]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 1995 Linus Torvalds |
3 | * Copyright (C) 2001,2002 Andi Kleen, SuSE Labs. | |
4 | */ | |
5 | ||
1da177e4 LT |
6 | #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> | |
1da177e4 LT |
16 | #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> | |
1eeb66a1 | 21 | #include <linux/vmalloc.h> |
1da177e4 | 22 | #include <linux/module.h> |
0f2fbdcb | 23 | #include <linux/kprobes.h> |
ab2bf0c1 | 24 | #include <linux/uaccess.h> |
1eeb66a1 | 25 | #include <linux/kdebug.h> |
1da177e4 LT |
26 | |
27 | #include <asm/system.h> | |
1da177e4 LT |
28 | #include <asm/pgalloc.h> |
29 | #include <asm/smp.h> | |
30 | #include <asm/tlbflush.h> | |
31 | #include <asm/proto.h> | |
1da177e4 | 32 | #include <asm-generic/sections.h> |
1da177e4 | 33 | |
33cb5243 HH |
34 | /* |
35 | * Page fault error code bits | |
36 | * bit 0 == 0 means no page found, 1 means protection fault | |
37 | * bit 1 == 0 means read, 1 means write | |
38 | * bit 2 == 0 means kernel, 1 means user-mode | |
39 | * bit 3 == 1 means use of reserved bit detected | |
40 | * bit 4 == 1 means fault was an instruction fetch | |
41 | */ | |
8a19da7b | 42 | #define PF_PROT (1<<0) |
66c58156 | 43 | #define PF_WRITE (1<<1) |
8a19da7b IM |
44 | #define PF_USER (1<<2) |
45 | #define PF_RSVD (1<<3) | |
66c58156 AK |
46 | #define PF_INSTR (1<<4) |
47 | ||
74a0b576 | 48 | static inline int notify_page_fault(struct pt_regs *regs) |
1bd858a5 | 49 | { |
33cb5243 | 50 | #ifdef CONFIG_KPROBES |
74a0b576 CH |
51 | int ret = 0; |
52 | ||
53 | /* kprobe_running() needs smp_processor_id() */ | |
54 | if (!user_mode(regs)) { | |
55 | preempt_disable(); | |
56 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
57 | ret = 1; | |
58 | preempt_enable(); | |
59 | } | |
1bd858a5 | 60 | |
74a0b576 | 61 | return ret; |
74a0b576 | 62 | #else |
74a0b576 | 63 | return 0; |
74a0b576 | 64 | #endif |
33cb5243 | 65 | } |
1bd858a5 | 66 | |
1dc85be0 HH |
67 | /* |
68 | * X86_32 | |
69 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
70 | * Check that here and ignore it. | |
71 | * | |
72 | * X86_64 | |
73 | * Sometimes the CPU reports invalid exceptions on prefetch. | |
74 | * Check that here and ignore it. | |
75 | * | |
76 | * Opcode checker based on code by Richard Brunner | |
77 | */ | |
78 | static int is_prefetch(struct pt_regs *regs, unsigned long addr, | |
79 | unsigned long error_code) | |
33cb5243 | 80 | { |
ab2bf0c1 | 81 | unsigned char *instr; |
1da177e4 | 82 | int scan_more = 1; |
33cb5243 | 83 | int prefetch = 0; |
f1290ec9 | 84 | unsigned char *max_instr; |
1da177e4 | 85 | |
1dc85be0 | 86 | #ifdef CONFIG_X86_32 |
1dc85be0 HH |
87 | if (unlikely(boot_cpu_data.x86_vendor == X86_VENDOR_AMD && |
88 | boot_cpu_data.x86 >= 6)) { | |
89 | /* Catch an obscure case of prefetch inside an NX page. */ | |
90 | if (nx_enabled && (error_code & PF_INSTR)) | |
91 | return 0; | |
92 | } else { | |
93 | return 0; | |
94 | } | |
1dc85be0 | 95 | #else |
1da177e4 | 96 | /* If it was a exec fault ignore */ |
66c58156 | 97 | if (error_code & PF_INSTR) |
1da177e4 | 98 | return 0; |
1dc85be0 HH |
99 | #endif |
100 | ||
f2857ce9 | 101 | instr = (unsigned char *)convert_ip_to_linear(current, regs); |
