fault.c 8.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343
  1. /*
  2. * This file is subject to the terms and conditions of the GNU General Public
  3. * License. See the file "COPYING" in the main directory of this archive
  4. * for more details.
  5. *
  6. * Copyright (C) 1995 - 2000 by Ralf Baechle
  7. */
  8. #include <linux/context_tracking.h>
  9. #include <linux/signal.h>
  10. #include <linux/sched.h>
  11. #include <linux/interrupt.h>
  12. #include <linux/kernel.h>
  13. #include <linux/errno.h>
  14. #include <linux/string.h>
  15. #include <linux/types.h>
  16. #include <linux/ptrace.h>
  17. #include <linux/ratelimit.h>
  18. #include <linux/mman.h>
  19. #include <linux/mm.h>
  20. #include <linux/smp.h>
  21. #include <linux/module.h>
  22. #include <linux/kprobes.h>
  23. #include <linux/perf_event.h>
  24. #include <linux/uaccess.h>
  25. #include <asm/branch.h>
  26. #include <asm/mmu_context.h>
  27. #include <asm/ptrace.h>
  28. #include <asm/highmem.h> /* For VMALLOC_END */
  29. #include <linux/kdebug.h>
  30. int show_unhandled_signals = 1;
  31. /*
  32. * This routine handles page faults. It determines the address,
  33. * and the problem, and then passes it off to one of the appropriate
  34. * routines.
  35. */
  36. static void __kprobes __do_page_fault(struct pt_regs *regs, unsigned long write,
  37. unsigned long address)
  38. {
  39. struct vm_area_struct * vma = NULL;
  40. struct task_struct *tsk = current;
  41. struct mm_struct *mm = tsk->mm;
  42. const int field = sizeof(unsigned long) * 2;
  43. siginfo_t info;
  44. int fault;
  45. unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
  46. static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
  47. #if 0
  48. printk("Cpu%d[%s:%d:%0*lx:%ld:%0*lx]\n", raw_smp_processor_id(),
  49. current->comm, current->pid, field, address, write,
  50. field, regs->cp0_epc);
  51. #endif
  52. #ifdef CONFIG_KPROBES
  53. /*
  54. * This is to notify the fault handler of the kprobes. The
  55. * exception code is redundant as it is also carried in REGS,
  56. * but we pass it anyhow.
  57. */
  58. if (notify_die(DIE_PAGE_FAULT, "page fault", regs, -1,
  59. (regs->cp0_cause >> 2) & 0x1f, SIGSEGV) == NOTIFY_STOP)
  60. return;
  61. #endif
  62. info.si_code = SEGV_MAPERR;
  63. /*
  64. * We fault-in kernel-space virtual memory on-demand. The
  65. * 'reference' page table is init_mm.pgd.
  66. *
  67. * NOTE! We MUST NOT take any locks for this case. We may
  68. * be in an interrupt or a critical region, and should
  69. * only copy the information from the master page table,
  70. * nothing more.
  71. */
  72. #ifdef CONFIG_64BIT
  73. # define VMALLOC_FAULT_TARGET no_context
  74. #else
  75. # define VMALLOC_FAULT_TARGET vmalloc_fault
  76. #endif
  77. if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END))
  78. goto VMALLOC_FAULT_TARGET;
  79. #ifdef MODULE_START
  80. if (unlikely(address >= MODULE_START && address < MODULE_END))
  81. goto VMALLOC_FAULT_TARGET;
  82. #endif
  83. /*
  84. * If we're in an interrupt or have no user
  85. * context, we must not take the fault..
  86. */
  87. if (faulthandler_disabled() || !mm)
  88. goto bad_area_nosemaphore;
  89. if (user_mode(regs))
  90. flags |= FAULT_FLAG_USER;
  91. retry:
  92. down_read(&mm->mmap_sem);
  93. vma = find_vma(mm, address);
  94. if (!vma)
  95. goto bad_area;
  96. if (vma->vm_start <= address)
  97. goto good_area;
  98. if (!(vma->vm_flags & VM_GROWSDOWN))
  99. goto bad_area;
  100. if (expand_stack(vma, address))
  101. goto bad_area;
  102. /*
  103. * Ok, we have a good vm_area for this memory access, so
  104. * we can handle it..
