]>
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 | |
d1163651 | 8 | #include <asm/set_memory.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> |
c138d811 | 21 | #include <asm/hypervisor.h> |
c7ad5ad2 | 22 | #include <asm/cpufeature.h> |
aa8c6248 | 23 | #include <asm/pti.h> |
9518e0e4 | 24 | |
d17d8f9d DH |
25 | /* |
26 | * We need to define the tracepoints somewhere, and tlb.c | |
27 | * is only compied when SMP=y. | |
28 | */ | |
29 | #define CREATE_TRACE_POINTS | |
30 | #include <trace/events/tlb.h> | |
31 | ||
5c51bdbe YL |
32 | #include "mm_internal.h" |
33 | ||
281d4078 JG |
34 | /* |
35 | * Tables translating between page_cache_type_t and pte encoding. | |
c709feda | 36 | * |
d5dc861b TK |
37 | * The default values are defined statically as minimal supported mode; |
38 | * WC and WT fall back to UC-. pat_init() updates these values to support | |
39 | * more cache modes, WC and WT, when it is safe to do so. See pat_init() | |
40 | * for the details. Note, __early_ioremap() used during early boot-time | |
41 | * takes pgprot_t (pte encoding) and does not use these tables. | |
c709feda IM |
42 | * |
43 | * Index into __cachemode2pte_tbl[] is the cachemode. | |
44 | * | |
45 | * Index into __pte2cachemode_tbl[] are the caching attribute bits of the pte | |
46 | * (_PAGE_PWT, _PAGE_PCD, _PAGE_PAT) at index bit positions 0, 1, 2. | |
281d4078 JG |
47 | */ |
48 | uint16_t __cachemode2pte_tbl[_PAGE_CACHE_MODE_NUM] = { | |
c709feda | 49 | [_PAGE_CACHE_MODE_WB ] = 0 | 0 , |
9cd25aac | 50 | [_PAGE_CACHE_MODE_WC ] = 0 | _PAGE_PCD, |
c709feda IM |
51 | [_PAGE_CACHE_MODE_UC_MINUS] = 0 | _PAGE_PCD, |
52 | [_PAGE_CACHE_MODE_UC ] = _PAGE_PWT | _PAGE_PCD, | |
53 | [_PAGE_CACHE_MODE_WT ] = 0 | _PAGE_PCD, | |
54 | [_PAGE_CACHE_MODE_WP ] = 0 | _PAGE_PCD, | |
281d4078 | 55 | }; |
31bb7723 | 56 | EXPORT_SYMBOL(__cachemode2pte_tbl); |
c709feda | 57 | |
281d4078 | 58 | uint8_t __pte2cachemode_tbl[8] = { |
c709feda | 59 | [__pte2cm_idx( 0 | 0 | 0 )] = _PAGE_CACHE_MODE_WB, |
9cd25aac | 60 | [__pte2cm_idx(_PAGE_PWT | 0 | 0 )] = _PAGE_CACHE_MODE_UC_MINUS, |
c709feda IM |
61 | [__pte2cm_idx( 0 | _PAGE_PCD | 0 )] = _PAGE_CACHE_MODE_UC_MINUS, |
62 | [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | 0 )] = _PAGE_CACHE_MODE_UC, | |
63 | [__pte2cm_idx( 0 | 0 | _PAGE_PAT)] = _PAGE_CACHE_MODE_WB, | |
9cd25aac | 64 | [__pte2cm_idx(_PAGE_PWT | 0 | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS, |
c709feda | 65 | [__pte2cm_idx(0 | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS, |
281d4078 JG |
66 | [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC, |
67 | }; | |
31bb7723 | 68 | EXPORT_SYMBOL(__pte2cachemode_tbl); |
281d4078 | 69 | |
cf470659 YL |
70 | static unsigned long __initdata pgt_buf_start; |
71 | static unsigned long __initdata pgt_buf_end; | |
72 | static unsigned long __initdata pgt_buf_top; | |
f765090a | 73 | |
9985b4c6 YL |
74 | static unsigned long min_pfn_mapped; |
75 | ||
c9b3234a YL |
76 | static bool __initdata can_use_brk_pgt = true; |
77 | ||
ddd3509d SS |
78 | /* |
79 | * Pages returned are already directly mapped. | |
80 | * | |
81 | * Changing that is likely to break Xen, see commit: | |
82 | * | |
83 | * 279b706 x86,xen: introduce x86_init.mapping.pagetable_reserve | |
84 | * | |
85 | * for detailed information. | |
86 | */ | |
22c8ca2a | 87 | __ref void *alloc_low_pages(unsigned int num) |
5c51bdbe YL |
88 | { |
89 | unsigned long pfn; | |
22c8ca2a | 90 | int i; |
5c51bdbe | 91 | |
5c51bdbe | 92 | if (after_bootmem) { |
22c8ca2a | 93 | unsigned int order; |
5c51bdbe | 94 | |
22c8ca2a | 95 | order = get_order((unsigned long)num << PAGE_SHIFT); |
75f296d9 | 96 | return (void *)__get_free_pages(GFP_ATOMIC | __GFP_ZERO, order); |
5c51bdbe | 97 | } |
5c51bdbe | 98 | |
c9b3234a | 99 | if ((pgt_buf_end + num) > pgt_buf_top || !can_use_brk_pgt) { |
5c51bdbe YL |
100 | unsigned long ret; |
101 | if (min_pfn_mapped >= max_pfn_mapped) | |
d4dd100f | 102 | panic("alloc_low_pages: ran out of memory"); |
5c51bdbe YL |
103 | ret = memblock_find_in_range(min_pfn_mapped << PAGE_SHIFT, |
104 | max_pfn_mapped << PAGE_SHIFT, | |
22c8ca2a | 105 | PAGE_SIZE * num , PAGE_SIZE); |
5c51bdbe | 106 | if (!ret) |
d4dd100f | 107 | panic("alloc_low_pages: can not alloc memory"); |
22c8ca2a | 108 | memblock_reserve(ret, PAGE_SIZE * num); |
5c51bdbe | 109 | pfn = ret >> PAGE_SHIFT; |
22c8ca2a YL |
110 | } else { |
111 | pfn = pgt_buf_end; | |
