]>
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> |
66441bd3 | 9 | #include <asm/e820/api.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> |
0483e1fa | 20 | #include <asm/kaslr.h> |
9518e0e4 | 21 | |
d17d8f9d DH |
22 | /* |
23 | * We need to define the tracepoints somewhere, and tlb.c | |
24 | * is only compied when SMP=y. | |
25 | */ | |
26 | #define CREATE_TRACE_POINTS | |
27 | #include <trace/events/tlb.h> | |
28 | ||
5c51bdbe YL |
29 | #include "mm_internal.h" |
30 | ||
281d4078 JG |
31 | /* |
32 | * Tables translating between page_cache_type_t and pte encoding. | |
c709feda | 33 | * |
d5dc861b TK |
34 | * The default values are defined statically as minimal supported mode; |
35 | * WC and WT fall back to UC-. pat_init() updates these values to support | |
36 | * more cache modes, WC and WT, when it is safe to do so. See pat_init() | |
37 | * for the details. Note, __early_ioremap() used during early boot-time | |
38 | * takes pgprot_t (pte encoding) and does not use these tables. | |
c709feda IM |
39 | * |
40 | * Index into __cachemode2pte_tbl[] is the cachemode. | |
41 | * | |
42 | * Index into __pte2cachemode_tbl[] are the caching attribute bits of the pte | |
43 | * (_PAGE_PWT, _PAGE_PCD, _PAGE_PAT) at index bit positions 0, 1, 2. | |
281d4078 JG |
44 | */ |
45 | uint16_t __cachemode2pte_tbl[_PAGE_CACHE_MODE_NUM] = { | |
c709feda | 46 | [_PAGE_CACHE_MODE_WB ] = 0 | 0 , |
9cd25aac | 47 | [_PAGE_CACHE_MODE_WC ] = 0 | _PAGE_PCD, |
c709feda IM |
48 | [_PAGE_CACHE_MODE_UC_MINUS] = 0 | _PAGE_PCD, |
49 | [_PAGE_CACHE_MODE_UC ] = _PAGE_PWT | _PAGE_PCD, | |
50 | [_PAGE_CACHE_MODE_WT ] = 0 | _PAGE_PCD, | |
51 | [_PAGE_CACHE_MODE_WP ] = 0 | _PAGE_PCD, | |
281d4078 | 52 | }; |
31bb7723 | 53 | EXPORT_SYMBOL(__cachemode2pte_tbl); |
c709feda | 54 | |
281d4078 | 55 | uint8_t __pte2cachemode_tbl[8] = { |
c709feda | 56 | [__pte2cm_idx( 0 | 0 | 0 )] = _PAGE_CACHE_MODE_WB, |
9cd25aac | 57 | [__pte2cm_idx(_PAGE_PWT | 0 | 0 )] = _PAGE_CACHE_MODE_UC_MINUS, |
c709feda IM |
58 | [__pte2cm_idx( 0 | _PAGE_PCD | 0 )] = _PAGE_CACHE_MODE_UC_MINUS, |
59 | [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | 0 )] = _PAGE_CACHE_MODE_UC, | |
60 | [__pte2cm_idx( 0 | 0 | _PAGE_PAT)] = _PAGE_CACHE_MODE_WB, | |
9cd25aac | 61 | [__pte2cm_idx(_PAGE_PWT | 0 | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS, |
c709feda | 62 | [__pte2cm_idx(0 | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS, |
281d4078 JG |
63 | [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC, |
64 | }; | |
31bb7723 | 65 | EXPORT_SYMBOL(__pte2cachemode_tbl); |
281d4078 | 66 | |
cf470659 YL |
67 | static unsigned long __initdata pgt_buf_start; |
68 | static unsigned long __initdata pgt_buf_end; | |
69 | static unsigned long __initdata pgt_buf_top; | |
f765090a | 70 | |
9985b4c6 YL |
71 | static unsigned long min_pfn_mapped; |
72 | ||
c9b3234a YL |
73 | static bool __initdata can_use_brk_pgt = true; |
74 | ||
ddd3509d SS |
75 | /* |
76 | * Pages returned are already directly mapped. | |
77 | * | |
78 | * Changing that is likely to break Xen, see commit: | |
79 | * | |
80 | * 279b706 x86,xen: introduce x86_init.mapping.pagetable_reserve | |
81 | * | |
82 | * for detailed information. | |
83 | */ | |
22c8ca2a | 84 | __ref void *alloc_low_pages(unsigned int num) |
5c51bdbe YL |
85 | { |
86 | unsigned long pfn; | |
22c8ca2a | 87 | int i; |
5c51bdbe | 88 | |
5c51bdbe | 89 | if (after_bootmem) { |
22c8ca2a | 90 | unsigned int order; |
5c51bdbe | 91 | |
22c8ca2a YL |
92 | order = get_order((unsigned long)num << PAGE_SHIFT); |
93 | return (void *)__get_free_pages(GFP_ATOMIC | __GFP_NOTRACK | | |
94 | __GFP_ZERO, order); | |
5c51bdbe | 95 | } |
5c51bdbe | 96 | |
c9b3234a | 97 | if ((pgt_buf_end + num) > pgt_buf_top || !can_use_brk_pgt) { |
5c51bdbe YL |
98 | unsigned long ret; |
99 | if (min_pfn_mapped >= max_pfn_mapped) | |
d4dd100f | 100 | panic("alloc_low_pages: ran out of memory"); |
5c51bdbe YL |
101 | ret = memblock_find_in_range(min_pfn_mapped << PAGE_SHIFT, |
102 | max_pfn_mapped << PAGE_SHIFT, | |
22c8ca2a | 103 | PAGE_SIZE * num , PAGE_SIZE); |
5c51bdbe | 104 | if (!ret) |
d4dd100f | 105 | panic("alloc_low_pages: can not alloc memory"); |
