aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/mm/init_64.c
diff options
context:
space:
mode:
authorThomas Gleixner <tglx@linutronix.de>2008-01-30 07:34:10 -0500
committerIngo Molnar <mingo@elte.hu>2008-01-30 07:34:10 -0500
commit14a62c34b134d24f7fedb8fd66028ecdcffb32c8 (patch)
tree58198c293144cfad46f8ed0cd816d8a9e3867cb2 /arch/x86/mm/init_64.c
parenta3828064be4ed8e95907d3943e7af13cb709694d (diff)
x86: unify ioremap
Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Diffstat (limited to 'arch/x86/mm/init_64.c')
-rw-r--r--arch/x86/mm/init_64.c301
1 files changed, 175 insertions, 126 deletions
diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index 8a7b725ce3c7..bb732bb79b4a 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -47,10 +47,10 @@
47#include <asm/numa.h> 47#include <asm/numa.h>
48 48
49#ifndef Dprintk 49#ifndef Dprintk
50#define Dprintk(x...) 50# define Dprintk(x...)
51#endif 51#endif
52 52
53const struct dma_mapping_ops* dma_ops; 53const struct dma_mapping_ops *dma_ops;
54EXPORT_SYMBOL(dma_ops); 54EXPORT_SYMBOL(dma_ops);
55 55
56static unsigned long dma_reserve __initdata; 56static unsigned long dma_reserve __initdata;
@@ -67,22 +67,26 @@ void show_mem(void)
67{ 67{
68 long i, total = 0, reserved = 0; 68 long i, total = 0, reserved = 0;
69 long shared = 0, cached = 0; 69 long shared = 0, cached = 0;
70 pg_data_t *pgdat;
71 struct page *page; 70 struct page *page;
71 pg_data_t *pgdat;
72 72
73 printk(KERN_INFO "Mem-info:\n"); 73 printk(KERN_INFO "Mem-info:\n");
74 show_free_areas(); 74 show_free_areas();
75 printk(KERN_INFO "Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10)); 75 printk(KERN_INFO "Free swap: %6ldkB\n",
76 nr_swap_pages << (PAGE_SHIFT-10));
76 77
77 for_each_online_pgdat(pgdat) { 78 for_each_online_pgdat(pgdat) {
78 for (i = 0; i < pgdat->node_spanned_pages; ++i) { 79 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
79 /* this loop can take a while with 256 GB and 4k pages 80 /*
80 so update the NMI watchdog */ 81 * This loop can take a while with 256 GB and
81 if (unlikely(i % MAX_ORDER_NR_PAGES == 0)) { 82 * 4k pages so defer the NMI watchdog:
83 */
84 if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
82 touch_nmi_watchdog(); 85 touch_nmi_watchdog();
83 } 86
84 if (!pfn_valid(pgdat->node_start_pfn + i)) 87 if (!pfn_valid(pgdat->node_start_pfn + i))
85 continue; 88 continue;
89
86 page = pfn_to_page(pgdat->node_start_pfn + i); 90 page = pfn_to_page(pgdat->node_start_pfn + i);
87 total++; 91 total++;
88 if (PageReserved(page)) 92 if (PageReserved(page))
@@ -91,32 +95,37 @@ void show_mem(void)
91 cached++; 95 cached++;
92 else if (page_count(page)) 96 else if (page_count(page))
93 shared += page_count(page) - 1; 97 shared += page_count(page) - 1;
94 } 98 }
95 } 99 }
96 printk(KERN_INFO "%lu pages of RAM\n", total); 100 printk(KERN_INFO "%lu pages of RAM\n", total);
97 printk(KERN_INFO "%lu reserved pages\n",reserved); 101 printk(KERN_INFO "%lu reserved pages\n", reserved);
98 printk(KERN_INFO "%lu pages shared\n",shared); 102 printk(KERN_INFO "%lu pages shared\n", shared);
99 printk(KERN_INFO "%lu pages swap cached\n",cached); 103 printk(KERN_INFO "%lu pages swap cached\n", cached);
100} 104}
101 105
102int after_bootmem; 106int after_bootmem;
103 107
104static __init void *spp_getpage(void) 108static __init void *spp_getpage(void)
105{ 109{
106 void *ptr; 110 void *ptr;
111
107 if (after_bootmem) 112 if (after_bootmem)
108 ptr = (void *) get_zeroed_page(GFP_ATOMIC); 113 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
109 else 114 else
110 ptr = alloc_bootmem_pages(PAGE_SIZE); 115 ptr = alloc_bootmem_pages(PAGE_SIZE);
111 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) 116
