Commit | Line | Data |
---|---|---|
5a0e3ad6 | 1 | #include <linux/gfp.h> |
2c1b284e | 2 | #include <linux/initrd.h> |
540aca06 | 3 | #include <linux/ioport.h> |
e5b2bb55 | 4 | #include <linux/swap.h> |
a9ce6bc1 | 5 | #include <linux/memblock.h> |
17623915 | 6 | #include <linux/bootmem.h> /* for max_low_pfn */ |
540aca06 | 7 | |
e5b2bb55 | 8 | #include <asm/cacheflush.h> |
f765090a | 9 | #include <asm/e820.h> |
4fcb2083 | 10 | #include <asm/init.h> |
e5b2bb55 | 11 | #include <asm/page.h> |
540aca06 | 12 | #include <asm/page_types.h> |
e5b2bb55 | 13 | #include <asm/sections.h> |
49834396 | 14 | #include <asm/setup.h> |
f765090a | 15 | #include <asm/tlbflush.h> |
9518e0e4 | 16 | #include <asm/tlb.h> |
76c06927 | 17 | #include <asm/proto.h> |
17623915 | 18 | #include <asm/dma.h> /* for MAX_DMA_PFN */ |
cd745be8 | 19 | #include <asm/microcode.h> |
9518e0e4 | 20 | |
d17d8f9d DH |
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 | ||
5c51bdbe YL |
28 | #include "mm_internal.h" |
29 | ||
281d4078 JG |
30 | /* |
31 | * Tables translating between page_cache_type_t and pte encoding. | |
c709feda IM |
32 | * |
33 | * Minimal supported modes are defined statically, they are modified | |
34 | * during bootup if more supported cache modes are available. | |
35 | * | |
36 | * Index into __cachemode2pte_tbl[] is the cachemode. | |
37 | * | |
38 | * Index into __pte2cachemode_tbl[] are the caching attribute bits of the pte | |
39 | * (_PAGE_PWT, _PAGE_PCD, _PAGE_PAT) at index bit positions 0, 1, 2. | |
281d4078 JG |
40 | */ |
41 | uint16_t __cachemode2pte_tbl[_PAGE_CACHE_MODE_NUM] = { | |
c709feda | 42 | [_PAGE_CACHE_MODE_WB ] = 0 | 0 , |
9cd25aac | 43 | [_PAGE_CACHE_MODE_WC ] = 0 | _PAGE_PCD, |
c709feda IM |
44 | [_PAGE_CACHE_MODE_UC_MINUS] = 0 | _PAGE_PCD, |
45 | [_PAGE_CACHE_MODE_UC ] = _PAGE_PWT | _PAGE_PCD, | |
46 | [_PAGE_CACHE_MODE_WT ] = 0 | _PAGE_PCD, | |
47 | [_PAGE_CACHE_MODE_WP ] = 0 | _PAGE_PCD, | |
281d4078 | 48 | }; |
31bb7723 | 49 | EXPORT_SYMBOL(__cachemode2pte_tbl); |
c709feda | 50 | |
281d4078 | 51 | uint8_t __pte2cachemode_tbl[8] = { |
c709feda | 52 | [__pte2cm_idx( 0 | 0 | 0 )] = _PAGE_CACHE_MODE_WB, |
9cd25aac | 53 | [__pte2cm_idx(_PAGE_PWT | 0 | 0 )] = _PAGE_CACHE_MODE_UC_MINUS, |
c709feda IM |
54 | [__pte2cm_idx( 0 | _PAGE_PCD | 0 )] = _PAGE_CACHE_MODE_UC_MINUS, |
55 | [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | 0 )] = _PAGE_CACHE_MODE_UC, | |
56 | [__pte2cm_idx( 0 | 0 | _PAGE_PAT)] = _PAGE_CACHE_MODE_WB, | |
9cd25aac | 57 | [__pte2cm_idx(_PAGE_PWT | 0 | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS, |
c709feda | 58 | [__pte2cm_idx(0 | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS, |
281d4078 JG |
59 | [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC, |
60 | }; | |
31bb7723 | 61 | EXPORT_SYMBOL(__pte2cachemode_tbl); |
281d4078 | 62 | |
cf470659 YL |
63 | static unsigned long __initdata pgt_buf_start; |
64 | static unsigned long __initdata pgt_buf_end; | |
65 | static unsigned long __initdata pgt_buf_top; | |
f765090a | 66 | |
9985b4c6 YL |
67 | static unsigned long min_pfn_mapped; |
68 | ||
c9b3234a YL |
69 | static bool __initdata can_use_brk_pgt = true; |
70 | ||
ddd3509d SS |
71 | /* |
72 | * Pages returned are already directly mapped. | |
73 | * | |
74 | * Changing that is likely to break Xen, see commit: | |
75 | * | |
76 | * 279b706 x86,xen: introduce x86_init.mapping.pagetable_reserve | |
77 | * | |
78 | * for detailed information. | |
79 | */ | |
22c8ca2a | 80 | __ref void *alloc_low_pages(unsigned int num) |
5c51bdbe YL |
81 | { |
82 | unsigned long pfn; | |
22c8ca2a | 83 | int i; |
5c51bdbe | 84 | |
5c51bdbe | 85 | if (after_bootmem) { |
22c8ca2a | 86 | unsigned int order; |
5c51bdbe | 87 | |
22c8ca2a YL |
88 | order = get_order((unsigned long)num << PAGE_SHIFT); |
