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