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
20 unsigned long __initdata pgt_buf_start;
21 unsigned long __meminitdata pgt_buf_end;
22 unsigned long __meminitdata pgt_buf_top;
23
24 int after_bootmem;
25
26 int direct_gbpages
27 #ifdef CONFIG_DIRECT_GBPAGES
28 = 1
29 #endif
30 ;
31
32 struct map_range {
33 unsigned long start;
34 unsigned long end;
35 unsigned page_size_mask;
36 };
37
38 /*
39 * First calculate space needed for kernel direct mapping page tables to cover
40 * mr[0].start to mr[nr_range - 1].end, while accounting for possible 2M and 1GB
41 * pages. Then find enough contiguous space for those page tables.
42 */
find_early_table_space(struct map_range * mr,int nr_range)43 static void __init find_early_table_space(struct map_range *mr, int nr_range)
44 {
45 int i;
46 unsigned long puds = 0, pmds = 0, ptes = 0, tables;
47 unsigned long start = 0, good_end;
48 unsigned long pgd_extra = 0;
49 phys_addr_t base;
50
51 for (i = 0; i < nr_range; i++) {
52 unsigned long range, extra;
53
54 if ((mr[i].end >> PGDIR_SHIFT) - (mr[i].start >> PGDIR_SHIFT))
55 pgd_extra++;
56
57 range = mr[i].end - mr[i].start;
58 puds += (range + PUD_SIZE - 1) >> PUD_SHIFT;
59
60 if (mr[i].page_size_mask & (1 << PG_LEVEL_1G)) {
61 extra = range - ((range >> PUD_SHIFT) << PUD_SHIFT);
62 pmds += (extra + PMD_SIZE - 1) >> PMD_SHIFT;
63 } else {
64 pmds += (range + PMD_SIZE - 1) >> PMD_SHIFT;
65 }
66
67 if (mr[i].page_size_mask & (1 << PG_LEVEL_2M)) {
68 extra = range - ((range >> PMD_SHIFT) << PMD_SHIFT);
69 #ifdef CONFIG_X86_32
70 extra += PMD_SIZE;
71 #endif
72 ptes += (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
73 } else {
74 ptes += (range + PAGE_SIZE - 1) >> PAGE_SHIFT;
75 }
76 }
77
78 tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
79 tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
80 tables += roundup(ptes * sizeof(pte_t), PAGE_SIZE);
81 tables += (pgd_extra * PAGE_SIZE);
82
83 #ifdef CONFIG_X86_32
84 /* for fixmap */
85 tables += roundup(__end_of_fixed_addresses * sizeof(pte_t), PAGE_SIZE);
86 #endif
87 good_end = max_pfn_mapped << PAGE_SHIFT;
88
89 base = memblock_find_in_range(start, good_end, tables, PAGE_SIZE);
90 if (!base)
91 panic("Cannot find space for the kernel page tables");
92
93 pgt_buf_start = base >> PAGE_SHIFT;
94 pgt_buf_end = pgt_buf_start;
95 pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT);
96
97 printk(KERN_DEBUG "kernel direct mapping tables up to %#lx @ [mem %#010lx-%#010lx]\n",
98 mr[nr_range - 1].end - 1, pgt_buf_start << PAGE_SHIFT,
99 (pgt_buf_top << PAGE_SHIFT) - 1);
100 }
101
native_pagetable_reserve(u64 start,u64 end)102 void __init native_pagetable_reserve(u64 start, u64 end)
103 {
104 memblock_reserve(start, end - start);
105 }
106
107 #ifdef CONFIG_X86_32
108 #define NR_RANGE_MR 3
109 #else /* CONFIG_X86_64 */
110 #define NR_RANGE_MR 5
111 #endif
112
save_mr(struct map_range * mr,int nr_range,unsigned long start_pfn,unsigned long end_pfn,unsigned long page_size_mask)113 static int __meminit save_mr(struct map_range *mr, int nr_range,
114 unsigned long start_pfn, unsigned long end_pfn,
115 unsigned long page_size_mask)
116 {
117 if (start_pfn < end_pfn) {
118 if (nr_range >= NR_RANGE_MR)
119 panic("run out of range for init_memory_mapping\n");
120 mr[nr_range].start = start_pfn<<PAGE_SHIFT;
121 mr[nr_range].end = end_pfn<<PAGE_SHIFT;
122 mr[nr_range].page_size_mask = page_size_mask;
123 nr_range++;
124 }
125
126 return nr_range;
127 }
128
129 /*
130 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
131 * This runs before bootmem is initialized and gets pages directly from
132 * the physical memory. To access them they are temporarily mapped.
