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> | |
c61e211d HH |
21 | #include <linux/highmem.h> |
22 | #include <linux/bootmem.h> /* for max_low_pfn */ | |
1eeb66a1 | 23 | #include <linux/vmalloc.h> |
1da177e4 | 24 | #include <linux/module.h> |
0f2fbdcb | 25 | #include <linux/kprobes.h> |
ab2bf0c1 | 26 | #include <linux/uaccess.h> |
1eeb66a1 | 27 | #include <linux/kdebug.h> |
1da177e4 LT |
28 | |
29 | #include <asm/system.h> | |
c61e211d HH |
30 | #include <asm/desc.h> |
31 | #include <asm/segment.h> | |
1da177e4 LT |
32 | #include <asm/pgalloc.h> |
33 | #include <asm/smp.h> | |
34 | #include <asm/tlbflush.h> | |
35 | #include <asm/proto.h> | |
1da177e4 | 36 | #include <asm-generic/sections.h> |
1da177e4 | 37 | |
33cb5243 HH |
38 | /* |
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 | */ | |
8a19da7b | 46 | #define PF_PROT (1<<0) |
66c58156 | 47 | #define PF_WRITE (1<<1) |
8a19da7b IM |
48 | #define PF_USER (1<<2) |
49 | #define PF_RSVD (1<<3) | |
66c58156 AK |
50 | #define PF_INSTR (1<<4) |
51 | ||
74a0b576 | 52 | static inline int notify_page_fault(struct pt_regs *regs) |
1bd858a5 | 53 | { |
33cb5243 | 54 | #ifdef CONFIG_KPROBES |
74a0b576 CH |
55 | int ret = 0; |
56 | ||
57 | /* kprobe_running() needs smp_processor_id() */ | |
f8c2ee22 HH |
58 | #ifdef CONFIG_X86_32 |
59 | if (!user_mode_vm(regs)) { | |
60 | #else | |
74a0b576 | 61 | if (!user_mode(regs)) { |
f8c2ee22 | 62 | #endif |
74a0b576 CH |
63 | preempt_disable(); |
64 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
65 | ret = 1; | |
66 | preempt_enable(); | |
67 | } | |
1bd858a5 | 68 | |
74a0b576 | 69 | return ret; |
74a0b576 | 70 | #else |
74a0b576 | 71 | return 0; |
74a0b576 | 72 | #endif |
33cb5243 | 73 | } |
1bd858a5 | 74 | |
1dc85be0 HH |
75 | /* |
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 | */ | |
86 | static int is_prefetch(struct pt_regs *regs, unsigned long addr, | |
87 | unsigned long error_code) | |
33cb5243 | 88 | { |
ab2bf0c1 | 89 | unsigned char *instr; |
1da177e4 | 90 | int scan_more = 1; |
33cb5243 | 91 | int prefetch = 0; |
f1290ec9 | 92 | unsigned char *max_instr; |
1da177e4 | 93 | |
3085354d IM |
94 | /* |
95 | * If it was a exec (instruction fetch) fault on NX page, then | |
96 | * do not ignore the fault: | |
97 | */ | |
66c58156 | 98 | if (error_code & PF_INSTR) |
1da177e4 | 99 | return 0; |
1dc85be0 | 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 | ||
1156e098 | 179 | #ifdef CONFIG_X86_64 |
33cb5243 HH |
180 | static int bad_address(void *p) |
181 | { | |
1da177e4 | 182 | unsigned long dummy; |
ab2bf0c1 | 183 | return probe_kernel_address((unsigned long *)p, dummy); |
33cb5243 | 184 | } |
1156e098 | 185 | #endif |
1da177e4 | 186 | |
cae30f82 | 187 | static void dump_pagetable(unsigned long address) |
1da177e4 | 188 | { |
1156e098 HH |
189 | #ifdef CONFIG_X86_32 |
190 | __typeof__(pte_val(__pte(0))) page; | |
191 | ||
192 | page = read_cr3(); | |
193 | page = ((__typeof__(page) *) __va(page))[address >> PGDIR_SHIFT]; | |
194 | #ifdef CONFIG_X86_PAE | |
195 | printk("*pdpt = %016Lx ", page); | |
196 | if ((page >> PAGE_SHIFT) < max_low_pfn | |
197 | && page & _PAGE_PRESENT) { | |
198 | page &= PAGE_MASK; | |
199 | page = ((__typeof__(page) *) __va(page))[(address >> PMD_SHIFT) | |
200 | & (PTRS_PER_PMD - 1)]; | |
201 | printk(KERN_CONT "*pde = %016Lx ", page); | |
202 | page &= ~_PAGE_NX; | |
203 | } | |
204 | #else | |
205 | printk("*pde = %08lx ", page); | |
206 | #endif | |
207 | ||
208 | /* | |
209 | * We must not directly access the pte in the highpte | |
210 | * case if the page table is located in highmem. | |
211 | * And let's rather not kmap-atomic the pte, just in case | |
212 | * it's allocated already. | |
213 | */ | |
214 | if ((page >> PAGE_SHIFT) < max_low_pfn | |
215 | && (page & _PAGE_PRESENT) | |
216 | && !(page & _PAGE_PSE)) { | |
217 | page &= PAGE_MASK; | |
218 | page = ((__typeof__(page) *) __va(page))[(address >> PAGE_SHIFT) | |
219 | & (PTRS_PER_PTE - 1)]; | |
220 | printk("*pte = %0*Lx ", sizeof(page)*2, (u64)page); | |
221 | } | |
