blob: 3e10152abbf0f5873194e383cfcd8845f202b252 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Initialize MMU support.
3 *
4 * Copyright (C) 1998-2003 Hewlett-Packard Co
5 * David Mosberger-Tang <davidm@hpl.hp.com>
6 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07007#include <linux/kernel.h>
8#include <linux/init.h>
9
10#include <linux/bootmem.h>
11#include <linux/efi.h>
12#include <linux/elf.h>
13#include <linux/mm.h>
14#include <linux/mmzone.h>
15#include <linux/module.h>
16#include <linux/personality.h>
17#include <linux/reboot.h>
18#include <linux/slab.h>
19#include <linux/swap.h>
20#include <linux/proc_fs.h>
21#include <linux/bitops.h>
Bob Picco139b8302007-01-30 02:11:09 -080022#include <linux/kexec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
24#include <asm/a.out.h>
25#include <asm/dma.h>
26#include <asm/ia32.h>
27#include <asm/io.h>
28#include <asm/machvec.h>
29#include <asm/numa.h>
30#include <asm/patch.h>
31#include <asm/pgalloc.h>
32#include <asm/sal.h>
33#include <asm/sections.h>
34#include <asm/system.h>
35#include <asm/tlb.h>
36#include <asm/uaccess.h>
37#include <asm/unistd.h>
38#include <asm/mca.h>
39
40DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
41
42extern void ia64_tlb_init (void);
43
44unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL;
45
46#ifdef CONFIG_VIRTUAL_MEM_MAP
47unsigned long vmalloc_end = VMALLOC_END_INIT;
48EXPORT_SYMBOL(vmalloc_end);
49struct page *vmem_map;
50EXPORT_SYMBOL(vmem_map);
51#endif
52
Robin Holtfde740e2005-04-25 13:13:16 -070053struct page *zero_page_memmap_ptr; /* map entry for zero page */
Linus Torvalds1da177e2005-04-16 15:20:36 -070054EXPORT_SYMBOL(zero_page_memmap_ptr);
55
Linus Torvalds1da177e2005-04-16 15:20:36 -070056void
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -070057__ia64_sync_icache_dcache (pte_t pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -070058{
59 unsigned long addr;
60 struct page *page;
Zhang, Yanmin5e485212006-02-23 11:07:20 +080061 unsigned long order;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 page = pte_page(pte);
64 addr = (unsigned long) page_address(page);
65
66 if (test_bit(PG_arch_1, &page->flags))
67 return; /* i-cache is already coherent with d-cache */
68
Zhang, Yanmin5e485212006-02-23 11:07:20 +080069 if (PageCompound(page)) {
Christoph Lameterd85f3382007-05-06 14:49:39 -070070 order = compound_order(page);
Zhang, Yanmin5e485212006-02-23 11:07:20 +080071 flush_icache_range(addr, addr + (1UL << order << PAGE_SHIFT));
72 }
73 else
74 flush_icache_range(addr, addr + PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 set_bit(PG_arch_1, &page->flags); /* mark page as clean */
76}
77
Jan Beulichcde14bb2007-02-05 18:46:40 -080078/*
79 * Since DMA is i-cache coherent, any (complete) pages that were written via
80 * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
81 * flush them when they get mapped into an executable vm-area.
82 */
83void
84dma_mark_clean(void *addr, size_t size)
85{
86 unsigned long pg_addr, end;
87
88 pg_addr = PAGE_ALIGN((unsigned long) addr);
89 end = (unsigned long) addr + size;
90 while (pg_addr + PAGE_SIZE <= end) {
91 struct page *page = virt_to_page(pg_addr);
92 set_bit(PG_arch_1, &page->flags);
93 pg_addr += PAGE_SIZE;
94 }
95}
96
Linus Torvalds1da177e2005-04-16 15:20:36 -070097inline void
98ia64_set_rbs_bot (void)
99{
100 unsigned long stack_size = current->signal->rlim[RLIMIT_STACK].rlim_max & -16;
101
102 if (stack_size > MAX_USER_STACK_SIZE)
103 stack_size = MAX_USER_STACK_SIZE;
KAMEZAWA Hiroyuki83d2cd32007-03-23 12:17:46 +0900104 current->thread.rbs_bot = PAGE_ALIGN(current->mm->start_stack - stack_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105}
106
107/*
108 * This performs some platform-dependent address space initialization.
