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>
15 #include <asm/proto.h>
17 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
19 unsigned long __initdata e820_table_start;
20 unsigned long __meminitdata e820_table_end;
21 unsigned long __meminitdata e820_table_top;
26 #ifdef CONFIG_DIRECT_GBPAGES
33 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
34 static int disable_nx __cpuinitdata;
39 * Control non-executable mappings for processes.
44 static int __init noexec_setup(char *str)
48 if (!strncmp(str, "on", 2)) {
49 __supported_pte_mask |= _PAGE_NX;
51 } else if (!strncmp(str, "off", 3)) {
53 __supported_pte_mask &= ~_PAGE_NX;
57 early_param("noexec", noexec_setup);
61 static void __init set_nx(void)
63 unsigned int v[4], l, h;
65 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
66 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
68 if ((v[3] & (1 << 20)) && !disable_nx) {
69 rdmsr(MSR_EFER, l, h);
71 wrmsr(MSR_EFER, l, h);
73 __supported_pte_mask |= _PAGE_NX;
78 static inline void set_nx(void)
84 void __cpuinit check_efer(void)
88 rdmsrl(MSR_EFER, efer);
89 if (!(efer & EFER_NX) || disable_nx)
90 __supported_pte_mask &= ~_PAGE_NX;
94 static void __init find_early_table_space(unsigned long end, int use_pse,
97 unsigned long puds, pmds, ptes, tables, start;
99 puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
100 tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
105 extra = end - ((end>>PUD_SHIFT) << PUD_SHIFT);
106 pmds = (extra + PMD_SIZE - 1) >> PMD_SHIFT;
108 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
110 tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
115 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
119 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
121 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
123 tables += roundup(ptes * sizeof(pte_t), PAGE_SIZE);
127 tables += roundup(__end_of_fixed_addresses * sizeof(pte_t), PAGE_SIZE);
131 * RED-PEN putting page tables only on node 0 could
132 * cause a hotspot and fill up ZONE_DMA. The page tables
133 * need roughly 0.5KB per GB.
140 e820_table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT,
142 if (e820_table_start == -1UL)
143 panic("Cannot find space for the kernel page tables");
145 e820_table_start >>= PAGE_SHIFT;
146 e820_table_end = e820_table_start;
147 e820_table_top = e820_table_start + (tables >> PAGE_SHIFT);
149 printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
150 end, e820_table_start << PAGE_SHIFT, e820_table_top << PAGE_SHIFT);
156 unsigned page_size_mask;
160 #define NR_RANGE_MR 3
161 #else /* CONFIG_X86_64 */
162 #define NR_RANGE_MR 5
165 static int __meminit save_mr(struct map_range *mr, int nr_range,
166 unsigned long start_pfn, unsigned long end_pfn,
167 unsigned long page_size_mask)
169 if (start_pfn < end_pfn) {
170 if (nr_range >= NR_RANGE_MR)
171 panic("run out of range for init_memory_mapping\n");
172 mr[nr_range].start = start_pfn<<PAGE_SHIFT;
173 mr[nr_range].end = end_pfn<<PAGE_SHIFT;
174 mr[nr_range].page_size_mask = page_size_mask;
182 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
183 * This runs before bootmem is initialized and gets pages directly from
184 * the physical memory. To access them they are temporarily mapped.
186 unsigned long __init_refok init_memory_mapping(unsigned long start,
189 unsigned long page_size_mask = 0;
190 unsigned long start_pfn, end_pfn;
191 unsigned long ret = 0;
194 struct map_range mr[NR_RANGE_MR];
196 int use_pse, use_gbpages;
198 printk(KERN_INFO "init_memory_mapping: %016lx-%016lx\n", start, end);
200 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
202 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
203 * This will simplify cpa(), which otherwise needs to support splitting
204 * large pages into small in interrupt context, etc.