222 | ||
223 | printk("\n"); | |
224 | #else /* CONFIG_X86_64 */ | |
1da177e4 LT |
225 | pgd_t *pgd; |
226 | pud_t *pud; | |
227 | pmd_t *pmd; | |
228 | pte_t *pte; | |
229 | ||
f51c9452 | 230 | pgd = (pgd_t *)read_cr3(); |
1da177e4 | 231 | |
33cb5243 | 232 | pgd = __va((unsigned long)pgd & PHYSICAL_PAGE_MASK); |
1da177e4 | 233 | pgd += pgd_index(address); |
1da177e4 | 234 | if (bad_address(pgd)) goto bad; |
d646bce4 | 235 | printk("PGD %lx ", pgd_val(*pgd)); |
33cb5243 | 236 | if (!pgd_present(*pgd)) goto ret; |
1da177e4 | 237 | |
d2ae5b5f | 238 | pud = pud_offset(pgd, address); |
1da177e4 LT |
239 | if (bad_address(pud)) goto bad; |
240 | printk("PUD %lx ", pud_val(*pud)); | |
b5360222 AK |
241 | if (!pud_present(*pud) || pud_large(*pud)) |
242 | goto ret; | |
1da177e4 LT |
243 | |
244 | pmd = pmd_offset(pud, address); | |
245 | if (bad_address(pmd)) goto bad; | |
246 | printk("PMD %lx ", pmd_val(*pmd)); | |
b1992df3 | 247 | if (!pmd_present(*pmd) || pmd_large(*pmd)) goto ret; |
1da177e4 LT |
248 | |
249 | pte = pte_offset_kernel(pmd, address); | |
250 | if (bad_address(pte)) goto bad; | |
33cb5243 | 251 | printk("PTE %lx", pte_val(*pte)); |
1da177e4 LT |
252 | ret: |
253 | printk("\n"); | |
254 | return; | |
255 | bad: | |
256 | printk("BAD\n"); | |
1156e098 HH |
257 | #endif |
258 | } | |
259 | ||
260 | #ifdef CONFIG_X86_32 | |
261 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
262 | { | |
263 | unsigned index = pgd_index(address); | |
264 | pgd_t *pgd_k; | |
265 | pud_t *pud, *pud_k; | |
266 | pmd_t *pmd, *pmd_k; | |
267 | ||
268 | pgd += index; | |
269 | pgd_k = init_mm.pgd + index; | |
270 | ||
271 | if (!pgd_present(*pgd_k)) | |
272 | return NULL; | |
273 | ||
274 | /* | |
275 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
276 | * and redundant with the set_pmd() on non-PAE. As would | |
277 | * set_pud. | |
278 | */ | |
279 | ||
280 | pud = pud_offset(pgd, address); | |
281 | pud_k = pud_offset(pgd_k, address); | |
282 | if (!pud_present(*pud_k)) | |
283 | return NULL; | |
284 | ||
285 | pmd = pmd_offset(pud, address); | |
286 | pmd_k = pmd_offset(pud_k, address); | |
287 | if (!pmd_present(*pmd_k)) | |
288 | return NULL; | |
289 | if (!pmd_present(*pmd)) { | |
290 | set_pmd(pmd, *pmd_k); | |
291 | arch_flush_lazy_mmu_mode(); | |
292 | } else | |
293 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); | |
294 | return pmd_k; | |
1da177e4 | 295 | } |
1156e098 | 296 | #endif |
1da177e4 | 297 | |
1dc85be0 | 298 | #ifdef CONFIG_X86_64 |
33cb5243 | 299 | static const char errata93_warning[] = |
1da177e4 LT |
300 | KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n" |
301 | KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n" | |
302 | KERN_ERR "******* Please consider a BIOS update.\n" | |
303 | KERN_ERR "******* Disabling USB legacy in the BIOS may also help.\n"; | |
fdfe8aa8 | 304 | #endif |
1da177e4 LT |
305 | |
306 | /* Workaround for K8 erratum #93 & buggy BIOS. | |
307 | BIOS SMM functions are required to use a specific workaround | |
33cb5243 HH |
308 | to avoid corruption of the 64bit RIP register on C stepping K8. |
309 | A lot of BIOS that didn't get tested properly miss this. | |
1da177e4 LT |
310 | The OS sees this as a page fault with the upper 32bits of RIP cleared. |
311 | Try to work around it here. | |
fdfe8aa8 HH |
312 | Note we only handle faults in kernel here. |
313 | Does nothing for X86_32 | |
314 | */ | |
33cb5243 | 315 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 316 | { |
fdfe8aa8 | 317 | #ifdef CONFIG_X86_64 |
1da177e4 | 318 | static int warned; |
65ea5b03 | 319 | if (address != regs->ip) |
1da177e4 | 320 | return 0; |
33cb5243 | 321 | if ((address >> 32) != 0) |
1da177e4 LT |
322 | return 0; |
323 | address |= 0xffffffffUL << 32; | |
33cb5243 HH |
324 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
325 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
1da177e4 | 326 | if (!warned) { |
33cb5243 | 327 | printk(errata93_warning); |
1da177e4 LT |
328 | warned = 1; |
329 | } | |
65ea5b03 | 330 | regs->ip = address; |
