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 */
8 #include <asm/cacheflush.h>
12 #include <asm/page_types.h>
13 #include <asm/sections.h>
14 #include <asm/setup.h>
15 #include <asm/tlbflush.h>
17 #include <asm/proto.h>
18 #include <asm/dma.h> /* for MAX_DMA_PFN */
20 unsigned long __initdata pgt_buf_start;
21 unsigned long __meminitdata pgt_buf_end;
22 unsigned long __meminitdata pgt_buf_top;
27 #ifdef CONFIG_DIRECT_GBPAGES
35 unsigned page_size_mask;
38 static int page_size_mask;
40 static void __init probe_page_size_mask(void)
42 #if !defined(CONFIG_DEBUG_PAGEALLOC) && !defined(CONFIG_KMEMCHECK)
44 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
45 * This will simplify cpa(), which otherwise needs to support splitting
46 * large pages into small in interrupt context, etc.
49 page_size_mask |= 1 << PG_LEVEL_1G;
51 page_size_mask |= 1 << PG_LEVEL_2M;
54 /* Enable PSE if available */
56 set_in_cr4(X86_CR4_PSE);
58 /* Enable PGE if available */
60 set_in_cr4(X86_CR4_PGE);
61 __supported_pte_mask |= _PAGE_GLOBAL;
64 void __init native_pagetable_reserve(u64 start, u64 end)
66 memblock_reserve(start, end - start);
71 #else /* CONFIG_X86_64 */
75 static int __meminit save_mr(struct map_range *mr, int nr_range,
76 unsigned long start_pfn, unsigned long end_pfn,
77 unsigned long page_size_mask)
79 if (start_pfn < end_pfn) {
80 if (nr_range >= NR_RANGE_MR)
81 panic("run out of range for init_memory_mapping\n");
82 mr[nr_range].start = start_pfn<<PAGE_SHIFT;
83 mr[nr_range].end = end_pfn<<PAGE_SHIFT;
84 mr[nr_range].page_size_mask = page_size_mask;
91 static int __meminit split_mem_range(struct map_range *mr, int nr_range,
95 unsigned long start_pfn, end_pfn;
99 /* head if not big page alignment ? */
100 start_pfn = start >> PAGE_SHIFT;
101 pos = start_pfn << PAGE_SHIFT;
104 * Don't use a large page for the first 2/4MB of memory
105 * because there are often fixed size MTRRs in there
106 * and overlapping MTRRs into large pages can cause
110 end_pfn = 1<<(PMD_SHIFT - PAGE_SHIFT);
112 end_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
113 << (PMD_SHIFT - PAGE_SHIFT);
114 #else /* CONFIG_X86_64 */
115 end_pfn = ((pos + (PMD_SIZE - 1)) >> PMD_SHIFT)
116 << (PMD_SHIFT - PAGE_SHIFT);
118 if (end_pfn > (end >> PAGE_SHIFT))
119 end_pfn = end >> PAGE_SHIFT;
120 if (start_pfn < end_pfn) {
121 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
122 pos = end_pfn << PAGE_SHIFT;
125 /* big page (2M) range */
126 start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
127 << (PMD_SHIFT - PAGE_SHIFT);
129 end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
130 #else /* CONFIG_X86_64 */
131 end_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
132 << (PUD_SHIFT - PAGE_SHIFT);
133 if (end_pfn > ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT)))
134 end_pfn = ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT));
137 if (start_pfn < end_pfn) {
138 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
139 page_size_mask & (1<<PG_LEVEL_2M));
140 pos = end_pfn << PAGE_SHIFT;
144 /* big page (1G) range */
145 start_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
146 << (PUD_SHIFT - PAGE_SHIFT);
147 end_pfn = (end >> PUD_SHIFT) << (PUD_SHIFT - PAGE_SHIFT);
148 if (start_pfn < end_pfn) {
149 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
151 ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G)));
152 pos = end_pfn << PAGE_SHIFT;
155 /* tail is not big page (1G) alignment */
156 start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
157 << (PMD_SHIFT - PAGE_SHIFT);
158 end_pfn = (end >> PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
159 if (start_pfn < end_pfn) {
160 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
161 page_size_mask & (1<<PG_LEVEL_2M));
162 pos = end_pfn << PAGE_SHIFT;
166 /* tail is not big page (2M) alignment */
167 start_pfn = pos>>PAGE_SHIFT;
168 end_pfn = end>>PAGE_SHIFT;
169 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
171 /* try to merge same page size and continuous */
172 for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
173 unsigned long old_start;
174 if (mr[i].end != mr[i+1].start ||
175 mr[i].page_size_mask != mr[i+1].page_size_mask)
178 old_start = mr[i].start;
179 memmove(&mr[i], &mr[i+1],
180 (nr_range - 1 - i) * sizeof(struct map_range));
181 mr[i--].start = old_start;
185 for (i = 0; i < nr_range; i++)
186 printk(KERN_DEBUG " [mem %#010lx-%#010lx] page %s\n",
187 mr[i].start, mr[i].end - 1,
188 (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":(
189 (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k"));
195 * First calculate space needed for kernel direct mapping page tables to cover
196 * mr[0].start to mr[nr_range - 1].end, while accounting for possible 2M and 1GB
197 * pages. Then find enough contiguous space for those page tables.
