blob: ca3335ea56ccf9dae52d19ddbe619315f1134225 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <asm/dma.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <asm/io.h>
26#include <asm/machvec.h>
27#include <asm/numa.h>
28#include <asm/patch.h>
29#include <asm/pgalloc.h>
30#include <asm/sal.h>
31#include <asm/sections.h>
32#include <asm/system.h>
33#include <asm/tlb.h>
34#include <asm/uaccess.h>
35#include <asm/unistd.h>
36#include <asm/mca.h>
Isaku Yamahatadd97d5c2009-03-04 21:05:34 +090037#include <asm/paravirt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
40
41extern void ia64_tlb_init (void);
42
43unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL;
44
45#ifdef CONFIG_VIRTUAL_MEM_MAP
Tejun Heo126b3fc2009-10-02 13:28:55 +090046unsigned long VMALLOC_END = VMALLOC_END_INIT;
47EXPORT_SYMBOL(VMALLOC_END);
Linus Torvalds1da177e2005-04-16 15:20:36 -070048struct page *vmem_map;
49EXPORT_SYMBOL(vmem_map);
50#endif
51
Robin Holtfde740e2005-04-25 13:13:16 -070052struct page *zero_page_memmap_ptr; /* map entry for zero page */
Linus Torvalds1da177e2005-04-16 15:20:36 -070053EXPORT_SYMBOL(zero_page_memmap_ptr);
54
Linus Torvalds1da177e2005-04-16 15:20:36 -070055void
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -070056__ia64_sync_icache_dcache (pte_t pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -070057{
58 unsigned long addr;
59 struct page *page;
60
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 page = pte_page(pte);
62 addr = (unsigned long) page_address(page);
63
64 if (test_bit(PG_arch_1, &page->flags))
65 return; /* i-cache is already coherent with d-cache */
66
Christoph Lameter273988f2008-04-09 13:05:41 -070067 flush_icache_range(addr, addr + (PAGE_SIZE << compound_order(page)));
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 set_bit(PG_arch_1, &page->flags); /* mark page as clean */
69}
70
Jan Beulichcde14bb2007-02-05 18:46:40 -080071/*
72 * Since DMA is i-cache coherent, any (complete) pages that were written via
73 * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
74 * flush them when they get mapped into an executable vm-area.
75 */
76void
77dma_mark_clean(void *addr, size_t size)
78{
79 unsigned long pg_addr, end;
80
81 pg_addr = PAGE_ALIGN((unsigned long) addr);
82 end = (unsigned long) addr + size;
83 while (pg_addr + PAGE_SIZE <= end) {
84 struct page *page = virt_to_page(pg_addr);
85 set_bit(PG_arch_1, &page->flags);
86 pg_addr += PAGE_SIZE;
87 }
88}
89
Linus Torvalds1da177e2005-04-16 15:20:36 -070090inline void
91ia64_set_rbs_bot (void)
92{
Jiri Slaby02b763b2010-01-06 16:24:30 +010093 unsigned long stack_size = rlimit_max(RLIMIT_STACK) & -16;
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
95 if (stack_size > MAX_USER_STACK_SIZE)
96 stack_size = MAX_USER_STACK_SIZE;
KAMEZAWA Hiroyuki83d2cd32007-03-23 12:17:46 +090097 current->thread.rbs_bot = PAGE_ALIGN(current->mm->start_stack - stack_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098}
99
100/*
101 * This performs some platform-dependent address space initialization.
102 * On IA-64, we want to setup the VM area for the register backing
103 * store (which grows upwards) and install the gateway page which is
104 * used for signal trampolines, etc.
105 */
106void
107ia64_init_addr_space (void)
108{
109 struct vm_area_struct *vma;
110
111 ia64_set_rbs_bot();
112
113 /*
114 * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore
115 * the problem. When the process attempts to write to the register backing store
116 * for the first time, it will get a SEGFAULT in this case.