206 use_pse = use_gbpages = 0;
208 use_pse = cpu_has_pse;
209 use_gbpages = direct_gbpages;
214 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
216 /* Enable PSE if available */
218 set_in_cr4(X86_CR4_PSE);
220 /* Enable PGE if available */
222 set_in_cr4(X86_CR4_PGE);
223 __supported_pte_mask |= _PAGE_GLOBAL;
227 page_size_mask |= 1 << PG_LEVEL_1G;
229 page_size_mask |= 1 << PG_LEVEL_2M;
231 memset(mr, 0, sizeof(mr));
234 /* head if not big page alignment ? */
235 start_pfn = start >> PAGE_SHIFT;
236 pos = start_pfn << PAGE_SHIFT;
239 * Don't use a large page for the first 2/4MB of memory
240 * because there are often fixed size MTRRs in there
241 * and overlapping MTRRs into large pages can cause
245 end_pfn = 1<<(PMD_SHIFT - PAGE_SHIFT);
247 end_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
248 << (PMD_SHIFT - PAGE_SHIFT);
249 #else /* CONFIG_X86_64 */
250 end_pfn = ((pos + (PMD_SIZE - 1)) >> PMD_SHIFT)
251 << (PMD_SHIFT - PAGE_SHIFT);
253 if (end_pfn > (end >> PAGE_SHIFT))
254 end_pfn = end >> PAGE_SHIFT;
255 if (start_pfn < end_pfn) {
256 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
257 pos = end_pfn << PAGE_SHIFT;
260 /* big page (2M) range */
261 start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
262 << (PMD_SHIFT - PAGE_SHIFT);
264 end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
265 #else /* CONFIG_X86_64 */
266 end_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
267 << (PUD_SHIFT - PAGE_SHIFT);
268 if (end_pfn > ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT)))
269 end_pfn = ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT));
272 if (start_pfn < end_pfn) {
273 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
274 page_size_mask & (1<<PG_LEVEL_2M));
275 pos = end_pfn << PAGE_SHIFT;
279 /* big page (1G) range */
280 start_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
281 << (PUD_SHIFT - PAGE_SHIFT);
282 end_pfn = (end >> PUD_SHIFT) << (PUD_SHIFT - PAGE_SHIFT);
283 if (start_pfn < end_pfn) {
284 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
286 ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G)));
287 pos = end_pfn << PAGE_SHIFT;
290 /* tail is not big page (1G) alignment */
291 start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
292 << (PMD_SHIFT - PAGE_SHIFT);
293 end_pfn = (end >> PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
294 if (start_pfn < end_pfn) {
295 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
296 page_size_mask & (1<<PG_LEVEL_2M));
297 pos = end_pfn << PAGE_SHIFT;
301 /* tail is not big page (2M) alignment */
302 start_pfn = pos>>PAGE_SHIFT;
303 end_pfn = end>>PAGE_SHIFT;
304 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
306 /* try to merge same page size and continuous */
307 for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
308 unsigned long old_start;
309 if (mr[i].end != mr[i+1].start ||
310 mr[i].page_size_mask != mr[i+1].page_size_mask)
313 old_start = mr[i].start;
314 memmove(&mr[i], &mr[i+1],
315 (nr_range - 1 - i) * sizeof(struct map_range));
316 mr[i--].start = old_start;
320 for (i = 0; i < nr_range; i++)
321 printk(KERN_DEBUG " %010lx - %010lx page %s\n",
322 mr[i].start, mr[i].end,
323 (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":(
324 (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k"));
327 * Find space for the kernel direct mapping tables.
329 * Later we should allocate these tables in the local node of the
330 * memory mapped. Unfortunately this is done currently before the
331 * nodes are discovered.
334 find_early_table_space(end, use_pse, use_gbpages);
337 for (i = 0; i < nr_range; i++)
338 kernel_physical_mapping_init(mr[i].start, mr[i].end,
339 mr[i].page_size_mask);
341 #else /* CONFIG_X86_64 */
342 for (i = 0; i < nr_range; i++)
343 ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
344 mr[i].page_size_mask);
348 early_ioremap_page_table_range_init();
350 load_cr3(swapper_pg_dir);
354 if (!after_bootmem && !start) {
358 mmu_cr4_features = read_cr4();
361 * _brk_end cannot change anymore, but it and _end may be
362 * located on different 2M pages. cleanup_highmap(), however,
363 * can only consider _end when it runs, so destroy any
364 * mappings beyond _brk_end here.
366 pud = pud_offset(pgd_offset_k(_brk_end), _brk_end);
367 pmd = pmd_offset(pud, _brk_end - 1);
368 while (++pmd <= pmd_offset(pud, (unsigned long)_end - 1))
374 if (!after_bootmem && e820_table_end > e820_table_start)
375 reserve_early(e820_table_start << PAGE_SHIFT,
376 e820_table_end << PAGE_SHIFT, "PGTABLE");
379 early_memtest(start, end);
381 return ret >> PAGE_SHIFT;
386 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
387 * is valid. The argument is a physical page number.
390 * On x86, access has to be given to the first megabyte of ram because that area
391 * contains bios code and data regions used by X and dosemu and similar apps.
392 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
393 * mmio resources as well as potential bios/acpi data regions.
395 int devmem_is_allowed(unsigned long pagenr)
399 if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
401 if (!page_is_ram(pagenr))
406 void free_init_pages(char *what, unsigned long begin, unsigned long end)
408 unsigned long addr = begin;
414 * If debugging page accesses then do not free this memory but
415 * mark them not present - any buggy init-section access will
416 * create a kernel page fault:
418 #ifdef CONFIG_DEBUG_PAGEALLOC
419 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
420 begin, PAGE_ALIGN(end));
421 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
424 * We just marked the kernel text read only above, now that
425 * we are going to free part of that, we need to make that
428 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
430 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
432 for (; addr < end; addr += PAGE_SIZE) {
433 ClearPageReserved(virt_to_page(addr));
434 init_page_count(virt_to_page(addr));
435 memset((void *)(addr & ~(PAGE_SIZE-1)),
436 POISON_FREE_INITMEM, PAGE_SIZE);
443 void free_initmem(void)
445 free_init_pages("unused kernel memory",
446 (unsigned long)(&__init_begin),
447 (unsigned long)(&__init_end));
450 #ifdef CONFIG_BLK_DEV_INITRD
451 void free_initrd_mem(unsigned long start, unsigned long end)
453 free_init_pages("initrd memory", start, end);