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> | |
0fd0e3da | 13 | #include <linux/mmiotrace.h> |
1da177e4 LT |
14 | #include <linux/mman.h> |
15 | #include <linux/mm.h> | |
16 | #include <linux/smp.h> | |
1da177e4 LT |
17 | #include <linux/interrupt.h> |
18 | #include <linux/init.h> | |
19 | #include <linux/tty.h> | |
20 | #include <linux/vt_kern.h> /* For unblank_screen() */ | |
21 | #include <linux/compiler.h> | |
c61e211d HH |
22 | #include <linux/highmem.h> |
23 | #include <linux/bootmem.h> /* for max_low_pfn */ | |
1eeb66a1 | 24 | #include <linux/vmalloc.h> |
1da177e4 | 25 | #include <linux/module.h> |
0f2fbdcb | 26 | #include <linux/kprobes.h> |
ab2bf0c1 | 27 | #include <linux/uaccess.h> |
1eeb66a1 | 28 | #include <linux/kdebug.h> |
1da177e4 LT |
29 | |
30 | #include <asm/system.h> | |
c61e211d HH |
31 | #include <asm/desc.h> |
32 | #include <asm/segment.h> | |
1da177e4 LT |
33 | #include <asm/pgalloc.h> |
34 | #include <asm/smp.h> | |
35 | #include <asm/tlbflush.h> | |
36 | #include <asm/proto.h> | |
1da177e4 | 37 | #include <asm-generic/sections.h> |
1da177e4 | 38 | |
33cb5243 HH |
39 | /* |
40 | * Page fault error code bits | |
41 | * bit 0 == 0 means no page found, 1 means protection fault | |
42 | * bit 1 == 0 means read, 1 means write | |
43 | * bit 2 == 0 means kernel, 1 means user-mode | |
44 | * bit 3 == 1 means use of reserved bit detected | |
45 | * bit 4 == 1 means fault was an instruction fetch | |
46 | */ | |
8a19da7b | 47 | #define PF_PROT (1<<0) |
66c58156 | 48 | #define PF_WRITE (1<<1) |
8a19da7b IM |
49 | #define PF_USER (1<<2) |
50 | #define PF_RSVD (1<<3) | |
66c58156 AK |
51 | #define PF_INSTR (1<<4) |
52 | ||
0fd0e3da | 53 | static inline int kmmio_fault(struct pt_regs *regs, unsigned long addr) |
86069782 | 54 | { |
10c43d2e | 55 | #ifdef CONFIG_MMIOTRACE_HOOKS |
0fd0e3da PP |
56 | if (unlikely(is_kmmio_active())) |
57 | if (kmmio_handler(regs, addr) == 1) | |
58 | return -1; | |
86069782 | 59 | #endif |
0fd0e3da | 60 | return 0; |
86069782 PP |
61 | } |
62 | ||
74a0b576 | 63 | static inline int notify_page_fault(struct pt_regs *regs) |
1bd858a5 | 64 | { |
33cb5243 | 65 | #ifdef CONFIG_KPROBES |
74a0b576 CH |
66 | int ret = 0; |
67 | ||
68 | /* kprobe_running() needs smp_processor_id() */ | |
f8c2ee22 | 69 | if (!user_mode_vm(regs)) { |
74a0b576 CH |
70 | preempt_disable(); |
71 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
72 | ret = 1; | |
73 | preempt_enable(); | |
74 | } | |
1bd858a5 | 75 | |
74a0b576 | 76 | return ret; |
74a0b576 | 77 | #else |
74a0b576 | 78 | return 0; |
74a0b576 | 79 | #endif |
33cb5243 | 80 | } |
1bd858a5 | 81 | |
1dc85be0 HH |
82 | /* |
83 | * X86_32 | |
84 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
85 | * Check that here and ignore it. | |
86 | * | |
87 | * X86_64 | |
88 | * Sometimes the CPU reports invalid exceptions on prefetch. | |
89 | * Check that here and ignore it. | |
90 | * | |
91 | * Opcode checker based on code by Richard Brunner | |
92 | */ | |
93 | static int is_prefetch(struct pt_regs *regs, unsigned long addr, | |
94 | unsigned long error_code) | |
33cb5243 | 95 | { |
ab2bf0c1 | 96 | unsigned char *instr; |
1da177e4 | 97 | int scan_more = 1; |
33cb5243 | 98 | int prefetch = 0; |
f1290ec9 | 99 | unsigned char *max_instr; |
1da177e4 | 100 | |
3085354d IM |
101 | /* |
102 | * If it was a exec (instruction fetch) fault on NX page, then | |
103 | * do not ignore the fault: | |
104 | */ | |
66c58156 | 105 | if (error_code & PF_INSTR) |
1da177e4 | 106 | return 0; |
1dc85be0 | 107 | |
f2857ce9 | 108 | instr = (unsigned char *)convert_ip_to_linear(current, regs); |
f1290ec9 | 109 | max_instr = instr + 15; |
1da177e4 | 110 | |
76381fee | 111 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE) |
1da177e4 LT |
112 | return 0; |
113 | ||
33cb5243 | 114 | while (scan_more && instr < max_instr) { |
1da177e4 LT |
115 | unsigned char opcode; |
116 | unsigned char instr_hi; | |
117 | unsigned char instr_lo; | |
118 | ||
ab2bf0c1 | 119 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 120 | break; |
1da177e4 | 121 | |
33cb5243 HH |
122 | instr_hi = opcode & 0xf0; |
123 | instr_lo = opcode & 0x0f; | |
1da177e4 LT |
124 | instr++; |
125 | ||
33cb5243 | 126 | switch (instr_hi) { |
1da177e4 LT |
127 | case 0x20: |
128 | case 0x30: | |
33cb5243 HH |
129 | /* |
130 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
