]>
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. | |
32 | * Minimal supported modes are defined statically, modified if more supported | |
33 | * cache modes are available. | |
34 | * Index into __cachemode2pte_tbl is the cachemode. | |
35 | * Index into __pte2cachemode_tbl are the caching attribute bits of the pte | |
36 | * (_PAGE_PWT, _PAGE_PCD, _PAGE_PAT) at index bit positions 0, 1, 2. | |
37 | */ | |
38 | uint16_t __cachemode2pte_tbl[_PAGE_CACHE_MODE_NUM] = { | |
39 | [_PAGE_CACHE_MODE_WB] = 0, | |
40 | [_PAGE_CACHE_MODE_WC] = _PAGE_PWT, | |
41 | [_PAGE_CACHE_MODE_UC_MINUS] = _PAGE_PCD, | |
42 | [_PAGE_CACHE_MODE_UC] = _PAGE_PCD | _PAGE_PWT, | |
43 | [_PAGE_CACHE_MODE_WT] = _PAGE_PCD, | |
44 | [_PAGE_CACHE_MODE_WP] = _PAGE_PCD, | |
45 | }; | |
31bb7723 | 46 | EXPORT_SYMBOL(__cachemode2pte_tbl); |
281d4078 JG |
47 | uint8_t __pte2cachemode_tbl[8] = { |
48 | [__pte2cm_idx(0)] = _PAGE_CACHE_MODE_WB, | |
49 | [__pte2cm_idx(_PAGE_PWT)] = _PAGE_CACHE_MODE_WC, | |
50 | [__pte2cm_idx(_PAGE_PCD)] = _PAGE_CACHE_MODE_UC_MINUS, | |
51 | [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD)] = _PAGE_CACHE_MODE_UC, | |
52 | [__pte2cm_idx(_PAGE_PAT)] = _PAGE_CACHE_MODE_WB, | |
53 | [__pte2cm_idx(_PAGE_PWT | _PAGE_PAT)] = _PAGE_CACHE_MODE_WC, | |
54 | [__pte2cm_idx(_PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS, | |
55 | [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC, | |
56 | }; | |
31bb7723 | 57 | EXPORT_SYMBOL(__pte2cachemode_tbl); |
281d4078 | 58 | |
cf470659 YL |
59 | static unsigned long __initdata pgt_buf_start; |
60 | static unsigned long __initdata pgt_buf_end; | |
61 | static unsigned long __initdata pgt_buf_top; | |
f765090a | 62 | |
9985b4c6 YL |
63 | static unsigned long min_pfn_mapped; |
64 | ||
c9b3234a YL |
65 | static bool __initdata can_use_brk_pgt = true; |
66 | ||
ddd3509d SS |
67 | /* |
68 | * Pages returned are already directly mapped. | |
69 | * | |
70 | * Changing that is likely to break Xen, see commit: | |
71 | * | |
72 | * 279b706 x86,xen: introduce x86_init.mapping.pagetable_reserve | |
73 | * | |
74 | * for detailed information. | |
75 | */ | |
22c8ca2a | 76 | __ref void *alloc_low_pages(unsigned int num) |
5c51bdbe YL |
77 | { |
78 | unsigned long pfn; | |
22c8ca2a | 79 | int i; |
5c51bdbe | 80 | |
5c51bdbe | 81 | if (after_bootmem) { |
22c8ca2a | 82 | unsigned int order; |
5c51bdbe | 83 | |
22c8ca2a YL |
84 | order = get_order((unsigned long)num << PAGE_SHIFT); |
85 | return (void *)__get_free_pages(GFP_ATOMIC | __GFP_NOTRACK | | |
86 | __GFP_ZERO, order); | |
5c51bdbe | 87 | } |
5c51bdbe | 88 | |
c9b3234a | 89 | if ((pgt_buf_end + num) > pgt_buf_top || !can_use_brk_pgt) { |
5c51bdbe YL |
90 | unsigned long ret; |
91 | if (min_pfn_mapped >= max_pfn_mapped) | |
d4dd100f | 92 | panic("alloc_low_pages: ran out of memory"); |
5c51bdbe YL |
93 | ret = memblock_find_in_range(min_pfn_mapped << PAGE_SHIFT, |
94 | max_pfn_mapped << PAGE_SHIFT, | |
22c8ca2a | 95 | PAGE_SIZE * num , PAGE_SIZE); |
5c51bdbe | 96 | if (!ret) |
d4dd100f | 97 | panic("alloc_low_pages: can not alloc memory"); |
22c8ca2a | 98 | memblock_reserve(ret, PAGE_SIZE * num); |
5c51bdbe | 99 | pfn = ret >> PAGE_SHIFT; |
22c8ca2a YL |
100 | } else { |
101 | pfn = pgt_buf_end; | |
102 | pgt_buf_end += num; | |
c9b3234a YL |
103 | printk(KERN_DEBUG "BRK [%#010lx, %#010lx] PGTABLE\n", |
104 | pfn << PAGE_SHIFT, (pgt_buf_end << PAGE_SHIFT) - 1); | |
22c8ca2a YL |
105 | } |
106 | ||
107 | for (i = 0; i < num; i++) { | |
108 | void *adr; | |