117 */
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800118 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 if (vma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 vma->vm_mm = current->mm;
121 vma->vm_start = current->thread.rbs_bot & PAGE_MASK;
122 vma->vm_end = vma->vm_start + PAGE_SIZE;
Hugh Dickins46dea3d2005-10-29 18:16:20 -0700123 vma->vm_flags = VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT;
Coly Li3ed75eb2007-10-18 23:39:15 -0700124 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 down_write(&current->mm->mmap_sem);
126 if (insert_vm_struct(current->mm, vma)) {
127 up_write(&current->mm->mmap_sem);
128 kmem_cache_free(vm_area_cachep, vma);
129 return;
130 }
131 up_write(&current->mm->mmap_sem);
132 }
133
134 /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */
135 if (!(current->personality & MMAP_PAGE_ZERO)) {
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800136 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 if (vma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 vma->vm_mm = current->mm;
139 vma->vm_end = PAGE_SIZE;
140 vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT);
141 vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO | VM_RESERVED;
142 down_write(&current->mm->mmap_sem);
143 if (insert_vm_struct(current->mm, vma)) {
144 up_write(&current->mm->mmap_sem);
145 kmem_cache_free(vm_area_cachep, vma);
146 return;
147 }
148 up_write(&current->mm->mmap_sem);
149 }
150 }
151}
152
153void
154free_initmem (void)
155{
156 unsigned long addr, eaddr;
157
158 addr = (unsigned long) ia64_imva(__init_begin);
159 eaddr = (unsigned long) ia64_imva(__init_end);
160 while (addr < eaddr) {
161 ClearPageReserved(virt_to_page(addr));
Nick Piggin7835e982006-03-22 00:08:40 -0800162 init_page_count(virt_to_page(addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 free_page(addr);
164 ++totalram_pages;
165 addr += PAGE_SIZE;
166 }
167 printk(KERN_INFO "Freeing unused kernel memory: %ldkB freed\n",
168 (__init_end - __init_begin) >> 10);
169}
170
Chen, Kenneth Wdae28062006-03-22 16:54:15 -0800171void __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172free_initrd_mem (unsigned long start, unsigned long end)
173{
174 struct page *page;
175 /*
176 * EFI uses 4KB pages while the kernel can use 4KB or bigger.
177 * Thus EFI and the kernel may have different page sizes. It is
178 * therefore possible to have the initrd share the same page as
179 * the end of the kernel (given current setup).
180 *
181 * To avoid freeing/using the wrong page (kernel sized) we:
182 * - align up the beginning of initrd
183 * - align down the end of initrd
184 *
185 * | |
186 * |=============| a000
187 * | |
188 * | |
189 * | | 9000
190 * |/////////////|
191 * |/////////////|
192 * |=============| 8000
193 * |///INITRD////|
194 * |/////////////|
195 * |/////////////| 7000
196 * | |
197 * |KKKKKKKKKKKKK|
198 * |=============| 6000
199 * |KKKKKKKKKKKKK|
200 * |KKKKKKKKKKKKK|
201 * K=kernel using 8KB pages
202 *
203 * In this example, we must free page 8000 ONLY. So we must align up
204 * initrd_start and keep initrd_end as is.
205 */
206 start = PAGE_ALIGN(start);
207 end = end & PAGE_MASK;
208
209 if (start < end)
210 printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10);
211
212 for (; start < end; start += PAGE_SIZE) {
213 if (!virt_addr_valid(start))
214 continue;
215 page = virt_to_page(start);
216 ClearPageReserved(page);
Nick Piggin7835e982006-03-22 00:08:40 -0800217 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 free_page(start);
219 ++totalram_pages;
220 }
221}
222
223/*
224 * This installs a clean page in the kernel's page table.
