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/*
31 * Tables translating between page_cache_type_t and pte encoding.
32 *
33 * Minimal supported modes are defined statically, they are modified
34 * during bootup if more supported cache modes are available.
35 *
36 *   Index into __cachemode2pte_tbl[] is the cachemode.
37 *
38 *   Index into __pte2cachemode_tbl[] are the caching attribute bits of the pte
39 *   (_PAGE_PWT, _PAGE_PCD, _PAGE_PAT) at index bit positions 0, 1, 2.
40 */
41uint16_t __cachemode2pte_tbl[_PAGE_CACHE_MODE_NUM] = {
42	[_PAGE_CACHE_MODE_WB      ]	= 0         | 0        ,
43	[_PAGE_CACHE_MODE_WC      ]	= _PAGE_PWT | 0        ,
44	[_PAGE_CACHE_MODE_UC_MINUS]	= 0         | _PAGE_PCD,
45	[_PAGE_CACHE_MODE_UC      ]	= _PAGE_PWT | _PAGE_PCD,
46	[_PAGE_CACHE_MODE_WT      ]	= 0         | _PAGE_PCD,
47	[_PAGE_CACHE_MODE_WP      ]	= 0         | _PAGE_PCD,
48};
49EXPORT_SYMBOL(__cachemode2pte_tbl);
50
51uint8_t __pte2cachemode_tbl[8] = {
52	[__pte2cm_idx( 0        | 0         | 0        )] = _PAGE_CACHE_MODE_WB,
53	[__pte2cm_idx(_PAGE_PWT | 0         | 0        )] = _PAGE_CACHE_MODE_WC,
54	[__pte2cm_idx( 0        | _PAGE_PCD | 0        )] = _PAGE_CACHE_MODE_UC_MINUS,
55	[__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | 0        )] = _PAGE_CACHE_MODE_UC,
56	[__pte2cm_idx( 0        | 0         | _PAGE_PAT)] = _PAGE_CACHE_MODE_WB,
57	[__pte2cm_idx(_PAGE_PWT | 0         | _PAGE_PAT)] = _PAGE_CACHE_MODE_WC,
58	[__pte2cm_idx(0         | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS,
59	[__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC,
60};
61EXPORT_SYMBOL(__pte2cachemode_tbl);
62
63static unsigned long __initdata pgt_buf_start;
64static unsigned long __initdata pgt_buf_end;
65static unsigned long __initdata pgt_buf_top;
66
67static unsigned long min_pfn_mapped;
68
69static bool __initdata can_use_brk_pgt = true;
70
71/*
72 * Pages returned are already directly mapped.
73 *
74 * Changing that is likely to break Xen, see commit:
75 *
76 *    279b706 x86,xen: introduce x86_init.mapping.pagetable_reserve
77 *
78 * for detailed information.