109 | ||
110 | adr = __va((pfn + i) << PAGE_SHIFT); | |
111 | clear_page(adr); | |
112 | } | |
5c51bdbe | 113 | |
22c8ca2a | 114 | return __va(pfn << PAGE_SHIFT); |
5c51bdbe YL |
115 | } |
116 | ||
527bf129 YL |
117 | /* need 3 4k for initial PMD_SIZE, 3 4k for 0-ISA_END_ADDRESS */ |
118 | #define INIT_PGT_BUF_SIZE (6 * PAGE_SIZE) | |
8d57470d YL |
119 | RESERVE_BRK(early_pgt_alloc, INIT_PGT_BUF_SIZE); |
120 | void __init early_alloc_pgt_buf(void) | |
121 | { | |
122 | unsigned long tables = INIT_PGT_BUF_SIZE; | |
123 | phys_addr_t base; | |
124 | ||
125 | base = __pa(extend_brk(tables, PAGE_SIZE)); | |
126 | ||
127 | pgt_buf_start = base >> PAGE_SHIFT; | |
128 | pgt_buf_end = pgt_buf_start; | |
129 | pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT); | |
130 | } | |
131 | ||
f765090a PE |
132 | int after_bootmem; |
133 | ||
134 | int direct_gbpages | |
135 | #ifdef CONFIG_DIRECT_GBPAGES | |
136 | = 1 | |
137 | #endif | |
138 | ; | |
139 | ||
148b2098 YL |
140 | static void __init init_gbpages(void) |
141 | { | |
142 | #ifdef CONFIG_X86_64 | |
143 | if (direct_gbpages && cpu_has_gbpages) | |
144 | printk(KERN_INFO "Using GB pages for direct mapping\n"); | |
145 | else | |
146 | direct_gbpages = 0; | |
147 | #endif | |
148 | } | |
149 | ||
844ab6f9 JS |
150 | struct map_range { |
151 | unsigned long start; | |
152 | unsigned long end; | |
153 | unsigned page_size_mask; | |
154 | }; | |
155 | ||
fa62aafe | 156 | static int page_size_mask; |
f765090a | 157 | |
22ddfcaa | 158 | static void __init probe_page_size_mask(void) |
fa62aafe | 159 | { |
148b2098 YL |
160 | init_gbpages(); |
161 | ||
fa62aafe YL |
162 | #if !defined(CONFIG_DEBUG_PAGEALLOC) && !defined(CONFIG_KMEMCHECK) |
163 | /* | |
164 | * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages. | |
165 | * This will simplify cpa(), which otherwise needs to support splitting | |
166 | * large pages into small in interrupt context, etc. | |
167 | */ | |
168 | if (direct_gbpages) | |
169 | page_size_mask |= 1 << PG_LEVEL_1G; | |
170 | if (cpu_has_pse) | |
171 | page_size_mask |= 1 << PG_LEVEL_2M; | |
172 | #endif | |
173 | ||
174 | /* Enable PSE if available */ | |
175 | if (cpu_has_pse) | |
176 | set_in_cr4(X86_CR4_PSE); | |
177 | ||
178 | /* Enable PGE if available */ | |
179 | if (cpu_has_pge) { | |
180 | set_in_cr4(X86_CR4_PGE); | |
181 | __supported_pte_mask |= _PAGE_GLOBAL; | |
182 | } | |
183 | } | |
279b706b | 184 | |
f765090a PE |
185 | #ifdef CONFIG_X86_32 |
186 | #define NR_RANGE_MR 3 | |
187 | #else /* CONFIG_X86_64 */ | |
188 | #define NR_RANGE_MR 5 | |
189 | #endif | |
190 | ||
dc9dd5cc JB |
191 | static int __meminit save_mr(struct map_range *mr, int nr_range, |
192 | unsigned long start_pfn, unsigned long end_pfn, | |
193 | unsigned long page_size_mask) | |
f765090a PE |
194 | { |
195 | if (start_pfn < end_pfn) { | |
196 | if (nr_range >= NR_RANGE_MR) | |
197 | panic("run out of range for init_memory_mapping\n"); | |
198 | mr[nr_range].start = start_pfn<<PAGE_SHIFT; | |
199 | mr[nr_range].end = end_pfn<<PAGE_SHIFT; | |
200 | mr[nr_range].page_size_mask = page_size_mask; | |
201 | nr_range++; | |
202 | } | |
203 | ||
204 | return nr_range; | |
205 | } | |
206 | ||
aeebe84c YL |
207 | /* |
208 | * adjust the page_size_mask for small range to go with | |
209 | * big page size instead small one if nearby are ram too. | |
210 | */ | |
211 | static void __init_refok adjust_range_page_size_mask(struct map_range *mr, | |
212 | int nr_range) | |
213 | { | |
214 | int i; | |