f1290ec9 | 102 | max_instr = instr + 15; |
1da177e4 | 103 | |
76381fee | 104 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE) |
1da177e4 LT |
105 | return 0; |
106 | ||
33cb5243 | 107 | while (scan_more && instr < max_instr) { |
1da177e4 LT |
108 | unsigned char opcode; |
109 | unsigned char instr_hi; | |
110 | unsigned char instr_lo; | |
111 | ||
ab2bf0c1 | 112 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 113 | break; |
1da177e4 | 114 | |
33cb5243 HH |
115 | instr_hi = opcode & 0xf0; |
116 | instr_lo = opcode & 0x0f; | |
1da177e4 LT |
117 | instr++; |
118 | ||
33cb5243 | 119 | switch (instr_hi) { |
1da177e4 LT |
120 | case 0x20: |
121 | case 0x30: | |
33cb5243 HH |
122 | /* |
123 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
124 | * In X86_64 long mode, the CPU will signal invalid | |
125 | * opcode if some of these prefixes are present so | |
126 | * X86_64 will never get here anyway | |
127 | */ | |
1da177e4 LT |
128 | scan_more = ((instr_lo & 7) == 0x6); |
129 | break; | |
33cb5243 | 130 | #ifdef CONFIG_X86_64 |
1da177e4 | 131 | case 0x40: |
33cb5243 HH |
132 | /* |
133 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
134 | * Need to figure out under what instruction mode the | |
135 | * instruction was issued. Could check the LDT for lm, | |
136 | * but for now it's good enough to assume that long | |
137 | * mode only uses well known segments or kernel. | |
138 | */ | |
76381fee | 139 | scan_more = (!user_mode(regs)) || (regs->cs == __USER_CS); |
1da177e4 | 140 | break; |
33cb5243 | 141 | #endif |
1da177e4 LT |
142 | case 0x60: |
143 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
144 | scan_more = (instr_lo & 0xC) == 0x4; | |
33cb5243 | 145 | break; |
1da177e4 | 146 | case 0xF0: |
1dc85be0 | 147 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ |
1da177e4 | 148 | scan_more = !instr_lo || (instr_lo>>1) == 1; |
33cb5243 | 149 | break; |
1da177e4 LT |
150 | case 0x00: |
151 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
152 | scan_more = 0; | |
f2857ce9 | 153 | |
ab2bf0c1 | 154 | if (probe_kernel_address(instr, opcode)) |
1da177e4 LT |
155 | break; |
156 | prefetch = (instr_lo == 0xF) && | |
157 | (opcode == 0x0D || opcode == 0x18); | |
33cb5243 | 158 | break; |
1da177e4 LT |
159 | default: |
160 | scan_more = 0; | |
161 | break; | |
33cb5243 | 162 | } |
1da177e4 LT |
163 | } |
164 | return prefetch; | |
165 | } | |
166 | ||
c4aba4a8 HH |
167 | static void force_sig_info_fault(int si_signo, int si_code, |
168 | unsigned long address, struct task_struct *tsk) | |
169 | { | |
170 | siginfo_t info; | |
171 | ||
172 | info.si_signo = si_signo; | |
173 | info.si_errno = 0; | |
174 | info.si_code = si_code; | |
175 | info.si_addr = (void __user *)address; | |
176 | force_sig_info(si_signo, &info, tsk); | |
177 | } | |
178 | ||
33cb5243 HH |
179 | static int bad_address(void *p) |
180 | { | |
1da177e4 | 181 | unsigned long dummy; |
ab2bf0c1 | 182 | return probe_kernel_address((unsigned long *)p, dummy); |
33cb5243 | 183 | } |
1da177e4 LT |
184 | |
185 | void dump_pagetable(unsigned long address) | |
186 | { | |
187 | pgd_t *pgd; | |
188 | pud_t *pud; | |
189 | pmd_t *pmd; | |
190 | pte_t *pte; | |
191 | ||
