Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 1995 Linus Torvalds |
3 | */ | |
4 | ||
5 | #include <linux/signal.h> | |
6 | #include <linux/sched.h> | |
7 | #include <linux/kernel.h> | |
8 | #include <linux/errno.h> | |
9 | #include <linux/string.h> | |
10 | #include <linux/types.h> | |
11 | #include <linux/ptrace.h> | |
12 | #include <linux/mman.h> | |
13 | #include <linux/mm.h> | |
14 | #include <linux/smp.h> | |
1da177e4 LT |
15 | #include <linux/interrupt.h> |
16 | #include <linux/init.h> | |
17 | #include <linux/tty.h> | |
18 | #include <linux/vt_kern.h> /* For unblank_screen() */ | |
19 | #include <linux/highmem.h> | |
28609f6e | 20 | #include <linux/bootmem.h> /* for max_low_pfn */ |
1eeb66a1 | 21 | #include <linux/vmalloc.h> |
1da177e4 | 22 | #include <linux/module.h> |
3d97ae5b | 23 | #include <linux/kprobes.h> |
11a4180c | 24 | #include <linux/uaccess.h> |
1eeb66a1 | 25 | #include <linux/kdebug.h> |
1da177e4 LT |
26 | |
27 | #include <asm/system.h> | |
1da177e4 | 28 | #include <asm/desc.h> |
78be3706 | 29 | #include <asm/segment.h> |
1da177e4 | 30 | |
33cb5243 HH |
31 | /* |
32 | * Page fault error code bits | |
33 | * bit 0 == 0 means no page found, 1 means protection fault | |
34 | * bit 1 == 0 means read, 1 means write | |
35 | * bit 2 == 0 means kernel, 1 means user-mode | |
36 | * bit 3 == 1 means use of reserved bit detected | |
37 | * bit 4 == 1 means fault was an instruction fetch | |
38 | */ | |
6f4d368e | 39 | #define PF_PROT (1<<0) |
33cb5243 | 40 | #define PF_WRITE (1<<1) |
6f4d368e HH |
41 | #define PF_USER (1<<2) |
42 | #define PF_RSVD (1<<3) | |
33cb5243 HH |
43 | #define PF_INSTR (1<<4) |
44 | ||
74a0b576 | 45 | static inline int notify_page_fault(struct pt_regs *regs) |
b71b5b65 | 46 | { |
33cb5243 | 47 | #ifdef CONFIG_KPROBES |
74a0b576 CH |
48 | int ret = 0; |
49 | ||
50 | /* kprobe_running() needs smp_processor_id() */ | |
51 | if (!user_mode_vm(regs)) { | |
52 | preempt_disable(); | |
53 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
54 | ret = 1; | |
55 | preempt_enable(); | |
56 | } | |
b71b5b65 | 57 | |
74a0b576 | 58 | return ret; |
74a0b576 | 59 | #else |
74a0b576 | 60 | return 0; |
74a0b576 | 61 | #endif |
33cb5243 | 62 | } |
b71b5b65 | 63 | |
33cb5243 | 64 | /* |
1dc85be0 | 65 | * X86_32 |
1da177e4 LT |
66 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. |
67 | * Check that here and ignore it. | |
1dc85be0 HH |
68 | * |
69 | * X86_64 | |
70 | * Sometimes the CPU reports invalid exceptions on prefetch. | |
71 | * Check that here and ignore it. | |
72 | * | |
73 | * Opcode checker based on code by Richard Brunner | |
1da177e4 | 74 | */ |
1dc85be0 HH |
75 | static int is_prefetch(struct pt_regs *regs, unsigned long addr, |
76 | unsigned long error_code) | |
33cb5243 | 77 | { |
1dc85be0 | 78 | unsigned char *instr; |
1da177e4 | 79 | int scan_more = 1; |
33cb5243 | 80 | int prefetch = 0; |
1dc85be0 HH |
81 | unsigned char *max_instr; |
82 | ||
83 | #ifdef CONFIG_X86_32 | |
1dc85be0 HH |
84 | if (unlikely(boot_cpu_data.x86_vendor == X86_VENDOR_AMD && |
85 | boot_cpu_data.x86 >= 6)) { | |
86 | /* Catch an obscure case of prefetch inside an NX page. */ | |
87 | if (nx_enabled && (error_code & PF_INSTR)) | |
88 | return 0; | |
89 | } else { | |
90 | return 0; | |
91 | } | |