112 | pgt_buf_end += num; | |
c9b3234a YL |
113 | printk(KERN_DEBUG "BRK [%#010lx, %#010lx] PGTABLE\n", |
114 | pfn << PAGE_SHIFT, (pgt_buf_end << PAGE_SHIFT) - 1); | |
22c8ca2a YL |
115 | } |
116 | ||
117 | for (i = 0; i < num; i++) { | |
118 | void *adr; | |
119 | ||
120 | adr = __va((pfn + i) << PAGE_SHIFT); | |
121 | clear_page(adr); | |
122 | } | |
5c51bdbe | 123 | |
22c8ca2a | 124 | return __va(pfn << PAGE_SHIFT); |
5c51bdbe YL |
125 | } |
126 | ||
fb754f95 TG |
127 | /* |
128 | * By default need 3 4k for initial PMD_SIZE, 3 4k for 0-ISA_END_ADDRESS. | |
129 | * With KASLR memory randomization, depending on the machine e820 memory | |
130 | * and the PUD alignment. We may need twice more pages when KASLR memory | |
131 | * randomization is enabled. | |
132 | */ | |
133 | #ifndef CONFIG_RANDOMIZE_MEMORY | |
134 | #define INIT_PGD_PAGE_COUNT 6 | |
135 | #else | |
136 | #define INIT_PGD_PAGE_COUNT 12 | |
137 | #endif | |
138 | #define INIT_PGT_BUF_SIZE (INIT_PGD_PAGE_COUNT * PAGE_SIZE) | |
8d57470d YL |
139 | RESERVE_BRK(early_pgt_alloc, INIT_PGT_BUF_SIZE); |
140 | void __init early_alloc_pgt_buf(void) | |
141 | { | |
142 | unsigned long tables = INIT_PGT_BUF_SIZE; | |
143 | phys_addr_t base; | |
144 | ||
145 | base = __pa(extend_brk(tables, PAGE_SIZE)); | |
146 | ||
147 | pgt_buf_start = base >> PAGE_SHIFT; | |
148 | pgt_buf_end = pgt_buf_start; | |
149 | pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT); | |
150 | } | |
151 | ||
f765090a PE |
152 | int after_bootmem; |
153 | ||
10971ab2 | 154 | early_param_on_off("gbpages", "nogbpages", direct_gbpages, CONFIG_X86_DIRECT_GBPAGES); |
148b2098 | 155 | |
844ab6f9 JS |
156 | struct map_range { |
157 | unsigned long start; | |
158 | unsigned long end; | |
159 | unsigned page_size_mask; | |
160 | }; | |
161 | ||
fa62aafe | 162 | static int page_size_mask; |
f765090a | 163 | |
c313ec66 DH |
164 | static void enable_global_pages(void) |
165 | { | |
166 | if (!static_cpu_has(X86_FEATURE_PTI)) | |
167 | __supported_pte_mask |= _PAGE_GLOBAL; | |
168 | } | |
169 | ||
22ddfcaa | 170 | static void __init probe_page_size_mask(void) |
fa62aafe | 171 | { |
fa62aafe | 172 | /* |
4675ff05 | 173 | * For pagealloc debugging, identity mapping will use small pages. |
fa62aafe YL |
174 | * This will simplify cpa(), which otherwise needs to support splitting |
175 | * large pages into small in interrupt context, etc. | |
176 | */ | |
4675ff05 | 177 | if (boot_cpu_has(X86_FEATURE_PSE) && !debug_pagealloc_enabled()) |
fa62aafe | 178 | page_size_mask |= 1 << PG_LEVEL_2M; |
d9ee35ac VB |
179 | else |
180 | direct_gbpages = 0; | |
fa62aafe YL |
181 | |
182 | /* Enable PSE if available */ | |
16bf9226 | 183 | if (boot_cpu_has(X86_FEATURE_PSE)) |
375074cc | 184 | cr4_set_bits_and_update_boot(X86_CR4_PSE); |
fa62aafe YL |
185 | |
186 | /* Enable PGE if available */ | |
c313ec66 | 187 | __supported_pte_mask &= ~_PAGE_GLOBAL; |
c109bf95 | 188 | if (boot_cpu_has(X86_FEATURE_PGE)) { |
375074cc | 189 | cr4_set_bits_and_update_boot(X86_CR4_PGE); |
c313ec66 DH |
190 | enable_global_pages(); |
191 | } | |
e61980a7 IM |
192 | |
193 | /* Enable 1 GB linear kernel mappings if available: */ | |
b8291adc | 194 | if (direct_gbpages && boot_cpu_has(X86_FEATURE_GBPAGES)) { |
e61980a7 IM |
195 | printk(KERN_INFO "Using GB pages for direct mapping\n"); |
196 | page_size_mask |= 1 << PG_LEVEL_1G; | |
197 | } else { | |
198 | direct_gbpages = 0; | |
199 | } | |
fa62aafe | 200 | } |
279b706b | 201 | |
c7ad5ad2 AL |
202 | static void setup_pcid(void) |
203 | { | |
6cff64b8 DH |
204 | if (!IS_ENABLED(CONFIG_X86_64)) |
205 | return; | |
206 | ||
207 | if (!boot_cpu_has(X86_FEATURE_PCID)) | |
208 | return; | |
209 | ||
210 | if (boot_cpu_has(X86_FEATURE_PGE)) { | |
211 | /* | |
212 | * This can't be cr4_set_bits_and_update_boot() -- the | |
213 | * trampoline code can't handle CR4.PCIDE and it wouldn't | |
214 | * do any good anyway. Despite the name, | |
215 | * cr4_set_bits_and_update_boot() doesn't actually cause | |
216 | * the bits in question to remain set all the way through | |
217 | * the secondary boot asm. | |
218 | * | |
219 | * Instead, we brute-force it and set CR4.PCIDE manually in | |
220 | * start_secondary(). | |
221 | */ | |
222 | cr4_set_bits(X86_CR4_PCIDE); | |
223 | ||
224 | /* | |
225 | * INVPCID's single-context modes (2/3) only work if we set | |
226 | * X86_CR4_PCIDE, *and* we INVPCID support. It's unusable | |