131 | * In X86_64 long mode, the CPU will signal invalid | |
132 | * opcode if some of these prefixes are present so | |
133 | * X86_64 will never get here anyway | |
134 | */ | |
1da177e4 LT |
135 | scan_more = ((instr_lo & 7) == 0x6); |
136 | break; | |
33cb5243 | 137 | #ifdef CONFIG_X86_64 |
1da177e4 | 138 | case 0x40: |
33cb5243 HH |
139 | /* |
140 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
141 | * Need to figure out under what instruction mode the | |
142 | * instruction was issued. Could check the LDT for lm, | |
143 | * but for now it's good enough to assume that long | |
144 | * mode only uses well known segments or kernel. | |
145 | */ | |
76381fee | 146 | scan_more = (!user_mode(regs)) || (regs->cs == __USER_CS); |
1da177e4 | 147 | break; |
33cb5243 | 148 | #endif |
1da177e4 LT |
149 | case 0x60: |
150 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
151 | scan_more = (instr_lo & 0xC) == 0x4; | |
33cb5243 | 152 | break; |
1da177e4 | 153 | case 0xF0: |
1dc85be0 | 154 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ |
1da177e4 | 155 | scan_more = !instr_lo || (instr_lo>>1) == 1; |
33cb5243 | 156 | break; |
1da177e4 LT |
157 | case 0x00: |
158 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
159 | scan_more = 0; | |
f2857ce9 | 160 | |
ab2bf0c1 | 161 | if (probe_kernel_address(instr, opcode)) |
1da177e4 LT |
162 | break; |
163 | prefetch = (instr_lo == 0xF) && | |
164 | (opcode == 0x0D || opcode == 0x18); | |
33cb5243 | 165 | break; |
1da177e4 LT |
166 | default: |
167 | scan_more = 0; | |
168 | break; | |
33cb5243 | 169 | } |
1da177e4 LT |
170 | } |
171 | return prefetch; | |
172 | } | |
173 | ||
c4aba4a8 HH |
174 | static void force_sig_info_fault(int si_signo, int si_code, |
175 | unsigned long address, struct task_struct *tsk) | |
176 | { | |
177 | siginfo_t info; | |
178 | ||
179 | info.si_signo = si_signo; | |
180 | info.si_errno = 0; | |
181 | info.si_code = si_code; | |
182 | info.si_addr = (void __user *)address; | |
183 | force_sig_info(si_signo, &info, tsk); | |
184 | } | |
185 | ||
1156e098 | 186 | #ifdef CONFIG_X86_64 |
33cb5243 HH |
187 | static int bad_address(void *p) |
188 | { | |
1da177e4 | 189 | unsigned long dummy; |
ab2bf0c1 | 190 | return probe_kernel_address((unsigned long *)p, dummy); |
33cb5243 | 191 | } |
1156e098 | 192 | #endif |
1da177e4 | 193 | |
cae30f82 | 194 | static void dump_pagetable(unsigned long address) |
1da177e4 | 195 | { |
1156e098 HH |
196 | #ifdef CONFIG_X86_32 |
197 | __typeof__(pte_val(__pte(0))) page; | |
198 | ||
199 | page = read_cr3(); | |
200 | page = ((__typeof__(page) *) __va(page))[address >> PGDIR_SHIFT]; | |
201 | #ifdef CONFIG_X86_PAE | |
202 | printk("*pdpt = %016Lx ", page); | |
203 | if ((page >> PAGE_SHIFT) < max_low_pfn | |
204 | && page & _PAGE_PRESENT) { | |
205 | page &= PAGE_MASK; | |
206 | page = ((__typeof__(page) *) __va(page))[(address >> PMD_SHIFT) | |
207 | & (PTRS_PER_PMD - 1)]; | |
208 | printk(KERN_CONT "*pde = %016Lx ", page); | |
209 | page &= ~_PAGE_NX; | |
210 | } | |
211 | #else | |
212 | printk("*pde = %08lx ", page); | |
213 | #endif | |
214 | ||
215 | /* | |
216 | * We must not directly access the pte in the highpte | |
217 | * case if the page table is located in highmem. | |
218 | * And let's rather not kmap-atomic the pte, just in case | |
219 | * it's allocated already. | |
220 | */ | |
221 | if ((page >> PAGE_SHIFT) < max_low_pfn | |
222 | && (page & _PAGE_PRESENT) | |
223 | && !(page & _PAGE_PSE)) { | |
224 | page &= PAGE_MASK; | |
225 | page = ((__typeof__(page) *) __va(page))[(address >> PAGE_SHIFT) | |
226 | & (PTRS_PER_PTE - 1)]; | |
227 | printk("*pte = %0*Lx ", sizeof(page)*2, (u64)page); | |
228 | } | |
229 | ||
230 | printk("\n"); | |
231 | #else /* CONFIG_X86_64 */ | |
1da177e4 LT |
232 | pgd_t *pgd; |
233 | pud_t *pud; | |
234 | pmd_t *pmd; | |
235 | pte_t *pte; | |
236 | ||
f51c9452 | 237 | pgd = (pgd_t *)read_cr3(); |
1da177e4 | 238 | |
33cb5243 | 239 | pgd = __va((unsigned long)pgd & PHYSICAL_PAGE_MASK); |
1da177e4 | 240 | pgd += pgd_index(address); |
1da177e4 | 241 | if (bad_address(pgd)) goto bad; |
d646bce4 | 242 | printk("PGD %lx ", pgd_val(*pgd)); |
33cb5243 | 243 | if (!pgd_present(*pgd)) goto ret; |
1da177e4 | 244 | |
d2ae5b5f | 245 | pud = pud_offset(pgd, address); |
1da177e4 LT |
246 | if (bad_address(pud)) goto bad; |
247 | printk("PUD %lx ", pud_val(*pud)); | |
b5360222 AK |
