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/system.h>
16 #include <asm/tlbflush.h>
18 #include <asm/proto.h>
19 #include <asm/dma.h> /* for MAX_DMA_PFN */
21 unsigned long __initdata pgt_buf_start;
22 unsigned long __meminitdata pgt_buf_end;
23 unsigned long __meminitdata pgt_buf_top;
28 #ifdef CONFIG_DIRECT_GBPAGES
33 static void __init find_early_table_space(unsigned long end, int use_pse,
36 unsigned long puds, pmds, ptes, tables, start = 0, good_end = end;
39 puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
40 tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
45 extra = end - ((end>>PUD_SHIFT) << PUD_SHIFT);
46 pmds = (extra + PMD_SIZE - 1) >> PMD_SHIFT;
48 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
50 tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
55 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
59 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
61 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
63 tables += roundup(ptes * sizeof(pte_t), PAGE_SIZE);
67 tables += roundup(__end_of_fixed_addresses * sizeof(pte_t), PAGE_SIZE);
69 good_end = max_pfn_mapped << PAGE_SHIFT;
71 base = memblock_find_in_range(start, good_end, tables, PAGE_SIZE);
73 panic("Cannot find space for the kernel page tables");
75 pgt_buf_start = base >> PAGE_SHIFT;
76 pgt_buf_end = pgt_buf_start;
77 pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT);
79 printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
80 end, pgt_buf_start << PAGE_SHIFT, pgt_buf_top << PAGE_SHIFT);
83 void __init native_pagetable_reserve(u64 start, u64 end)
85 memblock_reserve(start, end - start);
91 unsigned page_size_mask;
96 #else /* CONFIG_X86_64 */
100 static int __meminit save_mr(struct map_range *mr, int nr_range,
101 unsigned long start_pfn, unsigned long end_pfn,
102 unsigned long page_size_mask)
104 if (start_pfn < end_pfn) {
105 if (nr_range >= NR_RANGE_MR)
106 panic("run out of range for init_memory_mapping\n");
107 mr[nr_range].start = start_pfn<<PAGE_SHIFT;
108 mr[nr_range].end = end_pfn<<PAGE_SHIFT;
109 mr[nr_range].page_size_mask = page_size_mask;
117 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
118 * This runs before bootmem is initialized and gets pages directly from
119 * the physical memory. To access them they are temporarily mapped.
121 unsigned long __init_refok init_memory_mapping(unsigned long start,
124 unsigned long page_size_mask = 0;
125 unsigned long start_pfn, end_pfn;
126 unsigned long ret = 0;
129 struct map_range mr[NR_RANGE_MR];
131 int use_pse, use_gbpages;
133 printk(KERN_INFO "init_memory_mapping: %016lx-%016lx\n", start, end);
135 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
137 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
138 * This will simplify cpa(), which otherwise needs to support splitting
139 * large pages into small in interrupt context, etc.
141 use_pse = use_gbpages = 0;
143 use_pse = cpu_has_pse;
144 use_gbpages = direct_gbpages;
147 /* Enable PSE if available */
149 set_in_cr4(X86_CR4_PSE);
151 /* Enable PGE if available */
153 set_in_cr4(X86_CR4_PGE);
154 __supported_pte_mask |= _PAGE_GLOBAL;
158 page_size_mask |= 1 << PG_LEVEL_1G;
160 page_size_mask |= 1 << PG_LEVEL_2M;
162 memset(mr, 0, sizeof(mr));
165 /* head if not big page alignment ? */
166 start_pfn = start >> PAGE_SHIFT;
167 pos = start_pfn << PAGE_SHIFT;
170 * Don't use a large page for the first 2/4MB of memory
171 * because there are often fixed size MTRRs in there
172 * and overlapping MTRRs into large pages can cause
176 end_pfn = 1<<(PMD_SHIFT - PAGE_SHIFT);
178 end_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
179 << (PMD_SHIFT - PAGE_SHIFT);
180 #else /* CONFIG_X86_64 */
181 end_pfn = ((pos + (PMD_SIZE - 1)) >> PMD_SHIFT)
182 << (PMD_SHIFT - PAGE_SHIFT);
184 if (end_pfn > (end >> PAGE_SHIFT))
185 end_pfn = end >> PAGE_SHIFT;
186 if (start_pfn < end_pfn) {
187 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
188 pos = end_pfn << PAGE_SHIFT;
191 /* big page (2M) range */
192 start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
193 << (PMD_SHIFT - PAGE_SHIFT);
195 end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
196 #else /* CONFIG_X86_64 */
197 end_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
198 << (PUD_SHIFT - PAGE_SHIFT);
199 if (end_pfn > ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT)))
200 end_pfn = ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT));
203 if (start_pfn < end_pfn) {
204 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
205 page_size_mask & (1<<PG_LEVEL_2M));
206 pos = end_pfn << PAGE_SHIFT;
210 /* big page (1G) range */
211 start_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
212 << (PUD_SHIFT - PAGE_SHIFT);
213 end_pfn = (end >> PUD_SHIFT) << (PUD_SHIFT - PAGE_SHIFT);
214 if (start_pfn < end_pfn) {
215 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
217 ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G)));
218 pos = end_pfn << PAGE_SHIFT;
221 /* tail is not big page (1G) alignment */
222 start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
223 << (PMD_SHIFT - PAGE_SHIFT);
224 end_pfn = (end >> PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
225 if (start_pfn < end_pfn) {
226 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
227 page_size_mask & (1<<PG_LEVEL_2M));
228 pos = end_pfn << PAGE_SHIFT;
232 /* tail is not big page (2M) alignment */
233 start_pfn = pos>>PAGE_SHIFT;
234 end_pfn = end>>PAGE_SHIFT;
235 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
237 /* try to merge same page size and continuous */
238 for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
239 unsigned long old_start;
240 if (mr[i].end != mr[i+1].start ||
241 mr[i].page_size_mask != mr[i+1].page_size_mask)
244 old_start = mr[i].start;
245 memmove(&mr[i], &mr[i+1],
246 (nr_range - 1 - i) * sizeof(struct map_range));
247 mr[i--].start = old_start;
251 for (i = 0; i < nr_range; i++)
252 printk(KERN_DEBUG " %010lx - %010lx page %s\n",
253 mr[i].start, mr[i].end,
254 (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":(
255 (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k"));
258 * Find space for the kernel direct mapping tables.