1da177e4 LT |
331 | return 1; |
332 | } | |
fdfe8aa8 | 333 | #endif |
1da177e4 | 334 | return 0; |
33cb5243 | 335 | } |
1da177e4 | 336 | |
35f3266f HH |
337 | /* |
338 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps to illegal | |
339 | * addresses >4GB. We catch this in the page fault handler because these | |
340 | * addresses are not reachable. Just detect this case and return. Any code | |
341 | * segment in LDT is compatibility mode. | |
342 | */ | |
343 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
344 | { | |
345 | #ifdef CONFIG_X86_64 | |
346 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && | |
347 | (address >> 32)) | |
348 | return 1; | |
349 | #endif | |
350 | return 0; | |
351 | } | |
352 | ||
29caf2f9 HH |
353 | void do_invalid_op(struct pt_regs *, unsigned long); |
354 | ||
355 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) | |
356 | { | |
357 | #ifdef CONFIG_X86_F00F_BUG | |
358 | unsigned long nr; | |
359 | /* | |
360 | * Pentium F0 0F C7 C8 bug workaround. | |
361 | */ | |
362 | if (boot_cpu_data.f00f_bug) { | |
363 | nr = (address - idt_descr.address) >> 3; | |
364 | ||
365 | if (nr == 6) { | |
366 | do_invalid_op(regs, 0); | |
367 | return 1; | |
368 | } | |
369 | } | |
370 | #endif | |
371 | return 0; | |
372 | } | |
373 | ||
b3279c7f HH |
374 | static void show_fault_oops(struct pt_regs *regs, unsigned long error_code, |
375 | unsigned long address) | |
376 | { | |
1156e098 HH |
377 | #ifdef CONFIG_X86_32 |
378 | if (!oops_may_print()) | |
379 | return; | |
fd40d6e3 | 380 | #endif |
1156e098 HH |
381 | |
382 | #ifdef CONFIG_X86_PAE | |
383 | if (error_code & PF_INSTR) { | |
93809be8 | 384 | unsigned int level; |
1156e098 HH |
385 | pte_t *pte = lookup_address(address, &level); |
386 | ||
387 | if (pte && pte_present(*pte) && !pte_exec(*pte)) | |
388 | printk(KERN_CRIT "kernel tried to execute " | |
389 | "NX-protected page - exploit attempt? " | |
390 | "(uid: %d)\n", current->uid); | |
391 | } | |
392 | #endif | |
1156e098 | 393 | |
19f0dda9 | 394 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 395 | if (address < PAGE_SIZE) |
19f0dda9 | 396 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 397 | else |
19f0dda9 | 398 | printk(KERN_CONT "paging request"); |
fd40d6e3 HH |
399 | #ifdef CONFIG_X86_32 |
400 | printk(KERN_CONT " at %08lx\n", address); | |
401 | #else | |
19f0dda9 | 402 | printk(KERN_CONT " at %016lx\n", address); |
fd40d6e3 | 403 | #endif |
19f0dda9 | 404 | printk(KERN_ALERT "IP:"); |
b3279c7f HH |
405 | printk_address(regs->ip, 1); |
406 | dump_pagetable(address); | |
407 | } | |
408 | ||
1156e098 | 409 | #ifdef CONFIG_X86_64 |
1da177e4 LT |
410 | static noinline void pgtable_bad(unsigned long address, struct pt_regs *regs, |
411 | unsigned long error_code) | |
412 | { | |
1209140c | 413 | unsigned long flags = oops_begin(); |
6e3f3617 | 414 | struct task_struct *tsk; |
1209140c | 415 | |
1da177e4 LT |
416 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
417 | current->comm, address); | |
418 | dump_pagetable(address); | |
6e3f3617 JB |
419 | tsk = current; |
420 | tsk->thread.cr2 = address; | |
421 | tsk->thread.trap_no = 14; | |
422 | tsk->thread.error_code = error_code; | |
22f5991c JB |
423 | if (__die("Bad pagetable", regs, error_code)) |
424 | regs = NULL; | |
425 | oops_end(flags, regs, SIGKILL); | |
1da177e4 | 426 | } |
1156e098 | 427 | #endif |
1da177e4 | 428 | |
d8b57bb7 TG |
429 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
430 | { | |
431 | if ((error_code & PF_WRITE) && !pte_write(*pte)) | |
432 | return 0; | |
433 | if ((error_code & PF_INSTR) && !pte_exec(*pte)) | |
434 | return 0; | |
435 | ||
436 | return 1; | |
437 | } | |
438 | ||
5b727a3b JF |
439 | /* |
440 | * Handle a spurious fault caused by a stale TLB entry. This allows | |
441 | * us to lazily refresh the TLB when increasing the permissions of a | |
442 | * kernel page (RO -> RW or NX -> X). Doing it eagerly is very | |
443 | * expensive since that implies doing a full cross-processor TLB | |