248 | if (!pud_present(*pud) || pud_large(*pud)) |
249 | goto ret; | |
1da177e4 LT |
250 | |
251 | pmd = pmd_offset(pud, address); | |
252 | if (bad_address(pmd)) goto bad; | |
253 | printk("PMD %lx ", pmd_val(*pmd)); | |
b1992df3 | 254 | if (!pmd_present(*pmd) || pmd_large(*pmd)) goto ret; |
1da177e4 LT |
255 | |
256 | pte = pte_offset_kernel(pmd, address); | |
257 | if (bad_address(pte)) goto bad; | |
33cb5243 | 258 | printk("PTE %lx", pte_val(*pte)); |
1da177e4 LT |
259 | ret: |
260 | printk("\n"); | |
261 | return; | |
262 | bad: | |
263 | printk("BAD\n"); | |
1156e098 HH |
264 | #endif |
265 | } | |
266 | ||
267 | #ifdef CONFIG_X86_32 | |
268 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
269 | { | |
270 | unsigned index = pgd_index(address); | |
271 | pgd_t *pgd_k; | |
272 | pud_t *pud, *pud_k; | |
273 | pmd_t *pmd, *pmd_k; | |
274 | ||
275 | pgd += index; | |
276 | pgd_k = init_mm.pgd + index; | |
277 | ||
278 | if (!pgd_present(*pgd_k)) | |
279 | return NULL; | |
280 | ||
281 | /* | |
282 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
283 | * and redundant with the set_pmd() on non-PAE. As would | |
284 | * set_pud. | |
285 | */ | |
286 | ||
287 | pud = pud_offset(pgd, address); | |
288 | pud_k = pud_offset(pgd_k, address); | |
289 | if (!pud_present(*pud_k)) | |
290 | return NULL; | |
291 | ||
292 | pmd = pmd_offset(pud, address); | |
293 | pmd_k = pmd_offset(pud_k, address); | |
294 | if (!pmd_present(*pmd_k)) | |
295 | return NULL; | |
296 | if (!pmd_present(*pmd)) { | |
297 | set_pmd(pmd, *pmd_k); | |
298 | arch_flush_lazy_mmu_mode(); | |
299 | } else | |
300 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); | |
301 | return pmd_k; | |
1da177e4 | 302 | } |
1156e098 | 303 | #endif |
1da177e4 | 304 | |
1dc85be0 | 305 | #ifdef CONFIG_X86_64 |
33cb5243 | 306 | static const char errata93_warning[] = |
1da177e4 LT |
307 | KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n" |
308 | KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n" | |
309 | KERN_ERR "******* Please consider a BIOS update.\n" | |
310 | KERN_ERR "******* Disabling USB legacy in the BIOS may also help.\n"; | |
fdfe8aa8 | 311 | #endif |
1da177e4 LT |
312 | |
313 | /* Workaround for K8 erratum #93 & buggy BIOS. | |
314 | BIOS SMM functions are required to use a specific workaround | |
33cb5243 HH |
315 | to avoid corruption of the 64bit RIP register on C stepping K8. |
316 | A lot of BIOS that didn't get tested properly miss this. | |
1da177e4 LT |
317 | The OS sees this as a page fault with the upper 32bits of RIP cleared. |
318 | Try to work around it here. | |
fdfe8aa8 HH |
319 | Note we only handle faults in kernel here. |
320 | Does nothing for X86_32 | |
321 | */ | |
33cb5243 | 322 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 323 | { |
fdfe8aa8 | 324 | #ifdef CONFIG_X86_64 |
1da177e4 | 325 | static int warned; |
65ea5b03 | 326 | if (address != regs->ip) |
1da177e4 | 327 | return 0; |
33cb5243 | 328 | if ((address >> 32) != 0) |
1da177e4 LT |
329 | return 0; |
330 | address |= 0xffffffffUL << 32; | |
33cb5243 HH |
331 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
332 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
1da177e4 | 333 | if (!warned) { |
33cb5243 | 334 | printk(errata93_warning); |
1da177e4 LT |
335 | warned = 1; |
336 | } | |
65ea5b03 | 337 | regs->ip = address; |
1da177e4 LT |
338 | return 1; |
339 | } | |
fdfe8aa8 | 340 | #endif |
1da177e4 | 341 | return 0; |
33cb5243 | 342 | } |
1da177e4 | 343 | |
35f3266f HH |
344 | /* |
345 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps to illegal | |
346 | * addresses >4GB. We catch this in the page fault handler because these | |
347 | * addresses are not reachable. Just detect this case and return. Any code | |
348 | * segment in LDT is compatibility mode. | |
349 | */ | |
350 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
351 | { | |
352 | #ifdef CONFIG_X86_64 | |
353 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && | |
354 | (address >> 32)) | |
355 | return 1; | |
356 | #endif | |
357 | return 0; | |
358 | } | |
359 | ||
29caf2f9 HH |
360 | void do_invalid_op(struct pt_regs *, unsigned long); |
361 | ||
362 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) | |
363 | { | |
364 | #ifdef CONFIG_X86_F00F_BUG | |
365 | unsigned long nr; | |
366 | /* | |
367 | * Pentium F0 0F C7 C8 bug workaround. | |
368 | */ | |
