blob: 3fff490254a96fd24ea4bbf15d77a5163d191052 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1995 Linus Torvalds
3 * Copyright (C) 2001,2002 Andi Kleen, SuSE Labs.
4 */
5
Linus Torvalds1da177e2005-04-16 15:20:36 -07006#include <linux/signal.h>
7#include <linux/sched.h>
8#include <linux/kernel.h>
9#include <linux/errno.h>
10#include <linux/string.h>
11#include <linux/types.h>
12#include <linux/ptrace.h>
13#include <linux/mman.h>
14#include <linux/mm.h>
15#include <linux/smp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/interrupt.h>
17#include <linux/init.h>
18#include <linux/tty.h>
19#include <linux/vt_kern.h> /* For unblank_screen() */
20#include <linux/compiler.h>
Harvey Harrisonc61e2112008-01-30 13:34:11 +010021#include <linux/highmem.h>
22#include <linux/bootmem.h> /* for max_low_pfn */
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -070023#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/module.h>
Prasanna S Panchamukhi0f2fbdc2005-09-06 15:19:28 -070025#include <linux/kprobes.h>
Andi Kleenab2bf0c2006-12-07 02:14:06 +010026#include <linux/uaccess.h>
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -070027#include <linux/kdebug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
29#include <asm/system.h>
Harvey Harrisonc61e2112008-01-30 13:34:11 +010030#include <asm/desc.h>
31#include <asm/segment.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <asm/pgalloc.h>
33#include <asm/smp.h>
34#include <asm/tlbflush.h>
35#include <asm/proto.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <asm-generic/sections.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Harvey Harrison33cb5242008-01-30 13:32:19 +010038/*
39 * Page fault error code bits
40 * bit 0 == 0 means no page found, 1 means protection fault
41 * bit 1 == 0 means read, 1 means write
42 * bit 2 == 0 means kernel, 1 means user-mode
43 * bit 3 == 1 means use of reserved bit detected
44 * bit 4 == 1 means fault was an instruction fetch
45 */
Ingo Molnar8a19da72008-01-30 13:32:53 +010046#define PF_PROT (1<<0)
Andi Kleen66c58152006-01-11 22:44:09 +010047#define PF_WRITE (1<<1)
Ingo Molnar8a19da72008-01-30 13:32:53 +010048#define PF_USER (1<<2)
49#define PF_RSVD (1<<3)
Andi Kleen66c58152006-01-11 22:44:09 +010050#define PF_INSTR (1<<4)
51
Christoph Hellwig74a0b572007-10-16 01:24:07 -070052static inline int notify_page_fault(struct pt_regs *regs)
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070053{
Harvey Harrison33cb5242008-01-30 13:32:19 +010054#ifdef CONFIG_KPROBES
Christoph Hellwig74a0b572007-10-16 01:24:07 -070055 int ret = 0;
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070056
Christoph Hellwig74a0b572007-10-16 01:24:07 -070057 /* kprobe_running() needs smp_processor_id() */
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +010058#ifdef CONFIG_X86_32
59 if (!user_mode_vm(regs)) {
60#else
Christoph Hellwig74a0b572007-10-16 01:24:07 -070061 if (!user_mode(regs)) {
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +010062#endif
Christoph Hellwig74a0b572007-10-16 01:24:07 -070063 preempt_disable();
64 if (kprobe_running() && kprobe_fault_handler(regs, 14))
65 ret = 1;
66 preempt_enable();
67 }
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070068
Christoph Hellwig74a0b572007-10-16 01:24:07 -070069 return ret;
Christoph Hellwig74a0b572007-10-16 01:24:07 -070070#else
Christoph Hellwig74a0b572007-10-16 01:24:07 -070071 return 0;
Christoph Hellwig74a0b572007-10-16 01:24:07 -070072#endif
Harvey Harrison33cb5242008-01-30 13:32:19 +010073}
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070074
Harvey Harrison1dc85be2008-01-30 13:32:35 +010075/*
76 * X86_32
77 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
78 * Check that here and ignore it.
79 *
80 * X86_64
81 * Sometimes the CPU reports invalid exceptions on prefetch.
82 * Check that here and ignore it.
