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 | 32 | * |
d5dc861b TK |
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. | |
c709feda IM |
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. | |
281d4078 JG |
43 | */ |
44 | uint16_t __cachemode2pte_tbl[_PAGE_CACHE_MODE_NUM] = { | |
c709feda | 45 | [_PAGE_CACHE_MODE_WB ] = 0 | 0 , |
9cd25aac | 46 | [_PAGE_CACHE_MODE_WC ] = 0 | _PAGE_PCD, |
c709feda IM |
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, | |
281d4078 | 51 | }; |
31bb7723 | 52 | EXPORT_SYMBOL(__cachemode2pte_tbl); |
c709feda | 53 | |
281d4078 | 54 | uint8_t __pte2cachemode_tbl[8] = { |
c709feda | 55 | [__pte2cm_idx( 0 | 0 | 0 )] = _PAGE_CACHE_MODE_WB, |
9cd25aac | 56 | [__pte2cm_idx(_PAGE_PWT | 0 | 0 )] = _PAGE_CACHE_MODE_UC_MINUS, |
c709feda IM |
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, | |
9cd25aac | 60 | [__pte2cm_idx(_PAGE_PWT | 0 | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS, |
c709feda | 61 | [__pte2cm_idx(0 | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS, |
281d4078 JG |
62 | [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC, |
63 | }; | |
31bb7723 | 64 | EXPORT_SYMBOL(__pte2cachemode_tbl); |
281d4078 | 65 | |
cf470659 YL |
66 | static unsigned long __initdata pgt_buf_start; |
67 | static unsigned long __initdata pgt_buf_end; | |
68 | static unsigned long __initdata pgt_buf_top; | |
f765090a | 69 | |
9985b4c6 YL |
70 | static unsigned long min_pfn_mapped; |
71 | ||
c9b3234a YL |
72 | static bool __initdata can_use_brk_pgt = true; |
73 | ||
ddd3509d SS |
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 | */ | |
22c8ca2a | 83 | __ref void *alloc_low_pages(unsigned int num) |
5c51bdbe YL |
84 | { |
85 | unsigned long pfn; | |
22c8ca2a | 86 | int i; |
5c51bdbe | 87 | |
5c51bdbe | 88 | if (after_bootmem) { |
22c8ca2a | 89 | unsigned int order; |
5c51bdbe | 90 | |
22c8ca2a YL |
91 | order = get_order((unsigned long)num << PAGE_SHIFT); |
92 | return (void *)__get_free_pages(GFP_ATOMIC | __GFP_NOTRACK | | |
93 | __GFP_ZERO, order); | |
5c51bdbe | 94 | } |
5c51bdbe | 95 | |
c9b3234a | 96 | if ((pgt_buf_end + num) > pgt_buf_top || !can_use_brk_pgt) { |
5c51bdbe YL |
97 | unsigned long ret; |
98 | if (min_pfn_mapped >= max_pfn_mapped) | |
d4dd100f | 99 | panic("alloc_low_pages: ran out of memory"); |
5c51bdbe YL |
100 | ret = memblock_find_in_range(min_pfn_mapped << PAGE_SHIFT, |
101 | max_pfn_mapped << PAGE_SHIFT, | |
22c8ca2a | 102 | PAGE_SIZE * num , PAGE_SIZE); |
5c51bdbe | 103 | if (!ret) |
d4dd100f | 104 | panic("alloc_low_pages: can not alloc memory"); |
22c8ca2a | 105 | memblock_reserve(ret, PAGE_SIZE * num); |
5c51bdbe | 106 | pfn = ret >> PAGE_SHIFT; |
22c8ca2a YL |
107 | } else { |
108 | pfn = pgt_buf_end; | |
109 | pgt_buf_end += num; | |
c9b3234a YL |
110 | printk(KERN_DEBUG "BRK [%#010lx, %#010lx] PGTABLE\n", |
111 | pfn << PAGE_SHIFT, (pgt_buf_end << PAGE_SHIFT) - 1); | |
22c8ca2a YL |
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 | } | |
5c51bdbe | 120 | |
22c8ca2a | 121 | return __va(pfn << PAGE_SHIFT); |
5c51bdbe YL |
122 | } |
123 | ||
527bf129 YL |
124 | /* need 3 4k for initial PMD_SIZE, 3 4k for 0-ISA_END_ADDRESS */ |
125 | #define INIT_PGT_BUF_SIZE (6 * PAGE_SIZE) | |
8d57470d YL |
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 | ||
f765090a PE |
139 | int after_bootmem; |
140 | ||