369 | if (boot_cpu_data.f00f_bug) { | |
370 | nr = (address - idt_descr.address) >> 3; | |
371 | ||
372 | if (nr == 6) { | |
373 | do_invalid_op(regs, 0); | |
374 | return 1; | |
375 | } | |
376 | } | |
377 | #endif | |
378 | return 0; | |
379 | } | |
380 | ||
b3279c7f HH |
381 | static void show_fault_oops(struct pt_regs *regs, unsigned long error_code, |
382 | unsigned long address) | |
383 | { | |
1156e098 HH |
384 | #ifdef CONFIG_X86_32 |
385 | if (!oops_may_print()) | |
386 | return; | |
fd40d6e3 | 387 | #endif |
1156e098 HH |
388 | |
389 | #ifdef CONFIG_X86_PAE | |
390 | if (error_code & PF_INSTR) { | |
93809be8 | 391 | unsigned int level; |
1156e098 HH |
392 | pte_t *pte = lookup_address(address, &level); |
393 | ||
394 | if (pte && pte_present(*pte) && !pte_exec(*pte)) | |
395 | printk(KERN_CRIT "kernel tried to execute " | |
396 | "NX-protected page - exploit attempt? " | |
397 | "(uid: %d)\n", current->uid); | |
398 | } | |
399 | #endif | |
1156e098 | 400 | |
19f0dda9 | 401 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 402 | if (address < PAGE_SIZE) |
19f0dda9 | 403 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 404 | else |
19f0dda9 | 405 | printk(KERN_CONT "paging request"); |
f294a8ce | 406 | printk(KERN_CONT " at %p\n", (void *) address); |
19f0dda9 | 407 | printk(KERN_ALERT "IP:"); |
b3279c7f HH |
408 | printk_address(regs->ip, 1); |
409 | dump_pagetable(address); | |
410 | } | |
411 | ||
1156e098 | 412 | #ifdef CONFIG_X86_64 |
1da177e4 LT |
413 | static noinline void pgtable_bad(unsigned long address, struct pt_regs *regs, |
414 | unsigned long error_code) | |
415 | { | |
1209140c | 416 | unsigned long flags = oops_begin(); |
6e3f3617 | 417 | struct task_struct *tsk; |
1209140c | 418 | |
1da177e4 LT |
419 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
420 | current->comm, address); | |
421 | dump_pagetable(address); | |
6e3f3617 JB |
422 | tsk = current; |
423 | tsk->thread.cr2 = address; | |
424 | tsk->thread.trap_no = 14; | |
425 | tsk->thread.error_code = error_code; | |
22f5991c JB |
426 | if (__die("Bad pagetable", regs, error_code)) |
427 | regs = NULL; | |
428 | oops_end(flags, regs, SIGKILL); | |
1da177e4 | 429 | } |
1156e098 | 430 | #endif |
1da177e4 | 431 | |
d8b57bb7 TG |
432 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
433 | { | |
434 | if ((error_code & PF_WRITE) && !pte_write(*pte)) | |
435 | return 0; | |
436 | if ((error_code & PF_INSTR) && !pte_exec(*pte)) | |
437 | return 0; | |
438 | ||
439 | return 1; | |
440 | } | |
441 | ||
5b727a3b JF |
442 | /* |
443 | * Handle a spurious fault caused by a stale TLB entry. This allows | |
444 | * us to lazily refresh the TLB when increasing the permissions of a | |
445 | * kernel page (RO -> RW or NX -> X). Doing it eagerly is very | |
446 | * expensive since that implies doing a full cross-processor TLB | |
447 | * flush, even if no stale TLB entries exist on other processors. | |
448 | * There are no security implications to leaving a stale TLB when | |
449 | * increasing the permissions on a page. | |
450 | */ | |
451 | static int spurious_fault(unsigned long address, | |
452 | unsigned long error_code) | |
453 | { | |
454 | pgd_t *pgd; | |
455 | pud_t *pud; | |
456 | pmd_t *pmd; | |
457 | pte_t *pte; | |
458 | ||
459 | /* Reserved-bit violation or user access to kernel space? */ | |
460 | if (error_code & (PF_USER | PF_RSVD)) | |
461 | return 0; | |
462 | ||
463 | pgd = init_mm.pgd + pgd_index(address); | |
464 | if (!pgd_present(*pgd)) | |
465 | return 0; | |
466 | ||
467 | pud = pud_offset(pgd, address); | |
468 | if (!pud_present(*pud)) | |
469 | return 0; | |
470 | ||
d8b57bb7 TG |
471 | if (pud_large(*pud)) |
472 | return spurious_fault_check(error_code, (pte_t *) pud); | |
473 | ||
5b727a3b JF |
474 | pmd = pmd_offset(pud, address); |
475 | if (!pmd_present(*pmd)) | |
476 | return 0; | |
477 | ||
d8b57bb7 TG |
478 | if (pmd_large(*pmd)) |
479 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
480 | ||
5b727a3b JF |
481 | pte = pte_offset_kernel(pmd, address); |
482 | if (!pte_present(*pte)) | |
483 | return 0; | |
484 | ||
d8b57bb7 | 485 | return spurious_fault_check(error_code, pte); |
5b727a3b JF |
486 | } |
487 | ||
1da177e4 | 488 | /* |
f8c2ee22 HH |
489 | * X86_32 |
490 | * Handle a fault on the vmalloc or module mapping area | |
491 | * | |
492 | * X86_64 | |
f95190b2 | 493 | * Handle a fault on the vmalloc area |
