~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/x86/mm/init.c

Version: ~ [ linux-5.12-rc1 ] ~ [ linux-5.11.2 ] ~ [ linux-5.10.19 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.101 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.177 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.222 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.258 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.258 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 #include <linux/gfp.h>
  2 #include <linux/initrd.h>
  3 #include <linux/ioport.h>
  4 #include <linux/swap.h>
  5 #include <linux/memblock.h>
  6 #include <linux/bootmem.h>      /* for max_low_pfn */
  7 
  8 #include <asm/cacheflush.h>
  9 #include <asm/e820.h>
 10 #include <asm/init.h>
 11 #include <asm/page.h>
 12 #include <asm/page_types.h>
 13 #include <asm/sections.h>
 14 #include <asm/setup.h>
 15 #include <asm/tlbflush.h>
 16 #include <asm/tlb.h>
 17 #include <asm/proto.h>
 18 #include <asm/dma.h>            /* for MAX_DMA_PFN */
 19 #include <asm/microcode.h>
 20 
 21 /*
 22  * We need to define the tracepoints somewhere, and tlb.c
 23  * is only compied when SMP=y.
 24  */
 25 #define CREATE_TRACE_POINTS
 26 #include <trace/events/tlb.h>
 27 
 28 #include "mm_internal.h"
 29 
 30 /*
 31  * Tables translating between page_cache_type_t and pte encoding.
 32  *
 33  * The default values are defined statically as minimal supported mode;
 34  * WC and WT fall back to UC-.  pat_init() updates these values to support
 35  * more cache modes, WC and WT, when it is safe to do so.  See pat_init()
 36  * for the details.  Note, __early_ioremap() used during early boot-time
 37  * takes pgprot_t (pte encoding) and does not use these tables.
 38  *
 39  *   Index into __cachemode2pte_tbl[] is the cachemode.
 40  *
 41  *   Index into __pte2cachemode_tbl[] are the caching attribute bits of the pte
 42  *   (_PAGE_PWT, _PAGE_PCD, _PAGE_PAT) at index bit positions 0, 1, 2.
 43  */
 44 uint16_t __cachemode2pte_tbl[_PAGE_CACHE_MODE_NUM] = {
 45         [_PAGE_CACHE_MODE_WB      ]     = 0         | 0        ,
 46         [_PAGE_CACHE_MODE_WC      ]     = 0         | _PAGE_PCD,
 47         [_PAGE_CACHE_MODE_UC_MINUS]     = 0         | _PAGE_PCD,
 48         [_PAGE_CACHE_MODE_UC      ]     = _PAGE_PWT | _PAGE_PCD,
 49         [_PAGE_CACHE_MODE_WT      ]     = 0         | _PAGE_PCD,
 50         [_PAGE_CACHE_MODE_WP      ]     = 0         | _PAGE_PCD,
 51 };
 52 EXPORT_SYMBOL(__cachemode2pte_tbl);
 53 
 54 uint8_t __pte2cachemode_tbl[8] = {
 55         [__pte2cm_idx( 0        | 0         | 0        )] = _PAGE_CACHE_MODE_WB,
 56         [__pte2cm_idx(_PAGE_PWT | 0         | 0        )] = _PAGE_CACHE_MODE_UC_MINUS,
 57         [__pte2cm_idx( 0        | _PAGE_PCD | 0        )] = _PAGE_CACHE_MODE_UC_MINUS,
 58         [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | 0        )] = _PAGE_CACHE_MODE_UC,
 59         [__pte2cm_idx( 0        | 0         | _PAGE_PAT)] = _PAGE_CACHE_MODE_WB,
 60         [__pte2cm_idx(_PAGE_PWT | 0         | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS,
 61         [__pte2cm_idx(0         | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS,
 62         [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC,
 63 };
 64 EXPORT_SYMBOL(__pte2cachemode_tbl);
 65 
 66 static unsigned long __initdata pgt_buf_start;
 67 static unsigned long __initdata pgt_buf_end;
 68 static unsigned long __initdata pgt_buf_top;
 69 
 70 static unsigned long min_pfn_mapped;
 71 
 72 static bool __initdata can_use_brk_pgt = true;
 73 
 74 /*
 75  * Pages returned are already directly mapped.