112 panic("set_pte_phys: cannot allocate page data %s\n", after_bootmem?"after bootmem":""); 117 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
118 panic("set_pte_phys: cannot allocate page data %s\n",
119 after_bootmem ? "after bootmem" : "");
120 }
113 121
114 Dprintk("spp_getpage %p\n", ptr); 122 Dprintk("spp_getpage %p\n", ptr);
123
115 return ptr; 124 return ptr;
116} 125}
117 126
118static __init void set_pte_phys(unsigned long vaddr, 127static __init void
119 unsigned long phys, pgprot_t prot) 128set_pte_phys(unsigned long vaddr, unsigned long phys, pgprot_t prot)
120{ 129{
121 pgd_t *pgd; 130 pgd_t *pgd;
122 pud_t *pud; 131 pud_t *pud;
@@ -132,10 +141,11 @@ static __init void set_pte_phys(unsigned long vaddr,
132 } 141 }
133 pud = pud_offset(pgd, vaddr); 142 pud = pud_offset(pgd, vaddr);
134 if (pud_none(*pud)) { 143 if (pud_none(*pud)) {
135 pmd = (pmd_t *) spp_getpage(); 144 pmd = (pmd_t *) spp_getpage();
136 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER)); 145 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER));
137 if (pmd != pmd_offset(pud, 0)) { 146 if (pmd != pmd_offset(pud, 0)) {
138 printk("PAGETABLE BUG #01! %p <-> %p\n", pmd, pmd_offset(pud,0)); 147 printk("PAGETABLE BUG #01! %p <-> %p\n",
148 pmd, pmd_offset(pud, 0));
139 return; 149 return;
140 } 150 }
141 } 151 }
@@ -164,8 +174,8 @@ static __init void set_pte_phys(unsigned long vaddr,
164} 174}
165 175
166/* NOTE: this is meant to be run only at boot */ 176/* NOTE: this is meant to be run only at boot */
167void __init 177void __init
168__set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t prot) 178__set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t prot)
169{ 179{
170 unsigned long address = __fix_to_virt(idx); 180 unsigned long address = __fix_to_virt(idx);
171 181
@@ -180,18 +190,19 @@ static unsigned long __initdata table_start;
180static unsigned long __meminitdata table_end; 190static unsigned long __meminitdata table_end;
181 191
182static __meminit void *alloc_low_page(unsigned long *phys) 192static __meminit void *alloc_low_page(unsigned long *phys)
183{ 193{
184 unsigned long pfn = table_end++; 194 unsigned long pfn = table_end++;
185 void *adr; 195 void *adr;
186 196
187 if (after_bootmem) { 197 if (after_bootmem) {
188 adr = (void *)get_zeroed_page(GFP_ATOMIC); 198 adr = (void *)get_zeroed_page(GFP_ATOMIC);
189 *phys = __pa(adr); 199 *phys = __pa(adr);
200
190 return adr; 201 return adr;
191 } 202 }
192 203
193 if (pfn >= end_pfn) 204 if (pfn >= end_pfn)
194 panic("alloc_low_page: ran out of memory"); 205 panic("alloc_low_page: ran out of memory");
195 206
196 adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE); 207 adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE);
197 memset(adr, 0, PAGE_SIZE); 208 memset(adr, 0, PAGE_SIZE);
@@ -200,44 +211,49 @@ static __meminit void *alloc_low_page(unsigned long *phys)
200} 211}
201 212
202static __meminit void unmap_low_page(void *adr) 213static __meminit void unmap_low_page(void *adr)
203{ 214{
204
205 if (after_bootmem) 215 if (after_bootmem)
206 return; 216 return;
207 217
208 early_iounmap(adr, PAGE_SIZE); 218 early_iounmap(adr, PAGE_SIZE);
209} 219}
210 220
211/* Must run before zap_low_mappings */ 221/* Must run before zap_low_mappings */
212__meminit void *early_ioremap(unsigned long addr, unsigned long size) 222__meminit void *early_ioremap(unsigned long addr, unsigned long size)
213{ 223{
214 unsigned long vaddr;
215 pmd_t *pmd, *last_pmd; 224 pmd_t *pmd, *last_pmd;
225 unsigned long vaddr;
216 int i, pmds; 226 int i, pmds;
217 227
218 pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE; 228 pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