89 | return (void *)__get_free_pages(GFP_ATOMIC | __GFP_NOTRACK | | |
90 | __GFP_ZERO, order); | |
5c51bdbe | 91 | } |
5c51bdbe | 92 | |
c9b3234a | 93 | if ((pgt_buf_end + num) > pgt_buf_top || !can_use_brk_pgt) { |
5c51bdbe YL |
94 | unsigned long ret; |
95 | if (min_pfn_mapped >= max_pfn_mapped) | |
d4dd100f | 96 | panic("alloc_low_pages: ran out of memory"); |
5c51bdbe YL |
97 | ret = memblock_find_in_range(min_pfn_mapped << PAGE_SHIFT, |
98 | max_pfn_mapped << PAGE_SHIFT, | |
22c8ca2a | 99 | PAGE_SIZE * num , PAGE_SIZE); |
5c51bdbe | 100 | if (!ret) |
d4dd100f | 101 | panic("alloc_low_pages: can not alloc memory"); |
22c8ca2a | 102 | memblock_reserve(ret, PAGE_SIZE * num); |
5c51bdbe | 103 | pfn = ret >> PAGE_SHIFT; |
22c8ca2a YL |
104 | } else { |
105 | pfn = pgt_buf_end; | |
106 | pgt_buf_end += num; | |
c9b3234a YL |
107 | printk(KERN_DEBUG "BRK [%#010lx, %#010lx] PGTABLE\n", |
108 | pfn << PAGE_SHIFT, (pgt_buf_end << PAGE_SHIFT) - 1); | |
22c8ca2a YL |
109 | } |
110 | ||
111 | for (i = 0; i < num; i++) { | |
112 | void *adr; | |
113 | ||
114 | adr = __va((pfn + i) << PAGE_SHIFT); | |
115 | clear_page(adr); | |
116 | } | |
5c51bdbe | 117 | |
22c8ca2a | 118 | return __va(pfn << PAGE_SHIFT); |
5c51bdbe YL |
119 | } |
120 | ||
527bf129 YL |
121 | /* need 3 4k for initial PMD_SIZE, 3 4k for 0-ISA_END_ADDRESS */ |
122 | #define INIT_PGT_BUF_SIZE (6 * PAGE_SIZE) | |
8d57470d YL |
123 | RESERVE_BRK(early_pgt_alloc, INIT_PGT_BUF_SIZE); |
124 | void __init early_alloc_pgt_buf(void) | |
125 | { | |
126 | unsigned long tables = INIT_PGT_BUF_SIZE; | |
127 | phys_addr_t base; | |
128 | ||
129 | base = __pa(extend_brk(tables, PAGE_SIZE)); | |
130 | ||
131 | pgt_buf_start = base >> PAGE_SHIFT; | |
132 | pgt_buf_end = pgt_buf_start; | |
133 | pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT); | |
134 | } | |
135 | ||
f765090a PE |
136 | int after_bootmem; |
137 | ||
10971ab2 | 138 | early_param_on_off("gbpages", "nogbpages", direct_gbpages, CONFIG_X86_DIRECT_GBPAGES); |
148b2098 | 139 | |
844ab6f9 JS |
140 | struct map_range { |
141 | unsigned long start; | |
142 | unsigned long end; | |
143 | unsigned page_size_mask; | |
144 | }; | |
145 | ||
fa62aafe | 146 | static int page_size_mask; |
f765090a | 147 | |
22ddfcaa | 148 | static void __init probe_page_size_mask(void) |
fa62aafe YL |
149 | { |
150 | #if !defined(CONFIG_DEBUG_PAGEALLOC) && !defined(CONFIG_KMEMCHECK) | |
151 | /* | |
152 | * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages. | |
153 | * This will simplify cpa(), which otherwise needs to support splitting | |
154 | * large pages into small in interrupt context, etc. | |
155 | */ | |
fa62aafe YL |
156 | if (cpu_has_pse) |
157 | page_size_mask |= 1 << PG_LEVEL_2M; | |
158 | #endif | |
159 | ||
160 | /* Enable PSE if available */ | |
161 | if (cpu_has_pse) | |
375074cc | 162 | cr4_set_bits_and_update_boot(X86_CR4_PSE); |
fa62aafe YL |
163 | |
164 | /* Enable PGE if available */ | |
165 | if (cpu_has_pge) { | |
375074cc | 166 | cr4_set_bits_and_update_boot(X86_CR4_PGE); |
fa62aafe | 167 | __supported_pte_mask |= _PAGE_GLOBAL; |
0cdb81be JB |
168 | } else |
169 | __supported_pte_mask &= ~_PAGE_GLOBAL; | |
e61980a7 IM |
170 | |
171 | /* Enable 1 GB linear kernel mappings if available: */ | |
172 | if (direct_gbpages && cpu_has_gbpages) { | |
173 | printk(KERN_INFO "Using GB pages for direct mapping\n"); | |
174 | page_size_mask |= 1 << PG_LEVEL_1G; | |
175 | } else { | |
176 | direct_gbpages = 0; | |
177 | } | |
fa62aafe | 178 | } |
279b706b | 179 | |
f765090a PE |
180 | #ifdef CONFIG_X86_32 |
181 | #define NR_RANGE_MR 3 | |
182 | #else /* CONFIG_X86_64 */ | |
183 | #define NR_RANGE_MR 5 | |
184 | #endif | |
185 | ||
dc9dd5cc JB |
186 | static int __meminit save_mr(struct map_range *mr, int nr_range, |
187 | unsigned long start_pfn, unsigned long end_pfn, | |
