2 #include <linux/initrd.h>
3 #include <linux/ioport.h>
4 #include <linux/swap.h>
5 #include <linux/memblock.h>
7 #include <asm/cacheflush.h>
11 #include <asm/page_types.h>
12 #include <asm/sections.h>
13 #include <asm/setup.h>
14 #include <asm/system.h>
15 #include <asm/tlbflush.h>
17 #include <asm/proto.h>
19 unsigned long __initdata pgt_buf_start;
20 unsigned long __meminitdata pgt_buf_end;
21 unsigned long __meminitdata pgt_buf_top;
26 #ifdef CONFIG_DIRECT_GBPAGES
34 unsigned page_size_mask;
38 * First calculate space needed for kernel direct mapping page tables to cover
39 * mr[0].start to mr[nr_range - 1].end, while accounting for possible 2M and 1GB
40 * pages. Then find enough contiguous space for those page tables.
42 static void __init find_early_table_space(struct map_range *mr, int nr_range)
45 unsigned long puds = 0, pmds = 0, ptes = 0, tables;
46 unsigned long start = 0, good_end;
47 unsigned long pgd_extra = 0;
50 for (i = 0; i < nr_range; i++) {
51 unsigned long range, extra;
53 if ((mr[i].end >> PGDIR_SHIFT) - (mr[i].start >> PGDIR_SHIFT))
56 range = mr[i].end - mr[i].start;
57 puds += (range + PUD_SIZE - 1) >> PUD_SHIFT;
59 if (mr[i].page_size_mask & (1 << PG_LEVEL_1G)) {
60 extra = range - ((range >> PUD_SHIFT) << PUD_SHIFT);
61 pmds += (extra + PMD_SIZE - 1) >> PMD_SHIFT;
63 pmds += (range + PMD_SIZE - 1) >> PMD_SHIFT;
66 if (mr[i].page_size_mask & (1 << PG_LEVEL_2M)) {
67 extra = range - ((range >> PMD_SHIFT) << PMD_SHIFT);
71 ptes += (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
73 ptes += (range + PAGE_SIZE - 1) >> PAGE_SHIFT;
77 tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
78 tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
79 tables += roundup(ptes * sizeof(pte_t), PAGE_SIZE);
80 tables += (pgd_extra * PAGE_SIZE);
84 tables += roundup(__end_of_fixed_addresses * sizeof(pte_t), PAGE_SIZE);
86 good_end = max_pfn_mapped << PAGE_SHIFT;
88 base = memblock_find_in_range(start, good_end, tables, PAGE_SIZE);
89 if (base == MEMBLOCK_ERROR)
90 panic("Cannot find space for the kernel page tables");
92 pgt_buf_start = base >> PAGE_SHIFT;
93 pgt_buf_end = pgt_buf_start;
94 pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT);
96 printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
97 mr[nr_range - 1].end, pgt_buf_start << PAGE_SHIFT,
98 pgt_buf_top << PAGE_SHIFT);
101 void __init native_pagetable_reserve(u64 start, u64 end)
103 memblock_x86_reserve_range(start, end, "PGTABLE");
107 #define NR_RANGE_MR 3
108 #else /* CONFIG_X86_64 */
109 #define NR_RANGE_MR 5
112 static int __meminit save_mr(struct map_range *mr, int nr_range,
113 unsigned long start_pfn, unsigned long end_pfn,
114 unsigned long page_size_mask)
116 if (start_pfn < end_pfn) {
117 if (nr_range >= NR_RANGE_MR)
118 panic("run out of range for init_memory_mapping\n");
119 mr[nr_range].start = start_pfn<<PAGE_SHIFT;
120 mr[nr_range].end = end_pfn<<PAGE_SHIFT;
121 mr[nr_range].page_size_mask = page_size_mask;
129 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
130 * This runs before bootmem is initialized and gets pages directly from
131 * the physical memory. To access them they are temporarily mapped.
133 unsigned long __init_refok init_memory_mapping(unsigned long start,
136 unsigned long page_size_mask = 0;
137 unsigned long start_pfn, end_pfn;
138 unsigned long ret = 0;
141 struct map_range mr[NR_RANGE_MR];
143 int use_pse, use_gbpages;
145 printk(KERN_INFO "init_memory_mapping: %016lx-%016lx\n", start, end);
147 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
149 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
150 * This will simplify cpa(), which otherwise needs to support splitting
151 * large pages into small in interrupt context, etc.