444 | * flush, even if no stale TLB entries exist on other processors. | |
445 | * There are no security implications to leaving a stale TLB when | |
446 | * increasing the permissions on a page. | |
447 | */ | |
448 | static int spurious_fault(unsigned long address, | |
449 | unsigned long error_code) | |
450 | { | |
451 | pgd_t *pgd; | |
452 | pud_t *pud; | |
453 | pmd_t *pmd; | |
454 | pte_t *pte; | |
455 | ||
456 | /* Reserved-bit violation or user access to kernel space? */ | |
457 | if (error_code & (PF_USER | PF_RSVD)) | |
458 | return 0; | |
459 | ||
460 | pgd = init_mm.pgd + pgd_index(address); | |
461 | if (!pgd_present(*pgd)) | |
462 | return 0; | |
463 | ||
464 | pud = pud_offset(pgd, address); | |
465 | if (!pud_present(*pud)) | |
466 | return 0; | |
467 | ||
d8b57bb7 TG |
468 | if (pud_large(*pud)) |
469 | return spurious_fault_check(error_code, (pte_t *) pud); | |
470 | ||
5b727a3b JF |
471 | pmd = pmd_offset(pud, address); |
472 | if (!pmd_present(*pmd)) | |
473 | return 0; | |
474 | ||
d8b57bb7 TG |
475 | if (pmd_large(*pmd)) |
476 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
477 | ||
5b727a3b JF |
478 | pte = pte_offset_kernel(pmd, address); |
479 | if (!pte_present(*pte)) | |
480 | return 0; | |
481 | ||
d8b57bb7 | 482 | return spurious_fault_check(error_code, pte); |
5b727a3b JF |
483 | } |
484 | ||
1da177e4 | 485 | /* |
f8c2ee22 HH |
486 | * X86_32 |
487 | * Handle a fault on the vmalloc or module mapping area | |
488 | * | |
489 | * X86_64 | |
f95190b2 | 490 | * Handle a fault on the vmalloc area |
3b9ba4d5 AK |
491 | * |
492 | * This assumes no large pages in there. | |
1da177e4 LT |
493 | */ |
494 | static int vmalloc_fault(unsigned long address) | |
495 | { | |
fdfe8aa8 HH |
496 | #ifdef CONFIG_X86_32 |
497 | unsigned long pgd_paddr; | |
498 | pmd_t *pmd_k; | |
499 | pte_t *pte_k; | |
b29c701d HN |
500 | |
501 | /* Make sure we are in vmalloc area */ | |
502 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
503 | return -1; | |
504 | ||
fdfe8aa8 HH |
505 | /* |
506 | * Synchronize this task's top level page-table | |
507 | * with the 'reference' page table. | |
508 | * | |
509 | * Do _not_ use "current" here. We might be inside | |
510 | * an interrupt in the middle of a task switch.. | |
511 | */ | |
512 | pgd_paddr = read_cr3(); | |
513 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
514 | if (!pmd_k) | |
515 | return -1; | |
516 | pte_k = pte_offset_kernel(pmd_k, address); | |
517 | if (!pte_present(*pte_k)) | |
518 | return -1; | |
519 | return 0; | |
520 | #else | |
1da177e4 LT |
521 | pgd_t *pgd, *pgd_ref; |
522 | pud_t *pud, *pud_ref; | |
523 | pmd_t *pmd, *pmd_ref; | |
524 | pte_t *pte, *pte_ref; | |
525 | ||
cf89ec92 HH |
526 | /* Make sure we are in vmalloc area */ |
527 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
528 | return -1; | |
529 | ||
1da177e4 LT |
530 | /* Copy kernel mappings over when needed. This can also |
531 | happen within a race in page table update. In the later | |
532 | case just flush. */ | |
533 | ||
534 | pgd = pgd_offset(current->mm ?: &init_mm, address); | |
535 | pgd_ref = pgd_offset_k(address); | |
536 | if (pgd_none(*pgd_ref)) | |
537 | return -1; | |
538 | if (pgd_none(*pgd)) | |
539 | set_pgd(pgd, *pgd_ref); | |
8c914cb7 | 540 | else |
46a82b2d | 541 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
1da177e4 LT |
542 | |
543 | /* Below here mismatches are bugs because these lower tables | |
544 | are shared */ | |
545 | ||
546 | pud = pud_offset(pgd, address); | |
547 | pud_ref = pud_offset(pgd_ref, address); | |
548 | if (pud_none(*pud_ref)) | |
549 | return -1; | |
46a82b2d | 550 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) |
1da177e4 LT |
551 | BUG(); |
552 | pmd = pmd_offset(pud, address); | |
553 | pmd_ref = pmd_offset(pud_ref, address); | |
554 | if (pmd_none(*pmd_ref)) | |
555 | return -1; | |
556 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) | |
557 | BUG(); | |
558 | pte_ref = pte_offset_kernel(pmd_ref, address); | |
559 | if (!pte_present(*pte_ref)) | |