f51c9452 | 192 | pgd = (pgd_t *)read_cr3(); |
1da177e4 | 193 | |
33cb5243 | 194 | pgd = __va((unsigned long)pgd & PHYSICAL_PAGE_MASK); |
1da177e4 | 195 | pgd += pgd_index(address); |
1da177e4 | 196 | if (bad_address(pgd)) goto bad; |
d646bce4 | 197 | printk("PGD %lx ", pgd_val(*pgd)); |
33cb5243 | 198 | if (!pgd_present(*pgd)) goto ret; |
1da177e4 | 199 | |
d2ae5b5f | 200 | pud = pud_offset(pgd, address); |
1da177e4 LT |
201 | if (bad_address(pud)) goto bad; |
202 | printk("PUD %lx ", pud_val(*pud)); | |
203 | if (!pud_present(*pud)) goto ret; | |
204 | ||
205 | pmd = pmd_offset(pud, address); | |
206 | if (bad_address(pmd)) goto bad; | |
207 | printk("PMD %lx ", pmd_val(*pmd)); | |
b1992df3 | 208 | if (!pmd_present(*pmd) || pmd_large(*pmd)) goto ret; |
1da177e4 LT |
209 | |
210 | pte = pte_offset_kernel(pmd, address); | |
211 | if (bad_address(pte)) goto bad; | |
33cb5243 | 212 | printk("PTE %lx", pte_val(*pte)); |
1da177e4 LT |
213 | ret: |
214 | printk("\n"); | |
215 | return; | |
216 | bad: | |
217 | printk("BAD\n"); | |
218 | } | |
219 | ||
1dc85be0 | 220 | #ifdef CONFIG_X86_64 |
33cb5243 | 221 | static const char errata93_warning[] = |
1da177e4 LT |
222 | KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n" |
223 | KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n" | |
224 | KERN_ERR "******* Please consider a BIOS update.\n" | |
225 | KERN_ERR "******* Disabling USB legacy in the BIOS may also help.\n"; | |
fdfe8aa8 | 226 | #endif |
1da177e4 LT |
227 | |
228 | /* Workaround for K8 erratum #93 & buggy BIOS. | |
229 | BIOS SMM functions are required to use a specific workaround | |
33cb5243 HH |
230 | to avoid corruption of the 64bit RIP register on C stepping K8. |
231 | A lot of BIOS that didn't get tested properly miss this. | |
1da177e4 LT |
232 | The OS sees this as a page fault with the upper 32bits of RIP cleared. |
233 | Try to work around it here. | |
fdfe8aa8 HH |
234 | Note we only handle faults in kernel here. |
235 | Does nothing for X86_32 | |
236 | */ | |
33cb5243 | 237 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 238 | { |
fdfe8aa8 | 239 | #ifdef CONFIG_X86_64 |
1da177e4 | 240 | static int warned; |
65ea5b03 | 241 | if (address != regs->ip) |
1da177e4 | 242 | return 0; |
33cb5243 | 243 | if ((address >> 32) != 0) |
1da177e4 LT |
244 | return 0; |
245 | address |= 0xffffffffUL << 32; | |
33cb5243 HH |
246 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
247 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
1da177e4 | 248 | if (!warned) { |
33cb5243 | 249 | printk(errata93_warning); |
1da177e4 LT |
250 | warned = 1; |
251 | } | |
65ea5b03 | 252 | regs->ip = address; |
1da177e4 LT |
253 | return 1; |
254 | } | |
fdfe8aa8 | 255 | #endif |
1da177e4 | 256 | return 0; |
33cb5243 | 257 | } |
1da177e4 | 258 | |
1da177e4 LT |
259 | static noinline void pgtable_bad(unsigned long address, struct pt_regs *regs, |
260 | unsigned long error_code) | |
261 | { | |
1209140c | 262 | unsigned long flags = oops_begin(); |
6e3f3617 | 263 | struct task_struct *tsk; |
1209140c | 264 | |
1da177e4 LT |
265 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
266 | current->comm, address); | |
267 | dump_pagetable(address); | |
