1 #include <linux/initrd.h>
2 #include <linux/ioport.h>
3 #include <linux/swap.h>
5 #include <asm/cacheflush.h>
9 #include <asm/page_types.h>
10 #include <asm/sections.h>
11 #include <asm/setup.h>
12 #include <asm/system.h>
13 #include <asm/tlbflush.h>
16 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
18 unsigned long __initdata e820_table_start;
19 unsigned long __meminitdata e820_table_end;
20 unsigned long __meminitdata e820_table_top;
25 #ifdef CONFIG_DIRECT_GBPAGES
32 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
33 static int disable_nx __cpuinitdata;
38 * Control non-executable mappings for processes.
43 static int __init noexec_setup(char *str)
47 if (!strncmp(str, "on", 2)) {
48 __supported_pte_mask |= _PAGE_NX;
50 } else if (!strncmp(str, "off", 3)) {
52 __supported_pte_mask &= ~_PAGE_NX;
56 early_param("noexec", noexec_setup);
60 static void __init set_nx(void)
62 unsigned int v[4], l, h;
64 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
65 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
67 if ((v[3] & (1 << 20)) && !disable_nx) {
68 rdmsr(MSR_EFER, l, h);
70 wrmsr(MSR_EFER, l, h);
72 __supported_pte_mask |= _PAGE_NX;
77 static inline void set_nx(void)
83 void __cpuinit check_efer(void)
87 rdmsrl(MSR_EFER, efer);
88 if (!(efer & EFER_NX) || disable_nx)
89 __supported_pte_mask &= ~_PAGE_NX;
93 static void __init find_early_table_space(unsigned long end, int use_pse,
96 unsigned long puds, pmds, ptes, tables, start;
98 puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
99 tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
104 extra = end - ((end>>PUD_SHIFT) << PUD_SHIFT);
105 pmds = (extra + PMD_SIZE - 1) >> PMD_SHIFT;
107 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
109 tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
114 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
118 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
120 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
122 tables += roundup(ptes * sizeof(pte_t), PAGE_SIZE);
126 tables += roundup(__end_of_fixed_addresses * sizeof(pte_t), PAGE_SIZE);
130 * RED-PEN putting page tables only on node 0 could
131 * cause a hotspot and fill up ZONE_DMA. The page tables
132 * need roughly 0.5KB per GB.
139 e820_table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT,
141 if (e820_table_start == -1UL)
142 panic("Cannot find space for the kernel page tables");
144 e820_table_start >>= PAGE_SHIFT;
145 e820_table_end = e820_table_start;
146 e820_table_top = e820_table_start + (tables >> PAGE_SHIFT);
148 printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
149 end, e820_table_start << PAGE_SHIFT, e820_table_top << PAGE_SHIFT);
155 unsigned page_size_mask;
159 #define NR_RANGE_MR 3
160 #else /* CONFIG_X86_64 */
161 #define NR_RANGE_MR 5
164 static int __meminit save_mr(struct map_range *mr, int nr_range,
165 unsigned long start_pfn, unsigned long end_pfn,
166 unsigned long page_size_mask)
168 if (start_pfn < end_pfn) {
169 if (nr_range >= NR_RANGE_MR)
170 panic("run out of range for init_memory_mapping\n");
171 mr[nr_range].start = start_pfn<<PAGE_SHIFT;
172 mr[nr_range].end = end_pfn<<PAGE_SHIFT;
173 mr[nr_range].page_size_mask = page_size_mask;
181 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
182 * This runs before bootmem is initialized and gets pages directly from
183 * the physical memory. To access them they are temporarily mapped.
185 unsigned long __init_refok init_memory_mapping(unsigned long start,
188 unsigned long page_size_mask = 0;
189 unsigned long start_pfn, end_pfn;
190 unsigned long ret = 0;
193 struct map_range mr[NR_RANGE_MR];
195 int use_pse, use_gbpages;
197 printk(KERN_INFO "init_memory_mapping: %016lx-%016lx\n", start, end);
199 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
201 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
202 * This will simplify cpa(), which otherwise needs to support splitting
203 * large pages into small in interrupt context, etc.