 76  *
 77  * Changing that is likely to break Xen, see commit:
 78  *
 79  *    279b706 x86,xen: introduce x86_init.mapping.pagetable_reserve
 80  *
 81  * for detailed information.
 82  */
 83 __ref void *alloc_low_pages(unsigned int num)
 84 {
 85         unsigned long pfn;
 86         int i;
 87 
 88         if (after_bootmem) {
 89                 unsigned int order;
 90 
 91                 order = get_order((unsigned long)num << PAGE_SHIFT);
 92                 return (void *)__get_free_pages(GFP_ATOMIC | __GFP_NOTRACK |
 93                                                 __GFP_ZERO, order);
 94         }
 95 
 96         if ((pgt_buf_end + num) > pgt_buf_top || !can_use_brk_pgt) {
 97                 unsigned long ret;
 98                 if (min_pfn_mapped >= max_pfn_mapped)
 99                         panic("alloc_low_pages: ran out of memory");
100                 ret = memblock_find_in_range(min_pfn_mapped << PAGE_SHIFT,
101                                         max_pfn_mapped << PAGE_SHIFT,
102                                         PAGE_SIZE * num , PAGE_SIZE);
103                 if (!ret)
104                         panic("alloc_low_pages: can not alloc memory");
105                 memblock_reserve(ret, PAGE_SIZE * num);
106                 pfn = ret >> PAGE_SHIFT;
107         } else {
108                 pfn = pgt_buf_end;
109                 pgt_buf_end += num;
110                 printk(KERN_DEBUG "BRK [%#010lx, %#010lx] PGTABLE\n",
111                         pfn << PAGE_SHIFT, (pgt_buf_end << PAGE_SHIFT) - 1);
112         }
113 
114         for (i = 0; i < num; i++) {
115                 void *adr;
116 
117                 adr = __va((pfn + i) << PAGE_SHIFT);
118                 clear_page(adr);
119         }
120 
121         return __va(pfn << PAGE_SHIFT);
122 }
123 
124 /* need 3 4k for initial PMD_SIZE,  3 4k for 0-ISA_END_ADDRESS */
125 #define INIT_PGT_BUF_SIZE       (6 * PAGE_SIZE)
126 RESERVE_BRK(early_pgt_alloc, INIT_PGT_BUF_SIZE);
127 void  __init early_alloc_pgt_buf(void)
128 {
129         unsigned long tables = INIT_PGT_BUF_SIZE;
130         phys_addr_t base;
131 
132         base = __pa(extend_brk(tables, PAGE_SIZE));
133 
134         pgt_buf_start = base >> PAGE_SHIFT;
135         pgt_buf_end = pgt_buf_start;
136         pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT);
137 }
138 
139 int after_bootmem;
140 
141 early_param_on_off("gbpages", "nogbpages", direct_gbpages, CONFIG_X86_DIRECT_GBPAGES);
142 
143 struct map_range {
144         unsigned long start;
145         unsigned long end;
146         unsigned page_size_mask;
147 };
148 
149 static int page_size_mask;
150 
151 static void __init probe_page_size_mask(void)
152 {
153 #if !defined(CONFIG_KMEMCHECK)
154         /*
155          * For CONFIG_KMEMCHECK or pagealloc debugging, identity mapping will
156          * use small pages.
157          * This will simplify cpa(), which otherwise needs to support splitting
158          * large pages into small in interrupt context, etc.