219 vaddr = __START_KERNEL_map; 229 vaddr = __START_KERNEL_map;
220 pmd = level2_kernel_pgt; 230 pmd = level2_kernel_pgt;
221 last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1; 231 last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1;
232
222 for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) { 233 for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) {
223 for (i = 0; i < pmds; i++) { 234 for (i = 0; i < pmds; i++) {
224 if (pmd_present(pmd[i])) 235 if (pmd_present(pmd[i]))
225 goto next; 236 goto continue_outer_loop;
226 } 237 }
227 vaddr += addr & ~PMD_MASK; 238 vaddr += addr & ~PMD_MASK;
228 addr &= PMD_MASK; 239 addr &= PMD_MASK;
240
229 for (i = 0; i < pmds; i++, addr += PMD_SIZE) 241 for (i = 0; i < pmds; i++, addr += PMD_SIZE)
230 set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC)); 242 set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
231 __flush_tlb_all(); 243 __flush_tlb_all();
244
232 return (void *)vaddr; 245 return (void *)vaddr;
233 next: 246continue_outer_loop:
234 ; 247 ;
235 } 248 }
236 printk("early_ioremap(0x%lx, %lu) failed\n", addr, size); 249 printk("early_ioremap(0x%lx, %lu) failed\n", addr, size);
250
237 return NULL; 251 return NULL;
238} 252}
239 253
240/* To avoid virtual aliases later */ 254/*
255 * To avoid virtual aliases later:
256 */
241__meminit void early_iounmap(void *addr, unsigned long size) 257__meminit void early_iounmap(void *addr, unsigned long size)
242{ 258{
243 unsigned long vaddr; 259 unsigned long vaddr;
@@ -247,8 +263,10 @@ __meminit void early_iounmap(void *addr, unsigned long size)
247 vaddr = (unsigned long)addr; 263 vaddr = (unsigned long)addr;
248 pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE; 264 pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
249 pmd = level2_kernel_pgt + pmd_index(vaddr); 265 pmd = level2_kernel_pgt + pmd_index(vaddr);
266
250 for (i = 0; i < pmds; i++) 267 for (i = 0; i < pmds; i++)
251 pmd_clear(pmd + i); 268 pmd_clear(pmd + i);
269
252 __flush_tlb_all(); 270 __flush_tlb_all();
253} 271}
254 272
@@ -262,9 +280,10 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
262 pmd_t *pmd = pmd_page + pmd_index(address); 280 pmd_t *pmd = pmd_page + pmd_index(address);
263 281
264 if (address >= end) { 282 if (address >= end) {
265 if (!after_bootmem) 283 if (!after_bootmem) {
266 for (; i < PTRS_PER_PMD; i++, pmd++) 284 for (; i < PTRS_PER_PMD; i++, pmd++)
267 set_pmd(pmd, __pmd(0)); 285 set_pmd(pmd, __pmd(0));
286 }
268 break; 287 break;
269 } 288 }
270 289
@@ -280,19 +299,19 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
280static void __meminit 299static void __meminit
281phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end) 300phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
282{ 301{
283 pmd_t *pmd = pmd_offset(pud,0); 302 pmd_t *pmd = pmd_offset(pud, 0);
284 spin_lock(&init_mm.page_table_lock); 303 spin_lock(&init_mm.page_table_lock);
285 phys_pmd_init(pmd, address, end); 304 phys_pmd_init(pmd, address, end);
286 spin_unlock(&init_mm.page_table_lock); 305 spin_unlock(&init_mm.page_table_lock);
287 __flush_tlb_all(); 306 __flush_tlb_all();
288} 307}
289 308
290static void __meminit phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end) 309static void __meminit
291{ 310phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
311{
292 int i = pud_index(addr); 312 int i = pud_index(addr);
293 313
294 314 for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
295 for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE ) {
296 unsigned long pmd_phys; 315 unsigned long pmd_phys;
297 pud_t *pud = pud_page + pud_index(addr); 316 pud_t *pud = pud_page + pud_index(addr);
298 pmd_t *pmd; 317 pmd_t *pmd;