153 use_pse = use_gbpages = 0;
155 use_pse = cpu_has_pse;
156 use_gbpages = direct_gbpages;
159 /* Enable PSE if available */
161 set_in_cr4(X86_CR4_PSE);
163 /* Enable PGE if available */
165 set_in_cr4(X86_CR4_PGE);
166 __supported_pte_mask |= _PAGE_GLOBAL;
170 page_size_mask |= 1 << PG_LEVEL_1G;
172 page_size_mask |= 1 << PG_LEVEL_2M;
174 memset(mr, 0, sizeof(mr));
177 /* head if not big page alignment ? */
178 start_pfn = start >> PAGE_SHIFT;
179 pos = start_pfn << PAGE_SHIFT;
182 * Don't use a large page for the first 2/4MB of memory
183 * because there are often fixed size MTRRs in there
184 * and overlapping MTRRs into large pages can cause
188 end_pfn = 1<<(PMD_SHIFT - PAGE_SHIFT);
190 end_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
191 << (PMD_SHIFT - PAGE_SHIFT);
192 #else /* CONFIG_X86_64 */
193 end_pfn = ((pos + (PMD_SIZE - 1)) >> PMD_SHIFT)
194 << (PMD_SHIFT - PAGE_SHIFT);
196 if (end_pfn > (end >> PAGE_SHIFT))
197 end_pfn = end >> PAGE_SHIFT;
198 if (start_pfn < end_pfn) {
199 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
200 pos = end_pfn << PAGE_SHIFT;
203 /* big page (2M) range */
204 start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
205 << (PMD_SHIFT - PAGE_SHIFT);
207 end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
208 #else /* CONFIG_X86_64 */
209 end_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
210 << (PUD_SHIFT - PAGE_SHIFT);
211 if (end_pfn > ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT)))
212 end_pfn = ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT));
215 if (start_pfn < end_pfn) {
216 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
217 page_size_mask & (1<<PG_LEVEL_2M));
218 pos = end_pfn << PAGE_SHIFT;
222 /* big page (1G) range */
223 start_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
224 << (PUD_SHIFT - PAGE_SHIFT);
225 end_pfn = (end >> PUD_SHIFT) << (PUD_SHIFT - PAGE_SHIFT);
226 if (start_pfn < end_pfn) {
227 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
229 ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G)));
230 pos = end_pfn << PAGE_SHIFT;
233 /* tail is not big page (1G) alignment */
234 start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
235 << (PMD_SHIFT - PAGE_SHIFT);
236 end_pfn = (end >> PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
237 if (start_pfn < end_pfn) {
238 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
239 page_size_mask & (1<<PG_LEVEL_2M));
240 pos = end_pfn << PAGE_SHIFT;
244 /* tail is not big page (2M) alignment */
245 start_pfn = pos>>PAGE_SHIFT;
246 end_pfn = end>>PAGE_SHIFT;
247 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
249 /* try to merge same page size and continuous */
250 for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
251 unsigned long old_start;
252 if (mr[i].end != mr[i+1].start ||
253 mr[i].page_size_mask != mr[i+1].page_size_mask)
256 old_start = mr[i].start;
257 memmove(&mr[i], &mr[i+1],
258 (nr_range - 1 - i) * sizeof(struct map_range));
259 mr[i--].start = old_start;
263 for (i = 0; i < nr_range; i++)
264 printk(KERN_DEBUG " %010lx - %010lx page %s\n",
265 mr[i].start, mr[i].end,
266 (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":(
267 (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k"));
270 * Find space for the kernel direct mapping tables.
272 * Later we should allocate these tables in the local node of the
273 * memory mapped. Unfortunately this is done currently before the
274 * nodes are discovered.
277 find_early_table_space(mr, nr_range);
279 for (i = 0; i < nr_range; i++)
280 ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
281 mr[i].page_size_mask);
284 early_ioremap_page_table_range_init();
286 load_cr3(swapper_pg_dir);
292 * Reserve the kernel pagetable pages we used (pgt_buf_start -
293 * pgt_buf_end) and free the other ones (pgt_buf_end - pgt_buf_top)
294 * so that they can be reused for other purposes.
296 * On native it just means calling memblock_x86_reserve_range, on Xen it
297 * also means marking RW the pagetable pages that we allocated before
298 * but that haven't been used.
300 * In fact on xen we mark RO the whole range pgt_buf_start -
301 * pgt_buf_top, because we have to make sure that when
302 * init_memory_mapping reaches the pagetable pages area, it maps
303 * RO all the pagetable pages, including the ones that are beyond
304 * pgt_buf_end at that time.
306 if (!after_bootmem && pgt_buf_end > pgt_buf_start)
307 x86_init.mapping.pagetable_reserve(PFN_PHYS(pgt_buf_start),
308 PFN_PHYS(pgt_buf_end));
311 early_memtest(start, end);
313 return ret >> PAGE_SHIFT;
318 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
319 * is valid. The argument is a physical page number.
322 * On x86, access has to be given to the first megabyte of ram because that area
323 * contains bios code and data regions used by X and dosemu and similar apps.
324 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
325 * mmio resources as well as potential bios/acpi data regions.
327 int devmem_is_allowed(unsigned long pagenr)
331 if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
333 if (!page_is_ram(pagenr))
338 void free_init_pages(char *what, unsigned long begin, unsigned long end)
341 unsigned long begin_aligned, end_aligned;
343 /* Make sure boundaries are page aligned */
344 begin_aligned = PAGE_ALIGN(begin);
345 end_aligned = end & PAGE_MASK;
347 if (WARN_ON(begin_aligned != begin || end_aligned != end)) {
348 begin = begin_aligned;
358 * If debugging page accesses then do not free this memory but
359 * mark them not present - any buggy init-section access will
360 * create a kernel page fault:
362 #ifdef CONFIG_DEBUG_PAGEALLOC
363 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
365 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
368 * We just marked the kernel text read only above, now that
369 * we are going to free part of that, we need to make that
370 * writeable and non-executable first.
372 set_memory_nx(begin, (end - begin) >> PAGE_SHIFT);
373 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
375 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
377 for (; addr < end; addr += PAGE_SIZE) {
378 ClearPageReserved(virt_to_page(addr));
379 init_page_count(virt_to_page(addr));
380 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
387 void free_initmem(void)
389 free_init_pages("unused kernel memory",
390 (unsigned long)(&__init_begin),
391 (unsigned long)(&__init_end));
394 #ifdef CONFIG_BLK_DEV_INITRD
395 void free_initrd_mem(unsigned long start, unsigned long end)
398 * end could be not aligned, and We can not align that,
399 * decompresser could be confused by aligned initrd_end
400 * We already reserve the end partial page before in
401 * - i386_start_kernel()
402 * - x86_64_start_kernel()
403 * - relocate_initrd()
404 * So here We can do PAGE_ALIGN() safely to get partial page to be freed
406 free_init_pages("initrd memory", start, PAGE_ALIGN(end));