3b9ba4d5 AK |
494 | * |
495 | * This assumes no large pages in there. | |
1da177e4 LT |
496 | */ |
497 | static int vmalloc_fault(unsigned long address) | |
498 | { | |
fdfe8aa8 HH |
499 | #ifdef CONFIG_X86_32 |
500 | unsigned long pgd_paddr; | |
501 | pmd_t *pmd_k; | |
502 | pte_t *pte_k; | |
b29c701d HN |
503 | |
504 | /* Make sure we are in vmalloc area */ | |
505 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
506 | return -1; | |
507 | ||
fdfe8aa8 HH |
508 | /* |
509 | * Synchronize this task's top level page-table | |
510 | * with the 'reference' page table. | |
511 | * | |
512 | * Do _not_ use "current" here. We might be inside | |
513 | * an interrupt in the middle of a task switch.. | |
514 | */ | |
515 | pgd_paddr = read_cr3(); | |
516 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
517 | if (!pmd_k) | |
518 | return -1; | |
519 | pte_k = pte_offset_kernel(pmd_k, address); | |
520 | if (!pte_present(*pte_k)) | |
521 | return -1; | |
522 | return 0; | |
523 | #else | |
1da177e4 LT |
524 | pgd_t *pgd, *pgd_ref; |
525 | pud_t *pud, *pud_ref; | |
526 | pmd_t *pmd, *pmd_ref; | |
527 | pte_t *pte, *pte_ref; | |
528 | ||
cf89ec92 HH |
529 | /* Make sure we are in vmalloc area */ |
530 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
531 | return -1; | |
532 | ||
1da177e4 LT |
533 | /* Copy kernel mappings over when needed. This can also |
534 | happen within a race in page table update. In the later | |
535 | case just flush. */ | |
536 | ||
537 | pgd = pgd_offset(current->mm ?: &init_mm, address); | |
538 | pgd_ref = pgd_offset_k(address); | |
539 | if (pgd_none(*pgd_ref)) | |
540 | return -1; | |
541 | if (pgd_none(*pgd)) | |
542 | set_pgd(pgd, *pgd_ref); | |
8c914cb7 | 543 | else |
46a82b2d | 544 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
1da177e4 LT |
545 | |
546 | /* Below here mismatches are bugs because these lower tables | |
547 | are shared */ | |
548 | ||
549 | pud = pud_offset(pgd, address); | |
550 | pud_ref = pud_offset(pgd_ref, address); | |
551 | if (pud_none(*pud_ref)) | |
552 | return -1; | |
46a82b2d | 553 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) |
1da177e4 LT |
554 | BUG(); |
555 | pmd = pmd_offset(pud, address); | |
556 | pmd_ref = pmd_offset(pud_ref, address); | |
557 | if (pmd_none(*pmd_ref)) | |
558 | return -1; | |
559 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) | |
560 | BUG(); | |
561 | pte_ref = pte_offset_kernel(pmd_ref, address); | |
562 | if (!pte_present(*pte_ref)) | |
563 | return -1; | |
564 | pte = pte_offset_kernel(pmd, address); | |
3b9ba4d5 AK |
565 | /* Don't use pte_page here, because the mappings can point |
566 | outside mem_map, and the NUMA hash lookup cannot handle | |
567 | that. */ | |
568 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
1da177e4 | 569 | BUG(); |
1da177e4 | 570 | return 0; |
fdfe8aa8 | 571 | #endif |
1da177e4 LT |
572 | } |
573 | ||
abd4f750 | 574 | int show_unhandled_signals = 1; |
1da177e4 LT |
575 | |
576 | /* | |
577 | * This routine handles page faults. It determines the address, | |
578 | * and the problem, and then passes it off to one of the appropriate | |
579 | * routines. | |
1da177e4 | 580 | */ |
f8c2ee22 HH |
581 | #ifdef CONFIG_X86_64 |
582 | asmlinkage | |
583 | #endif | |
584 | void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code) | |
1da177e4 LT |
585 | { |
586 | struct task_struct *tsk; | |
587 | struct mm_struct *mm; | |
33cb5243 | 588 | struct vm_area_struct *vma; |
1da177e4 | 589 | unsigned long address; |
f8c2ee22 HH |
590 | int write, si_code; |
591 | int fault; | |
592 | #ifdef CONFIG_X86_64 | |
1209140c | 593 | unsigned long flags; |
f8c2ee22 | 594 | #endif |
1da177e4 | 595 | |
143a5d32 PZ |
596 | /* |
597 | * We can fault from pretty much anywhere, with unknown IRQ state. | |
598 | */ | |
599 | trace_hardirqs_fixup(); | |
600 | ||
a9ba9a3b AV |
601 | tsk = current; |
602 | mm = tsk->mm; | |
603 | prefetchw(&mm->mmap_sem); | |
604 | ||
1da177e4 | 605 | /* get the address */ |
f51c9452 | 606 | address = read_cr2(); |
1da177e4 | 607 | |
c4aba4a8 | 608 | si_code = SEGV_MAPERR; |
1da177e4 | 609 | |
608566b4 HH |
610 | if (notify_page_fault(regs)) |
611 | return; | |
0fd0e3da | 612 | if (unlikely(kmmio_fault(regs, address))) |
86069782 | 613 | return; |
1da177e4 LT |
614 | |
615 | /* | |
616 | * We fault-in kernel-space virtual memory on-demand. The | |