225 */
Chen, Kenneth Wdae28062006-03-22 16:54:15 -0800226static struct page * __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
228{
229 pgd_t *pgd;
230 pud_t *pud;
231 pmd_t *pmd;
232 pte_t *pte;
233
234 if (!PageReserved(page))
235 printk(KERN_ERR "put_kernel_page: page at 0x%p not in reserved memory\n",
236 page_address(page));
237
238 pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */
239
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 {
241 pud = pud_alloc(&init_mm, pgd, address);
242 if (!pud)
243 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 pmd = pmd_alloc(&init_mm, pud, address);
245 if (!pmd)
246 goto out;
Hugh Dickins872fec12005-10-29 18:16:21 -0700247 pte = pte_alloc_kernel(pmd, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 if (!pte)
249 goto out;
Hugh Dickins872fec12005-10-29 18:16:21 -0700250 if (!pte_none(*pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 set_pte(pte, mk_pte(page, pgprot));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 }
Hugh Dickins872fec12005-10-29 18:16:21 -0700254 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 /* no need for flush_tlb */
256 return page;
257}
258
Chen, Kenneth W914a4ea2006-03-12 09:08:26 -0800259static void __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260setup_gate (void)
261{
Isaku Yamahatae4ff5b82009-03-04 21:05:42 +0900262 void *gate_section;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 struct page *page;
264
265 /*
David Mosberger-Tangad597bd2005-06-08 10:45:00 -0700266 * Map the gate page twice: once read-only to export the ELF
267 * headers etc. and once execute-only page to enable
268 * privilege-promotion via "epc":
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 */
Isaku Yamahatae4ff5b82009-03-04 21:05:42 +0900270 gate_section = paravirt_get_gate_section();
271 page = virt_to_page(ia64_imva(gate_section));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
273#ifdef HAVE_BUGGY_SEGREL
Isaku Yamahatae4ff5b82009-03-04 21:05:42 +0900274 page = virt_to_page(ia64_imva(gate_section + PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
276#else
277 put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
David Mosberger-Tangad597bd2005-06-08 10:45:00 -0700278 /* Fill in the holes (if any) with read-only zero pages: */
279 {
280 unsigned long addr;
281
282 for (addr = GATE_ADDR + PAGE_SIZE;
283 addr < GATE_ADDR + PERCPU_PAGE_SIZE;
284 addr += PAGE_SIZE)
285 {
286 put_kernel_page(ZERO_PAGE(0), addr,
287 PAGE_READONLY);
288 put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE,
289 PAGE_READONLY);
290 }
291 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292#endif
293 ia64_patch_gate();
294}
295
296void __devinit
297ia64_mmu_init (void *my_cpu_data)
298{
Chen, Kenneth W00b65982006-10-13 10:08:13 -0700299 unsigned long pta, impl_va_bits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 extern void __devinit tlb_init (void);
301
302#ifdef CONFIG_DISABLE_VHPT
303# define VHPT_ENABLE_BIT 0
304#else
305# define VHPT_ENABLE_BIT 1
306#endif
307
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 /*
309 * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
310 * address space. The IA-64 architecture guarantees that at least 50 bits of
311 * virtual address space are implemented but if we pick a large enough page size
312 * (e.g., 64KB), the mapped address space is big enough that it will overlap with
313 * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages,
314 * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
315 * problem in practice. Alternatively, we could truncate the top of the mapped
316 * address space to not permit mappings that would overlap with the VMLPT.
317 * --davidm 00/12/06
318 */
319# define pte_bits 3
320# define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
321 /*
322 * The virtual page table has to cover the entire implemented address space within
323 * a region even though not all of this space may be mappable. The reason for
324 * this is that the Access bit and Dirty bit fault handlers perform
325 * non-speculative accesses to the virtual page table, so the address range of the
326 * virtual page table itself needs to be covered by virtual page table.
327 */
328# define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits)
329# define POW2(n) (1ULL << (n))
330
331 impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
332
333 if (impl_va_bits < 51 || impl_va_bits > 61)
334 panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
Peter Chubb6cf07a82005-08-23 20:07:00 -0700335 /*
336 * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
337 * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
338 * the test makes sure that our mapped space doesn't overlap the
339 * unimplemented hole in the middle of the region.
340 */
341 if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
342 (mapped_space_bits > impl_va_bits - 1))
343 panic("Cannot build a big enough virtual-linear page table"
344 " to cover mapped address space.\n"
345 " Try using a smaller page size.\n");
346
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
348 /* place the VMLPT at the end of each page-table mapped region: */
349 pta = POW2(61) - POW2(vmlpt_bits);
350
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 /*
352 * Set the (virtually mapped linear) page table address. Bit
353 * 8 selects between the short and long format, bits 2-7 the
354 * size of the table, and bit 0 whether the VHPT walker is
355 * enabled.