109 * On IA-64, we want to setup the VM area for the register backing
110 * store (which grows upwards) and install the gateway page which is
111 * used for signal trampolines, etc.
112 */
113void
114ia64_init_addr_space (void)
115{
116 struct vm_area_struct *vma;
117
118 ia64_set_rbs_bot();
119
120 /*
121 * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore
122 * the problem. When the process attempts to write to the register backing store
123 * for the first time, it will get a SEGFAULT in this case.
124 */
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800125 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 if (vma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 vma->vm_mm = current->mm;
128 vma->vm_start = current->thread.rbs_bot & PAGE_MASK;
129 vma->vm_end = vma->vm_start + PAGE_SIZE;
130 vma->vm_page_prot = protection_map[VM_DATA_DEFAULT_FLAGS & 0x7];
Hugh Dickins46dea3d2005-10-29 18:16:20 -0700131 vma->vm_flags = VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 down_write(&current->mm->mmap_sem);
133 if (insert_vm_struct(current->mm, vma)) {
134 up_write(&current->mm->mmap_sem);
135 kmem_cache_free(vm_area_cachep, vma);
136 return;
137 }
138 up_write(&current->mm->mmap_sem);
139 }
140
141 /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */
142 if (!(current->personality & MMAP_PAGE_ZERO)) {
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800143 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 if (vma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 vma->vm_mm = current->mm;
146 vma->vm_end = PAGE_SIZE;
147 vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT);
148 vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO | VM_RESERVED;
149 down_write(&current->mm->mmap_sem);
150 if (insert_vm_struct(current->mm, vma)) {
151 up_write(&current->mm->mmap_sem);
152 kmem_cache_free(vm_area_cachep, vma);
153 return;
154 }
155 up_write(&current->mm->mmap_sem);
156 }
157 }
158}
159
160void
161free_initmem (void)
162{
163 unsigned long addr, eaddr;
164
165 addr = (unsigned long) ia64_imva(__init_begin);
166 eaddr = (unsigned long) ia64_imva(__init_end);
167 while (addr < eaddr) {
168 ClearPageReserved(virt_to_page(addr));
Nick Piggin7835e982006-03-22 00:08:40 -0800169 init_page_count(virt_to_page(addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 free_page(addr);
171 ++totalram_pages;
172 addr += PAGE_SIZE;
173 }
174 printk(KERN_INFO "Freeing unused kernel memory: %ldkB freed\n",
175 (__init_end - __init_begin) >> 10);
176}
177
Chen, Kenneth Wdae28062006-03-22 16:54:15 -0800178void __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179free_initrd_mem (unsigned long start, unsigned long end)
180{
181 struct page *page;
182 /*
183 * EFI uses 4KB pages while the kernel can use 4KB or bigger.
184 * Thus EFI and the kernel may have different page sizes. It is
185 * therefore possible to have the initrd share the same page as
186 * the end of the kernel (given current setup).
187 *
188 * To avoid freeing/using the wrong page (kernel sized) we:
189 * - align up the beginning of initrd
190 * - align down the end of initrd
191 *
192 * | |
193 * |=============| a000
194 * | |
195 * | |
196 * | | 9000
197 * |/////////////|
198 * |/////////////|
199 * |=============| 8000
200 * |///INITRD////|
201 * |/////////////|
202 * |/////////////| 7000
203 * | |
204 * |KKKKKKKKKKKKK|
205 * |=============| 6000
206 * |KKKKKKKKKKKKK|
207 * |KKKKKKKKKKKKK|
208 * K=kernel using 8KB pages
209 *
210 * In this example, we must free page 8000 ONLY. So we must align up
211 * initrd_start and keep initrd_end as is.
212 */
213 start = PAGE_ALIGN(start);
214 end = end & PAGE_MASK;
215
216 if (start < end)
217 printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10);
218
219 for (; start < end; start += PAGE_SIZE) {
220 if (!virt_addr_valid(start))
221 continue;
222 page = virt_to_page(start);
223 ClearPageReserved(page);
Nick Piggin7835e982006-03-22 00:08:40 -0800224 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 free_page(start);
226 ++totalram_pages;
227 }
228}
229
230/*
231 * This installs a clean page in the kernel's page table.