227 | * on systems that have X86_CR4_PCIDE clear, or that have | |
228 | * no INVPCID support at all. | |
229 | */ | |
230 | if (boot_cpu_has(X86_FEATURE_INVPCID)) | |
231 | setup_force_cpu_cap(X86_FEATURE_INVPCID_SINGLE); | |
232 | } else { | |
233 | /* | |
234 | * flush_tlb_all(), as currently implemented, won't work if | |
235 | * PCID is on but PGE is not. Since that combination | |
236 | * doesn't exist on real hardware, there's no reason to try | |
237 | * to fully support it, but it's polite to avoid corrupting | |
238 | * data if we're on an improperly configured VM. | |
239 | */ | |
240 | setup_clear_cpu_cap(X86_FEATURE_PCID); | |
c7ad5ad2 | 241 | } |
c7ad5ad2 AL |
242 | } |
243 | ||
f765090a PE |
244 | #ifdef CONFIG_X86_32 |
245 | #define NR_RANGE_MR 3 | |
246 | #else /* CONFIG_X86_64 */ | |
247 | #define NR_RANGE_MR 5 | |
248 | #endif | |
249 | ||
dc9dd5cc JB |
250 | static int __meminit save_mr(struct map_range *mr, int nr_range, |
251 | unsigned long start_pfn, unsigned long end_pfn, | |
252 | unsigned long page_size_mask) | |
f765090a PE |
253 | { |
254 | if (start_pfn < end_pfn) { | |
255 | if (nr_range >= NR_RANGE_MR) | |
256 | panic("run out of range for init_memory_mapping\n"); | |
257 | mr[nr_range].start = start_pfn<<PAGE_SHIFT; | |
258 | mr[nr_range].end = end_pfn<<PAGE_SHIFT; | |
259 | mr[nr_range].page_size_mask = page_size_mask; | |
260 | nr_range++; | |
261 | } | |
262 | ||
263 | return nr_range; | |
264 | } | |
265 | ||
aeebe84c YL |
266 | /* |
267 | * adjust the page_size_mask for small range to go with | |
268 | * big page size instead small one if nearby are ram too. | |
269 | */ | |
bd721ea7 | 270 | static void __ref adjust_range_page_size_mask(struct map_range *mr, |
aeebe84c YL |
271 | int nr_range) |
272 | { | |
273 | int i; | |
274 | ||
275 | for (i = 0; i < nr_range; i++) { | |
276 | if ((page_size_mask & (1<<PG_LEVEL_2M)) && | |
277 | !(mr[i].page_size_mask & (1<<PG_LEVEL_2M))) { | |
278 | unsigned long start = round_down(mr[i].start, PMD_SIZE); | |
279 | unsigned long end = round_up(mr[i].end, PMD_SIZE); | |
280 | ||
281 | #ifdef CONFIG_X86_32 | |
282 | if ((end >> PAGE_SHIFT) > max_low_pfn) | |
283 | continue; | |
284 | #endif | |
285 | ||
286 | if (memblock_is_region_memory(start, end - start)) | |
287 | mr[i].page_size_mask |= 1<<PG_LEVEL_2M; | |
288 | } | |
289 | if ((page_size_mask & (1<<PG_LEVEL_1G)) && | |
290 | !(mr[i].page_size_mask & (1<<PG_LEVEL_1G))) { | |
291 | unsigned long start = round_down(mr[i].start, PUD_SIZE); | |
292 | unsigned long end = round_up(mr[i].end, PUD_SIZE); | |
293 | ||
294 | if (memblock_is_region_memory(start, end - start)) | |
295 | mr[i].page_size_mask |= 1<<PG_LEVEL_1G; | |
296 | } | |
297 | } | |
298 | } | |
299 | ||
f15e0518 DH |
300 | static const char *page_size_string(struct map_range *mr) |
301 | { | |
302 | static const char str_1g[] = "1G"; | |
303 | static const char str_2m[] = "2M"; | |
304 | static const char str_4m[] = "4M"; | |
305 | static const char str_4k[] = "4k"; | |
306 | ||
307 | if (mr->page_size_mask & (1<<PG_LEVEL_1G)) | |
308 | return str_1g; | |
309 | /* | |
310 | * 32-bit without PAE has a 4M large page size. | |
311 | * PG_LEVEL_2M is misnamed, but we can at least | |
312 | * print out the right size in the string. | |
313 | */ | |
314 | if (IS_ENABLED(CONFIG_X86_32) && | |
315 | !IS_ENABLED(CONFIG_X86_PAE) && | |
316 | mr->page_size_mask & (1<<PG_LEVEL_2M)) | |
317 | return str_4m; | |
318 | ||
319 | if (mr->page_size_mask & (1<<PG_LEVEL_2M)) | |
320 | return str_2m; | |
321 | ||
322 | return str_4k; | |
323 | } | |
324 | ||
4e33e065 YL |
325 | static int __meminit split_mem_range(struct map_range *mr, int nr_range, |
326 | unsigned long start, | |
327 | unsigned long end) | |
f765090a | 328 | { |
2e8059ed | 329 | unsigned long start_pfn, end_pfn, limit_pfn; |
1829ae9a | 330 | unsigned long pfn; |
4e33e065 | 331 | int i; |
f765090a | 332 | |
2e8059ed YL |
333 | limit_pfn = PFN_DOWN(end); |
334 | ||
f765090a | 335 | /* head if not big page alignment ? */ |
1829ae9a | 336 | pfn = start_pfn = PFN_DOWN(start); |
f765090a PE |
337 | #ifdef CONFIG_X86_32 |
338 | /* | |
339 | * Don't use a large page for the first 2/4MB of memory | |
340 | * because there are often fixed size MTRRs in there | |
341 | * and overlapping MTRRs into large pages can cause | |
342 | * slowdowns. | |
343 | */ | |
1829ae9a | 344 | if (pfn == 0) |
