Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/arch/parisc/mm/init.c
4 *
5 * Copyright (C) 1995 Linus Torvalds
6 * Copyright 1999 SuSE GmbH
7 * changed by Philipp Rumpf
8 * Copyright 1999 Philipp Rumpf (prumpf@tux.org)
9 * Copyright 2004 Randolph Chung (tausq@debian.org)
10 * Copyright 2006-2007 Helge Deller (deller@gmx.de)
11 *
12 */
13
14
15#include <linux/module.h>
16#include <linux/mm.h>
17#include <linux/memblock.h>
18#include <linux/gfp.h>
19#include <linux/delay.h>
20#include <linux/init.h>
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 */
26#include <linux/compat.h>
27#include <linux/execmem.h>
28
29#include <asm/pgalloc.h>
30#include <asm/tlb.h>
31#include <asm/pdc_chassis.h>
32#include <asm/mmzone.h>
33#include <asm/sections.h>
34#include <asm/msgbuf.h>
35#include <asm/sparsemem.h>
36#include <asm/asm-offsets.h>
37#include <asm/shmbuf.h>
38
39extern int data_start;
40extern void parisc_kernel_start(void); /* Kernel entry point in head.S */
41
42#if CONFIG_PGTABLE_LEVELS == 3
43pmd_t pmd0[PTRS_PER_PMD] __section(".data..vm0.pmd") __attribute__ ((aligned(PAGE_SIZE)));
44#endif
45
46pgd_t swapper_pg_dir[PTRS_PER_PGD] __section(".data..vm0.pgd") __attribute__ ((aligned(PAGE_SIZE)));
47pte_t pg0[PT_INITIAL * PTRS_PER_PTE] __section(".data..vm0.pte") __attribute__ ((aligned(PAGE_SIZE)));
48
49static struct resource data_resource = {
50 .name = "Kernel data",
51 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
52};
53
54static struct resource code_resource = {
55 .name = "Kernel code",
56 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
57};
58
59static struct resource pdcdata_resource = {
60 .name = "PDC data (Page Zero)",
61 .start = 0,
62 .end = 0x9ff,
63 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
64};
65
66static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __ro_after_init;
67
68/* The following array is initialized from the firmware specific
69 * information retrieved in kernel/inventory.c.
70 */
71
72physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __initdata;
73int npmem_ranges __initdata;
74
75#ifdef CONFIG_64BIT
76#define MAX_MEM (1UL << MAX_PHYSMEM_BITS)
77#else /* !CONFIG_64BIT */
78#define MAX_MEM (3584U*1024U*1024U)
79#endif /* !CONFIG_64BIT */
80
81static unsigned long mem_limit __read_mostly = MAX_MEM;
82
83static void __init mem_limit_func(void)
84{
85 char *cp, *end;
86 unsigned long limit;
87
88 /* We need this before __setup() functions are called */
89
90 limit = MAX_MEM;
91 for (cp = boot_command_line; *cp; ) {
92 if (memcmp(cp, "mem=", 4) == 0) {
93 cp += 4;
94 limit = memparse(cp, &end);
95 if (end != cp)
96 break;
97 cp = end;
98 } else {
99 while (*cp != ' ' && *cp)
100 ++cp;
101 while (*cp == ' ')
102 ++cp;
103 }
104 }
105
106 if (limit < mem_limit)
107 mem_limit = limit;
108}
109
110#define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
111
112static void __init setup_bootmem(void)
113{
114 unsigned long mem_max;
115#ifndef CONFIG_SPARSEMEM
116 physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
117 int npmem_holes;
118#endif
119 int i, sysram_resource_count;
120
121 disable_sr_hashing(); /* Turn off space register hashing */
122
123 /*
124 * Sort the ranges. Since the number of ranges is typically
125 * small, and performance is not an issue here, just do
126 * a simple insertion sort.
127 */
128
129 for (i = 1; i < npmem_ranges; i++) {
130 int j;
131
132 for (j = i; j > 0; j--) {
133 if (pmem_ranges[j-1].start_pfn <
134 pmem_ranges[j].start_pfn) {
135
136 break;
137 }
138 swap(pmem_ranges[j-1], pmem_ranges[j]);
139 }
140 }
141
142#ifndef CONFIG_SPARSEMEM
143 /*
144 * Throw out ranges that are too far apart (controlled by
145 * MAX_GAP).
146 */
147
148 for (i = 1; i < npmem_ranges; i++) {
149 if (pmem_ranges[i].start_pfn -
150 (pmem_ranges[i-1].start_pfn +
151 pmem_ranges[i-1].pages) > MAX_GAP) {
152 npmem_ranges = i;
153 printk("Large gap in memory detected (%ld pages). "
154 "Consider turning on CONFIG_SPARSEMEM\n",
155 pmem_ranges[i].start_pfn -
156 (pmem_ranges[i-1].start_pfn +
157 pmem_ranges[i-1].pages));
158 break;
159 }
160 }
161#endif
162
163 /* Print the memory ranges */
164 pr_info("Memory Ranges:\n");
165
166 for (i = 0; i < npmem_ranges; i++) {
167 struct resource *res = &sysram_resources[i];
168 unsigned long start;
169 unsigned long size;
170
171 size = (pmem_ranges[i].pages << PAGE_SHIFT);
172 start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
173 pr_info("%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
174 i, start, start + (size - 1), size >> 20);
175
176 /* request memory resource */
177 res->name = "System RAM";
178 res->start = start;
179 res->end = start + size - 1;
180 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
181 request_resource(&iomem_resource, res);
182 }
183
184 sysram_resource_count = npmem_ranges;
185
186 /*
187 * For 32 bit kernels we limit the amount of memory we can
188 * support, in order to preserve enough kernel address space
189 * for other purposes. For 64 bit kernels we don't normally
190 * limit the memory, but this mechanism can be used to
191 * artificially limit the amount of memory (and it is written
192 * to work with multiple memory ranges).
193 */
194
195 mem_limit_func(); /* check for "mem=" argument */
196
197 mem_max = 0;
198 for (i = 0; i < npmem_ranges; i++) {
199 unsigned long rsize;
200
201 rsize = pmem_ranges[i].pages << PAGE_SHIFT;
202 if ((mem_max + rsize) > mem_limit) {
203 printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
204 if (mem_max == mem_limit)
205 npmem_ranges = i;
206 else {
207 pmem_ranges[i].pages = (mem_limit >> PAGE_SHIFT)
208 - (mem_max >> PAGE_SHIFT);
209 npmem_ranges = i + 1;
210 mem_max = mem_limit;
211 }
212 break;
213 }
214 mem_max += rsize;
215 }
216
217 printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
218
219#ifndef CONFIG_SPARSEMEM
220 /* Merge the ranges, keeping track of the holes */
221 {
222 unsigned long end_pfn;
223 unsigned long hole_pages;
224
225 npmem_holes = 0;
226 end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
227 for (i = 1; i < npmem_ranges; i++) {
228
229 hole_pages = pmem_ranges[i].start_pfn - end_pfn;
230 if (hole_pages) {
231 pmem_holes[npmem_holes].start_pfn = end_pfn;
232 pmem_holes[npmem_holes++].pages = hole_pages;
233 end_pfn += hole_pages;
234 }
235 end_pfn += pmem_ranges[i].pages;
236 }
237
238 pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
239 npmem_ranges = 1;
240 }
241#endif
242
243 /*
244 * Initialize and free the full range of memory in each range.
245 */
246
247 max_pfn = 0;
248 for (i = 0; i < npmem_ranges; i++) {
249 unsigned long start_pfn;
250 unsigned long npages;
251 unsigned long start;
252 unsigned long size;
253
254 start_pfn = pmem_ranges[i].start_pfn;
255 npages = pmem_ranges[i].pages;
256
257 start = start_pfn << PAGE_SHIFT;
258 size = npages << PAGE_SHIFT;
259
260 /* add system RAM memblock */
261 memblock_add(start, size);
262
263 if ((start_pfn + npages) > max_pfn)
264 max_pfn = start_pfn + npages;
265 }
266
267 /*
268 * We can't use memblock top-down allocations because we only
269 * created the initial mapping up to KERNEL_INITIAL_SIZE in
270 * the assembly bootup code.
271 */
272 memblock_set_bottom_up(true);
273
274 /* IOMMU is always used to access "high mem" on those boxes
275 * that can support enough mem that a PCI device couldn't
276 * directly DMA to any physical addresses.
277 * ISA DMA support will need to revisit this.