560 | return -1; | |
561 | pte = pte_offset_kernel(pmd, address); | |
3b9ba4d5 AK |
562 | /* Don't use pte_page here, because the mappings can point |
563 | outside mem_map, and the NUMA hash lookup cannot handle | |
564 | that. */ | |
565 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
1da177e4 | 566 | BUG(); |
1da177e4 | 567 | return 0; |
fdfe8aa8 | 568 | #endif |
1da177e4 LT |
569 | } |
570 | ||
abd4f750 | 571 | int show_unhandled_signals = 1; |
1da177e4 LT |
572 | |
573 | /* | |
574 | * This routine handles page faults. It determines the address, | |
575 | * and the problem, and then passes it off to one of the appropriate | |
576 | * routines. | |
1da177e4 | 577 | */ |
f8c2ee22 HH |
578 | #ifdef CONFIG_X86_64 |
579 | asmlinkage | |
580 | #endif | |
581 | void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code) | |
1da177e4 LT |
582 | { |
583 | struct task_struct *tsk; | |
584 | struct mm_struct *mm; | |
33cb5243 | 585 | struct vm_area_struct *vma; |
1da177e4 | 586 | unsigned long address; |
f8c2ee22 HH |
587 | int write, si_code; |
588 | int fault; | |
589 | #ifdef CONFIG_X86_64 | |
1209140c | 590 | unsigned long flags; |
f8c2ee22 | 591 | #endif |
1da177e4 | 592 | |
143a5d32 PZ |
593 | /* |
594 | * We can fault from pretty much anywhere, with unknown IRQ state. | |
595 | */ | |
596 | trace_hardirqs_fixup(); | |
597 | ||
a9ba9a3b AV |
598 | tsk = current; |
599 | mm = tsk->mm; | |
600 | prefetchw(&mm->mmap_sem); | |
601 | ||
1da177e4 | 602 | /* get the address */ |
f51c9452 | 603 | address = read_cr2(); |
1da177e4 | 604 | |
c4aba4a8 | 605 | si_code = SEGV_MAPERR; |
1da177e4 | 606 | |
608566b4 HH |
607 | if (notify_page_fault(regs)) |
608 | return; | |
1da177e4 LT |
609 | |
610 | /* | |
611 | * We fault-in kernel-space virtual memory on-demand. The | |
612 | * 'reference' page table is init_mm.pgd. | |
613 | * | |
614 | * NOTE! We MUST NOT take any locks for this case. We may | |
615 | * be in an interrupt or a critical region, and should | |
616 | * only copy the information from the master page table, | |
617 | * nothing more. | |
618 | * | |
619 | * This verifies that the fault happens in kernel space | |
620 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 621 | * protection error (error_code & 9) == 0. |
1da177e4 | 622 | */ |
f8c2ee22 HH |
623 | #ifdef CONFIG_X86_32 |
624 | if (unlikely(address >= TASK_SIZE)) { | |
cf89ec92 HH |
625 | #else |
626 | if (unlikely(address >= TASK_SIZE64)) { | |
627 | #endif | |
f8c2ee22 HH |
628 | if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) && |
629 | vmalloc_fault(address) >= 0) | |
630 | return; | |
5b727a3b JF |
631 | |
632 | /* Can handle a stale RO->RW TLB */ | |
633 | if (spurious_fault(address, error_code)) | |
634 | return; | |
635 | ||
f8c2ee22 HH |
636 | /* |
637 | * Don't take the mm semaphore here. If we fixup a prefetch | |
638 | * fault we could otherwise deadlock. | |
639 | */ | |
640 | goto bad_area_nosemaphore; | |
641 | } | |
642 | ||
cf89ec92 HH |
643 | |
644 | #ifdef CONFIG_X86_32 | |
f8c2ee22 HH |
645 | /* It's safe to allow irq's after cr2 has been saved and the vmalloc |
646 | fault has been handled. */ | |
6b6891f9 | 647 | if (regs->flags & (X86_EFLAGS_IF | X86_VM_MASK)) |
f8c2ee22 HH |
648 | local_irq_enable(); |
649 | ||
650 | /* | |
651 | * If we're in an interrupt, have no user context or are running in an | |
652 | * atomic region then we must not take the fault. | |
653 | */ | |
654 | if (in_atomic() || !mm) | |
655 | goto bad_area_nosemaphore; | |
656 | #else /* CONFIG_X86_64 */ | |
65ea5b03 | 657 | if (likely(regs->flags & X86_EFLAGS_IF)) |
8c914cb7 JB |
658 | local_irq_enable(); |
659 | ||
66c58156 | 660 | if (unlikely(error_code & PF_RSVD)) |
1da177e4 LT |
661 | pgtable_bad(address, regs, error_code); |
662 | ||
663 | /* | |
33cb5243 HH |
664 | * If we're in an interrupt, have no user context or are running in an |
665 | * atomic region then we must not take the fault. | |
1da177e4 LT |
666 | */ |
667 | if (unlikely(in_atomic() || !mm)) | |
668 | goto bad_area_nosemaphore; | |
669 | ||