79 */
80__ref void *alloc_low_pages(unsigned int num)
81{
82	unsigned long pfn;
83	int i;
84
85	if (after_bootmem) {
86		unsigned int order;
87
88		order = get_order((unsigned long)num << PAGE_SHIFT);
89		return (void *)__get_free_pages(GFP_ATOMIC | __GFP_NOTRACK |
90						__GFP_ZERO, order);
91	}
92
93	if ((pgt_buf_end + num) > pgt_buf_top || !can_use_brk_pgt) {
94		unsigned long ret;
95		if (min_pfn_mapped >= max_pfn_mapped)
96			panic("alloc_low_pages: ran out of memory");
97		ret = memblock_find_in_range(min_pfn_mapped << PAGE_SHIFT,
98					max_pfn_mapped << PAGE_SHIFT,
99					PAGE_SIZE * num , PAGE_SIZE);
100		if (!ret)
101			panic("alloc_low_pages: can not alloc memory");
102		memblock_reserve(ret, PAGE_SIZE * num);
103		pfn = ret >> PAGE_SHIFT;
104	} else {
105		pfn = pgt_buf_end;
106		pgt_buf_end += num;
107		printk(KERN_DEBUG "BRK [%#010lx, %#010lx] PGTABLE\n",
108			pfn << PAGE_SHIFT, (pgt_buf_end << PAGE_SHIFT) - 1);
109	}
110
111	for (i = 0; i < num; i++) {
112		void *adr;
113
114		adr = __va((pfn + i) << PAGE_SHIFT);
115		clear_page(adr);
116	}
117
118	return __va(pfn << PAGE_SHIFT);
119}
120
121/* need 3 4k for initial PMD_SIZE,  3 4k for 0-ISA_END_ADDRESS */
122#define INIT_PGT_BUF_SIZE	(6 * PAGE_SIZE)
123RESERVE_BRK(early_pgt_alloc, INIT_PGT_BUF_SIZE);
124void  __init early_alloc_pgt_buf(void)
125{
126	unsigned long tables = INIT_PGT_BUF_SIZE;
127	phys_addr_t base;
128
129	base = __pa(extend_brk(tables, PAGE_SIZE));
130
131	pgt_buf_start = base >> PAGE_SHIFT;
132	pgt_buf_end = pgt_buf_start;
133	pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT);
134}
135
136int after_bootmem;
137
138early_param_on_off("gbpages", "nogbpages", direct_gbpages, CONFIG_X86_DIRECT_GBPAGES);
139
140struct map_range {
141	unsigned long start;
142	unsigned long end;
143	unsigned page_size_mask;
144};
145
146static int page_size_mask;
147
148static void __init probe_page_size_mask(void)
149{
150#if !defined(CONFIG_DEBUG_PAGEALLOC) && !defined(CONFIG_KMEMCHECK)
151	/*
152	 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
153	 * This will simplify cpa(), which otherwise needs to support splitting
154	 * large pages into small in interrupt context, etc.
155	 */
156	if (cpu_has_pse)
157		page_size_mask |= 1 << PG_LEVEL_2M;
158#endif
159
160	/* Enable PSE if available */
161	if (cpu_has_pse)
162		cr4_set_bits_and_update_boot(X86_CR4_PSE);
163
164	/* Enable PGE if available */
165	if (cpu_has_pge) {
166		cr4_set_bits_and_update_boot(X86_CR4_PGE);
167		__supported_pte_mask |= _PAGE_GLOBAL;
168	} else
169		__supported_pte_mask &= ~_PAGE_GLOBAL;
170
171	/* Enable 1 GB linear kernel mappings if available: */
172	if (direct_gbpages && cpu_has_gbpages) {
173		printk(KERN_INFO "Using GB pages for direct mapping\n");
174		page_size_mask |= 1 << PG_LEVEL_1G;
175	} else {
176		direct_gbpages = 0;
177	}
178}
179
180#ifdef CONFIG_X86_32
181#define NR_RANGE_MR 3
182#else /* CONFIG_X86_64 */
183#define NR_RANGE_MR 5
184#endif
185
186static int __meminit save_mr(struct map_range *mr, int nr_range,
187			     unsigned long start_pfn, unsigned long end_pfn,
188			     unsigned long page_size_mask)
189{
190	if (start_pfn < end_pfn) {
191		if (nr_range >= NR_RANGE_MR)
192			panic("run out of range for init_memory_mapping\n");
193		mr[nr_range].start = start_pfn<<PAGE_SHIFT;
194		mr[nr_range].end   = end_pfn<<PAGE_SHIFT;
195		mr[nr_range].page_size_mask = page_size_mask;
196		nr_range++;
197	}
198
199	return nr_range;
200}
201
202/*
203 * adjust the page_size_mask for small range to go with
204 *	big page size instead small one if nearby are ram too.