83 *
84 * Opcode checker based on code by Richard Brunner
85 */
86static int is_prefetch(struct pt_regs *regs, unsigned long addr,
87 unsigned long error_code)
Harvey Harrison33cb5242008-01-30 13:32:19 +010088{
Andi Kleenab2bf0c2006-12-07 02:14:06 +010089 unsigned char *instr;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 int scan_more = 1;
Harvey Harrison33cb5242008-01-30 13:32:19 +010091 int prefetch = 0;
Andi Kleenf1290ec2005-04-16 15:24:59 -070092 unsigned char *max_instr;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
Harvey Harrison1dc85be2008-01-30 13:32:35 +010094#ifdef CONFIG_X86_32
Harvey Harrisonb406ac62008-01-30 13:34:11 +010095 if (!(__supported_pte_mask & _PAGE_NX))
Harvey Harrison1dc85be2008-01-30 13:32:35 +010096 return 0;
Harvey Harrisonb406ac62008-01-30 13:34:11 +010097#endif
98
Harvey Harrisonc61e2112008-01-30 13:34:11 +010099 /* If it was a exec fault on NX page, ignore */
Andi Kleen66c58152006-01-11 22:44:09 +0100100 if (error_code & PF_INSTR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 return 0;
Harvey Harrison1dc85be2008-01-30 13:32:35 +0100102
Harvey Harrisonf2857ce2008-01-30 13:33:12 +0100103 instr = (unsigned char *)convert_ip_to_linear(current, regs);
Andi Kleenf1290ec2005-04-16 15:24:59 -0700104 max_instr = instr + 15;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105
Vincent Hanquez76381fe2005-06-23 00:08:46 -0700106 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 return 0;
108
Harvey Harrison33cb5242008-01-30 13:32:19 +0100109 while (scan_more && instr < max_instr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 unsigned char opcode;
111 unsigned char instr_hi;
112 unsigned char instr_lo;
113
Andi Kleenab2bf0c2006-12-07 02:14:06 +0100114 if (probe_kernel_address(instr, opcode))
Harvey Harrison33cb5242008-01-30 13:32:19 +0100115 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
Harvey Harrison33cb5242008-01-30 13:32:19 +0100117 instr_hi = opcode & 0xf0;
118 instr_lo = opcode & 0x0f;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 instr++;
120
Harvey Harrison33cb5242008-01-30 13:32:19 +0100121 switch (instr_hi) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 case 0x20:
123 case 0x30:
Harvey Harrison33cb5242008-01-30 13:32:19 +0100124 /*
125 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
126 * In X86_64 long mode, the CPU will signal invalid
127 * opcode if some of these prefixes are present so
128 * X86_64 will never get here anyway
129 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 scan_more = ((instr_lo & 7) == 0x6);
131 break;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100132#ifdef CONFIG_X86_64
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 case 0x40:
Harvey Harrison33cb5242008-01-30 13:32:19 +0100134 /*
135 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
136 * Need to figure out under what instruction mode the
137 * instruction was issued. Could check the LDT for lm,
138 * but for now it's good enough to assume that long
139 * mode only uses well known segments or kernel.
140 */
Vincent Hanquez76381fe2005-06-23 00:08:46 -0700141 scan_more = (!user_mode(regs)) || (regs->cs == __USER_CS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 break;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100143#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 case 0x60:
145 /* 0x64 thru 0x67 are valid prefixes in all modes. */
146 scan_more = (instr_lo & 0xC) == 0x4;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100147 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 case 0xF0:
Harvey Harrison1dc85be2008-01-30 13:32:35 +0100149 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 scan_more = !instr_lo || (instr_lo>>1) == 1;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100151 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 case 0x00:
153 /* Prefetch instruction is 0x0F0D or 0x0F18 */
154 scan_more = 0;
Harvey Harrisonf2857ce2008-01-30 13:33:12 +0100155
Andi Kleenab2bf0c2006-12-07 02:14:06 +0100156 if (probe_kernel_address(instr, opcode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 break;
158 prefetch = (instr_lo == 0xF) &&
159 (opcode == 0x0D || opcode == 0x18);
Harvey Harrison33cb5242008-01-30 13:32:19 +0100160 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 default:
162 scan_more = 0;
163 break;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100164 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 }
166 return prefetch;
167}
168
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100169static void force_sig_info_fault(int si_signo, int si_code,
170 unsigned long address, struct task_struct *tsk)
171{
172 siginfo_t info;
173
174 info.si_signo = si_signo;
175 info.si_errno = 0;
176 info.si_code = si_code;
177 info.si_addr = (void __user *)address;
178 force_sig_info(si_signo, &info, tsk);
179}
180
Harvey Harrison1156e092008-01-30 13:34:10 +0100181#ifdef CONFIG_X86_64
Harvey Harrison33cb5242008-01-30 13:32:19 +0100182static int bad_address(void *p)
183{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 unsigned long dummy;
Andi Kleenab2bf0c2006-12-07 02:14:06 +0100185 return probe_kernel_address((unsigned long *)p, dummy);
Harvey Harrison33cb5242008-01-30 13:32:19 +0100186}
Harvey Harrison1156e092008-01-30 13:34:10 +0100187#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189void dump_pagetable(unsigned long address)
190{
Harvey Harrison1156e092008-01-30 13:34:10 +0100191#ifdef CONFIG_X86_32
192 __typeof__(pte_val(__pte(0))) page;
193
194 page = read_cr3();
195 page = ((__typeof__(page) *) __va(page))[address >> PGDIR_SHIFT];
196#ifdef CONFIG_X86_PAE
197 printk("*pdpt = %016Lx ", page);
198 if ((page >> PAGE_SHIFT) < max_low_pfn
199 && page & _PAGE_PRESENT) {
200 page &= PAGE_MASK;
201 page = ((__typeof__(page) *) __va(page))[(address >> PMD_SHIFT)
202 & (PTRS_PER_PMD - 1)];
203 printk(KERN_CONT "*pde = %016Lx ", page);
204 page &= ~_PAGE_NX;
205 }
206#else
207 printk("*pde = %08lx ", page);
208#endif
209
210 /*
211 * We must not directly access the pte in the highpte
212 * case if the page table is located in highmem.
213 * And let's rather not kmap-atomic the pte, just in case
214 * it's allocated already.
