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

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

Version: ~ [ linux-5.19-rc8 ] ~ [ linux-5.18.14 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.57 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.133 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.207 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.253 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.289 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.324 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ 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 #include <linux/swapfile.h>
  8 #include <linux/swapops.h>
  9 
 10 #include <asm/cacheflush.h>
 11 #include <asm/e820.h>
 12 #include <asm/init.h>
 13 #include <asm/page.h>
 14 #include <asm/page_types.h>
 15 #include <asm/sections.h>
 16 #include <asm/setup.h>
 17 #include <asm/tlbflush.h>
 18 #include <asm/tlb.h>
 19 #include <asm/proto.h>
 20 #include <asm/dma.h>            /* for MAX_DMA_PFN */
 21 #include <asm/microcode.h>
 22 
 23 /*
 24  * We need to define the tracepoints somewhere, and tlb.c
 25  * is only compied when SMP=y.
 26  */
 27 #define CREATE_TRACE_POINTS
 28 #include <trace/events/tlb.h>
 29 
 30 #include "mm_internal.h"
 31 
 32 /*
 33  * Tables translating between page_cache_type_t and pte encoding.
 34  *
 35  * The default values are defined statically as minimal supported mode;
 36  * WC and WT fall back to UC-.  pat_init() updates these values to support
 37  * more cache modes, WC and WT, when it is safe to do so.  See pat_init()
 38  * for the details.  Note, __early_ioremap() used during early boot-time
 39  * takes pgprot_t (pte encoding) and does not use these tables.
 40  *
 41  *   Index into __cachemode2pte_tbl[] is the cachemode.
 42  *
 43  *   Index into __pte2cachemode_tbl[] are the caching attribute bits of the pte
 44  *   (_PAGE_PWT, _PAGE_PCD, _PAGE_PAT) at index bit positions 0, 1, 2.
 45  */
 46 uint16_t __cachemode2pte_tbl[_PAGE_CACHE_MODE_NUM] = {
 47         [_PAGE_CACHE_MODE_WB      ]     = 0         | 0        ,
 48         [_PAGE_CACHE_MODE_WC      ]     = 0         | _PAGE_PCD,
 49         [_PAGE_CACHE_MODE_UC_MINUS]     = 0         | _PAGE_PCD,
 50         [_PAGE_CACHE_MODE_UC      ]     = _PAGE_PWT | _PAGE_PCD,
 51         [_PAGE_CACHE_MODE_WT      ]     = 0         | _PAGE_PCD,
 52         [_PAGE_CACHE_MODE_WP      ]     = 0         | _PAGE_PCD,
 53 };
 54 EXPORT_SYMBOL(__cachemode2pte_tbl);
 55 
 56 uint8_t __pte2cachemode_tbl[8] = {
 57         [__pte2cm_idx( 0        | 0         | 0        )] = _PAGE_CACHE_MODE_WB,
 58         [__pte2cm_idx(_PAGE_PWT | 0         | 0        )] = _PAGE_CACHE_MODE_UC_MINUS,
 59         [__pte2cm_idx( 0        | _PAGE_PCD | 0        )] = _PAGE_CACHE_MODE_UC_MINUS,
 60         [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | 0        )] = _PAGE_CACHE_MODE_UC,
 61         [__pte2cm_idx( 0        | 0         | _PAGE_PAT)] = _PAGE_CACHE_MODE_WB,
 62         [__pte2cm_idx(_PAGE_PWT | 0         | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS,
 63         [__pte2cm_idx(0         | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS,
 64         [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC,
 65 };
 66 EXPORT_SYMBOL(__pte2cachemode_tbl);
 67 
 68 static unsigned long __initdata pgt_buf_start;
 69 static unsigned long __initdata pgt_buf_end;
 70 static unsigned long __initdata pgt_buf_top;
 71 
 72 static unsigned long min_pfn_mapped;
 73 
 74 static bool __initdata can_use_brk_pgt = true;
 75 
 76 /*
 77  * Pages returned are already directly mapped.
 78  *
 79  * Changing that is likely to break Xen, see commit:
 80  *
 81  *    279b706 x86,xen: introduce x86_init.mapping.pagetable_reserve
 82  *
 83  * for detailed information.