22c8ca2a | 106 | memblock_reserve(ret, PAGE_SIZE * num); |
5c51bdbe | 107 | pfn = ret >> PAGE_SHIFT; |
22c8ca2a YL |
108 | } else { |
109 | pfn = pgt_buf_end; | |
110 | pgt_buf_end += num; | |
c9b3234a YL |
111 | printk(KERN_DEBUG "BRK [%#010lx, %#010lx] PGTABLE\n", |
112 | pfn << PAGE_SHIFT, (pgt_buf_end << PAGE_SHIFT) - 1); | |
22c8ca2a YL |
113 | } |
114 | ||
115 | for (i = 0; i < num; i++) { | |
116 | void *adr; | |
117 | ||
118 | adr = __va((pfn + i) << PAGE_SHIFT); | |
119 | clear_page(adr); | |
120 | } | |
5c51bdbe | 121 | |
22c8ca2a | 122 | return __va(pfn << PAGE_SHIFT); |
5c51bdbe YL |
123 | } |
124 | ||
fb754f95 TG |
125 | /* |
126 | * By default need 3 4k for initial PMD_SIZE, 3 4k for 0-ISA_END_ADDRESS. | |
127 | * With KASLR memory randomization, depending on the machine e820 memory | |
128 | * and the PUD alignment. We may need twice more pages when KASLR memory | |
129 | * randomization is enabled. | |
130 | */ | |
131 | #ifndef CONFIG_RANDOMIZE_MEMORY | |
132 | #define INIT_PGD_PAGE_COUNT 6 | |
133 | #else | |
134 | #define INIT_PGD_PAGE_COUNT 12 | |
135 | #endif | |
136 | #define INIT_PGT_BUF_SIZE (INIT_PGD_PAGE_COUNT * PAGE_SIZE) | |
8d57470d YL |
137 | RESERVE_BRK(early_pgt_alloc, INIT_PGT_BUF_SIZE); |
138 | void __init early_alloc_pgt_buf(void) | |
139 | { | |
140 | unsigned long tables = INIT_PGT_BUF_SIZE; | |
141 | phys_addr_t base; | |
142 | ||
143 | base = __pa(extend_brk(tables, PAGE_SIZE)); | |
144 | ||
145 | pgt_buf_start = base >> PAGE_SHIFT; | |
146 | pgt_buf_end = pgt_buf_start; | |
147 | pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT); | |
148 | } | |
149 | ||
f765090a PE |
150 | int after_bootmem; |
151 | ||
10971ab2 | 152 | early_param_on_off("gbpages", "nogbpages", direct_gbpages, CONFIG_X86_DIRECT_GBPAGES); |
148b2098 | 153 | |
844ab6f9 JS |
154 | struct map_range { |
155 | unsigned long start; | |
156 | unsigned long end; | |
157 | unsigned page_size_mask; | |
158 | }; | |
159 | ||
fa62aafe | 160 | static int page_size_mask; |
f765090a | 161 | |
22ddfcaa | 162 | static void __init probe_page_size_mask(void) |
fa62aafe | 163 | { |
288cf3c6 | 164 | #if !defined(CONFIG_KMEMCHECK) |
fa62aafe | 165 | /* |
288cf3c6 CB |
166 | * For CONFIG_KMEMCHECK or pagealloc debugging, identity mapping will |
167 | * use small pages. | |
fa62aafe YL |
168 | * This will simplify cpa(), which otherwise needs to support splitting |
169 | * large pages into small in interrupt context, etc. | |
170 | */ | |
16bf9226 | 171 | if (boot_cpu_has(X86_FEATURE_PSE) && !debug_pagealloc_enabled()) |
fa62aafe YL |
172 | page_size_mask |= 1 << PG_LEVEL_2M; |
173 | #endif | |
174 | ||
175 | /* Enable PSE if available */ | |
16bf9226 | 176 | if (boot_cpu_has(X86_FEATURE_PSE)) |
375074cc | 177 | cr4_set_bits_and_update_boot(X86_CR4_PSE); |
fa62aafe YL |
178 | |
179 | /* Enable PGE if available */ | |
c109bf95 | 180 | if (boot_cpu_has(X86_FEATURE_PGE)) { |
375074cc | 181 | cr4_set_bits_and_update_boot(X86_CR4_PGE); |
fa62aafe | 182 | __supported_pte_mask |= _PAGE_GLOBAL; |
0cdb81be JB |
183 | } else |
184 | __supported_pte_mask &= ~_PAGE_GLOBAL; | |
e61980a7 IM |
185 | |
186 | /* Enable 1 GB linear kernel mappings if available: */ | |
b8291adc | 187 | if (direct_gbpages && boot_cpu_has(X86_FEATURE_GBPAGES)) { |
e61980a7 IM |
188 | printk(KERN_INFO "Using GB pages for direct mapping\n"); |
189 | page_size_mask |= 1 << PG_LEVEL_1G; | |
190 | } else { | |
191 | direct_gbpages = 0; | |
192 | } | |
fa62aafe | 193 | } |
279b706b | 194 | |
f765090a PE |
195 | #ifdef CONFIG_X86_32 |
196 | #define NR_RANGE_MR 3 | |
197 | #else /* CONFIG_X86_64 */ | |
198 | #define NR_RANGE_MR 5 | |
199 | #endif | |
200 | ||
dc9dd5cc JB |
201 | static int __meminit save_mr(struct map_range *mr, int nr_range, |
202 | unsigned long start_pfn, unsigned long end_pfn, | |
203 | unsigned long page_size_mask) | |
f765090a PE |
204 | { |
205 | if (start_pfn < end_pfn) { | |
206 | if (nr_range >= NR_RANGE_MR) | |
207 | panic("run out of range for init_memory_mapping\n"); | |
208 | mr[nr_range].start = start_pfn<<PAGE_SHIFT; | |
209 | mr[nr_range].end = end_pfn<<PAGE_SHIFT; | |