6e3f3617 JB |
268 | tsk = current; |
269 | tsk->thread.cr2 = address; | |
270 | tsk->thread.trap_no = 14; | |
271 | tsk->thread.error_code = error_code; | |
22f5991c JB |
272 | if (__die("Bad pagetable", regs, error_code)) |
273 | regs = NULL; | |
274 | oops_end(flags, regs, SIGKILL); | |
1da177e4 LT |
275 | } |
276 | ||
277 | /* | |
f95190b2 | 278 | * Handle a fault on the vmalloc area |
3b9ba4d5 AK |
279 | * |
280 | * This assumes no large pages in there. | |
1da177e4 LT |
281 | */ |
282 | static int vmalloc_fault(unsigned long address) | |
283 | { | |
fdfe8aa8 HH |
284 | #ifdef CONFIG_X86_32 |
285 | unsigned long pgd_paddr; | |
286 | pmd_t *pmd_k; | |
287 | pte_t *pte_k; | |
288 | /* | |
289 | * Synchronize this task's top level page-table | |
290 | * with the 'reference' page table. | |
291 | * | |
292 | * Do _not_ use "current" here. We might be inside | |
293 | * an interrupt in the middle of a task switch.. | |
294 | */ | |
295 | pgd_paddr = read_cr3(); | |
296 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
297 | if (!pmd_k) | |
298 | return -1; | |
299 | pte_k = pte_offset_kernel(pmd_k, address); | |
300 | if (!pte_present(*pte_k)) | |
301 | return -1; | |
302 | return 0; | |
303 | #else | |
1da177e4 LT |
304 | pgd_t *pgd, *pgd_ref; |
305 | pud_t *pud, *pud_ref; | |
306 | pmd_t *pmd, *pmd_ref; | |
307 | pte_t *pte, *pte_ref; | |
308 | ||
309 | /* Copy kernel mappings over when needed. This can also | |
310 | happen within a race in page table update. In the later | |
311 | case just flush. */ | |
312 | ||
313 | pgd = pgd_offset(current->mm ?: &init_mm, address); | |
314 | pgd_ref = pgd_offset_k(address); | |
315 | if (pgd_none(*pgd_ref)) | |
316 | return -1; | |
317 | if (pgd_none(*pgd)) | |
318 | set_pgd(pgd, *pgd_ref); | |
8c914cb7 | 319 | else |
46a82b2d | 320 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
1da177e4 LT |
321 | |
322 | /* Below here mismatches are bugs because these lower tables | |
323 | are shared */ | |
324 | ||
325 | pud = pud_offset(pgd, address); | |
326 | pud_ref = pud_offset(pgd_ref, address); | |
327 | if (pud_none(*pud_ref)) | |
328 | return -1; | |
46a82b2d | 329 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) |
1da177e4 LT |
330 | BUG(); |
331 | pmd = pmd_offset(pud, address); | |
332 | pmd_ref = pmd_offset(pud_ref, address); | |
333 | if (pmd_none(*pmd_ref)) | |
334 | return -1; | |
335 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) | |
336 | BUG(); | |
337 | pte_ref = pte_offset_kernel(pmd_ref, address); | |
338 | if (!pte_present(*pte_ref)) | |
339 | return -1; | |
340 | pte = pte_offset_kernel(pmd, address); | |
3b9ba4d5 AK |
341 | /* Don't use pte_page here, because the mappings can point |
342 | outside mem_map, and the NUMA hash lookup cannot handle | |
343 | that. */ | |
344 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
1da177e4 | 345 | BUG(); |
1da177e4 | 346 | return 0; |
fdfe8aa8 | 347 | #endif |
1da177e4 LT |
348 | } |
349 | ||
abd4f750 | 350 | int show_unhandled_signals = 1; |
1da177e4 LT |
351 | |
352 | /* | |
353 | * This routine handles page faults. It determines the address, | |
354 | * and the problem, and then passes it off to one of the appropriate | |