260 * Later we should allocate these tables in the local node of the
261 * memory mapped. Unfortunately this is done currently before the
262 * nodes are discovered.
265 find_early_table_space(end, use_pse, use_gbpages);
267 for (i = 0; i < nr_range; i++)
268 ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
269 mr[i].page_size_mask);
272 early_ioremap_page_table_range_init();
274 load_cr3(swapper_pg_dir);
280 * Reserve the kernel pagetable pages we used (pgt_buf_start -
281 * pgt_buf_end) and free the other ones (pgt_buf_end - pgt_buf_top)
282 * so that they can be reused for other purposes.
284 * On native it just means calling memblock_reserve, on Xen it also
285 * means marking RW the pagetable pages that we allocated before
286 * but that haven't been used.
288 * In fact on xen we mark RO the whole range pgt_buf_start -
289 * pgt_buf_top, because we have to make sure that when
290 * init_memory_mapping reaches the pagetable pages area, it maps
291 * RO all the pagetable pages, including the ones that are beyond
292 * pgt_buf_end at that time.
294 if (!after_bootmem && pgt_buf_end > pgt_buf_start)
295 x86_init.mapping.pagetable_reserve(PFN_PHYS(pgt_buf_start),
296 PFN_PHYS(pgt_buf_end));
299 early_memtest(start, end);
301 return ret >> PAGE_SHIFT;
306 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
307 * is valid. The argument is a physical page number.
310 * On x86, access has to be given to the first megabyte of ram because that area
311 * contains bios code and data regions used by X and dosemu and similar apps.
312 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
313 * mmio resources as well as potential bios/acpi data regions.
315 int devmem_is_allowed(unsigned long pagenr)
319 if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
321 if (!page_is_ram(pagenr))
326 void free_init_pages(char *what, unsigned long begin, unsigned long end)
329 unsigned long begin_aligned, end_aligned;
331 /* Make sure boundaries are page aligned */
332 begin_aligned = PAGE_ALIGN(begin);
333 end_aligned = end & PAGE_MASK;
335 if (WARN_ON(begin_aligned != begin || end_aligned != end)) {
336 begin = begin_aligned;
346 * If debugging page accesses then do not free this memory but
347 * mark them not present - any buggy init-section access will
348 * create a kernel page fault:
350 #ifdef CONFIG_DEBUG_PAGEALLOC
351 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
353 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
356 * We just marked the kernel text read only above, now that
357 * we are going to free part of that, we need to make that
358 * writeable and non-executable first.
360 set_memory_nx(begin, (end - begin) >> PAGE_SHIFT);
361 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
363 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
365 for (; addr < end; addr += PAGE_SIZE) {
366 ClearPageReserved(virt_to_page(addr));
367 init_page_count(virt_to_page(addr));
368 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
375 void free_initmem(void)
377 free_init_pages("unused kernel memory",
378 (unsigned long)(&__init_begin),
379 (unsigned long)(&__init_end));
382 #ifdef CONFIG_BLK_DEV_INITRD
383 void free_initrd_mem(unsigned long start, unsigned long end)
386 * end could be not aligned, and We can not align that,
387 * decompresser could be confused by aligned initrd_end
388 * We already reserve the end partial page before in
389 * - i386_start_kernel()
390 * - x86_64_start_kernel()
391 * - relocate_initrd()
392 * So here We can do PAGE_ALIGN() safely to get partial page to be freed
394 free_init_pages("initrd memory", start, PAGE_ALIGN(end));
398 void __init zone_sizes_init(void)
400 unsigned long max_zone_pfns[MAX_NR_ZONES];
402 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
404 #ifdef CONFIG_ZONE_DMA
405 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
407 #ifdef CONFIG_ZONE_DMA32
408 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
410 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
411 #ifdef CONFIG_HIGHMEM
412 max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
415 free_area_init_nodes(max_zone_pfns);