[TG3]: Set minimal hw interrupt mitigation.
[linux-2.6/verdex.git] / arch / i386 / mm / init.c
blob7a7ea3737265284eceafdcdc82000f3c87d06690
1 /*
2 * linux/arch/i386/mm/init.c
4 * Copyright (C) 1995 Linus Torvalds
6 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
7 */
9 #include <linux/config.h>
10 #include <linux/module.h>
11 #include <linux/signal.h>
12 #include <linux/sched.h>
13 #include <linux/kernel.h>
14 #include <linux/errno.h>
15 #include <linux/string.h>
16 #include <linux/types.h>
17 #include <linux/ptrace.h>
18 #include <linux/mman.h>
19 #include <linux/mm.h>
20 #include <linux/hugetlb.h>
21 #include <linux/swap.h>
22 #include <linux/smp.h>
23 #include <linux/init.h>
24 #include <linux/highmem.h>
25 #include <linux/pagemap.h>
26 #include <linux/bootmem.h>
27 #include <linux/slab.h>
28 #include <linux/proc_fs.h>
29 #include <linux/efi.h>
31 #include <asm/processor.h>
32 #include <asm/system.h>
33 #include <asm/uaccess.h>
34 #include <asm/pgtable.h>
35 #include <asm/dma.h>
36 #include <asm/fixmap.h>
37 #include <asm/e820.h>
38 #include <asm/apic.h>
39 #include <asm/tlb.h>
40 #include <asm/tlbflush.h>
41 #include <asm/sections.h>
43 unsigned int __VMALLOC_RESERVE = 128 << 20;
45 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
46 unsigned long highstart_pfn, highend_pfn;
48 static int noinline do_test_wp_bit(void);
51 * Creates a middle page table and puts a pointer to it in the
52 * given global directory entry. This only returns the gd entry
53 * in non-PAE compilation mode, since the middle layer is folded.
55 static pmd_t * __init one_md_table_init(pgd_t *pgd)
57 pud_t *pud;
58 pmd_t *pmd_table;
60 #ifdef CONFIG_X86_PAE
61 pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
62 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
63 pud = pud_offset(pgd, 0);
64 if (pmd_table != pmd_offset(pud, 0))
65 BUG();
66 #else
67 pud = pud_offset(pgd, 0);
68 pmd_table = pmd_offset(pud, 0);
69 #endif
71 return pmd_table;
75 * Create a page table and place a pointer to it in a middle page
76 * directory entry.
78 static pte_t * __init one_page_table_init(pmd_t *pmd)
80 if (pmd_none(*pmd)) {
81 pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
82 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
83 if (page_table != pte_offset_kernel(pmd, 0))
84 BUG();
86 return page_table;
89 return pte_offset_kernel(pmd, 0);
93 * This function initializes a certain range of kernel virtual memory
94 * with new bootmem page tables, everywhere page tables are missing in
95 * the given range.
99 * NOTE: The pagetables are allocated contiguous on the physical space
100 * so we can cache the place of the first one and move around without
101 * checking the pgd every time.
103 static void __init page_table_range_init (unsigned long start, unsigned long end, pgd_t *pgd_base)
105 pgd_t *pgd;
106 pud_t *pud;
107 pmd_t *pmd;
108 int pgd_idx, pmd_idx;
109 unsigned long vaddr;
111 vaddr = start;
112 pgd_idx = pgd_index(vaddr);
113 pmd_idx = pmd_index(vaddr);
114 pgd = pgd_base + pgd_idx;
116 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
117 if (pgd_none(*pgd))
118 one_md_table_init(pgd);
119 pud = pud_offset(pgd, vaddr);
120 pmd = pmd_offset(pud, vaddr);
121 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
122 if (pmd_none(*pmd))
123 one_page_table_init(pmd);
125 vaddr += PMD_SIZE;
127 pmd_idx = 0;
131 static inline int is_kernel_text(unsigned long addr)
133 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
134 return 1;
135 return 0;
139 * This maps the physical memory to kernel virtual address space, a total
140 * of max_low_pfn pages, by creating page tables starting from address
141 * PAGE_OFFSET.