210 | mr[nr_range].page_size_mask = page_size_mask; | |
211 | nr_range++; | |
212 | } | |
213 | ||
214 | return nr_range; | |
215 | } | |
216 | ||
aeebe84c YL |
217 | /* |
218 | * adjust the page_size_mask for small range to go with | |
219 | * big page size instead small one if nearby are ram too. | |
220 | */ | |
bd721ea7 | 221 | static void __ref adjust_range_page_size_mask(struct map_range *mr, |
aeebe84c YL |
222 | int nr_range) |
223 | { | |
224 | int i; | |
225 | ||
226 | for (i = 0; i < nr_range; i++) { | |
227 | if ((page_size_mask & (1<<PG_LEVEL_2M)) && | |
228 | !(mr[i].page_size_mask & (1<<PG_LEVEL_2M))) { | |
229 | unsigned long start = round_down(mr[i].start, PMD_SIZE); | |
230 | unsigned long end = round_up(mr[i].end, PMD_SIZE); | |
231 | ||
232 | #ifdef CONFIG_X86_32 | |
233 | if ((end >> PAGE_SHIFT) > max_low_pfn) | |
234 | continue; | |
235 | #endif | |
236 | ||
237 | if (memblock_is_region_memory(start, end - start)) | |
238 | mr[i].page_size_mask |= 1<<PG_LEVEL_2M; | |
239 | } | |
240 | if ((page_size_mask & (1<<PG_LEVEL_1G)) && | |
241 | !(mr[i].page_size_mask & (1<<PG_LEVEL_1G))) { | |
242 | unsigned long start = round_down(mr[i].start, PUD_SIZE); | |
243 | unsigned long end = round_up(mr[i].end, PUD_SIZE); | |
244 | ||
245 | if (memblock_is_region_memory(start, end - start)) | |
246 | mr[i].page_size_mask |= 1<<PG_LEVEL_1G; | |
247 | } | |
248 | } | |
249 | } | |
250 | ||
f15e0518 DH |
251 | static const char *page_size_string(struct map_range *mr) |
252 | { | |
253 | static const char str_1g[] = "1G"; | |
254 | static const char str_2m[] = "2M"; | |
255 | static const char str_4m[] = "4M"; | |
256 | static const char str_4k[] = "4k"; | |
257 | ||
258 | if (mr->page_size_mask & (1<<PG_LEVEL_1G)) | |
259 | return str_1g; | |
260 | /* | |
261 | * 32-bit without PAE has a 4M large page size. | |
262 | * PG_LEVEL_2M is misnamed, but we can at least | |
263 | * print out the right size in the string. | |
264 | */ | |
265 | if (IS_ENABLED(CONFIG_X86_32) && | |
266 | !IS_ENABLED(CONFIG_X86_PAE) && | |
267 | mr->page_size_mask & (1<<PG_LEVEL_2M)) | |
268 | return str_4m; | |
269 | ||
270 | if (mr->page_size_mask & (1<<PG_LEVEL_2M)) | |
271 | return str_2m; | |
272 | ||
273 | return str_4k; | |
274 | } | |
275 | ||
4e33e065 YL |
276 | static int __meminit split_mem_range(struct map_range *mr, int nr_range, |
277 | unsigned long start, | |
278 | unsigned long end) | |
f765090a | 279 | { |
2e8059ed | 280 | unsigned long start_pfn, end_pfn, limit_pfn; |
1829ae9a | 281 | unsigned long pfn; |
4e33e065 | 282 | int i; |
f765090a | 283 | |
2e8059ed YL |
284 | limit_pfn = PFN_DOWN(end); |
285 | ||
f765090a | 286 | /* head if not big page alignment ? */ |
1829ae9a | 287 | pfn = start_pfn = PFN_DOWN(start); |
f765090a PE |
288 | #ifdef CONFIG_X86_32 |
289 | /* | |
290 | * Don't use a large page for the first 2/4MB of memory | |
291 | * because there are often fixed size MTRRs in there | |
292 | * and overlapping MTRRs into large pages can cause | |
293 | * slowdowns. | |
294 | */ | |
1829ae9a | 295 | if (pfn == 0) |
84d77001 | 296 | end_pfn = PFN_DOWN(PMD_SIZE); |
f765090a | 297 | else |
1829ae9a | 298 | end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 299 | #else /* CONFIG_X86_64 */ |
1829ae9a | 300 | end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 301 | #endif |
2e8059ed YL |
302 | if (end_pfn > limit_pfn) |
303 | end_pfn = limit_pfn; | |
f765090a PE |
304 | if (start_pfn < end_pfn) { |
305 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); | |
1829ae9a | 306 | pfn = end_pfn; |
f765090a PE |
307 | } |
308 | ||
309 | /* big page (2M) range */ | |
1829ae9a | 310 | start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 311 | #ifdef CONFIG_X86_32 |
2e8059ed | 312 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 313 | #else /* CONFIG_X86_64 */ |
1829ae9a | 314 | end_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE)); |
2e8059ed YL |
315 | if (end_pfn > round_down(limit_pfn, PFN_DOWN(PMD_SIZE))) |
316 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); | |
f765090a PE |
317 | #endif |
318 | ||
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)); | |