215 | ||
216 | for (i = 0; i < nr_range; i++) { | |
217 | if ((page_size_mask & (1<<PG_LEVEL_2M)) && | |
218 | !(mr[i].page_size_mask & (1<<PG_LEVEL_2M))) { | |
219 | unsigned long start = round_down(mr[i].start, PMD_SIZE); | |
220 | unsigned long end = round_up(mr[i].end, PMD_SIZE); | |
221 | ||
222 | #ifdef CONFIG_X86_32 | |
223 | if ((end >> PAGE_SHIFT) > max_low_pfn) | |
224 | continue; | |
225 | #endif | |
226 | ||
227 | if (memblock_is_region_memory(start, end - start)) | |
228 | mr[i].page_size_mask |= 1<<PG_LEVEL_2M; | |
229 | } | |
230 | if ((page_size_mask & (1<<PG_LEVEL_1G)) && | |
231 | !(mr[i].page_size_mask & (1<<PG_LEVEL_1G))) { | |
232 | unsigned long start = round_down(mr[i].start, PUD_SIZE); | |
233 | unsigned long end = round_up(mr[i].end, PUD_SIZE); | |
234 | ||
235 | if (memblock_is_region_memory(start, end - start)) | |
236 | mr[i].page_size_mask |= 1<<PG_LEVEL_1G; | |
237 | } | |
238 | } | |
239 | } | |
240 | ||
4e33e065 YL |
241 | static int __meminit split_mem_range(struct map_range *mr, int nr_range, |
242 | unsigned long start, | |
243 | unsigned long end) | |
f765090a | 244 | { |
2e8059ed | 245 | unsigned long start_pfn, end_pfn, limit_pfn; |
1829ae9a | 246 | unsigned long pfn; |
4e33e065 | 247 | int i; |
f765090a | 248 | |
2e8059ed YL |
249 | limit_pfn = PFN_DOWN(end); |
250 | ||
f765090a | 251 | /* head if not big page alignment ? */ |
1829ae9a | 252 | pfn = start_pfn = PFN_DOWN(start); |
f765090a PE |
253 | #ifdef CONFIG_X86_32 |
254 | /* | |
255 | * Don't use a large page for the first 2/4MB of memory | |
256 | * because there are often fixed size MTRRs in there | |
257 | * and overlapping MTRRs into large pages can cause | |
258 | * slowdowns. | |
259 | */ | |
1829ae9a | 260 | if (pfn == 0) |
84d77001 | 261 | end_pfn = PFN_DOWN(PMD_SIZE); |
f765090a | 262 | else |
1829ae9a | 263 | end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 264 | #else /* CONFIG_X86_64 */ |
1829ae9a | 265 | end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 266 | #endif |
2e8059ed YL |
267 | if (end_pfn > limit_pfn) |
268 | end_pfn = limit_pfn; | |
f765090a PE |
269 | if (start_pfn < end_pfn) { |
270 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); | |
1829ae9a | 271 | pfn = end_pfn; |
f765090a PE |
272 | } |
273 | ||
274 | /* big page (2M) range */ | |
1829ae9a | 275 | start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 276 | #ifdef CONFIG_X86_32 |
2e8059ed | 277 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); |
f765090a | 278 | #else /* CONFIG_X86_64 */ |
1829ae9a | 279 | end_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE)); |
2e8059ed YL |
280 | if (end_pfn > round_down(limit_pfn, PFN_DOWN(PMD_SIZE))) |
281 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); | |
f765090a PE |
282 | #endif |
283 | ||
284 | if (start_pfn < end_pfn) { | |
285 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, | |
286 | page_size_mask & (1<<PG_LEVEL_2M)); | |
1829ae9a | 287 | pfn = end_pfn; |
f765090a PE |
288 | } |
289 | ||
290 | #ifdef CONFIG_X86_64 | |
291 | /* big page (1G) range */ | |
1829ae9a | 292 | start_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE)); |
2e8059ed | 293 | end_pfn = round_down(limit_pfn, PFN_DOWN(PUD_SIZE)); |
f765090a PE |
294 | if (start_pfn < end_pfn) { |
295 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, | |
296 | page_size_mask & | |
297 | ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G))); | |
1829ae9a | 298 | pfn = end_pfn; |
f765090a PE |