dbe3ed1c LT |
670 | /* |
671 | * User-mode registers count as a user access even for any | |
672 | * potential system fault or CPU buglet. | |
673 | */ | |
674 | if (user_mode_vm(regs)) | |
675 | error_code |= PF_USER; | |
f8c2ee22 HH |
676 | again: |
677 | #endif | |
1da177e4 LT |
678 | /* When running in the kernel we expect faults to occur only to |
679 | * addresses in user space. All other faults represent errors in the | |
676b1855 | 680 | * kernel and should generate an OOPS. Unfortunately, in the case of an |
80f7228b | 681 | * erroneous fault occurring in a code path which already holds mmap_sem |
1da177e4 LT |
682 | * we will deadlock attempting to validate the fault against the |
683 | * address space. Luckily the kernel only validly references user | |
684 | * space from well defined areas of code, which are listed in the | |
685 | * exceptions table. | |
686 | * | |
687 | * As the vast majority of faults will be valid we will only perform | |
676b1855 | 688 | * the source reference check when there is a possibility of a deadlock. |
1da177e4 LT |
689 | * Attempt to lock the address space, if we cannot we then validate the |
690 | * source. If this is invalid we can skip the address space check, | |
691 | * thus avoiding the deadlock. | |
692 | */ | |
693 | if (!down_read_trylock(&mm->mmap_sem)) { | |
66c58156 | 694 | if ((error_code & PF_USER) == 0 && |
65ea5b03 | 695 | !search_exception_tables(regs->ip)) |
1da177e4 LT |
696 | goto bad_area_nosemaphore; |
697 | down_read(&mm->mmap_sem); | |
698 | } | |
699 | ||
700 | vma = find_vma(mm, address); | |
701 | if (!vma) | |
702 | goto bad_area; | |
f8c2ee22 | 703 | if (vma->vm_start <= address) |
1da177e4 LT |
704 | goto good_area; |
705 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
706 | goto bad_area; | |
33cb5243 | 707 | if (error_code & PF_USER) { |
6f4d368e HH |
708 | /* |
709 | * Accessing the stack below %sp is always a bug. | |
710 | * The large cushion allows instructions like enter | |
711 | * and pusha to work. ("enter $65535,$31" pushes | |
712 | * 32 pointers and then decrements %sp by 65535.) | |
03fdc2c2 | 713 | */ |
65ea5b03 | 714 | if (address + 65536 + 32 * sizeof(unsigned long) < regs->sp) |
1da177e4 LT |
715 | goto bad_area; |
716 | } | |
717 | if (expand_stack(vma, address)) | |
718 | goto bad_area; | |
719 | /* | |
720 | * Ok, we have a good vm_area for this memory access, so | |
721 | * we can handle it.. | |
722 | */ | |
723 | good_area: | |
c4aba4a8 | 724 | si_code = SEGV_ACCERR; |
1da177e4 | 725 | write = 0; |
66c58156 | 726 | switch (error_code & (PF_PROT|PF_WRITE)) { |
33cb5243 HH |
727 | default: /* 3: write, present */ |
728 | /* fall through */ | |
729 | case PF_WRITE: /* write, not present */ | |
730 | if (!(vma->vm_flags & VM_WRITE)) | |
731 | goto bad_area; | |
732 | write++; | |
733 | break; | |
734 | case PF_PROT: /* read, present */ | |
735 | goto bad_area; | |
736 | case 0: /* read, not present */ | |
737 | if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))) | |
1da177e4 | 738 | goto bad_area; |
1da177e4 LT |
739 | } |
740 | ||
f8c2ee22 HH |
741 | #ifdef CONFIG_X86_32 |
742 | survive: | |
743 | #endif | |
1da177e4 LT |
744 | /* |
745 | * If for any reason at all we couldn't handle the fault, | |
746 | * make sure we exit gracefully rather than endlessly redo | |
747 | * the fault. | |
748 | */ | |
83c54070 NP |
749 | fault = handle_mm_fault(mm, vma, address, write); |
750 | if (unlikely(fault & VM_FAULT_ERROR)) { | |
751 | if (fault & VM_FAULT_OOM) | |
752 | goto out_of_memory; | |
753 | else if (fault & VM_FAULT_SIGBUS) | |
754 | goto do_sigbus; | |
755 | BUG(); | |
1da177e4 | 756 | } |
83c54070 NP |
757 | if (fault & VM_FAULT_MAJOR) |
758 | tsk->maj_flt++; | |
759 | else | |
760 | tsk->min_flt++; | |
d729ab35 HH |
761 | |
762 | #ifdef CONFIG_X86_32 | |
763 | /* | |
764 | * Did it hit the DOS screen memory VA from vm86 mode? | |
765 | */ | |
766 | if (v8086_mode(regs)) { | |
767 | unsigned long bit = (address - 0xA0000) >> PAGE_SHIFT; | |
768 | if (bit < 32) | |