1829ae9a | 322 | pfn = end_pfn; |
f765090a PE |
323 | } |
324 | ||
325 | #ifdef CONFIG_X86_64 | |
326 | /* big page (1G) range */ | |
1829ae9a | 327 | start_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE)); |
2e8059ed | 328 | end_pfn = round_down(limit_pfn, PFN_DOWN(PUD_SIZE)); |
f765090a PE |
329 | if (start_pfn < end_pfn) { |
330 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, | |
331 | page_size_mask & | |
332 | ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G))); | |
1829ae9a | 333 | pfn = end_pfn; |
f765090a PE |
334 | } |
335 | ||
336 | /* tail is not big page (1G) alignment */ | |
1829ae9a | 337 | start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
2e8059ed | 338 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); |
f765090a PE |
339 | if (start_pfn < end_pfn) { |
340 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, | |
341 | page_size_mask & (1<<PG_LEVEL_2M)); | |
1829ae9a | 342 | pfn = end_pfn; |
f765090a PE |
343 | } |
344 | #endif | |
345 | ||
346 | /* tail is not big page (2M) alignment */ | |
1829ae9a | 347 | start_pfn = pfn; |
2e8059ed | 348 | end_pfn = limit_pfn; |
f765090a PE |
349 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); |
350 | ||
7de3d66b YL |
351 | if (!after_bootmem) |
352 | adjust_range_page_size_mask(mr, nr_range); | |
353 | ||
f765090a PE |
354 | /* try to merge same page size and continuous */ |
355 | for (i = 0; nr_range > 1 && i < nr_range - 1; i++) { | |
356 | unsigned long old_start; | |
357 | if (mr[i].end != mr[i+1].start || | |
358 | mr[i].page_size_mask != mr[i+1].page_size_mask) | |
359 | continue; | |
360 | /* move it */ | |
361 | old_start = mr[i].start; | |
362 | memmove(&mr[i], &mr[i+1], | |
363 | (nr_range - 1 - i) * sizeof(struct map_range)); | |
364 | mr[i--].start = old_start; | |
365 | nr_range--; | |
366 | } | |
367 | ||
368 | for (i = 0; i < nr_range; i++) | |
c9cdaeb2 | 369 | pr_debug(" [mem %#010lx-%#010lx] page %s\n", |
365811d6 | 370 | mr[i].start, mr[i].end - 1, |
f15e0518 | 371 | page_size_string(&mr[i])); |
f765090a | 372 | |
4e33e065 YL |
373 | return nr_range; |
374 | } | |
375 | ||
08b46d5d | 376 | struct range pfn_mapped[E820_MAX_ENTRIES]; |
0e691cf8 | 377 | int nr_pfn_mapped; |
66520ebc JS |
378 | |
379 | static void add_pfn_range_mapped(unsigned long start_pfn, unsigned long end_pfn) | |
380 | { | |
08b46d5d | 381 | nr_pfn_mapped = add_range_with_merge(pfn_mapped, E820_MAX_ENTRIES, |
66520ebc | 382 | nr_pfn_mapped, start_pfn, end_pfn); |
08b46d5d | 383 | nr_pfn_mapped = clean_sort_range(pfn_mapped, E820_MAX_ENTRIES); |
66520ebc JS |
384 | |
385 | max_pfn_mapped = max(max_pfn_mapped, end_pfn); | |
386 | ||
387 | if (start_pfn < (1UL<<(32-PAGE_SHIFT))) | |
388 | max_low_pfn_mapped = max(max_low_pfn_mapped, | |
389 | min(end_pfn, 1UL<<(32-PAGE_SHIFT))); | |
390 | } | |
391 | ||
392 | bool pfn_range_is_mapped(unsigned long start_pfn, unsigned long end_pfn) | |
393 | { | |
394 | int i; | |
395 | ||
396 | for (i = 0; i < nr_pfn_mapped; i++) | |
397 | if ((start_pfn >= pfn_mapped[i].start) && | |
398 | (end_pfn <= pfn_mapped[i].end)) | |
399 | return true; | |
400 | ||
401 | return false; | |
402 | } | |
403 | ||
4e33e065 YL |
404 | /* |
405 | * Setup the direct mapping of the physical memory at PAGE_OFFSET. | |
406 | * This runs before bootmem is initialized and gets pages directly from | |
407 | * the physical memory. To access them they are temporarily mapped. | |
408 | */ | |
bd721ea7 | 409 | unsigned long __ref init_memory_mapping(unsigned long start, |
4e33e065 YL |
410 | unsigned long end) |
411 | { | |
412 | struct map_range mr[NR_RANGE_MR]; | |
413 | unsigned long ret = 0; | |
414 | int nr_range, i; | |
415 | ||
c9cdaeb2 | 416 | pr_debug("init_memory_mapping: [mem %#010lx-%#010lx]\n", |
4e33e065 YL |
417 | start, end - 1); |
418 | ||
419 | memset(mr, 0, sizeof(mr)); | |
420 | nr_range = split_mem_range(mr, 0, start, end); | |
421 | ||
f765090a PE |
422 | for (i = 0; i < nr_range; i++) |
423 | ret = kernel_physical_mapping_init(mr[i].start, mr[i].end, | |
424 | mr[i].page_size_mask); | |
f765090a | 425 | |
66520ebc JS |