 84  */
 85 __ref void *alloc_low_pages(unsigned int num)
 86 {
 87         unsigned long pfn;
 88         int i;
 89 
 90         if (after_bootmem) {
 91                 unsigned int order;
 92 
 93                 order = get_order((unsigned long)num << PAGE_SHIFT);
 94                 return (void *)__get_free_pages(GFP_ATOMIC | __GFP_NOTRACK |
 95                                                 __GFP_ZERO, order);
 96         }
 97 
 98         if ((pgt_buf_end + num) > pgt_buf_top || !can_use_brk_pgt) {
 99                 unsigned long ret;
100                 if (min_pfn_mapped >= max_pfn_mapped)
101                         panic("alloc_low_pages: ran out of memory");
102                 ret = memblock_find_in_range(min_pfn_mapped << PAGE_SHIFT,
103                                         max_pfn_mapped << PAGE_SHIFT,
104                                         PAGE_SIZE * num , PAGE_SIZE);
105                 if (!ret)
106                         panic("alloc_low_pages: can not alloc memory");
107                 memblock_reserve(ret, PAGE_SIZE * num);
108                 pfn = ret >> PAGE_SHIFT;
109         } else {
110                 pfn = pgt_buf_end;
111                 pgt_buf_end += num;
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_DEBUG_PAGEALLOC) && !defined(CONFIG_KMEMCHECK)
154         /*
155          * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
156          * This will simplify cpa(), which otherwise needs to support splitting
157          * large pages into small in interrupt context, etc.
158          */
159         if (cpu_has_pse)
160                 page_size_mask |= 1 << PG_LEVEL_2M;
161 #endif
162 
163         /* Enable PSE if available */
164         if (cpu_has_pse)
165                 cr4_set_bits_and_update_boot(X86_CR4_PSE);
166 
167         /* Enable PGE if available */
168         if (cpu_has_pge && !kaiser_enabled) {
169                 cr4_set_bits_and_update_boot(X86_CR4_PGE);
170                 __supported_pte_mask |= _PAGE_GLOBAL;
171         } else
172                 __supported_pte_mask &= ~_PAGE_GLOBAL;
173 
174         /* Enable 1 GB linear kernel mappings if available: */
175         if (direct_gbpages && cpu_has_gbpages) {
176                 printk(KERN_INFO "Using GB pages for direct mapping\n");
177                 page_size_mask |= 1 << PG_LEVEL_1G;
178         } else {
179                 direct_gbpages = 0;
180         }
181 }
182 
183 #ifdef CONFIG_X86_32
184 #define NR_RANGE_MR 3
185 #else /* CONFIG_X86_64 */
186 #define NR_RANGE_MR 5
187 #endif
188 
189 static int __meminit save_mr(struct map_range *mr, int nr_range,
190                              unsigned long start_pfn, unsigned long end_pfn,
191                              unsigned long page_size_mask)
192 {
193         if (start_pfn < end_pfn) {
194                 if (nr_range >= NR_RANGE_MR)
195                         panic("run out of range for init_memory_mapping\n");
196                 mr[nr_range].start = start_pfn<<PAGE_SHIFT;
197                 mr[nr_range].end   = end_pfn<<PAGE_SHIFT;
198                 mr[nr_range].page_size_mask = page_size_mask;
199                 nr_range++;
200         }
201 
202         return nr_range;
203 }
204 
205 /*
206  * adjust the page_size_mask for small range to go with
207  *      big page size instead small one if nearby are ram too.