188 | unsigned long page_size_mask) | |
f765090a PE |
189 | { |
190 | if (start_pfn < end_pfn) { | |
191 | if (nr_range >= NR_RANGE_MR) | |
192 | panic("run out of range for init_memory_mapping\n"); | |
193 | mr[nr_range].start = start_pfn<<PAGE_SHIFT; | |
194 | mr[nr_range].end = end_pfn<<PAGE_SHIFT; | |
195 | mr[nr_range].page_size_mask = page_size_mask; | |
196 | nr_range++; | |
197 | } | |
198 | ||
199 | return nr_range; | |
200 | } | |
201 | ||
aeebe84c YL |
202 | /* |
203 | * adjust the page_size_mask for small range to go with | |
204 | * big page size instead small one if nearby are ram too. | |
205 | */ | |
206 | static void __init_refok adjust_range_page_size_mask(struct map_range *mr, | |
207 | int nr_range) | |
208 | { | |
209 | int i; | |
210 | ||
211 | for (i = 0; i < nr_range; i++) { | |
212 | if ((page_size_mask & (1<<PG_LEVEL_2M)) && | |
213 | !(mr[i].page_size_mask & (1<<PG_LEVEL_2M))) { | |
214 | unsigned long start = round_down(mr[i].start, PMD_SIZE); | |
215 | unsigned long end = round_up(mr[i].end, PMD_SIZE); | |
216 | ||
217 | #ifdef CONFIG_X86_32 | |
218 | if ((end >> PAGE_SHIFT) > max_low_pfn) | |
219 | continue; | |
220 | #endif | |
221 | ||
222 | if (memblock_is_region_memory(start, end - start)) | |
223 | mr[i].page_size_mask |= 1<<PG_LEVEL_2M; | |
224 | } | |
225 | if ((page_size_mask & (1<<PG_LEVEL_1G)) && | |
226 | !(mr[i].page_size_mask & (1<<PG_LEVEL_1G))) { | |
227 | unsigned long start = round_down(mr[i].start, PUD_SIZE); | |
228 | unsigned long end = round_up(mr[i].end, PUD_SIZE); | |
229 | ||
230 | if (memblock_is_region_memory(start, end - start)) | |
231 | mr[i].page_size_mask |= 1<<PG_LEVEL_1G; | |
232 | } | |
233 | } | |
234 | } | |
235 | ||
f15e0518 DH |
236 | static const char *page_size_string(struct map_range *mr) |
237 | { | |
238 | static const char str_1g[] = "1G"; | |
239 | static const char str_2m[] = "2M"; | |
240 | static const char str_4m[] = "4M"; | |
241 | static const char str_4k[] = "4k"; | |
242 | ||
243 | if (mr->page_size_mask & (1<<PG_LEVEL_1G)) | |
244 | return str_1g; | |
245 | /* | |
246 | * 32-bit without PAE has a 4M large page size. | |
247 | * PG_LEVEL_2M is misnamed, but we can at least | |
248 | * print out the right size in the string. | |
249 | */ | |
250 | if (IS_ENABLED(CONFIG_X86_32) && | |
251 | !IS_ENABLED(CONFIG_X86_PAE) && | |
252 | mr->page_size_mask & (1<<PG_LEVEL_2M)) | |
253 | return str_4m; | |
254 | ||
255 | if (mr->page_size_mask & (1<<PG_LEVEL_2M)) | |
256 | return str_2m; | |
257 | ||
258 | return str_4k; | |
259 | } | |
260 | ||
4e33e065 YL |
261 | static int __meminit split_mem_range(struct map_range *mr, int nr_range, |
262 | unsigned long start, | |
263 | unsigned long end) | |
f765090a | 264 | { |
2e8059ed | 265 | unsigned long start_pfn, end_pfn, limit_pfn; |
1829ae9a | 266 | unsigned long pfn; |
4e33e065 | 267 | int i; |
f765090a | 268 | |
2e8059ed YL |
269 | limit_pfn = PFN_DOWN(end); |
270 | ||
f765090a | 271 | /* head if not big page alignment ? */ |
1829ae9a | 272 | pfn = start_pfn = PFN_DOWN(start); |
f765090a PE |
273 | #ifdef CONFIG_X86_32 |
274 | /* | |
275 | * Don't use a large page for the first 2/4MB of memory | |
276 | * because there are often fixed size MTRRs in there | |
277 | * and overlapping MTRRs into large pages can cause | |
278 | * slowdowns. | |
279 | */ | |
1829ae9a | 280 | if (pfn == 0) |
84d77001 | 281 | end_pfn = PFN_DOWN(PMD_SIZE); |
f765090a | 282 | else |
1829ae9a | 283 | end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 284 | #else /* CONFIG_X86_64 */ |
1829ae9a | 285 | end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 286 | #endif |
2e8059ed YL |
287 | if (end_pfn > limit_pfn) |
288 | end_pfn = limit_pfn; | |
f765090a PE |
289 | if (start_pfn < end_pfn) { |
290 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); | |