10971ab2 | 141 | early_param_on_off("gbpages", "nogbpages", direct_gbpages, CONFIG_X86_DIRECT_GBPAGES); |
148b2098 | 142 | |
844ab6f9 JS |
143 | struct map_range { |
144 | unsigned long start; | |
145 | unsigned long end; | |
146 | unsigned page_size_mask; | |
147 | }; | |
148 | ||
fa62aafe | 149 | static int page_size_mask; |
f765090a | 150 | |
22ddfcaa | 151 | static void __init probe_page_size_mask(void) |
fa62aafe | 152 | { |
288cf3c6 | 153 | #if !defined(CONFIG_KMEMCHECK) |
fa62aafe | 154 | /* |
288cf3c6 CB |
155 | * For CONFIG_KMEMCHECK or pagealloc debugging, identity mapping will |
156 | * use small pages. | |
fa62aafe YL |
157 | * This will simplify cpa(), which otherwise needs to support splitting |
158 | * large pages into small in interrupt context, etc. | |
159 | */ | |
288cf3c6 | 160 | if (cpu_has_pse && !debug_pagealloc_enabled()) |
fa62aafe YL |
161 | page_size_mask |= 1 << PG_LEVEL_2M; |
162 | #endif | |
163 | ||
164 | /* Enable PSE if available */ | |
165 | if (cpu_has_pse) | |
375074cc | 166 | cr4_set_bits_and_update_boot(X86_CR4_PSE); |
fa62aafe YL |
167 | |
168 | /* Enable PGE if available */ | |
169 | if (cpu_has_pge) { | |
375074cc | 170 | cr4_set_bits_and_update_boot(X86_CR4_PGE); |
fa62aafe | 171 | __supported_pte_mask |= _PAGE_GLOBAL; |
0cdb81be JB |
172 | } else |
173 | __supported_pte_mask &= ~_PAGE_GLOBAL; | |
e61980a7 IM |
174 | |
175 | /* Enable 1 GB linear kernel mappings if available: */ | |
176 | if (direct_gbpages && cpu_has_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 | } | |
fa62aafe | 182 | } |
279b706b | 183 | |
f765090a PE |
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 | ||
dc9dd5cc JB |
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) | |
f765090a PE |
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 | ||
aeebe84c YL |
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 | ||
f15e0518 DH |
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 | ||
4e33e065 YL |
265 | static int __meminit split_mem_range(struct map_range *mr, int nr_range, |
266 | unsigned long start, | |
267 | unsigned long end) | |
f765090a | 268 | { |
2e8059ed | 269 | unsigned long start_pfn, end_pfn, limit_pfn; |
1829ae9a | 270 | unsigned long pfn; |
4e33e065 | 271 | int i; |
f765090a | 272 | |
2e8059ed YL |
273 | limit_pfn = PFN_DOWN(end); |
274 | ||
f765090a | 275 | /* head if not big page alignment ? */ |
1829ae9a | 276 | pfn = start_pfn = PFN_DOWN(start); |
f765090a PE |
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 | */ | |
1829ae9a | 284 | if (pfn == 0) |
84d77001 | 285 | end_pfn = PFN_DOWN(PMD_SIZE); |
f765090a | 286 | else |
1829ae9a | 287 | end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 288 | #else /* CONFIG_X86_64 */ |
1829ae9a | 289 | end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 290 | #endif |
2e8059ed YL |
291 | if (end_pfn > limit_pfn) |
292 | end_pfn = limit_pfn; | |
f765090a PE |
293 | if (start_pfn < end_pfn) { |
294 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); | |
1829ae9a | 295 | pfn = end_pfn; |
f765090a PE |
296 | } |
297 | ||
298 | /* big page (2M) range */ | |
1829ae9a | 299 | start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 300 | #ifdef CONFIG_X86_32 |
2e8059ed | 301 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 302 | #else /* CONFIG_X86_64 */ |
1829ae9a | 303 | end_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE)); |