@@ -300,10 +319,11 @@ static void __meminit phys_pud_init(pud_t *pud_page, unsigned long addr, unsigne
300 if (addr >= end) 319 if (addr >= end)
301 break; 320 break;
302 321
303 if (!after_bootmem && !e820_any_mapped(addr,addr+PUD_SIZE,0)) { 322 if (!after_bootmem &&
304 set_pud(pud, __pud(0)); 323 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
324 set_pud(pud, __pud(0));
305 continue; 325 continue;
306 } 326 }
307 327
308 if (pud_val(*pud)) { 328 if (pud_val(*pud)) {
309 phys_pmd_update(pud, addr, end); 329 phys_pmd_update(pud, addr, end);
@@ -311,14 +331,16 @@ static void __meminit phys_pud_init(pud_t *pud_page, unsigned long addr, unsigne
311 } 331 }
312 332
313 pmd = alloc_low_page(&pmd_phys); 333 pmd = alloc_low_page(&pmd_phys);
334
314 spin_lock(&init_mm.page_table_lock); 335 spin_lock(&init_mm.page_table_lock);
315 set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE)); 336 set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE));
316 phys_pmd_init(pmd, addr, end); 337 phys_pmd_init(pmd, addr, end);
317 spin_unlock(&init_mm.page_table_lock); 338 spin_unlock(&init_mm.page_table_lock);
339
318 unmap_low_page(pmd); 340 unmap_low_page(pmd);
319 } 341 }
320 __flush_tlb_all(); 342 __flush_tlb_all();
321} 343}
322 344
323static void __init find_early_table_space(unsigned long end) 345static void __init find_early_table_space(unsigned long end)
324{ 346{
@@ -329,11 +351,13 @@ static void __init find_early_table_space(unsigned long end)
329 tables = round_up(puds * sizeof(pud_t), PAGE_SIZE) + 351 tables = round_up(puds * sizeof(pud_t), PAGE_SIZE) +
330 round_up(pmds * sizeof(pmd_t), PAGE_SIZE); 352 round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
331 353
332 /* RED-PEN putting page tables only on node 0 could 354 /*
333 cause a hotspot and fill up ZONE_DMA. The page tables 355 * RED-PEN putting page tables only on node 0 could
334 need roughly 0.5KB per GB. */ 356 * cause a hotspot and fill up ZONE_DMA. The page tables
335 start = 0x8000; 357 * need roughly 0.5KB per GB.
336 table_start = find_e820_area(start, end, tables); 358 */
359 start = 0x8000;
360 table_start = find_e820_area(start, end, tables);
337 if (table_start == -1UL) 361 if (table_start == -1UL)
338 panic("Cannot find space for the kernel page tables"); 362 panic("Cannot find space for the kernel page tables");
339 363
@@ -345,20 +369,23 @@ static void __init find_early_table_space(unsigned long end)
345 (table_start << PAGE_SHIFT) + tables); 369 (table_start << PAGE_SHIFT) + tables);
346} 370}
347 371
348/* Setup the direct mapping of the physical memory at PAGE_OFFSET. 372/*
349 This runs before bootmem is initialized and gets pages directly from the 373 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
350 physical memory. To access them they are temporarily mapped. */ 374 * This runs before bootmem is initialized and gets pages directly from
375 * the physical memory. To access them they are temporarily mapped.
376 */
351void __init_refok init_memory_mapping(unsigned long start, unsigned long end) 377void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
352{ 378{
353 unsigned long next; 379 unsigned long next;
354 380
355 Dprintk("init_memory_mapping\n"); 381 Dprintk("init_memory_mapping\n");
356 382
357 /* 383 /*
358 * Find space for the kernel direct mapping tables. 384 * Find space for the kernel direct mapping tables.
359 * Later we should allocate these tables in the local node of the memory 385 *
360 * mapped. Unfortunately this is done currently before the nodes are 386 * Later we should allocate these tables in the local node of the
361 * discovered. 387 * memory mapped. Unfortunately this is done currently before the
388 * nodes are discovered.