208  */
209 static void __init_refok adjust_range_page_size_mask(struct map_range *mr,
210                                                          int nr_range)
211 {
212         int i;
213 
214         for (i = 0; i < nr_range; i++) {
215                 if ((page_size_mask & (1<<PG_LEVEL_2M)) &&
216                     !(mr[i].page_size_mask & (1<<PG_LEVEL_2M))) {
217                         unsigned long start = round_down(mr[i].start, PMD_SIZE);
218                         unsigned long end = round_up(mr[i].end, PMD_SIZE);
219 
220 #ifdef CONFIG_X86_32
221                         if ((end >> PAGE_SHIFT) > max_low_pfn)
222                                 continue;
223 #endif
224 
225                         if (memblock_is_region_memory(start, end - start))
226                                 mr[i].page_size_mask |= 1<<PG_LEVEL_2M;
227                 }
228                 if ((page_size_mask & (1<<PG_LEVEL_1G)) &&
229                     !(mr[i].page_size_mask & (1<<PG_LEVEL_1G))) {
230                         unsigned long start = round_down(mr[i].start, PUD_SIZE);
231                         unsigned long end = round_up(mr[i].end, PUD_SIZE);
232 
233                         if (memblock_is_region_memory(start, end - start))
234                                 mr[i].page_size_mask |= 1<<PG_LEVEL_1G;
235                 }
236         }
237 }
238 
239 static const char *page_size_string(struct map_range *mr)
240 {
241         static const char str_1g[] = "1G";
242         static const char str_2m[] = "2M";
243         static const char str_4m[] = "4M";
244         static const char str_4k[] = "4k";
245 
246         if (mr->page_size_mask & (1<<PG_LEVEL_1G))
247                 return str_1g;
248         /*
249          * 32-bit without PAE has a 4M large page size.
250          * PG_LEVEL_2M is misnamed, but we can at least
251          * print out the right size in the string.
252          */
253         if (IS_ENABLED(CONFIG_X86_32) &&
254             !IS_ENABLED(CONFIG_X86_PAE) &&
255             mr->page_size_mask & (1<<PG_LEVEL_2M))
256                 return str_4m;
257 
258         if (mr->page_size_mask & (1<<PG_LEVEL_2M))
259                 return str_2m;
260 
261         return str_4k;
262 }
263 
264 static int __meminit split_mem_range(struct map_range *mr, int nr_range,
265                                      unsigned long start,
266                                      unsigned long end)
267 {
268         unsigned long start_pfn, end_pfn, limit_pfn;
269         unsigned long pfn;
270         int i;
271 
272         limit_pfn = PFN_DOWN(end);
273 
274         /* head if not big page alignment ? */
275         pfn = start_pfn = PFN_DOWN(start);
276 #ifdef CONFIG_X86_32
277         /*
278          * Don't use a large page for the first 2/4MB of memory
279          * because there are often fixed size MTRRs in there
280          * and overlapping MTRRs into large pages can cause
281          * slowdowns.
282          */
283         if (pfn == 0)
284                 end_pfn = PFN_DOWN(PMD_SIZE);
285         else
286                 end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
287 #else /* CONFIG_X86_64 */
288         end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
289 #endif
290         if (end_pfn > limit_pfn)
291                 end_pfn = limit_pfn;
292         if (start_pfn < end_pfn) {
293                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
294                 pfn = end_pfn;
295         }
296 
297         /* big page (2M) range */
298         start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
299 #ifdef CONFIG_X86_32
300         end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE));
301 #else /* CONFIG_X86_64 */
302         end_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE));
303         if (end_pfn > round_down(limit_pfn, PFN_DOWN(PMD_SIZE)))
304                 end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE));
305 #endif
306 
307         if (start_pfn < end_pfn) {
308                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
309                                 page_size_mask & (1<<PG_LEVEL_2M));
310                 pfn = end_pfn;
311         }
312 
313 #ifdef CONFIG_X86_64
314         /* big page (1G) range */
315         start_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE));
316         end_pfn = round_down(limit_pfn, PFN_DOWN(PUD_SIZE));
317         if (start_pfn < end_pfn) {
318                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
319                                 page_size_mask &
320                                  ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G)));
321                 pfn = end_pfn;
322         }
323 
324         /* tail is not big page (1G) alignment */
325         start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
326         end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE));
327         if (start_pfn < end_pfn) {
328                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
329                                 page_size_mask & (1<<PG_LEVEL_2M));
330                 pfn = end_pfn;
331         }
332 #endif
333 
334         /* tail is not big page (2M) alignment */
335         start_pfn = pfn;
336         end_pfn = limit_pfn;
337         nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
338 
339         if (!after_bootmem)
340                 adjust_range_page_size_mask(mr, nr_range);
341 
342         /* try to merge same page size and continuous */
343         for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
344                 unsigned long old_start;
345                 if (mr[i].end != mr[i+1].start ||
346                     mr[i].page_size_mask != mr[i+1].page_size_mask)
347                         continue;
348                 /* move it */
349                 old_start = mr[i].start;
350                 memmove(&mr[i], &mr[i+1],
351                         (nr_range - 1 - i) * sizeof(struct map_range));
352                 mr[i--].start = old_start;
353                 nr_range--;
354         }
355 
356         for (i = 0; i < nr_range; i++)
357                 pr_debug(" [mem %#010lx-%#010lx] page %s\n",
358                                 mr[i].start, mr[i].end - 1,
359                                 page_size_string(&mr[i]));
360 
361         return nr_range;
362 }
363 
364 struct range pfn_mapped[E820_X_MAX];
365 int nr_pfn_mapped;
366 
367 static void add_pfn_range_mapped(unsigned long start_pfn, unsigned long end_pfn)
368 {
369         nr_pfn_mapped = add_range_with_merge(pfn_mapped, E820_X_MAX,
370                                              nr_pfn_mapped, start_pfn, end_pfn);
371         nr_pfn_mapped = clean_sort_range(pfn_mapped, E820_X_MAX);
372 
373         max_pfn_mapped = max(max_pfn_mapped, end_pfn);
374 
375         if (start_pfn < (1UL<<(32-PAGE_SHIFT)))
376                 max_low_pfn_mapped = max(max_low_pfn_mapped,
377                                          min(end_pfn, 1UL<<(32-PAGE_SHIFT)));
378 }
379 
380 bool pfn_range_is_mapped(unsigned long start_pfn, unsigned long end_pfn)
381 {
382         int i;
383 
384         for (i = 0; i < nr_pfn_mapped; i++)
385                 if ((start_pfn >= pfn_mapped[i].start) &&
386                     (end_pfn <= pfn_mapped[i].end))
387                         return true;
388 
389         return false;
390 }
391 
392 /*
393  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
394  * This runs before bootmem is initialized and gets pages directly from
395  * the physical memory. To access them they are temporarily mapped.
396  */
397 unsigned long __init_refok init_memory_mapping(unsigned long start,
398                                                unsigned long end)
399 {
400         struct map_range mr[NR_RANGE_MR];
401         unsigned long ret = 0;
402         int nr_range, i;
403 
404         pr_debug("init_memory_mapping: [mem %#010lx-%#010lx]\n",
405                start, end - 1);
406 
407         memset(mr, 0, sizeof(mr));
408         nr_range = split_mem_range(mr, 0, start, end);
409 
410         for (i = 0; i < nr_range; i++)
411                 ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
412                                                    mr[i].page_size_mask);
413 
414         add_pfn_range_mapped(start >> PAGE_SHIFT, ret >> PAGE_SHIFT);
415 
416         return ret >> PAGE_SHIFT;
417 }
418 
419 /*
420  * We need to iterate through the E820 memory map and create direct mappings
421  * for only E820_RAM and E820_KERN_RESERVED regions. We cannot simply
422  * create direct mappings for all pfns from [0 to max_low_pfn) and
423  * [4GB to max_pfn) because of possible memory holes in high addresses
424  * that cannot be marked as UC by fixed/variable range MTRRs.
425  * Depending on the alignment of E820 ranges, this may possibly result
426  * in using smaller size (i.e. 4K instead of 2M or 1G) page tables.
427  *
428  * init_mem_mapping() calls init_range_memory_mapping() with big range.
429  * That range would have hole in the middle or ends, and only ram parts
430  * will be mapped in init_range_memory_mapping().
431  */
432 static unsigned long __init init_range_memory_mapping(
433                                            unsigned long r_start,
434                                            unsigned long r_end)
435 {
436         unsigned long start_pfn, end_pfn;
437         unsigned long mapped_ram_size = 0;
438         int i;
439 
440         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
441                 u64 start = clamp_val(PFN_PHYS(start_pfn), r_start, r_end);
442                 u64 end = clamp_val(PFN_PHYS(end_pfn), r_start, r_end);
443                 if (start >= end)
444                         continue;
445 
446                 /*
447                  * if it is overlapping with brk pgt, we need to
448                  * alloc pgt buf from memblock instead.
449                  */
450                 can_use_brk_pgt = max(start, (u64)pgt_buf_end<<PAGE_SHIFT) >=
451                                     min(end, (u64)pgt_buf_top<<PAGE_SHIFT);
452                 init_memory_mapping(start, end);
453                 mapped_ram_size += end - start;
454                 can_use_brk_pgt = true;
455         }
456 
457         return mapped_ram_size;
458 }
459 
460 static unsigned long __init get_new_step_size(unsigned long step_size)
461 {
462         /*
463          * Initial mapped size is PMD_SIZE (2M).
464          * We can not set step_size to be PUD_SIZE (1G) yet.
465          * In worse case, when we cross the 1G boundary, and
466          * PG_LEVEL_2M is not set, we will need 1+1+512 pages (2M + 8k)
467          * to map 1G range with PTE. Hence we use one less than the
468          * difference of page table level shifts.
469          *
470          * Don't need to worry about overflow in the top-down case, on 32bit,
471          * when step_size is 0, round_down() returns 0 for start, and that
472          * turns it into 0x100000000ULL.
473          * In the bottom-up case, round_up(x, 0) returns 0 though too, which
474          * needs to be taken into consideration by the code below.
475          */
476         return step_size << (PMD_SHIFT - PAGE_SHIFT - 1);
477 }
478 
479 /**
480  * memory_map_top_down - Map [map_start, map_end) top down
481  * @map_start: start address of the target memory range
482  * @map_end: end address of the target memory range
483  *
484  * This function will setup direct mapping for memory range
485  * [map_start, map_end) in top-down. That said, the page tables
486  * will be allocated at the end of the memory, and we map the
487  * memory in top-down.
488  */
489 static void __init memory_map_top_down(unsigned long map_start,
490                                        unsigned long map_end)
491 {
492         unsigned long real_end, start, last_start;
493         unsigned long step_size;
494         unsigned long addr;
495         unsigned long mapped_ram_size = 0;
496 
497         /* xen has big range in reserved near end of ram, skip it at first.*/
498         addr = memblock_find_in_range(map_start, map_end, PMD_SIZE, PMD_SIZE);
499         real_end = addr + PMD_SIZE;
500 
501         /* step_size need to be small so pgt_buf from BRK could cover it */
502         step_size = PMD_SIZE;
503         max_pfn_mapped = 0; /* will get exact value next */
504         min_pfn_mapped = real_end >> PAGE_SHIFT;
505         last_start = start = real_end;
506 
507         /*
508          * We start from the top (end of memory) and go to the bottom.
509          * The memblock_find_in_range() gets us a block of RAM from the
510          * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages
511          * for page table.
512          */
513         while (last_start > map_start) {
514                 if (last_start > step_size) {
515                         start = round_down(last_start - 1, step_size);
516                         if (start < map_start)
517                                 start = map_start;
518                 } else
519                         start = map_start;
520                 mapped_ram_size += init_range_memory_mapping(start,
521                                                         last_start);
522                 last_start = start;
523                 min_pfn_mapped = last_start >> PAGE_SHIFT;
524                 if (mapped_ram_size >= step_size)
525                         step_size = get_new_step_size(step_size);
526         }
527 
528         if (real_end < map_end)
529                 init_range_memory_mapping(real_end, map_end);
530 }
531 
532 /**
533  * memory_map_bottom_up - Map [map_start, map_end) bottom up
534  * @map_start: start address of the target memory range
535  * @map_end: end address of the target memory range
536  *
537  * This function will setup direct mapping for memory range
538  * [map_start, map_end) in bottom-up. Since we have limited the
539  * bottom-up allocation above the kernel, the page tables will
540  * be allocated just above the kernel and we map the memory
541  * in [map_start, map_end) in bottom-up.
542  */
543 static void __init memory_map_bottom_up(unsigned long map_start,
544                                         unsigned long map_end)
545 {
546         unsigned long next, start;
547         unsigned long mapped_ram_size = 0;
548         /* step_size need to be small so pgt_buf from BRK could cover it */
549         unsigned long step_size = PMD_SIZE;
550 
551         start = map_start;
552         min_pfn_mapped = start >> PAGE_SHIFT;
553 
554         /*
555          * We start from the bottom (@map_start) and go to the top (@map_end).
556          * The memblock_find_in_range() gets us a block of RAM from the
557          * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages
558          * for page table.
559          */
560         while (start < map_end) {
561                 if (step_size && map_end - start > step_size) {
562                         next = round_up(start + 1, step_size);
563                         if (next > map_end)
564                                 next = map_end;
565                 } else {
566                         next = map_end;
567                 }
568 
569                 mapped_ram_size += init_range_memory_mapping(start, next);
570                 start = next;
571 
572                 if (mapped_ram_size >= step_size)
573                         step_size = get_new_step_size(step_size);
574         }
575 }
576 
577 void __init init_mem_mapping(void)
578 {
579         unsigned long end;
580 
581         probe_page_size_mask();
582 
583 #ifdef CONFIG_X86_64
584         end = max_pfn << PAGE_SHIFT;
585 #else
586         end = max_low_pfn << PAGE_SHIFT;
587 #endif
588 
589         /* the ISA range is always mapped regardless of memory holes */
590         init_memory_mapping(0, ISA_END_ADDRESS);
591 
592         /*
593          * If the allocation is in bottom-up direction, we setup direct mapping
594          * in bottom-up, otherwise we setup direct mapping in top-down.
595          */
596         if (memblock_bottom_up()) {
597                 unsigned long kernel_end = __pa_symbol(_end);
598 
599                 /*
600                  * we need two separate calls here. This is because we want to
601                  * allocate page tables above the kernel. So we first map
602                  * [kernel_end, end) to make memory above the kernel be mapped
603                  * as soon as possible. And then use page tables allocated above
604                  * the kernel to map [ISA_END_ADDRESS, kernel_end).
605                  */
606                 memory_map_bottom_up(kernel_end, end);
607                 memory_map_bottom_up(ISA_END_ADDRESS, kernel_end);
608         } else {
609                 memory_map_top_down(ISA_END_ADDRESS, end);
610         }
611 
612 #ifdef CONFIG_X86_64
613         if (max_pfn > max_low_pfn) {
614                 /* can we preseve max_low_pfn ?*/
615                 max_low_pfn = max_pfn;
616         }
617 #else
618         early_ioremap_page_table_range_init();
619 #endif
620 
621         load_cr3(swapper_pg_dir);
622         __flush_tlb_all();
623 
624         early_memtest(0, max_pfn_mapped << PAGE_SHIFT);
625 }
626 
627 /*
628  * devmem_is_allowed() checks to see if /dev/mem access to a certain address
629  * is valid. The argument is a physical page number.
630  *
631  * On x86, access has to be given to the first megabyte of RAM because that
632  * area traditionally contains BIOS code and data regions used by X, dosemu,
633  * and similar apps. Since they map the entire memory range, the whole range
634  * must be allowed (for mapping), but any areas that would otherwise be
635  * disallowed are flagged as being "zero filled" instead of rejected.
636  * Access has to be given to non-kernel-ram areas as well, these contain the
637  * PCI mmio resources as well as potential bios/acpi data regions.
638  */
639 int devmem_is_allowed(unsigned long pagenr)
640 {
641         if (page_is_ram(pagenr)) {
642                 /*
643                  * For disallowed memory regions in the low 1MB range,
644                  * request that the page be shown as all zeros.
645                  */
646                 if (pagenr < 256)
647                         return 2;
648 
649                 return 0;
650         }
651 
652         /*
653          * This must follow RAM test, since System RAM is considered a
654          * restricted resource under CONFIG_STRICT_IOMEM.
655          */
656         if (iomem_is_exclusive(pagenr << PAGE_SHIFT)) {
657                 /* Low 1MB bypasses iomem restrictions. */
658                 if (pagenr < 256)
659                         return 1;
660 
661                 return 0;
662         }
663 
664         return 1;
665 }
666 
667 void free_init_pages(char *what, unsigned long begin, unsigned long end)
668 {
669         unsigned long begin_aligned, end_aligned;
670 
671         /* Make sure boundaries are page aligned */
672         begin_aligned = PAGE_ALIGN(begin);
673         end_aligned   = end & PAGE_MASK;
674 
675         if (WARN_ON(begin_aligned != begin || end_aligned != end)) {
676                 begin = begin_aligned;
677                 end   = end_aligned;
678         }
679 
680         if (begin >= end)
681                 return;
682 
683         /*
684          * If debugging page accesses then do not free this memory but
685          * mark them not present - any buggy init-section access will
686          * create a kernel page fault:
687          */
688 #ifdef CONFIG_DEBUG_PAGEALLOC
689         printk(KERN_INFO "debug: unmapping init [mem %#010lx-%#010lx]\n",
690                 begin, end - 1);
691         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
692 #else
693         /*
694          * We just marked the kernel text read only above, now that
695          * we are going to free part of that, we need to make that
696          * writeable and non-executable first.
697          */
698         set_memory_nx(begin, (end - begin) >> PAGE_SHIFT);
699         set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
700 
701         free_reserved_area((void *)begin, (void *)end, POISON_FREE_INITMEM, what);
702 #endif
703 }
704 
705 void free_initmem(void)
706 {
707         free_init_pages("unused kernel",
708                         (unsigned long)(&__init_begin),
709                         (unsigned long)(&__init_end));
710 }
711 
712 #ifdef CONFIG_BLK_DEV_INITRD
713 void __init free_initrd_mem(unsigned long start, unsigned long end)
714 {
715         /*
716          * Remember, initrd memory may contain microcode or other useful things.
717          * Before we lose initrd mem, we need to find a place to hold them
718          * now that normal virtual memory is enabled.
719          */
720         save_microcode_in_initrd();
721 
722         /*
723          * end could be not aligned, and We can not align that,
724          * decompresser could be confused by aligned initrd_end
725          * We already reserve the end partial page before in
726          *   - i386_start_kernel()
727          *   - x86_64_start_kernel()
728          *   - relocate_initrd()
729          * So here We can do PAGE_ALIGN() safely to get partial page to be freed
730          */
731         free_init_pages("initrd", start, PAGE_ALIGN(end));
732 }
733 #endif
734 
735 void __init zone_sizes_init(void)
736 {
737         unsigned long max_zone_pfns[MAX_NR_ZONES];
738 
739         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
740 
741 #ifdef CONFIG_ZONE_DMA
742         max_zone_pfns[ZONE_DMA]         = min(MAX_DMA_PFN, max_low_pfn);
743 #endif
744 #ifdef CONFIG_ZONE_DMA32
745         max_zone_pfns[ZONE_DMA32]       = min(MAX_DMA32_PFN, max_low_pfn);
746 #endif
747         max_zone_pfns[ZONE_NORMAL]      = max_low_pfn;
748 #ifdef CONFIG_HIGHMEM
749         max_zone_pfns[ZONE_HIGHMEM]     = max_pfn;
750 #endif
751 
752         free_area_init_nodes(max_zone_pfns);
753 }
754 
755 DEFINE_PER_CPU_SHARED_ALIGNED(struct tlb_state, cpu_tlbstate) = {
756         .active_mm = &init_mm,
757         .state = 0,
758         .cr4 = ~0UL,    /* fail hard if we screw up cr4 shadow initialization */
759 };
760 EXPORT_PER_CPU_SYMBOL(cpu_tlbstate);
761 
762 void update_cache_mode_entry(unsigned entry, enum page_cache_mode cache)
763 {
764         /* entry 0 MUST be WB (hardwired to speed up translations) */
765         BUG_ON(!entry && cache != _PAGE_CACHE_MODE_WB);
766 
767         __cachemode2pte_tbl[cache] = __cm_idx2pte(entry);
768         __pte2cachemode_tbl[entry] = cache;
769 }
770 
771 #ifdef CONFIG_SWAP
772 unsigned long max_swapfile_size(void)
773 {
774         unsigned long pages;
775 
776         pages = generic_max_swapfile_size();
777 
778         if (boot_cpu_has_bug(X86_BUG_L1TF)) {
779                 /* Limit the swap file size to MAX_PA/2 for L1TF workaround */
780                 unsigned long long l1tf_limit = l1tf_pfn_limit();
781                 /*
782                  * We encode swap offsets also with 3 bits below those for pfn
783                  * which makes the usable limit higher.
784                  */
785 #if CONFIG_PGTABLE_LEVELS > 2
786                 l1tf_limit <<= PAGE_SHIFT - SWP_OFFSET_FIRST_BIT;
787 #endif
788                 pages = min_t(unsigned long long, l1tf_limit, pages);
789         }
790         return pages;
791 }
792 #endif
793 

~ [ 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