blob: 25c350264a41012bba72d3d992674aea17d32fba [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>
Tejun Heo98e4ae82011-12-08 10:22:08 -080013#include <linux/memblock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/mm.h>
15#include <linux/mmzone.h>
16#include <linux/module.h>
17#include <linux/personality.h>
18#include <linux/reboot.h>
19#include <linux/slab.h>
20#include <linux/swap.h>
21#include <linux/proc_fs.h>
22#include <linux/bitops.h>
Bob Picco139b8302007-01-30 02:11:09 -080023#include <linux/kexec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <asm/dma.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <asm/io.h>
27#include <asm/machvec.h>
28#include <asm/numa.h>
29#include <asm/patch.h>
30#include <asm/pgalloc.h>
31#include <asm/sal.h>
32#include <asm/sections.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#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
Linus Torvalds1da177e2005-04-16 15:20:36 -070039extern void ia64_tlb_init (void);
40
41unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL;
42
43#ifdef CONFIG_VIRTUAL_MEM_MAP
Tejun Heo126b3fc2009-10-02 13:28:55 +090044unsigned long VMALLOC_END = VMALLOC_END_INIT;
45EXPORT_SYMBOL(VMALLOC_END);
Linus Torvalds1da177e2005-04-16 15:20:36 -070046struct page *vmem_map;
47EXPORT_SYMBOL(vmem_map);
48#endif
49
Robin Holtfde740e2005-04-25 13:13:16 -070050struct page *zero_page_memmap_ptr; /* map entry for zero page */
Linus Torvalds1da177e2005-04-16 15:20:36 -070051EXPORT_SYMBOL(zero_page_memmap_ptr);
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053void
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -070054__ia64_sync_icache_dcache (pte_t pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -070055{
56 unsigned long addr;
57 struct page *page;
58
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 page = pte_page(pte);
60 addr = (unsigned long) page_address(page);
61
62 if (test_bit(PG_arch_1, &page->flags))
63 return; /* i-cache is already coherent with d-cache */
64
Christoph Lameter273988f2008-04-09 13:05:41 -070065 flush_icache_range(addr, addr + (PAGE_SIZE << compound_order(page)));
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 set_bit(PG_arch_1, &page->flags); /* mark page as clean */
67}
68
Jan Beulichcde14bb2007-02-05 18:46:40 -080069/*
70 * Since DMA is i-cache coherent, any (complete) pages that were written via
71 * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
72 * flush them when they get mapped into an executable vm-area.
73 */
74void
75dma_mark_clean(void *addr, size_t size)
76{
77 unsigned long pg_addr, end;
78
79 pg_addr = PAGE_ALIGN((unsigned long) addr);
80 end = (unsigned long) addr + size;
81 while (pg_addr + PAGE_SIZE <= end) {
82 struct page *page = virt_to_page(pg_addr);
83 set_bit(PG_arch_1, &page->flags);
84 pg_addr += PAGE_SIZE;
85 }
86}
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088inline void
89ia64_set_rbs_bot (void)
90{
Jiri Slaby02b763b2010-01-06 16:24:30 +010091 unsigned long stack_size = rlimit_max(RLIMIT_STACK) & -16;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
93 if (stack_size > MAX_USER_STACK_SIZE)
94 stack_size = MAX_USER_STACK_SIZE;
KAMEZAWA Hiroyuki83d2cd32007-03-23 12:17:46 +090095 current->thread.rbs_bot = PAGE_ALIGN(current->mm->start_stack - stack_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096}
97
98/*
99 * This performs some platform-dependent address space initialization.
100 * On IA-64, we want to setup the VM area for the register backing
101 * store (which grows upwards) and install the gateway page which is
102 * used for signal trampolines, etc.
103 */
104void
105ia64_init_addr_space (void)
106{
107 struct vm_area_struct *vma;
108
109 ia64_set_rbs_bot();
110
111 /*
112 * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore
113 * the problem. When the process attempts to write to the register backing store
114 * for the first time, it will get a SEGFAULT in this case.