215 */
216 if ((page >> PAGE_SHIFT) < max_low_pfn
217 && (page & _PAGE_PRESENT)
218 && !(page & _PAGE_PSE)) {
219 page &= PAGE_MASK;
220 page = ((__typeof__(page) *) __va(page))[(address >> PAGE_SHIFT)
221 & (PTRS_PER_PTE - 1)];
222 printk("*pte = %0*Lx ", sizeof(page)*2, (u64)page);
223 }
224
225 printk("\n");
226#else /* CONFIG_X86_64 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 pgd_t *pgd;
228 pud_t *pud;
229 pmd_t *pmd;
230 pte_t *pte;
231
Glauber de Oliveira Costaf51c9452007-07-22 11:12:29 +0200232 pgd = (pgd_t *)read_cr3();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
Harvey Harrison33cb5242008-01-30 13:32:19 +0100234 pgd = __va((unsigned long)pgd & PHYSICAL_PAGE_MASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 pgd += pgd_index(address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 if (bad_address(pgd)) goto bad;
Jan Beulichd646bce2006-02-03 21:51:47 +0100237 printk("PGD %lx ", pgd_val(*pgd));
Harvey Harrison33cb5242008-01-30 13:32:19 +0100238 if (!pgd_present(*pgd)) goto ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Andi Kleend2ae5b52006-06-26 13:57:56 +0200240 pud = pud_offset(pgd, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 if (bad_address(pud)) goto bad;
242 printk("PUD %lx ", pud_val(*pud));
243 if (!pud_present(*pud)) goto ret;
244
245 pmd = pmd_offset(pud, address);
246 if (bad_address(pmd)) goto bad;
247 printk("PMD %lx ", pmd_val(*pmd));
Jan Beulichb1992df2007-10-19 20:35:03 +0200248 if (!pmd_present(*pmd) || pmd_large(*pmd)) goto ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
250 pte = pte_offset_kernel(pmd, address);
251 if (bad_address(pte)) goto bad;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100252 printk("PTE %lx", pte_val(*pte));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253ret:
254 printk("\n");
255 return;
256bad:
257 printk("BAD\n");
Harvey Harrison1156e092008-01-30 13:34:10 +0100258#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259}
260
Harvey Harrison1156e092008-01-30 13:34:10 +0100261#ifdef CONFIG_X86_32
262static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
263{
264 unsigned index = pgd_index(address);
265 pgd_t *pgd_k;
266 pud_t *pud, *pud_k;
267 pmd_t *pmd, *pmd_k;
268
269 pgd += index;
270 pgd_k = init_mm.pgd + index;
271
272 if (!pgd_present(*pgd_k))
273 return NULL;
274
275 /*
276 * set_pgd(pgd, *pgd_k); here would be useless on PAE
277 * and redundant with the set_pmd() on non-PAE. As would
278 * set_pud.
279 */
280
281 pud = pud_offset(pgd, address);
282 pud_k = pud_offset(pgd_k, address);
283 if (!pud_present(*pud_k))
284 return NULL;
285
286 pmd = pmd_offset(pud, address);
287 pmd_k = pmd_offset(pud_k, address);
288 if (!pmd_present(*pmd_k))
289 return NULL;
290 if (!pmd_present(*pmd)) {
291 set_pmd(pmd, *pmd_k);
292 arch_flush_lazy_mmu_mode();
293 } else
294 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
295 return pmd_k;
296}
297#endif
298
Harvey Harrison1dc85be2008-01-30 13:32:35 +0100299#ifdef CONFIG_X86_64
Harvey Harrison33cb5242008-01-30 13:32:19 +0100300static const char errata93_warning[] =
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
302KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n"
303KERN_ERR "******* Please consider a BIOS update.\n"
304KERN_ERR "******* Disabling USB legacy in the BIOS may also help.\n";
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100305#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306
307/* Workaround for K8 erratum #93 & buggy BIOS.
308 BIOS SMM functions are required to use a specific workaround
Harvey Harrison33cb5242008-01-30 13:32:19 +0100309 to avoid corruption of the 64bit RIP register on C stepping K8.
310 A lot of BIOS that didn't get tested properly miss this.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 The OS sees this as a page fault with the upper 32bits of RIP cleared.
312 Try to work around it here.
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100313 Note we only handle faults in kernel here.
314 Does nothing for X86_32
315 */
Harvey Harrison33cb5242008-01-30 13:32:19 +0100316static int is_errata93(struct pt_regs *regs, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317{
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100318#ifdef CONFIG_X86_64
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 static int warned;
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100320 if (address != regs->ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 return 0;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100322 if ((address >> 32) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 return 0;
324 address |= 0xffffffffUL << 32;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100325 if ((address >= (u64)_stext && address <= (u64)_etext) ||
326 (address >= MODULES_VADDR && address <= MODULES_END)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 if (!warned) {
Harvey Harrison33cb5242008-01-30 13:32:19 +0100328 printk(errata93_warning);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 warned = 1;
330 }
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100331 regs->ip = address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 return 1;
333 }
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100334#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 return 0;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100336}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
Harvey Harrison35f32662008-01-30 13:34:09 +0100338/*
339 * Work around K8 erratum #100 K8 in compat mode occasionally jumps to illegal
340 * addresses >4GB. We catch this in the page fault handler because these
341 * addresses are not reachable. Just detect this case and return. Any code
342 * segment in LDT is compatibility mode.