2e8059ed YL |
304 | if (end_pfn > round_down(limit_pfn, PFN_DOWN(PMD_SIZE))) |
305 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); | |
f765090a PE |
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)); | |
1829ae9a | 311 | pfn = end_pfn; |
f765090a PE |
312 | } |
313 | ||
314 | #ifdef CONFIG_X86_64 | |
315 | /* big page (1G) range */ | |
1829ae9a | 316 | start_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE)); |
2e8059ed | 317 | end_pfn = round_down(limit_pfn, PFN_DOWN(PUD_SIZE)); |
f765090a PE |
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))); | |
1829ae9a | 322 | pfn = end_pfn; |
f765090a PE |
323 | } |
324 | ||
325 | /* tail is not big page (1G) alignment */ | |
1829ae9a | 326 | start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
2e8059ed | 327 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); |
f765090a PE |
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)); | |
1829ae9a | 331 | pfn = end_pfn; |
f765090a PE |
332 | } |
333 | #endif | |
334 | ||
335 | /* tail is not big page (2M) alignment */ | |
1829ae9a | 336 | start_pfn = pfn; |
2e8059ed | 337 | end_pfn = limit_pfn; |
f765090a PE |
338 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); |
339 | ||
7de3d66b YL |
340 | if (!after_bootmem) |
341 | adjust_range_page_size_mask(mr, nr_range); | |
342 | ||
f765090a PE |
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++) | |
c9cdaeb2 | 358 | pr_debug(" [mem %#010lx-%#010lx] page %s\n", |
365811d6 | 359 | mr[i].start, mr[i].end - 1, |
f15e0518 | 360 | page_size_string(&mr[i])); |
f765090a | 361 | |
4e33e065 YL |
362 | return nr_range; |
363 | } | |
364 | ||
0e691cf8 YL |
365 | struct range pfn_mapped[E820_X_MAX]; |
366 | int nr_pfn_mapped; | |
66520ebc JS |
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 | ||
4e33e065 YL |
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 | ||
c9cdaeb2 | 405 | pr_debug("init_memory_mapping: [mem %#010lx-%#010lx]\n", |
4e33e065 YL |
406 | start, end - 1); |
407 | ||
408 | memset(mr, 0, sizeof(mr)); | |
409 | nr_range = split_mem_range(mr, 0, start, end); | |
410 | ||
f765090a PE |
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); | |
f765090a | 414 | |
66520ebc JS |
415 | add_pfn_range_mapped(start >> PAGE_SHIFT, ret >> PAGE_SHIFT); |
416 | ||
c14fa0b6 YL |
417 | return ret >> PAGE_SHIFT; |
418 | } | |
419 | ||
66520ebc | 420 | /* |
cf8b166d ZY |
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(). | |
66520ebc | 432 | */ |
8d57470d | 433 | static unsigned long __init init_range_memory_mapping( |
b8fd39c0 YL |
434 | unsigned long r_start, |
435 | unsigned long r_end) | |
66520ebc JS |
436 | { |
437 | unsigned long start_pfn, end_pfn; | |
8d57470d | 438 | unsigned long mapped_ram_size = 0; |
66520ebc JS |
439 | int i; |
440 | ||
66520ebc | 441 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) { |
b8fd39c0 YL |
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) | |
66520ebc JS |
445 | continue; |
446 | ||
c9b3234a YL |
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); | |
f763ad1d | 453 | init_memory_mapping(start, end); |
8d57470d | 454 | mapped_ram_size += end - start; |
c9b3234a | 455 | can_use_brk_pgt = true; |
66520ebc | 456 | } |
8d57470d YL |
457 | |
458 | return mapped_ram_size; | |
66520ebc JS |
459 | } |
460 | ||
6979287a YL |
461 | static unsigned long __init get_new_step_size(unsigned long step_size) |