84d77001 | 345 | end_pfn = PFN_DOWN(PMD_SIZE); |
f765090a | 346 | else |
1829ae9a | 347 | end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 348 | #else /* CONFIG_X86_64 */ |
1829ae9a | 349 | end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 350 | #endif |
2e8059ed YL |
351 | if (end_pfn > limit_pfn) |
352 | end_pfn = limit_pfn; | |
f765090a PE |
353 | if (start_pfn < end_pfn) { |
354 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); | |
1829ae9a | 355 | pfn = end_pfn; |
f765090a PE |
356 | } |
357 | ||
358 | /* big page (2M) range */ | |
1829ae9a | 359 | start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 360 | #ifdef CONFIG_X86_32 |
2e8059ed | 361 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 362 | #else /* CONFIG_X86_64 */ |
1829ae9a | 363 | end_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE)); |
2e8059ed YL |
364 | if (end_pfn > round_down(limit_pfn, PFN_DOWN(PMD_SIZE))) |
365 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); | |
f765090a PE |
366 | #endif |
367 | ||
368 | if (start_pfn < end_pfn) { | |
369 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, | |
370 | page_size_mask & (1<<PG_LEVEL_2M)); | |
1829ae9a | 371 | pfn = end_pfn; |
f765090a PE |
372 | } |
373 | ||
374 | #ifdef CONFIG_X86_64 | |
375 | /* big page (1G) range */ | |
1829ae9a | 376 | start_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE)); |
2e8059ed | 377 | end_pfn = round_down(limit_pfn, PFN_DOWN(PUD_SIZE)); |
f765090a PE |
378 | if (start_pfn < end_pfn) { |
379 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, | |
380 | page_size_mask & | |
381 | ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G))); | |
1829ae9a | 382 | pfn = end_pfn; |
f765090a PE |
383 | } |
384 | ||
385 | /* tail is not big page (1G) alignment */ | |
1829ae9a | 386 | start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
2e8059ed | 387 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); |
f765090a PE |
388 | if (start_pfn < end_pfn) { |
389 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, | |
390 | page_size_mask & (1<<PG_LEVEL_2M)); | |
1829ae9a | 391 | pfn = end_pfn; |
f765090a PE |
392 | } |
393 | #endif | |
394 | ||
395 | /* tail is not big page (2M) alignment */ | |
1829ae9a | 396 | start_pfn = pfn; |
2e8059ed | 397 | end_pfn = limit_pfn; |
f765090a PE |
398 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); |
399 | ||
7de3d66b YL |
400 | if (!after_bootmem) |
401 | adjust_range_page_size_mask(mr, nr_range); | |
402 | ||
f765090a PE |
403 | /* try to merge same page size and continuous */ |
404 | for (i = 0; nr_range > 1 && i < nr_range - 1; i++) { | |
405 | unsigned long old_start; | |
406 | if (mr[i].end != mr[i+1].start || | |
407 | mr[i].page_size_mask != mr[i+1].page_size_mask) | |
408 | continue; | |
409 | /* move it */ | |
410 | old_start = mr[i].start; | |
411 | memmove(&mr[i], &mr[i+1], | |
412 | (nr_range - 1 - i) * sizeof(struct map_range)); | |
413 | mr[i--].start = old_start; | |
414 | nr_range--; | |
415 | } | |
416 | ||
417 | for (i = 0; i < nr_range; i++) | |
c9cdaeb2 | 418 | pr_debug(" [mem %#010lx-%#010lx] page %s\n", |
365811d6 | 419 | mr[i].start, mr[i].end - 1, |
f15e0518 | 420 | page_size_string(&mr[i])); |
f765090a | 421 | |
4e33e065 YL |
422 | return nr_range; |
423 | } | |
424 | ||
08b46d5d | 425 | struct range pfn_mapped[E820_MAX_ENTRIES]; |
0e691cf8 | 426 | int nr_pfn_mapped; |
66520ebc JS |
427 | |
428 | static void add_pfn_range_mapped(unsigned long start_pfn, unsigned long end_pfn) | |
429 | { | |
08b46d5d | 430 | nr_pfn_mapped = add_range_with_merge(pfn_mapped, E820_MAX_ENTRIES, |
66520ebc | 431 | nr_pfn_mapped, start_pfn, end_pfn); |
08b46d5d | 432 | nr_pfn_mapped = clean_sort_range(pfn_mapped, E820_MAX_ENTRIES); |
66520ebc JS |
433 | |
434 | max_pfn_mapped = max(max_pfn_mapped, end_pfn); | |
435 | ||
436 | if (start_pfn < (1UL<<(32-PAGE_SHIFT))) | |
437 | max_low_pfn_mapped = max(max_low_pfn_mapped, | |
438 | min(end_pfn, 1UL<<(32-PAGE_SHIFT))); | |
439 | } | |
440 | ||
441 | bool pfn_range_is_mapped(unsigned long start_pfn, unsigned long end_pfn) | |
442 | { | |
443 | int i; | |
444 | ||
445 | for (i = 0; i < nr_pfn_mapped; i++) | |
446 | if ((start_pfn >= pfn_mapped[i].start) && | |
447 | (end_pfn <= pfn_mapped[i].end)) | |
448 | return true; | |
449 | ||
450 | return false; | |
451 | } | |
452 | ||
4e33e065 YL |
453 | /* |
454 | * Setup the direct mapping of the physical memory at PAGE_OFFSET. | |