143 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
145 unsigned long pfn;
146 pgd_t *pgd;
147 pmd_t *pmd;
148 pte_t *pte;
149 int pgd_idx, pmd_idx, pte_ofs;
151 pgd_idx = pgd_index(PAGE_OFFSET);
152 pgd = pgd_base + pgd_idx;
153 pfn = 0;
155 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
156 pmd = one_md_table_init(pgd);
157 if (pfn >= max_low_pfn)
158 continue;
159 for (pmd_idx = 0; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) {
160 unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET;
162 /* Map with big pages if possible, otherwise create normal page tables. */
163 if (cpu_has_pse) {
164 unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1;
166 if (is_kernel_text(address) || is_kernel_text(address2))
167 set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
168 else
169 set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
170 pfn += PTRS_PER_PTE;
171 } else {
172 pte = one_page_table_init(pmd);
174 for (pte_ofs = 0; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) {
175 if (is_kernel_text(address))
176 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
177 else
178 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
185 static inline int page_kills_ppro(unsigned long pagenr)
187 if (pagenr >= 0x70000 && pagenr <= 0x7003F)
188 return 1;
189 return 0;
192 extern int is_available_memory(efi_memory_desc_t *);
194 static inline int page_is_ram(unsigned long pagenr)
196 int i;
197 unsigned long addr, end;
199 if (efi_enabled) {
200 efi_memory_desc_t *md;
202 for (i = 0; i < memmap.nr_map; i++) {
203 md = &memmap.map[i];
204 if (!is_available_memory(md))
205 continue;
206 addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT;
207 end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT;
209 if ((pagenr >= addr) && (pagenr < end))
210 return 1;
212 return 0;
215 for (i = 0; i < e820.nr_map; i++) {
217 if (e820.map[i].type != E820_RAM) /* not usable memory */
218 continue;
220 * !!!FIXME!!! Some BIOSen report areas as RAM that
221 * are not. Notably the 640->1Mb area. We need a sanity
222 * check here.
224 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
225 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
226 if ((pagenr >= addr) && (pagenr < end))
227 return 1;
229 return 0;
232 #ifdef CONFIG_HIGHMEM
233 pte_t *kmap_pte;
234 pgprot_t kmap_prot;
236 #define kmap_get_fixmap_pte(vaddr) \
237 pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), (vaddr)), (vaddr))
239 static void __init kmap_init(void)
241 unsigned long kmap_vstart;
243 /* cache the first kmap pte */
244 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
245 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
247 kmap_prot = PAGE_KERNEL;
250 static void __init permanent_kmaps_init(pgd_t *pgd_base)
252 pgd_t *pgd;
253 pud_t *pud;
254 pmd_t *pmd;
255 pte_t *pte;
256 unsigned long vaddr;
258 vaddr = PKMAP_BASE;
259 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
261 pgd = swapper_pg_dir + pgd_index(vaddr);
262 pud = pud_offset(pgd, vaddr);
263 pmd = pmd_offset(pud, vaddr);
264 pte = pte_offset_kernel(pmd, vaddr);
265 pkmap_page_table = pte;
268 void __init one_highpage_init(struct page *page, int pfn, int bad_ppro)
270 if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
271 ClearPageReserved(page);
272 set_bit(PG_highmem, &page->flags);
273 set_page_count(page, 1);
274 __free_page(page);
275 totalhigh_pages++;
276 } else
277 SetPageReserved(page);
280 #ifndef CONFIG_DISCONTIGMEM
281 static void __init set_highmem_pages_init(int bad_ppro)
283 int pfn;
284 for (pfn = highstart_pfn; pfn < highend_pfn; pfn++)
285 one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
286 totalram_pages += totalhigh_pages;
288 #else
289 extern void set_highmem_pages_init(int);
290 #endif /* !CONFIG_DISCONTIGMEM */
292 #else
293 #define kmap_init() do { } while (0)
294 #define permanent_kmaps_init(pgd_base) do { } while (0)
295 #define set_highmem_pages_init(bad_ppro) do { } while (0)
296 #endif /* CONFIG_HIGHMEM */
298 unsigned long long __PAGE_KERNEL = _PAGE_KERNEL;
299 unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
301 #ifndef CONFIG_DISCONTIGMEM
302 #define remap_numa_kva() do {} while (0)
303 #else
304 extern void __init remap_numa_kva(void);
305 #endif
307 static void __init pagetable_init (void)
309 unsigned long vaddr;
310 pgd_t *pgd_base = swapper_pg_dir;
312 #ifdef CONFIG_X86_PAE
313 int i;
314 /* Init entries of the first-level page table to the zero page */
315 for (i = 0; i < PTRS_PER_PGD; i++)
316 set_pgd(pgd_base + i, __pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
317 #endif
319 /* Enable PSE if available */
320 if (cpu_has_pse) {
321 set_in_cr4(X86_CR4_PSE);
324 /* Enable PGE if available */
325 if (cpu_has_pge) {
326 set_in_cr4(X86_CR4_PGE);
327 __PAGE_KERNEL |= _PAGE_GLOBAL;
328 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
331 kernel_physical_mapping_init(pgd_base);
332 remap_numa_kva();
335 * Fixed mappings, only the page table structure has to be
336 * created - mappings will be set by set_fixmap():
338 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
339 page_table_range_init(vaddr, 0, pgd_base);
341 permanent_kmaps_init(pgd_base);
343 #ifdef CONFIG_X86_PAE
345 * Add low memory identity-mappings - SMP needs it when
346 * starting up on an AP from real-mode. In the non-PAE
347 * case we already have these mappings through head.S.