426 | add_pfn_range_mapped(start >> PAGE_SHIFT, ret >> PAGE_SHIFT); |
427 | ||
c14fa0b6 YL |
428 | return ret >> PAGE_SHIFT; |
429 | } | |
430 | ||
66520ebc | 431 | /* |
cf8b166d | 432 | * We need to iterate through the E820 memory map and create direct mappings |
09821ff1 | 433 | * for only E820_TYPE_RAM and E820_KERN_RESERVED regions. We cannot simply |
cf8b166d ZY |
434 | * create direct mappings for all pfns from [0 to max_low_pfn) and |
435 | * [4GB to max_pfn) because of possible memory holes in high addresses | |
436 | * that cannot be marked as UC by fixed/variable range MTRRs. | |
437 | * Depending on the alignment of E820 ranges, this may possibly result | |
438 | * in using smaller size (i.e. 4K instead of 2M or 1G) page tables. | |
439 | * | |
440 | * init_mem_mapping() calls init_range_memory_mapping() with big range. | |
441 | * That range would have hole in the middle or ends, and only ram parts | |
442 | * will be mapped in init_range_memory_mapping(). | |
66520ebc | 443 | */ |
8d57470d | 444 | static unsigned long __init init_range_memory_mapping( |
b8fd39c0 YL |
445 | unsigned long r_start, |
446 | unsigned long r_end) | |
66520ebc JS |
447 | { |
448 | unsigned long start_pfn, end_pfn; | |
8d57470d | 449 | unsigned long mapped_ram_size = 0; |
66520ebc JS |
450 | int i; |
451 | ||
66520ebc | 452 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) { |
b8fd39c0 YL |
453 | u64 start = clamp_val(PFN_PHYS(start_pfn), r_start, r_end); |
454 | u64 end = clamp_val(PFN_PHYS(end_pfn), r_start, r_end); | |
455 | if (start >= end) | |
66520ebc JS |
456 | continue; |
457 | ||
c9b3234a YL |
458 | /* |
459 | * if it is overlapping with brk pgt, we need to | |
460 | * alloc pgt buf from memblock instead. | |
461 | */ | |
462 | can_use_brk_pgt = max(start, (u64)pgt_buf_end<<PAGE_SHIFT) >= | |
463 | min(end, (u64)pgt_buf_top<<PAGE_SHIFT); | |
f763ad1d | 464 | init_memory_mapping(start, end); |
8d57470d | 465 | mapped_ram_size += end - start; |
c9b3234a | 466 | can_use_brk_pgt = true; |
66520ebc | 467 | } |
8d57470d YL |
468 | |
469 | return mapped_ram_size; | |
66520ebc JS |
470 | } |
471 | ||
6979287a YL |
472 | static unsigned long __init get_new_step_size(unsigned long step_size) |
473 | { | |
474 | /* | |
132978b9 | 475 | * Initial mapped size is PMD_SIZE (2M). |
6979287a YL |
476 | * We can not set step_size to be PUD_SIZE (1G) yet. |
477 | * In worse case, when we cross the 1G boundary, and | |
478 | * PG_LEVEL_2M is not set, we will need 1+1+512 pages (2M + 8k) | |
132978b9 JB |
479 | * to map 1G range with PTE. Hence we use one less than the |
480 | * difference of page table level shifts. | |
6979287a | 481 | * |
132978b9 JB |
482 | * Don't need to worry about overflow in the top-down case, on 32bit, |
483 | * when step_size is 0, round_down() returns 0 for start, and that | |
484 | * turns it into 0x100000000ULL. | |
485 | * In the bottom-up case, round_up(x, 0) returns 0 though too, which | |
486 | * needs to be taken into consideration by the code below. | |
6979287a | 487 | */ |
132978b9 | 488 | return step_size << (PMD_SHIFT - PAGE_SHIFT - 1); |
6979287a YL |
489 | } |
490 | ||
0167d7d8 TC |
491 | /** |
492 | * memory_map_top_down - Map [map_start, map_end) top down | |
493 | * @map_start: start address of the target memory range | |
494 | * @map_end: end address of the target memory range | |
495 | * | |
496 | * This function will setup direct mapping for memory range | |
497 | * [map_start, map_end) in top-down. That said, the page tables | |
498 | * will be allocated at the end of the memory, and we map the | |
499 | * memory in top-down. | |
500 | */ | |
501 | static void __init memory_map_top_down(unsigned long map_start, | |
502 | unsigned long map_end) | |
c14fa0b6 | 503 | { |
0167d7d8 | 504 | unsigned long real_end, start, last_start; |
8d57470d YL |
505 | unsigned long step_size; |
506 | unsigned long addr; | |
507 | unsigned long mapped_ram_size = 0; | |
ab951937 | 508 | |
98e7a989 | 509 | /* xen has big range in reserved near end of ram, skip it at first.*/ |
0167d7d8 | 510 | addr = memblock_find_in_range(map_start, map_end, PMD_SIZE, PMD_SIZE); |
8d57470d YL |