455 | * This runs before bootmem is initialized and gets pages directly from | |
456 | * the physical memory. To access them they are temporarily mapped. | |
457 | */ | |
bd721ea7 | 458 | unsigned long __ref init_memory_mapping(unsigned long start, |
4e33e065 YL |
459 | unsigned long end) |
460 | { | |
461 | struct map_range mr[NR_RANGE_MR]; | |
462 | unsigned long ret = 0; | |
463 | int nr_range, i; | |
464 | ||
c9cdaeb2 | 465 | pr_debug("init_memory_mapping: [mem %#010lx-%#010lx]\n", |
4e33e065 YL |
466 | start, end - 1); |
467 | ||
468 | memset(mr, 0, sizeof(mr)); | |
469 | nr_range = split_mem_range(mr, 0, start, end); | |
470 | ||
f765090a PE |
471 | for (i = 0; i < nr_range; i++) |
472 | ret = kernel_physical_mapping_init(mr[i].start, mr[i].end, | |
473 | mr[i].page_size_mask); | |
f765090a | 474 | |
66520ebc JS |
475 | add_pfn_range_mapped(start >> PAGE_SHIFT, ret >> PAGE_SHIFT); |
476 | ||
c14fa0b6 YL |
477 | return ret >> PAGE_SHIFT; |
478 | } | |
479 | ||
66520ebc | 480 | /* |
cf8b166d | 481 | * We need to iterate through the E820 memory map and create direct mappings |
09821ff1 | 482 | * for only E820_TYPE_RAM and E820_KERN_RESERVED regions. We cannot simply |
cf8b166d ZY |
483 | * create direct mappings for all pfns from [0 to max_low_pfn) and |
484 | * [4GB to max_pfn) because of possible memory holes in high addresses | |
485 | * that cannot be marked as UC by fixed/variable range MTRRs. | |
486 | * Depending on the alignment of E820 ranges, this may possibly result | |
487 | * in using smaller size (i.e. 4K instead of 2M or 1G) page tables. | |
488 | * | |
489 | * init_mem_mapping() calls init_range_memory_mapping() with big range. | |
490 | * That range would have hole in the middle or ends, and only ram parts | |
491 | * will be mapped in init_range_memory_mapping(). | |
66520ebc | 492 | */ |
8d57470d | 493 | static unsigned long __init init_range_memory_mapping( |
b8fd39c0 YL |
494 | unsigned long r_start, |
495 | unsigned long r_end) | |
66520ebc JS |
496 | { |
497 | unsigned long start_pfn, end_pfn; | |
8d57470d | 498 | unsigned long mapped_ram_size = 0; |
66520ebc JS |
499 | int i; |
500 | ||
66520ebc | 501 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) { |
b8fd39c0 YL |
502 | u64 start = clamp_val(PFN_PHYS(start_pfn), r_start, r_end); |
503 | u64 end = clamp_val(PFN_PHYS(end_pfn), r_start, r_end); | |
504 | if (start >= end) | |
66520ebc JS |
505 | continue; |
506 | ||
c9b3234a YL |
507 | /* |
508 | * if it is overlapping with brk pgt, we need to | |
509 | * alloc pgt buf from memblock instead. | |
510 | */ | |
511 | can_use_brk_pgt = max(start, (u64)pgt_buf_end<<PAGE_SHIFT) >= | |
512 | min(end, (u64)pgt_buf_top<<PAGE_SHIFT); | |
f763ad1d | 513 | init_memory_mapping(start, end); |
8d57470d | 514 | mapped_ram_size += end - start; |
c9b3234a | 515 | can_use_brk_pgt = true; |
66520ebc | 516 | } |
8d57470d YL |
517 | |
518 | return mapped_ram_size; | |
66520ebc JS |
519 | } |
520 | ||
6979287a YL |
521 | static unsigned long __init get_new_step_size(unsigned long step_size) |
522 | { | |
523 | /* | |
132978b9 | 524 | * Initial mapped size is PMD_SIZE (2M). |
6979287a YL |
525 | * We can not set step_size to be PUD_SIZE (1G) yet. |
526 | * In worse case, when we cross the 1G boundary, and | |
527 | * PG_LEVEL_2M is not set, we will need 1+1+512 pages (2M + 8k) | |
132978b9 JB |
528 | * to map 1G range with PTE. Hence we use one less than the |
529 | * difference of page table level shifts. | |
6979287a | 530 | * |
132978b9 JB |
531 | * Don't need to worry about overflow in the top-down case, on 32bit, |
532 | * when step_size is 0, round_down() returns 0 for start, and that | |
533 | * turns it into 0x100000000ULL. | |
534 | * In the bottom-up case, round_up(x, 0) returns 0 though too, which | |
535 | * needs to be taken into consideration by the code below. | |
6979287a | 536 | */ |
132978b9 | 537 | return step_size << (PMD_SHIFT - PAGE_SHIFT - 1); |
6979287a YL |
538 | } |
539 | ||
0167d7d8 TC |
540 | /** |
541 | * memory_map_top_down - Map [map_start, map_end) top down | |
542 | * @map_start: start address of the target memory range | |
543 | * @map_end: end address of the target memory range | |
544 | * | |
545 | * This function will setup direct mapping for memory range | |
546 | * [map_start, map_end) in top-down. That said, the page tables | |
547 | * will be allocated at the end of the memory, and we map the | |