362 */ 389 */
363 if (!after_bootmem) 390 if (!after_bootmem)
364 find_early_table_space(end); 391 find_early_table_space(end);
@@ -367,8 +394,8 @@ void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
367 end = (unsigned long)__va(end); 394 end = (unsigned long)__va(end);
368 395
369 for (; start < end; start = next) { 396 for (; start < end; start = next) {
370 unsigned long pud_phys;
371 pgd_t *pgd = pgd_offset_k(start); 397 pgd_t *pgd = pgd_offset_k(start);
398 unsigned long pud_phys;
372 pud_t *pud; 399 pud_t *pud;
373 400
374 if (after_bootmem) 401 if (after_bootmem)
@@ -377,13 +404,13 @@ void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
377 pud = alloc_low_page(&pud_phys); 404 pud = alloc_low_page(&pud_phys);
378 405
379 next = start + PGDIR_SIZE; 406 next = start + PGDIR_SIZE;
380 if (next > end) 407 if (next > end)
381 next = end; 408 next = end;
382 phys_pud_init(pud, __pa(start), __pa(next)); 409 phys_pud_init(pud, __pa(start), __pa(next));
383 if (!after_bootmem) 410 if (!after_bootmem)
384 set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys)); 411 set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys));
385 unmap_low_page(pud); 412 unmap_low_page(pud);
386 } 413 }
387 414
388 if (!after_bootmem) 415 if (!after_bootmem)
389 mmu_cr4_features = read_cr4(); 416 mmu_cr4_features = read_cr4();
@@ -396,6 +423,7 @@ void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
396void __init paging_init(void) 423void __init paging_init(void)
397{ 424{
398 unsigned long max_zone_pfns[MAX_NR_ZONES]; 425 unsigned long max_zone_pfns[MAX_NR_ZONES];
426
399 memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); 427 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
400 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; 428 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
401 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; 429 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
@@ -407,39 +435,48 @@ void __init paging_init(void)
407} 435}
408#endif 436#endif
409 437
410/* Unmap a kernel mapping if it exists. This is useful to avoid prefetches 438/*
411 from the CPU leading to inconsistent cache lines. address and size 439 * Unmap a kernel mapping if it exists. This is useful to avoid
412 must be aligned to 2MB boundaries. 440 * prefetches from the CPU leading to inconsistent cache lines.
413 Does nothing when the mapping doesn't exist. */ 441 * address and size must be aligned to 2MB boundaries.
414void __init clear_kernel_mapping(unsigned long address, unsigned long size) 442 * Does nothing when the mapping doesn't exist.
443 */
444void __init clear_kernel_mapping(unsigned long address, unsigned long size)
415{ 445{
416 unsigned long end = address + size; 446 unsigned long end = address + size;
417 447
418 BUG_ON(address & ~LARGE_PAGE_MASK); 448 BUG_ON(address & ~LARGE_PAGE_MASK);
419 BUG_ON(size & ~LARGE_PAGE_MASK); 449 BUG_ON(size & ~LARGE_PAGE_MASK);
420 450
421 for (; address < end; address += LARGE_PAGE_SIZE) { 451 for (; address < end; address += LARGE_PAGE_SIZE) {
422 pgd_t *pgd = pgd_offset_k(address); 452 pgd_t *pgd = pgd_offset_k(address);
423 pud_t *pud; 453 pud_t *pud;
424 pmd_t *pmd; 454 pmd_t *pmd;
455
425 if (pgd_none(*pgd)) 456 if (pgd_none(*pgd))
426 continue; 457 continue;
458
427 pud = pud_offset(pgd, address); 459 pud = pud_offset(pgd, address);
428 if (pud_none(*pud)) 460 if (pud_none(*pud))
429 continue; 461 continue;
462
430 pmd = pmd_offset(pud, address); 463 pmd = pmd_offset(pud, address);
431 if (!pmd || pmd_none(*pmd)) 464 if (!pmd || pmd_none(*pmd))
432 continue; 465 continue;
433 if (0 == (pmd_val(*pmd) & _PAGE_PSE)) { 466
434 /* Could handle this, but it should not happen currently. */ 467 if (!(pmd_val(*pmd) & _PAGE_PSE)) {
435 printk(KERN_ERR 468 /*
436 "clear_kernel_mapping: mapping has been split. will leak memory\n"); 469 * Could handle this, but it should not happen