356 */
357 ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
358
359 ia64_tlb_init();
360
361#ifdef CONFIG_HUGETLB_PAGE
362 ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
363 ia64_srlz_d();
364#endif
365}
366
367#ifdef CONFIG_VIRTUAL_MEM_MAP
Bob Piccoe44e41d2006-06-28 12:55:43 -0400368int vmemmap_find_next_valid_pfn(int node, int i)
369{
370 unsigned long end_address, hole_next_pfn;
371 unsigned long stop_address;
372 pg_data_t *pgdat = NODE_DATA(node);
373
374 end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
375 end_address = PAGE_ALIGN(end_address);
376
377 stop_address = (unsigned long) &vmem_map[
378 pgdat->node_start_pfn + pgdat->node_spanned_pages];
379
380 do {
381 pgd_t *pgd;
382 pud_t *pud;
383 pmd_t *pmd;
384 pte_t *pte;
385
386 pgd = pgd_offset_k(end_address);
387 if (pgd_none(*pgd)) {
388 end_address += PGDIR_SIZE;
389 continue;
390 }
391
392 pud = pud_offset(pgd, end_address);
393 if (pud_none(*pud)) {
394 end_address += PUD_SIZE;
395 continue;
396 }
397
398 pmd = pmd_offset(pud, end_address);
399 if (pmd_none(*pmd)) {
400 end_address += PMD_SIZE;
401 continue;
402 }
403
404 pte = pte_offset_kernel(pmd, end_address);
405retry_pte:
406 if (pte_none(*pte)) {
407 end_address += PAGE_SIZE;
408 pte++;
409 if ((end_address < stop_address) &&
410 (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
411 goto retry_pte;
412 continue;
413 }
414 /* Found next valid vmem_map page */
415 break;
416 } while (end_address < stop_address);
417
418 end_address = min(end_address, stop_address);
419 end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
420 hole_next_pfn = end_address / sizeof(struct page);
421 return hole_next_pfn - pgdat->node_start_pfn;
422}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
Matthew Wilcoxe088a4a2009-05-22 13:49:49 -0700424int __init create_mem_map_page_table(u64 start, u64 end, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425{
426 unsigned long address, start_page, end_page;
427 struct page *map_start, *map_end;
428 int node;
429 pgd_t *pgd;
430 pud_t *pud;
431 pmd_t *pmd;
432 pte_t *pte;
433
434 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
435 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
436
437 start_page = (unsigned long) map_start & PAGE_MASK;
438 end_page = PAGE_ALIGN((unsigned long) map_end);
439 node = paddr_to_nid(__pa(start));
440
441 for (address = start_page; address < end_page; address += PAGE_SIZE) {
442 pgd = pgd_offset_k(address);
443 if (pgd_none(*pgd))
444 pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
445 pud = pud_offset(pgd, address);
446
447 if (pud_none(*pud))
448 pud_populate(&init_mm, pud, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
449 pmd = pmd_offset(pud, address);
450
451 if (pmd_none(*pmd))
452 pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
453 pte = pte_offset_kernel(pmd, address);
454
455 if (pte_none(*pte))
456 set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
457 PAGE_KERNEL));
458 }
459 return 0;
460}
461
462struct memmap_init_callback_data {
463 struct page *start;
464 struct page *end;
465 int nid;
466 unsigned long zone;
467};
468
Adrian Bunk18b8bef2007-10-29 13:49:47 +0100469static int __meminit
Matthew Wilcoxe088a4a2009-05-22 13:49:49 -0700470virtual_memmap_init(u64 start, u64 end, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
472 struct memmap_init_callback_data *args;
473 struct page *map_start, *map_end;
474
475 args = (struct memmap_init_callback_data *) arg;
476 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
477 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
478
479 if (map_start < args->start)
480 map_start = args->start;
481 if (map_end > args->end)
482 map_end = args->end;
483
484 /*
485 * We have to initialize "out of bounds" struct page elements that fit completely
486 * on the same pages that were allocated for the "in bounds" elements because they
487 * may be referenced later (and found to be "reserved").