299 | } |
300 | ||
301 | /* tail is not big page (1G) alignment */ | |
1829ae9a | 302 | start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); |
2e8059ed | 303 | end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); |
f765090a PE |
304 | if (start_pfn < end_pfn) { |
305 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, | |
306 | page_size_mask & (1<<PG_LEVEL_2M)); | |
1829ae9a | 307 | pfn = end_pfn; |
f765090a PE |
308 | } |
309 | #endif | |
310 | ||
311 | /* tail is not big page (2M) alignment */ | |
1829ae9a | 312 | start_pfn = pfn; |
2e8059ed | 313 | end_pfn = limit_pfn; |
f765090a PE |
314 | nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); |
315 | ||
7de3d66b YL |
316 | if (!after_bootmem) |
317 | adjust_range_page_size_mask(mr, nr_range); | |
318 | ||
f765090a PE |
319 | /* try to merge same page size and continuous */ |
320 | for (i = 0; nr_range > 1 && i < nr_range - 1; i++) { | |
321 | unsigned long old_start; | |
322 | if (mr[i].end != mr[i+1].start || | |
323 | mr[i].page_size_mask != mr[i+1].page_size_mask) | |
324 | continue; | |
325 | /* move it */ | |
326 | old_start = mr[i].start; | |
327 | memmove(&mr[i], &mr[i+1], | |
328 | (nr_range - 1 - i) * sizeof(struct map_range)); | |
329 | mr[i--].start = old_start; | |
330 | nr_range--; | |
331 | } | |
332 | ||
333 | for (i = 0; i < nr_range; i++) | |
365811d6 BH |
334 | printk(KERN_DEBUG " [mem %#010lx-%#010lx] page %s\n", |
335 | mr[i].start, mr[i].end - 1, | |
f765090a PE |
336 | (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":( |
337 | (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k")); | |
338 | ||
4e33e065 YL |
339 | return nr_range; |
340 | } | |
341 | ||
0e691cf8 YL |
342 | struct range pfn_mapped[E820_X_MAX]; |
343 | int nr_pfn_mapped; | |
66520ebc JS |
344 | |
345 | static void add_pfn_range_mapped(unsigned long start_pfn, unsigned long end_pfn) | |
346 | { | |
347 | nr_pfn_mapped = add_range_with_merge(pfn_mapped, E820_X_MAX, | |
348 | nr_pfn_mapped, start_pfn, end_pfn); | |
349 | nr_pfn_mapped = clean_sort_range(pfn_mapped, E820_X_MAX); | |
350 | ||
351 | max_pfn_mapped = max(max_pfn_mapped, end_pfn); | |
352 | ||
353 | if (start_pfn < (1UL<<(32-PAGE_SHIFT))) | |
354 | max_low_pfn_mapped = max(max_low_pfn_mapped, | |
355 | min(end_pfn, 1UL<<(32-PAGE_SHIFT))); | |
356 | } | |
357 | ||
358 | bool pfn_range_is_mapped(unsigned long start_pfn, unsigned long end_pfn) | |
359 | { | |
360 | int i; | |
361 | ||
362 | for (i = 0; i < nr_pfn_mapped; i++) | |
363 | if ((start_pfn >= pfn_mapped[i].start) && | |
364 | (end_pfn <= pfn_mapped[i].end)) | |
365 | return true; | |
366 | ||
367 | return false; | |
368 | } | |
369 | ||
4e33e065 YL |
370 | /* |
371 | * Setup the direct mapping of the physical memory at PAGE_OFFSET. | |
372 | * This runs before bootmem is initialized and gets pages directly from | |
373 | * the physical memory. To access them they are temporarily mapped. | |
374 | */ | |
375 | unsigned long __init_refok init_memory_mapping(unsigned long start, | |
376 | unsigned long end) | |
377 | { | |
378 | struct map_range mr[NR_RANGE_MR]; | |
379 | unsigned long ret = 0; | |
380 | int nr_range, i; | |
381 | ||
382 | pr_info("init_memory_mapping: [mem %#010lx-%#010lx]\n", | |
383 | start, end - 1); | |
384 | ||
385 | memset(mr, 0, sizeof(mr)); | |
386 | nr_range = split_mem_range(mr, 0, start, end); | |
387 | ||
f765090a PE |
388 | for (i = 0; i < nr_range; i++) |
389 | ret = kernel_physical_mapping_init(mr[i].start, mr[i].end, | |