1dc85be0 HH |
92 | #else |
93 | /* If it was a exec fault ignore */ | |
94 | if (error_code & PF_INSTR) | |
95 | return 0; | |
1dc85be0 | 96 | #endif |
1da177e4 | 97 | |
f2857ce9 | 98 | instr = (unsigned char *)convert_ip_to_linear(current, regs); |
1dc85be0 HH |
99 | max_instr = instr + 15; |
100 | ||
1dc85be0 HH |
101 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE) |
102 | return 0; | |
1dc85be0 HH |
103 | |
104 | while (scan_more && instr < max_instr) { | |
1da177e4 LT |
105 | unsigned char opcode; |
106 | unsigned char instr_hi; | |
107 | unsigned char instr_lo; | |
108 | ||
11a4180c | 109 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 110 | break; |
1da177e4 | 111 | |
33cb5243 HH |
112 | instr_hi = opcode & 0xf0; |
113 | instr_lo = opcode & 0x0f; | |
1da177e4 LT |
114 | instr++; |
115 | ||
33cb5243 | 116 | switch (instr_hi) { |
1da177e4 LT |
117 | case 0x20: |
118 | case 0x30: | |
33cb5243 HH |
119 | /* |
120 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
121 | * In X86_64 long mode, the CPU will signal invalid | |
122 | * opcode if some of these prefixes are present so | |
123 | * X86_64 will never get here anyway | |
124 | */ | |
1da177e4 LT |
125 | scan_more = ((instr_lo & 7) == 0x6); |
126 | break; | |
33cb5243 HH |
127 | #ifdef CONFIG_X86_64 |
128 | case 0x40: | |
129 | /* | |
130 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
131 | * Need to figure out under what instruction mode the | |
132 | * instruction was issued. Could check the LDT for lm, | |
133 | * but for now it's good enough to assume that long | |
134 | * mode only uses well known segments or kernel. | |
135 | */ | |
136 | scan_more = (!user_mode(regs)) || (regs->cs == __USER_CS); | |
137 | break; | |
138 | #endif | |
1da177e4 LT |
139 | case 0x60: |
140 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
141 | scan_more = (instr_lo & 0xC) == 0x4; | |
33cb5243 | 142 | break; |
1da177e4 | 143 | case 0xF0: |
33cb5243 | 144 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ |
1da177e4 | 145 | scan_more = !instr_lo || (instr_lo>>1) == 1; |
33cb5243 | 146 | break; |
1da177e4 LT |
147 | case 0x00: |
148 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
149 | scan_more = 0; | |
f2857ce9 | 150 | |
11a4180c | 151 | if (probe_kernel_address(instr, opcode)) |
1da177e4 LT |
152 | break; |
153 | prefetch = (instr_lo == 0xF) && | |
154 | (opcode == 0x0D || opcode == 0x18); | |
33cb5243 | 155 | break; |
1da177e4 LT |
156 | default: |
157 | scan_more = 0; | |
158 | break; | |
33cb5243 | 159 | } |
1da177e4 LT |
160 | } |
161 | return prefetch; | |
162 | } | |
163 | ||
c4aba4a8 | 164 | static void force_sig_info_fault(int si_signo, int si_code, |
869f96a0 IM |
165 | unsigned long address, struct task_struct *tsk) |
166 | { | |
167 | siginfo_t info; | |
168 | ||
169 | info.si_signo = si_signo; | |
170 | info.si_errno = 0; | |
171 | info.si_code = si_code; | |
172 | info.si_addr = (void __user *)address; | |
173 | force_sig_info(si_signo, &info, tsk); | |
174 | } | |
175 | ||
e66a9512 HH |
176 | void dump_pagetable(unsigned long address) |
177 | { | |
178 | __typeof__(pte_val(__pte(0))) page; | |
179 | ||
180 | page = read_cr3(); | |
181 | page = ((__typeof__(page) *) __va(page))[address >> PGDIR_SHIFT]; | |
182 | #ifdef CONFIG_X86_PAE | |
