fault.c 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533
  1. /*
  2. * PowerPC version
  3. * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
  4. *
  5. * Derived from "arch/i386/mm/fault.c"
  6. * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
  7. *
  8. * Modified by Cort Dougan and Paul Mackerras.
  9. *
  10. * Modified for PPC64 by Dave Engebretsen (engebret@ibm.com)
  11. *
  12. * This program is free software; you can redistribute it and/or
  13. * modify it under the terms of the GNU General Public License
  14. * as published by the Free Software Foundation; either version
  15. * 2 of the License, or (at your option) any later version.
  16. */
  17. #include <linux/signal.h>
  18. #include <linux/sched.h>
  19. #include <linux/kernel.h>
  20. #include <linux/errno.h>
  21. #include <linux/string.h>
  22. #include <linux/types.h>
  23. #include <linux/ptrace.h>
  24. #include <linux/mman.h>
  25. #include <linux/mm.h>
  26. #include <linux/interrupt.h>
  27. #include <linux/highmem.h>
  28. #include <linux/module.h>
  29. #include <linux/kprobes.h>
  30. #include <linux/kdebug.h>
  31. #include <linux/perf_event.h>
  32. #include <linux/magic.h>
  33. #include <linux/ratelimit.h>
  34. #include <linux/context_tracking.h>
  35. #include <asm/firmware.h>
  36. #include <asm/page.h>
  37. #include <asm/pgtable.h>
  38. #include <asm/mmu.h>
  39. #include <asm/mmu_context.h>
  40. #include <asm/uaccess.h>
  41. #include <asm/tlbflush.h>
  42. #include <asm/siginfo.h>
  43. #include <asm/debug.h>
  44. #include <mm/mmu_decl.h>
  45. #include "icswx.h"
  46. #ifdef CONFIG_KPROBES
  47. static inline int notify_page_fault(struct pt_regs *regs)
  48. {
  49. int ret = 0;
  50. /* kprobe_running() needs smp_processor_id() */
  51. if (!user_mode(regs)) {
  52. preempt_disable();
  53. if (kprobe_running() && kprobe_fault_handler(regs, 11))
  54. ret = 1;
  55. preempt_enable();
  56. }
  57. return ret;
  58. }
  59. #else
  60. static inline int notify_page_fault(struct pt_regs *regs)
  61. {
  62. return 0;
  63. }
  64. #endif
  65. /*
  66. * Check whether the instruction at regs->nip is a store using
  67. * an update addressing form which will update r1.
  68. */
  69. static int store_updates_sp(struct pt_regs *regs)
  70. {
  71. unsigned int inst;
  72. if (get_user(inst, (unsigned int __user *)regs->nip))
  73. return 0;
  74. /* check for 1 in the rA field */
  75. if (((inst >> 16) & 0x1f) != 1)
  76. return 0;
  77. /* check major opcode */
  78. switch (inst >> 26) {
  79. case 37: /* stwu */
  80. case 39: /* stbu */
  81. case 45: /* sthu */
  82. case 53: /* stfsu */
  83. case 55: /* stfdu */
  84. return 1;
  85. case 62: /* std or stdu */
  86. return (inst & 3) == 1;
  87. case 31:
  88. /* check minor opcode */
  89. switch ((inst >> 1) & 0x3ff) {
  90. case 181: /* stdux */
  91. case 183: /* stwux */
  92. case 247: /* stbux */
  93. case 439: /* sthux */
  94. case 695: /* stfsux */
  95. case 759: /* stfdux */
  96. return 1;
  97. }
  98. }
  99. return 0;
  100. }
  101. /*
  102. * do_page_fault error handling helpers
  103. */
  104. #define MM_FAULT_RETURN 0
  105. #define MM_FAULT_CONTINUE -1
  106. #define MM_FAULT_ERR(sig) (sig)
  107. static int do_sigbus(struct pt_regs *regs, unsigned long address)
  108. {
  109. siginfo_t info;
  110. up_read(&current->mm->mmap_sem);
  111. if (user_mode(regs)) {
  112. current->thread.trap_nr = BUS_ADRERR;
  113. info.si_signo = SIGBUS;
  114. info.si_errno = 0;
  115. info.si_code = BUS_ADRERR;
  116. info.si_addr = (void __user *)address;
  117. force_sig_info(SIGBUS, &info, current);
  118. return MM_FAULT_RETURN;
  119. }
  120. return MM_FAULT_ERR(SIGBUS);
  121. }
  122. static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault)
  123. {
  124. /*
  125. * Pagefault was interrupted by SIGKILL. We have no reason to
  126. * continue the pagefault.