437 pmd_ERROR(*pmd); 470 * currently:
471 */
472 printk(KERN_ERR "clear_kernel_mapping: "
473 "mapping has been split. will leak memory\n");
474 pmd_ERROR(*pmd);
438 } 475 }
439 set_pmd(pmd, __pmd(0)); 476 set_pmd(pmd, __pmd(0));
440 } 477 }
441 __flush_tlb_all(); 478 __flush_tlb_all();
442} 479}
443 480
444/* 481/*
445 * Memory hotplug specific functions 482 * Memory hotplug specific functions
@@ -466,16 +503,13 @@ int arch_add_memory(int nid, u64 start, u64 size)
466 unsigned long nr_pages = size >> PAGE_SHIFT; 503 unsigned long nr_pages = size >> PAGE_SHIFT;
467 int ret; 504 int ret;
468 505
469 init_memory_mapping(start, (start + size -1)); 506 init_memory_mapping(start, start + size-1);
470 507
471 ret = __add_pages(zone, start_pfn, nr_pages); 508 ret = __add_pages(zone, start_pfn, nr_pages);
472 if (ret) 509 if (ret)
473 goto error; 510 printk("%s: Problem encountered in __add_pages!\n", __func__);
474 511
475 return ret; 512 return ret;
476error:
477 printk("%s: Problem encountered in __add_pages!\n", __func__);
478 return ret;
479} 513}
480EXPORT_SYMBOL_GPL(arch_add_memory); 514EXPORT_SYMBOL_GPL(arch_add_memory);
481 515
@@ -489,8 +523,8 @@ EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
489 523
490#endif /* CONFIG_MEMORY_HOTPLUG */ 524#endif /* CONFIG_MEMORY_HOTPLUG */
491 525
492static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel, kcore_modules, 526static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
493 kcore_vsyscall; 527 kcore_modules, kcore_vsyscall;
494 528
495void __init mem_init(void) 529void __init mem_init(void)
496{ 530{
@@ -518,7 +552,6 @@ void __init mem_init(void)
518#endif 552#endif
519 reservedpages = end_pfn - totalram_pages - 553 reservedpages = end_pfn - totalram_pages -
520 absent_pages_in_range(0, end_pfn); 554 absent_pages_in_range(0, end_pfn);
521
522 after_bootmem = 1; 555 after_bootmem = 1;
523 556
524 codesize = (unsigned long) &_etext - (unsigned long) &_text; 557 codesize = (unsigned long) &_etext - (unsigned long) &_text;
@@ -526,15 +559,16 @@ void __init mem_init(void)
526 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; 559 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
527 560
528 /* Register memory areas for /proc/kcore */ 561 /* Register memory areas for /proc/kcore */
529 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT); 562 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
530 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, 563 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
531 VMALLOC_END-VMALLOC_START); 564 VMALLOC_END-VMALLOC_START);
532 kclist_add(&kcore_kernel, &_stext, _end - _stext); 565 kclist_add(&kcore_kernel, &_stext, _end - _stext);
533 kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN); 566 kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
534 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START, 567 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
535 VSYSCALL_END - VSYSCALL_START); 568 VSYSCALL_END - VSYSCALL_START);
536 569
537 printk("Memory: %luk/%luk available (%ldk kernel code, %ldk reserved, %ldk data, %ldk init)\n", 570 printk("Memory: %luk/%luk available (%ldk kernel code, "
571 "%ldk reserved, %ldk data, %ldk init)\n",
538 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), 572 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
539 end_pfn << (PAGE_SHIFT-10), 573 end_pfn << (PAGE_SHIFT-10),
540 codesize >> 10, 574 codesize >> 10,
@@ -561,6 +595,7 @@ void free_init_pages(char *what, unsigned long begin, unsigned long end)
561 set_memory_np(begin, (end - begin) >> PAGE_SHIFT); 595 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
562#else 596#else
563 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10); 597 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
598
564 for (addr = begin; addr < end; addr += PAGE_SIZE) { 599 for (addr = begin; addr < end; addr += PAGE_SIZE) {