205 */
206static void __init_refok adjust_range_page_size_mask(struct map_range *mr,
207							 int nr_range)
208{
209	int i;
210
211	for (i = 0; i < nr_range; i++) {
212		if ((page_size_mask & (1<<PG_LEVEL_2M)) &&
213		    !(mr[i].page_size_mask & (1<<PG_LEVEL_2M))) {
214			unsigned long start = round_down(mr[i].start, PMD_SIZE);
215			unsigned long end = round_up(mr[i].end, PMD_SIZE);
216
217#ifdef CONFIG_X86_32
218			if ((end >> PAGE_SHIFT) > max_low_pfn)
219				continue;
220#endif
221
222			if (memblock_is_region_memory(start, end - start))
223				mr[i].page_size_mask |= 1<<PG_LEVEL_2M;
224		}
225		if ((page_size_mask & (1<<PG_LEVEL_1G)) &&
226		    !(mr[i].page_size_mask & (1<<PG_LEVEL_1G))) {
227			unsigned long start = round_down(mr[i].start, PUD_SIZE);
228			unsigned long end = round_up(mr[i].end, PUD_SIZE);
229
230			if (memblock_is_region_memory(start, end - start))
231				mr[i].page_size_mask |= 1<<PG_LEVEL_1G;
232		}
233	}
234}
235
236static const char *page_size_string(struct map_range *mr)
237{
238	static const char str_1g[] = "1G";
239	static const char str_2m[] = "2M";
240	static const char str_4m[] = "4M";
241	static const char str_4k[] = "4k";
242
243	if (mr->page_size_mask & (1<<PG_LEVEL_1G))
244		return str_1g;
245	/*
246	 * 32-bit without PAE has a 4M large page size.
247	 * PG_LEVEL_2M is misnamed, but we can at least
248	 * print out the right size in the string.
249	 */
250	if (IS_ENABLED(CONFIG_X86_32) &&
251	    !IS_ENABLED(CONFIG_X86_PAE) &&
252	    mr->page_size_mask & (1<<PG_LEVEL_2M))
253		return str_4m;
254
255	if (mr->page_size_mask & (1<<PG_LEVEL_2M))
256		return str_2m;
257
258	return str_4k;
259}
260
261static int __meminit split_mem_range(struct map_range *mr, int nr_range,
262				     unsigned long start,
263				     unsigned long end)
264{
265	unsigned long start_pfn, end_pfn, limit_pfn;
266	unsigned long pfn;
267	int i;
268
269	limit_pfn = PFN_DOWN(end);
270
271	/* head if not big page alignment ? */
272	pfn = start_pfn = PFN_DOWN(start);
273#ifdef CONFIG_X86_32
274	/*
275	 * Don't use a large page for the first 2/4MB of memory
276	 * because there are often fixed size MTRRs in there
277	 * and overlapping MTRRs into large pages can cause
278	 * slowdowns.
279	 */
280	if (pfn == 0)
281		end_pfn = PFN_DOWN(PMD_SIZE);
282	else
283		end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
284#else /* CONFIG_X86_64 */
285	end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
286#endif
287	if (end_pfn > limit_pfn)
288		end_pfn = limit_pfn;
289	if (start_pfn < end_pfn) {
290		nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
291		pfn = end_pfn;
292	}
293
294	/* big page (2M) range */
295	start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
296#ifdef CONFIG_X86_32
297	end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE));
298#else /* CONFIG_X86_64 */
299	end_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE));
300	if (end_pfn > round_down(limit_pfn, PFN_DOWN(PMD_SIZE)))
301		end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE));
302#endif
303
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));
307		pfn = end_pfn;
308	}
309
310#ifdef CONFIG_X86_64
311	/* big page (1G) range */
312	start_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE));
313	end_pfn = round_down(limit_pfn, PFN_DOWN(PUD_SIZE));
314	if (start_pfn < end_pfn) {
315		nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
316				page_size_mask &
317				 ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G)));
318		pfn = end_pfn;
319	}
320
321	/* tail is not big page (1G) alignment */