343 */
344static int is_errata100(struct pt_regs *regs, unsigned long address)
345{
346#ifdef CONFIG_X86_64
347 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) &&
348 (address >> 32))
349 return 1;
350#endif
351 return 0;
352}
353
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100354void do_invalid_op(struct pt_regs *, unsigned long);
355
356static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
357{
358#ifdef CONFIG_X86_F00F_BUG
359 unsigned long nr;
360 /*
361 * Pentium F0 0F C7 C8 bug workaround.
362 */
363 if (boot_cpu_data.f00f_bug) {
364 nr = (address - idt_descr.address) >> 3;
365
366 if (nr == 6) {
367 do_invalid_op(regs, 0);
368 return 1;
369 }
370 }
371#endif
372 return 0;
373}
374
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100375static void show_fault_oops(struct pt_regs *regs, unsigned long error_code,
376 unsigned long address)
377{
Harvey Harrison1156e092008-01-30 13:34:10 +0100378#ifdef CONFIG_X86_32
379 if (!oops_may_print())
380 return;
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100381#endif
Harvey Harrison1156e092008-01-30 13:34:10 +0100382
383#ifdef CONFIG_X86_PAE
384 if (error_code & PF_INSTR) {
Harvey Harrison93809be2008-02-01 17:49:43 +0100385 unsigned int level;
Harvey Harrison1156e092008-01-30 13:34:10 +0100386 pte_t *pte = lookup_address(address, &level);
387
388 if (pte && pte_present(*pte) && !pte_exec(*pte))
389 printk(KERN_CRIT "kernel tried to execute "
390 "NX-protected page - exploit attempt? "
391 "(uid: %d)\n", current->uid);
392 }
393#endif
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100394
Harvey Harrison1156e092008-01-30 13:34:10 +0100395 printk(KERN_ALERT "BUG: unable to handle kernel ");
396 if (address < PAGE_SIZE)
397 printk(KERN_CONT "NULL pointer dereference");
398 else
399 printk(KERN_CONT "paging request");
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100400#ifdef CONFIG_X86_32
Harvey Harrison1156e092008-01-30 13:34:10 +0100401 printk(KERN_CONT " at %08lx\n", address);
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100402#else
Harvey Harrison19f0dda2008-01-30 13:34:10 +0100403 printk(KERN_CONT " at %016lx\n", address);
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100404#endif
Harvey Harrison19f0dda2008-01-30 13:34:10 +0100405 printk(KERN_ALERT "IP:");
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100406 printk_address(regs->ip, 1);
407 dump_pagetable(address);
408}
409
Harvey Harrison1156e092008-01-30 13:34:10 +0100410#ifdef CONFIG_X86_64
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411static noinline void pgtable_bad(unsigned long address, struct pt_regs *regs,
412 unsigned long error_code)
413{
Jan Beulich12091402005-09-12 18:49:24 +0200414 unsigned long flags = oops_begin();
Jan Beulich6e3f3612006-01-11 22:42:14 +0100415 struct task_struct *tsk;
Jan Beulich12091402005-09-12 18:49:24 +0200416
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
418 current->comm, address);
419 dump_pagetable(address);
Jan Beulich6e3f3612006-01-11 22:42:14 +0100420 tsk = current;
421 tsk->thread.cr2 = address;
422 tsk->thread.trap_no = 14;
423 tsk->thread.error_code = error_code;
Jan Beulich22f59912008-01-30 13:31:23 +0100424 if (__die("Bad pagetable", regs, error_code))
425 regs = NULL;
426 oops_end(flags, regs, SIGKILL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427}
Harvey Harrison1156e092008-01-30 13:34:10 +0100428#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429
430/*
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100431 * Handle a spurious fault caused by a stale TLB entry. This allows
432 * us to lazily refresh the TLB when increasing the permissions of a
433 * kernel page (RO -> RW or NX -> X). Doing it eagerly is very
434 * expensive since that implies doing a full cross-processor TLB
435 * flush, even if no stale TLB entries exist on other processors.
436 * There are no security implications to leaving a stale TLB when
437 * increasing the permissions on a page.
438 */
439static int spurious_fault(unsigned long address,
440 unsigned long error_code)
441{
442 pgd_t *pgd;
443 pud_t *pud;
444 pmd_t *pmd;
445 pte_t *pte;
446
447 /* Reserved-bit violation or user access to kernel space? */
448 if (error_code & (PF_USER | PF_RSVD))
449 return 0;
450
451 pgd = init_mm.pgd + pgd_index(address);
452 if (!pgd_present(*pgd))
453 return 0;
454
455 pud = pud_offset(pgd, address);
456 if (!pud_present(*pud))
457 return 0;
458
459 pmd = pmd_offset(pud, address);
460 if (!pmd_present(*pmd))
461 return 0;
462
463 pte = pte_offset_kernel(pmd, address);
464 if (!pte_present(*pte))
465 return 0;
466
467 if ((error_code & PF_WRITE) && !pte_write(*pte))
468 return 0;
469 if ((error_code & PF_INSTR) && !pte_exec(*pte))
470 return 0;
471
472 return 1;
473}
474
475/*
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100476 * X86_32
477 * Handle a fault on the vmalloc or module mapping area
478 *
479 * X86_64
Andi Kleenf95190b2006-01-11 22:44:00 +0100480 * Handle a fault on the vmalloc area
Andi Kleen3b9ba4d2005-05-16 21:53:31 -0700481 *
482 * This assumes no large pages in there.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 */
484static int vmalloc_fault(unsigned long address)
485{
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100486#ifdef CONFIG_X86_32
487 unsigned long pgd_paddr;
488 pmd_t *pmd_k;
489 pte_t *pte_k;
490 /*
491 * Synchronize this task's top level page-table
492 * with the 'reference' page table.