617 | * 'reference' page table is init_mm.pgd. | |
618 | * | |
619 | * NOTE! We MUST NOT take any locks for this case. We may | |
620 | * be in an interrupt or a critical region, and should | |
621 | * only copy the information from the master page table, | |
622 | * nothing more. | |
623 | * | |
624 | * This verifies that the fault happens in kernel space | |
625 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 626 | * protection error (error_code & 9) == 0. |
1da177e4 | 627 | */ |
f8c2ee22 HH |
628 | #ifdef CONFIG_X86_32 |
629 | if (unlikely(address >= TASK_SIZE)) { | |
cf89ec92 HH |
630 | #else |
631 | if (unlikely(address >= TASK_SIZE64)) { | |
632 | #endif | |
f8c2ee22 HH |
633 | if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) && |
634 | vmalloc_fault(address) >= 0) | |
635 | return; | |
5b727a3b JF |
636 | |
637 | /* Can handle a stale RO->RW TLB */ | |
638 | if (spurious_fault(address, error_code)) | |
639 | return; | |
640 | ||
f8c2ee22 HH |
641 | /* |
642 | * Don't take the mm semaphore here. If we fixup a prefetch | |
643 | * fault we could otherwise deadlock. | |
644 | */ | |
645 | goto bad_area_nosemaphore; | |
646 | } | |
647 | ||
cf89ec92 HH |
648 | |
649 | #ifdef CONFIG_X86_32 | |
f8c2ee22 HH |
650 | /* It's safe to allow irq's after cr2 has been saved and the vmalloc |
651 | fault has been handled. */ | |
6b6891f9 | 652 | if (regs->flags & (X86_EFLAGS_IF | X86_VM_MASK)) |
f8c2ee22 HH |
653 | local_irq_enable(); |
654 | ||
655 | /* | |
656 | * If we're in an interrupt, have no user context or are running in an | |
657 | * atomic region then we must not take the fault. | |
658 | */ | |
659 | if (in_atomic() || !mm) | |
660 | goto bad_area_nosemaphore; | |
661 | #else /* CONFIG_X86_64 */ | |
65ea5b03 | 662 | if (likely(regs->flags & X86_EFLAGS_IF)) |
8c914cb7 JB |
663 | local_irq_enable(); |
664 | ||
66c58156 | 665 | if (unlikely(error_code & PF_RSVD)) |
1da177e4 LT |
666 | pgtable_bad(address, regs, error_code); |
667 | ||
668 | /* | |
33cb5243 HH |
669 | * If we're in an interrupt, have no user context or are running in an |
670 | * atomic region then we must not take the fault. | |
1da177e4 LT |
671 | */ |
672 | if (unlikely(in_atomic() || !mm)) | |
673 | goto bad_area_nosemaphore; | |
674 | ||
dbe3ed1c LT |
675 | /* |
676 | * User-mode registers count as a user access even for any | |
677 | * potential system fault or CPU buglet. | |
678 | */ | |
679 | if (user_mode_vm(regs)) | |
680 | error_code |= PF_USER; | |
f8c2ee22 HH |
681 | again: |
682 | #endif | |
1da177e4 LT |
683 | /* When running in the kernel we expect faults to occur only to |
684 | * addresses in user space. All other faults represent errors in the | |
676b1855 | 685 | * kernel and should generate an OOPS. Unfortunately, in the case of an |
80f7228b | 686 | * erroneous fault occurring in a code path which already holds mmap_sem |
1da177e4 LT |
687 | * we will deadlock attempting to validate the fault against the |
688 | * address space. Luckily the kernel only validly references user | |
689 | * space from well defined areas of code, which are listed in the | |
690 | * exceptions table. | |
691 | * | |
692 | * As the vast majority of faults will be valid we will only perform | |
676b1855 | 693 | * the source reference check when there is a possibility of a deadlock. |
1da177e4 LT |
694 | * Attempt to lock the address space, if we cannot we then validate the |
695 | * source. If this is invalid we can skip the address space check, | |
696 | * thus avoiding the deadlock. | |
697 | */ | |
698 | if (!down_read_trylock(&mm->mmap_sem)) { | |
66c58156 | 699 | if ((error_code & PF_USER) == 0 && |
65ea5b03 | 700 | !search_exception_tables(regs->ip)) |
1da177e4 LT |
701 | goto bad_area_nosemaphore; |
702 | down_read(&mm->mmap_sem); | |
703 | } | |
704 | ||
705 | vma = find_vma(mm, address); | |
706 | if (!vma) | |
707 | goto bad_area; | |
f8c2ee22 | 708 | if (vma->vm_start <= address) |
1da177e4 LT |
709 | goto good_area; |
710 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
711 | goto bad_area; | |
33cb5243 | 712 | if (error_code & PF_USER) { |
6f4d368e HH |
713 | /* |
714 | * Accessing the stack below %sp is always a bug. | |
715 | * The large cushion allows instructions like enter | |
716 | * and pusha to work. ("enter $65535,$31" pushes | |
717 | * 32 pointers and then decrements %sp by 65535.) | |
03fdc2c2 | 718 | */ |