1829ae9a | 291 | pfn = end_pfn; |
f765090a PE |
292 | } |
293 | ||
294 | /* big page (2M) range */ | |
1829ae9a | 295 | start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 296 | #ifdef CONFIG_X86_32 |
2e8059ed | 297 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 298 | #else /* CONFIG_X86_64 */ |
1829ae9a | 299 | end_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE)); |
2e8059ed YL |
300 | if (end_pfn > round_down(limit_pfn, PFN_DOWN(PMD_SIZE))) |
301 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); | |
f765090a PE |
302 | #endif |
303 | ||
304 | if (start_pfn < end_pfn) { | |
305 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, | |
306 | page_size_mask & (1<<PG_LEVEL_2M)); | |
1829ae9a | 307 | pfn = end_pfn; |
f765090a PE |
308 | } |
309 | ||
310 | #ifdef CONFIG_X86_64 | |
311 | /* big page (1G) range */ | |
1829ae9a | 312 | start_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE)); |
2e8059ed | 313 | end_pfn = round_down(limit_pfn, PFN_DOWN(PUD_SIZE)); |
f765090a PE |
314 | if (start_pfn < end_pfn) { |
315 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, | |
316 | page_size_mask & | |
317 | ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G))); | |
1829ae9a | 318 | pfn = end_pfn; |
f765090a PE |
319 | } |
320 | ||
321 | /* tail is not big page (1G) alignment */ | |
1829ae9a | 322 | start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
2e8059ed | 323 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); |
f765090a PE |
324 | if (start_pfn < end_pfn) { |
325 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, | |
326 | page_size_mask & (1<<PG_LEVEL_2M)); | |
1829ae9a | 327 | pfn = end_pfn; |
f765090a PE |
328 | } |
329 | #endif | |
330 | ||
331 | /* tail is not big page (2M) alignment */ | |
1829ae9a | 332 | start_pfn = pfn; |
2e8059ed | 333 | end_pfn = limit_pfn; |
f765090a PE |
334 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); |
335 | ||
7de3d66b YL |
336 | if (!after_bootmem) |
337 | adjust_range_page_size_mask(mr, nr_range); | |
338 | ||
f765090a PE |
339 | /* try to merge same page size and continuous */ |
340 | for (i = 0; nr_range > 1 && i < nr_range - 1; i++) { | |
341 | unsigned long old_start; | |
342 | if (mr[i].end != mr[i+1].start || | |
343 | mr[i].page_size_mask != mr[i+1].page_size_mask) | |
344 | continue; | |
345 | /* move it */ | |
346 | old_start = mr[i].start; | |
347 | memmove(&mr[i], &mr[i+1], | |
348 | (nr_range - 1 - i) * sizeof(struct map_range)); | |
349 | mr[i--].start = old_start; | |
350 | nr_range--; | |
351 | } | |
352 | ||
353 | for (i = 0; i < nr_range; i++) | |
365811d6 BH |
354 | printk(KERN_DEBUG " [mem %#010lx-%#010lx] page %s\n", |
355 | mr[i].start, mr[i].end - 1, | |
f15e0518 | 356 | page_size_string(&mr[i])); |
f765090a | 357 | |
4e33e065 YL |
358 | return nr_range; |
359 | } | |
360 | ||
0e691cf8 YL |
361 | struct range pfn_mapped[E820_X_MAX]; |
362 | int nr_pfn_mapped; | |
66520ebc JS |
363 | |
364 | static void add_pfn_range_mapped(unsigned long start_pfn, unsigned long end_pfn) | |
365 | { | |
366 | nr_pfn_mapped = add_range_with_merge(pfn_mapped, E820_X_MAX, | |
367 | nr_pfn_mapped, start_pfn, end_pfn); | |
368 | nr_pfn_mapped = clean_sort_range(pfn_mapped, E820_X_MAX); | |
369 | ||
370 | max_pfn_mapped = max(max_pfn_mapped, end_pfn); | |
371 | ||
372 | if (start_pfn < (1UL<<(32-PAGE_SHIFT))) | |
373 | max_low_pfn_mapped = max(max_low_pfn_mapped, | |
374 | min(end_pfn, 1UL<<(32-PAGE_SHIFT))); | |
375 | } | |
376 | ||
377 | bool pfn_range_is_mapped(unsigned long start_pfn, unsigned long end_pfn) | |
378 | { | |
379 | int i; | |
380 | ||
381 | for (i = 0; i < nr_pfn_mapped; i++) | |
382 | if ((start_pfn >= pfn_mapped[i].start) && | |
383 | (end_pfn <= pfn_mapped[i].end)) | |
384 | return true; | |