115 */
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800116 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 if (vma) {
Rik van Riel5beb4932010-03-05 13:42:07 -0800118 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 vma->vm_mm = current->mm;
120 vma->vm_start = current->thread.rbs_bot & PAGE_MASK;
121 vma->vm_end = vma->vm_start + PAGE_SIZE;
Hugh Dickins46dea3d2005-10-29 18:16:20 -0700122 vma->vm_flags = VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT;
Coly Li3ed75eb2007-10-18 23:39:15 -0700123 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 down_write(&current->mm->mmap_sem);
125 if (insert_vm_struct(current->mm, vma)) {
126 up_write(&current->mm->mmap_sem);
127 kmem_cache_free(vm_area_cachep, vma);
128 return;
129 }
130 up_write(&current->mm->mmap_sem);
131 }
132
133 /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */
134 if (!(current->personality & MMAP_PAGE_ZERO)) {
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800135 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 if (vma) {
Rik van Riel5beb4932010-03-05 13:42:07 -0800137 INIT_LIST_HEAD(&vma->anon_vma_chain);
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);
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -0700141 vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO |
142 VM_DONTEXPAND | VM_DONTDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 down_write(&current->mm->mmap_sem);
144 if (insert_vm_struct(current->mm, vma)) {
145 up_write(&current->mm->mmap_sem);
146 kmem_cache_free(vm_area_cachep, vma);
147 return;
148 }
149 up_write(&current->mm->mmap_sem);
150 }
151 }
152}
153
154void
155free_initmem (void)
156{
Jiang Liu11199692013-07-03 15:02:48 -0700157 free_reserved_area(ia64_imva(__init_begin), ia64_imva(__init_end),
Jiang Liudbe67df2013-07-03 15:02:51 -0700158 -1, "unused kernel");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159}
160
Chen, Kenneth Wdae280662006-03-22 16:54:15 -0800161void __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162free_initrd_mem (unsigned long start, unsigned long end)
163{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 /*
165 * EFI uses 4KB pages while the kernel can use 4KB or bigger.
166 * Thus EFI and the kernel may have different page sizes. It is
167 * therefore possible to have the initrd share the same page as
168 * the end of the kernel (given current setup).
169 *
170 * To avoid freeing/using the wrong page (kernel sized) we:
171 * - align up the beginning of initrd
172 * - align down the end of initrd
173 *
174 * | |
175 * |=============| a000
176 * | |
177 * | |
178 * | | 9000
179 * |/////////////|
180 * |/////////////|
181 * |=============| 8000
182 * |///INITRD////|
183 * |/////////////|
184 * |/////////////| 7000
185 * | |
186 * |KKKKKKKKKKKKK|
187 * |=============| 6000
188 * |KKKKKKKKKKKKK|
189 * |KKKKKKKKKKKKK|
190 * K=kernel using 8KB pages
191 *
192 * In this example, we must free page 8000 ONLY. So we must align up
193 * initrd_start and keep initrd_end as is.
194 */
195 start = PAGE_ALIGN(start);
196 end = end & PAGE_MASK;
197
198 if (start < end)
199 printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10);
200
201 for (; start < end; start += PAGE_SIZE) {
202 if (!virt_addr_valid(start))
203 continue;
Jiang Liu66f62592013-04-29 15:06:39 -0700204 free_reserved_page(virt_to_page(start));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 }
206}
207
208/*
209 * This installs a clean page in the kernel's page table.
210 */
Chen, Kenneth Wdae280662006-03-22 16:54:15 -0800211static struct page * __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
213{
214 pgd_t *pgd;
215 pud_t *pud;
216 pmd_t *pmd;
217 pte_t *pte;
218
219 if (!PageReserved(page))
220 printk(KERN_ERR "put_kernel_page: page at 0x%p not in reserved memory\n",
221 page_address(page));
222
223 pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */
224
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 {
226 pud = pud_alloc(&init_mm, pgd, address);
227 if (!pud)
228 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 pmd = pmd_alloc(&init_mm, pud, address);
230 if (!pmd)
231 goto out;
Hugh Dickins872fec12005-10-29 18:16:21 -0700232 pte = pte_alloc_kernel(pmd, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 if (!pte)
234 goto out;
Hugh Dickins872fec12005-10-29 18:16:21 -0700235 if (!pte_none(*pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 set_pte(pte, mk_pte(page, pgprot));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 }