65ea5b03 | 719 | if (address + 65536 + 32 * sizeof(unsigned long) < regs->sp) |
1da177e4 LT |
720 | goto bad_area; |
721 | } | |
722 | if (expand_stack(vma, address)) | |
723 | goto bad_area; | |
724 | /* | |
725 | * Ok, we have a good vm_area for this memory access, so | |
726 | * we can handle it.. | |
727 | */ | |
728 | good_area: | |
c4aba4a8 | 729 | si_code = SEGV_ACCERR; |
1da177e4 | 730 | write = 0; |
66c58156 | 731 | switch (error_code & (PF_PROT|PF_WRITE)) { |
33cb5243 HH |
732 | default: /* 3: write, present */ |
733 | /* fall through */ | |
734 | case PF_WRITE: /* write, not present */ | |
735 | if (!(vma->vm_flags & VM_WRITE)) | |
736 | goto bad_area; | |
737 | write++; | |
738 | break; | |
739 | case PF_PROT: /* read, present */ | |
740 | goto bad_area; | |
741 | case 0: /* read, not present */ | |
742 | if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))) | |
1da177e4 | 743 | goto bad_area; |
1da177e4 LT |
744 | } |
745 | ||
f8c2ee22 HH |
746 | #ifdef CONFIG_X86_32 |
747 | survive: | |
748 | #endif | |
1da177e4 LT |
749 | /* |
750 | * If for any reason at all we couldn't handle the fault, | |
751 | * make sure we exit gracefully rather than endlessly redo | |
752 | * the fault. | |
753 | */ | |
83c54070 NP |
754 | fault = handle_mm_fault(mm, vma, address, write); |
755 | if (unlikely(fault & VM_FAULT_ERROR)) { | |
756 | if (fault & VM_FAULT_OOM) | |
757 | goto out_of_memory; | |
758 | else if (fault & VM_FAULT_SIGBUS) | |
759 | goto do_sigbus; | |
760 | BUG(); | |
1da177e4 | 761 | } |
83c54070 NP |
762 | if (fault & VM_FAULT_MAJOR) |
763 | tsk->maj_flt++; | |
764 | else | |
765 | tsk->min_flt++; | |
d729ab35 HH |
766 | |
767 | #ifdef CONFIG_X86_32 | |
768 | /* | |
769 | * Did it hit the DOS screen memory VA from vm86 mode? | |
770 | */ | |
771 | if (v8086_mode(regs)) { | |
772 | unsigned long bit = (address - 0xA0000) >> PAGE_SHIFT; | |
773 | if (bit < 32) | |
774 | tsk->thread.screen_bitmap |= 1 << bit; | |
775 | } | |
776 | #endif | |
1da177e4 LT |
777 | up_read(&mm->mmap_sem); |
778 | return; | |
779 | ||
780 | /* | |
781 | * Something tried to access memory that isn't in our memory map.. | |
782 | * Fix it, but check if it's kernel or user first.. | |
783 | */ | |
784 | bad_area: | |
785 | up_read(&mm->mmap_sem); | |
786 | ||
787 | bad_area_nosemaphore: | |
1da177e4 | 788 | /* User mode accesses just cause a SIGSEGV */ |
66c58156 | 789 | if (error_code & PF_USER) { |
e5e3c84b SR |
790 | /* |
791 | * It's possible to have interrupts off here. | |
792 | */ | |
793 | local_irq_enable(); | |
794 | ||
1156e098 HH |
795 | /* |
796 | * Valid to do another page fault here because this one came | |
797 | * from user space. | |
798 | */ | |
1da177e4 LT |
799 | if (is_prefetch(regs, address, error_code)) |
800 | return; | |
801 | ||
35f3266f | 802 | if (is_errata100(regs, address)) |
1da177e4 LT |
803 | return; |
804 | ||
abd4f750 MAS |
805 | if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) && |
806 | printk_ratelimit()) { | |
1da177e4 | 807 | printk( |
f294a8ce | 808 | "%s%s[%d]: segfault at %lx ip %p sp %p error %lx", |
6f4d368e | 809 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, |
f294a8ce VN |
810 | tsk->comm, task_pid_nr(tsk), address, |
811 | (void *) regs->ip, (void *) regs->sp, error_code); | |
03252919 AK |
812 | print_vma_addr(" in ", regs->ip); |
813 | printk("\n"); | |
1da177e4 | 814 | } |
33cb5243 | 815 | |
1da177e4 LT |
816 | tsk->thread.cr2 = address; |
817 | /* Kernel addresses are always protection faults */ | |
818 | tsk->thread.error_code = error_code | (address >= TASK_SIZE); | |
819 | tsk->thread.trap_no = 14; | |
c4aba4a8 | 820 | force_sig_info_fault(SIGSEGV, si_code, address, tsk); |
1da177e4 LT |
821 | return; |
822 | } | |
823 | ||
29caf2f9 HH |
824 | if (is_f00f_bug(regs, address)) |
825 | return; | |
826 | ||
1da177e4 | 827 | no_context: |
1da177e4 | 828 | /* Are we prepared to handle this kernel fault? */ |
33cb5243 | 829 | if (fixup_exception(regs)) |
1da177e4 | 830 | return; |
1da177e4 | 831 | |
33cb5243 | 832 | /* |
f8c2ee22 HH |
833 | * X86_32 |
834 | * Valid to do another page fault here, because if this fault | |
835 | * had been triggered by is_prefetch fixup_exception would have | |
836 | * handled it. | |
837 | * | |
838 | * X86_64 | |
1da177e4 LT |
839 | * Hall of shame of CPU/BIOS bugs. |
840 | */ | |
33cb5243 HH |
841 | if (is_prefetch(regs, address, error_code)) |