322	start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
323	end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE));
324	if (start_pfn < end_pfn) {
325		nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
326				page_size_mask & (1<<PG_LEVEL_2M));
327		pfn = end_pfn;
328	}
329#endif
330
331	/* tail is not big page (2M) alignment */
332	start_pfn = pfn;
333	end_pfn = limit_pfn;
334	nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
335
336	if (!after_bootmem)
337		adjust_range_page_size_mask(mr, nr_range);
338
339	/* try to merge same page size and continuous */
340	for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
341		unsigned long old_start;
342		if (mr[i].end != mr[i+1].start ||
343		    mr[i].page_size_mask != mr[i+1].page_size_mask)
344			continue;
345		/* move it */
346		old_start = mr[i].start;
347		memmove(&mr[i], &mr[i+1],
348			(nr_range - 1 - i) * sizeof(struct map_range));
349		mr[i--].start = old_start;
350		nr_range--;
351	}
352
353	for (i = 0; i < nr_range; i++)
354		printk(KERN_DEBUG " [mem %#010lx-%#010lx] page %s\n",
355				mr[i].start, mr[i].end - 1,
356				page_size_string(&mr[i]));
357
358	return nr_range;
359}
360
361struct range pfn_mapped[E820_X_MAX];
362int nr_pfn_mapped;
363
364static void add_pfn_range_mapped(unsigned long start_pfn, unsigned long end_pfn)
365{
366	nr_pfn_mapped = add_range_with_merge(pfn_mapped, E820_X_MAX,
367					     nr_pfn_mapped, start_pfn, end_pfn);
368	nr_pfn_mapped = clean_sort_range(pfn_mapped, E820_X_MAX);
369
370	max_pfn_mapped = max(max_pfn_mapped, end_pfn);
371
372	if (start_pfn < (1UL<<(32-PAGE_SHIFT)))
373		max_low_pfn_mapped = max(max_low_pfn_mapped,
374					 min(end_pfn, 1UL<<(32-PAGE_SHIFT)));
375}
376
377bool pfn_range_is_mapped(unsigned long start_pfn, unsigned long end_pfn)
378{
379	int i;
380
381	for (i = 0; i < nr_pfn_mapped; i++)
382		if ((start_pfn >= pfn_mapped[i].start) &&
383		    (end_pfn <= pfn_mapped[i].end))
384			return true;
385
386	return false;
387}
388
389/*
390 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
391 * This runs before bootmem is initialized and gets pages directly from
392 * the physical memory. To access them they are temporarily mapped.
393 */
394unsigned long __init_refok init_memory_mapping(unsigned long start,
395					       unsigned long end)
396{
397	struct map_range mr[NR_RANGE_MR];
398	unsigned long ret = 0;
399	int nr_range, i;
400
401	pr_info("init_memory_mapping: [mem %#010lx-%#010lx]\n",
402	       start, end - 1);
403
404	memset(mr, 0, sizeof(mr));
405	nr_range = split_mem_range(mr, 0, start, end);
406
407	for (i = 0; i < nr_range; i++)
408		ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
409						   mr[i].page_size_mask);
410
411	add_pfn_range_mapped(start >> PAGE_SHIFT, ret >> PAGE_SHIFT);
412
413	return ret >> PAGE_SHIFT;
414}
415
416/*
417 * We need to iterate through the E820 memory map and create direct mappings
418 * for only E820_RAM and E820_KERN_RESERVED regions. We cannot simply
419 * create direct mappings for all pfns from [0 to max_low_pfn) and
420 * [4GB to max_pfn) because of possible memory holes in high addresses
421 * that cannot be marked as UC by fixed/variable range MTRRs.
422 * Depending on the alignment of E820 ranges, this may possibly result
423 * in using smaller size (i.e. 4K instead of 2M or 1G) page tables.
424 *
425 * init_mem_mapping() calls init_range_memory_mapping() with big range.
426 * That range would have hole in the middle or ends, and only ram parts
427 * will be mapped in init_range_memory_mapping().