183 | printk("*pdpt = %016Lx ", page); | |
184 | if ((page >> PAGE_SHIFT) < max_low_pfn | |
185 | && page & _PAGE_PRESENT) { | |
186 | page &= PAGE_MASK; | |
187 | page = ((__typeof__(page) *) __va(page))[(address >> PMD_SHIFT) | |
188 | & (PTRS_PER_PMD - 1)]; | |
189 | printk(KERN_CONT "*pde = %016Lx ", page); | |
190 | page &= ~_PAGE_NX; | |
191 | } | |
192 | #else | |
193 | printk("*pde = %08lx ", page); | |
194 | #endif | |
195 | ||
196 | /* | |
197 | * We must not directly access the pte in the highpte | |
198 | * case if the page table is located in highmem. | |
199 | * And let's rather not kmap-atomic the pte, just in case | |
200 | * it's allocated already. | |
201 | */ | |
202 | if ((page >> PAGE_SHIFT) < max_low_pfn | |
203 | && (page & _PAGE_PRESENT) | |
204 | && !(page & _PAGE_PSE)) { | |
205 | page &= PAGE_MASK; | |
206 | page = ((__typeof__(page) *) __va(page))[(address >> PAGE_SHIFT) | |
207 | & (PTRS_PER_PTE - 1)]; | |
208 | printk("*pte = %0*Lx ", sizeof(page)*2, (u64)page); | |
209 | } | |
210 | ||
211 | printk("\n"); | |
212 | } | |
213 | ||
75604d7f | 214 | void do_invalid_op(struct pt_regs *, unsigned long); |
1da177e4 | 215 | |
101f12af JB |
216 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) |
217 | { | |
218 | unsigned index = pgd_index(address); | |
219 | pgd_t *pgd_k; | |
220 | pud_t *pud, *pud_k; | |
221 | pmd_t *pmd, *pmd_k; | |
222 | ||
223 | pgd += index; | |
224 | pgd_k = init_mm.pgd + index; | |
225 | ||
226 | if (!pgd_present(*pgd_k)) | |
227 | return NULL; | |
228 | ||
229 | /* | |
230 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
231 | * and redundant with the set_pmd() on non-PAE. As would | |
232 | * set_pud. | |
233 | */ | |
234 | ||
235 | pud = pud_offset(pgd, address); | |
236 | pud_k = pud_offset(pgd_k, address); | |
237 | if (!pud_present(*pud_k)) | |
238 | return NULL; | |
239 | ||
240 | pmd = pmd_offset(pud, address); | |
241 | pmd_k = pmd_offset(pud_k, address); | |
242 | if (!pmd_present(*pmd_k)) | |
243 | return NULL; | |
8b14cb99 | 244 | if (!pmd_present(*pmd)) { |
101f12af | 245 | set_pmd(pmd, *pmd_k); |
8b14cb99 ZA |
246 | arch_flush_lazy_mmu_mode(); |
247 | } else | |
101f12af JB |
248 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
249 | return pmd_k; | |
250 | } | |
251 | ||
1dc85be0 HH |
252 | #ifdef CONFIG_X86_64 |
253 | static const char errata93_warning[] = | |
254 | KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n" | |
255 | KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n" | |
256 | KERN_ERR "******* Please consider a BIOS update.\n" | |
257 | KERN_ERR "******* Disabling USB legacy in the BIOS may also help.\n"; | |
fdfe8aa8 | 258 | #endif |
1dc85be0 HH |
259 | |
260 | /* Workaround for K8 erratum #93 & buggy BIOS. | |
261 | BIOS SMM functions are required to use a specific workaround | |
262 | to avoid corruption of the 64bit RIP register on C stepping K8. | |
263 | A lot of BIOS that didn't get tested properly miss this. | |
264 | The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
265 | Try to work around it here. | |
fdfe8aa8 HH |
266 | Note we only handle faults in kernel here. |
267 | Does nothing for X86_32 | |
268 | */ | |
1dc85be0 HH |
269 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
270 | { | |