842 | return; | |
1da177e4 LT |
843 | |
844 | if (is_errata93(regs, address)) | |
33cb5243 | 845 | return; |
1da177e4 LT |
846 | |
847 | /* | |
848 | * Oops. The kernel tried to access some bad page. We'll have to | |
849 | * terminate things with extreme prejudice. | |
850 | */ | |
5394f80f JF |
851 | check_for_bios_corruption(); |
852 | ||
f8c2ee22 HH |
853 | #ifdef CONFIG_X86_32 |
854 | bust_spinlocks(1); | |
fd40d6e3 HH |
855 | #else |
856 | flags = oops_begin(); | |
857 | #endif | |
f8c2ee22 HH |
858 | |
859 | show_fault_oops(regs, error_code, address); | |
1da177e4 | 860 | |
f8c2ee22 HH |
861 | tsk->thread.cr2 = address; |
862 | tsk->thread.trap_no = 14; | |
863 | tsk->thread.error_code = error_code; | |
fd40d6e3 HH |
864 | |
865 | #ifdef CONFIG_X86_32 | |
f8c2ee22 HH |
866 | die("Oops", regs, error_code); |
867 | bust_spinlocks(0); | |
868 | do_exit(SIGKILL); | |
fd40d6e3 | 869 | #else |
22f5991c JB |
870 | if (__die("Oops", regs, error_code)) |
871 | regs = NULL; | |
1da177e4 LT |
872 | /* Executive summary in case the body of the oops scrolled away */ |
873 | printk(KERN_EMERG "CR2: %016lx\n", address); | |
22f5991c | 874 | oops_end(flags, regs, SIGKILL); |
f8c2ee22 | 875 | #endif |
1da177e4 LT |
876 | |
877 | /* | |
878 | * We ran out of memory, or some other thing happened to us that made | |
879 | * us unable to handle the page fault gracefully. | |
880 | */ | |
881 | out_of_memory: | |
882 | up_read(&mm->mmap_sem); | |
f8c2ee22 HH |
883 | if (is_global_init(tsk)) { |
884 | yield(); | |
fd40d6e3 | 885 | #ifdef CONFIG_X86_32 |
f8c2ee22 HH |
886 | down_read(&mm->mmap_sem); |
887 | goto survive; | |
f8c2ee22 | 888 | #else |
1da177e4 | 889 | goto again; |
f8c2ee22 | 890 | #endif |
fd40d6e3 HH |
891 | } |
892 | ||
1da177e4 | 893 | printk("VM: killing process %s\n", tsk->comm); |
318aa296 | 894 | if (error_code & PF_USER) |
021daae2 | 895 | do_group_exit(SIGKILL); |
1da177e4 LT |
896 | goto no_context; |
897 | ||
898 | do_sigbus: | |
899 | up_read(&mm->mmap_sem); | |
900 | ||
901 | /* Kernel mode? Handle exceptions or die */ | |
66c58156 | 902 | if (!(error_code & PF_USER)) |
1da177e4 | 903 | goto no_context; |
f8c2ee22 HH |
904 | #ifdef CONFIG_X86_32 |
905 | /* User space => ok to do another page fault */ | |
906 | if (is_prefetch(regs, address, error_code)) | |
907 | return; | |
908 | #endif | |
1da177e4 LT |
909 | tsk->thread.cr2 = address; |
910 | tsk->thread.error_code = error_code; | |
911 | tsk->thread.trap_no = 14; | |
c4aba4a8 | 912 | force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk); |
1da177e4 | 913 | } |
9e43e1b7 | 914 | |
8c914cb7 | 915 | DEFINE_SPINLOCK(pgd_lock); |
2bff7383 | 916 | LIST_HEAD(pgd_list); |
8c914cb7 JB |
917 | |
918 | void vmalloc_sync_all(void) | |
919 | { | |
1156e098 | 920 | #ifdef CONFIG_X86_32 |
67350a5c | 921 | unsigned long start = VMALLOC_START & PGDIR_MASK; |
1156e098 HH |
922 | unsigned long address; |
923 | ||
924 | if (SHARED_KERNEL_PMD) | |
925 | return; | |
926 | ||
927 | BUILD_BUG_ON(TASK_SIZE & ~PGDIR_MASK); | |
928 | for (address = start; address >= TASK_SIZE; address += PGDIR_SIZE) { | |
67350a5c JF |
929 | unsigned long flags; |
930 | struct page *page; | |
931 | ||
932 | spin_lock_irqsave(&pgd_lock, flags); | |
933 | list_for_each_entry(page, &pgd_list, lru) { | |
934 | if (!vmalloc_sync_one(page_address(page), | |
935 | address)) | |
936 | break; | |
1156e098 | 937 | } |
67350a5c | 938 | spin_unlock_irqrestore(&pgd_lock, flags); |
1156e098 HH |
939 | } |
940 | #else /* CONFIG_X86_64 */ | |
67350a5c | 941 | unsigned long start = VMALLOC_START & PGDIR_MASK; |
8c914cb7 JB |
942 | unsigned long address; |
943 | ||
944 | for (address = start; address <= VMALLOC_END; address += PGDIR_SIZE) { | |
67350a5c JF |
945 | const pgd_t *pgd_ref = pgd_offset_k(address); |
946 | unsigned long flags; | |
947 | struct page *page; | |
948 | ||
949 | if (pgd_none(*pgd_ref)) | |
950 | continue; | |
951 | spin_lock_irqsave(&pgd_lock, flags); | |
952 | list_for_each_entry(page, &pgd_list, lru) { | |
953 | pgd_t *pgd; | |
954 | pgd = (pgd_t *)page_address(page) + pgd_index(address); | |
955 | if (pgd_none(*pgd)) | |
956 | set_pgd(pgd, *pgd_ref); | |
957 | else | |
958 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); | |
8c914cb7 | 959 | } |
67350a5c | 960 | spin_unlock_irqrestore(&pgd_lock, flags); |
8c914cb7 | 961 | } |
1156e098 | 962 | #endif |
8c914cb7 | 963 | } |