  127. */
  128. if (fatal_signal_pending(current)) {
  129. /*
  130. * If we have retry set, the mmap semaphore will have
  131. * alrady been released in __lock_page_or_retry(). Else
  132. * we release it now.
  133. */
  134. if (!(fault & VM_FAULT_RETRY))
  135. up_read(&current->mm->mmap_sem);
  136. /* Coming from kernel, we need to deal with uaccess fixups */
  137. if (user_mode(regs))
  138. return MM_FAULT_RETURN;
  139. return MM_FAULT_ERR(SIGKILL);
  140. }
  141. /* No fault: be happy */
  142. if (!(fault & VM_FAULT_ERROR))
  143. return MM_FAULT_CONTINUE;
  144. /* Out of memory */
  145. if (fault & VM_FAULT_OOM) {
  146. up_read(&current->mm->mmap_sem);
  147. /*
  148. * We ran out of memory, or some other thing happened to us that
  149. * made us unable to handle the page fault gracefully.
  150. */
  151. if (!user_mode(regs))
  152. return MM_FAULT_ERR(SIGKILL);
  153. pagefault_out_of_memory();
  154. return MM_FAULT_RETURN;
  155. }
  156. /* Bus error. x86 handles HWPOISON here, we'll add this if/when
  157. * we support the feature in HW
  158. */
  159. if (fault & VM_FAULT_SIGBUS)
  160. return do_sigbus(regs, addr);
  161. /* We don't understand the fault code, this is fatal */
  162. BUG();
  163. return MM_FAULT_CONTINUE;
  164. }
  165. /*
  166. * For 600- and 800-family processors, the error_code parameter is DSISR
  167. * for a data fault, SRR1 for an instruction fault. For 400-family processors
  168. * the error_code parameter is ESR for a data fault, 0 for an instruction
  169. * fault.
  170. * For 64-bit processors, the error_code parameter is
  171. * - DSISR for a non-SLB data access fault,
  172. * - SRR1 & 0x08000000 for a non-SLB instruction access fault
  173. * - 0 any SLB fault.
  174. *
  175. * The return value is 0 if the fault was handled, or the signal
  176. * number if this is a kernel fault that can't be handled here.
  177. */
  178. int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
  179. unsigned long error_code)
  180. {
  181. enum ctx_state prev_state = exception_enter();
  182. struct vm_area_struct * vma;
  183. struct mm_struct *mm = current->mm;
  184. unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
  185. int code = SEGV_MAPERR;
  186. int is_write = 0;
  187. int trap = TRAP(regs);
  188. int is_exec = trap == 0x400;
  189. int fault;
  190. int rc = 0;
  191. #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
  192. /*
  193. * Fortunately the bit assignments in SRR1 for an instruction
  194. * fault and DSISR for a data fault are mostly the same for the
  195. * bits we are interested in. But there are some bits which
  196. * indicate errors in DSISR but can validly be set in SRR1.
  197. */
  198. if (trap == 0x400)
  199. error_code &= 0x48200000;
  200. else
  201. is_write = error_code & DSISR_ISSTORE;
  202. #else
  203. is_write = error_code & ESR_DST;
  204. #endif /* CONFIG_4xx || CONFIG_BOOKE */
  205. if (is_write)
  206. flags |= FAULT_FLAG_WRITE;
  207. #ifdef CONFIG_PPC_ICSWX
  208. /*
  209. * we need to do this early because this "data storage
  210. * interrupt" does not update the DAR/DEAR so we don't want to
  211. * look at it
  212. */
  213. if (error_code & ICSWX_DSI_UCT) {
  214. rc = acop_handle_fault(regs, address, error_code);
  215. if (rc)
  216. goto bail;
  217. }
  218. #endif /* CONFIG_PPC_ICSWX */
  219. if (notify_page_fault(regs))
  220. goto bail;
  221. if (unlikely(debugger_fault_handler(regs)))
  222. goto bail;
  223. /* On a kernel SLB miss we can only check for a valid exception entry */
  224. if (!user_mode(regs) && (address >= TASK_SIZE)) {
  225. rc = SIGSEGV;
  226. goto bail;
  227. }
  228. #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE) || \
  229. defined(CONFIG_PPC_BOOK3S_64))
  230. if (error_code & DSISR_DABRMATCH) {
  231. /* breakpoint match */
  232. do_break(regs, address, error_code);
  233. goto bail;
  234. }
  235. #endif
  236. /* We restore the interrupt state now */
  237. if (!arch_irq_disabled_regs(regs))
  238. local_irq_enable();
  239. if (in_atomic() || mm == NULL) {
  240. if (!user_mode(regs)) {
  241. rc = SIGSEGV;
  242. goto bail;
  243. }
  244. /* in_atomic() in user mode is really bad,
  245. as is current->mm == NULL. */
  246. printk(KERN_EMERG "Page fault in user mode with "
  247. "in_atomic() = %d mm = %p\n", in_atomic(), mm);
  248. printk(KERN_EMERG "NIP = %lx MSR = %lx\n",
  249. regs->nip, regs->msr);
  250. die("Weird page fault", regs, SIGSEGV);
  251. }
  252. perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
  253. /* When running in the kernel we expect faults to occur only to
  254. * addresses in user space. All other faults represent errors in the
  255. * kernel and should generate an OOPS. Unfortunately, in the case of an
  256. * erroneous fault occurring in a code path which already holds mmap_sem
  257. * we will deadlock attempting to validate the fault against the
  258. * address space. Luckily the kernel only validly references user
  259. * space from well defined areas of code, which are listed in the
  260. * exceptions table.