fdfe8aa8 | 271 | #ifdef CONFIG_X86_64 |
1dc85be0 HH |
272 | static int warned; |
273 | if (address != regs->ip) | |
274 | return 0; | |
275 | if ((address >> 32) != 0) | |
276 | return 0; | |
277 | address |= 0xffffffffUL << 32; | |
278 | if ((address >= (u64)_stext && address <= (u64)_etext) || | |
279 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
280 | if (!warned) { | |
281 | printk(errata93_warning); | |
282 | warned = 1; | |
283 | } | |
284 | regs->ip = address; | |
285 | return 1; | |
286 | } | |
fdfe8aa8 | 287 | #endif |
1dc85be0 HH |
288 | return 0; |
289 | } | |
fdfe8aa8 | 290 | |
1dc85be0 | 291 | |
101f12af JB |
292 | /* |
293 | * Handle a fault on the vmalloc or module mapping area | |
294 | * | |
295 | * This assumes no large pages in there. | |
296 | */ | |
297 | static inline int vmalloc_fault(unsigned long address) | |
298 | { | |
fdfe8aa8 | 299 | #ifdef CONFIG_X86_32 |
101f12af JB |
300 | unsigned long pgd_paddr; |
301 | pmd_t *pmd_k; | |
302 | pte_t *pte_k; | |
303 | /* | |
304 | * Synchronize this task's top level page-table | |
305 | * with the 'reference' page table. | |
306 | * | |
307 | * Do _not_ use "current" here. We might be inside | |
308 | * an interrupt in the middle of a task switch.. | |
309 | */ | |
310 | pgd_paddr = read_cr3(); | |
311 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
312 | if (!pmd_k) | |
313 | return -1; | |
314 | pte_k = pte_offset_kernel(pmd_k, address); | |
315 | if (!pte_present(*pte_k)) | |
316 | return -1; | |
317 | return 0; | |
fdfe8aa8 HH |
318 | #else |
319 | pgd_t *pgd, *pgd_ref; | |
320 | pud_t *pud, *pud_ref; | |
321 | pmd_t *pmd, *pmd_ref; | |
322 | pte_t *pte, *pte_ref; | |
323 | ||
324 | /* Copy kernel mappings over when needed. This can also | |
325 | happen within a race in page table update. In the later | |
326 | case just flush. */ | |
327 | ||
328 | pgd = pgd_offset(current->mm ?: &init_mm, address); | |
329 | pgd_ref = pgd_offset_k(address); | |
330 | if (pgd_none(*pgd_ref)) | |
331 | return -1; | |
332 | if (pgd_none(*pgd)) | |
333 | set_pgd(pgd, *pgd_ref); | |
334 | else | |
335 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); | |
336 | ||
337 | /* Below here mismatches are bugs because these lower tables | |
338 | are shared */ | |
339 | ||
340 | pud = pud_offset(pgd, address); | |
341 | pud_ref = pud_offset(pgd_ref, address); | |
342 | if (pud_none(*pud_ref)) | |
343 | return -1; | |
344 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) | |
345 | BUG(); | |
346 | pmd = pmd_offset(pud, address); | |
347 | pmd_ref = pmd_offset(pud_ref, address); | |
348 | if (pmd_none(*pmd_ref)) | |
349 | return -1; | |
350 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) | |
351 | BUG(); | |
352 | pte_ref = pte_offset_kernel(pmd_ref, address); | |
353 | if (!pte_present(*pte_ref)) | |
354 | return -1; | |
355 | pte = pte_offset_kernel(pmd, address); | |
356 | /* Don't use pte_page here, because the mappings can point | |
357 | outside mem_map, and the NUMA hash lookup cannot handle | |
358 | that. */ | |
359 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
360 | BUG(); | |
361 | return 0; | |
362 | #endif | |
101f12af JB |
363 | } |
364 | ||
abd4f750 MAS |
365 | int show_unhandled_signals = 1; |
366 | ||
1da177e4 LT |
367 | /* |