348 * All user-space mappings are explicitly cleared after
349 * SMP startup.
351 pgd_base[0] = pgd_base[USER_PTRS_PER_PGD];
352 #endif
355 #if defined(CONFIG_PM_DISK) || defined(CONFIG_SOFTWARE_SUSPEND)
357 * Swap suspend & friends need this for resume because things like the intel-agp
358 * driver might have split up a kernel 4MB mapping.
360 char __nosavedata swsusp_pg_dir[PAGE_SIZE]
361 __attribute__ ((aligned (PAGE_SIZE)));
363 static inline void save_pg_dir(void)
365 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
367 #else
368 static inline void save_pg_dir(void)
371 #endif
373 void zap_low_mappings (void)
375 int i;
377 save_pg_dir();
380 * Zap initial low-memory mappings.
382 * Note that "pgd_clear()" doesn't do it for
383 * us, because pgd_clear() is a no-op on i386.
385 for (i = 0; i < USER_PTRS_PER_PGD; i++)
386 #ifdef CONFIG_X86_PAE
387 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
388 #else
389 set_pgd(swapper_pg_dir+i, __pgd(0));
390 #endif
391 flush_tlb_all();
394 static int disable_nx __initdata = 0;
395 u64 __supported_pte_mask = ~_PAGE_NX;
398 * noexec = on|off
400 * Control non executable mappings.
402 * on Enable
403 * off Disable
405 void __init noexec_setup(const char *str)
407 if (!strncmp(str, "on",2) && cpu_has_nx) {
408 __supported_pte_mask |= _PAGE_NX;
409 disable_nx = 0;
410 } else if (!strncmp(str,"off",3)) {
411 disable_nx = 1;
412 __supported_pte_mask &= ~_PAGE_NX;
416 int nx_enabled = 0;
417 #ifdef CONFIG_X86_PAE
419 static void __init set_nx(void)
421 unsigned int v[4], l, h;
423 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
424 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
425 if ((v[3] & (1 << 20)) && !disable_nx) {
426 rdmsr(MSR_EFER, l, h);
427 l |= EFER_NX;
428 wrmsr(MSR_EFER, l, h);
429 nx_enabled = 1;
430 __supported_pte_mask |= _PAGE_NX;
436 * Enables/disables executability of a given kernel page and
437 * returns the previous setting.
439 int __init set_kernel_exec(unsigned long vaddr, int enable)
441 pte_t *pte;
442 int ret = 1;
444 if (!nx_enabled)
445 goto out;
447 pte = lookup_address(vaddr);
448 BUG_ON(!pte);
450 if (!pte_exec_kernel(*pte))
451 ret = 0;
453 if (enable)
454 pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
455 else
456 pte->pte_high |= 1 << (_PAGE_BIT_NX - 32);
457 __flush_tlb_all();
458 out:
459 return ret;
462 #endif
465 * paging_init() sets up the page tables - note that the first 8MB are
466 * already mapped by head.S.
468 * This routines also unmaps the page at virtual kernel address 0, so
469 * that we can trap those pesky NULL-reference errors in the kernel.
471 void __init paging_init(void)
473 #ifdef CONFIG_X86_PAE
474 set_nx();
475 if (nx_enabled)
476 printk("NX (Execute Disable) protection: active\n");
477 #endif
479 pagetable_init();
481 load_cr3(swapper_pg_dir);
483 #ifdef CONFIG_X86_PAE
485 * We will bail out later - printk doesn't work right now so
486 * the user would just see a hanging kernel.
488 if (cpu_has_pae)
489 set_in_cr4(X86_CR4_PAE);
490 #endif
491 __flush_tlb_all();
493 kmap_init();
497 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
498 * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
499 * used to involve black magic jumps to work around some nasty CPU bugs,
500 * but fortunately the switch to using exceptions got rid of all that.