199 static void __init find_early_table_space(unsigned long start, unsigned long end)
202 unsigned long puds = 0, pmds = 0, ptes = 0, tables;
203 unsigned long good_end;
205 struct map_range mr[NR_RANGE_MR];
208 memset(mr, 0, sizeof(mr));
210 nr_range = split_mem_range(mr, nr_range, start, end);
212 for (i = 0; i < nr_range; i++) {
213 unsigned long range, extra;
215 range = mr[i].end - mr[i].start;
216 puds += (range + PUD_SIZE - 1) >> PUD_SHIFT;
218 if (mr[i].page_size_mask & (1 << PG_LEVEL_1G)) {
219 extra = range - ((range >> PUD_SHIFT) << PUD_SHIFT);
220 pmds += (extra + PMD_SIZE - 1) >> PMD_SHIFT;
222 pmds += (range + PMD_SIZE - 1) >> PMD_SHIFT;
225 if (mr[i].page_size_mask & (1 << PG_LEVEL_2M)) {
226 extra = range - ((range >> PMD_SHIFT) << PMD_SHIFT);
230 ptes += (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
232 ptes += (range + PAGE_SIZE - 1) >> PAGE_SHIFT;
236 tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
237 tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
238 tables += roundup(ptes * sizeof(pte_t), PAGE_SIZE);
242 tables += roundup(__end_of_fixed_addresses * sizeof(pte_t), PAGE_SIZE);
243 good_end = max_pfn_mapped << PAGE_SHIFT;
246 base = memblock_find_in_range(start, good_end, tables, PAGE_SIZE);
248 panic("Cannot find space for the kernel page tables");
250 pgt_buf_start = base >> PAGE_SHIFT;
251 pgt_buf_end = pgt_buf_start;
252 pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT);
254 printk(KERN_DEBUG "kernel direct mapping tables up to %#lx @ [mem %#010lx-%#010lx]\n",
255 mr[nr_range - 1].end - 1, pgt_buf_start << PAGE_SHIFT,
256 (pgt_buf_top << PAGE_SHIFT) - 1);
260 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
261 * This runs before bootmem is initialized and gets pages directly from
262 * the physical memory. To access them they are temporarily mapped.
264 unsigned long __init_refok init_memory_mapping(unsigned long start,
267 struct map_range mr[NR_RANGE_MR];
268 unsigned long ret = 0;
271 pr_info("init_memory_mapping: [mem %#010lx-%#010lx]\n",
274 memset(mr, 0, sizeof(mr));
275 nr_range = split_mem_range(mr, 0, start, end);
277 for (i = 0; i < nr_range; i++)
278 ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
279 mr[i].page_size_mask);
282 early_ioremap_page_table_range_init();
284 load_cr3(swapper_pg_dir);
289 return ret >> PAGE_SHIFT;
292 void __init init_mem_mapping(void)
294 probe_page_size_mask();
297 * Find space for the kernel direct mapping tables.