159          */
160         if (boot_cpu_has(X86_FEATURE_PSE) && !debug_pagealloc_enabled())
161                 page_size_mask |= 1 << PG_LEVEL_2M;
162 #endif
163 
164         /* Enable PSE if available */
165         if (boot_cpu_has(X86_FEATURE_PSE))
166                 cr4_set_bits_and_update_boot(X86_CR4_PSE);
167 
168         /* Enable PGE if available */
169         if (boot_cpu_has(X86_FEATURE_PGE)) {
170                 cr4_set_bits_and_update_boot(X86_CR4_PGE);
171                 __supported_pte_mask |= _PAGE_GLOBAL;
172         } else
173                 __supported_pte_mask &= ~_PAGE_GLOBAL;
174 
175         /* Enable 1 GB linear kernel mappings if available: */
176         if (direct_gbpages && boot_cpu_has(X86_FEATURE_GBPAGES)) {
177                 printk(KERN_INFO "Using GB pages for direct mapping\n");
178                 page_size_mask |= 1 << PG_LEVEL_1G;
179         } else {
180                 direct_gbpages = 0;
181         }
182 }
183 
184 #ifdef CONFIG_X86_32
185 #define NR_RANGE_MR 3
186 #else /* CONFIG_X86_64 */
187 #define NR_RANGE_MR 5
188 #endif
189 
190 static int __meminit save_mr(struct map_range *mr, int nr_range,
191                              unsigned long start_pfn, unsigned long end_pfn,
192                              unsigned long page_size_mask)
193 {
194         if (start_pfn < end_pfn) {
195                 if (nr_range >= NR_RANGE_MR)
196                         panic("run out of range for init_memory_mapping\n");
197                 mr[nr_range].start = start_pfn<<PAGE_SHIFT;
198                 mr[nr_range].end   = end_pfn<<PAGE_SHIFT;
199                 mr[nr_range].page_size_mask = page_size_mask;
200                 nr_range++;
201         }
202 
203         return nr_range;
204 }
205 
206 /*
207  * adjust the page_size_mask for small range to go with
208  *      big page size instead small one if nearby are ram too.
209  */
210 static void __init_refok adjust_range_page_size_mask(struct map_range *mr,
211                                                          int nr_range)
212 {
213         int i;
214 
215         for (i = 0; i < nr_range; i++) {
216                 if ((page_size_mask & (1<<PG_LEVEL_2M)) &&
217                     !(mr[i].page_size_mask & (1<<PG_LEVEL_2M))) {
218                         unsigned long start = round_down(mr[i].start, PMD_SIZE);
219                         unsigned long end = round_up(mr[i].end, PMD_SIZE);
220 
221 #ifdef CONFIG_X86_32
222                         if ((end >> PAGE_SHIFT) > max_low_pfn)
223                                 continue;
224 #endif
225 
226                         if (memblock_is_region_memory(start, end - start))
227                                 mr[i].page_size_mask |= 1<<PG_LEVEL_2M;
228                 }
229                 if ((page_size_mask & (1<<PG_LEVEL_1G)) &&
230                     !(mr[i].page_size_mask & (1<<PG_LEVEL_1G))) {
231                         unsigned long start = round_down(mr[i].start, PUD_SIZE);
232                         unsigned long end = round_up(mr[i].end, PUD_SIZE);
233 
234                         if (memblock_is_region_memory(start, end - start))
235                                 mr[i].page_size_mask |= 1<<PG_LEVEL_1G;
236                 }
237         }
238 }
239 
240 static const char *page_size_string(struct map_range *mr)
241 {
242         static const char str_1g[] = "1G";
243         static const char str_2m[] = "2M";
244         static const char str_4m[] = "4M";
245         static const char str_4k[] = "4k";
246 
247         if (mr->page_size_mask & (1<<PG_LEVEL_1G))
248                 return str_1g;
249         /*
250          * 32-bit without PAE has a 4M large page size.
251          * PG_LEVEL_2M is misnamed, but we can at least
252          * print out the right size in the string.
253          */
254         if (IS_ENABLED(CONFIG_X86_32) &&
255             !IS_ENABLED(CONFIG_X86_PAE) &&
256             mr->page_size_mask & (1<<PG_LEVEL_2M))
257                 return str_4m;
258 
259         if (mr->page_size_mask & (1<<PG_LEVEL_2M))
260                 return str_2m;
261 
262         return str_4k;
263 }
264 
265 static int __meminit split_mem_range(struct map_range *mr, int nr_range,
266                                      unsigned long start,
267                                      unsigned long end)
268 {
269         unsigned long start_pfn, end_pfn, limit_pfn;
270         unsigned long pfn;
271         int i;
272 
273         limit_pfn = PFN_DOWN(end);
274 
275         /* head if not big page alignment ? */
276         pfn = start_pfn = PFN_DOWN(start);
277 #ifdef CONFIG_X86_32
278         /*
279          * Don't use a large page for the first 2/4MB of memory
280          * because there are often fixed size MTRRs in there
281          * and overlapping MTRRs into large pages can cause
282          * slowdowns.