368 | * This routine handles page faults. It determines the address, | |
369 | * and the problem, and then passes it off to one of the appropriate | |
370 | * routines. | |
1da177e4 | 371 | */ |
75604d7f | 372 | void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code) |
1da177e4 LT |
373 | { |
374 | struct task_struct *tsk; | |
375 | struct mm_struct *mm; | |
33cb5243 | 376 | struct vm_area_struct *vma; |
1da177e4 | 377 | unsigned long address; |
869f96a0 | 378 | int write, si_code; |
83c54070 | 379 | int fault; |
1da177e4 | 380 | |
143a5d32 PZ |
381 | /* |
382 | * We can fault from pretty much anywhere, with unknown IRQ state. | |
383 | */ | |
384 | trace_hardirqs_fixup(); | |
385 | ||
608566b4 HH |
386 | tsk = current; |
387 | mm = tsk->mm; | |
388 | prefetchw(&mm->mmap_sem); | |
389 | ||
1da177e4 | 390 | /* get the address */ |
33cb5243 | 391 | address = read_cr2(); |
1da177e4 | 392 | |
869f96a0 | 393 | si_code = SEGV_MAPERR; |
1da177e4 | 394 | |
608566b4 HH |
395 | if (notify_page_fault(regs)) |
396 | return; | |
397 | ||
1da177e4 LT |
398 | /* |
399 | * We fault-in kernel-space virtual memory on-demand. The | |
400 | * 'reference' page table is init_mm.pgd. | |
401 | * | |
402 | * NOTE! We MUST NOT take any locks for this case. We may | |
403 | * be in an interrupt or a critical region, and should | |
404 | * only copy the information from the master page table, | |
405 | * nothing more. | |
406 | * | |
407 | * This verifies that the fault happens in kernel space | |
408 | * (error_code & 4) == 0, and that the fault was not a | |
101f12af | 409 | * protection error (error_code & 9) == 0. |
1da177e4 | 410 | */ |
101f12af | 411 | if (unlikely(address >= TASK_SIZE)) { |
318aa296 HH |
412 | if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) && |
413 | vmalloc_fault(address) >= 0) | |
101f12af | 414 | return; |
101f12af | 415 | /* |
1da177e4 LT |
416 | * Don't take the mm semaphore here. If we fixup a prefetch |
417 | * fault we could otherwise deadlock. | |
418 | */ | |
419 | goto bad_area_nosemaphore; | |
101f12af JB |
420 | } |
421 | ||
101f12af JB |
422 | /* It's safe to allow irq's after cr2 has been saved and the vmalloc |
423 | fault has been handled. */ | |
65ea5b03 | 424 | if (regs->flags & (X86_EFLAGS_IF|VM_MASK)) |
101f12af | 425 | local_irq_enable(); |
1da177e4 | 426 | |
1da177e4 LT |
427 | /* |
428 | * If we're in an interrupt, have no user context or are running in an | |
33cb5243 | 429 | * atomic region then we must not take the fault. |
1da177e4 LT |
430 | */ |
431 | if (in_atomic() || !mm) | |
432 | goto bad_area_nosemaphore; | |
433 | ||
434 | /* When running in the kernel we expect faults to occur only to | |
435 | * addresses in user space. All other faults represent errors in the | |
27b46d76 | 436 | * kernel and should generate an OOPS. Unfortunately, in the case of an |
80f7228b | 437 | * erroneous fault occurring in a code path which already holds mmap_sem |
1da177e4 LT |
438 | * we will deadlock attempting to validate the fault against the |
439 | * address space. Luckily the kernel only validly references user | |
440 | * space from well defined areas of code, which are listed in the | |
441 | * exceptions table. | |
442 | * | |
443 | * As the vast majority of faults will be valid we will only perform | |