428 */
429static unsigned long __init init_range_memory_mapping(
430					   unsigned long r_start,
431					   unsigned long r_end)
432{
433	unsigned long start_pfn, end_pfn;
434	unsigned long mapped_ram_size = 0;
435	int i;
436
437	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
438		u64 start = clamp_val(PFN_PHYS(start_pfn), r_start, r_end);
439		u64 end = clamp_val(PFN_PHYS(end_pfn), r_start, r_end);
440		if (start >= end)
441			continue;
442
443		/*
444		 * if it is overlapping with brk pgt, we need to
445		 * alloc pgt buf from memblock instead.
446		 */
447		can_use_brk_pgt = max(start, (u64)pgt_buf_end<<PAGE_SHIFT) >=
448				    min(end, (u64)pgt_buf_top<<PAGE_SHIFT);
449		init_memory_mapping(start, end);
450		mapped_ram_size += end - start;
451		can_use_brk_pgt = true;
452	}
453
454	return mapped_ram_size;
455}
456
457static unsigned long __init get_new_step_size(unsigned long step_size)
458{
459	/*
460	 * Initial mapped size is PMD_SIZE (2M).
461	 * We can not set step_size to be PUD_SIZE (1G) yet.
462	 * In worse case, when we cross the 1G boundary, and
463	 * PG_LEVEL_2M is not set, we will need 1+1+512 pages (2M + 8k)
464	 * to map 1G range with PTE. Hence we use one less than the
465	 * difference of page table level shifts.
466	 *
467	 * Don't need to worry about overflow in the top-down case, on 32bit,
468	 * when step_size is 0, round_down() returns 0 for start, and that
469	 * turns it into 0x100000000ULL.
470	 * In the bottom-up case, round_up(x, 0) returns 0 though too, which
471	 * needs to be taken into consideration by the code below.
472	 */
473	return step_size << (PMD_SHIFT - PAGE_SHIFT - 1);
474}
475
476/**
477 * memory_map_top_down - Map [map_start, map_end) top down
478 * @map_start: start address of the target memory range
479 * @map_end: end address of the target memory range
480 *
481 * This function will setup direct mapping for memory range
482 * [map_start, map_end) in top-down. That said, the page tables
483 * will be allocated at the end of the memory, and we map the
484 * memory in top-down.
485 */
486static void __init memory_map_top_down(unsigned long map_start,
487				       unsigned long map_end)
488{
489	unsigned long real_end, start, last_start;
490	unsigned long step_size;
491	unsigned long addr;
492	unsigned long mapped_ram_size = 0;
493
494	/* xen has big range in reserved near end of ram, skip it at first.*/
495	addr = memblock_find_in_range(map_start, map_end, PMD_SIZE, PMD_SIZE);
496	real_end = addr + PMD_SIZE;
497
498	/* step_size need to be small so pgt_buf from BRK could cover it */
499	step_size = PMD_SIZE;
500	max_pfn_mapped = 0; /* will get exact value next */
501	min_pfn_mapped = real_end >> PAGE_SHIFT;
502	last_start = start = real_end;
503
504	/*
505	 * We start from the top (end of memory) and go to the bottom.
506	 * The memblock_find_in_range() gets us a block of RAM from the
507	 * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages
508	 * for page table.
509	 */
510	while (last_start > map_start) {
511		if (last_start > step_size) {
512			start = round_down(last_start - 1, step_size);
513			if (start < map_start)
514				start = map_start;
515		} else
516			start = map_start;
517		mapped_ram_size += init_range_memory_mapping(start,
518							last_start);
519		last_start = start;
520		min_pfn_mapped = last_start >> PAGE_SHIFT;
521		if (mapped_ram_size >= step_size)
522			step_size = get_new_step_size(step_size);
523	}
524
525	if (real_end < map_end)
526		init_range_memory_mapping(real_end, map_end);
527}
528
529/**
530 * memory_map_bottom_up - Map [map_start, map_end) bottom up
531 * @map_start: start address of the target memory range
532 * @map_end: end address of the target memory range
533 *
534 * This function will setup direct mapping for memory range
535 * [map_start, map_end) in bottom-up. Since we have limited the
536 * bottom-up allocation above the kernel, the page tables will
537 * be allocated just above the kernel and we map the memory
538 * in [map_start, map_end) in bottom-up.