385 | ||
386 | return false; | |
387 | } | |
388 | ||
4e33e065 YL |
389 | /* |
390 | * Setup the direct mapping of the physical memory at PAGE_OFFSET. | |
391 | * This runs before bootmem is initialized and gets pages directly from | |
392 | * the physical memory. To access them they are temporarily mapped. | |
393 | */ | |
394 | unsigned long __init_refok init_memory_mapping(unsigned long start, | |
395 | unsigned long end) | |
396 | { | |
397 | struct map_range mr[NR_RANGE_MR]; | |
398 | unsigned long ret = 0; | |
399 | int nr_range, i; | |
400 | ||
401 | pr_info("init_memory_mapping: [mem %#010lx-%#010lx]\n", | |
402 | start, end - 1); | |
403 | ||
404 | memset(mr, 0, sizeof(mr)); | |
405 | nr_range = split_mem_range(mr, 0, start, end); | |
406 | ||
f765090a PE |
407 | for (i = 0; i < nr_range; i++) |
408 | ret = kernel_physical_mapping_init(mr[i].start, mr[i].end, | |
409 | mr[i].page_size_mask); | |
f765090a | 410 | |
66520ebc JS |
411 | add_pfn_range_mapped(start >> PAGE_SHIFT, ret >> PAGE_SHIFT); |
412 | ||
c14fa0b6 YL |
413 | return ret >> PAGE_SHIFT; |
414 | } | |
415 | ||
66520ebc | 416 | /* |
cf8b166d ZY |
417 | * We need to iterate through the E820 memory map and create direct mappings |
418 | * for only E820_RAM and E820_KERN_RESERVED regions. We cannot simply | |
419 | * create direct mappings for all pfns from [0 to max_low_pfn) and | |
420 | * [4GB to max_pfn) because of possible memory holes in high addresses | |
421 | * that cannot be marked as UC by fixed/variable range MTRRs. | |
422 | * Depending on the alignment of E820 ranges, this may possibly result | |
423 | * in using smaller size (i.e. 4K instead of 2M or 1G) page tables. | |
424 | * | |
425 | * init_mem_mapping() calls init_range_memory_mapping() with big range. | |
426 | * That range would have hole in the middle or ends, and only ram parts | |
427 | * will be mapped in init_range_memory_mapping(). | |
66520ebc | 428 | */ |
8d57470d | 429 | static unsigned long __init init_range_memory_mapping( |
b8fd39c0 YL |
430 | unsigned long r_start, |
431 | unsigned long r_end) | |
66520ebc JS |
432 | { |
433 | unsigned long start_pfn, end_pfn; | |
8d57470d | 434 | unsigned long mapped_ram_size = 0; |
66520ebc JS |
435 | int i; |
436 | ||
66520ebc | 437 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) { |
b8fd39c0 YL |
438 | u64 start = clamp_val(PFN_PHYS(start_pfn), r_start, r_end); |
439 | u64 end = clamp_val(PFN_PHYS(end_pfn), r_start, r_end); | |
440 | if (start >= end) | |
66520ebc JS |
441 | continue; |
442 | ||
c9b3234a YL |
443 | /* |
444 | * if it is overlapping with brk pgt, we need to | |
445 | * alloc pgt buf from memblock instead. | |
446 | */ | |
447 | can_use_brk_pgt = max(start, (u64)pgt_buf_end<<PAGE_SHIFT) >= | |
448 | min(end, (u64)pgt_buf_top<<PAGE_SHIFT); | |
f763ad1d | 449 | init_memory_mapping(start, end); |
8d57470d | 450 | mapped_ram_size += end - start; |
c9b3234a | 451 | can_use_brk_pgt = true; |
66520ebc | 452 | } |
8d57470d YL |
453 | |
454 | return mapped_ram_size; | |
66520ebc JS |
455 | } |
456 | ||
6979287a YL |
457 | static unsigned long __init get_new_step_size(unsigned long step_size) |
458 | { | |
459 | /* | |
132978b9 | 460 | * Initial mapped size is PMD_SIZE (2M). |
6979287a YL |
461 | * We can not set step_size to be PUD_SIZE (1G) yet. |
462 | * In worse case, when we cross the 1G boundary, and | |
463 | * PG_LEVEL_2M is not set, we will need 1+1+512 pages (2M + 8k) | |
132978b9 JB |
464 | * to map 1G range with PTE. Hence we use one less than the |
465 | * difference of page table level shifts. | |
6979287a | 466 | * |
132978b9 JB |
467 | * Don't need to worry about overflow in the top-down case, on 32bit, |
468 | * when step_size is 0, round_down() returns 0 for start, and that | |
469 | * turns it into 0x100000000ULL. | |