283          */
284         if (pfn == 0)
285                 end_pfn = PFN_DOWN(PMD_SIZE);
286         else
287                 end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
288 #else /* CONFIG_X86_64 */
289         end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
290 #endif
291         if (end_pfn > limit_pfn)
292                 end_pfn = limit_pfn;
293         if (start_pfn < end_pfn) {
294                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
295                 pfn = end_pfn;
296         }
297 
298         /* big page (2M) range */
299         start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
300 #ifdef CONFIG_X86_32
301         end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE));
302 #else /* CONFIG_X86_64 */
303         end_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE));
304         if (end_pfn > round_down(limit_pfn, PFN_DOWN(PMD_SIZE)))
305                 end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE));
306 #endif
307 
308         if (start_pfn < end_pfn) {
309                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
310                                 page_size_mask & (1<<PG_LEVEL_2M));
311                 pfn = end_pfn;
312         }
313 
314 #ifdef CONFIG_X86_64
315         /* big page (1G) range */
316         start_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE));
317         end_pfn = round_down(limit_pfn, PFN_DOWN(PUD_SIZE));
318         if (start_pfn < end_pfn) {
319                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
320                                 page_size_mask &
321                                  ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G)));
322                 pfn = end_pfn;
323         }
324 
325         /* tail is not big page (1G) alignment */
326         start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
327         end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE));
328         if (start_pfn < end_pfn) {
329                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
330                                 page_size_mask & (1<<PG_LEVEL_2M));
331                 pfn = end_pfn;
332         }
333 #endif
334 
335         /* tail is not big page (2M) alignment */
336         start_pfn = pfn;
337         end_pfn = limit_pfn;
338         nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
339 
340         if (!after_bootmem)
341                 adjust_range_page_size_mask(mr, nr_range);
342 
343         /* try to merge same page size and continuous */
344         for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
345                 unsigned long old_start;
346                 if (mr[i].end != mr[i+1].start ||
347                     mr[i].page_size_mask != mr[i+1].page_size_mask)
348                         continue;
349                 /* move it */
350                 old_start = mr[i].start;
351                 memmove(&mr[i], &mr[i+1],
352                         (nr_range - 1 - i) * sizeof(struct map_range));
353                 mr[i--].start = old_start;
354                 nr_range--;
355         }
356 
357         for (i = 0; i < nr_range; i++)
358                 pr_debug(" [mem %#010lx-%#010lx] page %s\n",
359                                 mr[i].start, mr[i].end - 1,
360                                 page_size_string(&mr[i]));
361 
362         return nr_range;
363 }
364 
365 struct range pfn_mapped[E820_X_MAX];
366 int nr_pfn_mapped;
367 
368 static void add_pfn_range_mapped(unsigned long start_pfn, unsigned long end_pfn)
369 {
370         nr_pfn_mapped = add_range_with_merge(pfn_mapped, E820_X_MAX,
371                                              nr_pfn_mapped, start_pfn, end_pfn);
372         nr_pfn_mapped = clean_sort_range(pfn_mapped, E820_X_MAX);
373 
374         max_pfn_mapped = max(max_pfn_mapped, end_pfn);
375 
376         if (start_pfn < (1UL<<(32-PAGE_SHIFT)))
377                 max_low_pfn_mapped = max(max_low_pfn_mapped,
378                                          min(end_pfn, 1UL<<(32-PAGE_SHIFT)));
379 }
380 
381 bool pfn_range_is_mapped(unsigned long start_pfn, unsigned long end_pfn)
382 {
383         int i;
384 
385         for (i = 0; i < nr_pfn_mapped; i++)
386                 if ((start_pfn >= pfn_mapped[i].start) &&
387                     (end_pfn <= pfn_mapped[i].end))
388                         return true;
389 
390         return false;
391 }
392 
393 /*
394  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
395  * This runs before bootmem is initialized and gets pages directly from
396  * the physical memory. To access them they are temporarily mapped.