511 | real_end = addr + PMD_SIZE; |
512 | ||
513 | /* step_size need to be small so pgt_buf from BRK could cover it */ | |
514 | step_size = PMD_SIZE; | |
515 | max_pfn_mapped = 0; /* will get exact value next */ | |
516 | min_pfn_mapped = real_end >> PAGE_SHIFT; | |
517 | last_start = start = real_end; | |
cf8b166d ZY |
518 | |
519 | /* | |
520 | * We start from the top (end of memory) and go to the bottom. | |
521 | * The memblock_find_in_range() gets us a block of RAM from the | |
522 | * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages | |
523 | * for page table. | |
524 | */ | |
0167d7d8 | 525 | while (last_start > map_start) { |
8d57470d YL |
526 | if (last_start > step_size) { |
527 | start = round_down(last_start - 1, step_size); | |
0167d7d8 TC |
528 | if (start < map_start) |
529 | start = map_start; | |
8d57470d | 530 | } else |
0167d7d8 | 531 | start = map_start; |
132978b9 | 532 | mapped_ram_size += init_range_memory_mapping(start, |
8d57470d YL |
533 | last_start); |
534 | last_start = start; | |
535 | min_pfn_mapped = last_start >> PAGE_SHIFT; | |
132978b9 | 536 | if (mapped_ram_size >= step_size) |
6979287a | 537 | step_size = get_new_step_size(step_size); |
8d57470d YL |
538 | } |
539 | ||
0167d7d8 TC |
540 | if (real_end < map_end) |
541 | init_range_memory_mapping(real_end, map_end); | |
542 | } | |
543 | ||
b959ed6c TC |
544 | /** |
545 | * memory_map_bottom_up - Map [map_start, map_end) bottom up | |
546 | * @map_start: start address of the target memory range | |
547 | * @map_end: end address of the target memory range | |
548 | * | |
549 | * This function will setup direct mapping for memory range | |
550 | * [map_start, map_end) in bottom-up. Since we have limited the | |
551 | * bottom-up allocation above the kernel, the page tables will | |
552 | * be allocated just above the kernel and we map the memory | |
553 | * in [map_start, map_end) in bottom-up. | |
554 | */ | |
555 | static void __init memory_map_bottom_up(unsigned long map_start, | |
556 | unsigned long map_end) | |
557 | { | |
132978b9 | 558 | unsigned long next, start; |
b959ed6c TC |
559 | unsigned long mapped_ram_size = 0; |
560 | /* step_size need to be small so pgt_buf from BRK could cover it */ | |
561 | unsigned long step_size = PMD_SIZE; | |
562 | ||
563 | start = map_start; | |
564 | min_pfn_mapped = start >> PAGE_SHIFT; | |
565 | ||
566 | /* | |
567 | * We start from the bottom (@map_start) and go to the top (@map_end). | |
568 | * The memblock_find_in_range() gets us a block of RAM from the | |
569 | * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages | |
570 | * for page table. | |
571 | */ | |
572 | while (start < map_end) { | |
132978b9 | 573 | if (step_size && map_end - start > step_size) { |
b959ed6c TC |
574 | next = round_up(start + 1, step_size); |
575 | if (next > map_end) | |
576 | next = map_end; | |
132978b9 | 577 | } else { |
b959ed6c | 578 | next = map_end; |
132978b9 | 579 | } |
b959ed6c | 580 | |
132978b9 | 581 | mapped_ram_size += init_range_memory_mapping(start, next); |
b959ed6c TC |
582 | start = next; |
583 | ||
132978b9 | 584 | if (mapped_ram_size >= step_size) |
b959ed6c | 585 | step_size = get_new_step_size(step_size); |
b959ed6c TC |
586 | } |
587 | } | |
588 | ||
0167d7d8 TC |
589 | void __init init_mem_mapping(void) |
590 | { | |
591 | unsigned long end; | |
592 | ||
593 | probe_page_size_mask(); | |
594 | ||
595 | #ifdef CONFIG_X86_64 | |
596 | end = max_pfn << PAGE_SHIFT; | |
597 | #else | |
598 | end = max_low_pfn << PAGE_SHIFT; | |
599 | #endif | |
600 | ||
601 | /* the ISA range is always mapped regardless of memory holes */ | |
602 | init_memory_mapping(0, ISA_END_ADDRESS); | |
603 | ||
b234e8a0 TG |
604 | /* Init the trampoline, possibly with KASLR memory offset */ |
605 | init_trampoline(); | |
606 | ||
b959ed6c TC |
607 | /* |
608 | * If the allocation is in bottom-up direction, we setup direct mapping | |
609 | * in bottom-up, otherwise we setup direct mapping in top-down. | |
610 | */ | |
611 | if (memblock_bottom_up()) { | |
612 | unsigned long kernel_end = __pa_symbol(_end); | |
613 | ||
614 | /* | |
615 | * we need two separate calls here. This is because we want to | |