278 */
279 max_low_pfn = max_pfn;
280
281 /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
282
283#define PDC_CONSOLE_IO_IODC_SIZE 32768
284
285 memblock_reserve(0UL, (unsigned long)(PAGE0->mem_free +
286 PDC_CONSOLE_IO_IODC_SIZE));
287 memblock_reserve(__pa(KERNEL_BINARY_TEXT_START),
288 (unsigned long)(_end - KERNEL_BINARY_TEXT_START));
289
290#ifndef CONFIG_SPARSEMEM
291
292 /* reserve the holes */
293
294 for (i = 0; i < npmem_holes; i++) {
295 memblock_reserve((pmem_holes[i].start_pfn << PAGE_SHIFT),
296 (pmem_holes[i].pages << PAGE_SHIFT));
297 }
298#endif
299
300#ifdef CONFIG_BLK_DEV_INITRD
301 if (initrd_start) {
302 printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
303 if (__pa(initrd_start) < mem_max) {
304 unsigned long initrd_reserve;
305
306 if (__pa(initrd_end) > mem_max) {
307 initrd_reserve = mem_max - __pa(initrd_start);
308 } else {
309 initrd_reserve = initrd_end - initrd_start;
310 }
311 initrd_below_start_ok = 1;
312 printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
313
314 memblock_reserve(__pa(initrd_start), initrd_reserve);
315 }
316 }
317#endif
318
319 data_resource.start = virt_to_phys(&data_start);
320 data_resource.end = virt_to_phys(_end) - 1;
321 code_resource.start = virt_to_phys(_text);
322 code_resource.end = virt_to_phys(&data_start)-1;
323
324 /* We don't know which region the kernel will be in, so try
325 * all of them.
326 */
327 for (i = 0; i < sysram_resource_count; i++) {
328 struct resource *res = &sysram_resources[i];
329 request_resource(res, &code_resource);
330 request_resource(res, &data_resource);
331 }
332 request_resource(&sysram_resources[0], &pdcdata_resource);
333
334 /* Initialize Page Deallocation Table (PDT) and check for bad memory. */
335 pdc_pdt_init();
336
337 memblock_allow_resize();
338 memblock_dump_all();
339}
340
341static bool kernel_set_to_readonly;
342
343static void __ref map_pages(unsigned long start_vaddr,
344 unsigned long start_paddr, unsigned long size,
345 pgprot_t pgprot, int force)
346{
347 pmd_t *pmd;
348 pte_t *pg_table;
349 unsigned long end_paddr;
350 unsigned long start_pmd;
351 unsigned long start_pte;
352 unsigned long tmp1;
353 unsigned long tmp2;
354 unsigned long address;
355 unsigned long vaddr;
356 unsigned long ro_start;
357 unsigned long ro_end;
358 unsigned long kernel_start, kernel_end;
359
360 ro_start = __pa((unsigned long)_text);
361 ro_end = __pa((unsigned long)&data_start);
362 kernel_start = __pa((unsigned long)&__init_begin);
363 kernel_end = __pa((unsigned long)&_end);
364
365 end_paddr = start_paddr + size;
366
367 /* for 2-level configuration PTRS_PER_PMD is 0 so start_pmd will be 0 */
368 start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
369 start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
370
371 address = start_paddr;
372 vaddr = start_vaddr;
373 while (address < end_paddr) {
374 pgd_t *pgd = pgd_offset_k(vaddr);
375 p4d_t *p4d = p4d_offset(pgd, vaddr);
376 pud_t *pud = pud_offset(p4d, vaddr);
377
378#if CONFIG_PGTABLE_LEVELS == 3
379 if (pud_none(*pud)) {
380 pmd = memblock_alloc(PAGE_SIZE << PMD_TABLE_ORDER,
381 PAGE_SIZE << PMD_TABLE_ORDER);
382 if (!pmd)
383 panic("pmd allocation failed.\n");
384 pud_populate(NULL, pud, pmd);
385 }
386#endif
387
388 pmd = pmd_offset(pud, vaddr);
389 for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) {
390 if (pmd_none(*pmd)) {
391 pg_table = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
392 if (!pg_table)
393 panic("page table allocation failed\n");
394 pmd_populate_kernel(NULL, pmd, pg_table);
395 }
396
397 pg_table = pte_offset_kernel(pmd, vaddr);
398 for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) {
399 pte_t pte;
400 pgprot_t prot;
401 bool huge = false;
402
403 if (force) {
404 prot = pgprot;
405 } else if (address < kernel_start || address >= kernel_end) {
406 /* outside kernel memory */
407 prot = PAGE_KERNEL;
408 } else if (!kernel_set_to_readonly) {
409 /* still initializing, allow writing to RO memory */
410 prot = PAGE_KERNEL_RWX;
411 huge = true;
412 } else if (address >= ro_start) {
413 /* Code (ro) and Data areas */
414 prot = (address < ro_end) ?
415 PAGE_KERNEL_EXEC : PAGE_KERNEL;
416 huge = true;
417 } else {
418 prot = PAGE_KERNEL;
419 }
420
421 pte = __mk_pte(address, prot);
422 if (huge)
423 pte = pte_mkhuge(pte);
424
425 if (address >= end_paddr)
426 break;
427
428 set_pte(pg_table, pte);
429
430 address += PAGE_SIZE;
431 vaddr += PAGE_SIZE;
432 }
433 start_pte = 0;
434
435 if (address >= end_paddr)
436 break;
437 }
438 start_pmd = 0;
439 }
440}
441
442void __init set_kernel_text_rw(int enable_read_write)
443{
444 unsigned long start = (unsigned long) __init_begin;
445 unsigned long end = (unsigned long) &data_start;
446
447 map_pages(start, __pa(start), end-start,
448 PAGE_KERNEL_RWX, enable_read_write ? 1:0);
449
450 /* force the kernel to see the new page table entries */
451 flush_cache_all();
452 flush_tlb_all();
453}
454
455void free_initmem(void)
456{
457 unsigned long init_begin = (unsigned long)__init_begin;
458 unsigned long init_end = (unsigned long)__init_end;
459 unsigned long kernel_end = (unsigned long)&_end;
460
461 /* Remap kernel text and data, but do not touch init section yet. */
462 map_pages(init_end, __pa(init_end), kernel_end - init_end,
463 PAGE_KERNEL, 0);
464
465 /* The init text pages are marked R-X. We have to
466 * flush the icache and mark them RW-
467 *
468 * Do a dummy remap of the data section first (the data
469 * section is already PAGE_KERNEL) to pull in the TLB entries
470 * for map_kernel */
471 map_pages(init_begin, __pa(init_begin), init_end - init_begin,
472 PAGE_KERNEL_RWX, 1);
473 /* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
474 * map_pages */
475 map_pages(init_begin, __pa(init_begin), init_end - init_begin,
476 PAGE_KERNEL, 1);
477
478 /* force the kernel to see the new TLB entries */
479 __flush_tlb_range(0, init_begin, kernel_end);
480
481 /* finally dump all the instructions which were cached, since the
482 * pages are no-longer executable */
483 flush_icache_range(init_begin, init_end);
484
485 free_initmem_default(POISON_FREE_INITMEM);
486
487 /* set up a new led state on systems shipped LED State panel */
488 pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
489}
490
491
492#ifdef CONFIG_STRICT_KERNEL_RWX
493void mark_rodata_ro(void)
494{
495 unsigned long start = (unsigned long) &__start_rodata;
496 unsigned long end = (unsigned long) &__end_rodata;
497
498 pr_info("Write protecting the kernel read-only data: %luk\n",
499 (end - start) >> 10);
500
501 kernel_set_to_readonly = true;
502 map_pages(start, __pa(start), end - start, PAGE_KERNEL, 0);
503
504 /* force the kernel to see the new page table entries */
505 flush_cache_all();
506 flush_tlb_all();
507}
508#endif
509
510
511/*
512 * Just an arbitrary offset to serve as a "hole" between mapping areas
513 * (between top of physical memory and a potential pcxl dma mapping
514 * area, and below the vmalloc mapping area).
515 *
516 * The current 32K value just means that there will be a 32K "hole"
517 * between mapping areas. That means that any out-of-bounds memory
518 * accesses will hopefully be caught. The vmalloc() routines leaves
519 * a hole of 4kB between each vmalloced area for the same reason.