  261. *
  262. * As the vast majority of faults will be valid we will only perform
  263. * the source reference check when there is a possibility of a deadlock.
  264. * Attempt to lock the address space, if we cannot we then validate the
  265. * source. If this is invalid we can skip the address space check,
  266. * thus avoiding the deadlock.
  267. */
  268. if (!down_read_trylock(&mm->mmap_sem)) {
  269. if (!user_mode(regs) && !search_exception_tables(regs->nip))
  270. goto bad_area_nosemaphore;
  271. retry:
  272. down_read(&mm->mmap_sem);
  273. } else {
  274. /*
  275. * The above down_read_trylock() might have succeeded in
  276. * which case we'll have missed the might_sleep() from
  277. * down_read():
  278. */
  279. might_sleep();
  280. }
  281. vma = find_vma(mm, address);
  282. if (!vma)
  283. goto bad_area;
  284. if (vma->vm_start <= address)
  285. goto good_area;
  286. if (!(vma->vm_flags & VM_GROWSDOWN))
  287. goto bad_area;
  288. /*
  289. * N.B. The POWER/Open ABI allows programs to access up to
  290. * 288 bytes below the stack pointer.
  291. * The kernel signal delivery code writes up to about 1.5kB
  292. * below the stack pointer (r1) before decrementing it.
  293. * The exec code can write slightly over 640kB to the stack
  294. * before setting the user r1. Thus we allow the stack to
  295. * expand to 1MB without further checks.
  296. */
  297. if (address + 0x100000 < vma->vm_end) {
  298. /* get user regs even if this fault is in kernel mode */
  299. struct pt_regs *uregs = current->thread.regs;
  300. if (uregs == NULL)
  301. goto bad_area;
  302. /*
  303. * A user-mode access to an address a long way below
  304. * the stack pointer is only valid if the instruction
  305. * is one which would update the stack pointer to the
  306. * address accessed if the instruction completed,
  307. * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
  308. * (or the byte, halfword, float or double forms).
  309. *
  310. * If we don't check this then any write to the area
  311. * between the last mapped region and the stack will
  312. * expand the stack rather than segfaulting.
  313. */
  314. if (address + 2048 < uregs->gpr[1]
  315. && (!user_mode(regs) || !store_updates_sp(regs)))
  316. goto bad_area;
  317. }
  318. if (expand_stack(vma, address))
  319. goto bad_area;
  320. good_area:
  321. code = SEGV_ACCERR;
  322. #if defined(CONFIG_6xx)
  323. if (error_code & 0x95700000)
  324. /* an error such as lwarx to I/O controller space,
  325. address matching DABR, eciwx, etc. */
  326. goto bad_area;
  327. #endif /* CONFIG_6xx */
  328. #if defined(CONFIG_8xx)
  329. /* 8xx sometimes need to load a invalid/non-present TLBs.
  330. * These must be invalidated separately as linux mm don't.
  331. */
  332. if (error_code & 0x40000000) /* no translation? */
  333. _tlbil_va(address, 0, 0, 0);
  334. /* The MPC8xx seems to always set 0x80000000, which is
  335. * "undefined". Of those that can be set, this is the only
  336. * one which seems bad.
  337. */
  338. if (error_code & 0x10000000)
  339. /* Guarded storage error. */
  340. goto bad_area;
  341. #endif /* CONFIG_8xx */
  342. if (is_exec) {
  343. #ifdef CONFIG_PPC_STD_MMU
  344. /* Protection fault on exec go straight to failure on
  345. * Hash based MMUs as they either don't support per-page
  346. * execute permission, or if they do, it's handled already
  347. * at the hash level. This test would probably have to
  348. * be removed if we change the way this works to make hash
  349. * processors use the same I/D cache coherency mechanism
  350. * as embedded.
  351. */
  352. if (error_code & DSISR_PROTFAULT)
  353. goto bad_area;
  354. #endif /* CONFIG_PPC_STD_MMU */
  355. /*
  356. * Allow execution from readable areas if the MMU does not
  357. * provide separate controls over reading and executing.