355 | * routines. | |
1da177e4 | 356 | */ |
0f2fbdcb PP |
357 | asmlinkage void __kprobes do_page_fault(struct pt_regs *regs, |
358 | unsigned long error_code) | |
1da177e4 LT |
359 | { |
360 | struct task_struct *tsk; | |
361 | struct mm_struct *mm; | |
33cb5243 | 362 | struct vm_area_struct *vma; |
1da177e4 | 363 | unsigned long address; |
83c54070 | 364 | int write, fault; |
1209140c | 365 | unsigned long flags; |
c4aba4a8 | 366 | int si_code; |
1da177e4 | 367 | |
143a5d32 PZ |
368 | /* |
369 | * We can fault from pretty much anywhere, with unknown IRQ state. | |
370 | */ | |
371 | trace_hardirqs_fixup(); | |
372 | ||
a9ba9a3b AV |
373 | tsk = current; |
374 | mm = tsk->mm; | |
375 | prefetchw(&mm->mmap_sem); | |
376 | ||
1da177e4 | 377 | /* get the address */ |
f51c9452 | 378 | address = read_cr2(); |
1da177e4 | 379 | |
c4aba4a8 | 380 | si_code = SEGV_MAPERR; |
1da177e4 | 381 | |
608566b4 HH |
382 | if (notify_page_fault(regs)) |
383 | return; | |
1da177e4 LT |
384 | |
385 | /* | |
386 | * We fault-in kernel-space virtual memory on-demand. The | |
387 | * 'reference' page table is init_mm.pgd. | |
388 | * | |
389 | * NOTE! We MUST NOT take any locks for this case. We may | |
390 | * be in an interrupt or a critical region, and should | |
391 | * only copy the information from the master page table, | |
392 | * nothing more. | |
393 | * | |
394 | * This verifies that the fault happens in kernel space | |
395 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 396 | * protection error (error_code & 9) == 0. |
1da177e4 | 397 | */ |
84929801 | 398 | if (unlikely(address >= TASK_SIZE64)) { |
f95190b2 AK |
399 | /* |
400 | * Don't check for the module range here: its PML4 | |
401 | * is always initialized because it's shared with the main | |
402 | * kernel text. Only vmalloc may need PML4 syncups. | |
403 | */ | |
66c58156 | 404 | if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) && |
f95190b2 | 405 | ((address >= VMALLOC_START && address < VMALLOC_END))) { |
8c914cb7 JB |
406 | if (vmalloc_fault(address) >= 0) |
407 | return; | |
1da177e4 LT |
408 | } |
409 | /* | |
410 | * Don't take the mm semaphore here. If we fixup a prefetch | |
411 | * fault we could otherwise deadlock. | |
412 | */ | |
413 | goto bad_area_nosemaphore; | |
414 | } | |
415 | ||
65ea5b03 | 416 | if (likely(regs->flags & X86_EFLAGS_IF)) |
8c914cb7 JB |
417 | local_irq_enable(); |
418 | ||
66c58156 | 419 | if (unlikely(error_code & PF_RSVD)) |
1da177e4 LT |
420 | pgtable_bad(address, regs, error_code); |
421 | ||
422 | /* | |
33cb5243 HH |
423 | * If we're in an interrupt, have no user context or are running in an |
424 | * atomic region then we must not take the fault. | |
1da177e4 LT |
425 | */ |
426 | if (unlikely(in_atomic() || !mm)) | |
427 | goto bad_area_nosemaphore; | |
428 | ||
dbe3ed1c LT |
429 | /* |
430 | * User-mode registers count as a user access even for any | |
431 | * potential system fault or CPU buglet. | |
432 | */ | |
433 | if (user_mode_vm(regs)) | |
434 | error_code |= PF_USER; | |
435 | ||
1da177e4 LT |
436 | again: |
437 | /* When running in the kernel we expect faults to occur only to | |
438 | * addresses in user space. All other faults represent errors in the | |