493 *
494 * Do _not_ use "current" here. We might be inside
495 * an interrupt in the middle of a task switch..
496 */
497 pgd_paddr = read_cr3();
498 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
499 if (!pmd_k)
500 return -1;
501 pte_k = pte_offset_kernel(pmd_k, address);
502 if (!pte_present(*pte_k))
503 return -1;
504 return 0;
505#else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 pgd_t *pgd, *pgd_ref;
507 pud_t *pud, *pud_ref;
508 pmd_t *pmd, *pmd_ref;
509 pte_t *pte, *pte_ref;
510
Harvey Harrisoncf89ec92008-02-04 16:47:56 +0100511 /* Make sure we are in vmalloc area */
512 if (!(address >= VMALLOC_START && address < VMALLOC_END))
513 return -1;
514
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 /* Copy kernel mappings over when needed. This can also
516 happen within a race in page table update. In the later
517 case just flush. */
518
519 pgd = pgd_offset(current->mm ?: &init_mm, address);
520 pgd_ref = pgd_offset_k(address);
521 if (pgd_none(*pgd_ref))
522 return -1;
523 if (pgd_none(*pgd))
524 set_pgd(pgd, *pgd_ref);
Jan Beulich8c914cb2006-03-25 16:29:40 +0100525 else
Dave McCracken46a82b22006-09-25 23:31:48 -0700526 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
528 /* Below here mismatches are bugs because these lower tables
529 are shared */
530
531 pud = pud_offset(pgd, address);
532 pud_ref = pud_offset(pgd_ref, address);
533 if (pud_none(*pud_ref))
534 return -1;
Dave McCracken46a82b22006-09-25 23:31:48 -0700535 if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 BUG();
537 pmd = pmd_offset(pud, address);
538 pmd_ref = pmd_offset(pud_ref, address);
539 if (pmd_none(*pmd_ref))
540 return -1;
541 if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref))
542 BUG();
543 pte_ref = pte_offset_kernel(pmd_ref, address);
544 if (!pte_present(*pte_ref))
545 return -1;
546 pte = pte_offset_kernel(pmd, address);
Andi Kleen3b9ba4d2005-05-16 21:53:31 -0700547 /* Don't use pte_page here, because the mappings can point
548 outside mem_map, and the NUMA hash lookup cannot handle
549 that. */
550 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 return 0;
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100553#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554}
555
Masoud Asgharifard Sharbianiabd4f752007-07-22 11:12:28 +0200556int show_unhandled_signals = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
558/*
559 * This routine handles page faults. It determines the address,
560 * and the problem, and then passes it off to one of the appropriate
561 * routines.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 */
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100563#ifdef CONFIG_X86_64
564asmlinkage
565#endif
566void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567{
568 struct task_struct *tsk;
569 struct mm_struct *mm;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100570 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 unsigned long address;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100572 int write, si_code;
573 int fault;
574#ifdef CONFIG_X86_64
Jan Beulich12091402005-09-12 18:49:24 +0200575 unsigned long flags;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100576#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
Peter Zijlstra143a5d32007-10-25 14:01:10 +0200578 /*
579 * We can fault from pretty much anywhere, with unknown IRQ state.
580 */
581 trace_hardirqs_fixup();
582
Arjan van de Vena9ba9a32006-03-25 16:30:10 +0100583 tsk = current;
584 mm = tsk->mm;
585 prefetchw(&mm->mmap_sem);
586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 /* get the address */
Glauber de Oliveira Costaf51c9452007-07-22 11:12:29 +0200588 address = read_cr2();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100590 si_code = SEGV_MAPERR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
Harvey Harrison608566b2008-01-30 13:33:12 +0100592 if (notify_page_fault(regs))
593 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
595 /*
596 * We fault-in kernel-space virtual memory on-demand. The
597 * 'reference' page table is init_mm.pgd.
598 *
599 * NOTE! We MUST NOT take any locks for this case. We may
600 * be in an interrupt or a critical region, and should
601 * only copy the information from the master page table,
602 * nothing more.
603 *
604 * This verifies that the fault happens in kernel space
605 * (error_code & 4) == 0, and that the fault was not a
Jan Beulich8b1bde92006-01-11 22:42:23 +0100606 * protection error (error_code & 9) == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 */
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100608#ifdef CONFIG_X86_32
609 if (unlikely(address >= TASK_SIZE)) {
Harvey Harrisoncf89ec92008-02-04 16:47:56 +0100610#else
611 if (unlikely(address >= TASK_SIZE64)) {
612#endif
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100613 if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) &&
614 vmalloc_fault(address) >= 0)
615 return;
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100616
617 /* Can handle a stale RO->RW TLB */
618 if (spurious_fault(address, error_code))
619 return;
620
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100621 /*
622 * Don't take the mm semaphore here. If we fixup a prefetch
623 * fault we could otherwise deadlock.