769 | tsk->thread.screen_bitmap |= 1 << bit; | |
770 | } | |
771 | #endif | |
1da177e4 LT |
772 | up_read(&mm->mmap_sem); |
773 | return; | |
774 | ||
775 | /* | |
776 | * Something tried to access memory that isn't in our memory map.. | |
777 | * Fix it, but check if it's kernel or user first.. | |
778 | */ | |
779 | bad_area: | |
780 | up_read(&mm->mmap_sem); | |
781 | ||
782 | bad_area_nosemaphore: | |
1da177e4 | 783 | /* User mode accesses just cause a SIGSEGV */ |
66c58156 | 784 | if (error_code & PF_USER) { |
e5e3c84b SR |
785 | /* |
786 | * It's possible to have interrupts off here. | |
787 | */ | |
788 | local_irq_enable(); | |
789 | ||
1156e098 HH |
790 | /* |
791 | * Valid to do another page fault here because this one came | |
792 | * from user space. | |
793 | */ | |
1da177e4 LT |
794 | if (is_prefetch(regs, address, error_code)) |
795 | return; | |
796 | ||
35f3266f | 797 | if (is_errata100(regs, address)) |
1da177e4 LT |
798 | return; |
799 | ||
abd4f750 MAS |
800 | if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) && |
801 | printk_ratelimit()) { | |
1da177e4 | 802 | printk( |
6f4d368e | 803 | #ifdef CONFIG_X86_32 |
edcd8119 | 804 | "%s%s[%d]: segfault at %lx ip %08lx sp %08lx error %lx", |
6f4d368e | 805 | #else |
03252919 | 806 | "%s%s[%d]: segfault at %lx ip %lx sp %lx error %lx", |
6f4d368e HH |
807 | #endif |
808 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, | |
809 | tsk->comm, task_pid_nr(tsk), address, regs->ip, | |
810 | regs->sp, error_code); | |
03252919 AK |
811 | print_vma_addr(" in ", regs->ip); |
812 | printk("\n"); | |
1da177e4 | 813 | } |
33cb5243 | 814 | |
1da177e4 LT |
815 | tsk->thread.cr2 = address; |
816 | /* Kernel addresses are always protection faults */ | |
817 | tsk->thread.error_code = error_code | (address >= TASK_SIZE); | |
818 | tsk->thread.trap_no = 14; | |
c4aba4a8 | 819 | force_sig_info_fault(SIGSEGV, si_code, address, tsk); |
1da177e4 LT |
820 | return; |
821 | } | |
822 | ||
29caf2f9 HH |
823 | if (is_f00f_bug(regs, address)) |
824 | return; | |
825 | ||
1da177e4 | 826 | no_context: |
1da177e4 | 827 | /* Are we prepared to handle this kernel fault? */ |
33cb5243 | 828 | if (fixup_exception(regs)) |
1da177e4 | 829 | return; |
1da177e4 | 830 | |
33cb5243 | 831 | /* |
f8c2ee22 HH |
832 | * X86_32 |
833 | * Valid to do another page fault here, because if this fault | |
834 | * had been triggered by is_prefetch fixup_exception would have | |
835 | * handled it. | |
836 | * | |
837 | * X86_64 | |
1da177e4 LT |
838 | * Hall of shame of CPU/BIOS bugs. |
839 | */ | |
33cb5243 HH |
840 | if (is_prefetch(regs, address, error_code)) |
841 | return; | |
1da177e4 LT |
842 | |
843 | if (is_errata93(regs, address)) | |
33cb5243 | 844 | return; |
1da177e4 LT |
845 | |
846 | /* | |
847 | * Oops. The kernel tried to access some bad page. We'll have to | |
848 | * terminate things with extreme prejudice. | |
849 | */ | |
f8c2ee22 HH |
850 | #ifdef CONFIG_X86_32 |
851 | bust_spinlocks(1); | |
fd40d6e3 HH |
852 | #else |
853 | flags = oops_begin(); | |
854 | #endif | |
f8c2ee22 HH |
855 | |
856 | show_fault_oops(regs, error_code, address); | |
1da177e4 | 857 | |
f8c2ee22 HH |
858 | tsk->thread.cr2 = address; |
859 | tsk->thread.trap_no = 14; | |
860 | tsk->thread.error_code = error_code; | |
fd40d6e3 HH |
861 | |
862 | #ifdef CONFIG_X86_32 | |
f8c2ee22 HH |
863 | die("Oops", regs, error_code); |
864 | bust_spinlocks(0); | |
865 | do_exit(SIGKILL); | |
fd40d6e3 | 866 | #else |
22f5991c JB |
867 | if (__die("Oops", regs, error_code)) |
868 | regs = NULL; | |
1da177e4 LT |
869 | /* Executive summary in case the body of the oops scrolled away */ |
870 | printk(KERN_EMERG "CR2: %016lx\n", address); | |
22f5991c | 871 | oops_end(flags, regs, SIGKILL); |
f8c2ee22 | 872 | #endif |
1da177e4 LT |
873 | |
874 | /* | |
875 | * We ran out of memory, or some other thing happened to us that made | |
876 | * us unable to handle the page fault gracefully. | |
877 | */ | |
878 | out_of_memory: | |
879 | up_read(&mm->mmap_sem); | |
f8c2ee22 HH |