548 | * memory in top-down. | |
549 | */ | |
550 | static void __init memory_map_top_down(unsigned long map_start, | |
551 | unsigned long map_end) | |
c14fa0b6 | 552 | { |
0167d7d8 | 553 | unsigned long real_end, start, last_start; |
8d57470d YL |
554 | unsigned long step_size; |
555 | unsigned long addr; | |
556 | unsigned long mapped_ram_size = 0; | |
ab951937 | 557 | |
98e7a989 | 558 | /* xen has big range in reserved near end of ram, skip it at first.*/ |
0167d7d8 | 559 | addr = memblock_find_in_range(map_start, map_end, PMD_SIZE, PMD_SIZE); |
8d57470d YL |
560 | real_end = addr + PMD_SIZE; |
561 | ||
562 | /* step_size need to be small so pgt_buf from BRK could cover it */ | |
563 | step_size = PMD_SIZE; | |
564 | max_pfn_mapped = 0; /* will get exact value next */ | |
565 | min_pfn_mapped = real_end >> PAGE_SHIFT; | |
566 | last_start = start = real_end; | |
cf8b166d ZY |
567 | |
568 | /* | |
569 | * We start from the top (end of memory) and go to the bottom. | |
570 | * The memblock_find_in_range() gets us a block of RAM from the | |
571 | * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages | |
572 | * for page table. | |
573 | */ | |
0167d7d8 | 574 | while (last_start > map_start) { |
8d57470d YL |
575 | if (last_start > step_size) { |
576 | start = round_down(last_start - 1, step_size); | |
0167d7d8 TC |
577 | if (start < map_start) |
578 | start = map_start; | |
8d57470d | 579 | } else |
0167d7d8 | 580 | start = map_start; |
132978b9 | 581 | mapped_ram_size += init_range_memory_mapping(start, |
8d57470d YL |
582 | last_start); |
583 | last_start = start; | |
584 | min_pfn_mapped = last_start >> PAGE_SHIFT; | |
132978b9 | 585 | if (mapped_ram_size >= step_size) |
6979287a | 586 | step_size = get_new_step_size(step_size); |
8d57470d YL |
587 | } |
588 | ||
0167d7d8 TC |
589 | if (real_end < map_end) |
590 | init_range_memory_mapping(real_end, map_end); | |
591 | } | |
592 | ||
b959ed6c TC |
593 | /** |
594 | * memory_map_bottom_up - Map [map_start, map_end) bottom up | |
595 | * @map_start: start address of the target memory range | |
596 | * @map_end: end address of the target memory range | |
597 | * | |
598 | * This function will setup direct mapping for memory range | |
599 | * [map_start, map_end) in bottom-up. Since we have limited the | |
600 | * bottom-up allocation above the kernel, the page tables will | |
601 | * be allocated just above the kernel and we map the memory | |
602 | * in [map_start, map_end) in bottom-up. | |
603 | */ | |
604 | static void __init memory_map_bottom_up(unsigned long map_start, | |
605 | unsigned long map_end) | |
606 | { | |
132978b9 | 607 | unsigned long next, start; |
b959ed6c TC |
608 | unsigned long mapped_ram_size = 0; |
609 | /* step_size need to be small so pgt_buf from BRK could cover it */ | |
610 | unsigned long step_size = PMD_SIZE; | |
611 | ||
612 | start = map_start; | |
613 | min_pfn_mapped = start >> PAGE_SHIFT; | |
614 | ||
615 | /* | |
616 | * We start from the bottom (@map_start) and go to the top (@map_end). | |
617 | * The memblock_find_in_range() gets us a block of RAM from the | |
618 | * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages | |
619 | * for page table. | |
620 | */ | |
621 | while (start < map_end) { | |
132978b9 | 622 | if (step_size && map_end - start > step_size) { |
b959ed6c TC |
623 | next = round_up(start + 1, step_size); |
624 | if (next > map_end) | |
625 | next = map_end; | |
132978b9 | 626 | } else { |
b959ed6c | 627 | next = map_end; |
132978b9 | 628 | } |
b959ed6c | 629 | |
132978b9 | 630 | mapped_ram_size += init_range_memory_mapping(start, next); |
b959ed6c TC |
631 | start = next; |
632 | ||
132978b9 | 633 | if (mapped_ram_size >= step_size) |
b959ed6c | 634 | step_size = get_new_step_size(step_size); |
b959ed6c TC |
635 | } |
636 | } | |
637 | ||
0167d7d8 TC |
638 | void __init init_mem_mapping(void) |
639 | { | |
640 | unsigned long end; | |
641 | ||
aa8c6248 | 642 | pti_check_boottime_disable(); |
0167d7d8 | 643 | probe_page_size_mask(); |
c7ad5ad2 | 644 | setup_pcid(); |
0167d7d8 TC |
645 | |
646 | #ifdef CONFIG_X86_64 | |
647 | end = max_pfn << PAGE_SHIFT; | |
648 | #else | |
649 | end = max_low_pfn << PAGE_SHIFT; | |
650 | #endif | |
651 | ||
652 | /* the ISA range is always mapped regardless of memory holes */ | |
653 | init_memory_mapping(0, ISA_END_ADDRESS); | |
654 | ||
b234e8a0 TG |