27b46d76 | 444 | * the source reference check when there is a possibility of a deadlock. |
1da177e4 LT |
445 | * Attempt to lock the address space, if we cannot we then validate the |
446 | * source. If this is invalid we can skip the address space check, | |
447 | * thus avoiding the deadlock. | |
448 | */ | |
449 | if (!down_read_trylock(&mm->mmap_sem)) { | |
33cb5243 | 450 | if ((error_code & PF_USER) == 0 && |
65ea5b03 | 451 | !search_exception_tables(regs->ip)) |
1da177e4 LT |
452 | goto bad_area_nosemaphore; |
453 | down_read(&mm->mmap_sem); | |
454 | } | |
455 | ||
456 | vma = find_vma(mm, address); | |
457 | if (!vma) | |
458 | goto bad_area; | |
459 | if (vma->vm_start <= address) | |
460 | goto good_area; | |
461 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
462 | goto bad_area; | |
33cb5243 | 463 | if (error_code & PF_USER) { |
1da177e4 | 464 | /* |
65ea5b03 | 465 | * Accessing the stack below %sp is always a bug. |
21528454 CE |
466 | * The large cushion allows instructions like enter |
467 | * and pusha to work. ("enter $65535,$31" pushes | |
65ea5b03 | 468 | * 32 pointers and then decrements %sp by 65535.) |
1da177e4 | 469 | */ |
65ea5b03 | 470 | if (address + 65536 + 32 * sizeof(unsigned long) < regs->sp) |
1da177e4 LT |
471 | goto bad_area; |
472 | } | |
473 | if (expand_stack(vma, address)) | |
474 | goto bad_area; | |
475 | /* | |
476 | * Ok, we have a good vm_area for this memory access, so | |
477 | * we can handle it.. | |
478 | */ | |
479 | good_area: | |
869f96a0 | 480 | si_code = SEGV_ACCERR; |
1da177e4 | 481 | write = 0; |
33cb5243 HH |
482 | switch (error_code & (PF_PROT|PF_WRITE)) { |
483 | default: /* 3: write, present */ | |
484 | /* fall through */ | |
485 | case PF_WRITE: /* write, not present */ | |
486 | if (!(vma->vm_flags & VM_WRITE)) | |
487 | goto bad_area; | |
488 | write++; | |
489 | break; | |
490 | case PF_PROT: /* read, present */ | |
491 | goto bad_area; | |
492 | case 0: /* read, not present */ | |
493 | if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))) | |
1da177e4 | 494 | goto bad_area; |
1da177e4 LT |
495 | } |
496 | ||
497 | survive: | |
498 | /* | |
499 | * If for any reason at all we couldn't handle the fault, | |
500 | * make sure we exit gracefully rather than endlessly redo | |
501 | * the fault. | |
502 | */ | |
83c54070 NP |
503 | fault = handle_mm_fault(mm, vma, address, write); |
504 | if (unlikely(fault & VM_FAULT_ERROR)) { | |
505 | if (fault & VM_FAULT_OOM) | |
1da177e4 | 506 | goto out_of_memory; |
83c54070 NP |
507 | else if (fault & VM_FAULT_SIGBUS) |
508 | goto do_sigbus; | |
509 | BUG(); | |
1da177e4 | 510 | } |
83c54070 NP |
511 | if (fault & VM_FAULT_MAJOR) |
512 | tsk->maj_flt++; | |
513 | else | |
514 | tsk->min_flt++; | |
1da177e4 | 515 | |
d729ab35 | 516 | #ifdef CONFIG_X86_32 |
1da177e4 LT |
517 | /* |
518 | * Did it hit the DOS screen memory VA from vm86 mode? | |
519 | */ | |
d729ab35 | 520 | if (v8086_mode(regs)) { |
1da177e4 LT |
521 | unsigned long bit = (address - 0xA0000) >> PAGE_SHIFT; |
522 | if (bit < 32) | |
523 | tsk->thread.screen_bitmap |= 1 << bit; | |
524 | } | |
d729ab35 | 525 | #endif |
1da177e4 LT |
526 | up_read(&mm->mmap_sem); |
527 | return; | |
528 | ||
529 | /* | |
530 | * Something tried to access memory that isn't in our memory map.. | |