676b1855 | 439 | * kernel and should generate an OOPS. Unfortunately, in the case of an |
80f7228b | 440 | * erroneous fault occurring in a code path which already holds mmap_sem |
1da177e4 LT |
441 | * we will deadlock attempting to validate the fault against the |
442 | * address space. Luckily the kernel only validly references user | |
443 | * space from well defined areas of code, which are listed in the | |
444 | * exceptions table. | |
445 | * | |
446 | * As the vast majority of faults will be valid we will only perform | |
676b1855 | 447 | * the source reference check when there is a possibility of a deadlock. |
1da177e4 LT |
448 | * Attempt to lock the address space, if we cannot we then validate the |
449 | * source. If this is invalid we can skip the address space check, | |
450 | * thus avoiding the deadlock. | |
451 | */ | |
452 | if (!down_read_trylock(&mm->mmap_sem)) { | |
66c58156 | 453 | if ((error_code & PF_USER) == 0 && |
65ea5b03 | 454 | !search_exception_tables(regs->ip)) |
1da177e4 LT |
455 | goto bad_area_nosemaphore; |
456 | down_read(&mm->mmap_sem); | |
457 | } | |
458 | ||
459 | vma = find_vma(mm, address); | |
460 | if (!vma) | |
461 | goto bad_area; | |
462 | if (likely(vma->vm_start <= address)) | |
463 | goto good_area; | |
464 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
465 | goto bad_area; | |
33cb5243 | 466 | if (error_code & PF_USER) { |
6f4d368e HH |
467 | /* |
468 | * Accessing the stack below %sp is always a bug. | |
469 | * The large cushion allows instructions like enter | |
470 | * and pusha to work. ("enter $65535,$31" pushes | |
471 | * 32 pointers and then decrements %sp by 65535.) | |
03fdc2c2 | 472 | */ |
65ea5b03 | 473 | if (address + 65536 + 32 * sizeof(unsigned long) < regs->sp) |
1da177e4 LT |
474 | goto bad_area; |
475 | } | |
476 | if (expand_stack(vma, address)) | |
477 | goto bad_area; | |
478 | /* | |
479 | * Ok, we have a good vm_area for this memory access, so | |
480 | * we can handle it.. | |
481 | */ | |
482 | good_area: | |
c4aba4a8 | 483 | si_code = SEGV_ACCERR; |
1da177e4 | 484 | write = 0; |
66c58156 | 485 | switch (error_code & (PF_PROT|PF_WRITE)) { |
33cb5243 HH |
486 | default: /* 3: write, present */ |
487 | /* fall through */ | |
488 | case PF_WRITE: /* write, not present */ | |
489 | if (!(vma->vm_flags & VM_WRITE)) | |
490 | goto bad_area; | |
491 | write++; | |
492 | break; | |
493 | case PF_PROT: /* read, present */ | |
494 | goto bad_area; | |
495 | case 0: /* read, not present */ | |
496 | if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))) | |
1da177e4 | 497 | goto bad_area; |
1da177e4 LT |
498 | } |
499 | ||
500 | /* | |
501 | * If for any reason at all we couldn't handle the fault, | |
502 | * make sure we exit gracefully rather than endlessly redo | |
503 | * the fault. | |
504 | */ | |
83c54070 NP |
505 | fault = handle_mm_fault(mm, vma, address, write); |
506 | if (unlikely(fault & VM_FAULT_ERROR)) { | |
507 | if (fault & VM_FAULT_OOM) | |
508 | goto out_of_memory; | |
509 | else if (fault & VM_FAULT_SIGBUS) | |
510 | goto do_sigbus; | |
511 | BUG(); | |
1da177e4 | 512 | } |
83c54070 NP |
513 | if (fault & VM_FAULT_MAJOR) |
514 | tsk->maj_flt++; | |