655 | /* Init the trampoline, possibly with KASLR memory offset */ |
656 | init_trampoline(); | |
657 | ||
b959ed6c TC |
658 | /* |
659 | * If the allocation is in bottom-up direction, we setup direct mapping | |
660 | * in bottom-up, otherwise we setup direct mapping in top-down. | |
661 | */ | |
662 | if (memblock_bottom_up()) { | |
663 | unsigned long kernel_end = __pa_symbol(_end); | |
664 | ||
665 | /* | |
666 | * we need two separate calls here. This is because we want to | |
667 | * allocate page tables above the kernel. So we first map | |
668 | * [kernel_end, end) to make memory above the kernel be mapped | |
669 | * as soon as possible. And then use page tables allocated above | |
670 | * the kernel to map [ISA_END_ADDRESS, kernel_end). | |
671 | */ | |
672 | memory_map_bottom_up(kernel_end, end); | |
673 | memory_map_bottom_up(ISA_END_ADDRESS, kernel_end); | |
674 | } else { | |
675 | memory_map_top_down(ISA_END_ADDRESS, end); | |
676 | } | |
8d57470d | 677 | |
f763ad1d YL |
678 | #ifdef CONFIG_X86_64 |
679 | if (max_pfn > max_low_pfn) { | |
680 | /* can we preseve max_low_pfn ?*/ | |
681 | max_low_pfn = max_pfn; | |
682 | } | |
719272c4 YL |
683 | #else |
684 | early_ioremap_page_table_range_init(); | |
8170e6be PA |
685 | #endif |
686 | ||
719272c4 YL |
687 | load_cr3(swapper_pg_dir); |
688 | __flush_tlb_all(); | |
719272c4 | 689 | |
f72e38e8 | 690 | x86_init.hyper.init_mem_mapping(); |
c138d811 | 691 | |
c14fa0b6 | 692 | early_memtest(0, max_pfn_mapped << PAGE_SHIFT); |
22ddfcaa | 693 | } |
e5b2bb55 | 694 | |
540aca06 PE |
695 | /* |
696 | * devmem_is_allowed() checks to see if /dev/mem access to a certain address | |
697 | * is valid. The argument is a physical page number. | |
698 | * | |
a4866aa8 KC |
699 | * On x86, access has to be given to the first megabyte of RAM because that |
700 | * area traditionally contains BIOS code and data regions used by X, dosemu, | |
701 | * and similar apps. Since they map the entire memory range, the whole range | |
702 | * must be allowed (for mapping), but any areas that would otherwise be | |
703 | * disallowed are flagged as being "zero filled" instead of rejected. | |
704 | * Access has to be given to non-kernel-ram areas as well, these contain the | |
705 | * PCI mmio resources as well as potential bios/acpi data regions. | |
540aca06 PE |
706 | */ |
707 | int devmem_is_allowed(unsigned long pagenr) | |
708 | { | |
a4866aa8 KC |
709 | if (page_is_ram(pagenr)) { |
710 | /* | |
711 | * For disallowed memory regions in the low 1MB range, | |
712 | * request that the page be shown as all zeros. | |
713 | */ | |
714 | if (pagenr < 256) | |
715 | return 2; | |
716 | ||
717 | return 0; | |
718 | } | |
719 | ||
720 | /* | |
721 | * This must follow RAM test, since System RAM is considered a | |
722 | * restricted resource under CONFIG_STRICT_IOMEM. | |
723 | */ | |
724 | if (iomem_is_exclusive(pagenr << PAGE_SHIFT)) { | |
725 | /* Low 1MB bypasses iomem restrictions. */ | |
726 | if (pagenr < 256) | |
727 | return 1; | |
728 | ||
540aca06 | 729 | return 0; |
a4866aa8 KC |
730 | } |
731 | ||
732 | return 1; | |
540aca06 PE |
733 | } |
734 | ||
e5b2bb55 PE |
735 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
736 | { | |
c967da6a | 737 | unsigned long begin_aligned, end_aligned; |
e5b2bb55 | 738 | |
c967da6a YL |
739 | /* Make sure boundaries are page aligned */ |
740 | begin_aligned = PAGE_ALIGN(begin); | |
741 | end_aligned = end & PAGE_MASK; | |
742 | ||
743 | if (WARN_ON(begin_aligned != begin || end_aligned != end)) { | |
744 | begin = begin_aligned; | |
745 | end = end_aligned; | |
746 | } | |
747 | ||
748 | if (begin >= end) | |
e5b2bb55 PE |
749 | return; |
750 | ||
751 | /* | |
752 | * If debugging page accesses then do not free this memory but | |
753 | * mark them not present - any buggy init-section access will | |
754 | * create a kernel page fault: | |
755 | */ | |
a75e1f63 CB |
756 | if (debug_pagealloc_enabled()) { |
757 | pr_info("debug: unmapping init [mem %#010lx-%#010lx]\n", | |
758 | begin, end - 1); | |
759 | set_memory_np(begin, (end - begin) >> PAGE_SHIFT); | |
760 | } else { | |
761 | /* | |
762 | * We just marked the kernel text read only above, now that | |
763 | * we are going to free part of that, we need to make that | |
764 | * writeable and non-executable first. | |
765 | */ | |
766 | set_memory_nx(begin, (end - begin) >> PAGE_SHIFT); | |