133 */
init_memory_mapping(unsigned long start,unsigned long end)134 unsigned long __init_refok init_memory_mapping(unsigned long start,
135 unsigned long end)
136 {
137 unsigned long page_size_mask = 0;
138 unsigned long start_pfn, end_pfn;
139 unsigned long ret = 0;
140 unsigned long pos;
141
142 struct map_range mr[NR_RANGE_MR];
143 int nr_range, i;
144 int use_pse, use_gbpages;
145
146 printk(KERN_INFO "init_memory_mapping: %016lx-%016lx\n", start, end);
147
148 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
149 /*
150 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
151 * This will simplify cpa(), which otherwise needs to support splitting
152 * large pages into small in interrupt context, etc.
153 */
154 use_pse = use_gbpages = 0;
155 #else
156 use_pse = cpu_has_pse;
157 use_gbpages = direct_gbpages;
158 #endif
159
160 /* Enable PSE if available */
161 if (cpu_has_pse)
162 set_in_cr4(X86_CR4_PSE);
163
164 /* Enable PGE if available */
165 if (cpu_has_pge) {
166 set_in_cr4(X86_CR4_PGE);
167 __supported_pte_mask |= _PAGE_GLOBAL;
168 }
169
170 if (use_gbpages)
171 page_size_mask |= 1 << PG_LEVEL_1G;
172 if (use_pse)
173 page_size_mask |= 1 << PG_LEVEL_2M;
174
175 memset(mr, 0, sizeof(mr));
176 nr_range = 0;
177
178 /* head if not big page alignment ? */
179 start_pfn = start >> PAGE_SHIFT;
180 pos = start_pfn << PAGE_SHIFT;
181 #ifdef CONFIG_X86_32
182 /*
183 * Don't use a large page for the first 2/4MB of memory
184 * because there are often fixed size MTRRs in there
185 * and overlapping MTRRs into large pages can cause
186 * slowdowns.
187 */
188 if (pos == 0)
189 end_pfn = 1<<(PMD_SHIFT - PAGE_SHIFT);
190 else
191 end_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
192 << (PMD_SHIFT - PAGE_SHIFT);
193 #else /* CONFIG_X86_64 */
194 end_pfn = ((pos + (PMD_SIZE - 1)) >> PMD_SHIFT)
195 << (PMD_SHIFT - PAGE_SHIFT);
196 #endif
197 if (end_pfn > (end >> PAGE_SHIFT))
198 end_pfn = end >> PAGE_SHIFT;
199 if (start_pfn < end_pfn) {
200 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
201 pos = end_pfn << PAGE_SHIFT;
202 }
203
204 /* big page (2M) range */
205 start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
206 << (PMD_SHIFT - PAGE_SHIFT);
207 #ifdef CONFIG_X86_32
208 end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
209 #else /* CONFIG_X86_64 */
210 end_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
211 << (PUD_SHIFT - PAGE_SHIFT);
212 if (end_pfn > ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT)))
213 end_pfn = ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT));
214 #endif
215
216 if (start_pfn < end_pfn) {
217 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
218 page_size_mask & (1<<PG_LEVEL_2M));
219 pos = end_pfn << PAGE_SHIFT;
220 }
221
222 #ifdef CONFIG_X86_64
223 /* big page (1G) range */
224 start_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
225 << (PUD_SHIFT - PAGE_SHIFT);
226 end_pfn = (end >> PUD_SHIFT) << (PUD_SHIFT - PAGE_SHIFT);
227 if (start_pfn < end_pfn) {
228 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
229 page_size_mask &
230 ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G)));
231 pos = end_pfn << PAGE_SHIFT;
232 }
233
234 /* tail is not big page (1G) alignment */
235 start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
236 << (PMD_SHIFT - PAGE_SHIFT);
237 end_pfn = (end >> PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
238 if (start_pfn < end_pfn) {
239 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
240 page_size_mask & (1<<PG_LEVEL_2M));
241 pos = end_pfn << PAGE_SHIFT;
242 }
243 #endif
244
245 /* tail is not big page (2M) alignment */
246 start_pfn = pos>>PAGE_SHIFT;
247 end_pfn = end>>PAGE_SHIFT;
248 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
249
250 /* try to merge same page size and continuous */
251 for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
252 unsigned long old_start;
253 if (mr[i].end != mr[i+1].start ||
254 mr[i].page_size_mask != mr[i+1].page_size_mask)
255 continue;
256 /* move it */
257 old_start = mr[i].start;
258 memmove(&mr[i], &mr[i+1],
259 (nr_range - 1 - i) * sizeof(struct map_range));
260 mr[i--].start = old_start;
261 nr_range--;
262 }
263
264 for (i = 0; i < nr_range; i++)
265 printk(KERN_DEBUG " %010lx - %010lx page %s\n",
266 mr[i].start, mr[i].end,
267 (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":(
268 (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k"));
269
270 /*
271 * Find space for the kernel direct mapping tables.
272 *
273 * Later we should allocate these tables in the local node of the
274 * memory mapped. Unfortunately this is done currently before the
275 * nodes are discovered.