520 */
521
522 /* Leave room for gateway page expansion */
523#if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
524#error KERNEL_MAP_START is in gateway reserved region
525#endif
526#define MAP_START (KERNEL_MAP_START)
527
528#define VM_MAP_OFFSET (32*1024)
529#define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
530 & ~(VM_MAP_OFFSET-1)))
531
532void *parisc_vmalloc_start __ro_after_init;
533EXPORT_SYMBOL(parisc_vmalloc_start);
534
535void __init mem_init(void)
536{
537 /* Do sanity checks on IPC (compat) structures */
538 BUILD_BUG_ON(sizeof(struct ipc64_perm) != 48);
539#ifndef CONFIG_64BIT
540 BUILD_BUG_ON(sizeof(struct semid64_ds) != 80);
541 BUILD_BUG_ON(sizeof(struct msqid64_ds) != 104);
542 BUILD_BUG_ON(sizeof(struct shmid64_ds) != 104);
543#endif
544#ifdef CONFIG_COMPAT
545 BUILD_BUG_ON(sizeof(struct compat_ipc64_perm) != sizeof(struct ipc64_perm));
546 BUILD_BUG_ON(sizeof(struct compat_semid64_ds) != 80);
547 BUILD_BUG_ON(sizeof(struct compat_msqid64_ds) != 104);
548 BUILD_BUG_ON(sizeof(struct compat_shmid64_ds) != 104);
549#endif
550
551 /* Do sanity checks on page table constants */
552 BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
553 BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
554 BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
555 BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
556 > BITS_PER_LONG);
557#if CONFIG_PGTABLE_LEVELS == 3
558 BUILD_BUG_ON(PT_INITIAL > PTRS_PER_PMD);
559#else
560 BUILD_BUG_ON(PT_INITIAL > PTRS_PER_PGD);
561#endif
562
563#ifdef CONFIG_64BIT
564 /* avoid ldil_%L() asm statements to sign-extend into upper 32-bits */
565 BUILD_BUG_ON(__PAGE_OFFSET >= 0x80000000);
566 BUILD_BUG_ON(TMPALIAS_MAP_START >= 0x80000000);
567#endif
568
569 high_memory = __va((max_pfn << PAGE_SHIFT));
570 set_max_mapnr(max_low_pfn);
571 memblock_free_all();
572
573#ifdef CONFIG_PA11
574 if (boot_cpu_data.cpu_type == pcxl2 || boot_cpu_data.cpu_type == pcxl) {
575 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
576 parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
577 + PCXL_DMA_MAP_SIZE);
578 } else
579#endif
580 parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
581
582#if 0
583 /*
584 * Do not expose the virtual kernel memory layout to userspace.
585 * But keep code for debugging purposes.
586 */
587 printk("virtual kernel memory layout:\n"
588 " vmalloc : 0x%px - 0x%px (%4ld MB)\n"
589 " fixmap : 0x%px - 0x%px (%4ld kB)\n"
590 " memory : 0x%px - 0x%px (%4ld MB)\n"
591 " .init : 0x%px - 0x%px (%4ld kB)\n"
592 " .data : 0x%px - 0x%px (%4ld kB)\n"
593 " .text : 0x%px - 0x%px (%4ld kB)\n",
594
595 (void*)VMALLOC_START, (void*)VMALLOC_END,
596 (VMALLOC_END - VMALLOC_START) >> 20,
597
598 (void *)FIXMAP_START, (void *)(FIXMAP_START + FIXMAP_SIZE),
599 (unsigned long)(FIXMAP_SIZE / 1024),
600
601 __va(0), high_memory,
602 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
603
604 __init_begin, __init_end,
605 ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
606
607 _etext, _edata,
608 ((unsigned long)_edata - (unsigned long)_etext) >> 10,
609
610 _text, _etext,
611 ((unsigned long)_etext - (unsigned long)_text) >> 10);
612#endif
613}
614
615unsigned long *empty_zero_page __ro_after_init;
616EXPORT_SYMBOL(empty_zero_page);
617
618/*
619 * pagetable_init() sets up the page tables
620 *
621 * Note that gateway_init() places the Linux gateway page at page 0.
622 * Since gateway pages cannot be dereferenced this has the desirable
623 * side effect of trapping those pesky NULL-reference errors in the
624 * kernel.
625 */
626static void __init pagetable_init(void)
627{
628 int range;
629
630 /* Map each physical memory range to its kernel vaddr */
631
632 for (range = 0; range < npmem_ranges; range++) {
633 unsigned long start_paddr;
634 unsigned long size;
635
636 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
637 size = pmem_ranges[range].pages << PAGE_SHIFT;
638
639 map_pages((unsigned long)__va(start_paddr), start_paddr,
640 size, PAGE_KERNEL, 0);
641 }
642
643#ifdef CONFIG_BLK_DEV_INITRD
644 if (initrd_end && initrd_end > mem_limit) {
645 printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
646 map_pages(initrd_start, __pa(initrd_start),
647 initrd_end - initrd_start, PAGE_KERNEL, 0);
648 }
649#endif
650
651 empty_zero_page = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
652 if (!empty_zero_page)
653 panic("zero page allocation failed.\n");
654
655}
656
657static void __init gateway_init(void)
658{
659 unsigned long linux_gateway_page_addr;
660 /* FIXME: This is 'const' in order to trick the compiler
661 into not treating it as DP-relative data. */
662 extern void * const linux_gateway_page;
663
664 linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
665
666 /*
667 * Setup Linux Gateway page.
668 *
669 * The Linux gateway page will reside in kernel space (on virtual
670 * page 0), so it doesn't need to be aliased into user space.
671 */
672
673 map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
674 PAGE_SIZE, PAGE_GATEWAY, 1);
675}
676
677static void __init fixmap_init(void)
678{
679 unsigned long addr = FIXMAP_START;
680 unsigned long end = FIXMAP_START + FIXMAP_SIZE;
681 pgd_t *pgd = pgd_offset_k(addr);
682 p4d_t *p4d = p4d_offset(pgd, addr);
683 pud_t *pud = pud_offset(p4d, addr);
684 pmd_t *pmd;
685
686 BUILD_BUG_ON(FIXMAP_SIZE > PMD_SIZE);
687
688#if CONFIG_PGTABLE_LEVELS == 3
689 if (pud_none(*pud)) {
690 pmd = memblock_alloc(PAGE_SIZE << PMD_TABLE_ORDER,
691 PAGE_SIZE << PMD_TABLE_ORDER);
692 if (!pmd)
693 panic("fixmap: pmd allocation failed.\n");
694 pud_populate(NULL, pud, pmd);
695 }
696#endif
697
698 pmd = pmd_offset(pud, addr);
699 do {
700 pte_t *pte = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
701 if (!pte)
702 panic("fixmap: pte allocation failed.\n");
703
704 pmd_populate_kernel(&init_mm, pmd, pte);
705
706 addr += PAGE_SIZE;
707 } while (addr < end);
708}
709
710static void __init parisc_bootmem_free(void)
711{
712 unsigned long max_zone_pfn[MAX_NR_ZONES] = { 0, };
713
714 max_zone_pfn[0] = memblock_end_of_DRAM();
715
716 free_area_init(max_zone_pfn);
717}
718
719void __init paging_init(void)
720{
721 setup_bootmem();
722 pagetable_init();
723 gateway_init();
724 fixmap_init();
725 flush_cache_all_local(); /* start with known state */
726 flush_tlb_all_local(NULL);
727
728 sparse_init();
729 parisc_bootmem_free();
730}
731
732static void alloc_btlb(unsigned long start, unsigned long end, int *slot,
733 unsigned long entry_info)
734{
735 const int slot_max = btlb_info.fixed_range_info.num_comb;
736 int min_num_pages = btlb_info.min_size;
737 unsigned long size;
738
739 /* map at minimum 4 pages */
740 if (min_num_pages < 4)
741 min_num_pages = 4;
742
743 size = HUGEPAGE_SIZE;
744 while (start < end && *slot < slot_max && size >= PAGE_SIZE) {
745 /* starting address must have same alignment as size! */
746 /* if correctly aligned and fits in double size, increase */
747 if (((start & (2 * size - 1)) == 0) &&
748 (end - start) >= (2 * size)) {
749 size <<= 1;
750 continue;
751 }
752 /* if current size alignment is too big, try smaller size */
753 if ((start & (size - 1)) != 0) {
754 size >>= 1;
755 continue;
756 }
757 if ((end - start) >= size) {
758 if ((size >> PAGE_SHIFT) >= min_num_pages)
759 pdc_btlb_insert(start >> PAGE_SHIFT, __pa(start) >> PAGE_SHIFT,
760 size >> PAGE_SHIFT, entry_info, *slot);
761 (*slot)++;
762 start += size;
763 continue;
764 }
765 size /= 2;
766 continue;
767 }
768}
769
770void btlb_init_per_cpu(void)
771{
772 unsigned long s, t, e;
773 int slot;
774
775 /* BTLBs are not available on 64-bit CPUs */
776 if (IS_ENABLED(CONFIG_PA20))
777 return;
778 else if (pdc_btlb_info(&btlb_info) < 0) {
779 memset(&btlb_info, 0, sizeof btlb_info);
780 }
781
782 /* insert BLTLBs for code and data segments */
783 s = (uintptr_t) dereference_function_descriptor(&_stext);
784 e = (uintptr_t) dereference_function_descriptor(&_etext);
785 t = (uintptr_t) dereference_function_descriptor(&_sdata);
786 BUG_ON(t != e);
787
788 /* code segments */
789 slot = 0;
790 alloc_btlb(s, e, &slot, 0x13800000);
791
792 /* sanity check */
793 t = (uintptr_t) dereference_function_descriptor(&_edata);
794 e = (uintptr_t) dereference_function_descriptor(&__bss_start);
795 BUG_ON(t != e);
796
797 /* data segments */
798 s = (uintptr_t) dereference_function_descriptor(&_sdata);
799 e = (uintptr_t) dereference_function_descriptor(&__bss_stop);
800 alloc_btlb(s, e, &slot, 0x11800000);
801}
802
803#ifdef CONFIG_PA20
804
805/*
806 * Currently, all PA20 chips have 18 bit protection IDs, which is the
807 * limiting factor (space ids are 32 bits).