880 | if (is_global_init(tsk)) { |
881 | yield(); | |
fd40d6e3 | 882 | #ifdef CONFIG_X86_32 |
f8c2ee22 HH |
883 | down_read(&mm->mmap_sem); |
884 | goto survive; | |
f8c2ee22 | 885 | #else |
1da177e4 | 886 | goto again; |
f8c2ee22 | 887 | #endif |
fd40d6e3 HH |
888 | } |
889 | ||
1da177e4 | 890 | printk("VM: killing process %s\n", tsk->comm); |
318aa296 | 891 | if (error_code & PF_USER) |
021daae2 | 892 | do_group_exit(SIGKILL); |
1da177e4 LT |
893 | goto no_context; |
894 | ||
895 | do_sigbus: | |
896 | up_read(&mm->mmap_sem); | |
897 | ||
898 | /* Kernel mode? Handle exceptions or die */ | |
66c58156 | 899 | if (!(error_code & PF_USER)) |
1da177e4 | 900 | goto no_context; |
f8c2ee22 HH |
901 | #ifdef CONFIG_X86_32 |
902 | /* User space => ok to do another page fault */ | |
903 | if (is_prefetch(regs, address, error_code)) | |
904 | return; | |
905 | #endif | |
1da177e4 LT |
906 | tsk->thread.cr2 = address; |
907 | tsk->thread.error_code = error_code; | |
908 | tsk->thread.trap_no = 14; | |
c4aba4a8 | 909 | force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk); |
1da177e4 | 910 | } |
9e43e1b7 | 911 | |
8c914cb7 | 912 | DEFINE_SPINLOCK(pgd_lock); |
2bff7383 | 913 | LIST_HEAD(pgd_list); |
8c914cb7 JB |
914 | |
915 | void vmalloc_sync_all(void) | |
916 | { | |
1156e098 HH |
917 | #ifdef CONFIG_X86_32 |
918 | /* | |
919 | * Note that races in the updates of insync and start aren't | |
920 | * problematic: insync can only get set bits added, and updates to | |
921 | * start are only improving performance (without affecting correctness | |
922 | * if undone). | |
923 | */ | |
924 | static DECLARE_BITMAP(insync, PTRS_PER_PGD); | |
925 | static unsigned long start = TASK_SIZE; | |
926 | unsigned long address; | |
927 | ||
928 | if (SHARED_KERNEL_PMD) | |
929 | return; | |
930 | ||
931 | BUILD_BUG_ON(TASK_SIZE & ~PGDIR_MASK); | |
932 | for (address = start; address >= TASK_SIZE; address += PGDIR_SIZE) { | |
933 | if (!test_bit(pgd_index(address), insync)) { | |
934 | unsigned long flags; | |
935 | struct page *page; | |
936 | ||
937 | spin_lock_irqsave(&pgd_lock, flags); | |
e3ed910d | 938 | list_for_each_entry(page, &pgd_list, lru) { |
1156e098 | 939 | if (!vmalloc_sync_one(page_address(page), |
e3ed910d | 940 | address)) |
1156e098 | 941 | break; |
e3ed910d | 942 | } |
1156e098 HH |
943 | spin_unlock_irqrestore(&pgd_lock, flags); |
944 | if (!page) | |
945 | set_bit(pgd_index(address), insync); | |
946 | } | |
947 | if (address == start && test_bit(pgd_index(address), insync)) | |
948 | start = address + PGDIR_SIZE; | |
949 | } | |
950 | #else /* CONFIG_X86_64 */ | |
6f4d368e HH |
951 | /* |
952 | * Note that races in the updates of insync and start aren't | |
953 | * problematic: insync can only get set bits added, and updates to | |
954 | * start are only improving performance (without affecting correctness | |
955 | * if undone). | |
956 | */ | |
8c914cb7 JB |
957 | static DECLARE_BITMAP(insync, PTRS_PER_PGD); |
958 | static unsigned long start = VMALLOC_START & PGDIR_MASK; | |
959 | unsigned long address; | |
960 | ||
961 | for (address = start; address <= VMALLOC_END; address += PGDIR_SIZE) { | |
962 | if (!test_bit(pgd_index(address), insync)) { | |
963 | const pgd_t *pgd_ref = pgd_offset_k(address); | |
58d5d0d8 | 964 | unsigned long flags; |
8c914cb7 JB |
965 | struct page *page; |
966 | ||
967 | if (pgd_none(*pgd_ref)) | |
968 | continue; | |
58d5d0d8 | 969 | spin_lock_irqsave(&pgd_lock, flags); |
2bff7383 | 970 | list_for_each_entry(page, &pgd_list, lru) { |
8c914cb7 JB |
971 | pgd_t *pgd; |
972 | pgd = (pgd_t *)page_address(page) + pgd_index(address); | |
973 | if (pgd_none(*pgd)) | |
974 | set_pgd(pgd, *pgd_ref); | |
975 | else | |
46a82b2d | 976 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
8c914cb7 | 977 | } |
58d5d0d8 | 978 | spin_unlock_irqrestore(&pgd_lock, flags); |
8c914cb7 JB |
979 | set_bit(pgd_index(address), insync); |
980 | } | |
981 | if (address == start) | |
982 | start = address + PGDIR_SIZE; | |
983 | } | |
1156e098 | 984 | #endif |
8c914cb7 | 985 | } |