565 ClearPageReserved(virt_to_page(addr)); 600 ClearPageReserved(virt_to_page(addr));
566 init_page_count(virt_to_page(addr)); 601 init_page_count(virt_to_page(addr));
@@ -596,7 +631,7 @@ void mark_rodata_ro(void)
596#ifdef CONFIG_KPROBES 631#ifdef CONFIG_KPROBES
597 start = (unsigned long)__start_rodata; 632 start = (unsigned long)__start_rodata;
598#endif 633#endif
599 634
600 end = (unsigned long)__end_rodata; 635 end = (unsigned long)__end_rodata;
601 start = (start + PAGE_SIZE - 1) & PAGE_MASK; 636 start = (start + PAGE_SIZE - 1) & PAGE_MASK;
602 end &= PAGE_MASK; 637 end &= PAGE_MASK;
@@ -627,17 +662,21 @@ void free_initrd_mem(unsigned long start, unsigned long end)
627} 662}
628#endif 663#endif
629 664
630void __init reserve_bootmem_generic(unsigned long phys, unsigned len) 665void __init reserve_bootmem_generic(unsigned long phys, unsigned len)
631{ 666{
632#ifdef CONFIG_NUMA 667#ifdef CONFIG_NUMA
633 int nid = phys_to_nid(phys); 668 int nid = phys_to_nid(phys);
634#endif 669#endif
635 unsigned long pfn = phys >> PAGE_SHIFT; 670 unsigned long pfn = phys >> PAGE_SHIFT;
671
636 if (pfn >= end_pfn) { 672 if (pfn >= end_pfn) {
637 /* This can happen with kdump kernels when accessing firmware 673 /*
638 tables. */ 674 * This can happen with kdump kernels when accessing
675 * firmware tables:
676 */
639 if (pfn < end_pfn_map) 677 if (pfn < end_pfn_map)
640 return; 678 return;
679
641 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n", 680 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n",
642 phys, len); 681 phys, len);
643 return; 682 return;
@@ -645,9 +684,9 @@ void __init reserve_bootmem_generic(unsigned long phys, unsigned len)
645 684
646 /* Should check here against the e820 map to avoid double free */ 685 /* Should check here against the e820 map to avoid double free */
647#ifdef CONFIG_NUMA 686#ifdef CONFIG_NUMA
648 reserve_bootmem_node(NODE_DATA(nid), phys, len); 687 reserve_bootmem_node(NODE_DATA(nid), phys, len);
649#else 688#else
650 reserve_bootmem(phys, len); 689 reserve_bootmem(phys, len);
651#endif 690#endif
652 if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) { 691 if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
653 dma_reserve += len / PAGE_SIZE; 692 dma_reserve += len / PAGE_SIZE;
@@ -655,46 +694,49 @@ void __init reserve_bootmem_generic(unsigned long phys, unsigned len)
655 } 694 }
656} 695}
657 696
658int kern_addr_valid(unsigned long addr) 697int kern_addr_valid(unsigned long addr)
659{ 698{
660 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT; 699 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
661 pgd_t *pgd; 700 pgd_t *pgd;
662 pud_t *pud; 701 pud_t *pud;
663 pmd_t *pmd; 702 pmd_t *pmd;
664 pte_t *pte; 703 pte_t *pte;
665 704
666 if (above != 0 && above != -1UL) 705 if (above != 0 && above != -1UL)
667 return 0; 706 return 0;
668 707
669 pgd = pgd_offset_k(addr); 708 pgd = pgd_offset_k(addr);
670 if (pgd_none(*pgd)) 709 if (pgd_none(*pgd))
671 return 0; 710 return 0;
672 711
673 pud = pud_offset(pgd, addr); 712 pud = pud_offset(pgd, addr);
674 if (pud_none(*pud)) 713 if (pud_none(*pud))
675 return 0; 714 return 0;
676 715
677 pmd = pmd_offset(pud, addr); 716 pmd = pmd_offset(pud, addr);
678 if (pmd_none(*pmd)) 717 if (pmd_none(*pmd))
679 return 0; 718 return 0;
719
680 if (pmd_large(*pmd)) 720 if (pmd_large(*pmd))
681 return pfn_valid(pmd_pfn(*pmd)); 721 return pfn_valid(pmd_pfn(*pmd));
682 722
683 pte = pte_offset_kernel(pmd, addr); 723 pte = pte_offset_kernel(pmd, addr);
684 if (pte_none(*pte)) 724 if (pte_none(*pte))
685 return 0; 725 return 0;
726
686 return pfn_valid(pte_pfn(*pte)); 727 return pfn_valid(pte_pfn(*pte));
687} 728}
688 729
689/* A pseudo VMA to allow ptrace access for the vsyscall page. This only 730/*
690 covers the 64bit vsyscall page now. 32bit has a real VMA now and does 731 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