808 */
809
810#define NR_SPACE_IDS 262144
811
812#else
813
814/*
815 * Currently we have a one-to-one relationship between space IDs and
816 * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
817 * support 15 bit protection IDs, so that is the limiting factor.
818 * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
819 * probably not worth the effort for a special case here.
820 */
821
822#define NR_SPACE_IDS 32768
823
824#endif /* !CONFIG_PA20 */
825
826#define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
827#define SID_ARRAY_SIZE (NR_SPACE_IDS / (8 * sizeof(long)))
828
829static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
830static unsigned long dirty_space_id[SID_ARRAY_SIZE];
831static unsigned long space_id_index;
832static unsigned long free_space_ids = NR_SPACE_IDS - 1;
833static unsigned long dirty_space_ids;
834
835static DEFINE_SPINLOCK(sid_lock);
836
837unsigned long alloc_sid(void)
838{
839 unsigned long index;
840
841 spin_lock(&sid_lock);
842
843 if (free_space_ids == 0) {
844 if (dirty_space_ids != 0) {
845 spin_unlock(&sid_lock);
846 flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
847 spin_lock(&sid_lock);
848 }
849 BUG_ON(free_space_ids == 0);
850 }
851
852 free_space_ids--;
853
854 index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
855 space_id[BIT_WORD(index)] |= BIT_MASK(index);
856 space_id_index = index;
857
858 spin_unlock(&sid_lock);
859
860 return index << SPACEID_SHIFT;
861}
862
863void free_sid(unsigned long spaceid)
864{
865 unsigned long index = spaceid >> SPACEID_SHIFT;
866 unsigned long *dirty_space_offset, mask;
867
868 dirty_space_offset = &dirty_space_id[BIT_WORD(index)];
869 mask = BIT_MASK(index);
870
871 spin_lock(&sid_lock);
872
873 BUG_ON(*dirty_space_offset & mask); /* attempt to free space id twice */
874
875 *dirty_space_offset |= mask;
876 dirty_space_ids++;
877
878 spin_unlock(&sid_lock);
879}
880
881
882#ifdef CONFIG_SMP
883static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
884{
885 int i;
886
887 /* NOTE: sid_lock must be held upon entry */
888
889 *ndirtyptr = dirty_space_ids;
890 if (dirty_space_ids != 0) {
891 for (i = 0; i < SID_ARRAY_SIZE; i++) {
892 dirty_array[i] = dirty_space_id[i];
893 dirty_space_id[i] = 0;
894 }
895 dirty_space_ids = 0;
896 }
897
898 return;
899}
900
901static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
902{
903 int i;
904
905 /* NOTE: sid_lock must be held upon entry */
906
907 if (ndirty != 0) {
908 for (i = 0; i < SID_ARRAY_SIZE; i++) {
909 space_id[i] ^= dirty_array[i];
910 }
911
912 free_space_ids += ndirty;
913 space_id_index = 0;
914 }
915}
916
917#else /* CONFIG_SMP */
918
919static void recycle_sids(void)
920{
921 int i;
922
923 /* NOTE: sid_lock must be held upon entry */
924
925 if (dirty_space_ids != 0) {
926 for (i = 0; i < SID_ARRAY_SIZE; i++) {
927 space_id[i] ^= dirty_space_id[i];
928 dirty_space_id[i] = 0;
929 }
930
931 free_space_ids += dirty_space_ids;
932 dirty_space_ids = 0;
933 space_id_index = 0;
934 }
935}
936#endif
937
938/*
939 * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
940 * purged, we can safely reuse the space ids that were released but
941 * not flushed from the tlb.
942 */
943
944#ifdef CONFIG_SMP
945
946static unsigned long recycle_ndirty;
947static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
948static unsigned int recycle_inuse;
949
950void flush_tlb_all(void)
951{
952 int do_recycle;
953
954 do_recycle = 0;
955 spin_lock(&sid_lock);
956 __inc_irq_stat(irq_tlb_count);
957 if (dirty_space_ids > RECYCLE_THRESHOLD) {
958 BUG_ON(recycle_inuse); /* FIXME: Use a semaphore/wait queue here */
959 get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
960 recycle_inuse++;
961 do_recycle++;
962 }
963 spin_unlock(&sid_lock);
964 on_each_cpu(flush_tlb_all_local, NULL, 1);
965 if (do_recycle) {
966 spin_lock(&sid_lock);
967 recycle_sids(recycle_ndirty,recycle_dirty_array);
968 recycle_inuse = 0;
969 spin_unlock(&sid_lock);
970 }
971}
972#else
973void flush_tlb_all(void)
974{
975 spin_lock(&sid_lock);
976 __inc_irq_stat(irq_tlb_count);
977 flush_tlb_all_local(NULL);
978 recycle_sids();
979 spin_unlock(&sid_lock);
980}
981#endif
982
983static const pgprot_t protection_map[16] = {
984 [VM_NONE] = PAGE_NONE,
985 [VM_READ] = PAGE_READONLY,
986 [VM_WRITE] = PAGE_NONE,
987 [VM_WRITE | VM_READ] = PAGE_READONLY,
988 [VM_EXEC] = PAGE_EXECREAD,
989 [VM_EXEC | VM_READ] = PAGE_EXECREAD,
990 [VM_EXEC | VM_WRITE] = PAGE_EXECREAD,
991 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_EXECREAD,
992 [VM_SHARED] = PAGE_NONE,
993 [VM_SHARED | VM_READ] = PAGE_READONLY,
994 [VM_SHARED | VM_WRITE] = PAGE_WRITEONLY,
995 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED,
996 [VM_SHARED | VM_EXEC] = PAGE_EXECREAD,
997 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_EXECREAD,
998 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_RWX,
999 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_RWX
1000};
1001DECLARE_VM_GET_PAGE_PROT
1002
1003#ifdef CONFIG_EXECMEM
1004static struct execmem_info execmem_info __ro_after_init;
1005
1006struct execmem_info __init *execmem_arch_setup(void)
1007{
1008 execmem_info = (struct execmem_info){
1009 .ranges = {
1010 [EXECMEM_DEFAULT] = {
1011 .start = VMALLOC_START,
1012 .end = VMALLOC_END,
1013 .pgprot = PAGE_KERNEL_RWX,
1014 .alignment = 1,
1015 },
1016 },
1017 };
1018
1019 return &execmem_info;
1020}
1021#endif /* CONFIG_EXECMEM */
1/*
2 * linux/arch/parisc/mm/init.c
3 *
4 * Copyright (C) 1995 Linus Torvalds
5 * Copyright 1999 SuSE GmbH
6 * changed by Philipp Rumpf
7 * Copyright 1999 Philipp Rumpf (prumpf@tux.org)
8 * Copyright 2004 Randolph Chung (tausq@debian.org)
9 * Copyright 2006-2007 Helge Deller (deller@gmx.de)
10 *
11 */
12
13
14#include <linux/module.h>
15#include <linux/mm.h>
16#include <linux/bootmem.h>
17#include <linux/gfp.h>
18#include <linux/delay.h>
19#include <linux/init.h>
20#include <linux/pci.h> /* for hppa_dma_ops and pcxl_dma_ops */
21#include <linux/initrd.h>
22#include <linux/swap.h>
23#include <linux/unistd.h>
24#include <linux/nodemask.h> /* for node_online_map */
25#include <linux/pagemap.h> /* for release_pages and page_cache_release */
26
27#include <asm/pgalloc.h>
28#include <asm/pgtable.h>
29#include <asm/tlb.h>
30#include <asm/pdc_chassis.h>
31#include <asm/mmzone.h>
32#include <asm/sections.h>
33
34extern int data_start;
35
36#ifdef CONFIG_DISCONTIGMEM
37struct node_map_data node_data[MAX_NUMNODES] __read_mostly;
38unsigned char pfnnid_map[PFNNID_MAP_MAX] __read_mostly;
39#endif
40
41static struct resource data_resource = {
42 .name = "Kernel data",
43 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
44};
45
46static struct resource code_resource = {
47 .name = "Kernel code",
48 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
49};
50
51static struct resource pdcdata_resource = {
52 .name = "PDC data (Page Zero)",
53 .start = 0,
54 .end = 0x9ff,
55 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
56};
57
58static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __read_mostly;
59
60/* The following array is initialized from the firmware specific
61 * information retrieved in kernel/inventory.c.
62 */
63
64physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __read_mostly;
65int npmem_ranges __read_mostly;
66
67#ifdef CONFIG_64BIT
68#define MAX_MEM (~0UL)
69#else /* !CONFIG_64BIT */
70#define MAX_MEM (3584U*1024U*1024U)
71#endif /* !CONFIG_64BIT */
72
73static unsigned long mem_limit __read_mostly = MAX_MEM;
74
75static void __init mem_limit_func(void)
76{
77 char *cp, *end;
78 unsigned long limit;
79
80 /* We need this before __setup() functions are called */
81
82 limit = MAX_MEM;
83 for (cp = boot_command_line; *cp; ) {
84 if (memcmp(cp, "mem=", 4) == 0) {
85 cp += 4;
86 limit = memparse(cp, &end);
87 if (end != cp)
88 break;
89 cp = end;
90 } else {
91 while (*cp != ' ' && *cp)
92 ++cp;
93 while (*cp == ' ')
94 ++cp;
95 }
96 }
97
98 if (limit < mem_limit)
99 mem_limit = limit;
100}
101
102#define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
103
104static void __init setup_bootmem(void)
105{
106 unsigned long bootmap_size;
107 unsigned long mem_max;
108 unsigned long bootmap_pages;
109 unsigned long bootmap_start_pfn;
110 unsigned long bootmap_pfn;
111#ifndef CONFIG_DISCONTIGMEM
112 physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
113 int npmem_holes;
114#endif
115 int i, sysram_resource_count;
116
117 disable_sr_hashing(); /* Turn off space register hashing */
118
119 /*
120 * Sort the ranges. Since the number of ranges is typically
121 * small, and performance is not an issue here, just do
122 * a simple insertion sort.
123 */
124
125 for (i = 1; i < npmem_ranges; i++) {
126 int j;
127
128 for (j = i; j > 0; j--) {
129 unsigned long tmp;
130
131 if (pmem_ranges[j-1].start_pfn <
132 pmem_ranges[j].start_pfn) {
133
134 break;
135 }
136 tmp = pmem_ranges[j-1].start_pfn;
137 pmem_ranges[j-1].start_pfn = pmem_ranges[j].start_pfn;
138 pmem_ranges[j].start_pfn = tmp;
139 tmp = pmem_ranges[j-1].pages;
140 pmem_ranges[j-1].pages = pmem_ranges[j].pages;
141 pmem_ranges[j].pages = tmp;
142 }
143 }
144
145#ifndef CONFIG_DISCONTIGMEM
146 /*
147 * Throw out ranges that are too far apart (controlled by
148 * MAX_GAP).
149 */
150
151 for (i = 1; i < npmem_ranges; i++) {
152 if (pmem_ranges[i].start_pfn -
153 (pmem_ranges[i-1].start_pfn +
154 pmem_ranges[i-1].pages) > MAX_GAP) {
155 npmem_ranges = i;
156 printk("Large gap in memory detected (%ld pages). "
157 "Consider turning on CONFIG_DISCONTIGMEM\n",
158 pmem_ranges[i].start_pfn -
159 (pmem_ranges[i-1].start_pfn +
160 pmem_ranges[i-1].pages));
161 break;
162 }
163 }
164#endif
165
166 if (npmem_ranges > 1) {
167
168 /* Print the memory ranges */
169
170 printk(KERN_INFO "Memory Ranges:\n");
171
172 for (i = 0; i < npmem_ranges; i++) {
173 unsigned long start;
174 unsigned long size;
175
176 size = (pmem_ranges[i].pages << PAGE_SHIFT);
177 start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
178 printk(KERN_INFO "%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
179 i,start, start + (size - 1), size >> 20);
180 }
181 }
182
183 sysram_resource_count = npmem_ranges;
184 for (i = 0; i < sysram_resource_count; i++) {
185 struct resource *res = &sysram_resources[i];
186 res->name = "System RAM";
187 res->start = pmem_ranges[i].start_pfn << PAGE_SHIFT;
188 res->end = res->start + (pmem_ranges[i].pages << PAGE_SHIFT)-1;
189 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
190 request_resource(&iomem_resource, res);
191 }
192
193 /*
194 * For 32 bit kernels we limit the amount of memory we can
195 * support, in order to preserve enough kernel address space
196 * for other purposes. For 64 bit kernels we don't normally
197 * limit the memory, but this mechanism can be used to
198 * artificially limit the amount of memory (and it is written
199 * to work with multiple memory ranges).
200 */
201
202 mem_limit_func(); /* check for "mem=" argument */
203
204 mem_max = 0;
205 num_physpages = 0;
206 for (i = 0; i < npmem_ranges; i++) {
207 unsigned long rsize;
208
209 rsize = pmem_ranges[i].pages << PAGE_SHIFT;
210 if ((mem_max + rsize) > mem_limit) {
211 printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
212 if (mem_max == mem_limit)
213 npmem_ranges = i;
214 else {
215 pmem_ranges[i].pages = (mem_limit >> PAGE_SHIFT)
216 - (mem_max >> PAGE_SHIFT);
217 npmem_ranges = i + 1;
218 mem_max = mem_limit;
219 }
220 num_physpages += pmem_ranges[i].pages;
221 break;
222 }
223 num_physpages += pmem_ranges[i].pages;
224 mem_max += rsize;
225 }
226
227 printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
228
229#ifndef CONFIG_DISCONTIGMEM
230 /* Merge the ranges, keeping track of the holes */
231
232 {
233 unsigned long end_pfn;
234 unsigned long hole_pages;
235
236 npmem_holes = 0;
237 end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
238 for (i = 1; i < npmem_ranges; i++) {
239
240 hole_pages = pmem_ranges[i].start_pfn - end_pfn;
241 if (hole_pages) {
242 pmem_holes[npmem_holes].start_pfn = end_pfn;
243 pmem_holes[npmem_holes++].pages = hole_pages;
244 end_pfn += hole_pages;
245 }
246 end_pfn += pmem_ranges[i].pages;
247 }
248
249 pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
250 npmem_ranges = 1;
251 }
252#endif
253
254 bootmap_pages = 0;
255 for (i = 0; i < npmem_ranges; i++)
256 bootmap_pages += bootmem_bootmap_pages(pmem_ranges[i].pages);
257
258 bootmap_start_pfn = PAGE_ALIGN(__pa((unsigned long) &_end)) >> PAGE_SHIFT;
259
260#ifdef CONFIG_DISCONTIGMEM
261 for (i = 0; i < MAX_PHYSMEM_RANGES; i++) {
262 memset(NODE_DATA(i), 0, sizeof(pg_data_t));
263 NODE_DATA(i)->bdata = &bootmem_node_data[i];
264 }
265 memset(pfnnid_map, 0xff, sizeof(pfnnid_map));
266
267 for (i = 0; i < npmem_ranges; i++) {
268 node_set_state(i, N_NORMAL_MEMORY);
269 node_set_online(i);
270 }
271#endif
272
273 /*
274 * Initialize and free the full range of memory in each range.
275 * Note that the only writing these routines do are to the bootmap,
276 * and we've made sure to locate the bootmap properly so that they
277 * won't be writing over anything important.
278 */
279
280 bootmap_pfn = bootmap_start_pfn;
281 max_pfn = 0;
282 for (i = 0; i < npmem_ranges; i++) {
283 unsigned long start_pfn;
284 unsigned long npages;
285
286 start_pfn = pmem_ranges[i].start_pfn;
287 npages = pmem_ranges[i].pages;
288
289 bootmap_size = init_bootmem_node(NODE_DATA(i),
290 bootmap_pfn,
291 start_pfn,
292 (start_pfn + npages) );
293 free_bootmem_node(NODE_DATA(i),
294 (start_pfn << PAGE_SHIFT),
295 (npages << PAGE_SHIFT) );
296 bootmap_pfn += (bootmap_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
297 if ((start_pfn + npages) > max_pfn)
298 max_pfn = start_pfn + npages;
299 }
300
301 /* IOMMU is always used to access "high mem" on those boxes
302 * that can support enough mem that a PCI device couldn't
303 * directly DMA to any physical addresses.
304 * ISA DMA support will need to revisit this.
305 */
306 max_low_pfn = max_pfn;
307
308 /* bootmap sizing messed up? */
309 BUG_ON((bootmap_pfn - bootmap_start_pfn) != bootmap_pages);
310
311 /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
312
313#define PDC_CONSOLE_IO_IODC_SIZE 32768
314
315 reserve_bootmem_node(NODE_DATA(0), 0UL,
316 (unsigned long)(PAGE0->mem_free +
317 PDC_CONSOLE_IO_IODC_SIZE), BOOTMEM_DEFAULT);
318 reserve_bootmem_node(NODE_DATA(0), __pa((unsigned long)_text),
319 (unsigned long)(_end - _text), BOOTMEM_DEFAULT);
320 reserve_bootmem_node(NODE_DATA(0), (bootmap_start_pfn << PAGE_SHIFT),
321 ((bootmap_pfn - bootmap_start_pfn) << PAGE_SHIFT),
322 BOOTMEM_DEFAULT);
323
324#ifndef CONFIG_DISCONTIGMEM
325
326 /* reserve the holes */
327
328 for (i = 0; i < npmem_holes; i++) {
329 reserve_bootmem_node(NODE_DATA(0),
330 (pmem_holes[i].start_pfn << PAGE_SHIFT),
331 (pmem_holes[i].pages << PAGE_SHIFT),
332 BOOTMEM_DEFAULT);
333 }
334#endif
335
336#ifdef CONFIG_BLK_DEV_INITRD
337 if (initrd_start) {
338 printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
339 if (__pa(initrd_start) < mem_max) {
340 unsigned long initrd_reserve;
341
342 if (__pa(initrd_end) > mem_max) {
343 initrd_reserve = mem_max - __pa(initrd_start);
344 } else {
345 initrd_reserve = initrd_end - initrd_start;
346 }
347 initrd_below_start_ok = 1;
348 printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
349
350 reserve_bootmem_node(NODE_DATA(0), __pa(initrd_start),
351 initrd_reserve, BOOTMEM_DEFAULT);
352 }
353 }
354#endif
355
356 data_resource.start = virt_to_phys(&data_start);
357 data_resource.end = virt_to_phys(_end) - 1;
358 code_resource.start = virt_to_phys(_text);
359 code_resource.end = virt_to_phys(&data_start)-1;
360
361 /* We don't know which region the kernel will be in, so try
362 * all of them.
363 */
364 for (i = 0; i < sysram_resource_count; i++) {
365 struct resource *res = &sysram_resources[i];
366 request_resource(res, &code_resource);
367 request_resource(res, &data_resource);
368 }
369 request_resource(&sysram_resources[0], &pdcdata_resource);
370}
371
372static void __init map_pages(unsigned long start_vaddr,
373 unsigned long start_paddr, unsigned long size,
374 pgprot_t pgprot, int force)
375{
376 pgd_t *pg_dir;
377 pmd_t *pmd;
378 pte_t *pg_table;
379 unsigned long end_paddr;
380 unsigned long start_pmd;
381 unsigned long start_pte;
382 unsigned long tmp1;
383 unsigned long tmp2;
384 unsigned long address;
385 unsigned long vaddr;
386 unsigned long ro_start;
387 unsigned long ro_end;
388 unsigned long fv_addr;
389 unsigned long gw_addr;
390 extern const unsigned long fault_vector_20;
391 extern void * const linux_gateway_page;
392
393 ro_start = __pa((unsigned long)_text);
394 ro_end = __pa((unsigned long)&data_start);
395 fv_addr = __pa((unsigned long)&fault_vector_20) & PAGE_MASK;
396 gw_addr = __pa((unsigned long)&linux_gateway_page) & PAGE_MASK;
397
398 end_paddr = start_paddr + size;
399
400 pg_dir = pgd_offset_k(start_vaddr);
401
402#if PTRS_PER_PMD == 1
403 start_pmd = 0;
404#else
405 start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
406#endif
407 start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
408
409 address = start_paddr;
410 vaddr = start_vaddr;
411 while (address < end_paddr) {
412#if PTRS_PER_PMD == 1
413 pmd = (pmd_t *)__pa(pg_dir);
414#else
415 pmd = (pmd_t *)pgd_address(*pg_dir);
416
417 /*
418 * pmd is physical at this point
419 */
420
421 if (!pmd) {
422 pmd = (pmd_t *) alloc_bootmem_low_pages_node(NODE_DATA(0), PAGE_SIZE << PMD_ORDER);
423 pmd = (pmd_t *) __pa(pmd);
424 }
425
426 pgd_populate(NULL, pg_dir, __va(pmd));
427#endif
428 pg_dir++;
429
430 /* now change pmd to kernel virtual addresses */
431
432 pmd = (pmd_t *)__va(pmd) + start_pmd;
433 for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) {
434
435 /*
436 * pg_table is physical at this point
437 */
438
439 pg_table = (pte_t *)pmd_address(*pmd);
440 if (!pg_table) {
441 pg_table = (pte_t *)
442 alloc_bootmem_low_pages_node(NODE_DATA(0), PAGE_SIZE);
443 pg_table = (pte_t *) __pa(pg_table);
444 }
445
446 pmd_populate_kernel(NULL, pmd, __va(pg_table));
447
448 /* now change pg_table to kernel virtual addresses */
449
450 pg_table = (pte_t *) __va(pg_table) + start_pte;
451 for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) {
452 pte_t pte;
453
454 /*
455 * Map the fault vector writable so we can
456 * write the HPMC checksum.
457 */
458 if (force)
459 pte = __mk_pte(address, pgprot);
460 else if (core_kernel_text(vaddr) &&
461 address != fv_addr)
462 pte = __mk_pte(address, PAGE_KERNEL_EXEC);
463 else
464#if defined(CONFIG_PARISC_PAGE_SIZE_4KB)
465 if (address >= ro_start && address < ro_end
466 && address != fv_addr
467 && address != gw_addr)
468 pte = __mk_pte(address, PAGE_KERNEL_RO);
469 else
470#endif
471 pte = __mk_pte(address, pgprot);
472
473 if (address >= end_paddr) {
474 if (force)
475 break;
476 else
477 pte_val(pte) = 0;
478 }
479
480 set_pte(pg_table, pte);
481
482 address += PAGE_SIZE;
483 vaddr += PAGE_SIZE;
484 }
485 start_pte = 0;
486
487 if (address >= end_paddr)
488 break;
489 }
490 start_pmd = 0;
491 }
492}
493
494void free_initmem(void)
495{
496 unsigned long addr;
497 unsigned long init_begin = (unsigned long)__init_begin;
498 unsigned long init_end = (unsigned long)__init_end;
499
500 /* The init text pages are marked R-X. We have to
501 * flush the icache and mark them RW-
502 *
503 * This is tricky, because map_pages is in the init section.
504 * Do a dummy remap of the data section first (the data
505 * section is already PAGE_KERNEL) to pull in the TLB entries
506 * for map_kernel */
507 map_pages(init_begin, __pa(init_begin), init_end - init_begin,
508 PAGE_KERNEL_RWX, 1);
509 /* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
510 * map_pages */
511 map_pages(init_begin, __pa(init_begin), init_end - init_begin,
512 PAGE_KERNEL, 1);
513
514 /* force the kernel to see the new TLB entries */
515 __flush_tlb_range(0, init_begin, init_end);
516 /* Attempt to catch anyone trying to execute code here
517 * by filling the page with BRK insns.
518 */
519 memset((void *)init_begin, 0x00, init_end - init_begin);
520 /* finally dump all the instructions which were cached, since the
521 * pages are no-longer executable */
522 flush_icache_range(init_begin, init_end);
523
524 for (addr = init_begin; addr < init_end; addr += PAGE_SIZE) {
525 ClearPageReserved(virt_to_page(addr));
526 init_page_count(virt_to_page(addr));
527 free_page(addr);
528 num_physpages++;
529 totalram_pages++;
530 }
531
532 /* set up a new led state on systems shipped LED State panel */
533 pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
534
535 printk(KERN_INFO "Freeing unused kernel memory: %luk freed\n",
536 (init_end - init_begin) >> 10);
537}
538
539
540#ifdef CONFIG_DEBUG_RODATA
541void mark_rodata_ro(void)
542{
543 /* rodata memory was already mapped with KERNEL_RO access rights by
544 pagetable_init() and map_pages(). No need to do additional stuff here */
545 printk (KERN_INFO "Write protecting the kernel read-only data: %luk\n",
546 (unsigned long)(__end_rodata - __start_rodata) >> 10);
547}
548#endif
549
550
551/*
552 * Just an arbitrary offset to serve as a "hole" between mapping areas
553 * (between top of physical memory and a potential pcxl dma mapping
554 * area, and below the vmalloc mapping area).