232 */
Chen, Kenneth Wdae28062006-03-22 16:54:15 -0800233static struct page * __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
235{
236 pgd_t *pgd;
237 pud_t *pud;
238 pmd_t *pmd;
239 pte_t *pte;
240
241 if (!PageReserved(page))
242 printk(KERN_ERR "put_kernel_page: page at 0x%p not in reserved memory\n",
243 page_address(page));
244
245 pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */
246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 {
248 pud = pud_alloc(&init_mm, pgd, address);
249 if (!pud)
250 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 pmd = pmd_alloc(&init_mm, pud, address);
252 if (!pmd)
253 goto out;
Hugh Dickins872fec12005-10-29 18:16:21 -0700254 pte = pte_alloc_kernel(pmd, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 if (!pte)
256 goto out;
Hugh Dickins872fec12005-10-29 18:16:21 -0700257 if (!pte_none(*pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 set_pte(pte, mk_pte(page, pgprot));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 }
Hugh Dickins872fec12005-10-29 18:16:21 -0700261 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 /* no need for flush_tlb */
263 return page;
264}
265
Chen, Kenneth W914a4ea2006-03-12 09:08:26 -0800266static void __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267setup_gate (void)
268{
269 struct page *page;
270
271 /*
David Mosberger-Tangad597bd2005-06-08 10:45:00 -0700272 * Map the gate page twice: once read-only to export the ELF
273 * headers etc. and once execute-only page to enable
274 * privilege-promotion via "epc":
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 */
276 page = virt_to_page(ia64_imva(__start_gate_section));
277 put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
278#ifdef HAVE_BUGGY_SEGREL
279 page = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE));
280 put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
281#else
282 put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
David Mosberger-Tangad597bd2005-06-08 10:45:00 -0700283 /* Fill in the holes (if any) with read-only zero pages: */
284 {
285 unsigned long addr;
286
287 for (addr = GATE_ADDR + PAGE_SIZE;
288 addr < GATE_ADDR + PERCPU_PAGE_SIZE;
289 addr += PAGE_SIZE)
290 {
291 put_kernel_page(ZERO_PAGE(0), addr,
292 PAGE_READONLY);
293 put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE,
294 PAGE_READONLY);
295 }
296 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297#endif
298 ia64_patch_gate();
299}
300
301void __devinit
302ia64_mmu_init (void *my_cpu_data)
303{
Chen, Kenneth W00b65982006-10-13 10:08:13 -0700304 unsigned long pta, impl_va_bits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 extern void __devinit tlb_init (void);
306
307#ifdef CONFIG_DISABLE_VHPT
308# define VHPT_ENABLE_BIT 0
309#else
310# define VHPT_ENABLE_BIT 1
311#endif
312
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 /*
314 * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
315 * address space. The IA-64 architecture guarantees that at least 50 bits of
316 * virtual address space are implemented but if we pick a large enough page size
317 * (e.g., 64KB), the mapped address space is big enough that it will overlap with
318 * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages,
319 * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
320 * problem in practice. Alternatively, we could truncate the top of the mapped
321 * address space to not permit mappings that would overlap with the VMLPT.
322 * --davidm 00/12/06
323 */
324# define pte_bits 3
325# define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
326 /*
327 * The virtual page table has to cover the entire implemented address space within
328 * a region even though not all of this space may be mappable. The reason for
329 * this is that the Access bit and Dirty bit fault handlers perform
330 * non-speculative accesses to the virtual page table, so the address range of the
331 * virtual page table itself needs to be covered by virtual page table.
332 */
333# define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits)
334# define POW2(n) (1ULL << (n))
335
336 impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
337
338 if (impl_va_bits < 51 || impl_va_bits > 61)
339 panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
Peter Chubb6cf07a82005-08-23 20:07:00 -0700340 /*
341 * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
342 * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
343 * the test makes sure that our mapped space doesn't overlap the
344 * unimplemented hole in the middle of the region.
345 */
346 if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
347 (mapped_space_bits > impl_va_bits - 1))
348 panic("Cannot build a big enough virtual-linear page table"
349 " to cover mapped address space.\n"
350 " Try using a smaller page size.\n");
351
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
353 /* place the VMLPT at the end of each page-table mapped region: */
354 pta = POW2(61) - POW2(vmlpt_bits);
355
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 /*
357 * Set the (virtually mapped linear) page table address. Bit
358 * 8 selects between the short and long format, bits 2-7 the
359 * size of the table, and bit 0 whether the VHPT walker is
360 * enabled.