Hugh Dickins872fec12005-10-29 18:16:21 -0700239 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 /* no need for flush_tlb */
241 return page;
242}
243
Chen, Kenneth W914a4ea2006-03-12 09:08:26 -0800244static void __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245setup_gate (void)
246{
Isaku Yamahatae4ff5b82009-03-04 21:05:42 +0900247 void *gate_section;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 struct page *page;
249
250 /*
David Mosberger-Tangad597bd2005-06-08 10:45:00 -0700251 * Map the gate page twice: once read-only to export the ELF
252 * headers etc. and once execute-only page to enable
253 * privilege-promotion via "epc":
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 */
Isaku Yamahatae4ff5b82009-03-04 21:05:42 +0900255 gate_section = paravirt_get_gate_section();
256 page = virt_to_page(ia64_imva(gate_section));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
258#ifdef HAVE_BUGGY_SEGREL
Isaku Yamahatae4ff5b82009-03-04 21:05:42 +0900259 page = virt_to_page(ia64_imva(gate_section + PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
261#else
262 put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
David Mosberger-Tangad597bd2005-06-08 10:45:00 -0700263 /* Fill in the holes (if any) with read-only zero pages: */
264 {
265 unsigned long addr;
266
267 for (addr = GATE_ADDR + PAGE_SIZE;
268 addr < GATE_ADDR + PERCPU_PAGE_SIZE;
269 addr += PAGE_SIZE)
270 {
271 put_kernel_page(ZERO_PAGE(0), addr,
272 PAGE_READONLY);
273 put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE,
274 PAGE_READONLY);
275 }
276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277#endif
278 ia64_patch_gate();
279}
280
Greg Kroah-Hartman5b5e76e2012-12-21 14:05:13 -0800281void ia64_mmu_init(void *my_cpu_data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282{
Chen, Kenneth W00b659852006-10-13 10:08:13 -0700283 unsigned long pta, impl_va_bits;
Greg Kroah-Hartman5b5e76e2012-12-21 14:05:13 -0800284 extern void tlb_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
286#ifdef CONFIG_DISABLE_VHPT
287# define VHPT_ENABLE_BIT 0
288#else
289# define VHPT_ENABLE_BIT 1
290#endif
291
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 /*
293 * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
294 * address space. The IA-64 architecture guarantees that at least 50 bits of
295 * virtual address space are implemented but if we pick a large enough page size
296 * (e.g., 64KB), the mapped address space is big enough that it will overlap with
297 * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages,
298 * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
299 * problem in practice. Alternatively, we could truncate the top of the mapped
300 * address space to not permit mappings that would overlap with the VMLPT.
301 * --davidm 00/12/06
302 */
303# define pte_bits 3
304# define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
305 /*
306 * The virtual page table has to cover the entire implemented address space within
307 * a region even though not all of this space may be mappable. The reason for
308 * this is that the Access bit and Dirty bit fault handlers perform
309 * non-speculative accesses to the virtual page table, so the address range of the
310 * virtual page table itself needs to be covered by virtual page table.
311 */
312# define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits)
313# define POW2(n) (1ULL << (n))
314
315 impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
316
317 if (impl_va_bits < 51 || impl_va_bits > 61)
318 panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
Peter Chubb6cf07a82005-08-23 20:07:00 -0700319 /*
320 * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
321 * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
322 * the test makes sure that our mapped space doesn't overlap the
323 * unimplemented hole in the middle of the region.
324 */
325 if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
326 (mapped_space_bits > impl_va_bits - 1))
327 panic("Cannot build a big enough virtual-linear page table"
328 " to cover mapped address space.\n"
329 " Try using a smaller page size.\n");
330
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331
332 /* place the VMLPT at the end of each page-table mapped region: */
333 pta = POW2(61) - POW2(vmlpt_bits);
334
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 /*
336 * Set the (virtually mapped linear) page table address. Bit
337 * 8 selects between the short and long format, bits 2-7 the
338 * size of the table, and bit 0 whether the VHPT walker is
339 * enabled.