276 */
277 if (!after_bootmem)
278 find_early_table_space(mr, nr_range);
279
280 for (i = 0; i < nr_range; i++)
281 ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
282 mr[i].page_size_mask);
283
284 #ifdef CONFIG_X86_32
285 early_ioremap_page_table_range_init();
286
287 load_cr3(swapper_pg_dir);
288 #endif
289
290 __flush_tlb_all();
291
292 /*
293 * Reserve the kernel pagetable pages we used (pgt_buf_start -
294 * pgt_buf_end) and free the other ones (pgt_buf_end - pgt_buf_top)
295 * so that they can be reused for other purposes.
296 *
297 * On native it just means calling memblock_reserve, on Xen it also
298 * means marking RW the pagetable pages that we allocated before
299 * but that haven't been used.
300 *
301 * In fact on xen we mark RO the whole range pgt_buf_start -
302 * pgt_buf_top, because we have to make sure that when
303 * init_memory_mapping reaches the pagetable pages area, it maps
304 * RO all the pagetable pages, including the ones that are beyond
305 * pgt_buf_end at that time.
306 */
307 if (!after_bootmem && pgt_buf_end > pgt_buf_start)
308 x86_init.mapping.pagetable_reserve(PFN_PHYS(pgt_buf_start),
309 PFN_PHYS(pgt_buf_end));
310
311 if (!after_bootmem)
312 early_memtest(start, end);
313
314 return ret >> PAGE_SHIFT;
315 }
316
317
318 /*
319 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
320 * is valid. The argument is a physical page number.
321 *
322 *
323 * On x86, access has to be given to the first megabyte of ram because that area
324 * contains bios code and data regions used by X and dosemu and similar apps.
325 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
326 * mmio resources as well as potential bios/acpi data regions.
327 */
devmem_is_allowed(unsigned long pagenr)328 int devmem_is_allowed(unsigned long pagenr)
329 {
330 if (pagenr <= 256)
331 return 1;
332 if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
333 return 0;
334 if (!page_is_ram(pagenr))
335 return 1;
336 return 0;
337 }
338
free_init_pages(char * what,unsigned long begin,unsigned long end)339 void free_init_pages(char *what, unsigned long begin, unsigned long end)
340 {
341 unsigned long addr;
342 unsigned long begin_aligned, end_aligned;
343
344 /* Make sure boundaries are page aligned */
345 begin_aligned = PAGE_ALIGN(begin);
346 end_aligned = end & PAGE_MASK;
347
348 if (WARN_ON(begin_aligned != begin || end_aligned != end)) {
349 begin = begin_aligned;
350 end = end_aligned;
351 }
352
353 if (begin >= end)
354 return;
355
356 addr = begin;
357
358 /*
359 * If debugging page accesses then do not free this memory but
360 * mark them not present - any buggy init-section access will
361 * create a kernel page fault:
362 */
363 #ifdef CONFIG_DEBUG_PAGEALLOC
364 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
365 begin, end);
366 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
367 #else
368 /*
369 * We just marked the kernel text read only above, now that
370 * we are going to free part of that, we need to make that
371 * writeable and non-executable first.
372 */
373 set_memory_nx(begin, (end - begin) >> PAGE_SHIFT);
374 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
375
376 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
377
378 for (; addr < end; addr += PAGE_SIZE) {
379 ClearPageReserved(virt_to_page(addr));
380 init_page_count(virt_to_page(addr));
381 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
382 free_page(addr);
383 totalram_pages++;
384 }
385 #endif
386 }
387
free_initmem(void)388 void free_initmem(void)
389 {
390 free_init_pages("unused kernel memory",
391 (unsigned long)(&__init_begin),
392 (unsigned long)(&__init_end));
393 }
394
395 #ifdef CONFIG_BLK_DEV_INITRD
free_initrd_mem(unsigned long start,unsigned long end)396 void free_initrd_mem(unsigned long start, unsigned long end)
397 {
398 /*
399 * end could be not aligned, and We can not align that,
400 * decompresser could be confused by aligned initrd_end
401 * We already reserve the end partial page before in
402 * - i386_start_kernel()
403 * - x86_64_start_kernel()
404 * - relocate_initrd()
405 * So here We can do PAGE_ALIGN() safely to get partial page to be freed
406 */
407 free_init_pages("initrd memory", start, PAGE_ALIGN(end));
408 }
409 #endif
410
zone_sizes_init(void)411 void __init zone_sizes_init(void)
412 {
413 unsigned long max_zone_pfns[MAX_NR_ZONES];
414
415 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
416
417 #ifdef CONFIG_ZONE_DMA
418 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
419 #endif
420 #ifdef CONFIG_ZONE_DMA32
421 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
422 #endif
423 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
424 #ifdef CONFIG_HIGHMEM
425 max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
426 #endif
427
428 free_area_init_nodes(max_zone_pfns);
429 }
430
431