361 */
362 ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
363
364 ia64_tlb_init();
365
366#ifdef CONFIG_HUGETLB_PAGE
367 ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
368 ia64_srlz_d();
369#endif
370}
371
372#ifdef CONFIG_VIRTUAL_MEM_MAP
Bob Piccoe44e41d2006-06-28 12:55:43 -0400373int vmemmap_find_next_valid_pfn(int node, int i)
374{
375 unsigned long end_address, hole_next_pfn;
376 unsigned long stop_address;
377 pg_data_t *pgdat = NODE_DATA(node);
378
379 end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
380 end_address = PAGE_ALIGN(end_address);
381
382 stop_address = (unsigned long) &vmem_map[
383 pgdat->node_start_pfn + pgdat->node_spanned_pages];
384
385 do {
386 pgd_t *pgd;
387 pud_t *pud;
388 pmd_t *pmd;
389 pte_t *pte;
390
391 pgd = pgd_offset_k(end_address);
392 if (pgd_none(*pgd)) {
393 end_address += PGDIR_SIZE;
394 continue;
395 }
396
397 pud = pud_offset(pgd, end_address);
398 if (pud_none(*pud)) {
399 end_address += PUD_SIZE;
400 continue;
401 }
402
403 pmd = pmd_offset(pud, end_address);
404 if (pmd_none(*pmd)) {
405 end_address += PMD_SIZE;
406 continue;
407 }
408
409 pte = pte_offset_kernel(pmd, end_address);
410retry_pte:
411 if (pte_none(*pte)) {
412 end_address += PAGE_SIZE;
413 pte++;
414 if ((end_address < stop_address) &&
415 (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
416 goto retry_pte;
417 continue;
418 }
419 /* Found next valid vmem_map page */
420 break;
421 } while (end_address < stop_address);
422
423 end_address = min(end_address, stop_address);
424 end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
425 hole_next_pfn = end_address / sizeof(struct page);
426 return hole_next_pfn - pgdat->node_start_pfn;
427}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
Chen, Kenneth Wdae28062006-03-22 16:54:15 -0800429int __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430create_mem_map_page_table (u64 start, u64 end, void *arg)
431{
432 unsigned long address, start_page, end_page;
433 struct page *map_start, *map_end;
434 int node;
435 pgd_t *pgd;
436 pud_t *pud;
437 pmd_t *pmd;
438 pte_t *pte;
439
440 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
441 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
442
443 start_page = (unsigned long) map_start & PAGE_MASK;
444 end_page = PAGE_ALIGN((unsigned long) map_end);
445 node = paddr_to_nid(__pa(start));
446
447 for (address = start_page; address < end_page; address += PAGE_SIZE) {
448 pgd = pgd_offset_k(address);
449 if (pgd_none(*pgd))
450 pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
451 pud = pud_offset(pgd, address);
452
453 if (pud_none(*pud))
454 pud_populate(&init_mm, pud, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
455 pmd = pmd_offset(pud, address);
456
457 if (pmd_none(*pmd))
458 pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
459 pte = pte_offset_kernel(pmd, address);
460
461 if (pte_none(*pte))
462 set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
463 PAGE_KERNEL));
464 }
465 return 0;
466}
467
468struct memmap_init_callback_data {
469 struct page *start;
470 struct page *end;
471 int nid;
472 unsigned long zone;
473};
474
475static int
476virtual_memmap_init (u64 start, u64 end, void *arg)
477{
478 struct memmap_init_callback_data *args;
479 struct page *map_start, *map_end;
480
481 args = (struct memmap_init_callback_data *) arg;
482 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
483 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
484
485 if (map_start < args->start)
486 map_start = args->start;
487 if (map_end > args->end)
488 map_end = args->end;
489
490 /*
491 * We have to initialize "out of bounds" struct page elements that fit completely
492 * on the same pages that were allocated for the "in bounds" elements because they
493 * may be referenced later (and found to be "reserved").