531 | * Fix it, but check if it's kernel or user first.. | |
532 | */ | |
533 | bad_area: | |
534 | up_read(&mm->mmap_sem); | |
535 | ||
536 | bad_area_nosemaphore: | |
537 | /* User mode accesses just cause a SIGSEGV */ | |
33cb5243 | 538 | if (error_code & PF_USER) { |
e5e3c84b SR |
539 | /* |
540 | * It's possible to have interrupts off here. | |
541 | */ | |
542 | local_irq_enable(); | |
543 | ||
33cb5243 HH |
544 | /* |
545 | * Valid to do another page fault here because this one came | |
1da177e4 LT |
546 | * from user space. |
547 | */ | |
548 | if (is_prefetch(regs, address, error_code)) | |
549 | return; | |
550 | ||
abd4f750 MAS |
551 | if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) && |
552 | printk_ratelimit()) { | |
6f4d368e HH |
553 | printk( |
554 | #ifdef CONFIG_X86_32 | |
edcd8119 | 555 | "%s%s[%d]: segfault at %lx ip %08lx sp %08lx error %lx", |
6f4d368e | 556 | #else |
03252919 | 557 | "%s%s[%d]: segfault at %lx ip %lx sp %lx error %lx", |
6f4d368e HH |
558 | #endif |
559 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, | |
560 | tsk->comm, task_pid_nr(tsk), address, regs->ip, | |
561 | regs->sp, error_code); | |
03252919 AK |
562 | print_vma_addr(" in ", regs->ip); |
563 | printk("\n"); | |
abd4f750 | 564 | } |
1da177e4 LT |
565 | tsk->thread.cr2 = address; |
566 | /* Kernel addresses are always protection faults */ | |
567 | tsk->thread.error_code = error_code | (address >= TASK_SIZE); | |
568 | tsk->thread.trap_no = 14; | |
869f96a0 | 569 | force_sig_info_fault(SIGSEGV, si_code, address, tsk); |
1da177e4 LT |
570 | return; |
571 | } | |
572 | ||
573 | #ifdef CONFIG_X86_F00F_BUG | |
574 | /* | |
575 | * Pentium F0 0F C7 C8 bug workaround. | |
576 | */ | |
577 | if (boot_cpu_data.f00f_bug) { | |
578 | unsigned long nr; | |
33cb5243 | 579 | |
1da177e4 LT |
580 | nr = (address - idt_descr.address) >> 3; |
581 | ||
582 | if (nr == 6) { | |
583 | do_invalid_op(regs, 0); | |
584 | return; | |
585 | } | |
586 | } | |
587 | #endif | |
588 | ||
589 | no_context: | |
590 | /* Are we prepared to handle this kernel fault? */ | |
591 | if (fixup_exception(regs)) | |
592 | return; | |
593 | ||
33cb5243 | 594 | /* |
1da177e4 | 595 | * Valid to do another page fault here, because if this fault |
33cb5243 | 596 | * had been triggered by is_prefetch fixup_exception would have |
1da177e4 LT |
597 | * handled it. |
598 | */ | |
33cb5243 HH |
599 | if (is_prefetch(regs, address, error_code)) |
600 | return; | |
1da177e4 | 601 | |
fdfe8aa8 HH |
602 | if (is_errata93(regs, address)) |
603 | return; | |
604 | ||
1da177e4 LT |
605 | /* |
606 | * Oops. The kernel tried to access some bad page. We'll have to | |
607 | * terminate things with extreme prejudice. | |
608 | */ | |
609 | ||
610 | bust_spinlocks(1); | |
611 | ||
dd287796 | 612 | if (oops_may_print()) { |
28609f6e JB |
613 | |
614 | #ifdef CONFIG_X86_PAE | |
318aa296 | 615 | if (error_code & PF_INSTR) { |
dd287796 AM |
616 | pte_t *pte = lookup_address(address); |
617 | ||
4c3c4b45 | 618 | if (pte && pte_present(*pte) && !pte_exec(*pte)) |
dd287796 AM |
619 | printk(KERN_CRIT "kernel tried to execute " |
620 | "NX-protected page - exploit attempt? " | |
621 | "(uid: %d)\n", current->uid); | |
622 | } | |
28609f6e | 623 | #endif |