  105. */
  106. good_area:
  107. info.si_code = SEGV_ACCERR;
  108. if (write) {
  109. if (!(vma->vm_flags & VM_WRITE))
  110. goto bad_area;
  111. flags |= FAULT_FLAG_WRITE;
  112. } else {
  113. if (cpu_has_rixi) {
  114. if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {
  115. #if 0
  116. pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] XI violation\n",
  117. raw_smp_processor_id(),
  118. current->comm, current->pid,
  119. field, address, write,
  120. field, regs->cp0_epc);
  121. #endif
  122. goto bad_area;
  123. }
  124. if (!(vma->vm_flags & VM_READ)) {
  125. #if 0
  126. pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] RI violation\n",
  127. raw_smp_processor_id(),
  128. current->comm, current->pid,
  129. field, address, write,
  130. field, regs->cp0_epc);
  131. #endif
  132. goto bad_area;
  133. }
  134. } else {
  135. if (!(vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)))
  136. goto bad_area;
  137. }
  138. }
  139. /*
  140. * If for any reason at all we couldn't handle the fault,
  141. * make sure we exit gracefully rather than endlessly redo
  142. * the fault.
  143. */
  144. fault = handle_mm_fault(mm, vma, address, flags);
  145. if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
  146. return;
  147. perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
  148. if (unlikely(fault & VM_FAULT_ERROR)) {
  149. if (fault & VM_FAULT_OOM)
  150. goto out_of_memory;
  151. else if (fault & VM_FAULT_SIGSEGV)
  152. goto bad_area;
  153. else if (fault & VM_FAULT_SIGBUS)
  154. goto do_sigbus;
  155. BUG();
  156. }
  157. if (flags & FAULT_FLAG_ALLOW_RETRY) {
  158. if (fault & VM_FAULT_MAJOR) {
  159. perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
  160. regs, address);
  161. tsk->maj_flt++;
  162. } else {
  163. perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
  164. regs, address);
  165. tsk->min_flt++;
  166. }
  167. if (fault & VM_FAULT_RETRY) {
  168. flags &= ~FAULT_FLAG_ALLOW_RETRY;
  169. flags |= FAULT_FLAG_TRIED;
  170. /*
  171. * No need to up_read(&mm->mmap_sem) as we would
  172. * have already released it in __lock_page_or_retry
  173. * in mm/filemap.c.
  174. */
  175. goto retry;
  176. }
  177. }
  178. up_read(&mm->mmap_sem);
  179. return;
  180. /*
  181. * Something tried to access memory that isn't in our memory map..
  182. * Fix it, but check if it's kernel or user first..
  183. */
  184. bad_area:
  185. up_read(&mm->mmap_sem);
  186. bad_area_nosemaphore:
  187. /* User mode accesses just cause a SIGSEGV */
  188. if (user_mode(regs)) {
  189. tsk->thread.cp0_badvaddr = address;
  190. tsk->thread.error_code = write;
  191. if (show_unhandled_signals &&
  192. unhandled_signal(tsk, SIGSEGV) &&
  193. __ratelimit(&ratelimit_state)) {
  194. pr_info("\ndo_page_fault(): sending SIGSEGV to %s for invalid %s %0*lx",
  195. tsk->comm,
  196. write ? "write access to" : "read access from",
  197. field, address);
  198. pr_info("epc = %0*lx in", field,
  199. (unsigned long) regs->cp0_epc);
  200. print_vma_addr(" ", regs->cp0_epc);
  201. pr_info("ra = %0*lx in", field,
  202. (unsigned long) regs->regs[31]);
  203. print_vma_addr(" ", regs->regs[31]);
  204. pr_info("\n");
  205. }
  206. info.si_signo = SIGSEGV;
  207. info.si_errno = 0;
  208. /* info.si_code has been set above */
  209. info.si_addr = (void __user *) address;
  210. force_sig_info(SIGSEGV, &info, tsk);
  211. return;
  212. }
  213. no_context:
  214. /* Are we prepared to handle this kernel fault? */
  215. if (fixup_exception(regs)) {
  216. current->thread.cp0_baduaddr = address;
  217. return;
  218. }
  219. /*
  220. * Oops. The kernel tried to access some bad page. We'll have to
  221. * terminate things with extreme prejudice.