624 */
625 goto bad_area_nosemaphore;
626 }
627
Harvey Harrisoncf89ec92008-02-04 16:47:56 +0100628
629#ifdef CONFIG_X86_32
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100630 /* It's safe to allow irq's after cr2 has been saved and the vmalloc
631 fault has been handled. */
632 if (regs->flags & (X86_EFLAGS_IF|VM_MASK))
633 local_irq_enable();
634
635 /*
636 * If we're in an interrupt, have no user context or are running in an
637 * atomic region then we must not take the fault.
638 */
639 if (in_atomic() || !mm)
640 goto bad_area_nosemaphore;
641#else /* CONFIG_X86_64 */
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100642 if (likely(regs->flags & X86_EFLAGS_IF))
Jan Beulich8c914cb2006-03-25 16:29:40 +0100643 local_irq_enable();
644
Andi Kleen66c58152006-01-11 22:44:09 +0100645 if (unlikely(error_code & PF_RSVD))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 pgtable_bad(address, regs, error_code);
647
648 /*
Harvey Harrison33cb5242008-01-30 13:32:19 +0100649 * If we're in an interrupt, have no user context or are running in an
650 * atomic region then we must not take the fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 */
652 if (unlikely(in_atomic() || !mm))
653 goto bad_area_nosemaphore;
654
Linus Torvaldsdbe3ed12007-09-19 11:37:14 -0700655 /*
656 * User-mode registers count as a user access even for any
657 * potential system fault or CPU buglet.
658 */
659 if (user_mode_vm(regs))
660 error_code |= PF_USER;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100661again:
662#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 /* When running in the kernel we expect faults to occur only to
664 * addresses in user space. All other faults represent errors in the
Simon Arlott676b1852007-10-20 01:25:36 +0200665 * kernel and should generate an OOPS. Unfortunately, in the case of an
Adrian Bunk80f72282006-06-30 18:27:16 +0200666 * erroneous fault occurring in a code path which already holds mmap_sem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 * we will deadlock attempting to validate the fault against the
668 * address space. Luckily the kernel only validly references user
669 * space from well defined areas of code, which are listed in the
670 * exceptions table.
671 *
672 * As the vast majority of faults will be valid we will only perform
Simon Arlott676b1852007-10-20 01:25:36 +0200673 * the source reference check when there is a possibility of a deadlock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 * Attempt to lock the address space, if we cannot we then validate the
675 * source. If this is invalid we can skip the address space check,
676 * thus avoiding the deadlock.
677 */
678 if (!down_read_trylock(&mm->mmap_sem)) {
Andi Kleen66c58152006-01-11 22:44:09 +0100679 if ((error_code & PF_USER) == 0 &&
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100680 !search_exception_tables(regs->ip))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 goto bad_area_nosemaphore;
682 down_read(&mm->mmap_sem);
683 }
684
685 vma = find_vma(mm, address);
686 if (!vma)
687 goto bad_area;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100688 if (vma->vm_start <= address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 goto good_area;
690 if (!(vma->vm_flags & VM_GROWSDOWN))
691 goto bad_area;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100692 if (error_code & PF_USER) {
Harvey Harrison6f4d3682008-01-30 13:33:13 +0100693 /*
694 * Accessing the stack below %sp is always a bug.
695 * The large cushion allows instructions like enter
696 * and pusha to work. ("enter $65535,$31" pushes
697 * 32 pointers and then decrements %sp by 65535.)
Chuck Ebbert03fdc2c2006-06-26 13:59:50 +0200698 */
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100699 if (address + 65536 + 32 * sizeof(unsigned long) < regs->sp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 goto bad_area;
701 }
702 if (expand_stack(vma, address))
703 goto bad_area;
704/*
705 * Ok, we have a good vm_area for this memory access, so
706 * we can handle it..
707 */
708good_area:
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100709 si_code = SEGV_ACCERR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 write = 0;
Andi Kleen66c58152006-01-11 22:44:09 +0100711 switch (error_code & (PF_PROT|PF_WRITE)) {
Harvey Harrison33cb5242008-01-30 13:32:19 +0100712 default: /* 3: write, present */
713 /* fall through */
714 case PF_WRITE: /* write, not present */
715 if (!(vma->vm_flags & VM_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 goto bad_area;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100717 write++;
718 break;
719 case PF_PROT: /* read, present */
720 goto bad_area;
721 case 0: /* read, not present */
722 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
723 goto bad_area;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 }
725
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100726#ifdef CONFIG_X86_32
727survive:
728#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 /*
730 * If for any reason at all we couldn't handle the fault,
731 * make sure we exit gracefully rather than endlessly redo
732 * the fault.