488 */
489 map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
490 map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
491 / sizeof(struct page));
492
493 if (map_start < map_end)
494 memmap_init_zone((unsigned long)(map_end - map_start),
Dave Hansena2f3aa02007-01-10 23:15:30 -0800495 args->nid, args->zone, page_to_pfn(map_start),
496 MEMMAP_EARLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 return 0;
498}
499
Adrian Bunk18b8bef2007-10-29 13:49:47 +0100500void __meminit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501memmap_init (unsigned long size, int nid, unsigned long zone,
502 unsigned long start_pfn)
503{
504 if (!vmem_map)
Dave Hansena2f3aa02007-01-10 23:15:30 -0800505 memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 else {
507 struct page *start;
508 struct memmap_init_callback_data args;
509
510 start = pfn_to_page(start_pfn);
511 args.start = start;
512 args.end = start + size;
513 args.nid = nid;
514 args.zone = zone;
515
516 efi_memmap_walk(virtual_memmap_init, &args);
517 }
518}
519
520int
521ia64_pfn_valid (unsigned long pfn)
522{
523 char byte;
524 struct page *pg = pfn_to_page(pfn);
525
526 return (__get_user(byte, (char __user *) pg) == 0)
527 && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
528 || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
529}
530EXPORT_SYMBOL(ia64_pfn_valid);
531
Matthew Wilcoxe088a4a2009-05-22 13:49:49 -0700532int __init find_largest_hole(u64 start, u64 end, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533{
534 u64 *max_gap = arg;
535
536 static u64 last_end = PAGE_OFFSET;
537
538 /* NOTE: this algorithm assumes efi memmap table is ordered */
539
540 if (*max_gap < (start - last_end))
541 *max_gap = start - last_end;
542 last_end = end;
543 return 0;
544}
Mel Gorman05e0caa2006-09-27 01:49:54 -0700545
Bob Picco139b8302007-01-30 02:11:09 -0800546#endif /* CONFIG_VIRTUAL_MEM_MAP */
547
Matthew Wilcoxe088a4a2009-05-22 13:49:49 -0700548int __init register_active_ranges(u64 start, u64 len, int nid)
Mel Gorman05e0caa2006-09-27 01:49:54 -0700549{
Zoltan Menyhart98075d22008-04-11 15:21:35 -0700550 u64 end = start + len;
Bob Picco139b8302007-01-30 02:11:09 -0800551
Bob Picco139b8302007-01-30 02:11:09 -0800552#ifdef CONFIG_KEXEC
553 if (start > crashk_res.start && start < crashk_res.end)
554 start = crashk_res.end;
555 if (end > crashk_res.start && end < crashk_res.end)
556 end = crashk_res.start;
557#endif
558
559 if (start < end)
560 add_active_range(nid, __pa(start) >> PAGE_SHIFT,
561 __pa(end) >> PAGE_SHIFT);
Mel Gorman05e0caa2006-09-27 01:49:54 -0700562 return 0;
563}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
Chen, Kenneth Wdae28062006-03-22 16:54:15 -0800565static int __init
Matthew Wilcoxe088a4a2009-05-22 13:49:49 -0700566count_reserved_pages(u64 start, u64 end, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567{
568 unsigned long num_reserved = 0;
569 unsigned long *count = arg;
570
571 for (; start < end; start += PAGE_SIZE)
572 if (PageReserved(virt_to_page(start)))
573 ++num_reserved;
574 *count += num_reserved;
575 return 0;
576}
577
Zou Nan haia3f5c332007-03-20 13:41:57 -0700578int
Matthew Wilcoxe088a4a2009-05-22 13:49:49 -0700579find_max_min_low_pfn (u64 start, u64 end, void *arg)
Zou Nan haia3f5c332007-03-20 13:41:57 -0700580{
581 unsigned long pfn_start, pfn_end;
582#ifdef CONFIG_FLATMEM
583 pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
584 pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
585#else
586 pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
587 pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
588#endif
589 min_low_pfn = min(min_low_pfn, pfn_start);
590 max_low_pfn = max(max_low_pfn, pfn_end);
591 return 0;
592}
593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594/*
595 * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
596 * system call handler. When this option is in effect, all fsyscalls will end up bubbling
597 * down into the kernel and calling the normal (heavy-weight) syscall handler. This is
598 * useful for performance testing, but conceivably could also come in handy for debugging
599 * purposes.
600 */
601
Chen, Kenneth W03906ea2006-03-12 09:10:59 -0800602static int nolwsys __initdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
604static int __init
605nolwsys_setup (char *s)
606{
607 nolwsys = 1;
608 return 1;
609}
610
611__setup("nolwsys", nolwsys_setup);
612
Chen, Kenneth Wdae28062006-03-22 16:54:15 -0800613void __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614mem_init (void)
615{
616 long reserved_pages, codesize, datasize, initsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 pg_data_t *pgdat;
618 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
Robin Holtfde740e2005-04-25 13:13:16 -0700620 BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
621 BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
622 BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
623
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624#ifdef CONFIG_PCI
625 /*
626 * This needs to be called _after_ the command line has been parsed but _before_
627 * any drivers that may need the PCI DMA interface are initialized or bootmem has
628 * been freed.