470 | * In the bottom-up case, round_up(x, 0) returns 0 though too, which | |
471 | * needs to be taken into consideration by the code below. | |
6979287a | 472 | */ |
132978b9 | 473 | return step_size << (PMD_SHIFT - PAGE_SHIFT - 1); |
6979287a YL |
474 | } |
475 | ||
0167d7d8 TC |
476 | /** |
477 | * memory_map_top_down - Map [map_start, map_end) top down | |
478 | * @map_start: start address of the target memory range | |
479 | * @map_end: end address of the target memory range | |
480 | * | |
481 | * This function will setup direct mapping for memory range | |
482 | * [map_start, map_end) in top-down. That said, the page tables | |
483 | * will be allocated at the end of the memory, and we map the | |
484 | * memory in top-down. | |
485 | */ | |
486 | static void __init memory_map_top_down(unsigned long map_start, | |
487 | unsigned long map_end) | |
c14fa0b6 | 488 | { |
0167d7d8 | 489 | unsigned long real_end, start, last_start; |
8d57470d YL |
490 | unsigned long step_size; |
491 | unsigned long addr; | |
492 | unsigned long mapped_ram_size = 0; | |
ab951937 | 493 | |
98e7a989 | 494 | /* xen has big range in reserved near end of ram, skip it at first.*/ |
0167d7d8 | 495 | addr = memblock_find_in_range(map_start, map_end, PMD_SIZE, PMD_SIZE); |
8d57470d YL |
496 | real_end = addr + PMD_SIZE; |
497 | ||
498 | /* step_size need to be small so pgt_buf from BRK could cover it */ | |
499 | step_size = PMD_SIZE; | |
500 | max_pfn_mapped = 0; /* will get exact value next */ | |
501 | min_pfn_mapped = real_end >> PAGE_SHIFT; | |
502 | last_start = start = real_end; | |
cf8b166d ZY |
503 | |
504 | /* | |
505 | * We start from the top (end of memory) and go to the bottom. | |
506 | * The memblock_find_in_range() gets us a block of RAM from the | |
507 | * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages | |
508 | * for page table. | |
509 | */ | |
0167d7d8 | 510 | while (last_start > map_start) { |
8d57470d YL |
511 | if (last_start > step_size) { |
512 | start = round_down(last_start - 1, step_size); | |
0167d7d8 TC |
513 | if (start < map_start) |
514 | start = map_start; | |
8d57470d | 515 | } else |
0167d7d8 | 516 | start = map_start; |
132978b9 | 517 | mapped_ram_size += init_range_memory_mapping(start, |
8d57470d YL |
518 | last_start); |
519 | last_start = start; | |
520 | min_pfn_mapped = last_start >> PAGE_SHIFT; | |
132978b9 | 521 | if (mapped_ram_size >= step_size) |
6979287a | 522 | step_size = get_new_step_size(step_size); |
8d57470d YL |
523 | } |
524 | ||
0167d7d8 TC |
525 | if (real_end < map_end) |
526 | init_range_memory_mapping(real_end, map_end); | |
527 | } | |
528 | ||
b959ed6c TC |
529 | /** |
530 | * memory_map_bottom_up - Map [map_start, map_end) bottom up | |
531 | * @map_start: start address of the target memory range | |
532 | * @map_end: end address of the target memory range | |
533 | * | |
534 | * This function will setup direct mapping for memory range | |
535 | * [map_start, map_end) in bottom-up. Since we have limited the | |
536 | * bottom-up allocation above the kernel, the page tables will | |
537 | * be allocated just above the kernel and we map the memory | |
538 | * in [map_start, map_end) in bottom-up. | |
539 | */ | |
540 | static void __init memory_map_bottom_up(unsigned long map_start, | |
541 | unsigned long map_end) | |
542 | { | |
132978b9 | 543 | unsigned long next, start; |
b959ed6c TC |
544 | unsigned long mapped_ram_size = 0; |
545 | /* step_size need to be small so pgt_buf from BRK could cover it */ | |
546 | unsigned long step_size = PMD_SIZE; | |
547 | ||
548 | start = map_start; | |
549 | min_pfn_mapped = start >> PAGE_SHIFT; | |
550 | ||
551 | /* | |
552 | * We start from the bottom (@map_start) and go to the top (@map_end). | |
553 | * The memblock_find_in_range() gets us a block of RAM from the | |