462 | { | |
463 | /* | |
132978b9 | 464 | * Initial mapped size is PMD_SIZE (2M). |
6979287a YL |
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) | |
132978b9 JB |
468 | * to map 1G range with PTE. Hence we use one less than the |
469 | * difference of page table level shifts. | |
6979287a | 470 | * |
132978b9 JB |
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. | |
6979287a | 476 | */ |
132978b9 | 477 | return step_size << (PMD_SHIFT - PAGE_SHIFT - 1); |
6979287a YL |
478 | } |
479 | ||
0167d7d8 TC |
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) | |
c14fa0b6 | 492 | { |
0167d7d8 | 493 | unsigned long real_end, start, last_start; |
8d57470d YL |
494 | unsigned long step_size; |
495 | unsigned long addr; | |
496 | unsigned long mapped_ram_size = 0; | |
ab951937 | 497 | |
98e7a989 | 498 | /* xen has big range in reserved near end of ram, skip it at first.*/ |
0167d7d8 | 499 | addr = memblock_find_in_range(map_start, map_end, PMD_SIZE, PMD_SIZE); |
8d57470d YL |
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; | |
cf8b166d ZY |
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 | */ | |
0167d7d8 | 514 | while (last_start > map_start) { |
8d57470d YL |
515 | if (last_start > step_size) { |
516 | start = round_down(last_start - 1, step_size); | |
0167d7d8 TC |
517 | if (start < map_start) |
518 | start = map_start; | |
8d57470d | 519 | } else |
0167d7d8 | 520 | start = map_start; |
132978b9 | 521 | mapped_ram_size += init_range_memory_mapping(start, |
8d57470d YL |
522 | last_start); |
523 | last_start = start; | |
524 | min_pfn_mapped = last_start >> PAGE_SHIFT; | |
132978b9 | 525 | if (mapped_ram_size >= step_size) |
6979287a | 526 | step_size = get_new_step_size(step_size); |
8d57470d YL |
527 | } |
528 | ||
0167d7d8 TC |
529 | if (real_end < map_end) |
530 | init_range_memory_mapping(real_end, map_end); | |
531 | } | |
532 | ||
b959ed6c TC |
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 | { | |
132978b9 | 547 | unsigned long next, start; |
b959ed6c TC |
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) { | |
132978b9 | 562 | if (step_size && map_end - start > step_size) { |
b959ed6c TC |
563 | next = round_up(start + 1, step_size); |
564 | if (next > map_end) | |
565 | next = map_end; | |
132978b9 | 566 | } else { |
b959ed6c | 567 | next = map_end; |
132978b9 | 568 | } |
b959ed6c | 569 | |
132978b9 | 570 | mapped_ram_size += init_range_memory_mapping(start, next); |
b959ed6c TC |
571 | start = next; |
572 | ||
132978b9 | 573 | if (mapped_ram_size >= step_size) |
b959ed6c | 574 | step_size = get_new_step_size(step_size); |
b959ed6c TC |
575 | } |
576 | } | |
577 | ||
0167d7d8 TC |
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 | ||
b959ed6c TC |
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 | } | |
8d57470d | 612 | |
f763ad1d YL |
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 | } | |
719272c4 YL |
618 | #else |
619 | early_ioremap_page_table_range_init(); | |
8170e6be PA |
620 | #endif |
621 | ||
719272c4 YL |
622 | load_cr3(swapper_pg_dir); |
623 | __flush_tlb_all(); | |
719272c4 | 624 | |
c14fa0b6 | 625 | early_memtest(0, max_pfn_mapped << PAGE_SHIFT); |
22ddfcaa | 626 | } |
e5b2bb55 | 627 | |
540aca06 PE |
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 | |
801a5591 | 634 | * contains BIOS code and data regions used by X and dosemu and similar apps. |