539 */
540static void __init memory_map_bottom_up(unsigned long map_start,
541					unsigned long map_end)
542{
543	unsigned long next, start;
544	unsigned long mapped_ram_size = 0;
545	/* step_size need to be small so pgt_buf from BRK could cover it */
546	unsigned long step_size = PMD_SIZE;
547
548	start = map_start;
549	min_pfn_mapped = start >> PAGE_SHIFT;
550
551	/*
552	 * We start from the bottom (@map_start) and go to the top (@map_end).
553	 * The memblock_find_in_range() gets us a block of RAM from the
554	 * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages
555	 * for page table.
556	 */
557	while (start < map_end) {
558		if (step_size && map_end - start > step_size) {
559			next = round_up(start + 1, step_size);
560			if (next > map_end)
561				next = map_end;
562		} else {
563			next = map_end;
564		}
565
566		mapped_ram_size += init_range_memory_mapping(start, next);
567		start = next;
568
569		if (mapped_ram_size >= step_size)
570			step_size = get_new_step_size(step_size);
571	}
572}
573
574void __init init_mem_mapping(void)
575{
576	unsigned long end;
577
578	probe_page_size_mask();
579
580#ifdef CONFIG_X86_64
581	end = max_pfn << PAGE_SHIFT;
582#else
583	end = max_low_pfn << PAGE_SHIFT;
584#endif
585
586	/* the ISA range is always mapped regardless of memory holes */
587	init_memory_mapping(0, ISA_END_ADDRESS);
588
589	/*
590	 * If the allocation is in bottom-up direction, we setup direct mapping
591	 * in bottom-up, otherwise we setup direct mapping in top-down.
592	 */
593	if (memblock_bottom_up()) {
594		unsigned long kernel_end = __pa_symbol(_end);
595
596		/*
597		 * we need two separate calls here. This is because we want to
598		 * allocate page tables above the kernel. So we first map
599		 * [kernel_end, end) to make memory above the kernel be mapped
600		 * as soon as possible. And then use page tables allocated above
601		 * the kernel to map [ISA_END_ADDRESS, kernel_end).
602		 */
603		memory_map_bottom_up(kernel_end, end);
604		memory_map_bottom_up(ISA_END_ADDRESS, kernel_end);
605	} else {
606		memory_map_top_down(ISA_END_ADDRESS, end);
607	}
608
609#ifdef CONFIG_X86_64
610	if (max_pfn > max_low_pfn) {
611		/* can we preseve max_low_pfn ?*/
612		max_low_pfn = max_pfn;
613	}
614#else
615	early_ioremap_page_table_range_init();
616#endif
617
618	load_cr3(swapper_pg_dir);
619	__flush_tlb_all();
620
621	early_memtest(0, max_pfn_mapped << PAGE_SHIFT);
622}
623
624/*
625 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
626 * is valid. The argument is a physical page number.
627 *
628 *
629 * On x86, access has to be given to the first megabyte of ram because that area
630 * contains BIOS code and data regions used by X and dosemu and similar apps.
631 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
632 * mmio resources as well as potential bios/acpi data regions.