397  */
398 unsigned long __init_refok init_memory_mapping(unsigned long start,
399                                                unsigned long end)
400 {
401         struct map_range mr[NR_RANGE_MR];
402         unsigned long ret = 0;
403         int nr_range, i;
404 
405         pr_debug("init_memory_mapping: [mem %#010lx-%#010lx]\n",
406                start, end - 1);
407 
408         memset(mr, 0, sizeof(mr));
409         nr_range = split_mem_range(mr, 0, start, end);
410 
411         for (i = 0; i < nr_range; i++)
412                 ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
413                                                    mr[i].page_size_mask);
414 
415         add_pfn_range_mapped(start >> PAGE_SHIFT, ret >> PAGE_SHIFT);
416 
417         return ret >> PAGE_SHIFT;
418 }
419 
420 /*
421  * We need to iterate through the E820 memory map and create direct mappings
422  * for only E820_RAM and E820_KERN_RESERVED regions. We cannot simply
423  * create direct mappings for all pfns from [0 to max_low_pfn) and
424  * [4GB to max_pfn) because of possible memory holes in high addresses
425  * that cannot be marked as UC by fixed/variable range MTRRs.
426  * Depending on the alignment of E820 ranges, this may possibly result
427  * in using smaller size (i.e. 4K instead of 2M or 1G) page tables.
428  *
429  * init_mem_mapping() calls init_range_memory_mapping() with big range.
430  * That range would have hole in the middle or ends, and only ram parts
431  * will be mapped in init_range_memory_mapping().
432  */
433 static unsigned long __init init_range_memory_mapping(
434                                            unsigned long r_start,
435                                            unsigned long r_end)
436 {
437         unsigned long start_pfn, end_pfn;
438         unsigned long mapped_ram_size = 0;
439         int i;
440 
441         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
442                 u64 start = clamp_val(PFN_PHYS(start_pfn), r_start, r_end);
443                 u64 end = clamp_val(PFN_PHYS(end_pfn), r_start, r_end);
444                 if (start >= end)
445                         continue;
446 
447                 /*
448                  * if it is overlapping with brk pgt, we need to
449                  * alloc pgt buf from memblock instead.
450                  */
451                 can_use_brk_pgt = max(start, (u64)pgt_buf_end<<PAGE_SHIFT) >=
452                                     min(end, (u64)pgt_buf_top<<PAGE_SHIFT);
453                 init_memory_mapping(start, end);
454                 mapped_ram_size += end - start;
455                 can_use_brk_pgt = true;
456         }
457 
458         return mapped_ram_size;
459 }
460 
461 static unsigned long __init get_new_step_size(unsigned long step_size)
462 {
463         /*
464          * Initial mapped size is PMD_SIZE (2M).
465          * We can not set step_size to be PUD_SIZE (1G) yet.
466          * In worse case, when we cross the 1G boundary, and
467          * PG_LEVEL_2M is not set, we will need 1+1+512 pages (2M + 8k)
468          * to map 1G range with PTE. Hence we use one less than the
469          * difference of page table level shifts.
470          *
471          * Don't need to worry about overflow in the top-down case, on 32bit,
472          * when step_size is 0, round_down() returns 0 for start, and that
473          * turns it into 0x100000000ULL.
474          * In the bottom-up case, round_up(x, 0) returns 0 though too, which
475          * needs to be taken into consideration by the code below.
476          */
477         return step_size << (PMD_SHIFT - PAGE_SHIFT - 1);
478 }
479 
480 /**
481  * memory_map_top_down - Map [map_start, map_end) top down
482  * @map_start: start address of the target memory range
483  * @map_end: end address of the target memory range
484  *
485  * This function will setup direct mapping for memory range
486  * [map_start, map_end) in top-down. That said, the page tables
487  * will be allocated at the end of the memory, and we map the
488  * memory in top-down.
489  */
490 static void __init memory_map_top_down(unsigned long map_start,
491                                        unsigned long map_end)
492 {
493         unsigned long real_end, start, last_start;
494         unsigned long step_size;
495         unsigned long addr;
496         unsigned long mapped_ram_size = 0;
497 
498         /* xen has big range in reserved near end of ram, skip it at first.*/
499         addr = memblock_find_in_range(map_start, map_end, PMD_SIZE, PMD_SIZE);
500         real_end = addr + PMD_SIZE;
501 
502         /* step_size need to be small so pgt_buf from BRK could cover it */
503         step_size = PMD_SIZE;
504         max_pfn_mapped = 0; /* will get exact value next */
505         min_pfn_mapped = real_end >> PAGE_SHIFT;
506         last_start = start = real_end;
507 
508         /*
509          * We start from the top (end of memory) and go to the bottom.