767 | set_memory_rw(begin, (end - begin) >> PAGE_SHIFT); | |
e5b2bb55 | 768 | |
a75e1f63 CB |
769 | free_reserved_area((void *)begin, (void *)end, |
770 | POISON_FREE_INITMEM, what); | |
771 | } | |
e5b2bb55 PE |
772 | } |
773 | ||
18278229 | 774 | void __ref free_initmem(void) |
e5b2bb55 | 775 | { |
0c6fc11a | 776 | e820__reallocate_tables(); |
47533968 | 777 | |
c88442ec | 778 | free_init_pages("unused kernel", |
e5b2bb55 PE |
779 | (unsigned long)(&__init_begin), |
780 | (unsigned long)(&__init_end)); | |
781 | } | |
731ddea6 PE |
782 | |
783 | #ifdef CONFIG_BLK_DEV_INITRD | |
0d26d1d8 | 784 | void __init free_initrd_mem(unsigned long start, unsigned long end) |
731ddea6 | 785 | { |
c967da6a YL |
786 | /* |
787 | * end could be not aligned, and We can not align that, | |
788 | * decompresser could be confused by aligned initrd_end | |
789 | * We already reserve the end partial page before in | |
790 | * - i386_start_kernel() | |
791 | * - x86_64_start_kernel() | |
792 | * - relocate_initrd() | |
793 | * So here We can do PAGE_ALIGN() safely to get partial page to be freed | |
794 | */ | |
c88442ec | 795 | free_init_pages("initrd", start, PAGE_ALIGN(end)); |
731ddea6 PE |
796 | } |
797 | #endif | |
17623915 | 798 | |
4270fd8b IM |
799 | /* |
800 | * Calculate the precise size of the DMA zone (first 16 MB of RAM), | |
801 | * and pass it to the MM layer - to help it set zone watermarks more | |
802 | * accurately. | |
803 | * | |
804 | * Done on 64-bit systems only for the time being, although 32-bit systems | |
805 | * might benefit from this as well. | |
806 | */ | |
807 | void __init memblock_find_dma_reserve(void) | |
808 | { | |
809 | #ifdef CONFIG_X86_64 | |
810 | u64 nr_pages = 0, nr_free_pages = 0; | |
811 | unsigned long start_pfn, end_pfn; | |
812 | phys_addr_t start_addr, end_addr; | |
813 | int i; | |
814 | u64 u; | |
815 | ||
816 | /* | |
817 | * Iterate over all memory ranges (free and reserved ones alike), | |
818 | * to calculate the total number of pages in the first 16 MB of RAM: | |
819 | */ | |
820 | nr_pages = 0; | |
821 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) { | |
822 | start_pfn = min(start_pfn, MAX_DMA_PFN); | |
823 | end_pfn = min(end_pfn, MAX_DMA_PFN); | |
824 | ||
825 | nr_pages += end_pfn - start_pfn; | |
826 | } | |
827 | ||
828 | /* | |
829 | * Iterate over free memory ranges to calculate the number of free | |
830 | * pages in the DMA zone, while not counting potential partial | |
831 | * pages at the beginning or the end of the range: | |
832 | */ | |
833 | nr_free_pages = 0; | |
834 | for_each_free_mem_range(u, NUMA_NO_NODE, MEMBLOCK_NONE, &start_addr, &end_addr, NULL) { | |
835 | start_pfn = min_t(unsigned long, PFN_UP(start_addr), MAX_DMA_PFN); | |
836 | end_pfn = min_t(unsigned long, PFN_DOWN(end_addr), MAX_DMA_PFN); | |
837 | ||
838 | if (start_pfn < end_pfn) | |
839 | nr_free_pages += end_pfn - start_pfn; | |
840 | } | |
841 | ||
842 | set_dma_reserve(nr_pages - nr_free_pages); | |
843 | #endif | |
844 | } | |
845 | ||
17623915 PE |
846 | void __init zone_sizes_init(void) |
847 | { | |
848 | unsigned long max_zone_pfns[MAX_NR_ZONES]; | |
849 | ||
850 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
851 | ||
852 | #ifdef CONFIG_ZONE_DMA | |
c072b90c | 853 | max_zone_pfns[ZONE_DMA] = min(MAX_DMA_PFN, max_low_pfn); |
17623915 PE |
854 | #endif |
855 | #ifdef CONFIG_ZONE_DMA32 | |
c072b90c | 856 | max_zone_pfns[ZONE_DMA32] = min(MAX_DMA32_PFN, max_low_pfn); |
17623915 PE |
857 | #endif |
858 | max_zone_pfns[ZONE_NORMAL] = max_low_pfn; | |
859 | #ifdef CONFIG_HIGHMEM | |
860 | max_zone_pfns[ZONE_HIGHMEM] = max_pfn; | |
861 | #endif | |
862 | ||
863 | free_area_init_nodes(max_zone_pfns); | |
864 | } | |
865 | ||
6fd166aa | 866 | __visible DEFINE_PER_CPU_SHARED_ALIGNED(struct tlb_state, cpu_tlbstate) = { |
3d28ebce | 867 | .loaded_mm = &init_mm, |
10af6235 | 868 | .next_asid = 1, |
1e02ce4c AL |
869 | .cr4 = ~0UL, /* fail hard if we screw up cr4 shadow initialization */ |
870 | }; | |
1e547681 | 871 | EXPORT_PER_CPU_SYMBOL(cpu_tlbstate); |
1e02ce4c | 872 | |
bd809af1 JG |
873 | void update_cache_mode_entry(unsigned entry, enum page_cache_mode cache) |
874 | { | |
875 | /* entry 0 MUST be WB (hardwired to speed up translations) */ | |
876 | BUG_ON(!entry && cache != _PAGE_CACHE_MODE_WB); | |
877 | ||
878 | __cachemode2pte_tbl[cache] = __cm_idx2pte(entry); | |
879 | __pte2cachemode_tbl[entry] = cache; | |
880 | } |