299 * Later we should allocate these tables in the local node of the
300 * memory mapped. Unfortunately this is done currently before the
301 * nodes are discovered.
304 find_early_table_space(0, max_pfn<<PAGE_SHIFT);
306 find_early_table_space(0, max_low_pfn<<PAGE_SHIFT);
308 max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
309 max_pfn_mapped = max_low_pfn_mapped;
312 if (max_pfn > max_low_pfn) {
313 max_pfn_mapped = init_memory_mapping(1UL<<32,
314 max_pfn<<PAGE_SHIFT);
315 /* can we preseve max_low_pfn ?*/
316 max_low_pfn = max_pfn;
320 * Reserve the kernel pagetable pages we used (pgt_buf_start -
321 * pgt_buf_end) and free the other ones (pgt_buf_end - pgt_buf_top)
322 * so that they can be reused for other purposes.
324 * On native it just means calling memblock_reserve, on Xen it also
325 * means marking RW the pagetable pages that we allocated before
326 * but that haven't been used.
328 * In fact on xen we mark RO the whole range pgt_buf_start -
329 * pgt_buf_top, because we have to make sure that when
330 * init_memory_mapping reaches the pagetable pages area, it maps
331 * RO all the pagetable pages, including the ones that are beyond
332 * pgt_buf_end at that time.
334 if (pgt_buf_end > pgt_buf_start)
335 x86_init.mapping.pagetable_reserve(PFN_PHYS(pgt_buf_start),
336 PFN_PHYS(pgt_buf_end));
338 /* stop the wrong using */
341 early_memtest(0, max_pfn_mapped << PAGE_SHIFT);
345 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
346 * is valid. The argument is a physical page number.
349 * On x86, access has to be given to the first megabyte of ram because that area
350 * contains bios code and data regions used by X and dosemu and similar apps.
351 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
352 * mmio resources as well as potential bios/acpi data regions.
354 int devmem_is_allowed(unsigned long pagenr)
358 if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
360 if (!page_is_ram(pagenr))
365 void free_init_pages(char *what, unsigned long begin, unsigned long end)
368 unsigned long begin_aligned, end_aligned;
370 /* Make sure boundaries are page aligned */
371 begin_aligned = PAGE_ALIGN(begin);
372 end_aligned = end & PAGE_MASK;
374 if (WARN_ON(begin_aligned != begin || end_aligned != end)) {
375 begin = begin_aligned;
385 * If debugging page accesses then do not free this memory but
386 * mark them not present - any buggy init-section access will
387 * create a kernel page fault:
389 #ifdef CONFIG_DEBUG_PAGEALLOC
390 printk(KERN_INFO "debug: unmapping init [mem %#010lx-%#010lx]\n",
392 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
395 * We just marked the kernel text read only above, now that
396 * we are going to free part of that, we need to make that
397 * writeable and non-executable first.
399 set_memory_nx(begin, (end - begin) >> PAGE_SHIFT);
400 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
402 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
404 for (; addr < end; addr += PAGE_SIZE) {
405 ClearPageReserved(virt_to_page(addr));
406 init_page_count(virt_to_page(addr));
407 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
414 void free_initmem(void)
416 free_init_pages("unused kernel memory",
417 (unsigned long)(&__init_begin),
418 (unsigned long)(&__init_end));
421 #ifdef CONFIG_BLK_DEV_INITRD
422 void __init free_initrd_mem(unsigned long start, unsigned long end)
425 * end could be not aligned, and We can not align that,
426 * decompresser could be confused by aligned initrd_end
427 * We already reserve the end partial page before in
428 * - i386_start_kernel()
429 * - x86_64_start_kernel()
430 * - relocate_initrd()
431 * So here We can do PAGE_ALIGN() safely to get partial page to be freed
433 free_init_pages("initrd memory", start, PAGE_ALIGN(end));
437 void __init zone_sizes_init(void)
439 unsigned long max_zone_pfns[MAX_NR_ZONES];
441 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
443 #ifdef CONFIG_ZONE_DMA
444 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
446 #ifdef CONFIG_ZONE_DMA32
447 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
449 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
450 #ifdef CONFIG_HIGHMEM
451 max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
454 free_area_init_nodes(max_zone_pfns);