205 use_pse = use_gbpages = 0;
207 use_pse = cpu_has_pse;
208 use_gbpages = direct_gbpages;
213 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
215 /* Enable PSE if available */
217 set_in_cr4(X86_CR4_PSE);
219 /* Enable PGE if available */
221 set_in_cr4(X86_CR4_PGE);
222 __supported_pte_mask |= _PAGE_GLOBAL;
226 page_size_mask |= 1 << PG_LEVEL_1G;
228 page_size_mask |= 1 << PG_LEVEL_2M;
230 memset(mr, 0, sizeof(mr));
233 /* head if not big page alignment ? */
234 start_pfn = start >> PAGE_SHIFT;
235 pos = start_pfn << PAGE_SHIFT;
238 * Don't use a large page for the first 2/4MB of memory
239 * because there are often fixed size MTRRs in there
240 * and overlapping MTRRs into large pages can cause
244 end_pfn = 1<<(PMD_SHIFT - PAGE_SHIFT);
246 end_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
247 << (PMD_SHIFT - PAGE_SHIFT);
248 #else /* CONFIG_X86_64 */
249 end_pfn = ((pos + (PMD_SIZE - 1)) >> PMD_SHIFT)
250 << (PMD_SHIFT - PAGE_SHIFT);
252 if (end_pfn > (end >> PAGE_SHIFT))
253 end_pfn = end >> PAGE_SHIFT;
254 if (start_pfn < end_pfn) {
255 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
256 pos = end_pfn << PAGE_SHIFT;
259 /* big page (2M) range */
260 start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
261 << (PMD_SHIFT - PAGE_SHIFT);
263 end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
264 #else /* CONFIG_X86_64 */
265 end_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
266 << (PUD_SHIFT - PAGE_SHIFT);
267 if (end_pfn > ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT)))
268 end_pfn = ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT));
271 if (start_pfn < end_pfn) {
272 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
273 page_size_mask & (1<<PG_LEVEL_2M));
274 pos = end_pfn << PAGE_SHIFT;
278 /* big page (1G) range */
279 start_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
280 << (PUD_SHIFT - PAGE_SHIFT);
281 end_pfn = (end >> PUD_SHIFT) << (PUD_SHIFT - PAGE_SHIFT);
282 if (start_pfn < end_pfn) {
283 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
285 ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G)));
286 pos = end_pfn << PAGE_SHIFT;
289 /* tail is not big page (1G) alignment */
290 start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
291 << (PMD_SHIFT - PAGE_SHIFT);
292 end_pfn = (end >> PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
293 if (start_pfn < end_pfn) {
294 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
295 page_size_mask & (1<<PG_LEVEL_2M));
296 pos = end_pfn << PAGE_SHIFT;
300 /* tail is not big page (2M) alignment */
301 start_pfn = pos>>PAGE_SHIFT;
302 end_pfn = end>>PAGE_SHIFT;
303 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
305 /* try to merge same page size and continuous */
306 for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
307 unsigned long old_start;
308 if (mr[i].end != mr[i+1].start ||
309 mr[i].page_size_mask != mr[i+1].page_size_mask)
312 old_start = mr[i].start;
313 memmove(&mr[i], &mr[i+1],
314 (nr_range - 1 - i) * sizeof(struct map_range));
315 mr[i--].start = old_start;
319 for (i = 0; i < nr_range; i++)
320 printk(KERN_DEBUG " %010lx - %010lx page %s\n",
321 mr[i].start, mr[i].end,
322 (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":(
323 (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k"));
326 * Find space for the kernel direct mapping tables.
328 * Later we should allocate these tables in the local node of the
329 * memory mapped. Unfortunately this is done currently before the
330 * nodes are discovered.
333 find_early_table_space(end, use_pse, use_gbpages);
336 for (i = 0; i < nr_range; i++)
337 kernel_physical_mapping_init(mr[i].start, mr[i].end,
338 mr[i].page_size_mask);
340 #else /* CONFIG_X86_64 */
341 for (i = 0; i < nr_range; i++)
342 ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
343 mr[i].page_size_mask);
347 early_ioremap_page_table_range_init();
349 load_cr3(swapper_pg_dir);
353 if (!after_bootmem && !start) {
357 mmu_cr4_features = read_cr4();
360 * _brk_end cannot change anymore, but it and _end may be
361 * located on different 2M pages. cleanup_highmap(), however,
362 * can only consider _end when it runs, so destroy any
363 * mappings beyond _brk_end here.
365 pud = pud_offset(pgd_offset_k(_brk_end), _brk_end);
366 pmd = pmd_offset(pud, _brk_end - 1);
367 while (++pmd <= pmd_offset(pud, (unsigned long)_end - 1))
373 if (!after_bootmem && e820_table_end > e820_table_start)
374 reserve_early(e820_table_start << PAGE_SHIFT,
375 e820_table_end << PAGE_SHIFT, "PGTABLE");
378 early_memtest(start, end);
380 return ret >> PAGE_SHIFT;
385 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
386 * is valid. The argument is a physical page number.
389 * On x86, access has to be given to the first megabyte of ram because that area
390 * contains bios code and data regions used by X and dosemu and similar apps.
391 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
392 * mmio resources as well as potential bios/acpi data regions.
394 int devmem_is_allowed(unsigned long pagenr)
398 if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
400 if (!page_is_ram(pagenr))
405 void free_init_pages(char *what, unsigned long begin, unsigned long end)
407 unsigned long addr = begin;
413 * If debugging page accesses then do not free this memory but
414 * mark them not present - any buggy init-section access will
415 * create a kernel page fault:
417 #ifdef CONFIG_DEBUG_PAGEALLOC
418 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
419 begin, PAGE_ALIGN(end));
420 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
423 * We just marked the kernel text read only above, now that
424 * we are going to free part of that, we need to make that
427 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
429 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
431 for (; addr < end; addr += PAGE_SIZE) {
432 ClearPageReserved(virt_to_page(addr));
433 init_page_count(virt_to_page(addr));
434 memset((void *)(addr & ~(PAGE_SIZE-1)),
435 POISON_FREE_INITMEM, PAGE_SIZE);
442 void free_initmem(void)
444 free_init_pages("unused kernel memory",
445 (unsigned long)(&__init_begin),
446 (unsigned long)(&__init_end));
449 #ifdef CONFIG_BLK_DEV_INITRD
450 void free_initrd_mem(unsigned long start, unsigned long end)
452 free_init_pages("initrd memory", start, end);