633 */
634int devmem_is_allowed(unsigned long pagenr)
635{
636	if (pagenr < 256)
637		return 1;
638	if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
639		return 0;
640	if (!page_is_ram(pagenr))
641		return 1;
642	return 0;
643}
644
645void free_init_pages(char *what, unsigned long begin, unsigned long end)
646{
647	unsigned long begin_aligned, end_aligned;
648
649	/* Make sure boundaries are page aligned */
650	begin_aligned = PAGE_ALIGN(begin);
651	end_aligned   = end & PAGE_MASK;
652
653	if (WARN_ON(begin_aligned != begin || end_aligned != end)) {
654		begin = begin_aligned;
655		end   = end_aligned;
656	}
657
658	if (begin >= end)
659		return;
660
661	/*
662	 * If debugging page accesses then do not free this memory but
663	 * mark them not present - any buggy init-section access will
664	 * create a kernel page fault:
665	 */
666#ifdef CONFIG_DEBUG_PAGEALLOC
667	printk(KERN_INFO "debug: unmapping init [mem %#010lx-%#010lx]\n",
668		begin, end - 1);
669	set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
670#else
671	/*
672	 * We just marked the kernel text read only above, now that
673	 * we are going to free part of that, we need to make that
674	 * writeable and non-executable first.
675	 */
676	set_memory_nx(begin, (end - begin) >> PAGE_SHIFT);
677	set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
678
679	free_reserved_area((void *)begin, (void *)end, POISON_FREE_INITMEM, what);
680#endif
681}
682
683void free_initmem(void)
684{
685	free_init_pages("unused kernel",
686			(unsigned long)(&__init_begin),
687			(unsigned long)(&__init_end));
688}
689
690#ifdef CONFIG_BLK_DEV_INITRD
691void __init free_initrd_mem(unsigned long start, unsigned long end)
692{
693#ifdef CONFIG_MICROCODE_EARLY
694	/*
695	 * Remember, initrd memory may contain microcode or other useful things.
696	 * Before we lose initrd mem, we need to find a place to hold them
697	 * now that normal virtual memory is enabled.
698	 */
699	save_microcode_in_initrd();
700#endif
701
702	/*
703	 * end could be not aligned, and We can not align that,
704	 * decompresser could be confused by aligned initrd_end
705	 * We already reserve the end partial page before in
706	 *   - i386_start_kernel()
707	 *   - x86_64_start_kernel()
708	 *   - relocate_initrd()
709	 * So here We can do PAGE_ALIGN() safely to get partial page to be freed
710	 */
711	free_init_pages("initrd", start, PAGE_ALIGN(end));
712}
713#endif
714
715void __init zone_sizes_init(void)
716{
717	unsigned long max_zone_pfns[MAX_NR_ZONES];
718
719	memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
720
721#ifdef CONFIG_ZONE_DMA
722	max_zone_pfns[ZONE_DMA]		= min(MAX_DMA_PFN, max_low_pfn);
723#endif
724#ifdef CONFIG_ZONE_DMA32
725	max_zone_pfns[ZONE_DMA32]	= min(MAX_DMA32_PFN, max_low_pfn);
726#endif
727	max_zone_pfns[ZONE_NORMAL]	= max_low_pfn;
728#ifdef CONFIG_HIGHMEM
729	max_zone_pfns[ZONE_HIGHMEM]	= max_pfn;
730#endif
731
732	free_area_init_nodes(max_zone_pfns);
733}
734
735DEFINE_PER_CPU_SHARED_ALIGNED(struct tlb_state, cpu_tlbstate) = {
736#ifdef CONFIG_SMP
737	.active_mm = &init_mm,
738	.state = 0,
739#endif
740	.cr4 = ~0UL,	/* fail hard if we screw up cr4 shadow initialization */
741};
742EXPORT_SYMBOL_GPL(cpu_tlbstate);
743
744void update_cache_mode_entry(unsigned entry, enum page_cache_mode cache)
745{
746	/* entry 0 MUST be WB (hardwired to speed up translations) */
747	BUG_ON(!entry && cache != _PAGE_CACHE_MODE_WB);
748
749	__cachemode2pte_tbl[cache] = __cm_idx2pte(entry);
750	__pte2cachemode_tbl[entry] = cache;
751}
752