340 */
341 ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
342
343 ia64_tlb_init();
344
345#ifdef CONFIG_HUGETLB_PAGE
346 ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
347 ia64_srlz_d();
348#endif
349}
350
351#ifdef CONFIG_VIRTUAL_MEM_MAP
Bob Piccoe44e41d2006-06-28 12:55:43 -0400352int vmemmap_find_next_valid_pfn(int node, int i)
353{
354 unsigned long end_address, hole_next_pfn;
355 unsigned long stop_address;
356 pg_data_t *pgdat = NODE_DATA(node);
357
358 end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
359 end_address = PAGE_ALIGN(end_address);
Xishi Qiu64080682013-11-12 15:07:17 -0800360 stop_address = (unsigned long) &vmem_map[pgdat_end_pfn(pgdat)];
Bob Piccoe44e41d2006-06-28 12:55:43 -0400361
362 do {
363 pgd_t *pgd;
364 pud_t *pud;
365 pmd_t *pmd;
366 pte_t *pte;
367
368 pgd = pgd_offset_k(end_address);
369 if (pgd_none(*pgd)) {
370 end_address += PGDIR_SIZE;
371 continue;
372 }
373
374 pud = pud_offset(pgd, end_address);
375 if (pud_none(*pud)) {
376 end_address += PUD_SIZE;
377 continue;
378 }
379
380 pmd = pmd_offset(pud, end_address);
381 if (pmd_none(*pmd)) {
382 end_address += PMD_SIZE;
383 continue;
384 }
385
386 pte = pte_offset_kernel(pmd, end_address);
387retry_pte:
388 if (pte_none(*pte)) {
389 end_address += PAGE_SIZE;
390 pte++;
391 if ((end_address < stop_address) &&
392 (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
393 goto retry_pte;
394 continue;
395 }
396 /* Found next valid vmem_map page */
397 break;
398 } while (end_address < stop_address);
399
400 end_address = min(end_address, stop_address);
401 end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
402 hole_next_pfn = end_address / sizeof(struct page);
403 return hole_next_pfn - pgdat->node_start_pfn;
404}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Matthew Wilcoxe088a4a2009-05-22 13:49:49 -0700406int __init create_mem_map_page_table(u64 start, u64 end, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407{
408 unsigned long address, start_page, end_page;
409 struct page *map_start, *map_end;
410 int node;
411 pgd_t *pgd;
412 pud_t *pud;
413 pmd_t *pmd;
414 pte_t *pte;
415
416 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
417 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
418
419 start_page = (unsigned long) map_start & PAGE_MASK;
420 end_page = PAGE_ALIGN((unsigned long) map_end);
421 node = paddr_to_nid(__pa(start));
422
423 for (address = start_page; address < end_page; address += PAGE_SIZE) {
424 pgd = pgd_offset_k(address);
425 if (pgd_none(*pgd))
426 pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
427 pud = pud_offset(pgd, address);
428
429 if (pud_none(*pud))
430 pud_populate(&init_mm, pud, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
431 pmd = pmd_offset(pud, address);
432
433 if (pmd_none(*pmd))
434 pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
435 pte = pte_offset_kernel(pmd, address);
436
437 if (pte_none(*pte))
438 set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
439 PAGE_KERNEL));
440 }
441 return 0;
442}
443
444struct memmap_init_callback_data {
445 struct page *start;
446 struct page *end;
447 int nid;
448 unsigned long zone;
449};
450
Adrian Bunk18b8bef2007-10-29 13:49:47 +0100451static int __meminit
Matthew Wilcoxe088a4a2009-05-22 13:49:49 -0700452virtual_memmap_init(u64 start, u64 end, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
454 struct memmap_init_callback_data *args;
455 struct page *map_start, *map_end;
456
457 args = (struct memmap_init_callback_data *) arg;
458 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
459 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
460
461 if (map_start < args->start)
462 map_start = args->start;
463 if (map_end > args->end)
464 map_end = args->end;
465
466 /*
467 * We have to initialize "out of bounds" struct page elements that fit completely
468 * on the same pages that were allocated for the "in bounds" elements because they
469 * may be referenced later (and found to be "reserved").
470 */
471 map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
472 map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
473 / sizeof(struct page));
474
475 if (map_start < map_end)
476 memmap_init_zone((unsigned long)(map_end - map_start),
Dave Hansena2f3aa022007-01-10 23:15:30 -0800477 args->nid, args->zone, page_to_pfn(map_start),
478 MEMMAP_EARLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 return 0;
480}
481
Adrian Bunk18b8bef2007-10-29 13:49:47 +0100482void __meminit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483memmap_init (unsigned long size, int nid, unsigned long zone,
484 unsigned long start_pfn)
485{
486 if (!vmem_map)