  222. */
  223. bust_spinlocks(1);
  224. printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at "
  225. "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n",
  226. raw_smp_processor_id(), field, address, field, regs->cp0_epc,
  227. field, regs->regs[31]);
  228. die("Oops", regs);
  229. out_of_memory:
  230. /*
  231. * We ran out of memory, call the OOM killer, and return the userspace
  232. * (which will retry the fault, or kill us if we got oom-killed).
  233. */
  234. up_read(&mm->mmap_sem);
  235. if (!user_mode(regs))
  236. goto no_context;
  237. pagefault_out_of_memory();
  238. return;
  239. do_sigbus:
  240. up_read(&mm->mmap_sem);
  241. /* Kernel mode? Handle exceptions or die */
  242. if (!user_mode(regs))
  243. goto no_context;
  244. else
  245. /*
  246. * Send a sigbus, regardless of whether we were in kernel
  247. * or user mode.
  248. */
  249. #if 0
  250. printk("do_page_fault() #3: sending SIGBUS to %s for "
  251. "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
  252. tsk->comm,
  253. write ? "write access to" : "read access from",
  254. field, address,
  255. field, (unsigned long) regs->cp0_epc,
  256. field, (unsigned long) regs->regs[31]);
  257. #endif
  258. tsk->thread.cp0_badvaddr = address;
  259. info.si_signo = SIGBUS;
  260. info.si_errno = 0;
  261. info.si_code = BUS_ADRERR;
  262. info.si_addr = (void __user *) address;
  263. force_sig_info(SIGBUS, &info, tsk);
  264. return;
  265. #ifndef CONFIG_64BIT
  266. vmalloc_fault:
  267. {
  268. /*
  269. * Synchronize this task's top level page-table
  270. * with the 'reference' page table.
  271. *
  272. * Do _not_ use "tsk" here. We might be inside
  273. * an interrupt in the middle of a task switch..
  274. */
  275. int offset = __pgd_offset(address);
  276. pgd_t *pgd, *pgd_k;
  277. pud_t *pud, *pud_k;
  278. pmd_t *pmd, *pmd_k;
  279. pte_t *pte_k;
  280. pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset;
  281. pgd_k = init_mm.pgd + offset;
  282. if (!pgd_present(*pgd_k))
  283. goto no_context;
  284. set_pgd(pgd, *pgd_k);
  285. pud = pud_offset(pgd, address);
  286. pud_k = pud_offset(pgd_k, address);
  287. if (!pud_present(*pud_k))
  288. goto no_context;
  289. pmd = pmd_offset(pud, address);
  290. pmd_k = pmd_offset(pud_k, address);
  291. if (!pmd_present(*pmd_k))
  292. goto no_context;
  293. set_pmd(pmd, *pmd_k);
  294. pte_k = pte_offset_kernel(pmd_k, address);
  295. if (!pte_present(*pte_k))
  296. goto no_context;
  297. return;
  298. }
  299. #endif
  300. }
  301. asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
  302. unsigned long write, unsigned long address)
  303. {
  304. enum ctx_state prev_state;
  305. prev_state = exception_enter();
  306. __do_page_fault(regs, write, address);
  307. exception_exit(prev_state);
  308. }