540aca06 PE |
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 | { | |
73e8f3d7 | 640 | if (pagenr < 256) |
540aca06 PE |
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 | ||
e5b2bb55 PE |
649 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
650 | { | |
c967da6a | 651 | unsigned long begin_aligned, end_aligned; |
e5b2bb55 | 652 | |
c967da6a YL |
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) | |
e5b2bb55 PE |
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 | */ | |
a75e1f63 CB |
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); | |
e5b2bb55 | 682 | |
a75e1f63 CB |
683 | free_reserved_area((void *)begin, (void *)end, |
684 | POISON_FREE_INITMEM, what); | |
685 | } | |
e5b2bb55 PE |
686 | } |
687 | ||
688 | void free_initmem(void) | |
689 | { | |
c88442ec | 690 | free_init_pages("unused kernel", |
e5b2bb55 PE |
691 | (unsigned long)(&__init_begin), |
692 | (unsigned long)(&__init_end)); | |
693 | } | |
731ddea6 PE |
694 | |
695 | #ifdef CONFIG_BLK_DEV_INITRD | |
0d26d1d8 | 696 | void __init free_initrd_mem(unsigned long start, unsigned long end) |
731ddea6 | 697 | { |
cd745be8 FY |
698 | /* |
699 | * Remember, initrd memory may contain microcode or other useful things. | |
700 | * Before we lose initrd mem, we need to find a place to hold them | |
701 | * now that normal virtual memory is enabled. | |
702 | */ | |
703 | save_microcode_in_initrd(); | |
cd745be8 | 704 | |
c967da6a YL |
705 | /* |
706 | * end could be not aligned, and We can not align that, | |
707 | * decompresser could be confused by aligned initrd_end | |
708 | * We already reserve the end partial page before in | |
709 | * - i386_start_kernel() | |
710 | * - x86_64_start_kernel() | |
711 | * - relocate_initrd() | |
712 | * So here We can do PAGE_ALIGN() safely to get partial page to be freed | |
713 | */ | |
c88442ec | 714 | free_init_pages("initrd", start, PAGE_ALIGN(end)); |
731ddea6 PE |
715 | } |
716 | #endif | |
17623915 PE |
717 | |
718 | void __init zone_sizes_init(void) | |
719 | { | |
720 | unsigned long max_zone_pfns[MAX_NR_ZONES]; | |
721 | ||
722 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
723 | ||
724 | #ifdef CONFIG_ZONE_DMA | |
c072b90c | 725 | max_zone_pfns[ZONE_DMA] = min(MAX_DMA_PFN, max_low_pfn); |
17623915 PE |
726 | #endif |
727 | #ifdef CONFIG_ZONE_DMA32 | |
c072b90c | 728 | max_zone_pfns[ZONE_DMA32] = min(MAX_DMA32_PFN, max_low_pfn); |
17623915 PE |
729 | #endif |
730 | max_zone_pfns[ZONE_NORMAL] = max_low_pfn; | |
731 | #ifdef CONFIG_HIGHMEM | |
732 | max_zone_pfns[ZONE_HIGHMEM] = max_pfn; | |
733 | #endif | |
734 | ||
735 | free_area_init_nodes(max_zone_pfns); | |
736 | } | |
737 | ||
1e02ce4c AL |
738 | DEFINE_PER_CPU_SHARED_ALIGNED(struct tlb_state, cpu_tlbstate) = { |
739 | #ifdef CONFIG_SMP | |
740 | .active_mm = &init_mm, | |
741 | .state = 0, | |
742 | #endif | |
743 | .cr4 = ~0UL, /* fail hard if we screw up cr4 shadow initialization */ | |
744 | }; | |
745 | EXPORT_SYMBOL_GPL(cpu_tlbstate); | |
746 | ||
bd809af1 JG |
747 | void update_cache_mode_entry(unsigned entry, enum page_cache_mode cache) |
748 | { | |
749 | /* entry 0 MUST be WB (hardwired to speed up translations) */ | |
750 | BUG_ON(!entry && cache != _PAGE_CACHE_MODE_WB); | |
751 | ||
752 | __cachemode2pte_tbl[cache] = __cm_idx2pte(entry); | |
753 | __pte2cachemode_tbl[entry] = cache; | |
754 | } |