616 | * allocate page tables above the kernel. So we first map | |
617 | * [kernel_end, end) to make memory above the kernel be mapped | |
618 | * as soon as possible. And then use page tables allocated above | |
619 | * the kernel to map [ISA_END_ADDRESS, kernel_end). | |
620 | */ | |
621 | memory_map_bottom_up(kernel_end, end); | |
622 | memory_map_bottom_up(ISA_END_ADDRESS, kernel_end); | |
623 | } else { | |
624 | memory_map_top_down(ISA_END_ADDRESS, end); | |
625 | } | |
8d57470d | 626 | |
f763ad1d YL |
627 | #ifdef CONFIG_X86_64 |
628 | if (max_pfn > max_low_pfn) { | |
629 | /* can we preseve max_low_pfn ?*/ | |
630 | max_low_pfn = max_pfn; | |
631 | } | |
719272c4 YL |
632 | #else |
633 | early_ioremap_page_table_range_init(); | |
8170e6be PA |
634 | #endif |
635 | ||
719272c4 YL |
636 | load_cr3(swapper_pg_dir); |
637 | __flush_tlb_all(); | |
719272c4 | 638 | |
c14fa0b6 | 639 | early_memtest(0, max_pfn_mapped << PAGE_SHIFT); |
22ddfcaa | 640 | } |
e5b2bb55 | 641 | |
540aca06 PE |
642 | /* |
643 | * devmem_is_allowed() checks to see if /dev/mem access to a certain address | |
644 | * is valid. The argument is a physical page number. | |
645 | * | |
a4866aa8 KC |
646 | * On x86, access has to be given to the first megabyte of RAM because that |
647 | * area traditionally contains BIOS code and data regions used by X, dosemu, | |
648 | * and similar apps. Since they map the entire memory range, the whole range | |
649 | * must be allowed (for mapping), but any areas that would otherwise be | |
650 | * disallowed are flagged as being "zero filled" instead of rejected. | |
651 | * Access has to be given to non-kernel-ram areas as well, these contain the | |
652 | * PCI mmio resources as well as potential bios/acpi data regions. | |
540aca06 PE |
653 | */ |
654 | int devmem_is_allowed(unsigned long pagenr) | |
655 | { | |
a4866aa8 KC |
656 | if (page_is_ram(pagenr)) { |
657 | /* | |
658 | * For disallowed memory regions in the low 1MB range, | |
659 | * request that the page be shown as all zeros. | |
660 | */ | |
661 | if (pagenr < 256) | |
662 | return 2; | |
663 | ||
664 | return 0; | |
665 | } | |
666 | ||
667 | /* | |
668 | * This must follow RAM test, since System RAM is considered a | |
669 | * restricted resource under CONFIG_STRICT_IOMEM. | |
670 | */ | |
671 | if (iomem_is_exclusive(pagenr << PAGE_SHIFT)) { | |
672 | /* Low 1MB bypasses iomem restrictions. */ | |
673 | if (pagenr < 256) | |
674 | return 1; | |
675 | ||
540aca06 | 676 | return 0; |
a4866aa8 KC |
677 | } |
678 | ||
679 | return 1; | |
540aca06 PE |
680 | } |
681 | ||
e5b2bb55 PE |
682 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
683 | { | |
c967da6a | 684 | unsigned long begin_aligned, end_aligned; |
e5b2bb55 | 685 | |
c967da6a YL |
686 | /* Make sure boundaries are page aligned */ |
687 | begin_aligned = PAGE_ALIGN(begin); | |
688 | end_aligned = end & PAGE_MASK; | |
689 | ||
690 | if (WARN_ON(begin_aligned != begin || end_aligned != end)) { | |
691 | begin = begin_aligned; | |
692 | end = end_aligned; | |
693 | } | |
694 | ||
695 | if (begin >= end) | |
e5b2bb55 PE |
696 | return; |
697 | ||
698 | /* | |
699 | * If debugging page accesses then do not free this memory but | |
700 | * mark them not present - any buggy init-section access will | |
701 | * create a kernel page fault: | |
702 | */ | |
a75e1f63 CB |
703 | if (debug_pagealloc_enabled()) { |
704 | pr_info("debug: unmapping init [mem %#010lx-%#010lx]\n", | |
705 | begin, end - 1); | |
706 | set_memory_np(begin, (end - begin) >> PAGE_SHIFT); | |
707 | } else { | |
708 | /* | |
709 | * We just marked the kernel text read only above, now that | |
710 | * we are going to free part of that, we need to make that | |
711 | * writeable and non-executable first. | |
712 | */ | |
713 | set_memory_nx(begin, (end - begin) >> PAGE_SHIFT); | |
714 | set_memory_rw(begin, (end - begin) >> PAGE_SHIFT); | |
e5b2bb55 | 715 | |
a75e1f63 CB |
716 | free_reserved_area((void *)begin, (void *)end, |
717 | POISON_FREE_INITMEM, what); | |
718 | } | |
e5b2bb55 PE |
719 | } |
720 | ||
18278229 | 721 | void __ref free_initmem(void) |