Dave Hansena2f3aa022007-01-10 23:15:30 -0800487 memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 else {
489 struct page *start;
490 struct memmap_init_callback_data args;
491
492 start = pfn_to_page(start_pfn);
493 args.start = start;
494 args.end = start + size;
495 args.nid = nid;
496 args.zone = zone;
497
498 efi_memmap_walk(virtual_memmap_init, &args);
499 }
500}
501
502int
503ia64_pfn_valid (unsigned long pfn)
504{
505 char byte;
506 struct page *pg = pfn_to_page(pfn);
507
508 return (__get_user(byte, (char __user *) pg) == 0)
509 && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
510 || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
511}
512EXPORT_SYMBOL(ia64_pfn_valid);
513
Matthew Wilcoxe088a4a2009-05-22 13:49:49 -0700514int __init find_largest_hole(u64 start, u64 end, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515{
516 u64 *max_gap = arg;
517
518 static u64 last_end = PAGE_OFFSET;
519
520 /* NOTE: this algorithm assumes efi memmap table is ordered */
521
522 if (*max_gap < (start - last_end))
523 *max_gap = start - last_end;
524 last_end = end;
525 return 0;
526}
Mel Gorman05e0caa2006-09-27 01:49:54 -0700527
Bob Picco139b8302007-01-30 02:11:09 -0800528#endif /* CONFIG_VIRTUAL_MEM_MAP */
529
Matthew Wilcoxe088a4a2009-05-22 13:49:49 -0700530int __init register_active_ranges(u64 start, u64 len, int nid)
Mel Gorman05e0caa2006-09-27 01:49:54 -0700531{
Zoltan Menyhart98075d22008-04-11 15:21:35 -0700532 u64 end = start + len;
Bob Picco139b8302007-01-30 02:11:09 -0800533
Bob Picco139b8302007-01-30 02:11:09 -0800534#ifdef CONFIG_KEXEC
535 if (start > crashk_res.start && start < crashk_res.end)
536 start = crashk_res.end;
537 if (end > crashk_res.start && end < crashk_res.end)
538 end = crashk_res.start;
539#endif
540
541 if (start < end)
Tejun Heo98e4ae82011-12-08 10:22:08 -0800542 memblock_add_node(__pa(start), end - start, nid);
Mel Gorman05e0caa2006-09-27 01:49:54 -0700543 return 0;
544}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Zou Nan haia3f5c332007-03-20 13:41:57 -0700546int
Matthew Wilcoxe088a4a2009-05-22 13:49:49 -0700547find_max_min_low_pfn (u64 start, u64 end, void *arg)
Zou Nan haia3f5c332007-03-20 13:41:57 -0700548{
549 unsigned long pfn_start, pfn_end;
550#ifdef CONFIG_FLATMEM
551 pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
552 pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
553#else
554 pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
555 pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
556#endif
557 min_low_pfn = min(min_low_pfn, pfn_start);
558 max_low_pfn = max(max_low_pfn, pfn_end);
559 return 0;
560}
561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562/*
563 * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
564 * system call handler. When this option is in effect, all fsyscalls will end up bubbling
565 * down into the kernel and calling the normal (heavy-weight) syscall handler. This is
566 * useful for performance testing, but conceivably could also come in handy for debugging
567 * purposes.
568 */
569
Chen, Kenneth W03906ea2006-03-12 09:10:59 -0800570static int nolwsys __initdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
572static int __init
573nolwsys_setup (char *s)
574{
575 nolwsys = 1;
576 return 1;
577}
578
579__setup("nolwsys", nolwsys_setup);
580
Chen, Kenneth Wdae280662006-03-22 16:54:15 -0800581void __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582mem_init (void)
583{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Robin Holtfde740e2005-04-25 13:13:16 -0700586 BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
587 BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
588 BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
589
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590#ifdef CONFIG_PCI
591 /*
592 * This needs to be called _after_ the command line has been parsed but _before_
593 * any drivers that may need the PCI DMA interface are initialized or bootmem has
594 * been freed.
595 */
596 platform_dma_init();
597#endif
598
Bob Picco2d4b1fa2005-10-04 15:13:57 -0400599#ifdef CONFIG_FLATMEM
Stoyan Gaydarov80a03e22009-03-10 00:10:30 -0500600 BUG_ON(!mem_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601#endif
602
Jiang Liub57b63a2013-07-03 15:04:25 -0700603 set_max_mapnr(max_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 high_memory = __va(max_low_pfn * PAGE_SIZE);
Jiang Liub57b63a2013-07-03 15:04:25 -0700605 free_all_bootmem();
Jiang Liude4bcdd2013-07-03 15:03:58 -0700606 mem_init_print_info(NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
608 /*
609 * For fsyscall entrpoints with no light-weight handler, use the ordinary
610 * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
611 * code can tell them apart.