510          * The memblock_find_in_range() gets us a block of RAM from the
511          * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages
512          * for page table.
513          */
514         while (last_start > map_start) {
515                 if (last_start > step_size) {
516                         start = round_down(last_start - 1, step_size);
517                         if (start < map_start)
518                                 start = map_start;
519                 } else
520                         start = map_start;
521                 mapped_ram_size += init_range_memory_mapping(start,
522                                                         last_start);
523                 last_start = start;
524                 min_pfn_mapped = last_start >> PAGE_SHIFT;
525                 if (mapped_ram_size >= step_size)
526                         step_size = get_new_step_size(step_size);
527         }
528 
529         if (real_end < map_end)
530                 init_range_memory_mapping(real_end, map_end);
531 }
532 
533 /**
534  * memory_map_bottom_up - Map [map_start, map_end) bottom up
535  * @map_start: start address of the target memory range
536  * @map_end: end address of the target memory range
537  *
538  * This function will setup direct mapping for memory range
539  * [map_start, map_end) in bottom-up. Since we have limited the
540  * bottom-up allocation above the kernel, the page tables will
541  * be allocated just above the kernel and we map the memory
542  * in [map_start, map_end) in bottom-up.
543  */
544 static void __init memory_map_bottom_up(unsigned long map_start,
545                                         unsigned long map_end)
546 {
547         unsigned long next, start;
548         unsigned long mapped_ram_size = 0;
549         /* step_size need to be small so pgt_buf from BRK could cover it */
550         unsigned long step_size = PMD_SIZE;
551 
552         start = map_start;
553         min_pfn_mapped = start >> PAGE_SHIFT;
554 
555         /*
556          * We start from the bottom (@map_start) and go to the top (@map_end).
557          * The memblock_find_in_range() gets us a block of RAM from the
558          * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages
559          * for page table.
560          */
561         while (start < map_end) {
562                 if (step_size && map_end - start > step_size) {
563                         next = round_up(start + 1, step_size);
564                         if (next > map_end)
565                                 next = map_end;
566                 } else {
567                         next = map_end;
568                 }
569 
570                 mapped_ram_size += init_range_memory_mapping(start, next);
571                 start = next;
572 
573                 if (mapped_ram_size >= step_size)
574                         step_size = get_new_step_size(step_size);
575         }
576 }
577 
578 void __init init_mem_mapping(void)
579 {
580         unsigned long end;
581 
582         probe_page_size_mask();
583 
584 #ifdef CONFIG_X86_64
585         end = max_pfn << PAGE_SHIFT;
586 #else
587         end = max_low_pfn << PAGE_SHIFT;
588 #endif
589 
590         /* the ISA range is always mapped regardless of memory holes */
591         init_memory_mapping(0, ISA_END_ADDRESS);
592 
593         /*
594          * If the allocation is in bottom-up direction, we setup direct mapping
595          * in bottom-up, otherwise we setup direct mapping in top-down.
596          */
597         if (memblock_bottom_up()) {
598                 unsigned long kernel_end = __pa_symbol(_end);
599 
600                 /*
601                  * we need two separate calls here. This is because we want to
602                  * allocate page tables above the kernel. So we first map
603                  * [kernel_end, end) to make memory above the kernel be mapped
604                  * as soon as possible. And then use page tables allocated above
605                  * the kernel to map [ISA_END_ADDRESS, kernel_end).
606                  */
607                 memory_map_bottom_up(kernel_end, end);
608                 memory_map_bottom_up(ISA_END_ADDRESS, kernel_end);
609         } else {
610                 memory_map_top_down(ISA_END_ADDRESS, end);
611         }
612 
613 #ifdef CONFIG_X86_64
614         if (max_pfn > max_low_pfn) {
615                 /* can we preseve max_low_pfn ?*/
616                 max_low_pfn = max_pfn;
617         }
618 #else
619         early_ioremap_page_table_range_init();
620 #endif
621 
622         load_cr3(swapper_pg_dir);
623         __flush_tlb_all();
624 
625         early_memtest(0, max_pfn_mapped << PAGE_SHIFT);
626 }
627 
628 /*
629  * devmem_is_allowed() checks to see if /dev/mem access to a certain address
630  * is valid. The argument is a physical page number.