555 *
556 * The current 32K value just means that there will be a 32K "hole"
557 * between mapping areas. That means that any out-of-bounds memory
558 * accesses will hopefully be caught. The vmalloc() routines leaves
559 * a hole of 4kB between each vmalloced area for the same reason.
560 */
561
562 /* Leave room for gateway page expansion */
563#if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
564#error KERNEL_MAP_START is in gateway reserved region
565#endif
566#define MAP_START (KERNEL_MAP_START)
567
568#define VM_MAP_OFFSET (32*1024)
569#define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
570 & ~(VM_MAP_OFFSET-1)))
571
572void *parisc_vmalloc_start __read_mostly;
573EXPORT_SYMBOL(parisc_vmalloc_start);
574
575#ifdef CONFIG_PA11
576unsigned long pcxl_dma_start __read_mostly;
577#endif
578
579void __init mem_init(void)
580{
581 int codesize, reservedpages, datasize, initsize;
582
583 /* Do sanity checks on page table constants */
584 BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
585 BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
586 BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
587 BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
588 > BITS_PER_LONG);
589
590 high_memory = __va((max_pfn << PAGE_SHIFT));
591
592#ifndef CONFIG_DISCONTIGMEM
593 max_mapnr = page_to_pfn(virt_to_page(high_memory - 1)) + 1;
594 totalram_pages += free_all_bootmem();
595#else
596 {
597 int i;
598
599 for (i = 0; i < npmem_ranges; i++)
600 totalram_pages += free_all_bootmem_node(NODE_DATA(i));
601 }
602#endif
603
604 codesize = (unsigned long)_etext - (unsigned long)_text;
605 datasize = (unsigned long)_edata - (unsigned long)_etext;
606 initsize = (unsigned long)__init_end - (unsigned long)__init_begin;
607
608 reservedpages = 0;
609{
610 unsigned long pfn;
611#ifdef CONFIG_DISCONTIGMEM
612 int i;
613
614 for (i = 0; i < npmem_ranges; i++) {
615 for (pfn = node_start_pfn(i); pfn < node_end_pfn(i); pfn++) {
616 if (PageReserved(pfn_to_page(pfn)))
617 reservedpages++;
618 }
619 }
620#else /* !CONFIG_DISCONTIGMEM */
621 for (pfn = 0; pfn < max_pfn; pfn++) {
622 /*
623 * Only count reserved RAM pages
624 */
625 if (PageReserved(pfn_to_page(pfn)))
626 reservedpages++;
627 }
628#endif
629}
630
631#ifdef CONFIG_PA11
632 if (hppa_dma_ops == &pcxl_dma_ops) {
633 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
634 parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
635 + PCXL_DMA_MAP_SIZE);
636 } else {
637 pcxl_dma_start = 0;
638 parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
639 }
640#else
641 parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
642#endif
643
644 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init)\n",
645 nr_free_pages() << (PAGE_SHIFT-10),
646 num_physpages << (PAGE_SHIFT-10),
647 codesize >> 10,
648 reservedpages << (PAGE_SHIFT-10),
649 datasize >> 10,
650 initsize >> 10
651 );
652
653#ifdef CONFIG_DEBUG_KERNEL /* double-sanity-check paranoia */
654 printk("virtual kernel memory layout:\n"
655 " vmalloc : 0x%p - 0x%p (%4ld MB)\n"
656 " memory : 0x%p - 0x%p (%4ld MB)\n"
657 " .init : 0x%p - 0x%p (%4ld kB)\n"
658 " .data : 0x%p - 0x%p (%4ld kB)\n"
659 " .text : 0x%p - 0x%p (%4ld kB)\n",
660
661 (void*)VMALLOC_START, (void*)VMALLOC_END,
662 (VMALLOC_END - VMALLOC_START) >> 20,
663
664 __va(0), high_memory,
665 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
666
667 __init_begin, __init_end,
668 ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
669
670 _etext, _edata,
671 ((unsigned long)_edata - (unsigned long)_etext) >> 10,
672
673 _text, _etext,
674 ((unsigned long)_etext - (unsigned long)_text) >> 10);
675#endif
676}
677
678unsigned long *empty_zero_page __read_mostly;
679EXPORT_SYMBOL(empty_zero_page);
680
681void show_mem(unsigned int filter)
682{
683 int i,free = 0,total = 0,reserved = 0;
684 int shared = 0, cached = 0;
685
686 printk(KERN_INFO "Mem-info:\n");
687 show_free_areas(filter);
688#ifndef CONFIG_DISCONTIGMEM
689 i = max_mapnr;
690 while (i-- > 0) {
691 total++;
692 if (PageReserved(mem_map+i))
693 reserved++;
694 else if (PageSwapCache(mem_map+i))
695 cached++;
696 else if (!page_count(&mem_map[i]))
697 free++;
698 else
699 shared += page_count(&mem_map[i]) - 1;
700 }
701#else
702 for (i = 0; i < npmem_ranges; i++) {
703 int j;
704
705 for (j = node_start_pfn(i); j < node_end_pfn(i); j++) {
706 struct page *p;
707 unsigned long flags;
708
709 pgdat_resize_lock(NODE_DATA(i), &flags);
710 p = nid_page_nr(i, j) - node_start_pfn(i);
711
712 total++;
713 if (PageReserved(p))
714 reserved++;
715 else if (PageSwapCache(p))
716 cached++;
717 else if (!page_count(p))
718 free++;
719 else
720 shared += page_count(p) - 1;
721 pgdat_resize_unlock(NODE_DATA(i), &flags);
722 }
723 }
724#endif
725 printk(KERN_INFO "%d pages of RAM\n", total);
726 printk(KERN_INFO "%d reserved pages\n", reserved);
727 printk(KERN_INFO "%d pages shared\n", shared);
728 printk(KERN_INFO "%d pages swap cached\n", cached);
729
730
731#ifdef CONFIG_DISCONTIGMEM
732 {
733 struct zonelist *zl;
734 int i, j;
735
736 for (i = 0; i < npmem_ranges; i++) {
737 zl = node_zonelist(i, 0);
738 for (j = 0; j < MAX_NR_ZONES; j++) {
739 struct zoneref *z;
740 struct zone *zone;
741
742 printk("Zone list for zone %d on node %d: ", j, i);
743 for_each_zone_zonelist(zone, z, zl, j)
744 printk("[%d/%s] ", zone_to_nid(zone),
745 zone->name);
746 printk("\n");
747 }
748 }
749 }
750#endif
751}
752
753/*
754 * pagetable_init() sets up the page tables
755 *
756 * Note that gateway_init() places the Linux gateway page at page 0.
757 * Since gateway pages cannot be dereferenced this has the desirable
758 * side effect of trapping those pesky NULL-reference errors in the
759 * kernel.
760 */
761static void __init pagetable_init(void)
762{
763 int range;
764
765 /* Map each physical memory range to its kernel vaddr */
766
767 for (range = 0; range < npmem_ranges; range++) {
768 unsigned long start_paddr;
769 unsigned long end_paddr;
770 unsigned long size;
771
772 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
773 end_paddr = start_paddr + (pmem_ranges[range].pages << PAGE_SHIFT);
774 size = pmem_ranges[range].pages << PAGE_SHIFT;
775
776 map_pages((unsigned long)__va(start_paddr), start_paddr,
777 size, PAGE_KERNEL, 0);
778 }
779
780#ifdef CONFIG_BLK_DEV_INITRD
781 if (initrd_end && initrd_end > mem_limit) {
782 printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
783 map_pages(initrd_start, __pa(initrd_start),
784 initrd_end - initrd_start, PAGE_KERNEL, 0);
785 }
786#endif
787
788 empty_zero_page = alloc_bootmem_pages(PAGE_SIZE);
789 memset(empty_zero_page, 0, PAGE_SIZE);
790}
791
792static void __init gateway_init(void)
793{
794 unsigned long linux_gateway_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 linux_gateway_page;
798
799 linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
800
801 /*
802 * Setup Linux Gateway page.
803 *
804 * The Linux gateway page will reside in kernel space (on virtual
805 * page 0), so it doesn't need to be aliased into user space.