e5b2bb55 | 722 | { |
0c6fc11a | 723 | e820__reallocate_tables(); |
47533968 | 724 | |
c88442ec | 725 | free_init_pages("unused kernel", |
e5b2bb55 PE |
726 | (unsigned long)(&__init_begin), |
727 | (unsigned long)(&__init_end)); | |
728 | } | |
731ddea6 PE |
729 | |
730 | #ifdef CONFIG_BLK_DEV_INITRD | |
0d26d1d8 | 731 | void __init free_initrd_mem(unsigned long start, unsigned long end) |
731ddea6 | 732 | { |
c967da6a YL |
733 | /* |
734 | * end could be not aligned, and We can not align that, | |
735 | * decompresser could be confused by aligned initrd_end | |
736 | * We already reserve the end partial page before in | |
737 | * - i386_start_kernel() | |
738 | * - x86_64_start_kernel() | |
739 | * - relocate_initrd() | |
740 | * So here We can do PAGE_ALIGN() safely to get partial page to be freed | |
741 | */ | |
c88442ec | 742 | free_init_pages("initrd", start, PAGE_ALIGN(end)); |
731ddea6 PE |
743 | } |
744 | #endif | |
17623915 | 745 | |
4270fd8b IM |
746 | /* |
747 | * Calculate the precise size of the DMA zone (first 16 MB of RAM), | |
748 | * and pass it to the MM layer - to help it set zone watermarks more | |
749 | * accurately. | |
750 | * | |
751 | * Done on 64-bit systems only for the time being, although 32-bit systems | |
752 | * might benefit from this as well. | |
753 | */ | |
754 | void __init memblock_find_dma_reserve(void) | |
755 | { | |
756 | #ifdef CONFIG_X86_64 | |
757 | u64 nr_pages = 0, nr_free_pages = 0; | |
758 | unsigned long start_pfn, end_pfn; | |
759 | phys_addr_t start_addr, end_addr; | |
760 | int i; | |
761 | u64 u; | |
762 | ||
763 | /* | |
764 | * Iterate over all memory ranges (free and reserved ones alike), | |
765 | * to calculate the total number of pages in the first 16 MB of RAM: | |
766 | */ | |
767 | nr_pages = 0; | |
768 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) { | |
769 | start_pfn = min(start_pfn, MAX_DMA_PFN); | |
770 | end_pfn = min(end_pfn, MAX_DMA_PFN); | |
771 | ||
772 | nr_pages += end_pfn - start_pfn; | |
773 | } | |
774 | ||
775 | /* | |
776 | * Iterate over free memory ranges to calculate the number of free | |
777 | * pages in the DMA zone, while not counting potential partial | |
778 | * pages at the beginning or the end of the range: | |
779 | */ | |
780 | nr_free_pages = 0; | |
781 | for_each_free_mem_range(u, NUMA_NO_NODE, MEMBLOCK_NONE, &start_addr, &end_addr, NULL) { | |
782 | start_pfn = min_t(unsigned long, PFN_UP(start_addr), MAX_DMA_PFN); | |
783 | end_pfn = min_t(unsigned long, PFN_DOWN(end_addr), MAX_DMA_PFN); | |
784 | ||
785 | if (start_pfn < end_pfn) | |
786 | nr_free_pages += end_pfn - start_pfn; | |
787 | } | |
788 | ||
789 | set_dma_reserve(nr_pages - nr_free_pages); | |
790 | #endif | |
791 | } | |
792 | ||
17623915 PE |
793 | void __init zone_sizes_init(void) |
794 | { | |
795 | unsigned long max_zone_pfns[MAX_NR_ZONES]; | |
796 | ||
797 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
798 | ||
799 | #ifdef CONFIG_ZONE_DMA | |
c072b90c | 800 | max_zone_pfns[ZONE_DMA] = min(MAX_DMA_PFN, max_low_pfn); |
17623915 PE |
801 | #endif |
802 | #ifdef CONFIG_ZONE_DMA32 | |
c072b90c | 803 | max_zone_pfns[ZONE_DMA32] = min(MAX_DMA32_PFN, max_low_pfn); |
17623915 PE |
804 | #endif |
805 | max_zone_pfns[ZONE_NORMAL] = max_low_pfn; | |
806 | #ifdef CONFIG_HIGHMEM | |
807 | max_zone_pfns[ZONE_HIGHMEM] = max_pfn; | |
808 | #endif | |
809 | ||
810 | free_area_init_nodes(max_zone_pfns); | |
811 | } | |
812 | ||
1e02ce4c AL |
813 | DEFINE_PER_CPU_SHARED_ALIGNED(struct tlb_state, cpu_tlbstate) = { |
814 | #ifdef CONFIG_SMP | |
815 | .active_mm = &init_mm, | |
816 | .state = 0, | |
817 | #endif | |
818 | .cr4 = ~0UL, /* fail hard if we screw up cr4 shadow initialization */ | |
819 | }; | |
820 | EXPORT_SYMBOL_GPL(cpu_tlbstate); | |
821 | ||
bd809af1 JG |
822 | void update_cache_mode_entry(unsigned entry, enum page_cache_mode cache) |
823 | { | |
824 | /* entry 0 MUST be WB (hardwired to speed up translations) */ | |
825 | BUG_ON(!entry && cache != _PAGE_CACHE_MODE_WB); | |
826 | ||
827 | __cachemode2pte_tbl[cache] = __cm_idx2pte(entry); | |
828 | __pte2cachemode_tbl[entry] = cache; | |
829 | } |