515 | else | |
516 | tsk->min_flt++; | |
1da177e4 LT |
517 | up_read(&mm->mmap_sem); |
518 | return; | |
519 | ||
520 | /* | |
521 | * Something tried to access memory that isn't in our memory map.. | |
522 | * Fix it, but check if it's kernel or user first.. | |
523 | */ | |
524 | bad_area: | |
525 | up_read(&mm->mmap_sem); | |
526 | ||
527 | bad_area_nosemaphore: | |
1da177e4 | 528 | /* User mode accesses just cause a SIGSEGV */ |
66c58156 | 529 | if (error_code & PF_USER) { |
e5e3c84b SR |
530 | |
531 | /* | |
532 | * It's possible to have interrupts off here. | |
533 | */ | |
534 | local_irq_enable(); | |
535 | ||
1da177e4 LT |
536 | if (is_prefetch(regs, address, error_code)) |
537 | return; | |
538 | ||
539 | /* Work around K8 erratum #100 K8 in compat mode | |
540 | occasionally jumps to illegal addresses >4GB. We | |
541 | catch this here in the page fault handler because | |
542 | these addresses are not reachable. Just detect this | |
543 | case and return. Any code segment in LDT is | |
544 | compatibility mode. */ | |
545 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && | |
546 | (address >> 32)) | |
547 | return; | |
548 | ||
abd4f750 MAS |
549 | if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) && |
550 | printk_ratelimit()) { | |
1da177e4 | 551 | printk( |
6f4d368e | 552 | #ifdef CONFIG_X86_32 |
edcd8119 | 553 | "%s%s[%d]: segfault at %lx ip %08lx sp %08lx error %lx", |
6f4d368e | 554 | #else |
03252919 | 555 | "%s%s[%d]: segfault at %lx ip %lx sp %lx error %lx", |
6f4d368e HH |
556 | #endif |
557 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, | |
558 | tsk->comm, task_pid_nr(tsk), address, regs->ip, | |
559 | regs->sp, error_code); | |
03252919 AK |
560 | print_vma_addr(" in ", regs->ip); |
561 | printk("\n"); | |
1da177e4 | 562 | } |
33cb5243 | 563 | |
1da177e4 LT |
564 | tsk->thread.cr2 = address; |
565 | /* Kernel addresses are always protection faults */ | |
566 | tsk->thread.error_code = error_code | (address >= TASK_SIZE); | |
567 | tsk->thread.trap_no = 14; | |
c4aba4a8 HH |
568 | |
569 | force_sig_info_fault(SIGSEGV, si_code, address, tsk); | |
1da177e4 LT |
570 | return; |
571 | } | |
572 | ||
573 | no_context: | |
1da177e4 | 574 | /* Are we prepared to handle this kernel fault? */ |
33cb5243 | 575 | if (fixup_exception(regs)) |
1da177e4 | 576 | return; |
1da177e4 | 577 | |
33cb5243 | 578 | /* |
1da177e4 LT |
579 | * Hall of shame of CPU/BIOS bugs. |
580 | */ | |
581 | ||
33cb5243 HH |
582 | if (is_prefetch(regs, address, error_code)) |
583 | return; | |
1da177e4 LT |
584 | |
585 | if (is_errata93(regs, address)) | |
33cb5243 | 586 | return; |
1da177e4 LT |
587 | |
588 | /* | |
589 | * Oops. The kernel tried to access some bad page. We'll have to | |
590 | * terminate things with extreme prejudice. | |
591 | */ | |
592 | ||
1209140c | 593 | flags = oops_begin(); |
1da177e4 LT |
594 | |
595 | if (address < PAGE_SIZE) | |
596 | printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference"); | |
597 | else | |
598 | printk(KERN_ALERT "Unable to handle kernel paging request"); | |
33cb5243 | 599 | printk(" at %016lx RIP: \n" KERN_ALERT, address); |
bc850d6b | 600 | printk_address(regs->ip, regs->bp); |