629 */
630 platform_dma_init();
631#endif
632
Bob Picco2d4b1fa2005-10-04 15:13:57 -0400633#ifdef CONFIG_FLATMEM
Stoyan Gaydarov80a03e22009-03-10 00:10:30 -0500634 BUG_ON(!mem_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 max_mapnr = max_low_pfn;
636#endif
637
638 high_memory = __va(max_low_pfn * PAGE_SIZE);
639
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -0800640 for_each_online_pgdat(pgdat)
bob.picco564601a2005-06-30 09:52:00 -0700641 if (pgdat->bdata->node_bootmem_map)
642 totalram_pages += free_all_bootmem_node(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
644 reserved_pages = 0;
645 efi_memmap_walk(count_reserved_pages, &reserved_pages);
646
647 codesize = (unsigned long) _etext - (unsigned long) _stext;
648 datasize = (unsigned long) _edata - (unsigned long) _etext;
649 initsize = (unsigned long) __init_end - (unsigned long) __init_begin;
650
651 printk(KERN_INFO "Memory: %luk/%luk available (%luk code, %luk reserved, "
Geert Uytterhoevencc013a82009-09-21 17:02:36 -0700652 "%luk data, %luk init)\n", nr_free_pages() << (PAGE_SHIFT - 10),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 num_physpages << (PAGE_SHIFT - 10), codesize >> 10,
654 reserved_pages << (PAGE_SHIFT - 10), datasize >> 10, initsize >> 10);
655
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
657 /*
658 * For fsyscall entrpoints with no light-weight handler, use the ordinary
659 * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
660 * code can tell them apart.
661 */
662 for (i = 0; i < NR_syscalls; ++i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 extern unsigned long sys_call_table[NR_syscalls];
Isaku Yamahatadd97d5c2009-03-04 21:05:34 +0900664 unsigned long *fsyscall_table = paravirt_get_fsyscall_table();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
666 if (!fsyscall_table[i] || nolwsys)
667 fsyscall_table[i] = sys_call_table[i] | 1;
668 }
669 setup_gate();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670}
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900671
672#ifdef CONFIG_MEMORY_HOTPLUG
Yasunori Gotobc02af92006-06-27 02:53:30 -0700673int arch_add_memory(int nid, u64 start, u64 size)
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900674{
675 pg_data_t *pgdat;
676 struct zone *zone;
677 unsigned long start_pfn = start >> PAGE_SHIFT;
678 unsigned long nr_pages = size >> PAGE_SHIFT;
679 int ret;
680
Yasunori Gotobc02af92006-06-27 02:53:30 -0700681 pgdat = NODE_DATA(nid);
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900682
683 zone = pgdat->node_zones + ZONE_NORMAL;
Gary Hadec04fc582009-01-06 14:39:14 -0800684 ret = __add_pages(nid, zone, start_pfn, nr_pages);
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900685
686 if (ret)
687 printk("%s: Problem encountered in __add_pages() as ret=%d\n",
Harvey Harrisond4ed8082008-03-04 15:15:00 -0800688 __func__, ret);
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900689
690 return ret;
691}
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900692#endif
Huang, Xiaolan839052d2008-05-15 10:18:41 +0800693
694/*
695 * Even when CONFIG_IA32_SUPPORT is not enabled it is
696 * useful to have the Linux/x86 domain registered to
697 * avoid an attempted module load when emulators call
698 * personality(PER_LINUX32). This saves several milliseconds
699 * on each such call.
700 */
701static struct exec_domain ia32_exec_domain;
702
703static int __init
704per_linux32_init(void)
705{
706 ia32_exec_domain.name = "Linux/x86";
707 ia32_exec_domain.handler = NULL;
708 ia32_exec_domain.pers_low = PER_LINUX32;
709 ia32_exec_domain.pers_high = PER_LINUX32;
710 ia32_exec_domain.signal_map = default_exec_domain.signal_map;
711 ia32_exec_domain.signal_invmap = default_exec_domain.signal_invmap;
712 register_exec_domain(&ia32_exec_domain);
713
714 return 0;
715}
716
717__initcall(per_linux32_init);