494 */
495 map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
496 map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
497 / sizeof(struct page));
498
499 if (map_start < map_end)
500 memmap_init_zone((unsigned long)(map_end - map_start),
Dave Hansena2f3aa02007-01-10 23:15:30 -0800501 args->nid, args->zone, page_to_pfn(map_start),
502 MEMMAP_EARLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 return 0;
504}
505
506void
507memmap_init (unsigned long size, int nid, unsigned long zone,
508 unsigned long start_pfn)
509{
510 if (!vmem_map)
Dave Hansena2f3aa02007-01-10 23:15:30 -0800511 memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 else {
513 struct page *start;
514 struct memmap_init_callback_data args;
515
516 start = pfn_to_page(start_pfn);
517 args.start = start;
518 args.end = start + size;
519 args.nid = nid;
520 args.zone = zone;
521
522 efi_memmap_walk(virtual_memmap_init, &args);
523 }
524}
525
526int
527ia64_pfn_valid (unsigned long pfn)
528{
529 char byte;
530 struct page *pg = pfn_to_page(pfn);
531
532 return (__get_user(byte, (char __user *) pg) == 0)
533 && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
534 || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
535}
536EXPORT_SYMBOL(ia64_pfn_valid);
537
Chen, Kenneth Wdae28062006-03-22 16:54:15 -0800538int __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539find_largest_hole (u64 start, u64 end, void *arg)
540{
541 u64 *max_gap = arg;
542
543 static u64 last_end = PAGE_OFFSET;
544
545 /* NOTE: this algorithm assumes efi memmap table is ordered */
546
547 if (*max_gap < (start - last_end))
548 *max_gap = start - last_end;
549 last_end = end;
550 return 0;
551}
Mel Gorman05e0caa2006-09-27 01:49:54 -0700552
Bob Picco139b8302007-01-30 02:11:09 -0800553#endif /* CONFIG_VIRTUAL_MEM_MAP */
554
Mel Gorman05e0caa2006-09-27 01:49:54 -0700555int __init
Tony Luck8b9c1062006-12-12 11:18:55 -0800556register_active_ranges(u64 start, u64 end, void *arg)
Mel Gorman05e0caa2006-09-27 01:49:54 -0700557{
Bob Picco139b8302007-01-30 02:11:09 -0800558 int nid = paddr_to_nid(__pa(start));
559
560 if (nid < 0)
561 nid = 0;
562#ifdef CONFIG_KEXEC
563 if (start > crashk_res.start && start < crashk_res.end)
564 start = crashk_res.end;
565 if (end > crashk_res.start && end < crashk_res.end)
566 end = crashk_res.start;
567#endif
568
569 if (start < end)
570 add_active_range(nid, __pa(start) >> PAGE_SHIFT,
571 __pa(end) >> PAGE_SHIFT);
Mel Gorman05e0caa2006-09-27 01:49:54 -0700572 return 0;
573}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
Chen, Kenneth Wdae28062006-03-22 16:54:15 -0800575static int __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576count_reserved_pages (u64 start, u64 end, void *arg)
577{
578 unsigned long num_reserved = 0;
579 unsigned long *count = arg;
580
581 for (; start < end; start += PAGE_SIZE)
582 if (PageReserved(virt_to_page(start)))
583 ++num_reserved;
584 *count += num_reserved;
585 return 0;
586}
587
Zou Nan haia3f5c332007-03-20 13:41:57 -0700588int
589find_max_min_low_pfn (unsigned long start, unsigned long end, void *arg)
590{
591 unsigned long pfn_start, pfn_end;
592#ifdef CONFIG_FLATMEM
593 pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
594 pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
595#else
596 pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
597 pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
598#endif
599 min_low_pfn = min(min_low_pfn, pfn_start);
600 max_low_pfn = max(max_low_pfn, pfn_end);
601 return 0;
602}
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604/*
605 * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
606 * system call handler. When this option is in effect, all fsyscalls will end up bubbling
607 * down into the kernel and calling the normal (heavy-weight) syscall handler. This is
608 * useful for performance testing, but conceivably could also come in handy for debugging
609 * purposes.
610 */
611
Chen, Kenneth W03906ea2006-03-12 09:10:59 -0800612static int nolwsys __initdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
614static int __init
615nolwsys_setup (char *s)
616{
617 nolwsys = 1;
618 return 1;
619}
620
621__setup("nolwsys", nolwsys_setup);
622
Chen, Kenneth Wdae28062006-03-22 16:54:15 -0800623void __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624mem_init (void)
625{
626 long reserved_pages, codesize, datasize, initsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 pg_data_t *pgdat;
628 int i;
629 static struct kcore_list kcore_mem, kcore_vmem, kcore_kernel;
630
Robin Holtfde740e2005-04-25 13:13:16 -0700631 BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
632 BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
633 BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
634
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635#ifdef CONFIG_PCI
636 /*
637 * This needs to be called _after_ the command line has been parsed but _before_
638 * any drivers that may need the PCI DMA interface are initialized or bootmem has
639 * been freed.