554 | * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages | |
555 | * for page table. | |
556 | */ | |
557 | while (start < map_end) { | |
132978b9 | 558 | if (step_size && map_end - start > step_size) { |
b959ed6c TC |
559 | next = round_up(start + 1, step_size); |
560 | if (next > map_end) | |
561 | next = map_end; | |
132978b9 | 562 | } else { |
b959ed6c | 563 | next = map_end; |
132978b9 | 564 | } |
b959ed6c | 565 | |
132978b9 | 566 | mapped_ram_size += init_range_memory_mapping(start, next); |
b959ed6c TC |
567 | start = next; |
568 | ||
132978b9 | 569 | if (mapped_ram_size >= step_size) |
b959ed6c | 570 | step_size = get_new_step_size(step_size); |
b959ed6c TC |
571 | } |
572 | } | |
573 | ||
0167d7d8 TC |
574 | void __init init_mem_mapping(void) |
575 | { | |
576 | unsigned long end; | |
577 | ||
578 | probe_page_size_mask(); | |
579 | ||
580 | #ifdef CONFIG_X86_64 | |
581 | end = max_pfn << PAGE_SHIFT; | |
582 | #else | |
583 | end = max_low_pfn << PAGE_SHIFT; | |
584 | #endif | |
585 | ||
586 | /* the ISA range is always mapped regardless of memory holes */ | |
587 | init_memory_mapping(0, ISA_END_ADDRESS); | |
588 | ||
b959ed6c TC |
589 | /* |
590 | * If the allocation is in bottom-up direction, we setup direct mapping | |
591 | * in bottom-up, otherwise we setup direct mapping in top-down. | |
592 | */ | |
593 | if (memblock_bottom_up()) { | |
594 | unsigned long kernel_end = __pa_symbol(_end); | |
595 | ||
596 | /* | |
597 | * we need two separate calls here. This is because we want to | |
598 | * allocate page tables above the kernel. So we first map | |
599 | * [kernel_end, end) to make memory above the kernel be mapped | |
600 | * as soon as possible. And then use page tables allocated above | |
601 | * the kernel to map [ISA_END_ADDRESS, kernel_end). | |
602 | */ | |
603 | memory_map_bottom_up(kernel_end, end); | |
604 | memory_map_bottom_up(ISA_END_ADDRESS, kernel_end); | |
605 | } else { | |
606 | memory_map_top_down(ISA_END_ADDRESS, end); | |
607 | } | |
8d57470d | 608 | |
f763ad1d YL |
609 | #ifdef CONFIG_X86_64 |
610 | if (max_pfn > max_low_pfn) { | |
611 | /* can we preseve max_low_pfn ?*/ | |
612 | max_low_pfn = max_pfn; | |
613 | } | |
719272c4 YL |
614 | #else |
615 | early_ioremap_page_table_range_init(); | |
8170e6be PA |
616 | #endif |
617 | ||
719272c4 YL |
618 | load_cr3(swapper_pg_dir); |
619 | __flush_tlb_all(); | |
719272c4 | 620 | |
c14fa0b6 | 621 | early_memtest(0, max_pfn_mapped << PAGE_SHIFT); |
22ddfcaa | 622 | } |
e5b2bb55 | 623 | |
540aca06 PE |
624 | /* |
625 | * devmem_is_allowed() checks to see if /dev/mem access to a certain address | |
626 | * is valid. The argument is a physical page number. | |
627 | * | |
628 | * | |
629 | * On x86, access has to be given to the first megabyte of ram because that area | |
801a5591 | 630 | * contains BIOS code and data regions used by X and dosemu and similar apps. |
540aca06 PE |
631 | * Access has to be given to non-kernel-ram areas as well, these contain the PCI |
632 | * mmio resources as well as potential bios/acpi data regions. | |
633 | */ | |
634 | int devmem_is_allowed(unsigned long pagenr) | |
635 | { | |
73e8f3d7 | 636 | if (pagenr < 256) |
540aca06 PE |
637 | return 1; |
638 | if (iomem_is_exclusive(pagenr << PAGE_SHIFT)) | |
639 | return 0; | |
640 | if (!page_is_ram(pagenr)) | |
641 | return 1; | |
642 | return 0; | |
643 | } | |
644 | ||
e5b2bb55 PE |
645 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
646 | { | |
c967da6a | 647 | unsigned long begin_aligned, end_aligned; |
e5b2bb55 | 648 | |
c967da6a YL |
649 | /* Make sure boundaries are page aligned */ |
650 | begin_aligned = PAGE_ALIGN(begin); | |
651 | end_aligned = end & PAGE_MASK; | |
652 | ||
653 | if (WARN_ON(begin_aligned != begin || end_aligned != end)) { | |
