_ids--;
931
932 index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
933 space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1)));
934 space_id_index = index;
935
936 spin_unlock(&sid_lock);
937
938 return index << SPACEID_SHIFT;
939}
940
941void free_sid(unsigned long spaceid)
942{
943 unsigned long index = spaceid >> SPACEID_SHIFT;
944 unsigned long *dirty_space_offset;
945
946 dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG);
947 index &= (BITS_PER_LONG - 1);
948
949 spin_lock(&sid_lock);
950
2fd83038 951 BUG_ON(*dirty_space_offset & (1L << index)); /* attempt to free space id twice */
1da177e4
LT		 952
953 *dirty_space_offset |= (1L << index);
954 dirty_space_ids++;
955
956 spin_unlock(&sid_lock);
957}
958
959
960#ifdef CONFIG_SMP
961static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
962{
963 int i;
964
965 /* NOTE: sid_lock must be held upon entry */
966
967 *ndirtyptr = dirty_space_ids;
968 if (dirty_space_ids != 0) {
969 for (i = 0; i < SID_ARRAY_SIZE; i++) {
970 dirty_array[i] = dirty_space_id[i];
971 dirty_space_id[i] = 0;
972 }
973 dirty_space_ids = 0;
974 }
975
976 return;
977}
978
979static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
980{
981 int i;
982
983 /* NOTE: sid_lock must be held upon entry */
984
985 if (ndirty != 0) {
986 for (i = 0; i < SID_ARRAY_SIZE; i++) {
987 space_id[i] ^= dirty_array[i];
988 }
989
990 free_space_ids += ndirty;
991 space_id_index = 0;
992 }
993}
994
995#else /* CONFIG_SMP */
996
997static void recycle_sids(void)
998{
999 int i;
1000
1001 /* NOTE: sid_lock must be held upon entry */
1002
1003 if (dirty_space_ids != 0) {
1004 for (i = 0; i < SID_ARRAY_SIZE; i++) {
1005 space_id[i] ^= dirty_space_id[i];
1006 dirty_space_id[i] = 0;
1007 }
1008
1009 free_space_ids += dirty_space_ids;
1010 dirty_space_ids = 0;
1011 space_id_index = 0;
1012 }
1013}
1014#endif
1015
1016/*
1017 * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
1018 * purged, we can safely reuse the space ids that were released but
1019 * not flushed from the tlb.
1020 */
1021
1022#ifdef CONFIG_SMP
1023
1024static unsigned long recycle_ndirty;
1025static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
2fd83038 1026static unsigned int recycle_inuse;
1da177e4
LT		 1027
1028void flush_tlb_all(void)
1029{
1030 int do_recycle;
1031
416821d3 1032 __inc_irq_stat(irq_tlb_count);
1da177e4
LT		 1033 do_recycle = 0;
1034 spin_lock(&sid_lock);
1035 if (dirty_space_ids > RECYCLE_THRESHOLD) {
2fd83038 1036 BUG_ON(recycle_inuse); /* FIXME: Use a semaphore/wait queue here */
1da177e4
LT		 1037 get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
1038 recycle_inuse++;
1039 do_recycle++;
1040 }
1041 spin_unlock(&sid_lock);
15c8b6c1 1042 on_each_cpu(flush_tlb_all_local, NULL, 1);
1da177e4
LT		 1043 if (do_recycle) {
1044 spin_lock(&sid_lock);
1045 recycle_sids(recycle_ndirty,recycle_dirty_array);
1046 recycle_inuse = 0;
1047 spin_unlock(&sid_lock);
1048 }
1049}
1050#else
1051void flush_tlb_all(void)
1052{
416821d3 1053 __inc_irq_stat(irq_tlb_count);
1da177e4 1054 spin_lock(&sid_lock);
1b2425e3 1055 flush_tlb_all_local(NULL);
1da177e4
LT		 1056 recycle_sids();
1057 spin_unlock(&sid_lock);
1058}
1059#endif
1060
1061#ifdef CONFIG_BLK_DEV_INITRD
1062void free_initrd_mem(unsigned long start, unsigned long end)
1063{
7d2c7747 1064 free_reserved_area((void *)start, (void *)end, -1, "initrd");
1da177e4
LT		 1065}
1066#endif
Linux 6.x block layer and io_uring tree(s)