dd287796 AM |
624 | if (address < PAGE_SIZE) |
625 | printk(KERN_ALERT "BUG: unable to handle kernel NULL " | |
626 | "pointer dereference"); | |
627 | else | |
628 | printk(KERN_ALERT "BUG: unable to handle kernel paging" | |
629 | " request"); | |
33cb5243 | 630 | printk(" at virtual address %08lx\n", address); |
65ea5b03 | 631 | printk(KERN_ALERT "printing ip: %08lx ", regs->ip); |
28609f6e | 632 | |
e66a9512 | 633 | dump_pagetable(address); |
28609f6e JB |
634 | } |
635 | ||
4f339ecb AN |
636 | tsk->thread.cr2 = address; |
637 | tsk->thread.trap_no = 14; | |
638 | tsk->thread.error_code = error_code; | |
1da177e4 LT |
639 | die("Oops", regs, error_code); |
640 | bust_spinlocks(0); | |
641 | do_exit(SIGKILL); | |
642 | ||
643 | /* | |
644 | * We ran out of memory, or some other thing happened to us that made | |
645 | * us unable to handle the page fault gracefully. | |
646 | */ | |
647 | out_of_memory: | |
648 | up_read(&mm->mmap_sem); | |
b460cbc5 | 649 | if (is_global_init(tsk)) { |
1da177e4 LT |
650 | yield(); |
651 | down_read(&mm->mmap_sem); | |
652 | goto survive; | |
653 | } | |
654 | printk("VM: killing process %s\n", tsk->comm); | |
318aa296 | 655 | if (error_code & PF_USER) |
dcca2bde | 656 | do_group_exit(SIGKILL); |
1da177e4 LT |
657 | goto no_context; |
658 | ||
659 | do_sigbus: | |
660 | up_read(&mm->mmap_sem); | |
661 | ||
662 | /* Kernel mode? Handle exceptions or die */ | |
33cb5243 | 663 | if (!(error_code & PF_USER)) |
1da177e4 LT |
664 | goto no_context; |
665 | ||
666 | /* User space => ok to do another page fault */ | |
667 | if (is_prefetch(regs, address, error_code)) | |
668 | return; | |
669 | ||
670 | tsk->thread.cr2 = address; | |
671 | tsk->thread.error_code = error_code; | |
672 | tsk->thread.trap_no = 14; | |
869f96a0 | 673 | force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk); |
101f12af | 674 | } |
1da177e4 | 675 | |
101f12af JB |
676 | void vmalloc_sync_all(void) |
677 | { | |
678 | /* | |
679 | * Note that races in the updates of insync and start aren't | |
680 | * problematic: insync can only get set bits added, and updates to | |
681 | * start are only improving performance (without affecting correctness | |
682 | * if undone). | |
683 | */ | |
684 | static DECLARE_BITMAP(insync, PTRS_PER_PGD); | |
685 | static unsigned long start = TASK_SIZE; | |
686 | unsigned long address; | |
1da177e4 | 687 | |
5311ab62 JF |
688 | if (SHARED_KERNEL_PMD) |
689 | return; | |
690 | ||
101f12af JB |
691 | BUILD_BUG_ON(TASK_SIZE & ~PGDIR_MASK); |
692 | for (address = start; address >= TASK_SIZE; address += PGDIR_SIZE) { | |
693 | if (!test_bit(pgd_index(address), insync)) { | |
694 | unsigned long flags; | |
695 | struct page *page; | |
696 | ||
697 | spin_lock_irqsave(&pgd_lock, flags); | |
698 | for (page = pgd_list; page; page = | |
699 | (struct page *)page->index) | |
700 | if (!vmalloc_sync_one(page_address(page), | |
701 | address)) { | |
702 | BUG_ON(page != pgd_list); | |
703 | break; | |
704 | } | |
705 | spin_unlock_irqrestore(&pgd_lock, flags); | |
706 | if (!page) | |
707 | set_bit(pgd_index(address), insync); | |
708 | } | |
709 | if (address == start && test_bit(pgd_index(address), insync)) | |
710 | start = address + PGDIR_SIZE; | |
1da177e4 LT |
711 | } |
712 | } |