390 | mr[i].page_size_mask); | |
f765090a | 391 | |
66520ebc JS |
392 | add_pfn_range_mapped(start >> PAGE_SHIFT, ret >> PAGE_SHIFT); |
393 | ||
c14fa0b6 YL |
394 | return ret >> PAGE_SHIFT; |
395 | } | |
396 | ||
66520ebc | 397 | /* |
cf8b166d ZY |
398 | * We need to iterate through the E820 memory map and create direct mappings |
399 | * for only E820_RAM and E820_KERN_RESERVED regions. We cannot simply | |
400 | * create direct mappings for all pfns from [0 to max_low_pfn) and | |
401 | * [4GB to max_pfn) because of possible memory holes in high addresses | |
402 | * that cannot be marked as UC by fixed/variable range MTRRs. | |
403 | * Depending on the alignment of E820 ranges, this may possibly result | |
404 | * in using smaller size (i.e. 4K instead of 2M or 1G) page tables. | |
405 | * | |
406 | * init_mem_mapping() calls init_range_memory_mapping() with big range. | |
407 | * That range would have hole in the middle or ends, and only ram parts | |
408 | * will be mapped in init_range_memory_mapping(). | |
66520ebc | 409 | */ |
8d57470d | 410 | static unsigned long __init init_range_memory_mapping( |
b8fd39c0 YL |
411 | unsigned long r_start, |
412 | unsigned long r_end) | |
66520ebc JS |
413 | { |
414 | unsigned long start_pfn, end_pfn; | |
8d57470d | 415 | unsigned long mapped_ram_size = 0; |
66520ebc JS |
416 | int i; |
417 | ||
66520ebc | 418 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) { |
b8fd39c0 YL |
419 | u64 start = clamp_val(PFN_PHYS(start_pfn), r_start, r_end); |
420 | u64 end = clamp_val(PFN_PHYS(end_pfn), r_start, r_end); | |
421 | if (start >= end) | |
66520ebc JS |
422 | continue; |
423 | ||
c9b3234a YL |
424 | /* |
425 | * if it is overlapping with brk pgt, we need to | |
426 | * alloc pgt buf from memblock instead. | |
427 | */ | |
428 | can_use_brk_pgt = max(start, (u64)pgt_buf_end<<PAGE_SHIFT) >= | |
429 | min(end, (u64)pgt_buf_top<<PAGE_SHIFT); | |
f763ad1d | 430 | init_memory_mapping(start, end); |
8d57470d | 431 | mapped_ram_size += end - start; |
c9b3234a | 432 | can_use_brk_pgt = true; |
66520ebc | 433 | } |
8d57470d YL |
434 | |
435 | return mapped_ram_size; | |
66520ebc JS |
436 | } |
437 | ||
6979287a YL |
438 | static unsigned long __init get_new_step_size(unsigned long step_size) |
439 | { | |
440 | /* | |
132978b9 | 441 | * Initial mapped size is PMD_SIZE (2M). |
6979287a YL |
442 | * We can not set step_size to be PUD_SIZE (1G) yet. |
443 | * In worse case, when we cross the 1G boundary, and | |
444 | * PG_LEVEL_2M is not set, we will need 1+1+512 pages (2M + 8k) | |
132978b9 JB |
445 | * to map 1G range with PTE. Hence we use one less than the |
446 | * difference of page table level shifts. | |
6979287a | 447 | * |
132978b9 JB |
448 | * Don't need to worry about overflow in the top-down case, on 32bit, |
449 | * when step_size is 0, round_down() returns 0 for start, and that | |
450 | * turns it into 0x100000000ULL. | |
451 | * In the bottom-up case, round_up(x, 0) returns 0 though too, which | |
452 | * needs to be taken into consideration by the code below. | |
6979287a | 453 | */ |
132978b9 | 454 | return step_size << (PMD_SHIFT - PAGE_SHIFT - 1); |
6979287a YL |
455 | } |
456 | ||
0167d7d8 TC |
457 | /** |
458 | * memory_map_top_down - Map [map_start, map_end) top down | |
459 | * @map_start: start address of the target memory range | |
460 | * @map_end: end address of the target memory range | |
461 | * | |
462 | * This function will setup direct mapping for memory range | |