631  *
632  *
633  * On x86, access has to be given to the first megabyte of ram because that area
634  * contains BIOS code and data regions used by X and dosemu and similar apps.
635  * Access has to be given to non-kernel-ram areas as well, these contain the PCI
636  * mmio resources as well as potential bios/acpi data regions.
637  */
638 int devmem_is_allowed(unsigned long pagenr)
639 {
640         if (pagenr < 256)
641                 return 1;
642         if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
643                 return 0;
644         if (!page_is_ram(pagenr))
645                 return 1;
646         return 0;
647 }
648 
649 void free_init_pages(char *what, unsigned long begin, unsigned long end)
650 {
651         unsigned long begin_aligned, end_aligned;
652 
653         /* Make sure boundaries are page aligned */
654         begin_aligned = PAGE_ALIGN(begin);
655         end_aligned   = end & PAGE_MASK;
656 
657         if (WARN_ON(begin_aligned != begin || end_aligned != end)) {
658                 begin = begin_aligned;
659                 end   = end_aligned;
660         }
661 
662         if (begin >= end)
663                 return;
664 
665         /*
666          * If debugging page accesses then do not free this memory but
667          * mark them not present - any buggy init-section access will
668          * create a kernel page fault:
669          */
670         if (debug_pagealloc_enabled()) {
671                 pr_info("debug: unmapping init [mem %#010lx-%#010lx]\n",
672                         begin, end - 1);
673                 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
674         } else {
675                 /*
676                  * We just marked the kernel text read only above, now that
677                  * we are going to free part of that, we need to make that
678                  * writeable and non-executable first.
679                  */
680                 set_memory_nx(begin, (end - begin) >> PAGE_SHIFT);
681                 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
682 
683                 free_reserved_area((void *)begin, (void *)end,
684                                    POISON_FREE_INITMEM, what);
685         }
686 }
687 
688 void free_initmem(void)
689 {
690         free_init_pages("unused kernel",
691                         (unsigned long)(&__init_begin),
692                         (unsigned long)(&__init_end));
693 }
694 
695 #ifdef CONFIG_BLK_DEV_INITRD
696 void __init free_initrd_mem(unsigned long start, unsigned long end)
697 {
698         /*
699          * end could be not aligned, and We can not align that,
700          * decompresser could be confused by aligned initrd_end
701          * We already reserve the end partial page before in
702          *   - i386_start_kernel()
703          *   - x86_64_start_kernel()
704          *   - relocate_initrd()
705          * So here We can do PAGE_ALIGN() safely to get partial page to be freed
706          */
707         free_init_pages("initrd", start, PAGE_ALIGN(end));
708 }
709 #endif
710 
711 void __init zone_sizes_init(void)
712 {
713         unsigned long max_zone_pfns[MAX_NR_ZONES];
714 
715         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
716 
717 #ifdef CONFIG_ZONE_DMA
718         max_zone_pfns[ZONE_DMA]         = min(MAX_DMA_PFN, max_low_pfn);
719 #endif
720 #ifdef CONFIG_ZONE_DMA32
721         max_zone_pfns[ZONE_DMA32]       = min(MAX_DMA32_PFN, max_low_pfn);
722 #endif
723         max_zone_pfns[ZONE_NORMAL]      = max_low_pfn;
724 #ifdef CONFIG_HIGHMEM
725         max_zone_pfns[ZONE_HIGHMEM]     = max_pfn;
726 #endif
727 
728         free_area_init_nodes(max_zone_pfns);
729 }
730 
731 DEFINE_PER_CPU_SHARED_ALIGNED(struct tlb_state, cpu_tlbstate) = {
732 #ifdef CONFIG_SMP
733         .active_mm = &init_mm,
734         .state = 0,
735 #endif
736         .cr4 = ~0UL,    /* fail hard if we screw up cr4 shadow initialization */
737 };
738 EXPORT_SYMBOL_GPL(cpu_tlbstate);
739 
740 void update_cache_mode_entry(unsigned entry, enum page_cache_mode cache)
741 {
742         /* entry 0 MUST be WB (hardwired to speed up translations) */
743         BUG_ON(!entry && cache != _PAGE_CACHE_MODE_WB);
744 
745         __cachemode2pte_tbl[cache] = __cm_idx2pte(entry);
746         __pte2cachemode_tbl[entry] = cache;
747 }
748 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp