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