463 | * [map_start, map_end) in top-down. That said, the page tables | |
464 | * will be allocated at the end of the memory, and we map the | |
465 | * memory in top-down. | |
466 | */ | |
467 | static void __init memory_map_top_down(unsigned long map_start, | |
468 | unsigned long map_end) | |
c14fa0b6 | 469 | { |
0167d7d8 | 470 | unsigned long real_end, start, last_start; |
8d57470d YL |
471 | unsigned long step_size; |
472 | unsigned long addr; | |
473 | unsigned long mapped_ram_size = 0; | |
ab951937 | 474 | |
98e7a989 | 475 | /* xen has big range in reserved near end of ram, skip it at first.*/ |
0167d7d8 | 476 | addr = memblock_find_in_range(map_start, map_end, PMD_SIZE, PMD_SIZE); |
8d57470d YL |
477 | real_end = addr + PMD_SIZE; |
478 | ||
479 | /* step_size need to be small so pgt_buf from BRK could cover it */ | |
480 | step_size = PMD_SIZE; | |
481 | max_pfn_mapped = 0; /* will get exact value next */ | |
482 | min_pfn_mapped = real_end >> PAGE_SHIFT; | |
483 | last_start = start = real_end; | |
cf8b166d ZY |
484 | |
485 | /* | |
486 | * We start from the top (end of memory) and go to the bottom. | |
487 | * The memblock_find_in_range() gets us a block of RAM from the | |
488 | * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages | |
489 | * for page table. | |
490 | */ | |
0167d7d8 | 491 | while (last_start > map_start) { |
8d57470d YL |
492 | if (last_start > step_size) { |
493 | start = round_down(last_start - 1, step_size); | |
0167d7d8 TC |
494 | if (start < map_start) |
495 | start = map_start; | |
8d57470d | 496 | } else |
0167d7d8 | 497 | start = map_start; |
132978b9 | 498 | mapped_ram_size += init_range_memory_mapping(start, |
8d57470d YL |
499 | last_start); |
500 | last_start = start; | |
501 | min_pfn_mapped = last_start >> PAGE_SHIFT; | |
132978b9 | 502 | if (mapped_ram_size >= step_size) |
6979287a | 503 | step_size = get_new_step_size(step_size); |
8d57470d YL |
504 | } |
505 | ||
0167d7d8 TC |
506 | if (real_end < map_end) |
507 | init_range_memory_mapping(real_end, map_end); | |
508 | } | |
509 | ||
b959ed6c TC |
510 | /** |
511 | * memory_map_bottom_up - Map [map_start, map_end) bottom up | |
512 | * @map_start: start address of the target memory range | |
513 | * @map_end: end address of the target memory range | |
514 | * | |
515 | * This function will setup direct mapping for memory range | |
516 | * [map_start, map_end) in bottom-up. Since we have limited the | |
517 | * bottom-up allocation above the kernel, the page tables will | |
518 | * be allocated just above the kernel and we map the memory | |
519 | * in [map_start, map_end) in bottom-up. | |
520 | */ | |
521 | static void __init memory_map_bottom_up(unsigned long map_start, | |
522 | unsigned long map_end) | |
523 | { | |
132978b9 | 524 | unsigned long next, start; |
b959ed6c TC |
525 | unsigned long mapped_ram_size = 0; |
526 | /* step_size need to be small so pgt_buf from BRK could cover it */ | |
527 | unsigned long step_size = PMD_SIZE; | |
528 | ||
529 | start = map_start; | |
530 | min_pfn_mapped = start >> PAGE_SHIFT; | |
531 | ||
532 | /* | |
533 | * We start from the bottom (@map_start) and go to the top (@map_end). | |
534 | * The memblock_find_in_range() gets us a block of RAM from the | |
535 | * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages | |
536 | * for page table. | |
537 | */ | |
538 | while (start < map_end) { | |
132978b9 | 539 | if (step_size && map_end - start > step_size) { |
b959ed6c TC |
540 | next = round_up(start + 1, step_size); |