733 */
Nick Piggin83c54072007-07-19 01:47:05 -0700734 fault = handle_mm_fault(mm, vma, address, write);
735 if (unlikely(fault & VM_FAULT_ERROR)) {
736 if (fault & VM_FAULT_OOM)
737 goto out_of_memory;
738 else if (fault & VM_FAULT_SIGBUS)
739 goto do_sigbus;
740 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 }
Nick Piggin83c54072007-07-19 01:47:05 -0700742 if (fault & VM_FAULT_MAJOR)
743 tsk->maj_flt++;
744 else
745 tsk->min_flt++;
Harvey Harrisond729ab32008-01-30 13:33:23 +0100746
747#ifdef CONFIG_X86_32
748 /*
749 * Did it hit the DOS screen memory VA from vm86 mode?
750 */
751 if (v8086_mode(regs)) {
752 unsigned long bit = (address - 0xA0000) >> PAGE_SHIFT;
753 if (bit < 32)
754 tsk->thread.screen_bitmap |= 1 << bit;
755 }
756#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 up_read(&mm->mmap_sem);
758 return;
759
760/*
761 * Something tried to access memory that isn't in our memory map..
762 * Fix it, but check if it's kernel or user first..
763 */
764bad_area:
765 up_read(&mm->mmap_sem);
766
767bad_area_nosemaphore:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 /* User mode accesses just cause a SIGSEGV */
Andi Kleen66c58152006-01-11 22:44:09 +0100769 if (error_code & PF_USER) {
Steven Rostedte5e3c842007-06-06 23:34:04 -0400770 /*
771 * It's possible to have interrupts off here.
772 */
773 local_irq_enable();
774
Harvey Harrison1156e092008-01-30 13:34:10 +0100775 /*
776 * Valid to do another page fault here because this one came
777 * from user space.
778 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 if (is_prefetch(regs, address, error_code))
780 return;
781
Harvey Harrison35f32662008-01-30 13:34:09 +0100782 if (is_errata100(regs, address))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 return;
784
Masoud Asgharifard Sharbianiabd4f752007-07-22 11:12:28 +0200785 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
786 printk_ratelimit()) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 printk(
Harvey Harrison6f4d3682008-01-30 13:33:13 +0100788#ifdef CONFIG_X86_32
Harvey Harrisonedcd8112008-01-30 13:33:16 +0100789 "%s%s[%d]: segfault at %lx ip %08lx sp %08lx error %lx",
Harvey Harrison6f4d3682008-01-30 13:33:13 +0100790#else
Andi Kleen03252912008-01-30 13:33:18 +0100791 "%s%s[%d]: segfault at %lx ip %lx sp %lx error %lx",
Harvey Harrison6f4d3682008-01-30 13:33:13 +0100792#endif
793 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
794 tsk->comm, task_pid_nr(tsk), address, regs->ip,
795 regs->sp, error_code);
Andi Kleen03252912008-01-30 13:33:18 +0100796 print_vma_addr(" in ", regs->ip);
797 printk("\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 }
Harvey Harrison33cb5242008-01-30 13:32:19 +0100799
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 tsk->thread.cr2 = address;
801 /* Kernel addresses are always protection faults */
802 tsk->thread.error_code = error_code | (address >= TASK_SIZE);
803 tsk->thread.trap_no = 14;
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100804 force_sig_info_fault(SIGSEGV, si_code, address, tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 return;
806 }
807
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100808 if (is_f00f_bug(regs, address))
809 return;
810
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811no_context:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 /* Are we prepared to handle this kernel fault? */
Harvey Harrison33cb5242008-01-30 13:32:19 +0100813 if (fixup_exception(regs))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
Harvey Harrison33cb5242008-01-30 13:32:19 +0100816 /*
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100817 * X86_32
818 * Valid to do another page fault here, because if this fault
819 * had been triggered by is_prefetch fixup_exception would have
820 * handled it.
821 *
822 * X86_64
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 * Hall of shame of CPU/BIOS bugs.
824 */
Harvey Harrison33cb5242008-01-30 13:32:19 +0100825 if (is_prefetch(regs, address, error_code))
826 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
828 if (is_errata93(regs, address))
Harvey Harrison33cb5242008-01-30 13:32:19 +0100829 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
831/*
832 * Oops. The kernel tried to access some bad page. We'll have to
833 * terminate things with extreme prejudice.
834 */
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100835#ifdef CONFIG_X86_32
836 bust_spinlocks(1);
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100837#else
838 flags = oops_begin();
839#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100841 show_fault_oops(regs, error_code, address);
842
843 tsk->thread.cr2 = address;
844 tsk->thread.trap_no = 14;
845 tsk->thread.error_code = error_code;
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100846
847#ifdef CONFIG_X86_32
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100848 die("Oops", regs, error_code);
849 bust_spinlocks(0);
850 do_exit(SIGKILL);
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100851#else
Jan Beulich22f59912008-01-30 13:31:23 +0100852 if (__die("Oops", regs, error_code))
853 regs = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 /* Executive summary in case the body of the oops scrolled away */
855 printk(KERN_EMERG "CR2: %016lx\n", address);
Jan Beulich22f59912008-01-30 13:31:23 +0100856 oops_end(flags, regs, SIGKILL);
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100857#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
859/*
860 * We ran out of memory, or some other thing happened to us that made
861 * us unable to handle the page fault gracefully.