654 | begin = begin_aligned; | |
655 | end = end_aligned; | |
656 | } | |
657 | ||
658 | if (begin >= end) | |
e5b2bb55 PE |
659 | return; |
660 | ||
661 | /* | |
662 | * If debugging page accesses then do not free this memory but | |
663 | * mark them not present - any buggy init-section access will | |
664 | * create a kernel page fault: | |
665 | */ | |
666 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
365811d6 BH |
667 | printk(KERN_INFO "debug: unmapping init [mem %#010lx-%#010lx]\n", |
668 | begin, end - 1); | |
e5b2bb55 PE |
669 | set_memory_np(begin, (end - begin) >> PAGE_SHIFT); |
670 | #else | |
671 | /* | |
672 | * We just marked the kernel text read only above, now that | |
673 | * we are going to free part of that, we need to make that | |
5bd5a452 | 674 | * writeable and non-executable first. |
e5b2bb55 | 675 | */ |
5bd5a452 | 676 | set_memory_nx(begin, (end - begin) >> PAGE_SHIFT); |
e5b2bb55 PE |
677 | set_memory_rw(begin, (end - begin) >> PAGE_SHIFT); |
678 | ||
c88442ec | 679 | free_reserved_area((void *)begin, (void *)end, POISON_FREE_INITMEM, what); |
e5b2bb55 PE |
680 | #endif |
681 | } | |
682 | ||
683 | void free_initmem(void) | |
684 | { | |
c88442ec | 685 | free_init_pages("unused kernel", |
e5b2bb55 PE |
686 | (unsigned long)(&__init_begin), |
687 | (unsigned long)(&__init_end)); | |
688 | } | |
731ddea6 PE |
689 | |
690 | #ifdef CONFIG_BLK_DEV_INITRD | |
0d26d1d8 | 691 | void __init free_initrd_mem(unsigned long start, unsigned long end) |
731ddea6 | 692 | { |
cd745be8 FY |
693 | #ifdef CONFIG_MICROCODE_EARLY |
694 | /* | |
695 | * Remember, initrd memory may contain microcode or other useful things. | |
696 | * Before we lose initrd mem, we need to find a place to hold them | |
697 | * now that normal virtual memory is enabled. | |
698 | */ | |
699 | save_microcode_in_initrd(); | |
700 | #endif | |
701 | ||
c967da6a YL |
702 | /* |
703 | * end could be not aligned, and We can not align that, | |
704 | * decompresser could be confused by aligned initrd_end | |
705 | * We already reserve the end partial page before in | |
706 | * - i386_start_kernel() | |
707 | * - x86_64_start_kernel() | |
708 | * - relocate_initrd() | |
709 | * So here We can do PAGE_ALIGN() safely to get partial page to be freed | |
710 | */ | |
c88442ec | 711 | free_init_pages("initrd", start, PAGE_ALIGN(end)); |
731ddea6 PE |
712 | } |
713 | #endif | |
17623915 PE |
714 | |
715 | void __init zone_sizes_init(void) | |
716 | { | |
717 | unsigned long max_zone_pfns[MAX_NR_ZONES]; | |
718 | ||
719 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
720 | ||
721 | #ifdef CONFIG_ZONE_DMA | |
c072b90c | 722 | max_zone_pfns[ZONE_DMA] = min(MAX_DMA_PFN, max_low_pfn); |
17623915 PE |
723 | #endif |
724 | #ifdef CONFIG_ZONE_DMA32 | |
c072b90c | 725 | max_zone_pfns[ZONE_DMA32] = min(MAX_DMA32_PFN, max_low_pfn); |
17623915 PE |
726 | #endif |
727 | max_zone_pfns[ZONE_NORMAL] = max_low_pfn; | |
728 | #ifdef CONFIG_HIGHMEM | |
729 | max_zone_pfns[ZONE_HIGHMEM] = max_pfn; | |
730 | #endif | |
731 | ||
732 | free_area_init_nodes(max_zone_pfns); | |
733 | } | |
734 | ||
1e02ce4c AL |
735 | DEFINE_PER_CPU_SHARED_ALIGNED(struct tlb_state, cpu_tlbstate) = { |
736 | #ifdef CONFIG_SMP | |
737 | .active_mm = &init_mm, | |
738 | .state = 0, | |
739 | #endif | |
740 | .cr4 = ~0UL, /* fail hard if we screw up cr4 shadow initialization */ | |
741 | }; | |
742 | EXPORT_SYMBOL_GPL(cpu_tlbstate); | |
743 | ||
bd809af1 JG |
744 | void update_cache_mode_entry(unsigned entry, enum page_cache_mode cache) |
745 | { | |
746 | /* entry 0 MUST be WB (hardwired to speed up translations) */ | |
747 | BUG_ON(!entry && cache != _PAGE_CACHE_MODE_WB); | |
748 | ||
749 | __cachemode2pte_tbl[cache] = __cm_idx2pte(entry); | |
750 | __pte2cachemode_tbl[entry] = cache; | |
751 | } |