541 | if (next > map_end) | |
542 | next = map_end; | |
132978b9 | 543 | } else { |
b959ed6c | 544 | next = map_end; |
132978b9 | 545 | } |
b959ed6c | 546 | |
132978b9 | 547 | mapped_ram_size += init_range_memory_mapping(start, next); |
b959ed6c TC |
548 | start = next; |
549 | ||
132978b9 | 550 | if (mapped_ram_size >= step_size) |
b959ed6c | 551 | step_size = get_new_step_size(step_size); |
b959ed6c TC |
552 | } |
553 | } | |
554 | ||
0167d7d8 TC |
555 | void __init init_mem_mapping(void) |
556 | { | |
557 | unsigned long end; | |
558 | ||
559 | probe_page_size_mask(); | |
560 | ||
561 | #ifdef CONFIG_X86_64 | |
562 | end = max_pfn << PAGE_SHIFT; | |
563 | #else | |
564 | end = max_low_pfn << PAGE_SHIFT; | |
565 | #endif | |
566 | ||
567 | /* the ISA range is always mapped regardless of memory holes */ | |
568 | init_memory_mapping(0, ISA_END_ADDRESS); | |
569 | ||
b959ed6c TC |
570 | /* |
571 | * If the allocation is in bottom-up direction, we setup direct mapping | |
572 | * in bottom-up, otherwise we setup direct mapping in top-down. | |
573 | */ | |
574 | if (memblock_bottom_up()) { | |
575 | unsigned long kernel_end = __pa_symbol(_end); | |
576 | ||
577 | /* | |
578 | * we need two separate calls here. This is because we want to | |
579 | * allocate page tables above the kernel. So we first map | |
580 | * [kernel_end, end) to make memory above the kernel be mapped | |
581 | * as soon as possible. And then use page tables allocated above | |
582 | * the kernel to map [ISA_END_ADDRESS, kernel_end). | |
583 | */ | |
584 | memory_map_bottom_up(kernel_end, end); | |
585 | memory_map_bottom_up(ISA_END_ADDRESS, kernel_end); | |
586 | } else { | |
587 | memory_map_top_down(ISA_END_ADDRESS, end); | |
588 | } | |
8d57470d | 589 | |
f763ad1d YL |
590 | #ifdef CONFIG_X86_64 |
591 | if (max_pfn > max_low_pfn) { | |
592 | /* can we preseve max_low_pfn ?*/ | |
593 | max_low_pfn = max_pfn; | |
594 | } | |
719272c4 YL |
595 | #else |
596 | early_ioremap_page_table_range_init(); | |
8170e6be PA |
597 | #endif |
598 | ||
719272c4 YL |
599 | load_cr3(swapper_pg_dir); |
600 | __flush_tlb_all(); | |
719272c4 | 601 | |
c14fa0b6 | 602 | early_memtest(0, max_pfn_mapped << PAGE_SHIFT); |
22ddfcaa | 603 | } |
e5b2bb55 | 604 | |
540aca06 PE |
605 | /* |
606 | * devmem_is_allowed() checks to see if /dev/mem access to a certain address | |
607 | * is valid. The argument is a physical page number. | |
608 | * | |
609 | * | |
610 | * On x86, access has to be given to the first megabyte of ram because that area | |
611 | * contains bios code and data regions used by X and dosemu and similar apps. | |
612 | * Access has to be given to non-kernel-ram areas as well, these contain the PCI | |
613 | * mmio resources as well as potential bios/acpi data regions. | |
614 | */ | |
615 | int devmem_is_allowed(unsigned long pagenr) | |
616 | { | |
73e8f3d7 | 617 | if (pagenr < 256) |
540aca06 PE |
618 | return 1; |
619 | if (iomem_is_exclusive(pagenr << PAGE_SHIFT)) | |
620 | return 0; | |
621 | if (!page_is_ram(pagenr)) | |
622 | return 1; | |
623 | return 0; | |
624 | } | |
625 | ||
e5b2bb55 PE |
626 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
627 | { | |
c967da6a | 628 | unsigned long begin_aligned, end_aligned; |
e5b2bb55 | 629 | |
c967da6a YL |
630 | /* Make sure boundaries are page aligned */ |
631 | begin_aligned = PAGE_ALIGN(begin); | |
632 | end_aligned = end & PAGE_MASK; | |
633 | ||
634 | if (WARN_ON(begin_aligned != begin || end_aligned != end)) { | |