806 */
807
808 map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
809 PAGE_SIZE, PAGE_GATEWAY, 1);
810}
811
812#ifdef CONFIG_HPUX
813void
814map_hpux_gateway_page(struct task_struct *tsk, struct mm_struct *mm)
815{
816 pgd_t *pg_dir;
817 pmd_t *pmd;
818 pte_t *pg_table;
819 unsigned long start_pmd;
820 unsigned long start_pte;
821 unsigned long address;
822 unsigned long hpux_gw_page_addr;
823 /* FIXME: This is 'const' in order to trick the compiler
824 into not treating it as DP-relative data. */
825 extern void * const hpux_gateway_page;
826
827 hpux_gw_page_addr = HPUX_GATEWAY_ADDR & PAGE_MASK;
828
829 /*
830 * Setup HP-UX Gateway page.
831 *
832 * The HP-UX gateway page resides in the user address space,
833 * so it needs to be aliased into each process.
834 */
835
836 pg_dir = pgd_offset(mm,hpux_gw_page_addr);
837
838#if PTRS_PER_PMD == 1
839 start_pmd = 0;
840#else
841 start_pmd = ((hpux_gw_page_addr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
842#endif
843 start_pte = ((hpux_gw_page_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
844
845 address = __pa(&hpux_gateway_page);
846#if PTRS_PER_PMD == 1
847 pmd = (pmd_t *)__pa(pg_dir);
848#else
849 pmd = (pmd_t *) pgd_address(*pg_dir);
850
851 /*
852 * pmd is physical at this point
853 */
854
855 if (!pmd) {
856 pmd = (pmd_t *) get_zeroed_page(GFP_KERNEL);
857 pmd = (pmd_t *) __pa(pmd);
858 }
859
860 __pgd_val_set(*pg_dir, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pmd);
861#endif
862 /* now change pmd to kernel virtual addresses */
863
864 pmd = (pmd_t *)__va(pmd) + start_pmd;
865
866 /*
867 * pg_table is physical at this point
868 */
869
870 pg_table = (pte_t *) pmd_address(*pmd);
871 if (!pg_table)
872 pg_table = (pte_t *) __pa(get_zeroed_page(GFP_KERNEL));
873
874 __pmd_val_set(*pmd, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pg_table);
875
876 /* now change pg_table to kernel virtual addresses */
877
878 pg_table = (pte_t *) __va(pg_table) + start_pte;
879 set_pte(pg_table, __mk_pte(address, PAGE_GATEWAY));
880}
881EXPORT_SYMBOL(map_hpux_gateway_page);
882#endif
883
884void __init paging_init(void)
885{
886 int i;
887
888 setup_bootmem();
889 pagetable_init();
890 gateway_init();
891 flush_cache_all_local(); /* start with known state */
892 flush_tlb_all_local(NULL);
893
894 for (i = 0; i < npmem_ranges; i++) {
895 unsigned long zones_size[MAX_NR_ZONES] = { 0, };
896
897 zones_size[ZONE_NORMAL] = pmem_ranges[i].pages;
898
899#ifdef CONFIG_DISCONTIGMEM
900 /* Need to initialize the pfnnid_map before we can initialize
901 the zone */
902 {
903 int j;
904 for (j = (pmem_ranges[i].start_pfn >> PFNNID_SHIFT);
905 j <= ((pmem_ranges[i].start_pfn + pmem_ranges[i].pages) >> PFNNID_SHIFT);
906 j++) {
907 pfnnid_map[j] = i;
908 }
909 }
910#endif
911
912 free_area_init_node(i, zones_size,
913 pmem_ranges[i].start_pfn, NULL);
914 }
915}
916
917#ifdef CONFIG_PA20
918
919/*
920 * Currently, all PA20 chips have 18 bit protection IDs, which is the
921 * limiting factor (space ids are 32 bits).
922 */
923
924#define NR_SPACE_IDS 262144
925
926#else
927
928/*
929 * Currently we have a one-to-one relationship between space IDs and
930 * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
931 * support 15 bit protection IDs, so that is the limiting factor.
932 * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
933 * probably not worth the effort for a special case here.
934 */
935
936#define NR_SPACE_IDS 32768
937
938#endif /* !CONFIG_PA20 */
939
940#define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
941#define SID_ARRAY_SIZE (NR_SPACE_IDS / (8 * sizeof(long)))
942
943static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
944static unsigned long dirty_space_id[SID_ARRAY_SIZE];
945static unsigned long space_id_index;
946static unsigned long free_space_ids = NR_SPACE_IDS - 1;
947static unsigned long dirty_space_ids = 0;
948
949static DEFINE_SPINLOCK(sid_lock);
950
951unsigned long alloc_sid(void)
952{
953 unsigned long index;
954
955 spin_lock(&sid_lock);
956
957 if (free_space_ids == 0) {
958 if (dirty_space_ids != 0) {
959 spin_unlock(&sid_lock);
960 flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
961 spin_lock(&sid_lock);
962 }
963 BUG_ON(free_space_ids == 0);
964 }
965
966 free_space_ids--;
967
968 index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
969 space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1)));
970 space_id_index = index;
971
972 spin_unlock(&sid_lock);
973
974 return index << SPACEID_SHIFT;
975}
976
977void free_sid(unsigned long spaceid)
978{
979 unsigned long index = spaceid >> SPACEID_SHIFT;
980 unsigned long *dirty_space_offset;
981
982 dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG);
983 index &= (BITS_PER_LONG - 1);
984
985 spin_lock(&sid_lock);
986
987 BUG_ON(*dirty_space_offset & (1L << index)); /* attempt to free space id twice */
988
989 *dirty_space_offset |= (1L << index);
990 dirty_space_ids++;
991
992 spin_unlock(&sid_lock);
993}
994
995
996#ifdef CONFIG_SMP
997static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
998{
999 int i;
1000
1001 /* NOTE: sid_lock must be held upon entry */
1002
1003 *ndirtyptr = dirty_space_ids;
1004 if (dirty_space_ids != 0) {
1005 for (i = 0; i < SID_ARRAY_SIZE; i++) {
1006 dirty_array[i] = dirty_space_id[i];
1007 dirty_space_id[i] = 0;
1008 }
1009 dirty_space_ids = 0;
1010 }
1011
1012 return;
1013}
1014
1015static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
1016{
1017 int i;
1018
1019 /* NOTE: sid_lock must be held upon entry */
1020
1021 if (ndirty != 0) {
1022 for (i = 0; i < SID_ARRAY_SIZE; i++) {
1023 space_id[i] ^= dirty_array[i];
1024 }
1025
1026 free_space_ids += ndirty;
1027 space_id_index = 0;
1028 }
1029}
1030
1031#else /* CONFIG_SMP */
1032
1033static void recycle_sids(void)
1034{
1035 int i;
1036
1037 /* NOTE: sid_lock must be held upon entry */
1038
1039 if (dirty_space_ids != 0) {
1040 for (i = 0; i < SID_ARRAY_SIZE; i++) {
1041 space_id[i] ^= dirty_space_id[i];
1042 dirty_space_id[i] = 0;
1043 }
1044
1045 free_space_ids += dirty_space_ids;
1046 dirty_space_ids = 0;
1047 space_id_index = 0;
1048 }
1049}
1050#endif
1051
1052/*
1053 * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
1054 * purged, we can safely reuse the space ids that were released but
1055 * not flushed from the tlb.
1056 */
1057
1058#ifdef CONFIG_SMP
1059
1060static unsigned long recycle_ndirty;
1061static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
1062static unsigned int recycle_inuse;
1063
1064void flush_tlb_all(void)
1065{
1066 int do_recycle;
1067
1068 do_recycle = 0;
1069 spin_lock(&sid_lock);
1070 if (dirty_space_ids > RECYCLE_THRESHOLD) {
1071 BUG_ON(recycle_inuse); /* FIXME: Use a semaphore/wait queue here */
1072 get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
1073 recycle_inuse++;
1074 do_recycle++;
1075 }
1076 spin_unlock(&sid_lock);
1077 on_each_cpu(flush_tlb_all_local, NULL, 1);
1078 if (do_recycle) {
1079 spin_lock(&sid_lock);
1080 recycle_sids(recycle_ndirty,recycle_dirty_array);
1081 recycle_inuse = 0;
1082 spin_unlock(&sid_lock);
1083 }
1084}
1085#else
1086void flush_tlb_all(void)
1087{
1088 spin_lock(&sid_lock);
1089 flush_tlb_all_local(NULL);
1090 recycle_sids();
1091 spin_unlock(&sid_lock);
1092}
1093#endif
1094
1095#ifdef CONFIG_BLK_DEV_INITRD
1096void free_initrd_mem(unsigned long start, unsigned long end)
1097{
1098 if (start >= end)
1099 return;
1100 printk(KERN_INFO "Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
1101 for (; start < end; start += PAGE_SIZE) {
1102 ClearPageReserved(virt_to_page(start));
1103 init_page_count(virt_to_page(start));
1104 free_page(start);
1105 num_physpages++;
1106 totalram_pages++;
1107 }
1108}
1109#endif