503 static void __init test_wp_bit(void)
505 printk("Checking if this processor honours the WP bit even in supervisor mode... ");
507 /* Any page-aligned address will do, the test is non-destructive */
508 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
509 boot_cpu_data.wp_works_ok = do_test_wp_bit();
510 clear_fixmap(FIX_WP_TEST);
512 if (!boot_cpu_data.wp_works_ok) {
513 printk("No.\n");
514 #ifdef CONFIG_X86_WP_WORKS_OK
515 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
516 #endif
517 } else {
518 printk("Ok.\n");
522 static void __init set_max_mapnr_init(void)
524 #ifdef CONFIG_HIGHMEM
525 num_physpages = highend_pfn;
526 #else
527 num_physpages = max_low_pfn;
528 #endif
529 #ifndef CONFIG_DISCONTIGMEM
530 max_mapnr = num_physpages;
531 #endif
534 static struct kcore_list kcore_mem, kcore_vmalloc;
536 void __init mem_init(void)
538 extern int ppro_with_ram_bug(void);
539 int codesize, reservedpages, datasize, initsize;
540 int tmp;
541 int bad_ppro;
543 #ifndef CONFIG_DISCONTIGMEM
544 if (!mem_map)
545 BUG();
546 #endif
548 bad_ppro = ppro_with_ram_bug();
550 #ifdef CONFIG_HIGHMEM
551 /* check that fixmap and pkmap do not overlap */
552 if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
553 printk(KERN_ERR "fixmap and kmap areas overlap - this will crash\n");
554 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
555 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, FIXADDR_START);
556 BUG();
558 #endif
560 set_max_mapnr_init();
562 #ifdef CONFIG_HIGHMEM
563 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
564 #else
565 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
566 #endif
568 /* this will put all low memory onto the freelists */
569 totalram_pages += free_all_bootmem();
571 reservedpages = 0;
572 for (tmp = 0; tmp < max_low_pfn; tmp++)
574 * Only count reserved RAM pages
576 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
577 reservedpages++;
579 set_highmem_pages_init(bad_ppro);
581 codesize = (unsigned long) &_etext - (unsigned long) &_text;
582 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
583 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
585 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
586 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
587 VMALLOC_END-VMALLOC_START);
589 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
590 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
591 num_physpages << (PAGE_SHIFT-10),
592 codesize >> 10,
593 reservedpages << (PAGE_SHIFT-10),
594 datasize >> 10,
595 initsize >> 10,
596 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
599 #ifdef CONFIG_X86_PAE
600 if (!cpu_has_pae)
601 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
602 #endif
603 if (boot_cpu_data.wp_works_ok < 0)
604 test_wp_bit();
607 * Subtle. SMP is doing it's boot stuff late (because it has to
608 * fork idle threads) - but it also needs low mappings for the
609 * protected-mode entry to work. We zap these entries only after
610 * the WP-bit has been tested.
612 #ifndef CONFIG_SMP
613 zap_low_mappings();
614 #endif
617 kmem_cache_t *pgd_cache;
618 kmem_cache_t *pmd_cache;
620 void __init pgtable_cache_init(void)
622 if (PTRS_PER_PMD > 1) {
623 pmd_cache = kmem_cache_create("pmd",
624 PTRS_PER_PMD*sizeof(pmd_t),
625 PTRS_PER_PMD*sizeof(pmd_t),
627 pmd_ctor,
628 NULL);
629 if (!pmd_cache)
630 panic("pgtable_cache_init(): cannot create pmd cache");
632 pgd_cache = kmem_cache_create("pgd",
633 PTRS_PER_PGD*sizeof(pgd_t),
634 PTRS_PER_PGD*sizeof(pgd_t),
636 pgd_ctor,
637 PTRS_PER_PMD == 1 ? pgd_dtor : NULL);
638 if (!pgd_cache)
639 panic("pgtable_cache_init(): Cannot create pgd cache");
643 * This function cannot be __init, since exceptions don't work in that
644 * section. Put this after the callers, so that it cannot be inlined.
646 static int noinline do_test_wp_bit(void)
648 char tmp_reg;
649 int flag;
651 __asm__ __volatile__(
652 " movb %0,%1 \n"
653 "1: movb %1,%0 \n"
654 " xorl %2,%2 \n"
655 "2: \n"
656 ".section __ex_table,\"a\"\n"
657 " .align 4 \n"
658 " .long 1b,2b \n"
659 ".previous \n"
660 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
661 "=q" (tmp_reg),
662 "=r" (flag)
663 :"2" (1)
664 :"memory");
666 return flag;
669 void free_initmem(void)
671 unsigned long addr;
673 addr = (unsigned long)(&__init_begin);
674 for (; addr < (unsigned long)(&__init_end); addr += PAGE_SIZE) {
675 ClearPageReserved(virt_to_page(addr));
676 set_page_count(virt_to_page(addr), 1);
677 memset((void *)addr, 0xcc, PAGE_SIZE);
678 free_page(addr);
679 totalram_pages++;
681 printk (KERN_INFO "Freeing unused kernel memory: %dk freed\n", (__init_end - __init_begin) >> 10);
684 #ifdef CONFIG_BLK_DEV_INITRD
685 void free_initrd_mem(unsigned long start, unsigned long end)
687 if (start < end)
688 printk (KERN_INFO "Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
689 for (; start < end; start += PAGE_SIZE) {
690 ClearPageReserved(virt_to_page(start));
691 set_page_count(virt_to_page(start), 1);
692 free_page(start);
693 totalram_pages++;
696 #endif