635 | begin = begin_aligned; | |
636 | end = end_aligned; | |
637 | } | |
638 | ||
639 | if (begin >= end) | |
e5b2bb55 PE |
640 | return; |
641 | ||
642 | /* | |
643 | * If debugging page accesses then do not free this memory but | |
644 | * mark them not present - any buggy init-section access will | |
645 | * create a kernel page fault: | |
646 | */ | |
647 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
365811d6 BH |
648 | printk(KERN_INFO "debug: unmapping init [mem %#010lx-%#010lx]\n", |
649 | begin, end - 1); | |
e5b2bb55 PE |
650 | set_memory_np(begin, (end - begin) >> PAGE_SHIFT); |
651 | #else | |
652 | /* | |
653 | * We just marked the kernel text read only above, now that | |
654 | * we are going to free part of that, we need to make that | |
5bd5a452 | 655 | * writeable and non-executable first. |
e5b2bb55 | 656 | */ |
5bd5a452 | 657 | set_memory_nx(begin, (end - begin) >> PAGE_SHIFT); |
e5b2bb55 PE |
658 | set_memory_rw(begin, (end - begin) >> PAGE_SHIFT); |
659 | ||
c88442ec | 660 | free_reserved_area((void *)begin, (void *)end, POISON_FREE_INITMEM, what); |
e5b2bb55 PE |
661 | #endif |
662 | } | |
663 | ||
664 | void free_initmem(void) | |
665 | { | |
c88442ec | 666 | free_init_pages("unused kernel", |
e5b2bb55 PE |
667 | (unsigned long)(&__init_begin), |
668 | (unsigned long)(&__init_end)); | |
669 | } | |
731ddea6 PE |
670 | |
671 | #ifdef CONFIG_BLK_DEV_INITRD | |
0d26d1d8 | 672 | void __init free_initrd_mem(unsigned long start, unsigned long end) |
731ddea6 | 673 | { |
cd745be8 FY |
674 | #ifdef CONFIG_MICROCODE_EARLY |
675 | /* | |
676 | * Remember, initrd memory may contain microcode or other useful things. | |
677 | * Before we lose initrd mem, we need to find a place to hold them | |
678 | * now that normal virtual memory is enabled. | |
679 | */ | |
680 | save_microcode_in_initrd(); | |
681 | #endif | |
682 | ||
c967da6a YL |
683 | /* |
684 | * end could be not aligned, and We can not align that, | |
685 | * decompresser could be confused by aligned initrd_end | |
686 | * We already reserve the end partial page before in | |
687 | * - i386_start_kernel() | |
688 | * - x86_64_start_kernel() | |
689 | * - relocate_initrd() | |
690 | * So here We can do PAGE_ALIGN() safely to get partial page to be freed | |
691 | */ | |
c88442ec | 692 | free_init_pages("initrd", start, PAGE_ALIGN(end)); |
731ddea6 PE |
693 | } |
694 | #endif | |
17623915 PE |
695 | |
696 | void __init zone_sizes_init(void) | |
697 | { | |
698 | unsigned long max_zone_pfns[MAX_NR_ZONES]; | |
699 | ||
700 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); | |
701 | ||
702 | #ifdef CONFIG_ZONE_DMA | |
c072b90c | 703 | max_zone_pfns[ZONE_DMA] = min(MAX_DMA_PFN, max_low_pfn); |
17623915 PE |
704 | #endif |
705 | #ifdef CONFIG_ZONE_DMA32 | |
c072b90c | 706 | max_zone_pfns[ZONE_DMA32] = min(MAX_DMA32_PFN, max_low_pfn); |
17623915 PE |
707 | #endif |
708 | max_zone_pfns[ZONE_NORMAL] = max_low_pfn; | |
709 | #ifdef CONFIG_HIGHMEM | |
710 | max_zone_pfns[ZONE_HIGHMEM] = max_pfn; | |
711 | #endif | |
712 | ||
713 | free_area_init_nodes(max_zone_pfns); | |
714 | } | |
715 | ||
bd809af1 JG |
716 | void update_cache_mode_entry(unsigned entry, enum page_cache_mode cache) |
717 | { | |
718 | /* entry 0 MUST be WB (hardwired to speed up translations) */ | |
719 | BUG_ON(!entry && cache != _PAGE_CACHE_MODE_WB); | |
720 | ||
721 | __cachemode2pte_tbl[cache] = __cm_idx2pte(entry); | |
722 | __pte2cachemode_tbl[entry] = cache; | |
723 | } |