691 not need special handling anymore. */ 732 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
692 733 * not need special handling anymore:
734 */
693static struct vm_area_struct gate_vma = { 735static struct vm_area_struct gate_vma = {
694 .vm_start = VSYSCALL_START, 736 .vm_start = VSYSCALL_START,
695 .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES << PAGE_SHIFT), 737 .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
696 .vm_page_prot = PAGE_READONLY_EXEC, 738 .vm_page_prot = PAGE_READONLY_EXEC,
697 .vm_flags = VM_READ | VM_EXEC 739 .vm_flags = VM_READ | VM_EXEC
698}; 740};
699 741
700struct vm_area_struct *get_gate_vma(struct task_struct *tsk) 742struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
@@ -709,14 +751,17 @@ struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
709int in_gate_area(struct task_struct *task, unsigned long addr) 751int in_gate_area(struct task_struct *task, unsigned long addr)
710{ 752{
711 struct vm_area_struct *vma = get_gate_vma(task); 753 struct vm_area_struct *vma = get_gate_vma(task);
754
712 if (!vma) 755 if (!vma)
713 return 0; 756 return 0;
757
714 return (addr >= vma->vm_start) && (addr < vma->vm_end); 758 return (addr >= vma->vm_start) && (addr < vma->vm_end);
715} 759}
716 760
717/* Use this when you have no reliable task/vma, typically from interrupt 761/*
718 * context. It is less reliable than using the task's vma and may give 762 * Use this when you have no reliable task/vma, typically from interrupt
719 * false positives. 763 * context. It is less reliable than using the task's vma and may give
764 * false positives:
720 */ 765 */
721int in_gate_area_no_task(unsigned long addr) 766int in_gate_area_no_task(unsigned long addr)
722{ 767{
@@ -736,8 +781,8 @@ const char *arch_vma_name(struct vm_area_struct *vma)
736/* 781/*
737 * Initialise the sparsemem vmemmap using huge-pages at the PMD level. 782 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
738 */ 783 */
739int __meminit vmemmap_populate(struct page *start_page, 784int __meminit
740 unsigned long size, int node) 785vmemmap_populate(struct page *start_page, unsigned long size, int node)
741{ 786{
742 unsigned long addr = (unsigned long)start_page; 787 unsigned long addr = (unsigned long)start_page;
743 unsigned long end = (unsigned long)(start_page + size); 788 unsigned long end = (unsigned long)(start_page + size);
@@ -752,6 +797,7 @@ int __meminit vmemmap_populate(struct page *start_page,
752 pgd = vmemmap_pgd_populate(addr, node); 797 pgd = vmemmap_pgd_populate(addr, node);
753 if (!pgd) 798 if (!pgd)
754 return -ENOMEM; 799 return -ENOMEM;
800
755 pud = vmemmap_pud_populate(pgd, addr, node); 801 pud = vmemmap_pud_populate(pgd, addr, node);
756 if (!pud) 802 if (!pud)
757 return -ENOMEM; 803 return -ENOMEM;
@@ -759,19 +805,22 @@ int __meminit vmemmap_populate(struct page *start_page,
759 pmd = pmd_offset(pud, addr); 805 pmd = pmd_offset(pud, addr);
760 if (pmd_none(*pmd)) { 806 if (pmd_none(*pmd)) {
761 pte_t entry; 807 pte_t entry;
762 void *p = vmemmap_alloc_block(PMD_SIZE, node); 808 void *p;
809
810 p = vmemmap_alloc_block(PMD_SIZE, node);
763 if (!p) 811 if (!p)
764 return -ENOMEM; 812 return -ENOMEM;
765 813
766 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL_LARGE); 814 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
815 PAGE_KERNEL_LARGE);
767 set_pmd(pmd, __pmd(pte_val(entry))); 816 set_pmd(pmd, __pmd(pte_val(entry)));
768 817
769 printk(KERN_DEBUG " [%lx-%lx] PMD ->%p on node %d\n", 818 printk(KERN_DEBUG " [%lx-%lx] PMD ->%p on node %d\n",
770 addr, addr + PMD_SIZE - 1, p, node); 819 addr, addr + PMD_SIZE - 1, p, node);
771 } else 820 } else {
772 vmemmap_verify((pte_t *)pmd, node, addr, next); 821 vmemmap_verify((pte_t *)pmd, node, addr, next);
822 }
773 } 823 }
774
775 return 0; 824 return 0;
776} 825}
777#endif 826#endif