1da177e4 | 601 | dump_pagetable(address); |
6e3f3617 JB |
602 | tsk->thread.cr2 = address; |
603 | tsk->thread.trap_no = 14; | |
604 | tsk->thread.error_code = error_code; | |
22f5991c JB |
605 | if (__die("Oops", regs, error_code)) |
606 | regs = NULL; | |
1da177e4 LT |
607 | /* Executive summary in case the body of the oops scrolled away */ |
608 | printk(KERN_EMERG "CR2: %016lx\n", address); | |
22f5991c | 609 | oops_end(flags, regs, SIGKILL); |
1da177e4 LT |
610 | |
611 | /* | |
612 | * We ran out of memory, or some other thing happened to us that made | |
613 | * us unable to handle the page fault gracefully. | |
614 | */ | |
615 | out_of_memory: | |
616 | up_read(&mm->mmap_sem); | |
b460cbc5 | 617 | if (is_global_init(current)) { |
1da177e4 LT |
618 | yield(); |
619 | goto again; | |
620 | } | |
621 | printk("VM: killing process %s\n", tsk->comm); | |
318aa296 | 622 | if (error_code & PF_USER) |
021daae2 | 623 | do_group_exit(SIGKILL); |
1da177e4 LT |
624 | goto no_context; |
625 | ||
626 | do_sigbus: | |
627 | up_read(&mm->mmap_sem); | |
628 | ||
629 | /* Kernel mode? Handle exceptions or die */ | |
66c58156 | 630 | if (!(error_code & PF_USER)) |
1da177e4 LT |
631 | goto no_context; |
632 | ||
633 | tsk->thread.cr2 = address; | |
634 | tsk->thread.error_code = error_code; | |
635 | tsk->thread.trap_no = 14; | |
c4aba4a8 | 636 | force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk); |
1da177e4 LT |
637 | return; |
638 | } | |
9e43e1b7 | 639 | |
8c914cb7 | 640 | DEFINE_SPINLOCK(pgd_lock); |
2bff7383 | 641 | LIST_HEAD(pgd_list); |
8c914cb7 JB |
642 | |
643 | void vmalloc_sync_all(void) | |
644 | { | |
6f4d368e HH |
645 | /* |
646 | * Note that races in the updates of insync and start aren't | |
647 | * problematic: insync can only get set bits added, and updates to | |
648 | * start are only improving performance (without affecting correctness | |
649 | * if undone). | |
650 | */ | |
8c914cb7 JB |
651 | static DECLARE_BITMAP(insync, PTRS_PER_PGD); |
652 | static unsigned long start = VMALLOC_START & PGDIR_MASK; | |
653 | unsigned long address; | |
654 | ||
655 | for (address = start; address <= VMALLOC_END; address += PGDIR_SIZE) { | |
656 | if (!test_bit(pgd_index(address), insync)) { | |
657 | const pgd_t *pgd_ref = pgd_offset_k(address); | |
658 | struct page *page; | |
659 | ||
660 | if (pgd_none(*pgd_ref)) | |
661 | continue; | |
662 | spin_lock(&pgd_lock); | |
2bff7383 | 663 | list_for_each_entry(page, &pgd_list, lru) { |
8c914cb7 JB |
664 | pgd_t *pgd; |
665 | pgd = (pgd_t *)page_address(page) + pgd_index(address); | |
666 | if (pgd_none(*pgd)) | |
667 | set_pgd(pgd, *pgd_ref); | |
668 | else | |
46a82b2d | 669 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
8c914cb7 JB |
670 | } |
671 | spin_unlock(&pgd_lock); | |
672 | set_bit(pgd_index(address), insync); | |
673 | } | |
674 | if (address == start) | |
675 | start = address + PGDIR_SIZE; | |
676 | } | |
677 | /* Check that there is no need to do the same for the modules area. */ | |
678 | BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL)); | |
33cb5243 | 679 | BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) == |
8c914cb7 JB |
680 | (__START_KERNEL & PGDIR_MASK))); |
681 | } |