612 */
613 for (i = 0; i < NR_syscalls; ++i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 extern unsigned long sys_call_table[NR_syscalls];
Isaku Yamahatadd97d5c2009-03-04 21:05:34 +0900615 unsigned long *fsyscall_table = paravirt_get_fsyscall_table();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
617 if (!fsyscall_table[i] || nolwsys)
618 fsyscall_table[i] = sys_call_table[i] | 1;
619 }
620 setup_gate();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621}
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900622
623#ifdef CONFIG_MEMORY_HOTPLUG
Yasunori Gotobc02af92006-06-27 02:53:30 -0700624int arch_add_memory(int nid, u64 start, u64 size)
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900625{
626 pg_data_t *pgdat;
627 struct zone *zone;
628 unsigned long start_pfn = start >> PAGE_SHIFT;
629 unsigned long nr_pages = size >> PAGE_SHIFT;
630 int ret;
631
Yasunori Gotobc02af92006-06-27 02:53:30 -0700632 pgdat = NODE_DATA(nid);
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900633
634 zone = pgdat->node_zones + ZONE_NORMAL;
Gary Hadec04fc582009-01-06 14:39:14 -0800635 ret = __add_pages(nid, zone, start_pfn, nr_pages);
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900636
637 if (ret)
638 printk("%s: Problem encountered in __add_pages() as ret=%d\n",
Harvey Harrisond4ed8082008-03-04 15:15:00 -0800639 __func__, ret);
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900640
641 return ret;
642}
Wen Congyang24d335c2013-02-22 16:32:58 -0800643
644#ifdef CONFIG_MEMORY_HOTREMOVE
645int arch_remove_memory(u64 start, u64 size)
646{
647 unsigned long start_pfn = start >> PAGE_SHIFT;
648 unsigned long nr_pages = size >> PAGE_SHIFT;
649 struct zone *zone;
650 int ret;
651
652 zone = page_zone(pfn_to_page(start_pfn));
653 ret = __remove_pages(zone, start_pfn, nr_pages);
654 if (ret)
655 pr_warn("%s: Problem encountered in __remove_pages() as"
656 " ret=%d\n", __func__, ret);
657
658 return ret;
659}
660#endif
Yasunori Goto1681b8e2006-01-07 11:50:38 +0900661#endif
Huang, Xiaolan839052d2008-05-15 10:18:41 +0800662
663/*
664 * Even when CONFIG_IA32_SUPPORT is not enabled it is
665 * useful to have the Linux/x86 domain registered to
666 * avoid an attempted module load when emulators call
667 * personality(PER_LINUX32). This saves several milliseconds
668 * on each such call.
669 */
670static struct exec_domain ia32_exec_domain;
671
672static int __init
673per_linux32_init(void)
674{
675 ia32_exec_domain.name = "Linux/x86";
676 ia32_exec_domain.handler = NULL;
677 ia32_exec_domain.pers_low = PER_LINUX32;
678 ia32_exec_domain.pers_high = PER_LINUX32;
679 ia32_exec_domain.signal_map = default_exec_domain.signal_map;
680 ia32_exec_domain.signal_invmap = default_exec_domain.signal_invmap;
681 register_exec_domain(&ia32_exec_domain);
682
683 return 0;
684}
685
686__initcall(per_linux32_init);
Mel Gormanaec6a882014-01-21 15:49:13 -0800687
688/**
689 * show_mem - give short summary of memory stats
690 *
691 * Shows a simple page count of reserved and used pages in the system.
692 * For discontig machines, it does this on a per-pgdat basis.
693 */
694void show_mem(unsigned int filter)
695{
696 int total_reserved = 0;
697 unsigned long total_present = 0;
698 pg_data_t *pgdat;
699
700 printk(KERN_INFO "Mem-info:\n");
701 show_free_areas(filter);
702 printk(KERN_INFO "Node memory in pages:\n");
703 for_each_online_pgdat(pgdat) {
704 unsigned long present;
705 unsigned long flags;
706 int reserved = 0;
707 int nid = pgdat->node_id;
708 int zoneid;
709
710 if (skip_free_areas_node(filter, nid))
711 continue;
712 pgdat_resize_lock(pgdat, &flags);
713
714 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
715 struct zone *zone = &pgdat->node_zones[zoneid];
716 if (!populated_zone(zone))
717 continue;
718
719 reserved += zone->present_pages - zone->managed_pages;
720 }
721 present = pgdat->node_present_pages;
722
723 pgdat_resize_unlock(pgdat, &flags);
724 total_present += present;
725 total_reserved += reserved;
726 printk(KERN_INFO "Node %4d: RAM: %11ld, rsvd: %8d, ",
727 nid, present, reserved);
728 }
729 printk(KERN_INFO "%ld pages of RAM\n", total_present);
730 printk(KERN_INFO "%d reserved pages\n", total_reserved);
731 printk(KERN_INFO "Total of %ld pages in page table cache\n",
732 quicklist_total_size());
733 printk(KERN_INFO "%ld free buffer pages\n", nr_free_buffer_pages());
734}