862 */
863out_of_memory:
864 up_read(&mm->mmap_sem);
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100865 if (is_global_init(tsk)) {
866 yield();
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100867#ifdef CONFIG_X86_32
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100868 down_read(&mm->mmap_sem);
869 goto survive;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100870#else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 goto again;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100872#endif
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100873 }
874
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 printk("VM: killing process %s\n", tsk->comm);
Harvey Harrison318aa292008-01-30 13:32:59 +0100876 if (error_code & PF_USER)
Will Schmidt021daae2007-07-21 17:11:17 +0200877 do_group_exit(SIGKILL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 goto no_context;
879
880do_sigbus:
881 up_read(&mm->mmap_sem);
882
883 /* Kernel mode? Handle exceptions or die */
Andi Kleen66c58152006-01-11 22:44:09 +0100884 if (!(error_code & PF_USER))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 goto no_context;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100886#ifdef CONFIG_X86_32
887 /* User space => ok to do another page fault */
888 if (is_prefetch(regs, address, error_code))
889 return;
890#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 tsk->thread.cr2 = address;
892 tsk->thread.error_code = error_code;
893 tsk->thread.trap_no = 14;
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100894 force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895}
Andi Kleen9e43e1b2005-11-05 17:25:54 +0100896
Jan Beulich8c914cb2006-03-25 16:29:40 +0100897DEFINE_SPINLOCK(pgd_lock);
Christoph Lameter2bff7382007-05-02 19:27:10 +0200898LIST_HEAD(pgd_list);
Jan Beulich8c914cb2006-03-25 16:29:40 +0100899
900void vmalloc_sync_all(void)
901{
Harvey Harrison1156e092008-01-30 13:34:10 +0100902#ifdef CONFIG_X86_32
903 /*
904 * Note that races in the updates of insync and start aren't
905 * problematic: insync can only get set bits added, and updates to
906 * start are only improving performance (without affecting correctness
907 * if undone).
908 */
909 static DECLARE_BITMAP(insync, PTRS_PER_PGD);
910 static unsigned long start = TASK_SIZE;
911 unsigned long address;
912
913 if (SHARED_KERNEL_PMD)
914 return;
915
916 BUILD_BUG_ON(TASK_SIZE & ~PGDIR_MASK);
917 for (address = start; address >= TASK_SIZE; address += PGDIR_SIZE) {
918 if (!test_bit(pgd_index(address), insync)) {
919 unsigned long flags;
920 struct page *page;
921
922 spin_lock_irqsave(&pgd_lock, flags);
Jeremy Fitzhardingee3ed9102008-01-30 13:34:11 +0100923 list_for_each_entry(page, &pgd_list, lru) {
Harvey Harrison1156e092008-01-30 13:34:10 +0100924 if (!vmalloc_sync_one(page_address(page),
Jeremy Fitzhardingee3ed9102008-01-30 13:34:11 +0100925 address))
Harvey Harrison1156e092008-01-30 13:34:10 +0100926 break;
Jeremy Fitzhardingee3ed9102008-01-30 13:34:11 +0100927 }
Harvey Harrison1156e092008-01-30 13:34:10 +0100928 spin_unlock_irqrestore(&pgd_lock, flags);
929 if (!page)
930 set_bit(pgd_index(address), insync);
931 }
932 if (address == start && test_bit(pgd_index(address), insync))
933 start = address + PGDIR_SIZE;
934 }
935#else /* CONFIG_X86_64 */
Harvey Harrison6f4d3682008-01-30 13:33:13 +0100936 /*
937 * Note that races in the updates of insync and start aren't
938 * problematic: insync can only get set bits added, and updates to
939 * start are only improving performance (without affecting correctness
940 * if undone).
941 */
Jan Beulich8c914cb2006-03-25 16:29:40 +0100942 static DECLARE_BITMAP(insync, PTRS_PER_PGD);
943 static unsigned long start = VMALLOC_START & PGDIR_MASK;
944 unsigned long address;
945
946 for (address = start; address <= VMALLOC_END; address += PGDIR_SIZE) {
947 if (!test_bit(pgd_index(address), insync)) {
948 const pgd_t *pgd_ref = pgd_offset_k(address);
949 struct page *page;
950
951 if (pgd_none(*pgd_ref))
952 continue;
953 spin_lock(&pgd_lock);
Christoph Lameter2bff7382007-05-02 19:27:10 +0200954 list_for_each_entry(page, &pgd_list, lru) {
Jan Beulich8c914cb2006-03-25 16:29:40 +0100955 pgd_t *pgd;
956 pgd = (pgd_t *)page_address(page) + pgd_index(address);
957 if (pgd_none(*pgd))
958 set_pgd(pgd, *pgd_ref);
959 else
Dave McCracken46a82b22006-09-25 23:31:48 -0700960 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
Jan Beulich8c914cb2006-03-25 16:29:40 +0100961 }
962 spin_unlock(&pgd_lock);
963 set_bit(pgd_index(address), insync);
964 }
965 if (address == start)
966 start = address + PGDIR_SIZE;
967 }
968 /* Check that there is no need to do the same for the modules area. */
969 BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL));
Harvey Harrison33cb5242008-01-30 13:32:19 +0100970 BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) ==
Jan Beulich8c914cb2006-03-25 16:29:40 +0100971 (__START_KERNEL & PGDIR_MASK)));
Harvey Harrison1156e092008-01-30 13:34:10 +0100972#endif
Jan Beulich8c914cb2006-03-25 16:29:40 +0100973}