640 */
641 platform_dma_init();
642#endif
643
Bob Picco2d4b1fa2005-10-04 15:13:57 -0400644#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 if (!mem_map)
646 BUG();
647 max_mapnr = max_low_pfn;
648#endif
649
650 high_memory = __va(max_low_pfn * PAGE_SIZE);
651
652 kclist_add(&kcore_mem, __va(0), max_low_pfn * PAGE_SIZE);
653 kclist_add(&kcore_vmem, (void *)VMALLOC_START, VMALLOC_END-VMALLOC_START);
654 kclist_add(&kcore_kernel, _stext, _end - _stext);
655
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -0800656 for_each_online_pgdat(pgdat)
bob.picco564601a2005-06-30 09:52:00 -0700657 if (pgdat->bdata->node_bootmem_map)
658 totalram_pages += free_all_bootmem_node(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
660 reserved_pages = 0;
661 efi_memmap_walk(count_reserved_pages, &reserved_pages);
662
663 codesize = (unsigned long) _etext - (unsigned long) _stext;
664 datasize = (unsigned long) _edata - (unsigned long) _etext;
665 initsize = (unsigned long) __init_end - (unsigned long) __init_begin;
666
667 printk(KERN_INFO "Memory: %luk/%luk available (%luk code, %luk reserved, "
668 "%luk data, %luk init)\n", (unsigned long) nr_free_pages() << (PAGE_SHIFT - 10),
669 num_physpages << (PAGE_SHIFT - 10), codesize >> 10,
670 reserved_pages << (PAGE_SHIFT - 10), datasize >> 10, initsize >> 10);
671
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
673 /*
674 * For fsyscall entrpoints with no light-weight handler, use the ordinary
675 * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
676 * code can tell them apart.
677 */
678 for (i = 0; i < NR_syscalls; ++i) {
679 extern unsigned long fsyscall_table[NR_syscalls];
680 extern unsigned long sys_call_table[NR_syscalls];
681
682 if (!fsyscall_table[i] || nolwsys)
683 fsyscall_table[i] = sys_call_table[i] | 1;
684 }
685 setup_gate();
686
687#ifdef CONFIG_IA32_SUPPORT
688 ia32_mem_init();
689#endif
690}
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900691
692#ifdef CONFIG_MEMORY_HOTPLUG
693void online_page(struct page *page)
694{
695 ClearPageReserved(page);
Nick Piggin7835e982006-03-22 00:08:40 -0800696 init_page_count(page);
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900697 __free_page(page);
698 totalram_pages++;
699 num_physpages++;
700}
701
Yasunori Gotobc02af92006-06-27 02:53:30 -0700702int arch_add_memory(int nid, u64 start, u64 size)
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900703{
704 pg_data_t *pgdat;
705 struct zone *zone;
706 unsigned long start_pfn = start >> PAGE_SHIFT;
707 unsigned long nr_pages = size >> PAGE_SHIFT;
708 int ret;
709
Yasunori Gotobc02af92006-06-27 02:53:30 -0700710 pgdat = NODE_DATA(nid);
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900711
712 zone = pgdat->node_zones + ZONE_NORMAL;
713 ret = __add_pages(zone, start_pfn, nr_pages);
714
715 if (ret)
716 printk("%s: Problem encountered in __add_pages() as ret=%d\n",
717 __FUNCTION__, ret);
718
719 return ret;
720}
KAMEZAWA Hiroyuki48e94192007-10-16 01:26:14 -0700721#ifdef CONFIG_MEMORY_HOTREMOVE
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900722int remove_memory(u64 start, u64 size)
723{
KAMEZAWA Hiroyukide33b822007-10-16 01:26:13 -0700724 unsigned long start_pfn, end_pfn;
725 unsigned long timeout = 120 * HZ;
726 int ret;
727 start_pfn = start >> PAGE_SHIFT;
728 end_pfn = start_pfn + (size >> PAGE_SHIFT);
729 ret = offline_pages(start_pfn, end_pfn, timeout);
730 if (ret)
731 goto out;
732 /* we can free mem_map at this point */
733out:
734 return ret;
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900735}
KAMEZAWA Hiroyuki9c576ff2006-04-27 05:25:00 -0400736EXPORT_SYMBOL_GPL(remove_memory);
KAMEZAWA Hiroyuki48e94192007-10-16 01:26:14 -0700737#endif /* CONFIG_MEMORY_HOTREMOVE */
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900738#endif