  358. *
  359. * Note: That code used to not be enabled for 4xx/BookE.
  360. * It is now as I/D cache coherency for these is done at
  361. * set_pte_at() time and I see no reason why the test
  362. * below wouldn't be valid on those processors. This -may-
  363. * break programs compiled with a really old ABI though.
  364. */
  365. if (!(vma->vm_flags & VM_EXEC) &&
  366. (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
  367. !(vma->vm_flags & (VM_READ | VM_WRITE))))
  368. goto bad_area;
  369. /* a write */
  370. } else if (is_write) {
  371. if (!(vma->vm_flags & VM_WRITE))
  372. goto bad_area;
  373. /* a read */
  374. } else {
  375. /* protection fault */
  376. if (error_code & 0x08000000)
  377. goto bad_area;
  378. if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
  379. goto bad_area;
  380. }
  381. /*
  382. * If for any reason at all we couldn't handle the fault,
  383. * make sure we exit gracefully rather than endlessly redo
  384. * the fault.
  385. */
  386. fault = handle_mm_fault(mm, vma, address, flags);
  387. if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) {
  388. rc = mm_fault_error(regs, address, fault);
  389. if (rc >= MM_FAULT_RETURN)
  390. goto bail;
  391. else
  392. rc = 0;
  393. }
  394. /*
  395. * Major/minor page fault accounting is only done on the
  396. * initial attempt. If we go through a retry, it is extremely
  397. * likely that the page will be found in page cache at that point.
  398. */
  399. if (flags & FAULT_FLAG_ALLOW_RETRY) {
  400. if (fault & VM_FAULT_MAJOR) {
  401. current->maj_flt++;
  402. perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
  403. regs, address);
  404. #ifdef CONFIG_PPC_SMLPAR
  405. if (firmware_has_feature(FW_FEATURE_CMO)) {
  406. preempt_disable();
  407. get_lppaca()->page_ins += (1 << PAGE_FACTOR);
  408. preempt_enable();
  409. }
  410. #endif /* CONFIG_PPC_SMLPAR */
  411. } else {
  412. current->min_flt++;
  413. perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
  414. regs, address);
  415. }
  416. if (fault & VM_FAULT_RETRY) {
  417. /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
  418. * of starvation. */
  419. flags &= ~FAULT_FLAG_ALLOW_RETRY;
  420. flags |= FAULT_FLAG_TRIED;
  421. goto retry;
  422. }
  423. }
  424. up_read(&mm->mmap_sem);
  425. goto bail;
  426. bad_area:
  427. up_read(&mm->mmap_sem);
  428. bad_area_nosemaphore:
  429. /* User mode accesses cause a SIGSEGV */
  430. if (user_mode(regs)) {
  431. _exception(SIGSEGV, regs, code, address);
  432. goto bail;
  433. }
  434. if (is_exec && (error_code & DSISR_PROTFAULT))
  435. printk_ratelimited(KERN_CRIT "kernel tried to execute NX-protected"
  436. " page (%lx) - exploit attempt? (uid: %d)\n",
  437. address, from_kuid(&init_user_ns, current_uid()));
  438. rc = SIGSEGV;
  439. bail:
  440. exception_exit(prev_state);
  441. return rc;
  442. }
  443. /*
  444. * bad_page_fault is called when we have a bad access from the kernel.
  445. * It is called from the DSI and ISI handlers in head.S and from some
  446. * of the procedures in traps.c.
  447. */
  448. void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
  449. {
  450. const struct exception_table_entry *entry;
  451. unsigned long *stackend;
  452. /* Are we prepared to handle this fault? */
  453. if ((entry = search_exception_tables(regs->nip)) != NULL) {
  454. regs->nip = entry->fixup;
  455. return;
  456. }
  457. /* kernel has accessed a bad area */
  458. switch (regs->trap) {
  459. case 0x300:
  460. case 0x380:
  461. printk(KERN_ALERT "Unable to handle kernel paging request for "
  462. "data at address 0x%08lx\n", regs->dar);
  463. break;
  464. case 0x400:
  465. case 0x480:
  466. printk(KERN_ALERT "Unable to handle kernel paging request for "
  467. "instruction fetch\n");
  468. break;
  469. default:
  470. printk(KERN_ALERT "Unable to handle kernel paging request for "
  471. "unknown fault\n");
  472. break;
  473. }
  474. printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
  475. regs->nip);
  476. stackend = end_of_stack(current);
  477. if (current != &init_task && *stackend != STACK_END_MAGIC)
  478. printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
  479. die("Kernel access of bad area", regs, sig);
  480. }